Journal of Financial Markets 6 (2003) 49–72
Who makes markets
$
Paul Schultz*
College of Business Administration, University of Notre Dame, Notre Dame, IN 46556, USA
Abstract
A dealer needs access to order flow and information to make a market profitably in a
Nasdaq stock. Several variables that proxy for the stocks that an individual market maker’s
brokerage customers trade, including volume, location, underwriting participation and analyst
coverage, are significant determinants of market making activity. Informational advantages
may also factor in the market making decision as evidenced by dealers specializing in
industries. These findings suggest that individual dealers have competitive advantages in
making markets in specific stocks, and that potential market making competition comes from
the dealers who share those advantages rather than all Nasdaq market makers.
r 2002
Elsevier Science B.V. All rights reserved.
JEL classification:
G24; D43; G14
Keywords:
Nasdaq; Market making; Brokerage; Underwriting
1. Introduction
A common view is that there are dozens, or even hundreds of potential market
makers for any Nasdaq stock. This paper makes the point that there may be far
fewer dealers who can effectively make a market in a stock because different dealers
have competitive advantages in making markets in different stocks. There are two
sources for these competitive advantages: access to order flow and access to
information. Customer orders in Nasdaq stocks are seldom sent to a market maker
$
I am grateful to John Affleck-Graves, Robert Battalio, Thomas Cosimano, Raymond Fishe, Marc
Huson, Aditya Kaul, Youngsoo Kim, Tim Loughran, Tim McCormick and seminar participants at The
University of Alberta, The University of Minnesota, The University of Notre Dame and the Securities and
Exchange Commission for helpful suggestions. I am indebted to I
=B=E=S for information on analyst
recommendations. All errors are my own.
*Tel.: +1-219-631-6370; fax: +1-219-31-5255.
E-mail address:
paul.h.schultz.19@nd.edu (P. Schultz).
1386-4181/02/$ - see front matter
r 2002 Elsevier Science B.V. All rights reserved.
PII: S 1 3 8 6 - 4 1 8 1 ( 0 2 ) 0 0 0 2 2 - 8
on the basis of his quoted price for the stock, so market makers must obtain order
flow through a brokerage business or by purchasing it from brokers.
1 Informational
advantages accrue to market makers through their underwriting activities, their
trading in other securities and the work of their analysts.
I use four variables in this paper to proxy for the ease of receiving order flow: the
stock’s trading volume, whether the market maker provides analyst coverage, the
location of the company and whether the market maker was part of the stock’s IPO
syndicate. A stock’s volume is used because a market maker with an institutional
brokerage business can expect to receive order flow only in the high volume stocks
that their customers trade. Analyst coverage of a stock is needed for order flow
because investors usually buy a stock through the broker that provides information
about it. For market makers with regional brokerage businesses location matters
because their brokerage customers are more likely to invest in local firms than in
stocks of more distant companies. Participation in a stock’s underwriting syndicate
makes it easier to obtain order flow in the stock because some of the market maker’s
brokerage customers will have purchased the stock in the offering.
Analyst coverage, company location and participation in a stock’s underwriting
syndicate are also associated with informational advantages. A market maker who
provides analyst coverage is more likely to have non-public information about the
company or to anticipate news releases than a market maker who does not provide
coverage. A market maker is also more likely to obtain information about a stock
through conversations with the firm’s employees or customers if the company is a
local one than if it is based in another state. Participation in an underwriting
syndicate may confer informational advantages to the market maker as a result of
the due diligence process or an ongoing consulting relation with the firm. An
additional variable, the firm’s industry, is also used as a proxy for informational
advantages in market making. Market makers may specialize by industry if
information learned in trading one stock provides an advantage in making markets
in other stocks in the same industry.
The evidence presented in this paper confirms that sources of order flow and access
to information provide competitive advantages to dealers for market making in
specific stocks. Market makers with regional brokerage businesses are far more likely
to make markets in local stocks than others. Dealers of all types are much more
likely to make a market in a stock they helped underwrite or a stock their analysts
follow. High trading volume is a particularly important requirement for market
making for dealers with institutional brokerage business or national retail businesses.
Many market makers do specialize in stocks in particular industries, even after
adjusting for the effects of location and IPO syndicate participation.
These results suggest that we need to change the way we look at competition
between dealers. It has been said that for a Nasdaq market maker, making a market
in a new stock is ‘‘as easy as turning on a light’’. Indeed, Wahal (1997) finds that
entry and exit from market making in specific Nasdaq stocks is common. However,
1
Ellis et al. (1999) note that ‘‘Preferencing agreements are particularly prevalent on the Nasdaq, where it
is estimated that in some stocks virtually all order flow is preferenced’’.
50
P. Schultz / Journal of Financial Markets 6 (2003) 49–72
making a market profitably requires more than simply quoting a stock. The market
maker must find a source of order flow, either through its own brokerage business or
by purchasing order flow. The market maker may be required to provide analyst
coverage of the stock. The ease or difficulty in making a market in a stock will
depend on the nature of the market maker’s brokerage business, its other market
making positions, and its investment banking business. Dealers who have
advantages in market making in particular stocks may appear to earn abnormally
large returns trading these stocks, at least in the short-run.
A second implication of these findings is that we need to reconsider the standard
assumption that market makers are uninformed. Dealers’ preference for making
markets in local stocks and stocks that they have underwritten, and especially their
tendency to specialize in particular industries suggests that they trade stocks where
they have an informational advantage. If this is true, dealers may exploit information
by executing orders for their own account selectively. They may attempt to conceal
information rather than posting quotes that fully reveal what they know. These
elements of dealer behavior are typically ignored in microstructure models.
The rest of the paper is organized as follows. The data used here is described in
Section 2. Section 3 provides evidence on the relation between a dealers’ sources of
order flow and their choices of stocks in which to make markets. Section 4 shows
that some dealers specialize in making markets in stocks in particular industries and
discusses whether dealers have informational advantages in making markets in some
stocks. In Section 5, I examine how different characteristics of a stock determine the
probability that a dealer will make a market in the stock. Section 6 offers a summary
and conclusions.
2. Data
The sample employed here consists of monthly observations for all Nasdaq stocks
over May 1995 through February 1998. Data is obtained from several sources.
Nasdaq provided total share volume reported by each dealer in every Nasdaq stock
each sample month. The location of the headquarters of every Nasdaq listed stock is
also provided by Nasdaq. Daily closing prices of the stocks are obtained from the
NYSE’s TAQ database. Information on market makers’ businesses (institutional,
wholesale, national retail, etc.) and on the location of their headquarters is obtained
from the Securities Industry Association member directory and the Standard and
Poor’s Guide to North American Securities Dealers. Data on participation in IPO
underwriting syndicates comes from Security Data Corporation (SDC). I
=B=E=S
provided information on analyst coverage and recommendations.
Table 1 provides summary statistics on the distribution across stocks of the
number of market makers in a stock, the stock’s Herfindahl index, the daily dollar
volume, and the daily number of trades. All stocks with market share data in the
Nasdaq market share database and with quote information in the TAQ database are
included. The number of stocks during a month varies from 5,669 to 6,340. Statistics
are reported for six months that span the sample period: May 1995, December 1995,
P. Schultz / Journal of Financial Markets 6 (2003) 49–72
51
June 1996, December 1996, June 1997 and February 1998. The last column of the
table provides the time series average of the statistics over the 34 months.
Panel A reports on the distribution of the number of market makers per stock.
While there is some slight variation from month to month, the mean number of
market makers is generally about 10.5, the median is about 9, and the first and third
quartile are generally around 5 and 14 market makers, respectively. The slight
increase in the number of market makers shown in February is a result of including
ECN quotes in Nasdaq.
The Herfindahl–Hirschman (henceforth Herfindahl) index is reported in Panel B
of Table 1. This is a measure of the concentration of market making. For stock
i in
month
t; it is
Herfindahl
i;t ¼X
N
n
¼1
p
2
n
;i;t; ð1Þ
Table 1
Summary statistics for Nasdaq stocks over May 1995 through February 1998
Volume and trades per day are obtained from the TAQ data. Nasdaq provided each market maker’s share
volume in every Nasdaq stock each month of the sample period. The Herfindahl measures and the number
of dealers are calculated monthly for each stock from this data. If a dealer reports any trading volume for
the month he is counted. Each month, the cross-sectional quartiles and mean of each variable are
calculated. This table reports the time-series means of these quartiles. Grand means of the 34 monthly
means and quartiles are reported in the last column.
May December June December June February Mean
1995 1995 1996 1996 1997 1998 5
=95–2=98
Panel A. The number of market makers
Mean 10.6 10.3 10.1 9.8 10.5 11.8 10.4
Third quartile 14 14 13 13 14 16 13.9
Median 9 9 8 8 9 10 8.8
First quartile 5 5 5 5 5 6 5.1
Panel B. Herfindahl indices
Mean 3,140 3,001 3,049 3,061 3,057 3,038 3,061.4
Third quartile 3,969 3,912 3,915 4,032 3,963 3,955 3,987.2
Median 2,487 2,392 2,439 2,447 2,444 2,443 2,444.5
First quartile 1,593 1,483 1,506 1,521 1,542 1,528 1,524.7
Panel C. Daily dollar volume
Mean 1,537 2,040 2,184 2,227 2,658 3,434 2,408.0
Third quartile 430.0 642.9 873.6 675.6 691.4 895.8 695.7
Median 81.5 128.7 169.2 146.7 137.3 164.4 137.8
First quartile 18.7 30.7 37.5 34.0 30.9 37.5 31.4
Panel D. Number of trades per day
Mean 36 52 59 58 63 89 59.9
Third quartile 23 35 41 39 38 54 37.0
Median 8 11 12 13 12 16 11.5
First quartile 3 4 4 4 4 5 4.2
52
P. Schultz / Journal of Financial Markets 6 (2003) 49–72
where
pn;i;t is the percentage of the volume in stock i handled by market maker n:
Note that if one market maker had 100% of the volume in stock
i; the Herfindahl
index would be 100
2 ¼ 10; 000: If four dealers each handled 1
4
of the total volume, the
Herfindahl index value would be 4
ð25Þ2 ¼ 2; 500:
The Herfindahl index increases as the number of market makers decreases or as
the proportion of business taken by the leading market maker increases. Thus a high
Herfindahl index is associated with a lack of competition.
2 This measure of
concentration varies little during the sample period. Table 1 shows that the mean
Herfindahl value ranges between 3,001 and 3,140, while the median is always
between 2,392 and 2,487. The first and third quartiles are usually around 1,500 and
4,000. When the mean Herfindahl index values are compared with the number of
dealers, it is apparent that the number of dealers in a stock tends to overstate the
degree of competition. The average number of market makers exceeds ten, but the
Herfindahl index is above 2,500, the value that would occur if four market makes
split all the volume equally. This is consistent with Ellis et al. (1999), who find that
one-third of dealers trade less than once per week.
Panel C shows the distribution of the daily dollar volume (in thousands of dollars)
across stocks. Dollar volume increases steadily over the period, with the mean,
median and quartiles of the distribution all approximately doubling between May
1995 and February 1998. Dollar volume is right-skewed. For example, the mean
daily dollar volume is $3,433,600 in February 1998, while the median is only
$164,400. Panel D of Table 1 shows that the mean median and quartiles of the
number of trades also double between May 1995 and February 1998. It appears that
the increase in dollar volume reported in Panel C is a result of more trades, not
bigger trades.
3. Sources of order flow and market making
While precise figures are unavailable, it is usually conceded that only a small
proportion of orders are sent to a market maker because he has the best quote.
Instead, most orders are preferenced. Preferencing can take two forms. First, the
market maker can internalize order flow from its brokerage customers. This is the
major source of order flow for most Nasdaq market makers. Alternatively, the
market maker can pay for order flow from brokers. This is the source of order flow
for wholesalers.
Because most Nasdaq order flow is preferenced, differences in dealers’ choices of
stocks to handle should be generated by differences in their brokerage customers, or
2
Kremer and Polkovnichenko (2000) find that Nasdaq stocks with higher Herfindahl indices have wider
quoted spreads even after adjustment for market capitalization and volatility. However, Dutta and
Madhavan (1997) suggest that collusion is most likely when all dealers have equal market shares. Dutta
and Madhavan consider a cartel of dealers. In their model, the dealer who undercuts the spread gets all the
order flow for one period but the cartel then collapses. The dealer with the smallest order flow has the least
to lose, so he would be the one to undercut the spread. Thus tacit collusion is more likely when all have
approximately equal market share.
P. Schultz / Journal of Financial Markets 6 (2003) 49–72
53
the customers of the firms that sell order flow to them. To examine this, I divide
market makers into four categories; institutional brokers, national retail brokers,
wholesale market makers, and regional brokerage firms. This last category includes
all market makers other than electronic communications networks (ECNs) that do
not fit into one of the other classifications. Categorizations are based on the firms’
self-descriptions in the Securities Industry Association member directory. The
categorizations are not perfect. Merrill Lynch for example is classified as a national
retail broker, but has a large institutional business as well. Dain Bosworth is
classified as a national retail broker but is concentrated primarily in the midwest.
Table 2 provides information on the market making activities of dealers in the
different categories for December 1996, a typical month in the middle of the sample
period. Panel A reports the number of markets made by different types of dealers.
For each dealer each month, I calculate the number of markets made. I then
calculate the cross-sectional mean, median, maximum and first and third quartiles
across all dealers and across all dealers with wholesale, national retail, institutional
or regional businesses. When all dealers are considered, the median number of
markets made is 20. Panel A reveals that regional brokers make up the great
majority of Nasdaq market makers. While the number of all dealers is 542, the
number of regional brokers is 470. The table also shows that regional brokers make
markets in far fewer stocks than others. The median number of markets made by
regional brokers is only 17, while the medians for wholesalers, national retail brokers
and institutional brokers all exceed 180.
3
Panel B provides data on the daily dollar volume of trading by different categories
of market makers. The median of all dealers’ daily volume is $422,000. When dollar
volumes are broken down by type of dealer, we see that regional brokers’ trading
volume is far lower than that of other dealers. While the median regional broker’s
daily dollar volume is $287,000, the median for wholesalers is $51,916,000 and the
median for institutional brokers is $75,799,000.
3.1. Trading volume and market making
Institutions tend to trade large, high-volume stocks that provide sufficient
liquidity to allow them to take large positions. Thus we would expect that market
makers whose brokerage business is primarily from institutions would receive order
flow in large, high volume stocks and would make markets in those stocks. National
retail brokers make the same recommendations to thousands of retail clients, and
would thus need to recommend stocks that are large and liquid enough to
accommodate their many investors. On the other hand, wholesale market makers
typically buy order flow from a number of small brokers. Because they are buying
orders from many different brokers with different clienteles and different
recommended securities, they must make markets in a large number of stocks.
Thus they will have a larger share of their market making positions in small stocks
than will institutional brokers. Regional brokers, unlike institutional brokers or
3
Ho and Macris (1985) find a similar distribution of number of markets made in the 1980s.
54
P. Schultz / Journal of Financial Markets 6 (2003) 49–72
Table 2
Panel A. The distribution of the number of markets made by dealers during December 1996
The 542 dealers all reported volume from market making in at least one Nasdaq stock during December 1996. Classifications are based on the firm’s selfdescription
in the Securities Industry Association member directory. All firms not otherwise described are classified as regional brokers.
Number of markets made
Number of dealers Mean Minimum 25th percentile Median 75th percentile Maximum
All dealers 542 110 1 8 20 62 4,884
Wholesalers 16 1,273 4 50 295 2,485 4,884
National retail brokers 30 377 12 171 290 570 1,156
Institutional brokers 26 220 2 42 185 342 618
Regional brokers 470 48 1 7 17 45 1,374
Panel B. The distribution of the dollar volume handled by dealers during December 1996
Total daily dollar volume for each dealer is approximated by multiplying the average daily closing share price of each stock over the month by the total share
volume reported by the dealer in every Nasdaq stock for the month divided by the number of trading days in the month. Daily dollar volumes are then
averaged across all dealers.
Dollar volume handled
Mean Minimum First quartile Median Third quartile Maximum
All dealers $24,113,000
o$1,000 $84,000 $422,000 $2,630,000 $906,673,000
Wholesalers $170,711,000 $299,000 $2,563,000 $51,916,000 $348,433,000 $656,913,000
National retail brokers $132,777,000 $8,000 $8,763,000 $38,015,000 $163,043,000 $906,673,000
Institutional brokers $168,619,000 $4,000 $1,118,000 $75,799,000 $269,260,000 $885,549,000
Regional brokers $4,193,000
o$1,000 $70,000 $287,000 $1,139,000 $214,680,000
P. Schultz / Journal of Financial Markets 6 (2003) 49–72
55
national retailers are not limited to making markets in large stocks. Because they
have a small number of clients, the low trading volume in small stocks provides
sufficient liquidity for their customers.
To see if the source of the market makers’ order flow determines which stocks they
handle, I divide stocks into dollar volume quintiles each month. I then calculate the
proportion of market making positions by all dealers in each dollar volume quintile.
I also calculate the proportion of market making positions in each dollar volume
quintile separately for institutional brokers, national retail market makers, wholesale
market makers, and regional brokers. Results are reported in Table 3.
The first row of Table 3 shows the proportion of market making positions by
dollar volume quintiles for all market makers. The second column gives the number
of market makers. The number of market makers is calculated separately for each of
the 34 months, and the first number, 530.3, is the average number of market makers
across the 34 months. The numbers in parentheses below are the minimum and
maximum number of dealers. The number ranges from 480 to 555. In the next
columns of the first row we see that the average, across months, of the proportion of
market making positions in the smallest dollar volume quintile of stocks is 10.15%.
The range is from 9.38% to 10.70%. On average 14.61% of market making positions
are in the second dollar volume quintile, 18.70% are in the third quintile, 23.28% are
in the fourth quintile and 33.27% are in the largest dollar volume quintile. Not
surprisingly, higher volume stocks have more market makers.
When the proportion of market making positions is calculated separately for
market makers who obtain order flow from different sources a pattern emerges.
Market makers who are institutional brokers have only about 2.5% of their
positions in the lowest volume quintile and about 63% of their positions in the
highest volume quintile of stocks. This is what we would expect if these market
makers chose stocks on the basis of order flow from their clients. National retail
brokers have 6.5% of their market making positions in the quintile of stocks with the
lowest volume and about 41.4% of their market making positions in the quintile of
stocks with the highest volume.
Table 3 also shows that the market makers who receive order flow in the less active
stocks are more likely to make markets in them than other dealers. Wholesale
market makers have 12.5% of their positions in the quintile of the least active stocks
as compared to 10.15% for all market makers. Only 27.9% of their positions are in
the most active stocks as compared to 33.3% for all market makers. Similarly,
regional brokers have 12.4% of their positions in the quintile of the least active
stocks while only 25.4% of their positions are in the most active stocks.
In Panel B of Table 3, relative market shares are shown for each market maker
type for each volume quintile. Each month, the relative market share for each market
maker type and volume quintile pair is obtained by dividing the average market
share of the market maker type in stocks in the dollar volume quintile by the average
market share of all dealers in the dollar volume quintile. The table reports the
average across all months of each relative market share as well as the minimum and
maximum mean relative market share across the 34 months. So for example, the first
row in the table shows that the average market share is 6.01% in the quintile of the
56
P. Schultz / Journal of Financial Markets 6 (2003) 49–72
Table 3
Panel A. The proportion of market makers’ positions in stocks in different dollar volume quintiles
Each month, stocks are sorted into quintiles based on dollar volume. Each dealer is placed in one of four mutually exclusive categories: institutional brokers,
national retail brokers, wholesale market makers, and regional brokerage firms. Dealer classifications are obtained from self-descriptions in the Securities
Industry Association member directory. The proportion of each dealer’s positions that are in each dollar volume quintile is calculated each month, and a crosssectional
average is calculated for each type of dealer. A grand average is then calculated across months from May 1995 through February 1998.The first
number in each cell is this grand average. The numbers in parentheses are the minimum and maximum across sample months.
Proportion of market making positions in each dollar volume quintile
Market maker Number of Lowest dollar Highest dollar
type market makers volume 2 3 4 volume
All 530.3 0.1015 0.1461 0.1870 0.2328 0.3327
(480–555) (0.0938–0.1070) (0.1321–0.1549) (0.1727–0.1969) (0.2199–0.2439) (0.3134–0.3633)
Institutional brokers 24.7 0.0257 0.0437 0.0971 0.2052 0.6283
(22–27) (0.0167–0.0388) (0.0371–0.0509) (0.0823–0.1098) (0.1755–0.2238) (0.5934–0.6606)
National retail brokers 29.2 0.0650 0.1102 0.1649 0.2455 0.4144
(27–30) (0.0578–0.0729) (0.0955–0.1206) (0.1457–0.1750) (0.2356–0.2617) (0.3974–0.4457)
Wholesale market makers 15.6 0.1249 0.1667 0.1985 0.2306 0.2794
(15–16) (0.1162–0.1321) (0.1529–.1779) (0.1835–0.2080) (0.2207–0.2442) (0.2584–0.3050)
Regional brokers 459.1 0.1236 0.1765 0.2130 0.2331 0.2538
(415–478) (0.1091–0.1428) (0.1578–0.1891) (0.1957–0.2253) (0.2168–0.2467) (0.2318–0.2783)
P. Schultz / Journal of Financial Markets 6 (2003) 49–72
57
Table 3 (
continued )
Panel B. Relative market shares for different market maker types by trading volume quintile
Average market shares are obtained for each dollar volume quintile by each market maker classification each month by taking a simple average across all
positions of all dealers in the category. The relative market share is obtained by dividing the average market share of a dealer type in stocks in a dollar volume
quintile by the average market share of all dealers in the dollar volume quintile. Simple average are taken across month in the sample periods. The table shows
averages across months with the minimum and maximum monthly averages underneath.
Market share
Market maker Number of Lowest dollar 2 3 4 Highest dollar
type market makers volume volume
All 530.3 0.1926 0.1345 0.1054 0.0851 0.0601
(480–555) (0.1761–0.2070) (0.1284–0.1409) (0.0960–0.1108) (0.0732–0.0927) (0.0470–0.0664)
Relative market share
Institutional brokers 24.7 1.386 1.463 1.620 1.710 1.474
(22–27) (1.004–1.622) (1.234–1.732) (1.350–1.830) (1.551–1.967) (1.338–1.800)
National retail brokers 29.2 1.314 1.447 1.438 1.383 1.310
(27–30) (1.199–1.446) (1.358–1.621) (1.363–1.608) (1.295–1.638) (1.206–1.495)
Wholesale market makers 15.6 0.889 0.864 0.854 0.809 0.771
(15–16) (0.831–1.015) (0.792–1.049) (0.764–1.100) (0.729–1.055) (0.683–0.962)
Regional brokers 459.1 1.012 0.976 0.924 0.870 0.751
(415–478) (0.964–1.045) (0.907–1.017) (0.851–1.004) (0.783–0.985) (0.670–0.846)
58
P. Schultz / Journal of Financial Markets 6 (2003) 49–72
highest dollar volume stocks. However, as shown in the second row of the table, on
average an institutional broker has a market share of 1.474 times as much, or 8.86%
in the quintile of stocks with the highest trading volumes.
Results in Panel B indicate that regional brokers’ market shares are high relative
to other market makers in low volume stocks, but on those occasions when they do
make markets in high volume stocks their market share is smaller than that of the
average dealer. Similarly, wholesaler’s relative market shares are smaller for more
active stocks.
To summarize, the results in Table 3 are consistent with dealers making markets in
the stocks in which they receive order flow. Market makers with a brokerage clientele
that consists mainly of large investors make markets in active stocks. Market makers
whose brokerage clientele consists of a small number of small investors are much
more likely to make markets in less active stocks.
3.2. Geographic location, order flow, and market making
It has long been known that investors are biased toward holding domestic stocks
over foreign shares. Recent work shows that in addition, investors prefer to buy
stocks in local companies rather than domestic companies that are located further
away. Coval and Moskowitz(1999) find that professional money managers in the
U.S. tend to invest disproportionately in local companies. They are particularly
likely to exhibit this bias for local businesses when investing in small firms. Similarly,
Huberman (1998) finds that investors are much more likely to hold shares in the
Regional Bell Operating Company that provides phone service to their area than
another Regional Bell. Investors’ preference for local companies suggests that
Nasdaq market makers with a regional brokerage businesses will get a disproportionate
share of their order flow in local stocks and will thus tend to make markets in
local stocks.
To examine the relation between location and market making activity, I calculate,
for each dealer each month, the number of market making positions in companies
located in the same state that would be expected if the decision to make a market was
independent of location. This is obtained by multiplying the total number of market
making positions by all dealers in in-state stocks by the proportion of all market
making positions held by in-state dealers. This number is summed across all regional
market makers, along with the actual number of positions taken in stocks of
companies located in the state. Results for May 1995, December 1995, June 1996,
December 1996, June 1997, and February 1998 are shown in the first two rows of
Table 4. In each month, the number of actual market making positions in same-state
stocks exceeds 4,300 while the expected number is always less than 1,500. In their
choices of stocks, regional market makers display a strong bias toward local
companies. In some cases the bias is extreme. There is a 72.5% probability that
Hilliard and Lyons will make a market in a stocks from their home state of
Kentucky, but only a 1.7% chance that they will make a market in another stock.
There is a 72.2% chance that John J. Kinnard will make a market in a stock from
P. Schultz / Journal of Financial Markets 6 (2003) 49–72
59
their home state of Minnesota but only a 3.1% chance that they will make a market
in a stock from another state.
There are 34 states, including the District of Columbia, with both regional market
makers and Nasdaq listed stocks. I sum the actual and expected number of samestate
market making positions separately for dealers in each of these states. I then
calculate the number of states whose dealers make markets in more same-state stocks
than expected. As shown in the third and fourth rows of the table, dealers in almost
all states make markets in more local stocks than would be expected. Furthermore,
as shown in the fifth row of the table, for most states the number of positions by
dealers in same state stocks is more than ten times the expected number.
The last two rows of the table show the mean market share for regional dealers in
same-state stocks, the mean market share for regional dealers in stocks from other
states, and the mean market share of all dealers in all stocks. The mean market share
of regional dealers in same-state stocks ranges from 11.05% to 12.87%, while the
same dealers’ market share in stocks from out-of-state ranges from 8.53% to 9.27%.
3.3. Analyst coverage and market making
If order flow from brokerage customers is an important determinant of a dealer’s
choice of market making positions, we would expect dealers to make markets in the
stocks that they are recommending to their brokerage customers. I obtain, from
I
=B=E=S; the buy/sell recommendations for Nasdaq stocks made by firms that made
markets in Nasdaq stocks over the sample period. The data includes the date of each
recommendation and whether it was a strong buy, buy, hold, underperform or sell
recommendation. Table 5 provides data on the relation between market making and
analyst coverage for May 1995, December 1995, June 1996, December 1996, June
1997 and February 1998.
All dealers with analyst recommendations in I
=B=E=S are included in the table.
For each dealer each month, I calculate the expected number of stocks in which the
Table 4
Market making activities by regional dealers in stocks located in the same and other states
The location of the stock’s headquarters and the location of the market makers’s headquarters are their
states. For each market maker, the expected number of positions in stocks based in the same state is
obtained by multiplying the number of market making positions by all dealers in stocks based in the state
by the proportion of all stocks in which the dealer is a market maker.
May
1995
Dec.
1995
June
1996
Dec.
1996
June
1997
Feb.
1998
Actual number of positions in same-state stocks 4,578 4,341 4,336 4,623 4,883 4,894
Exp. number of positions in same-state stocks 1,249 1,245 1,301 1,392 1,413 1,393
Number of states with regional dealers 32 32 32 34 34 34
Number of states, in-state positions
> expected 32 32 32 34 34 33
Number of states, in-state positions
> 10 expected 20 21 20 21 20 20
Mean mkt. share in same-state stocks 11.71% 11.84% 12.64% 12.87% 12.36% 11.05%
Mean mkt. share in other state stocks 8.53% 8.94% 9.18% 9.27% 9.05% 8.95%
60
P. Schultz / Journal of Financial Markets 6 (2003) 49–72
dealer would both make a market and provide analyst recommendations if the two
activities were independent. This is obtained by multiplying the total number of
Nasdaq stocks during the month by the proportion of stocks that the dealer makes a
market in and the proportion of stocks for which the dealer provides analyst
recommendations. The expected number, and actual number of stocks in which
dealers both made markets and provided analyst coverage is averaged across dealers.
As shown in the table, dealers are far more likely to make markets in the stock that
their analysts cover. For example, in May 1995 dealers both made markets and
provided analyst coverage for an average of 48.1 stocks. The expected number of
stocks with both activities is 2.7. Analyst recommendations are also, on average,
more favorable when the analyst works for a firm that makes a market in the stock.
With 1 as a strong buy, 2 as a buy, and 3 as a hold, the mean recommendation in
May 1995 is 1.92 if the firm made a market in the stock and 2.45 if it did not. Similar
results are reported for other months. Paired sample
t-tests for each month indicate
that the mean recommendation is always significantly more favorable, at the 1%
confidence level, for stocks in which the dealers made markets. Of course, it is not
clear whether market makers choose to make markets in stocks their analysts like, or
whether their analysts choose to like the stocks in which they make markets.
Table 5
Market making and analyst coverage
Analyst recommendations are obtained for Nasdaq stocks from I
=B=E=S: Recommendations are on a
5-point scale: 1 is a strong buy, 2 is a buy, 3 is a hold, 4 is underperform, and 5 is sell. The mean analyst
recommendations, market shares and dollar volumes are calculated for each dealer. Averages across
dealers are reported in the table.
t-Statistics for differences in dealer means are based on cross-sectional
standard deviations of differences in means.
5
=95 12=95 6=96 12=96 6=97 2=98
Number of dealers providing
analyst reports on I
=B=E=S
92 92 93 93 92 91
Mean number of stocks both analyst and dealer 48.1 45.2 39.7 35.6 32.2 25.7
Mean number of stocks market maker only 170.8 169.5 176.7 185.1 195.0 201.3
Mean number of stocks analyst only 10.5 14.0 16.4 17.8 18.4 17.0
Mean expected number both analyst and dealer 2.7 2.6 2.3 2.2 2.1 1.9
Number dealers chi-square test indicates
(at 1% level) that are both market
maker and analyst more often than if independent
91 92 90 90 89 90
Mean analyst recommendation when
the analyst is also a dealer
1.92 1.94 1.91 1.90 1.88 1.92
Mean analyst recommendation when
the analyst is not a dealer
2.45 2.43 2.35 2.30 2.27 2.21
t
-Statistic for difference in recommendations 8.93 8.03 8.14 8.18 8.15 6.19
Mean market share when both a dealer and analyst 12.0% 11.2% 11.2% 11.1% 10.8% 10.0%
Mean market share when a dealer only 10.9% 11.6% 11.6% 11.9% 11.4% 11.1%
Mean dollar volume when both a dealer and analyst $7.3m $8.4m $8.6m $10.8m $12.4m $13.3m
Mean daily dollar volume when a dealer only $2.6m $3.6m $4.3m $4.3 m $4.8m $5.5m
t
-Statistic for difference in volumes 5.26 4.86 3.04 4.57 4.13 3.66
P. Schultz / Journal of Financial Markets 6 (2003) 49–72
61
Interestingly, market share is often lower for stocks in which the dealer both
makes a market and provides analyst coverage than for stocks in which the dealer
only makes a market. This may be explained by dealers providing analyst coverage
in more active stocks and avoiding it in less active stocks. While market shares are
similar for stocks in which the dealers do and do not provide analyst coverage, mean
dollar volume tends to be more than twice as large when the market maker provides
analyst coverage.
The data reported in Table 5 are also compiled separately for institutional brokers,
national retail brokers and regional brokers (not shown). Results were similar for
each category of market maker. In each case, market makers were far more likely to
make markets in stocks their analysts covered. Analysts on average provide more
favorable assessments of the stocks that the firm made markets in. Average dollar
volume is always larger when the dealer provides analyst coverage than when the
dealer only makes a market.
3.4. Underwriting and order flow
Aggarwal and Conroy (2000), Ellis et al. (1999, 2000), and Schultzand Zaman
(1994) find that lead underwriters are almost always market makers in the
immediately following an IPO. In addition, Ellis et al. (2000) show that the lead
underwriter handles a large share of the aftermarket trading volume and that their
market making is usually profitable.
In this section of the paper I ask a related question. Rather than seeing if
underwriters make markets immediately following an IPO, I examine whether
market makers in the five years following an IPO are likely to have been part of the
IPO underwriting syndicate. A fundamental advantage that the underwriter has in
market making is that he will get order flow. Investors who bought shares in the IPO
are the market maker’s brokerage customers and are likely to sell their shares
through him. Also, during the IPO, it is likely that the underwriter will find investors
who will buy in the aftermarket.
I obtain ticker symbols, underwriting syndicate member’s names, and the offering
date for all IPOs from 1990 on from Securities Data Corporation (SDC). For
individual months over the sample time period, I identify all Nasdaq stocks that
went public during the previous five years. Then, for every dealer making a market in
any of the stocks during the month, I count the number of stocks in which the dealer
makes a market and participated in the underwriting and calculate the number of
stocks in which each dealer would be expected to be both a market maker and
underwriter if the activities were independent. This expected number is obtained as
the product of the proportion of stocks in which he makes a market, the proportion
that he helped underwrite and the total number of stocks that had IPOs in the
previous five years.
Table 6 presents results for six months; May 1995, December 1995, June 1996,
December 1996, June 1997 and February 1998 for dealers who made markets in at
least 20 stocks and who participated in at least 20 IPOs in the previous five years.
The number of dealers who meet the criteria increases over the period from 39 to 95.
62
P. Schultz / Journal of Financial Markets 6 (2003) 49–72
This likely reflects the large number of IPOs in the mid-1990s that allowed more
dealers to participate in at least 20 offerings. In every month, every one of the dealers
makes markets in more underwritten stocks than would be expected if the two
activities were independent. In every month except the last, chi-square tests reject
independence of market making and underwriting activities for every single dealer.
The cross-sectional average number of the stocks in which the dealer is both a
market maker and an underwriter shown in the fourth row of the table, ranges from
37 to 62, while the cross-sectional average of the expected number of stocks in which
the dealer is both underwriter and market maker, shown in the fifth row of the table,
ranges from 5.7 to 6.8. The sixth row shows the cross-sectional average of the
dealers’ mean market share for stocks in which they acted as an underwriter. It
ranges from 18.5% to 24.8%. As shown in the following row, the range is from 7.5%
to 9.4% for stocks in which they were not an underwriter. A
t-statistic testing for
differences in the mean market share is highly significant in each month.
To summarize, in this section of the paper, I show that dealers make markets in
the stocks in which they are likely to receive order flow. Dealers with institutional
brokerage clients are likely to receive order flow in high volume stocks, and those are
the types of stocks in which they make markets. Dealers who recommend stocks to
their brokerage clients are likely to receive orders in those stocks, and I find they are
Table 6
Summary statistics on the relation between underwriting and market making
For each month, the sample consists of all stocks that went public in the previous five years. For each
dealer, the expected number of stocks in which he is both underwriter and market maker is obtained by
multiplying the number of stocks in the sample by the proportion of all sample stocks in which the dealer
makes a market and the proportion of all sample stocks for which the dealer was a member of the
underwriting syndicate. Market share is the proportion of share volume in a stock handled by the market
maker.
5
=95 12=95 6=96 12=96 6=97 2=98
Number of dealers who made markets
in at least 20 Stocks and
underwrote 20+ stocks
39 42 46 51 77 95
Number who are both dealer and
underwriter in the same stock more
often than expected
39 42 46 51 77 95
Number of dealers
w2 test rejects
no correlation between market making
and underwriting
39 42 46 51 77 89
Mean number of stocks in which dealer
is also an underwriter
47.0 54.8 59.2 62.3 45.4 37.1
Mean expected number of stocks in which
the dealer is also an underwriter
5.7 6.6 6.7 6.8 5.9 6.7
Mean market share (%) when dealer
was also an underwriter
24.0 24.7 24.1 24.8 22.6 18.5
Mean market share (%) when dealer
was not an underwriter
8.4 9.0 9.4 9.0 8.0 7.5
t
-Statistic for difference in mean market shares 18.07 18.76 22.93 20.90 15.85 12.97
P. Schultz / Journal of Financial Markets 6 (2003) 49–72
63
more likely to make markets in these stocks. Several studies show that investors like
to buy shares in local companies. I find that market makers are likely to make
markets in stocks that are located in the same state as the market maker and its
customers. Finally, if a market maker was part of the underwriting syndicate for an
IPO, it is likely that its customers will be familiar with the stock and will hold shares
in the company. This would in turn suggest that the dealer would receive order flow
in the stock and I find that dealers are more likely to make markets in stocks if they
were part of the underwriting syndicate.
4. Informational advantages and market making
Dealers may also make markets in stocks where they have an informational
advantage, and some of the results of the previous section can be interpreted as
supporting that hypothesis. A preference for local stocks could reflect an ease of
obtaining public or private information about local companies. For example, Hau
(2001) finds that professionals who trade German stocks make more money if they
are located near the corporate headquarters of the stock they are trading. Likewise,
information obtained by analysts may be useful to a market maker as well as to his
brokerage clients. Participation in an underwriting syndicate seems a particularly
important way to acquire information because dealers may learn non-public
information as part of the due-diligence process. In addition, lead underwriters often
continue in an advisory role after the offering and are usually used as underwriters in
follow-on offerings. This continued relation with the company can provide the
market maker with a significant informational advantage over potential rivals.
Another source of informational advantages for market makers may be
information obtained from making markets in similar stocks. Information may
affect values of all stocks in an industry, and market makers who are made aware of
information while trading one stock may use it profitably to trade other stocks in the
industry.
In this section, I examine whether market makers specialize by industry. A stock’s
industry is defined by its two-digit SIC code from the CRSP tapes. To calculate the
number of stocks in an industry in which a dealer would be expected to makes
markets, I first count the number of market makers in each stock and then create
grand totals by summing across stocks in an industry. The grand totals for each
industry are then divided by the total number of market making positions in all
stocks during that month. This gives the percentage of all market making positions
that are in a particular industry. The expected number of stocks in an industry is
calculated for each dealer by multiplying the dealer’s total number of positions by
the percentage of all market making positions in the industry.
For each dealer, I calculate the absolute value of the difference between the
percentage of market making positions in an industry and the expected percentage of
positions if the dealer does not specialize. The expected number is the percentage of
all market making positions by all dealers in the industry times the total number of
markets made by the dealer. I then sum across all industries to get the percentage
64
P. Schultz / Journal of Financial Markets 6 (2003) 49–72
difference for market maker
m: That is, the percentage difference for market maker
m
is
%
Dm ¼X
I
i
¼1
j
oi;m ei;mj; ð2Þ
where
oi;m is the observed percentage of dealer m’s positions that are in industry i; ei;m
is the expected percentage of dealer
m’s positions in industry i and I is the total
number of industries. If a dealer specializes in a particular industry, the observed
percentage of positions in that industry will exceed the expected number. The
observed percentage of positions in other industries will be less than the expected
values and thus the absolute value of the differences will be large and positive. Thus
the more a dealer specializes in stocks in a particular industry, the larger the dealer
percentage difference.
I sum all dealers’ total percentage differences for a month to get the total
percentage difference. That is,
%
Dtotal ¼X
M
m
¼1
%
Dm: ð3Þ
I next simulate
%Dm and %Dtotal assuming that dealers do not specialize. I
randomly assign stocks to each dealer. Each dealer is assigned the number of stocks
he makes a market in that month and each stock is assigned to the number of dealers
who trade it. The stock is assigned a maximum of once to each dealer. I then
calculate the dealer percentage differences for each dealer and the total percentage
difference. This process is repeated 100 times for each of six months: May 1995,
December 1995, June 1996, December 1996, June 1997 and February 1998.
Table 7 reports results. The second column of Panel A reveals that for each
month, the total percentage difference
%Dtotal is larger than every one of the 100 total
differences simulated under the assumption that dealers do not specialize in
industries. This rejects the hypothesis that dealers as a whole do not specialize. The
next column examines percentage differences for individual dealers. For the average
dealer, 74–75% of the simulated percentages are less than the dealer’s percentage
difference. The next column of the table lists the total number of dealers each month
while the last column shows the number of dealers for which the actual percentage
difference is greater than each of the 100 simulated differences. Although the
totals vary slightly from month to month, we can say that there are usually about
500 market makers and about 145 of them have percentage differences between
the expected and observed number of positions in industries that are greater
than every one of the 100 simulated differences. It is hard to know whether the
dealers’ percentage differences reflect widespread specialization by many dealers,
or whether a small number of dealers are so specialized that others avoid the
industries in which they concentrate. However, the large number of dealers with
percentage differences between the expected and observed number of positions in
industries that are greater than every one of the 100 simulated differences suggests
specialization is widespread.
P. Schultz / Journal of Financial Markets 6 (2003) 49–72
65
While different dealers specialize in different industries, specialization in
companies with bank holding and related companies (SIC code 67) seems to be
particularly strong. For example, even though these stocks make up only about 7%
of Nasdaq listings, 109 of Keefe, Bruyette and Woods 205 market making positions
in December 1996 had SIC codes of 67.
4 Similarly, during the same month Sandler,
O’Neill and Partners made a market in 204 stocks and 124 of them had an SIC code
of 67.
In summary, the evidence is strong that dealers tend to concentrate their market
making in particular industries, and that this concentration is far greater than would
be expected by chance. It seems unlikely that the decision to concentrate market
making in one industry is driven by customer order flow. Brokerage clients will want
to hold diversified portfolios. Thus informational advantages that come from
making markets in similar stocks may explain the preference for making markets in
stocks in the same industry.
5. The probability of making a market
Are the effects of industry, location and underwriting on market making activity
really different effects? For example, Hambrecht and Quist underwrites a lot of
Table 7
Tests for market maker specialization by industry
A stock’s industry is given by its two-digit SIC code. For each market maker each month, I calculate the
difference between the actual percentage of market making positions in an industry and the expected
percentage. I then sum across industries. The sum is the dealer’s percentage difference. The sum of all
dealers’ percentage differences is the total difference. The expected number of positions in an industry is
obtained by dividing the total number of positions in the industry by all dealers by the total number of
market making positions. As a test for industry specialization, I assign all market making positions
randomly across dealers. Each dealer is assigned the same number of positions as he actually takes. No
dealer is assumed to make a market in the same stock more than once. The simulated total difference and
the simulated percentage difference for each dealer is then calculated. The simulation is run 100 times.
Date
Percentage of simulated
total differences less
than the actual
Average across dealers of the
percentage of simulated
differences less than the actual
Number of
dealers
Number of dealers with
percentage difference larger
than simulated percentage
difference every time
5
=95 100.0 75.4 482 135
12
=95 100.0 74.2 507 145
6
=96 100.0 74.3 535 142
12
=96 100.0 74.3 532 144
6
=97 100.0 74.4 535 149
2
=98 100.0 74.1 525 152
4
An incident involving the CEO of Keefe, Bruyette and Woods suggests that market makers do acquire
valuable non-public information. In January 2000, James McDermott was arranged on insider trading
charges for passing information on six pending takeovers of small banks to porn actress Marylin Star.
66
P. Schultz / Journal of Financial Markets 6 (2003) 49–72
technology stocks. Is their preference for technology stocks in market making merely
a reflection of their underwriting activity? Many of these technology companies are
based in Northern California where Hambrecht and Quist is located. Can
Hambrecht and Quist’s tendency to make markets in California companies be
explained by their underwriting activities?
To answer these questions, I run a separate logistic regression to determine the
likelihood of making a market in stocks with different characteristics for each
market maker that made markets in at least 20 stocks, for each of the sample
months. The dependent variable in the regressions takes a value of one if the dealer
made a market in the stock that month and zero otherwise. The independent
variables take values of one if the stock is in a specific volume quintile, if the market
maker and stock have headquarters in the same state and if the stock’s IPO took
place in the previous five years and the market maker was a member of the
underwriting syndicate. I also include dummy variables for two digit SIC codes 13,
28, 38, 60, 67 and 73.
5 While individual market makers may specialize in particular
industries, we would not expect that a type of market maker, like a national retail
broker, would specialize in stocks from certain industries. Thus the median
coefficients for industry dummies are not reported in the table.
Results are in Table 8 Panel A presents a summary of the logistic regressions for
market makers with national retail brokerage businesses. For each variable, the
cross-sectional median of the coefficients is reported. In parentheses below the
median coefficients is the Wilcoxin test statistic for the null hypothesis that the
median is zero. Asymptotically, it is normally distributed with a mean of zero and
standard deviation of one. Below the Wilcoxin statistic, in brackets, I report the
number of positive coefficients more than two standard deviations above zero, and
the number of negative coefficients at least two standard errors below zero.
The median coefficients on the high volume quintiles are positive, and the great
majority of the individual stock coefficients are significantly positive. Positive
coefficients in this model indicate an increased likelihood of making a market in a
stock, so the results imply that market makers with national brokerage businesses
are more likely to make markets in high volume stocks than the stocks in the low
volume quintile. The coefficients for other volume quintiles are also positive, but
decline with the volume quintile. This indicates that national retail brokers become
less and less likely to make a market in a stock for stocks with lower and lower
volume.
The following two rows indicate that national retailers are far more likely to make
markets in stocks based in the same state and in stocks they underwrote. It is
noteworthy that these coefficients and the coefficients on the volume quintiles are all
5
I use these six industries because they account for 38.5% of market making positions in December 1995
and 39.7% of market making positions in December 1996. Casual examination of the data suggest that
firms that specialize in particular industries often specialize in one of these six. A difficulty in using all
industries is that the procedure for getting maximum likelihood estimates the logistic regression
coefficients may fail to converge.
P. Schultz / Journal of Financial Markets 6 (2003) 49–72
67
Table 8
Cross-sectional medians of coefficients of logistic regressions of the determinants of market making
The dependent variable takes a value of one if the dealer made a market in a stock and zero otherwise.
Independent variables take on values of one for stocks in specific volume quintiles or industries, or if the
dealer and stock are located in the same state or if the dealer underwrote the stock’s IPO within the
previous five years. Industry dummies are included in the regression (but not shown in the table) for SIC
codes 13, 28, 38, 60 , 67, and 73. The probability of making a market is obtained by dividing the number of
stocks a dealer makes a market in by the total number of stocks on Nasdaq. The probability of making a
market in a large volume quintile underwritten stock, and the probability of making a market in a small
volume stock that the dealer did not underwrite are obtained from the logistic regression coefficients.
Numbers in parentheses under the coefficient medians are the Wilcoxin statistics for a test of a null
hypothesis that the median is zero. Their asymptotic distribution is unit normal. In brackets below the
Wilcoxin statistics the number of coefficients significantly positive at the 5% level, and the number
significantly negative at the 5% level are reported
May Dec. June Dec. June Feb.
1995 1995 1996 1996 1997 1998
Panel A. National retailers
Intercept
4.40 4.61 4.69 4.87 4.84 5.03
(
4.29) ( 4.46) ( 4.54) ( 4.46) ( 4.54) ( 4.37)
[0, 24] [0, 26] [0, 26] [0, 26] [0, 27] [0, 25]
High volume quintile 2.16 2.04 1.78 2.39 2.42 2.39
(3.31) (3.62) (3.80) (3.52) (4.20) (3.13)
[20, 3] [24, 2] [24, 2] [23, 2] [25, 2] [21, 3]
Second volume quintile 1.58 1.54 1.45 1.67 1.77 1.76
(3.69) (3.64) (4.37) (4.10) (3.96) (3.40)
[22, 1] [23, 1] [24, 1] [24, 2] [24, 2] [22, 2]
Third volume quintile 1.14 1.33 1.06 1.32 1.45 1.49
(4.29) (3.87) (3.58) (4.31) (4.40) (4.00)
[21, 0] [23, 0] [21, 0] [24, 1] [25, 1] [23, 1]
Fourth volume quintile 0.44 0.65 0.52 0.77 0.99 1.07
(3.86) (3.52) (3.60) (4.03) (4.23) (3.00)
[16, 1] [23, 2] [20, 1] [22, 1] [24, 1] [20, 2]
Stock and dealer same state 1.62 1.26 1.14 0.80 1.43 1.10
(2.77) (3.62) (4.04) (3.92) (3.63) (2.97)
½
19; 5 [22, 3] [23, 2] [22, 2] [22, 3] [18, 1]
Dealer underwrote IPO 6.52 6.45 6.45 5.43 2.84 1.95
(4.29) (4.46) (4.54) (4.46) (3.41) (3.67)
[14, 0] [15, 0] [16, 0] [23, 0] [24, 0] [23, 0]
R
2 0.1594 0.1604 0.1560 0.1654 0.1362 0.1077
Prob. of making a market 4.7% 4.6% 5.0% 4.9% 4.7% 4.9%
Prob. large underwritten 98.6% 98.5% 99.0% 96.3% 58.4% 37.3%
Prob. small not underwritten 1.2% 1.0% 0.9% 0.8% 0.8% 0.7%
Number of market makers 24 26 27 26 27 25
Panel B. Dealers with institutional brokerage businesses
Intercept
5.15 5.90 5.98 5.96 5.92 5.86
(
3.72) ( 3.72) ( 3.72) ( 3.82) ( 3.92) ( 3.82)
[0, 18] [0, 15] [0, 17] [0, 15] [0, 18] [0, 16]
High volume quintile 3.54 4.01 3.98 4.20 4.35 4.28
(3.72) (3.72) (3.72) (3.82) (3.92) (3.30)
[18, 0] [15, 0] [17, 0] [15, 0] [18, 0] [15, 0]
68
P. Schultz / Journal of Financial Markets 6 (2003) 49–72
Table 8 (
continued)
May Dec. June Dec. June Feb.
1995 1995 1996 1996 1997 1998
Second volume quintile 1.64 2.39 2.45 2.57 2.55 2.49
(3.72) (3.72) (3.72) (3.82) (3.92) (3.82)
[16, 0] [15, 0] [17, 0] [15, 0] [18, 0] [16, 0]
Third volume quintile 1.07 1.48 1.37 1.57 1.83 1.72
(3.55) (3.72) (3.72) (3.82) (3.92) (3.82)
[12, 0] [11, 0] [13, 0] [14, 0] [18, 0] [15, 0]
Fourth volume quintile 0.17 0.47 0.39 0.89 0.75 0.97
(0.72) (1.85) (2.03) (3.06) (2.43) (2.82)
[3, 3] [6, 2] [5, 1] [5, 1] [9, 0] [10, 0]
Stock and dealer same state 0.08 0.11
0.03 0.15 0.84 0.02
(0.20) (1.24) (0.89) (0.28) (0.63) (0.36)
[9, 6] [7, 6] [7, 6] [8, 8] [8, 7] [8, 5]
Dealer underwrote IPO 6.02 6.18 6.00 5.92 3.91 2.41
(3.59) (3.72) (3.72) (3.10) (3.81) (3.03)
[12, 0] [13, 0] [13, 0] [15, 0] [18, 0] [16, 0]
R
2 0.1893 0.1913 0.1977 0.2276 0.1895 0.1690
Prob. of making a market 5.3% 5.1% 4.4% 3.9% 4.3% 5.3%
Prob. large underwritten 98.5% 98.6% 98.9% 98.7% 88.5% 70.3%
Prob. small not underwritten 0.6% 0.3% 0.3% 0.3% 0.3% 0.3%
Number of market makers 18 18 18 19 20 19
Panel C. Wholesale market makers
Intercept
5.16 1.79 2.24 1.75 1.67 2.23
(
2.83) ( 2.98) ( 3.11) ( 3.11) ( 2.82) ( 3.11)
[1, 11] [1, 11] [1, 12] [1, 11] [1, 11] [1, 12]
High volume quintile 3.01 2.16 1.97 1.86 2.14 2.51
(3.04) (2.43) (1.22) (1.22) (2.12) (2.73)
[10, 1] [10, 1] [9, 3] [9, 2] [10, 2] [10, 1]
Second volume quintile 1.28 1.67 1.32 1.52 1.53 1.80
(3.18) (2.59) (2.90) (3.04) (2.12) (2.73)
[10, 0] [10, 1] [11, 1] [10, 1] [11, 1] [11, 1]
Third volume quintile 1.01 1.19 0.95 0.91 0.91 1.21
(1.92) (2.82) (2.69) (2.97) (3.06) (2.42)
[9, 0] [10, 0] [10, 0] [10, 0] [12, 0] [11, 1]
Fourth volume quintile 0.68 0.69 0.69 0.54 0.64 0.64
(1.71) (1.73) (2.76) (2.13) (3.06) (1.79)
[8, 0] [9, 0] [10, 0] [10, 0] [10, 0] [9, 1]
R
2 0.0594 0.0657 0.0509 0.0561 0.0749 0.0630
Prob. of making a market 1.9% 22.9% 11.7% 15.0% 28.2% 10.5%
Prob. large volume 8.9% 26.8% 14.0% 16.4% 30.4% 24.3%
Prob. small volume 0.6% 14.7% 9.6% 14.8% 16.1% 9.8%
Number of market makers 13 12 13 13 12 14
Panel D. Regional dealers
Intercept
5.43 5.73 5.86 5.75 5.90 5.91
(
7.87) ( 8.05) ( 8.46) ( 8.55) ( 9.43) ( 9.35)
[0, 76] [0, 72] [0, 85] [0, 91] [0, 106] [0, 104]
High volume quintile 1.34 1.46 1.17 1.03 1.49 1.32
(4.08) (4.99) (4.48) (4.64) (4.03) (3.95)
[54, 6] [46, 6] [50, 5] [54, 12] [62, 13] [62, 14]
P. Schultz / Journal of Financial Markets 6 (2003) 49–72
69
significant. Location, volume, and underwriting affect the likelihood of making a
market separately, as does the stock’s industry.
The fourth row from the bottom of the table shows the median unconditional
probability of making a market. It ranges across months from 4.7% to 5.0%. In the
row below is the logistic regression’s median predicted probability of making a
market in a stock that the dealer underwrote that is in the high volume quintile. It is
over 98% in May 1995, December 1995 and June 1996. The probability falls
dramatically to 37.3% at the end of the period. A possible explanation for this is that
dealers are more likely to make markets in recent IPOs and the large number of IPOs
over this period meant that by February 1998 many market makers had several IPOs
they had underwritten several years ago. The second row from the bottom reports
the probability of making a market in a low volume stock that the dealer did not
underwrite. It ranges from 0.7% to 1.2%.
Panel B of Table 8 reports results for market makers who are also institutional
brokers. Results are similar to those obtained for national retail brokers. High
volume is a particularly important determinant of market making activity for these
dealers. They are also far more likely to make markets in stocks they underwrote.
Institutional brokers differ from national retail brokers in that location is not an
important determinant of their choices of stocks for market making.
Table 8 (
continued)
May Dec. June Dec. June Feb.
1995 1995 1996 1996 1997 1998
Second volume quintile 1.13 1.46 1.28 0.98 1.36 1.25
(4.37) (5.71) (6.68) (5.76) (5.79) (6.33)
[47, 5] [47, 5] [54, 5] [56, 8] [70, 7] [74, 11]
Third volume quintile 0.70 1.10 1.16 0.85 1.01 1.02
(4.22) (5.87) (6.86) (7.34) (8.40) (6.90)
[40, 4] [42, 4] [52, 5] [55, 5] [65, 5] [62, 7]
Fourth volume quintile 0.32 0.85 0.81 0.59 0.68 0.66
(1.73) (4.55) (6.18) (4.63) (5.89) (5.83)
[28, 2] [35, 4] [46, 5] [37, 3] [50, 5] [40, 6]
Stock and dealer same state 1.57 1.33 1.17 0.92 1.18 1.09
(6.63) (6.02) (5.99) (5.51) (6.94) (7.28)
[64, 1] [65, 1] [67, 3] [67, 3] [89, 4] [86, 2]
Dealer underwrote IPO 23.30 21.07 7.23 6.90 4.56 3.41
(7.50) (7.84) (7.83) (7.88) (7.50) (8.54)
[22, 0] [30, 0] [44, 0] [54, 0] [79, 0] [90, 0]
R
2 0.1295 0.1223 0.1222 0.1338 0.0962 0.0933
Prob. of making a market 1.4% 1.4% 1.2% 1.2% 1.1% 1.2%
Prob. same state 100.0% 100.0% 97.7% 94.8% 54.6% 27.8%
underwritten
Prob. not underwritten 0.4% 0.3% 0.3% 0.3% 0.3% 0.3%
or same state
Number of market makers 82 86 95 97 118 116
70
P. Schultz / Journal of Financial Markets 6 (2003) 49–72
Results for wholesale market makers are presented in Panel C. These market
makers do not underwrite stock offerings, so the IPO variable is omitted, as is the
variable for market maker and stock headquarters in the same state. The coefficients
on volume quintiles indicate that the likelihood that a wholesale market makers will
handle a stock increases with the volume.
Panel D provides results for regional dealers. The median coefficients for the high
volume quintiles, while positive, are much smaller than their counterparts in the
other regressions. Volume is less important in the market making decision of a
regional broker than a wholesaler, institutional broker, or national retailer. Results
indicate that regional brokers are much more likely to make markets in stocks
located in the same state and in stocks they have underwritten.
To summarize, the logistic regressions indicate that trading volume, location and
underwriting are all important factors in the underwriting decision. For national
retail brokers, trading volume, the stock’s location and participation in the
underwriting syndicate are particularly important. For institutional brokers, trading
volume is critical in the market making decision, as is participation in the
underwriting syndicate. For wholesale market makers, volume is the most important
factor. For regional brokers, location and underwriting participation are the key
determinants of market making activity.
6. Summary and conclusions
In this paper, I posit two explanations for Nasdaq dealers’ choices of stocks in
which to make markets. The first is that Nasdaq dealers’ choices of stocks are
dictated by the order flow from their brokerage business or from the businesses of
the brokers who sell them order flow. The evidence presented here is consistent with
this explanation. Institutional brokers make markets in the high volume stocks that
their customers trade while small retail brokers make markets in less active stocks.
Regional brokers make markets in the local stocks that their customers own in
disproportionate amounts. Dealers are more likely to make a market in a stock if
they have sold shares to their customers in an IPO. Finally, if a dealer provides
analyst coverage of a stock for its customers, it will be much more likely to make a
market in the stock.
A second explanation for Nasdaq dealers’ choices of stocks in which to make
markets is that they pick stocks in which they have an informational advantage. This
hypothesis does not explain why market makers with institutional clients prefer to
deal in more active stocks. However, it is consistent with dealers making markets in
local stocks, stocks they have underwritten, and stocks that their analysts cover. In
addition, it may explain why market makers often specialize in stocks in certain
industries or avoid stocks in others. Of course, these hypotheses are not mutually
exclusive. It is possible that both sources of order flow and informational advantages
play a role in determining the stocks that a dealer trades.
These results should serve to caution researchers against examining Nasdaq
market making in isolation rather than as part of an integrated business. Entry and
P. Schultz / Journal of Financial Markets 6 (2003) 49–72
71
exit decisions for market making in individual stocks should be considered in the
context of the sources of the market makers’ order flow. Researchers who want to
estimate the profitability of making a market in a stock should consider the costs of
obtaining order flow in the stock and the value of any information obtained from the
market making that will make trading other stocks more profitable.
An interesting question for further research is why market making is typically
bundled with brokerage, analyst coverage and underwriting in the same firm. Why
are not these businesses separable? One possibility is that the information generated
in one of these activities is valuable in the others, and that the information is not
easily sold by, for example, a broker to a market maker. The organization of Nasdaq
and other securities markets is clearly an area that calls for additional work.
References
Aggarwal, R., Conroy, P., 2000. Price discovery in initial public offerings and the role of the lead
underwriter. Journal of Finance 55, 2903–2922.
Coval, J., Moskowitz, T., 1999. Home bias at home: local equity preference in domestic portfolios. Journal
of Finance 54, 2045–2074.
Dutta, P.K., Madhavan, A., 1997. Competition and collusion in dealer markets. Journal of Finance 52,
245–276.
Ellis, K., Michaely, R., O’Hara, M., 1999. The making of a dealer market: from entry to equilibrium in the
trading of Nasdaq stocks. Working paper, Cornell University.
Ellis, K., Michaely, R., O’Hara, M., 2000. When the underwriter is the market maker: an examination of
trading in the IPO aftermarket. Journal of Finance 55, 1039–1074.
Hau, H., 2001. Location matters: an examination of trading profits. Journal of Finance 56, 1959–1983.
Ho, T.S.Y., Macris, R.G., 1985. Dealer market structure and performance. In: Amihud, Y., Ho, T.S.Y.,
Schwartz, R. (Eds.), Market Making and the Changing Structure of the Securities Industry. Lexington
Books, Lexington, MA.
Huberman, G., 1998. Familiarity breeds investment. Working paper, Columbia University.
Kremer, I., Polkovnichenko, V., 2000. Competition in financial dealership markets, Working paper,
University of Minnesota.
Schultz, P., Zaman, M., 1994. Aftermarket support and underpricing of initial public offerings. Journal of
Financial Economics 35, 199–219.
Wahal, S., 1997. Entry, exit, market makers, and the bid–ask spread. Review of Financial Studies 10,
871–901.
72
P. Schultz / Journal of Financial Markets 6 (2003) 49–72
No comments:
Post a Comment