Professional Documents
Culture Documents
TRADING IN TRADINGAND
PORTFOLIO
MANAGEMENT
DAVID J. LEINWEBER
It is illegal to make unauthorized copies of this article, forward to an unauthorized user or to post electronically without Publisher permission.
40 USING
INFORMATION FROM TRADING
I N TRADING
AND PORTFOLIO
MANAGEMENT SIJMMER
1995
concerned with transaction costs. It is now generally tionate share of the costs. The trades expected to
understood that these costs have the potential to be low-cost “no-brainers” are not.
erode or eliminate the value added by money Larger trades, generally handled by higher-com-
managers. Trading is the implementation of invest- mission brokers, have lower than expected costs.
ment ideas, and the quality of the implementation is The expected relationships between manage-
as important as the cleverness of the idea. ment styles and costs are not found.
(until recently) because of the scarcity of data. In the Some trades produce transaction profits, which
absence of a well-researched body of empirical and partially offset costs.
The Journal of Investing 1995.4.2:40-50. Downloaded from www.iijournals.com by George Mason University on 08/15/16.
Costs can be measured after the fact, and used Information Technology and Markets
to plan future trading. They can also be measured as
a trade progresses and used to determine how it will The 1975 legislative mandate to develop a
be completed. This dynamic measurement of costs U.S. national market system encouraged ‘‘~~aximum
and feedback into ;I trading program is a valuable reliance on computer and communications technolo-
cost control technique available only to traders using gies.’’ This has happened with a vengeance, both in
electronic systems capable of keeping up with the the U.S. and around the world. The NYSE’s DOT
process in real time. Measurement is the first step to system handles two-thirds of the orders on the
control on multiple time scales. world’s largest stock exchange. Nasdaq used automa-
tion to transform itself from a sleepy market for small
Heat, Light, and Transaction Costs
companies to the second largest U.S. equity market.
This article began with a discussion of trading Instinet, the Crossing Network, POSIT, Lattice Trad-
It is illegal to make unauthorized copies of this article, forward to an unauthorized user or to post electronically without Publisher permission.
costs illustrated by comparing real and paper portfo- ing, Arizona Stock Exchange, CompBid, and others
have introduced entirely new types of purely elec-
The Journal of Investing 1995.4.2:40-50. Downloaded from www.iijournals.com by George Mason University on 08/15/16.
42 U S I N G INFORMATION F R O M TRADING
I N TRADING
AND P O R T F O L I O MANAGEMENT SUMMER 1995
EXHIBIT 2 very different ways from how we use today’s cars. It
COMPUTING POWER PER DOLLAR would totally change our lives, our economy, and
our investments.
This is precisely the magnitude of the chal-
lenge facing the securities industry regarding
computers: I f you had everything computationally, where
would you put it jnann‘ally? One place to put it is in
the improvement of the trading process, and a
reasonable question is: “How?” This is the motiva-
tion for our attempt to understand the nature of the
transaction process better and use that factor of 400
million to reduce the cost of transacting. Here are
It is illegal to make unauthorized copies of this article, forward to an unauthorized user or to post electronically without Publisher permission.
A $2 Billion Experiment
No technology in history has progressed as
rapidly as electronic computing. Let’s look at this in We examine 13,651 equity purchase transac-
more detail. Two reasonable ways to measure the tions, totaling nearly $2 billion, made by one of the
capacity of a computer are by its memory size (in largest U.S. corporate pension plans in 1991. The
megabytes) and its speed (in millions of instructions data are provided by the Plexus Group, a firm that
per second [MIPS]).2 A reasonable way to measure analyzes transaction costs, with the cooperation of
its cost is by its price (in current dollars). Thus, a the fund manager. Trade sizes range from 100 shares
good figure of merit for computers is (Memory x to blocks of more than 400,000. Both active and
Speed)/Price. passive management styles are represented. All orders
A 1960s-era PDP-1 computer filled a large in this sample are filled, some on the first day after
room. It operated at about 0.1 MIPS, had 0.008 the decision to trade, some up to twenty-one trading
megabytes of memory (which the salesman would days later. Because all orders are eventually filled, no
have called 8K), and cost $250,000. Our cost perfor- opportunity costs are observed here. This simplifies
mance measure turns out to be 0.0000000032. A the measurement and interpretation of costs, as there
modern high-end workstation (the kind you find by is no need to define opportunity costs.
the truckload on Wall Street) runs at 100 MIPS, has We set out to test the validity of the conven-
256 megabytes of memory, fits in a desk drawer, and tional wisdom in several areas: specifically, the rela-
costs about $20,000. Our figure of merit is now tionships between transaction costs and 1) trade size
1.28. This is an improvement from the early 1960s relative to market capitalization, 2) trade size relative
by a factor of 400 million (which readers of this to average trading volume, 3) management style, 4)
Journal might call 4,000,000,000,000 basis points). patience in trading, and 5) use of crossing networks.
This growth is shown in a fairly standard way by Some findings are surprising, although they
Exhibit 2. must be taken with the appropriate caveat: These
To appreciate what this really means, consider transactions all come from a single pension fund, and
what would happen if the same rate of progress had generalization to all funds may be premature.
occurred in the automobile industry during the same However, as these results have become more widely
p e r i ~ dLet’s
. ~ define a similar figure of merit for cars known, several other individuals, from the United
as (Mileage x Cruising Speed)/Cost. Two good States and Japan, have told me they have observed
things multiplied together and divided by a bad thing. very similar effects in their independent analyses of
We start with a typical 1960s Chevrolet. It trading data. They have sent in charts of thousands
gets 10 mpg, cruises at 75 mpg, and costs $2,000. of their trades that look remarkably like the charts in
There are an infinite number of ways to scale up the this article, so we believe it is safe to say these are not
car’s figure of merit by a factor of 400 million to completely unique observations.
match the computer’s progress. Here is one: the Costs and Trade Size
scaled-up car gets 2,000 miles per gallon, cruises at
7,500 mph, and costs twelve cents. This is not really What should we expect regarding costs and
what we would consider a car at all; it is a virtually block size? The most reasonable sounding model is
costless, disposable land rocket. We would use this in first described in Loeb [1983] and extended in Loeb
SUMMER
1995 THEJOURNAL OF INVESTING 43
EXHIBIT 3 tions. Our expectation is that a scatter plot of cost
- COSTS VERSUS BLOCKSIZE
EXPECTATIONS versus size as a percentage of shares outstanding for
the 13,651 transactions in our sample would lie
along these lines.
s
8 from our expectations. The largest trades have costs
c) 7- much lower than those predicted by the model. The
highest and lowest percentage costs are associated
z 5 with the smaller trades. Many trades, of all sizes, have
kE 4
negative costs; that is, they produce transaction prof-
.................. ................ its. Overall, costs rise slightly as trade size increases,
1E
.: _ _ . .____. .___. .___. _. _.-.---
--
but they rise less than expected. What is responsible
It is illegal to make unauthorized copies of this article, forward to an unauthorized user or to post electronically without Publisher permission.
: e __
_ . _.-..-..-..
_._ 1 for this? It may be that these are largely actual trades,
~
The Journal of Investing 1995.4.2:40-50. Downloaded from www.iijournals.com by George Mason University on 08/15/16.
.01 .03 .05 .W .09 .I1 .13 .15 .17 .19 .21 .23 .25 .27 2 9 not suppositions. Apparently, when real orders are
Block Size (Percentof Shares Outstanding) filled using real money, block traders can deliver
better performance than the Loeb model predicts.
Market Capitalization One objection to the use of percentage of
$10 Million shares outstanding as a standardized measure of order
$100 Million size is that it is a poor indicator of the actual difficul-
........ $1Billion
ty of a trade. Many shares are closely held .and do not
$10 Billion
trade very often. Looking at trade size as a percent-
$100 Billion
age of average daily volume may be more appropri-
ate. A scatter plot of the same transactions with the
[1991]. This very plausible model is based on queries size as a percentage of the three-day average volume
to brokers and market makers for blocks valued (prior to the decision date) is shown in Exhibit 5 .
between $5,000 and $20 mdion at various levels of This appears to be a better measure of size, but it still
market capitalization. The 1991 paper generalized fails to explain the surprising bulge in both trading
the earlier result by fitting an equation to the data. costs and profits for the smaller transactions. For
The Loeb function predicts transaction cost these smaller trades, the costs substantially offset the
(in percentage of value) as a function of trade size profits, suggesting that the cumulative price of "no-
expressed as a percentage of shares outstanding and brainer" executions is much higher than realized.
market capitalization. Exhibit 3 shows these predic- This is illustrated when we look a t net
EXHIBIT 4 EXHIBIT 5
- COSTSVERSUS BLOCKSIZE
OBSERVATIONS - COSTSVERSUSBLOCKSIZE
OBSERVATIONS
-1.5 20 .
e
--E*; - .*
-
l*
P
.
eo
. .':.*
. - e
.. 10
-*
. e .
. ' 3
" 0 -. .
'E.. -.. - H
8
a
-10
Y
60 -
t 50 - Value Value Low Low Low
Growth Value Low Low Low
Earnings
Surprise Information High High High
It is illegal to make unauthorized copies of this article, forward to an unauthorized user or to post electronically without Publisher permission.
Index Fund,
U
G25 25-50 50-75 75-100 100+ Combined
Large-Capa Passive Variable Variable High
The Journal of Investing 1995.4.2:40-50. Downloaded from www.iijournals.com by George Mason University on 08/15/16.
percentage trading cost by order size, as shown in aThe costs associated with some investment strategies that use
Exhibit 6. What does this mean in terms of actual futures can be low, despite high opportunity costs.
dollar cost to the pension fund? The answer is seen
in Exhibit 7, which illustrates the total transaction
costs by trade size. The small orders represent the tracking the S&P 500, are expected to depend on the
largest single contribution to total trading costs. funds’ use of index futures to manage costs.
Costs and Management Style These expectations are summarized in Exhib-
it 8. Observations of the actual transactions for
What do we expect to observe when we look different management styles are shown in Exhibit 9.
at transaction costs separated by management style? Management styles expected to exhibit the highest
An excellent discussion of this subject is found in costs have the lowest costs, and vice versa. This
Collins and Fabozzi [1991]. Patient disciplines, such surprising result suggests that something may be
as value and growth investing, have longer time hori- peculiar to this particular fund (as suggested by the
zons and are expected to have lower transaction costs. earlier caveats), or that our understanding of the rela-
Earnings strategies depend on quicker execution to tionship between management style and costs is in
capture the market’s reaction to difierences between need of refinement, or that fund managers may not
expected and actual earnings. Index funds tracking be rigorously adhering to their target style.
small-capitalization stocks would be expected to have Costs and Patience
larger costs because of the characteristics of the small-
er stocks comprising those indexes. Execution costs Our expectation is that patience will be
for high-capitalization index funds, such as those rewarded. Patient trading allows short-term market
EXHIBIT 9
EXHIBIT 7
TRADINGLoss BY ORDERSIZE($MIL.)
- COSTSVERSUS
OBSERVATIONS
MANAGEMENTSTYLE
.ti
k.60
-
- Expectation Low +
0 2.0 - .Y
1.5 -
Expectation High
3
’ 1.0
0.5
n
U
-
-
0-25 25-50 50-75 75-100 100+
Order Size (1,000 shares) Combined S&P500 XI Earnings Value
SUMMER
1995 THEJOURNAL OF INVESTING 45
EXHIBIT 10 shortfall rose dramatically, clear evid.ence that
-- COSTSVERSUS TIME
EXPECTATIONS patient trading can pay.
The lower chart shows that, after fifty days,
the traders were losing patience and became more
-- aggressive on size, increasing the percentage of
- remaining shares bid for from roughly 35% to 100%
lo
of the total remaining, and roughly dolibling this
“aggressiveness” on the price of their limit order^.^
8 -
u5
At Day 63, they really wanted to get this over with,
6
- and again increased their aggressiveness on price.
3 This set of transactions requires an extraordi-
It is illegal to make unauthorized copies of this article, forward to an unauthorized user or to post electronically without Publisher permission.
cost curve.
These expectations are reinforced by the
results of a real exercise in patient trading reported
in Bodurtha and Quinn [1990], as shown in Exhibit
11. This is an actual series of trades to construct a
350-stock portfolio to track the Russell 2000 Index.
The returns on the real and paper portfolios are
shown in the upper chart, along with the imple-
mentation shortfall (the difference between these “48 50 52 54 56 58 60 62 64 66
two curves). This shortfall is almost zero for the first Trade Day
fifty days of patient trading. During the period
Percent of RemainingShares Bid For
when the trading was done patiently, the real port-
0 Aggressiveness of LimiVBid Price
folio tracks the paper, with very low transaction
costs. When patience ran thin, the implementation Source: Bodurtha and Quinn [1990].
46 USING
INFORMATION FROM TRADING
I N TRADING
AND PORTFOLIO
MANAGEMENT !jUMMER 1995
EXHIBIT 12 trading/portfolio management support tools made
MEAN COST VERSUS possible by that factor of 400 million in technologi-
TRANSACTION
DECISION INTERVAL cal growth. Systems such as Quantex from the
Investment Technology Group (ITG) and First
Boston’s Lattice Trading can, and do, manage the
90
e task of controlling this type of electronic order
70 - e
working to achieve best execution in a broader sense.
e e
e With regard to our findings about patient trad-
e . 0
/ ing, our expectation of observing a downward sloping
e e curve resembling the execution cost curve in Exhibit
10
- e 10 has yet to be met. What we do observe is shown in
0
It is illegal to make unauthorized copies of this article, forward to an unauthorized user or to post electronically without Publisher permission.
EXHIBIT 13
- CROSSING
EXPECTATIONS AND
TRANSACTIONCOSTS
Note: Factors lowering costs for crossed trades include midquote or closing price, no concession, and low commissions.
SUMMER
1995 THEJOURNAL OF INVESTING 49
substitution order working and a variety of elec- ’Or, “meaningless index of processor speed.”
tronic order working strategies. 3This analogy was popularized by Gordon Moore, a
Instinet’s Order Working System provides access founder of Intel Corp. My apologies to readers who have heard me
or Moore or others tell this before. It is still new for enough of us to
to both the continuous Instinet market and the warrant its inclusion.
crossing networks, which cross orders after 4Aggressiveness on price is measured by the position of
market hours at closing prices. Order Working the limit order price within the quote prevailing at the time the order
is submitted. Aggressiveness on size is measured by the percentage of
System orders can float at the midquote (within shares remaining to buy, represented by the limit order size.
specified limits) for execution on Instinet, with SThis was first suggested (and implemented) by Evan
residuals being sent to the cross. Schulman at Batterymarch in the O OS, and is reflected in the First
BostodLattice Trading System that his firm now oper:ites.
The Arizona Stock Exchange (formerly the
6Efficient implementation is a key element of a strategy
Wunsch Auction System) is an important exam- that has returns exceeding the S&P 500 over a five-year period.
ple of the changes technology can bring to secu- 7Many thanks to John Dorian for his expert advice on the
It is illegal to make unauthorized copies of this article, forward to an unauthorized user or to post electronically without Publisher permission.
rities markets. It operates as a call (periodic) use of GPBs and eficient trading techniques in equity portfolio
management.
The Journal of Investing 1995.4.2:40-50. Downloaded from www.iijournals.com by George Mason University on 08/15/16.
Loeb, Thomas. “Is There a Gift from Small Stock Invcsting!” Finan-
SUMMARY cial AnalysfiJournal, January/February 1991, pp. 39-44.
ENDNOTES Treynor, Jack L. “What Does It Take to Win the Trading Game?”
Financial Analystsjournal,January/February 1981, pp. 5!5-60.
‘In fairness, some of this shortfall can be attributed to the
brief delay ValueLine imposes on its fund to avoid trading ahead of Wagner, Wayne. “Transactions Costs: Measurement and Control.”
its subscribers. Financial AnalystsJournal, September/October 1990.
50 USINGINFORMATIONFROM TRADING
I N TRADING
AND PORTFOLIO
MANAGEMENT