Welcome to Scribd. Sign in or start your free trial to enjoy unlimited e-books, audiobooks & documents.Find out more
Download
Standard view
Full view
of .
Look up keyword
Like this
2Activity
0 of .
Results for:
No results containing your search query
P. 1
Non stationary model for statistical arbitrage

Non stationary model for statistical arbitrage

Ratings: (0)|Views: 218|Likes:
Published by Will Bertram

More info:

Published by: Will Bertram on Sep 23, 2010
Copyright:Attribution Non-commercial

Availability:

Read on Scribd mobile: iPhone, iPad and Android.
download as PDF, TXT or read online from Scribd
See more
See less

09/11/2011

pdf

text

original

 
A non-stationary model for statisticalarbitrage trading
W.K. Bertram
a
,
a
ITG Australia Limited, Level 41, 1 Macquarie Place, Sydney NSW 2000,Australia 
Abstract
In this paper we present a non-stationary model for statistical arbitrage trading.This model represents the security price as the sum of an arithmetic Brownianmotion and an Ornstein-Uhlenbeck process. A continuous time trading strategy isderived for the process and expressions for the expected value and the variance of the return are formulated. Analytic solutions to problem are obtained in the casewhere the non-stationary component has zero drift.
Key words:
Econophysics, Stochastic Processes, Analytic Solutions
PACS:
89.65.Gh, 05.45.Tp, 05.40.Jc
1 Introduction
Statistical based trading strategies have become widely used by the financialindustry over the past decade [1–8]. Such strategies seek to profit by exploitingthe statistical behaviour financial data. Dependencies and correlations arecaused by inefficiencies such as liquidity are one example of this behaviour.These strategies are commonly associated with high frequency trading. Sincethe returns on individual trades are usually small, the benefit of a systematicapproach lies in being able to perform hundreds or thousands of trades per day,gaining profitability from executing the trade many times. Such an approachincurs a high level of trading cost and can usually only be implemented byinvestment banks and specialist market-makers whose transaction costs arevirtually negligible. In the context of formulating optimal trading strategiesit is essential to model real world data with a stochastic process that reflects
Corresponding author.
Email address:
william.bertram@yahoo.com.au
(W.K. Bertram).
Preprint submitted to PhysicaA 27 June 201
 
the appropriate physical properties of the data. A significant part of designingand implementing a strategy involves the use of phenomenological models forprice dynamics to model the observed behaviour of the market.A common approach attempts to profit from so called mean-reversion tradingwhere traders aim to profit from temporary extremes in pricing, entering atrade when the security reaches an extreme value and exiting when it revertsto some equilibrium level. In this approach, participants form linear combi-nations of stocks, futures, options and other derivatives in order to produce asynthetic asset whose price follows a mean-reverting stochastic process. Theability to identify the optimal trading points is central to employing such astrategy successfully. There have been several studies of such trading strate-gies in recent years [1,3,5,8]. Many of these studies model the log-price of the traded securities as stationary processes, such as the Ornstein-Uhlenbeckprocess. This assumption allows the use of common time invariant statisticalmeasures to estimate trade entry and exit points. However, when the dataexhibits non-stationary effects the entry and exit points cannot be reliablydetermined using methods and models based on stationary time series.Non-stationary behaviour is a widely observed phenomenon in financial timeseries [9,10,14,11,15–21]. Given a stationary time series it is relatively straight-forward to determine whether a given observation represents an extreme value.For data that displays non-stationary behaviour, the task of identifying ex-treme events becomes more difficult since one must distinguish transient ex-treme moves from non-stationary shifts in the data. Methods such as coin-tegration [22] have been proposed in order to remove non-stationary effectsfrom time series data, so that analysis may be performed using common statis-tics based on assumptions of stationarity. In the context of formulating tradingstrategies, synthetic assets are constructed to remove non-stationary behaviourfrom price series. In cases such as
dual-listed 
securities, where the same stockis traded on multiple exchanges, or
index arbitrage
, where a basket of stocksis traded against an index futures contract, the removal of non-stationary be-haviour is possible due to the existence of a deterministic relationship betweenthe assets being traded. However, in more general cases such as the popularmethod of 
pairs trading 
, when two similar securities are traded against eachother, it is not possible to guarantee the complete removal of all non-stationarybehaviour. If a synthetic asset is constructed where a deterministic relation-ship does not exist between the component assets, the resulting asset maystill retain residual non-stationary behaviour. The magnitude of the remain-ing non-stationary behaviour may have a significant effect on the return andrisk of the strategy.In this paper we present a method for the construction of optimal tradingstrategies on non-stationary securities. We consider the price of the securityto be a combination of stationary and non-stationary components. We propose2
 
a model for asset dynamics as an Ornstein-Uhlenbeck process with stochas-tic mean. The model exhibits mean-reverting behaviour around an arithmeticBrownian motion. The optimal trading strategy is calculated for the chosenstochastic process under the framework described in [7,8]. Under this model,the expected trade length is independent of the non-stationary behaviour. Ex-pressions for the expected profit rate and its variance are presented. In thiscase the expected profit rate is shown to be a function of the drift of the non-stationary component, while the variance of the profit rate is shown to be afunction of the instantaneous variance of the non-stationary component. Wederive analytic solutions for the expected return and the variance of the re-turn for the strategy in the case where the non-stationary component has zerodrift. The expected return is shown to be independent of the non-stationarybehaviour while the variance is shown to be a linear function of the instanta-neous variance of the non-stationary component. We derive analytic solutionsfor the strategy that maximises the expected return. We derive a closed formsolution for the variance associated with this optimal case. The results forthe mean and variance of the strategy are used to derive an expression forthe signal-to-noise ratio of the strategy. This ratio is used to determine anoptimal strategy in terms of the strength of the non-stationary component.The model is illustrated with a real-world example where the stationary andnon-stationary components are directly observable. The results for the optimalstrategies are calculated for this example.The rest of the paper is as follows. In the following section we present themodel and define the continuous trading strategy for the process. In Section3 we derive analytic solutions for the expected return and the variance of the return. An analytic solution for the problem of maximising the expectedreturn is also presented. Section 4 presents the results of applying the modelto real world data. Section 5 summarises the main results and concludes.
2 Non-stationary model for trading
In order to address the potential existence of non-stationary behaviour weassume that a synthetic asset consists of two components: a stationary processthat provides transient mean-reversion effects over short time scales; and non-stationary process that drives the long term behaviour of the asset. We expressthe price of the synthetic security
p
t
as,
 p
t
=
e
t
,
3

You're Reading a Free Preview

Download
scribd
/*********** DO NOT ALTER ANYTHING BELOW THIS LINE ! ************/ var s_code=s.t();if(s_code)document.write(s_code)//-->