Capital Market Frictions†

Market Maker Inventories and Stock Prices

By Terrence Hendershott and Mark S. Seasholes*

Empirical studies linking liquidity provision makers or “specialists”—over an 11-year period,

to asset prices follow naturally from inventory this paper contributes to a deeper understanding
models. Liquidity suppliers and market markers of inventory/asset price dynamics. The length of
profit from providing immediacy to less patient our sample enables us to confirm the underlying
investors, but have limited inventory-carrying causal mechanism—liquidity-­supplier inven-
and risk-bearing capacity. Similarly, limits to tory—behind attempts to link liquidity and
arbitrage arguments rely on certain market par­ stock returns through return reversals. Prior
tici­pants accommodating buying or selling pres- data on inventories typically cover relatively
sure. These liquidity suppliers/arbitrageurs are short periods of time and/or a limited number
willing to accommodate trades—and, therefore, of securities. While these limitations prevented
hold suboptimal portfolios—only if they are testing of inventory/price relationships at inter­
able to buy (sell) at a discount (premium) relative day horizons, the microstructure literature has
to future prices. Thus, large liquidity-supplier been quite successful in showing that order flow
inventories should coincide with large buying or and inventories play an important role in intra­
selling pressure, which causes price movements day trading and price formation.
that subsequently reverse themselves. We examine the relationship between closing
By identifying and studying the inventories of market-maker (specialist) inventories and future
traders who are central to the trading process stock prices at daily and weekly horizons. We
and whose primary roll is to provide liquid- find that specialist inventories are negatively
ity—New York Stock Exchange (NYSE) market correlated with contemporaneous returns at
both the aggregate market and individual stock
†
Discussants: Gideon Saar, Cornell University; Hongjun levels. This finding is consistent with special-
Yan, Yale University; Guillaume Plantin, London Business ists acting as dealers and accommodating buy-
School. ing and selling pressure. For the specialist to be
* Hendershott: Haas School of Business, University of compensated for taking on inventory, he must
California, Berkeley, 545 Student Services Bldg. #1900, unwind positions at better prices than those
Berkeley, CA 94720-1900 (e‑mail: [email protected]. prices at which the position was accumulated.
edu); Seasholes: Haas School of Business, University of
California, Berkeley, 545 Student Services Bldg. #1900,
Using returns calculated with quotes (to avoid
Berkeley, CA 94720-1900 (e-mail: mss-hass@emailias. bid-ask bounce), we find that a value-weighted
com). We thank the New York Stock Exchange for provid- portfolio of stocks where the specialist is long
ing data—especially Katharine Ross and Jennifer Chan. outperforms a portfolio of stocks where the
We thank Larry Glosten, Charles Jones, Marc Lipson, specialist is short by 10.3 basis points the day
Gideon Saar, and seminar participants at Stanford Institute
for Theoretical Economics (SITE) and the National Bureau following portfolio formation and 10.2 basis
of Economic Research for helpful comments. Hendershott points the second day after portfolio formation.
gratefully acknowledges support from the National Science Returns decline steadily to 3.4 basis points at
Foundation. Part of this research was conducted while day 5. All these returns are statistically signifi-
Hendershott was a visiting economist at the New York
Stock Exchange. This is a condensed version of a working
cant. At day 10, the long-short portfolio return is
paper with the same title. down to 2 basis points and is no longer statisti-

See Yakov Amihud and Haim Mendelson (1980), cally significant. The cumulative return of the
Thomas Ho and Hans R. Stoll (1981), Sanford J. Grossman long-short portfolio is 45.4 basis points over 10
and Merton H. Miller (1988), and others for inventory mod- days. While these returns seem large, special-
els that lead to reversals. Inventory reversals are empiri-
cally similar to, but on a larger scale than, reversals due to ists do not disclose their inventory positions, so
bid-ask bounce (Richard Roll 1984) and reversals following predictability based on inventories comes from
block trades (Alan Kraus and Hans R. Stoll 1978). nonpublic information.
210
VOL. 97 NO. 2 Market Maker Inventories and Stock Prices 211

Because inventory data have previously been $107 million. Absolute changes in the inventory
unavailable to study longer-horizon returns, position average $77 million each day with a
researchers have constructed proxies for market- standard deviation of $75 million.
maker inventories and limited risk-bearing Daily changes in aggregate inventory have a
capacity. Proxies such as order imbalances 20.71 correlation with contemporaneous mar-
and “liquidity shocks” capture the demand ket returns. Because inventories typically start
for liquidity, which the suppliers of liquid- the day above or below their average level and
ity presumably accommodate (Tarun Chordia, exhibit mean reversion, the correlation of returns
Roll, and Avanidhar Subrahmanyam 2002). over a day and inventory levels at the end of the
John Y. Campbell, Sanford J. Grossman, and day is somewhat lower at 20.57. The cross-
Jiang Wang (1993) examine how trading vol- sectional mean of individual stocks’ time-series
ume interacts with past returns in determining correlation between inventory levels and returns
future return reversals. Lubos Pastor and Robert is 20.23. The skewness of aggregate inventories
F. Stambaugh (2003) use a related measure to (0.71) and individual stock inventories (0.75 on
show that liquidity is a priced risk factor. Simple average) are both positive, implying that spe-
return reversals in individual stocks—Bruce N. cialists take larger positive positions than nega-
Lehmann (1990) and others—are also related to tive positions.
inventory effects. Our approach of directly mea- Each day we allow for time-varying target
suring a supply of available liquidity (i.e., spe- inventory levels (Ananth Madhavan and Seymour
cialist inventories) is complementary to these Smidt 1993) by calculating the moving average
studies. This paper broadens our understanding and standard deviation of each stock’s inventory
of the complex and dynamic process of demand- level over the past three months beginning ten
ing and supplying liquidity by studying it from days ago. We define the standardized inventory
the liquidity-supplier side. (z INV) as the dollar inventory minus its mean
divided by its standard deviation. The average
I. Inventory Data and Descriptive Statistics correlation of individual stocks’ dollar inven-
tory and standardized inventory is 0.76. The
Several datasets are used to construct our average correlation of individual stocks’ dollar
sam­ple of daily specialist inventories and prices inventory and returns is 20.24, while average
from 1994 through 2004. Center for Research in correlation of z INV and returns is 20.25.
Security Prices (CRSP) data are used to identify
common stocks and their trading volume, mar- II. Inventories and Future Returns
ket capitalization, stock splits/distributions, clos-
ing prices, and returns. The Trades and Quotes We now test another inventory model predic-
(TAQ) database is used to identify the closing tion—inventory levels forecast future return
quotes. Internal NYSE data from the special- reversals. While, prior to this paper, there is
ist summary file (SPETS) provide the specialist no direct evidence of empirical support for
closing inventories data for each stock on each the reversal prediction, it is commonly used to
day. We refer to the specialist inventory at the justify and examine the relationship between
end of the trading day simply as “inventory.” To liquidity and prices. Table 1 shows the impact
remove bid-ask bounce, close-to-close returns of inventories on subsequent prices. Following
are calculated using bid-ask quote midpoints. the standard portfolio-formation approach, we
The aggregate market inventory averages sort stocks into quintiles each day of our sample
about $200 million at the end of each day, but period based on two inventory measures. Panel
declines somewhat starting in late 2002. The A sorts by dollar-inventory levels and is labeled
volatility of the inventory levels increases over INV; panel B sorts by our standardized inven-
the beginning of the sample period. Aggregate tory measure and is labeled z INV. Portfolios are
inventory levels reach a maximum of $1 billion formed each day and returns are calculated using
dollars (long) and a minimum of 2$200 mil- closing mid-quote returns with market capital-
lion (short). The inventory level fluctuates with izations as weights. We use mid-quote returns,
a daily standard deviation of $137 million, and value weighting, and quintiles to minimize the
the standard deviation of inventory changes is impact of small illiquid stocks.
212 AEA PAPERS AND PROCEEDINGS MAY 2007

Table 1 

Sort Turnover Return MktCap rt+1

$VNVMBUJWF
QPSUGPMJP
WBMVF


quintile (bp) Stdev (%) ($ bn) (bp)
Panel A: Sort by INV 
Lo (2) 45 2.2 7.7 0.2
P2 40 2.8 2.3 2.5 
 
P3 38 2.8 1.9 4.5         
P4 42 2.4 3.0 7.6  

Hi (+) 49 2.3 9.8 8.6
Hi – Lo 8.5 

Panel B: Sort by z INV 

Lo (2) 40 2.5 4.2 20.6 %BZT
GSPN
TPSU
P2 45 2.6 4.9 2.1
P3 46 2.5 5.8 5.0 )J

[
*/7 1 1 -P
m
[
*/7
1
P4 44 2.4 5.5 7.5
Hi (+) 40 2.4 4.3 9.7
Figure 1. Post-Inventory Sort Portfolio Returns
Hi – Lo 10.3

Sorting by dollar inventory (panel A) puts the To quantify the duration of inventory effects
highest turnover, least volatile, and largest mar- on prices, Figure 1 shows the returns net of
ket capitalization stocks in the outer quintiles. the market for the 12 days after portfolio for-
This suggests that inventory is more manageable mation based on standardized inventory. The
in larger, more active stocks, so specialists are highest inventory portfolio (P5) increases by
willing to take larger positions in these stocks. 5 basis points on the first day, 4 basis points
While such a finding may be expected, it leads on day 2, and asymptotes to 19.0 basis points.
the quintile portfolios to have different stock The lowest inventory portfolio (P1) declines
characteristics. Sorting by the standardized by 5 basis points each of the first 2 days and
inventory measure (z INV) helps to ­distribute mar- decreases by approximately 3 basis points on
ket capitalization more evenly across quintiles. days 3 through 6, and ultimately declines to
Both inventory sorts provide qualitatively 26.4 basis points. The cumulative 5- and 10-
similar result in terms of predicting returns. The day return differences between the long- and
low-inventory portfolios have next-day returns short-­inventory portfolios are 33.0 and 45.4
close to or below zero, while the high-inven- basis points, respectively. Controlling for mar-
tory portfolios have returns between 8 and 10 ket returns, the Fama-French size factor, the
basis points the next day. Therefore, a portfolio Fama-French market-to-book factor, and a
long on the highest inventory stocks and short momentum factor have little effect on the pre-
the lowest inventory stocks yields 8.5 basis dictability results. Each of the first five days’
points (panel A) and 10.3 basis points (panel risk-adjusted return (intercept, often referred to
B) the next day. Both have Newey-West t-sta- as “alpha”) is ­significant. The tenth day’s alpha
tistics greater than nine. The raw returns dem- remains positive at 2.1 basis points, but the
onstrate that the inventory positions of liquidity t-statistic is only 1.8.
providers forecast future prices. When sorting The asymmetry between the returns of the
on dollar inventory, large firms are in the outer highest and lowest portfolios suggests that
portfolios, indicating that the inventory/reversal the specialist’s willingness to take larger long
effect is not a small stock phenomenon. Given positions than short positions translates into
that we are trying to isolate inventory effects differences in future prices. The largest posi-
from other stock characteristics, we focus on the tive positions lead to less mean reversion than
standardized inventory measure for the rest of the most negative positions. The difference in
the paper (although using the dollar inventory returns between the highest inventory and sec-
measure yields similar results.) ond highest inventory position is also smaller
VOL. 97 NO. 2 Market Maker Inventories and Stock Prices 213

than the difference in returns between the ­lowest 

inventory and second lowest inventory position.

$VNVMBUJWF
QPSUGPMJP
WBMVF

When the specialist is short, other traders must 
sell for the specialist to reduce his position (buy
back shares). Traders who do not already own 
the stock face short-sale constraints, potentially
limiting the number of sellers. 
                 
Our findings that long (short) inventories coin-
cide with negative (positive) returns and forecast m
positive (negative) returns the next day are con-
m
sistent with inventory and liquidity-provision
models. To examine the pre- and post-formation
m
price changes, Figure 2 extends the returns in %BZT
GSPN
TPSU
Figure 1 back six days in time by adding the
portfolio formation day as well as the prior -P
m
[
*/7 1 1 1 )J

[
*/7
five days. Note that the ordering of the high-
and low-inventory portfolios is switched when Figure 2. Pre- and Post-Inventory Sort
compared with Figure 1. The Y-axis measures Portfolio Returns
cumulative returns (prices), which means the
highest inventory portfolio is on top in Figure 1,
while the highest inventory portfolio is on the
bottom in Figure 2. The graphs are consistent to the specialist preferring long positions to
with the specialists acquiring their positions short positions. This preference leads to smaller
as they accommodate the liquidity demands of downward price changes by day 0 and smaller
other traders. The specialists then unwind their subsequent return reversals over days 1 to 12.
positions as prices reverse. While our results show that the marginal
The highest and lowest inventory portfolios additional dollar of inventory appears profit-
exhibit asymmetry prior to formation with the able, most of the large long (short) inventory
high-inventory portfolio falling 1.29 percent and positions occur on days when prices fall (rise).
the lowest portfolio rising 1.48 percent. As in Prices then show small mean reversion relative
Figure 1, the highest portfolio then reverses 19.0 to the pre-formation return, making these large
basis points, while the lowest portfolio reverses inventory positions appear unprofitable overall
26.4 basis points. The pre- and post-formation for the specialist.
returns show that price changes prior to port- Finally, we examine “day-of-the-week” effects
folio formation are many times larger than the in the inventory induced reversals. The predict-
reversal. Just as the asymmetry between the able reversal (over a week) of the high-­inventory
post-formation returns of the longest and short- minus low-inventory portfolio is 50 percent
est inventory positions does not naturally arise higher when sorting at the end of the calendar
in inventory models, neither do the asymmetric week versus on Wednesdays. The greater pre-
price movements in the pre-formation periods. dictability at the end of trading weeks is due
The asymmetry in long- and short-inventory to the specialists needing to hold suboptimal
size and pre- and post-formation returns points portfolios for a longer period of time—over the
weekend as opposed to overnight.


Specialists often give a different explanation for the
long-short asymmetry. They claim to be more sensitive to
preventing downward stock price movements and, there-

fore, take large long positions when others investors are This is consistent with the Joel Hasbrouck and George
net sellers. This explanation appears less plausible because Sofianos (1993) evidence that the specialists make most of
there is asymmetry in inventory levels, but not in changes their money at short horizons and are not profitable at lon-
in inventory. ger horizons.
214 AEA PAPERS AND PROCEEDINGS MAY 2007

III. Conclusion References

Liquidity and limits to arbitrage arguments Amihud, Yakov, and Haim Mendelson. 1980.
regarding asset prices rely on the idea that cer- “Dealership Market: Market Making with
tain market participants accommodate buying or Inventory.” Journal of Financial Economics,
selling pressure. These liquidity suppliers/arbi- 8(1): 31–53.
trageurs will hold suboptimal portfolios only if Campbell, John Y., Sanford J. Grossman, and
they are compensated by favorable subsequent Jiang Wang. 1993. “Trading Volume and Serial
price movements. Thus, when inventories are Correlation in Stock Returns.” Quarterly Jour­
large, liquidity suppliers have deviated from their nal of Economics, 108(4): 905–39.
optimal portfolios. Associated price changes Chordia, Tarun, Richard Roll, and Avanidhar
should subsequently reverse. Using a unique Subrahmanyam. 2002. “Order Imbalance,
11-year sample of NYSE specialist inventories, Liquidity, and Market Returns.” Journal of
this paper is able to test and confirm the under- Financial Economics, 65(1): 111–30.
lying causal mechanism—liquidity supplier Grossman, Sanford J., and Merton H. Miller.
inventory—behind attempts to link ­liquidity and 1988. “Liquidity and Market Structure.” Jour­
stock returns through return reversals. Consis­ nal of Finance, 43(3): 617–37.
tent with specialists acting as dealers and tem- Hasbrouck, Joel, and George Sofianos. 1993.
porarily accommodating buying and selling “The Trades of Market-Makers: An Analysis
pressure, we find that specialist inventories are of NYSE Specialists.” Journal of Finance,
negatively correlated with contemporaneous re-­ 48(5): 1565–93.
turns at both the aggregate market level and Ho, Thomas, and Hans R. Stoll. 1981. “Optimal
individual stock level. We find that specialists Dealer Pricing under Transactions and Return
are compensated for inventory risk by return Uncertainty.” Journal of Financial Economics,
reversals. 9(1): 47–73.
Substantial work remains to be done in under- Kraus, Alan, and Hans R. Stoll. 1972. “Price
standing the dynamics of specialist inventories. Impacts of Block Trading on the New York
Better knowledge of these dynamics should Stock Exchange.” Journal of Finance, 27(3):
help refine analysis of inventories and prices. 569–88.
First, the length of our sample can allow for a Lehmann, Bruce N. 1990. “Fads, Martingales,
detailed study of the mean reversion in inven- and Market Efficiency.” Quarterly Journal of
tories both cross-sectionally and over time. Economics, 105(1): 1–28.
Second, individual specialists trade a number Madhavan, Ananth, and Seymour Smidt. 1993.
of stocks referred to as a “panel.” Inventory and “An Analysis of Changes in Specialist Inven­
return dynamics within a given panel may be tories and Quotations.” Journal of Finance,
important for risk management. Third, special- 48(5): 1595–1628.
ists are part of larger firms. Inventory and return Pastor, Lubos, and Robert F. Stambaugh. 2003.
dynamics within a given firm may be important “Liquidity Risk and Expected Stock Returns.”
for risk management. A deeper understanding Journal of Political Economy, 111(3):
of market-making firms and the inventories of 642–85.
other financial intermediaries may generate Roll, Richard. 1984. “A Simple Implicit Measure
additional hypotheses regarding the dynamics of the Effective Bid-Ask Spread in an Efficient
of trading and prices. Market.” Journal of Finance, 39(4): 1127–39.