Market Microstructure
Market Microstructure
Ananth Madhavan
Knowledge of market microstructurehow investors latent or hidden
demands are ultimately translated into prices and volumeshas grown
explosively in recent years. This literature is of special interest to
practitioners because of the rapid transformation of the market
environment by technology, regulation, and globalization. Yet, for the most
part, the major theoretical insights and empirical results from academic
research have not been readily accessible to practitioners. I discuss the
practical implications of the literature, with a focus on price formation,
market structure, transparency, and applications to other areas of finance.
Market Microstructure
Bid Price
Midquote
0
Deviation from Desired Inventory
29
noise traders, dealers would not be willing to provide liquidity and markets would fail. (3) Given the
practical impossibility of identifying informed traders (who are not necessarily insiders), prices adjust
in the direction of money flow.
Empirical evidence on the extent to which
information traders affect the price process is complicated by the difficulties of identifying the effects
explicitly because of asymmetrical information.
Both inventory and information models predict that order flow will affect pricesbut for different reasons. In the traditional inventory model,
order flow affects dealers positions and dealers
adjust prices accordingly. In the information model,
order flow acts as a signal about future value and
causes a revision in beliefs. Stoll (1989) proposed a
method to distinguish the two effects by using
transaction data. But without inventory data, the
results of such indirect approaches are difficult to
verify. Madhavan and Smidt developed a dynamicprogramming model that incorporates inventorycontrol and asymmetric-information effects. The
market maker acts as a dealer and as an active
investor. As a dealer, the market maker quotes
prices that induce mean reversion toward inventory targets; as an active investor, the market maker
periodically adjusts the target levels toward which
inventories revert. The authors estimated the model
with daily specialist inventory data and found evidence of both inventory and information effects.
Inventory and information effects also explain
why excess volatility might be observed, in the
sense that market prices appear to move more often
than is warranted by fundamental news about
interest rates, dividends, and so on. An interesting
example is provided in Exhibit 3.
2002, AIMR
Market Microstructure
Exhibit 3. Does Trading Create Volatility?
French and Roll (1986) found that, on an hourly basis, the
variance during trading periods is 20 times larger than the
variance during nontrading periods. One explanation is that
public information arrives more frequently during business
hours, when the exchanges are open. An alternative explanation is that order flow is required to move prices to equilibrium levels. To distinguish between these explanations is
difficult, but a historical quirk in the form of weekday
exchange holidays that the NYSE declared at one point (to
catch up on a backlog of paperwork) provides a solution.
Because other markets and businesses were open, the public
information hypothesis predicted that the variance over this
two-day period, beginning with the close the day before the
exchange holiday, would be roughly double the variance of
returns on a normal trading day. In fact, however, the variance for the period of the weekday exchange holiday and the
next trading day was only 14 percent higher than the normal
one-day return. This evidence suggests that trading itself is
an important source of volatility; for markets to be efficient,
someone has to make them efficient.
Trade Time
Permanent
Temporary
Time
September/October 2002
the annual reconstitution of the Frank Russell Company indexes from 1996 through 2002. Decomposing returns into permanent and temporary effects
provides insights into such return anomalies. Specifically, permanent changes in prices are attributable to shifts in liquidity associated with changes in
index membership, whereas temporary effects are
related to transitory price pressure. The magnitude
of return reversals (temporary effects) following
index rebalancing suggests that liquidity pressures
help explain the return anomalies associated with
the annual reconstitution of the stock indexes of the
Frank Russell Company.
Pretrade-Cost Estimation. Intraday models
are essential for formulating accurate predictions
of trading costs. Traders use pretrade-cost models
to evaluate alternative trading strategies and form
benchmarks for evaluating the post-trade performance of traders and brokers. Such models can be
used as modules in autotrading strategies in which
computers automatically generate trades under
certain conditions.
Interest in pretrade-cost models is also motivated by the fact that transaction costs can significantly erode investment performance. Specifically,
the net alpha to an investment strategy is the
expected alpha less the product of turnover and
two-way trading costs. Because alpha is linear in
trade size but costs are typically nonlinear, many
investment managers use portfolio optimizers that
incorporate pretrade-cost models to avoid constructing portfolios that consist of large positions in
illiquid small-cap securities.
A successful model has three essential ingredients: (1) Because most investors break their orders
32
Market Microstructure
that all parcels of the order be on the same side as
the order itself.
However, although most traders agree that
linearity is too simplistic an assumption, they disagree over what form these functions really take.
Are they concave (i.e., rising at a decreasing rate in
size), convex (rising at an increasing rate), or linear?
Figure 3 shows three possible shapes. Loebs 1983
study of block quotations suggests a concave shape
to the functions. Similarly, Hasbrouck (1991) found
that square-root transformations of volume fit well
in modeling price impacts. But others argue for
convex functions because available liquidity is ultimately limited. Madhavan and Cheng (1997) suggested that the resolution of this problem might be
found in the fact that observed prices and volumes
reflect the operation of two economic mechanisms.
On the one hand, Keim and Madhavan argued that
the price functions are concave for upstairs trades,
where buyers and sellers are matched. They
showed that as block size increases, more counterparties are contacted by the upstairs broker, which
cushions the price impact, so the costs of trading
are relatively low for large trades. These results are
consistent with Loebs findings from interviews
with block traders; he reported a concave priceimpact function. Similar logic applies to trades in
external crossing systems. Because of commission
costs, upstairs trades are relatively uneconomical
for small trades.
against the limit-order book or the specialist, producing some price impact. A very large trade will
eat up all the liquidity on the book, and the specialist may demand a large price concession to accommodate the remainder of the trade from inventory,
as is sometimes observed at the close if a large order
imbalance exists. The result is a convex price function.
In general, traders who have the choice will
select the lowest-cost mechanism (Madhavan and
Cheng), so the proportion of upstairs trades will
rise with size. Unfortunately, publicly available
databases (the Trade and Quote database, for
example) do not distinguish between upstairs and
downstairs trades, so the true cost of large trades
directed to the downstairs market is underestimated. The result is, as Figure 4 illustrates, that a
concave price function appears to best fit the data.
The dashed line shows the price impact in the
downstairs (convex) market, and the dotted line
shows the price impact in the upstairs (concave)
market; the observed relationship is the lower
envelope of the two curves, the solid line. In addition, a concave relationship might emerge if an
order that moved prices substantially induced
other traders to enter the market as counterparties
to supply liquidity.
Downstairs
Price Impact
Concave
Upstairs
Linear
Convex
Trade Size
Trade Size
Table 1.
Architecture
Element
Continuous trading
Typical
ECN
NYSE
Open
Market
Dealer presence
34
Price discovery
Automation
Anonymity
Pretrade quotes
Post-trade reports
NYSE
Intraday
Trading
Paris
Bourse
POSIT
Chicago
Board of
Trade
Foreign
Exchange
Market
2002, AIMR
Market Microstructure
only serves to confirm this prediction. With asymmetrical information, rational, informed traders
will split their orders between the two markets,
providing incentives for liquidity traders to consolidate their trading. Intuitively, if two markets are
combined into one, the fraction of informed trading
volume will drop, resulting in a narrowing of
spreads. Even if the information signals are symmetrical, if they are diverse, then pooling orders will
provide prices that are informationally more efficient than decentralized trading among fragmented
markets. Indeed, even when multiple markets coexist, the primary market is often the source of all price
discovery (as shown by Hasbrouck 1995), with the
satellite markets merely matching quotes.
Despite strong arguments for consolidation
occurring, however, many markets are fragmented
and remain so for long periods of time. This puzzle
has two aspects: (1) the failure of a single market to
consolidate trading in time and (2) the failure of
diverse markets to consolidate in space (or cyberspace) by sharing information on prices, quotes, and
order flows. As for the first issue, theory suggests
that multilateral trading systems (such as singleprice call auctions) are efficient mechanisms to
aggregate diverse information. Consequently,
researchers are interested in how call auctions operate and whether such systems can be used more
widely to trade securities. The information aggregation argument suggests that call auctions are especially valuable when uncertainty over fundamentals
is large and market failure is a possibility. Indeed,
many continuous markets use single-price auction
mechanisms when uncertainty is large, such as at the
open, close, or reopening after a trading halt. Yet,
most trading systems are continuous and bilateral
(not periodic and multilateral), which suggests that
the benefits of being able to trade immediately at
known prices are extremely important.
With regard to the second issue, although consolidated markets pool information, whether they
will be more efficient than fragmented markets is
not clear; for example, efficiency will diminish if
some traders develop reputations based on their
trading histories. Indeed, one argument cited for
the growth of electronic crossing markets is that
traditional markets offer too much information
about a traders identity and motivations for trade.
Models emphasizing asymmetrical information
provide some rationale for the success of offmarket competitors in attracting order flow from
primary markets. Technology may eventually
resolve the fragmentation debate. Today, a variety
of systems are being built to route orders intelligently to the most liquid market centers.
September/October 2002
system of the type used by the NYSE, but the opposite seems to be the case, even after such factors as
company age, company size, risk, and price level
have been controlled for. One explanation, provided by Christie and Schultz (1994), is that dealers
on Nasdaq implicitly collude to set spreads wider
than those justified by competition. Institutional
practices such as order-flow preferencing (i.e.,
directing order flow to preferred brokers) and soft
dollar payments limit the ability and willingness of
dealers to compete with one another on the basis of
price and may explain why spreads are large
despite easy entry into market making.
Tests of market structure theories face a serious
problemthe absence of high-quality data that
would allow researchers to pose what if questions. However, some interesting natural experiments have been reported. For example, in the late
1990s, the Tel-Aviv Stock Exchange moved some
stocks from periodic trading to continuous trading,
which allowed researchers to investigate the effects
of market structure on asset values by using the
stocks that did not move as a control. Amihud,
Mendelson, and Lauterbach (1997) documented
large increases in asset values for the stocks that
moved to continuous trading. But such natural
instances for testing are few and far between.
Compounding the problem is the fact that
traders adjust their strategies in response to market
protocols and information, so assessing the impact
of market protocols is difficult. Furthermore,
empirical studies are limited by the scarcity of large
samples of events to study. An additional obstacle
to empirical research is that the changes in structure that markets make are often responses to perceived problems or competitor actions. Such
changes are often accompanied by design alterations in other dimensions, such as a switch to
automation or greater transparency.
A promising alternative to market-initiated
changes is laboratory or experimental studies that
allow tests of subtle theoretical predictions about
market design. For example, in a laboratory study
conducted by Bloomfield and OHara (2000),
human subjects trading in artificial markets
allowed the researchers to examine the effects of
various changes in protocols on measures of market quality. A key advantage of such laboratory
research is that researchers can accurately measure
quality metrics (e.g., deviation of price from intrinsic value, speed of convergence to full-information
prices) that cannot ordinarily be observed with real
data. Further benefits of laboratory experiments in
finance are discussed in Exhibit 6.
Practical issues of market design are central to
the subject of market microstructure. Although
2002, AIMR
Market Microstructure
Exhibit 6. Experimental Finance
The ability to frame controlled experiments in laboratory
markets allows researchers to analyze complex information
effects. The obvious focus is on metrics, such as the bidask
spread, market depth or liquidity, and volatility. But an
experimental study can also address variables that might not
otherwise be possible to observe, including data on traders
estimates of value over time, their beliefs regarding the dispersion of true prices, and the trading profits of various
classes of traders (e.g., informed versus uninformed, speculators versus hedgers).
Bloomfield and OHara used experimental markets to
analyze changes in disclosure rules. In their study, lab participants faced various disclosure regimes, and in some experiments, dealers (markets) could decide whether they
preferred transparency or not. Bloomfield and OHara found
that transparency has a large impact on market outcomes.
Several other interesting findings have emerged from lab
markets. For example, generating price bubbles is quite easy,
even if market participants are aware of bounds on fundamental value. Interestingly, prices in auction markets need
not always converge to full-information values; agents may
learn incorrectly and settle prices at the wrong value.
Information
Many of the informational issues about market
microstructure concern transparency and disclosure. Market transparency has been defined (see,
for example, OHara) as the ability of market participants to observe information about the trading
process. Information in this context can be knowledge about prices, quotes, volumes, the sources of
order flow, or the identities of market participants.
A useful way to think about transparency,
which has many aspects, is to divide it into pretrade
and post-trade dimensions. Pretrade transparency
refers to the wide dissemination of current bid and
September/October 2002
ask quotations, depths (and possibly also information about limit orders away from the best prices),
and other pertinent trade-related information, such
as the existence of large order imbalances. Posttrade transparency refers to the public and timely
disclosure of information on past trades, including
execution time, volume, price, and possibly information about buyer and seller identifications.
Issues of Market Transparency. Both preand post-trade transparency issues have been central to some recent policy debates. For example, the
delayed reporting of large trades in London has
been cited as a factor in intermarket competition
and order-flow migration.
In addition, transparency is a major factor in
debates about floor systems versus electronic systems. Floor systems, such as the Chicago futures
markets, generally do not display customer limit
orders unless they represent the best quote. In contrast, electronic limit-order-book systems, such as
the TSEs Computer Assisted Trading System
(CATS) and the Paris Bourses Cotation Assiste en
Continu (CAC) system typically disseminate not
only the current quotes but also information on
limit orders away from the best quotes. The trend
worldwide has been toward greater transparency.
(The NYSE recently created OpenBook, a real-time
view of the specialists limit-order book.)
The practical importance of market transparency has given rise to a great deal of theoretical and
empirical research. Several authors have examined
the effect of disclosing information about the identity of traders or their motives for trading. These
issues arise in many different contexts, including
post-trade transparency and reporting,
predisclosure of intentions to trade (known as
sunshine trading) or the revelation of order
imbalances at the open or during a trading halt,
dual-capacity trading, in which brokers can
also act as dealers,
front running, when brokers trade ahead of
customer orders,
upstairs and off-exchange trading,
hidden limit orders in automated trading systems,
counterparty trade disclosure, and
the choice of floor-based or automated trading
systems.
In a totally automated trading system, where
the components of order flow cannot be distinguished, transparency is not an issue. Most floorbased trading systems, however, offer some degree
of transparency regarding the composition of order
flow. For example, for trading on the NYSE, the
identity of the broker submitting an order may
37
Market Microstructure
where
p = an N 1 vector of price changes
X
= an N k matrix of order flows, current
and lagged, and other predetermined
variables affecting price movements
= a k 1 vector of coefficients
U = an N 1 vector of error terms
Returns to stock i may depend on current and
lagged flows in stock j. Commonality in order flows
is manifested in that, although X has full rank, only
a few sources of independent variation explain
most of the variation in the data.
Hasbrouck and Seppi (2001) used principal
components analysis to characterize the extent to
which common factors are present in returns and
order flows. Principal components analysis can be
viewed as a regression that tries to find a linear
combination of the columns of the data matrix X
that best describes the data, subject to normalization restrictions imposed to remove indeterminacy.
Hasbrouck and Seppi found that common factors
are present in both returns and order flows and that
common factors in order flows account for 50 percent of the commonality in returns. Whether such
factors can help predict short-run returns, variation
in intraday risk premiums, or the observed relationship between price variability and volume is
still an open question.
Another interesting application of market
microstructure in the asset-pricing area concerns
technical analysis, in which past price movements
39
Market Microstructure
structure and availability of information to some
extent, however, and could alter the nature of the
equilibrium, dramatically reducing volumes.
Another important aspect of the interaction
between microstructure and international finance
concerns segmentation in internal capital markets.
Such barriers to investment are important because
they may give rise to various documented anomalies, such as discounts on international closedend funds. They also may give rise to arbitrage
trading or other cross-border order flows and thus
affect market efficiency. Finally, an analysis of segmentation may shed light on the positive abnormal
stock returns observed following economic liberalizations. Exhibit 9 illustrates how microstructure
variables can be used to predict exchange rate
changes.
Exhibit 9. Exchange Rate Models
Economic theory suggests that exchange rate movements are
determined by macroeconomic factors. Yet, macroeconomic
exchange rate models, with R2s below 0.10, do not fit the data
well. Evans and Lyons (1999) proposed a microstructure
model of exchange rate dynamics based on portfolio shifts
that augments the standard macroeconomic variables with
(it i t ) is the change in the overnight interest rate differential between the two countries, and xt is the signed order flow.
They estimated their model for the mark/dollar and yen/
dollar exchange rates. As predicted, both 1 and 2 were
positive and significant. The estimated R2 improved substantially when signed order flow was included. More than 50
percent of the daily changes in the mark/dollar rate and 30
percent in the yen/dollar rate were explained by the model.
Applications include short-run exchange rate forecasting,
targeting of central bank intervention, and prediction of trading costs for large transactions.
A puzzling aspect of international segmentation arises when domestic companies issue different equity tranches aimed at different investors. For
example, countries as diverse as Mexico, China,
and Thailand have foreign ownership restrictions
that mandate different shares for foreign and
domestic investors. The objective of such a partition of otherwise identical shares is to ensure that
ownership of corporations rests in the hands of
domestic nationals. Interestingly, the prices of
these two equity tranches vary widely among companies and over time. The share price premiums or
discounts can be explained in terms of relative
trading costs: If both shares are otherwise equal but
one share has higher transaction costs, that share
would have to have a lower price if holding-period
returns are to be equal. Elimination of market segmentation should reduce costs, thus lowering the
September/October 2002
Conclusions
Several conclusions from this survey of the literature are especially relevant for practitioners.
First, markets are a great deal more complex
than commonly believed. One of the major achievements of the microstructure literature is success in
illuminating the black box by which prices and
quantities are determined in financial markets. The
recognition that order flows can have long-lasting
effects on prices has many practical implications.
For example, large price impacts may drive institutional traders to lower-cost venues, creating a
potential for alternative trading systems. It may also
explain the anomalous return behavior associated
with periodic index reconstitutions like those of the
Frank Russell Company indexes in June each year.
Second, microstructure matters. Under certain
protocols, markets may fail and large deviations
between fundamental value and price may
occur. These issues are especially relevant for
exchange officials, operators of trading systems,
regulators, and traders.
Third, one size fits all approaches to regulation and policy making should be avoided. For
example, greater transparency does not always
enhance liquidity.
Finally, the interface of microstructure with
other areas of finance is an exciting new area. A
more complete understanding of the time-varying
nature of liquidity and its relationship to expected
returns is needed; evidence is growing that liquidity is a factor in explaining stock returns. Differences in liquidity over time may explain variations
in the risk premium and may, therefore, influence
stock-price levels.
I thank Ian Domowitz, Margaret Forster, Larry Harris,
Don Keim, and Seymour Smidt for many helpful discussions over the years that influenced my thoughts. Of
course, any errors are entirely my own. The opinions
contained in this article reflect solely the personal views
of the author and do not necessarily reflect the views of
ITG Inc., its employees, officers, or affiliates.
41
References
Almgren, R., and N. Chriss. 1999. Optimal Execution of
Portfolio Transactions. Working paper, New York University.
Amihud, Y., and H. Mendelson. 1986. Asset Pricing and the
BidAsk Spread. Journal of Financial Economics, vol. 17, no. 2
(December):223249.
Amihud, Y., H. Mendelson, and B. Lauterbach. 1997. Market
Microstructure and Securities Values: Evidence from the Tel
Aviv Stock Exchange. Journal of Financial Economics, vol. 45, no.
3 (September):365390.
Bagehot, W. 1971. The Only Game in Town. Financial Analysts
Journal, vol. 27, no. 2 (March/April):1214, 22.
42
2002, AIMR