You are on page 1of 366

TRADING SYSTEMS THAT WORK
Building and Evaluating Effective Trading Systems

THOMAS STRIDSMAN

McGraw-Hill
New York San Francisco Washington, D.C. Auckland Bogota Caracas Lisbon London Madrid Mexico City Milan Montreal New Delhi San Juan Singapore Sydney Tokyo Toronto

Library of Congress Cataloging-in-Publication Data Stridsman, Thomas. Trading-systems that work : building and evaluating effective trading systems / by Thomas Stridsman. p. cm. 1SBN0-O7-13598O-X 1. Investment analysis. HG4529.S77 332.6—<lc21 00-055008 2000 2. Portfolio management. I. Title.

McGraw-Hill
A Division ofTheMcGravHiUCompanies Copyright ©2001 by The McGraw-Hill Companies, Inc. All rights reserved. Printed in the United States of America. Except as permitted under the United States Copyright Act of 1976, no part of this publication may be reproduced or distributed in any form or by any means, or stored in a data base or retrieval system, without the prior written permission of the publisher. 2 3 4 5 6 7 8 9 0 AGM/AGM 0-07-135980-X This book was set in Times New Roman by Patricia Wallenburg. Printed and bound by Quebecor World/Martinsburg. 0987654321

This publication is designed to provide accurate and authoritative information in regard to the subject matter covered. It is sold with the understanding that neither the author nor the publisher is engaged in rendering legal, accounting, futures/securities trading, or other professional service. If legal advice or other expert assistance is required, the services of a competent professional person should be sought. —From by a a Declaration of Principles jointly Committee adopted

Committee

of the American and a

Bar Association of Publishers

McGraw-Hill books arc available at special quantity discounts to use as premiums and sales promotions, or for use in corporate training programs. For more information, please write to the Director of Special Sales, Professional Publishing, McGraw-Hill, Two Perm Plaza, New York, NY 10121-2298. Or contact your local bookstore. This book is printed on recycled, acid-free paper containing a minimum of 50% recycled, de-inked fiber.

This book is dedicated to my mother, Britt-Marie Stridsman. Thank you for your support and always being there for all of us. There are not enough words in the world to describe how much I love you. All my forthcoming proceeds from this book will be donated to Children's Wish List in Chicago, and similar voluntary organizations, dedicated to help kids with medical and other urgent needs, to help them grow up as healthy and happy individuals.

DISCLAIMER
Hypothetical performance results have many inherent limitations, s o m e of which are described below. No representation is b e i n g m a d e that any account will or is likely to achieve profits or losses similar to those shown. In fact, there are frequently sharp differences between hypothetical performance results and the actual results achieved by any particular trading p r o g r a m . O n e of the limitations of hypothetical performance results is that they are generally prepared with the benefit of hindsight. In addition, hypothetical trading does not involve financial risk, and no hypothetical trading record can completely account for the impact of financial risk in actual trading. For example, the ability to withstand losses or to adhere to a particular trading p r o g r a m in spite of trading losses are material points which can also adversely affect actual trading results. T h e r e are n u m e r o u s other factors related to the markets in general or to the implementation of any specific trading p r o g r a m which cannot be fully accounted for in the preparation of hypothetical performance results and all of which can adversely affect actual trading results. All trading systems and strategies in this book are intended for educational purposes only, to provide a perspective of different market concepts. T h e y are not meant to r e c o m m e n d or promote any trading system or approach. You are advised to do your o w n research and testing to determine the validity of a trading idea.

CONTENTS

INTRODUCTION xv ACKNOWLEDGMENTS

xxiii

PART ONE
EVALUATING P E R F O R M A N C E 1

Chapter 1

Performance Measures

3

Total N e t Profit 4 M a x i m u m Intraday D r a w d o w n 5 A c c o u n t Size Required and Return on A c c o u n t 6 Average Trade 7 Largest W i n n i n g a n d L o s i n g Trades 7 Gross Profit and Gross Loss 8 The Profit Factor 8 Average W i n n i n g a n d Losing Trades 9 N u m b e r of (Winning/Losing) Trades a n d Average N u m b e r of Bars per Trade 9 M a x Consecutive Winners a n d Losers a n d Percent of Profitable Trades 10 C o d e 11

ix

X Contents Chapter 2 Better Measures 13 Slippage and C o m m i s s i o n 17 Profit per Trade 17 Largest Winning/Losing Trade 20 Cumulative Profit and Equity Tops Drawdown 23 Flat Time and Run Ups 25 Code 26 Chapter 3 Futures Contract Data 31 21 Nonadjusted Data 31 Point-based Back-adjusted Data Ratio-adjusted Data ( R A D ) 37 Multimarket Portfolios 38 Perpetual Contracts 41 32 A Few Final Thoughts About Part 1 45 PART TWO SYSTEM CONCEPTS Chapter 4 Picking Tops and Bottoms Chapter 5 Data Mining 55 51 47 Better Use of Your Data 62 Basic Exit Techniques 69 Chapter 6 Trade or Not to Trade 83 .

Contents xi Chapter 7 Following the Trend 85 M o v i n g Averages 86 D y n a m i c Breakout Systems Standard Deviation B r e a k o u t 97 107 115 A Few Final Thoughts A b o u t Part 2 PART THREE GETTING OUT Chapter 8 Efficient Trades Drawdowns Chapter 9 Sweeney's MAE/MFE Beyond M A E / M F E Chapter 10 Adding Long-term Exits 167 156 143 121 119 117 T h e D y n a m i c Breakout System 168 T h e Standard Deviation Breakout System Chapter 11 Working with Random Entries 187 177 T h e Gold D i g g e r System 191 T h e M e a n d e r System (Weekly Data) T h e Black Jack System 204 A Few Final T h o u g h t s A b o u t Part 3 197 223 .

xii Contents PART FOUR HIGH-PROBABILITY FILTERS 225 Chapter 12 Filtering 227 228 Trend Filters for Short-term Systems T h e Gold Digger System 236 T h e M e a n d e r System 239 T h e Black Jack System 240 Chapter 13 Long-term Volatility Filters T h e D B S System 2 4 6 T h e S D B System 2 4 7 T h e Directional Slope System Chapter 14 What Makes a Trend 255 241 252 A F e w Final T h o u g h t s A b o u t Part 4 267 PART FIVE MONEY MANAGEMENT AND P O R T F O L I O C O M P O S I T I O N 269 Chapter 15 Money Management 271 282 Practical Applications for M o n e y M a n a g e m e n t Short-term Systems 2 8 5 T h e Long-term Strategies 305 The SDB Strategy 305 The Directional Slope Strategy 313 The DBS Strategy 319 .

Contents xiii Chapter 16 Portfolio Composition 321 Total Equity Contribution I 321 Correlations and Covariances 324 Total Equity Contribution II 325 Chapter 17 Increasing Your Confidence Out-of-sample Testing 331 Altering the Inputs 334 A Few Final T h o u g h t s About Part 5 Back to the Future 342 339 331 INDEX 345 .

INTRODUCTION

After I had been an editor, writer, and technical analysis expert for Futures magazine for two years it b e c a m e apparent to me that the main mistake most people m a k e in the trading industry is to believe that there is nothing to it; that it is a piece of cake to put together a trading strategy and then go out there and m a k e the big bucks, k n o w i n g nothing about risk m a n a g e m e n t , m o n e y m a n a g e m e n t , why certain systems simply do not work when traded on certain markets, why your worst drawdown is still to c o m e or w h y a system can work as it should but still break you. And horrifyingly enough, this attitude s e e m s almost equally as c o m m o n a m o n g longtime professionals as a m o n g wanna-be amateurs. A n o t h e r c o m m o n mistake, m a d e especially by smaller investors, is the thought that lesser money needs less efficient and less thoroughly researched trading methods. That there will be plenty of time to learn about m o r e "sophisticated" strategies and market intricacies as their accounts grow. This is not so. M o n e y is money; you are equally as likely to lose it, whether you are a long-time, multimillion dollar m o n e y manager or are new to the g a m e with a $ 10,000 savings account, if the methods you are using won't allow you to succeed in the first place. Think about it; why would you be any m o r e successful trading $10,000 in a basic moving average crossover system, with no m o n e y m a n a g e m e n t attached (just because this is all you know about trading and all the m o n e y that you have), than if you were sitting on millions of dollars, knowing everything there is to know about trading? If you think that you do not have enough money or knowledge to do it the right way from the very beginning, you should not trade. Period. Because if you do, you will find that there will in fact be plenty of time to learn about the more "sophisticated" m e t h o d s , but not while trading. Ask m e — I had the opportunity to experience this first hand. Trading Systems That Work is for all of you traders out there, no matter what your level of experience is, w h o have c o m e to realize that it is not as easy at it seems,
XV

xvi

Introduction

but still haven't been able to pinpoint the missing pieces that will solve the whole puzzle. I would guess that what you believe is missing, or what is keeping you from succeeding as a trader, is the overall understanding of h o w it all ties together and what really constitutes a m o r e sophisticated strategy with a higher likelihood for success. This book is my most sincere effort to try to provide you with this overall understanding, to help you solve the whole puzzle. Call me when you d o . Today, I am primarily an analyst and writer—not a trader, although I have traded in the past and currently have a few systems out there, traded by others. M a n y misinformed traders think that just because you are not a trader, you do not know anything about h o w the markets work. In my opinion, nothing could be further from the truth. Just as good math skill is no guarantee of good language skill, or good driving skill is no guarantee of good mechanical skill, good trading skill is no guarantee of good analytical skill—and vice versa for that matter. Furthermore, being an analyst and writer also is a matter of choice: I simply cannot resist sitting down in front of the computer and tinkering with something, then later trying to write about it in one form or another. As an analyst and specializing in systems a n d mechanical trading, I believe it follows that I have the necessary trading skills—as long as I follow my o w n strategies. If you do not believe m e , please continue reading and let me prove you wrong. As you turn the pages, I will provide you with a set of useful tools based on my personal and, I believe, highly innovative analysis techniques; all on a level of sophistication a n d market knowledge widely surpassing most of what you can find "out t h e r e " today. Although I sometimes use scientific and academic jargon—words and phrases such as standard deviation, kurtosis, and mathematical expectancy—and s o m e times even indulge in the mathematics and calculus behind it all, I do not purport to be a scientist, statistician, or psychiatrist, or to hold any other degree from which I might have stolen s o m e of the techniques used throughout this book. In fact, I am convinced that several of you will identify several situations where I expose my ignorance in more topics than one. But that does not matter. The most important thing that I will show you is that you do not need to be a rocket scientist to analyze the markets successfully, but you must know a bit m o r e than what you have been able to pick up from other books, and you must dare to think "outside of the b o x " a little. A n d for that I m a k e no excuse. W h e n trying to explain something new, a teacher often finds himself in a dilemma. To explain subject C, he first must explain subjects A and B, but often these subjects interweave with each other. So to learn A we need to first understand B and C. For the p u r p o s e of this book I have tried hard to keep it all in some logical order and as simple as possible, but invariably there are instances where I must talk about and explain two new subjects at the s a m e time. In other instances, I have simply left a certain topic unexplained and gotten back to it later, in a m o r e thorough analysis and explanation. On a few occasions, I also cheated a bit by

Introduction

xvii

oversimplifying things. I apologize for this, w i t h the explanation that I simply cannot give away the w h o l e store in one book. I h o p e you will bear with me and find it worthwhile reading j u s t the s a m e . Although I h o p e that you will perceive this as a practical book, it is not a b o o k about a n e w set of ready-made technical analysis indicators, or a book about foolproof trading systems, because there is no such thing. N o r is it a book that will help you identify unique trading opportunities and show y o u how to exploit those opportunities with a swift decision and then go do something else. And although it might seem like it, it isn't even a book about n e w techniques or n e w ways to develop trading strategies. Instead, this is a book about how to go about reasoning even before you start a development process that eventually and hopefully will allow you to put together a trading strategy that is exactly that—a strategy, a long-term work process, as opposed to a series of single, isolated decisions. With the risk of sounding way too pretentious right off the bat, this book is about philosophy: that is, the philosophy behind the work of putting together a good, working trading strategy; the philosophy behind what drives markets and what makes them behave as they do; and finally, the philosophy that allows us to understand why a strategy works, or why it doesn't. T h e key point I h o p e to get across is that a trading strategy is a "process machine," where each decision m a d e automatically and immediately leads to the next one, and next o n e , and next o n e . . . producing a long string of interacting decisions that form a never-ending p r o c e s s — a l m o s t aperpetuum mobile. And, just like a perpetuum mobile, the trading strategy is a very sensitive m a c h i n e , consisting of a few parts that, by necessity, form a w h o l e greater than those parts and can allow for no loss of energy. Ideally, for this to be possible, each part must be put together or constructed with all the other parts in m i n d but also be a part of all other parts. (This is the philosophical aspect of this book.) It is almost.. . N o , in fact, it is exactly as if each part is both one of the parts that m a k e a car work, and the car itself, m a d e up of all the parts. However, just as a perpetuum mobile is an impossibility, putting together a good trading strategy is very m u c h like putting together, or buying, a good and well-balanced car. While doing all this work, you also must think about your own needs and current status in life, financially and otherwise. For instance, even if you can afford it, y o u do not go and b u y yourself a N A S C A R racer for your Sunday afternoon sightseeing drives with the family. N o r would y o u go and b u y a road grader for the s a m e purpose. Instead you probably try to find a car that fulfills its p u r p o s e for your everyday life, although you might think a station w a g o n or a mini van a little boring when you know you really are the Lamborghini type. The s a m e goes with trading strategies. First, you m u s t c o m e to grips with w h o you are and what type of strategy is suitable for you, even if it is a little boring a n d leaves little r o o m for fast "overtaking m a n e u v e r s " or any other cool improvisations "behind the wheel." In fact, my experience is that trading, at least systematic trading, is as

xviii

Introduction

boring as watching baseball without a beer in your hand a n d a few in your belly. But the s a m e holds true for both the public roads and the trading rooms. They are no places for fun a n d g a m e s . W h e n putting together a trading strategy, this is h o w I look at it: the actual system with its buy and sell rules is the engine that could be anything from a fast intraday, market specific system (comparable to an 800 h.p., fast, N A S C A R - t y p e engine) to a slow, long-term, universal system (comparable to a slow but steady road grader engine). W h e n this is done, I take a look at the m o n e y m a n a g e m e n t , which is equal to the gearbox and the transmission. With the purpose of the system in mind, I now try to find a way to get its force down to the ground as efficiently and safely as possible. Continuing with the car analogy, you know it is a recipe for disaster if y o u try to fit the engine of a N A S C A R racer with the transmission of a station wagon. O n c e the engine (the system) and the transmission (the m o n e y m a n a g e m e n t ) seem to be working well in a balanced h a r m o n y with each other it's time to think about the coach and the chassis. In a trading strategy, this is comparable to the question of which markets to trade. In the case of a market-specific system, this is already done (and a case of when subject C c o m e s before subject A ) . But whether a strategy is market specific or not, it still is a good idea to make sure that it works in as m a n y markets as possible. In the case of a multimarket strategy, it is equally as important not to start curve-fitting the strategy by optimizing the markets in regard to the system, as it is not to curve-fit the system to the markets. There is a vast difference between a good system and a profitable system. It is paramount to understand this difference. It also is very important to understand that a good system can always be turned into a profitable system, while the opposite does not necessarily hold true. This is because a well working system will usually work well on a multitude of markets, catching the same type of moves, measured in percentage terms or s o m e other universal measurer. A profitable system, on the other hand, is a well working system that generates a surplus of dollars when applied in the context of a complete strategy to a specific market or portfolio of markets. W h e n you have put together the car (the overall strategy) with its engine (the system), the transmission (the money m a n a g e m e n t ) , a n d the chassis (the portfolio of markets) there still are a few things missing: the fuel and the driver. T h e fuel is your time and money. The driver is you. But before you fill the baby up and j u m p into the driving seat, m a k e sure that this really is the vehicle for you (even though y o u know in your heart that you really are the Lamborghini type). Although Trading Systems That Work is not one of those 13-in-a-dozen books on day trading that s e e m s to s w a m p the market these days, everything said here could be translated to day trading techniques as well. Neither is it a book specifically aimed toward any specific type of market, although I have m a d e extensive use of the c o m m o d i t y futures markets in my examples. Instead, words such as stocks, markets, and contracts should be looked at as s y n o n y m s .

Introduction

xix

Part 1 takes a close look at how to measure the performance of a system using a set of basic a n d universal m e a s u r e m e n t s and h o w to expand the analysis further by incorporating a spreadsheet program, such as MS Excel or Lotus 1-2-3. This part also takes a closer look at different types of data series a n d discusses when and how to use them. This is especially important to understand if you are a futures trader, but even if y o u have decided to stick to the stock market, this section should provide you with valuable insight on why so many of the systems y o u have built so far have ceased to work as s o o n as you have taken them live. In Part 2 , 1 put together a set of basic long-term and short-term trading syst e m s , using different types of data d e p e n d i n g on what I am trying to achieve. S o m e of the systems are market specific, others are suitable for a basket of markets. A lot of the analysis is done in a spreadsheet program, a n d for this we need the code developed in Part 1. Other than the specific entry techniques that we will continue to use throughout the b o o k , the most important things to learn from Part 2 are that a good working system does not necessarily have to be a profitable one and that s o m e systems are d o o m e d to fail when traded on specific markets. In Part 3, we take the systems that we put together in Part 2 a n d examine them further for their statistical characteristics and h o w they can be improved upon, using J o h n Sweeney's maximum adverse excursion ( M A E ) and maximum favorable excursion ( M F E ) analysis techniques a n d different ways of splicing up the drawdown. We also take a closer look at a few additional measurements, such as kurtosis and skew. In m a n y ways, Part 3 is the most important part, thanks to the exits that we later attach to a fixed fractional m o n e y m a n a g e m e n t regimen that boosts the results ten-fold—at least. With the entries from Part 2 and the exits from Part 3, Part 4 takes a closer look at different ways of filtering out favorable market situations and set-ups. A lot of the work is done using r a n d o m entries to generate as many trades as possible. Using r a n d o m entries, we can take only a few y e a r s ' worth of data to produce as m a n y years of u n i q u e trading sequences as we d e e m necessary. (In fact, after writing this book, using the techniques described in this section, I tested a system on the last ten years of data for all the D o w - 3 0 stocks, producing a total of m o r e than 3,000,000 years of u n i q u e trading sequences. If that doesn't produce robust results, nothing will.) At the end of this part a m o r e theoretically oriented chapter provides the framework for understanding for what m a k e s a trend and why it is my conviction that systematic trading is likely to work in the first place. In Part 5, we tie it all together by combining all the systems with different fixed fractional m o n e y m a n a g e m e n t strategies. We also take a closer look at how to put together a portfolio consisting of several market/system combinations, which help each other produce results that would have b e e n impossible to achieve without all systems working in t a n d e m under a mutually shared m o n e y m a n a g e ment regimen. To coin a new word, what we will do is to "optisize" rather than optimize the system. T h e main difference between "optisizing" and optimizing is

XX

Introduction

that w i t h optimizing we are fitting the system to the data; with "optisizing" we are fitting the bet size to the system. T h e less optimized your system is, the m o r e "optisized" you can m a k e the strategy. " O p t i s i z i n g " will almost always do m o r e for your bottom line than optimizing. M o s t of the work is with the help of a spreadsheet program, a n d there will be plenty of e x a m p l e formulas a n d code that you can copy and use in your own work. Before we round off the entire b o o k , we also take a brief look at how to test a system for robustness to further increase our confidence in it before we start using it in real life trading. A few people have asked me w h y I am giving away my ideas instead of m a k ing the most of them myself. Well, the truth is that I think I am m a k i n g the most of t h e m , because even if this b o o k all of a sudden tops The New York Times bestseller list, these systems and strategies won't be widespread enough to arbitrage away what I will show you over the next few pages. W h y ? B e c a u s e , first of all, not everyone w h o reads this will use it, because they w o n ' t feel comfortable w i t h it or b e c a u s e they simply don't like working w i t h other people's ideas, but like to c o m e up with ideas of their own. Second, even if they do use t h e m , m a n y will still not be able to m a k e any m o n e y out of them, because invariably they will c o m e up with unthought-of ways of screwing things up anyway. A n d last but not least, because the markets are m u c h bigger than any of us can fathom, these strategies will still only m a k e up a m e r e fraction of all strategies out there—strategies that help m a k e these work. In fact, I h o p e that these strategies will be widespread e n o u g h to reinforce and feed on themselves, m a k i n g them even m o r e profitable. If y o u look at it from that perspective, y o u are not my e n e m y in this, but my best friend and accomplice. My only true enemy in this g a m e is myself. As in almost any other situation in life there are a thousand-and-one ways I will be able to screw things up without your help. So, the bottom line is that even if I give this away to you, the danger that p o s e s to my future wealth and well-being amounts to nothing, compared to the danger I p o s e to myself. L a s t but not least, please read the following quotation by the renowned m o n e y m a n a g e r Ralph Vince, then p u t Trading Systems That Work aside for awhile and contemplate it. Because if y o u do not understand, or do not a g r e e with it, there is no need for you to continue: " T h e key to ensure that you have a positive mathematical expectation in the future is to not restrict your system's degrees of freed o m . . . This is accomplished not only by eliminating, or at least minimizing, the n u m b e r of optimizable parameters, but also by eliminating, or at least minimizing, as many of the system rules as possible. Every parameter you add, every rule you add, every little adjustment and qualification you add to your system diminishes its degrees of freedom. Ideally, you will have a system that is very

Introduction

xxi

primitive and simple, and that continually grinds out marginal profits over time in almost all the different markets. Again, it is important that you realize that it really doesn't matter how profitable the system is [by itself], so long as it is profitable. The m o n e y you will m a k e trading will be m a d e by h o w effective the m o n e y m a n a g e m e n t you employ is. The trading system is simply a vehicle to give you a positive mathematical expectation on which to use money management. Systems that work [show at least a marginal profit] on only one or a few markets, or have different rules or parameters for different markets, probably w o n ' t work real-time for very l o n g . . . " W h i l e you are reading through the rest of the book, please feel free to visit me at my Web site, at w w w . T h o m a s S t r i d s m a n . c o m , for updates on how to order additional material, to help me help the kids at the Children's M e m o r i a l Medical Center, or to j u s t leave me a c o m m e n t or two. Thomas Stridsman

I

T h a n k s to N e l s o n Freeburg and his family. A special thank you to my friends Jim and Margie Kharouf for helping me get settled in Chicago. Inevitably. thanking everybody would result in a thank-you note longer than the rest of the book. Thanks also to Jonas Vikstrom and Johan Ljung at Vikstrom. despite my limited English. A last minute thanks to Patty Wallenburg for being a great sport and of tremendous help during the final editorial stages of putting this book together. and to M a x von Liechtenstein for contributing with the very important " W h a t makes a trend" chapter in Part 4 and for helping me refresh my micro. To all the rest of y o u that I have learned to k n o w and learned from throughout the years. xxiii . Thanks to Ginger Szala and Jamey Holter and all the other editors at Futures magazine for hiring me and believing in me as an editor at Futures. A very w a r m "thank y o u " also to Melissa L a n g e for teaching me everything there is to k n o w about w h a t an open and honest relationship is all about. both professionally and personally. for providing me with a state-of-the-art computer. and to Dan and M a r y a n n e Gramza. know that I am forever grateful.and macroeconomics proficiencies. so I just have to settle for a few very special and recent people.ACKNOWLEDGMENTS Never in my whole life did I think that I would get a chance to thank everybody w h o has m e a n t so m u c h to m e . Ljung & Partners. for believing in me and r e c o m m e n d i n g me for the j o b as a technical analysis editor for Futures magazine.

Other p r o g r a m s exist. especially when it c o m e s to the evaluation p r o c e s s — w h e r e it shares the s a m e weaknesses found in m a n y other programs. several p r o g r a m vendors now offer p r o g r a m s with w h i c h you can build and test your own trading strategies. M o r e and m o r e people also try to do this in a systematic fashion. A couple of the most popular p r o g r a m s for this are O m e g a Research's TradeStation and E q u i s ' MetaStock. S o m e of t h e m you will be able to derive directly from TradeStation or MetaStock's performance s u m m a r i e s . a n d if the information is not there. because before you can start to build a n d investigate any type of trading strategy you must know what type of information to look for. M o s t professional market analysts (myself included) probably have a strange love-hate relationship to TradeStation. c o m e up with a way to p r o d u c e it yourself. trying to m a k e a living (and hopefully a fortune). for further analysis in Excel or any i . these two programs are the m o s t popular.PART ONE Evaluating Performance W i t h the stock market in an unprecedented up-trend and with ever decreasing prices for state-of-the-art computers. you will learn which m e a s u r e m e n t s are m o s t important. On the one hand. using one or several mechanical trading strategies. a i m e d toward the retail customer. it provides you with the most possibilities of any program. T h a n k s to its PowerEditor and EasyLanguage. but on the other hand. In this first part. but thanks to their built-in p r o g r a m m i n g capabilities. or even b u y and plug in strategies m a d e by others. This is very important. Others m u s t be exported into a text file with the help of each program's built-in p r o g r a m m i n g capabilities. In the backwater of this. m o r e people than ever before are playing the markets. TradeStation probably provides the m o s t professional way available today for coding and evaluating your own trading strategy. its m a n y possibilities also reveal a few embarrassing weaknesses.

For c o m m o d i t y futures traders. it also is very important to use the right type of data a n d to r e m e m b e r that n o t all time series should be treated the same. .2 PART 1 Evaluating Performance other spreadsheet program.

CHAPTER 1 Performance Measures W h i c h system-testing m e a s u r e s are likely to work and which are not? By necessity all system testing and design m u s t be m a d e on historical data.50 will rise by 1. 12. or 5/8.75 equals 1.000 would grow to 3 . If you think this difference isn't m u c h to worry about. you probably understand what I am hinting at and should have little trouble understanding this part of the book. you probably are a little too anxiously chasing that elusive dollar.50 stock. then. This chapter presents a r u n d o w n of the m o s t c o m m o n l y used measures in TradeStation's performance s u m m a r y a n d explains which have some value. This m e a n s that for every 500 $20 stocks you buy you can use the s a m e a m o u n t of m o n e y to buy. is to m a k e as good use as possible out of this data. being able to use the profits from each trade in the next one? T h e n your initial $10. which one would you choose? If you answer the $12.000 worth of the $20 stock. 800 $12.000 worth of the $12. and which can be modified with the help of a spreadsheet program. If however. one currently priced at $12. B u t before we go on. you answered the $20 stock. If you can choose between b u y i n g two different stocks.50 stocks.50 divided by 20 equals 0.50 stock. and to m a k e your evaluation m e a s ures as forward-looking as possible.625. 800 times 1. let us start with a little quiz. what if you could choose a m o n g 20 trades like this for the rest of the year. which don't.50 and the other at $20. Just do the math. 500 times 2.75 points over the next couple of days. and you know for sure that the one priced at $12. while the one priced at $20 will increase 2.400 points (or 14%) in profits if you b u y $10.60 points equals 1.300 points (or 13%) in profits if you b u y $10. In this case.60 points (almost a full point more) over the same period of time. T h e trick.

231 if you only b o u g h t the $20 stock. W h a t applies to the one-market case also holds true for a portfolio. Often.534 if you only b o u g h t the $20 stock.50 stock. m e a n s that the total net profit is unevenly distributed through t i m e and mostly influenced by the very latest market action. the total net profit tells you nothing about h o w well diversified your portfolio is.959.000 would grow to $15. In a market with several distinctive u p . the total net profit is of very little value when it c o m e s to evaluating a trading strategy's estimated future performance. Unfortunately. or loss of capital. You cannot. This. it is used together w i t h maximum intraday drawdown. or o n e contract per c o m m o d i t y in the c o m m o d ity futures market. and end at any time. diversify a one-contract trade in the S & P 500 futures markets with a one-contract trade in corn. they illustrate t h e point that it pays to take it easy and do the m a t h before y o u j u m p into a t r a d e . calculated over both open and closed out profits. 1 1 Your maximum equity loss. at any day. A l t h o u g h these n u m b e r s are idealized.435 if you only bought the $12. In a down-trend the o p p o site holds true. at any day.d o w n trends. the total net profit tells you nothing about w h e n your profits occurred and h o w large they were in relation to each other. In the multimarket case. that the trend of the market says nothing about whether the system has b e c o m e m o r e robust or not. in turn. but to $25. The term intraday indicates that this drop in equity can start at any time. but to $137.300. This is especially true if you stick to trading an equal a m o u n t of shares for all equities. This is especially important if the market you are interested in is p r o n e to trending. the larger the impact on the total net profit of the portfolio. no matter h o w rigorous the testing or how robust the system. For instance.4 PART 1 Evaluating Performance $115. for instance. A n d w h a t if you could do this for three years straight? T h e n your initial $10. no matter h o w well your system s e e m s to work in each market. A n d it is exactly this t y p e of m a t h that no m a r k e t analysis a n d trading software p a c k a g e s allow y o u to d o .187 if you only bought the $12.000 after only 60 trades.50 stock. Notice however. . if the market has been in a prolonged up-trend it is likely that the dollar value of each trade has increased with the increasing dollar value of the market. however. the larger the market value of the market. This is because w h a t are considered h u g e dollar moves in some stocks or markets are only considered to be ripples on the surface in others.a n d . TOTAL NET PROFIT Probably the m o s t frequently used optimization m e a s u r e is total net profit. In the one-market case.000. Simply stated. this matter is even m o r e complex. This is a difference of m o r e than $10. T h e reason for this is twofold and depends on whether you p r e fer to work with only one market at a time or with a basket of markets and/or systems to m a k e up a portfolio.

w h e n we need to compare different markets. something that can happen to anybody. hypothetically back-tested profit has decreased to $50.000 drawdown with the market trading at 1. it does not have to happen first t h i n g tomorrow. to value a system by its total net profit can take on absurd consequences w h e n a c o m p a n y decides to do a stock split. B u t this is not necessarily the best way to look at the d r a w d o w n w h e n it c o m e s to building a robust system. over different periods of time. the d r a w d o w n is calculated as the closed trade drawdown ( C T D ) . In Part 3 we also take a look at . and traded with different systems. and from this e x a m ple it is easy to see that after the split you m u s t trade the stock in lots of 300 stocks p e r trade to m a k e the new results comparable with old pre-split results. m a k i n g up the total equity drawdown ( T E D ) at that point in time. provided you have d o n e your h o m e w o r k correctly. For instance. M a n y times.000. To calculate the true expected drawdown. there is a vast difference between a $20.000 drawdown is similar to being caught in a series of bad trades for about 5% of current market value. an equally large drawdown in dollars would amount to approximately 16% of the market value at that point in time. because the number given is in dollars.000. D o e s this m e a n that the system all of a sudden is three t i m e s as bad as the day before? Of course not. T h e following chapters teach you how to work around this dilemma and build a welldiversified portfolio that is likely to hold up in the future and continue to provide g o o d risk protection. without any relation to where and when this bad trade sequence struck. it's exactly the s a m e system. however. when evaluating your drawdown. the historical.350 a 16% drawdown translates to $54. MAXIMUM INTRADAY DRAWDOWN H o w many of us have not heard the old adage "your worst drawdown is still to c o m e ? " While this is likely to c o m e true sooner or later. a $20. with a point value of $250 for a S & P 500 futures contract.000 drawdown when the market was trading around 500 and a $20.000. plus the open trade drawdown ( O T D ) . Unfortunately. Tomorrow. you must first find the largest drawdown—in percentage terms—in relation to the market value when it occurred and then translate that percentage into today's market value. you m u s t know exactly what you are looking at. it is not always this easy. A l s o . however. that so many traders blow out before they even have a chance to get started. hypothetically back-tested profit of $150. In TradeStation's case. In the first case. With the market trading at 1. there is no way for you to find that out. say that a stock currently is trading at $90. the information most system testing software provides you will not be enough. however. In the latter case. (We will explore the reasons for this throughout this book).CHAPTER 1 Performance Measures 5 In the stock market. No wonder then. after the stock has been split 3:1 and the stock is trading at $30. and a trading system that consistently buys and sells 100 stocks per trade shows a historical. For instance. With only the information from your regular performance summary.350.

1. To calculate the return on account. ACCOUNT SIZE REQUIRED AND RETURN ON ACCOUNT T h e account size required a n d return on account are probably the m o s t deceiving n u m b e r s of t h e m all. N o t until this is d o n e should you address the C T D for a better overall performance. As you can see by looking at Figure 1. and there certainly is no guarantee for this value to hold up in the future.6 PART 1 Evaluating Performance how to divide the O T D into start trade drawdown ( S T D ) a n d end trade d r a w d o w n ( E T D ) a n d h o w to analyze t h e m for better entries and exits. there is no way for you to k n o w that n u m b e r before you start trading. TradeStation simply divides the total net profit by the account size required. . T h e m a i n thing w r o n g with this n u m b e r is that FIGURE 1. But this is a purely theoretical figure produced with the help of hindsight. the account size required is the s a m e as the value for the m a x intraday drawdown. which shows the performance s u m m a r y for an early version of Black J a c k / M e a n d e r system in TradeStation.1 The performance summary for an early version of Black Jack/Meander system In TradeStation.

while the true expected average profit per trade in today's market is $1. the values for the largest winning trade and the largest losing trade essentially . neither TradeStation's nor MetaStock's performance summaries give you any such forward-looking information. TradeStation will tell you that the average trade is $150. the next trade is expected to show a profit of $250.000. and most recently.500. o n e for $ 1 0 0 when the markets was trading at 1. A m o r e disturbing example. when the market has rallied and is trading considerably higher.269 (as of October 1999). will impact the value of the average trade too m u c h . the reasoning is that the trades that happened "way back when. if the market has traded from 1. o n e for $150 when the market w a s trading at 1. But this is not what you can expect the average trade to be now. As is the case with so m a n y other of the most popular measures. however. can be seen in Figure 1." when the market was trading at a completely different level. T h u s . F r o m this example it is not too difficult to figure out that with the market trading at the 2. however. you have no idea about how large your trade account m u s t be and consequently no way of estimating the return on your account either. the information about the largest losing trade is m u c h more important than the drawdown for m o n e y m a n a g e m e n t purposes. at least as long as it is based solely on historical dollar measures.500 level. For instance. W h e n it c o m e s to the average profit per trade. Perhaps even m o r e important from a real-world point of view.1.000 points to 2. LARGEST WINNING AND LOSING TRADES Although the estimated largest drawdown holds information about how large your account size must be and can give you an indication of whether you have the psychological profile to trade the system in question. another for $200 w h e n the markets w a s trading at 2. Basically. AVERAGE TRADE O n e of the m o s t important things to consider before starting to trade any system is the estimated average profit per trade in the future. H e r e the average profit per trade is $ 5 7 7 . and y o u ' r e trading a system based on three profitable trades. no trader in his right m i n d would start out with a trading account that is only expected to cover his worst historical drawdown. Therefore.000. however. because there is no way to k n o w your exact largest future drawdown or ending equity.CHAPTER 1 Performance Measures 7 it is supposed to be calculated at one point in time (before you start trading) with the hindsight from two totally different points in time (during trading for the worst d r a w d o w n and when finished trading for total net profit). especially because this drawdown figure has no connection to the future whatsoever and is very likely to be exceeded. Unfortunately. what can be said about the total net profit also can be said about the average profit per trade. these two figures are completely u n n e c essary and provide you with no information whatsoever.500 points.

That is. (Fixed fractional m o n e y m a n a g e m e n t is discussed in Part 5. the profit factor is simply the relationship between dollars lost and dollars gained and. For the profit factor to work. This t i m e you w i n $2. it should follow that so are the gross profit and the gross loss. It is even m o r e important to determine how robust the profit factor is. simply divide the gross profit by the gross loss. Your gross profit is therefore $2. 20 divided by 10 also equals two. For instance. ending up with a total of $ 3 . Now. A n d in a portfolio. because it is a ratio.8 PART 1 Evaluating Performance hold no value as long as you can't put t h e m in relation to w h e n they happened and where the market was trading at the time. ending up with $30. than h o w high it is. you lose and your gross loss is $ 1. it is not that simple and the answer is " y e s — a n d — n o . the larger the market value of the market. the values of the winning and losing trades are likely to vary with the value of the market. It is obvious that the higher the profit factor the better the system. The Profit Factor To calculate the profit factor. in a market prone to trending. This information is derived via the profit factor. T h e first time you try. hoping to w i n $2 m o r e . w h i c h is your profit factor. it is a way of normalizing your results to m a k e t h e m comparable between time frames and markets. you take another chance. it is likely that the gross profit and the gross loss will be influenced in the very same way as the total net profit. the answer tells y o u h o w m a n y dollars you are likely to win for every dollar you lose. That is. Hence. provided that the profits and the losses are evenly distributed through time and the relationship between t h e m stays approximately the same over time. right? Unfortunately. the largest losing trade can be used to set up a fixed fractional m o n e y m a n a g e m e n t strategy that m o s t certainly will prove m u c h m o r e important for your b o t t o m line than the system itself.) GROSS PROFIT AND GROSS LOSS If the total net profit is of little good as a performance evaluator. For the same reason. do the same betting sequence all over again. " L o o k e d at separately. the larger the impact on the gross profit and the gross loss of the portfolio. it also is possible to use the profit factor as a c o m p a r i s o n between different systems a n d m a r k e t s . ending up with $4. W i t h your last $ 1. Once this is known. but this time multiply all values by 10. However. h o w likely is it that the profit factor will hold up in the future a n d in different m a r k e t situations? . it does not matter if the market has b e e n trending or not as long as it is reasonable to a s s u m e that your profits a n d losses have fluctuated at an equal p a c e a n d are evenly distributed through time. say that y o u have $2 a n d place $1 in a bet. they can hold a wealth of information that is very useful in your initial performance evaluation. This time you first lose $10 and then w i n $20. Two divided by one equals two.

For instance. A g a i n . . But this is very important information that gives you the first clue to whether the system is suitable for you. Therefore. or even lower. because they k n o w from experience that the profit factors for all their systems will decrease considerably w h e n the system is traded live on u n s e e n data. but the truth is that a specific system will not suit everybody. the less time you can spend in the market to reach a certain profit. if y o u k n o w that your average profit per trade equals $ 4 0 0 . y o u will not feel comfortable trading it. the average w i n n i n g and losing trades can h o l d s o m e v e r y valuable information if u n d e r s t o o d a n d treated correctly. however. 5 0 0 / 4 0 0 ) + 1). the estimated n u m b e r of trades to get you out of the d r a w d o w n is seven ( I N T ( 2 . m a k e sure that the underlying logic is sound and simple. t h e trick is to m e a s u r e t h e m at today's m a r k e t value. and that the gross profit and the gross loss are evenly distributed through time and in relation to each other. is h o w m u c h time the system is expected to stay in the market. y o u will be surprised at how m u c h you can expect to p r o d u c e from a system with a profit factor as low as 1. If you m a n a g e to do all this. If nowhere else.5. no matter how profitable it is. W h a t is even m o r e important. To build a robust trading system that is likely to continue to perform in the future. you can calculate the m i n i m u m e s t i m a t e d n u m b e r of trades (and time) it s h o u l d take to r e a c h a n e w equity high. Probably the only reason for this r e c o m m e n d a t i o n is that they do not understand how to build a robust trading strategy in the first place. this b e c o m e s evident w h e n they continue to use the total net profit and drawdowns as their m o s t important performance evaluators. a n d y o u currently are in a $2. To calculate the relative time spent in the market. For instance. AVERAGE WINNING AND LOSING TRADES As is the case w i t h gross profit and g r o s s loss.500 d r a w d o w n . NUMBER OF (WINNING/LOSING) TRADES AND AVERAGE NUMBER OF BARS PER TRADE M a n y traders and analysts pay very little or no attention to the n u m b e r of trades a system is likely to generate. that the trading rules are as simple and as few as possible. T h e questions you m u s t ask yourself are " d o e s this system trade often e n o u g h " and " d o e s it keep me in the market enough to satisfy my need for a c t i o n ? " T h e s e are seemingly silly questions at first glance. the better off you are. if y o u currently are in a d r a w d o w n a n d k n o w the values of y o u r average w i n n e r a n d loser and the frequency with w h i c h they are likely to occur.CHAPTER 1 Performance Measures 9 M a n y system vendors and trading experts believe that you should not trade a system with a hypothetically back-tested profit factor below three. however. If it does not fit your personality or style of trading. This is because time spent in the market equals risk assumed.

Again. divide by the total n u m b e r of b a r s examined. low. high. and closing price. everything else being equal. 1 hour.500 / 700) + 1). it can be expected to take at least ten trades before you are in the black again—provided that you experience two winners in a row. multiplied by the average n u m b e r of bars for the losers. Finally. For a correctly built system* however. a n d start out with two winning trades. Continuing with the $2. the percent profitable trades is the only performance m e a s u r e that can be derived immediately from the performance s u m m a r y a n d has any value w h e n it c o m e s to estimating the system's future performance. Once y o u k n o w this. however. the estimated n u m b e r of trades w o u l d be 16. If the answer is it m o s t likely h a p p e n s once in a blue m o o n . these n u m b e r s hold very little value and should be looked upon m o r e as freak occurrences than anything else. U s i n g the performance s u m m a r y above. If it does. ask yourself h o w likely it is for this to happen.10 PART 1 Evaluating Performance multiply the n u m b e r of w i n n i n g trades by the average n u m b e r of b a r s for the w i n ners. Obviously. and that your average loser is for $300. if you like to feel comfortable with trading the system. you also will be better off the fewer and shorter the losing trades are. . with a relatively high n u m b e r of profitable trades. it will take you four w i n ning trades in a row to reach a n e w equity h i g h if your average winning trade is w o r t h $700 (INT(2. A bar usually shows the opening. it is very important to know whether the system has a tendency to produce strings of trades with similar outcomes. that it is highly likely that the losers will c o m e in pairs. this equals approximately 0.500 drawdown example. Together with the profit factor. T h e n you k n o w that in the best of worlds. such as 1 minute. 1 day. never experience m o r e than two losers in a row. 2 MAX CONSECUTIVE WINNERS AND LOSERS AND PERCENT OF PROFITABLE TRADES You should try to keep the m a x i m u m n u m b e r s of consecutive winners and losers as low as possible a n d the percent of profitable trades as high as possible. It can be specified to cover any time period. W h e n you examine your system.4. T h e n u m ber of consecutive losers is especially important. w h i c h m e a n s that you will be in a trade only four days out of ten. 1 week. you need to keep this information in m i n d when you are designing the m o n e y m a n a g e m e n t strategy to go with the system. A d d this to the n u m b e r of losing trades. It also is a good idea. three times in a row. to keep track of h o w m a n y consecutive w i n n e r s it might take to bring you out of a drawdown. this assumes that the underlying logic is sound and 2 A bar is a marker in a chart that shows the trader how the price of the market fluctuated over a specific time period. w h e n monitoring your real-time performance.100. If you instead start out with two losing trades. that the m o s t winning trades y o u can expect in a row are two. If you do not m a n age to get rid of this tendency. etc. the system and the market are still holding valuable information that you haven't yet exploited. taking the d r a w d o w n down to $3.

" This equation shows that trading is a very deceptive business. sometimes it even is a good thing to keep this n u m b e r down in favor of a higher n u m b e r of profitable m o n t h s for the overall strategy or portfolio. only s e e m s to require "just a few good trades. we now can put together a m o r e generic formula to calculate the estimated n u m b e r of trades. T h e only two measures that you can put to immediate use are the profit factor a n d the percent profitable trades. if the average trade length is five days and the total a m o u n t of time spent in the market is 3 3 % . In fact. and that we h a d better m a k e sure that we know w h a t we are doing. you can keep track of which version of the system each row refers to. CODE T h i s chapter discussed which of TradeStation and MetaStock's performance s u m m a r y measures are likely to work and w h i c h are not.33) to get out of the red again. N. then it will take you an estimated 250 trading days (17 5 + 0. and the likelihood of a w i n n e r is 4 5 % . not counting w e e k e n d s and holidays. to get out of a drawdown: N = I N T ( D D A / (X * AW . By interchanging the SysVer input with all your other inputs. with this n u m b e r in hand. With the c o d e below. Furthermore.500 drawdown is 17.CHAPTER 1 Performance Measures 11 v that the system can be considered robust. Obviously you should try to keep this n u m b e r as high as possible to feel comfortable with the strategy. a m o r e robust n u m b e r is better than a high one w h e n it c o m e s to the system's tradability and when picking the best m o n e y m a n a g e m e n t strategy. between zero and one AW = Average w i n n e r AL = Average loser For example: If the average winning trade is worth $700. this matter b e c o m e s even m o r e complex. want to compare the same system on a w i d e variety of markets. together with the time spent in the market. . That is almost a full year's worth of trading to get out of a situation that. but. then the estimated n u m b e r of trades it will take to bring you out of a $2. N o t e also that in this e x a m ple we are working with dollar amounts and a fixed n u m b e r of contracts traded on one market only. for instance. you can create a file that exports this information to a spreadsheet program. the average losing trade $ 3 0 0 . as is the case with the profit factor. w h e r e DDA = Drawdown amount X = Likelihood of a winner.(1 . or several variations of the s a m e system on o n e market during the initial steps of the building process.X) * A L ) ) + 1. In u p c o m i n g sections. T h i s can c o m e in h a n d y if you. where we start working with percentages a n d an ever-changing n u m b e r of contracts traded on a w i d e variety of both markets and systems. at a first glance. Nonetheless.

." + NumToStr(PFactor. 2) + "." + "% in trade" + NewLine. TradeStrl(""). 2) + "." + "Version" + ". End." + NumToStr(SysVer. FileAppend("C:\Temp\Chap 1 -1 . TradeStrl). Inputs: SysVer(O). End. WTrades = NuraWinTrades * 100 / TotalTrades. If CurrentBar = 1 Then Begin TradeStrl = "Market" + ". If LastBarOnChart Then Begin PFactor = GrossProfit / -GrossLoss." + "P factor" + ".12 PART 1 Evaluating Performance In the next chapter we show h o w to use TradeStation's E a s y L a n g u a g e to export m o r e data into Excel for further analysis. 0) + ". 2) + ". TotBars(O). TradeStrl). WTrades(O).csv". 0) + NewLine. TradeStrl = LeftStr(GetSymbolName. FileAppend("C :\Temp\Chap 1-1 . Vars: PFactor(O)." + "% Winners" + ".csv"." + NumToStr(WTrades. a n d h o w to c o m e up with a new set of forward-looking evaluation measures that are better suited to holding up in the future and helping you to build m o r e robust and reliable trading strategies." + NumToStr(TotBars. TotBars = (TotalBarsLosTrades + TotalBarsWinTrades) * 100 / BarNumber.

while the percentage moves are likely to remain approximately the same. (For proof of this. say that a market (the S & P 500) currently is trading at 1. you will be able to perform m o r e accurate comparisons of how a system is likely to work in different markets and time frames. In this way. what you m u s t do is to calculate all the necessary values in percentage terms rather than in dollars or points. For instance. however. O n c e you are done with your percentage-based calculations. This is all fine and well if you are only interested in how m u c h money you could have m a d e had you been able to trade your strategy in the past. In essence. we take a closer look at how to export the necessary data from your system trading software into a text file and how to calculate this new set of performance measures in a spreadsheet program." w h e n the market was trading at 2 5 0 . In this chapter. 13 . is that it tells you very little about how well your strategy is likely to hold up in the future. by using percentage-based calculations. see Figures 2.375 (1. it is likely that the dollar moves increase as well. T h e same reasoning can also be applied to a comparison of different markets with different market values. a whole new set of performance measures must be put together. The main disadvantage with this way of testing. If the market value increases. a 1% move was only worth $625 (250 X 0.) T h u s .1 through 2. that move is worth $3. But if the s a m e percentage move happened " w a y b a c k when. which are m o r e likely to hold up in the future.01 X 250). If this market rises 1%.350 X 0.01 X 2 5 0 ) .350 and that each point m o v e is worth $250. To achieve this. most testing packages (and back-adjusted time series for the commodity futures markets) have only allowed you to back-test your strategies using dollar values.CHAPTER 2 Better Measures So far. you give each and every trade an opportunity to influence your strategy to an equal degree and you are able to build better and m o r e reliable strategies. you can transform t h e m into dollar values based on w h e r e the market is trading today.3.

.14 PART 1 Evaluating Performance FIGURE 2. the higher the point-based volatility.1 The higher the market value.

.CHAPTER 2 Better Measures 15 FIGURE 2.2 Least-square regression line of the point-based volatility.

.16 PART 1 Evaluating Performance FIGURE 2.3 Least-square regression line of the percentage-based volatility.

If this dollar value still looks good enough. the lower the value of the market. it can prove tricky to c o m e up with a system that beats a simple buy-and-hold strategy. while the profit per contract traded only decreased 4 0 % ? To assume the s a m e effective risk as in the buy-and-hold strategy. measured in dollars. T h e n deduct the proper amount for slippage and c o m m i s s i o n . the more risk you're assuming. which favor systems with few and long-lasting trades. could c o m e up with a strategy that only kept you in the market 5 0 % of the time. about different performance measures. only results in the risk for a suboptimal solution that can prove to be less profitable and robust w h e n you r u n your system real-time. then transform that m o v e into dollar t e r m s in today's market by multiplying by today's market level and point value. with the same a m o u n t of contracts for the entire period. when you're back-testing on historical data. In a trending market. For instance. in most cases you shouldn't even use dollar-based calculations when y o u ' r e putting your system together. This sounds strange. instead. but rather try to capture as many and as large favorable moves as possible. is still 2 0 % higher than it would have been for the buy-and-hold strategy. we looked at how the dollar value of the moves involved is likely to increase with the value of the market.CHAPTER 2 Better Measures 17 SLIPPAGE AND COMMISSION Although there is no way around these costs once the system is up and running. First of all. B u t with a buy-and-hold strategy. T h u s . while spending as little time in the market as necessary. you should not try to squeeze out as many dollars or points as you can. A n d finally. W h a t if you.4 shows you the dollar profit for every trade m a d e w i t h a simple stock index and b o n d system traded on historical data for the S & P 500 stock index future contract. you now can trade twice as m a n y contracts per time units spent in the market. T h e final net outcome. In Chapter 1. Subtracting the same cost for slippage and c o m m i s s i o n across all trades in such a market only lowers the impact of the low-value trades even further. if you are building a system to trade the S & P 500 stock index futures contract. you should not concern yourself with slippage and commission when building and researching a trading strategy. T h e s a m e goes for comparing different markets with each other. but there are several reasons for this. This m a k e s the dollar-based slippage and c o m m i s s i o n assumptions obsolete as well. using dollar-based calculations. Notice h o w the profits and loss- . PROFIT PER TRADE Figure 2. but with only a 4 0 % drop in return per contract traded. the slippage and c o m m i s s i o n settings have a larger impact on your b o t t o m line. you are in the market 1 0 0 % of the time. because the m o r e time you spend in the market. you should take the trade. as already mentioned. First calculate the expected percentage move you are likely to catch. taking into consideration slippage and c o m m i s s i o n at this point.

which shows the s a m e sequence of trades but this time with percentage-based calculations. N o t $ 7 5 5 . the percentage returns for both the profits and the losses stays the s a m e no matter what the market is doing.4 The dollar value of a series of Individual trades. O n c e the system building process is over. contrast this to Figure 2. T h a t is. In this case. 2 6 9 . this is the average profit p e r trade that you could expect to m a k e in the i m m e d i a t e future. for a robust system.18 PART 1 Evaluating Performance FIGURE 2. the average profit per trade c o m e s out to $ 7 5 5 .500 (1. M e a s u r i n g the profit per trade in p e r c e n t a g e s . Notice h o w the profits a n d losses per trade now are m u c h m o r e evenly distributed in time and c o m e out to $ 1 . If you w e r e to start using this s y s t e m today (and provided that it is robust a n d your reasoning is sound). es per trade have increased dramatically for the last h u n d r e d trades or so. (in today's marketplace). which also can be seen in Figure 1. . as implied by the performance summary. it is easy to transform the average percentage return into a dollar value applicable to today's m a r k e t situation. based on a 0 . of 337. the value for both the profits a n d the losses increases (decreases) at the s a m e rate at w h i c h the m a r k e t trades higher (lower).5. however. Now. all trades are given an equal weighting.1.350 X 250). For a robust system. by m e a s u r i n g the profit per trade in dollars. T h i s in turn gives m o r e weight to the m o r e distant results and consequently lowers (raises) the value of the average trade. 3 8 % average profit per trade and a m a r k e t value.

type in the following formula at the b o t t o m of the column w h e r e the data are stored: =AVERAGE(A1 :AX) w h e r e A denotes the column for data stored and X denotes the n u m b e r of rows/trades. the E a s y L a n g u a g e code is as follows: Vars: Prof(0). O n c e in Excel (or the spreadsheet p r o g r a m of your choice).4 but with percentage-based calculations.5 The same sequence of trades as in Figure 2. TotTr = TotalTrades. File Append("C:\Temp\Chap 1 -2. If TotTr > TotTr[l] Then Begin Prof = 1 + PositionProfit(l)/(EntryPrice(l) X BigPointValue). To export your data to a text file. TradeStr2). To calculate the percentage profit per trade in TradeStation you can use the following Easy L a n g u a g e code: Vars: TotTr(O). Prof(0). 2) + NewLine.csv".CHAPTER 2 Better Measures 19 FIGURE 2.1) XI00. TradeStr2 = NumToStr((Prof . . TradeStr2("").

">0") If you also would like to calculate these values for the short and the long side respectively. Finally. 0) + ". type in the following formula in Excel: =STDEV(A1:AX) To get the two standard deviation interval. the formula for the average dollar value p e r trade would look like: =AVERAGE(A 1: AX)/100 X 250 X1. If typed into the s a m e cell.350. or $781 ± 11. you m u s t first m u s t modify your E a s y L a n g u a g e export function to: TradeStr2 = NumToStr(MarketPosition(l). simply multiply by two. to calculate the dollar value of all trades into today's market value. It also is very important.326 in dollar terms in today's . it is of importance to know your standard deviation intervals. and X denotes the n u m b e r of rows/trades. a n d X denotes the n u m ber of rows/trades.350 is the current trading level for the S & P 500. this interval c o m e s out to 0 . multiply the results by the current trading level and dollar value per trade. To calculate your largest w i n ning trade.">0")/COUNTIF(A1 :AX. 2) + NewLine. either in the s a m e or in an adjacent cell. 2 3 % ± 3. to keep all trades as uniform as possible. type in the following formula in the spreadsheet: =MAX(A1:AX) w h e r e A denotes the column for w h e r e the data is stored. In our case.1) X100. w h e r e 250 is the dollar value per point a n d 1.36%. it is m u c h m o r e important to know your largest losing trade than to k n o w your drawdown. Therefore. the formula is: =MIN(A1:AX) w h e r e A denotes the c o l u m n for where the data are stored. A n d for your largest losing trade." + NumToStr((Prof . with the system traded on the S & P 500 stock index futures contract. To calculate the standard deviation interval for all trades. LARGEST WINNING/LOSING TRADE F r o m a m o n e y m a n a g e m e n t standpoint. from a m o n e y m a n a g e m e n t standpoint.20 PART 1 Evaluating Performance To calculate your average w i n n e r ( " > 0 " ) and loser ( " < = 0 " ) use the following formula: =SUMIF(A1 :AX.

This explains why a seemingly good system suddenly can start to p r o d u c e large losers.545 to $12. This m e a n s . it comes out to 0.CHAPTER 2 Better Measures 21 market. 1 CUMULATIVE PROFIT AND EQUITY TOPS T h e cumulative profit isn't really of any interest at this point.107)). however. To get a better feel for w h a t you can expect from your system (less any freak occurrences) and depending on w h a t type of system you're building/trading. TradeStr2(""). TradeStr2 = NumToStr((CumProf . Obviously." + NumToStr((ETop . 2) + ". or 4 6 % (= 10. this too is a risk measure that you should try to keep as low as possible. the m o r e uniform the trades and the less risky the system. It also is questionable to w e e d out too m a n y (if any at all) losers and. Usually trades that are more than three standard deviations away from the average trade are considered outlayers. To calculate and export the cumulative profit and equity tops in TradeStation's E a s y L a n g u a g e you can use the following formulas: Vars: CumProf(l). CumProf = CumProf x Prof. it also may be a good idea to w e e d out all outlayer trades before you continue to investigate it further. to m a k e up for long strings of expected small losses. future value of the average trade easily could be negative. that in this case the interval stretches well into negative territory. the true average profit for this system should lie somewhere within the interval —$10. ETop = MaxList(ETop. In fact. With this information in hand. I believe that the system that I ' m using here for demonstration p u r p o s e s is v e r y robust. divided by the total distance between the two standard deviation boundaries. but with a large n u m b e r of trades m a d e on unseen data. . with very uniform trades. that s o m e s y s t e m s ' sole p u r p o s e is to catch as m a n y "outlayers" as possible. ETop(0). however.545 / (10. and that it will continue to stay in positive territory for m a n y years to c o m e .545 + 12. To keep the standard deviation interval completely in positive territory is virtually impossible. you also can calculate the likelihood of failure as the distance between the zero-profit line down to the lower standard deviation boundary. In this case. fool yourself into believing that your system is better than it really is. a n d I have yet to see such a non-curve fitted system. although it undoubtedly has a few very scary features as well. hence.1) An outlayer trade is a trade that is judged to be so far away from the average trade that it can be considered a "freak occurrence". M a n y t i m e s the system has not b r o k e n down. it simply has m o v e d closer to its true average per trade. Notice. T h e smaller the interval. CumProf). especially considering that you cannot rule out that the true. R e m e m b e r .46.107. but is only needed to calculate other important measures such as drawdowns and flat times (the time spent b e t w e e n two equity highs). with 9 5 % certainty. which happens to be negative.1) X 100.

especially not in the futures markets. is to compare different systems and markets with each other on an equal basis. This is not all that correct. the final equity g r o w t h is only a result of how aggressively you choose to trade it and/or what type of m o n e y m a n a g e m e n t strategy you use. the upper straight line depicts the latest equity peak. percentage-based equity curve for a version of the Black Jack system. T h e reason the cumulative profit is of lesser interest at this point is that as long as the system is considered to be robust. provided that you can b u y fractions of a share.NewLine. is being reinvested each time.6 shows the cumulative. . The trick is to m a k e all lines coincide with each other as m u c h as possible and keep the " b o x e s " as small and as few as possible.csv". with a stable profit factor a n d a constant relative a m o u n t of winners. T h e plot in Figure 2. a n d has a positive expectancy p e r trade b a s e d on today's market value a n d less expected costs for slippage and commission. Also. percentage-based equity curve for a version of the Black Jack system. or test how FIGURE 2. in the calculation above we are assuming that the entire equity.TradeStr2). T h e squiggly line is the cumulative profit. including previously m a d e profits.6 The cumulative.2) . however. FileAppend("C:\Temp\Chapl-2. W h a t it does allow you to do.22 PART 1 Evaluating Performance X 100. but could be so in the stock market. and the lower straight line is the latest equity drawdown.

O n c e this is done. for instance.000. From this chart you can see that the worst equity dip for this system is about 8%. time frames. A good way to get a feel for the drawdown for any trading strategy is to chart it as a so-called underwater equity curve.000 in today's market value (1. This is achieved by a trade-by-trade efficiency analysis primarily using John Sweeney's m a x i m u m adverse excursion ( M A E ) and m a x i m u m favorable excursion ( M F E ) techniques {Campaign Trading: Tactics and Strategies to Exploit the Markets [Wiley Finance Editions. as in Figure 2.178. no matter what entry technique you use. including both closed trade drawdown ( C T D ) and open trade d r a w d o w n ( O T D ) in the same figure. you should have a trade m a n a g e m e n t technique that will let you trade (at least marginally) profitably over a wide variety of markets. As mentioned. First. But that doesn't free you from the responsibility to investigate it as thoroughly as you can to make sure that you're properly capitalized when you start trading. or total equity drawdown ( T E D ) .1. 1997]).000 level).7.000 were you to start trad- ." you think: "I can stomach a drawdown of less than $20." But nowhere does it say when in time this drawdown happened or how large it was in relation to where the market was trading at the time (in this case it happened during mid-1998.000. when the market was trading around the 1. T h e n . a n d only then. one-third of your equity right off the bat. and market conditions. If you do not want to experience a drawdown of. should you finalize your entry technique and start to look for a high probability filter technique to c o m b i n e with your trade-by-trade management. which shows the depth of the historical drawdown.08). To build a robust trading system. "Great.000. you must m a k e each and every trade as efficient as possible. by dividing the O T D into the start trade drawdown ( S T D ) and the end trade drawdown ( E T D ) . the worst expected drawdown will only make up a third of my initial capital. The drawdown n u m b e r s you then c o m e up with should be for closed out trades only. this is not the way to go about finding the optimal solution. DRAWDOWN M a n y traders and analysts prefer to calculate the drawdown as the m a x i m u m intratrade drawdown. Take another look atTradeStation's performance summary in Figure 1. T h e problem really consists of several parts. This. 1996] and Maximum Adverse Excursion: Analyzing Price Fluctuations for Trading Management [Wiley Trader's Advantage Series.CHAPTER 2 Better Measures 23 your strategy would have held up compared to a buy-and-hold strategy over the same time period. you would need an initial trading capital of at least $81. which corresponds to approximately $27. however (as already mentioned in Chapter 1).350 X 250 X 0. This system has a drawdown of $18. the drawdown really isn't that interesting when it comes to deciding whether a system is robust and profitable enough to trade. as mentioned above. in turn would have taken away close to 5 0 % of your initial equity of $60. and if I start out with $60.

In fact. w h o thrive on stomach aches and sleepless nights. EBot = MinList(EBot. the probabilities have it that we all will be w i p e d out sooner or later. CumProf). the drawdowns must be measured in percentages and then transformed into today's market value. ing this system today.) This little piece of TradeStation code shows you h o w you can export the n e c essary drawdown information for your closed out trades to a text file for further analysis in Excel: Vars: CumProf(l). (Or. It is only a question of h o w long you stick around. EDraw(O). not only is your worst drawdown still to c o m e — t h e truth is that you will eventually be completely w i p e d out. If we all stick around long enough. is to do your homework as well as possible so that you can keep disaster in front of you and quit while you are still ahead. avoid getting started in the first place. O n l y the most risk-seeking people. To c o m e to the necessary conclusions. R e m e m b e r also that your worst drawdown is always still to c o m e . EBot(O).7 The underwater equity curve measures the depth of the historial drawdowns. would trade a system with an expected drawdown closing in on 5 0 % of their equity. T h e trick then. . as in the case above. That is in addition to all those w h o do not know that the expected d r a w d o w n of their system usually is m u c h higher than w h a t their system trading software's performance s u m m a r y tells them.24 PART 1 Evaluating Performance FIGURE 2. It is j u s t a matter of time. TradeStr2("").

CHAPTER 2 Better Measures 25 EDraw = CumProf / ETop. flat time m e a n s the time spent in between trades.6 again. FileAppend("C:\Temp\Chapl-2.TradeStr2)." + NumToStr((EDraw .8 The flat time measured as the time spent between two equity highs. But h o w long do you have to spend between two equity highs? This is revealed in Figure 2. this alternative interpretation only m a k e s up a part of the total flat time. From Figure 2.1) X 100. 2) + '*. or in a directionally neutral position.csv".8. (There is a risk of confusion of terms here. .1) X 100. the t i m e spent between two equity highs. In this book. however. as FIGURE 2. For m a n y analyst a n d traders. which shows the so-called flat time. 2) + NewLine. FLAT TIME AND RUN UPS Take a look at Figure 2. This looks like a pretty decent equity curve with a steady upward slope and tolerable drawdowns. TradeStr2 = NumToStr((EBot .7 it also is evident that this system will keep you in the red an awfully long time and only on occasion will you be able to get b a c k to the surface and set a n e w equity high. Even the best of systems will spend most of their t i m e in drawdowns.

with a flat time of little m o r e than 4 0 0 days. In this case. on average. are only of interest from a psychological point of view. a n d eventually broke even. TradeStr2 = NumToStr(TopInt. TradeStr2). T h e average time for this system spent in a flat is 86 bars. But d o n ' t despair. Botlnt = CurrentBar .csv". So. or a m a x i m u m flat time of m o r e than 18 months.) A good system should have a m a x i m u m flat time of no longer than 18 months. 0) + "." + NumToStr(BotInt. and the average time between equity lows is 66 bars. we c o m e pretty close to w h a t is tolerable. In "psychological t e r m s " this m e a n s that only about four times per year. drawdowns are simply something you have to learn to live with and the truth of the matter is that even the m o s t robust and profitable trading strategies will keep you feeling on your way d o w n m o r e often than on your way u p . IfEBot < EBot[l]Then BotBar = CurrentBar.26 PART 1 Evaluating Performance we define it. To chart the time spent between two equity highs and to calculate the average. However. Toplnt = CurrentBar — TopBar. EBot(O). you m u s t add the following E a s y L a n g u a g e code (which also allows you to calculate the time spent between equity lows) to the system: Vars: Etop(0). FileAppend("C:\Temp\Chap 1 -2. you will spend five times every year feeling really rotten. will you be allowed to feel like a really successful trader.BotBar. M o s t institutional investors will not tolerate a drawdown over 30 to 3 5 % of total equity. BotBar(O).5 years just to get your m o n e y back. like the expected d r a w d o w n and expected flat time. TopInt(O). 0) + NewLine. Others. setting a new equity low before the markets finally start to move your way again. IfETop > ETop[l]Then TopBar = CurrentBar. y o u would have started out losing. If you started to trade this system just prior to this drawdown. CODE In this chapter we looked at how to build a new set of system testing measures that give you m o r e realistic results that are m o r e likely to hold up in the future. TopBar(O). if you are aspiring to selling services to the "big money. as long as you can keep t h e m within tolerable levels. T h e most important m e a s u r e s w h e n you are building your system are average expected profit p e r trade and largest expected losing trade. Botlnt(O). setting a n e w equity high. stuck it out for approximately 1." these are the measures within which you m u s t keep yourself. all vari- .

Toplnt = CurrentBar . IfEBot<EBot[l]Then BotBar = CurrentBar." + NumToStr(TopInt. 2) + ". you probably are b o u n d to lose. B e l o w is all the code used in this chapter. 0) + ".CHAPTER 2 Better Measures 27 ables m u s t be e x a m i n e d equally thoroughly. 0) + ".csv".1) X 100.1) X 100. no matter w h a t the market is doing or h o w well the s a m e system s e e m s to be w o r k i n g for others. EBot(l). 0) + ". EBot = ETop." + "X Price" + ". If TotTr > TotTr[l] Then Begin Prof = 1 + PositionProfit(l)/(EntryPrice(l) X BigPointValue)." + "Run up" + ". TotTr = TotalTrades. CumProf(l). 2) + "." + NumToStr(ExitPrice(l). 2) + ".1) X 100." + "Drawdown" + NewLine. 0) + ". TradeStr2 = NumToStr(EntryDate(l). TotTr(O). 2) + NewLine. FileDelete(FName).m . prof. EDraw(l). ETop(l). Botlnt = CurrentBar . put together into one piece: Vars: FName("").1) X 100." + NumToStr((Prof ." + "E-Bottom" + ". Prof(0). If CurrentBar = 1 Then Begin FName = "C:\Temp\" + LeftStr(GetSymbolName. TradeStr2( )." + NumToStr(EntryPrice(l). End. If you do not feel comfortable with all of them." + "Profit" + "." + NumToStr((EBot ." + "Cum." + "." + NumToStr(MarketPosition(l)." + "E Price" + ". 0) + ".TopBar." + NumToStr(ExitDate(l). 2) + ". TradeStr2 . TopBar(O)." + NumToStr((CumProf . Tf ETop > ETop[l] Then Begin TopBar = CurrentBar. Toplnt(O)." + NumToStr((ETop ." + "X Date" + "." + "Flat time" + ". 2) + ". Botlnt(O). BotBar(O). CumProf = CumProf X Prof. CumProf). 2) + ". FileAppend(FName. EDraw = CumProf / ETop." + NumToStr(BotInt. ETop = MaxList(ETop." + NumToStr((EDraw . .BotBar. End. TradeStr2). EBot = MinList(EBot. CumProf).1) X 100."E Date" + "." + "Position" + "." + "E-Top" + ". 2) + ".

in cell F ( X + 5 ) : =GX/100 w h e r e G denotes the c o l u m n w h e r e you have stored the cumulative profit. F r o m n o w on. you can use it to put together your own performance s u m mary. TABLE 2. in cell F ( X + 3 ) : =MAX(F$2:FX)/100 For the average (%) winner.33% 16.1 Example of performance summary. For the standard deviation in percent. in cell I ( X + 2 ) : .">0")/100 For the cumulative (%) profit. in cell F ( X + 2 ) : =COUNTIF(F$2:FX. and X denotes the n u m b e r of rows/trades.28 PART 1 Evaluating Performance FileAppend(FName. >0").269 5. To put together a performance s u m m a r y like this. 1 .">0")/COUNTIF(F$2:FX. TradeStr2).38% 1.667 250.373 -4.80 1.61% 150 1.76% 38.470 4. Total trades Profit factor Avg profit St Dev 0. For the largest (%) winner. which will be m u c h m o r e telling about your system than any other performance s u m m a r y you are likely to find in any other program.74% -1.22% -7.421 Winners Lrg winner Avg winner Cum profit 92 4..88 1.67% -19. For the n u m b e r of losers. O n c e you have exported all the necessary data into Excel or any other spreadsheet program. End.38 74. type the following formulas into your spreadsheet p r o g r a m : For the average percentage profit.155 Losers Lrg loser Avg loser Drawdown 58 -5. in cell C ( X + 4 ) : =AVERAGE(F$2: FX)/100 where F denotes the c o l u m n w h e r e you have stored the profit from each individual trade. in cell F ( X + 4 ) : =SUMIF(F$2:FX. we will work extensively with a type of performance s u m m a r y that resembles that in Table 2 . in cell C ( X + 5 ) : = STDEV(F$2:FX)/100 For the n u m b e r of winners.190 .121 -26.12 61.

in cell D ( X + 4 ) : = C ( X X 4 ) X 1350X250 w h e r e 1350 denotes today's index value." <=0")/COUNTIF(F$2:FX. in cell G ( X + 3 ) : = F ( X + 3 ) X 1350X250 For today's dollar value of average winner. and 250 denotes dollar value per point." < = 0 " ) / 1 0 0 For the largest (%) drawdown. in cell I ( X + 5 ) : =MIN(L$2:LX)/100 w h e r e L denotes the c o l u m n for w h e r e you have stored the drawdown. in cell G ( X + 4 ) : = F ( X + 4 ) X 1350X250 For today's dollar value of the cumulative profit.CHAPTER 2 Better Measures 29 =COUNTIF(F$2:FX. For today's dollar value of the average trade. in cell I ( X + 3 ) : MIN(F$2:FX)/100 For the average (%) loser. in cell D ( X + 2 ) : =F(X+2)+I(X+2) For the percentage of winners. For today's dollar value of the standard deviation. in cell D ( X + 5 ) : = C ( X + 5 ) X 1350X250 For the n u m b e r of trades. in cell J ( X + 2 ) : =I(X+2)/D(X+2) For today's dollar value of largest loser. in cell G ( X + 5 ) : = F ( X + 5 ) X 1350X250 For the percentage of losers. in cell J ( X + 4 ) : = I ( X + 4 ) X 1350X250 . in cell I ( X + 4 ) : =SUMIF(F$2 :FX."<=0") For the largest (%) loser. in cell G ( X + 2 ) : =F(X+2)/D(X+2) For today's dollar value of largest winner. in cell J ( X + 3 ) : = I ( X + 3 ) X 1350X250 For today's dollar value of average loser.

in cell D ( X + 3 ) : = A B S((F(X+2) X G(G+4))/(I(X+2) X J(X+4))) . in cell J ( X + 5 ) : = I ( X + 5 ) X 1350X250 For the profit factor.30 PART 1 Evaluating Performance For today's dollar value of the largest drawdown.

we take a closer look at these different m e t h o d s . one of the m a i n hurdles to overcome w h e n testing trading strategies on historical futures data is the limited life span of a futures contract. using the percentage-based calculations described in the previous chapter. look at their pros and cons. and the perpetual-adjustment method. we had to base our calculations on percentages rather than dollars or points.CHAPTER 3 Futures Contract Data In Chapters 1 and 2 we looked at those performance measures that could be derived directly from TradeStation's performance summary. this coincides with w h e n the market as a whole moves from one contract to the next. and explore suggestions on w h e n and how to use them. the back-adjustment method. T h e back-adjustment m e t h o d can further be subdivided into point-based adjusted a n d ratio-adjusted. Usually. If you are a c o m m o d i t y futures trader. resulting in 31 . To overcome this. There are three different m e t h o d s to go about splicing contracts together. you simply stop charting o n e contract when it expires or w h e n you otherwise d e e m it justifiable to do so and continue to chart the next contract in line. This is especially important when it c o m e s to m a k i n g your system reports as forward-looking as possible. and those that we had to export into a spreadsheet p r o g r a m for further analysis to m a k e t h e m m o r e forward looking. In this chapter. w h i c h n o w is the new front contract. various m e t h o d s have been invented to splice several contracts together to form a longer time series. NONADJUSTED DATA With the nonadjustment method. In Chapter 2 we also concluded that to achieve the latter. T h e nonadjustment method.

In percentage terms. If the new front contract is trading at a p r e m i u m c o m p a r e d to the old contract.296. with the latest roll m a d e in September 1999. the entire historical time series leading up to the roll is adjusted upward by that distance.296. m e a n i n g that the new front contract usually trades slightly higher than the old one on the day of the roll. the new contract closes at 1. Over time. if. the market fell a total of 152 points. 6 % (152 / 333) of the total market value ( r e m e m b e r these n u m b e r s ) . M a n y traders. it shows you exactly how the front contract at that particular time w a s traded.309. at each point in time. the high on October 2 is at 567.309. the point-based back-adjusted contract was invented. T h e m a i n disadvantage of the nonadjustment m e t h o d is the difference in price that frequently appears on the day of the roll.318.5 — 1.2 shows w h a t the October 1987 market action looks like using a point-based back-adjusted contract.000 (152 X 2 5 0 ) to the low of 181 on October 20. resulting in a new artificial value of 1.60 points (1. If there happens to be a difference in price between the new and the old contract. Figure 3. which will distort your backtesting results.9). With the TradeStation code provided in Chapter 1 and 2 it is easy to export all trading results from several different and separately tested contracts into the same spreadsheet for a c o m b i n e d analysis. To overcome the distortions on the historical systems testing results induced by the nonadjustment method.6.331. this equals a drop of 4 5 . nonadjusted data is best for day traders or other c o m modity futures traders with very short trading horizons.5 and the old contract at 1. Therefore.1 illustrates w h a t this looked like for the D e c e m b e r contract on the S & P 500 index during the crash of 1987. Figure 3. a n d the day before at 1. T h e m a i n advantage of the nonadjusted time series is that. with all trading levels and price relationships intact and exactly as they appeared in real life. this difference would have been added or subtracted to the result of that particular trade. this error adds u p . however.9.6 + 1. In this chart. during your historical profit testing. prefer to roll over on a certain day of the month or when there are a certain n u m b e r of days left on the life of the old contract. H a d you been in an open position at this time. F r o m the high of 333 on October 2. on the day of the roll. it usually trades at a p r e m i u m . or $38.6).35 and the low on October 20 is at . A similar but opposite adjustment takes place each time the n e w contract is traded at a discount.32 PART 1 Evaluating Performance an increase of the open interest for the n e w front contract surpassing the open interest for the old contract. For instance.2 for the last close used from the old contract (12. POINT-BASED BACK-ADJUSTED DATA M o s t markets trade at cither a p r e m i u m or a discount the longer in t i m e you look.318. it is left unadjusted. w h o m o r e often than not close out all their trades at the end of the day. the entire time series up to that point of the roll is adjusted upward by 12. If we continue to use the S & P 500 as an example.

CHAPTER 3 Futures Contract Data 33 FIGURE 3.1 The crash of 1967 viewed through the nonadjusted contract .

2 The crash of 1987 viewed through the point-based back-adjusted contract.34 PART 1 Evaluating Performance FIGURE 3. .

B e c a u s e of the transition upward. with the market trading around 1. however. the original percentage difference of 4 5 .311 (0. Viewed in percentage terms. Yes. but this is only a natural consequence of the market's trading higher. indicating that the market has b e c o m e m o r e volatile and difficult to trade. they m u s t be viewed in percentage terms.268 X 333 X 250) instead of $38. This is only a consequence of the increasing market value. This is illustrated by Figures 2. 3 5 . a $5. as the regression line in Figure 2. those w h o claim that the volatility of the stock market has increased over the years and.2 indicates. the volatility has indeed increased considerably over the years. T h e straight lines that run through the two latter charts are best-fit regression lines. the volatility. if the 2 6 . if anything volatility has decreased somewhat.3. 6 % n o w has decreased to 2 6 . we would c o m e up with $22.35). covering the period late 1985 to late 1999.000 trailing stop (approximately 3% of current market value) were added to a standard breakout system. 8 % d r o p — a s implied by the back-adjusted contract—were to be used to calculate the dollar value of the drop from the actual top of 3 3 3 .1 through 2.or point-based volatility has increased considerably. one important thing to r e m e m b e r is never to use dollar. however. By the s a m e token. the dollar. If we look at Figure 2 . Consider.CHAPTER 3 Futures Contract Data 35 4 1 5 . For instance. W h a t immediate consequences does this have for you as a trader? Well. presumably upward. which has stayed approximately constant (the slope is slightly down) around 3 . In fact. measured in percentages. measured in dollars or points. 8 % (152 / 567. for a total difference of 152 points. the volatility has stayed approximately the s a m e since late 1985. As you can see. But.000 m o n e y m a n a g e m e n t stop (equal to approximately 1.456 X 333 X 250). resulting in an ever-decreasing percentage difference. it seems that the price bars have b e c o m e wider and wider in recent time. This continuing decrease (increase) of the relative importance of historical moves in an up-trending (down-trending) market a n d markets that are traded at a p r e m i u m (discount) is the m a i n disadvantage of the point-based back-adjusted contract. do not k n o w what they are talking about. What's m o r e .350) a n d a $10. has b e c o m e m o r e difficult to trade. In back-testing this system on nonadjusted S & P 500 . if anything. together with the absolute monthly changes in points and percentages. Contrast this to the regression line for the absolute percentage change. which show a nonadjusted monthly chart of the nearest future on the S&P 500 index.or point-based stops. therefore. for instance the following example: U s i n g TradeStation. a n d should prove no problem if dealt with properly. 5 % p e r month.000 (0. for each roll all these contract values will continue to change slightly. Therefore. aside from those topics discussed in the previous chapters. actually has decreased somewhat. relatively. the regression line for the absolute point changes has increased over the years from close to 0 to over 30 points p e r month. 1 . to keep the relative importance of all historical moves constant.5% of current market value. That is.

3 The crash of 1987 viewed through the RAD contract .36 PART 1 Evaluating Performance FIGURE 3.

the hits b e c a m e m o r e frequent.15. the market fell from a July high of 1199. is that the percentage-based move stays intact at 4 5 . the percentage relationship stays the s a m e . e C — T h e close of the n e w contract on the day of the roll. January 1999) that a better wa y to adjust a spliced-together time series would be to use percentages rather t h a n points or dollars. " Futures m a g azine.1).CHAPTER 3 Futures Contract Data 37 data.4 points (as . This h a p p e n e d in the a u t u m n of 1998. c T h old price i bars ago. for a point difference of 221. This c o m p a r i s o n is d o n e w i t h the help of the following formula: Cmew = c Where: c i o l d X(l+(C-c)/C) inew ioid = = n e w price i bars ago. covering the period January 1983 to April 1999. implicitly suggesting that this w a s equally as bad or at least w a s soon to be. only showing up again in M a r c h 1996." Futures m a g azine. After this. D u r i n g the a u t u m n of 1998. however. D u r i n g October 1987. 25 out of 33 trades were stopped out. June 1998) and also later in a second article ("Truth Be Told. or $ 3 8 . RATIO-ADJUSTED DATA (RAD) To overcome the d i l e m m a induced by the point-based back-adjusted contract. or 4 5 . 0 0 0 . I suggested in an article for Futures m a g a z i n e ("Data Pros and C o n s . it t o o k another 35 trades before they were hit again in October 1990. this we know. 6 % . with the latest roll m a d e in September 1999. This was a n n o u n c e d as big n e w s by the media.264. c = T h e close of the old contract on the day of the roll. To get a better feel for the benefits of the R A D contract. this n e w time series will also vary in level as c o m p a r e d to w h e n the true contract was traded. Figure 3. This increase in magnitude measured in points is due to the upward transition because of the many rolls. If we do so. 6 % ((486 . for a total drop of 270. After the crash of 1987 it t o o k close to 11 years before the S & P 500 experie n c e d a larger d r o p m e a s u r e d in points. 6 % . T h e important thing. F r o m t h e n on. however. This time the high of October 2 is at 486 and the low of October 20 at 264. with s o m e analysts and market writers even c o m p a r i n g it to the crash of 1987. the market fell by 152 points. but with a percentage difference once again equal to 4 5 . 13 out of 13 trades w e r e stopped out this way. but instead of keeping the point. it took 55 trades (out of a total of 197 trades) before any of these stops got hit in October 1987. and from July 1998 to present.85 points.3 shows what the October 1987 crash looks like through the perspective of the ratio-adjusted data ( R A D ) contract.4 to an October low of 929. for the period D e c e m b e r 1996 to present. let us take a closer look at this recent event and c o m p a r e it to the crash of 1987 (see Figure 2.or dollar-based relationship between two points in time intact.15) X 100 / 486).

therefore. In dollar t e r m s . $38. A n d while at it. this equaled a drop of $67.4 X 250). We discuss multimarket portfolios and trading strategies in m o r e detail in Part 5. there is no saying that there will be no larger drops c o m ing.38 PART 1 Evaluating Performance measured on the nonadjusted contract). the 1998 drop would have had to continue all the way down to the 652. Furthermore. Figure .350 X 250 X 0. however. MULTIMARKET PORTFOLIOS T h e benefits of the R A D contract also b e c o m e evident when you want to put together a multimarket portfolio.477 (0. In fact. or $136. the drop of 1998 only equaled 2 2 . system vendor.625 (218 X 250) N o n e of the above If you answer " n o n e of the above." what is the correct answer? T h e answer is: N o n e of the above. T h a t is almost twice as m u c h as the October 1987 drop. with the market trading at the 1. do not forget to get rid of those dollar-based stops and profit targets as well. the only way to do that is to use the R A D contract in combination with the percentage-based performance m e a s u r e s described in the previous chapters. what is the expected worst drop in equity for the S & P 500 if it were to equal the drop of 1987 and with the market trading at the 1. 5 % of total market value. For now we only state that the percentage-based calculations do not take into consideration h o w m a n y contracts y o u ' r e trading and.5 level.456). No wonder then that your software supplier. 6 % . I h y p o thetically traded the Japanese yen and the S & P 500 over a period of 10 years. to m a k e your systems m o r e robust and give t h e m a better chance to hold up in the future. That is. a n d if you.000 (152 X 250) $23. just because the drop in October 1987 stopped at 4 5 .900 (1. B a s e d on what we have learned so far the expected worst drop in equity. Let's stop for a little quiz: B a s e d on the information so far.282 X 333 X 250) $54. the crash of 1987 is still the largest relative drop in equity in modern times. for a total drop of 546.600 (270. In percentage terms. C. D. would like to treat it as such. B.350-level? Is it A. the drop in 1998 was not even half as b a d as the crash of 1987. resulting in 71 trades for the Japanese yen and 79 trades for the S & P 500.350 level is $153. T h u s . Let us look at the following example: U s i n g a regular 20-day breakout system that is always in the market. to equal the crash of 1987 in relative t e r m s . in your system building and analysis work. give each market an equal weighting in the portfolio. or market guru of choice says that your worst drawdown is still to c o m e — a n d they do not even know how to calculate it.9 points.725.

is worth approximately $3.200). They all represent foregone opportunities. you can't.000. However.4 The results from a multi-market portfolio traded on a one-contract basis.000. had you been able to place all trades in today's markets. trading o n e contract. resulting in an overall profit for the portfolio of about $80. no matter how you do it? N o .275 / 1. Now. but only $1.000. to m a k e results comparable between the two markets and at the same time relevant to today's markets. . can you place any of these trades for real. you would have had to trade approximately five Japanese yen contracts for every two S & P 500 contracts (3. so that is a good and valid question. rather than in a ten-year-old market situation to get a feel for how the situation was back FIGURE 3. those results are purely hypothetical. Isn't it better then to at least place t h e m h y p o thetically in today's marketplace to get a feel for what might happen today. H o w can I place all the trades in the same market at presumably the s a m e point in time? Well. we m u s t look at how m u c h a percentage m o v e in each market is worth today. In short.4 shows what the individual and c o m b i n e d equity curves would have looked like had we b e e n able to place all these trades for real.000. you say. wait a minute. It turns out that a 1% move in the S & P 500.200 in Japanese yen. in today's market.275. which in turn would have resulted in an overall profit of approximately $240. the Japanese yen would have e n d e d up with a profit of close to $60.CHAPTER 3 Futures Contract Data 39 3. and the S & P 500 with a profit of approximately $20. H e n c e . but let me ask you. of course not.

40 PART 1 Evaluating Performance then? I think we all can agree that it is better to k n o w w h a t might happen today. By m a k i n g the profits from the two markets m o r e comparable to each other we have raised the total profit from $80.0.) FIGURE 3.5. we m u s t put together a chart based on the (cumulative) percentage moves we would have been able to catch.4 (1 + 0. As you can see from Figure 3. rather than what h a p p e n e d ten years ago. something we have not taken into consideration in the above example.05 = 1 X (1 .5 The results from a multi-market portfolio traded on a percentage basis with compounded returns.75 (1 . but in an almost 2 5 % c o m b i n e d move in the w r o n g direction (the market moved against our position) for the S & P 500. W h a t we forget. resulting in an overall c o m b i n e d move that barely m a n aged to reach a profit of 5 % . A n d because both the dollar moves and the contract relationships between different markets are constantly changing. however.4) and then by 0. this is looking good.2 5 % = 1 X 1.000 to $240.05) = 1 + 5%. from the previous chapters we k n o w that only looking at hypothetical dollars m a d e is not the m o s t efficient way to j u d g e a trading strategy.75 = 1.0.25). multiply 1 by 1. (To understand how a 4 0 % positive m o v e and a 2 5 % negative m o v e can end up being close to zero.4 X 0. . that is: 1 + 4 0 % . Nonetheless. is that this relationship changes with time.000. Furthermore. this results in an approximately 4 0 % c o m b i n e d move in the right direction (the market moved in favor of our position) for the Japanese yen.

it stays the same. trading the S & P 500 from the short side. Worse yet. W h e n the squiggly line is below its fast and slow moving averages. they might not work at all because the underlying assumption for your strategy was not there in the first place. That is when you m u s t c o m p a r e the developm e n t of two different markets. this looks like a smooth ride with few erroneous signals a n d whipsaw trades. trigger a short position in the S & P 500 futures. Overall. which in turn can lead to perceptual changes in the historical relationship between the markets. which also adds s o m e smoothing to the historical data. but each is recalculated each t i m e there is a roll from one contract to the next. To work a r o u n d this problem. W h a t looked like a fairly smooth ride w h e n looked at through the pointb a s e d back-adjusted contracts. T h i s is because none of these contracts stays stationary in time. as it is perceived to have looked in D e c e m b e r 1984.7. PERPETUAL CONTRACTS No matter h o w good the back-adjusted contracts are there is a certain situation w h e n n o n e of them should be used. the trend for the spread is down a n d the T-bill can be expected to outperform the stock market and. p e r h a p s to find the one that seems to have the strongest relative strength. in this case. turns out to be a close-to-losing endeavor w h e n giving both markets an equal percentage weighting and c o m p o u n d i n g the results. turns out to be a little bit m o r e j a g g e d a n d dangerous w h e n looked at through the original contracts as they actually were trading at the time. through the actual contracts traded at the time. you m u s t put together a so-called perpetual contract that stays stationary in time. look at Figure 3. This can be very deceiving if you're not aware of the problem and might m a k e you build trading systems that will not work as well as they seemed to have done in the past. despite the rollovers. which shows you the s a m e market action. with m o r e j a g g e d curves and several w h i p saw trades. what at a first glance seemed to be a profitable portfolio w h e n tested on a one-contract basis in dollar terms. once a historical value is calculated. Now.6 shows a ratio spread between the S&P 500 and the T-bill. That is. T h e m e t h o d for calculating a perpetual series m a d e up of two contracts (the nearest contract and the second nearest contract) is as follows: First set a forward quote day a certain n u m b e r of days in the future and keep this n u m b e r of days constant as time passes. as it looked in D e c e m b e r 1984.CHAPTER 3 Futures Contract Data 41 T h u s . using the point-based back-adjusted contract. w h i c h leaves plenty of r o o m for a quick buck. This time the picture is no longer as clear. This is what you actually would have seen had you monitored this spread at the time. T h e difference in the two pictures is caused by the back-adjustment process in combination with the ratio calculation. Then calculate the n u m b e r of days . Figure 3.

.6 A spread between two markets put together using point-based back-adjusted contracts.42 PART 1 Evaluating Performance FIGURE 3.

CHAPTER 3 Futures Contract Data 43 FIGURE 3.7 A spread between two markets put together using the actual contracts at the time. .

. W h e n it expires it b e c o m e s 0. the less weight it gets. roll it forward one position for every new bar) the relative distance for the two contracts in relation to this day changes for each day. T h e closer the nearest contract c o m e s to expiration. W h e n the distance to the forward quote day is kept constant (i. multiply the relative distance for both contracts by their respective prices and add t h e m together. Finally. the second nearest contract is now the nearest contract. and the third nearest contract is the n e w second nearest contract. for today's perpetual price.e.44 PART 1 Evaluating Performance between the expiration days for the two contracts a n d the forward quote day.

not all data can be used all the time. it also is important to use the right type of data. no matter where and when they are derived. Later chapters take a closer look at how to evaluate a system or a strategy that is already up and running. To properly evaluate a trading system. As we have seen. and why it could be a good idea to expand the analysis work a bit with the help of a spreadsheet program. like MS Excel or Lotus 1-2-3. it is of paramount importance to use a set of universal measures that give an equal weighting to all the trades. To accomplish this. knowing when to use what is vital in building a robust and profitable trading system. to m a k e sure that it stays on track and performs as it should.PART ONE A Few Final Thoughts About Part 1 This part started out by taking a closer look at which system testing measures are m o r e useful than others. Before we go any further in the evaluation process. 45 . Building a system suited for a specific type of market action or trading perspective is the major topic in the next section. we need a couple of systems to trade.

and something completely different for s o m e o n e else. personality. or should we try to capture the shorter-term moves within a previously defined trend? Or are we trying to a c c o m plish something completely different. we must understand what it should try to accomplish. however. Aside from the psychological a s p e c t s — whether the system fits your style of trading.PART TWO System Concepts In the first part. which must be addressed a n d — i f not answered—at least contemplated. h o w should it be built and tested so that all the markets have an equal impact on the final result? Should we try to capture a certain type of trending move. Before we can start designing and investigating any trading system. r e a d y . W h a t is long t e r m and short term? Or what is high and low? There really are no generic. Whatever your answer is. We must know if the trading methodology should be long-term. it does not stop there. or perhaps even intraday. we discussed a few important concepts to understand and think about before the system building process can begin. there is a set of basic questions of a m o r e philosophical character. There is m o r e than one thing to think about w h e n trying to put together a mechanical trading strategy. short-term. or buying or selling volatility with the help of options? T h e n there is the question of what type of m o n e y m a n a g e m e n t strategy to use. like arbitraging o n e market or contract month against another. and w h e n this should be taken into consideration. But even before these fundamental questions can be answered. Unfortunately. trading budget—several other m o r e technical questions m u s t be answered as well. 47 . it is correct for you and not for anyone else to question. because long t e r m for me might be short t e r m for you.m a d e answers to these questions. Should the system be market specific or tradable over a wide variety of markets? If the system is a multimarket system.

48

PART 2

System Concepts

Before you answer, however, here is some food for thought: w h a t if, instead of measuring long-term or short-term market activity in regular time units, such as minutes, hours, or days, you were to m e a s u r e it in n u m b e r of bars (market time units) encapsulating as m u c h time (regular time units) as you d e e m necessary? Or perhaps, long t e r m and short term simply should be defined by the system, where a breakout-type system is always considered long term while a top-and-bottom picking system is always considered short term, no matter for how long the trade actually lasts, whether measured in regular time units or market t i m e units? A n o t h e r important thing to consider before researching a trading system (and a c o m m o n l y debated question a m o n g system builders) is whether it is a good idea to save some of the data for out-of-sample testing. T h e answer is that it depends on the system and its underlying assumptions; m a n y times it shouldn't m a k e any difference. In a correctly built system, y o u ' r e interested in exploiting something that should, on average, work well in several different markets and over several different t i m e periods, not with any specific type of anomaly that might or might not still exist in a particular market. By saving some of your data for out-of-sample testing, you can strengthen your trust in the m o d e l , but in doing so, you also use up valuable data that could have added even further to the robustness of the system. Only you can m a k e that choice. T h a t said, in m a n y of these examples, we work with an out-of-sample period for comparative reasons. Whatever you choose, it is important to r e m e m b e r that by necessity all system testing and design m u s t be m a d e on historical data. T h e trick, then, is to m a k e as good use as possible out of these data and m a k e your evaluation measures as forward looking as possible. Building systems is no easy matter, and you h a d better k n o w the o d d s against you before you get started. For instance, if you only traded the S & P 500 index approximately every fifth day (I picked out all days randomly 20 times, calculated the c o m p o u n d e d returns, and averaged the results), between M a y 1992, and October 1999, and if you could have been absolutely sure that all these days would have been d o w n days, you would have m a d e a w h o p p i n g 71 times the index. If you instead h a d b e e n able to pick the same a m o u n t of up moves, you would have m a d e an incredibly 161 t i m e s the index. This tells us that we are m u c h better at going with the underlying trend (as it undoubtedly has been up during the last few years), and if we can only t e m p e r ourselves and trade only on a few selected days, the rewards can be v e r y good. No wonder day trading is such an alluring business. But rewards like this do not c o m e without a price. T h e above examples assume that we can be 1 0 0 % sure of the move, but what if we can only be 5 0 % sure? Well, then the expected return sinks to only 0.25 t i m e s index, if going long only, or to a mere 0.11 times index, if going short only. Similarly, if y o u ' r e a long-term trader and prefer to c o m p a r e yourself to a buy-and-hold strategy, you could have increased you return to 4.9 times the index, had you been able to stay out of the market every 20th d a y — i f you also were absolutely sure that day also was a down day. But if you only were 5 0 % sure about

CHAPTER

4

Picking Tops and Bottoms

W h e n it c o m e s to trying to trade shorter term, or picking tops and bottoms, m a n y system traders usually start out trying to use any or all of the m a n y oscillator-type indicators that have swamped the field of technical analysis in recent years. Examples of such indicators are RSI, Stochastic, Momentum, Rate-of-change, MACD, and Plus and Minus DMI. M a n y others exist, and the list can be m a d e m u c h longer, but those mentioned above are the m o s t c o m m o n and popular ones. T h e s e indicators are believed to be g o o d top and bottom pickers because they're believed to be able to anticipate or even predict the market by moving in and out of overbought and oversold regions, or simply by not " c o n f i r m i n g " the moves of the market. I believe, however, that one main reason you have picked up this book in the first place is that you have c o m e to realize that m a n y of these indicators simply do not work nearly as well in real-time as they seem to do when looked at on a historical price chart. If all these indicators do not work, how then should one go about finding profitable short-term trading opportunities? Well, probably the most important task for you as a trader, especially if you prefer to trade fairly short term, is to mine the data from which you hope to m a k e a living. If, for nothing else, it m u s t be done to get an in-depth feel for the markets you are trading, which in turn should help you to better interpret the current market situation and the indicators you are using to g a u g e it. But before you can go about doing this, you must know exactly what it is you should look for and how to measure it. To build a system that works equally well in all markets, you must decide what type of data to use. With the ordinary point-based back-adjusted contract, you can estimate how m u c h profit your system would have made historically in each market,
51

52

PART 2

System Concepts

but because of the way it is constructed you cannot channel any information from one market to another. To give all moves an equal importance, and to derive the necessary input for your calculations, it is of paramount importance that you only use percentage-based calculations—and if you are a futures trader, also use the newly invented R A D contract (see "Data Pros and Cons," Futures magazine, June 1998, and "Truth Be Told," Futures magazine, January 1999). In a trending market, the dollar value of the magnitude of the moves changes with the market value, so that if the market is in an up trend, with a steadily increasing market value, the dollar- or point-based move is likely to increase as well. T h e relationship between the moves and the market value stays the same, resulting in a average percentage move that is m o r e likely to stay the same, no matter at what level the market is trading. This is also the contract you need as soon as you start to work with any type of percentagebased decisions, such as rate-of-change analysis or percentage-based stops. Notice that the key words here are "equally w e l l " instead of "equally profitable." To build a trading system that continues to work equally well in the future as it did during the building process, w h e n it w a s hypothetically traded on historical data, it also is important to understand the difference between a good working system and a profitable system. Furthermore, once this is fully comprehended, it is even m o r e important to understand w h y a good working system is not necessarily a profitable system, but why a profitable system always m u s t be a good system. T h e s e key concepts that will be stressed continuously throughout this book. W h e t h e r a good working system will be profitable or not has nothing to do with the system, but rather with the level at which the market is currently trading a n d the dollar value of that level. For instance, suppose the dollar value of the S & P 500 futures contract were to be lowered from today's $250 per point to $2.50 per point. H o w many S & P 500 systems would still be profitable in the future, or even test profitable in hypothetical back-testing? Probably not many. T h e point is that the actual value of the market and its move is a technicality decided by the exchange, which has nothing to do with whether the system is good at capturing the moves of the market. D u r i n g the building process, the actual dollar values are of no concern. Instead, we focus on the universal measures: the profit factor, the percentage move, and the n u m b e r profitable trades. To m a k e sure the results do not differ too m u c h between markets, we also look at the standard deviations of all these measures. T h u s , a well-working s y s t e m — a good system—is a system that captures as m a n y and as large moves as possible in any and all market environments, as m e a s ured with a universal m e t h o d like those mentioned. For a system to be profitable, however, it must be traded in a market where the moves it is designed to catch also are worth taking. This has nothing to do with the system, but rather with the level at which the market currently is trading and the dollar value of that level. Once you k n o w what to look for in terms of the data and the different measurements, you also must figure out exactly what it is you would like to do. That is,

CHAPTER 4

Picking Tops and Bottoms

53

should you only take trades that go in the s a m e direction as the longer-term underlying trend, or should you also look for counter-trend moves? Whatever your answer, you then m u s t ask yourself whether you should try to anticipate the turning point with an early limit-order entry; or play it a little safer and look for proof that the preceding m o v e has run its course, and then enter with a stop? Furthermore, once you're in the trade, should you then try to ride it for as long as possible, or should you only go with the short-term thrust and perhaps exit m a n y of the winners with a limit order at a specific price or after a move of a certain magnitude? In this chapter, we took a closer look at a few different system ideas that all have a high likelihood of producing a large n u m b e r of profitable trades for a decent overall profit. T h e first two are a couple of market-specific data m i n i n g systems, w h e r e we only are interested in the specific characteristics of that particular market, or group of related markets, although it is fully possible and realistic to apply both the research technique and the systems to any market. T h e next is based on a proprietary indicator perhaps best described as a mix b e t w e e n percentage-based analysis, Bollinger b a n d s , and pivot-points. T h e last system, which really is only a demonstration of how to c o m e up with a good set of exit rules, I have given the n a m e Black Jack, because of its ability to m a k e u s e of small statistical advantages, w h i c h are the s a m e for all markets, in an attempt to p r o d u c e long-term profits out of several smaller a n d sometimes just barely profitable trades. It is especially designed to w o r k well together with m o r e advanced m o n e y m a n a g e m e n t strategies, s u c h as Ralph Vince's Optimal f, or fixed fractional investing. (In Part 3, we will continue to work with different ways of deriving exits, which we then apply to all our systems.)

CHAPTER

5

Data Mining

W h a t will the market do today? E a c h day we ask our fellow trader friends and ourselves the same old question, and each day the market answers us at the end of the day. One would think that after having asked the question and observed the answer, say, a couple of thousand times or so, we either would learn the answer or stop wasting time pondering the question. Yet, we keep on asking and each day the market baffles us, by always behaving m o r e or less out of sync with what we had expected. O n e day the stock market is behaving exactly as anticipated, except for that little dip after lunch that had us stopped out with a loss. W h i c h , by the way, reminded almost everybody about how the T-bond used to behave a couple of years ago, w h e n the typical characteristics of the European currency crisis were casting their shadow over the financial world. Then, the next day, the very same market behaves in a way never seen before. Except for your b u d d y Joel that is, w h o traded the grain and meat markets back in the 1970s and therefore recognizes the similarities to how the corn market used to behave back then. T h e point I am trying to m a k e here is that it's very difficult to distinguish the behavior of one market from that of another, and the very s a m e second that we think w e ' v e nailed it, it all changes and the experience gained loses almost all its value. Let's face it: can you really say that the stock market today behaved exactly as the stock market usually does, or w a s it m o r e like the recent coffee market or the lumber market in the 1980s? And, if it behaved like the coffee market, is the recent behavior for the coffee market really typical for coffee at this time, and on and on and on? Is there really a specific, consistent behavior for each and every market, and, if so, can it be traded profitably? To find out you m u s t e x a m i n e each and every market very carefully and ask yourself an a b u n d a n c e of questions:
55

56

PART 2

System Concepts

For h o w long does a typical trend last? W h a t are the typical characteristics of any corrections? W h e n can a m o v e be considered to have g o n e too far? W h a t is the likelihood for a certain amount of days in a row in the s a m e direction? A n d m o s t important: h o w can you benefit from this information and i m p l e m e n t it in your arsenal of existing trading tools? T h o s e are only s o m e of the questions to ask when you start m i n i n g your data to c o m e up with high-probability trading opportunities; or at least, in trying to avoid the m o s t obvious bad ones. For example, if the market has been in a down trend for four m o n t h s straight but you k n o w that a typical d o w n trend should last for only two months, it probably isn't too good an idea to a d d to your position w h e n your breakout system signals that you once again should go short. Granted, the market might have changed from w h a t is typical for o n e market to what is typical for another, but at least you still have the long-term statistics on your side. Alternatively, if you already are short, perhaps it is a good idea to start scaling back, no matter what your system is telling you. Or, if you k n o w that only 2 2 % of all down moves, measured on daily data, last for m o r e than two days, you could speculate on the high statistical likelihood for a short up trend to follow, set up a contrarian trading system, and p r o b e the market with a small position whenever the market has fallen for two days or m o r e . Similarly, if you k n o w that only 9% of all d o w n moves result in a decline of 8% or m o r e , measured on monthly data, you could take a long position, betting on an extended up m o v e to follow, as soon as the market has declined that amount. To get a feel for relationships like these, it is a good idea to put together a set of tables like Table 5.1 through Table 5.3, which serve as good starting points for experimentation. All tables have been put together using the R A D contract for the S & P 500 futures market, with data expanding from January 1985, to D e c e m b e r 1994, and January 1995 to October 1999. (The m o s t up-to-date data, from January 1995 to October 1999, w e r e used for out-of-sample testing, and are placed within parentheses.) F r o m Table 5.1 you can see that during the period January 1, 1985, to D e c e m b e r 3 1 , 1994, the average up m o v e for the S & P 500, measured on weekly data, went on for 2.29 weeks, for an average total gain of 3 . 3 % , while the average down move, measured on monthly data, went on for 1.35 m o n t h s with an average total decline of 4 . 8 2 % . F r o m Table 5.2 you can see that measured with daily data, 2 5 % of all up moves lasted for longer than two weeks, while only 6% of all down moves could be expected to last for three weeks or m o r e . In Table 5.3 the monthly amplitudes are within the parentheses in the headers. M e a s u r e d with weekly data, 4 2 % of all up moves resulted in an increase of the market value of m o r e than 3 % . M e a s u r e d with monthly data, 4 6 % of all up moves resulted in an increase of the market value of m o r e than 6%. To put together Tables 5.1 through 5.3 you m u s t be able to export your data into a text file, which can be opened with the help of MS Excel or any other spreadsheet program. O n c e in Excel, you first must calculate the percentage change of the closing price with the following formula:

CHAPTER 5

Data Mining

57

TABLE

5.1

Data mining summary. Move per period Periods in move Daily data All moves Up moves Down moves 0.68% 0.69% (0.77%) 0.68% (0.71%) 1.95 2.04 (2.17) 1.86 (1.93) Weekly data All moves Up moves Down moves 1.52% 1.44% (1.78%) 1.64% (1.67%) 1.97 2.29 (2.21) 1.65 (1.52) Monthly data All moves Up moves Down moves 3.32% 3.22% (3.47%) 3.56% (3.50%) 1.72 2.09 (3.31) 1.35 (1.25) 5.73% 6.72% (11.93%) 4.82% (4.36%) 2.99% 3.30% (3.98%) 2.70% (2.52%) 1.34% 1.41% (1.68%) 1.26% (1.36%) Amplitude

TABLE

5.2

Periods of moves of certain length. 1 >1 Daily data All moves Up moves Down moves 50% 4 6 % (43%) 5 3 % (53%) 50% 54% (57%) 4 7 % (47%) Weekly data All moves Up moves Down moves 50% 4 1 % (43%) 6 0 % (67%) 50% 5 9 % (57%) 4 0 % (33%) Monthly data All moves Up moves Down moves 57% 3 7 % (38%) 7 6 % (83%) 43% 6 3 % (61%) 24% (17%) 16% 2 3 % (46%) 9 % (8%) 6% 9 % (46%) 3 % (0%) 25% 3 3 % (28%) 17% (13%) 11% 17% (15%) 6% (4%) 23% 2 5 % (33%) 2 2 % (25%) 11% 13% (15%) 10% (12%) >2 >3

AND(F3 < 1 .">0")) M . continue the calculation in the next column: =IF(G3>=G4.3 Percentage of moves of certain amplitude. To calculate h o w m a n y periods in a row a certain move lasts.SIGN(F3-1)XG3.PRODUCT(INDEX(F:F. (For the down moves. type in the following sets of formulae at the b o t t o m of the spreadsheet to derive the necessary n u m b e r s for the tables.ROW()-ABS(H3)+l. 1 .ROW().l))-1.58 PART 2 System Concepts TABLE 5. first use the following formula in the adjacent column: =IF(OR(AND(F3 > 1 . in the last column."") After you have filled in all the calculations."") Finally. 1(2)% >1(2)% >2(4)% Daily data All moves Up moves Down moves 55% 5 1 % (42%) 5 9 % (57%) 45% 4 9 % (58%) 4 1 % (43%) 20% 2 2 % (28%) 18% (25%) Weekly data All moves Up moves Down moves 26% 2 3 % (10%) 2 8 % (28%) 74% 7 7 % (90%) 7 2 % (72%) 53% 6 1 % (75%) 4 5 % (43%) 36% 4 2 % (59%) 3 1 % (27%) 25% 2 8 % (41%) 2 2 % (21%) 16% 18% (24%) 14% (18%) 9% 1 1 % (18%) 8 % (11%) 4% 4 % (11%) 4 % (6%) 2% 2 % (4%) 2 % (3%) >3(6)% >4(8)% >5(10)% Monthly data All moves Up moves Down moves 25% 2 6 % (0%) 2 4 % (25%) 75% 7 6 % (75%) 45% 3 2 % (33%) 35% 4 6 % (62%) 2 4 % (17%) 19% 2 9 % (46%) 9% (8%) 12% 1 7 % (46%) 6 % (8%) 7 4 % (100%) 5 7 % (92%) =E3/E2 w h e r e E denotes the column where the data is stored. ) To calculate the total n u m b e r of periods u p : =ABS(SUMIF(H$3:H4429.F2 > 1)).G2 + 1 ) T h e n . use the following formula to calculate the percentage move: =IF(H3<> ".l): INDEX(F:F.">0")) To calculate the total n u m b e r of moves up: =ABS(COUNTIF(H$3:H4429.F2 < 1). simply change " > " t o " < = " .

set up a system that only goes long as soon as you have a down day that follows a down week that follows a down month. together with the R A D contract for exporting the results into a spreadsheet program. the requirements for a short position could be two up days. U s i n g the trade-by-trade export function from Part 1. looks something like this: Condition 1 = CloseM(l) > C and CloseW(l) > C and C[l] > C. If Condition 1 = True and MarketPosition = 0 Then Buy ("Go long") at open. and two up m o n t h s . A .">0. for instance. T h e TradeStation code for this simple system. . you can. Because of the natural u p w a r d drift in the stock market and because of w h a t the data mining has shown about up moves going on for a longer period of time. If C[2] < C[l] and C[l] < C Then ExitLong ("Exit long") at close..4 and 5. two up w e e k s ..CHAPTER 5 Data Mining 59 To calculate the average n u m b e r of periods in an up move: =H4431/H4432 To calculate the average percentage amplitude for an up move: =SUMIF(I$3:I4429.">0 ) To calculate the average percentage amplitude for each period within the up move: =((H4434+1) (1/H4433)-1) To calculate the likelihood for an up m o v e to last for two periods or m o r e : =COUNTIF(H$3 :H4429. Condition2 = CloseM(2) < CloseM(l) and CloseM(l) < C and CloseW(2) < CloseW(l) and CloseW(l) < C and C[2] < C[l] and C[l] < C.01")/H4432 To test a simple trading system that m a k e s use of information like this.5. If Condition2 = True and MarketPosition = 0 Then Sell ("Go short") at open. I f C [ l ] > CThen ExitShort ("Exit short") at close. you can put together a performance s u m m a r y table like those in Tables 5. . and then using the Excel formulae we also derived in Part 1." > 1 ")/H4432 To calculate the likelihood for the amplitude of an up m o v e to be greater than 1%: =COUNTIF(I$3:I4429. G o l d D i g g e r I. >0")/COTJNTIF(I$3:I4429.

928 189. w a s reinvested at each new trade.5 Performance summary for Gold Digger I. or c o m p a r e h o w your system would have held up against a b u y . for an average gain p e r trade of 0 .45 1.123 Losers Lrg loser Avg loser Drawdown 9 -19. Despite the not too b a d results. while the largest loss h a d d e c r e a s e d to 4 . January 1985-December 1994.02 -7.10 66. including previously m a d e profits. January 1995-October 1999.4 Performance summary for Gold Digger I. Total trades Profit factor Avg profit St Dev 0.18% 2. Table 5. we are a s s u m i n g that the entire equity.65% -64.35% 26.190 3. provided that you can b u y fractions of a share.33% -14.74 63.43% 1.350).76 1.27 1.530 -4. especially not in the futures m a r k e t s . is to c o m p a r e different systems a n d m a r k e t s with each other on an equal basis.392 p e r contract traded ($1.467 — $ 7 5 ) in the i m m e diate future.684 161.a n d . you can expect this system to generate approximately $1. This particular system is n o t a good system to trade if you compare the statistical characteristics in Tables 5.456 -24.175 in today's m a r ket value).16 56. 4 6 7 . 2 0 % (or $14. but w i t h an expected $75 in slippage and c o m m i s s i o n .719 Winners Lrg winner Avg winner Cum profit 7 3.5 also shows that t h e d r a w d o w n decreased substantially from close to 2 7 % .3 for the m a r k e t dur- TA B L E 5.71 36.739 -90. with the S & P 500 trading around 1.60 PART 2 System Concepts TABLE 5. For d r a w d o w n a n d cumulative profit values to be correct. the average profit h a d increased to $ 1 .40% 105 2. W h e n tested on out-of-sample data.67% 11.h o l d strategy over the s a m e t i m e period.12 -1. No m o n e y w a s deducted for slippage a n d c o m m i s s i o n .146 Over the period January 1. to 7% for the out-of-sample period. to D e c e m b e r 3 1 .338 Losers Lrg loser Avg loser Drawdown 3 -4. 1 8 % (or $597 in today's m a r k e t value.01% 251 1. but it could be so in the stock market. A fairly low d r a w d o w n and standard deviation also m a k e us want to continue the research. however. As m e n t i o n e d in Part 1. a m o n g w h i c h close to 6 4 % w e r e profitable. however.644 3. this is s e l d o m possible. W h a t this does allow you to do. for the in-sample period. 1985.467 4.20 -1.1 to 5. An ever-low standard deviation a n d a h i g h p e r c e n t a g e profitable trades k e e p us interested. Gold D i g g e r I p r o duced 251 trades. it is important to r e m e m b e r that Gold Digger I is put together merely to illustrate a simple system for taking advantage of a market's statistical characteristics.34 597 6. Total trades Profit factor Avg profit St Dev 0.799 Winners Lrg winner Avg winner C u m profit 15 7.300 .20 33.175 -3. 1994.40 -26.09 47.

C o m p a r i n g the in. the only way to build a mechanical trading system that holds up and behaves the s a m e way in the future as it does during testing—no matter what the longert e r m underlying trend looks like—is to focus on the shorter time perspective. 7 2 % to 3. This suggests that when the market changes its characteristics or m o d e . This is two percentage points less than what could be expected looking at the period ending in 1994.31 m o n t h s and 1 1 . Or. If Conditionl =True and MarketPosition = 0 Then Buy ("Go long") at open. This is a very important conclusion. put differently: no matter what long-term m o d e the market is in. you notice the average m a g n i t u d e of an up move. a modified version of the original system considers only daily and weekly data.APTER 5 Data Mining 61 ing the first 10-year period. the short-term statistical characteristics are likely to still look the s a m e and be close to stationary in nature. Also. w h e r e the percentage of up moves with an amplitude of m o r e than 5 % . IfC[2] < C[l] and C[l] < C T h e n ExitLong ("Exit long") at close. for the period up until 1994 there was only a 9% chance for an up move to last for m o r e than three m o n t h s . Hence. when looking at the m o n t h ly data. with trades lasting for no longer than approximately a week and using as little historical data for the signals as possible. C o n d i t i o n = CloseW(2) < CloseW(l) and CloseW(l) < C and C[2] < C [ l ] and C[l] < C. To test if this could be true. there was a 4 6 % chance for the s a m e type of move. . with the characteristics for the latest four years. has grown from 2. measured on weekly data. these changes can be very hard to detect w h e n looking at a shorter time frame within the bigger picture. but for the period spanning January 1995 to October 1999. in the table describing the likelihood for periods of moves of certain length. which also does not differentiate between up trends and down trends: Conditionl = CloseW(2) > CloseW(l) and CloseW(l) > C and C[2] > C[l] and C[l] > C. as suggested by the following TradeStation code. the n u m b e r of persistent up trends measured on monthly data has skyrocketed. 9 3 % . the m o s t recent study suggests that 5 7 % of up moves m e a s ured on weekly data go on for more than one week. measured on monthly data. while the n u m b e r of persistent up trends measured on weekly data has decreased. In the s a m e table.a n d out-of-sample periods with each other shows that ever-sosmall differences between the daily and weekly data soon add up to quite large differences in the monthly time perspective. while the n u m b e r of up moves with an amplitude of m o r e than 10%. has skyrocketed by 1 7 0 % (from 1 7 % to 4 6 % ) .09 m o n t h s and 6 . For instance. measured on monthly data. because if this is true. has only increased by 3 3 % (from 1 8 % to 2 4 % ) . T h e s a m e p h e n o m e n o n also can be seen in the table for percentage of moves of certain amplitude.

6. s e e m to be a little m o r e robust. Therefore. the lower standard deviation. or if in a long position. (We will talk m o r e about stops a n d exits in Part 3. It does. but with d o w n days of greater m a g n i t u d e than up days.840 -24. F r o m the TradeStation code. for trades in the immediate future. IfC[l] > CThen ExitShort ("Exit short") at close. B e c a u s e Gold D i g g e r II is a simpler system (less curvefitted) than G o l d D i g g e r I.84 -7. for instance.31 1. it should follow that you should m a k e as g o o d use of those bars as possible. every other day is an up day and every other day is a d o w n day. happen if the m a r k e t immediately takes off in the w r o n g direction.710 Losers Lrg loser Avg loser Drawdown 4 -4. as j u d g e d by a lower standard deviation.26 37.671 3. its performance is not quite as good.) BETTER USE OF YOUR DATA If you are better off using as few bars as possible in your system. 3 1 % (or $1.38% 107 1. No m o n e y was deducted for slippage and commission. not takTABLE 5.703 -2.62% 24.31% 1. while we will exit a short trade as soon as we have a down day.62 PART 2 System Concepts If Condition2 = True and MarketPosition = 0 Then Sell ("Go short") at open.38% -13. a m o v ing average usually only looks at one particular price within a t i m e series.666 Winners Lrg winner Avg winner Cum profit 6 7. but this is easily accounted for. however. January 1995-October 1999.98 1. in combination with Gold D i g g e r IPs b e i n g m o r e symmetrical in nature (less curve-fitted). B e c a u s e the market does not always behave as we want it to. Total trades Profit factor Avg profit St Dev 0. as can be seen from the values for the largest a n d average losing trades. the losses can be quite severe before we are allowed to exit the trade. with 6 3 % profitable trades and an average profit of 0 .503 . however. This can.6 Performance summary for Gold Digger II. Tested on the out-of-sample period this system h a d 107 trades. as j u d g e d by a slightly lower profit factor and percentage profitable trades and a m u c h lower cumulative profit.045 in today's market value). For instance. A l t h o u g h the average profit and profit factor are slightly lower for Gold D i g g e r II w h e n c o m p a r e d to Gold D i g g e r I. T h e results for this version of the system can be seen in Table 5.364 127. you can see that the only exit criteria we have on the long side is two up closes in a row. the lack of properly developed stops still m a k e s this a dangerous system.00 37.06 -0. by deducting an appropriate a m o u n t from the expected value of the average trade. m a k e s it clear that Gold Digger II is the preferable system.84 62.045 4.

Aside from the fact that you will not always have that m u c h data to experiment with. if and w h e n the market trades at the same level again. where you are measuring the height of a random sample of m e n that you meet throughout the day. T h e one standard deviation boundaries and the two standard deviation boundaries hold approximately 6 8 % and 9 5 % . and very tall (above 6 feet 6 inches).) Let us say that the average for all 20 m e n was 6 feet and the standard deviation was 3 inches. which is the m i n i m u m amount required to make any statistical conclusions. both in time and between different markets. Using regular m o v i n g averages a n d Bollingerb a n d . all open. To develop and use the same type of research in the markets.000. it is important to note not at what level the markets are trading. average height (between 5 feet 9 inches to 6 feet 3 inches). and depending on w h e r e they are done. tall (between 6 feet 3 inches to 6 feet 6 inches). we m u s t use the R A D contract. a value of 1. very short (below 5 feet 6 inches). That is. the shortest man. 5 % (0.t y p e indicators to identify trading opportunities also present a problem in that the readings are not always comparable over time. short (between 5 feet 6 inches to 5 feet 9 inches). B u t for each investigation. while still others say 100 or more. it is close to 1. these levels will vary slightly. of the m e n m e a s u r e d and the expected height of all m e n to be measured in the future. you also know that the c h a n c e of meeting two such m e n in a row is a m e r e 2 . which keeps the percentage changes constant instead of using the actual point or dollar values. respectively. a n d at the s a m e time m a k e it possible to m a k e all of the indicator observations comparable with each other. at least s o m e of the previous information will be useful in estimating the future outcome or what can be expected for the next day. Wouldn't it be great to c o m e up with an indicator that can m a k e use of all four of these price extremes.68) X 100 / 2). For each and every investigation d o n e this way. After five days you have 20 observations. the tallest man. high. To calculate the move rather than . but m a y not be so a couple of years from now. With this information at hand y o u now can classify all m e n that you met as. It does not matter in what order you write t h e m down. for instance. some say 20. and the last man you have measured that day. For e x a m p l e . (This number varies. For instance. In the medical field. At the end of the day you write down the height of the first man. regardless of w h e n they happened or at w h a t level the market w a s trading at the time? Contrast the above with a statistical examination. because the chances of meeting a tall or very tall m a n are approximately 1 6 % ((1 — 0. others say 30. and low prices usually are ignored.16 X 100). w h e n calculating a moving average of closing prices. It all depends on where the market is c o m i n g from at that point in time. To m a k e each move comparable to all others. there is no reason why you shouldn't treat your m o n e y any less seriously.16 X 0.350 for the S & P 500 index m a y very well be below its moving average and standard deviation bands today. but the amplitude of the moves. shorten the lookback period by 7 5 % .CHAPTER 5 Data Mining 63 ing into account that each b a r actually holds three other important price levels.

for a total of 20 data points for the last five bars: Input: VSStd(l). no matter where and w h e n they happened. End. For SumArr = 0 To 19 Begin If SumArr = 0 Then SumVS = 0. the obvious level from which to normalize a person's height is the ground on which he stands.000 move equals only a 6% move. it m a k e s little or no sense to measure each man's height from sea level instead of from the ground where he is standing. DiffArr(O). H e r e each move has been normalized to the previous bar's closing price. DiffVS(O). In the first case.C[SetArr + 1]) / C[SetArr + 1]. compared to the s a m e dollar move if the market is trading at 2 5 0 . a $20. StdVS(O). For SetArr = 0 To 4 Begin VS[SetArr * 4 + 0] = (0[SetArr] . That is.350 level.C[SetArr + 1]) / C[SetArr + I]. it is necessary to normalize the data. VS[SetArr * 4 + 1] = (H[SetArr] . you will be 6 feet tall whether you are standing on a pair of skis in Vail or lying down on a beach towel in Hawaii. while in the latter the s a m e dollar move a m o u n t s to over 3 0 % . FNameC*"). In m a r k e t research. TradeStrl(""). it also is necessary to relate each move to a base or ground level.C[SetArr + 1]) / C[SetArr + 1]. Array: VS[20](0). VSLow(O). for a comparison to be meaningful. VS[SetArr * 4 + 2] = (LfSetArr] . SumArr(O).000 if the market is trading at the 1.64 PART 2 System Concepts the level. It would m a k e little or no sense to c o m p a r e your height in percentage terms to h o w high you are from the ocean. A n o t h e r difference between m e a s u r i n g m e n and market action is the use of absolute and percentage values. four observations were kept each day. B u t in the market there is a huge difference between a move worth $20. VSHigh(O). This can be done for the markets as well with what I call the M e a n d e r indicator. SetArr(O). Continuing the comparison with measuring m e n on the street. If SumArr = 19 Then . AvgVS(O). If you are 6 feet tall. moves to the highs.C[SetArr + 1]) / C[SetArr + 1].m e n example. Vars: SumVS(O). SumVS = SumVS + VS[SumArr]. like the previous bar's close or the average price of the previous bar. there are a few levels to which e a c h m o v e can be normalized. VS[SetArr * 4 + 3] = (C[SetArr] . it is of p a r a m o u n t importance to m e a s u r e market moves in percentages rather than point or dollar values so that you can c o m p a r e different types of moves. The following TradeStation c o d e sets up an array containing all the opening moves. moves to the lows. T h u s . VSMid(O). and moves to the close. In our m e a s u r i n g .

DiffVS = DiffVS + Square(VS[DiffArr] . 2) + ". If CurrentBar > 5 then Begin TradeStrl = NumToStr(Date. All moves are measured with the previous bar's close as a base. End." + "VS Low" + "." + NumToStr(VSLow[l]. TradeStrl). Plot2(VSMid. If DiffArr = 19 Then StdVS = SquareRoot(DiffVS / 20). "VS Mid"). TradeStrl)." + NumToStr(VSMid[l]. Plotl(VSLow. you have an indicator that m a k e s the m o s t out of each bar's price action and allows you to trade intraday using only end-of-day data or intraweek. FileAppend(FName. End." + NumToStr(Low. If CurrentBar = 1 Then Begin FName = "C:\Temp\" + LeftStr(GetSymbolName. 2) + ". VSHigh = C * (1 + (AvgVS + StdVS * VSStd))." + "VS Mid" + ".csv"." + NumToStr(High. End. T h e last part of the code holds the n e c e s s a r y instructions for exporting all the data into a text file for further analysis . 2) + ". End.StdVS * VSStd))." + "Close" + ". 2) + NewLine. For DiffArr = 0 To 19 Begin If DiffArr = OThen DiffVS = 0. 2) + ". W h e n the calculation is completed. 2) + ". the c o d e calculates the average move and the standard deviation before the data are denormalized to fit with the latest price action." + NumToStr(Open." + NumToStr(VSHigh[l]." + "Low" + "." + "Open" + ". VSLow = C * (1 + (AvgVS .AvgVS)." + "High" + ". FileDelete(FName). using a m i x of weekly a n d daily data. FileAppend(FName. 0) + ". 2) + ". VSMid = C * (1 + AvgVS). "VS High"). Plot3(VSHigh. TradeStrl = "Date" + "." + "VS High" + NewLine. O n c e the data are collected.CHAPTER 5 Data Mininj 65 AvgVS = SumVS / 20. 2) + ". "VS Low")." + NumToStr(Close.

With the necessary data imported into Excel. the long risk level (stop loss). the Meander indicator consists of three lines.I2*(1-H$1212/100). T h e M e a n d e r indicator works exactly the s a m e for all markets. you can type in the following formula in adjacent c o l u m n s to calculate the long entry level. reverse the calculations for the short side. typed into cell H1212.") C o l u m n K denotes the long exit level. column H the VS High level."") C o l u m n I denotes the long entry level a n d cell HI212 refers to the cell where you have typed in your percentage risk."<>-") C o l u m n L denotes the percentage result for each trade. such as Excel.(K2 . As you can see. column C denotes the High. percent winning trades. and column E the closing price. C o l u m n B denotes the opening price. =COUNTIF(L2:L1208."<>-")/L1210 =L1212*1350*250 T h e percentage risk that you are willing to take is."") C o l u m n J denotes the long risk level.) = IF(B2<F2.IF(AND(C2>H2. (Again. a m o n g which the upper (VS High) and lower ( V S L o w ) lines can be chartered one or two standard deviations away from the middle (VS M i d ) line. At the bottom of the spreadsheet type in the following formulas to calculate the total n u m b e r of trades.E2<H2). the long exit level.t y p e indicator looks like w h e n charted together with the latest price action for the S & P 500 futures contract.E2)).IF(D2<J2.B2.F2.1 shows w h a t this pivotp o i n t . =IF(I2<>"".) =COUNT1F(L2:L1208.I2)/I2."")).66 PART 2 System Concepts in a spreadsheet p r o g r a m .">0")/L1210 =SUMIF(L2:L1207. . This simple system puts on a long (short) trade immediately at the open if the opening price is below (above) the VS Low ( V S High) level. or at the VS Low n . and c o l u m n D the Low. and the result of the trade. = IF(I2<>"'*. Figure 5. in this case.IF(D2<F2.J2. =IF(I2<>"**. c o l u m n F denotes the VS L o w level. average percentage profit per trade and average dollar profit per trade.H2. (Reverse the calculations in the next four c o l u m n s for the short side.

1 The Meander indicator makes the most out of your data. .CHAPTER 5 Data Mining 67 FIGURE 5.

R e m e m b e r also that because we played it safe by always considering a trade to be a loser if it reached the stop loss level. with the very same settings. it increases the likelihood that we have run out of tall m e n . as suggested by the closing price. there is a high likelihood for the percentage profitable trades and the average profit to be even higher than shown. the M e a n d e r had over 5 0 % profitable trades on the long side. Now. the standard deviation was set to 1. the highest probability trades are those that enter immediately at the open. the average profit for the long side will be $ 2 1 2 . TABLE 5. w h i c h m e a n s that the system triggers a trade fairly frequently. the system generates fewer trades. let us take a look at the very s a m e system. analogous to the chance of meeting two tall m e n in a row. B e c a u s e this is an intraday system based on daily data. This is because if the open is below (above) the trigger level.79% 0. Long trades Number of trades Percent winners Average profit (%) Average profit ($) 522 52. That is.09% 287. 7 5 % . which increases the probability for a move in the opposite direction. The percentage risked at each trade w a s 0 .7 Basic results for Meander I trading system on the S&P 500.8. for a total average profit in today's m a r k e t value of $287. we m u s t play it safe and look for the losing trades first.16 Short trades Number of trades Percent winners Average profit (%) Average profit ($) 459 43. With the standard deviation set to 2. In this case. but this time traded on corn.68 PART 2 System Concepts (VS High) level if the market trades below (above) that level. and sometimes we do not k n o w what came first. if we continue on the tall m e n analogy. but with a higher probability for success. N o t too b a d for an intraday trading system on a one-contract basis.7 shows the results for this system broken down into long and short trades. the low (high) will be. the high or the low. or with a profit if it trades above (below) the VS H i g h ( V S L o w ) level but then moves b a c k down again. No m o n e y has been deducted for slippage and c o m m i s s i o n . In this version. there will be two low probability occurrences in a row. All trades are also closed out at the end of the day. the system continues to work fairly well if we only look at the percentage average profit and the n u m b e r of profitable trades. Table 5. tested on the S & P 500 futures contract.30% 0. Or rather. It exits with a loss if the market moves against the position with a certain percentage. for the period January 1995 to October 1999. although we very well might have reached the profit target first. too. As you can see in Table 5. A l s o .81 .03% 112. With an estimated $75 deducted for slippage and commission.

07% 6. BASIC EXIT TECHNIQUES T h e Black Jack Trading System derives its n a m e from the popular Las Vegas card g a m e of the s a m e n a m e .8 Basic results for Meander I trading system on com. but only to trade those markets in which the percentage-based moves also equal a high enough dollar profit. To accomplish this you must know .02 But when it comes to the dollar profits. in continuing our analogy to measuring men. T h e conclusion from this simple test is to test the same percentage-based system on several different markets to m a k e sure that it works equally well on all markets. To put it in trading terms: just because a system is equally good at capturing all 2% moves in corn as it is in S & P 500. if you were to sell pants to as m a n y of the m e n you have measured as possible. For a system to be profitable. it does not m e a n it will be tradable on all markets. the M e a n d e r does not even c o m e close. W h e t h e r a good working system will be profitable or not has nothing to do with the system. because not every system will produce big enough dollar profits to overc o m e the costs. But r e m e m b e r that this does not m e a n that M e a n d e r is a b a d system. this is a technicality decided by the exchange. when to ask to be hit with yet another card.75% 0. Black Jack is probably the only betting g a m e where you can m a k e sure that the long-term odds work in your favor and therefore allow you to beat the house in a consistent manner.CHAPTER 5 Data Mining 69 TABLE 5. but rather at what level the market currently is trading and the dollar value of that level. this doesn't m e a n that it will be equally as profitable in both markets. it is essential to know w h e n to hold. This is because there is a huge difference between a good working system and a profitable system. it needs to be traded in a market w h e r e the moves it is designed to catch also are worth taking. as previously noted.97% 0. which adds to the never-ending debate on whether or not to trade the s a m e system on several different markets. This is an interesting observation. W h e n you are playing Black Jack.07% 7. and w h e n to bet m o r e or less aggressively (if at all). choosing your sizes and markets would be beneficial in selling as many pairs as possible. Or. Because a system is working well on all markets. Long trades Number of trades Percent winners Average profit (%) Average profit ($) 439 48. w h i c h has nothing to do with whether the system is good at capturing the moves of the market. In the case of corn. the results look less promising.88 Short trades Number of trades Percent winners Average profit (%) Average profit ($) 476 44. And.

there are probably as m a n y ways to go about designing them as there are traders (and then some). especially if the dealer's card is high as well. N u m b e r three is the stop loss for those occurrences when the trade goes against you right off the bat. To continue to keep it as simple as possible. If you would like to look into m o r e sophisticated long-term techniques. which lets you exit the market with a limit order at the m a x i m u m open profit. You are trying to keep your average trade (hand) in positive territory with just enough money to make it worthwhile. and know exactly when to exit and why. Therefore. in the long run it will not matter what cards you are being dealt. this version of Black Jack was put together with a very simple top-and-bottom-picking entry technique paired with a short-term breakout system in case the retracement w a s n ' t deep e n o u g h to trigger an entry. exit technique n u m b e r one is the profit target. we will substitute these entries with random entries. and because of the reasoning above. we also need to understand what it is we are trying to achieve. In Part 3.70 PART 2 System Concepts the current composition of the deck and how to analyze your odds correctly as compared to the dealer's. it is not only essential that your entry rule allows you to take as many trades as possible. T h e basic rules are to bet m o r e aggressively. In the best of worlds. for a set of good short-term entry ideas. In the academic c o m m u n i t y and a m o n g fundamental analysts w h o pay h o m a g e to the efficient market hypothesis. so instead of delving any deeper into this matter. which c o m e s into play when the market is starting to move against you after you have accumulated an open profit. and consequently. where we focus m o r e explicitly on efficient exits and overall trade efficiency. you can do so under four different assumptions. all this is comparable to being able to trade profitably. If you can accomplish this. is the 200-days moving average. In the trading system Black Jack. but hold at a lower n u m b e r if there are m o r e high cards left in the deck. possibly in combination with any of exit techniques one to three. Finally. w h i c h also has withstood the test of time. W h e n it c o m e s to the entries and trade triggers. Probably the m o s t simple trend filter. If you know how to do this. the rest is just a matter of trading as frequently (playing as many hands) as possible in as short a time as possible." For decades. Picture your- . Exit technique n u m b e r two is the trailing stop. but even more important that your trades don't last too long. That said. how to go about doing it. Finally. this is also what we use for Black Jack. no matter what the entry rule looks like. this drift has been measured by technical analysts with the help of moving averages of various lengths. as long as you only take trades in the direction of the underlying trend. d o n ' t overlook the work of N e l s o n Freeburg. look into the work of Linda Bradford Raschke or T o m D e M a r k . Before we can put together the system. W h e n exiting a trade. you also can exit with a time-based stop. in recent years even these people have had to cave in and admit that there is something that they call "a r a n d o m walk with a drift. there is the trend filter. However. for the longest time there was no such thing as a trend. you will end up being a winner anyway.

By staying away from all of them. W h a t about the big winners? D o e s n ' t the old adage say that you should let your profits run? Yes. T h e problem is that in the time you spent processing this unit you could have processed three average units. N o w w h a t exactly does this m e a n ? It m e a n s that to c o m e up with a reliable system with a positive average trade and a standard deviation that is as low as p o s sible. it is of p a r a m o u n t importance that the system work. T h e really big winners are usually few and far between. without k n o w i n g it. you would like the assembly line to run as fast as possible. but in trading it also has its price. But wait a minute. while at the s a m e time allowing yourself to step in and out of other m a r k e t s as well. This o n e shows all the right features for being a real m o n e y maker. In fact. and it is probably good advice to listen to every now and then. these features grow even bigger. which in the end should prove itself m o r e profitable. the lower your payoff will be. T h e price is that you lose track of what is average. which in the end resulted in less overall profits. you are freeing up time and m o n e y that can be spent on other average trades—trades with which you are familiar and k n o w how to handle without panicking. equally well on several different markets over . all of a sudden its value-added features start to diminish. Without wasting any time. p r o m i s i n g to m a k e this unit at least three times as valuable as the average unit. A l o n g c o m e s an obviously defective unit that is bound to lower your results. You freeze in panic. but only after you have examined one unit can the next unit be placed on the line. the rest are just a b u n c h of look-alikes there to confuse you. and lowered the n u m b e r of units produced. you e x a m i n e it for a while longer than you did the previous one. and while y o u ' r e watching it. Translated into trading t e r m s . and not until its value has sunk b a c k to twice the value of the average unit do you take it off the line. but to be sure. so the longer the time spent examining a single unit. and with as m a n y profitable trades as possible. you toss it b e h i n d your back right away. increased the standard deviation of the o u t c o m e . for a trading system to be both as robust and as profitable as possible you should strive to m a k e each trade as similar to all other trades as possible. with this average unit you spent an average a m o u n t of time. even if it m e a n s that you have to cut your profits short every now and then. T h e next unit looks pretty average. freeing up the line for another unit to be placed upon it. Your pay (final estimated profit) for this j o b is dependent on how m a n y units (number of trades) you and your machine (system) can produce. But the third unit that c o m e s along is m u c h trickier. it does. By doing this. you are freeing both your m i n d and capital to milk that big w i n n e r several times over in a sequence of average trades. on average.CHAPTER 5 Data Mining 71 self as a quality controller standing at the end of an assembly line. but also on how little you can m a k e the look of all these units deviate from each other or from the average look (a low standard deviation). W h a t w a s it that h a p p e n e d here? By trying to increase your profits from an individual unit you put yourself in unfamiliar territory. you say. W h a t to do? Should you let it grow even bigger or should you free up the line for m o r e units? But then. w h i c h m a d e you panic. Obviously.

End. BSLevel = C. if the underlying assumption isn't market specific. corn. copper. Japanese yen. Hence. 6) and H < XAverage(H. TotProf(0). T h e s e are: coffee. 6) Then Begin Buy ("Buy Support") at C. always m a k e sure that your system runs at least marginally profitably.") at C. Therefore. if and w h e n this happens. ExpVar(O). TotTr(O). 6) Then Begin Sell ("Sell Resist. We can modify the code from Part 1 before we attach it to the b o t t o m of the system. If C < Trigger and MarketPosition = 0 and (H > XAverage(H. 6) or H Crosses below XAverage(H. D-mark. Eurodollar. This holds true even if you only are going to trade it on o n e market. natural gas. 6)) and C Crosses above XAverage(C. . such as how m a n y other people are playing. T-bond. C R B index. We know that we should look for as m a n y profitable trades as possible to generate an as high as possible average profitable trade in percentage terms. a n d the T-bill. If you wish. h o w the deck is shuffled. End. from 14 different markets. w h o the dealer is. crude oil. orange j u i c e . in percentage terms. this could be comparable to m a k i n g sure that your Black Jack card playing skills stay independent from any external factors. because you never k n o w w h e n the characteristics of this market will change to resemble m o r e closely those of another market. Vars: Trigger(O). PT(0). SL(0). BSLevel = C. but with am as low as possible standard deviation. so that we can export the right data into a text file for further analysis in Excel. MP(0). The TradeStation code for the Black Jack I trading system follows: Inputs: BarNo(O). you must have taken this into consideration beforehand by m a k i n g sure that the system (at the least) will not go berserk in this new market environment. or the n a m e of the hotel. on as m a n y markets as possible. 6) and L > XAverage(L. Prof(0). Filters *****} Trigger = Average(C. TradeStrl (""). 6) or L Crosses above XAverage(L. live cattle. cotton. 200). I ***** Retracement Entry *****} If C > Trigger and MarketPosition = 0 and (L < XAverage(L. BSLevel(O). For this system we use data covering the period January 1985 to October 1999. 6)) and C Crosses below XAverage(C.72 PART 2 System Concepts several different time periods. if ever so slightly.

6) Then Begin Buy ("Buy Break") at C. 6) and C Crosses above XAverage(C. Therefore. If C < Trigger and MarketPosition = 0 and L Crosses below XAverage(L. End. This is because we would like to activate t h e m one at a time. 6) Then Begin Sell ("Sell Break") at C.01) Stop. {***** Exit techniques *****} If BarsSinceEntry >= BarNo and BarNo <> 0 Then Begin ExitLong ("Long time") at C. ExitShort ("Short profit") tomorrow at BSLevel * (1 . IfTS < > OThen Begin If C > BSLevel * (1 + MP*0.01) Stop. End.MP*0. T h e first thing we must do is find the optimal length of our trades. we start by adding the following piece of code to the bottom of the system: TotTr = TotalTrades.PT*0. ExitShort ("Short loss") tomorrow at BSLevel * (1 . 6) and C Crosses below XAverage(C.MP*0. so far all input values are denoted by zeros. If C < BSLevel * (1 .01) Stop. If TotTr > TotTr[l] Then Begin Prof = PositionProfit(l) / (EntryPrice(l) * BigPointValue). End.01)Then ExitShort ("Short stop") tomorrow at BSLevel * (1 .01) Limit. If SL <> OThen Begin ExitLong ("Long loss") tomorrow at BSLevel * (1 + SL*0. . ExitShort ("Short time") at C.01)Then ExitLong ("Long stop") tomorrow at BSLevel * (1 + MP*0. As you can see from the code. End. End.01) Stop. BSLevel = C. IfPT < > OThen Begin ExitLong ("Long profit") tomorrow at BSLevel * (1 + PT*0.SL*0.CHAPTER 5 Data Mining 73 {***** Breakout Entry *****} If C > Trigger and MarketPosition = 0 and H Crosses above XAverage(H. End. BSLevel = C.01) Limit.

it also exports the data we need into a text file for further analysis in Excel. If LastBarOnChart Then Begin ExpVar = BarNo. FileAppend("D:\Temp\B J. 2) + ". then type in the following formulae at the b o t t o m of each group." + NumToStr(TotProf * 100. 2) + ".cvs".74 PART 2 System Concepts TotProf = TotProf + Prof. start by sorting and separating the data by n u m b e r s of days in trade. To calculate the percentage profitable trades for a specific trade length for all markets: =AVERAGE(D1:D14) w h e r e c o l u m n D denotes the percentage profitable trades for a specific market and trade length. In Excel. 2) + NewLine. To calculate the standard deviation of the percentage profit for a specific trade length for all markets: =STDEV(C1:C14) To calculate the ratio between the average profit and standard deviation of outcomes: =E14/E15 w h e r e cell El 4 denotes the average profit and cell El5 denotes the standard deviation. End." + NumToStr(Percentprofit. End. TradeStrl). N e v e r m i n d the optimization report in TradeStation. To calculate the average percentage profit for a specific trade length for all markets: =AVERAGE(C1:C14) w h e r e column C denotes the average profit for a specific market a n d trade length. we can r u n an optimization process in TradeStation." + NumToStr(ExpVar. TradeStrl = LeftStr(GetSymbolName. 2) + ". To calculate the standard deviation of the percentage profitable trades for a specific trade length for all markets: = STDEV(D1:D14) . For e a c h t i m e TradeStation runs a test. O n c e this is done. TotProf = TotProf / TotalTrades.

32 0.15 0. which is triggered at a certain n u m b e r of days.22 53.54 5.96 8.96 4.36 53.01 53.32 5.20 0.16 0.70 6 0.51 52.28 4 0. especially considering that the ratio between the percentage profitable trades and its standard deviation also is slightly higher for these trade lengths than for the 14day alternative.21 5.34 8 0.28 12.9. the m o r e robust the system.35 0.82 4.05 10.30 3.05 0.64 4.30 3.08 52.12 4.01 0. If we were to set a m a x i m u m trade length of either 10 or 11 days.24 0.03 0.00 0. the actual average trade length is slightly shorter w h e n the other stops are added.08 0. also s e e m s to have a decent risk/reward relationship.99 4.89 13.18 0.37 12.12 4. however. no matter what.9 Summary tradelength statistics for Black Jack I trading system.36 0.38 0.58 53.94 7 0.08 12.25 0. A trade length of ten to 11 days.39 12 0.30 0.70 14.10 0.21 52. Therefore.31 5 0.17 0.35 0.10 0.45 53. In this case. S u m m e d up in a table for easier interpretation.04 3.64 10 0.29 53.49 15 0. w h i c h shows the data for the time-based stop.26 0.15 10.21 54.50 54.78 11.91 3 0. the results resemble those in Table 5.9 we can see that the ratio between the average profit and its standard deviation is at its highest with a trade length of 14 days.01 10.91 4.14 0. we are looking for a high average profit and a high percentage of profitable trades.32 51. Days in trade 2 Average profit Standard deviation Ratio Percent profitable Standard deviation Ratio 0.21 -0.40 13 0.56 53.79 11.27 9 Average profit Standard deviation Ratio Percent profitable Standard deviation Ratio 0.61 5. T h e higher the ratio between the critical variable and its standard deviation.CHAPTER 5 Data Mining 75 To calculate the ratio between the percentage profitable trades and its standard deviation: =F147F15 w h e r e cell F14 denotes the percentage profitable trades a n d cell F15 denotes the standard deviation.13 52.13 0.04 0.05 0.04 0.11 0.38 11 0. however. TABLE 5.91 13. we will change the time-based stop slightly and turn it into a very tight trailing stop that will be triggered if the trade lasts for 11 days or m o r e .12 14 0.23 12.32 12.60 . F r o m Table 5.

30 54.48 0.69 10. we d o n ' t want our profits to run for too long either.71 11.41 3. likely to produce good results in the future.47 4.08 0. even though the optimal level might m o v e around slightly.10 0.13 0.55 11.36 53.13 0.06 13.15 0.45 0.54 9 0. End.75 7 0.43 0. TABLE 5.97 13. we aim for the 5 to 1 0 % region. it is time to set a profit target.59 9. 5 % .72 4.91 7.49 0.91 10.43 0.42 0.63 11.21 55.31 53.52 5. With a time-based trailing stop that kicks in after 11 days in place.55 4. we decide that we will add a profit target of 9 .45 0.44 6 0.41 0.43 0.64 4.5 0.76 PART 2 System Concepts If the trade runs on for too long.13 0.5 0. From Table 5.46 0.10 shows what our results look like.32 12.72 .90 10.5 0.47 0.41 4.99 10.27 54.92 10 0.5 0. Table 5.10 0.10 0. but because the entire idea behind Black Jack is to keep all trades as uniform as possible.13 0. our profit target eventually stops it out instead.32 53. Therefore. T h e TradeStation code for the time-based trailing stop is : If BarsSinceEntry > = BarNo Then Begin ExitLong ("Long time") tomorrow at C Stop. which indicates that this is a robust level.10 Summary profit target statistics for Black Jack I trading system Percentage profit target 5 Average profit Standard deviation Ratio Percent profitable Standard deviation Ratio 0.20 55.23 55.05 4.17 54.90 4.5 0. This value also seems to be surrounded by other fairly high values. 5 % to the system.10 we can see that the ratio b e t w e e n the average profit and its standard deviation is at its highest with a profit target of 9 .14 4. In the TradeStation code for the export function simply change the line "ExpVar = B a r N o " to "ExpVar = P T " .99 10. In this case.07 13.45 0. R e m e m b e r that the profit target is only there to catch the m o s t extreme trades.29 53.10 4.21 4.69 11.91 4.13 0.33 53.96 6.13 0.5 0.09 0.97 8.21 54. ExitShort ("Short time") tomorrow at C Stop.77 8 Average profit Standard deviation Ratio Percent profitable Standard deviation Ratio 0.

12 0.54 62.5 0.5 0.40 0. as m e a s u r e d by the ratio between the average profit a n d its standard deviation.35 5. From Table 5.5 percent region.02 12.28 56.30 55.17 0.93 2 0.69 4.1 1 Summary minimum profit level statistics for Black Jack I trading system. but also that the uncertainty. it s e e m s as if we should go with a fairly low accepted m i n i m u m profit. we decide to aim for a m i n i m u m accepted profit of 1%.17 4.08 .30 54. ) TABLE 5.11 shows the result for these levels. the h i g h e r t h e ratio b e t w e e n t h e critical variable a n d its s t a n d a r d deviation.23 0.66 2.83 1 0.58 5.10 3 Average profit Standard deviation Ratio Percent profitable Standard deviation Ratio 0. the lower the average trade.5 0.11 we can see that the higher the m i n i m u m accepted profit.25 0.12 0. T h e m i n i m u m profit stop we try to place somewhere between zero profits and halfway up to the profit target. is increasing.CHAPTER 5 Data Mining 77 With a time-based trailing stop and a profit target in place.73 12.11 0.40 0.21 3. but at the s a m e time m a k e sure that we give the trade s o m e initial slack.65 65. I n l o o k i n g a t Table 5.16 0.12 shows our findings.33 58.41 0. For this system. w e i m m e d i a t e l y can m a k e o n e v e r y i m p o r t a n t o b s e r v a t i o n .14 0. T h e stop loss.20 3.14 0.43 0.25 0.13 0.5 Average profit Standard deviation Ratio Percent profitable Standard deviation Ratio 0.39 5.26 0.52 3.33 57.86 14.81 10.5 0.98 5 0. T h e ratio b e t w e e n t h e a v e r a g e profit a n d its s t a n d a r d d e v i a t i o n n o w is a b o v e 1 in m o s t i n s t a n c e s . In doing so we also can expect a high percentage of profitable trades even though the insecurity about this n u m b e r decreases with a lower m i n i m u m accepted profit.5 to 2.18 13.11 0.73 4 0.21 13.84 11.43 3.04 13. we would like to keep as tight as possible.98 13. To be sure.86 4. the s a m e m a n e u vers are repeated for the m i n i m u m profit stop a n d the stop loss.83 14.61 60. Let us see what we can find in the 0. which m e a n s that we must look in the 0 to 5% region.33 0.47 5.33 55. Percentage minimum target 0.69 69. Table 5. t h e m o r e r o b u s t the s y s t e m .07 4.12. finally.16 0.72 1. Table 5.35 0. ( R e m e m b e r .

with the stop loss at 0 .00 0.14 1.75 Average profit Standard deviation Ratio Percent profitable Standard deviation Ratio 0.53 9.50 0.00 0.03 51. but that this c o m e s at the price of a deteriorating profit factor.02 1.25 6.35 5.79 5.16 1.40 11.18 43.99 1. As you can see. Table 5.50 7.95 55.75 0.32 2.16 0. the uncertainty (as measured by the standard deviation) about the o u t c o m e has decreased at the s a m e time as the average expected profit has increased somewhat for the tightest stops.15 1. 3 1 % .15 0.90 1.99 8.14%.17 0.12 Summary stop loss statistics for Black Jack I trading system.76 2. This m e a n s that with 6 8 % certainty.18 0.25 0.15 0. together with its current dollar value.16 1.16 0.73 7.50 Average profit Standard deviation Ratio Percent profitable Standard deviation Ratio 0. s e e m s to be to place the stop in the m i d d l e of the interval.57 7.12 1.14 0. it is n o w t i m e to look at what performance can be expected from each of the individual markets.94 56.18 0. the average percentage profitable trade is not big enough to generate a big enough dollar profit to be worthwhile trading. A good c o m p r o m i s e between these two p h e n o m e n a .93 9. That this is a compromise is emphasized by the fact that n o n e of the markets tested actually had the 1. we can say that the average profit will be positive. By the s a m e token.25 1.93 11.5% level. we can say that the true value for the average profit will lie somewhere in the interval 0. 5 % .5% level as its best alternative.17 0.92 5. A l t h o u g h Black Jack I w o r k s well on almost all markets.08 46. Table 5.16 1. we can see that the percentage of profitable trades s e e m s to increase somewhat the further away we place the stop. for most of our 15 markets. therefore. the average profit is 0 . at the 1. after adding the stops. however.17 0.16 2. Percentage stop loss 0.13 0.50 0. T h e price for this is a lower percentage of profitable trades.24 5.90 0.13 shows the average percentage profit for each market.75 0.37 6.12 also reveals that.93 6.14 1.00 49.77 With 6 8 % certainty. With all exit levels in place.97 4. because not every system produces big .78 PART 2 System Concepts TABLE 5. 1 7 % and the standard deviation 0.25 0.01 57.03 to 0 .35 4.04 3.88 6.84 53.54 10.50 2.13 0. this does not m e a n it will be tradable on all markets.24 38. For instance.15 0.27 33.

97 52. for some outof-sample testing. H e n c e .71 43.02 0.13 Market summary for Black Jack I trading system.01 0. perhaps the most popular of all markets.13 37. it would have come out to approximately $387.14 e n o u g h dollar profits to overcome the costs. . this is because all of these packages look at how many dollars you could have made. of all the markets tested. not how m a n y dollars you can expect to make.52 119.01 0.40 -88. Eurodollar.18%. live cattle.48 60. the historical value will almost always (providing a steady working model) be lower than the actual value for today's market.15 0.17 58.96 56. a good working m o d e l is not the s a m e as a profitable model.25 52. at this particular point in time.30 0. The bottom row in Table 5. but simply that these markets.73 29. but at the s a m e t i m e only feed your b r o k e r — a r e natural gas.13 0.48 243. This does not m e a n that Black Jack I is a b a d system.02 47.18 0. because the way TradeStation and several other testing packages calculate this number. This is very interesting information.13 0.64 48. while not ruining you. however.93 82.97 13. the only market that might be worthwhile trading is T-bonds and even this market is a border case.38 40. are not trading at the appropriate price levels and with a high enough point value to generate a profit after slippage and commission.37 -0. coffee. with an average percentage profit of 0. in today's market value.13 shows the initial findings for how our model would have fared on the S & P 500 futures contract.55 Percent profitable 46.50 48. Other m a r k e t s that. The S & P 500 was not a m o n g the markets chosen for the testing process.78 13.21 0.22 0. and crude oil.74 51. This does not mean. that the m o d e l is not working for these markets.89 -7. equal to an average dollar profit of $579. Market Orange Juice Copper Live cattle Japanese yen Corn T-bond Eurodollar Natural gas D-mark Coffee CRB-index Cotton T-bill Crude oil All markets S&P 500 Average profit 0.68 99.04 0.13 49. it would have done fairly well.24 80.11 0.18 Dollar value 28. because I wanted to save this. In fact.33 0.69 36. Again. A n d as you can see.86 — 579.70 76. in an up-trending market.20 44.28 -0. not counting slippage and commission.93 49.CHAPTER 5 Data Mining 79 TABLE 5.26 55.

"your largest drawdown is still to come.80 PART 2 System Concepts On a separate note.958 Drawdown . Black Jack can hardly be considered a tradable m o d e l — a t this point.521 180.662 Losers Lrg loser Avg loser 132 -2." Remember. however. No wonder they say.594 Winners Lrg winner Avg winner C u m profit 148 9. which is what you're assuming w h e n you're doing it the old-fashioned way. the market was trading at a m u c h lower level than today. you can take this sequence of trades. price for price. and transfer t h e m to today's market level for an estimate of what the dollar value would be today. that in TradeStation this drawdown figure would only have c o m e out to $43.25 581 6.52% 47.991 5. Total trades Profit factor Avg profit St Dev 0. However. If neither of these two assumptions is likely to happen again. 4 7 % .695 on a one-contract basis. measured in percentages.3 1 . you should not bother with slippage and c o m m i s s i o n until it is time to trade the system with real-time data and then decide if the expected profit is worth the expected cost.69% 55. you cannot use this value when comparing your system with other ready-made systems or the results from a CTA.275 -4. as presented by Table 5. you probably are thinking. Applying slippage and c o m m i s s i o n at this point will result in a suboptimal solution and m a k e the system m o r e unreliable when traded on realtime data.18% 2. it is once again worthwhile m e n t i o n i n g that the reason you don't want to adjust your results for slippage and c o m m i s s i o n while you're building the system is that you would like to find a solution that can trade as efficiently as possible (squeeze out the m o s t of the move) with as few constraints (rules) as possible. W h y is this? It's because when this drawdown happened. how likely is it that all trades would be repeated from t o d a y ? " Well.02% 280 1. but even so.38% 52. is it m u c h better than any value any technical analysis or system-trading package will give you. Let us use our detailed export function from Part 1 to take a closer look at h o w Black Jack would have fared trading the S & P 500. w h o m you might contemplate giving your m o n e y to instead. Tested on the S & P 500 with data from January 1985 to October 1999. Ideally. and consequently the dollar moves were smaller as well. move for move. Black Jack I produced 280 trades. with a drawdown of m o r e than $102. Therefore. " a m I TABLE 5. ask yourself. As such.23% -1.14. it's just as likely (or shall we say unlikely) that the market will repeat itself level for level. with an average profit of $581 in today's market value. however.86% 30. But because the percentage moves stay approximately constant.1 0 2 .50% 1. January 1985-October 1999.14% -7. Instead. "Yeah. it should only be used for your own research during the building process. Now.000. 6 6 2 . Note. that this value assumed that all trades would have been taken from today's market value and provided that all profits and losses could have been reinvested.14 Performance summary for Black Jack traded on the S&P 500.

As we will see in later chapters. . there are m u c h better ways of building a system like Black Jack than this example. to get a feel for h o w m u c h I might be m a k i n g t o d a y ? " Nonetheless. or am I m o r e interested in getting to know h o w I can m a k e my results as forwardlooking as possible.CHAPTER 5 Data Mining 81 m o r e interested in k n o w i n g what profits or losses I could have m a d e historically. in its current version Black Jack would be very difficult to trade. But do not despair.

the more volatile the market. indexes. the macroeconomic-oriented markets. which m e a n s that these markets seldom produce a half-weekly average true range of above 1%. and orange j u i c e . For coffee a $1. Let us take a closer look at some markets in this table. a 1% move is worth $ 3 5 9 . For instance. respectively.CHAPTER 6 Trade or Not to Trade If a specific system is not worth trading on a specific market. On the other hand. natural gas. such as corn. as c o m p a r e d to m a n y other markets. 4 5 3 .459 move equals a 4 .343 and $ 1 . on a percentage basis. for coffee half the average weekly true range for the last ten years and for the last year are worth $1. As it turns out. This is especially interesting if it is a short-term system. the higher the n u m b e r in the rightmost comparative volatility column. At the s a m e time. all have very low comparative volatility. such as the currencies and interest rates. 83 . have even higher volatility. and interest rates. most macro-markets are still viable trading alternatives. wheat and soybeans. then it b e c o m e s interesting to know h o w m u c h the market in which you are interested really is worth. Table 6. and if the risks are worth taking considering the volatility of the market and other viable trading alternatives. have fairly high comparative volatility. With true ranges this low it is doubtful that one can trade a market like this profitably with a short-term system if the current market level and price value aren't high enough. 0 5 % move. the weekly moves of most agricultural markets have too small dollar values. as c o m p a r e d to the other markets. Even m o r e weather-sensitive markets (and perhaps traditional gambling hot spots). many times below 1. such as currencies. It is interesting to see that basically all weather-sensitive markets (like the A m e r i c a n agricultural markets). Hence. Fortunately. although it works as well on that market as on any other markets. with trades lasting no longer than a week or two. such as coffee.1 shows how m u c h half the average percentage weekly true range is worth in today's market value in several different markets.

With a profit factor of 2.47 2.390 616 574 406 325 162 1. In the best of worlds. produce a profit of $3.59 0. on average.51 4.206 186 272 251 1.020 229 992 2. With 5 5 % profitable trades.65 2. and you are down to $90 per contract traded.617 478 391 756 884 1. you can expect the average winning trade to be half the average true range of that market.2 and an average winning trade of $476 (see Table 6. Deduct from that another $75 to cover slippage and commission.292 (11 X $476 . making only a little over a $100 a week simply is not worth sticking your neck out and running the risk of losing it all.96 2.037 680 818 280 — — But even many highly volatile markets can be difficult to trade if the dollar values of their moves are not high enough.063 859 825 139 526 393 941 1.628 1.07 0.36 2.34 3.767 614 444 299 213 192 1.83 0.51 148 409 982 4.087 322 874 55 134 248 3.75 0. which in turn breaks down to an average profit of $165 per trade ($3.2 that is expected to produce 5 5 % profitable trades. .453 492 218 570 1.115 1.05 2.311 6. you have only traded it on T-bonds.76 4.61 3.32 1.825 780 385 650 745 1.69 4. N o w you would like to trade this system on a m o r e volatile market.387 641 477 347 209 171 1. which is a very liquid and modestly volatile market for an average profit per trade of $294 (with $75 deducted for slippage and commission).29 1. Vol.55 5.542 1.374 1. the average losing trade is $216 ($476 / 2.2).84 PART 2 System Concepts TABLE 6.19 1.66 1.347 398 294 257 1.045 5.06 1.88 2.42 2. and their comparative volatility (AWTR/% move) Market Bean oil Cacao Canada dollar Coffee Copper Corn Cotton CRB index Crude oil Dollar index Eurodollar Feeder cattle Gold Heating oil Japanese yen Lean hogs Live cattle Lumber Municipal bonds Natural gas Nikkei Oats Orange juice Pork bellies S&P 500 Silver Soy beans Soy meal Sugar T-bills T-bonds T-notes Unleaded gas Wheat 10 years 203 231 345 1.046 676 836 316 1 year 272 278 447 1.22 0. 1% move.343 452 187 502 825 703 866 204 473 339 776 1.92 0. or less than one third of what you are making in the T-bonds.45 3. a sequence of 20 trades will. 2.292 / 20).18 2.645 135 476 1.9 x $216). So far. For instance.20 0.13 1.264 261 241 147 77 2.1 Dollar value of half average weekly true range (AWTRO.66 4. say that you have a short-term system with a profit factor as high as 2.38 2.04 3.030 659 979 321 1% 98 87 680 359 199 100 267 1.512 490 208 499 828 708 781 173 573 310 763 1.1). Considering that you only get one or two chances like this per week.78 3.425 413 345 714 840 5 years 206 200 369 1.084 268 128 Comp.723 149 428 1.79 1. like orange juice.

the longer the holding period. After an optimal n u m b e r of time units for the system in question. especially if the market you are trading simply does not p r o d u c e the moves you seek. the only way to trade these markets is to go with less frequent trading a n d longer trades that raise the value of the average trade to a level worth trading.1 shows the average profit per trade from a 20-day (four-week) breakout system. you should use weekly (or perhaps even monthly) data during the building and research process so that you can work with as few market time units (bars) as possible. but also very risky. Table 7. It might be a good idea to always try to be as short t e r m as you can w h e n it c o m e s to market time units (bars). W h e n y o u ' r e designing a long-term system.CHAPTER 7 Following the Trend From the previous chapters we have seen that short-term trading can be incredibly rewarding. the higher the average profit. measured in regular time units (days). As long as the ratio is increasing as well. In the previous chapters we discussed how to classify trades and what might constitute long t e r m and short term. This can be measured by the ratio between the average profit and the standard deviation. and with all trades lasting for a certain amount of days (weeks) no matter what. as can be seen from Table 7 . 1 . the further away from the signal you move. traded on the long side only on the S & P 500 stock index futures contract over the period June 1983 to August 1999. indicating that the uncertainty about the results also is increasing. F r o m the top half of this table you can see that. however. the less profitable and reliable your results will be. in this case. But with a higher average profit comes a higher standard deviation of the individual outcomes. the average profit is growing faster 85 . B e c a u s e not every market can produce a shortterm average profit big enough in dollar t e r m s .

316 0. than a slow moving average.022 50 1. weekly part of the table.026 60 1. the ratio seems to level off after about 50 days.164 0. To cure this.842 0. and vice versa.339 0. as measured by the average profit and the standard deviation.205 0. we can see that the weekly data produce both m o r e profitable and less risky results.231 0. Ratio Profit/week Profit/day 0.913 0.147 0. What is even more important is that by comparing the weekly data with the corresponding daily data. That said.293 4. or that of a faster moving average crossing a slower one.028 40 0.854 3.041 6 1.037 4 0.152 2.468 4. This is confirmed by the average profit m a d e per day. profit St.262 0. In this case.366 2. or a shorter lookback period. Dev.309 0.183 0. indicating a decreasing relative risk. Days in trade 10 Avg. Dev.1 Trade/time reliability comparison.148 0. perc. T h e crossover signal is produced either by the price crossing o n e or several averages. most of us probably have produced fantastic results on historical back-testing.462 2.086 3.780 4.86 PART 2 System Concepts TABLE 7.254 0.304 4.571 0.375 0.029 12 1.559 0. m o s t of t h e m focusing on the basic crossover signal.746 0. Ratio Profit/day 0.022 Weeks in trade 2 Avg. perc.433 0.171 0. m a n y try to filter out the actual buy/sell signals. Similar observations also can be made from the bottom.036 8 1.) W i t h a basic system like this. either by adding additional requirements like only taking the trade after a retracement and/or on the .023 30 0.075 0. profit St.151 0.136 0.030 than the standard deviation.263 0.030 10 1.886 3.982 0.651 0. MOVING AVERAGES M o r e than one book or article has been written on the topic of moving averages.033 0.181 0. (A fast moving average is an average that is calculated with fewer data points.321 0. indicating that there is little to be gained by staying any longer in the trade. which starts to decrease after 30 days.368 0.818 2.015 20 0.205 3. in the following long-term systems we work both with weekly and daily data. only to discover that reality looks less glamorous when the model is traded live or tested on previously unseen data.822 0.

the trend of the moving average will follow. A n o t h e r c o m m o n trading technique. using this method. Notice that for the Japanese yen. Therefore. and from 10 to 3 2 % with the 100-day moving average. As an example. From its inception in M a y 1986 to October 1999. but only to require the price to cross the fast average to exit. This could be a good idea for markets with a natural upward drift. With the directional slope method. W h e n the two contradict e a c h other. for a long-term system to be stable over time it m u s t be constructed in such a way that it leaves you with enough slack. is by hindsight. w h e n it c o m e s to moving averages. and consequently only use it when the market is trending. you stay out. but only 160 occurrences where the average has changed its direction from one day to another. there have been 184 crossovers and 127 directional changes from M a y 1972 to October 1999. with the 200-day moving average. like the Japanese yen. W h e n it c o m e s to m o v i n g averages. you also can build a system that uses less and m o r e up-to-date data. . the directional slope m e t h o d is a better choice than the basic crossover method. W h e n it c o m e s to the directional slope technique. we can apply a 100-day moving average to the C R B index. T h e problem is that the only way to classify the market as either trending or nontrending. W h e n you test such a combination. For instance. and from 31 to 4 0 % with the 100-day moving average. For the Japanese yen the s a m e n u m b e r s went from 17 to 3 6 % with the 200-day moving average. but the slope of a faster moving average for the exit (Exit M A ) . B u t that is not all. Yet another way of doing this is to use one moving average (or one set of averages) for the long side and another average (or set of averages) for the short side. with the 200-day m o v ing average the n u m b e r of crossovers is still larger than the n u m b e r of directional changes for the 100-day average. such as the stock market. the same effect can be achieved by using the slope of a slower moving average for the entry (Entry M A ) . which is a notoriously choppy market. A n o t h e r strategy that looks good on paper is to avoid the m o v i n g average model completely w h e n the market is in a trading range. instead of the actual crossover. so you are not stopped out time and time again during less advantageous times. the s a m e n u m b e r s c o m e out to 122 and 82 for the C R B index. or by adding additional m o v i n g averages or other indicators of various lengths.CHAPTER 7 Following the Trend 87 second crossover. and still achieve the same effect. is to u s e two moving averages and to require the fast moving average to cross over the slow one before entering a trade. and 170 and 92 for the Japanese yen. there have been 214 occurrences where the closing price has crossed over the average. On a trendier market. W h a t does this m e a n ? If it is reasonable to a s s u m e that in a prolonged trend. With a 200-days average. you should be able to reduce the n u m b e r of w r o n g signals considerably by using the slope of the average as a trigger for your trade. which is a luxury that few of us are granted in actual real-time trading. for the C R B index the percentage profitable trades went from 1 6 % for the crossover m e t h o d to 3 3 % for the directional slope.

consisting of two m o v i n g averages that will be the s a m e for both the long a n d the short side and that will work equally as well on as m a n y markets as possible. 2) + ". EntryVal = Average(Close. Conditionl = EntryVal > EntryVal[1]. percentage profitable trades. T h e m o d e l w a s tested on all moving average combinations on 16 different markets over the period from January 1979 to October 1994. we will put it together using weekly data substituting a 200day moving average with a 40-bar moving average." + NumToStr(WTrades. 2) + "." + NumToStr(TotBars. traded with two different systems. If Conditionl = True and Condition2 = True Then Buy at Close. FileAppend('"D:\Temp\MaDirect. ExitMA). If LastBarOnChart Then Begin PFactor = GrossProfit / — GrossLoss. T h e TradeStation code for this system follows: Input: EntryMA(lO). If Condition2 = False Then ExitLong at Close. TradeStrl). TotBars(O). End. If Conditionl = False and Condition2 = False Then Sell at Close. At the end of the code we attach our export function for the profit factor. ExitVal = Average(Close. 2) + ". w h e n the systems contradict each other. If Condition2 = True Then ExitShort at Close. 0) + NewLine. in accordance with our finding that we will be better off using as few bars as possible when examining the system. Condition2 = ExitVal > ExitVal[l]. TradeStrl = LeftStr(GetSymbolName. PFactor(O). EntryMA). 0) + ". WTrades(O). WTrades = NumWinTrades * 100 / TotalTrades.88 PART 2 System Concepts r e m e m b e r to treat each side as two different markets." + NumToStr(ExitMA. 0)+ ". TotBars = (TotalBarsLosTrades + TotalBarsWinTrades) * 100 / BarNumber. you can either choose to stay out or trade in smaller positions. A l s o . and time-in-market. T h e rest of the data . ExitMA(5). ExitVal(O). and tailor it so that it also exports the current moving average combination.'* + NumToStr(PFactor." + NumToStr(EntryMA. TradeStrl(""). Let's p u t together a m o d e l using the directional slope method.txt". Vars: EntryVal(O).

To calculate the ratio between the profit factor a n d its standard deviation: =IF(G2<>"".1 shows what this can look like. orange j u i c e . Starting at the top row in the next colu m n .2+((ROW()-2)*16)) where column B denotes the Entry M A . we build a matrix as a base for a surface chart. .2+((ROW()-2)* 16)) where column H denotes the ratio between the profit factor and its standard deviation. dollar index."") where column C denotes the Exit MA and c o l u m n D the profit factor for each moving average combination and market. corn. and coffee. S & P 500. In cell Q2 fill in the following formula and c a r r y it out all the way to cell AA2: =INDEX($K:$K. Eurodollar. copper.2+((ROW()-2)*16)). T h e data w a s put together using the ordinary point-based back-adjustment method. in an adjacent column. T h e value 16 stems from the fact that we are testing on 16 markets. live cattle. Figure 7. C R B index. fill in the values 0 to 11 in the following c o l u m n s . crude oil. gold.2+((ROW()-2)* 16)) =INDEX(H:H. cotton. C a n a d a dollar. =INDEX(C:C.2+(P1*1)) w h e r e column K denotes the Exit MA and cell PI denotes the value zero. = INDEX(G:G. lumber. No m o n e y w a s deducted for slippage and commission. W h e n this is done. To calculate the average profit factor for each moving average combination: =IF(C17<>C18. This lets us graphically examine our data and look for robust and stable moving average combinations. which are likely to continue to hold up in the future. Start by using the data-sort function to sort first by the Entry MA and then by the Exit M A .CHAPTER 7 Following the Trend 89 were saved for some out-of-sample testing. Japanese yen. T h e 16 markets tested were: T-bonds. and fill d o w n to the bottom of the spreadsheet.G2/STDEV(D2:D17).1 shows a sample matrix created in a spreadsheet p r o g r a m after we exported the necessary data from TradeStation and applied the suggested formulas. Then type in the following formulae in two adjacent c o l u m n s . Figure 7.AVERAGE(D2:D17)."") where column G denotes the previously calculated average profit factor. With all the data exported into Excel. leave the top cell blank and fill in the values 0 to 11 in the following rows. =INDEX(B:B.

However. T h e m o s t robust combinations seem to be with an 8-bar Exit M A . T h e most profitable Exit M A s seem to be the 8-bars and 12-bar Exit M A s . the better the system. we also can see that the ratio between the profit factor and its standard deviation seems to be the highest for the 8-bar Exit M A .2.2+(02*11)) w h e r e column J denotes the Entry MA and cell 02 denotes the value zero. From Figure 7. T h e higher the profit factor. T h e ratio between the profit factor and its standard deviation measures the robustness of the system. . we can see that an Entry MA in the 16. down to cell P I 3 : =INDEX(J:J.90 PART 2 System Concepts FIGURE 7. in cell Q3 fill in the following formula. or even of the standard deviation of any of the above.to 8-bar region. but also with an Exit MA in the 7. as yet another measure over the robustness of the system. the two charts in conjunction tell us to go with an 18-bar Entry MA and an 8-bar Exit M A . out and down. T h e higher the ratio. Finally. it is now easy to create a surface chart similar to that in Figure 7. together with a 12. Similar charts also can be m a d e for the n u m b e r of profitable trades or the time spent in the market.0) w h e r e column L denotes the average profit factor. either together with an Entry MA around 12 bars or with an Entry MA around 18 bars.or 18-bar Entry M A . 1 : =OFFSET($L$ 1 .to 20-bar region s e e m s to produce very high profit factors with almost any Exit MA from 10 bars and u p .2. which shows how the average profit factor for all markets varies with the length of the moving averages that m a k e up the system. all the way to cell A A 1 3 to complete the matrix in Figure 7 .3. the m o r e robust the system. In cell P3 fill in following formula.1 The average profit factor for different MA combinations.($02*$AA$ 1)+Q$ 1 . from Figure 7. Therefore. With the matrix ready.

Figure 7.4 we can see that all Entry M A s between 14 and 20 bars were a m o n g the best 200 combinations at least 15 times each.CHAPTER 7 Following the Trend 91 FIGURE 7. for instance. This strengthens our conclusion that an Entry MA of 18 bars will be a robust and stable choice that is likely to hold up in the future. the top 200 for all markets. the top 10 combinations for each market or a m o n g . you should go for an Exit MA in the 6. A n o t h e r m e t h o d is to calculate how often a specific moving average c o m b i nation happens to be a m o n g . T h e jewel of the crown is the 8-bar M A . As you can see. in Figure 7.2 The average profit factor for different MA combinations. for instance. if y o u ' r e looking to b o o s t the n u m b e r of profitable trades. with the 16-bar MA topping the crowd with a total of 27 occurrences.5 shows yet another e x a m p l e of this technique.to 10-bar region. with 27 occurrences. . but this time we have sorted on the percent profitable trades c o l u m n and for how m a n y times we could find any Exit M A s a m o n g the best 200 c o m b i n a t i o n s . Using our Entry MA as an e x a m p l e .

the average profit factor was 2.13. the profit factor w a s slightly higher for the out-of-sample data than for the in-sample data. 7 3 . Tables 7. D u r i n g the out-of-sample period.30 (1.09 / 0. which m e a n s that we can be 6 8 % sure that the true average profit factor will be no lower than 1.61 (1.73). . Hence.91 .09. the average profit factor w a s 1.91.3 show how the directional slope system. the standard deviation c o m e s out to 0.86 (2. With the ratio between the profit factor and its standard deviation equal to 3. with an 18-bar Entry MA and an 8-bar Exit M A . but the price for this came in the form of a smaller ratio between the profit factor and its standard deviation.0.61). the standard deviation c o m e s out to 2. w h i c h m e a n s that we no longer can be 6 8 % sure that the true average profit factor will be above 1. With the ratio between the profit factor and its standard deviation equal to 0 .3 The ratio between the average profit factor and the standard deviation shows the robustness of the system. For the in-sample period.92 PART 2 System Concepts Exit MA FIGURE 7.91 / 3.13). would have performed on the individual markets on both the in-sample data and the out-of-sample data. which m e a n s that the system has b e c o m e less robust.2 and 7.

.5 Number of top 200 percentage winners for different exit MAs. FIGURE 7.4 Number of top 200 profit factor occurrences for different entry MAs.CHAPTER 7 Following the Trend 93 FIGURE 7.

00 77.00 76. Data from October 1994 to October 1999.99 46.94 PART 2 System Concepts TABLE 7.83 42.23 40 52.79 0.73 % Winners 69.48 1.72 1.85 4.14 40.57 37.93 .38 45.86 48.92 66.67 6.93 28.13 % Winners 36.07 2.00 76.85 1.71 42.00 77.00 76. with 18/8 setting Market Corn Canada dollar Crude oil CRB index Cotton Dollar index Eurodollar Gold Copper Japanese yen Coffee Lumber Live cattle Orange juice S&P 500 T-bonds Average Ratio Entry MA 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 Exit MA 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 P factor 12.16 1.91 41.06 1.98 % In trade 75 76 80 71 80 73 74 80 85 76 83 76 74 83 82 78 77.00 81.2 Performance summary I for the directional slope system.5 0.73 1.72 0.62 26.73 1.63 1.00 78.00 77.71 44.3 Performance summary II for the directional slope system. Data from January 1979 to October 1994.86 35.16 33.00 76.67 25 52.29 54.00 77.88 18.68 TABLE 7.33 45.35 33.99 1.37 0.61 2.00 81.51 2.70 2.94 1.00 72.13 0.00 42.96 0.25 54.00 77.5 47.28 2.00 77.00 78.17 40 25.58 1.23 2.11 35.36 1.87 0.19 2.33 45.45 24.02 2.43 % In trade 77.67 41.09 0.00 79.59 1.03 42.84 41.29 2. with 18/8 setting Market Corn Canada dollar Crude oil CRB index Cotton Dollar index Eurodollar Gold Copper Japanese yen Coffee Lumber Live cattle Orange juice S&P 500 T-bonds Average Ratio Entry MA 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 Exit MA 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 P factor 2.28 2.38 0.91 3.19 36.92 1.

Therefore. which won't m a k e up for the slippage and commission on real trading. This information can be derived from the fact that the percentage of profitable trades only decreased from 4 3 % for the in-sample period to 4 2 % for the out-of-sample period. where corn and Japanese yen have extremely high profit factors—especially corn—while almost all of the others are losing money. which again confirms the fact that a few markets have m a d e up for all the others. but as it turns out the profit factor for the in-sample period w a s 1. which m e a n s that not only has the 18/8 combination b e c o m e less robust.73 for the out-of-sample period. so you just have to take my w o r d for it. most markets simply have not been in strong enough trends.3. and the 18/8 setting. T h u s . to a mere 1. and from 4 0 % to 3 9 % for the entire methodology (not shown). This is also confirmed by the differences between Figures 7. Even so.2 and 7. be seen in Table 7. This also can be seen from Figures 7. Another interpretation is that during the out-of-sample period. but so has the entire logic for this trend-following m e t h o d (and perhaps all trend-following methods). with a few markets m a k i n g bigger profits than previously and m a k i n g up for m a n y small losses and below previously expected average profits m a d e by other markets. the system still seems to be able to catch the right types of moves. Again. O n e other interesting observation that can be m a d e from comparing the insample charts and the out-of-sample charts is how well a 12-bar Exit MA holds up in both periods. This can.63 for the in-sample period.CHAPTER 7 Following the Trend 95 Just because the system has b e c o m e less robust.7. for instance. these moves simply have not b e e n strong enough.3. This is not shown. and thanks to a couple of good moves in a few markets. does not have to m e a n that it is no longer working as well. T h e main difference between Figure 7. the profit factor has b e c o m e m o r e heterogeneous. It could be that it is still just as good in catching the right types of moves. you would have m a d e a profit during the out-of-sample period as well. as indicated by a profit factor below or very close to 1. while others have been trending extremely well. however. but for m o s t markets.77. it can be interesting to look a little closer at the 18/12 .6 a n d 7. but what is interesting is that the ratio between the profit factor and the standard deviation has decreased from 2. which are derived the s a m e way as Figures 7. there is a higher standard deviation for the out-of-sample period. for both the in-sample period and the out-of-sample period. N o t m u c h of a difference. but there simply have been no such moves to catch. T h e ride would just have been a little choppier.7. O n e way to find out is to look at the overall average profit factor and its standard deviation ratio for all m a r k e t s and moving average combinations. but indeed also in a m u c h higher standard deviation.07 for the out-of-sample period.6 is that during the out-of-sample period. Notice also that Figure 7. except this time with data for the period October 1994 to October 1999.2 a n d 7. as c o m p a r e d to 1. as indicated by an average profit factor above 2.3 indicates that the trends have not been strong enough. This results in a higher profit factor. however.3 and 7. not that they have disappeared all together.

Not top values by any m e a n s . but good enough. with a ratio ranging from 2.2 and 7.0 to 3.7. T h e s e values are perhaps a little lower than for the 18/8 and 18/12 combinations.0.96 PART 2 System Concepts 5 6 7 8 9 10 11 12 13 14 15 Exit MA The average profit factor for different MA combinations for the out-of-sample period. But this is not the most interesting observation that can be m a d e when studying all four charts. especially considering that these values do not change that m u c h in Figures 7. That is. and from 1. resulting in a very low stability (high standard deviation). FIGURE 7. a few very profitable markets keep the overall average profit factor very high.6 and 7. these findings are the s a m e as those for the 18/8 combination. As it turns out. but with all the other markets showing very low or no profits at all. .6 combination as well.3 show that this entire area has a profit factor ranging from approximately 1.5 to 2. Figures 7. respectively.5. but the relative robustness over time m a k e s this an interesting area to consider. Yet another very interesting finding is the relative robustness surrounding the area of the 6-bar Exit M A .0 to 2. w h e r e the s a m e values range from approximately 1.9.6 to 1.0.

but simply that they haven't worked that well on that m a n y m a r k e t s recently. DYNAMIC BREAKOUT SYSTEMS Ever since the mid-eighties w h e n Richard Dennis and Bill Eckhardt m a d e a h u g e success with their turtle traders.7 The ratio between the average profit factor and the standard deviation for the out-of-sample period. R e m e m b e r also that the sole p u r p o s e of m a n y trend-following systems is to catch the few winners that will m a k e up for m a n y of the losers that occur w h e n trends aren't strong enough to generate decent profits.CHAPTER 7 Following the Trend 97 FIGURE 7. Dennis named the group "turtles" after having visited turtle farms in Singapore and having decided to grow traders as the farms grew turtles. This is exactly the situation we have h a d during our out-of-sample period. . As for the 18/8 and 18/12 combinations: Just because they have not worked as well nor been as stable on the out-of-sample data does not m e a n that they d o n ' t w o r k in the long run. breakout systems have been on everybody's 1 A group of aspiring young traders trained by Richard Dennis.

follows: Inputs: MaxLB(60). or if the m a r k e t takes out the lowest l o w for the last n days then sell. do the opposite. Vars: HistVol(O). MinLB). T h e concept is easy enough: never m i n d trying to pick tops and b o t t o m s — if the m a r k e t is going up. w h i c h then results in the system's also allowing for flat or neutral market situations in times of consolidation. EntryLB = MaxList(EntryLB. EntryLB(O). 30). the highest high for the last n days. T h e logic behind the D B S system. Sell Tomorrow at Lowest(Low. DeltaHistVol = (HistVol . DeltaHistVol(O). In fact. or a short position after a breakout on the short side. w h i c h can be." T h e TradeStation code for the D B S system with a variable lookback period of 20 to 60 days. EntryLB) Stop.5. EntryLB = YestEntryLB * (1 + DeltaHistVol). is that in times of relative high volatility the n u m b e r of "false" breakouts increases together with "false" trend reversals a n d deeper than expected corrections. if the market takes out the highest high for the last n days then buy. Buy Tomorrow at Highest(High. and if it is going down. HistVol = StdDev(C.98 PART 2 System Concepts m i n d . as it is called for short. you sell. ExitLB(O). Therefore. ExitLB = EntryLB * 0. EntryLB = MinList(EntryLB. YestEntryLB = EntryLB. you buy. instead of trying to sell the tops and buy the b o t t o m s . assigning t h e m different lookback periods. either with a long position anticipating an up-trend after a breakout on the long side. or to have a separate lookback period handling the exits. to m a k e it harder both to enter and exit a trade "too soon. With this basic system. times a volatility factor. YestEntryLB(O). MinLB(20). A m o n g these is to treat the long and the short side separately. EntryLB) Stop.YestHistVol) / HistVol. Yet another way is to keep the lookback period constant but add a volatility factor to the breakout level. y o u are in the market all the time. A n o t h e r refinement is to m a k e the lookback period d y n a m i c and sensitive to the historical volatility of the market. If CurrentBar = 1 Then EntryLB = 20. YestHistVol = HistVol. a volatility comp o n e n t increases the lookback period. To this basic system several improvements or refinements can be m a d e that could be used either separately or together with others. . for instance. YestHistVol(O). MaxLB). O n e such volatility-sensitive system is the D y n a m i c Breakout System that has been featured in several articles for Futures m a g a z i n e during 1996 and 1998. That is.

if you so wish). T h e g o o d n e w s is that. a n d continued higher over the next few days for a total move worth $14. ExitLB) Stop. the system gave a buy signal at 0. O n e way to change the stop-loss d i l e m m a is to apply the same logic to the exits as to the entries. For the C R B index the n u m b e r of trades decreased from 125 to 112 trades. the volatility of the market increased the lookback period to its m a x i m u m 60 days. with a stop loss placed at 0. you can see that the lookback period for the exits will always be half the lookback period for the entries at that particular point in time. F r o m the code (we will call this version l a ) . the TradeStation code for the D B S system can be rewritten as follows: . usually when the market takes off in a trend. is that the variable lookback period does not always work as intended.250. That is. w h i c h is a g o o d thing. if the m a r k e t takes off in either direction.764. and the Japanese yen over the period M a y 1972 to October 1999. Hence.822 (the solid u p p e r line). when traded on the C R B index over the period June 1986 to October 1999. W h e n the m a r k e t t o o k off on October 7. shouldn't the s a m e reasoning then m a k e it easier to exit the market (or m o r e difficult to stay in) as well? If you adhere to that theory. the volatility increases with the direction of the trend.781 (the dotted lower line). w h i c h in turn resulted in a lowering of the stop loss to 0. 1998.500. a n d for the Japanese yen the same n u m b e r s were 191 with the variable lookback period and 181 without it.8 shows h o w this affected the breakout and exit levels for the D B S system w h e n viewed with the ordinary point-based backadjusted contract. O n e example of this p h e n o m e n o n occurred October 7. if higher volatility—a sign of increasing insecurity and higher r i s k — m a k e s it m o r e difficult to enter the m a r k e t (or easier to stay out. for a total value at stake of $5. Also. T h e s a m e n u m b e r s for the C R B index were 26 and 2 8 % respectively. H e n c e . ExitLB) Stop. A n o t h e r less desirable feature with this system is that with an increasing lookback period y o u r u n the risk of having the stops move away from the market instead of with it. or $15. During the next couple of days. w h e n the Japanese yen took off with a m o v e w o r t h almost $10. Figure 7.000. the variable lookback period m a n a g e d to p r o d u c e a few m o r e profitable t r a d e s — 5 2 % as compared to 4 6 % with the 2 0 day stationary lookback period. the n u m b e r of trades actually decreased somewhat for b o t h markets. T h e problem with this version.875 from the close of 0. ExitShort Tomorrow at Highest(High.891 at October 13. the first part of this little study contradicts the logic behind the system. good volatility that works for your position and bad volatility that works against it. for a total value at stake from the entry point of $7.125 on a one-contract basis. if the variable lookback period was substituted for a stationary lookback period of 20 days.CHAPTER 7 Following the Trend 99 ExitLong Tomorrow at Lowest(Low. For instance. however. at least for the Japanese yen. while the results from the second part can only be considered inconclusive. there are two types of volatility.

8 The DBS system applied on the Japanese yen.100 PART 2 System Concepts FIGURE 7. .

MinExitLB). 30). MinEntryLB). YestExitLB = ExitLB. If CurrentBar = 1 Then EntryLB = 20. YestEntryLB = EntryLB. there also are situations w h e n the l o o k b a c k p e r i o d for the entry is shorter than for the exit. as well as plain exits into a neutral position. YestEntryLB(O). EntryLB(O). Sell Tomorrow at Lowest(Low. t h e q u e s t i o n still r e m a i n s : is t h e r e a way to u s e the historical volatility both to d e c r e a s e the n u m b e r of t r a d e s a n d increase the n u m b e r o f w i n n e r s ? By letting the volatility alter the lookback period. ExitLB) Stop. MinExitLB(lO). the n u m b e r of t r a d e s w h e n t r a d e d on the J a p a n e s e yen over the period M a y 1972 t o O c t o b e r 1999 w a s 2 0 1 . w h i c h m e a n s that there also will be several stop-and-reversal situations. Buy Tomorrow at Highest(High. In this version of the c o d e (version l b ) . ExitLB = YestExitLB * (1 . MaxExitLB(30). MinEntryLB(20). YestHistVol = HistVol. EntryLB = MinList(EntryLB. ExitShort Tomorrow at Highest(High. b e c a u s e these results are i n c o n c l u s i v e . ExitLong Tomorrow at Lowest(Low. DeltaHistVol(O). A m o r e direct way to do this is to let the volatility explicitly alter the level that triggers the trade. ExitLB = MaxList(ExitLB. An attempt at such a syst e m coded in TradeStation (version 2a) can look something like this: .CHAPTER 7 Following the Trend 101 Inputs: MaxEntryLB(60). EntryLB = MaxList(EntryLB. For t h e C R B index. the s a m e n u m b e r s over the period J u n e 1986 to O c t o b e r 1999 w e r e 3 1 % profitable t r a d e s out of a total of 114 t r a d e s . Vars: HistVol(O). we only implicitly alter the actual level at which a trade is triggered. MaxEntryLB).DeltaHistVol). HistVol = StdDev(C. MaxExitLB). a m o n g w h i c h 5 3 % w e r e profitable. as c o m p a r e d to w h a t would be d e s i r a b l e . EntryLB) Stop. DeltaHistVol = (HistVol . w h e r e t h e l o o k b a c k p e r i o d for the entry always is longer than for t h e exit. P e r h a p s this c o u l d b e c o n s i d e r e d slightly m o r e " d y n a m i c " than the original s y s t e m . EntryLB) Stop. YestHistVol(O). s o m e of these extra t r a d e s can be explained by the fact that a faster exit also allows the s y s t e m to enter the m a r k e t a s e c o n d t i m e in the s a m e direction. EntryLB = YestEntryLB * (1 + DeltaHistVol). however. YestExitLB(O). ExitLB(O). To be sure.YestHistVol) / HistVol. ExitLB) Stop. N o n e t h e l e s s . ExitLB = MinList(ExitLB. In this c a s e .

percent profitable trades. Eurodollar. In doing so. LongExit(O). Vars: HistVol(O). simply change the plus sign to a m i n u s sign when calculating the long exit. To m a k e it easier for the system to exit in times of high volatility. Japanese yen. If VolMethod > 0 Then HistVol = StdDev(C. orange j u i c e . and T-bonds. or flat. ShortExit(O). LongEntry = Highest(High. coffee. however. we use the ordinary point-based back-adjusted contract. .5). 20) . which is a s s u m e d to be a pivotal support (resistance) level. the system exits a trade before the market reaches down (up) to the lowest low (highest high) of the lookback period. Another way to achieve essentially the s a m e thing is to substitute all the historical volatility factors. With this system. gold. ShortExit = Highest(High.HistVol. LongEntry(O). 20) . we add half the 30-day historical volatility to the breakout level. 20) + HalfHistVol. m a k i n g it relatively harder to get in than to get out. Buy Tomorrow at LongEntry Stop. ExitShort Tomorrow at ShortExit Stop. crude oil. Because we are not interested in any percentage-based calculations or relationships. ShortEntry = Lowest(Low. and time in market.HalfHistVol. ShortEntry(O). CRB index. but also allow for neutral. LongExit = Lowest(Low. from Part 1. lumber. copper. as measured with the standard deviation. HalfHistVol(O). VolFactor(0. and the minus sign to a p l u s sign w h e n calculating the short exit (this will be version 2b). We do this by using our export function for the profit factor. You can do this in the code by changing the VolMethod input to zero. live cattle. we still leave the system with some slack so as not to be stopped out too soon. with the average true range for the period.5. Canada dollar. HalfHistVol = HistVol * 0. L e t us see h o w all these systems would have performed on a portfolio of 16 markets over the period from January 1980 to October 1999. Sell Tomorrow at ShortEntry Stop. corn. 30) * VolFactor Else HistVol = AvgTrueRange(30) * VolFactor.102 PART 2 System Concepts Inputs: VolMethod(l). positions in case the volatility gets too rough. cotton. but only a quarter to the exit level. 20) + HistVol. No m o n e y has been deducted for slippage and c o m m i s s i o n to avoid c o m i n g up with any suboptimal solutions. In this way. ExitLong Tomorrow at LongExit Stop. T h e 16 m a r k e t s we will use are S & P 500. dollar index.

58 .36 Ratio % 5. they also s e e m to be m o r e robust when c o m p a r e d b e t w e e n different markets.16 5.46 / 2.39 In market 79 80 88 76 90 81 Ratio 28.67 15. T h a t is.46 and a ratio of 2. not only do the version b systems show results that are m o r e profitable. TotTr(O). If CurrentBar = 1 Then Begin FileName = "DATempV + LeftStr(GetSymbolName.74 = 0. FileAppend(FileName. For instance." + "Profit" + NewLine.10 7. all version b systems have higher ratios than their version A counterparts.29 33.52 2. T h e code for this version of the export function follows: Vars: FileName(" ).65 2.78 7. TradeStr2). L e t us take a closer look at version lb in Tables 7.90 39.36 1. with 6 8 % certainty.71 2.35 1. we also m u s t switch over to the R A D contract. Version VeMa VeMb Ver 2a (HV) Ver 2b (HV) Ver 2a (ATR) Ver 2b (ATR) P Factor 1.02 39. the profit factor and the percentage profitable trades.36 1. In addition. End. w h i c h depicts the ratio between the profit factor a n d its standard deviation.93). to p r o d u c e a profit factor above 1 (1. 2) + ". TradeStr2("").36 Ratio 2.24 40. b u t in choosing b e t w e e n the two. For this we also use a variation of our trade-by-trade export function.36 / 3. w h e n it c o m e s to the two m o s t important readings. and see h o w it performed on all the individual markets.6.39 7. This also holds true for version lb (1.92 Trades 145 140 100 116 107 119 Ratio 7. version lb c o m e s out on top because of its higher stability. except for the t i m e spent in the market.65 27. with a profit factor of 1. all version 2b systems also show better results.64 2.4 Trade statistics summary for different DBS systems. as c o m p a r e d to version l b .73 .CHAPTER 7 Following the Trend 103 Table 7.txt".36 .72 6. 37.46 — 1.24 39.89 7. w h i c h we also developed in Part 1.46 1.5 and 7. FileDelete(FileName). To do so.54 5.74 3.52 = 0. T h e higher the ratio.1. Prof(0). D B S system version la cannot be expected.64 8. in m o s t instances.49 % Prof.95 42.97).38 1. Interestingly enough. M TABLE 7.21 40. TotTr = TotalTrades. TradeStr2 = "Position" + ".50 21.4 shows that the original system (version l a ) indeed h a d the highest profit factor.74. but that this also c o m e s at the expense of a higher standard deviation. the m o r e h o m o g e n o u s the behavior between the m a r k e t s . This is illustrated by the first ratio column.48 6.

72 1. as is the case with m a n y trend-following breakout-type systems. Copper." + NumToStr(BigPointValue.62 1.27 0.70 47. End. FileAppend(FileName. F r o m this table we can see that the average losses for most markets are quite substantial in percentage t e r m s .1) * 100.87 42. for instance. TABLE 7. this version of the D B S system works very well on a handful of markets.38 1.05 0. while others barely break even or lose money.03 1. Prof = 1 + PositionProfit(l) / (EntryPrice(l) * BigPointValue)." + NumToStr((Prof . 38. version lb. If LastBarOnChart Then Begin TradeStr2 = NumToStr(Close.14 41.01 44. TradeStr2). are these markets still worthy of being traded in a portfolio? Table 7.47 0.14 30.69 43.82 39. End.35 39. TradeStr2).83 1.53 2.26 50. TradeStr2 = NumToStr(MarketPosition(l).21 46.65 1. FileAppend(FileName.36 35. 4) + ".45 36. Table 7.33 Trades 152 155 158 144 148 114 117 95 127 141 144 157 147 152 139 146 % Prof. even if at a not-too-alarming rate. 2) + NewLine.104 PART 2 System Concepts If TotTr > TotTr[l] Then Begin. 2) + NewLine.84 % In market 78 79 80 81 84 78 82 83 83 80 78 79 78 79 78 85 .5 shows that.03 40.6 gives a m o r e detailed picture of the same p h e n o m e n o n . 0) + ". will on average allow for a 4 .00 1.78 1. T h e question is.00 35.5 Trade statistics summary for DBS system. 2 5 % move against the trade before it allows you to exit. Market Canada dollar Coffee Copper Corn Cotton CRB index Crude oil Dollar index Eurodollar Gold Japanese yen Live cattle Lumber Orange juice S&P 500 T-bonds P Factor 1.56 1.60 1.46 43.90 1.

76 -4.499 Let us also investigate how version lb would have performed over two different time periods and see if we can add to or confirm any of the conclusions we m a d e about our directional slope system in the previous chapter.793 2. but another indication that there are less profitable trades can be derived from the time in market column and the n u m b e r of trades column.057 2.89 9.01 0.44 -0.017 -1. T h e n u m b e r of profitable trades. from the average profit factor for the two periods.156 12. N o t e also that the profit factor differs somewhat from our original findings.53 4. we can say that this version of the D B S system has not performed as well recently as in the past. version lb.78 -2.04 1.7) and January 1990 to October 1999 (Table 7.060 -2. Here I used the ratio-adjusted contract and calculated the profit factor with the help of the average weighted values of the winners and losers.67 0.178 749 .378 1.59 5.093 255 (%) 1.37 1.64 -5. has decreased ever so slightly.09 0. for the last ten years.90 -2.25 -3. the results look even less promising.30 3. although this version of the D B S system trades profitably in m o s t markets. As shown in Table 7. This can be seen. Table 7. but also has a higher standard deviation than in the earlier period. In this case. Avg.13 0.78 -5.38 0.814 -1.51 -0. Loss (%) -0.6 Trade statistics summary for DBS system.48 -3.43 -2.848 4.40 -1.46 5.8).74 10.52 2. Profit Avg. As for all other numbers. for example. the profit factor is somewhat on the low side.065 -848 -349 -1.092 4. due to the different contracts used.427 -677 -876 -2.64 3.239 -1.98 -1.250 ($) -545 -2. As w a s the case with the directional slope system.225 3 93 162 -1. the .33 0.CHAPTER 7 Following the Trend 105 TABLE 7.188 1.340 2.91 7.02 0.29 -2.61 13.81 ($) 1.80 -5.72 -0. we will look at the periods January 1980 to October 1989 (Table 7. Trade Market Canada dollar Coffee Copper Corn Cotton CRB index Crude oil Dollar index Eurodollar Gold Japanese yen Live cattle Lumber Orange juice S&P 500 T-bonds (%) 0. which not only is slightly lower for the later period.8.21 -0.34 1.294 -716 -1.76 2.59 1.00 0.09 -4. they too are not as good for the later period.277 -592 -9.74 3.7 shows that.830 1.94 3.368 1.628 5.24 ($) 26 541 -67 100 194 -532 545 663 208 175 1.81 -1. for instance.463 3.100 1.23 Avg.47 8. Usually.43 8.

84 .89 45.66 0.11 1.65 1.38 2.97 1.22 42.76 28 55 35.48 2.11 Trades 72 76 74 68 75 83 74 58 38 70 69 81 73 80 79 66 71.62 % In trade 79 79 82 84 86 74 84 82 82 77 78 77 81 81 75 85 80.71 1.35 1.59 36 33.34 40 40 49.47 37.07 5.5 42.85 0.01 1.32 0.48 32.41 1.31 1.89 39.75 34.60 5.48 2.91 1.16 0.25 1.83 0.25 36.65 56.72 60.44 1.29 42.73 43.88 25.92 1.33 1.8 Performance summary for version lb.73 33. January 1990-October 1999.00 6.12 1.57 1.53 1.84 2.9 1. Market Canada dollar Coffee Copper Corn Cotton CRB index Crude oil Dollar index Eurodollar Gold Japanese yen Live cattle Lumber Orange juice S&P 500 T-bonds Average Ratio P factor 1.106 PART 2 System Concepts Performance summary for version lb.18 39.37 3.24 51.39 0.62 1.38 22.29 43. January 1980-October 1989.71 51.71 Trades 78 76 80 71 68 25 40 28 41 68 71 70 70 69 56 73 61.1 39.60 TABLE 7.04 1.28 39.57 % Winners 38. Market Canada dollar Coffee Copper Corn Cotton CRB index Crude oil Dollar index Eurodollar Gold Japanese yen Live cattle Lumber Orange juice S&P 500 T-bonds Average Ratio P factor 0.44 0.47 42.18 46.95 1.40 2.44 % Winners 37.47 41.50 3.53 44.58 % In trade 75 78 76 76 83 78 80 80 85 81 77 80 75 76 79 83 78.97 41.24 2.

Usually the Bollinger bands indicator is charted together with the price. Even in version l b . T h e TradeStation code for this is as follows: Inputs: BandLen(60). the m o r e time spent in the market. LoBand(O). experience has shown that you should look for a lookback period for the moving average and the standard deviation calculations somewhere between 50 a n d 100 days. For a turtle it is vital to cut b a c k in size during times of high volatility and drawdown to preserve as m u c h capital as possible in expectation of the next major move. Vars: BandDevi(O).9 it is important to r e m e m b e r that j u s t because the price penetrates the b a n d s . which m a k e s it difficult to trade this system with a fixed fractional m o n e y m a n a g e m e n t strategy. Equally as often it will be the b a n d that catches up with the price. simultaneously. NoStDev(2). For a shorter-term system. It can be used both in short-term oscillator-type systems. which can c o m e in any market. however. MidBand(O).CHAPTER 7 Following the Trend 107 higher the profit factor. BandDevi = StdDev(Close. C o m p a r e d to the directional slope system. while the percentage of profitable trades is approximately the same. looking for overbought/oversold situations or in a long-term breakout-type system. the profit factor also is considerably lower. STANDARD DEVIATION BREAKOUT T h e Bollinger bands indicator is one of the m o s t versatile indicators there is. . R o t h the D B S system and the directional slope system have in c o m m o n the fact that they do not have any fixed stop levels. For a long-term breakout system. and you have to c o m p r o m i s e on one performance measure in favor of another. you should aim for about 20 to 40 days. MidBand = Average(Close. It will be interesting to see if we can improve on these n u m bers in the u p c o m i n g sections. UpBand(O). BandLen) * NoStDev. there always is a price to pay. This is an interesting observation that suggests that whatever system you chose. Although the directional slope system lacks a natural stop level completely. because what is even m o r e important for a turtle than w h e n or what to trade. In fact. albeit with a lower standard deviation. first calculate an appropriate-length moving average. is h o w m u c h to trade. This takes us back to the beginning of this section where we talked about the turtles-. the D B S system at least has a d y n a m i c level.9. W h e n you look at Figure 7. it d o e s n ' t have to c o m e back from there. the fewer the trades and. BandLen). and then add the n u m b e r of standard deviations for the s a m e time period to that average for the upper band or deduct the n u m b e r of standard deviations for the lower band. as shown in Figure 7. the Bollinger b a n d s indicator often m a k e s up the foundation for some excellent breakout systems. this level is running the risk of moving away from the market. To calculate the indicator.

108 PART 2 System Concepts FIGURE 7.9 The Bollinger bands indicator applied on the Deutsche Mark. .

b u t this t i m e with the Bollinger bands oscillator added at the bottom. with an ordinary R S I or stochastic indicator. UpBand(O). O n e of t h e m we stressed w h e n w e discussed the M e a n d e r indicator: you can m a k e n o comparative assumptions b e t w e e n two different readings. for instance.LoBand) * 100/ (UpBand .BandDevi.BandDevi. "Position").9. Vars: BandDevi(O). "UpBand"). respectively. LoBand = MidBand . it also shares several of their disadvantages. Plotl (BandPos. alternatively. T h i s is especially useful if y o u are primarily interested in catching the short-term m o v e against the underlying trend. This m e a n s that you will be able to take m u c h m o r e data into consideration but still achieve the s a m e effect as.LoBand). NoStDev(2). MidBand(O). w h e r e you keep the upper a n d lower bands stationary at. "LoBand"). As you can see. Plot3(LoBand. although the Bollinger bands indicator takes twice as m u c h data into consideration. . UpBand = MidBand + BandDevi. Figure 7. BandLen) * NoStDev. 100 a n d 0. BandDevi = StdDev(Close. BandPos(O). "UpBand"). A l t h o u g h the Bollinger bands indicator has several advantages c o m p a r e d to m o s t other indicators. for instance. BandPos = (AvgPrice . Plotl (MidBand. In this case it might be better to chart the indicator as a n o r m a l i z e d oscillator instead. M a n y t i m e s it can be difficult to see what is actually h a p p e n i n g w h e n the price reaches o n e of the b a n d s a n d then starts to h u g the b a n d in a trending motion. use the s a m e amount of data for a faster reacting indicator. BandLen). charted on top of each other. "LoBand"). Plot2(100. This is shown in Figure 7. whether you look at the s a m e market at two different points in time or at two different markets at the s a m e point in time. respectively. LoBand(O). O n e interesting p h e n o m e n o n of the Bollinger bands indicator w h e n used as an oscillator is h o w fast it catches the m o s t recent moves despite a relatively long lookback period. depicting the same price activity on the s a m e market. This time the upper and lower boundaries for the Bollinger bands oscillator are set to 70 a n d 30. "MidBand"). w h i c h u s e s the s a m e chart as in Figure 7. Plot3(0. Plot2(UpBand.CHAPTER 7 Following the Trend 109 UpBand = MidBand + BandDevi. Or. it still moves m u c h faster between the overbought and oversold regions than the R S I indicator.10. MidBand = Average(Close.11 shows a 20-day Bollinger oscillator together with a 10day RSI. T h e code for the oscillator follows: Inputs: BandLen(60). LoBand = MidBand .

.10 The Bollinger bands indicator also can be charted as an oscillator.110 PART 2 System Concepts FIGURE 7.

11 The Bollinger bands indicator reacts much faster than RSI.CHAPTER 7 Following the Trend 111 FIGURE 7. .

robust setting you could use the same technique as we did for the directional slope system. for the short exit. T h e TradeStation code follows: Inputs: BandLen(60). T h e m o d e l w a s tested on data from January 1980 to D e c e m b e r 1992. respectively. as is the case with m a n y breakout-type systems. it already is relatively difficult for an ordinary highest high/lowest low breakout system to enter the market. .112 PART 2 System Concepts W h e n it c o m e s to using the Bollinger bands in a trading system. we also substituted the standard moving average for an exponential one. BandLen) . most Bollinger bands systems only seem to take the volatility into consideration when entering the market but not w h e n exiting. we have substituted the closing price with the low. because of the large distance between these entry levels. ExitLong tomorrow at XAverage(Low. and the high. BandLen). Instead the exit usually is governed only by a crossover of the mid-band. Buy tomorrow at UpBand Stop. we hope that these slight changes to the original setup will w e e d out a few trades that were b o u n d to go nowhere and only catch those signals that have a high enough m o m e n t u m to follow through and develop into a profitable trade. w h e r e we m a k e it even harder than usual to enter. UpBand = XAverage(High. Vars: UpBand(O). and for the lower band. here is an alternative long-term system. We also have m a d e it a little m o r e difficult to exit by substituting the closing price with the low. with relatively high historical volatility readings. BandLen). As you can see from the code. which most of the time is a standard moving average. BandLen) Stop. copper. lumber. ExitShort tomorrow at XAverage(High. if you currently do not have a position on and the market is in a relatively w i d e trading range. To find an optimal. w h e r e we exported the different moving average combinations and then put together a couple of so-called surface charts. after we added a few additional stops and s o m e sort of filter. BandLen) + 2* StdDev(High. Also. All in all. as compared to what happens in a basic breakout system. LoBand(O). To c o m p e n s a t e s o m e w h a t for both the slower/delayed entries and exits. for the upper band.2 * StdDev(Low. The period January 1993 to October 1999 w a s saved for some out-of-sample testing. Nonetheless. LoBand = XAverage(Low. But to a Bollinger band-type system you also have to add the distance to the upper a n d lower bands which. could be placed at a considerable distance away from the highest high and lowest low. for the long exit. another disadvantage is that you will take the volatility into consideration twice. in section three and four. we have substituted the closing price with the high. crude oil. Sell tomorrow at LoBand Stop. T h e 16 markets tested were D-mark. because of the volatility. For instance. T h e lookback period is set to 60 days and has not been optimized in any way. BandLen) Stop.

as Table 7. cotton. Table 7. dollar index. 7 4 % profitable trades per market. wheat.28 2. and r o u g h rice. we added the export function for the profit factor.76 73.41 3.9.71 38.9 Long-term Bollinger bands system. with an average of 5 1 .3 2. for further analysis in a spreadsheet program.94 2. with profit factors well above 1 in m a n y cases a n d with the average profit factor as high as 3.61 40. and percentage profitable trades to the code above. T-bonds. this version of the Bollinger bands system s e e m s to have several desirable features.54 41. In fact.61 47. natural gas. Because we used the ordinary point-based back-adjusted contract. TABLE 7.32 1.41 1.86 64.59 60 66.91 32. such as a high profit factor in m o s t markets. Japanese yen. January 1980-December 1992.45 43. days in trade.29 1.92 1.01 10.21 66. T-bills.62 5. No m o n e y has been deducted for slippage a n d commission. For a long-term system.42 Trades 34 23 46 37 43 17 34 33 39 40 6 38 15 33 34 6 % Winners 61. m o s t markets also h a d a very high percentage of profitable trades. a high percentage of profitable trades.9 shows the results. coffee.82.CHAPTER 7 Following the Trend 113 gold.67 60. and a fairly low percentage of the time spent in the market.06 50 % In trade 59 61 53 52 55 54 44 52 59 62 82 53 59 62 50 63 .21 2. live cattle.9 shows. As y o u can see from Table 7. Market Deutsche mark Crude oil Lumber Copper Gold Dollar index Live cattle T-bonds Cotton Japanese yen Natural gas Wheat Rough rice Timber Coffee Nikkei index P factor 3. attributable to the slight changes we m a d e to the original system.67 3. all markets traded profitably.65 3.32 3.24 45.37 12. Nikkei.67 34.

So is this second system m o r e or less curve fitted that the first one? 115 . w h e r e we will continue to develop and e x a m i n e t h e m further." For a well-working system to be profitable it needs to be applied to a market that can exhibit a high enough dollar value in relation to the moves the system is designed to catch.PART TWO A Few Final Thoughts About Part 2 In this part. but this does not m e a n that the system is a b a d system. but w h a t if you reverse all the rules? In other words. Is there a way of measuring curve fitting? Is it as easy as saying that the better the result. the m o r e curve-fitted the system. we created a few basic trading systems to take with us to the u p c o m ing sections. for every rule you had. you will add o n e rule. This does not always have to be the case. In fact." not " b e profitable. we have shown that it is possible to exploit different types of market behavior and create a trading system that is likely to continue to work well in the future and on several different m a r k e t s — s o m e t i m e s even on markets the system has not been tested on previously. say that you just have finished building a system with thousands of rules. W h a t will the historical results from this system be? T h e y will be disastrous. 1 would like to leave you with a few questions and a little food for thought. the m o r e curve-fitted the system? For instance. the whole key to m a k i n g a system robust is to keep it as simple as p o s sible. which will capture every historical daily m o v e in the S & P 500 futures contract perfectly. Before we go on to Part 3 . with as few rules as possible. or the m o r e rules. Notice once again that the key words here are "work well. while at the s a m e time m a k i n g sure it works well on several different markets. ending up with twice as many. A l t h o u g h all the systems have been very basic. No doubt this system is curve fitted.

For every change you m a k e . you lose money. Have I d o n e this myself? N o p e . and losing m o n e y is not a bright thing to do. in fact.116 PART 2 System Concepts H e r e is a suggestion on h o w to measure the degree of curve fittedness: take the market or m a r k e t s on which you would like to trade. As you will discover. manually pick out all those market segments you would like for your system to catch. Look at it this way: if you are long in the market and it goes against you. it is very difficult. or less than ( < ) signs I a m . that is a bad thing. a n d string them together in a fantasy equity curve or several curves. adding and deducting rules in an effort to m i m i c the fantasy results. here is what I consider to be a rule: every t i m e I n e e d to use any equal to ( = ) . At least not in a systematic fashion. adding stops and exits is a necessary evil and should be looked u p o n m o r e as a c o m m o n s e n s e proceeding than actual curve fitting. For m e . It just c a m e to m e . If you have to add a rule to increase the correlation. and that is w h a t this is all about. m o r e than ( > ) . Although it is true that every rule you add to the system m e a n s that you probably are curve fitting it more. but nonetheless. Then start building your system. b u t if you m a n a g e to increase the correlation by deducting a rule. formulating a rule. In the next part. that is a good thing. preferably below five for the entire strategy. we will see if we can improve each system's performance further by adding a few stops to protect us from too severe losses and lock in profits. if you are being honest about what constitutes a sign. . the trick then is to keep all these signs down to a m i n i m u m . measure the difference between the fantasy curves and those on which you are working by looking at the correlation between t h e m .

you can look at any basic system. or neutral position. But in essence. starting out with the entry is not always the best way to go.PART THREE Getting Out O n c e you have a basic idea about what you would like to achieve with your system and which type of moves you would like it to catch. A filter efficiently keeps you away from the markets as m u c h as possible. which we hope signals a high-probability trading opportunity. Notice. short. like those in the previous section. the trick is to consistently exit with profits that are bigger than the losses. Therefore. and work b a c k w a r d to c o m e up with a simple entry technique that matches your exit criteria a n d the underlying assumptions for the trade. that this does not by any m e a n s m e a n that the work you do looking into different entries is a waste of time. the basic system tells you w h e n it is time to enter into a long." F r o m this it follows that m o s t of the time the exits are m u c h m o r e important than the entries. But as so m a n y successful traders have said. a basic system is a set of entry rules. you might already have c o m e up with a set of excellent entry techniques without giving the exits any second thoughts. it is time to start experimenting with different types of stops a n d exits. however. "anyone can enter the market. S o m e t i m e s . B u t to k n o w about it you n e e d to have e x a m i n e d and tested it on a previous occasion. it should "only" be a question of examining or adding a set of exits to improve performance even further. In Part 4. because m a n y times it will turn out that there will be a better entry technique than the one with w h i c h you started. no m o r e no less. however. Of course. as a set of entry rules only. but to be successful in the long run. If you wish. Of course. 117 . In that case. we continue the development process by adding some sort of filter technique. That is. entering into a short position usually also m e a n s exiting a long position. it is an even better idea to start out w i t h the stops and exits.

but only tell you whether this position. A d d a set of appropriate. is as follows: • Start out with a basic. keeps up with what could be expected and that it doesn't run a m o k with your account. or neutral) also signal the exit for any other type of position. Exits. • • . on the other h a n d never try to forecast the market. If necessary. standalone exits that form a good c o m p r o m i s e between risk and reward. a n d let the entry for one position (long. short. and robust set of entries. A good exit should also let you use your entry and filter techniques as often as possible without letting such use affect the bottom line. change the entry technique and/or a d d a filter to sift out lowprobability trades. as it looks so far. By necessity (and whether you like it or not). stable.118 PART 3 Getting Out only letting you place a trade according to your original or modified triggers after you have examined and constructed the exit techniques. T h e normal order of w o r k then. filters and entries always to s o m e extent try to forecast the market.

the TradeStation performance s u m m a r y will no longer be enough. TradeStation again m a k e s the mistake of calculating all moves in dollar terms. as well as the overall efficiency of your trades as R I N A syst e m s a n d O m e g a Research define t h e m . In the case of efficiency analysis.Lowest price) Entry efficiency = (Highest price .Lowest price) For Short Trades Total efficiency = (Entry price — Exit price) / (Highest price . TradeStation's efficiency calculations work very well. but if you would like to expand your research and try to optimize the possibilities these techniques offer. For Long Trades Total efficiency = (Exit price — Entry price) / (Highest price . rather than in percentages. Let us take a look at how TradeStation does it.Lowest price) 119 . With the highly improved systems performance s u m m a r y in TradeStation 2000i.CHAPTER 8 Efficient Trades The idea is to try to m a x i m i z e the efficiency of each trade as m u c h as possible.Entry price) / (Highest price . As an initial m e a s u r e a n d starting point.Lowest price) / (Highest price .Lowest price) Exit efficiency = (Exit price .Lowest price) Entry efficiency = (Entry price . this is OK w h e n looking at the numbers. but not if you would like to do something about them. you can now get a feel for both the entry and exit efficiencies.Lowest price) / (Highest price . This is a concept popularized by R I N A systems.

1 shows an individual long trade. before we recaptured s o m e of that o p e n position loss. before we exit at 1. What really happened then. It was the m i d d l e part of the trade that failed. reach a low at 1. This. and a total efficiency of 5 0 % . the exit efficiency 17%. Inserting these n u m b e r s in the above formula.390. we would have started out with a profit. Had this instead been a short trade. that in the case of a short trade. as FIGURE 8. and the total efficiency —50%.330. and then a high at 1. therefore illustrates the importance of dividing the trade and especially the d r a w d o w n into several subcategories. however. was that both the entry and the exit were quite good. we get an entry efficiency of 6 7 % . .120 PART 3 Getting Out Exit efficiency = (Highest price — Exit price) / (Highest price — Lowest price) For All Trades Average total efficiency = Sum of total efficiencies / Number of trades Average entry efficiency = Sum of entry efficiencies / Number of trades Average exit efficiency = Sum of exit efficiencies / Number of trades Figure 8.1 An Individual long trade. does not show up in the performance s u m m a r y and. N o t e . then moved into a substantial loss. however. an exit efficiency of 8 3 % . In this case we enter the market at 1.380. the entry efficiency would have been 3 3 % .350.

leading down to the low. the end trade drawdown ( E T D ) . With this also c o m e s the necessity of using the R A D contract that is p u t together using percentage differences rather than point differences. there is the E T D . But what m a n y traders do not know. For instance. As we discussed. say that you. Unfortunately. they work well as comparison measures. leading up to the high. but you m u s t know w h a t you are doing and w h a t you are investigating. on the day for the roll from one contract to the next. you cannot calculate that stop in dollar t e r m s . As you can see. like adding different exit techniques or add-on entries. A l m o s t every trade can be divided into three distinct parts. all values are expressed as percentages or ratios of the distance between the low and high values for the trade. B u t if you would like to do something about them.m a k i n g phase. you m u s t look at all the price changes and relationships in percentages as well. A l t h o u g h this last conclusion is very important. d o n ' t have any positions and that your last trade was a $3. First. it is not the m a i n reason why these efficiency calculations are good to look at but not to tinker with. another major error m o s t system designers m a k e when they are building a n d evaluating systems is only to look at the overall total equity drawdown ( T E D ) . b e t w e e n both different markets and different t i m e frames. DRAWDOWNS A h . T r u e . this is because we want all our trades to be affected in the s a m e way. but have to do it in percentage t e r m s .000 winner that also took your account . For instance. Finally. This is calculated using b o t h the open equity and the already closed out equity on your account. whether the stop is applied to a m a r k e t that has been trending heavily or to several different m a r k e t s that trade at different levels and point values. if you would like to add a stop loss to this system. is that the drawdown is highly overrated as a syst e m testing decision variable. Aside from not placing the drawdown in relation to the market situation at the time. however. the information that can be derived from m o s t system-testing p a c k a g e s will not be enough. B e c a u s e they are ratios. there is the STD. leading d o w n to the exit level and the final profit. the closed trade drawdown ( C T D ) . d r a w d o w n s ! There is plenty that can be said about this intriguing subject. For o n e thing. the d r a w d o w n should be neither neglected completely nor investigated less than thoroughly. and the total equity drawd o w n ( T E D ) . however. T h e n there is the actual m o n e y .CHAPTER 8 Efficient Trades 121 already touched upon in Part 1: the start trade d r a w d o w n ( S T D ) . at a specific point in time. the estimated largest drawdown holds valuable information about h o w large your account size m u s t be and can give you an indication about whether y o u have the psychological profile to trade the system in question. because the n u m ber is given in dollars without any relation to w h e r e and w h e n this b a d sequence of trades struck you.

500.2 A look at different types of drawdown. you n o w are $1. we are dealing with the start trade drawdown (STD) that measures how m u c h the trade went against us after the entry and before it started to go our way. finally. T h e n it gave back $1. leaving you $6. . At point 5. and your total equity drawdown would now be a $1. A c c o r d i n g to m o s t analysis packages.000 to your closed out equity.000. we again have to do with the E T D . However. and your d r a w d o w n is $5. FIGURE 8. At point 4.500 in the hole.500. But before you m a n a g e d to exit the position. it gave back $1.000 in the hole as c o m p a r e d to your latest total equity high. they consist of three different types of drawdowns.500 loss before. Figure 8.500. if your next trade turns out to be a loser that immediately had you stopped out with a $4.000 of its open equity. before it took off. if you look closely at these n u m b e r s . we are looking at the closed trade drawdown ( C T D ) that measures the distance between the entry and exit points without taking into consideration w h a t is going on within the trade.000.500 of the open profit. let's say. A n d at point 7. Furthermore. $9.000 w i n n e r which. started out going the w r o n g way with an additional $1.2 shows what this sequence of trades could have looked like. we are dealing with the end trade drawdown ( E T D ) that tells us how m u c h of the open profit we h a d to give back before we were allowed to exit a specific trade.122 PART 3 Getting Out to a n e w total equity high at. At point 3. this m e a n s that even t h o u g h you managed to a d d $3. Then you have a $5. your closed out equity n o w is $4.

But w h e n only examining the overall drawdown n u m b e r (the T E D ) and blindly trying to do something about it leaves us with no way of knowing w h a t it is we're really doing and what is actually changing within the system w h e n we're m a k i n g any changes to the input parameters. M o r e recently. 1997) came up with the concepts of maximum adverse excursion ( M A E ) and maximum favorable excursion ( M F E ) . while m o s t of the trades with an S T D / M A E less than 1% turned around and produced a profit. and how often will that happen? For instance. technical analysis editor at Technical Analysis of Stocks and Commodities. But the question still s e e m s to r e m a i n open about whether this analysis is necessary and how one can gain from it. in recent years a few system developers and market analysts have addressed this issue in various ways. M o s t of the trades that produced an S T D / M A E over 2% also ended up as losers. L o n g e r . To be sure. m a n y of your trades will experience a start trade drawdown ( S T D ) before they start going your way.CHAPTER 8 Efficient Trades 123 Of course we would like to keep all these d r a w d o w n n u m b e r s as small as possible. our S T D will be the distance between the entry point (the open) and the low of the day. In this case. you m u s t c o m e to grips with the STD. David Stendahl at RINA Systems has taken this a little further and developed a method to calculate the efficiency of a trade. who. E T D . that in this particular case we are assuming that the low c o m e s before the high. F u r t h e r m o r e . w h e n they enter with a stop order on the highest high or lowest low over the lookback period. With version 2000i of O m e g a Research's TradeStation. E T D . which is not always the case for a system like this. the efficient trade analysis also has been implemented in the systems and optimization report. 1996) and Maximum Adverse Excursion (Wiley Trader's Advantage Series. Usually.t e r m b r e a k o u t systems often also experience an STD. Depending on what type of entry technique you are using. however. this level coincides with a pivotal resistance or support level in the . Figure 8. with the S T D charted on the horizontal x-axis and the result of the trade on the vertical yaxis. where you enter with a limit order. and CTD. the T E D is m a d e up of the STD. At any given m o m e n t . One of these analysts is John Sweeney. F r o m this chart we can see that only one trade with an S T D / M A E of m o r e than or equal to 3% m a n a g e d to produce a slight profit. and C T D . This is especially true for short-term top and bottom picking systems. there still remains the question of h o w to go about m a k i n g the m o s t out of it. This m e a n s that to c o m e to grips with the T E D . but as far as I k n o w nobody has really nailed it down when it comes to how one must go about examining the markets and the systems in an appropriate and scientific fashion. N o t e . the only way to avoid an S T D is to enter at the absolute low or absolute high.3 shows this. let's simplify the code for our intraday M e a n d e r system slightly to go long on the open. If we open below the VS M i d level and then stay in the trade all the way to the close the same day. in his two books Campaign Trading (Wiley Finance Editions. if it is necessary.

LongEntryDate(O). ExitAvg(O). because this system. from which. the m a r k e t usually m a k e s yet another correction before the final penetration and follow-through takes us away (we h o p e ) in a major trending m o v e . Figure 8. ShortEntry(O). as it is constructed so far. To export the necessary data for the directional slope syst e m into a spreadsheet program. getting a feel for the S T D is crucial.124 PART 3 Getting Out FIGURE 8. operates not only completely without any actual levels for w h e n a trade gets triggered. LongExitDate(O). a n d the reason w h y the trade is triggered. (Note that in this section we work with an 18-bar Entry MA a n d a 12-bar Exit M A . other s y s t e m s might completely lack any specific levels around which to place any natural orders. Vars: EntryAvg(O). ExitMA(12). LongEntry(O). for instance (which b a s e d its entries on the slope of the m o v i n g average rather than the crossover).4 shows that there is no connection b e t w e e n the level of the price or the m o v i n g average.3 The final profit In relation to the STD/MAE for the intraday Meander system. ) To start looking for the S T D / M A E and E T D / M F E levels that should work well on several different markets and time periods. market. In our directional slope system. LongExit(O). ShortEntryDate(O). use the following TradeStation code: Input: EntryMA(18). after this level has been tested a n d y o u ' v e entered the trade. but also completely without any exit rules as we define t h e m . Yet. we must do all our research with the R A D contract. . ShortExit(O).

4 The directional slope system applied to the Japanese yen. .CHAPTER 8 Efficient Trades 125 FIGURE 8.

TotTr(O). TradeStr2(""). Entry Avg = Average(Close. TradeStr3(""). ShortEntryBar = BarNumber. If MarketPosition <> — 1 Then Begin ShortEntry = Close. MAEfe(O). LongEntryBar = BarNumber. DateForEntry(O). HighToLow(O). ShortExitBar(O). Conditionl = EntryAvg > EntryAvg[l]. MFEBar(O). PriceForEntry(O). Condition2 = ExitAvg > ExitAvg[l]. End. If Condition2 = True Then . If MarketPosition = — 1 Then Begin ShortExit = Close. LongEntryDate = Date. PosProf(0). LongExitBar = BarNumber. PrelMAEfe(O). LowestLow(Low). LongEntryBar(O). End. If Conditionl = True and Condition2 = True and MarketPosition = 0 Then Buy at Close. PriceForExit(O). HighestHigh(High). LowToHigh(O). MP(0). BarsForTrade(O). ExitMA). ShortExitDate = Date. FTE(O). End. DateForExit(O). ShortEntryBar(O). ETD(O). If MarketPosition = 1 Then Begin LongExit = Close. ShortEntryDate = Date.126 PART 3 Getting Out ShortExitDate(O). If Condition2 = False Then ExitLong at Close. MFE(O). ExitAvg = Average(Close. If MarketPosition <> 1 Then Begin LongEntry = Close. TradeStrl (""). ShortExitBar = BarNumber. LongExitDate = Date. LongExitBar(O). End. EntryMA). MAE(O). If Conditionl = False and Condition2 = False and MarketPosition = 0 Then Sell at Close.

HighestHighfl]) / HighestHighfl]. TradeStr2 = NumToStr(PosProf. HighToLow = Lowest(Low." + NumToStr(PosProf. End. TradeStr2 = NumToStr(PosProf. PrelMAEfe = (HighToLow . . MFE = (HighestHigh — LongEntry) / LongEntry. HighToLow = 0. (MFEBar MaxList(MFEBar[ 1 ]. PosProf = (Close — LongEntry) / LongEntry. If High > HighestHigh Then Begin HighestHigh = High. TotTr = TotalTrades. MP = MarketPosition. End. TradeStr2 = TradeStr2 + '*. MFEBar = BarNumber. If MarketPosition = — 1 Then Begin If BarsSinceEntry = 1 Then Begin LowestLow = ShortEntry. LowToHigh = 0. End. LongEntryBar))). If Low < LowestLow Then Begin LowestLow = Low. If MarketPosition = 1 Then Begin If BarsSinceEntry = 1 Then Begin LowestLow = LongEntry. If PrelMAEfe < MAEfe Then MAEfe = PrelMAEfe. HighestHigh = LongEntry. MAE = (LowestLow — LongEntry) / LongEntry. 4). MAEfe = 0. 4). End.CHAPTER 8 Efficient Trades 127 ExitShort at Close. MAE = 0. MFEBar = BarNumber. 4). HighestHigh = ShortEntry. MFE = 0.

MAE = (ShortEntry . End. MAEfe = 0. End. MFEBar = BarNumber. DateForEntry = LongEntryDatefl]. FTE = (PriceForExit .HighestHigh[l]) / HighestHigh[l] Then MAEfe = (PriceForExit .HighestHigh) / ShortEntry. PosProf = (ShortEntry . LowToHigh = Highest(High. 4). MFE = (ShortEntry . If High > HighestHigh Then Begin HighestHigh = High. MFEBar = BarNumber.128 PART 3 Getting Out MAE = 0. If PrelMAEfe < MAEfe Then MAEfe = PrelMAEfe.PriceForEntry) / PriceForEntry. BarsForTrade = LongExitBar[ 1 ] . TradeStr2 = TradeStr2 + ".LongEntryBar[l]. ShortEntryBar))). End. I f M P [ l ] = -1 Then Begin PriceForEntry = ShortEntry[1]. DateForExit = LongExitDate[l]. (MFEBar MaxList(MFEBar[ 1 ]." + NumToStr(PosProf. PrelMAEfe = (LowestLow[l] — LowToHigh) / LowestLow[l]. . If MAEfe[l] > (PriceForExit . PriceForExit = LongExit[l].HighestHigh[l]) / HighestHighfl].Close) / ShortEntry.HighestHigh[ll) / Highesthigh[l]. ETD = (PriceForExit . If TotTr > TotTrfl] Then Begin I f M P [ l ] = 1 Then Begin PriceForEntry = LongEntryfl]. MFE = 0. End.LowestLow) / ShortEntry. If Low < LowestLow Then Begin LowestLow = Low. End.

0). 0) + ". ExitLong ("Profit") tomorrow at EntryPrice * 1. If FTE < MAE[1] Then MAE = FTE Else MAE = MAE[1].PriceForExit) / PriceForEntry.CHAPTER 8 Efficient Trades 129 PriceForExit = ShortExit[l]. 0) + ". DateForExit = ShortExitDate[l].PriceForExit) / LowestLow[l] Then MAEfe = (LowestLow[l] — PriceForExit) / LowestLow[l]. However. consequently.02 limit. 4) + NumToStr(FTE." + TradeStr2[l] + NewLine. DateForEntry = ShortEntryDatefl]. FTE = (PriceForEntry . apply the following c o d e to any market: Buy at Close." + NumToStr(MFE.ShortEntryBar[l].98 stop. . 2) + ". 0) + "." + NumToStr(PriceForExit.98 stop. 4) + ". its length is also because of one of the stupidest features in TradeStation—the way the p r o g r a m registers an open trade and its entry price a n d counts the n u m b e r of bars in trade." + NumToStr(ETD." + NumToStr(DateForExit. I am not the best p r o g r a m m e r around. does n o t calculate the n u m b e r of bars in the trade correctly. 4 ) + ".TradeStr3)." + NumToStr(MP[l]. ExitLong ("Loss") tomorrow at EntryPrice * 0." + NumToStr(DateForEntry. Unfortunately. TradeStr3 = TradeStrl + ". End." + NumToStr(BarsForTrade. 4) + "." + NumToStr(MAEfe. TradeStrl = LeftStr(GetSymbolName. T h e result is that TradeStation will exit each trade at the o p e n of the next bar. 4) + ". If FTE > MFE[1] Then MFE = FTE Else MFE = MFE[1]. TradeStation does not note that a trade is o p e n until after the first full bar in the trade and. ExitLong ("Profit") tomorrow at EntryPrice * 1. For instance." + NumToStr(PriceForEntry.csv". BarsForTrade = ShortExitBarfl] . which is probably revealed by the code above. FileAppend("D:\Temp\DSS.02 limit." + NumToStr(MAE. 4) + + ". ETD = (LowestLowfl] — PriceForExit) / LowestLow [1]. ExitLong ("Loss") tomorrow at EntryPrice * 0. If MAEfe[l] > (LowestLow[l] . B u t if you change the code to the following: Buy at Open. 4) + ". End. no matter what.

zero-profit trades. for that to happen we simply must find a way around the problem. You will not be able to exit on the close. Together. If BarsSinceEntry > = 1 Then Begin ExitLong ("Loss") tomorrow at EntryPrice * 0. Another example of the same flaw is when you. Hence. N o r is either one of t h e m the s a m e as R I N A S y s t e m s ' entry efficiency. O n c e we are in the spreadsheet p r o g r a m it is easy to create a scatter chart like Figures 8. for instance. or intrabar of that same bar. B u t then TradeStation simply skips a bar before it starts to track the trade and does not let you exit the trade until bar three (counting the b a r for the entry as bar one). O n e way to work around this is to rewrite the code to: Buy at close. the M A E and S T D m a k e up the entry efficiency. which obviously increases the risk for other types of errors and logical blunders. it can be difficult to distinguish between the two.) To m e . w h e n the formula get too complicated. (That is why I have chosen to do all the research for the intraday version of the Meander system within Excel. It is still your m o n e y we are talking about here. E n o u g h said. such as a stop loss. T h e s e . End. I do not k n o w about you. it is important to understand that the two are not the same.02 limit. by the way. it is my (your) m o n e y we are dealing with here and we should all expect things to work properly. T h e point is that you should at least be aware of the differences and exactly what it is you are trying to achieve. everything else is a loss. without Microsoft's letting you k n o w that sometimes. After all. while the M A E should primarily be adjusted with a specific exit technique. it loses track a n d simply m a k e s up an answer.98 stop. no matter h o w m u c h the market moves against you. but to me a w i n n e r is a trade that ends up with a profit. Unfortunately. . which w o n ' t c h a n g e even if you decide to enter at the open instead of the close. we end up with a large c h u n k of extra spaghetti code.3 a n d 8. M a n y times. are all counted as winners in the performance summary. this is the equivalent of Excel's purporting to calculate everything in a spreadsheet correctly. let us work with what we have. ExitLong ("Profit") tomorrow at EntryPrice * 1. T h e difference is that the S T D should primarily be adjusted w i t h the entry technique. Although we might be using the same techniques to derive the necessary data for the S T D as for the M A E .5 that compare the S T D to the final profit. rather than in TradeStation. as suggested by the closing price of that same bar. enter a trade intrabar on a breakout. Notice also the word "primarily.130 PART 3 Getting Out T h e result is that TradeStation enters a n d exits each trade on the very same open." because in a g o o d working system all parts should and will intervene to form a w h o l e larger than the parts. A n d setting the BarsSinceEntry function equal to or greater than zero does not work either. leaving you with a b u n c h of zero-bar.

M F E is the m a x i m u m open profit of the position. T h e ETD. . In the case of the directional slope system. perhaps all that m u s t be d o n e to the entry is to wait a couple of ticks before you enter.5 shows the S T D / M A E for the directional slope system traded on the Japanese yen. E T D is the difference between the M F E a n d the exit point. on the other hand. or at least not enter immediately at the o p e n w h e n the volatility can be high. o n e solution could be to wait for a pullback and then enter on a b r e a k through the price level that causes the Entry MA to change its direction. In the case of the M e a n d e r system. and the amount you are giving back to the market before the syst e m signals the time to exit. after you have m a d e sure that you have optimized your exit techniques and taken every other step necessary.5 The final profit in relation to STD/MAE for the directional slope system. should be managed with different types of exit techniques and stops. This chart shows that of all trades with an S T D / M A E of m o r e than or equal to 4% only two did not end up as losers. although we might be using essentially the s a m e techniques to derive the necessary information. T h e M F E and the final profit should therefore be m a x imized with the amount invested and the n u m b e r s of contracts traded. while most of the trades with a n S T D / M A E o f 2 % o r less e n d e d u p a s winners. Figure 8.CHAPTER 8 Efficient Trades 131 FIGURE 8. Just as the S T D is not the same as the M A E . the end trade drawdown ( E T D ) is not the s a m e as the m a x i m u m favorable excursion ( M F E ) .

Sure. From Figure 8.132 PART 3 Getting Out such as a trailing stop. This has been done in Figures 8. we must look at the E T D in relation to the MFE instead of in relation to the final profit.6 and 8. while Figure 8. Figure 8.7 illustrates the final profit in relation to the M F E for the directional slope system traded on the Japanese yen. It shows that not until we get an M F E above approximately 5% does the system manage to signal an exit before we have given away all the open equity. or a m o r e conservative entry technique (to address S T D ) most of these losses could have been reduced considerably. 5 % very often resulted in a loss greater than 1%. or profit target stop.6. FIGURE 8. Looking at Figures 8. With a stop loss or trailing stop at an appropriate level (to address E T D ) . time-based stop. and the S T D and E T D b e c o m e s the same. The only way to completely avoid any E T D is to exit all trades with a limit order that is triggered at the top of the equity curve.7. but what? To get a better feel for this.8 and 8.7 shows the relationship between the M F E and the final profit for the directional slope system.6 The final profit in relation to ETD for the Meander system. This happens when the trade goes immediately against us. the higher the M F E . we can see that an E T D of m o r e than 0 .5%. Figure 8. Figure 8. we realize that something needs to be done somewhere around the 5% level for the directional slope system. Because this modified version of the M e a n d e r system operates with a d y n a m i c profit target that triggers an exit with a limit order.8 shows that for all trades with an M F E above 0.9. the lower the E T D . .6. it is not too difficult to see that it is easier to handle the information from Figure 8.6 shows the relationship between the E T D and the final profit for the simplified M e a n d e r system.

CHAPTER 8 Efficient Trades 133 FIGURE 8.8 The ETD in relation to MFE for the Meander system. .7 The final profit in relation to the MFE for the directional slope system. FIGURE 8.

N o t e that in this case the E T D is denoted with negative values. m a n y of these losses can be avoided. especially for a longer-term trend-following system that is p r o n e to give back a substantial part of the open profit for a trade before it allows you to exit. as the differences can be quite substantial. Figure 8. Because m a n y times the C T D also is the s a m e as the STD. there are still several that have a larger E T D than M F E .134 PART 3 Getting Out however. the directional slope system s e e m s to give away a substantial part of the profit before it allows us to exit.9 shows that both trades with an M F E immediately above 2 0 % did not allow us to exit until the market had taken back about a third of our open profit. the C T D also can be a function of the ETD. given the entry trigger you use. Especially for long-term trend-following systems. It also is important to distinguish the C T D from the T E D . As is the case with so m a n y other long-term trend-following systems. T h e closed trade drawdown ( C T D ) is the drawdown you experience on your account from one trade to another for all your losing trades. With a profit protection or breakeven stop.9 The ETD in relation to MFE for the directional slope system. you m u s t c o m e to t e r m s with the C T D by limiting the n u m b e r of trades with an overriding filter. For instance. which lets through only those setups and long-term market situations that show the highest likelihood of producing a w i n n i n g trade. . which m e a n s it also should be matched with an appropriate exit tech- FIGURE 8.

m o s t of the time it does so right off the bat. separated into w i n n i n g and losing trades. only on a few occasions w a s the final p r o f i t / C T D significantly different FIGURE 8. From Figures 8. a n d not clutter t h e analysis w i t h u n n e c e s s a r y information. T h e s a m e p h e n o m e n o n also can be seen in Figures 8. To get a feel for the C T D . we also can see that of all winning trades. the relationship between the final profit and S T D / M A E isn't as clear-cut as for the losers. . which in m o s t instances also produced a losing trade of m o r e than 2 % . you can chart the final profit. or at.10 and 8. such as a trailing stop.10 shows that if a trade within the directional slope system turned out to be a loser.CHAPTER 8 Efficient Trades 135 nique.12. against both the S T D / M A E . This suggests that w h e n a trade goes bad. w h i c h show the C T D respectively. At the other e n d of the spectrum there were 16 losing trades with an S T D of m o r e than 2 % . of which only three m a n a g e d to produce a profit of m o r e than 10%. w h i c h indicates that to c o m e to grips w i t h the C T D .11 shows that for the w i n n i n g trades. m o r e often than not. only six had an S T D of m o r e than 2 % .11. As you can see from Figure 8. In fact. only on a few occasions w a s the final p r o f i t / C T D significantly different from the S T D / M A E . if the trade turned out to be a loser.10 The final profit for all losing trades in relation to STD/MAE for the directional slope system. it is important to analyze it separated from w i n n i n g trades. the final profit for the M e a n d e r system. Figure 8. it also m a n a g e d to be closed out very close t o . the lowest low of the trade.13.12 a n d 8. and against the M F E for all losing trades. if the trade turned out to be a loser. Figure 8. in relation to the S T D / M A E .

m o s t of the time it does so right off the bat.136 PART 3 Getting Out FIGURE 8. and could this have been dealt with before entering the trade in the first place? W h e n we are comparing the final profit for the trades within the directional slope system with the M F E we can see that there are a few trades that seem not to be stopped out until there has been an adverse move in the neighborhood of 5% or more. and only one came close to 2 % . F r o m Figures 8. F r o m Figure 8. m a n y of these trades seem to have been closed out at or very near their lowest open equity. which confirms that to c o m e to grips with the C T D it is important to analyze it separated from winning trades. what was the current market situation like to produce these bad trades. which m a k e s it a good idea to . A n d just as w a s the case for the directional slope system.13 you again can see that the similarity between the long-term directional slope system and the short-term M e a n d e r system is striking. with no unnecessary information. T h e question is.11 The final profit for all winning trades In relation to STD/MAE for the directional slope system. Obviously. 5 % . this is too m u c h to give back. On the other hand.12 and 8. there are plenty of trades with an S T D of more than 1% that also produced a final loss greater than 1%.13 we also can see that very few trades with an STD of m o r e than 2% m a n a g e d to p r o d u c e a profit above 0 . This suggests that w h e n a trade goes bad. from the S T D / M A E . T h e relationship between the winning trades and their S T D / M A E aren't as clear-cut as for the losers.

CHAPTER 8 Efficient Trades 137 FIGURE 8. FIGURE 8.12 The CTD in relation to STD/MAE for the Meander system. .13 The final profit for all winning trades in relation to STD/MAE for the Meander system.

and the E T D / f i n a l profit in o n e s e q u e n c e . T h e s a m e holds true for the M e a n d e r system in Figure 8. as they are a s s u m e d to h a v e taken place.5% or m o r e before w e ' r e allowed to exit on the close of the bar. no matter in w h a t time perspective w e ' r e trading. the larger the M F E . which has quite a few trades that start out with a profit but then allow for an adverse move of more than 1. There also are plenty of trades for which m o s t of the losers failed to p r o d u c e an M F E of more than 0. it is likely to be persistent and lead to a profitable trade. whether we're trading long t e r m or intraday. Again. the smaller the final loss. Figure 8.138 PART 3 Getting Out e x a m i n e the possibility of adding s o m e form of a trailing stop. when a trend gets hold of the market it is likely to be persistent and lead to a profitable trade. we could have avoided some of the losses with a stop loss of s o m e sort. In the former case. In the latter case. m a n y trades could probably have been avoided completely with some sort of a filter. F i g u r e 8. Also.14 The CTD in relation to MFE for the directional slope system. the smaller the final loss. .15. the M F E . This indicates that w h e n a trend gets hold of the market. With a good working filter m a n y of these trades could have been weeded out as well. this indicates that most losers go b a d from the point of inception or very soon thereafter. A n o t h e r way to s u m m a r i z e the entire life s p a n of a trade is to chart the S T D / M A E . A l s o .14 also shows that a total of 17 out of 24 losing trades for the directional slope system only m a n a g e d to produce an M F E of 2% or less. Again. the larger the M F E .2%).16 attempts to s h o w the s u m m a r i z e d life FIGURE 8.

m o s t M F E s h a p p e n e d after the S T D s . this assumes that the course of events is as outlined. and we can register an M F E of about 0 . after we have given b a c k m o r e than 5 0 % o f the o p e n profit.17. T h e a s s u m p t i o n we m a k e is that the further we get from the point of origin. in the case of the directional slope system. which leads to a higher standard deviation of the outcomes. It is reasonable to assume that the further we get from the point of origin. before it takes off in the right direction to register an M F E of about 7 . 1 % . w h i c h leads to a h i g h e r standard deviation of the o u t c o m e s . the m o r e divergent the p a t h s the trade could have taken.17 shows that the average trade starts out with a wrong move of approximately 0. T h i s is not entirely correct b e c a u s e not all t r a d e s register their M F E s after their M A E s . the m o r e divergent paths the trade could have taken. it is r e a s o n a b l e to a s s u m e that. which shows the summarized life of a trade in the M e a n d e r system. which m e a n s that we have to be a little m o r e careful interpreting our S T D and E T D findings. 6 % and a final profit of about 0 .15 The CTD in relation to MFE for the Meander system. T h e s a m e p h e n o m e n o n can be seen again in Figure 8.ER 8 Efficient Trades 139 FIGURE 8. we can see from Figure 8. Again.16 that the average trade for the directional slope s y s t e m on the J a p a n e s e yen first starts out by going against us with about 2% (the m i d d l e line).6% before things start to go our way. B u t a s s u m i n g this is the c a s e . 5 % . it is downhill again. before we are stopped out for a final profit of approximately 3 % . for a trade in the directional slope system. Therefore. Figure 8. however. however. . F r o m then on. This seems not to be the case for the M e a n d e r system.

17 The summarized life for a trade in the Meander system.140 PART 3 Getting Out FIGURE 8.16 The summarized life for a trade in the directional slope system. . FIGURE 8.

where we take a closer look at Sweeney's M A E and M F E . R e m e m b e r from Part 2 that a trade that is too good to be true is not necessarily a good thing. this does not happen in this chart. Although the old adage "your worst drawd o w n is still to c o m e " is likely to c o m e true sooner or later. as well as better ways to depict the life of a trade. with its standard deviation interval. Nonetheless. are dealt with in the next chapter. individually examining each point.5 . in the case of the directional slope system we can expect that 6 8 % of all trades should produce an S T D / M A E somewhere between . it does not have to h a p pen first thing tomorrow. provided you have done your homework correctly.5%. Different methods for handling these trades. also can give us some clues to where and how to place our stops. For instance.3 . 7 to 0 % . Trades that do not match these criteria could then be assumed to be better or worse than acceptable. which are likely to m o v e further apart from each other the further away from the entry point we get. and a final profit between .CHAPTER 8 Efficient Trades 141 That this is not always the case is noticeable in the standard deviation interval (the upper and lower lines). . 5 and 11. an M F E in the interval -2 to 17%.

For a gunslinger or a sniper.CHAPTER 9 Sweeney's MAE/MFE Look at it this way: W h e t h e r your trading style resembles a gunslinger shooting from the hip or a well-aimed sniper lying in ambush. but that every trade also has a unique set of characteristics and traits that distinguish the winners from the losers. The M A E does not take into consideration whether this happens immediately at the inception of the trade. k n o w i n g where your trades are heading could m e a n the difference b e t w e e n riding into the sunset or lying fatally w o u n d e d on a dusty street at high n o o n . To stay alive you m u s t k n o w when to draw a n d when to run. is find w h e r e your targets have been and how you have performed in the past. which corresponds to the highest open equity. The M A E is defined as the most negative intraday price movement against your position. Although this hardly is possible in a shootout that d e m a n d s split second decisions. W h e n you do. which corresponds to the lowest open equity during the lifespan of the trade. This chapter illustrates one way of interpreting and m a k i n g better use of Mr. All of these characteristics often can be spotted w h e n you trade using John Sweeney's M A E and M F E techniques. the ultimate advantage c o m e s from being able to calculate the trajectory of the bullet. The M F E is defined as the most positive price movement for your position. To do this you must take a deep look into your population of trades and examine each individual trade closely. you will discover that there are several similarities between them. after the trade might have lost all its open profit and turned into a loser. T h e first thing you m u s t do. Sweeney's excellent work. or just before exiting. during the lifespan of 143 . as illustrated in Figure 9 . 1 . it can be d o n e in the financial markets where you can estimate the trajectory of your trades in relation to their targets.

.1 MAE and MFE levels for two trades in the Japanese yen.144 PART 3 Getting Out FIGURE 9.

we can examine our systems for high-probability exit techniques and even for situations when it could be a good idea to add to a trade.) This helps us examine our systems for additional trading rules. If we look at the long trade and refer back to the different drawdowns just discussed. in our directional slope system from Part 2. do not forget to change from the ordinary point-based back-adjusted contract to the R A D contract. lumber. If . that in Figure 9 . the M F E targets are the bad guys to be aimed at. as before. As an analogy. And. for instance. both M A E s occurred after the M F E s . crude oil. In this case. But once in the trade anyway. we start by looking at all trades (winners and losers. and coffee. corn. With the M A E and M F E techniques. C R B index. the distance between the entry and the M A E level would in fact make up the S T D and as such should preferably have been avoided completely with the help of some sort of a filter. (A sort of trajectory analysis if you like. while the M A E targets are the "innocent p e o p l e " w h o should be avoided. the adverse move following the M F E should have been limited by a trailing stop or a stop loss. and for the short trade also in conjunction with the exit of the trade.CHAPTER 9 Sweeney's MAE/MFE 145 the trade. practical. over the time period January 1980 to October 1999. 1 . T h e markets tested are T-bonds. c o m pare the M A E / M F E values with the pop-up targets at a shooting range. gold. because this will only result in a suboptimal solution. or if it could be a good idea to throw away a couple of additional shots in the same direction w h e n we have o n e of the bad guys in our sights. Additionally. O n c e in your spreadsheet program. T h e trajectory analysis helps us determine if we should "freeze" the bullet (the trade) mid-air if we mistakenly shoot at one of the good guys (or even take a shot in the completely opposite direction). respectively. live cattle. If you have this workspace saved already. or even fade (to trade against the trend) a previously given signal. Canada dollar. we continue to use the code from the previous chapter. copper. The M F E does not take into consideration whether this happens early on in the trade or just before exiting with or without a profit. Japanese yen. longs and shorts) for all markets. Eurodollar. albeit with a little backwards reasoning. cotton. do not deduct any m o n e y for slippage and commission. and then divide our analysis into winners and losers. Notice. and beneficial form by which this can be done? To export the entire necessary data into a spreadsheet. each curve of the open equity is a mirror image of the price development of the market. orange j u i c e . depending on whether the trade stays within the boundaries set by the M A E / M F E analysis. In this case. which m a k e s it possible to compare each trade's ongoing price action with what actually happened during similar situations in the past. isn't this the ultimate. m a k e three copies of the spreadsheet. with the trajectory analysis. S & P 500. Let us apply this code to all markets. dollar index. As a c o m p l e m e n t and a starting point to the M A E / M F E analysis we also take a closer look at the entire lifespan and development of a trade. If technical analysis is the study of the price action from the past on the assumption that it will repeat itself in the future.

Although we already can see s o m e tendencies and form some basic guidelines. already at this point we can distinguish a few differences between the winners and the losers. For o n e thing.3.2 The development for a set of trades for the directional slope system. while the winners seem to live on for much longer. . together with its standard deviation boundaries. " this is not enough to formulate any systematic and mechanical rules that m o v e us away from "trading with the gut. the longer the trade lasts. approximately by bar 1 8 " or "around bar 30. There also seem to be a bias (trend) to the upside. To emphasize and m a k e it easier to see this tendency we can channel in the entire bouquet. like "a winner should be in positive territory. as in Figure 9." To be able to do this. T h e first thing to do is calculate the average percentage m o v e for each bar.2. it is fairly easy to see that the life span for several of the losers is fairly short. which shows the development for a set of trades. we can create a chart similar to Figure 9. Although just charting all trades on top of each other m a k e s little sense and looks m o r e like the inside of a magpie's nest. With the data from the spreadsheet. we must simplify things. At the b o t t o m of the spreadsheet. type in the following formula and fill t h e m out to the right for as long as you have any open trades: FIGURE 9.146 PART 3 Getting Out you also would like to examine short and long trades separate from each other. it is r e c o m m e n d e d that you m a k e separate copies for this as well. most trades should show a profit between 10 to 3 0 % .

y o u s h o u l d be able to p r o d u c e a chart like Figure 9.CHAPTER 9 Sweeney's MAE/MFE 147 FIGURE 9.3 The development for a set of trades with manually added trend lines. the fewer the trades. T h e s t a n d a r d deviation b o u n d a r i e s d e n o t e the area in w h i c h 6 8 % of all trades can be expected to fall. the less reliable our results.4. we are n o w d o w n to four. this h a p p e n s at b a r 4 8 . I also have added the regression line (middle straight line) for the average trade. m o r e and the m o r e trades are closed out. To this chart. In this case. m a k i n g for m u c h easier interpretation and rule making. T h e regression equation for this line is . However. T h e vertical dashed line m a r k s out the point w h e r e we have fewer than 20 open trades.STDEV(H 1: H1240) W h e n d o n e . =AVERAGE(H1:H1240) w h e r e H denotes the column w h e r e we stored the percentage move for bar o n e . To this y o u also could a d d a least square regression line for t h e average trade. w h i c h s h o w s the d e v e l o p m e n t of the average trade (the m i d d l e squiggly line) together w i t h the p l u s (upper line) a n d m i n u s (lower line) standard deviation b o u n d a r i e s . = H 1 2 4 6 + S T D E V(H 1 :H 1240) =H1246 . Instead of l o o k i n g at h u n d r e d s of lines that m a k e little sense. T h e further away from the entry point we get.

7 4 % per day. 0.148 PART 3 Getting Out FIGURE 9. T h e dashed line at bar 14 m a r k s out the day after which we have less than 20 trades still alive.5 shows the development of the average losing trade." If we also divide all the trades into groups of winners and losers. which s e e m s to go nowhere fast.5 can give rise to guidelines such as "if we're not profitable by b a r 14. If a trade is not at least marginally profitable at this time. the trade is about to b e c o m e better than average and can be worth adding t o " or "if we cross the u p p e r standard deviation line from below the trade is too good to be true and we h a d better take some profit. and no m o r e than an estimated 1 6 % of all losers ((1 . the trade is probably a loser and should be ditched.4 and 9. as indicated by the u p p e r 1 standard deviation line around bar 16. This is too few to be able to m a k e any statistical conclusions.68)/2) m a n a g e to produce an open profit of m o r e than approximately 7%. which m e a n s that the average trade increases in value by approximately 0 . Figure 9. the average loser hardly m a n a g e s to produce an open profit of m o r e than 2 % . As you can see. With a chart like this we can start to set up some m o r e definite guidelines like "if we cross the average line from below.4 The development of the average trade (the middle squiggly line) together with the plus (upper line) and minus (lower line) standard deviation boundaries. Together.0. your m o n e y is probably better used elsewhere.0074X." . Figures 9. we get an even better feel for our rule making.

while the winners. w h i c h show the same thing as Figure 9. there . T h e m o s t obvious thing w h e n c o m p a r i n g these charts is h o w difficult it is for the losers to take off in any direction. clearly have a u p w a r d bias.5 The development of the average losing trade. 5 % (which is the lower 1 standard deviation b o u n d a r y for the M A E ) is t o o far away from the average trade and therefore b o u n d to be a loser. the dispersion of the o u t c o m e s has not yet had a c h a n c e to b e c o m e that large.CHAPTER 9 Sweeney's MAE/MFE 149 FIGURE 9." T h e standard deviation interval is narrower for the exit than for the point of the M F E because the exit for m a n y losers h a p p e n s m u c h sooner after the entry than t h e t i m e it takes for m o s t w i n n e r s to p r o d u c e an M F E reading in the first place. B e c a u s e of the short distance traveled from the entry for m a n y of these trades.8. By n a r r o w i n g d o w n our analysis as m u c h as this. w h i c h also gives us s o m e additional insight into the system. Figure 9.240 trades in the directional slope syst e m can be seen in Figure 9. like "a trade that quotes an M A E b e l o w 7 .3.7 a n d 9. By the way. we n o w can s u m m a r i z e the life of our trades using only three lines with only four data points each. but this time with the w i n n e r s and the losers separate from each other. on the other hand. with the M A E occurring before the M F E .4 also supports our a s s u m p t i o n from the previous chapter about the typical look of a trade.6. T h a t there definitely is a difference b e t w e e n winners and losers can be seen from Figures 9. W h a t this looks like for all of our 1.

with the highest likelihood based on previous experience. however. are plenty m o r e winners taking off to higher grounds than there are losers taking the express elevator to the basement. In essence. To do this we must perform the same type of analysis as we did in the previous chapter about different types of drawdowns. they quickly take off with a bold m o v e in the right direction. respectively. m o s t losers do not last half that long. we can estimate the outcome of the trade.6 The summarized life for the average trade in the directional slope system.9 shows that the least square regression line for this M A E .10 which show the final profit from all trades in all 16 markets (1. A n o t h e r compelling difference is the lengths of the trades. we can see that m o r e often than not. Figure 9. they give back way too m u c h of the open profit before the system allows us to exit with s o m e of the profits. In fact.150 PART 3 Getting Out F I G U R E 9. Too often. while m o s t losers usually go nowhere fast. Although most w i n n e r s last well above 30 bars. For the directional slope system this may resemble the charts in Figures 9. if we just let it run without any stops or exits. and those that m a n a g e to m o v e into positive territory quickly end their lives with a swift m o v e in the w r o n g direction.9 and 9. The purpose of the M A E / M F E analysis is to find those points in the trade where. a winning trade usually starts out very well right off the bat and continues to develop favorably for quite some time.240 trades in total) in relation to each trade's M A E and M F E readings. As for the winners. very few of the losers really take off in any direction.

7 The development for a set of losing trades.8 The development for a set of winning trades.CHAPTER 9 Sweeney's MAE/MFE 151 FIGURE 9. FIGURE 9. .

H e n c e . This also can be seen from the regression equation. as long as the M A E stays above a certain level. That is. there must be a way to keep more of that open profit? F r o m Figure 9. To do this. T h e exact point where the cumulative value of the losers will be greater than that of the w i n n e r s can be calculated by subtracting the value of all stopped-out losing trades from the value of all winning trades. the larger the loser. which shows that the final profit can be expected to reach approximately 5 7 % of the value of the MFE. type in the following formula at the b o t t o m of your spreadsheet: FIGURE 9.9 The final profit in relation to MAE for 16 markets in the directional slope system.9 we can see that only a handful of trades with an M A E larger than 1 0 % m a n a g e d to end up as winners. seems m o r e to resemble an exponential one. T h e true relationship. which is w h e n the value of losing trades has gotten larger than the value of all winning trades. T h e least square regression line for the M F E analysis in Figure 9. the directional slope system regularly gives back as m u c h as 40 to 5 0 % of the m a x i m u m open profit in the trade before it allows us to exit. dotted trend line. this value eventually gets below zero. you are no longer gaining anything by tightening the stop further. the larger the winner. Surely. That is. we would have m a d e all these trades into 1 0 % losers.10 hints that the larger the M F E .152 PART 3 Getting Out analysis hints that the larger the M A E . however. By setting a stop loss at 10%. as suggested by the handdrawn. As you tighten the stop. . the final outcome of the trade is very likely to be profitable.

0. ." < = .005". the average profit for all those trades would have equaled that level as well. =SUMIF(G$1:G$1240.005")+0." < = .CHAPTER 9 Sweeney's MAE/MFE 153 FIGURE 9. and column / denotes the value of the final profit.1.0."< = -0.005 Using this method.5%). had they been stopped out at the M A E level. to calculate the difference between the results at a specific M A E level. Obviously.I$1:I$1240)/ COUNTIF(G$ 1 :G$ 1240. depending on whether or not you have assumed the trades to be stopped out. which shows the differences in the average profits from all trades corresponding to a certain M A E . Therefore." < = . simply rewrite the above formula to: =SUMIF(G$ 1 :G$ 1240.I$ 1 :I$ 1240)/ COUNTIF(G$ 1 :G$ 1240.005 ". Just calculating the M A E point where the losers are likely to b e c o m e larger than the winners is not enough (in this case the point will happen with an M A E larger than or equal to 1. N o t e that this is the arithmetically calculated value. The next step is to calculate the difference between what would have happened had these trades been stopped out and what would have happened had they not. you can put together a table resembling Table 9.005") w h e r e column G denotes the value for the M A E .10 The final profit in relation to MFE for 16 markets in the directional slope system.0.

That is.24% -0.34 points 2.00% -1.55 points 1.71 points 2.26 points 1. the larger the percentage point difference a m o n g the results from using the stop w h e n c o m p a r e d to not using the stop. 1 .95 points 1. For instance. MAE -0. Such is the nature of this g a m e . but with the stop it obviously would have been —3.00% -8.54% -1.00% -6.154 PART 3 Getting Out TABLE 9. An interesting observation is that generally the larger the M A E . Hence.34 points 1. original profits and differences in outcome for 16 markets in the directional slope system.93 points 1.00% -5.50% -4. we would have done worse with a 3% stop loss than without.84 points 2.50% -1.03 points 1.00% -9.47 points 2.25% -0. As Table 9.05% -1. By m a k i n g sure that you know exactly how m u c h you are expected to lose in a worst-case scenario you decrease the standard deviation of outcomes for the system and increase your feeling of certainty and well-being for that particular trade.42 points 2.29% -6.50% -9. But.84% 0.05%.65% -5. this holds true at the other end of the spectrum as well.25 points 1.96 points 2.50% -7.1 MAE levels. no matter where you place the stop you are worse off than had you not used a stop at all. T h e price you pay for this is the higher likelih o o d for a lower accumulated result over a longer sequence of trades. in this case.50% -10.00% -7.75 points 2. for all trades with an M A E of at least 3% the arithmetically calculated average profit for all trades would have been —1.0%.95% -1. we now can compare the original final profits for all trades corresponding to a certain M A E level to the s a m e set of trades as if they h a d been stopped out at that level.2 shows you.94% -4.69% -6.50% -3.08% ^1.66% -2.00% -2.85 points 2.81 points 3.50% -5.07% -3.50% -2.50% -6.50% -8.12% -6.26 points With these values in place.06 points 2. as can be seen from Table 9 .00% Original profit 0.44 points 1.56% -0.74% Difference 1.75% -3. by taking profits w h e n a trade has reached a certain level we once .38 points 2.16% -4.47% -3.53% -6.00% -3.00% -4.

w h e r e to place the stops.98% 30. So.93% 18.72% 23.05% Difference 3.98% 71.62 points 10.08% 12.93% 36.93 points 6. and differences for 16 markets In the directional slope system.98 points 5.2. T h e point is.00 .44% 63.35 0. by keeping the standard deviation of the outcomes lower. while at the s a m e time also decreasing the ratio b e t w e e n the open a n d closed out profits.72 points 3. " S o w h a t is the point w i t h all t h i s ? " you m i g h t ask.00% 97.40 0. as suggested by Tables 9. original profits.19 points 9. we m a k e things as " b a c k w a r d s " as possible and place a stop loss at 4% a n d a profit target at 6 0 % .30 0.05 0. TABLE 9.85 0.45% 43.34% 60.00 points 7.80 0.94 points 15.20 0.44 points 17.62% 50.CHAPTER 9 Sweeney's MAE/MFE 155 again lower the standard deviation of o u t c o m e s .00 points 2.00% 102. Original profit 8.65 0.1 and 9.45 0.55 0.19% 86.50 0. In this case.34 points 15.94% 80.25 0.47 points 16.93 points 3.98 points 16. if we even should have any? That is pretty m u c h a question only you can answer depending on h o w you evaluate the risk/reward relationships between different stop or profit target levels for the particular system on w h i c h you currently are working.05 points .90 0.08 points 2. no matter where we place the profit target level.75 0. simply because those are the levels (within reason) w h e r e we lose the m o s t in the long run.45 points 8. we can be m o r e aggressive w h e n it c o m e s to the n u m b e r of contracts traded in our portfolio.70 0.00% 97.53 points 12." T h e price we pay is a lower accumulated profit over a longer sequence of trades.95 1.44% 77.10 0.15 0.19% 89.44 points 13.53% 97.60 0.00 points 2. This should also m e a n that these levels will help us gain the m o s t w h e n we apply the system to a fixed fractional m o n e y m a n a g e m e n t strategy.2 MFE levels.19 points 11. according to an optimal / or fixed fractional trading strategy. MFE 0.47% 86. m a k i n g the system "less risky.

85% 11.82% 6. MFE 9.96% 17. MAEfe -2.00% -8.00% -9.00% -7.10% 6.04% 6.32% 7. we n o w turn our attention to the trailing stop. but rather any of the previous m a x i m u m favorable excursions preceding the largest retracement in the opposite direction that did or did not result in an exit. In this c a s e .31% 14.39% 12. TABLE 9. and theoretical profits for 16 markets In the directional slope system.02% 16. we c o m p a r e the M A E f e w i t h the average c o r r e s p o n d i n g M F E for all trades a n d then calculate the theoretical result as if the M A E f e h a d b e e n the end trade d r a w d o w n .54% 7.00% -10.50% -6.156 PART 3 Getting Out BEYOND MAE/MFE W i t h a stop loss and a profit target in place.89% 6.98% 6.50% -11.08% Profits 7.50% -7.50% -4. T h e difficulty w i t h t h e M A E f e is that we do not k n o w w h e r e the m a r k e t w a s trading before the M A E f e (at least n o t in this analysis).00% -5.34% 13.02% 6.50% -5.78% 10.43% .98% 7. We call this point M A E f e .50% -9.24% 20.00% -11.66% 13.53% 7.02% 12. leading into the end trade drawdown.00% -3.50% -10.88% 6.3 MAEfe levels. Therefore to c o m e u p w i t h s o m e meaningful results w e m u s t p r o c e e d differently f r o m the way w e w o r k the profit target a n d the stop loss analysis. w h i c h is likely to be the very last favorable excursion.62% 7. 3 .84% 7.74% 7. T h e basic Excel formulae are t h e s a m e as for the stop loss a n d the profit target.51% 22.06% 19. T h e results can be seen in Table 9 .97% 6.68% 6.24% 18.69% 6.28% 10.50% -12.00% Avg. average MFE levels.00% ^.86% 15.50% -3.94% 6.71% 14. N o t e that this does not have to be the m a x i m u m favorable excursion point.50% -8.14% 20.92% 21.33% 16.03% 7. We would like to find the m a x i m u m adverse excursion ( M A E ) from the latest favorable excursion (FE) within e a c h trade.00% -6.

if a trade stays or turns negative after this bar. on average. 7 4 % . and then goes nowhere from there. is a likely reversal candidate (perhaps after a prolonged m o v e in the same direction). All trades that did not produce a retracement of 4% will continue to produce the same total profit as they would have before. Table 9. this open profit will. it could be a g o o d idea to c o m p l e m e n t the initial stop loss with a time-based stop. This finding. N o t e also that for both the stop loss and the trailing stop. we only start giving back large parts of the open profit before we're allowed to exit in accordance with the original system. which is triggered if the trade hasn't reached a certain profit within the stipulated n u m b e r of bars.5. in conjunction with what we already have discussed.CHAPTER 9 Sweeney's MAE/MFE 157 F r o m Table 9. A n o t h e r way of lowering the standard deviation of the o u t c o m e is to work with time-based stops as a c o m p l e m e n t to any of the stops discussed so far.96%) .5 we can see that all trades that last for 40 bars or longer have an average open profit of 2 5 . A trailing stop of 4% from the latest favorable excursion also m a k e s sense from a symmetrical point of view w h e n c o m p a r e d to the stop loss and what could be a general market behavior. Also. there is a bottom for the theoretical profits. With a trailing stop at this level. it will take an average o p e n profit of 8. which should be positive after bar 14.4 shows that if we would like to take the expected end trade drawd o w n ( E T D ) into consideration to calculate the estimated final profit for a trade.3 we can see that w h e n the M A E f e is at 4 % . we employ a strategy similar to that we used for the stop loss and profit target and put together a table like Table 9. For the average trade. on average. the average profit from all stopped out trades should now be slightly below 7 % . To investigate this. to streamline the entire group of trades and lower even further the standard deviation of the outcomes of all trades. This m e a n s that although the average open profit for bar 14 is 8. From Table 9. If it is reasonable to assume that a market that first takes out the latest resistance/support level signals that it is ready to take off in o n e direction. but p r o d u c e an average final profit of 31. the next worse level s e e m s to be very close to 8%. That is. we let the trade live on if it is below this level. be reached on b a r 14. as long as it does not m o v e into negative territory. if we let a trade go on for too long. m a k e s it seem a good idea to put a time-based stop loss at this bar. At the other end of the trade. we exit with a market order at the close. if we assume that the breakout and initial entry usually coincides with a test of a previous high (low) that acts as a pivotal resistance (support) level. and a s s u m i n g our reasoning m a k e s sense. w h i c h further confirms the theory that there s e e m s to be some kind of symmetrical behavior that results in. Note that it is the final profit in which we are interested. 4% retracements (or multiples thereof) after the market has set a new high/low in the direction of the underlying trend. a time-based stop also could work as a c o m plement to the profit target.59% before we can expect the average trade to e n d with a small profit at s o m e bar in the future.58.

and differences for 16 markets in the directional slope system.06% -8.07% 6.96% -8.22% -8.6 . Bars in Trade 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 Open profit 0.55% -8.4 .37% 22.31% 8.3 . the time-based stop also could be implemented into the trailing stop technique.5 .42% 16.29 points 6. ETD.60% -8.74% ETD -5.66% 19. the trailing stop will probably . O n e way of doing this is to m a k e use of the findings from a chart resembling Figure 9. 9 1 points 0.46 points 10. is lost in a lower expected final profit.66% -9. as suggested by the original unadjusted trade.16% 2.59% -8.28% -9.74% -7.98% 13.94% -8. as we do at bar 14.63 points at some future date. or a time-based stop. W h a t is gained in increased certainty regarding the individual outcome and lower standard deviation.4 Bars in trade.90% -7. then the regression line from the basic system with no stops added could work as a trailing stop level.43% 3.67% 23. For a trade that currently is not doing as well as the average trade.04% -9. 0 4 points . 3 4 points . as suggested by the least square regression line. but has not yet been stopped out for any other reasons.90% 24.10 points 19.59% 10.25% 14.20% -7.79 points 13.08 points 5. Finally.83% 26. for instance.4. 1 8 points . then you m u s t think about giving these stops ample leeway as well. Cutting all trades short lowers the total result over a longer string of trades. but only if the retracement also m e a n s that the trade has moved below the regression line.18 points 17. For instance.158 PART 3 Getting Out TABLE 9.17% 1.11% -8.16% 22. a combination of several techniques could be used that. open profits.1 . if the current trade is doing better than what could be expected.36% 11.24% -8.20% -8.35 points 2. stops out all trades that retrace m o r e than what is allowed by the trailing stop.91% -8.76 points 14.35 points 8.16 points 4.78% -7.86 points 15. As we can see.5 .55 points 17. there is no such thing as a free lunch at the markets. we could use the regular trailing stop.11% Difference . 5 8 points . Alternatively. If you also are using a stop loss.20% -7. If you do not.50 points 13. 0 4 points .75% 5.78% 25.66% 27.

1 .63% 31.66% 19.37% 22. Enter into a neutral position if the 12-bar average changes direction.96% 33.5 Bars in trade.42% 16.4 .74% 28. 5 3 points .59% 0. 3 3 points have you stopped out too soon during the initial stage of the trade.94% 17. 2 5 points .37% 34. 0 2 points .33% Open profit 6.83% 26.50% 42.3 .3 .18% 27.90% 24. 1 8 points .2 .31% 8. Exit with a profit if the trade lasts for 40 bars or m o r e .2 .90% 31. 2 4 points .5 .73% 24. 7 9 points . 7 9 points .CHAPTER 9 Sweeney's MAE/MFE 159 TABLE 9.98% 13.26% 38. 0 3 points .69% 14.2 . and differences for 16 markets in the directional slope system. 4 7 points .3 . 6 9 points .4 . Let us c o m p a r e the results for the original directional slope system with s o m e of the findings we have m a d e in this section. if the market moves the w r o n g way by 4% or m o r e and ends up below the regression line for the original average trade.84% 11.94% 22.3 .00% Difference .3 .08% 27.2 . w h e n it is the most vulnerable. Exit with a loss if the m a r k e t moves the w r o n g way by 4 % . 2 2 points .35% 12.30% 15.81% 42.25% 14.2 .69% 28.66% 26.66% 25. T h e rules for the modified syst e m a n d the corresponding TradeStation code are: • • • • Enter long/short w h e n the 18-bar average turns up/down.98% 44.67% 19. 3 5 points . Exit with a profit target if the market moves by 6 0 % or m o r e . 7 6 points . after bar 14.67% 23.16% 22.6 .4 . Exit with a trailing stop.96% 35. Exit with a small loss/breakeven if the market moves into (stays in) negative territory at or after bar 14. 3 3 points .36% 11.2 .84% 37. final profit open profits. 2 8 points .2 . • • • .2 . 3 1 points .4 .70% 25.97% 31. 4 3 points . 0 6 points . 3 2 points . Bars in Trade 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 Final profit 8.

there will be a few trades that we m u s t adjust manually within the spreadsheet program. End. BarsSinceEntry) * 1. therefore. BarsSinceEntry) * 0. We also must recalculate the cumulative profit.04 Stop. BarsSinceEntry) * 0. BarsSinceEntry) * 1. ExitLong at EntryPrice Stop.60 Limit.96 Stop. Obviously. ExitShort at EntryPrice Stop. If MarketPosition = — 1 Then Begin ExitShort at EntryPrice * 0.04 Stop. To this code. If BarsSinceEntry >= 40 Then ExitLong at Close. If BarsSinceEntry >= 40 Then ExitShort at Close. we have a s s u m e d that all losses larger than 5% are erroneous and consequently adjusted t h e m down to 4 % . Finally (because of another little TradeStation feature). we also m u s t attach the trade-by-trade export function from Part 1. End. and drawdown.0. If BarsSinceEntry >= 14 Then Begin If Highest(High. End. We m u s t use the trade-by-trade function because we are now dealing with ratio adjusted data and can.160 PART 3 Getting Out T h e TradeStation code for our additional exit techniques is: If MarketPosition = 1 Then Begin ExitLong at EntryPrice * 1. I am cheating a little bit here and obvious- .96 < EntryPrice * ( 1 + 0.04 > EntryPrice * ( 1 . which truly did have a loss greater than 4% because of legitimate opening gaps. We also run the risk of inadvertently changing the results to the better for a few trades. ExitShort at EntryPrice * 1.0074 * BarsSinceEntry) Then ExitLong at Highest(High.0074 * BarsSinceEntry) Then ExitShort at Lowest(Low. This happens w h e n the trade moves immediately and swiftly against us and.96 Stop. ExitLong at EntryPrice * 0. End. should have b e e n stopped out sooner than TradeStation allows. no longer rely on the performance summary from TradeStation. In this case. latest equity high. therefore.40 Limit. If BarsSinceEntry >= 14 Then Begin If Lowest(Low.

05 2.36% 10.18 -776.20 1. and then trans- TABLE 9. T h e largest loser would have b e e n 1 6 . To recalculate the drawdown.980. to fill all the cells in the c o l u m n .692. copper had a profit factor of 1.7 shows that with the modified system. we s u m m a r i z e the differences for each market.231.49% 12.54 -8.46 24.231 in today's m a r k e t value. To recalculate the cumulative profit in Excel.6 a n d 9.46% -3.CHAPTER 9 Sweeney's MAE/MFE 161 ly.254. 4 % .32% 128.46 Drawdown . To recalculate the equity top.60 286. In Table 9. Table 9. to fill all the cells in the c o l u m n . but there's not r o o m e n o u g h here to go into details. or $13. type into cell L 2 : =((1 + G2/100)/( 1 + H2/100) -1 )* 100 then drag d o w n to the b o t t o m of the sheet. you will n e e d to be m o r e careful in your o w n research.698. 0 2 % .0) In cell H 3 : =MAX(H2. Table 9. calculated on the average percentage w i n n e r and loser.54% 13.429. T h e largest w i n n e r would have b e e n 7 1 . before exits (original system) Total trades Profit factor Avg profit St Dev 1.6 shows that with the original system. Copper. type in the following formula on top of the values in the cumulative profits column: In cell G 2 : =M2 In cell G 3 : =((1+G2/100)*(1+F3/100)-1)*100 then drag d o w n to the b o t t o m of the sheet.8 and 9.67% 96 1. type: In cell H 2 : =MAX(G2. or —$3. and Tables 9.15 Losers Lrg loser Avg loser 59 -16.72% 38. 0 1 . 8 % .91 Winners Lrg winner Avg winner C u m profit 37 71. copper had a profit factor of 2 .692 in today's market value.9 with a composite of the m o s t important m e a s u r e m e n t s .6 Example of performance summary. to fill all the cells in the column.7 for all of the individual markets. W h e n you are d o n e exporting and calculating in Excel.G3) then drag d o w n to the b o t t o m of the sheet.10.60.05% 61.84% -4.4 3 . you can put together a set of tables resembling Tables 9.

89 -565.64 -11.26 -^.60 -37.653.57 Drawdown .991.390. Copper.21 239.68 -4.32 -5.04 -6.05 243.58 1.309.845. Before exits (original system) Market Corn S&P 500 Orange juice Live cattle Lumber Coffee Japanese yen Copper Gold Eurodollar Dollar index Cotton CRB index Crude oil Canada dollar T-bonds P.38 -6.72 Drawdown -3.88 1. dev.462. T h e largest loser would have been 4 .227.891.254.09 1.49 -22.14 Winners Lrg winner Avg winner C u m profit 46 60.41 3.50 1. trade 285.62% 363.843.29 3.413.90 1.604.46 1.7 31.97 5.007.40 1. A high profit factor.4 0 .14 681.90 Losers Lrg loser Avg loser 67 -4.19 19.66 3. 8 8 % formed into today's market value.00 194.59 1.79 286.11 . From Table 9.680.285.772.00 3.31 4.826. in today's market.22 2.126.47 0. 7 % (thanks to the stop loss).20 2.28 -8.502.95% 59. after exits (modified system) Total trades Profit factor Avg profit St Dev 1.33 584.859.158.83 3.540.7 Individual market performance summary.8 we can see that all markets but one had a profit factor above 1 in the original system. or $11.33 816.28 -38.31 521.92 19.34 2.944.96 5.40 10.996.004.82 4.95 37.436.72 7. or —$900 in today's market value.63 149.43 1.15 7.00 -390.162 PART 3 Getting Out TABLE 9. and deduct a proper amount for slippage and c o m m i s s i o n .019.16 -14.01 1.51 1.72 -7.71% 11. we m u s t also look at the value of the average trade.671.29 7.797.8 Performance measurements for all markets with the original system. factor 2.96% 113 2.331.839.76% 9.513 in today's market value.174.63 218. T h e largest winner would have been 6 0 % (thanks to the profit target).60 1.69% -2.17 69.34 Avg. TABLE 9. 62.46 -9. however.24 2 St.254.00% 8.443.93 -136.375.64% 40. A rule of t h u m b could be that the remaining value of the average trade should be at least two times the estimated slippage a n d commission.512. For this conclusion.25 -8.01 337.70 2.383.29% -899. does not necessarily mean that the market in question is worth trading.84 2.74 2.617.135.953.11 -7.

54 1.843. after the exits w e r e implemented.363.02 0. but only three had a better average trade.38 26.04 -28.44 917. which suggests that the system has b e c o m e m o r e stable a n d less risky to trade. consequently.12 -3.04 1.802.57 -10.15 Drawdown -2. After exits (modified system) Market Corn S&P 500 Orange juice Live cattle Lumber Coffee Japanese yen Copper Gold Eurodollar Dollar index Cotton CRB index Crude oil Canada dollar T-bonds P.036.550.28 Avg. it should be c o m p a r e d to the standard deviation of the o u t c o m e . are S & P 500.8. Of these five markets. A l s o .80 to m a k e r o o m for a decrease in performance and so that the system will continue to be profitable in case the m a r k e t declines and.913.21 3.875.707.14 -33.289.01 1.90 248.461.08 4.28 -6.28 3.822.62 0. respectively.95 10. With a slippage and c o m m i s s i o n of $ 7 5 . in this case.9 contain 11 and 12 markets.07 238.78 -5. M o s t remarkable is that orange j u i c e goes from being a non-tradable market to a tradable one.000. From Table 9. Of the remaining 11 markets that were tradable with the original system.00 399.94 758.544. six h a d a better profit factor with the modified system.427.83 -8.06 2.898. the cost of doing business will also be m u c h less.68 -8.21 1.204.98 337.8 and 9.208.74 2.67 2.10 we also can see that out of the 12 tradable markets with the modified system (including orange juice). three also h a d a lower profit factor a n d average trade. that are w o r t h trading.78 829. In essence.938.669.282.32 7.346.38 -15. but at the same time.42 2 St. factor 2.08 2.438.90 6.72 -1.373.17 -177.95 11.32 -6. Tables 9. trade 266. orange juice.215.25 9.97 2.00 2.67 2. and C a n a d a dollar.551.21 93.36 1.965. just to look at the average trade by itself is not always that useful.78 4.71 7.08 5.80 -7. seven (corn. live cattle.87 -5.384. lumber.982.9 Performance measurements for all markets.75 1.991.43 -6.21 116.742.38 2.59 234.48 -11. dev. . the value of the average trade b e c o m e s lower as well.416.77 -258.72 2.947. in Table 9.271.CHAPTER 9 Sweeney's MAE/MFE 163 TABLE 9. orange j u i c e . Instead. In most cases the drawdown also has decreased dramatically.42 462.63 3. 2. T h e five m a r k e t s not worth trading.23 2. this m e a n s that we will not m a k e as m u c h m o n e y per contract traded with the modified system as with the original system. C R B index.82 2.200. 15 out of 16 markets h a d a lower standard deviation.882. copper.

dev.12% -40. Differences Market Corn S&P 500 Orange juice Live cattle Lumber Coffee Japanese yen Copper Gold Eurodollar Dollar index Cotton CRB index Crude oil Canada dollar T-bonds Better P. orange juice.04% 42. a better way could be to examine only one exit technique at a time and then rerun the system before each new implementation. please r e m e m b e r that this is a sample system only and that we used all the exit techniques previously discussed.11% -18. in accordance with a fixed fractional money management strategy.99% 25. has decreased for eight (corn.06% -36. copper.37% 4 2 St.11% -8.25% 7. Finally.48% -16. trade -6.19% 11.68% -16.03% -29.78% 10. T-bonds) out of the 12 worth trading.54% -15.35% 29. Or. coffee.43% -24.00% -0.09% -7.15% -24.30% -23.66% -6257. before/after exits.02% 28. and it might be better. while at the same time we can feel more sure about the outcome and sleep better thanks to a lower expected drawdown.42% -42.21% 18.66% -20. Instead.45% 7.83% -4.59% 4.68% 0.39% -48. This might not be the most optimal solution.48% -31.86% 29.35% -7. with the modified system the cost of doing business has decreased. In this case (and all the following). This is thanks to the fact that a m o r e stable system allows us to trade more aggressively with more contracts. we still should be able to make more money over any given period of time. T h e drawback is that the value of the average trade has decreased. crude oil and T-bonds) also show a positive relationship when it comes to how m u c h the average trade has changed in relation to the standard deviation. Japanese yen. but because of all the benefits.79% 10 Better 3 1 4 2 3 3 2 4 3 2 1 1 3 3 1 2 — Eurodollar.70% 8.43% -40. That is. this is not necessarily the best way to do it.35% -21. factor 7.59% -10.27% 61.63% -11. crude oil. -9. set up the modified system as we have done . finally.26% -17.94% -10.47% -10.58% 1.98% 8.94% -17. Because the statistical traits for the system will change with every change we m a k e .75% -46.68% -4.19% -14.00% -0.48% 71.13% 2.46% 15 Drawdown -33. lumber.13% -37.71% 133. excluding one or several of the exits.05% -33.53% -24.87% -13.40% -32.11% 9 Avg. I added all the exits directly on top of the results from the original system.05% -2. The drawdown.78% -9.69% -30.92% -33.164 PART 3 Getting Out TABLE 9.10 Summarized differences.

. A l s o . in this and the other long-term systems to follow. modifying or excluding one exit type at a time. because of a variety of u n w a n t e d TradeStation "features" I have had to m a k e a few modifications to the TradeStation c o d e and simplifications in the Excel spreadsheet that might have diluted the results somewhat.CHAPTER 9 Sweeney's MAE/MFE 165 here a n d work b a c k w a r d s from there.

With the M A E / M F E technique it also is possible to add to trades that show tendencies of being better than w h a t could be expected. I have c h o sen not to look into this technique in this book. w h i c h is a very specific request 167 . As I have stressed throughout this b o o k . In doing so.CHAPTER 10 Adding Long-term Exits There are probably several w a y s of locking in a profit or taking a loss. you want it to be a long-term trend-following system. For instance. you must treat it as such a n d allow p r o p e r leeway for each trade. However. first we w a n t the system to trigger a trade in one direction. Therefore. I am painfully aware that this m i g h t not be the optimal solution. although it h a s its merits. E v e n this m e t h o d can p r o d u c e catastrophic results if not u n d e r s t o o d and i m p l e m e n t e d properly. call me a b a d syst e m s developer. A n o t h e r possibility I have not explored is to fade the original system if the trade moves against us by a certain amount. T h e important thing to r e m e m b e r is w h a t it is you w a n t your system to d o . one of the m o s t important things you m u s t do is m a k e sure that you understand exactly w h a t it is you are m e a s u r i n g a n d h o w to go about doing it. so that you do not inadvertently c h a n g e it into a shorter-term system. In the following chapters. I will r u n d o w n w h a t we have learned and i m p l e m e n t it on the other two long-term systems with w h i c h we have worked so far. b u t one technique that has great scientific value is J o h n Sweeney's m a x i m u m adverse ( M A E ) and m a x i m u m favorable excursion ( M F E ) . and you m i g h t be better off excluding one or several of the exits we discuss here. This b e c o m e s apparent if you formulate the rules in plain English. but I have yet to c o m e across such a system that is capable of delivering results and is thoroughly robust and stable. I believe that such a technique. starts to c o m e awfully close to curve fitting. If for instance. if you like.

w h i c h are two other very specific requests. the greater the M A E f e value. if you so wish).3 shows that. and the least square regression line from Figure 1 0 . In this case. m a k e s it m o r e difficult to enter the market. T h e n we want the market to move against us by a certain amount within a limited t i m e frame. To find the m o s t logical level for w h e r e to place a stop loss or trailing stop you can put together a set of tables resembling Tables 10.1 shows the average open profit for all trades in relation to the n u m b e r of bars in trade. which h a d a "reversed" volatility relationship for the exits. From this chart we can see that the average profit is expected to increase with about 0 . and buts for my comfort. In fact. w h e n the historical volatility increased. In this chapter we apply that system (version l b ) to the s a m e 16 markets as in Part 2. the less we benefit . basically the s a m e reasoning goes for the trailing stop as for the stop loss. F u r t h e r m o r e . as an indication of higher risk.2. Because there seem to be no local m i n i m a . for m u c h easier interpretation and rule making.1 and 10. then a higher volatility also should m a k e it m o r e difficult to stay in (or easier to get out. indicating s o m e kind of market inefficiency that we could have tried to exploit. T h e export function from TradeStation into Excel is essentially the s a m e for the D B S system as for the directional slope system.168 PART 3 Getting Out in the first place. Table 10. ifs. we are n o w down to four. A n o t h e r reason I chose to work with this system is simply that the original system is closer in resemblance to the standard deviation breakout system that follows. That is. THE DYNAMIC BREAKOUT SYSTEM In Part 2 we discussed several different versions of the D y n a m i c Breakout System (the D B S system) before we settled for version l b . A little too m a n y rules. Figure 10. the lookback period for the exit decreased. we would have benefited from placing a stop loss at j u s t about any level. the trailing stop in accordance with the M A E f e . As you can see from Table 10. Instead of looking at h u n d r e d s of lines that m a k e little sense. T h e first things we will examine are the stop loss in accordance with the M A E . we might even add another criterion or two that might stipulate that the m a r k e t is allowed or not allowed to behave in a certain fashion. we simply m u s t place the stop w h e r e it m a k e s m o s t sense. T h e t i m e period covered is from January 1980 to October 1999. 2 3 % per bar. the m o r e we would gain h a d we decided to place a stop at a specific level. 1 . for this version of the D B S system. the greater the M A E . to examine the possibility of increasing its performance by adding a few stops and exits in accordance with our findings regarding John Sweeney's M A E / M F E methods. ands. to take all the other exits a n d risk levels into consideration. T h e underlying reasoning w a s that if a higher volatility.1. which is too few to m a k e any meaningful statistical conclusions. To the right of the dotted line we have fewer than 20 trades that remain open.

67% -5.50% -1.50% -3. original profits.43% ^.29 -1.93 -1.1 MAE levels.2 . 3 5 points -0.50% -8.00 points points points points points points points points.17 -1.50% -5.56% Difference .50% -7.08% -11.2 . 4 2 points . points points points points .76% -8.00% -2.58 -1.0 .00% -9.1 Tthe average open profit for all trades in relation to the number of bars In trade.58% -7.00% -7.96 -2.65% -10.50% -10.39 -1.29% -5.20% ^.35% -1.50% -2.20 -1.00% -9. MAE -0.00% -8.00% -5.21 -2.CHAPTER 10 Adding Long-term Exits 169 FIGURE 10.46% -9.08% -10.39% -11. and differences in outcome for 16 markets in the DBS system.50% ^1.71% -12.00% Original profit -0. 1 5 points .76 -1.98% -2.00% -6.92% -1. 0 8 points -1.00% -1. TABLE 10.00% -4.89 -2.56 points points points points .0 .89% -6.00% -3.64% -3.08 -2.64 -0.27% -7.50% -9.50% -6.77 -1.48 -0.

02% 8. you simply must place the stop where it m a k e s the m o s t sense to you.49% 12.00% -11.01% 11.3.69% 0.70% 3.59% 11.76% 1.49% 13. which says that if the open profit for a trade stays or turns negative at or after bar 30.29% Profits 4.58% from a stop at a specific level.00% -6.1.50% -5.00% -3.11% 12. Exactly h o w deep this retracement m u s t be before we exit we d o n ' t know.53% 14. A n o t h e r way of analyzing the expected profits is to look at the average open profit at a specific bar a n d then deduct the expected E T D for trades lasting longer. This is shown in Table 10.00% -7.08% 7. average MFE levels. so again. No local m a x i m a indicate any market inefficiency that we could have tried to exploit. T h e trailing stop c o m e s into play at the same time as the time-based breakeven stop and only kicks in if the decrease in open profit also takes us below the regression line from Figure 10.49% 10. using the s a m e b a c k w a r d s reasoning as we did for the stop . MFE 6.46% 1.83% 12.98% 9. but we do k n o w that as long as the retracement does not bring us below the regression line.52% 2.42% 3.4 shows that we would n o t have benefited from placing a profit target anywhere. For the sake of simplicity. With the information from Figure 10.50% -8.68% 7.24% 0.75% 3. However.00% -9.11% 1.53% 8.170 PART 3 Getting Out TABLE 10.2 MAEfe levels.50% -4. we exit with a market order at the close.13% 2.50% -11.75% 10.00% Avg.07% 10.17% 2. the trade is still doing better than average and could be worth staying in.75% 2.50% -7.50% -10.00% -4.3.13% 13.68% 0. then.47% 0.65% 3. MAEfe -2.84% 11. Table 10.87% 3.91% 2.1 and Table 10.44% 9.51% 8.00% -5. we place a stop loss and a trailing stop at 4 % .54% 3. which shows that final profit is not expected to turn positive until after b a r 30.50% -3.50% -6. and theoretical profits for 16 markets in the DBS system.50% -12.02% 3.00% -8.00% -10. we can formulate our third exit rule.50% -9.

65% 0.36% -5. 7 1 points . the " b e s t " result s e e m s to be around the 35 to 4 0 % area.13% 6.59% -5. T h e p u r p o s e of this is to streamline all trades as m u c h as possible and keep the standard deviation as low as possible.2 .83 points 2. what we lose with the profit target stop.90 points 1.36% -5.38% 0. U s i n g the s a m e reasoning as for the profit target.06 points 0.48% -5.CHAPTER 10 Adding Long-term Exits 171 TABLE 10. Bars in trade 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 Final profit 0. which.52% -5.28% -5. 7 5 points . we h o p e to regain from m o r e aggressive trading.31% -5.15% 5. 5 2 points . 2 3 points 0.33% -5. 0 8 points . we place it at bar 100.31 points loss and profit target in the directional slope system.34% 1.2 . as soon we reach an open profit of 7 0 % or m o r e .61 points 0. Table 10.50% 7.5 . R e m e m b e r . 0 7 points . Just as w a s the case for the profit target. 3 5 points .4 . A n o t h e r way to achieve parts of the s a m e effect is to apply a time-based stop.4 .44 points 1. a n d not to place the stop too close to the entry point. By keeping the standard deviation down. 6 1 points .69% 3. 9 7 points .72% 2.58% -5.21% 4.0 .67% -5. . open profits. 0 7 points .61% 5.5 shows that the D B S system s e e m s to have two different trade lengths at which this would be the least opportune m o m e n t .30% -5.3 . we are decreasing the overall risk of the system and should be able to trade it m o r e aggressively with m o r e contracts within a fixed fractional m o n e y m a n a g e m e n t strategy.28% -5.14% 2. and differences for 16 markets in the DBS system.61% Difference .1 .1 . 5 9 points .31% -5.0 . 1 3 points .33% -5. T h e first o n e c o m e s after about 50 bars and the second after about 100 bars. again indicates that there is some sort of s y m m e t r y in h o w the markets behave. ETD.32% -5.25% 3.3 Bars in trade.92% Open profit -5.27% -5.75% 4.26% 0. we exit with a limit order.02% 7.38% -5. 1 9 points . 9 4 points .3 .97% 1.48% 7.54% 6. Because this s e e m s to be a little too close to the entry.12% 0.5 . Therefore.31% -5. That is.3 .4 .30% -5. by the way. we will instead choose to go with the second "best" level at around 7 0 % .

00% 55.00% 15.85% 78.00% 20. if the m a r k e t moves the w r o n g way by 4% or m o r e and ends up below the regression line for the original average trade.96% 55.30 points 6.07 points 0.07% 150. just to c o m e up with s o m e motivation to use these stops at all.24 points 1.08 points 9. Difference 1.30% 51.00% 25.07% 95.07% 95. • • W h e n you are done exporting and calculating in Excel.00% 100.61" points 1.54 points 8.24% 95.86 points 3.20 points 5. the suggested exit levels for the D B S system are as follows: • • • Exit with a loss if the market moves the w r o n g way by 4 % .62 points 6. Hence.00% 75. finally we summarize the differences for each market. that this is very b a c k w a r d reasoning and for demonstration purposes only.27 points 2.54% 48. Exit with a limit order if the open profit exceeds 7 0 % . you can put together a set of tables like Tables 10.20% 35.08% 44.00% 60. Exit with a small loss/breakeven if the m a r k e t moves into (stays in) negative territory at or after bar 30.96 points 1.27% 22.85 points 8.07 points 50. and Tables 10. original profits.8 and 10.77 points 3.77% 53. In Table 10.00% 45.10.62% 81.00% 30.00% 95.96 points 0.9 with a composite of the most important measurements.62% however.6 and 10.00% 65.61% 16. after bar 30.00% 85.00% 50.6 shows that with the .172 PART 3 Getting Out TABLE 10. Tabic 10.00% 90. Exit with a profit if the trade lasts for m o r e than 100 bars.62 points WIFE 5.24 points 10.24% 81. and differences for 16 markets in the DBS system.14 points 1.49 points 5.00% 10.00% 80.00% 70.86% 28. Exit with a trailing stop.4 MFE levels.00% 35.07 points 5.96% 61.14% 11.00% 40.49% 70.00% Original profit 6.7 for all of the individual markets.

1 .0 .263 at today's market value.03 points . 2 1 .90 points 0. 5 5 points .35% 22.07 points .58 points 0. 4 9 points .25% 23.29% 22.36% 24.67% 24.1 . and differences for 16 markets in the DBS system.54% 6.83 5. 7 3 points . 0 5 points .25 -2.12% 21. or .87% 53. and then transformed into today's market value.09 299.91% 21.72% 12.52% 18.147.12% 17. final profit. The largest winner would have been 2 1 .43% 16. T h e largest winner would have been 17.97% 9.60% 46.41 points .70% Difference .74% 11.12% 11.74% 7.10 points 0.CHAPTER 10 Adding Long-term Exits 173 TABLE 10.06% 28. 0 7 points .36% 13.20%. TABLE 10. Bars in trade 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100 105 110 115 120 125 Final profit 5.0 .3 .0 .05% 28.0 .00 Drawdown .29% 19. 5 % . or $25.262.29 points 1.60 points 2.1 . Japanese yen h a d a profit factor of 2 .84% 12.6 Individual market performance summary.00.43% 12. 0 5 points 0.00 1. 6 6 points original system.95% 16.38% 11. 3 1 points 0.57% -6. Japanese yen had a profit factor of 2.63 points 0. 7 5 points .1 2 .709.56% 6.42% 14.00% 4. before exits (original system) Total trades Profit factor Avg profit St Dev 1.27 -14.43% 25.50 5.50% 4.0 . Table 10. T h e largest loser would have been 5 .173. T h e largest loser would have been 4.61 points 0.56% 17.$ 6 . or $20.62% 10.36% Open profit 5.73% 11.70% 7. 7 1 % .71% -1.92% 9.00 Losers Lrg loser Avg loser 75 -5.91% 13.7 shows that with the modified system. 0 4 % .66% 18.060.178. or —$4. calculated on the average percentage winner and loser.41% 140 2.76% 27.155.14% 26.32% 12. open profits.56% 12.668 at today's market value.5 Bars in trade.194.0 .935 at today's market value. 0 2 points 0. Japanese yen.31% 254.98 Winners Lrg winner Avg winner C u m profit 65 21.91% 14. 7 0 9 at today's market value.6%.

935 -1.15 1.80 1.902 -13.13% 143 2.576. To stay on the safe side and to m a k e r o o m for future changes in the value of the average trade. the value of the average trade also must be high enough.157.58 5. the D B S system has several markets with considerably lower profit factors and several m a r k e t s with an average trade too low to m a k e it worthwhile trading after we have deducted the necessary a m o u n t for slippage and c o m m i s s i o n .27 0.85% -4.82 178.236 4.8 Performance measurements for all markets.92 10.78 -9.05% 4.446.81 -517.46 29.80 Drawdown -2.96 173.898.904.70 509.159 Losers Lrg loser Avg loser 77 -4.319. trade 113.08 -14.72 -7.57 2.898 351.44 2.45 -16.78 737.701 Drawdown . copper.17 -9.974.06 1.74 -9.90 2 St.86 5.137.86% 46.50 1.022.59 -9.24 26. Before exits (original system) Market Corn S&P 500 Orange juice Live cattle Lumber Coffee Japanese yen Copper Gold Eurodollar Dollar index Cotton CRB index Crude oil Canada dollar T-bonds P.11 Avg.549.8 we can see that c o m p a r e d to the directional slope system.07 1.23 55.86 161.53 5. however.98 2.87 226. 1.1 1 .62% 53.34 1. For a system to be profitable on a specific market.147.220.027.88 569. factor 1.10 -166.549.91 1. after exits (modified system) Total trades Profit factor Avg profit St Dev 1.09 1.79 5.76 3.03 8.21 1.80 28.357.02 3.990.865.15% 20.075.83 -46.50 -14.04 -56.58 2.388.59 1.095.04 1.34 -1. and CRB index) h a d a profit factor above one.281.00 -30.7 Individual market performance summary.159.057.20% -1.00 -12.00 0.08 .705.61 2.29 174.668 4.972.59 -10.174 PART 3 Getting Out TABLE 10.217.59% 4.64 5.848 Winners Lrg winner Avg winner Cum profit 66 17.20 -8.72 11. all but three ( S & P 500.470.282.80 -16. Japanese yen.230.455. which is dependent on the current level at which the m a r k e t is trading and TABLE 10. Of all 16 m a r k e t s traded with the original system.032.51 0.17% 298.32 1.84 1.931. 6 6 % F r o m Table 10. dev.173.808.

938.123.202.38 0.48 -10.843.504.733.662.24 Avg.83 1.81 -8.71 248.23 1.13 -502.49 -10.02 1.19 131.65 1. should be two times the expected costs.18 8.13 1.58 5.131.08 332.688.964.58 3. as well as smoother.17 3.289.442.14 141.53 -12. TABLE 10.46 -56.431. dev.29 10.696.12 1. the gross value of the average trade should be $ 2 2 5 .9 shows that after we have added the stops. For instance.90 -8.CHAPTER 10 Adding Long-term Exits 175 the point value of the market.24 .76 -886.578.236. trade 82. thus.018.13 -25. W h e n tested on historical c o m m o d i t y futures data.97 544.848.42 2.bonds P.635.01 2.45 1.65 26.14 -8.556.34 3.35 -167.72 40.035.738. w h e n looking at the standard deviation of the trades and the drawdown.60 2.49 1.08 -13.841.873. a rule of t h u m b is that the value of the average trade. a n d p r e cious metals.91 2 St. factor 1. U s i n g this rule of t h u m b . with a p r o p e r m o n e y m a n a g e m e n t strategy: it m i g h t be uncorrelated e n o u g h with all or s o m e of the other m a r k e t s to p e r f o r m satisfactorily w h e n the other m a r k e t s are p e r f o r m i n g badly.249.11 -25. the changes in the syst e m are not that large in the profit factor and average trade.581.61 194.98 -10.95 -23. 1.819. it is w o r t h m e n t i o n i n g that j u s t b e c a u s e a m a r k e t s e e m s not to be profitable in itself.96 2.54 74.20 4.25 -7.04 -7.700.087. we can see that the improvements for the portfolio as a whole are quite dramatic.90 9. if you expect slippage a n d c o m m i s s i o n s to be $ 7 5 .21 0. Table 10.48 5. it still could be w o r t h trading in a portfolio of m a r kets.18 1.684.150.24 0. k e e p i n g the equity level u p .62 7.05 Drawdown -3.91 -7.93 702. this version of the D B S system holds five markets that are worth trading both with the original system and w i t h the modified version. net after expected costs. such t e n d e n c i e s can m a n y t i m e s be found in such m a r k e t s as short-term interest rates. for the future w h e n the other m a r k e t s start to p e r f o r m better. live cattle. However.10 1.69 2.53 2.203.94 1.84 4. lumber.862. On a separate note.9 Performance measurements for all markets.50 2. After exits (modified system) Market Corn S&P 500 Orange juice Live cattle Lumber Coffee Japanese yen Copper Gold Eurodollar Dollar index Cotton CRB index Crude oil Canada dollar T.

which should give the individual trades a little m o r e leeway. A l t h o u g h this system overall seems to be less profitable than the directional slope system. trade -26. we can see that the exits are still doing their j o b w h e n it comes to streamlining the trades.15% -16.37% 14 Drawdown 11. Differences Market Corn S&P 500 Orange juice Live cattle Lumber Coffee Japanese yen Copper Gold Eurodollar Dollar index Cotton CRB index Crude oil Canada dollar T-bonds Better P.50% -28. while as m a n y as nine m a n a g e d to improve at least three categories.81% 1. indicating that the system has b e c o m e m o r e stable and easier to trade.01% 10 Better 1 2 2 1 4 1 4 3 4 3 3 2 1 3 3 4 — Table 10.39% -16.76% -3.19% 1. This is all g o o d news.80% 40.70% -58.01% -25.02% -4. Of all m a r k e t s tested. while a total of 14 also m a n a g e d to end up with lower standard deviations.10 shows that of all 16 markets examined.71% 0.01% -9. This m e a n s that the cost of doing business has decreased considerably.65% 10.63% 6 2 St.77% -4.47% -0.81% 16. This is further confirmed by the profit factor.04% -4.48% -27. -7.16% 3.83% -26.91% 5.98% -27.29% 5. Other things that could increase performance might be to change b a c k to the original D B S system (version l a ) .86% -17.60% -8. factor -8.81% -21.03% -6. .58% -1.22% -6.46% 7. we m a n a g e d to lower the d r a w d o w n for 10 and lower the standard deviation of the trades for 14.96% 0.17% -28. A l s o .14% 2.61% 5. 6 ended up with a higher average trade after the exits were added.85% -13.90% 11 Avg.56% 0. only four increased their performance in only one category.33% 0.42% -18. which has increased for a total of 11 markets after the exits have been added.10% -6.37% -2.38% -97.68% 11.31% 4. and/or continue to experim e n t with different types of stops other than the profit target stop and the timeb a s e d stop. before/after exits.84% -14. w h i c h should m a k e it possible to trade it m o r e aggressively with the proper m o n e y m a n a g e m e n t .34% 9.28% -4.176 PART 3 Getting Out TABLE 10.25% -71.81% -2.88% -4.28% -22. and although the average trade did not improve as m u c h as we would have liked.02% -15.23% 33.79% -29.90% 21.10 Summarized differences.40% 38. dev.90% 105.31% -44. of all markets.67% -2.34% 11.41% 3.

w h i c h might start m o v i n g away from the market somewhat before you are allowed to exit. w h e n we put together a trailing stop.2 shows that the o p e n equity for the average trade is exceptionally smooth all the way up to over 150 bars. the Standard Deviation Breakout ( S D B ) system is very similar to the original D B S system (version l a ) . There w a s no optimization influencing this decision. but despite the similarities it also has a few specific features that m a k e s it very suitable to form the base in a portfolio of systems. This is. it often ends with close to 7 0 % profitable m o n t h s a n d fairly short flat times (time spent between two equity highs). O n e major feature of this system is that it is incredibly robust and trades profitably on a w i d e variety of markets. copper. N i k k e i index. however. of lesser concern for a professional trader. We will m a k e use of this finding later. cotton. coffee. I settled for a system with a lookback period of 60 bars. a n d wheat. 1 8 % per bar. For our p u r p o s e s . r o u g h rice. it traded profitably on all 16 markets. T-bonds.CHAPTER 10 Adding Long-term Exits 177 THE STANDARD DEVIATION BREAKOUT SYSTEM Let us do one m o r e long-term system in this meticulous m a n n e r before we m o v e on to the shorter-term systems and yet another way of examining h o w best to exit a trade. and that we can expect the average trade to increase its o p e n profit with about 0 . A n d because the basic d y n a m i c breakout system usually does a g o o d j o b in letting the profits r u n while cutting the losses short. natural gas. D m a r k (as a p r o x y for the E u r o ) . almost regardless of the lookback period for the indicators and breakout levels. . there are a few trades that go on for m u c h longer than that. T h e 16 m a r kets traded were crude oil. or as a starting point for diversifying out into other types of systems. gold. As we mentioned in a previous chapter. as indicated by the system at that point. This w a s for no particular reason other than that I k n o w it works fairly well. W h e n traded on 16 different markets. T h e data for the period N o v e m b e r 1992 to October 1999 were saved for some out-of-sample testing in later sections. lumber. This h a p p e n s at b a r 132. on data from J a n u a r y 1980 to October 1992. Japanese yen. you never can be completely sure about h o w m u c h to risk a n d h o w m a n y contracts to trade in a fixed fractional m o n e y m a n a g e m e n t strategy. dollar index. T h e dotted line denotes the n u m b e r of bars from w h i c h point on there are fewer than 20 open trades. A less than ideal feature is a relatively low n u m ber of profitable trades. live cattle. T h e export function from TradeStation into Excel is essentially the s a m e for the S D B system as for the directional slope syst e m . using the R A D contract. T-bills. Figure 10. because the m a x i m u m loss. such as short-term systems that trade both with and against the prevailing trend. O n e disadvantage with this system is that it b a s e s its exits on a moving average. This is especially important to r e m e m b e r at the time of the entry. might not be the s a m e as after a few days. As you can see. w h o usually is m o r e interested in the n u m b e r of profitable m o n t h s for a system traded on a portfolio of markets.

5 % stop. Furthermore. it is very important to examine the system for any natural stop loss levels based on percentage data. hoping that what we might lose in a few individual trades on a one-contract basis we will win b a c k with the proper m o n e y management.2 The average profit for all open trades in relation to the number of bars in trade. would then be to place the stop at 2 % . seems to form a local m a x i m i u m . the most obvious solution doesn't necessarily have to be the best one in the long run. on the other hand. wanted a n d unwanted features. T h e natural choice. in accordance with our previous discussion about everything in trading being a trade-off between good and bad. 5 % M A E level. 5 % . Therefore. the larger the e x p e c t e d profit.12 s h o w s t h a t the larger t h e M A E f e .5% would have ended with an average profit of — 1. The 5 . what we lose on the swings. Table 10. Obviously. had we decided to put a stop in at this level. the average profit would have been — 5 . the overall decrease in performance would have been 0 .178 PART 3 Getting Out rluURc 10. we might m a k e up on the merry-go-round. with higher values on each side of it.90%. Table 10.11 shows that all trades with an M A E equal to or greater than 5. It also seems that the 2% M A E forms s o m e sort of local m i n i m i u m . simply because this seems to be the most obvious place for w h e r e not place a stop. we decide to go with the 5 . For all trades with an M A E of 2 % . which would have resulted in an overall decrease of performance of 3. w h i c h is t h e last retrace- . a s s u m i n g that the M A E f e c o r r e s p o n d s to the E T D . and definitely one that should be investigated further.60%) points. However. 9 9 % points for these trades. with lower values on each side of it.

60 points 2.15% -3. give back 7. but before a trade that goes on for this long is allowed to exit it will.50% -9. B e c a u s e o f this w e p r o b a b l y w o u l d b e better off w i t h o u t a trailing stop. p r o v i d e d that a r e t r a c e m e n t this large also results in a d o w n s i d e b r e a k t h r o u g h of t h e r e g r e s s i o n line for the average trade.50% -5.84 points 2. Thus not until this b a r can we expect the average trade to e n d with a profit. Table 10.01% -1.00% -1.50% -1. we do not w a n t our trades to go on for too long either.77 points 3.85 points 4.60 points 2. It j u s t so h a p p e n s that o n e unsatisfactory feature of the S D B system is its tendency to give back a . At the other end of the spectrum.74%.89% -2.33 points 1.38% -2. W h e n it c o m e s to the time-based stop. on average.00% Original profit -1.52 points 1.65% -12.86 points 2. the larger the resulting retracement is likely to be as well.02% -7.52% -1.52% -0. 5 7 points m e n t d o w n from t h e M F E level.00% -4.05 points 2.17% -1.99 points 1.77 points 3.00% -2. we d e c i d e to p l a c e o n e a n y w a y at the 5 . original profits.00% -5.2 .85 points .63. MAE -0.40% -2.00% -9. 5 % level.50% -7. for all trades that go on for m o r e than 40 bars.45% -2.00% -3.24% -1. we will cut all trades that are not profitable at this point on the assumption that they will only turn even worse because of the E T D that is still likely to come. a trade m u s t accumulate a certain open profit before it can be expected to close with a profit.97 points 0.48 points 1. because the larger the o p e n profit.00% -7.50% -2.13 shows that the average profit on b a r 40.61 points 1.62 points 3.CHAPTER 10 Adding Long-term Exits 179 TABLE 10. is 7. resulting in a less-than-stellar final profit.64% -4.27 points 0.73% -4.50% -10.23% -1.00% -6.50% -3. but b e c a u s e we w a n t to m a k e u s e of o u r t e c h n i q u e s .50% -6.00% -0.53% -8.00% -8.16% -4. and differences in outcome for 16 markets in the SDB system.57% Difference 1.98 points 1.11 MAE levels. Because of the expected E T D .50% -4.50% -8.50 points 1.90% -3. Therefore.

65% 15. . average MFE levels. so just for the heck of it we place it at 7 5 % .94% 9.39% 26.42% 17.71% 11. W h e n you put together your own systems. Avg. T h i s can be seen from Table 10.00% -11.50% -7.00% -8.58 percentage points.96% 10. I am merely pointing t h e m out and using t h e m as examples for you to learn from.50% -3. one would think.12% 27.68% 13.50% -8. although we are dealing with two trend-following strategies.16% 9. and theoretical profits for 16 markets in the SDB system.77% 7.61% 7.) A l s o .05% Profits 7.21% 7. we seem to be better off no matter w h e r e we place a profit target.15% 7.12 MAEfe levels.11% 11.01% 8.35% 18.52% 16. but do better with a profit target—the exact opposite of what we found for the D B S system. which should result in a performance increase of about 25.06% 20.50% 22.00% -7.00% -3. N o b o d y else can m a k e these decisions for you. are likely to p r o d u c e only an average profit of 4 5 .14 which.180 PART 3 Getting Out TABLE 10. notice the differences between this system and the not too different D B S system in the previous chapter.15% 24.22% 10.50% -10. MFE 9.77% 8.00% -4.50% -11.50% -12. For the S D B system.58% 10. we seemingly do worse using a stop loss and a trailing stop.81% 21.50% -4. y o u must think about all these subtleties and m a k e sure that they w o r k the way y o u want t h e m to and in a way that m a k e sense to you.00% -9.00% -6.85% 9.00% -5.80% 7.44% 14. With the S D B system.80% MAEfe -2.32% 7. There seems to be no local m a x i m a or m i n i m a on w h i c h to base our reasoning (backwards or otherwise).65% 20. shows that all trades with an M F E of 6 0 % or m o r e .62% 11.22% 7.96% 10. for instance. 6 3 % . ( R e m e m b e r that these systems are for demonstration p u r p o s e s only.50% -9. which.27% 8.86% 23.55% 9.00% substantial part of the o p e n profit before it signals time to exit. would act pretty m u c h the same.07% 14.79% 12.47% 11.00% -10.50% -5.50% -6.

Table 10.63% -7.30 points 1.10% -7.1 . To c o m e to a c o n c l u s i o n for this we can c o m p a r e the average o p e n profit for a specific b a r with the average final profit for all t r a d e s that last longer. but also b u i l d s up an o p e n profit that it is likely to give b a c k a substantial part of before the s y s t e m signals an exit.89% 5.83% -7. so we d e c i d e to go w i t h a m a x i m u m t r a d e length of 105 days.49% 4. the exit techniques for the S D B system are as follows: • • Exit with a loss if the market moves the w r o n g way by 5 .CHAPTER 10 Adding Long-term Exits 181 TABLE 10. .3 .86% 8. In this case.61% -7.42% 2.91 points .62% -7.11 points 0.83% 9. Exit with a small loss/breakeven if the market moves into (stays in) negative territory at or after bar 4 0 . 8 7 points . 7 2 points .91% -6.53% -7.56% 2.74% 7.01% 2. it could be a g o o d idea to cut all profitable t r a d e s that go on p a s t a certain n u m b e r of b a r s .4 .5 .60% -7. 3 0 points .4 .24% 7.2 . 4 0 points 0.02% -7.55 points 1.23% 3.97% -7.2 .63% -7. 5 % . 5 5 points .88% 7.92% -6. Bars in trade 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 Open profit 1.60% -7. no m a t t e r what.70% Difference .81% 4.47% -7.1 .07 points 1. 3 8 points .98 points W h e n it c o m e s to n o t letting a w i n n i n g t r a d e go o n e for t o o long we also c o u l d d e c i d e to cut all w i n n e r s at a certain bar. T h e r e f o r e .69% -7.2 . open profits.2 .21% -7.26% 5.72% 6.29% 4.25% 8.15 s h o w s that t h e r e s e e m s to be s o m e k i n d of local m a x i m u m s u r r o u n d i n g the 105-day region. and differences for 16 markets in the SDB system.68% ETD -6.76% -7. 2 7 points . 3 5 points —4.30% 6.64% 3. 8 8 points . 5 9 points . 9 8 points .28% -7.44% -7.0 . With all the suggested exit levels in place.0 .13 Bars in trade. ETD.3 . A t r a d e that g o e s on for t o o long not only i n c r e a s e s t h e s t a n d a r d deviation of o u t c o m e s for the entire s y s t e m .24 points 0. 9 5 points . 4 1 points .13% 9.

5 0 .65% 42. 3 7 points .00% 100.25 points .16 and 10.00% 55.00% 50.88% 32.00% 35.00% 90.00% 25.85 and an average profit of 1.41% 49.00% 70.00% 30. which is good news.2 .27%.4 5 .10% 47.02 points .00% 40. which shows that no market ended up with a profit factor below 1 or an average trade less .00% 10.9 .17 also show that the added exits have decreased both the drawdown and the standard deviation.65% 41.00% 45. 3 5 points .03 points 1.7 . 1 3 % .48%.00% 20. original profits.00% 60.16 points 0.3 . Without the exits.17 show the hypothetical results for wheat traded with and without the exits.03% 16.97% 45.14 MFE levels. 1 2 points 2. the largest winner would have been 3 2 .66 and an average profit p e r trade of 1.3 9 . With the exits. T h a t the S D B system is very robust can be seen from Table 10.41% Difference 2. 5 9 points . MFE 5. equal to $307 in today's marketplace.3 0 . Exit with a limit order if the open profit exceeds 7 5 % . 5 % or m o r e and ends up below the regression line for the original average trade. Exit with a profit if the trade lasts for m o r e than 105 bars. 5 9 points • Exit with a trailing stop.42% 49.69% 49. 5 0 % (thanks to the stop loss). 0 3 points .0 .34 points 2. 4 8 points .63% 47. • • Tables 10. wheat had a profit factor of 1.18.4 4 .42% 45.182 PART 3 Getting Out TABLE 10.0 .00% Original profit 7.16% 20.00% 15. For the modified system. 3 5 points . 5 8 points .2 5 .34% 12.00% 75.1 7 . wheat had a profit factor of 1.60% 37. 4 0 points .16 and 10.2 2 .00% 85.52% 49. 3 1 points . 3 7 points .00% 95.665 at today's market value.141 at today's market value.00% 80. 9 0 points . if the market moves the w r o n g way by 5 .02% 24. or —$1.69% 45.00% 65. 5 8 points .25% 34. Tables 10. and differences for 16 markets in the SDB system.1 4 . after bar 40. T h e largest loser would have been 5 . equal to $277 in today's marketplace. 3 1 points . or $6.63% 45.

18% -25. because of no disaster markets.04 points 1.10% 18.58 .27% 7.92% 11.107.26 points 0. 2 7 points 0.40 -5.62% 13. Table 10.41% 20.93% 21. S o m e markets.54 1.51% 12.15 Bars in trade.16% 28.29 10.02% 28.76% 60.00% 24. final profit.82 points than three times the expected costs of $75 p e r trade.43% 27.12% 17.07 points 1.75% 20.71% 23.678.57% 18. Bars in trade 55 60 65 70 75 80 85 90 95 100 105 110 115 120 125 130 135 140 145 150 Final profit 10.46% 24.05% 21. but unfortunately the increased security about each trade's individual o u t c o m e also results in a lower average trade for m o s t markets.47% 23.44% 14. 3 9 points 0.17% 27.35% 16.81% 23.37% 22.11% 8.09% 23.62 points 0.59% -3.65 points 1.22 points 1.88 -690.53% -1. However.99% Difference 0.759.36% 13.64 points 0. the overall result confirms the stability of the S D B TABLE 10.21% 15.649. all markets still have a profit factor above 1.78% 23.73% 3 1.85% 16.43 points 0.599.71% 18. open profits. and differences.00% 13.19 shows that with the exits added.09% 46.52 points 0.34 points 1.47% 6.71% 20.77 points 1.0 . before exits (original system) Total trades Profit factor Avg profit St Dev 1.34% 23.02 points .762.7 Winners Lrg winner Avg winner C u m profit 15 31.5 1.94 Losers Lrg loser Avg loser Drawdown 23 -7.49% 14.0 .97 points 0.69% 23.85 points 1.24 points .37 points 0.87% 14.50% 39.93% 11. even showed amazingly g o o d results with profit factors well above 10. like rough rice a n d natural gas.17% Open profit 10. Wheat.50% 17.90% 12.6 276.88 points 0.76% 19.16 Individual market performance summary.CHAPTER 10 Adding Long-term Exits 183 TABLE 10.

35 1.46 12.357.46 3.96 14.5 Winners Lrg winner Avg winner C u m profit 16 32.51 5. trade 987.67 6.513.17 Individual market performance summary.02 5.71 755.11 2.212.50% -2.750.11 276.52% 3 1.00 3.847.49 2.200.787.27 -7.23 2.781.05 1.20 we can see that a total of six markets m a n a g e d to increase their performance in at least three categories. 0 6 % system.92 -610.58 2 St.2 3 .964.27 -6.54 535.08 1.03 -5.11 917. Wheat.03 1.140.97 3.19 -546.89 1.13% 7.65 10.747.699.96 5.03% 6.29 1.39% 41. F r o m Table 10.918.596. and crude oil and cotton their profit factors.00 -6.444. Japanese yen. dollar index and D m a r k only managed to increase their average trades.557.42 1.30 333. dev.29 -7.45 1.94 378.34 2. Interestingly enough.463.089. the markets that did get a higher average trade a n d profit factor were markets that usually do well in almost any trend-following system.93 Losers Lrg loser Avg loser 23 -5.95 9.675.599. but also that as m a n y as seven only m a n a g e d to increase one category or less.95 2.678.18 Performance measurements for all markets.37 639.64 662.22 7.06 -28.184 PART 3 Getting Out TABLE 10. both with TABLE 10.73 6. These markets were Tb o n d s .303.932. Before exits (original system) Market Crude oil T-bonds T-bills Rough rice Nikkei index Natural gas Live cattle Lumber Coffee Japanese yen Copper Gold Dollar index D mark (Euro) Cotton Wheat P.53 -3.732.94 -11. factor 3. 5.46 -8.25 -2.24 Drawdown .19 10.665.89 1. T h e one market that did not improve its performance at all w a s natural gas.75 -4.72 228.771.62 1.54 Drawdown -4. and wheat. but even so.55 2.818.3 1.48% 7.00 4.825.816.16 17.58 .85 -5.36 5.66 Avg.51 3.784.23 -4.958. after exits (modified system) Total trades Profit factor Avg profit St Dev 1.693.86 -4.357.84% 61.430.583.11 1.734.76 -6.8 307. even after we have "polluted" it with all the exits.85 2.042.97% -1.241.275.181.524. Nikkei index.87 20.559.94% 58.626. it was highly profitable.07 2.

46 -12.458.09 1.40 87.49 4.886. we m a n a g e d to decrease the standard deviation for 13 and the drawdown for 8.558.451.19 Performance measurements for all markets.41 1.441.118.68 125. Of all 16 markets tested.38 2 St.64 5.472. After exits (modified system) Market Crude oil T-bonds T-bills Rough rice Nikkei index Natural gas Live cattle Lumber Coffee Japanese yen Copper Gold Dollar index D mark (Euro) Cotton Wheat P.04 Drawdown -1.42 1.610.042.626.376.72 1. TABLE 10.357.24 .72 -4.15 1.882.47 -1.38 267.890.78 -4.157.572.49 4.575. optimize the lookback a n d breakout period.01 248.11 3. In conclusion.96 2.24 9.009.29 2.73 -3.32 -5.76 24.36 -19.44 12.447. trade 747.70 -5. we did not m a n a g e to increase the profit factor and the average trade for as m a n y as we wanted.074.182.557.23 4.87 676.86 ^1.768.904. we could c o m e to the same conclusion as we did for the D B S system.018.529.22 670.97 6.784.98 -6. dev. 4.03 440.99 -4. factor 3.10 -8.98 1.925.019.49 307.44 1.68 2.15 1. to possibly increase performance. Unfortunately.78 1.94 1.27 2.45 -8.919.26 4.829.CHAPTER 10 Adding Long-term Exits 185 and without the exits.77 6.711.09 6. which is indicated by the fact that we m a n a g e d to lower the drawdown for eight markets and the standard deviation for 13 markets.803.049.51 -10.25 2.54 48.84 96. for the S D B system.041.95 3.176.10 3.119. we can see that the exits are still doing their j o b w h e n it c o m e s to streamlining the trades. This should m a k e it possible to trade it m o r e aggressively with p r o p er m o n e y m a n a g e m e n t . although the average trade did not improve as m u c h as we would have liked.98 8.98 10.85 Avg.43 2.89 -5.28 1. and/or continue to experiment with different types of stops a n d exit levels.68 1.93 6. That is.

38% -18.14% 4 2 St. Differences Market Crude oil T-bonds T-bills Rough rice Nikkei index Natural gas Live cattle Lumber Coffee Japanese yen Copper Gold Dollar index D mark (Euro) Cotton Wheat Better P. before/after exits.95% -39.02% -10.58% -11.08% -62.20% 4.28% -0.88% -29.00% 45.186 PART 3 Getting Out Summarized differences.06% -0.07% 6.16% 137.82% -30.55% 0.12% -59.50% 29.49% 4.96% -49.55% 4.65% 15.24% -4. trade -24.10% 18.74% -16.40% -31. -24.19% 0.25% -3.48% -7.96% -4. factor 11.57% 6 Avg.31% 90.68% -18.80% 15.10% 13 Drawdown -57.60% -26.29% 2.26% -92.39% 16.79% -86.51% 114.23% 5.70% 11.76% -87.88% -2.41% -12.33% -27.45% 89.23% -2.24% 5.17% -14.17% -82.27% 0.20% -24.65% 19. dev.56% 8 Better 3 4 2 1 3 0 1 2 1 4 1 1 1 2 3 4 — .77% -57.93% -30.48% -6.23% -78.81% -19.21% 11.34% -18.27% -10.31% -37.

0 5 % chance that I will b u m p into two such m e n in a row. Figure 11. we can estimate and m a k e use of the likelihood of a certain occurrence." W h e n doing so. we take a slightly different approach. 2 7 % chance that the next m a n I b u m p into will be very tall a n d only a 0 . a r a n d o m variable (like that in Figure 11. we have a s s u m e d that all our variables. such as the average profit and the standard deviation of the outcomes. M o r e often than not. as we did by m e a s u r i n g m e n ' s height for the M e a n d e r system in Part 2.1. and assuming the height of m e n is normally distributed. So far. The beauty of this assumption is that the familiar bell-shaped n o r m a l distribution is both easy to understand and. we are relating the r a n d o m variable directly to its m e a n a n d standard deviation. w h e n we have used several of our statistical m e a s u r e m e n t s . N o t e . continuous r a n d o m variables. we can start saying things like " u n d e r n o r m a l circumstances.CHAPTER 11 Working with Random Entries For our shorter-term systems. at the s a m e time. close to normally distributed variables as well. This chart is simply created with the help of the r a n d o m function in Excel.2 shows. That is. that for a variable to be r a n d o m it does not have to be normally distributed. have b e e n normally distributed a n d statistically independent.1 shows what such a distribution can look like. Figure 11. not only w h e n building systems but also when analyzing markets in general. there is only a 2 . as Figure 11. We look m o r e closely into the world of descriptive statistics and learn about a few new measurers that can c o m e in handy. With descriptive statistics. lends itself to approximate other.2. is created by taking a 10-period average of the r a n d o m variable in Figure 11. in turn.2) that is an average of another r a n d o m variable (like 187 . such as the percentage profit per trade. however.

which shows the daily percentage returns for the S & P 500 index over the period April 1982 to October 1999. and 9 9 . 9 5 . For any normally distributed variable. However.1 The bell-shaped curve of a normal distribution.3. a normal distribution can very well be both higher and narrower. the n o r m a l distribution has b e e n the statistical distribution m o s t analysts used to calculate market returns. with equally as m a n y observations to the left as to the right of the mean. although it has long been evident that the m a r k e t does not follow a n o r m a l distribution. it has a single m o d e (one value that appears m o r e frequently than the others) and is symmetrical. regardless of the distribution of the original variable.188 PART 3 Getting Out FIGURE 11. For the longest time. Most important. As you can see from Figure 11. If a distribution has a relatively peaked and narrow body a n d fatter tails than a normal distribution. and it s e e m s to have m u c h fatter tails than the n o r m a l distribution curve in Figure 11. it is said to .1) is distributed approximately as a n o r m a l r a n d o m variable. 2 7 % of all values lie within 1 standard deviation of the mean. it is said to be leptokurtic.1. 7 3 % of the values lie within 3 standard deviations of the mean. 4 6 % of the values lie within 2 standard deviations. that in Figure 11. If the opposite hold true. than that in Figure 11. This can be seen from Figure 11. To calculate the standard deviation of a sample you can use the standard deviation formula in Excel. it holds that 6 8 . or lower and fatter.3.1. this curve is not entirely symmetrical.

if all values are sorted from the smallest to the largest). That is. W h a t probably cannot be seen from Figure 11. in that a positive value for the kurtosis also might indicate that we are doing a g o o d j o b of keeping all our trades as similar as possible to the average trade. a negative value m e a n s that it is platykurtic. To calculate the kurtosis of a variable you can use the kurt function in Excel. T h e tail to the left of the b o d y stretches out further than the tail to the right. so that we m a k e sure that the system does not stand or fall with any large out Iyer trades. while the dispersion of these trades stays the same. 0 3 1 1 % . the skew equals .2 . This is a twoedged sword. be platykurtic. In the case of the daily distributions of returns for the S & P 500 index. with the average return p e r day equal to 0 . we are not changing the levels at which we take our profits a n d losses. If that is the case.3 is that it also is slightly skewed to the left (a negative skew) with a m e a n smaller than the m e d i a n (the middle value. A positive value m e a n s that the distribution is leptokurtic. To calculate the degree of skew of the distribution y o u can use the skew function in Excel.2 A random variable. not normally distributed.CHAPTER 11 Working with Random Entries 189 FIGURE 11. 0 1 . W h e n building a trading system we would prefer for . the kurtosis should increase with the n u m b e r of trades from w h i c h we can draw any statistical conclusions. In this case. 0 2 9 6 % and the m e d i a n equal to 0 . however. kurtosis equals 56. W h e n building a trading system we would prefer for the distribution of returns to be platykurtic.

Ideally. so that we m a k e sure that w e ' r e cutting our losses short and letting our profits run. the returns to be positively skewed (to the right) with the tail to the right stretching out further from the b o d y than the tail to the left. In this case. for instance. If the t r i m m e a n is larger than the average return. You can do this in Excel with the t r i m m e a n function. having to rely on a set of not-yet-seen out-Iyer trades to find your way out of a drawdown. A n o t h e r way to m a k e sure that a system is not too dependent on any out-lyers is to exclude t h e m w h e n calculating the average return. it m e a n s that historically the positive out-lyers have been larger than the negative out-lyers. 2 7 % percent of all observations will be within 1 standard deviation of the m e a n . If we d o n ' t k n o w for sure whether the distribution is normal. but on the other hand. which excludes a certain percentage of observations from each end of the distribution. whatever you think is best depends on what you expect and how you would like the system to perform in the future.190 PART 3 Getting Out FIGURE 11. w h e n building a trading system we would prefer for the t r i m m e a n to remain above zero. Instead we . so that we m a k e sure that the bulk of our trades are behaving as they should. it could be comforting to k n o w that the positive out-lyers historically have a tendency to be larger than the negative ones. If the t r i m m e a n is smaller than the average return. say that 6 8 . seems to be awfully close to gambling with your life and the need to rely on Lady Luck. it m e a n s that historically the positive out-lyers have been smaller than the negative out-lyers.3 A random variable skewed to the left (a negative skew) with a mean smaller than the median. After a few bad trades. we cannot use our n o r m a l distribution m e a s u r e m e n t s that.

Note that this method cannot give an exact value.16. as compared to Gold Digger I.41. F u r t h e r m o r e . Let us take a closer look at how we can rebuild and better the exit techniques for our short-term systems using this newly found knowledge. For a system to be completely symmetrical in both the long and short entries. from Part 2. we also start out with the assumption that no trade will be allowed to go on for longer than five days. we k n o w that 9 5 . we would exit a long position. which says that for a k greater than or equal to one (k 2= 1). that it lacks any type of stops other than the entry in the opposite direction. such as Gold Digger II. is that Gold Digger II is m o r e symmetrical in nature. however. T h e advantage with this system. or if we were short. Let us see if we can m a k e it completely s y m m e t r i c and at the s a m e time e x a m i n e the possibility of adding some sort of stop-loss techniques as well.CHAPTER 11 Working with Random Entries 191 have to use Chebychef's theorem. 6 7 . T h e exit techniques also were very simple. we needed two up days and two up weeks. For instance.74. the long and short exit should be m i r r o r images of each other. If the market proved us right by going our way for two consecutive days. You will find the original TradeStation code for this system in Part 2. Because we want this to continue to be a short-term strategy. . A A THE GOLD DIGGER SYSTEM T h e second version of our data m i n i n g system. T h e first step is to modify the TradeStation code to the following: Conditionl = CloseW(2) > CloseW(l) and CloseW(l) > C and C[2] > C[l] and C [ l ] > C. T h e reasoning behind this strategy w a s that a move in a certain direction cannot go on forever and that a short-term rebound is likely to happen sooner or later. at least (1 . respectively. A partly symmetrical system. We have to estimate this from our kurtosis and skew measurements. and 9 0 % of all observations. but if we d o n ' t k n o w what the distribution looks like we can only say that at least 7 5 % (1 — 1 / 2 2 ) will lie within 2 standard deviations. told us to go long as soon as we had two down days and two down weeks in a row. C o n d i t i o n = CloseW(2) < CloseW(l) and CloseW(l) < C and C[2] < C [ l ] and C[1]<C. not completely so. Gold Digger II. To go short. 1. Similarly. the standard deviations m u s t be 1.1 / k 2 ) will lie within k standard deviations of their m e a n value. we cannot say if they will be equally distributed around the mean. has either symmetrical entries or exits. if we have a n o r m a l distribution. 4 6 % of all observations will lie within 2 standard deviations. to encapsulate at least 50. but because of the different exit techniques. T h e danger with this strategy is. which can be quite a substantial amount of time and m o n e y away. without k n o w i n g the exact distribution. we would exit at the first close that went our way. and 3.

for instance. T h i s is good.559 30. this totally symmetrical version of Gold Digger does not look as good as its predecessor.483 Winners Lrg winner Avg winner C u m profit 4 8. suggesting that the right-hand tail stretches out further than the o n e on the left-hand side. B u t for now. looked for an optimal value in the s a m e m a n n e r as we did with the Black Jack system in Part 2. Total trades Profit factor Avg profit StDev 0. U s i n g the trade-by-trade export function from Part 1 p r o d u c e d the results in Table 11.13% 2. this new version of Gold Digger is not as g o o d as the original version. O n e not so g o o d thing.62 -17. In all.192 PART 3 Getting Out If Conditionl = True and MarketPosition = 0 Then Buy ("Go long") at open. however. b e c a u s e despite the fact that we are cutting all trades at day five the system by itself is still letting the profits run while cutting losses short. we will test this strategy on the R A D contract for the S & P 500 index futures contract. If BarsSinceEntry 5= 5 Then Begin ExitLong ("Exit long") at close.18 53. Of course. If we c o m p a r e this system to the original version of Gold Digger II.961 . Instead. we can see that the standard deviation for this version is m u c h higher. covering the period January 1995 to October 1999. these results are not normally distributed at all.983 Losers Lrg loser Avg loser Drawdown 3 -4.22% 80 1.4.25% -14.54 1. January 1995-October 1999. ExitShort ("Exit short") at close. is the positive value TABLE 11. A lower profit factor also indicates that the cost of doing business has increased.614 -5. we also could have used the measures for skew and kurtosis that seemed to be producing the m o s t stable results. which m e a n s that this version is m u c h riskier to trade.1 Performance summary for the modified version of Gold Digger II. while the average profit is m u c h lower. As w a s the case with the original system.47 46. Overall.65 9. If that had been the case. As can be deduced from Table 11. End.33 -1. we have a m e d i a n that is smaller than the m e a n and a positive skew.1. If Condition2 = True and MarketPosition = 0 Then Sell ("Go short") at open. let us settle for the five-day limit to a trade.2 a n d Figure 11. we might have decided on a different m a x i m u m trade length and.18 453 7.75% 28. T h e drawdown also has increased considerably.823 5. although it has its benefits and perhaps also a few strengths that can be improved upon even further using the right analysis tools.480 -58.

2 Descriptive statistics for Gold Digger II (modified version). with the p r o p e r set of exit techniques we should be able to lower the kurtosis into negative territory while k e e p i n g a positive skew. However.TER 11 Working with Random Entries 193 TABLE 11.21 2. Mean Median Kurtosis Skew Trimmean (20%) 0. F r o m Figure 11. hence. does not p r o d u c e stable returns.49 0. w h i c h suggests that so far the s y s t e m is leptokurtic and. .4 The distribution of returns for the modified Gold Digger II system. indicating that. T h e trick then is to see if we can improve on the results somewhat.4. the positive outlyers have been larger than the negative ones.04 for the kurtosis. historically. w h i c h is smaller than the m e a n . while also m a k i n g the distribution of the outcomes less leptokurtic and perhaps slightly m o r e skewed to the right. T h a t this should be possible is suggested by the 2 0 % trimm e a n .83 0. we also can see FIGURE 11.13 0.

we will.33 X 0. Because we know that the returns from this version of Gold Digger are not normally distributed. which indicates instability.014 * EntryPrice stop. An alternative m e t h o d could be to look for a different a m o u n t of trades that the standard deviation boundaries should encapsulate. suggests that the results also are a little too toppy.) T h e additional TradeStation c o d e follows: If MarketPosition = 1 Then Begin ExitLong ("Long profit") tomorrow at 1. let us say that a m o n g these we should cut four times as many losers as winners. For instance. which in this case would mean approximately 21 losers (80 X 0. (It also might be a good idea to add a m i n i m u m profit objective.0% and a profit target at about 3 . but because the average profit from the original system is somewhat low. 8 % .80) and 5 winners (80 X 0. we have to use Chebychev's theorem that says that the two standard deviation boundaries will hold for at least 7 5 % of all observations. respectively. which will be triggered if the market first closes with a larger profit but then moves down below this level again. If MarketPosition = -1 Then Begin ExitShort ("Short profit") tomorrow at 0.33 X 0. by adding these stops to the code.99 * EntryPrice stop. then. But because we do not know the exact distribution. if we would like to encapsulate 6 8 % of all trades. Instead. T h e reasoning.01 * EntryPrice stop. T h e positive kurtosis. instead.194 PART 3 Getting Out that the distribution of returns indicates that the right-hand tail is stretching out further than the left-hand side. This will be at an open profit of 1. This we can only estimate with the help of the other measures. ExitShort ("Short loss") tomorrow at 1. we should leave out about 1 6 % of all trades on each side. Doing so in the Gold Digger system would result in a stop loss at about 1. we also do not know how many of the remaining 2 5 % will be located in the right-hand tail or the left-hand tail. End. as suggested by the regular standard deviation formula.4%. a larger quantity of the 2 5 % remaining trades should be located to the left of the mean.014 * EntryPrice Then ExitLong ("Long prot") tomorrow at 1. we cannot simply cut off all trades that go against us by a certain value.038 * EntryPrice limit. aim for half the distance between the stop loss and the profit target. If Close > 1. For argument's sake then we could say that 20 percentage points should be located in the right tail with the remaining 5 percentage points in the left tail. is that because of the positive skew. Let us say that we would like to cut all trades that go against us with more than two standard deviations. . Let us see w h e r e this takes us.962 * EntryPrice limit.20). Just for argument's sake. This is generally good and an indication that we are letting the profits run. ExitLong ("Long loss") tomorrow at 0. however.

T h i s looks really bad. C o m p a r e d to the original system. 2 5 % and a drawdown no higher than 1 5 % . B e c a u s e m o s t t r a d e s s e e m to be m o v i n g a g a i n s t us immediately.5% r e g i o n . and perhaps the entry criteria m i g h t even be too strict? Let's take a look. Before we m o v e on with the next strategy.21% 106 1.013 -4. As you can see. m o s t n u m b e r s are slightly worse in Table 11. Specifically. As t h i n g s are right now.986 * EntryPrice Then ExitShort ("Short prot.6 and Figure 11. T h e only positive thing is that we m a n a g e d to decrease the drawdown quite a bit.986 * EntryPrice stop.08 229 7.446 Winners Lrg winner Avg winner Cum profit 39 5.79% 17. tested over the period January 1995 to October 1999 (see Table 5. T h e trick is to f i g u r e o u t why.35% Drawdown -12. F r o m Table 11.462 15.01% Losers Lrg loser 63. m o s t n o t a b l y w h e n i t c o m e s t o the still-positive k u r t o s i s a n d the j u s t b a r e l y positive skew.72% 36.30% Avg loser -1.3 Performance summary for the second modified version of Gold Digger II. T h e results can be found in Tables 11. these stops most certainly did not improve results.930 67 -8.") tomorrow at 0. As we can see from F i g u r e 11.6). W h a t do you k n o w ! ? All of a s u d d e n the system seems to be working fine. T h a t is a q u e s t i o n for u p c o m i n g c h a p t e r s . Total trades Profit factor Avg profit StDev 0. with an average profit of a m e r e 0 .14% 2. End. although the standard deviation stayed approximately the same. let us see h o w this system would have fared on older data.5 and 11. O n e simple TABLE 11. m o s t t r a d e s will be s t o p p e d out by the stop loss.6.5. we will look at the period January 1985 to D e c e m b e r 1994. with an average profit of 0 .4 a n d F i g u r e 11. leading up to the optimization period.CHAPTER 11 Working with Random Entries 195 If Close < 0. But because we believe in the original concept (please say "yes") we will not give u p .5 it is o b v i o u s that m o s t t r a d e s m o v e against us right off the bat. N o t only did the stops.348 8. though. they also h e l p e d t o w o r s e n t h e statistical m e a s u r e s . the profit p r o t e c t i o n stop a n d profit target s e e m to be c o m p l e t e l y u s e l e s s .07% 2.72% after five years of trading.51% 4. January 1995-October 1999. however.564 -40. exits. a fact m o r e evident b e c a u s e of the way we h a v e split the final profit into fewer p o s s i b l e o u t c o m e s .534 .3 shows the result with these stops in place. Table 11.5 we also can see that the b a d n e w s c o n t i n u e s w h e n we l o o k at t h e statistical m e a s u r e s . a n d profit targets h e l p t o c o m p l e t e l y s c r e w u p the actual results.21% -28. Perhaps we might be able in the end to increase the profits s o m e w h a t by substituting these exits with m o r e generic and less curve-fitted ones. A profit p r o t e c t i o n exit also m a k e s sure that m o r e t r a d e s will have their o u t c o m e s in the 1. 0 7 % and a cumulative profit of 4.

That is.5 The distribution of returns for the second modified version of Gold Digger II system. . This also is the conclusion to be m a d e from studying the distribution of returns chart in Figure 11. is a positive because it m e a n s that m o s t o u t c o m e s now are centered around the m e a n .6. w h i c h is m o r e than twice as long and consequently leaves a lot m o r e r o o m for b a d things to happen. as c o m p a r e d to Table 11.07 -1.14 0. FIGURE 11. indicating that the system is doing a good j o b catching those situations when the market is poised to take off in a swift m o v e — o n l y thing. these moves s e e m to be in the w r o n g direction. It is just not profitable enough.4.15 0.4 Descriptive statistics for Gold Digger II (second modified version).00 reason for this is the test period.196 PART 3 Getting Out TABLE 11.00 1. Mean Median Kurtosis Skew Trimmean 0. A positive value for the kurtosis. the system is doing a good j o b doing the s a m e thing over and over again. A n o t h e r negative reflection is that the skew is now negative.

DifrYS(O). N o w we take a closer look at the s a m e system using a combination of weekly and daily data.6 Descriptive statistics for Gold Digger II. we will build the indicator and its trigger levels using weekly data and then try to trade and monitor it using daily data. Second. A s s u m i n g we are once again trying to pick. TABLE 11.04% ^42.16 87. we can m a k e two interesting observations. T h e best way to go about achieving this is to use the weekly data series as data 2 in TradeStation.528 Losers Lrg loser Avg loser Drawdown 15 -12.981 -50.96% 25. very few trades are stopped out by the profit limit. First.25% 2. SetArr(O). That is. and with the trigger levels set to two standard deviations away from the m e a n . VSHigh(O). T h e big discrepancy in n u m b e r of trades. SumArr(O).06 . Vars: SumVS(O).996 L o o k i n g at the distribution of trades.the tops and bottoms.019 0. most losers are stopped out by the stop loss.11 57. Total trades Profit factor Avg profit St Dev 0. DifTArr(O).718 7.00% 270 1. January 1985-December 1994 (second modified version). when looking at trades stopped out with a loss and trades stopped out at the m a x i m u m profit.5 Performance summary for the second modified version of Gold Digger II. simply w a s n ' t working in favor of the trade.00 6.18 -15. the TradeStation code looks as follows: Input: VSStd(2). Mean Median Kurtosis Skew Trimmean 0. VSLow(O). from n o w on I simply refer to this system as the Gold Digger. VSMid(O).38 859 6. and only on occasion does a trade end up as a loser otherwise.00 -. Array: VS[20](0). StdVS(O). AvgVS(O). indicates that the m o m e n t u m .61 -1.559 -3. stay with the trade for five days (no matter w h a t ) .745 Winners Lrg winner Avg winner Cum profit 11 7. with the M e a n d e r system we have only looked at daily data.25 -1. For simplicity's sake. We will try to do something about that later.86 42. w h e n the trade was taken.283 296. THE MEANDER SYSTEM (WEEKLY DATA) So far.62 2.CHAPTER 11 Working with Random Entries 197 TABLE 11. January 1985-December 1994.

198 PART 3 Getting Out FIGURE 11. VSfSetArr * 4 + 1] = (H[SetArr] Data2 . For SetArr = 0 To 4 Begin VSfSetArr * 4 + 0] = (OfSetArr] Data2 .CfSetArr + 1] Data2) / C[SetArr + 1] Data2.CfSetArr + 1] Data2) / CfSetArr + 1] Data2. For SumArr = 0 To 19 Begin If SumArr = 0 Then SumVS = 0.CfSetArr + 1] Data2) / CfSetArr + 1] Data2.CfSetArr + 1] Data2) / CfSetArr + 1] Data2. End. SumVS = SumVS + VSfSumArr]. VS[SetArr * 4 + 2] = (L[SetArr] Data2 . If SumArr = 19 Then AvgVS = SumVS / 20.6 The distribution of returns for the second modified version of Gold Digger II. VSfSetArr * 4 + 3] = (CfSetArr] Data2 . between January 1985-December 1994. For DiffArr = 0 To 19 Begin If DiffArr = OThen .

7 Performance summary for the modified version of Meander.1 5 .7 and 11. DiffVS = DiffVS + Square(VS[DiffArr] .365 -51.443 Losers Lrg loser Avg loser 26 -4. End. a negative factor at this point.70% 41.02% 52. in Table 11. End. Total trades Profit factor Avg profit St Dev 0. T h e drawd o w n is a little too high. Sell ("Go short") tomorrow at VSHigh limit.694 -5.CHAPTER 11 Working with Random Entries 199 DifTVS = 0. c o m p a r e d to Gold Digger at the s a m e stage. F r o m Table 11. A high profit factor is c o m p l e m e n t e d by a high average trade. IfDiffArr = 19 Then StdVS = SquareRoot(DiffVS / 20). m a n a g e s to produce a higher risk-adjusted return than Gold Digger.73% 47.569 Winners Lrg winner Avg winner Cum profit 29 13. TABLE 11.7 show the results for this version of the Meander. which m e a n s that the Meander.739 Drawdown .270 10. but could probably be worked on with a well-working stop loss and some type of filtering technique (to be added later). January 1995-October 1999. A n o t h e r positive factor is the positive value for the skew.StdVS * VSStd)). VSLow = C Data2 * (1 + (AvgVS . VSMid = C Data2 * (1 + AvgVS).8 and Figure 11. w h e n tested on data for the period January 1995 to October 1999.65% -1.90 2.014 9. however.AvgVS).13% 55 1. 3 3 % . VSHigh = C Data2 * (1 + (AvgVS + StdVS * VSStd)).116 138. and with the trade-by-trade export function from Part 1 attached to the code. indicating that the individual outcomes are a little too centered around the mean.59% 47. If BarsSinceEntry >= 5 Then Begin ExitLong ("Long time") at close.27% -15. T h e very high positive value for the kurtosis is. too lets its profits r u n while cutting the losses short. End. If MarketPosition = 0 Then Begin Buy ("Go long") tomorrow at VSLow limit. Tables 11.8. at this stage. T h e standard deviation is somewhat high.93% 2. End.7 we see that this is not a b a d start at all.67% 3. ExitShort ("Short time") at close. but a slightly higher ratio between the average profit a n d the standard deviation shows that the Meander.

which shows the distribution of trades. or to use a time-based stop that forces us to exit before profits turn even worse. as the high n u m b e r of very profitable trades. or j u s t below. . The only way to limit these losses even further is to either exit the trade with a small profit (if and w h e n that h a p p e n s ) . Trades ending up in this area are likely not to be stopped out by a stop loss or a profit target. U s i n g the s a m e reasoning as for Gold Digger.7 The distribution of returns for the modified version of Meander.67 0.37 T h e clustering of individual outcomes around.200 PART 3 Getting Out TABLE 11. a profit protection stop will do it. Hopefully.10 5.7.55 1. but this time assuming that all o u t c o m e s are evenly distributed around the m e a n .78 0. at the 3.4 and 5% levels.8 Descriptive statistics for Meander (modified version). the mean also can be seen in Figure 11. indicates that the M e a n d e r is doing a good j o b of working with the m o m e n t u m of the market. Mean Median Kurtosis Skew Trimmean 0. we place a stop loss approxi- FIGURE 11.

If MarketPosition = -1 Then Begin ExitShort ("Short profit") tomorrow at 0.19% 73 1. ExitShort ("Short loss") tomorrow at 1. and Figure 11.373 -6.52% 2.018 * EntryPrice stop. with three very different and clear-cut scenarios that have crystallized: o n e highly profitable.81% 43. If Close > 1.032 * EntryPrice limit.88% 1.74% -1. o n e modestly profitable.968 * EntryPrice limit. 8 1 % .9936 * EntryPrice Then ExitShort ("Short prot.587 -36.484 Drawdown .982 * EntryPrice stop. A r e we good. 7 5 % .CHAPTER 11 Working with Random Entries 201 mately 12 trades in both on the left-hand tail and on the right-hand tail.64 or 3.758 Losers Lrg loser Avg loser 25 -5.388 Winners Lrg winner Avg winner Cum profit 48 4. A negative kurtosis m e a n s that we have spread out (or perhaps "plateaued o u t " is a better expression) the individual outcomes a little m o r e .9 and 11.8% and a profit target at about 3.9 Performance summary for the second modified version of Meander.10. Although this system is even m o r e distinct. End. T h e good news continues with a negative kurtosis.113 147. we are no longer cutting our losses as short as we probably should. This m e a n s that we place a stop loss (using the same type of code and principles as for G o l d Digger) at about 1. which could be interpreted as our target area has grown and.25% -19. Tables 11. 1 9 % and the drawdown to 1 0 . ExitLong ("Long loss") tomorrow at 0. January 1995-October 1999. therefore.764 7.8% loser.2%.78 1. and with only a few trades scattered around.0064 * EntryPrice stop. From Figure 11. End.1 0 . If Close < 0.64%. Total trades Profit factor Avg profit St Dev 0.75% 16. TABLE 11.2% winner. and one with a negative return.95% 34.8 show the results of using these stops and exits in accordance with the following TradeStation code: If MarketPosition = 1 Then Begin ExitLong ("Long profit") tomorrow at 1. T h e n u m b e r of profitable trades also increased dramatically to 6 5 .8. we see that we have transformed the distribution of returns to resemble Gold Digger.78% 65.9936 * EntryPrice stop. 8 1 % . With the skew in negative territory.0064 * EntryPrice Then ExitLong ("Long prot") tomorrow at 1. or what? As can be seen from Table 11.470 6. Either the trade b e c o m e s a 1. Furthermore. should be easier to hit. using the profit for the average trade as a guide we also add a m i n i m u m profit level at 0. however.") tomorrow at 0. although the average trade and profit factor declined somewhat.9. or a 0. we still m a n a g e d to lower the standard deviation to 2 .

8 % that ended up with a profit of 0. 2 % winner. on average.49 T h e interesting thing is that these different scenarios are so distinctly separate from each other that we almost can start calculating the likelihood for a certain scenario to happen. Mean Median Kurtosis Skew Trimmean 0. 5 % that ended up s o m e w h e r e in between. or at least every fourth trade. from Figure 11. 3 2 . That is. 5 2 % winner.52 0.8 %) that ended up with a profit of 3. we also can calculate the expected average return for all the trades that fall outside of the three major scenarios. we k n o w that on close to every third trade. With the average trade being a 0 . and only 9 . 2 8 . .8.64%.8% losers or worse.8 The distribution of returns for the second modified version of Meander.2% or m o r e .10 Descriptive statistics for Meander (second modified version).56 -0. we should end up with a 3 . which will c o m e out to approximately FIGURE 11.03 0. For instance. we can see that there are a total of 21 out of 73 trades (or 28.64 -0. 9 % that ended up as 1.202 PART 3 Getting Out TABLE 11.

But even with such a m o d e s t system as Gold Digger.834 -6. let us see h o w this strategy would have performed on data covering the period January 1985 to D e c e m b e r 1994. January 1985-December 1994. T h e good n e w s also continues w h e n we look at Table 11. also is positive in that it indicates that the system is doing a very good j o b in producing a very similar set of trades.03% 121 1. but because the basic characteristics are still there.11. T h e m e a n and m e d i a n values for the out-of-sample data are exactly the s a m e as for the insample data. This indicates that the system is very robust and therefore very likely to hold up in the future as well.12 which once again shows that the skew has moved into positive territory.980 6.8% losers than there are 3. A l s o . too bad for what is falling through the net.21% -2. while not quite as m a n y trades m a n a g e to be closed out with a limit order at the profit target level. C o m p a r e d to the optimization period. with an average profit of 0.11 Performance summary for the second modified version of Meander.820 Losers Lrg loser Avg loser 43 -11. we did not do quite as well. as shown in Table 11. as c o m p a r e d to Table 11. TABLE 11.232 Winners Lrg winner Avg winner Cum profit 78 21. A n d there is plenty of h o p e for the M e a n d e r as well. although the out-of-sample data did not completely stand up against those for the optimization period. that if something s e e m s to be too g o o d to be true during optimization.820 Drawdown .12. they did c o m e in remarkably close.54% -37.52% 3.71 1.92% 1.2% winners does not matter as long as the winners are worth so m u c h m o r e in dollar t e r m s . T h i s can also be seen from Figure 11. it likely is. 0 3 % . which shows the distribution of returns. we definitely did well enough. indicating that we are once m o r e letting our profits r u n while cutting the losses short. Nonetheless. It is obvious the market has changed its behavior somewhat.93% 69. which indicates that the system is working basically the s a m e over different time periods. That there are plenty m o r e 1. the m i n i m u m profit trades n o w m a k e up the majority of all trades. W h a t do the M e a n d e r and Gold D i g g e r experiences teach us? It is very p o s sible to c o m e up with a profitable trading system that is likely to w o r k equally as well on both the optimization period a n d the out-of-sample data. W h e n tested on out-of-sample data.28% 64. 0 4 8 % (or approximately $161 in today's market value). as long as we do not aim for the stars. Total trades Profit factor Avg profit St Dev 0. we still can work wonders after we improve it by adding a filter that helps us w e e d out the m o s t obvious losers.745 10.1 3 .CHAPTER 11 Working with Random Entries 203 0 . b u t even so. 2 8 % .46% 73. Before we move on with the next system.04% 35. T h e high value for the kurtosis.9. the results look very robust and promising.52% and a standard deviation of 3 .507 233. N o t .894 -44.

the n a m e implying that the system should be able to exploit minute market discrepancies and that if we only stay at it long enough we should be able to make a profit—not m u c h from each trade. using TradeStation's random number generator. Mean Median Kurtosis Skew Trimmean 0. which we called Black Jack. We started with a couple of preset entry and filter criteria. . but instead enter each trade randomly.12 Descriptive statistics for Meander. By entering randomly.27 2. but slowly and surely build up a stake worth retiring on. This time. based upon c o m m o n sense and market knowledge.62 0.9 The distribution of returns for the second modified version of Meander between January 1985-December 1994. we can test all markets several times and each FIGURE 11.52 0. we will reverse our procedure and start with no such criteria at all.64 21.33 THE BLACK JACK SYSTEM In Part 2 we built a system.204 PART 3 Getting Out TABLE 11.

for each run through the data. hours. As for the stop loss. starting out with the time-based stop and the stop loss. whatever the reason for entering in the first place. T h e quickest way to do the search is to declare both the trade length and the stop loss as inputs. we k n o w from experience that we probably are best off looking in the 0.5 to 2% region. T h e stop loss is obviously necessary to tell us when we are most likely to be wrong. in steps of one. For this version of the Black Jack system. To m i m i c approximate general market conditions that keep the market in a trading range approximately 6 0 % of the time. R e m e m b e r . To do this. for the m a x i m u m trade . because at the point of exit. the m a x i m u m trade length shows us for how m a n y bars (days. there are two slightly different ways to go about building a system like this. we set the r a n d o m n u m b e r generator to signal a long or short position approximately every fifth day.CHAPTER 11 Working with Random Entries 205 time c o m e up with a new and unique sequence of trades. In the optimization dialog boxes we tell TradeStation to r u n through every combination of our input variables. W h e n the market starts to discount other factors as m o r e important. we use the s a m e 16 markets as we did for the Standard Deviation Breakout ( S D B ) system. Do not bother using the optimization report in TradeStation. so this is what we will d o . no more. This is the amount of time we would like to stay with the trade. export the necessary information into a spreadsheet program for further analysis.or downtrend approximately 2 0 % of the time in each direction. etc. Basically. we can run the s a m e test on the s a m e market as m a n y times as we wish and always c o m e up with a new and different result. we must know the value of the average trade for each market and each stop loss/trade length c o m bination. the reason for our entry is no longer there. It is a g o o d idea to start out with the stop loss/trade length combination because if it happens that we enter exactly w h e n the market takes off in one direction and we happen to be right. and in an u p . I chose to go with two days instead of o n e day because of the s a m e TradeStation b u g I already have described and will rant further about below. starting out with the time-based stop. This is what we do in this example. (With all trades entered randomly there invariably will be situations w h e n we need s o m e type of exit to take us out of the market no matter what the underlying market conditions look like at the time.) the market r e m e m b e r s a certain event or whatever external factor that triggered the move. we want to exit and wait for the next move that might c o m e our way. thanks to the fact that each trade will be taken randomly. run t h e m through the optimization procedure and. from two to ten. if w e ' r e not in a position already. which later could be removed. because it is tailored to run on point-based back-adjusted contracts only. no less. for every time we run that particular combination on that particular m a r ket.) Or you could examine all exits in pairs. we restrict our search to the area between two to ten days. You could examine for one type of exit at a time. Because we want this to be a short-term system. covering the time period January 1980 to D e c e m b e r 1992.

but you also generate so m a n y data that a single Excel spreadsheet w o n ' t be large e n o u g h to hold it all. ExitShort ("Short Target") at EntryPrice * (1 . for the stop loss.01)) limit.) T h e TradeStation c o d e for this and all other runs follows: Inputs: Counter(O).01)) stop. 2 5 6 M b memory.01)) stop. .} ExitLong ("Long Loss") at EntryPrice * (1 . even on the fastest of computers. MP(0). SumETD(O).(TrailingStop * 0. We also declare a third input variable. the testing b e c o m e s extremely slow (now we're talking days). If you perform the analysis by exporting the result for each trade separately. FName( ). Note: This takes a hell of a lot of time. but hopefully this will be e n o u g h to form s o m e solid statistical conclusions.01)) Then ExitLong ("Long Trailing") at EntryPrice * (1 + (TrailingStop * 0.(StopLoss * 0. this produces something like 230.01)) stop. FTE(0). each one holding about ten years' w o r t h of data. I work on a dual processor. End. I run each combination ten times.5 to two.440 loops for each market.l).206 PART 3 Getting Out length. 5 0 0 M H z Pentium III.01)) Then ExitShort ("Short Trailing") at EntryPrice * (1 . (One Excel spreadsheet has close to 17 million cells [65.536 rows X 256 columns]. If Close > EntryPrice * (1 + (TrailingStop * 0. a n d from 0. LowestLow(O). If PositionGenerator = 4 Then Sell at Close. If BarsSinceEntry > = 1 Then Begin { ExitLong ("Long Target") at Entry Price * (1 + (ProfitTarget * 0. SumMAE(O). End. in steps of 0 . ProfitTarget(2).m . a n d I still have a few hours to go do other things when I do runs like this. TotTr(O). MFE(0). TradeLength(7). SumMFE(O). I have not bothered to count how m a n y trades we will do. This results in 1. SumFTE(O). PositionGenerator = IntPortion(Random(5)). StopLoss(l. 1 . ExitShort ("Short Loss") at EntryPrice * (1 + (StopLoss * 0.TrailingStop(O. TradeStrl("").01)) stop. Vars: PositionGenerator(O). If Close < EntryPrice * (1 . HighestHigh(O).01)) limit. ETD(0). With 16 markets to test.(ProfitTarget * 0.l).000 u n i q u e yearly trading sequences. which we can use to r u n the same stop loss/trade length combination as m a n y times as we wish. If MarketPosition = 0 Then Begin If PositionGenerator = 3 Then Buy at Close. MAE(0).(TrailingStop * 0.

. If MarketPosition = 1 Then Begin If BarsSinceEntry = 1 Then Begin LowestLow = EntryPrice. End. MFE = 0. HighestHigh = EntryPrice. MAE = (EntryPrice — HighestHigh) / EntryPrice. End. MAE = 0. HighestHigh = EntryPrice.EntryPrice) / EntryPrice. MFE = (EntryPrice — LowestLow) / EntryPrice. End. If High > HighestHigh Then Begin HighestHigh = High. If MarketPosition = — 1 Then Begin If BarsSinceEntry = 1 Then Begin LowestLow = EntryPrice. If Low < LowestLow Then Begin LowestLow = Low. MAE = (LowestLow — EntryPrice) / EntryPrice. End. ExitShort ("Short Time") at Close. MFE = 0. TotTr = TotalTrades. End. If High > HighestHigh Then Begin HighestHigh = High. End.CHAPTER 11 Working with Random Entries 207 If BarsSinceEntry >= TradeLength Then Begin ExitLong ("Long Time") at Close. If Low < LowestLow Then Begin LowestLow = Low. MFE = (HighestHigh . End. MP = MarketPosition. MAE = 0.

MAE = SumMAE/TotalTrades." + NumToStr(FTE. If CurrentBar = 1 Then Begin FName = "D:\Temp\BJS. SumMAE = SumMAE + MAE." + NumToStr(ProfitTarget. SumMFE = SumMFE + MFE.208 PART 3 Getting Out End. ETD = (ExitPrice(l) . SumFTE = SumFTE + FTE. each time . TradeStrl)." + NumToStr(TrailingStop." + NumToStr(StopLoss. a n d C T D values. If FTE < MAE[1] Then MAE = FTE Else MAE = MAE[1]." + NumToStr(TradeLength. 2) + ".EntryPrice(l)) / EntryPrice(l). MFE = SumMFE / TotalTrades. 4)+ V + NumToStr(MFE." + NumToStr(MAE. TradeStrl = LeftStr(GetSymbolName.ExitPrice(l)) / EntryPrice(l). 4) + ". FileAppend(FName. and the value of the input parameters for that particular run. End.ExitPrice(l)) / LowestLowfl].HighestHigh[l]) / Highesthigh[l].(ExitPrice(l) . SumETD = SumETD + ETD. If TotTr > TotTr[l] Then Begin I f M P [ l ] = I T h e n Begin FTE . ETD = SumETD / TotalTrades. End. ETD = (LowestLc-wfl] .csv". 2) + ". 2) + ". together with the average profit. End. I f M P [ l ] = -1 Then Begin FTE = (EntryPrice(l) . M F E ." + NumToStr(ETD. End. If LastBarOnChart Then Begin FTE = SumFTE / TotalTrades. End. This code exports the M A E . If FTE > MFE[1] Then MFE = FTE Else MFE = MFE[1]. 4) + ". 2) + ". 4) + ". End. 2) + NewLine.

we should select the one with the narrowest stop and shortest trade length. As you can see. Figure 11. with a stop loss of 1. like in the export function and on. and on. Because of this extra piece of code. analysis techniques.10 there are two areas that look particularly promising. we encounter another problem. if you entered at the close the day before) the syst e m does not recognize that y o u have a trade going. we m u s t move on. T h e others are there for other. the average profit c o m e s out to somewhere between —0. In this chart we would like to find an area that is as large as possible. Figure 11. the higher the standard deviation. we substitute for the r a n d o m trade generator a high probability entry technique. 7 % stop loss a n d the other one around the 1. w h e n we examined different moving average length combinations. T h e trick is to try to find a squared area that is as large as possible. on the day you wait for this condition to c o m e true (the first whole day in the trade. whether the trade is (or might have been) a w i n n e r or loser.2% stop loss.1 and 0%. no matter w h a t the underlying market conditions look like. . shows the average percentage profit as a function of the stop loss in percent and the trade length in days. In this case we can see that.3% and a trade length of eight days. This really should not be necessary. . a n d o n . T h e first one is around the 0 .CHAPTER 11 Working with Random Entries 209 the p r o g r a m loops through the market. but similar. but without this line TradeStation sometimes exits the long on the first o p e n following the entry. T h e first chart. O n e way to work around this is to declare your o w n entry price variables. but that causes you a multitude of other p r o b l e m s further d o w n in the code. T h e code above also illustrates another TradeStation feature. you can use the s a m e charting technique as we did in Part 2. . we also need the standard deviation that we choose to coincide with an area of positive returns in Figure 11.11 shows the standard deviation of the returns. with a value for the profitable trade that is as high as possible. the longer the trade length and the w i d e r the stop loss. we only need the average profit and the value for the inputs to keep track of all combinations. generally. For instance. but has a value that is as low as possible. Okay.1% stop loss looks very interesting. . A surface chart is an excellent way to get a feel for how the output variable reacts to two different input variables a n d how the inputs are interacting with each other. T h e area surrounding the eight-day trade length and the 1. In Figure 11. Obviously. For our p u r p o s e s .10. which I already ranted about. That is. O n c e we are done testing. such as w h e n exporting each individual trade or examining one exit technique at a time. T h e two charts in combination tell us that if we have two similar alternatives to choose from w h e n it c o m e s to the average profit.10. k n o w i n g that we n o w should have a system that lets us trade marginally profitably or at least does not force us into too large disasters.1 to 1. After you have exported all the necessary data for the trade length and the stop loss into your spreadsheet program. for the stop loss we use the criteria If BarsSinceEntry 3= 1.

.11 The standard deviation of the percentage returns as a function of the trade length and stop loss level.210 PART 3 Getting Out FIGURE 11.10 The percentage return from random entries as a function of the trade length and stop loss level. FIGURE 11.

generally. In this case too. no matter what h a p p e n s with all the other input c o m b i n a tions. .2% stop loss looks m o s t promising. it o n c e again seems that the streak around the 1. which m e a n s that the longer the trade. F r o m Table 11. the better off we seem to be. A similar relationship does not to exist for the kurtosis.2% level. we also can look at separate kurtosis and skew tables for each input. W h e n it c o m e s to the stop loss both the kurtosis and the skew p e a k around the 1. F r o m Figures 11. Figure 11. My rule-of-thumb is to give a higher consideration to the standard deviation and give the kurtosis the absolute lowest consideration. It also s e e m s that the kurtosis generally behaves the opposite way from the standard deviation.10 through 11. it b e c o m e s difficult to keep it negative. respectively.13 and 11. we can see that. w h i c h should be as high as possible. the higher the kurtosis. Tables 11.12 shows the kurtosis for each stop loss/trade length combination.13 we can see that the skew increases with the trade length. H e r e we also would like to find an area as large as possible with a value as low as possible and preferably also in negative territory.13 and Tables 11.13 shows the skew. a n d each stop loss value applied to every trade length.13 and 11. In this case.12 The kurtosis of the returns as a function of the trade length and stop loss level. In this case.2% and seven to eight days FIGURE 11. but bad from a kurtosis point of view. Finally. the longer the trade length and the w i d e r the stop loss.14 show w h a t these measures look like for each trade length applied to all the different stop loss values. This is good w h e n it c o m e s to the skew. To find the individual stop loss or trade length that is likely to be the most robust input parameter.14 it seems as t h o u g h a stop loss/trade length combination of 1.1 to 1.CHAPTER 11 Working with Random Entries 211 Figure 11.

3825 3.0000 4.1974 7.13 Kurtosis and skew for the trade length. has a lot going for itself. this is w h e r e the system-building profession still is m o r e of an art form than a science and where previous experience.0000 10. Trade length 2.5569 3.6934 1. It also is important to always keep in mind w h o you are and if you actually are psychologically able to trade and cope with the system that you are about to put together.0000 5. TABLE 11.8914 Skew -0.0000 6.212 PART 3 Getting Out FIGURE 11. plays a vital role for the end result.4220 3. R e m e m b e r these values when we m o v e on to c o m p a r e the stop loss in relation to the profit target.7326 -0.3015 0.8803 4.0727 0.7154 1.0000 3.2584 . both as a systems designer and as a trader.6979 4.0000 7.7034 -0.13 The skew of the returns as a function of the trade length and stop loss level.3035 -0.0748 1. N o t e that despite all the fancy techniques and statistical m e a s u r e m e n t s .8781 4.0289 2.0000 9.0000 8.0000 Kurtosis 3.

0154 0.8000 1.9603 2.5000 0.8000 0. 4 % profit target and 1. with the 3 .6853 4.6%>. to increase the return for the profit target by 2 1 % .0000 1.1007 8.2000 1. which m e a n s that it probably will continue to produce good results even if the true best values happen .1659 9. To reach these returns.9000 1. That is. but this time for a total of over 400.000 unique yearly trade sequences.4000 1. T h e profit target/stop loss tests were m a d e the same way as the stop loss/trade length tests. while also keeping the lid on the losses? O n e advantage of the riskier combination.0000 Kurtosis 4.3000 1.1491 2. 8 % .4595 Figure 11.8757 0.8 to 3.2669 0. which took forever and a day. the profit target was tested for all values between 1 and 4% in increments of 0.6% stop loss.0508 3. The average return seems to benefit from keeping the distance between the stop and the profit target as w i d e as possible.4%.7710 3.14 y o u can see that generally. we m u s t take on proportionally m u c h more risk that if we choose to keep the profit target down to around 2 . from 2. from 1 to 1. As you can see.1058 0.4604 1. Stop loss 0. and the other with a profit target around 3. y o u m u s t sit d o w n and figure out which one will give you the m o s t b a n g for the buck.6000 1.5511 -0.0930 11. the m o r e distant the profit target and the wider the stop loss.9000 2. which is not such a desirable trade-off.CHAPTER 11 Working with Random Entries 213 TABLE 11.4169 1.9646 4. we must increase the risk by 6 0 % .0409 0.0517 1. With several alternatives like this.9998 10.0752 7.2%.7810 2. B u t there also are two smaller areas around the 1% stop loss that could be worth looking into: o n e with a profit target of about 2 .14 shows the average return as a function of the profit target and the stop loss.1000 1.1055 1. 8 % . For instance. which combination will give y o u the highest average profit and at the s a m e time keep the profit potential open for large positive out Iyer trades.14 kurtosis and skew for the stop loss.6000 0.8028 2.0647 2.2248 3.3240 2.7000 1.4%.7000 0.6028 4. however.7190 5.8537 Skew 1. the better the results.9926 1. is that it is surrounded by several other high-value areas.5000 1.0985 0. From Figure 11.

which also confirms our findings from the stop loss/trend length examination. 8 % . to move away from w h e r e we think they are. as noted. such as those for kurtosis and skew also indicate that we probably would be better off with a stop loss s o m e w h e r e in the 0.214 PART 3 Getting Out FIGURE 11.000 unique yearly trade sequences. But. testing all values between 0. Other charts and tables. To m a k e a choice like this. W h e n testing for the .1% region.2 and 1% in increments of 0 .1% and a profit target of 2 . 1 % . it also could be a good idea to look at the standard deviation chart which. which shows the average profit in relation to different m i n i m u m profit levels and time-based stops. there is a price to everything in the market and only you can m a k e the choices most suitable for you. This results in m o r e than 125. Parts of the result from this r u n can be seen in Figure 11. shows that the standard deviation also increases the w i d e r the stop loss and the m o r e distant the profit target (chart not shown). Let us go with a stop loss of 1.15. in this case. To c o m e up with a trailing stop or m i n i m u m profit level we will r u n it against the time-based stop.14 The percentage return from random entries as a function of the profit target and stop loss levels.9 to 1.

6 % produces fairly robust results. . the criterion is that the market m u s t have closed above a certain open profit and then moved below this level on a later bar for a trade to be closed out. This confirms what we already suspected when we analyzed Figure 1 1 . W h e n it c o m e s to the m i n i m u m profit level. 1 0 . FIGURE 11. This generally gives the system slightly m o r e slack and results in fewer trades getting closed out at the m i n i m u m profit level.15 The percentage returns from random entries as a function of the trade length and minimum profit level. no matter what the m a x i m u m trade length h a p p e n s to be. However. a m a x i m u m trade length of seven to nine days seems to be the most robust t i m e window. because they never m a n a g e d to fulfill the m o r e stringent criterion for the profit protection stop to kick in. A m i n i m u m profit of 0 . There also will be a b u n c h of trades closed out with a loss. and a bunch of trades s o m e w h e r e in between because of the time-based stop. slightly m o r e trades getting closed out at the m a x i m u m profit target.CHAPTER 11 Working with Random Entries 215 m i n i m u m profit level. you might also say that the market m u s t close with a low above the desired level for all long trades or the high below the desired level for all short trades.

8%). T h e summarized results from adding these exits to the system. it s e e m s that our best alternative is a mini m u m profit level of 0.1% Exit with a profit if the market closes with a 0.15 Trade characteristics for Black Jack system exits.3200 0. (Please r e m e m b e r that we are still placing all trades randomly and with no concern for the underlying trend. Another interesting observation is that the M F E value is a great deal larger than the absolute value for the M A E level. Exit with a limit order if the o p e n profit exceeds 2. As you can see. MAE Average St Dev -1. together with the export function for the R A D contract. T h u s . Table TABLE 11. the exits are doing a very good j o b in streamlining the trades. T h a t this does not necessarily have to be so can be seen both from the relatively low standard deviation interval for the close and the fact that the standard deviation intervals for the M A E a n d M F E seem to be approximately the same.0140 0. aside from the actual outcome. We h o p e to boost the results by applying this set of exits to one or several high-probability entry techniques. this indicates that the stops a n d exits indeed manage to create a positive bias out of nothing. ( R e m e m b e r that it is reasonable to a s s u m e the standard deviation should increase the further away we get from the entry. All in all.1203 . the final system has the following exit rules: • • • • Exit with a loss if the market moves the w r o n g way by 1. This confirms our previous findings about an eight-day m a x i m u m trading horizon.15. T h e next step in building a system using the Black Jack technique is obviously to attach it to a high-probability entry technique. but at least it's there.6% together with a trade length between seven to nine days.0841 0.216 PART 3 Getting Out As you can see from Figure 11. It isn't m u c h . the average final profit is only one hundredth of a percent. assuming that the sequence of events is as outlined. Always exit at the close at the eighth day of the trade. Let us see h o w these exits would have fared with the short-term entry techniques already discussed.3652 Final profit 0.) Whatever the case. please note h o w narrow the standard deviation interval is.6% open profit and then moves below this level.) Figure 11. but the point is that just by using a set of logical exits and stop techniques.15. that m e a n s that.3192 MFE 1. can be seen in Table 11. Of course. an expected return this small is not tradable. we managed to produce a slightly positive average return.16 shows w h a t the average trade could look like.

16 Results for Gold Digger/Black Jack.16 The summarized life for the average random trade.348 Winners Lrg winner Avg winner Cum profit 78 5. and therefore. This is c o m p e n s a t e d for by a lower standard deviation. A n d what do you know! This set of exits worked even better than the originals. T h e cumulative profit also is m u c h T A B LE 11.47% -15. January 1995-October 1999.853 . First.955 -52. we have a m u c h higher average trade. is likely to be profitable on a w i d e variety of m a r k e t s as long as the point values are high enough.348 5.16 shows the results for the Gold D i g g e r entry technique. for the period January 1995 to October 1999. also r e m e m b e r that w h e n we put together the Black Jack exits we didn't even look at the S & P 500.68% 55.66% 44.3.013 -4. which m e a n s that the risk-adjusted return n o w is m u c h higher. using the Black Jack exits.031 79.88% 141 1.32% 17.CHAPTER 11 Working with Random Entries 217 F I G U R E 11. If this system w o r k s on the S & P 500 as well.49% 23. Total trades Profit factor Avg profit St Dev 0. although it is still slightly on the low side.30% -1.68% -28. T h i s table can be c o m p a r e d directly to Table 11. it m e a n s that we are exploiting a type of market discrepancy that is the s a m e for all markets. but before you do that.26 569 6.17% 1.14% 1. 11. traded on the S & P 500.920 Losers Lrg loser Avg loser Drawdown 63 -8.

Because if you have built a system exactly right. Total trades Profit factor Avg profit St Dev 0. the Gold Digger/ Black Jack combination has a lower profit factor and average trade than the Gold Digger system alone. furthermore.12% 1. January 1985-December 1994.55% -42. W h e n we built the Gold D i g g e r system. you will have squeezed out every o u n c e of information the market holds and the system will therefore have a positive bias. B u t after you have traded the system for a while you realize that the average profit. as long as the system does not behave in this r a n d o m fashion.218 PART 3 Getting Out higher.30% 45. which naturally put the lid on the results w h e n c o m p a r e d to the optimization period. we used the period January 1985 to D e c e m b e r 1994 for s o m e out-of-sample testing. W h e n it comes to the percentage profitable trades and standard deviation. but behave as if the individual results were dropping in randomly. It also is interesting to TABLE 11.45% 25. Gold Digger/Black Jack excels. on the other hand.80% 54.864 Winners Lrg winner Avg winner Cum profit 159 7. Tested over the period January 1985 to December 1994.20 407 5. you do not know what you are doing. has b e e n closer to 2 % . examine the system closely a n d try to find out w h a t is causing these excess returns and adjust the system to take these causes into consideration. any market condition that does not exactly parallel the conditions under which you built the system will be unfavorable.74% 292 1.1 5 . Therefore. On the negative side. in fact.17 Results for Gold Digger/Black Jack.559 -4. C o m p a r e Tables 11. The truth is that if you have built a system exactly right. T h e above reasoning also puts a twist to another interesting question.61% -1. and if something is predictable there is m o r e information to squeeze out.31% 35. T h e out-of-sample period was.718 4. what might cause excess returns might also be the reason why you have to sell your h o u s e tomorrow.595 Drawdown . as is the percentage of winning trades. 8 8 % .62% 1. say that you have put together a system that is expected to p r o d u c e an average profit of 1%.825 Losers Lrg loser Avg loser 133 -12. W h a t to do? T h e answer is. Because the truth is that w h e n the system is not w o r k i n g as intended. w h i c h m e a n t that it should hold m o r e unfavorable m a r k e t conditions. and the system will now perform on a suboptimal level—even if it temporarily might produce better-than-expected results. the outcome is predictable. Is there such a thing as a system that (seemingly) could be doing too well? For instance. it is likely that a comparison with that period will c o m e closer to the truth regarding what the Gold Digger/Black Jack combination really is worth.412 120. we see that the drawdown is slightly higher as well. T h e reason for this is. If the conditions have changed from one time to another that m e a n s that there is m o r e and newer information to squeeze out.5. m u c h larger in size.381 -53.17 a n d 11.

First. W h e n it is tested on the out-of-sample period. there is a quite dramatic decrease of m a x i m u m drawdown. on average.32% -7. because the better the TABLE 11.67 1. Second.74% -1. Finally. For the M e a n d e r / B l a c k Jack combination. 8 % . w h e n we c o m p a r e Tables 11. 1 1 .30% -19. As you can see.CHAPTER 11 Working with Random Entries 219 note that the results for the out-of-sample period no longer differ that m u c h from those of the in-sample period. As you can see. T h e B l a c k J a c k exits are not as g o o d together with the M e a n d e r system as those developed specifically for this system.18 shows h o w this system would have performed during the optimization period. T h e m o s t obvious thing. On the positive side.39% 1. T h e question we then m u s t ask ourselves is if we should stick to the original exits because we k n o w they work well with this particular system.24% 55.470 5.321 6. but because the ratio between the average profit and the standard deviation still c o m e s out m u c h lower. the average profit n o w has decreased to 0 . however.32.8%. Table 11.190 . this system has a slightly lower profit factor and average profit per trade. Total trades Profit factor Avg profit St Dev 0. the profit target is m u c h further away from the entry in the original set of exits.9.70% 16. O n e positive thing is the slightly lower standard deviation. although the profit potential is somewhat low.379 Winners Lrg winner Avg winner Cum profit 44 4. the results are a little m o r e inconclusive. which m e a n s that.19 and 1 1 . with any type of entry technique? To answer this question. In doing so. at least the results are very robust over time. the original set of exits also operates without a time-based stop. January 1995-October 1999.90% 79 1.88% 1. this does not help m u c h . is that it continues to be outperformed by the original M e a n d e r system.75% 34. or switch to the Black Jack exits. while the profit factor is down to 1. This confirms the reasoning behind points n u m b e r one and two. w h i c h m e a n s that we are putting m o r e trust into the entry signal.440 -26. T h e drawdown also has increased to 15. and is directly comparable to Tabic 11. two sad things happen with the M e a n d e r / B l a c k Jack combination. This m e a n s that we are letting the profits r u n " m o r e fluently.373 -4. W h a t is even worse is that the results for the out-of-sample period are no longer similar to those from the in-sample period.18 Results for Meander/Black Jack." so to speak. T h e percentage of profitable trades also is lower. as c o m p a r e d to the original system. 2 2 % . m a k i n g the m o s t of w h a t s e e m s to be a very strong entry technique. with the original set of exits. from an already low 1 0 . 8 % to a very low 7 . we can see that there are three major differences. we m u s t go back and look at the differences between the two sets of exits.76% 44.560 Losers Lrg loser Avg loser Drawdown 35 -5. the stop loss is also further away from the entry in the original set of exits. because they should work well.902 115.

the m o r e trust we should give it a n d not w o r r y so m u c h if the market does not happen to take off in the right direction immediately.16 1.14 w h i c h shows the results for both these periods combined.534 .42% 47.535 Losers Lrg loser Avg loser 6 -11.799 -53. that's the set I will stick with.20 Results for Black Jack/Black Jack.291 Drawdown . This is also a matter of personal choice.1 5 .980 5.654 178. while Table 11.220 PART 3 Getting Out TABLE 11. there is nothing keeping you from trading each entry technique with several sets of exit techniques. ( B y the way.92% 1.21% 73. T h e answer then.834 -4. Therefore.20 and 11.21% -1.09% 1. we would not have traded it.76% 136 1.64% 52.00% 67.01% 32.352 Winners Lrg winner Avg winner C u m profit 59 3.695 5.20 shows what this Black Jack/ Black Jack combination looks like w h e n traded on the S & P 500 over the period January 1995 to October 1999.79% -37.788 -4. that's curve fitted. Obviously. 7 9 % entry signal.299 Winners Lrg winner Avg winner Cum profit 7 21. Table 11.541 -13.72% 25.32 731 9. the most obvious being that w h a t worked during one time period did m o s t definitely not work in the other.35% -4. h a d we put together an entry/exit combination like Black Jack/Black Jack.38% 53.59% 87 2.82% 10. Personally I really like the reasoning behind the Black Jack exits. let us c o m p a r e this new set of exits with those that we created for the original Black Jack system in Part 2.875 Losers Lrg loser Avg loser Drawdown 28 -2.22% 2. s e e m s to be to stick to the original set of exits.21 shows the results for the period January 1985 to D e c e m b e r 1994.793 86.429 4. j u s t by c o m p a r i n g Tables 11. w h e r e the scenario w a s exactly the opposite. with a profit factor of 1.21 we can c o m e to a few interesting conclusions.5.3 and 11. it is quite reasonable to assume that we would have started to trade the Gold Digger/Black Jack combi- TA B L E 11.19 Results for Meander/Black Jack. 0 8 % in early 1995. However. Total trades Profit factor Avg profit St Dev 0. Instead. This is very interesting and b e c o m e s even m o r e so if we also mix in what we discovered about the original Gold D i g g e r system in Tables 11. January 1995-October 1999.50% 1.03 and an average profit per trade of — 0 . Total trades Profit factor Avg profit St Dev 0.) Finally.90% -1. under the assumptions that these things will even out over time and if there is one thing that the version with the Black Jack exits is not. January 1985-December 1994.18% -9. anyway. N o t e that neither table is directly comparable to Table 5.

54% 202 0.08% 1. It also tells us that even a market/syst e m combination that we trust completely will go through dry spells and that we never should place all our eggs in o n e basket (more about that in Part 5).2 6 . we h a d no way of c o m i n g to that conclusion. and exactly the s a m e environment that s e e m s to be so friendly this very m o m e n t will eat you alive in five m i n u t e s if you haven't m a d e sure that you have other alternative routes of action w h e n danger strikes. January 1985-December 1994.98% -21.1 6 .12% 1. 3 5 % Drawdown .89 -257 5. in both cases you are taking a w a l k on the wild side.919 Cum profit . B e c a u s e — a s already pointed o u t — w h e t h e r a system works better or worse than expected.181 Losers Lrg loser Avg loser 10 -6. and that the input parameters we are using are robust e n o u g h to be trusted. But with the help of hindsight. 0 5 % nation.21 Results for Black Jack/Black Jack. equally as well over several time periods and markets. we n o w k n o w that we in fact should have thrown away the Gold Digger/Black Jack combination and started to trade the Black Jack/Black Jack combination.199 -55. .373 -87. Total trades Profit factor Avg profit St Dev -0.30% 51. So what does this teach u s ? It teaches us that it is p a r a m o u n t to m a k e sure that the system works.197 Winners Lrg winner Avg winner 9 4.29% -1.905 4. F r o m j u s t looking at the results for the period leading up to D e c e m b e r 1994.24% 48. on average.229 -4.02% 13.CHAPTER 11 Working with Random Entries 221 TABLE 11.

and that I have used almost no market-specific parameter settings whatsoever. if you are using an older version of TradeStation.) In Part 3. b a s ing a lot of our findings on r a n d o m entries.PART THREE A Few Final Thoughts About Part 3 In this part. T h e difference lies in how we 223 . I h o p e you also have noticed h o w almost every single system has been applied to a broad selection of markets. and m a x i m u m favorable excursion. T h e s e two m e a s u r e m e n t s will continue to be used in Part 4. After we explored t h o s e topics as well. m a x i m u m adverse excursion. m o s t of the short-term systems have only been applied to the S & P 500. or at the very least be aware of. we also introduced a set of new statistical measures to help us evaluate a system's performance quickly and efficiently. we m a d e use of kurtosis and skew. at the following Internet site: www.mechtrading. but the reason for that is simply that the S & P 500 is one of the few markets that is worth trading short-term.com/tradestation/random. derived with the help of TradeStation's r a n d o m n u m b e r generator. b u t a profitable system also is a well-working system. In particular. related topics such as trade efficiency. T h e n it does not matter h o w m u c h you might think you k n o w about other. we discussed the various forms of d r a w d o w n with w h i c h you m u s t be familiar.html. long-term market situations. ( B y the way. you do not k n o w what m e a n s you m u s t use to c o m e to grips with it. R e m e m b e r — a well-working system does not have to be a profitable system. True. courtesy of Dave D e L u c a at Trade Works Software. where we take a closer look at different ways of filtering out favorable. when y o u are examining your trading systems. If you are not aware of where and w h e n the drawdown happens. you should be able to d o w n l o a d a r a n d o m n u m b e r generator. we set out to better the s y s t e m s develo p e d in Part 2 by a d d i n g a set of tailor-made stops and exits to e a c h system.

ending up with a positive return m e a s u r e d in percentage t e r m s . namely a high e n o u g h dollar value p e r point in relation to w h e r e the m a r k e t is trading and low e n o u g h costs in relation to this level. By well-working. B e c a u s e all markets are trading at different levels and with different point values. . no system will trade profitably if the necessary prerequisites are not there. By a profitable system. we m e a n a system that does a g o o d j o b catching a specific type of m o v e in several different markets. not all well-working s y s t e m s will be profitable on all markets.224 PART 3 Getting Out define well-working a n d profitable. on the other hand. we m e a n a well-working system that achieves a positive return in dollar terms. But p e r h a p s even m o r e important. however. We use percentage terms. As far as the system goes. because that is a universal m e a s u r e that allows us to ignore other market technicalities such as the point value or the current trading level. it all c o m e s down to w h a t type of m o v e we w a n t it to catch. as well as in percentage terms.

225 . "Is it better to trade with the long-term trend a n d can the short-term volatility help to determ i n e the outcome of the t r a d e ? " This is an important question that all traders should ask themselves. w h e n not to take the signal or. If only there was a way to tell w h e n to take the signal. perhaps even w h e n to fade the signal. and the only way to find out is to do the necessary research. is very difficult a n d probably the first reason why so m a n y of us fail as traders. Basically. there are two ways of filtering out high-probability trading opportunities and the question y o u should ask yourself is.PART FOUR High-probability Filters To continue to trust your trading strategy after a long d r y spell. especially if at the same time. the market is breaking out to record highs or lows. heaven forbid.

Then stick to that prognosis over a prolonged period of time. m a k e their living explicitly out of you instead of out of the markets. y o u will have a sure winner. they might or might not k n o w better. A n d the fastest way they can do that is to show h i m the fancy looking indicator-piling technique. the most obvious thing to test for first is to see if there is anything to be gained from only going with the longer-term trend. are heading. (Presented later is a m o r e thorough discussion of what 227 . this inefficient a n d risky method is favored by m o s t trading p r o g r a m vendors. T h e longer-term trend can be determined in several different ways. you might ask. Never m i n d that they all give different answers." you actually have a 9 0 % chance of c o m i n g to the w r o n g conclusion if you l o o k at each one at a time. since you've read this far). indicator piling is especially popular a m o n g n e w traders a n d a horrifying n u m b e r of broker-type salespeople w h o . A r e n ' t these guys pros. Nonetheless.CHAPTER 12 Filtering Besides filtering. shouldn't they k n o w better? Well. Indicator piling is executed by piling several similar indicators and "expert c o m m e n t a r i e s " on top of each other in the high hopes that one day.) If you still are fairly n e w to this game. O n c e you're past the piling stage (which you probably are. please do not buy into this method. not to mention w h a t the odds are if you c o m b i n e t h e m all. a n d with 10 indicators on your screen and only one being the "right one. the simplest (but certainly not the easiest) being your own fundamental and discretionary j u d g m e n t about where the economy in general. You do not need all those indicators—even though vendors compete with each other to develop the program that has the most. w h e n they all move in unison. but the only way they can m a k e m o n e y out of the wanna-be trader is to sell him their trading program. in one way or another. and your markets or stocks in particular. (At least this is my experience from m o r e than three years of watching these guys and their marketing tactics from a very close distance.

Imagine what we could do if we were to optimize a trend-monitoring strategy with a set of well-researched entry and exit techniques. live cattle.29. a n d "In the Pudding. cotton. Each market w a s traded 12 times. all markets but one (natural gas) h a d an average profit factor above 1 and for a total of nine we also can say that we can be 6 8 % sure the true profit factor will be above 1. we cannot say with 6 8 % certainty that the true profit factor for rough rice is above 1.) Thus. " S k i m m i n g the T r e n d " a n d "Fading Away. In both strategies. only seven markets had an average profit factor above one.19. aspect to the way you might perceive your role as a market participant and what it is you are trying to achieve. Table 12. T-bills. in w h i c h you test which lookback period w o r k s best together with the original entry technique on a wide variety of markets. In a series of articles for Futures m a g a z i n e .88 to 1. creating a total of close to 3. indeed. For all markets combined. dollar index. with an average profit factor of 1 and a standard deviation of 0. copper.500 unique yearly trading sequences per strategy.13. the true profit factor is with 6 8 % certainty likely to be found somewhere within the interval 0. wheat. I concluded that trading only in the direction of a 200-day moving average does. . This m e a n s that we can be 6 8 % sure that the true profit factor for all market will lie somewhere in the interval 1. natural gas. lumber. I stayed in each trade for five days. with a simple test like this we prove the benefits of always trading with the longt e r m trend. Japanese yen. as compared to entering randomly but only in the s a m e direction as a 200-day moving average. b u t for none of t h e m can we say with 6 8 % certainty that the true profit factor is above 1. however. With the trend filter. (Similar observations that support the findings regarding the profit factor also can be m a d e for the percentage of profitable trades." (February 1999). c o m e rain or c o m e shine. crude oil. the average profit factor c o m e s out to 1.03 to 1. This holds true even if you enter the markets randomly. even w h e n we enter randomly and with such a simple exit technique as always exiting after five days. with a standard deviation of 0. as long as you take trades only in the direction of the underlying trend and stick to the same exit strategy for all trades. Without the trend filter. TREND FILTERS FOR SHORT-TERM SYSTEMS A basic moving average filter should form the base of any filter testing procedure. T h e 16 markets traded were D mark." (March 1999). coffee. T-bonds. For instance. Nikkei index. For all markets combined. and in the April 2 0 0 0 issue of Active Trader Magazine.) Alternatively. increase results significantly.228 PART 4 High-probability Filters makes a trend and how this knowledge might add a m o r e theoretical. or even philosophical. you could simply only trade in the s a m e direction as a longt e r m moving average.1 shows the result of entering 16 different markets randomly in any direction over the time period January 1980 to October 1999. gold.12. and rough rice.16.

81 54.26 50.10 0.12 With trend filter St Dev 1.11 0.08 56.03 54.11 1.23 1.1 A simple 200-day moving average trend filter.76 1.83 1.11 50.62 2.10 50.21 1.1 shows w h a t this can look like for the S & P 500.23 1.00 St Dev 0.19 0.13 0.24 49. w h e n charted together with the O B V indicator and its 2 0 0 day m o v i n g average. does not necessarily have to coincide with the actual trend of the price chart. Figure 12.22 1.11 0. For instance.25 2.76 51.11 0.78 2.60 49.05 56.60 55.00 1.09 0.42 54.03 1.68 50.67 Market D-mark Crude oil Lumber Copper Gold Dollar index Live cattle T-bonds Cotton Japanese yen Natural gas Wheat Nikkei index Coffee T-bills Rough rice All markets St Dev 2.11 0.45 50.13 % profitable 52.04 1.32 Profit factor 1. the time series increases/ decreases m o r e in value the m o r e volume there is behind the m o v e of the price.43 52.18 0.13 0.43 50.11 0. which also takes the v o l u m e behind the m o v e into consideration.13 0.04 0.75 1.52 2.65 Alternatively.06 1.97 1.04 1.87 51.98 0.65 52.40 1.15 0. I think this is one of the best and m o s t versatile indicators there is.76 1. creating a n e w time series that holds b o t h price and v o l u m e information.98 0.26 1.65 1.96 2.16 0. the psychology of the m a s s e s .64 1.79 1.01 1.97 1.97 0.00 1.18 1.26 53.21 53. you could incorporate the on-balance-volume ( O B V ) indicator that weights each day's price action with its volume. w h e n the O B V indicator is above its long-term moving average the trend is up and w h e n it is below.11 1.14 0.32 55.07 0.52 1. That is.09 52.06 1.23 52. as you decide to define it.10 0.53 50.66 % profitable 56.09 0. Personally.56 51.26 53.99 1. O n e important thing to r e m e m b e r w h e n looking at Figure 12.09 0.09 0. cover everything there is to k n o w and monitor in the field of technical analysis.57 1.01 1.11 0. because of its ability to in one single and simple calculation. In this case.12 0.22 0.95 2.98 50.46 1.96 1.94 1. the trend is down.15 0. the O B V indicator tells us that the volume on .11 0.15 50.99 1.24 1.06 0.09 1.11 0.25 1.15 1. the interpretation being the s a m e as for the moving average of price.88 3.CHAPTER 12 Filtering 229 TABLE 12.20 1.1 is that the trend.94 0.99 1.08 0.56 49.63 1.51 2. the trend for the stock market has been down since July 1998. In this way. according to this indicator.11 0.16 0.02 1.16 St Dev 0.13 0.19 50.22 1. namely.86 1.64 3.14 50. however.82 2.09 0.59 1.03 0. although we all k n o w that the prices have not fallen m u c h (if any). Without trend filter Profit factor 0.26 51.11 0.21 1.90 2.

1 The S&P 500 futures contract charted together with a 200-day moving average and the OBV indicator. .230 PART 4 High-probability Filters FIGURE 12.

especially if " y o u r m a r k e t " starts to decline as well. I can go long. This could present a good opportunity to go short. however." m a n y m a r k e t participants prefer the safe haven of interest rates instead of taking the c h a n c e of investing in a market that they d e e m to have a very low profit potential. This counts for a total of six different filter techniques. there might be plenty of contradicting signals—short-term entry signals saying " b u y " and long-term trend filters saying "sell. If you like to take a m o r e fundamental approach. as indicated by the filter. Finally. This could present a good opportunity to go long. the lookback period will vary from 20 to 120 days. in steps of 20 days. I use another method that sort of turns the entire reasoning around." To work around this. T h e 16 markets tested . and your m a r k e t is increasing faster in price than the T-bills. 10 different lengths of the moving average will be tested. m a k i n g it increasingly harder to trade any rallies successfully. For the basic price filter. sell (buy) signal does not collide with a new high (low). the O B V filter. With the three m e t h o d s above. we can take the trade. as long as the short-term limit order. which says that as soon as we m a k e a new high (low) the long-term trend is up (down) and. If the opposite holds true. but only in the s a m e direction as the underlying trend. m o r e a n d m o r e investors n o w prefer to b u y the market instead of staying in interest rates. ranging from 50 to 250 days. in steps of 10 days. E a c h trade is entered randomly. It is important to stay consistent with your beliefs and trust w h a t your research has shown you for that particular system. on a portfolio of different markets to see if there is anything to be gained from adding filters to the systems as they function so far. In this chapter we test a few versions and different lengths of these trend filters. but because m a n y short-term systems try to enter with limit orders. you instead could calculate a ratio between the market of your interest and the short-term interest rate. based on several different trend filters. With the help of a long-term moving average we can decide whether or not the m o n e y is flowing into or out of the market in w h i c h we are interested. you might formulate a rule that says. Again. there is nothing keeping you from having several different strategies. I can go short. "as long as the trend is not down." O n e way of doing this is to use a basic highest high/lowest low breakout indicator. we strive to trade in the direction of the underlying trend. T h e reasoning behind this indicator is that w h e n the price for Tbills is increasing faster than the price for " y o u r market. together with all our short-term systems. and each market is traded 10 times for G o d knows h o w m a n y trades and unique yearly trading sequences—certainly enough to let us c o m e up with some meaningful conclusions anyway. Also. while there has been a diminishing volume on the up side. T h e exit strategy for all m a r k e t s and filters is the generic exits a n d stops combination we derived in Part 3. the m e t h o d is the s a m e as for the basic moving average filter. For each m o v i n g average filter. For the breakout filter.CHAPTER 12 Filtering 231 the down side has increased. and as long as the trend is not up. and the T-bill ratio filter we test if it is better to go with the slope or the crossover of their respective moving averages. such as that of the T-bill.

" + NumToStr(FirstInput. MALen(O). 2) + NewLine. MALen) < Average(Close. In TradeStation.1 Then FTE =(EntryPrice(l) . RaLen)[l]) Or (SystemSwitch =9 and Close < Highest(Close. RaLen)) Or (SystemSwitch =8 and Average(Ratio. natural gas. cotton. 4) + ". crude oil. If TotTr > TotTrfl] Then Begin If MP[1] =1 Then FTE =(ExitPrice(l) . lumber. wheat. Nikkei index.EntryPrice(l)) / EntryPrice(l). together with a simple export function that exports the value of the average trade for each t i m e the system runs through the market. T-bonds. RaLen(O). Vars: TotTr(O). End. TradeStrl(""). live cattle. OBVLen)[l]) Or (SystemSwitch =5 and Close < Average(Close. TradeStrl =LeftStr(GetSymbolName.ExitPrice(l)) / EntryPrice(l). MP(0). If (SystemSwitch =3 and OBV < Average(OBV. If LastBarOnChart Then Begin If SystemSwitch =3 or SystemSwitch =4 Then Firstlnput = OBVLen. FTE = SumFTE / TotalTrades. MALen)[l]) Or (SystemSwitch =7 and Ratio < Average(Ratio. MALen)) Or (SystemSwitch =6 and Average(Close. TradeStrl). coffee. End.232 PART 4 High-probability Filters over the period January 1980 to October 1992 are D mark. I f M P [ l ] = . the code for taking a short position with any of these filters. Japanese yen. RaLen) < Average(Ratio. c o p per. BOLen(O). OBVLen)) Or (SystemSwitch =4 and Average(OBV. S & P 500.csv". 2) + ". gold. SumFTE = SumFTE + FTE. OBVLen) < Average(OBV. OBVLen(O). FileAppend("D:\Temp\LT-Filter. dollar index. and r o u g h rice. SystemSwitch(O). FTE(O). . If SystemSwitch =9 Then Firstlnput = BOLen. If SystemSwitch =5 or SystemSwitch =6 Then Firstlnput = MALen. looks something like this: Inputs: Counter(O). BOLen)) Then Sell at Close." + NumToStr(FTE. If SystemSwitch =7 or SystemSwitch =8 Then Firstlnput = RaLen. Firstlnput(O).

over the last couple of years there s e e m s to be at least one m e t h o d that has emerged as the m o s t popular a m o n g technical analysts. in turn. from a safety point of view. with both the standard deviation method and the average true range method is that none takes the direction of the volatility into consideration. high volatility also m e a n s a higher profit potential. T h e rule w a s not to take a trade in any direction if the most recent volatility (true range) was higher than the average volatility (average true range) over the lookback period. For instance. w h i c h is basically what you do if you are an options trader and. T h e true range is defined as the distance between today's high (low). w h e n the optimization is done. T h e m a i n disadvantage. of the volatility is to the up side. W h e n it c o m e s to volatility. it also could be a good idea to try to use some . and today's low (high) or yesterday's close. whichever p r o d u c e s the largest distance. no matter the direction of the potential trade. and sometimes even three times! Weed out these extra runs from your spreadsheets manually before you do any further analysis. because this can take a while. but twice. the lower the volatility the better. such as Black and Scholes or Black-76. however. Therefore. T h e trick is to find those short-term situations w h e n the volatility is high enough to m a k e it worthwhile to trade but not so high that only a fool would enter the market. in a strongly up-trending market that only climbs higher and higher. T h e problem is that especially for shorter-term trading. therefore. the p r o g r a m once again runs through and exports the results for the "best" input variable combination. how do you k n o w w h i c h m e t h o d will give the most accurate reading? T h e truth is. For some reason. using the same markets a n d set up as for the long-term filters. in any market. it is obvious that most. nobody knows. I set out to test this technique over several different lookback periods. work with different option pricing formulas. To test if there is any validity to this c o m m o n l y used technique. and run the system optimizer on all markets at once. is of course very good if you are long in the market but not so good if you are short. the market is about to calm down a n d b e c o m e less risky to trade.CHAPTER 12 Filtering 233 With this code you can set up a w o r k g r o u p containing all the markets you would like to test. and hence. This. in steps of three days. not o n c e . higher volatility also m e a n s higher risk because of the greater likelihood that the market can move against you. ranging from 10 to 40 days. But because there is m o r e than one way to measure volatility. go do something else for a change. However. This is the true range or average true range method that states that w h e n the true range of today's price action is below the average true range for a specified lookback period (sometimes multiplied by a chosen factor) the most recent volatility is lower than expected. While the c o m p u t e r is chewing away. be careful w h e n you are exporting data from TradeStation while using the optimizer. Obviously then. and your guess is as g o o d as m i n e . Probably the m o s t c o m m o n way is to calculate the standard deviation of the closes. By the way. if not all.

To test if these two indicators could serve as filters for our short-term trading m o d e l s . I decided to go with the 130-day alternative. respectively. the opposite holds true for the breakout filter. the DMI-plus and DMIminus lines. with the absolute best risk-reward alternative being the 110-day average. For instance. For the R S I indicator you also can experiment with different trigger levels and. using random entries and the Black J a c k exits were derived in Part 3. the RSI indicator and the ADX indicator. there are a few m e t h o d s that readily lend themselves to this purpose. For the A D X indicator. as shown in Tables 12. instead of using the A D X indicator. the upward volatility is stronger than the downward volatility. If you have studied the math behind m a n y of the m o s t c o m m o n l y used oscillators that you can find in m o s t technical analysis p r o g r a m s . the reasoning is simple enough. w h e n the market is about to shift from a higher down volatility to a higher up volatility. you will realize that two of these indicators are your regular everyday technical analysis tools. Because the true best average might m o v e around. T h e results for all filters were then s u m m a r i z e d in separate tables. and vice versa.m i n u s line. T h e TradeStation code for all the volatility filters basically looked the s a m e as for the trend filters. Therefore. With the D M I . w h i c h show the results for the basic moving average crossover filter a n d the breakout filter. This is how you m u s t reason in any type of optimization procedure. I set out to test t h e m with lookback periods ranging from 10 to 40 days in steps of three days. from trading all 16 markets 10 times. Fortunately. for the 50-day MA cross filter the average profit per trade w a s 0. 0 5 % . the upward volatility is greater than the downward volatility.12%. it is better to use two of the c o m p o n e n t s u p o n which it is built. and vice versa.2 and 12. or better yet. which also s e e m s to be the best alternative after we have m e a s u r e d in all the other variables. For this I used the s a m e markets a n d general setup as for the long-term filters a n d for the average true range filters. it is slightly m o r e complicated. with a 60-day lookback period and the exact s a m e set up as for the moving average filter. T h e m o s t robust results seem to be around the 110. leaving the area in between as a neutral zone. only allow long trades if the R S I reading is above 60 a n d short trades if the RSI reading is below 4 0 .p l u s line above the D M I . You also could add to that the usual overbought/oversold technique and not go long (short) if you consider the R S I to be too high (low). a n d vice versa. T h e rule was only to take a trade in a specific direction if the volatility reading suggested that the volatility w a s stronger in that direction as well. b e c a u s e a high A D X reading could m e a n either a strong upside volatility or a strong downside volatility. For instance. such as the . If the results look p r o m i s i n g for the regular moving average filter. W h e n the indicator reading is above 50. however. the breakout filter only m a n a g e d to p r o d u c e an average profit per trade of 0 .to 170-day area. for example.234 PART 4 High-probability Filters sort of volatility m e a s u r e that can tell the difference between up volatility a n d down volatility. In the R S I indicator.3.

6040 0.2071 0.1186 0.3125 0.1578 0.2106 standard deviation.6654 0.0419 St Dev 0.1578 0.0293 0. TABLE 12.3 Breakout filter.2145 -0.2312 6.1023 0.1722 0.5265 10.6071 18. Length 20 30 40 50 60 70 80 90 100 110 120 Average 0.0474 0.1058 0. Length 50 70 90 110 130 150 170 190 210 230 250 Average 0. T h e different filter/lookback periods I decided to test further can be seen in Table 12.0591 7.5799 0.1703 -0.0318 0.2 MA cross filter.2946 0. a n d the skew. j u s t b e c a u s e one filter s e e m s to be inferior to another at this stage of the testing p r o cedure it does n o t m e a n that it will not work in the end.0317 0.2450 4.1438 0.0456 0.1139 0. which I set out to test together with the complete system.2699 5.1230 -1.1478 0.2116 0.7399 0.2570 5.2437 0. January 1980-October 1992.4098 0. the D M I indicator proved m o s t useful with a lookback period of 34 days.9180 -0.5719 7.0328 4. For all different lookback periods for all different filters I p i c k e d out one lookback period from e a c h filter.2875 1.1657 1.7311 -0.0369 Skew 0.1318 Ratio 0.3112 0.5863 0. 1 3 % p e r trade.3598 0.1718 0.2649 0.1766 0.4122 Skew 2.1202 St Dev 0.2966 0.7634 6.3499 0.5606 0.0821 0.6641 0.1475 0.7590 -0.5955 8.1196 0.6157 0.5033 .0777 1.0408 0.1201 0.3088 0. As we will see.1715 0.0309 0.5181 0.7088 0. the kurtosis.8615 5.6563 0.1252 0.1400 0.6584 4.8590 0.1385 0.0401 0.3182 Kurtosis 5.2141 0.4910 4. w h i c h generated an average profit of a l m o s t 0 .9483 0.0511 0.2034 0.1131 0. however.5950 Kurtosis 6.3835 1.2053 4. For instance.1001 0.7597 1.5006 -2.1538 0.1525 0.1378 4.CHAPTER 12 Filtering 235 TABLE 12.1461 0.1267 0.0818 0.1597 0.0460 0.6751 8.3662 13.0252 10.4. January 1980-October 1992.6568 0.5680 -1.2020 Ratio 0.2543 0.3691 3.1486 0.

THE GOLD DIGGER SYSTEM W h e n e a c h filter w a s tested o n t h e S & P 5 0 0 t o g e t h e r w i t h t h e G o l d D i g g e r syst e m .0511 St Dev 0. it is interesting to note that the m o v i n g average crossover m e t h o d does not s e e m to work quite as well as the slope m e t h o d of half the length. T h e not-so-tantalizing results for the breakout filter probably are because of the b a c k w a r d s reasoning you m u s t use for a filter like this.1860 0.6 a n d directly c o m p a r e d to Tables 11. allow for a lot of trades with the market in no trend at all.3042 9.6645 -0. and the breakout method.1117 0. when c o m p a r e d to the results for the D M I and R S I indicators.2165 0.7590 T h o s e that did not work at all w e r e the true range method.3499 Kurtosis 4.16 and 11. over the two t i m e p e r i o d s J a n u a r y 1985 to D e c e m b e r 1994 a n d January 1995 to O c t o b e r 1999. .0467 0. as things are right now. Perhaps it worked a little t o o well: high profit factors and low drawdowns count for very little if the trades are too few to m a k e it possible to c o m e to any meaningful conclusions.1019 0.0086 1.6045 0.7958 1. T h e s e can be seen in Tables 12. Filter DMI True range RSI OBV-cross OBV-slope MA-cross MA-slope Ratio cross Ratio slope Breakout Length 34 37 28 70 70 130 70 70 90 60 Average 0. which.5 a n d 12. t h e ratio cross filter p r o d u c e d the b e s t results. in combination w i t h the r a n d o m entries.1261 0.1807 0.8047 -0. the T-bill ratio slope indicator. w h e r e we put together our weekly directional slope system.1287 0.2699 Skew 1.1984 0.1131 4.4 Chosen input parameters for different filters. or in consolidation m o d e .236 PART 4 High-probability Filters TABLE 12.8590 1.5805 0.5210 0.7510 0.1026 0.4997 0.2137 0. T h e reason for the very few trades is to be found in the original composition of the entry technique.1722 0.1418 1.2583 0.0776 0. T h e s a m e holds true for the true range method.5629 0. Finally. This confirms our findings from Part 2.3633 0.1461 Ratio 0.9234 4.0422 3.9592 4.17.6920 0.1128 0.6568 0.1679 0.9665 1. A few things worth mentioning from this test can c o m e in h a n d y for your own testing: the O B V and ratio indicators probably would have d o n e better had they been tested with shorter lookback periods.9766 6.1131 0.1767 0. which needs to be modified to make it worthwhile trading it together with a filter.6584 6.6123 5.1240 0.1686 0.8603 7.

80% 1.06% 55.713 .713 -3.51% 30 1. For the period January 1995 to October 1999.78% 9.18% 1.093 Winners Lrg winner Avg winner C u m profit 16 2.30.33% 9.17.986 -4.834 Losers Lrg loser Avg loser Drawdown 2 -1. can be seen in Tables 12.94%.713 -3.30 and an average profit p e r trade of 0 . N o w all that needs to be d o n e is to m a k e sure that this new filter/entry c o m bination indeed betters the results as c o m p a r e d to the new entry technique alone.86% 11. which m e a n t that the risk-adjusted return also decreased somewhat. v Total trades Profit factor Avg profit St Dev 0. you see that this latest version beats its predecessor in almost every single category.450 4. T h e results for trading the new entry technique only.584 Winners Lrg winner Avg winner C u m profit 7 2.19% -4.10% 22.16 and 11.050 5.7. while the average profit increased somewhat to 0 .10% -1.32 593 5. and as you can see these results are not as good as those with the filter.8.CHAPTER 12 Filtering 237 TABLE 12. with a profit factor of 1. which is an indication of h o w well the system will let you sleep at night.6 Gold Digger (original)/ratio cross filter. which c o m e s awfully close to a complete change of trend.5 Gold Digger (original)/ratio cross filter. T h e d r a w d o w n is a tolerable 7.10% -1. January 1985-December 1994. the increase for the average profit also was a c c o m p a n i e d by a slightly higher increase of the standard deviation.10.07% -1. especially when TABLE 12.16% 46. there is little use adding the filter.078 Losers Lrg loser Avg loser Drawdown 14 -2. This is shown in Table 12. If you c o m p a r e this version of the Gold D i g g e r — with its slightly loosened entry trigger and the attached filter—with the version depicted by Tables 11. To m a k e it a little easier to enter a trade we can loosen the original entry somewhat and only look at the latest week's market action.67% -6. T h e drawdown also increased to approximately 12%.268 37.21% 77. it turns out that the breakout filter produces the best results for the period covering January 1985 to D e c e m b e r 1994.37% 5. If it does not. However.20% 1. Total trades Profit factor Avg profit St Dev 1.613 17. After we modified the original entry technique.9 and 12.65% 9 5. the profit factor also came out to 1. without the filter.22% -3. not the least of w h i c h is the drawdown.450 6.002 -14.040 This is largely b e c a u s e the original entry trigger d e m a n d s two w e e k s in a row in the s a m e direction. before we are allowed to enter with a limit order in the opposite direction. 2 0 % p e r trade. 1 7 % .80% 1. the breakout filter was the m o s t profitable for the period January 1985 to D e c e m b e r 1994.91 4. January 1995-October 1999. If we do that. which is shown in Table 12.

72% 17. 1 1 % it c o m e s to the drawdown and percentage of profitable trades. T h e lower cumulative profit.654 Winners Lrg winner Avg winner Cum profit 242 7.62% 1.120 -3. In fact.9 Gold Digger (modified)/no filter. does not matter at this point.23% 46. Total trades Profit factor Avg profit St Dev 0. because this will be dealt with later by attaching the proper m o n e y m a n a g e m e n t during the "optisizi n g " process.59% 16.1 2 .28% -28.718 4. January 1995-October 1999. 2 9 % .559 -4. with the filter attached. we have not only m a n a g e d to consistently improve the original system from Part 2.949 -26. T h u s .106 Losers Lrg loser Avg loser 212 -12.30% 25. January 1985-December 1994.88% 188 1. F u r t h e r m o r e . even though this is a system specifically designed for the S & P 500.68% 53. as a final step we even took away some of the original c o m p o n e n t s .44% 46.68% 454 1.38% 77.135 Winners Lrg winner Avg winner Cum profit 193 5. January 1985-December 1994.48% -1. TABLE 12.871 Drawdown .8 Gold Digger (modified)/break filter. but we also have m a n aged to do so without stacking indicators on top of each other and curve fitting the final version to a specific market. Total trades Profit factor Avg profit St Dev 0.52% 367 1.642 261. ending up with a m o d e l that trades far m o r e often than its predecessor.729 Drawdown . by following the system building sequence outlined in Part 3. T h e profit factors and average trades.41% -15.17% 1.94% 47.61% -1.348 5.329 Winners Lrg winner Avg winner Cum profit 101 5.70% -42.26 513 5.7 Gold Digger (modlfied)/break filter.30 569 5. Total trades Profit factor Avg profit St Dev 0.30 685 6.147 -61.58% 52.670 Losers Lrg loser Avg loser 87 -8. we have traded it randomly.862 -40. which really was nothing m o r e than a very simple entry technique. are not as g o o d without the filter. m o s t of the time we haven't even used the S & P 500 index in our testing and w h e n we have.464 140.36% 86.55% 53. summarizing the results together with 15 other markets.013 -4.976 4.14% 1.20% 1.238 PART 4 High-probability Filters TABLE 12.15% 1.798 TABLE 12.62% 41. too.833 Losers Lrg loser Avg loser Drawdown 174 -4.03% 1.30% -1.17% -7.595 292.1 8 .

which are comparable to Tables 11.11 and 12.19% 46.999 -11.516 TABLE 12. altering the entry level by changing the n u m b e r of standard deviations away from the m e a n needed to signal an entry. for instance.59% -5.CHAPTER 12 Filtering 239 TABLE 12.14% 1. which is already taking the trend into consideration by looking at the last five weekly bars to calculate the entry levels.890 Losers Lrg loser Avg loser Drawdown 8 -1.77% 1.30% -1.80% 1.401 129.348 5.209 5.33% 9. January 1985-December 1994. the best filter w a s the moving average slope filter. Therefore. as it is.127 82.36 2. we will keep the M e a n d e r system.19. it turned out to be very difficult to find a filter that could p r o duce consistent results.37% 50.535 -3.15% 1. This w a s very difficult to work around by.41% 27. At this point.12 The Meander/NA slope filter. Total trades Profit factor Avg profit St Dev 0. N o r did it help to switch away from the Black Jack exits.11 The Meander/MA slope filter.10 Gold Digger (modified)/no filter.23 520 6.68% 60 0.939 Losers Lrg loser Avg loser Drawdown 28 -8.81% 236 1.05% -1.21% 1.18 and 11.495 THE MEANDER SYSTEM For the Meander. January 1995-October 1999.529 -45.56% 72. Tables 12. Total trades Profit factor Avg profit St Dev --0.609 5.12 show the results. which p r o d u c e d a very high profit factor for the latter of our two testing periods.169 -4. it also could be worthwhile pointing out the importance of trying to c o m e up with a logical reasoning and a way to understand why a specific TABLE 12.48% 49. T h e reason for this can probably be found in the composition of the original entry technique.13% 55. In fact. together with the Black Jack exits. but still failed miserably during the period January 1985 to D e c e m b e r 1994.67% -27.41% 14. also developed in Part 3.15% -28.498 Winners Lrg winner Avg winner C u m profit 21 4.34% -13.19% -3.095 Winners Lrg winner Avg winner C u m profit 120 5.685 Winners Lrg winner Avg winner Cum profit 32 2.64% -1.89 -250 5.60% 38.672 -13.40% -13. January 1995-October 1999.732 -44.63% 29 3.07% 1.013 -4.52% 24.499 Losers Lrg loser Avg loser Drawdown 116 -8.509 .450 3.85% 17. Total trades Profit factor Avg profit St Dev 0. back to the original exits.09% -4.

this matters very little.08% -10.635 4.838 Winners Lrg winner Avg winner C u m profit 117 4. and h e n c e m u c h m o r e likely to hold up in the future as well.720 TABLE 12. As you can see.02% 65. Total trades Profit factor Avg profit St Dev 0. January 1985-December 1994. because the system now is m u c h m o r e robust a n d reliable. as long as we can learn something and c o m e to s o m e conclusion about how or what to do to improve results in future experiments.14 and 11.775 168.41% 50.429 -4.46% 1.788 -4.14 Black Jack (modified)/OBV slope filter.038 146. as can be seen if you c o m p a r e Tables 12. THE BLACK JACK SYSTEM W h e n we built the Black Jack entries in Part 2.35% 57. the results are not quite as good as with the original system.485 Winners Lrg winner Avg winner Cum profit 60 4.78% -9.981 Losers Lrg loser Avg loser Drawdown 85 -3.036 -36. T h e results for the time period January 1985 to D e c e m b e r 1999 can be seen in Table 12.43% 202 1.09% -1. With this filter. T h e percentage of w i n n i n g trades also is very high.38 641 4.12% 1. To see if any of our other filter techniques would work any better and be m o r e consistent.240 PART 4 High-probability Filters filter/system combination does not work.13 and directly c o m p a r e d to those in Table 11.63% 92 2.818 Losers Lrg loser Avg loser Drawdown 32 -2. As a result. the system worked very well during the latter of our two testing periods but not at all during the former. in the larger scheme of things.04% 1.33% -5. the system started to produce fairly consistent results together with the O B V slope filter. I substituted the original filter for each o n e derived in this section.92% 13.20% -10.03% 34. R e m e m b e r that there is no such thing as a failed experiment. January 1995-October 1999. For the period January 1995 to D e c e m b e r 1999.00 1. Total trades Profit factor Avg profit St Dev 0.556 5.19% 1. 1 9 % .20. we have taken a negative average profit p e r trade and turned it into a positive one of 0 . Other positive signs are a low standard deviation and a tolerable drawdown. TABLE 12.478 -16.21. However.13 Black Jack (modified)/OBV slope filter.90% -1.976 .20% 43. we h a d already added a 200-day moving average filter.905 4.88% 42.22% 13.

short-term entry point as early as possible within this trend. In theory. as a standalone indicator. Instead. a n d vice versa. p e r definition. T h e range multiplier we varied from 0. the p a r a d o x is that for the long-term system filter we m u s t work with shorter-term data. which could not be used in a short-term system. we also can use the A D X indicator. because.5. this m e a n s that we no longer can trade the system randomly. For the longer-term system the trend is. in steps of 0. For long-term systems. For one thing. as we did for the short-term systems. as indicated by the 241 .5 to 2. it says nothing about the direction of the volatility. we can assume that a high A D X reading m e a n s a high volatility in the direction of the trend.CHAPTER 13 Long-term Volatility Filters For long-term systems. This time. already defined. we will do it a little differently a n d change the range multiplier as well as the lookback period. or situations w h e r e the direction of the volatility coincides with the anticipated m o v e of the market. the best way to achieve this is to look for either low-volatility situations in general. For short-term systems. For the lookback period we varied the length from 5 to 25 days.2. A g o o d indicator to use for the former m e t h o d is the average true range indicator. we m u s t find the most opportune. however. T h e added trading rule w a s not to take the trade if the true r a n g e for the day of the breakout surpassed the average true range t i m e s its multiplier. we m u s t think a little differently. the p u r p o s e of the filter w a s to identify the type of trend that holds as m a n y and as favorable short-term moves as possible. H e n c e . But if we use it together with any other long-term indicators. with the direction of the trend already defined. A two-variable test also m e a n s that we once again can m a k e use of the surface chart m e t h o d that we developed for the directional slope system in Part 2. in steps of 2 days.

in relation to the final profit of the trade. the M A E and S T D m a k e up the entry efficiency. should primarily be dealt with by filtering out the m o s t opportune trading situations. it is important to understand that the two are not the s a m e . because as the A D X indicator is their first choice when it c o m e s to measuring the strength of the trend. j u s t e x a m i n e the S T D w o u l d n ' t m a k e m u c h sense either. A l t h o u g h we might be using the s a m e techniques to derive the necessary data for the S T D as for the M A E . However. and r o u g h rice. For the lookback period we varied the length from 10 to 20 days. crude oil. T h e trigger level we varied from 15 to 35. dollar index. Nikkei index. which is what we are about to do now. on the other hand. T h e time period covered was January 1980. which was to find the m o s t opportune m o m e n t to enter in a trend that w a s already defined. because this would w e e d out far too m a n y trades and m a k e the results less meaningful. in steps of 1 day. wheat. Both system/filter combinations also m a k e use of the stops a n d exits for each respective system we developed in Part 3. T h e added trading rule was not to take the trade if the A D X reading for the day if the breakout was below the required trigger level. w h i c h we did in Part 3.5. natural gas. T h e m a r k e t s used for testing were D mark. because we still would like the trade to go on for as long as possible within the confines of the original system. lumber. gold. to October 1992. Therefore. T h e STD. N o r does either o n e of t h e m equal R I N A S y s t e m s ' entry efficiency. described in detail in Part 3. T-bills. w h e r e we also took care of all the other types of drawdowns and excursions. To do that would also contradict the reason we added the filter to the system in the first place. The A D X test also was d o n e as a two-variable test. although the M A E should primarily be adjusted with a specific exit technique. for a long-term system. coffee.242 PART 4 High-probability Filters second indicator. A n o t h e r important difference between the long-term a n d short-term testing procedures is that for the long-term systems we cannot look at the final profit for the trade. this m e a n s that we m u s t examine the entry efficiency of the trade. the best way is to try to minimize the STD. Figure 13. In Part 3 we concluded that. live cattle. Japanese yen. although the interpretations and treatm e n t s of the two are not the s a m e . copper. it is reasonable to a s s u m e that the S T D will be the s a m e as the m a x i m u m adverse excursion (MAE) and therefore can be derived the s a m e way. The difference is that the S T D should primarily be adjusted with the entry technique. in steps of 2. cotton. In essence. T-bonds. B e c a u s e this is not a . w h e r e we tested different lengths for the lookback period together with several different trigger levels. such as a stop loss. which for our purposes is also the original entry strategy—or this is the theory m a n y system vendors seem to adhere to. T h e M A E should primarily be dealt with by adjusting the stops. Together.1 shows that the D y n a m i c Breakout System ( D B S system) produced the highest entry efficiency in combination with the average true range filter with a seven-day lookback period and a multiplier of 0.

we should be able to go with a multiplier of 1. we should strive to place ourselves somewhere in the area around a lookback period of 12 days and a trigger level at or above 2 5 . In this way.2. Which means we will settle for a 9/1. the 1. we should strive to place ourselves s o m e w h e r e in the area around a lookback period of nine days and a multiplier around 1. In .2 it seems that. Figure 13.5. particularly robust solution.1 The entry efficiency for the DBS system in relation to the true range multiplier and its lookback period. for instance. we avoid any of the areas of lower efficiency around. however. In Figure 13.1 multiplier. it cannot be our preferred choice. Because this too is not a particularly robust solution. it cannot be our preferred choice. which depicts the standard deviation of the efficiency. Support for this also can be found in Figure 13. without getting too close to the higher standard deviation areas. if it j u s t so h a p p e n s that the true best combination might shift around a little in the future.CHAPTER 13 Long-term Volatility Filters 243 FIGURE 13. as far as the standard deviation goes. Instead.3 shows that the D B S system produced the highest entry efficiency in combination with the A D X filter with a 20-day lookback period and a trigger level of 3 3 . Instead.5.5 lookback/multiplier combination.

244

PART 4

High-probability Filters

FIGURE

13.2

The standard deviation of the entry efficiency for the DBS system and the true range filter.

this way, we avoid any of the areas of lower efficiency around, for instance, a trigger level of 21 and a lookback period of 11 days, if it happens that the true best combination might shift slightly in the future This is confirmed by Figure 13.4, w h i c h depicts the standard deviation of the efficiency. In Figure 13.4, it seems that, as far as the standard deviation goes, we should go with a trigger level of 25 and a lookback period of 11 days, without increasing the standard deviation too m u c h ; because this seems to be possible without running the risk of lowering the efficiency, this is our preferred choice. Table 13.1 summarizes all the preferred choices for our three long-term syst e m s and filters. T h e time period covered is from January 1980, for all systems, to October 1999 for the D B S and directional slope systems, and to October 1992 for the standard deviation breakout system. T h e trade-by-trade export function from Part 1 is used to export the results into an Excel spreadsheet for further evaluation and comparison with earlier, pre-filter results.

CHAPTER 13

Long-term Volatility FiIters

245

FIGURE

13.3

The entry efficiency for the DBS system in relation to the ADX trigger level and its lookback period.

TABLE

13.1

Preferred system/filter combinations. ADX filter System Dynamic Breakout Standard Deviation Breakout Directional Slope LB period 11 14 11 Trigger 25 21 25 True range filter LB period 9 17 11 Multiplier 1.5 0.9 1.7

246

PART 4

High-probability Filters

F I G U R E

1 3 . 4

The standard deviation of the entry efficiency for the DBS system and the ADX filter.

THE DBS SYSTEM
Tables 13.2 a n d 13.3 show the result for the DBS/true range combination, as outlined in Table 13.1. T h e s e tables can be directly c o m p a r e d to Tables 10.9 and 10.10. Table 13.2 shows that of all 16 markets, only five are worth trading (if we use our m l e of t h u m b that says that the gross average profit should be at least three times the expected slippage and commission, which we estimated at $75). This is no different from Table 10.9, so the filter did not help any in this regard, even though it did increase the value of the average trade in nine markets. W h e n it c o m e s to t h e profit factor, the filter left us with a profit factor above 1 in twelve m a r k e t s . T h i s , t o o , is the s a m e as w i t h o u t the filter, but b e c a u s e the filter only m a n a g e d to increase the profit factor in seven markets and, h e n c e , d e c r e a s e it in n i n e , it is o b v i o u s that this s y s t e m does not benefit from this filter. It d o e s not m a t t e r that the filter m a n a g e d to lower the drawdown for nine m a r k e t s .

CHAPTER 13

Long-term Volatility Filters

247

TABLE

13.2

Performance measurements for all markets for the DBS/true range combination. After filter (modified system with exits and filter) Market Corn S&P 500 Orange juice Live cattle Lumber Coffee Japanese yen Copper Gold Eurodollar Dollar index Cotton CRB index Crude oil Canada dollar T-bonds P. factor 1.31 0.94 1.35 0.98 1.20 1.19 2.13 0.87 1.23 1.77 2.23 1.32 0.59 2.22 1.20 1.18 Avg. trade 69.73 -289.83 104.50 -6.81 114.27 153.91 1,207.82 -52.83 107.37 263.05 776.99 165.24 -505.79 555.04 62.68 258.24 2 St. dev. 1,425.07 27,234.52 2,407.67 2,293.99 3,630.98 5,558.24 9,862.45 2,224.48 2,799.42 3,001.01 6,054.09 3,923.99 4,982.93 4,772.65 2,055.76 8,695.56 Drawdown -3,556.69 -144,606.61 -9,192.10 -12,059.97 -11,792.83 -19,517.05 -13,700.98 -11,294.01 -9,348.00 -7,073.90 -7,816.56 -10,654.24 -54,603.45 -7,144.46 -7,899.86 -24,953.48

Unfortunately, the depressing results continue when we look at the D B S / A D X combination, as illustrated by Tables 13.4 and 13.5. The only thing positive thai can be said about this combination is that it did m a n a g e to make one m o r e market (gold) worth trading, but other than that it only managed to increase the value of the average trade and the profit factor for five markets (lowering them for 11) and decrease the drawdown and standard deviation for five and eight markets, respectively. T h e s e are not very impressive n u m b e r s at all and certainly not impressive enough to warrant the extra set of rules that we would have to add to the original system. T h e conclusion that can be m a d e from these tests is, therefore, that this version of the D B S system, w h e n it c o m e s down to measuring p u r e profits or robustness of the results, does not benefit from being traded together with either of the two filter techniques discussed.

THE SDB SYSTEM
For the Standard Deviation Breakout ( S D B ) system c o m b i n e d with the true range filter, the results are, however, m o r e encouraging. Of all 16 markets tested, a total of 12 and 11 resulted in higher profit factors and average profits per trade. Tables 13.6 and 13.7 show the results for this system/filter combination, which also are

248

PART 4

High-probability Filters

TABLE

13.3

Summarized differences, before/after filter for the DBS/true range combination. Differences Market Corn S&P 500 Orange juice Live cattle Lumber Coffee Japanese yen Copper Gold Eurodollar Dollar index Cotton CRB index Crude oil Canada dollar T-bonds Better P. factor -5.02% 12.99% -10.02% -3.47% 6.61% 17.98% -3.59% -6.72% -15.00% 4.77% 6.45% 5.82% -1.00% -0.47% 6.67% -4.43% 7 Avg. trade -15.74% -67.29% -26.22% -190.30% 53.80% 1344.93% -2.29% 106.33% -44.85% 5.67% 10.65% 26.01% 0.56% 1.90% 56.39% -22.43% 9 2 St. dev. -0.46% 1.44% -1.44% 0.20% 1.39% 19.90% 1.71% -1.11% -5.57% -0.58% 5.60% 2.09% -2.75% 2.37% 6.07% 0.09% 6 Drawdown 11.03% -13.86% 5.85% 12.31% -2.02% -15.69% 0.00% 11.48% 9.25% -17.53% -0.58% -1.73% -3.90% -4.80% 9.69% -0.53% 9 Better 1 2 1 0 3 3 0 2 1 4 3 3 3 2 2 1

TABLE

13.4

Performance measurements for all markets for the DBS/ADX combination. After filter (modified system with exits and filter) Market Corn S&P 500 Orange juice Live cattle Lumber Coffee Japanese yen Copper Gold Eurodollar Dollar index Cotton CRB index Crude oil Canada dollar T-bonds P. factor 1.26 0.86 1.42 0.99 1.12 1.10 2.14 0.87 1.58 1.68 1.71 1.25 0.50 2.36 1.11 1.19 Avg. trade 58.29 -732.10 127.51 -4.31 68.83 79.02 1,205.34 -54.77 237.19 238.06 525.55 131.08 -647.78 600.54 37.30 264.28 2 St. dev. 1,396.50 27,240.46 2,562.59 2,293.65 3,572.99 4,792.56 9,841.21 2,166.07 2,916.27 2,992.40 5,854.09 3,910.42 4,884.03 4,868.78 1,929.15 8,514.21 Drawdown -3,355.61 -170,608.33 -10,495.37 -11,805.89 -13,732.89 -19,299.50 -16,746.74 -11,078.56 -6,728.37 -8,849.74 -10,174.83 -11,709.41 -55,666.35 -7,508.33 -6,386.36 -25,036.36

CHAPTER 13

Long-term Volatility Filters

249

TABLE

13.5

Summarized differences, before/after filter for the DBS/ADX combination. Differences Market Corn S&P 500 Orange juice Live cattle Lumber Coffee Japanese yen Copper Gold Eurodollar Dollar index Cotton CRB index Crude oil Canada dollar T-bonds Better P. factor -8.80% 3.28% -5.06% -2.88% -0.72% 8.14% -3.04% -7.71% 9.14% -0.97% -18,38% 0.07% -16.57% 5.58% -0.74% -4.06% 5 Avg. trade -29.57% -17.38% -9.97% -157.13% -7.36% 641.83% -2.49% 113.89% 21.81% -4.37% -25.16% -0.04% 28.79% 10.25% -6.92% -20.62% 5 2 St. dev. -2.46% 1.46% 4.90% 0.19% -0.23% 3.38% 1.49% -3.71% -1.63% -0.87% 2.11% 1.74% -4.68% 4.43% -0.47% -2.00% 8 Drawdown 4.75% 1.63% 20.85% 9.94% 14.10% -16.63% 22.23% 9.35% -21.37% 3.17% 29.41% 8.01% -2.03% 0.05% -11.33% -0.20% 5 Better 1 1 0 0 1 3 0 2 4 1 0 1 3 2 2 2

TABLE

13.6

Performance measurements for all markets for the SDB/true range combination. After filter (modified system with exits and filter) Market Crude oil T-bonds T-bills Rough rice Nikkei index Natural gas Live cattle Lumber Coffee Japanese yen Copper Gold Dollar index D mark (Euro) Cotton Wheat 1.28 1.50 1.73 1.64 2.29 1.92 2.89 2.93 4.03 1.42 P. factor 4.22 1.34 2.54 3.57 1.21 Avg. trade 808.63 517.60 865.65 1,266.12 770.16 8,817.78 160.80 376.28 274.34 966.81 765.87 693.90 1,418.88 1,693.64 948.52 202.91 2 St. dev. 4,276.38 11,124.55 6,978.05 9,152.78 22,715.90 21,619.06 3,353.59 5,895.38 4,278.94 12,290.12 7,659.45 5,423.84 7,538.99 9,198.07 4,829.12 3,382.67 Drawdown -2,192.25 -19,518.81 -5,794.19 -3,587.94 -9,832.09 0.00 -9,101.18 -4,591.37 -4,998.12 -13,073.75 -5,704.59 -5,592.07 -4,611.68 -11,567.57 -2,639.46 -6,670.94

250

PART 4

High-probability Filters

TABLE

13.7

Summarized differences, before/after filter for the SDB/true range combination. Differences Market Crude oil T-bonds T-bills Rough rice Nikkei index Natural gas Live cattle Lumber Coffee Japanese yen Copper Gold Dollar index D mark (Euro) Cotton Wheat Better 17.43% 5.31% 38.21% -31.83% 99.84% 50.94% 18.80% -1.15% 48.15% -23.62% 12 P. factor 22.73% 3.77% 18.00% 182.47% -70.47% Avg. trade 8.25% 17.41% 27.93% 906.66% -87.41% 466.14% 232.39% 40.72% 212.89% -43.52% 697.70% 179.49% -2.01% -7.41% 41.47% -33.99%' 11 2 St. dev. 2.23% 1.82% 8.17% 200.80% -5.39% 235.64% 11.10% 23.63% 74.05% -4.60% 89.14% 0.87% -5.99% -2.89% 6.60% 8.45% 4 Drawdown 21.56% -1.85% 2.98% -30.69% 100.47% -100.00% 5.69% -17.61% 20.22% 23.62% -17.55% -34.66% 5.83% -4.20% -32.16% 39.44% 8 Better 2 3 2 3 1 3 2 3 2 1 3 3 2 2 3 0

comparable to Tables 10.19 and 10.20. A n o t h e r positive factor is that the number of tradable markets, as measured by the value of the average trade, also has increased from 12 to 14. Less optimistic is the fact that the risk, as measured by the standard deviation, has increased for a total of 12 markets. But as long as the risk does not increase at a faster rate than the value of the average trade, the risk-adjusted return still increases, in 11 out of 12 instances. T h e only market that really does not like this filter strategy is the wheat market, w h i c h does not m a n a g e to improve by one single measure. Given the good performance of the S D B system together with the true range filter a n d the popularity of the A D X filter a m o n g the elite system builders, one would think that the S D B s y s t e m / A D X filter combination is b o u n d to be a success. W r o n g ! As Tables 13.8 a n d 13.9 show, the A D X filter did not work this time either, improving the profit factor for only seven and the value of the average trade for only eight, out of 16 markets. C o m p a r e d to Tables 10.19 and 10.20, this system/filter combination also m a d e a total of four markets untradable, as measured by the profit factor, by lowering it to 1.0 or lower. A m o n g the markets that really were hit by this combination are rough rice and Nikkei index. For the Nikkei index, for instance, the profit factor decreased

88% -70.47 2.56 -6.150.23 13.336.82 7.31% -2.00 1.892.22% 270.95 13.27% -18.96 1.61 1.78 TABLE 13.31% -1.33 -3.427.130.84% 30.594.378.44% -28.CHAPTER 13 Long-term Volatility Filters 251 TABLE 13.690.84% 2.50 -4.76 -5.97% 15.096.14% -0.90 2.77% 6.417.11 25.38% 271.000.47 2.92% 6.06 -7.04% -5.80% -1.74 1.91% 8.677.27 191.33 -8.28 -13.21 -7. Differences Market Crude oil T-bonds T-bills Rough rice Nikkei index Natural gas Live cattle Lumber Coffee Japanese yen Copper Gold Dollar index D mark (Euro) Cotton Wheat Better P.52 2.76% -97.98 920.978.39% -27.41% 83.70% 125.81% 36. before/after filter for the SDB/ADX combination.54 -4.50% 31.15% -13.9 Summarized differences.57 416.477.51 -3.11 -1.36 2.88 -13.893.97 8.96 -64.38 1. trade 685.02 -8.38 -6.406.59 1.502.8 Performance measurements for all markets for the SDB/ADX combination.937.821.89% 2.633.60% 2.59% 0.51% -99.42 3.34% 171.58% 16.923.59% -28.77 -34.29% -156.61% -19.84 Avg.66% 1.45% 8.17% 24.15% 12.02% 5 Drawdown -3.19% 97.29% -139.56 5.07% -10.87% -8.41% -151.94 877.70% 11.69% 31.294.82 2.36 3.76 5.70 2.353.89 -5.333.85% 22.03% 10.46 -8. 2.83% 342. factor 2.03 5.91% 7 Better 1 3 1 0 1 2 1 0 1 2 3 4 2 2 3 1 — .40% -17.20% -6.69 1.55 Drawdown -1.38 2 St.005.22 3.35 8.23 0.91 2.11% 13. 4.38% -15.90 0.332.27 0.51 3.738.53 6.302.787.63% 19.89% 858.357.83% 67.012.00% -31.119. dev.798.145.004. dev.93% 12.33 20. trade -8.25 6.83% 7 Avg.09 0.78 7. After filter (modified system with exits and filter) Market Crude oil T-bonds T-bills Rough rice Nikkei index Natural gas Live cattle Lumber Coffee Japanese yen Copper Gold Dollar index D mark (Euro) Cotton Wheat P. factor -15.85% -1.114.13% 8 2 St.85 692.76 332.62% -55.918.94% 210.46% 0.67% -0.104.

therefore. factor -13.58% 22.60% -9.93% 14.24% -10.88% -25.35% 5 Better 1 2 1 1 1 2 1 1 1 3 3 4 2 2 1 0 — .00% -19.23% -6.10 Summarized differences.99% 9.06% -1.02% 3 Avg. must be put aside in favor of the true range filter.86% -48. But to m e .90% -35.16% -35.10 shows the differences b e t w e e n trading with a n d w i t h o u t the true r a n g e filter.66% -19.55% -28. -7.44% -3.36% -17.74% 2.23% -19.81% -1.31% 6. like natural gas. Table 13.11 shows the differences b e t w e e n trading with a n d w i t h o u t the A D X filter. the above results do not rule out that the A D X indicator m i g h t work very well together with other systems. Differences Market Corn S&P 500 Orange juice Live cattle Lumber Coffee Japanese yen Copper Gold Eurodollar Dollar index Cotton CRB index Crude oil Canada dollar T-bonds Better P.42% -5.29% 4.10% 18. before/after filter.08% -6.45% -3. T h e s e tables are directly c o m p a r a b l e to Table 9.44% -10.48% -8. dev.38% -4.11 with the filter. THE DIRECTIONAL SLOPE SYSTEM For the directional slope s y s t e m I d e c i d e d to s h o w only those tables depicting the s u m m a r i z e d differences.43% 2.50% -4.40% 13 Drawdown 36.51% 154. Table 13.86% -58.42% -23. to 0.09 without the filter.52% -16.24% 4.21% -6.08% -9.85% -8. trade -23.81% -23. However.00% -27.38% 10. simply b e c a u s e neither of the two filters m a n a g e d to a d d to the p e r f o r m a n c e . it does not matter that other markets.60% -10.252 PART 4 High-probability Filters from 4.14% 2.47% 3.53% -2.92% -6. s o m e of the t i m e and/or traded on s o m e other markets.61% 24.11% -7.66% -1.01% 12.39% 59.99% 5 2 St. really excel.98% -21.55% 26. W i t h too few markets benefiting in the form of lower standard deviation of the o u t c o m e of the returns and the m a x i m u m drawdown.57% 9.55% -81.10. a test like this shows that this indicator TABLE 13.04% 0.70% -8.25% 0. T h e use of the A D X indicator as a filter in systems like these always is questionable.29% -10.49% 17. at best. T h e conclusion m u s t be that the A D X filter does not do its j o b together with this system and.

62% -67.60% -20.45% 60.38% -18.74% 12. W h a t finally m a d e me decide to do it was my belief that there is no such thing as a "failed" experiment.99% 6. In fact. the results are still worth m e n t i o n i n g so that others w o n ' t spend their time on unnecessary work.13% -60.98% 12.32% -30.11 Summarized differences.62% -25.22% 2 Avg.89% 8.57% -12. Differences Market Corn S&P 500 Orange juice Live cattle Lumber Coffee Japanese yen Copper Gold Eurodollar Dollar index Cotton CRB index Crude oil Canada dollar T-bonds Better P. before/after filter. I start to wonder if this g u y really knows what he is doing.38% 48. 13.17% -9.09% -8.90% ^13.56% 17.89% 34.94% -10.99% 28. I debated even bringing up the topic.39% 16.59% -17.58% -23. . 4. B e c a u s e I h a d a pretty good idea beforehand about the results of the longt e r m filters.42% 6 Drawdown 155.02% -11.30% 4.30% -36.63% -18.85% -1.19% 8.44% -22.05% -15.69% 3.93% -25.15% 5 2 St.44% -15.70% 86.03% -2.76% 42.32% -14.51% 28.31% -58.76% 7.CHAPTER 13 Long-term Volatility Filters 253 TABLE 13.46% 295. as soon as I see a system developer speaking at a seminar about filtering w i t h the A D X indicator.72% -18.07% -13.87% ^5.30% 12.84% -12. dev.07% 6 Better 0 2 0 1 1 0 1 1 1 2 1 3 3 1 0 2 — is not robust and reliable enough for me to allow it to m a k e its way into any of my systems.65% -11. trade -37.55% 47.91% -65.13% 46.34% 67.10% 19.57%.31% 21.38% -18.20% -19.18% -2. and that even when the hypothesis leading up to the test cannot be proved.25% -62. factor -38.06% -11.

CHAPTER 14 What Makes a Trend This chapter was cowritten with Max von Liechtenstein. trading limited to the act of isolated decisions is plain gambling with or without a favorable outcome. Japan. these results are usually not based on knowledge. Paris. But underneath these colorful stories there also runs a m o r e serious history of the development of the financial system as the heart of capitalist society. and that trading is a business and a craft carried out in a work process rather than as a set of individual and isolated decisions. 1926). while truly skillful trading is a work process. graduate student majoring in economics at Uppsala University. Bruno Latour. Knut Wicksell (d. 255 . Jesse Livermore. Edgar Peters. Granted. Professor. but without knowledge. Henry Pruden. chief investment strategist. Golden Gate University. calling t h e m " r a n d o m " is an easy way out. and George Soros. Tokyo. Popular histories from LaSalle and Wall Street tell the story of tycoons and market operators like N a t h a n Rotschild. Boston. That is. For those w h o do not want to understand the markets. France and Dr. Panagora Asset Management. Dr. Ecole Nationale Superieure des Mines. T h e s e stories also provide insights into why technical analysis and rule-based trading in general are profitable if understood and treated correctly. Lund University. Professor. but on luck. there always will be a few uninformed traders w h o will m a k e incredible profits. Profit Research Center. and influenced by the ideas of Richard Werner. and Dr. chief economist. However. Sweden. which interweaves tasks like analysis and m o n e y and portfolio m a n a g e m e n t . Sweden. the individual's chances of meaningful profits are greatly reduced. San Francisco. O n e lesson we learn from the tycoons and market operators is that failure in the financial markets is based on lack of discipline and knowledge. Professor.

which is caused and created by the aggregated work processes of individuals in the economy. We try to do this as it relates to how to isolate the underlying long. it is hard to discuss a philosophy without points of reference. If an individual wants to learn the work of rule-based trading. Therefore. T h e development of general knowledge of the financial markets has taken a curious route since the 1950s. We believe the concept of w o r k to be a better starting point for discussing profitable rule-based trading. T h a t is not so. especially within the subgroup of financial analysts. However. After the 1950s. and work and c o m p e t e to get wealthy from price changes. T h e p r o b l e m with the d o m i n a n c e of the . however. While the a c a d e m i c world has been searching for general a n d abstract theories that could be derived and proven mathematically. is that trading can be understood as a form of problem solving. Philosophy answers the question why. A better perspective is to assume and study individual work processes. the part missing in the equation is h o w this wealth is created in the first place. you m u s t put in effort. We deliberately u s e the word "work. w h e r e each decision is not only a decision by itself. a n d performance. and performance. Prices are a product and not a cause of wealth. A general p r o b l e m with today's finance is the preoccupation with a s s u m i n g and studying the separate decisions of individuals m a d e at a fixed point in time. or else j u s t g a m ble and lose. he m u s t develop an ability to solve the p r o b l e m of m a k i n g m o n e y in financial markets by using the four P's of speculation: philosophy. therefore.256 PART 4 High-probability Filters To trade mechanical trading systems. Prices and v o l u m e s are therefore just materialized wealth. in c o m parison with decision. It is. practitioners have concentrated on relevant principles that have proven practical to use in their own working environments. we m u s t understand that our w o r k process contains m u c h m o r e than the actual decision to b u y or sell something as signaled by the system. procedures. T h u s . T h e advantage with the concept of work." because w h e n today's general k n o w l edge of finance only focuses on the single decision and the allocation of wealth. the academic world alone to a large extent created the general knowledge of the financial markets. influenced what was taught in higher learning institutions. work is w h a t we do by carrying out different tasks in a working role. both practitioners and academicians participated in the creation of general knowledge. but also part of a larger plan that might take a certain amount of time to w o r k through. procedures. important k n o w l e d g e regarding the financial market from traders and technical analyst simply was overlooked. and explains the value of the suggested principles and procedures in the book. principles. essential to understand the development of general knowledge in the field of finance and w h e r e we stand today. O n e c o n s e q u e n c e of this is the widespread belief that market fluctuations are a source of wealth a n d that fortunes are m a d e or destroyed by prices going up and down.and short-term trends. whereas the rest of the book is about principles. Before the 1950s. T h e emphasis in this chapter is on philosophy. T h i s meant that practitioners.

Small firms a n d value strategies have shown this. price changes in financial markets are likely to and will trend from time to time. within the field of behavioral finance. Furthermore. it is limited to the concept of decision m a k i n g by only looking at social factors. too. depends on its risk relative to the market. C A P M states that an asset price. in an informationally efficient market. B u t even if behavioral finance is a good step forward. competition will cause n e w information regarding intrinsic values to be reflected instantaneously in actual prices. and the random walk hypothesis (RWH). then they are to some degree predictable. fundamental analysis can be used to beat the market. Robert Merton. according to the scholars. a truly efficient market is one where price changes are completely r a n d o m and unpredictable. which considers price changes to be essentially unpredictable. In the early 1960s. formed by M e r t o n Miller and Franco Modigliani. T h e implication is that investors should use probability theory to m a k e rational investment decisions in r a n d o m real and financial markets. H a r r y Markowitz showed how a stock portfolio. In the past few years s o m e scholars. price changes m u s t be u n p r e dictable a n d r a n d o m .CHAPTER 14 What Makes a Trend 257 scholars is the lack of relevant practical principles adapted to the "real-world" restrictions in the practitioners' everyday w o r k processes. Therefore. r a n d o m process rather than in an unpredictable r a n d o m w a l k and that the persistence in behavior of prices and the way they change are explained by social factors like ideas and h u m a n bias. T h e scholars' d o m i n a n c e in defining the general knowledge of the financial markets started in the area of portfolio theory. . T h e attackers believe that financial asset prices change in a predictable. T h u s . could be optimized to have the lowest possible variability in periodic return. with a certain objective of expected return. T h e scholars' d o m i n a n c e of general knowledge consolidated in the 1970s with a set of option pricing m o d e l s by Fisher Black. This. It has been s h o w n that asset prices tend to overreact and that. on average. can be profitable. Because of this. and M y r o n Scholes. T h e cornerstones of m o d e r n finance are the arbitrage principle. the academic society strengthened its d o m i n a n c e over financial theory through William Sharpe's Capital Asset Pricing Model (CAPM) and E u g e n e Fama's Efficient Market Hypothesis (EMH). arbitrage is always possible. biased. Arbitrage between competitors eliminates all profit opportunities by incorporating available information instantaneously into m a r k e t prices. A n o t h e r insight from practitioners is the value of rule-based trading. therefore. for instance. with a certain expected return. such as individual wants or needs at specific points in time. T h e randomness is achieved through competition between active rational investors seeking greater wealth. If changes in asset prices follow a biased r a n d o m walk. In the early 1950s. have attacked m o d e r n finance and received strong support from traders and technical analysts in the Wall Street community. In m o d e r n finance. relative strength strategies. contradicts m o d e r n finance. because investors can borrow freely. E M H states that.

This way of looking at trading also emphasizes the need for m o n e y m a n a g e m e n t and portfolio management. E d g a r Peters describes a j u d g e w h o goes m a d . or work process. good people b e c o m e victims due to local randomness in joint action with a global structure. But that is not all. What the judge missed is that most victims are still singled out by m o r e or less purely random events. such as individual preferences and willingness to change. But because of different social factors. however. b e c o m e s something completely different from just buying a n d holding the m o s t efficient portfolio in accordance with C A P M a n d E M H ." Together. which. Aramis or the Love of Technology (Harvard University Press. This is also true in technological and financial (technofinancial) systems. forming local randomness in joint action with global structure. Together they also form a complex process of at least partially predictable trends in several different markets intervening with each other. or flow of funds) create the necessary d e m a n d to create different markets for all types of g o o d s and services. On the one hand. In other words. there are m a n y complex processes (for example the weather) that consist of several elements acting as an entity. social factors only explain half the story and m u s t be complem e n t e d by m o r e material factors. they m u s t be the result of a conspiracy. you end up being right m o r e often than not and/or make m o r e m o n e y than you lose. In his book. This stresses the need to look at any type of trading as a rational work process rather than a quest for a set of optimal decisions at different fixed points in time. we need to define a n d integrate social and material factors in the s a m e model. T h e observation that bad people cause bad things to happen. outside of the world of rational investors. implies that there exists a stable and certain structure. Patterns in the Dark (Wiley. 1999). 1996) B r u n o Latour points out that if we want to understand material changes like price and v o l u m e changes in traded assests. other factors of production. in the economy. and financial assets. even w h e n there are n o n e . b a d people cause b a d things to happen. T h u s . T h e j u d g e ' s madness stems from a simple observation: every day. they also create local randomness. This story captures our ability and desire to i m p o s e a stable order and on specific explanations for what is happening around us in our everyday life. forming a global structure that explains the creation and flow of the necessary funds for materializing these " w a n t s " and "needs.258 PART 4 High-probability Filters However. In his book. Could so m a n y tragedies simply be a string of r a n d o m events? N o . the joint . forming a global structure based on similar material prerequisites and conditions. T h e following e x a m p l e illustrates what we mean by a trend b e i n g partially predictable. In nature. The complexity of it all causes the evolving entity to look like a conspiracy. T h e rule-based approach to trading is essentially a reflection of the idea that prices move in at least partially predictable trends and by following the s a m e strategy. the social factors (the wants) and the global structure (the funds. individuals act together. according to the j u d g e . although no mastermind is behind it.

because they determine the supply (allocation) of funds. where the net present value ( N P V ) equals the intrinsic value of an asset. as we all know. etc. b e c a u s e this is the only variable left that can explain the portfolio allocation decisions in the economy. such as investors' allocation of purchasing power. b e c a u s e financial crisis a n d bubbles otherwise tend to be explained only by irrational decision-making by individuals rather than by a market failure within the system. In both cases.. this has led to difficulties in explaining financial crisis. To solve this p r o b l e m s o m e scholars inspired by K n u t Wicksell's book. because arbitrage otherwise would be possible. Kelley. Up until recently.CHAPTER 14 What Makes a Trend 259 action between local randomness a n d global structure forms an at least partially certain and stable change of events within a complex process. change in G D P is explained only by social factors such as individual preferences. changes in both transaction flows and G D P can n o w not only be explained by individuals' portfolio allocation alone. This implies that e c o n o m i c trends are partly created by us as we interact and take part in forming these changes with our predictions and explanations. T h i s assumption also applies to the markets. This joint action forms a trend of some sort in something we try to predict or explain. With an emphasis on the material constraints in society. standard neoclassic e c o n o m i c s and m o d e r n finance have b e e n occupied with explaining e c o n o m i c fluctuations and g r o w t h in the gross domestic product ( G D P ) only as a result of social factors. in turn. growth. we cannot b o r r o w freely. This implies that intrinsic values c o m m a n d or determine prices and quantities in the markets. but also by the b a n k s ' credit creation process as an underlying force inducing the wants a n d needs of individuals. and this m a k e s the inclusion of individuals' budget or i n c o m e constraints in absolute terms a great fallacy. Because the savings themselves are a result of the creation and allocation of purchasing power. which. M . But. This logic depends on the assumption that market prices and quantities always equal the value they offer the individual. Interest and Prices ( A . These factors also create trends in the markets. T h e distinction between social and material factors also is important. and fluctuations in both G D P and asset prices and why these processes to s o m e degree are predictable. M 2 . Joe Sixpack acknowledges the p r o b l e m with allocating a nonexistent purchasing power. As a consequence. Two such important material factors that affect the allocation of purchasing power are demographics a n d fiscal policies. has its origin in a too-simple m o d e l b a s e d on a barter e c o n o m y instead of a credit-driven capitalist economy. 1965) have instead started to explain the creation of p u r c h a s i n g power as a creation of credit a n d b a n k s ' willingness to lend. the reasoning is justified by an assumption that individuals always can borrow freely. in neoclassic economics and m o d e r n finance because even Mr. . This problem stems from the neoclassic e c o n o m i c s ' definition of purchasing power and m o n e y as savings a n d deposit aggregates like M l . neglecting h o w this purchasing power is created in the first place.

to build a p y r a m i d — t h e n this capital loss is really lost labor. single individuals a n d the economy as a whole can obtain the necessary funds needed to materialize all our "wants. In our economy. only in what it r e p r e s e n t s — c o m m a n d of labor and material. credit that cannot be repaid hurts society. labor and material is c o m m a n d e d by money. we have several pharaohs seeking comm a n d of labor and material by m e a n s of credit. T h e econ o m i c world of Pharaoh w a s very simple. In a m o d e r n .260 PART 4 High-probability Filters By m e a n s of credit. because it discourages b a n k s ' creation of n e w credit—the c o m m a n d of labor and material for u p c o m i n g periods. He j u s t needed to use his whip and power to c o m m a n d labor and material to build pyramids. T h u s . short.and long-term trends are separated by whether the purchasing power is created inside the market. In place of Egypt's single pharaoh. However. T h e price borrowers have to pay to c o m m a n d labor and material is annual interest. Nevertheless. because it creates d e m a n d a n d transactions in the economy. the life cycle of m o n e y and credit is such that it is born when a b a n k issues a loan and is put to death as the person w h o borrows it pays it back. T h e importance of credit in understanding markets can be illustrated by comparing our capitalist society with the barter economy of ancient Egypt. however. the quantity of which will only change as m o r e gold or other precious material can be d u g up from the ground. at first. products are c o n s u m e d — h o u s e and cars are bought. is used. and m o n e y is spent abroad by foreign customers." as explained by behavioral finance. or money. there is no solitary pharaoh with the power to c o m m a n d labor and material by word alone." All our "wants. because they can survive only by new loans. " i m m o r t a l " asset. it does not really matter h o w this money. capitalist society. m a c h i n e s are created. if credit c o m m a n d s labor and material into something unproductive—for e x a m p l e . is a fabulous agent both for borrowers and others in the society. banks often require collateral for their lending. managed. and s o m e profit. our m o d e r n world is filled with thousands of "pyramids. In a barter economy. T h a n k s to the credit creation process.000. That is. m o n e y is viewed as a tangible. Further. repayment. There is neither value nor power in m o n e y itself. In our economy.000 days of h u m a n labor and all that was created after a period of 20 years was a frozen asset. or credit. T h e s e restrictions have a functional significance because they force borrowers to m a k e sure that their c o m m a n d of labor and material will be productive e n o u g h to generate income that pays for interest. credit. or c o m e from outside flows. At first. To increase the likelihood for the debt to be amortized (or put to death). credit constraints restrain pyramid building. will not cause the market to trend by itself. as investors speculate on margin. By m e a n s of credit. and fed with raw materials. T h e pyramid of Cheops c o n s u m e d 600." and the invisible ones are the m o s t dangerous. . T h u s . the ability of a market to trend is dependent on how m u c h a n d what type of credit it can attract. Instead.

and a low g r o w t h in consumer prices. w h i c h m e a n s a l o n g trend of m o n e y flowing out of t h e real m a r k e t s a n d into (allocated to) the financial m a r k e t s . coupled with a very low inflation in the real markets. services. All in all. an increase in d e m a n d is mainly m e t by an enlarged quantity supplied. c o m m a n d for labor and material) during the period that followed. b r o k e r s ' loans on the N e w York Stock E x c h a n g e alone increased by $5 billion. it caused the lenders substantial losses. credit could have been used for m o r e productive p u r p o s e s . Financial market transactions involve asset stocks. w h e r e the supply is m o r e limited. r e s u l t i n g in a d r a m a t i c increase in the ratio of savers to s p e n d e r s . A stock-exchange speculation with ever-increasing prices. "Towards a N e w M o n e t a r y P a r a d i g m " (Kredit und Kapital. in the real market. t h e r e has been a s h a r p shift in d e m o g r a p h i c s . Once again. and factors of production. Quantities of stocks are m o r e fixed in c o m p a r i s o n w i t h production flows. Transactions in real markets for g o o d s . such as that before the crash of 1929. In two years. as during the margin lending of the 1920s. Transactions in real markets are an e c o n o m i c activity measured as GDP. we o n c e a g a i n s p e a k of the " n e w e c o n o m y " as we e x p e r i e n c e a p e r i o d of h i g h g r o w t h in GDP. However. which only served to inflate stock prices. But even m o r e seriously. as central banks fight expected inflation and private banks d e m a n d a higher risk p r e m i u m for m o n e y lent for speculative purposes. Transactions in financial markets are a noncreative activity that is not a part of GDP. this has created a situation of high e c o n o m i c growth.CHAPTER 14 What Makes a Trend 261 In his article. T h e " 1 9 2 9 p y r a m i d " c o n s u m e d credit in an uncontrollable manner. b u t w a s not. a b o o m i n g stock m a r k e t . A lot of these savers have invested their s a v i n g s in m u t u a l funds. result in an increase of price. Today. this would result in higher interest rates. A d d to this that several governments throughout the world have both lowered their taxes and started to b u y b a c k h u g e a m o u n t s of b o n d s issued in the 1980s (the United States is supposed to have bought b a c k all its debt by 2 0 1 3 ) . is an invisible pyramid. This also transfers m o n e y that would have been spent in the real markets to the financial m a r k e t s . whereas in the financial market. the interest rates are b e i n g h a m p e r e d by competition a m o n g the banks. For one thing. involve income or production flows. total credit creation m u s t be subdivided into a real and a financial part. when changes in asset prices or G D P are explained or forecast. A t the s a m e t i m e . by necessity. Normally. but with highly inflated financial markets. Therefore. the b o o m i n g markets have m a d e banks lend heavily for speculative p u r p o s e s . Thus. an increase in d e m a n d will. w h i c h decreased production and e m p l o y m e n t (i. t h u s i n c r e a s i n g G D P w i t h o u t i n c r e a s i n g inflation. T h e pyramid caused a great n u m b e r of problems. Its stones are the supply of funds and h u m a n sentiment. . 1997) Richard Werner points out that there is a fundamental difference in outcome b e t w e e n transactions in the real and financial markets.e. This is because new and improved technology has made it p o s s i b l e t o m e e t t h e i n c r e a s e d d e m a n d for g o o d s and s e r v i c e s w i t h i n c r e a s e s i n productivity.

sentiment. then an increase in collateral values alleviates credit constraints. T h e paradox is. The creation-of-funds indicators include central b a n k s ' open market operations and private b a n k s ' loans to the total economy.262 PART 4 High-probability Filters governments decreasing outstanding stock of b o n d s . A very good sentiment indicator in the commodity futures markets is Commodity Futures Trading Commission's (CFTC) C o m m i t m e n t of Traders Report that monitors the positions of commercial and noncommercial traders. and the fact that there is no inflation to fight in the real economy. such as the margin lending of the 1920s or as is seen today. that while individual banks assume asset prices to be independent of their actions. . secondary offerings. Flow-of-funds analysis is concerned with trends in mutual fund cash positions and other major institutions such as pension funds and insurance companies. This school of thought in general. and a shorter-term sentiment indicator. as the current use of margin by larger market participants oftentimes can reflect a m o r e short-term use of what normally is d e e m e d stable or m o r e long-term money. and rule-based trading in particular. asset price inflation occurs. In times w h e n banks lend heavily for speculative p u r p o s e s . They monitor the emotions or expectations of investors moving from one extreme at a market bottom to another at a market top. subdivided into financial and real sectors. If banks extend loans with real or financial assets as collateral. however. T h e use of margin (or leverage) borders is both a long-term creation-of-funds indicator. Technical analysis is increasingly recognized for explaining price predictability based on a study of investors' actions by charting the development of traded volu m e and price of a financial asset. Other flow-of-funds indicators are new equity offerings. It is assumed that different groups of investors are consistent in their actions at major market turning points. T h u s . and market microstructure indicators. for example. Banks tend to extend loans with real or financial assets as collateral. which lead to a continued increased d e m a n d for financial assets. which are normally a source of cash on the b u y side. Technical analysis can be broken down into four essential areas: creation-of-funds. are two groups that are considered to be wrong more often than not at or just prior to major turning points. The sentiment indicators monitor the different market participants such as mutual funds and floor specialists. flow-of-funds. Advisory services and newspapers. perceive market behavior to be both deterministic and random. as the use of margin is a form of borrowing. and c u s t o m e r s ' free balances. although c o n s u m e r prices m a y hardly rise at all. T h e creation-offunds and flow-of-funds analyses represent the m o r e stable trends that are reflected in the sentiment and market microstructure analysis. Flow-of-funds indicators analyze the financial position of various investor groups in an attempt to measure their financial capacity for buying or selling stocks. rising asset prices cause extended asset-related loans. w h e r e an increase in collateral values alleviates credit constraints. invisible pyram i d s of extreme changes in asset prices always d e p e n d on increasing asset-related lending from the banks.

Dr. a p p l i e d either directly on price or to the n o w familiar O B V indicator. induce the necessary funds. In a consolidating phase. the d e m a n d and supply of assets are in a relative equilibrium. he m u s t sell a p o r t i o n of his h o l d i n g s . can be used to explain the rise and fall of financial asset prices. However. We have expanded Dr. W h e n e v e r the value of t h e leveraged trader's h o l d i n g s s h r i n k s . are strong enough to entice buying. is that the O B V indicator. Pruden argues that the adoption-diffusion m o d e l captures how innovations over t i m e are adopted in a techno-financial system. time. w e i g h t s e a c h day's price action w i t h its t r a d i n g v o l u m e a n d by d o i n g so b e c o m e s a better m e a s u r e of the relative a m o u n t of participation b e h i n d t h e trend. In essence. based on behavioral finance. to approxim a t e a society's short-term willingness to allocate m o n e y (flow of funds) toward . the g r e a t e r t h e a m o u n t of strong h a n d s a n d the larger the b a s e from w h i c h p r i c e s can rise.1 illustrates what this m a y look like. while at the s a m e time the social factors. An idealized market cycle begins with a consolidating phase. W h e n the p r o p o r t i o n of stocks in strong h a n d s is a b n o r m a l l y high. January 1999).t e r m m o v ing a v e r a g e . known as either a distribution or accumulation area.and short-term trends of changing supply/demand relationship in the specific market.CHAPTER 14 What Makes a Trend 263 We perceive the material factors to be captured by the creation-of-funds and flow-of-funds indicators. T h e r e f o r e . breadth. and rank these indicators' relative importance and h o w they affect different markets at different points in time. Pruden's thoughts to e n c o m p a s s m o r e material factors when we explain the existence of long. "Life Cycle Model of C r o w d Behavior" {Technical Analysis of Stocks and Commodities. a highly leveraged trader has a w e a k h a n d . the credit creation process together with other material factors. T h e S-shaped cumulative normal distribution curve in Figure 14. S h o r t . That is. the m a r k e t is in a strong p o s i t i o n . volume. w h i c h forces the m a r k e t prices b e y o n d w h a t the e c o n o m y can afford in the long run. because of the difficulty to m e a s u r e . In his article. In an accumulation (distribution) phase. we often m u s t use what we call market microstructure indicators. r a t h e r t h a n directly to the p r i c e . 1 . Henry Pruden describes how an a d o p tion-diffusion model. t h e longer the p e r i o d of a c c u m u l a t i o n . such as demographics and fiscal policies. as illustrated by the familiar bell-shaped normal distribution curve. with the trend creation process still in mind. and volatility are market microstructure indicators. such as all our wants and needs. T h e benefit of a p p l y i n g this t y p e of analysis to the O B V indicator. as in F i g u r e 1 2 . T h e r e f o r e .t e r m t r e n d s are c a u s e d b y i n v e s t o r s ' s p e c u l a t i o n o n m a r g i n . It also is possible. as s u g g e s t e d by its n a m e . while the social factors are captured by the sentiment indicators. But asset prices can only advance if investors have the purchasing power and courage to buy. quantify. all technical analysis indicators that are based on price. In real life this can be a p p r o x i m a t e d with the help of a l o n g . assets pass from weaker (stronger) to stronger (weaker) h a n d s .

This m e a n s that. for instance. For you as a trader. a certain t y p e of investment vehicle by looking at the relative p e r f o r m a n c e of the investment alternative in question as c o m p a r e d to the interest rate. unfortunately.264 PART 4 High-probability Filters F I G U R E 14. but faster than the increase in the p r i c e for wheat. the price of c o m is increasing. W h e t h e r the interest rate is currently increasing or decreasing does n o t matter in this analysis as long as it does so at a rate different than that of the alternative investment. B u t to do better than j u s t O K — t o be competitive—you also must do better t h a n y o u would have d o n e using any other investment alternative or method. Returning to the analysis of a trend. you are doing O K . For e x a m p l e . in this particular case. y o u should look for situations to short the w h e a t a n d g o long the c o m . it is reasonable to a s s u m e that investors prefer to invest in b o n d s rather than wheat. if the interest rate is decreasing m o r e slowly (bond p r i c e s m o v e up) than. T h a t is w h a t the rest of this b o o k is all about. W h e n sever- . it is seldom as simple as j u s t isolating o n e trend from all the others a n d then riding that wave. and a reasonable p r e m i u m for the risk you are assuming. the rate of inflation. this m e a n s that as long as you can keep a rate of increase in your investment or trading account that is higher t h a n a combination of the interest rate at w h i c h the invested m o n e y is a s s u m e d to have b e e n borrowed.1 The cummulative normal distribution of a dependent variable. but also that they favor corn over interest rates.

To isolate both the long-term and short-term shifts in the trend caused by. the longer the lookback FIGURE 14. a larger trend naturally consists of several smaller ones. In Figure 14. for instance. sentiment shifts a n d the a m o u n t of trading taking place on margin. forming the well-known halfway flag of technical analysts.CHAPTER 14 What Makes a Trend 265 al different ideas. In this case.2 and 14. T h e theoretical explanation behind these shorter-term trend and m o m e n t u m changes is illustrated by Figures 14. preferences. however. we can see h o w this long-term trend really consists of two distinctly different states of market behavior. m a k i n g life difficult for any trader.3. although there also is the underlying assumption that the material factors are m o r e long t e r m than the social factors. In this case they also are separated by a transition phase. we do not concern ourselves with whether the trend is driven mainly by material or social factors. the m o r e stable and long-term the trend. different moving averages applied directly to the price. To c o m e to grips with this. In this case. again on the assumption that the longer the average. m a n y analysts also use one or several oscillator-type indicators that could be applied either directly to the price or to any other indicator that attempts to capture the material or social factors m o r e explicitly. m a n y technical analysts try to isolate different trends from each other using. for instance. and n o r m s interweave.2 Two normal distributions intervening with each other. .3.

the higher the degree of uncertainty regarding the material factors and the m o r e unpredictable the markets. it could be argued that these stages of unpredictability coincide with periods of major shifts in the long-term.3 The cummulative effect of two normal distributions phased together. the m o r e significant and reliable the shift is assumed to be. because it then is a s s u m e d to be the result of m o r e material factors. as these players scramble to figure out w h e r e the larger players are heading. M o r e formally and in accordance with our reasoning so far. such as the creation and flow of funds and/or sentiment shifts a m o n g the larger players. stable trends and a high amount of margin trading (short-term credit creation and flow of funds) a m o n g the weaker hands. period for the oscillator. such as the pension funds a n d insurance c o m p a n i e s .266 PART 4 High-probability Filters F I G U R E 14. because these shorter credits inseminate a larger degree of fear and panic into the mixture of social factors affecting the traders. . the shorter the credits. That is.

Granted. single individuals. and for the p u r p o s e very unsophisticated. it is full of people. It could be other traders. suitable for the entry/exit triggers we currently are working with. the right framework for analyzing it a n d a set of generalized rules. regular workers. a n d friends. j u s t as you d o . Whatever your preferred time perspective it never hurts to get an understanding of w h i c h underlying forces are at play to create a n d maintain the moves we are interested in and are trying to capitalize from. But instead of being full of oil. you can look at it this way: Pretend that the e c o n o m y (or the part of the e c o n o m y that you are interested in) is a super tanker h e a d i n g in a certain direction on the ocean. W h a t it all c o m e s down to is to increase your chances to place a winning bet. both in the short and the long run. a m o n g w h o m you are one. farmers. b r o kers. we can b e c o m e good e n o u g h in recognizing the general differences between an uptrend and a downtrend. this is as good as it gets. and all these people's families. For the long-term perspective. T h e n you " o n l y " will n e e d to be able to k n o w at w h a t scale you are looking so that you k n o w h o w to distinguish one seemingly r a n d o m m o v e from another. analysts. the technical analysis and statistical tools we are using are very coarse a n d clumsy. B u t however clumsy the tools. and if the market seems to be in a state of predictable change or not. We n e e d to understand that the predictability and profitability is there if we could only find a way to m e a s u r e it a n d transform it into dollars m a d e . 267 . w h o in one way or another have an interest in what is going on. as those in a m e c h a n i c a l trading system. relatives.PART FOUR A Few Final Thoughts About Part 4 This section started out by taking a closer look at how to filter out those situations w h e n the m a r k e t s seem to greet us with a particularly friendly environment. a n d at least with a proper understanding for how a trend gets started.

. You do not mm a super tanker on a dime. like an individual stock. Place a long-term bet that tries to catch the beginning of the move. W h a t would you bet on: that it will continue in the same direction for another couple of miles. or minutes. has decided that the ship is heading in the w r o n g direction. pretend that you. Generally speaking. you are m u c h better off going with the current flow of events. Even if he puts the engines into full reverse. To be sure. or that the next thing you know. as indicated by high short-term volatility levels. instead are standing on the shore. you m u s t k n o w what to look for. On the financial markets. w h i c h s u m s up the p u r p o s e of this book and all the hard work w e ' v e d o n e leading up to and including the "optisizing" process in Part 5. it will have turned around? Of course. But they are just that—exceptions to the rule. Or to continue the analogy.268 PART 4 High-probability Filters Pretend further that the person in charge of this ship. the sheer mass and m o m e n t u m of the ship force it to continue in the original direction for many miles to c o m e . in the short run strange things happen every now and then— exceptions to the rule bring short-lived fame and fortune to a lucky few. but be prepared to get out as soon as it is clear it is not heading w h e r e you expected or w h e r e the situation is less clear. Wishful thinking has nothing to do with where the markets are heading or the o u t c o m e of your trading. holding exactly the s a m e information as everybody on board. Now. looking at this ship on the horizon. w h e n you can see bad weather approaching or w h e n mist or fog m a k e s it difficult to assess the situation. Not even if everybody on the ship agrees with his decision and wishes to turn around right away will the ship be able to turn around any sooner. not even if you close your eyes hard and wish. you are better off betting it will continue in the s a m e direction. To do that. or even any specific smaller part of it. or that he can see trouble ahead. A n d the s a m e goes for the economy. how to measure it and h o w to m a k e the most of it. as a trader. that is when your filters should signal to stay out. and decides to change course. for instance Alan Greenspan. or days. this m e a n s placing your trades in the s a m e direction as the trend is heading.

But if your account size is only $50. Because in a well-working system you really have no idea about whether your next trade will be a winner or a loser. But first I need to point out that there is no way I would have been able to c o m e up with the math a n d the logical reasoning behind this section myself. always and for each trade. a 2% loss equals only $1.PART FIVE Money Management and Portfolio Composition Now that we have finished building the engines. you will be best off.000.000 and you have decided that if the trade goes w r o n g . if your account size going into a particular trade is $100. let us move over to the intricate subject of putting together a gearbox and a chassis to get our babies going. w h o also coined m o s t of the terminology I use in this section.000. you will be willing to lose $2. I have merely taken what is in these b o o k s and tweaked it around a little bit to suit my needs. in the long run. 1992). w h e r e a s a $2. For this I rely solely on Ralph Vince. Vince's b o o k s . you can stand to lose 2% of that amount. in his b o o k s Portfolio Management Formulas (John Wiley & Sons. W h e n it c o m e s to the gearbox (the m o n e y m a n a g e m e n t ) .000 loss equals a fraction of 4 % . it is quite simple really. 269 . That is. If you are interested in learning m o r e about what is covered in the following section. 1990) and The Mathematics of Money Management (John Wiley & Sons. to risk the same fraction of your account. d o n ' t ask m e — g o buy Mr.000.

T h e trick is to put together a portfolio of systems and/or m a r k e t s with correlations as low as possible to each other. w h e r e a few m a r kets still zig w h e n others zag. y o u will not be able to practice the very s a m e techniques that already are in practice w i t h professional m o n e y m a n a g e r s . of your account going into each trade. A n o t h e r important point is that although these techniques add m o r e to your b o t t o m line than your actual systems. however.CHAPTER 15 Money Management The w h o l e idea of m o n e y m a n a g e m e n t is to always invest that fraction. it is fully possible to have individual market/system c o m binations that have a negative mathematical expectancy (a profit factor below 1 a n d a negative value for the average trade). your s y s t e m s m u s t have positive m a t h e matical expectations in the first place. w i t h the p r o p e r m o n e y m a n a g e m e n t you can w o r k wonders. or entry and exit triggers.o n e system. considering such factors as account size. If they just have that. As soon as you trade a portfolio of m a r k e t s and/or systems. they must have a profit factor above 1 and a positive value for the average trade to be worthwhile. T h a t is. a n d dependency a m o n g trades. flat times. or percentage. m o s t suitable to you. if they are only marginally positive. growth rates. rather than in flat dollar rates. it is w o r t h m e n t i o n i n g that this only h o l d s true for the special case of one m a r k e t . Otherwise. they do not p e r f o r m miracles. expected worst-case trades. w h i c h m a y raise the positive expectancy for the trading strategy as a w h o l e . drawdowns. T h e negative expectancy markets/systems help to hold up the b o t t o m line while you wait for the positive 271 . On a separate note. To be able to do this—to " o p t i s i z e " your tradi n g — y o u m u s t start to think in fractions a n d percentages. generating a small but steady profit in the long r u n . b u t with the zigging greater than the zagging. For these techniques to work.

the H P R is 1. you have 1. In the world of trading. T h e a m o u n t of risk you assume has nothing to do with which type of vehicle you are trading. respectively. with H P R s of 1. that for money m a n a g e m e n t purposes.000. loss.155 (1. the m o r e you are likely to gain. The T W R is the geometrically c o m p o u n d e d H P R for all trades. we immediately must address the fact. if a trade ends up with a 5% loss.155). A T W R of 1. We need to understand the t e r m holding period return ( H P R ) . plus one. F u r t h e r m o r e . the T W R is 1.155 means that after the two trades. w h a t might be devastating for some may only be a freak occurrence but still a fully expected outcome for others. T h e H P R is the percentage gained or lost for each specific trade. From this it follows that to fit as m a n y outcomes as possible into the boundaries of what could be considered expected. the only thing that is certain in trading is that we all will be blown out of the markets sooner or later.5 % ) = 1 + ( . It is just a matter of time.05 and 1. to use the old start-a-new-business cliche. ." If it is not. 5 % (1. Potential gain is not a straight-line function of potential risk. it is not true that the m o r e you risk. eventually you will die from being hit by a truck.272 PART 5 Money Management and Portfolio Composition expectancy markets/systems to m o v e b a c k in line a n d start performing again. It is as simple as that. On the other hand. m o n e y m a n a g e m e n t ( m a k i n g a business plan) is to h o p e for the best but prepare for the worst. 0 5 ) = 0.155 = 15. For e x a m p l e . "Potential gain is not a straight-line function of potential risk. If it h a p p e n s to you.05 X 1. If you are caught on the w r o n g side of such a m o v e you will suffer a severe. if a trade ends up with a 5% profit. the H P R is 0. which m e a n s that we are using multiplication rather than addition.155 times your original equity.05 (1 + 5% = 1 + 0.05 = 1. Now. For example. but that's j u s t the n a m e of the g a m e .155 .10 = 1. let us consider the next point in the list above. however. or your total profit is 1 5 . Still.05). you m u s t have considered and treated the whole subject of trading very defensively. no matter h o w m u c h they have tried to protect themselves. Diversification does not necessarily have to reduce drawdown. I ' m sorry.5%). because there will be no way out even if you had a stop loss at 1.350 to 1. because these moves happen. Of these four things.0 . Or to use a simple analogy: if you can only die by from being hit by a truck.95 (1 + ( .10. for instance 1. w h e n price does not behave in a rational m a n n e r it means that if it goes from. for some the catastrophe might c o m e first thing tomorrow. T h e second t e r m we must understand is terminal wealth relative ( T W R ) .95). Or.1 = 0. if you have two winning trades. then what is it? To answer that question we first m u s t define a few t e r m s . it does not have to stop at every tick in between. if not totally devastating. Before we go on there are a few things you must fully understand: • • • • Price does not behave in a rational manner.345.

If we only decide to optimize in relation to the speed with which we wish to m a k e the account grow.000 per unit traded). But. that we currently have $100.0975 = —9. you should trade m o r e and m o r e units. T h e n your T W R would be 0. stocks. For m o s t traders.5 / 10.75%). or from $1. but slightly less than that.) As your account grows. on average. because we have no control over whether the next trade will be profitable or not. but do have control over the quantity we can put on. the higher the ƒ the m o r e units we can trade.000 to $10. two 5% losses in a row d o n ' t equal a total loss of 10%.6 / 10.000 X 0.000 to $100. given our expected worst-case scenario and current account status.9025. if you can m a k e your account grow geometrically. being right about the direction of the move seems to be m o r e important than actually m a k i n g any m o n e y out of it. To understand this better.000.000 to $1. 7 5 % (0. T h u s . consider instead if the two trades above t u m e d out to be two losers of 5% e a c h — w h i c h was the fraction you were willing to risk. the s a m e a m o u n t of time [trades] to take it from $10.000.000 in our account. This clearly demonstrates the importance of always risking the s a m e fraction of your account. It has always a m a z e d me how most traders simply do not concern themselves with this.000. not to mention m a k i n g the optimal amount of money. but instead put all their attention in whether or not they m a n a g e d to call the right direction of the move. For instance. as it will to take it from $100.000. it will take you.CHAPTER 15 Money Management 273 Note that a profit of 5% and a profit of 1 0 % do not add up to a profit of 1 5 % . T h u s . etc. we then calculate the "optisized" n u m b e r of units (stocks or contracts) we can put on for any given trade. and that the worst-case scenario is a 1 0 % loss (which in this case happens to equal $10. your equity would have decreased to 0. This is because you were able to use the extra m o n e y you m a d e in trade one going into trade two. In this case. w o u l d n ' t it be better to focus our attention on "optisizing" that amount. suppose we already have calculated ƒ to be 0. . and at the same time it keeps you in the g a m e for m u c h longer during drawdown periods. You were able to reinvest your winnings. With this value for ƒ in our hand.9025 — 1 = —0. once we have a system that we can trust and with a positive expectation in the long run? T h e third t e r m we m u s t understand is the v a r i a b l e ƒ which is simply the fraction of account size. we can put on 5 units (100. Two major benefits of working with a fixed fractional m o n e y m a n a g e m e n t strategy.000 = 5). T h e variable ƒ can be optimized in several different ways: what might be optimal for you might not be optimal for m e . is that your account m e a s u r e d in dollars g r o w s geometrically.6.5 (we will see how to find the optimal ƒ value shortly).).000 X 0. but slightly m o r e .000. we could have put on six units (100. where you always reinvest your profits and losses. it is called the optimal ƒ. to keep the relative a m o u n t risked at a constant level.000 = 6).9025 times the original equity and your total loss is 9 .000. because the strategy forces you to scale back and trade fewer and fewer units (contracts. Had ƒ instead b e e n 0. as represented by your expected worst-case scenario for that particular trade. (By the way.

which tells you if your system has m o r e or fewer streaks of consecutive w i n s or losses than w h a t could have been expected if the trades had been randomly distributed. Had you done it the other way around. even if your systems seem to exhibit some serial correlation or dependency. e x a m i n e if there are any type of serial correlations or dependencies a m o n g the outcomes of your trade. the dependency needs in fact to be there. although in hindsight you would have been better off trading it according to some d e p e n d e n c y strategy. but also continues to do so in the future as well. while for a type I error it would only be a slower growth of equity. the penalty for m a k i n g a type II error would be a complete wipe out. rejected the hypothesis when you in fact should have accepted it (a type I error). This also relates to what we learned about the system building process and the importance of keeping it as simple as possible. O n e way of examining for dependency is to do a runs test. For instance. you still would have m a d e money. we should have formulated our entry and exit rules in such a way that the information the market w a s giving us. As we learned in Part 3. not proof—and for a m o n e y management strategy based on dependency and correlations to prove itself m o r e profitable than any other type of m o n e y m a n a g e m e n t regimen (or even just profitable). would have been dealt with at that time.274 PART 5 Money Management and Portfolio Composition To be sure. T h i s illustrates the paradox that with a system working optimally. if a system has any dependencies or correlations a m o n g the outcomes of the trades. albeit at a slower rate. That is. and traded it according to a fixed fractional m o n e y m a n a g e m e n t strategy. A n o t h e r way is to look at the linear correlation coefficient. you probably still are m u c h better off ignoring it as long as the evidence is not truly and really overwhelming "beyond any reasonable doubt. T h e above reasoning also applies to the special case of scaling back faster in times of drawdown than what is dictated by the fixed fractional strategy. there is no way to k n o w if the next trade will be a w i n n e r or a loser. we are curve fitting the system and making it less likely to . T h u s . This method results in a system that not only trades profitably on historical data. your m o n e y m a n a g e m e n t strategy will be suboptimal. If it is not. By adding rules to build away the (presumed) dependency. You form a hypothesis that states that the dependency is in fact there and. But what if it turns out that the d e p e n d e n c y were not there? T h e n you will have c o m m i t t e d a type II error (statistical jargon for accepting a hypothesis that in fact should have been rejected) and will lose m o n e y from trading the system either t o o aggressively or not aggressively enough. there are other ways of c o m i n g up with an optimal n u m b e r of units to trade. by signaling dependency. therefore. there is information there that we are failing to exploit at the system building stage." This is because the evidence is just that—evidence. based on as few rules as possible. To understand this. consider the following example: suppose you consider trading a system that shows evidence of dependency a m o n g the outc o m e s of the historical trades. Furthermore. the system should be traded with this in mind.

or you could take the geometric m e a n raised by the n u m b e r of trades (more about this shortly). In the previous e x a m p l e we really didn't state how m u c h we risked or lost in dollar terms. 0 0 0 .D $ 3 ) In cell F 6 : = E 5 / ( C $ 3 / . type in t h e following formulas and values into an empty spreadsheet: In cell B 3 : 100.000. . — P F I T = the profit or loss. .7 . 0 0 0 In cell D 3 : 0 . with the sign reversed = $5. 5 In cell B 5 : 5. T h e following formula tells us that the H P R is 1. .P F I T X 100 / W C S .1 0 .25: HPR = 1 + f x ( . W h e t h e r it's there or not.1. equal to a profit of $5. w h i c h is akin to the average H P R or growth factor per play.B 5 / C $ 3 ) (then drag down to fill all the cells C5 to C14) I n cell D 5 : = C 5 In cell D 6 : = D 5 * C 6 (then drag d o w n to fill all the cells D6 to D14) I n cell E 3 : = C O U N T I F ( B 5 : B 1 4 . the geometric mean. 7. Therefore.4 . on the trade. 0 0 0 . . I also have added another very useful variable.000 p e r contract. 0 0 0 and -3. you could either multiply all the H P R s .D $ 3 ) (then drag down to fill all the cells F6 to F14) W h e n you are done. but just stated the percentage value as a result of the formula . To arrive at any final T W R . a n d W C S = the worst-case scenario (always a negative n u m b e r ) = —$10. In cell C 5 : = 1 + D $ 3 * ( .000.000.5. Notice that the only difference between this and the previous formula for the H P R is that we now multiply the percentage m a d e on the trade with the factor ƒ and the worst-case a m o u n t is the a m o u n t risked.000. To do this in Excel. " < > 0 " ) In cell E 5 : = D 5 * B $ 3 (then drag d o w n to fill all the cells E5 to E l 4 ) I n cell F 3 : = D 1 4 ( 1 / E 3 ) I n cell F 5 : = B 3 / ( C $ 3 / . To the e x a m p l e above. per contract.2 .000. Now.000 Type in the following n u m b e r sequence (one below the other) into cells B6 t o B 1 4 : 2. 6. build your system with as few rules as possible that m a k e sense to you.1 0 . then trade it with a fixed fractional m o n e y m a n a g e m e n t strategy.CHAPTER 15 Money Management 275 work in the future.000. T h e price of not doing so is simply too high.000 I n cell C 3 : . 2. A .P F I T / WCS) w h e r e ƒ = the value we are using for f = 0.000. 0 0 0 . my recommendation is not to bother with the dependency or serial correlation at all. suppose that the trade in the previous example ended up in a profitable m o v e of 5 % . or a total profit of $25.000 or 2 5 % . your spreadsheet will resemble that in Figure 15.

276 PART 5 Money Management and Portfolio Composition FIGURE 15.2).1 How to use Excel to calculate your holding period returns and terminal wealth relative. if you change the value in cell D3 to 0 . You can manually change the value in cell D3 between 0 and 1 to try to find the highest value for T W R in cell D 1 4 . FIGURE 15. . T h e value in cell D3 that c o r r e s p o n d to the highest value in cell D 1 4 is the optimal f for the system.0889 (if y o u ' r e displaying four decimals). u n d e r the Tools m e n u (Figure 15. 1 . the value in cell D 1 4 changes to 1. A n o t h e r a n d quicker way is to use the Solver Add-in function in Excel.2 Use the solver add-in function in Excel to calculate your optimal f. For instance.

such as not letting the m a x i m u m d r a w d o w n or the standard deviations of the individual outcomes grow past specific values.52%. Figure 15.19%. This is a fantasy system similar to that depicted by the performance s u m m a ry in Table 12. 1. which you can use for single market/system combinations. In this case. eventually we also will be able to apply it to a market w h e r e it will be profitable as well. you should not risk m o r e than 3 1 % of your total equity going into the trade. not taking into consideration any costs or the fact that we cannot trade fractions of contracts. Consider the following expectancies for a system you are thinking about trading real-time: • • • • T h e percentage of winning trades are over 7 2 % . This is how I prefer to do it. For every losing trade you lose.36. For every winning trade you win. Besides. As long as we do that. seldom will you be able to trade your system at its optimal ƒ level anyway. just saving a little in case you happen to get into a drawdown.CHAPTER 15 Money Management 277 N o t e also that in this particular case we a s s u m e d the worst case to be constant and equal to the worst case in our sequence of trades. with a final T W R of approximately 1. Because the data that is exported into Excel has all outcomes measured in percentages rather than dollars. as is the case in the example above. 1.18. when it c o m e s to trading with a fixed fractional m o n e y m a n a g e m e n t . This is the optimal ƒ for this system w h e n we work under no other constraints than trying to m a x i m i z e the growth rate. or letting the stops of your system (as they pertain to each individual trade) determine an individual worst-case scenario for each trade. this is also how you should look for your system's o p t i m a l ƒ R e m e m b e r that we first and foremost want the system to be well w o r k i n g rather than profitable—we want it to catch a certain type of market m o v e as efficiently as possible. which you should m a k e up for easily considering the high percentage winning trades. including calculating the standard deviation of all o u t c o m e s and setting the worst trade to equal a certain distance from the average trade. (We will learn how to build a bar-by-bar export function for multiple market /system combinations and bar- . on average. I have used the trade-by-trade export function from Part 1. the calculations look a little different. at first glance it looks like you should bet almost your entire equity each time. A n d j u s t as we do not take slippage and c o m m i s s i o n into consideration when we are looking for the optimal variable setting for a system (so we do not c o m e up with a suboptimal solution). as will b e c o m e evident shortly. There are several other ways to do this.2 also hints at the fact that you could perform m o r e complex optimizations by adding a few other constraints. With a system like this. 3 1 . what could go w r o n g ? T h e losers are considerably smaller than the winners a n d with plenty m o r e w i n n e r s than losers.12. This m e a n s that for any particular trade. on average. T h e profit factor is 3. In this case. the optimal ƒ equals approximately 0 .

05. the optimal ƒ for this system turns out to be as high as 0.325 times your original equity. which generates a T W R of 3.1) * 100). corresponding to a total profit of equally w h o p p i n g 3 3 2 . you would have m a d e a w h o p p i n g 3. If your spreadsheet looks like that in Figure 15. as shown in Figure 15. you must insert a few empty rows at the top of the spreadsheet and a few empty columns directly to the right of the "Profit" column. 4 0 0 % ((3. .4 shows the T W R you would have been able to achieve depending on what ƒ level you traded. In cell G 5 : = 1 +G$3*(-(F5/100)/F$3) (then drag down to fill all the cells G5 to G219) In cell H 3 : = H 2 1 9 ( 1 /COUNTIF(F5 :F219. this is a fantasy system that you can substitute for any system of your choice). you can then type in the following formulas or values: I n cell F 3 : =MIN(F5:F219)/100.4.325. Sure enough. W h e n done.3 Using Excel to calculate TWR as a function of f.90.3. and copy each value into a separate c o l u m n .278 PART 5 Money Management and Portfolio Composition by-bar portfolio analysis later. you should have no problem creating a chart like that in Figure 15. w h e r e F219 denotes the last data row.) In this case. h a d you been able to trade this system at its optimal ƒ level. Figure 15.3 (remember.325 . in steps of 0. A FIGURE 15. That is." < >0")) In cell H 5 : =G5 In cell H 6 : =G6*H5 (then drag down to fill all the cells H6 to H219) T h e n manually enter into cell G3 all values for f between 0 a n d 1.

If you cut your risk (your f) in half. or h o w y o u think that you can keep a lower risk by only trading stocks and/or not selling short). A n d this is for one trade only. b a s e d on h o w large you expect the largest loser to be. the drawdown is even m o r e severe than that.e. W h e n that loser hits you. profits change geometrically. you lose that percentage a m o u n t of your equity. Your . as suggested by the ƒ level that you are using.) B u t there is m o r e to it than that. Instead you are a s s u m i n g m o r e risk than necessary to a c c o m p l i s h the same growth rate as you could have d o n e trading at a lower ƒ and you are also taking on a higher likelihood for going b r o k e the m o r e trades you place. if y o u change your ƒ arithmetically. T h e paradox is that the better the system a n d the higher the optimal ƒ the larger the drawdowns you suffer. In fact. trading at an ƒ level higher than the optimal ƒ does not m a k e your account grow any faster. h o w well diversified you think you are. If the next couple of trades turn out to be losers as well.CHAPTER 15 Money Management 279 FIGURE 15. you will cut the expected profits by m u c h m o r e than that. because this has nothing to do with h o w m a n y markets or what type of markets y o u trade (i. As y o u can see.4 The TWR of a fantasy system in relation to the fraction of your capital risked per trade. This relates to points three a n d four in the list at the beginning of this chapter. (See point t w o in the list at the beginning of this section. This is because the optimal ƒ is the percentage amount you are willing to lose on one trade.

A n d as things might turn out.280 PART 5 Money Management and Portfolio Composition worst d r a w d o w n will still be at least as large as your largest loser multiplied by the n u m b e r of contracts it takes to assume the total risk required by the ƒ level you are trading at. you might still end up trading to the right of the true optimal ƒ because it m i g h t have shifted to the left as well. F u r t h e r m o r e . unknowingly. w h i c h includes the largest loser. you are better off staying to the left of optimal ƒ because trading on a lower ƒ level results in the s a m e equity growth as trading at a level higher than the optimal ƒ but with a lower risk. B u t then a series of b a d trades. taking the equity below the initial level. k n o w that we have been working with several of these m o n e y m a n a g e m e n t parameters 10000 F I G U R E 15. if you were to trade at the optimal ƒ level. in real-time trading. hits us. at about trade 50 we m a d e approximately 10 times our initial equity.5 The growth in equity for a fantasy system traded with an f of 0. Did you. B u t to go from that conclusion to reject the entire t h e o r y behind fixed fractional trading is to take matters a little lightly. F r o m this perspective.9. for instance. traded at the optimal ƒ As you can see. . even if you a i m to trade to the left of the theoretical optimal ƒ you calculated using historical data. h o w sure could you be that this would be the m o s t optimal level in the future as well? Because you can't. Figure 15.5 shows what the development of the T W R would have looked like for this system. optimal ƒ is a purely a c a d e m i c figure with little practical value.

4 4 % ((1. To calculate the geometric m e a n . which corresponds to the T W R variable. Finally. except we have presented it as a percentage figure instead of a multiplier. also raised by 2. For instance. Then raise that by 2. From here." column. raise everything by 1 divided by 2.46%) and transforming it into a multiplier (1. Hence. the system will not m a k e any money. T h e s a m e goes for the cumulative profit. as we did.00442 (1.1) * 100). raise it by 1 divided by the n u m b e r of trades for the system and you will get 1. if the first parenthesis does not end up with a n u m b e r greater than 1.00442 . T h e only way for this to happen is for A P M to be sufficiently larger than A A A A A A A A A A A . take a look at the export function for the performance s u m m a r y for all systems based on the R A D contract and any of the spreadsheets you might already have produced with this function. where EGM = (APM 2 . before we deduct the standard deviation ( S D ) . To transform these percentage n u m b e r s (X) into multipliers (Y) do the following: Y = X /100 + 1 To transfer a multiplier to a percentage figure: X = (Y .5002 (50. By unweighted I m e a n that we have not taken the largest loser and the ƒ level into consideration when calculating it. It is The Equation. Let us call this multiplier A P M and the estimated geometric m e a n E G M . Perhaps not. then ETWR = (APM 2 . 4 5 % . in cells C5 to C 1 4 in the e x a m p l e above.S D 2 ) ( l / 2 ) . It shows a final cumulative profit of 5 0 . titled "Profit" by the export function. in the " C u m . the one that explains everything. This is the return per trade you could expect to m a k e h a d you b e e n able to reinvest all previous profits. This translates to approximately 0 . Into the individual profits c o l u m n .1) * 100 Take a look at the performance s u m m a r y in Table 12. in this case we calculate E G M to be 1. we can continue to estimate the T W R by raising everything by the n u m b e r of trades (please bear with m e ) .0046). which equals a T W R of 1.CHAPTER 15 Money Management 281 already throughout the entire b o o k ? If you have not thought about it. 0 2 % . Let us call the estimated T W R ETWR and the n u m b e r of trades N. or approximately 0 . but it does explain a lot. We also can estimate the geometric m e a n by taking the average profit (0.02 / 100 + 1).SD 2) (1/2) N = (APM 2 .5002 (1 / 92)). if you wish. for instance. we essentially have exported the unweighted H P R for each trade. 0 0 4 6 2 0 . prof. 0 1 6 3 2 ) ( l / 2 ) ) .00447 ( ( 1 .14.SD 2) (N/2) This final equation is so important that Ralph Vince dubbed it the fundamental equation of trading. That is: A A A A ETWR = EGM N.

or at least trade e a c h market on a separate level of aggressiveness.282 PART 5 Money Management and Portfolio Composition SD. Because the same system traded at different markets will c o m e up with a different optimal ƒ for each market. after we have gone to great lengths not to treat any of the markets any differently from all the others.b a s e d stops a n d exits. a n d almost at a n y t i m e . b a s e d on as few r u l e s as possible. in any t i m e frame. works equally well on all markets. but what really matters is how sure you can be that this relationship will hold up in the future. a fixed fractional trading strategy will d a m p e n it by forcing you to cut back on the n u m b e r of contracts traded. once again we would be curve fitting. That is. w e make sure that the s y s t e m will w o r k well in any m a r k e t . it s e e m s we should trade each market individually on a separate account. T h i s is the only way to c o m e close to a s y s t e m that not only trades profitably on historical data. simply because of the additional n u m b e r of trades e a c h n e w market brings to the portfolio. but instead a specific blend of separately weighted market/system combinations to the overall result of our portfolio. as suggested by each market/system combination's optimal ƒ Furthermore. B u t by w o r k i n g with techn i q u e s s u c h a s r a n d o m entries a n d p e r c e n t a g e . it seriously contradicts what we have done so far. T h e fundamental equation of trading also relates to what we discussed earlier about drawdown. PRACTICAL APPLICATIONS FOR MONEY MANAGEMENT All that we have discussed so far leaves us at a crossroads. By a d d i n g rules to build away individual m a r k e t d i s c r e p a n c i e s . You do that by m a k i n g sure that the system works well and/or trades profitably on as m a n y markets as possible. A n d h o w do you do that? O n c e again. This time we w o u l d n ' t be curve fitting a specific system to a specific market. That is. on average and over a long period of time. However. B e c a u s e as long as you can be sure of that. the larger the A P M in relation to the SD. and that we have treated each . If we decide to do this. while a larger n u m b e r of markets traded will help you to recover from drawdown at a faster rate. A l t h o u g h diversification does not necessarily reduce drawdown. why should we start doing it n o w ? Because if we did. Of course. but c o n t i n u e s to do so in the future. we c u r v e fit the s y s t e m a n d m a k e it less likely to w o r k in the future. m a k e N as large as possible. to m a k e sure that each and every one of our systems. if we blend in the Capital Asset Pricing M o d e l ( C A P M ) and the Efficient Market Hypothesis ( E M H ) we also should weigh each market/system combination against all others so that each market/system combination is delivering the s a m e risk-adjusted return as all others and so that the entire portfolio of market/system combinations balances on the efficient frontier. the only thing that will affect your bott o m line is h o w frequently you can trade. the better. this relates t o w h a t w e l e a r n e d a b o u t the system-building p r o c e s s a n d the i m p o r t a n c e of k e e p i n g it as s i m p l e as p o s s i b l e .

flat times. we still would have m a d e money. not risking m o r e than 1. you should treat each trade the s a m e way. if you would like to combine all your portfolios into one. Exactly the s a m e goes for a situation where you have several systems applied to one market. will be.CHAPTER 15 Money Management 283 and every market the same as all the others. If your systems are thoroughly researched and you " k n o w " they will continue to work. say the last 20 years or so. Another key phrase in the p a r a g r a p h above is "over a long period of time. I therefore suggest that for each portfolio of markets to which you apply the s a m e system. the lowest ƒ or the ƒ for the portfolio as a whole. while the average for the entire portfolio of all market/system combinations will still be approximately the same." not "equally as profitably. therefore. no matter in which market that particular trade h a p p e n s to be. using the fantasy system above. We will trade the system too aggressively when we shouldn't and not aggressively enough when wc should. based on historical observations.99 or higher. should be traded at its o w n / " If this turns out to be false. and worst-case scenarios. In this case the hypothesis. only at a slower rate. continue to do the same.4478. Had we. "that each market has its own specific features and statistical traits.015. but to optisize our betting size in accordance with a n u m b e r of constraints. . basically stating that a time series is a time series is a time s e r i e s . on the other hand rejected the hypothesis when we shouldn't have (a type I error) by trading all the market/system combinations with the s a m e ƒ as suggested by. depending on how you choose to define the constraints against which you are maximizing your utility. . for instance. But that is just pure coincidence. we will lose m o n e y and increase the likelihood of going bust. we would have curve fitted a system to a specific market. In the long run. and. which means a total equity growth of a mere 4 5 % . . Finally. then we are making a type II error by accepting a hypothesis that in fact should have been rejected. Over the next 20 years there might be other market/system combinations with higher fs than those that seem to be doing the best right now. than you would on a fixed contract basis. To keep it simple. N o t e that the key words here are "equally as well. the new T W R would equal 1." Of course we will end up with different optimal fs for different market/systems combinations within a limited t i m e frame. which is basically how far back we have tested the systems in this book. Not too . It also is possible to do the s a m e type of logical reasoning as we did w h e n we decided not to m a k e the amount invested dependent on any serial correlations a m o n g the trades. you still will m a k e tons m o r e m o n e y trading at a fixed fractional level as low 0.5% of your total equity per trade. calculated the optimal ƒ level to 0. then traded it real-time and g o n e broke tomorrow. a m o n g w h i c h a desired ending equity is one. For instance." If we only wanted to optimize historical profits. drawdowns. R e m e m b e r that we are not solely interested in optimizing our equity. and trade it at a fixed fractional level that m a k e s the m o s t sense to you. if we were to risk only 3 % . in regards to expected equity growth.

D o e s n ' t it m a k e sense then to go with what I expect to lose in this particular trade. because already at an ƒ equal to 0.6. this is as bad as I expect things to go. For each new trade I put on. If you trade 10 markets and it h a p p e n s that they all signal entry the same day. on average. As you can see. as the following e x a m p l e shows. A n d by allocating the m a x i m u m 1 0 % per m a r k e t you will go broke if all trades go against you. but on the other hand. Figure 15. as suggested by the original optimal ƒ but this fantasy figure describes the power of this method. N o t to risk going b r o k e you m u s t allocate less than that. say 2% p e r market. but that is as bad as I expect them to go.44. the n e w T W R equals 19. you w o n ' t be able to allocate m o r e than 1 0 % of your total capital. 4 0 0 % . to each market.10.7 shows you what the equity g r o w t h for this strategy would look like. By risking only 3% instead of 1 0 % of total equity I believe I have given myself all the slack I need. Now. rarely will you enter all markets at o n c e a n d rarely will they all go against you.844%. while the average return from the stock market over the last 15 to 20 years is no m o r e than 1 2 % or so. n o w it becomes even m o r e important that you do not overtrade. 0 1 1 " into cell F 3 . We can do a statistical type I/type II error analysis to confirm the importance of trading at an ƒ lower than the optimal ƒ This time the hypothesis is "the true ƒ for the future will be higher or the same as the optimal ƒ of the past. because the original system without optimal ƒ managed to produce a total of more than 2 1 6 % . m a k i n g it likely that we will go bust sooner rather than later.284 PART 5 Money Management and Portfolio Composition good. That is about 3 5 % per year. no matter what the o p t i m a l / m i g h t be for each individual market/syst e m combination. trading only 10 m a r k e t s isn't m u c h either. notice especially that the drawdown around trade 50 (although the scales are different) is no worse than any of the others. Anyway. not w h a t I might have lost in the past? Besides. Now. I will not be m a k i n g 3 3 2 .15. whether I choose to go with the largest loser (as suggested by the historical data) or with what I expect to lose the most in any particular trade (as suggested by the stop loss in the system) makes little difference. the best CTAs average around 2 0 % to 2 5 % per year.1%. therefore we should not use an ƒ lower than that ƒ" If we accept this hypothesis and it turns out to be false. w h i c h will give you a total risk of 2 0 % . the likelihood for going b r o k e is 100%. As a comparison. over the last 10 years. the new optim a l / f o r this overall strategy c o m e s out to 0. So from that point of view. which equals a total equity growth of 1. In that case. We will trade at an ƒ that is too high. if you trade the same system on a portfolio of m a r k e t s you w o u l d n ' t be able to trade it at the optimal ƒ level anyway. The fact is that I know if I do this long enough there will come along one trade that will blow me away completely. But what if instead we change the expected worst-case scenario to equal the stop loss for the system at 1. with the worst-case scenario set to equal the stop loss. then we again c o m m i t a type II error. In Figure 15. Granted. I know that sometimes things will go worse than that. with the relationship between different ƒs and the T W R resembling Figure 15. continuously compounded. by typing " — 0 . H a d we instead .7.

on a one-contract basis and -$13. For e x a m p l e . well. we have put together a spreadsheet like that shown in Figure 15. SHORT-TERM SYSTEMS For each market/system combination. rejected the hypothesis by trading at an ƒ sufficiently lower than the historical optim a l f.8. then we w o u l d n ' t have m a d e as m u c h m o n e y as we could have. T h e fixed fractional m o n e y m a n a g e m e n t strategy attached to this system told us to put on 56 contracts.50. Next.440. w h e n it turned out that we shouldn't have. T h e position was closed out on D e c e m b e r 30. we m u s t decide the "optisized" amount of contracts or shares we should feasibly trade.8 shows that on D e c e m b e r 24.6 The TWR of a fantasy system using an individual worst-case scenario for each trade. Figure 15. . w h i c h contains a daily s u m m a r y for that particular market/system combination. for a total profit of $13. A n d we will look at actual dollar figures instead of theoretical percentage n u m b e r s .CHAPTER 15 Money Management 285 FIGURE 15.020 for the overall portfolio. the system signaled to go long with a m a x i m u m risk of $578 per contract. this market/system combination w a s d o w n $232. 1985. with a $315 profit per contract. but we also would have increased the likelihood of still being in the g a m e . On the close of that day.

C37<0).8. 1 ). while the data in the three c o l u m n s u n d e r the "N contracts—calculated" are calculated in Excel. the data in the five c o l u m n s under the "1 Contract from T S " h e a d e r are imported directly from Trade Station. but can be altered within Excel as well.F36.H$32*H$33*MAX(INT($AN36* ( $B$47100)/C37).0) . type in the following formulas (disregard that some formulas hold references to other cells.7 The growth equity for a fantasy system with an individual worst-case scenario for each trade. All other cells also are imported directly from TradeStation. To perform the necessary calculations in columns F to H.0)) then drag d o w n to fill all the cells in that column. To calculate the open profit of all contracts: In cell G 3 6 : =ROUND(D36*F36.286 PART 5 Money Management and Portfolio Composition FIGURE 15. outside of Figure 15.8): In cell F 3 6 : 0 In cell F 3 7 : =IF(AND(C37<>C36.IF(E37=0. In Figure 15.

StdVS(O). For SumArr = 0 To 19 Begin IfSumArr = OThen SumVS = 0. TrueEntry(O).C[SetArr + 1] Data2) / C[SetArr + 1] Data2. Vars: VSStd(2). Direction("Both").0).QSetArr + 1] Data2) / C[SetArr + 1] Data2. To export t h e data needed. SumArr(O). AvgVS(O). PosProfit(O). FName(""). DiffVS(O). MarketWeight(l). MarginReq(l). DollarValue(l). then drag d o w n to fill all the cells in that c o l u m n .0) then drag d o w n to fill all the cells in that column.C[SetArr + 1] Data2) / C[SetArr + 1] Data2. To calculate the closed out profit from all contracts. SumVS(O).CHAPTER 15 Money Management 287 FIGURE 15. SystemName("MaxlOChar. M Array: VS[20](0)."). p u t t o g e t h e r a n o t h e r export function in TradeStation. For SetArr = 0 To 4 Begin VS[SetArr * 4 + 0] = (0[SetArr] Data2 .8 Using Excel to calculate an "optisized" number of contracts for a fixed fractional trading strategy. Missing(O). SystemWeight(l). VSMid(O). RiskValue(O). ClosedProfit(O). VSLow(O). SetArr(O). TradeStr2( "). in cell H 3 6 : =ROUND(IF(E36<>0. Vacuum(O). VSHigh(O). VS[SetArr * 4 + 1] = (H[SetArr] Data2 .C[SetArr + 1] Data2) / C[SetArr + 1] Data2. DiffArr(O).(E36-$B$3)*F35. VS[SetArr * 4 + 3] = (C[SetArr] Data2 ."). FillDay(O). . VS[SetArr * 4 + 2] = (L[SetArr] Data2 . MP(0). End. TotTr(O). We use the M e a n d e r system as an e x a m p l e : Inputs: MarketName("MaxlOChar.

01)) stop.288 PART 5 Money Management and Portfolio Composition SumVS = SumVS + VS[SumArr]. If Close > EntryPrice * (1 + (0. VSMid = C Data2 * (1 + AvgVS). End. End. If SumArr = 19 Then AvgVS = SumVS / 20.8 * 0. If DiffArr = 19 Then StdVS = SquareRoot(DiffVS / 20). If MarketPosition = 1 Then Begin ExitLong ("Long Target") at EntryPrice * (1 + (2.01)) stop.8 * 0. If MarketPosition = -1 Then Begin ExitShort ("Short Target") at EntryPrice * (1 . End.(2. ExitShort ("Short Time") at Close.01)) limit. If BarsSinceEntry > = 8 Then Begin ExitLong ("Long Time") at Close.6 * 0.(0.01)) Then ExitShort ("Short Trailing") at EntryPrice * (1 . ExitShort ("Short Loss") at EntryPrice * (1 + (1. End. ExitLong ("Long Loss") at EntryPrice * (1 .01)) stop.6 * 0.1 * 0.6 * 0.6 * 0. {***** Export function starts here *****} MP = MarketPosition. End.1 * 0. VSHigh = C Data2 * (1 + (AvgVS + StdVS * VSStd)).(0. . VSLow = C Data2 * (1 + (AvgVS . For DiffArr = 0 To 19 Begin If DiffArr = OThen DiffVS = 0.StdVS * VSStd)). Sell ("Go short") tomorrow at VSHigh limit. If MarketPosition = 0 Then Begin Buy ("Go long") tomorrow at VSLow limit. TotTr = TotalTrades.AvgVS).01)) limit. End.01)) Then ExitLong ("Long Trailing") at EntryPrice * (1 + (0.(1. If Close < EntryPrice * (1 .01)) stop. DiffVS = DiffVS + Square(VS[DiffArr] .

" + "0" + NewLine. 0) + "." + LeftStr(SystemName. TradeStr2 = "Market" + ". End.1 * 0. FileAppend(FName. End." + "." + NumToStr(MarketPosition. If DayOfWeek(FillDay) > 0 and DayOfWeek(FillDay) < 6 Then Begin TradeStr2 = NumToStr(FillDay.DateToJulian(Date[l])." + "Open ($)" + ". FileAppend(FName." + "Margin ($)" + ". FileAppend(FName." + "$ Value" + ". Vacuum = DateToJulian(Date) . ClosedProfit = 0. 0) + "." + NumToStr(RiskValue. PosProfit = 0.calculated" + NewLine. TradeStr2 = "1 Contract fromTS" + ". 2) + ". TradeStr2)." + ". TradeStr2)." + "Close ($)" + ". FileDelete(FName)." + NumToStr(MarginReq. If MarketPosition = 1 Then ." + NumToStr(PosProfit[l]. TradeStr2). 10) + ".csv"." + "N contracts ." + LeftStr(Direction.CHAPTER 15 Money Management 289 If CurrentBar = 1 Then Begin FName = "D:\Temp\MM-TM-" + LeftStr(GetSymbolName." + "." + "Open ($)" + ". TradeStr2)." + "System" + ". TradeStr2 = LeftStr(MarketName. 10) + ". RiskValue = TrueEntry * (1. If MP <> MP[1] and MarketPosition <> 0 Then Begin TrueEntry = C Data3 * (EntryPrice / C). TradeStr2 = "Date" + ". FileAppend(FName. FileAppend(FName. 2) + ". 10) + ". For Missing = 2 To Vacuum Begin FillDay = JulianToDate(DateToJulian(Date[l]) + (Missing-1)). 2) + NewLine." + "Contracts" + ". 0) + "." + ". 2) + "." + "." + "Close ($)" + NewLine." + "Position" + "." + NumToStr(MarketWeight." + "." + "Risk ($)" + "." + "Direction" + "." + NumToStr(DollarValue." + "System weight (%)" + ". RiskValue = 0. 2) + NewLine. End. 0) + ". TradeStr2).01) * BigPointValue." + "Market weight (%)" + ". End." + NumToStr(SystemWeight.

This information can be typed into the spreadsheet directly. fund m a n a g e r s . when we would like to have our results in dollars. FileAppend(FName.1) * TrueEntry[l] * BigPointValue. 2) + ".8. 2) + NewLine. R e m e m b e r from Part 1 that we learned we cannot m a k e any percentagebased calculations on the point-based back-adjusted contract and no dollar-based calculations on the R A D contract. If MarketPosition(l) = -1 Then ClosedProfit = -((ExitPrice(l) / EntryPrice(l)) . As you can see. This causes a problem now. where all individual front m o n t h s are attached to the time series one after the other. we have added the daily data for the unadj u s t e d time series as data 3. W h e n you start to put together portfolios. TradeStr2 = NumToStr(Date. 0) + ". End. because we would like to build hypothetical track records that can be compared to similar track records from different CTAs. to transfer our system's results back to the dollar-based world. and then export that value into the spreadsheet program. 0) + ". we can use percentage values to calculate the n u m b e r of points at risk and the open and closed-out profits in dollars for the unadjusted time series. If TotTr <> TotTrfl] Then Begin If MarketPosition(l) = 1 Then ClosedProfit = ((ExitPrice(l) / EntryPrice(l)) ." + NumToStr(ClosedProfit. all the data from all different markets m u s t be in sync. If MarketPosition = -1 Then PosProfit = -((C / EntryPrice) .290 PART 5 Money Management and Portfolio Composition PosProfit = ((C / EntryPrice) . A n o t h e r important feature of this code is the loop function.1) * TrueEntry[l] * BigPointValue. In the code above you can see that this has been done in a variable called "TrueEntry. In this example. but still are dependent on a set of percentage-based exit techniques to maintain them." + NumToStr(MarketPosition. TradeStr2)." + NumToStr(-RiskValue. we m u s t use some ingenuity and incorporate the unadjusted time series. row 33 holds information on which market and system this export stems. Therefore. look again at Figure 15.1) * TrueEntry * BigPointValue. or as inputs to the code." + NumToStr(PosProfit.1) * TrueEntry * BigPoint Value. I also have . We will work with dollar values now." which is calculated at the same time as the open and closed-out profits. and/or other system builders like ourselves. N o w that we k n o w h o w to export the data. w h i c h makes sure that you also export data for all days missing in your database or during holidays (except weekends). 2) + ". as well as information about point value and margin requirements. With this time series in place.

CHAPTER 15

Money Management

291

prepared both the export function and the spreadsheet for the possibility of experimenting with different systems and market weights. To give all market/system combinations an equal weighting keep the setting at 1, if you don't want to turn off that particular market/system combination completely, in which case you can set either o n e of the weights to 0. O n c e you have typed in the necessary information and formulae in c o l u m n s F to H, following Figure 15.8, you can either continue to add m o r e markets in adjacent c o l u m n s before you finish with the portfolio s u m m a r y calculations or, if you only are interested in this particular market, m o v e on directly to the portfolio s u m m a r y in c o l u m n s J to Q, as shown in Figure 15.9. To build the portfolio s u m m a r y type in the following formulas: In cell J36: =A36 then drag d o w n to fill all the cells in that column. In cell K 3 6 : =SUM(G36) then drag down to fill all the cells in that column. In cell L 3 6 : =SUM(H36) then drag down to fill all the cells in that column. In cell M 3 6 : =B2 I n cell M 3 7 : =M36+L37+(K37-K36) then drag down to fill all the cells in that c o l u m n .

FIGURE

15.9

Using Excel to calculate the aggregate results from trading several markets In a portfolio.

292

PART 5

Money Management and Portfolio Composition

In cell N 3 6 : =ROUND(K36* 100/M3 6,2) then drag down to fill all the cells in that c o l u m n . I n cell 0 3 6 : =M36 I n cell 0 3 7 : =MAX(M37,036) then drag down to fill all the cells in that c o l u m n . In cell P 3 6 : =ROUND((M36-036)* 100/036,2) then drag down to fill all the cells in that c o l u m n . In cell Q 3 6 : =IF(P36=0,0,Q35+1) then drag down to fill all the cells in that c o l u m n . Notice especially the use of the s u m function in c o l u m n s K and L. This is because h a d you also added data from other markets, that data would have been a d d e d to the overall portfolio results. In several of the above Excel formulae there are references to a set of cells in the B column, because this is w h e r e I stored the initial p a r a m e t e r settings, like the initial account balance, the dollar value to be deducted for slippage and commission, and the percentage risked per trade. This can be seen in Figure 15.10. In this case, we start out with $1,000,000 in initial equity, of which we will risk 3 . 2 5 % p e r trade. Before we add the closed-out profit to the overall equity of the portfolio, we deduct $75 p e r contract for slippage a n d commission. Now, with the data and all the formulas in place it is t i m e to build a performance s u m m a r y that resembles that in Figure 15.11. In this particular case it shows the results from trading the M e a n d e r system on the S & P 500, over the period M a r c h 1996 to October 1999, with $75 deducted for slippage a n d commission per contract traded. T h e fixed fraction of the total equity risked per trade was set to 5% and the initial capital w a s $1,000,000. As you can see, in this particular case we took the initial capital to an ending capital of $ 3 , 2 0 3 , 9 7 1 , w h i c h equals a yearly percentage return of approximately 3 6 . 4 % . T h e " p r i c e " for doing this c a m e in the form of having to spend approximately 16 out of every 100 days in the market, a m a x i m u m drawdown of close to 2 8 % , and a longest flat t i m e of 317 days. Before we go on, let us take a closer look at h o w each performance figure is calculated and what you need to type into each cell.

CHAPTER 15

Money Management

293

FIGURE

15.10

Example of initial parameter setting for all Excel calculations.

In cell E 3 : =ROUND(AN(X),0), w h e r e column AN refers to the column for w h e r e the portfolio equity is calculated ( c o l u m n M, in Figure 15.9, above) and Xrefers to the last row of data. In cell E 4 : =ROUND((E3/B2 - 1 ) * 100,0) In cell E 5 : =ROUND(((E3/B2) (l/B5)-1)* 100,2)
A

FIGURE

15.11

Strategy summary for a portfolio during the "optisizing" process in Excel.

294

PART 5

Money Management and Portfolio Composition

In cell E 6 : =ROUND(SUM(SUMIF(H36:H989,">0"),SUMIF(Q36:Q989, >0 ), SUMIF(Z36:Z989,">0"),SUMIF(AI36:AI989,">0"))/ABS(SUM(SUMIF (H36:H989; <=0 ),SUMIF(Q36:Q989, <=0"),SUMlF(Z36:Z989,"<=0 ), SUMIF(AI36:AI989,"<=0"))),2),
, n ,, M ,, ,,

where columns H, Q, Z, and AI refers to the closed out profit for each market/ syst e m in the portfolio (in this case I p r e p a r e d the fixed fractional portfolio calculations for four markets/systems, but temporarily turned off the other three). In cell E 9 : =ROUND(MIN(AQ36: AQ989),2), w h e r e c o l u m n AQ refers to the portfolio d r a w d o w n ( c o l u m n P, in Figure 15.9, above). I n cell E l 0 : =ROUND(MIN(AM36:AM989),0), w h e r e c o l u m n AM refers to the closed out equity for the portfolio (column L, in Figure 15.9, above). I n cell E l l : =ROU]S0^(SUM(COUNTIF(H36:H989, >0"),COUNTIF(Q36:Q989, •>0 ), COmTIF(Z36:Z989,*>0'0,COUNTIF(AI36:AI989,">0 ))*100/H3,2)
,, M , ,,

In cell H 3 : =SUM(COUNTIF(H36:H989, <>0"),COTJNTIF(Q36:Q989,"<>0 ), COUNTIF(Z36:Z989,"<>0"),COUNTIF(AI36:AI989, <>0")) In cell H 4 : =ROUND(SUM(SUM(H36:H989),SUM(Q36:Q989),SUM(Z36:Z989), SUM(AI36:AI989))/H3,0) In cell H 5 : =ROUND(H3/B5/B7,l) In cell H 6 : =ROUND(H5/12,l) In cell H 9 : =MAX(AR36:AR989), w h e r e c o l u m n AR refers to the flat time for the portfolio ( c o l u m n Q, in Figure 15.9, above).
M ,, ,,

CHAPTER 15

Money Management

295

In cell H 1 0 : =COUNTIF(AL36:AL989,"<>0")*100/COUNT(AL36:AL989), w h e r e column AL refers to the total open equity for all open trades in the portfolio (column K, in Figure 15.9, above). In cell HI 1: =ROUND(COUNTIF(AL36:AL989, <>0")/COUNTIF(AM36:AM989,">0 ),0) With all the formulae in place, it is easy to put together a set of charts like those in Figure 15.12 through 15.14, which show the total equity, the drawdown, and the open profit, respectively. Figure 15.12 shows the equity curve for a portfolio consisting of only one market/system combination. It corresponds to the performance s u m m a r y in Figure 15.11. Because of the very long initial flat period, it is not difficult to see that this equity curve is no good, despite the fact that the overall growth rate is as high as 3 6 . 5 % p e r year. At this point, it also is important to understand that the percentage moves of the market are not the same as the percentage increase or decrease of our account. That the system will have us stopped out if the market moves against the position by 1.1%, is not the s a m e as if we are risking only 1.1% of our capital at that trade. No matter h o w m u c h the market
M M

FIGURE

15.12

The equity curve from trading Meander on the S&P 500 with a 5% fixed fractional setting.

296

PART 5

Money Management and Portfolio Composition

FIGURE

15.13

The drawdown curve from trading Meander on the S&P 500 with a 5% fixed fractional setting.

moves, it is still possible to lose m o r e or less than that, in percentage t e r m s . In this case, we m a k e sure that we have e n o u g h m o n e y going into the trades so that we can afford to lose 5% of it if the market moves against us by 1.1%. T h e dangers of trading this market/system combination alone are confirmed by Figure 15.13, which shows the different drawdown periods. However, when looking at the open equity of all trades at each day, in Figure 15.14, we can see that one m a i n reason for this long flat period is not that the system is performing badly, but instead that it doesn't trade often enough; we have to suffer the pain of one bad trade for an exceptionally long time. To shorten this period of torture, we must diversify by trading the s a m e system on other m a r k e t s and/or apply other uncorr e c t e d systems to the s a m e market. By the way, Figure 15.14 fills an important function in that it allows you to keep track of the total open equity from all open positions in relation to the total equity (both closed-out and open) of the entire portfolio and compare it to historical values. Positive values m e a n that the profit is positive, while negative values obviously indicate the opposite. If the current open profit gets out of line when compared to what is expected, you k n o w that one or several market/system combinations have developed in such a way that you n o w are steering the entire portfolio into unfamiliar territory. In this case, very seldom will the profit exceed 10%, or

CHAPTER 15

Money Management

297

FIGURE

15.14

Profits per trade for Meander traded on the S&P 500 with a 5% fixed fractional setting.

surpass a loss of — 5 % (because that should have us closed out). Therefore, when this happens, you know this is an exceptional occurrence. A n d because exceptional occurrences, whether for better or worse, add m o r e to the insecurity and risk than to the bottom line, you might be better off taking a profit or cutting the loss, overriding the signal from the system for the best of the portfolio as a whole. As already concluded, the performance s u m m a r y in Figure 15.11 and the development of the total equity as depicted by Figures 15.12 and 15.13 are not all that impressive. But this portfolio consisted of only o n e market/system combination. To better our results we need to diversify into other market/system combinations. Table 15.1 shows the result of one such diversification. This time the portfolio consists of three different market/system combinations, the three short-term systems we built in the previous section of this book, all applied to the S & P 500 c o m m o d i t y futures contract, with the m a x i m u m percentage risked per trade set to only 1.5%. T h e initial capital was set to $1,000,000, with $75 deducted p e r contract traded. Notice that we n o w are m a k i n g over 5 0 % p e r year, with a largest d r a w d o w n of less that 2 0 % a n d a longest flat time of less than 6 months ( 1 1 3 / 2 1 = 5.4). A n d we are only risking 1.5% p e r trade! Figures 15.15 and 15.16 show the a c c o m p a n y i n g equity curve a n d historical drawdown for this short-term strategy. As you can see, although from the total

00 Trade statistics 316 12.166 383 52. the worst drawdown in percentages h a p p e n e d late in 1998.17 1. in dollar terms.050 56. the worst drawdown.298 PART 5 Money Management and Portfolio Composition TABLE 15.773 28.56 No Trades Avg Tr ($) T r / Mark /Year Tr / Month Time statistics 113 54.82 3.3 equity chart it looks like the worst d r a w d o w n in dollar t e r m s h a p p e n e d sometime during the s u m m e r of 1999.827. and rests on sound and simple principles. Strategy summaries Profitability End Eq ($) Total (%) Year(%) P factor Risk measurers Max DD (%) Lrg Loss ($) Winners (%) -19.96 Lng Flat (d) TIM (%) Avg Days 4. as the equity continues to grow.15 The combined equity curve from trading three different market/system combinations. . will F I G U R E 15.47 -307.5% fixed fractional portfolio performance summary for three systems trading S&P 500.1 1.1 2. Providing the overall strategy is robust.

you k n o w that the only reason why you suddenly get a higher drawdown figure is that your strategy is no longer working as it should (or at least is behaving in an unexpected m a n n e r ) . you k n o w exactly h o w to go back and correct what is w r o n g . w h i c h show two different ways of m e a s u r i n g the performance of the individual trades. 5 . This is because in both charts the m a g n i t u d e of the bars is expressed in percentage t e r m s . You m a y even c h o o s e to build a new system. If and w h e n this h a p p e n s .18 there seem to be an u p p e r and lower limit to h o w big each bar is likely to b e . .16 The combined drawdowns from trading three different market/system combinations. if the m a r k e t trend is up or down. all of w h i c h are develo p e d solely to work well together u n d e r a fixed fractional m o n e y m a n a g e m e n t regimen. C o m p a r e t h e m to Figures 2.18. because having r e a d the previous sections.4 and 2 . W i t h the above paragraph in mind. or any other reason.5 and 15.CHAPTER 15 Money Management 299 FIGURE 15. continue to grow as well. If it h a p p e n s to increase. especially suited to trade with a portfolio of other systems and markets. the size of your account. y o u no longer need to worry. 5 .17 a n d 15. In Figure 2 . look at Figures 15. thus forming an overall trading strategy w h e r e the whole is greater than the s u m of its parts. With the technique illustrated here. this will have nothing to do with the level the m a r k e t currently is trading. T h i s is not the case w h e n using any of the commercially available system testing p a c k a g e s . while the drawdown as a percentage of your equity is likely to stay the same. In Figure 2.

4 the dollar value per trade is increasing because the market is trading higher.14. on the other hand. T h a t is.17 can be fun to look at as long as everything is going the right way. if the closed-out profits seem to get out of line w h e n c o m p a r e d to w h a t could be expected.4 a n d 15. In Figure 2. we are looking at the dollar value from trading each m o v e with an increasing n u m b e r of contracts. In Figure 15.18. whether our equity is increasing or not has nothing to do with whether the market(s) are trending up. while in Figure 15. We know this now. However.18. the most important chart to keep track of is Figure 15.17.4. but there was no way of knowing this from just looking at Figure 2. do not use these relative comparisons. it is calculated as the value of a specific trade in relation to the total equity of the portfolio. In Figure 15.17 is that in Figure 2. Figures 2.300 PART 5 Money Management and Portfolio Composition FIGURE 15. or sideways. catching whatever type of trend or m o v e happens.17 the dollar value is increasing because our equity is increasing. but rather with the fact that our strategy is working as it should. T h e main difference between Figures 2.17 The dollar value for each trade from three different market/system combinations.4 and 15. for exactly the s a m e reasons as those for Figure 15. w h e n we started out putting our strategies together. the percentage is calculated as the value of a specific m o v e in relation to where the market currently is trading. above. down. we are looking at the dollar value from trading each m o v e with one contract. but chart the dollar values directly. you know . Although Figure 15.17.4.

It is important to realize.19. which is shown in Figure 15. say 5 0 % just in case. I have assumed a margin of $22. as time moves on.000 per contract. that one or several market/system combinations have developed in such a way that you n o w are steering the entire portfolio into unfamiliar territory. on account with our broker.CHAPTER 15 Money Management 301 FIGURE 15. with an expected m a x i m u m around 3 5 % . . the average margin to equity ratio should probably even out just below 2 0 % . we only need to keep. if this starts to h a p p e n m o r e and m o r e often. That is the margin to equity ratio. very seldom will the closed-out losses be any worse than j u s t a few percentage points. This chart shows how m u c h of your total equity you actually need to keep on account with your broker to do the necessary trades. In the case of the S & P 500 portfolio we have been working with so far. In this case. indeed. where we h o p e it can generate a higher return than the interest rate our broker can offer. it would be as if you were trading your portfolio with only half the equity. that you m u s t have this money readily available and committed to your strategy as well. T h e rest of it we can have elsewhere. Otherwise. W h e n working with the daily and bar-by-bar data there is one m o r e important chart to consider. This m e a n s that of the total equity we have at stake. you k n o w that o n e or several of your market/system combinations have started to behave in ways not in line w i t h w h a t could be expected and perhaps even against the logic of the system. In this case. though. but twice the ƒ a recipe for disaster.18 The percentage return for each trade from three different market/system combinations. Therefore.

1 and 15. c h a n c e s are that we m i g h t h a v e ditched the three-market/system portfolio for t h e b e n e f i t of the s i n g l e .000. while the three-market/system portfolio would have m a d e $331.500 on a one-contract basis.1 with Tables 15.19 The margin to equity ratio for a portfolio. a n d the d r a w d o w n all are better in F i g u r e 15.000. O t h e r interesting o b s e r v a t i o n s are that the profit factor. T h a t is quite a difference c o m p a r e d to the 5 2 % per year we could have m a d e only risking 1. a n d then traded that o n e m o r e aggressively instead.2 shows that the single market/system portfolio would only have m a d e $88. If a fixed fractional trading strategy is not doing any better than a one-contract. Before we m o v e on to the long-term systems.3.000. then what is the point? But we do not have to worry. if we h a d derived these n u m b e r s directly from T r a d e S t a t i o n ' s or any other c o m m e r c i a l l y available p r o g r a m ' s performance summary. .302 PART 5 Money Management and Portfolio Composition FIGURE 15. these are all i m p o r t a n t figures. the p e r c e n t a g e profitable trades.11 a n d Table 15.11 and Table 15. H a d we not k n o w n what we n o w know.m a r k e t / s y s t e m c o m b i n a t i o n . the value of the average trade. or approximately 8%> p e r year on an initial equity of $1.3 which show the results from trading both these portfolios on a one-contract basis. For the average s y s t e m builder. Table 15.2 than they are in Tables 15. let us compare the performance summaries in Figure 15.2 and 15.5% of total equity p e r trade.

68 -13. is that the best way to speed up the equity growth rate for a trading strategy with a positive expectancy is to trade it as frequently as possible.3 Single contract portfolio performance summary.5 There are several reasons w h y a portfolio of several market/systems combinations does better than only a single market/sy stem combination. by letting two or several market/system combinations with positive expectancies share the s a m e account.656 56.53 Lng Flat (d) TIM (%) Avg Days 1. Furthermore.213 6. as we learned from the fundamental equation of trading. A n d no matter h o w aggressively we trade a single market/system combination. going into each individual trade.088.82 3.29 1.1 2.5 0.3 .92 1. as it will if we trade several market/system combinations simultaneously.079 28.35 4. the actual n u m b e r of trades will not increase. there is a positive expectancy that there will be m o r e m o n e y to play with.85 -32.CHAPTER 15 Money Management TABLE 15.2 Single contract performance for the Meander system. even if we try to c o m p e n s a t e by trading the sole combination m o r e aggressively.96 Lng Flat (d) TIM (%) Avg Days 1.148 57.86 No Trades AvgTr($) Tr/Mark/Year Tr / Month Time statistics 275 16.59 No Trades Avg Tr($) Tr/Mark/Year Tr/Month Time statistics 112 54.00 Trade statistics 316 1. The m o s t obvious reason.00 Trade statistics 73 1.013 33 7. Strategy summaries Profitability End Eq ($) Total (%) Year (%) P factor Risk measurers Max DD (%) Lrg Loss ($) Winners (%) -1.331. TABLE 15.541 9 2. Strategy summaries Profitability End Eq ($) Total (%) Year(%) P factor Risk measurers Max DD (%) Lrg Loss ($) Winners (%) -3.

given the a s s u m e d risk as dictated by the largest expected amount to lose p e r contract and the largest expected amount to lose per trade (the fixed fractional trading level). A n d who in his right m i n d would have lent us that k i n d of m o n e y w i t h a track record showing d r a w d o w n s upwards of 8 0 % ? N o .148. results in an optim a l ƒ of less than 0. it could be fun to k n o w h o w m u c h we could have m a d e h a d we traded the three-market/system portfolio at or near its optimal ƒ Table 15.20 do you think is the worst. w h i c h also is illustrated by the fact that the m a x i m u m n u m b e r of contracts traded at o n e t i m e would have b e e n m o r e than 10. the decrease in efficiency as a result of trading several uncorrelated market/system combinations also has a major benefit. to take the reasoning through yet another circle. with an optimal ƒ of 0. however.5 for the strategy as a w h o l e . for instance.20 reveal that we would have taken the initial million to $114. trading two market/system combinations that are not perfectly correlated. so has the other. j u s t to cover margin. although the theory surrounding optimal ƒ is beautiful. However. w h i c h m e a n s that all losing trades leading into a new drawdown from a recent equity high hit you full force. (The .000. the price for this would have been to suffer through a drawdown of m o r e than 7 5 % of account equity a n d a largest single loss of close to $37.304 PART 5 Money Management and Portfolio Composition This is true as long as the different market/system combinations are not perfectly (100%) correlated and with each one having a positive expectancy. T h e r e is a price for doing this. two perfectly correlated market/system combinations. w h i c h corresponds to an average percentage return of 2 5 4 % per year.4 and Figure 15. as measured by the estimated geometric mean ( E G M ) . just for the h e c k if it. For instance. but instead of how risky it is. but also because they don't trade at the s a m e time. w h e r e a s a drawdown for an uncorrelated portfolio of market/system combinations is likely to be dampened because the combinations are low correlated. Finally. To get an indication of h o w high the return will be. W h e n o n e of them has a losing trade. as a percentage of total equity?) T h e s e are clearly not tolerable or tradable n u m b e r s . Consider. to trade at an unconstrained optimal ƒ level simply is not feasible.000 and that the average margin-to-equity ratio would have been close to 2 0 0 % . w h i c h one of all the drawdowns in Figure 15. we must look at the estim a t e d growth rate. But.000.5 each. We would have h a d to borrow the s a m e a m o u n t as we h a d on the account. This m e a n s that there also is a limit to h o w m a n y market/system combinations it is meaningful to trade. (By the way. It could also be true even if one or several of the market/system combinations have a negative expectancy. T h e lower the optimal ƒ or the fixed fractional trading level. and that is that the overall efficiency of the portfolio diminishes the m o r e market/syst e m combinations you add. the lower the risk and the shallower the expected d r a w d o w n s . All the above leads to one important conclusion: the optimal ƒ is not a measure of how high a certain portfolio's return is likely to be. as long as they are uncorrelated e n o u g h with the rest of the combinations and the portfolio as a w h o l e .000. This m e a n s that each losing trade decreases the n u m b e r of contracts that can be traded going into each consecutive trade.

1 2. is a key factor in the success of a fixed fractional trading strategy. going into the trade.3 worst d r a w d o w n as a percentage of total equity h a p p e n e d during the fall of 1997. in doing so it is vital that the system be robust a n d likely to trade profitably on as m a n y markets as possible.5% of total portfolio equity p e r trade.20.CHAPTER 15 Money Management 305 TABLE 15. the long-term systems have one major disadvantage c o m p a r e d to their short-term siblings: their lower frequency of trades.973 11. therefore should use the system together with as m a n y markets as possible to speed up the trading frequency. T h e starting balance . and shows up as a little blip about halfway t h r o u g h the time series in Figure 15. The SDB Strategy During the entire testing procedure for the S D B system we only used data up until October 1992.72 1. no matter what the current m a r k e t conditions are like or what market we are trading.315 253. To overcome this w e .5 shows the result of trading the entire portfolio on the in-sample data.00 Trade statistics 316 451.96 Lng Flat (d) TIM (%) Avg Days 114. Strategy summaries Profitability End Eq ($) Total (%) Year(%) P factor Risk measurers Max DD (%) Lrg Loss ($) Winners (%) -75. to achieve this the system m u s t work the same. which. using the same fixed fractional m o n e y m a n a g e m e n t for all markets.93 3.147. as we have learned from the fundamental equation of trading.67 -36. However.) THE LONG-TERM STRATEGIES W h e n it c o m e s to fixed fractional investing.724.33 No Trades Avg Tr ($) Tr/Mark/Year Tr / Month Time statistics 160 54.875 56. a n d risking no more than 1.4 7% fixed fractional (optiomal f) performance summary for a three market/system combination. this also m e a n s that we a s s u m e no significant statistical differences between the m a r kets and that they are all treated identically and traded with the same ƒ even though there might be significant differences between the fs w h e n we c o m p a r e different market/sy stem combination with each other. As we have seen. T h e rest of the data we saved for some out-of-sample testing. To be consistent. Table 15.612 28.

you can c o m p a r e the results in Table TABLE 15.75 Lng Flat (m) TIM (%) Avg Days 7. Just to m a k e sure that it is indeed beneficial to use a fixed fractional trading strategy.20 The equity curve from trading three different market/system combinations at their optimal f.000.000.614 2. even if we keep t h e / a s low as 0. Strategy summaries Profitability End Eq ($) Total (%) Year(%) P factor Risk measurers Max DD'(%) Lrg Loss ($) Winners (%) -17. with $75 deducted for slippage and c o m m i s s i o n per contract traded.462.81 20.17 -148.124 41.9 99.04 No Trades AvgTr($) Tr/Mark/Year Tr / Month Time statistics 25.5% fixed fractional trading with the SDB strategy (in-sample period). w a s set to $1.552 646 17.015.5 1.8 .6 2.00 Trade statistics 424 14.1 2.306 PART 5 Money Management and Portfolio Composition F I G U R E 15.

7 5 billion a year.81 99.5 billion.CHAPTER 15 Money Management 307 15. We will put together a set of formulae that produce monthly results. on a v e r a g e — i f y o u believe that is w o r t h sleepless n i g h t s .75 Lng Flat (m) TIM (%) Avg Days 1. w h i c h we will use for further input into a new set of formulae that produce results like those in Tables 15. T h a t we will do. the final r e t u r n is only a function of h o w well you w o u l d like to sleep at night. a largest loss close to $4. In this case we w o u l d be m a k i n g s o m e t h i n g like $ 1 . Strategy summaries Profitability End Eq ($) Total (%) Year(%) P factor Risk measurers Max DD (%) Lrg Loss ($) Winners (%) -2.32 1.00 Trade statistics 424 765 2. N o w that we k n o w the S D B system works fine within a m o r e complete trading strategy incorporating consistent m o n e y m a n a g e m e n t on a whole portfolio of market/system combinations. resembling those in Figure 15. b u t also a d r a w d o w n close to 9 5 % . T h e only major drawback with the statistics in Table 15. and the portfolio composition—interfere and influence each other.81 20. As s h o u l d be clear by now. the m o n e y m a n a g e m e n t . m a k i n g all three parts equally responsible for results.7 and 15.99 No Trades AvgTr($) Tr/ Mark /Year Tr / Month Time statistics 20. TABLE 15. 5 % or lowered to 1%. a n d a l o n g e s t flat t i m e of m o r e than 40 m o n t h s . By the way. T h e only thing that h a p p e n s is that with a higher ƒ c o m e s a higher drawdown and a higher return. t h e o p t i m a l ƒ for this portfolio can be found a r o u n d 1 4 % . research and experimentation show that it is m o r e a result of b a d portfolio composition than of the system itself.8. c o r r e s p o n d i n g to a yearly rate of return close to 1 2 5 % . although at this stage all three variables—the system. 5 billion.6. b e m y guest.6 Single-contract trading with the SDB strategy (in-sample period). w h i c h shows how m u c h you would have m a d e trading o n e contract only.21.329. but first we will add another few analysis techniques to our portfolio s u m m a r y spreadsheet.53 -6. Nonetheless.1 2. perhaps it is time to look at how the same strategy would have performed on previously unseen data. w h i c h p r o d u c e s a final profit of m o r e than $ 2 1 .060 33 2.8 . and vice versa.5 is a little too long flat period. However.5 with those in Table 15. in this case the longest flat period r e m a i n s the s a m e even if ƒ is increased to 2 .726 41.

26 8.14 20.17 N/A . EGM Sharpe ratio Avg.7 Cumulative percentage period returns.41 27.77 49.78 110.56 1.78 12.308 PART 5 Money Management and Portfolio Composition FIGURE 15.92 1. In c e l l E Z 3 1 : =IF(INT(E031/100)<INT(EO32/l 00).09 3.42 29.17 256.70 -6. In cell E Z 3 0 : =EO30 w h e r e column EO denotes the daily updated dates.43 13.26 36 79.13 -12.92 1.81 2.35 24 54.21 Using Excel to calculate monthly returns.04 149. To start.80 -3.19 0.36 3.46 69.63 -4.53 25. highlight the cell in the column.88 159. dev. o n e c o l u m n away from your portfolio calculations a n d at the s a m e row for w h e r e you start the portfolio calculations. losing 1 -0.21 4. winning Avg.33 3 3. In cell FA30: =ER30.05 36.14 17. In c e l l F A 3 1 : =IF(INT(EO31/100)<INT(EO32/100).17 -3.38 44.61 7.ER31.25 -9.21 -11.21.35 0.23 69.03 228.57 -2.64 1. n months rolling window (in-sample period).11 8.67 48 119.71 60.53 101.34 7. Cumulative Most recent Average Best Worst St. In Figure 15.95 4.48 -3.65 62. w h e r e column ER the daily updated total equity.82 110.56 70.12 1.40 63."") then drag down to fill all the cells in that c o l u m n .EO31.52 39.03 N/A 60 159.37 8.63 12 39.94 1.31 159. that is cell E Z 3 0 .97 116.21 -10.48 162.11 -1.97 11.54 15.76 -1."") TABLE 15.16 0.42 16.52 6 26.67 44.

FB30) then drag d o w n to fill all the cells in that c o l u m n .95 4.56 13.31 47.65 62.7 -6. You m u s t delete all the blank cells in columns EZ and FA and line up all the other cells containing information underneath each other so that the end result will look like Figure 15.52 19.12 48 21. 12-.00 then drag down to fill all the cells in that column.23 37.31 122.05 20. Annualized Most recent Average Best Worst St. 4 8 . 36-. dev. W h e n you are done.11 12. This can be done either manually or with the help of a macro.03 1. 24-.76 18.95 32. In cell F E 3 3 : =(FA33/FA30-1)*100 then drag down to fill all the cells in that column.71 18.88 13.96 24 24.6 1.48 36 21.58 345.21) continue to calculate the rolling 6-. n months rolling window (in-sample period).79 0.1 -68.7.00 60 21. . In c o l u m n s FD and F E . Sharpe % winners 1 -10..48 99.64 -20.8 Annualized percentage period returns.56 1.99 -0. In columns FF to FK (not shown in Figure 15.59 72.14 86.31 18.42 1.CHAPTER 15 Money Management 309 TABLE 15.59 2. With the above formulae and all other formulae we have used throughout the book.35 90.9 100.21 -40.69 18.04 -5.98 28.11 0.77 78.63 0.84 161.76 3 16. In cell F C 3 0 : =(FA30-FB30)*100/FB30 then drag down to fill all the cells in that c o l u m n . and 60-month rolling returns.54 15.76 20. continue to calculate the latest equity high and drawdown.62 100.07 1.38 34. In cell F D 3 1 : =(FA31/FA30-1)*100 then drag down to fill all the cells in that column. we calculate the rolling monthly and quarterly returns.14 17.11 6 59.13 58.6 11.74 23. it should be no problem to go on and produce a table like Table 15.21.47 12 39.32 63. In cell F B 3 0 : =FA30 In cell F B 3 1 : =MAX(FA31.

the starting equity was set to $ 1 . for instance.7 and 15. as long it decreases the standard deviation even m o r e . a n d the out-of-sample period from O c t o b e r 1992 to O c t o b e r 1999. 0 0 0 . 6 8 % . use the following formula i n cell K l 3 : =ROUND(((K3/l 00 +1 ) ( 12/K2) -1 )* 100.9 reveals that the e n d i n g equity n o w w o u l d have been close to $ 2 0 .8 is to create charts like those in Figures 15. 0 0 0 . This is such an important finding that it m u s t be explained further for a deeper understanding.) T h e positive side effect from all this is that we also are likely to increase the n u m b e r of trades. which boosts our profits even further according to the fundamental equation of trading. To m a k e sure that the n u m b e r of winning m o n t h s will be high. Let's take a l o o k at h o w the S D B strategy w o u l d have fared both during the i n . Table 15.2). 5 4 % and never did we produce a losing period lasting longer than 3 years. other trades to c o m e will have an easier time m a k i n g up for the recently produced loss. this might m e a n that a lower percentage of profitable trades will produce a higher percentage of profitable months. A n o t h e r way to illustrate several of the n u m b e r s in Tables 15. w h e r e cell K2 denotes the length in m o n t h s of the original period.s a m p l e period alone indicates that the strategy has A . Second. (Within reason that i s — t h e mathematical expectancy still must be positive and the average trade large e n o u g h to m a k e it worthwhile trading the system. to transfer it into an annualized figure in cell K 1 3 . 0 0 0 . simply raise each value by (12/PL). I think you will agree when I say that it is far more important to have a high percentage of profitable m o n t h s than it is to have profitable trades. a single loss w o n ' t go on forever. Because we have constructed the system to cut our losses short. w h e r e PL equals the length of the period. As before. the percentage of profitable m o n t h s was as high a s 6 4 % .22 and 15. We also can see that although the percentage of profitable trades w a s less than 4 2 % . but only to m a k e sure that it is sufficiently larger than the standard deviation. a n d $75 w a s d e d u c t e d for slippage a n d c o m m i s s i o n p e r contract traded.310 PART 5 Money Management and Portfolio Composition To transform the cumulatively c o m p o u n d e d values in Table 15. 0 0 0 .23. F r o m Tables 15.8. if the m o s t recent monthly return is stored in cell K 3 . This is the s a m e type of trade-off we m u s t do w h e n we compare the average trade to the standard deviation.8 we can. using such techniques as time-based stops we accomplish two things. every change we m a k e to the system that decreases the average trade is all right. T h e trick is not to have as high as possible an avera g e trade.s a m p l e p e r i o d from J a n u a r y 1980 to O c t o b e r 1992.7 to annualized dittos in Table 15. First. see that during no 12-month period did we lose any m o r e than 6 . weighting down the results for several m o n t h s in a row. In the end. For instance. O n c e you have taken your analysis to this point. T h e fact that this is slightly lower t h a n for the i n .7 and 15. you must have taken that into consideration at step one of your system-building process. c o r r e s p o n d i n g to a yearly return of 1 6 .

59 Lng Flat (m) TIM (%) Avg Days 19.895 16. Strategy summaries Profitability End Eq ($) Total (%) Year(%) P factor Risk measurers Max DD (%) Lrg Loss ($) Winners (%) -27.453 39.949.5% fixed fractional trading with the SDB strategy (both sample periods).9 99.22 A graphical presentation of cumulative monthly returns (in-sample period).CHAPTER 15 Money Management 311 FIGURE 15.52 No Trades AvgTr($) Tr/Mark/Year T r / Month Time statistics 25.1 .9 1.4 3. TABLE 15.3 -568.88 19.184 2.789 1.00 Trade statistics 730 25.68 1.

25 give you essentially the same type of information as Tables 15.23 A graphical presentation of annualized monthly returns (in-sample period). but also it has turned m o r e risky.5. and figures 15. Figures 15.8. T h i s is further c o n f i r m e d by a lower profit factor.9 than in Table 15. T h e percentage of winning m o n t h s also has decreased somewhat. and a lower p e r c e n t a g e profitable trades. As you can see. because it implies that not only has the strategy turned less profitable. .23. the average trade m e a s u r e d in dollars is m u c h higher in Table 15. but still not too far away from what could be considered tolerable for a professional m o n e y manager. this has n o t h i n g to do with t h e strategy's p e r f o r m i n g any better. as s h o u l d be clear by now.7 and 15. b u t simply b e c a u s e the available equity is so m u c h h i g h e r — w e can have m o r e m o n e y at stake at e a c h individual trade. the only thing that most definitely is not tolerable is the longest flat time. not w o r k e d as well during recent years.10 we also can see that the average trade.312 PART 5 Money Management and Portfolio Composition FIGURE 15. but. In fact.22 and 15. From Table 15.24 and 15. This is not a good sign. has decreased while at the same time the standard deviation has increased. although it has stayed intact during the out-of-sample period. All in all this is a decrease in performance. a larger m a x i m u m d r a w d o w n . measured in percentage terms.

000 in initial equity. winning Avg.10 Monthly returns for the SDB strategy (both sample periods). one major reason for this is that. we started out with $1. and Figures 15.3 4. As before.27 show the result of trading the directional slope system on all 16 markets used during the system-building process. As y o u can see. such as the S & P 500 index and the C R B index. Table 15.40 15. the point here is not to give you the very best systems around. Most recent Average Best Worst St. . dev.58 -13.63 4. that my research in this area is still at its initial stages and that only time will prove me right or wrong. losing % winners -7.11. or—perhaps even more important—to describe the underlying reasoning that you must consider before you sit down in front of the computer. Inevitably there will c o m e a day w h e n such trend-following workhorses such as the Japanese yen will stop working a n d start behaving like the C R B index (which is a notoriously difficult market with any strategy). Unfortunately.29 0. deducting $75 for slippage and c o m m i s s i o n per contract traded.70 1. but simply to outline a few tips and tricks on h o w to go about building one yourself. Nevertheless. mainly because of a largest d r a w d o w n closing in on 6 0 % and a longest flat period of m o r e than 28 m o n t h s .000. we used weekly data to filter out some of the daily noise.86 60. My initial research shows that this inevitably lowers the results somewhat.12 1. to be able to track the system in real time.52 The Directional Slope Strategy W h e n we built the Directional Slope System.CHAPTER 15 Money Management 313 TABLE 15. I did this to pollute the final parameter setting with as m a n y market conditions as possible to increase the likelihood for the m o d e l to hold up in the future. W h e n that day c o m e s I want to have m a d e sure that I have taken every possible action to keep myself from going broke. I must point out however. we now need to transfer it back to a daily strategy by multiplying all the input variables by five. I deliberately used markets that I knew from experience would not sit well with a trend-following strategy. but so far I have found no proof for ditching this system-building concept altogether. provided that original system logic has been sound and simple and that every effort has been m a d e to keep the system as robust as possible.18 -2.26 a n d 15. these results aren't too tantalizing. However. Geo Mean Sharpe ratio (Annualized) Avg. during the optimization process.

In this case.11 1.24 The cumulative monthly returns for the SDB strategy (both sample periods).12. w h e r e I simply divided the total closedout profit for each market by the total profit for the portfolio as a whole.06 7.06 No Trades AvgTr($) Tr/Mark/Year Tr / Month Time statistics 28.314 PART 5 Money Management and Portfolio Composition FIGURE 15.080 1.5% fixed fractional trading with the directional slope strategy.420 9. we can see that the S & P 500 lowers the result by close to 2 5 % . Strategy summaries Profitability End Eq ($) Total (%) Year(%) P factor Risk measurers Max DD (%) Lrg Loss ($) Winners (%) -57.56 -2.73 1.4 . To get a feel for how m u c h each market is contributing to the end result you can put together a table like Table 15.885 4.71 Lng Flat (m) TIM (%) Avg Days 17.3 12.374.435 29.00 Trade statistics 2.011.24 98. while the crude TABLE 15.601 15.

one m u s t be very careful not to m a k e any hasty conclusions.13 and Figure 15. Anyway. it could very well be that a market with a negative expectancy contributes positively to the portfolio as long as its winning trades tend to happen at the m o s t opportune m o m e n t . T h e nine mar- . let us take a look at Table 15. oil alone is m a k i n g up for m o r e than 9 0 % of the final profit. it still might add positively to the very s a m e bottom line by being uncorrelated enough to markets to keep the equity level up when the other markets are in a slump. however. W h e n interpreting these numbers. had we only traded only the nine markets that a d d e d the m o s t to the bottom line. in which ending equity is not a straight line function of the equity of the individual markets. Furthermore. it could be that particular market (no matter how profitable it seems to be by itself) simply happens to have all its profitable trades w h e n the portfolio as a whole is in a drawdown. Similarly. This is because these n u m b e r s are picked from the aggregate portfolio. which will result in a lesser n u m b e r of contracts traded w h e n compared to a market that always happens to work at its best when the portfolio has reached a new equity high. note that both the winning and losing percentages are surpassing 100%.CHAPTER 15 Money Management 315 FIGURE 15. A l s o . just because one market h a p p e n s to have a low or negative contribution factor doesn't m e a n it is inferior to other markets.28 to see h o w the directional slope system would have performed. even if it is reasonable to assume that a market will have a negative contribution to the portfolio result.25 The annualized monthly returns for the SDB strategy (both sample periods). For one thing. j u s t for the heck of it.

77 49.30 0.96 100.34 21.316 PART 5 Money Management and Portfolio Composition FIGURE 15.46 7.77 51.61 4.26 The equity curve for the directional slope strategy.12 Individual contribution factors for the directional slope strategy.49 7.82 -59. TABLE 15. Market Corn S&P 500 Juice Live cattle Lumber Coffee Japanese yen Copper Gold Eurodollar Dollar index Cotton CRB Crude oil Canada dollar T-bonds Total Contributing -24.01 .98 -24.71 -23.33 12.25 -3.16 -4.52 91.40 -5.

Z 7 The drawdown curve for the directional slope strategy. C a n a d a dollar.000. "with a long-term strategy you should trade as m a n y markets as possible. this is counterbalanced by a m a x i m u m flat period that is m u c h t o o long. As you can see. Japanese yen. because. cotton.000. Would you then decide to trade it with all the available markets or only the nine markets that. crude oil. coffee. " This is very b a c k w a r d reasoning.26 and 15. which is considerably better than it was from trading all markets. copper. A n o t h e r interesting observation that can be m a d e from comparing Figures 15.26 at o n e point h a d a m a x i m u m total equity of close to $40. Additionally. in hindsight.CHAPTER 15 Money Management 317 F I G U R E 1 5 . however. Honestly now. Unfortunately. o n e reason being that c o m m o n knowledge dictates that. pretend it is early 1998 and that you just finished building this very s a m e system using these very s a m e markets. before the result started to turn south in the middle of 1998. this time the m a x i m u m drawdown has decreased to 2 7 .28 is that Figure 15. dollar index. because all our funds are . kets were lumber. p r o d u c i n g that h u g e w i n n e r that will m a k e it possible to build a house on the m o o n . for o n e thing. considering the deep drawdown that started later that s a m e year? I bet m o s t of you would have gone with the original portfolio. the m o r e markets you trade the m o r e likely it is that you will be caught on the wrong side w h e n panic strikes any o n e of them. turned out to p r o d u c e the best result. 5 % . because you never k n o w w h i c h one will take off next. and T-bonds.

margin constraints might keep us from trading as m a n y markets as we wish.28 The equity curve for the directional slope strategy traded on the nine best markets. to m a k e sure that when the Japanese yen starts to perform just as " b a d l y " at least I have built in s o m e sort of catastrophe protection. It would be foolish to try to squeeze in markets like those mentioned. The argument against this last sentence is that. I know that there are s o m e differences (indicated by the different optimal fs). it m a k e s sense only to trade those that can provide us with the highest likelihood of success. however. I am aware that this last statement to some extent contradicts my previous statements that there is no difference between any two time series and that we should place as m a n y trades as possible. In ten years' time. Just because we used markets like the S & P 500 and C R B index to build the system does not m e a n we have to trade t h e m if there are m o r e viable alternatives. however. But that is in the long run. such as the likelihood of minimizing the d r a w d o w n or flat time by increasing the n u m b e r of trades. the picture might look very different and perhaps then I will be better off substituting all markets completely. I would be leaving a lot of m o n e y on the table while wait- . or m a r k e t s that are too correlated with each other. A n d that is why we are not trading at each market/system combination's historical optimal ƒ In the short run. limited. as long as I can't justify it for any other reasons. and is why we use markets like the S & P 500 and C R B index when we build the system.318 PART 5 Money Management and Portfolio Composition FIGURE 15. Therefore.

54 -1.509. a portfolio of markets.14.33 98. you have to decide what it is you are trying to achieve.971 16. b u t rather to j u m p on a market in motion once the trend has proved itself and then ride that m o v e until our stops and exit techniques tell us that it is no longer there.982 31. lowcorrelated markets. only repeating the discussion regarding the directional slop strategy. True. We are discussing trend-following strategies. we are not trying to pick tops and bottoms. My research shows that a portfolio of 12 to 18 different.1 ing for the evidence to be strong enough to warrant the switch.867 5.91 1.00 Trade statistics 1.83 Lng Flat (m) TIM (%) Avg Days 20. and I urge you to do your own research to c o m e up with a portfolio that m a k e s the m o s t sense to you. Rather than once again going through w h y this is the case.3 7. Strategy summaries Profitability End Eq ($) Total (%) Year(%) P factor Risk measurers Max DD (%) Lrg Loss ($) Winners (%) -27.646 10. lowering the optim a l ƒ for the portfolio as a w h o l e . This is indicated by Table 15. consisting of a m o n e y m a n a g e m e n t regim e n .CHAPTER 15 Money Management 319 TABLE 15.401 1.06 11.709. . let's j u s t move on. That is. The DBS Strategy For the D B S system the results are pretty m u c h the s a m e as those for the directional slope strategy. with the s a m e markets producing similar results. which shows that both the flat time and the drawdown are a little t o o steep. and a trend-following system. A n o t h e r practical reason why we shouldn't trade every market is that the m o r e markets we trade the m o r e inefficient the strategy will be. and the s a m e reasoning holds true for the complete strategy as it does for the single system. A n d because we like to have plenty of r o o m b e t w e e n the ƒ we are using a n d the true optimal ƒ we had better not b o g down the portfolio too m u c h . but my answer to that is the same as for the type I and type II errors: "Better safe than sorry." Furthermore. seems to work best. possibly traded with as m a n y as three different systems. But this conclusion is b a s e d on my personal preferences.13 Results for a preferred portfolio with the directional slope strategy.23 No Trades AvgTr($) Tr/Mark/Year Tr / Month Time statistics 26.

123.21 1. Strategy summaries Profitability End Eq ($) Total (%) Year (%) P factor Risk measurers Max DD (%) Lrg Loss ($) Winners (%) -42.9 Lng Flat (m) TIM (%) Avg Days 21.5% fixed fractional trading with th DBS strategy.787.581 11.00 Trade statistics 1.934 2.14 1.98 11.079 17.1 6.37 -1.425 5.8 .320 PART 5 Money Management and Portfolio Composition TABLE 15.734 33.11 No Trades AvgTr($) Tr/Mark/Year Tr / Month Time statistics 30.67 99.

a n d C a n a d a dollar. orange j u i c e . Tables 16.CHAPTER 16 Portfolio Composition In m a n y ways. cotton. dollar index. natural gas.2 show the result from trading a portfolio consisting of all the above 12 markets. A n o t h e r important factor to consider is to use the s a m e m a r k e t s in several different market/system combinations to cover a w i d e r range of scenarios a n d o u t c o m e s . if we decide to pick the 12 t o p markets from our long-term portfolios (four from each portfolio. crude oil. For instance.1 a n d 16. T-bonds. each o n e in combination with the three different long-term systems. TOTAL EQUITY CONTRIBUTION I O n e of the m o s t obvious (but m a y b e also one of the m o s t naive) ways of picking the m a r k e t s to trade in the future is simply to look at h o w they have performed in the past and select a g r o u p of top performers. but also that a market with a negative mathematical expectancy still might fill an important roll for the portfolio as a w h o l e . plus four short-term systems applied to the S & P 500 (the two entry techniques in the original Black Jack system w e r e divided into 321 . without picking the s a m e m a r k e t twice). we end up with the following 12: r o u g h rice. We discussed finding m a r k e t s that are not too correlated with each other. coffee.2. do some testing on historical data. such as the currencies. we have already covered a large part of the complexity of portfolio composition. and Figures 16. Consider. Japanese yen. t o o . not picking too m a n y markets out of the s a m e group of markets. a n d discuss a few pros and cons associated with e a c h method. D m a r k (Euro). In this chapter we look at three different ways of putting together a portfolio. and to diversify over different time frames j u s t as we should over different markets a n d systems. copper.1 a n d 16.

2 shows that the lowest end-of-month drawdown did not grow past 1 5 % .76 -23.5% fixed fractional trading with Charity I.81 17 . FIGURE 16. 5 % .322 PART 5 Money Management and Portfolio Composition FIGURE 16.000. $75 p e r contract traded was deducted for slipp a g e and c o m m i s s i o n . As usual.304 32.040. These are very good numbers. while Figure 16.1 we can see that the m a x i m u m intramonth drawdown peaked at close to 24 %. for a total of 40 different market/system combinations. for a total ending equity of close to $250.1 0.000. The time period covered is January 1980 to October 1999 and the fixed fractional trading level is set to 0 .528 24. As you can see Charity I m a d e close to 3 3 % per year.1 The equity curve for Charity I. Strategy summary End Eq ($) Total (%) Year (%) Max DD (%) Lng Flat (m) 244. two separate systems). I call this portfolio Charity I. From Table 16.

or nothing but the S & P 500.77 N/A 48 23.04 73.59 0.66 18.41 158. once again showing the importance of trading as frequently as possible.41 11.13 64. This m e a n s that at any particular time we never risk m o r e than 1.46 100.29 -41.32 95.51 0.89 36. For o n e thing.53 270.48 83. It is doubtful this would be a preferable solution for a real-life portfolio. dev. But.00 36 97.20 131. losing Annualized Most recent Average Best Worst St.49 4. which is no m o r e than in any of the single system portfolios discussed earlier.46 4.66 253.68 1. 500 M H z .37 48.15 109.03 37. The only not-so-good n u m b e r is the flat period that might be considered a little too long.5 34. T h e reason behind the relatively long flat time probably can be found in this portfolio's greatest drawback: the high correlation and similarity between several of the markets.2 131.32 6 -11. Unfortunately.76 17.56 -22.48 0.55 0.2 Percentage period returns. one other thing that increases exponentially is the computational time.5% of the total equity going into the trade.32 39.59 425.47 100. we are still raising the total equity tenfold or m o r e .63 N/A 60 24.00 indeed.22 28. there are several traders w h o use only o n e system.53 21.84 1. per market ( 2 % for the S & P 500).97 12 16.77 744.99 -8.95 8. EGM Sharpe Avg.49 41.CHAPTER 16 Portfolio Composition 323 TABLE 16. So.19 309.55 39.07 59.96 985.42 26.72 1. at least Charity I is far m o r e diversified than m a n y other real life examples.99 85.34 257.87 5.01 -12.35 175.98 -3.00 60 196.68 75.1 0.22 28.47 100.34 N/A 24 7. there are as m a n y as four currencies and two energies.06 2.76 309.05 24 15.80 -3.30 -4.67 93.14 34.37 444. Sharpe % winners 1 -11.39 69.28 70.63 1120.04 1 -75.94 19. however. on the other hand. I did all this work on a computer with two Pentium III.24 43.38 0.64 1.14 37. Cumulative Most recent Average Best Worst St.25 28.08 -75.5 t i m e s as m a n y system/market combinations as c o m p a r e d to the original long-term portfolios.98 500.07 15.03 6 -22.63 0.00 48 131.53 3 -3.14 1.5 90.31 3 -14. trading nothing but a few currencies.71 11.93 8.05 89.98 -11. 5 % per trade.59 N/A 36 25.04 500. especially since that we are working with such a favorable portfolio of markets. Despite the very low risk and the fact that we are only trading 2.72 27.52 2. dev.61 269.84 60.76 16.4 539.06 5.99 20.31 42.57 1.38 1. n months rolling window for Charity I.54 93.14 53.31 7.24 -11.27 1.60 180.99 -8.30 -2. in that regard.88 92.03 37. winning Avg.65 180.28 1.96 12 16.94 1.52 28.5 100.14 36. w h i c h m u s t be considered very low. O n e good thing is that we are not risking any m o r e than 0 .05 0.

Data!F3 :F 1036) T h e covariance between crude oil and British Pound in cell J 7 : .324 PART 5 Money Management and Portfolio Composition FIGURE 16.3. This is not something I recommend you to do unless you have plenty of computational power. T h e trick then is to put together a portfolio consisting of markets with as low correlation to or covariance from all the others as possible.2 The drawdown for Charity I. while at the s a m e t i m e m a k i n g sure that the m a r k e t s suggested by the correlation analysis also are likely to k e e p their positive m a t h e m a t i c a l expectations and profit potentials when traded on previously unseen data. T h e correlation between crude oil a n d British p o u n d in cell G 1 0 : =CORREL(Data! $L3 :$L 1036. but it still can take as long as half an hour or more for the computer to do all the necessary calculations. To analyze for correlations and covariances you could put together a spreadsheet like that in Figure 16. processors and 256 MB memory. I c h o s e to look at weekly data. CORRELATIONS AND COVARIANCES A m o r e sophisticated m e t h o d than just picking out the m o s t profitable markets could be to look at their correlations and covariances with each other. In this case.

Swiss franc. Again. coffee. and Figures 16.4 and 16. however.2. Taking $1. T-notes.000 in less than 20 years is not bad at all and plenty more than what the stock market has produced over the same time period.000.3.5.Data!L3 :L 1036)* STDEV(Data!$F3:$F1036)*STDEV(Data!L3:L1036) With the help of the spreadsheet depicted in Figure 16. however. you can look at this portfolio as a worst-case scenario of what we might have been able to produce had we failed to pick the most profitable markets but still used some c o m m o n sense in the portfolio construction process. The criteria for selecting the markets w a s — i n order of importance—to pick only two markets from the same group of markets. Still. =CORREL(Data! $F3 :$F 1036. platinum. to try to find previously unused markets. You can see the hypothetical results from trading Charity II in Table 16.3 Using Excel to calculate correlations. in hindsight. As you can see from Table 16.000 to close to $38. heating oil.3. If you wish. Nikkei index. the flat time is a little too long and this time the maximum drawdown also has moved closer to what could be considered tolerable. TOTAL EQUITY CONTRIBUTION II Instead of j u s t looking at the profit contribution. it also is possible to look at how m a n y of the m o v e m e n t s in a t i m e series can be explained by the movements in . we still would have made more than 2 0 % per year.CHAPTER 16 Portfolio Composition 325 FIGURE 16. I wanted to c o m e up with a m o r e or less new set of markets to see what the results would have looked like when traded on previously unseen and. Eurodollar. The long-term markets I selected were corn. Canada dollar. and wheat. I called this portfolio Charity II. if produced in real-time trading. even with these secondary markets. orange juice. there are plenty of professional traders and analysts—myself included—who would be very happy for numbers like these. not necessarily the most profitable markets.000.1 put together a second portfolio that I traded the very same way as Charity I. and to keep the correlations and covariances between different markets as low as possible.

4 The equity curve for Charity II. To do this we use the Pearson correlation coefficient which returns a value between 1 and — 1. another t i m e series.680 20.5% fixed fractional trading with Charity II.690. T h i s m e a n s that if two markets have a Pearson correlation coefficient close to 1. T h e difference between the Pearson correlation coefficient and the n o r m a l correlation coefficient is that the Pearson correlation coefficient requires that one of the variables be dependent on the other variable.3 0. almost the entire m o v e of the dependent market can be explained by what is TABLE 16. Strategy summary End Eq ($) Total (%) Year(%) Max DD (%) Lrg Loss ($) Lng Flat (m) 37. very m u c h like the correlation coefficient.326 PART 5 Money Management and Portfolio Composition FIGURE 16.59 -28.564 23.89 -1.797.489 3.29 .

provided a certain m o v e in the market. a stock w i t h a beta higher than 1. This is an especially good m e t h o d to use w h e n we have the possibility to trade the same market with several different systems and would like to narrow down the portfolio a little bit. In this case. because this might m a k e the portfolio too dependent on the performance of a few markets. Tests like these are very c o m m o n in the stock market. We are also careful to choose markets that seem to be too good to be true. A major . the two markets m o v e independently of each other. W h a t we look for is. is likely to m o v e twice as m u c h in the same direction as the rest of the market. For instance. markets that influence the total equity positively. N o t e that this is a way of trying to explain h o w m u c h of a particular stock's m o v e can be explained by the m o v e in the market.5 The drawdown for Charity II. W i t h a Pearson coefficient close to 0. w h i c h indicates h o w m u c h a stock is likely to move. going on in the independent market.CHAPTER 16 Portfolio Composition 327 FIGURE 16. b u t not all of that m o v e m u s t be explained by the m o v e of t h e m a r k e t — w h i c h is w h a t we m e a s u r e w i t h the Pearson coefficient. first of all. we do it the other way around and look at h o w m u c h of the total equity can be explained by the total equity from a certain market/system combination. which is not the s a m e as a stock's beta. w h e n traded on previously unseen data in real-time trading. w h e r e m a n y analysts look at h o w m a n y of the m o v e m e n t s in a particular stock can be explained by the movements of the market as a w h o l e . say 2.

5%. and cotton) Pearson correlation coefficients. Just as with the other techniques. $75 was deducted for slippage and commission. Nonetheless.4 shows the Pearson correlation coefficients together with the equity contribution factors for 23 different markets traded with the S D B system over the time period January 1980 to October 1999. feeder cattle.000 and the percentage risked per trade was set to 1. It is a drag.000. and wheat. or to default as h i d d e n all data streams but the first one. I will tell you. natural gas. this m e t h o d is very computationally intensive and time c o n s u m i n g once you have everything up a n d running in Excel. municipal b o n d s . soybeans. Table 16. N i k k e i index. or once a n d for all to define a particular contract as a futures contract. while at the s a m e time our overall m a r g i n requirements have decreased. To do all this it takes me at least 20 m o u s e clicks (conservatively counting) for each market to set up the chart with the necessary system. I deducted the six m a r k e t s with the lowest (Au dollar. so to speak. I m a g i n e doing this on several 100-market/system combinations. Table 16. Before we get started. ending up with a portfolio consisting of sugar. lean hogs. N o t only that. it takes me about a minute (and I ' m b e c o m i n g pretty quick at this) to set up one market/system combination from a set of predefined workspaces that I have built. From these 23 markets. but to get everything up and running you also need to export from TradeStation the trading results for every market/system combination you can think of. working backwards. crude oil.000. copper.5 shows the summarized results for this portfolio. So far so good. C a n a d a dollar. one at the time. this w o u l d n ' t be that bad if it w e r e n ' t for the fact that TradeStation does n o t have a way to apply the s a m e system to several markets (or otherwise manipulate several markets) all at once. cocoa. With these settings. The initial equity was set to $1. a n d British p o u n d ) and the three markets w i t h the highest (rough rice. corn.) .328 PART 5 Money Management and Portfolio Composition difference between this m e t h o d and the correlation/covariance method is that with this m e t h o d we are. lumber. platinum. (With fewer markets.000. Now. because we first trade every market/system combination we can think of in one m e g a portfolio and then deduct all the unwanted market/system combinations from that portfolio. I need to throw in another word of warning. however. live cattle. As usual. gold. A n o t h e r way to illustrate this include the fact that there seems to be no way to set the starting date once and for all. the strategy produced an ending equity of close to $29. coffee. I am not writing a b o o k like this j u s t because I think it is fun to tinker around with TradeStation. Again. but a m a x i m u m drawdown of close to 4 0 % and a longest flat time of m o r e than 47 months would have m a d e this a very difficult portfolio to trade in real life. because the strategy n o w should be a little m o r e efficient. corresponding to a percentage return of close to 1 9 % per year. the initial markets were picked in such a way as not to use those obvious ones already used in many of the previous examples. At this point. to say the least. T h e s e I then traded with a fixed fractional setting of 2 % . Eurodollar. we can increase the risk p e r trade somewhat.

8413 0.58 5.5 Initial strategy summary for Charity III.46 0.1241 0.6305 0.57 .6967 0.6 shows the result from trading this portfolio.64 20.19 1.52 15.86 3.9099 0.8741 0. 5 0 % p e r year.85 -22.17 -2.000.016. Market Au dollar Sugar Soybeans Rice Platinum Nikkei index Natural gas Municipal bonds Lean hogs Live cattle Lumber Coffee Copper Gold Feeder cattle Eurodollar Cotton Crude oil Canada dollar Cocoa Corn British Pound Wheat Contributing -5.40 5. T h e longest flat time also has decreased considerably but is still.8640 0.5608 0.9137 0.6489 0.6592 0.8362 0.66 -2.09 8.79 -4.050 47. corresponding to a percentage return of m o r e than 2 0 .29 -1.83 28.96 -39. Strategy summary End Eq ($) Total (%) Year(%) Max DD (%) Lrg Loss ($) Lng Flat (m) 29.9064 0.0561 -0. at close to 31 m o n t h s . a little too long.60 6.80 -0.313 2.000.802 18.4263 Table 16.54 Pearson -0.CHAPTER 16 Portfolio Composition 329 TABLE 16.9508 -0.97 32.3184 0.4445 0.8302 -0.55 12.7143 -0.82 -9.7810 0.2626 -0.29 -823.9098 0.84 0.4 ( Initial markets for Charity III. TABLE 16.13 -2. T h e d r a w d o w n has decreased to a tolerable 2 9 % . T h e ending equity is close to $40.93 7.

81 All in all. this is not too b a d for a o n e . traded on a b u n c h of markets that were basically chosen because they h a d not been used before or because they were plain average.s y s t e m strategy.330 PART 5 Money Management and Portfolio Composition TABLE 16.07 -1.757 20.565.579.72 -29.683 30. . however.334 3.6 Modified portfolio. strategy summary for Charity III. Strategy summary End Eq ($) Total (%) Year (%) Max DD (%) Lrg Loss ($) Lng Flat (m) 38.

all 16 markets were tradable according to our rule-of-thumb. If the system still holds up w h e n tested on this part of your data. To be sure. and the one I prefer. This table is directly comparable to Table 10.CHAPTER 17 Increasing Your Confidence Let us briefly look at a few techniques for h o w you can increase your confidence in your paper tigers. is to alter the different input variables one at a t i m e by approximately ± 5 0 % . chances are that particular variable might be too sensitive to be trusted with real-time data. but for the out-of-sample period. I also run a best.1 shows how the original S D B system (without exits and filters) fared w h e n tested on the out-of-sample data. If you compare the two.and a worst-case scenario a n d will say m o r e about this shortly.18. but that only four markets managed to increase the value for the average trade. we specifically put some data aside when we built the Standard Deviation Breakout (SDB) system. this indicates that the system is still 331 . you find that 12 markets had both a higher profit factor and a lower standard deviation when tested on the out-ofsample data. trading one contract only. For the in-sample period. covering the period October 1992 to October 1999. and that nine markets also had lower drawdowns. T h e second method. If the result differs too m u c h from the original result. Table 17. OUT-OF-SAMPLE TESTING For the first method. Overall. T h e most obvious m e t h o d is to save some of your data for outof-sample testing. only nine markets were tradable. and you m u s t go back to the drawing board. chances are very good that you have m a n a g e d to c o m e up with a m o d e l that does not j u s t m a n a g e to catch a p h e n o m e n o n only associated with the data you used for your testing and optimization.

75 4.19.39 959.4 and 17.2 and 17.29 7. Tables 17.3 show h o w the S D B system p e r f o r m e d on the out-ofsample data. a filter adds to the performance again.78 2.711.84 0.97 10.366. the value of the average trade continued to decrease for 11 markets.713.26 8. with both the exits and the filter added.16 1.5 show the out-of-sample results for the final system. Original system without exits and filters Market Crude oil T-bonds T-bills Rough rice Nikkei index Natural gas Live cattle Lumber Coffee Japanese yen Copper Gold Dollar index D mark (Euro) Cotton Wheat P.19 469. leaving only a few m a c r o e c o n o m i c oriented m a r k e t s with high point values as tradable alternatives.90 1. 7.342.64 2.012.61 -9.47 2. indicating a lower risk.231.00 -8.61 2 St.23 -14.34 -6.75 124. resulting in only 6 tradable m a r k e t s according to our rule of t h u m b .077.240.008.39 3. something must be done.217.27 0.66 278.918.16 1.51 -2.92 370.677.015. factor 4.28 6.014.414.42 8.2 also is comparable to Table 10.63 1. If the profits are not there.987.70 -6.924.369.26 1.479.29 1.978.13 Drawdown -2.02 -4.19 93.332 PART 5 Money Management and Portfolio Composition TABLE 17.39 14.80 -302. of course.845. but that it also is not as market insensitive as we would have liked.28 5.24 -13. On the m i n u s side is also the fact that we only m a n a g e d to decrease the drawd o w n for four markets. With the filters added.47 1.77 -2.18 4.057.45 -290. but unfortunately.16 doing a good j o b keeping down the risk and the cost of doing business.38 47.05 536.622.39 -3.083.45 -14.471. O n c e again we m a n a g e d to improve the profit factor for 12 markets.45 0. we once again man- .535. In the best of worlds.082.705.38 26. Tables 17. no g o o d a n d m i g h t indicate that we have set the stops and exits too tightly.16 1.13 3.980.06 -53. with the stops a n d exits added.61 -15.060. dev.96 203. T h e n it does not matter that the standard deviation decreased for a total of nine markets.13 2.00 -5. trade 1.78 2.672.89 4.1 The SDB system tested on out-of-sample data.75 2. which confirms the fact that the stops are not doing their j o b s as intended. T h i s is.475.35 1. Table 17.922.976. we should go b a c k to the drawing b o a r d to e x a m i n e w h a t h a p p e n s if we exclude or modify one or several of the exit techniques. Hopefully.23 -9.860.18 1.49 1.40 Avg.909.79 -2.

617.79 2.95 1.839. trade 1. 6. tested on in-sample data (Table 10. .57 2.4) to the very original system.94 0.09 0.504.86 5. however.81 2.36 1.88 -11.80 1.18).86 -3.16 2. Modified system with exits Market Crude oil T-bonds T-bills Rough rice Nikkei index Natural gas Live cattle Lumber Coffee Japanese yen Copper Gold Dollar index D mark (Euro) Cotton Wheat P. which in the end resulted in a total of 9 tradable markets.18 -2.815.477.35 -8.71 -3.30 -724. This also s e e m s to hold true if we c o m p a r e the final in-sample system (Table 13.2 The SDB system tested on out-of-sample data.986. dev.e.97 2.97 11. the decrease c a m e in markets that already w e r e quite profitable.52 -9.85 0.50 1. An ever-increasing profit factor for most markets also is a positive.12 -333.45 6.12 -16. lower standard deviations) for the final out-of-sample system.681. but in conclusion we can say that the fact that the system seems to m o v e toward a m o r e uniform behavior w h e n comparing the different markets is a good thing.56 -4.83 aged to improve the profit factor for 12 markets. This time.608. as are the facts that the drawdown and standard deviation of the value of the average trade seem to stay at relatively low levels.57 -11. but the value of the average trade also continued to decrease for 10 markets.7) with the final out-of-sample system (Table 17.31 411.93 7.035.407.99 10.927.90 1.172.73 0.51 0.79 0.578. than for its original in-sample counterpart. B u t such are the traits of a n o n . factor 3.796.83 12.09 13. we can see that all m e a s u r e s are both lower in value and m o r e uniform (i. T h e b a d n e w s is that several markets did not m a n a g e to p r o d u c e a tradable average profit p e r trade a n d that s o m e even turned negative.34 0.211.311.07 -12.616.m a r k e t .132.250.430.55 1.03 Avg. If we c o m p a r e the final system.044.80 -88.95 -9.914.s p e c i f i c m o d e l . It will work well.29 1.43 2 St.317.71 1.131.22 -5. traded on out-of-sample data (Table 17.91 -6.37 Drawdown -2.007.46 2.74 0.231.61 -18.13 -14.190.053.85 -69.826.75 209.770. Of course there are plenty m o r e research and comparative analyses that c o u l d — a n d s h o u l d — g o into this before you finally decide to start trading your m o d e l .149.4).50 2.05 2.656.90 5.89 4.445.287.99 4.33 596.245.359.44 -271.59 -49.20 -16.CHAPTER 17 Increasing Your Confidence 333 TABLE 17.72 -7.

05% -34.31% 43.59% -59.77% 13.33% 26. we then look at w h i c h individual parameter settings produced the best and the worst results.98% 19. T h e parameters we will c h a n g e are the breakout filter (60 days ± 3 0 days).21% 13.21% 0.90% 113. the trailing stop ( 0 .40% 2.92% -10. With these performance s u m m a r i e s as a base. on a one-contract basis. but there invariably will be times w h e n s o m e markets start to m o v e against you.62% -89. this can be done both before a n d after you have attached the m o n e y m a n a g e m e n t .40% 29.32% 25.44% 2.85% -124. and the stop loss ( 1 .03% 19. on all markets over a longer period of time.20% 4 Better 2 2 1 1 3 3 2 3 1 1 1 1 2 0 3 0 — on average.14% -79.18% -14.97% -63.36% -17.79% 15.25% 9.78% -45. dev. this m e a n s that we must r u n the system test eight times.47% 13.19% -2.04% -127.65% -24.334 PART 5 Money Management and Portfolio Composition TABLE 17. -10. trade -29.38% 11.4 percentage points).00% -18.67% 3. 8 % ± 1.87% -103.22% 5 2 St.07% -13.87% -28.48% 9 Drawdown -23.71% 76. We will change one p a r a m e t e r at the time keeping all the others constant on their original values. In this case.83% -14. and alter all the parameters at once to c o m e up with the worst and the best .42% -41. W i t h four parameters to e x a m i n e .09% 17. Differences Market Crude oil T-bonds T-bills Rough rice Nikkei index Natural gas Live cattle Lumber Coffee Japanese yen Copper Gold Dollar index D mark (Euro) Cotton Wheat Better P.74% -8.5 percentage points).01% 6.72% -189. 6 % ± 0.27% 11.13% 4.95% -23.32% -82. over the period January 1995 to October 1999.63% 19.42% 7.3 percentage points).97% -13.21% -194.71% -12. factor -28.99% 12 Avg.70% -7. As is the case with the out-of-sample testing. the profit target ( 2 . 1 % ± 0. ALTERING THE INPUTS Altering the inputs is another g o o d way to check the robustness of your system.77% 15.90% 30.64% -5. we will look at how the Gold Digger system would have performed on the S & P 500.40% -25.72% 12.48% 17.32% 2.3 Differences between the original system and system with exits.20% -83.20% 36.

75 0.588. y o u w o u l d c o m e up w i t h results that are m o r e reliable if y o u do the comparisons over a whole portfolio instead of o n e single market.228.757.91 1.d a y filter.82 1.48 Dollar index D mark (Euro) Cotton Wheat p a r a m e t e r settings. A n d of course.4 The SDB system tested on out-of-sample data.528.52 0.81 -3.898. a 2 .260. you c a n — a n d should—use other evaluation m e a s u r e s as well.6 a n d 17.49 3.69 -1.04 12. 3 % trailing stop.61 -58.91 1. T h e results are directly c o m p a r a b l e to Table 12. P.084.6%.91 905.48 1. Of course.158.CHAPTER 17 Increasing Your Confidence 335 TABLE 17.d a y filter.48 2 St.78 2. 6.956.11 -4.33 -37. even with the worst possible setting.14 -5.21 587. a profit target of 1.758.022.92 0.61 -8.468.79 -11.54 12.33 -3.11 2.89 774.90 1.42 0. In this case. trade 1.7 s h o w t h e results from t h e worst a n d t h e b e s t p a r a m eter settings respectively.472.50 2.72 -52.348.544. a t r a i l i n g stop of 0 .4%. only you can m a k e that decision and only t i m e a n d plenty of b a d system-building attempts help you gain the experience necessary.895.93 1.767. if we d e e m it profitable e n o u g h or if we would have liked the results to be slightly m o r e similar to those of the original p a r a m e t e r settings. T h e b e s t setting i n c l u d e d a 9 0 .88 Drawdown -2.91 -8. In this case.406.438.85 3. T h e question is.51 -4.926.87 -14.87 -265.8.68 8.332.29 240.485. w h i c h is a v e r y g o o d thing. 6 % stop l o s s .57 1. a n d a 0 .10 2. all p a r a m e t e r settings are j u d g e d by the average profit only.42 6.278.43 235.214. In real life.863.43 -13.556.83 -5.866.43 -10.726.57 0. 6 % a n d a stop loss of 1. possible.34 1.50 2. we should not lose our shirts. T h e w o r s t p o s s i b l e setting t u r n e d out t o b e o n e w i t h a 3 0 . dev.130. With only o n e m a r k e t tested over a very limited .00 1.02 Avg. indicating that even in the worst of times. Tables 17.160.25 -9.048. I would say that I w o u l d have liked to see the worst-case scenario a little better than it is.93 10.68 2.548.05 -6. a 0 .71 1. As y o u can see.526.341. Modified (final) system with exits and filter Market Crude oil T-bonds T-bills Rough rice Nikkei index Natural gas Live cattle Lumber Coffee Japanese yen Copper Gold .88 1.28 6. respectively.82 0.54 14.52 -120.89 6.273. however. factor 3. 8 % profit target.75 -95.42 5. the system is marginally profitable.763.

89% 3. dev.38% 1.275 -65.013 -6.63% 65.45% -29.96% -71.48% 71.24% -11.64% 3.39% 10 Better 2 3 3 1 2 3 2 3 2 0 2 2 2 0 3 0 — t i m e span and using one m e a s u r e r only.14% 24.55% -1.44% 2 Drawdown 8. This is actually good n e w s . 5 2 % .32% 24.91% 11. worst-case scenario.31% 46. factor 1.250 15. 9.07% 5.67% 3.49% -224.64% -1. January 1995-October 1999.792 Winners Lrg winner Avg winner Cum profit 118 5. because it m e a n s that the original settings are doing a good j o b supporting each other in their respective roles and p r o d u c i n g a final system w h e r e the whole is larger than its parts.33% 13.72% 194 1.18% -28.61% -46. although there might be t i m e s w h e n the true best-parameter settings differ slightly.42% -21.20% 181.94% 119. TABLE 17.07% 1.74% 196.255 Losers Lrg loser Avg loser 76 -8.92% -20.82% 17.52% 60.83% 14.05 127 5.86% 39.86% ^0.05% 2.68% ^t2. we do not have e n o u g h data to c o m e to a definite conclusion.348 4. trade 11. Total trades Profit factor Avg profit St Dev 0.22% 33.04% 1.5 Differences between SDB system with exits only and final system.79% 73.97% -40. As for the c o m b i n e d best-case scenario.49% -46.25% -2.66% 12 Avg. we can see that it too produces an average profit lower than that produced by the original p a r a m e t e r settings.68% -41.94% 22.20% -27.1 9 .80% -2176.46% 7.20% 8.43% 270.69% 7.70% 6.85% -129.26% -1.73% 6 2 St.82% -15.20% 19.14% 1.6 Gold Digger.35% 4.880 Drawdown .42% 0.73% 6.336 PART 5 Money Management and Portfolio Composition TABLE 17.25% -32.26% 4.10% 27.33% -63.20% 136.30% -34. Differences Market Crude oil T-bonds T-bills Rough rice Nikkei index Natural gas Live cattle Lumber Coffee Japanese yen Copper Gold Dollar index D mark (Euro) Cotton Wheat Better P.74% -50.43% -1414. however.74% -5.49% 29.17% -51.30% -1.

428 Losers Lrg loser Avg loser Drawdown 10 -8.00% 17.00% -28.76% 50.73% 206 1.17% 1.730 -49.874 126.31 572 5.CHAPTER 17 Increasing Your Confidence 337 TABLE 17.44% 37.11% -14.348 4.14% 1.46% 50.832 Winners Lrg winner Avg winner C u m profit 10 5.30% -1. best-case scenario.013 -3.7 Gold Digger. Total trades Profit factor Avg profit St Dev 0. January 1995-October 1999.815 .

including a system. but even so there have been a couple of occasions throughout this section were I have implicitly argued against him and 339 . not to mention all the hassle you have to go through if you want to test a portfolio! In Part 5 we took a closer look at how to put together a fixed fractional m o n e y m a n a g e m e n t regimen.PART FIVE A Few Final Thoughts About Part 5 The observant reader probably has noticed that up until Part 5 I have consistently talked about " s y s t e m s " rather than "strategies. a n d we did so using percentage-based calculations throughout. It is important to understand that we would not have been able to do this had we not d o n e our h o m e w o r k properly and g o n e through all the excruciating work in the previous chapters. a m o n e y m a n a g e m e n t regimen. is a complete trading plan. and a well-thoughtthrough reason for trading a particular set of markets." on the other hand. and the portfolio c o m p o s i t i o n — w e m u s t know exactly w h a t it is we want to achieve. You c a n ' t even do proper system testing on a single market. to build a trading strategy that is larger than its p a r t s — t h e system. There is no way you can use TradeStation as a standalone tool to build a complete strategy. A "strategy. This is because to m e . I am a big fan of Mr. Vince's work. a " s y s t e m " is nothing but the basic entry and exit rules." That is nothing but another ridiculous way for a software v e n d o r to try to cash in on a new b u z z word or distance itself from another word that might have lost a little bit of its sex appeal. a system without a proper m o n e y m a n a g e m e n t regimen attached to it can never be a strategy. It should now be clear that everything affects everything else. To m e . I have noticed that O m e g a research recently went from using the word "syst e m " to using the word "strategy. and there is no way of cheating." but that in Part 5 I have used the word "strategy" m o r e frequently. the m o n e y m a n agement.

or dollar-based data in heavily trending markets. That is. You should use as m a n y data as you possibly can to m a k e your results as robust as possible. Vince correctly. True. Furthermore. or w h o h a p p e n e d to be president at the time. T h e optimal ƒ is the ƒ that gives you the highest T W R value in accordance with the constraints you have applied to your model. Vince's b o o k s . John Wiley & Sons.340 PART 5 Money Management and Portfolio Composition what seems to have b e c o m e " c o m m o n w i s d o m . extremely high drawdowns. This has stuck in people's minds and m a n y have therefore dismissed the entire theory surrounding fixed fractional investing as something very dangerous. There is no way around this. but he does not address the problem of h o w to get there. and an impossible n u m b e r of contracts traded. I would like to take the opportunity to address these in m o r e detail. 83 ff.. Vince says that for a system to be robust. the largest historical loser being the constraint. then you are probably using too m u c h data anyway (The Mathematics of Money Management. 1992). trading with an / b a s e d on your largest loser is very dangerous. Mr. Before I end this book. Mr. p. it should work on several markets. if I understand Mr. using percentage-based calculations is the only way to m a k e your results comparable a m o n g different markets. times two. what the most recent trend looked like. T h e y read something a n d take it for face value without putting their own thinking caps on. most analysts and system designers consistently refer to the optimal ƒ as the ƒ corresponding to the largest historical loser. But this is only one way of many to relate optimal ƒ to a specific constraint. Unfortunately. but also says that if you have to use so m u c h data that the trend starts to b e c o m e a problem. it should not matter what data we use. It m u s t be a p h e n o m e n o n not dependent on any particular circumstances. because for a system to work in the future it needs to exploit a type of market a n o m a l y that is not dependent on w h a t level we are trading.. because of several practical reasons such as margin constraints. B u t we could equally as well have said that we do not want to experience a loser larger than our largest historical loser. Vince acknowledges the problem of using point.) Another key point throughout this section has been that you do not have to use your largest historical loser to calculate optimal f. (Granted that is not the p u r p o s e of any of Mr. Vince's intention to have everybody stare themselves blind on his way of doing it. C o n c e r n i n g percentage-based versus point-based data. N o t h i n g can be m o r e wrong. Instead of using your largest historical loser to calculate ƒ I suggest that you simply use the stop loss level for each individual trade and that you keep that stop . it is m o r e or less impossible to use an ƒ that high anyway if you like to trade a well-diversified portfolio. In this particular case I do not think it w a s Mr. " or otherwise argued for a different way of approaching a specific problem or topic. but that is the way people work. only referable to that particular point in time.. but as we have seen. because this would result in an even higher optimal ƒ than the optimal ƒ most commonly referred to. we do not want to experience a loser larger than our largest historical loser. but still.

we have stumbled u p o n and acknowledged several such [short-term] anomalies within this book). This also argues against the p o s sibility. we cannot deviate from it w h e n it is time to apply m o n e y management. Those systems would. or whether we are about to go long or short. then you m u s t build a system that exploits exactly that. If we do not believe that the statistical traits for all markets are the same. But this is not so. or even anticipating. especially in the stock market. system builders and analysts have advocated that rather than j u s t looking at the historical trades. A well-working system m a k e s no difference between trying to pick tops or bottoms. and from there calculate ƒ parametrically. whether the trend is up or down. and very. within the confines of a specific strategy. this is what we must assume. to risk m o r e going long (short) in an uptrend (downtrend) either by altering the stop loss level or by altering the bet size. because it would result in a suboptimal solution. curve fitted. no longer be universally applicable and therefore probably m u c h less reliable and robust than any of the systems we have worked with in this book. M a n y portfolio managers. this does not mean that several market-specific or even trend-specific characteristics or anomalies do not exist (in fact. especially for a market-specific syst e m built with point-based data. one market anomaly at a time. a few. to the system we are building.PART 5 Money Management and Portfolio Composition 341 loss level the same. entering on breakouts to the up side or down side. as far as I understand. Finally. However. However. because for a system to work equally as well on several different markets. no matter what m o d e the m a r k e t is in. b e c a u s e there is no guarantee that the future. u n k n o w n . A few also have argued that one should use different ƒs depending on whether the overall portfolio is in a drawdown. whether the market is in an uptrending or downtrending m o d e . the two cannot be combined. If these are the types of market anomalies you would like to exploit. m a k e sure that you actually use an ƒ far lower than that. then there is no way we can build systems that should work equally as well in all markets. but for our purposes. A n d because we assume this. the outcome of each individual trade. in doing so we m u s t assume that we h a d done a b a d j o b constructing the underlying system and that we think we can increase the return from the individual trades by tinkering with the m o n e y management. have to be market specific. however. ƒ w o n ' t be far lower as well. At first glance this m a k e s a lot of sense. m o r e theoretically inclined. we have to assume that there is no difference in the statistical traits of all these markets. or even in which markets all this occurs. because using the m a t h e matical expression for the expected distribution of our future trades would m a k e . This might or might not be so in real life. probably also have tried to combine a fixed fractional m o n e y m a n a g e m e n t regimen with the Capital Asset Pricing M o d e l ( C A P M ) a n d the Efficient Market Hypothesis ( E M H ) . out of necessity. O n c e y o u have calculated the optimal ƒ in accordance w i t h the constraints of your choice. it is better first to calculate the distribution of the outcomes of the trades. That said. not robust at all. very dangerous to trade using any m o n e y management regimen. because with a correctly built system there is no way of knowing. all systems would.

Transmitting b a d vibes. I would say that we had a b u n c h of b a d systems on our h a n d s . You only need a tad more knowledge than you hold about the topic already." "stronger than average. selling something m e a n s bad news and your broker does not want you to think about bad news or what might go wrong. one being the stop loss that we also used for our optimal ƒ calculation. especially if it turns out to be normally distributed. Instead. always at a . we should m a k e sure that each trade only has a set of very few. T h e systems a n d strategies we explored so far are simply "paper tigers." or "sell if it reaches. if they did. Unfortunately. But because neither you nor I can blow hundred dollar bills out of our noses. T h e good news is that you do not have to be a rocket scientist to do so. so they have no incentive to educate you. We looked into the distribution of our trades w h e n we put together the stops and exits for our short-term systems. but always. In this particular case. either with a profit or a loss." "buy." W h y is this? It is because in this industry. essentially leaving each trade to its own destiny as soon as we have entered into it. Go through all the recommendations from several analyst and brokerage companies and you will find a multitude of opinions such as "strong buy.342 PART 5 Money Management and Portfolio Composition our m o d e l s even m o r e forward looking than using the actual." "outperform. whether professionally right or not. very specific and clearly distinguished outcomes. we would not be able trust our systems to a fixed fractional m o n e y m a n a g e m e n t regimen in the first place. because what is devastating for your economy is a gold m i n e to theirs." The strongest phrases in this regard seem to be " m e d i u m perform" or "hold. Remember. That said." but seldom will you find the words "sell. is not good for business." "positive outlook." "long-term buy. the distributions we c a m e up with hardly lend themselves to any parametric calculations of ƒ In fact. because when it comes to basic trading and investing. getting out of a position is what nobody really talks about." "cheap. If it were not for these techniques. historical trades. we are probably doing a very b a d j o b in m a n a g i n g our trades. it is no coincidence that our discussion on stop placements has the most exhaustive one. to b u y something (or go long or short on a futures contract) usually m e a n s we need to get out of something else." BACK TO THE FUTURE T h e financial markets are not r a n d o m and there are ways to take advantage of this n o n r a n d o m n e s s with the help of mechanical trading strategies. Considering the above. knowing too little about a specific topic sometimes can be m o r e devastating than knowing nothing at all. T h e big fish in this industry realize that. it is thanks to our stop and exit techniques we are able to exploit non-market-specific anomalies." "sell if it drops to. using no stop or exit techniques to streamline the o u t c o m e . Because as soon as the distribution of our trades starts to resemble something that we can put a n a m e to. there still might be plenty of systems with normally distributed trades that work m u c h better than those in this book.

although the small. N o t only that. distinguishing u p . T h e stops and exits not only form the core in the trading process but also set the stage for a profitable m o n e y m a n a g e m e n t regimen. For instance. A n d as if this w e r e not enough. you also m u s t know h o w to calculate each trade's m a x i m u m adverse excursion ( M A E ) and m a x i m u m favorable excursion ( M F E ) . For each of our e x a m p l e systems we put together a set of stop and exit techniques that e n h a n c e d the performance for every single o n e of them.m a r k e t .e. down.PART 5 Money Management and Portfolio Composition 343 point where we believe our m o n e y can c o m e to better use elsewhere. the statistical traits b e c o m e d y n a m i c w i t h measurable differences. for practical reasons. Therefore. B e c a u s e this figure tells you nothing about w h a t to expect in the future. B u t the decision to get out of the same position is an ongoing process.s p e cific discrepancies that are likely to occur again a n d again in all markets. in another trading or investment opportunity. the end trade d r a w d o w n ( E T D ) . Unfortunately. and the total equity d r a w d o w n ( T E D ) . the closed trade d r a w d o w n ( C T D ) . but for our short-term syst e m s these techniques were generic and n o n . or under your mattress. and side- . nonmeasurable changes that might occur over time lead up to a n d form the longer-term trend. down. If y o u do not k n o w or cannot form an opinion regarding the direction of the long-term trend. in accordance with our research. you m u s t k n o w h o w to divide the d r a w d o w n into several different subcategories. by j u s t looking at the last five days of market action. approximately three to nine days) using as little historical data as possible for your decision making. T h e statistical traits for the markets are pretty m u c h the s a m e whether the trend is u p . This brings us to two other interesting conclusions: • • You get a long-term edge in almost every market if y o u only trade with the direction of the underlying trend. such as the start trade drawdown ( S T D ) .m a r k e t . j u s t as it gives y o u no clue whatsoever—absolutely n o n e — a b o u t h o w to go about doing something about it. This m e a n s that getting into a position is a one-time decision and once it is m a d e it is over a n d done with. you are better off trading short t e r m (i.s p e c i f i c best c o m p r o m i s e s . In Part 3 I showed you h o w to do all that a n d m o r e . to research how and w h e r e to place your stops and exits you m u s t k n o w a whole lot m o r e about your system than its m a x i m u m historical drawd o w n in dollar t e r m s . w h e r e you constantly ask yourself "will my m o n e y c o m e to better use elsewhere?" No matter if this " e l s e w h e r e " h a p p e n s to be in the b a n k account. you likely are not able to tell the direction of the underlying longt e r m trend. principally developed with the help of r a n d o m entries and later applied to a m a r ket other than those used during the testing procedure. or sideways. That is why no trading or investment technique is complete without a set of thoroughly researched stops and exits. For this. the short-term traits can be said to be static in nature. This m e a n s that there w a s virtually no curve fitting involved a n d that we are exploiting are n o n . At that level.

Because if we do not understand and acknowledge this. influenced by the increased security you feel about k n o w i n g the long-term trend.344 PART 5 Money Management and Portfolio Composition ways trends from each other. but "optisizing" your trading positions in relation to your overall strategy and other personal constraints and preferences. both in the short and the long run. is that you dare to w o r k with limit orders. or perhaps even scrap it all together if it turns out that we cannot c o m e up with a way to fit the filter to the rest of the machinery as efficiently as we would like to. To this set of rules and frameworks we also need to add the understanding of the trading process and the creation of a rule-based trading strategy as a flow of events and intervening decisions. Not until all the pieces from Parts 1 and 4 fit snugly together can we attach the m o n e y m a n a g e m e n t that not only transforms the system to a strategy. rather than individual decisions separated from each other in time and space. was what Part 4 was all about. we cannot see the benefits of the type of trading a n d that it is not a matter of optimizing your system's individual signals to the underlying data. and the t h e o r y w h y all this is happening in the first place. but only in the direction of the long-term trend. To figure out w h e r e the market is heading. trying to pick tops and b o t t o m s . but also will m a k e the whole picture larger than any of its parts. Another major advantage of trading short term. . W h e n attaching the filters we also might have to go back and adjust the original idea a little.

7 9 ( 0 . 233 Black Jack system. 213-214. 69-81. 2 4 0 ( 0 Gold Digger system vs. 257 asymmetric distribution. 213-214. maximum adverse excursion (MAE). 22. 53. 216 Meander system combined with. 31 best-fit regression lines in. 6-7 accumulation area in market analysis. 264 adoption-diffusion model in market analysis. 200-day moving average. 204-205 ratio of average profit to standard deviation in. 187-188. 217-218 entry techniques for. 214 average trade in. 74 performance summary of. 9 back-adjustment method of splicing. 149 average percentage profit. 241-242. 216-217.8 0 ( 0 . 214 average trade value.216(0. 263. 321 profit factor in. 77. back-adjusted method. 7 average true range indicator. 188-189 average dollar value per trade.. 204-221 average percentage of profitable trades in. percentage values. 241 average weekly true range (AWTRO/AWTR). 252-253 in filtering. 47. 74. 78-80. 217-221(0 portfolio composition of. 209. 256-257 account size vs. 35 beta value. 211. 148. 31.217. 219-221. 217 closed trade drawdown (CTD) in. 216. 7 5 . 23. MAE/MFE. 213-214. 2 1 2 ( 0 . 2 1 3 ( 0 maximum adverse excursion (MAE) for. 64 academic theories of market trends. depth of historical drawdowns. 216 maximum favorable excursion (MFE) for. 76. 28 in Black Jack. 74 ratio of percentage profitable trades and standard deviation in. 7 7 ( 0 percentage profitable trades in. 217-221. 7 6 ( 0 . 75 345 . 129-130 bars per trade. 70. 219-221(0 minimum profit level statistics for. 214. 233. average number of. 218 profit target in. 208. 32-37 ratio-adjusted method of splicing and. 208. 211. 208. 263 ADX indicator.INDEX Note: Boldface numbers indicate illustrations. 216. 228. 31. 327 Black and Scholes pricing formula. 153-154. 210 average percentage winner. 1 5 4 ( 0 average profit multiplier (APM).188 best-fit regression lines. 234 arbitrage. 229 absolute vs. 205. 83. 10(f) basic crossover signals. 84 average winning and losing trades. 209. 218-221(0 kurtosis in. 78 average percentage profit calculation in. 210 average return in. Black Jack. 20 average losing trade. trends in markets. Italic t indicates a table. 239-240. 37-38 bar count/calculation in TradeStation. 35 point-based adjusted method of splicing and. futures contracts. 28 average profit. 209. 218-221.148. 215-216 random number generator for. 220-221 exit techniques for. 9-10. 209. 209 cumulative profit and equity tops in.24 bell curve distribution. 22.219 filtering in. 87 filtering. return on account. 86-87 bear markets. 250. 281-282 average return.

210 testing.123 Capital Asset Pricing Model (CAPM). 266 credit policies and market trends. 214-215. 160. 47. 63. 206-208 trailing stops in. 70 trend filter in. 6 0 ( 0 . 134-135. 262-263. 174-175. 90-94. 191. 87-97.135. Dave.. 70 in Gold Digger I system.262 Crash of 1987. 261. 69-81 200-days moving average in. 23. 24 Directional Slope Method. 3 2 9 ( 0 . 17 Commitment of Traders Report. 325. 70 demographics and market trends. 123. 316(0. 77-78. 70 drift in. 28. 3 2 5 ( 0 covariances. 53. 282-283 building a system. 217-218. 62(0 in Meander system. 6 8 ( 0 . maximum number. 314(0. 124-129 data sort function in. 257. 55-81 absolute vs. 6 9 ( 0 normalization of market movements and. 57(0 regular moving averages in. 205-206. 205. 324-325. 63 calculations for. 89 Dynamic Breakout System(DBS) vs.. 322-324. 55-56 uncertainty measures in. 262 computer software for evaluating performance. 56-62. using historical data. 74. 108. 121. 74 standard deviation in. 327 Charity III portfolio. 209. 52-53 Charity I portfolio. 233 Bollinger-band indicators. 64-69. 86-87 cumulative profit.319(0. 10-11 correlations. 37-38 creation-of-funds analysis. 206 time-based stops in. 52-53 decision to trade. Fisher.. 209. 5 7 ( 0 . 208. 48. 216. 5. 259-260 Dennis. 83-84 DeLuca. 23. 324-325. 2 1 2 ( 0 . 205-206. 215 uncertainty measures in. 274. 215. 328-330. 7 8 ( 0 time-based stop for.211-212 surface chart for. 174(0. 88. 115-116. 176 .211-212 TradeStation code for. percentage values in. 23.. 324 Charity II portfolio. 76-77 trade-length statistics for. 257 Black-76 pricing formula. 107-113. 63 summary report for. 219 time-based trailing stop in. 260 data mining.317.318. 105. 89 code for. 223 DeMark. 258. 316. 77 Black. 29.8 0 ( 0 Bollinger-band type indicators in. 70 time-based trailing stop in. 62-69 in Black Jack I system. 76-77 trailing stop for.218 stop loss statistics in. 209. 74 standard deviation of percentage profitable trades in. 3 3 0 ( 0 Chebchev's theorem. 274 depth of historical drawdowns. 214-215. 64 percentage of moves of certain amplitude and. 136. 70 triggers for. 75-76. 167. 113(0 crossover signals. 194 closed trade drawdown (CTD). 3 2 6 ( 0 . 9 3 ( 0 . 262 Commodity Futures Trading Commission (CFTC).313 average profit factor in. 7 5 ( 0 . 217. 1-2 consecutive winners/losers. 9 4 ( 0 . 209. 70 profit target for. 210. 282. 122. 343 commission.) skew in. 80-81. 7 8 ( 0 . 59-62. 59 for tracking market moves. 22. Tactics and Strategies. 70. 2 1 3 ( 0 standard deviation of percentage profit in. 5 8 ( 0 periods of moves of certain lengths and. 77 data selection for analysis. 3 2 5 ( 0 Crash of 1929. 67. 341 catching market moves.110. 52 Campaign Trading. 77-78. 211. 212. Tom. 8 0 ( 0 . 21. 58-59 exit techniques and. 69-81. 58(0 trends in markets and. 281 curve fitting. trends in markets. 3 2 3 ( 0 .346 Index Black Jack system (cont. 64 better use of data in. Richard. 322. 5 7 ( 0 for testing a system.111. 7 5 . 97-98 dependencies. 313-319. 107. 70 stop loss for. 161.

9 4 ( 0 . 160. 136. 5-6.. 190-191 symmetric distribution in.223. 181(0. 30 dollar value of largest loser. 90-92. 23. 179. 279 DMI-plus/DMI-minus lines. 170 exit techniques in. 187-221. 279-280. 6. 23. 135-136.169. 87-88 Dynamic Breakout System (DBS). 247(0. 168 filtering in.124. 136.169. 169-176(0. 29 dollar value of largest drawdown. 123. 6. for market analysis. 23 run ups.. 246.319. 121-124. 29 drawdown. 9 6 . 144. 23-25. 179.319(0 out-of-sample testing of. 130-131. 180-181 . 191-197 kurtosis in. 9 7 trend strength vs. 5. 264 asymmetric distribution in.131. 159-160 money management in. 343 depth of historical drawdowns.179. 9 2 exit techniques for. 189-190 standard deviation in. 158(0. 161. 187-188. 97-107.188 Chebychef's theorem in. 9 3 ( 0 . 216. 99-101. 141 testing of.131.273. 188 diversification in future contract data. 168 MAEfe in. 136. 187-188. 161-165. 343 favorable excursion (FE) level (MAEfe). 168-177. 189 leptokurtic distribution curves in. 105. 9 7 distribution area in market analysis. 319. 9 1 .. 242-247.272. 38-41. 3 9 . 282 risk vs. 123 drift exit techniques and. 90 modifications to. 3 1 4 ( 0 .135. 6.. 316. 70 in market trends. 264 distribution curves {See also market analysis. 169-176(0. 190 platykurtic distribution curves in. 343 maximum favorable excursion (MFE). 320(0 code for. 140. 88-89.318. 234 dollar value of average loser. 95-97.124. 242. 101. 156-157. 140. 263. 9 6 . 131. 121-122. 29 dollar value of average trade. 2 5 2 ( 0 . 134-135. 23. 3 2 0 ( 0 performance summaries for. 123. 123 open trade drawdown (OTD). 243. 143-166. 172-177. 29 dollar value of largest winner. 188-189 bell curve distribution in. 29 dollar value of average winner. 130-131. 24 dollar value of largest drawdown. 343 maximum intraday drawdown in. 23. 191 distribution area in. 208-209.122.136. 95-97. 4 0 in money management. 173-176(0 profit analysis for. 9 6 . 272. 343 TradeStation and. 131-134. 25 largest percentage drawdown. 174-175. 89-90. 343 closed trade drawdown (CTD).135.244. 149 matrix of values for. 245. 122. 159-160 filtering in.) out-lyers and.100 MAE in. trends in markets). 156(0. 23 Meander system and. 90-92. 9 2 Exit MA for. 106(0. 180(0 flat time. 264 distribution curves in. 9 7 performance summaries for. 188-189 normal distribution curves in. 135-136. 123.317. 168-177. 23. 30 in efficient trades. 168 money management in. 170 Standard Deviation Breakout system (SDB) vs. 29 maximum adverse excursion (MAE).132.316(0. 263. 208-209. 188 Gold Digger II system in. 157-158.Index 347 Directional Slope Method {cont. 174(0 end trade drawdown (ETD) in. 98-99 directional slope method vs. 252-253. 2 5 3 ( 0 MAE/MFE and. 170. 133. 23. 188. 187-188. 107. 5. 138-139. 89 STD/MAE in. 143-166. 9 1 . 161-165(0 ratio of profit factor and standard deviation in. 141. 121. 25 start trade drawdown (STD).188 accumulation area in.) Entry MA for. 131-134. 343 total equity drawdown (TED). 141. 313-319. 208. 122. 23. 209.216. 4-5. 95-97. 189 skew in.138. 123-124. 5. 132-134. 248. 138-139. 188 drawdown {cont. 167. 121-141 end trade drawdown (ETD).135. 245(0. 249(0 lookback period in. 263. 167. 4-5. 177-186.

123 "equally well" vs. The. 136.131. 29 dollar value of largest drawdown. 171-172 trade statistics summary for. 136 in Black Jack. 99-101. 119-121 start trade drawdown (STD) in. 122. 140 final profit. 17 computer software for.135. 257.131. 138. 121-141 end trade drawdown (ETD). TradeStation equations for. Bill.) stop loss in. 134-135. 123. 123. 30 dollar value of largest loser. 140. 123. 123-124. 133. 29 diversity in. 122. 102-107. 220-221 closed trade drawdown (CTD) in. 121. TradeStation. 28 average trade value. 103-105. 138-139.135. 28 average percentage winner. 124-129 drawdowns for. 122. 161 estimated TWR (ETWR). 133. 136 directional slope system for.23 equity tops. 138-139. 136 money management and. 97-98 efficient frontier. 181(0. 216-217. 123-124.140 long trades.140. 105(0 trailing stops in. 160. 6.131. 179. 7-8 largest winning/losing trade. 282. 135-136. 23 commission. 120. 21. 271 start trade drawdown (STD) in. 6. 107 volatility in. 134-135. 168 stops in. 20-21 maximum adverse excursion (MAE). underwater equity curve. 132. 119-142 bar count/calculation in TradeStation. 25 futures contract data. 131-134. 258. 31-44 gross profit/loss. 21 flat time.122 TradeStation code for. 141 Meander system and.100 entry points. bad. 140 EasyLanguage. 281 Euro currency. 131-134. 157-158. 22. 8 largest percentage drawdown.120 maximum adverse excursion (MAE) in. 28 largest winning and losing trades. 130-131. 129-130 closed trade drawdown (CTD). 121-123.132. "equally profitably" defined. 158(0. good vs. 123. 170.124. 28. 55 evaluating performance.140 total equity drawdown (TED). 119 end trade drawdown (ETD). 103-106(0 time-based stop in. 135-136. 123-124 RINA system and. 123. 130-131. 23 maximum consecutive winners/losers. 103(0. 130-131. 4-5 dollar value of average loser. 23-25 end trade drawdown (ETD).100 test of. 136. 23 equity highs/tops. 138-139. 1-12 account size vs. 138-139. 29 dollar value of average winner.135. 134. 274 in Black Jack. 138-139. 9-10 closed trade drawdown (CTD). 341 efficient trading.132. 135 total equity drawdown (TED)in. 97-98.348 Index Dynamic Breakout System (cont. 1-2 cumulative profit. 122. 141 maximum favorable excursion (MFE) in. TradeStation equations for. 23. 281-282 Equis (See MetaStock) equity.132. 5. 131-134. 220-22lend trade drawdown (ETD) in. 104(0. 122. 29 largest percentage winner. 343 . 20 average percentage profit. 29 dollar value of average trade.135.124. 21. 6-7 average dollar value per trade. 123 overriding filter in. 124. 168 triggers of volatility for. 136. 133. 135-136. 121-124. 134.135. 131-134. 7 average winning and losing trades. 138-139. 282 Efficient Market Hypothesis (EMH). 123. 130 short trades and. 29 drawdown. average number of. 29 largest percentage loser. 29 dollar value of largest winner. in money management. 9 bars per trade. 134 pivotal resistance level/support levels in. 101-102 turtle traders in. 133. return on account. 99-101. 10-11 STD/MAE in. 216-217. 122. 1 Eckhardt. 134. 52 Equation. 119.

237-239(0 high/low breakout indicator in.3 0 MetaStock program for.4 profit factor. 2 3 4 . 6.Index 349 evaluating performance (cont. 7 1 . 179.134. 134-135. 29 open trade drawdown ( O T D ) . 29 start trade drawdown ( S T D ) . 245. 70 in Standard Deviation Breakout system ( S D B ) . 234 overriding. 2 2 8 . 70 time-based trailing stop in. 7 5 . 23 exit techniques. 23 percent profitable trades. 242 in Meander system. 342-344 200-days moving average in. 2 4 9 ( 0 in Gold Digger system. 70 trend filter in. 234 in short-term systems.9 . 2 3 9 . 8 . 70. 2 4 2 . Eugene. 122. 2 3 8 .217. 29 percentage of winning trades.8 0 ( 0 . 91. 2 4 0 ( 0 moving averages. 229 A D X indicator in. 9 . 1 3 2 . 123. 30 profit per trade. 234. 4—5 TradeStation program for.1 9 .2 4 7 .9 2 .6 .1 0 number of losers. 167-186 in Meander system. 246. 4 . 229. 6 . 29 performance summary report. 225 kurtosis in.1 7 6 ( 0 filter techniques. 1 6 8 . 2 3 6 . 23 maximum intraday drawdown in.8 1 . 179. 123. 274. 219.2 3 6 .1 3 4 .) stop loss for. 122. 135. 136 in Directional Slope Method. 7 8 ( 0 time-based stop for.1 7 6 ( 0 end trade drawdown ( E T D ) and. 233 profit factor and. 1 9 return on account.216(0.2 3 9 . 70 in Dynamic Breakout System ( D B S ) . 233 in Black Jack system. 2 5 2 . 243. 250.2 5 3 . 1 3 1 . 156(0. 1122-124. 2 4 1 . 17 standard deviation calculation. 205. exit techniques (cont. 9 0 . 6.1 3 6 . 117-118. 23 total net profit in. 282 multiple markets and. 234 in Dynamic Breakout System ( D B S ) . 235 long-term volatility. 6 9 . 136 . 1 6 9 .1 7 6 ( 0 . 267 200-day moving average in. 70 in Black Jack system. 14-16 percentage of losers. 124.7 2 profit target for.) maximum favorable excursion ( M F E ) .7 7 total equity drawdown ( T E D ) in.1 3 1 . 135. 220 money management and. 2 4 5 ( 0 . 2 4 7 ( 0 .7 8 . 248.2 5 4 lookback periods for. 2 4 0 ( 0 Black-76 pricing formula and. 199. 28. 25 slippage.2 4 2 . 236 oscillators in. 123 trailing stop for. TradeStation. 2 3 1 . 1 6 9 . 191 long-term exits and. 1—2 number (winning/losing) trades. 2 1 . 3 5 . 230.219 closed trade drawdown ( C T D ) and. 70 uncertainty measures in. 180(0 Standard Deviation Breakout system ( S D B ) .1 3 9 . 231 high-probability. 1-2 underwater equity curve. 23 total equity drawdown ( T E D ) . 242 RSI indicators in. 7 6 .2 3 6 92 drift in. 5. 3 . 134 pricing formulas and. 177 start trade drawdown ( S T D ) in. 180(0 Dynamic Breakout System ( D B S ) . 23 measures of performance in. 3 .2 3 2 . 13-14. 168-177. 1 6 9 . 1 3 1 . 241 Black and Scholes pricing formula and. 2 3 5 .216. 133. 168-177. 252(0. 70 triggers for. 257 favorable excursion level ( M A E f e ) . 2 3 6 ( 0 maximum adverse excursion ( M A E ) in.1 7 7 .2 4 0 . 244. 28 number of trades. 77 fading a system.2 3 5 on-balance volume ( O B V ) indicator in. 6 9 . 1 3 0 . 167 Fama. 117-118. 1 7 . 2 7 1 . 6 price vs. 1 6 9 . 2 2 7 . 177-186. 5.7 run ups.2 4 0 . 101. 252-253 average true range method in. 2 4 1 .169.8 1 . 233 in Directional Slope System. 1 3 8 . 140 in Gold Digger I I .169. 233. 253(0 DMI-plus/DMI-minus lines in. 7 7 .2 3 8 . value in. 156-157. 10-11 percentage calculation of performance. 228.

Excel.) skew in. 273. 29 largest percentage winner. 234 sentiment indicators in. 37-38 splicing contracts in. 4 3 point-based adjusted method of splicing in. 236-238.196(0. 233 volatility and. 31 volatility in. 189. 3 9 . 231-232 TradeStation code for. 7-8. 218 stop loss in. 259-260 fixed fractional money management method. 6 0 ( 0 . 225 holding period return (HPR). 38-41. 281 global market structures. 7 largest winning/losing trade. gain vs. 264. 275. 241-242. 31. 191 filtering in. 194. 51. 32-37 ratio-adjusted method of splicing. 263 oscillators. 231 high-probability filters. 29 largest percentage loser. 252-253 average true range indicator. nelson. 192. 262-263. 6 2 ( 0 . 279-280. altering inputs for. 235 in Standard Deviation Breakout (SDB) system. 3 3 7 ( 0 skew in. 194. 261 gross profit/loss. 8 halfway flags in market analysis. 234. 192(0. 218-221.339 flat time. risk. 241 DMI-plus/DMI-minus lines. 31-44 back-adjustment method of splicing in. 31 ratio-adjusted method of splicing in. 193(0 kurt function.) performance summary of. 59-62. 229. 275. 230. 195-197(0 profit factor in. risk in. 25 flow-of-funds analysis. 218-221(0 Chebchev's theorem in. 249-251(0 start trade drawdown (STD) in. 195. 234 RSI indicators. 281 indicator piling. 195-197. 2 1 2 ( 0 . 263 on-balance volume (OBV) indicator. 194-195 gross domestic product (GDP). 258-259 Gold Digger systems. 31 best-fit regression lines. 20 . 189 kurtosis.198 in Meander system. 53. 192-193. 203 largest percentage drawdown. 233. 193(0 standard deviation in. 272. 231-232 interest and amortization in market trends. 211. 70 futures contract data. 247-252. 31. 232-233 true range method in.218 robustness test of. 234. 35 multimarket portfolios in. 276. 242 T-bill ratio. 193 mean in. 237-239(0 kurtosis in. 265 high/low breakout indicator. 191-197 Black Jack vs. 4 2 . 272. 192-193. 41-44.. 223 in Black Jack. 35 point-based adjusted method of splicing. 196.350 Index filter techniques (cont. 2 1 3 ( 0 filtering and. 241-254 fiscal policies and market trends. 259. 334-336. 31-32 nonadjustment method of splicing contracts in. 193(0. 191 stops in. 227 ADX indicator.264 gain vs. 262 T-bill ratio filter. 31 perpetual adjustment method of splicing in. 192-193. 192. 336(0. 250. 235 in Gold Digger II.198 leptokurtic nature of. 282 geometric mean. 28 largest winning and losing trades. 235 standard deviation in. 194 exit techniques in. 31 perpetual contracts in. 266 Freeburg. 227-228 indicators. 20-21. 37-38 best-fit regression lines in. 199. 233. 231-232. 260-261. 231 market microstructure indicators. 236. 201. 194. 196(0. 21. 192-193. 35 Gold Digger systems (cont. 273. 193(0. 4 0 nonadjusted data in. 234 high/low breakout indicator. 196. 31. 211. 193 median in. 195 TradeStation code for.

306. 301 market analysis (See also distribution curves. 327 Capital Asset Pricing Model (CAPM) in. 265. 319.258 Commitment of Traders Report in. 8-9 TradeStation equations for. 259-260 flow-of-funds analysis in.313-319. 256. 261. 149 average winning and losing trades in. 257 beta value and. 283. 9-10 number of losers in. financial markets. 152.314. 29 gross profit/loss in. 242 momentum changes in market trends and. Bruno. 257. 121 Directional Slope System for.312. 83 margin to equity ratio. 148. 258-259 gross domestic product (GDP) in. 235-236. 51 macroeconomic-oriented markets. 241-254 lookback period Dynamic Breakout System. 316. 252(0. 29 dollar value of largest loser in. 259 New Economy and. 121. 247-252. 149-150. 250. 326-329 portfolio theory in.244. 305-313. 2 4 5 ( 0 .245.318.313(0. 248. 262-263.316(0. 261 . 9 bars per trade. 319 profit factor in. 150. dollar value of average loser in. 342-344 academic theories of. 311.152 maximum favorable excursion (MFE).151 maximum consecutive winners/losers in. 259. 252-253. 257 real vs. 264. 260-261. 305-313 optimal fin. 2 3 6 ( 0 losses/losing trades average dollar value per trade in. 257 random walk hypothesis (RWH) and.153 leptokurtic distribution curves. 266 money management in. 317. 266 credit policies and. 242-247. 265-266. 241 closed trade drawdown (CTD) in. 152. 261 on-balance volume (OBV) indicator in. 265. 120. 8 largest percentage loser in. 255. 256-257 accumulation area in. 249-251(0. 2 4 7 ( 0 . 308.Index 351 Latour.120 volatility filters and. value in. outcomes in. 260 demographics and. 241-242. 29 dollar value of average trade in. 262 Commodity Futures Trading Commission (CFTC) and. 9-10. 29 profit factor in.100 filtering. 264 Efficient Market Hypothesis (EMH) and. 261 halfway flags in.262 creation-of-funds analysis in. 122 losses/losing trades (cont. 167-186 maximum adverse excursion (MAE) in. 264 adoption-diffusion model in. 343-344 ADX indicator in. 264 market microstructure indicators in.266 money supply (Ml. 257. average number of. 262-263. trends in markets). 188-189. 266 global market structures and. 306-309(0.314(0. 283. 246. 318. 119. 7-8. 262 Crash of 1929. 252-253 average true range indicator in. 259-260 distribution area in. 253(0.311. 320(0 exits for. 257 price vs. 20-21 MAE/MFE and. 10-11 percentage of losers in. 249(0.) number (winning/losing) trades in. 265-266. 263. 28 percent profitable trades in. 258 fiscal policies and. 243. 242 Standard Deviation Breakout (SDB) system for. 265 interest and amortization in. 242 total equity drawdown (TED) in. 121-122. 20 average losing trade in.315 start trade drawdown (STD) in. 259 net present value (NPV) in. 10-11 MACD indicator. 29 largest winning and losing trades in. 263 Pearson correlation coefficient in. 255-266. 263 arbitrage principles and.148.319(0 Dynamic Breakout System (DBS) for. 99-101. 255 long-term investing systems. 193 Livermore. 263 momentum changes in. 258 least square regression maximum adverse excursion (MAE). M2) in. 263. Jesse.

The. 219 weekly data for. 152. 123 maximum consecutive winners/losers. 148. 148. 155(0 start trade drawdown (STD) in. 303(0. 242 least square regression in.140. 143. 199. 293. 143-166. 155 trailing stop and. 269. 220 filtering in. 154-155. 10-11 maximum favorable excursion (MFE). 138-139. 148 standard deviation in. 256. rules of.216 definition of. 208-209. 219-221.149 average percentage move for each bar in.136. 148 in Standard Deviation Breakout system (SDB). 149-150.. 306 out-of-sample testing for. 177-186. 23 mean in Gold Digger II. 295-302.139 mean in. 150.) rule-based trading in.133. 23. 145 market test for. 132-134. 216.147 regression equation for. 157-159.. Harry. 167. 9 . 257 Mathematics of Money Management. 23. 145 directional slope in. 202-203. 6 9 ( 0 . 156(0 filtering and. 340 maximum adverse excursion (MAE). 209. 145 directional slope in.216. 130-131. 5-6. 2 9 8 ( 0 . 146-147. 238-239.148. losers and. 147-148. 148 rules for. 155 time-based stop and. 201.132. 263 market-specific vs. 2 4 0 ( 0 kurtosis in. losers in. 203 performance summary of. 158(0 favorable excursion (FE) level (MAEfe) in. 47 Markowitz. 199.151 maximum intraday drawdown in. 131-134.150 least square regression in.169. 199.148. 23. 148.146. 197-204.134 exit techniques in. 7 average winning and losing trades. 168-177.179. 177-186. 199 money management in. 169-176(0 end trade drawdown (ETD) in. 28 average trade value.124. 136. 145 market test for. 6 8 ( 0 .. 145. 135-136. 6-7 average dollar value per trade. 3-30 account size vs.138. 145. 199. 4-5. 141 time-based stop in. 202 summarized life of trade in. 193 in Meander system. 258 sentiment indicators in. Analyzing Price Fluctuations. 147-148. 152-153. 149. 146-147. 154-155. 209. 123-124. 141.152 lifespan and development of trade using. 145 winners vs. 144. 167. 145. 2 0 2 ( 0 . 257 speculation and. 208.131. 256 technical analysis in. 219. 151 Maximum Adverse Excursion. 343 average losing trade in. 145 trajectory analysis in. 199 Meander system. 203 standard deviation in. 203 skew in. 2 0 3 ( 0 profit factors in. 143-166. 148 in Black Jack. 200-201. 131 Black Jack combined with.135. 200 measures of performance. 132-134. 143. 216 definition of. 64-69. 158(0. return on account. 157-159. 208-209.149 average percentage move for each bar in. diversified portfolios in. 149-150. 156-157. 159(0 trailing stop and. 67. 219-221(0 drawdowns and. 20 average percentage profit. 262 social factor influencing market trends and. 155(0 stop loss and.352 Index market analysis (cont. 159(0 trajectory analysis and. 343 in average losing trade. 157-158.179 standard deviation in. 145 winners vs. 199(0. 291. 146. 123. 141. 149.148 rules for. 146-147. 146-147.148 average profit calculation in. 153 lifespan and development of trade using. 2 0 1 ( 0 . 150 in Dynamic Breakout System (DBS). 139. 138-139. 140. 148. 287-290. 199 stop loss in. 158(0. 123 ETD/MFE in. 203 maximum favorable excursion (MFE) in. 152. 208. 145 stop loss and. 147 regression equation for. 256 market microstructure indicators. 154(0 in Black Jack. 262 wealth of markets and. 153-154.

value in. 274-275 runs test in. 282.Index 353 measures of performance (cont. 23 percent profitable trades. 303(0. Franco.313(0. 319. 277. 28 start trade drawdown (STD). 339 average profit multiplier (APM) in. 305-313. 38-41. 274. 23-25 end trade drawdown (ETD). 287. 275. 279. 257 momentum changes in market trends. 281-282 terminal wealth relative (TWR) in. 234-235 in Standard Deviation Breakout system (SDB).) bars per trade. 9-10 closed trade drawdown (CTD). 306 optimal fin. 287-290. 281 exit techniques and. 283. 278. 271. 17 cumulative profit. 286. 272 in long-term systems. 282 efficient market hypothesis (EMH) in. 316. 265-266. 10-11 maximum favorable excursion (MFE). 291. 23 equity tops. 5. 23 number (winning/losing) trades. 283-285. 275. 281-282 type I errors in. future contract data. 3-4 profit factor. 25 slippage. 274 directional slope strategy. 23 median. Robert.318. 23 maximum consecutive winners/losers.315 standard deviation in. 282-283 dependencies. 275. 1-2. 6. 17-19. 5. 275 risk. 304-306. 273. 21. average number of. 282-285 price fluctuations in. 282. 341 entry points and.312.316(0. 285. 20-21 maximum adverse excursion (MAE).19 return on account. 320(0 efficient frontier in. 21 flat time. 272. 3 1 4 ( 0 . 301 in Meander system. 6-7 run ups.39. 10-11 percentage calculation of performance.319(0 diversification. 284-285 money supply (Ml. Merton. 5. one-market systems. 6. 293. 47. 23 maximum intraday drawdown in. 271. 23 total net profit in. 281 curve fitting. 25 gross profit/loss. 284-285 type II error in. 257 Modigliani. 269. 280. 279-280. 2 9 8 ( 0 . 273. 279-281.311. 281-282 capital asset pricing model (CAPM) in. 265. 275. 311(0. 280. 9-10 open trade drawdown (OTD). 4-5. M2). 305-313 margin to equity ratio in. 280. 4 drawdown. 306-309(0.341 correlations in. 339 gain vs.293. 6. 291. risk in. 313-319. 4-6. 274. 283. 228 filter using. 319. 6 price vs. 282 rule-based trading systems. 340-342 in portfolios vs. 4-5 underwater equity curve. 303(0. 259 moving averages. 318. 8 largest winning and losing trades. 193(0 Merton. 272. 285-305. 272. 274 estimated TWR (ETWR) in. 272-273. 279. 282 geometric mean in. 274. 279-280. 17 standard deviation calculation. 306 in Standard Deviation Breakout (SDB). 22 diversity in. 282. 274 cumulative profit in. 282 drawdown. 283-284. 7 Miller. 279-280.266 momentum indicator. 281 holding period return (HPR) in.317. 276. 257 MetaStock. 21. 308. 23 commission. 295-302.314. 8-9 profit per trade. 23 total equity drawdown (TED). 286 The Equation for. 273 in Dynamic Breakout System (DBS). 271-272 practical applications for. 177-178 in tracking trends. 192-193. 277. 273. 272. 282 fixed fractional strategy for. 14-16 performance summary report. 295-302. 281 key points in. 51 money management. 274 in short-term systems. 298(0. 273. 285. 13-14. 86-97 multimarket portfolios. TradeStation. 271-320.40 . 272 profit/loss in. 276. 272. 7-8.

53 Bollinger bands in. 326-329 total equity contribution I. 227-228 pivot-points in. 51 MACD indicator in. "equally profitably" defined. 53 in building a system. 324-325. 330(0 correlations in. 233 standard deviation method. 203 percent profitable trades vs. 233 in filtering. 3 2 6 ( 0 .188 normalization of market movements. 153-154. 233 true range method. 52 fixed fractional investing and. 277. 30 . 30. 269 Patterns in the Dark. 51 momentum indicator in. 4 3 Peters. 51 technical analysis for. 231-232. 29 percentage of winning trades. 2 2 . 51 stochastic indicator in.354 Index net present value (NPV) in. 31. 28 average profit per trade over time in. 339 Charity I portfolio. 161. 70 in filtering. 52 catching market moves in. 31 perpetual contracts. 85-86. 273. 285. TradeStation. 229. 31 normal distribution curves.) rate-of-change indicator in. 322-324. 257 PowerEditor. 318. 255. 306. 19 profit/profit factor. Edgar. 52-53 "equally well" vs. 275. 319 in short-term systems. 51 optimal f in. 28 Omega Research (See TradeStation) on-balance volume (OBV) indicator. 134 picking tops and bottoms (cont. 340-342 in long-term systems. 236. profit target and. 279. 53 performance summary report. 279-285. 325-330 Portfolio Management Formulas.280. 51 RSI indicator in. 21 exit techniques in. 154(0 in Meander system. 53 percentage-based analysis in. 32-37 portfolio composition. 331-334. 286. 323(0. 332-336(r) outlyer trades. 8 MAE/MFE in. TradeStation. 325. 233 Black and Scholes pricing formula. 272 price vs. 23 optimal f. 41-44. 256. 64 number (winning/losing) trades. 218 commission vs. 324 Charity II portfolio. 8 6 ( 0 in Blackjack. 233 profit per trade. 21. 189 plus and minus DMI indicator. 6. 53 platykurtic distribution curves. 321-324. 263 open trade drawdown (OTD). 6 perpetual adjustment method of splicing. 322. 242 in Gold Digger II. 304-305 oscillator-type indicators (See also Bollingerbands). 190 overriding filter. 48. 10-11 profit factor defined in. 3 2 5 ( 0 Pearson correlation coefficient. 1 price fluctuations. 258 picking tops and bottoms. 8-9. 51 in testing a system. 259 New Economy and. 160. 17-19. 52-53 data selection for. 29 percentage-based analysis. 53 pivot-points in. 3 2 3 ( 0 . 258 Pearson correlation coefficient. 3 2 2 . 257 pricing formula average true range method. 233 Black-76 pricing formula. 328-330. 321-330.. 234 out-of-sample testing. 3 2 7 Charity III portfolio. 3 2 9 ( 0 . 8-9. 324 total equity contribution II. 324-325. indicators. 275 average percentage profit in. 10-11 percentage calculation of performance. 51-82 in Black Jack. 269. 326-329 percent profitable trades. 3-4. 218 gross profit/loss in. average profit calculation in. 17 cumulative profit in. 13-14.. 29 equity tops in. 5. 21(f). 53 indicators of. 107. 53 plus and minus DMI indicator in. 187-188. 51 point-based adjusted method of splicing. value. 28. 3 2 5 ( 0 covariances in. 199. 230. 14-16 percentage of losers. 51 portfolio theory. 52 piling. 4 2 . 9-10 number of losers. 261 nonadjustment method of splicing contracts in.

8 6 ( 0 . 233.. 85 volatility and. 2 4 0 ( 0 . 343-344 ADX indicator in. 234 rule building. data mining.315 Dynamic Breakout System(DBS) vs. 229. 238-239.313(0. 179 . 265-266. 119. 211. 21. 256. 287. 193(0 in Meander system. 257 sentiment indicators. 214. 230. 308. 273. 130. 119-121 trends in markets and. 234 skew. Myron.110. 2 9 8 ( 0 . outcomes in market trends. 63 return on account. 218 MAE/MFE. Nathan. 258. 337(0 Rothschild. 306-309(0. 255. 234 filtering in. 146-147. 304-305 oscillators in. 233. 189-190. 235 in Gold Digger II. 2 4 0 ( 0 closed trade drawdown (CTD) in. 283.) money management in. 218 Chebychef's theorem in. 51. 287-290.19 slippage vs. 336(/).) profit per trade in. 306 momentum changes in market trends and. 194 in filtering. 223 Black Jack. financial markets. 199 money management and. 188. 31. 37-38 real vs. 256 splicing contracts in. 295-302. 17 social factor influencing market trends.311(0. 257 Raschke. Linda B. 147-148. 236 optimal fin. 28-29. 231-232. 177-186. 2 1 3 ( 0 filtering and. 122 Scholes. 228-236 Gold Digger system in. 217. 6-7 RINA system. 255 RSI indicator. 148 in Meander system. Henry. 235 lookback periods for. 121 T-bill ratio filter in. 179. William. 155(0 maximum adverse excursion (MAE) and maximum favorable excursion (MFE). 116 rule-based trading. 178-179. 190-191 in Black Jack. 31 standard deviation. 239-240.293. 199. 191. 25 runs test. 298(0. 272.Index 355 profit/profit factor (cont. 194. 235 start trade drawdown (STD) in.314. 303(0. 279-280.148 regular moving averages. 265. 234 pricing formulas in. 291. 242 risk. 181(0 exit techniques for.312. 274 short-term investing systems (cont. 212. 234 skew in. 85-86. 237-239(0 kurtosis in. 282 robustness test. 334-336. 178. 70 in Black Jack. 111. George. 301 Meander system in. 70 rate-of-change indicator. 234-235 on-balance volume (OBV) indicator in. 231-232 total equity drawdown (TED) in. 85 Black Jack system in. 17 profit target. 216.108.. 274-275 run ups. altering inputs for. 210. 261 regression equation. 180 Pruden. 235 standard deviation in. 293. 107-113. 223 in Black Jack. 236-238. 209. 192-193.291. 263 random entries (See distribution curves) random number generator. 204-205 random walk hypothesis (RWH). 303(0.. 51 ratio-adjusted method of splicing. 247-252. 2 3 6 ( 0 . 154-155. 249-251(0 maximum adverse excursion (MAE) in. 17-19. 305-313. 121 DMI-plus/DMI-minus lines in. 192. 47-49. 213-214. 179. 257 short-term investing systems. 233 RSI indicators in. 281-282 Standard Deviation Breakout (SDB). 295-302. MAE/MFE. 181-182 favorable excursion (MAEfe) in. 121-122. 311. 283. 2 1 2 ( 0 . 266 TradeStation equations for. 180(0 filtering in. 255 speculation. rules of. 180-181 end trade drawdown (ETD) in. 235-236. 203 slippage. 257 Soros. 235 in Gold Digger II. 306 moving averages filter in. 285-305. 235 margin to equity ratio in. 215-216 in Standard Deviation Breakout system (SDB). 234 average profit per trade over time in. 262 Sharpe.

70. 119 equity highs in. John. 275. 90-94. 4 3 cumulative profit in. 152-153. 5. 130-131. 337(0 time-based stop. 29 dollar value of largest winner in. 1-2. 77-78. 48 picking tops and bottoms. 6. 263 terminal wealth relative (TWR). 161 data mining calculations and. 136. 23.169. 279. altering inputs for.169. 205. 55 technical analysis of markets. 324 total equity contribution II. 171-172 maximum adverse excursion (MAE) and. 205-206 TradeStation. 306-309(0. 181. 242. 305-313.311. 169-176(0 efficient trades in. 158(0. 283-284. 272-273. 331-334. 167 symmetric distribution.356 Index Standard Deviation Breakout (cont. 99-101. 52. 94(0 dollar value of average loser in. 160. 121-122. 336(0. 169-176(0.169. 47-50 arbitrage. 278. 75-80(0. 58-59 directional slope method in. 158(0. 343 total net profit in. 7 8 ( 0 . 30 dollar value of largest loser in. 168-177. 219 in Dynamic Breakout System(DBS). 206-208 Bollinger-bands indicator in. 159(0 in Black Jack. 29 dollar value of average winner in. 215.. 47-49 testing a system. 211-212 in Dynamic Breakout System (DBS). 214-215. 182-186. 48 market-specific vs. 47 out-of-sample testing in. 123. 155 in Meander system. 132 in Black Jack. 285. 183(0 money management for. 1 average percentage profit in. Black Jack. 160. 202 in Standard Deviation Breakout system (SDB). 28 average percentage winner in. 3 1 1 ( 0 . 262 Technical Analysis of Stocks and Commodities. 124. 334-336. 276. 29 dollar value of largest drawdown in.135. 28. 313(0. 51. 121-124. 58(0 trade length. 48. 59 out-of-sample testing. 47 volatility in. 343 stochastic indicator. 321-324. 129-130 Black Jack code in. 183-186(0 profit target for. 177 start trade drawdown (STD). 202(0. 169-176(0 in Gold Digger II. 323(0. 41-44. systems. 339 surface chart. 56-62. 155 maximum favorable excursion (MFE). 47. 312. 23. 7 bar count/calculation in. 25 drawdowns in. 210 Sweeney.) maximum trade length in. 157-159. 23. perpetual contracts vs. 98. 135-136. 29.315 moving averages and exits in. 188 system concepts. diversified portfolios in.314.and long-term methodologies defined for. 113(0 T-bill ratio filter. 280. 29 drawdown calculations in. 87-97. 209. 47 systems vs. 195 maximum adverse excursion (MAE). 143. 47. 339 T-bond. 47. 180 stop loss in. 47 money management strategies in. 178 strategies vs. 51 stop loss. 205-206. 178 testing of. 277. 57(0.100. 159(0 in Meander system. 2 0 1 ( 0 . Black Jack.286 testing a system. 209.122. 325-330 total equity drawdown (TED). 177-178 out-of-sample testing in. 93(0. 308.111. 70. 168-177. 331-334. 161 . 157-159. 141. 47 building a system. 76-77 total equity contribution I. 108. 69-81. 191. 200-201. 160-161 Dynamic Breakout System (DBS) code. 280. 51-82 short. 332-336(0 robustness test. 29 dollar value of average trade in. 48 trends.131. 4-5 tracking market moves.110. 219 time-based trailing stop. 145. 123. strategies. 322. 28 average trade value in. 332-336(0 performance summary of. 4 2 . 107-113. 231-232 T-bills.

263. 29 largest percentage winner in. 6 0 ( 0 . percentage values in. 83-84 triggers for exit techniques. 84 drift in. 8 6 ( 0 basic crossover signals in. 97-98. 86-87 Bollinger-band type indicators and. 87. 63. 157-159.108. 83. 261 normalization of market movements in. 228. 145 trend filter. 259.120 maximum intraday drawdown in. 191-197 largest percentage drawdown in. 67. 259 moving averages in. 145 trajectory analysis. 264 long-term systems and. 6 8 ( 0 . 87-97. 260-261.vs. 231-232.111. equations for. 266 money supply (Ml. 255-266 market microstructure indicators in. 168-177. 5-6 Meander system code. 259-260 flow-of-funds analysis in. long-term trading and. 228.. 263. 229 absolute vs. 28 long trades. 64 accumulation area in. 29 largest percentage loser in. 26 futures contract data and. 177-186 tracking market moves. 262-263. 284-285 type II error. 83-84 demographics and.169. 262 trends in markets (cont. M2) in. 132. 274. 85-113. 111. 69(0 number of losers in. 159(0 maximum favorable excursion (MFE). 223 short trades. 158(0. 227-228 interest and amortization in. 107-113.) creation-of-funds analysis in. 23 . 262 Crash of 1929 and. 28. 259-260 directional slope method in. 258-259 gross domestic product (GDP) in. 228 net present value (NPV) in. 160. 284-285 uncertainty measures. 232-233 flat time calculation in. 70 trends in markets (See also distribution curves. 94(0 distribution area in. 113(0.264. 11-12. 257 Standard Deviation Breakout system in. 236. 263 oscillator-type indicators in. 47. 6. 97-98 volatility in. 260 decision to trade vs. 97-107 exit techniques and.110. 157-159. 160 filtering in. 6 8 ( 0 . 343-344 social factors in. 169-176 maximum adverse excursion (MAE). 119-121 standard deviation in. 263 Meander system and.108. 262-263. 264 average profit per trade over time in. 129-130 percentage of losers in. MAE/MFE. 77 underwater equity curve. 229.) evaluating performance. 215 in Dynamic Breakout System (DBS). financial markets. 158(0. 204-205. 87-88 Dynamic Breakout Systems and. equations for. 67.110. 230. 5 7 ( 0 . 85 short-term systems and. 262 short. 214-215. 58-59 Trade Works Software. 265-266. 120.343-344 200-day moving average in. 29 trend-tracking calculations in. 266 credit policies and. 29 open trade and entry price in. 58-59 Commitment of Traders Report and. 55-56. 70 true range method. 229 fiscal policies and. 31-44 Gold Digger systems code. 261 indicator piling in. 86-97. 343-344 market analysis and. 119. 6 2 ( 0 . 85-86. 262 Commodity Futures Trading Commission (CFTC) and. 107 type I errors. 56-62. 107-113. 274. 59-62. 29 percentage of winning trades in. 5 8 ( 0 turtle traders in. 30 profit in. 265. 28 number of trades in. 69—81 filtering and. 90-94. 261. 264 dollar value of half average weekly true range (AWTRO/AWTR) in. 64-69. 113(0 calculations for. 161 random number generator in.266 global market structures and.Index 357 TradeStation (cont. 2 8 ( 0 profit factor in. 6 9 ( 0 momentum changes in. outcomes in. 259 New Economy and. 64-69. 223 trailing stop. 93(0. 63 sentiment indicators in. market analysis). 261 regular moving averages and. 26-28 exit techniques in. 107 real vs. 159(0 in Black Jack. 29 performance summary report in. filtering. 64 on-balance volume (OBV) indicator in. code for. 233 turtle traders. 70.

29 . 83 Werner. 29 dollar value of average winner in. 20-21 MAE/MFE and. 255 winning trades average dollar value per trade in. 99-101. 9 bars per trade in. 10-11 number (winning/losing) trades in. 20 winning trades (cont.151 maximum consecutive winners/losers in. 29 dollar value of largest winner in. 29 largest percentage winner in. 256 weather-sensitive markets. 28 largest winning and losing trades in. Richard.100 Dynamic Breakout System. 255 wealth of markets. bad. 269. good vs. triggers of. 241-254 future contract data and. 101-102 filtering and. 83-84 dollar value vs. Max. 233. percentage and. 10-11 percentage of winning trades in.) average percentage winner in. 9-10 percent profitable trades in. 9-10 dollar value of average trade in. 281. 234. 7-8 largest winning/losing trade in. average number of in.358 Index Vince. 53. 84 Dynamic Breakout System. 28 average winning and losing trades in. 35 von Liechtenstein. 47. 339-340 volatility. 149-150. Ralph.

trading consultant. Stridsman operated his own Web-based trading advisory service in Sweden and w a s chair of the Swedish Technical Analysts Federation. Before moving to Chicago in 1997. Previously. . and popular speaker at industry seminars a n d conferences. he wrote for Futures magazine.ABOUT THE AUTHOR Thomas Stridsman is a senior editor for Active Trader m a g a z i n e .

Other books in The Irwin Trader's Edge Series The Encyclopedia of Trading Strategies by Jeffrey Owen Katz and D o n n a L.E. M c C o r m i c k Technical Analysis for the Trading Professional by Constance B r o w n Agricultural Futures and Options by Richard D u n c a n The Options Edge by William Gallacher The Art of the Trade by R. M c M a s t e r .