Are Trading Commissions a Factor in IPO Allocation?
M. Nimalendran, Jay R. Ritter and Donghang Zhang∗
October 20, 2004
Nimalendran is from the Department of Finance, University of Florida, Gainesville, FL 32611-7168 and the
U.S. Securities and Exchange Commission, Ritter is from the University of Florida, and Zhang is from the
Moore School of Business, University of South Carolina, Columbia, SC 29208. Nimalendran can be reached at
(352) 392-9526 or firstname.lastname@example.org. Ritter can be reached at (352) 846-2837 or email@example.com. Zhang
can be reached at (803) 777-0242 or firstname.lastname@example.org. Zhang acknowledges financial support from the
University of South Carolina Research and Productive Scholarship Fund. We thank Shingo Goto, Paul Irvine,
Greg Niehaus, Eric Powers, Sergey Tsyplakov and seminar participants at the University of South Carolina for
comments. The Securities and Exchange Commission disclaims any responsibility for any private publication or
statement of any SEC employee or commissioner. This article expresses the author’s views and does not
necessarily reflect those of the Commission, the Commissioners, or other members of the staff.
Are Trading Commissions a Factor in IPO Allocation?
Underwriters using bookbuilding have discretionary power for allocating shares of initial
public offerings (IPOs). Commission revenue paid to underwriters by investors is one of the
determinants of IPO allocations. We test the hypothesis that investors trade liquid stocks in
order to affect their IPO allocations. Consistent with this hypothesis, we find that money left
on the table by IPOs has a significant impact on the contemporaneous trading volume of these
liquid stocks during 1993-2001. For an IPO that leaves $1 billion on the table, in the week
before trading commences there is abnormal volume in the 50 most liquid stocks of 2.7-4.0%.
Underwriters have discretion in allocating shares of initial public offerings (IPOs)
when bookbuilding is used. There are three main theories describing the allocation of shares
by investment bankers, which might be titled 1) the academic view, 2) the pitchbook view,
and 3) the agency view. The academic view, exposited by Benveniste and Spindt (1989),
argues that IPO allocations are the solution to a mechanism design problem, in which regular
(i.e., institutional) investors must be given inducements to honestly reveal their private
information about the valuation of a firm going public. The pitchbook view, exposited by
investment bankers in their presentations to prospective issuing firms, emphasizes that shares
will be allocated to buy-and-hold investors, as proxied by institutions who already are large
holders of similar firms. The agency view, exposited by Loughran and Ritter (2002, 2004),
argues that hot IPOs are allocated to investors who direct commission business to the
investment banking firm in return. This paper tests an implication of the agency view of IPO
From 1993 to 2001, firms going public left more than $93.5 billion on the table, where
the amount of money left on the table is defined as the product of the number of shares
offered and the difference between the first-day close price and the offer price. During 1999
and 2000, the Internet bubble period, there were 803 IPOs, and the total amount of money left
on the table was $63.5 billion. During this period, an average company would be underpriced
by 65.0%, and leave $79 million on the table. Who received the $93.5 billion?
Evidence from regulatory settlements indicates that investment banks have allocated
IPOs partly on the basis of trading commissions. For example, the January 22, 2002 Securities
and Exchange Commission (SEC) settlement with Credit Suisse First Boston (CSFB) states
that CSFB allocated hot IPOs to some clients and in return received commissions of up to
65% of the profits (money left on the table). The clients kicked back part of their profits by
paying unusually high commissions on stock trading that in some cases had no other
economic purpose. For instance, the usual commission for institutional investors is at most 6¢
per share, but CSFB’s clients paid as high as $3.00 per share in commissions for block trades
executed by CSFB brokers. CSFB paid a fine of $100 million and settled the case with the
SEC (without admitting or denying wrongdoing). On January 9, 2003, the former Robertson
Stephens securities unit of FleetBoston Financial Corp. also settled similar accusations with
the SEC and the National Association of Securities Dealers (NASD). Robertson Stephens Inc.
paid a fine of $28 million.1
SEC Commissioner Paul Atkins has publicly stated that it is permissible for IPO
allocations to be based on customer “relationships.”2 If the linkage between allocations of
underpriced IPOs and commission business is too direct, however, then it amounts to profit
sharing. This is against security laws and regulations, such as NASD Rule 2330, which
prohibits a NASD member firm from sharing, directly or indirectly, in the profits or losses in
any account of a customer. It is unclear, however, whether the regulatory cases against some
investment banks are exceptions, or whether the trading of hot IPOs for commission business
was pervasive. Our paper addresses this issue.
All IPO allocation data are confidential, and are carefully guarded by investment
banks. The biggest obstacle for any research on this topic is the availability of detailed IPO
allocation and commission data. In this study, we use an innovative empirical design to solve
this issue. As indicated by the CSFB and Robertson Stephens cases, investment banks in at
least some instances link IPO allocations to the commissions generated by clients. If an
investor wants to get a large share of a hot IPO allocation, it needs to generate enough
commissions. Although commissions generated over a period will be taken into account,
commissions generated more recently for the investment bank generally carry more weight in
See SEC press release 2002-14, and the SEC complaint against CSFB for evidence on CSFB IPO allocations.
See SEC press release 2003-3, the SEC complaint against Robertson Stephens, and the Letter of Acceptance,
Waiver and Consent (AWC) No. CAF030001 submitted to the NASD by Robertson Stephens for details on the
Robertson Stephens case.
As quoted in the Bloomberg News article of October 13, 2004 by Judy Mathewson, “U.S. SEC Proposes New
Initial Public Offering Rules.”
the investment bank’s ranking of clients.
To boost its commission ranking, an investor may engage in stock churning. This is
non-economic trading with commission generation as the only purpose. Other trading an
investor may engage in is subtler. For example, an investor could engage in excessive
portfolio rebalancing and time the trading around IPO dates. An investor could also move
routine trading that is not timing-sensitive to around IPO dates (by routine trading we mean
economic trading that is incurred for non-IPO-related reasons).3 For expositional
convenience, we use the term “IPO-related trading” for all trading with a whole or partial
purpose of generating commissions to affect IPO allocations.
If the purpose of IPO-related trading is to generate commissions which transfer profits
from IPO recipients back to the investment banking firm, there is an incentive to structure the
trades in a manner that minimizes the “leakage” to other market participants due to bid-ask
spreads, price impact, etc. A variety of trading strategies could be employed to minimize this
leakage. One mechanism would be to simultaneously submit buy and sell orders for the
identical block of stock with two different securities firms. Another strategy would be to trade
fewer shares, but to pay an abnormally high commission per share. For almost all conceivable
strategies, trading highly liquid stocks would be preferred, both to minimize bid-ask spreads
and price impact costs, and to avoid the attention that would come from trading a large block
of a less liquid stock. Thus, we use the trading volume of the 50 most actively traded stocks to
capture IPO-related trading.
Furthermore, if commission generation is an important factor in IPO allocation, there
should be a positive relation between the amount of money left on the table by IPOs and the
trading volume of liquid stocks around the IPO dates. Consequently, we can use publicly
In their study of institutional trading patterns, Goldstein, Irvine, Kandel, and Wiener (2004) provide evidence
showing that smaller institutional investors often trade unnecessarily more and pay higher commissions. They
interpret the evidence as being consistent with the fact that these institutional investors pay to get better access to
brokerage services, including allocations of hot IPOs. This supports our argument of using “IPO-related trading”
to capture commission generation.
available data to examine this issue.
We choose the top 50 stocks for each trading day based on the rank of a stock’s
average trading volume for the past 20 trading days. We exclude stocks with high volatility or
a price of below five dollars. We have 3,499 IPOs during our sample period from 1993 to
2001. We aggregate the money left on the table by IPOs by offer dates, and obtain a time
series of the daily amount of money left on the table. With controls for market movements, a
time trend, and calendar-related patterns in trading volume, we employ an autoregressive
model with four lags to study the relationship between the trading volume of the 50 liquid
stocks and the amount of money left on the table by IPOs.
Because of changes in IPO practices, we partition our sample period into three
sub-periods: the pre-Internet bubble period (1993 – 1998), the Internet bubble period (1999 –
2000), and the post-Internet bubble period (2001). For all three subperiods, each $1 billion left
on the table during the six trading days beginning on day t generates abnormal volume in the
50 liquid stocks of between 2.7% and 4.0% on day t, although only during the Internet bubble
period is this point estimate reliably different from zero.
The commissions generated from the increase in the trading volume are economically
important. During the internet bubble period, for a six-day window with only average IPO
activities, our point estimate suggests that IPO-related trading would cause an increase of 2%
in the trading volume of the 50 liquid stocks, and would result in an additional $656,410 per
day in commissions if we assume a 10¢ per share commission, based on the average daily
trading volume of the 50 most liquid stocks during this sub-period.
Our estimate is conservative. For the Internet bubble period, the $656,410 in
commissions only represents a 0.09% payback ratio of the money left on the table during the
following six trading days. This is much lower than the 30% or even 65% kickbacks revealed
in the CSFB regulatory case. Our estimates are also much lower than the 10% increase in
volume conjectured by Welch and Ritter (2002). There are a few reasons to believe that our
analysis only gives a lower bound estimate of the effect of underpriced IPOs on trading
volume. Investors may simply direct trading to a firm with IPOs to distribute rather than
another firm (such as an ECN or crossing network), with no incremental trading occurring, or
the same trade could be done at higher commissions. Both of these mechanisms suggest that
our measure is a lower bound estimate of the effect of the quid pro quo between commission
generation and IPO allocation. Furthermore, investors may in some cases trade less liquid
stocks, but submit buy and sell orders simultaneously with different brokers to mitigate
market impact. This also makes our measure only a lower bound estimate of the effect.
Our findings complement those of Reuter (2003), who reports that there is a positive
relation between the commissions paid to a given lead underwriter and the mutual fund’s
holdings of recent IPOs from that underwriter. Our empirical evidence suggests that there is
not only a relation between long-term commission business and IPO allocations (Reuter’s
finding), but also a short-run relation between the aggregate amount of money left on the table
and aggregate trading volume.
The rest of the paper is organized as follows. In the next section we develop our
hypothesis. In section II we describe the data and report summary statistics. In section III we
develop and analyze an empirical model relating money left on the table by IPOs to volume.
Section IV concludes.
I. The Hypothesis
Bookbuilding has been the dominant form of selling IPOs in the U.S. For
bookbuilding, the bookrunner is the lead underwriter if there is a sole lead (for some IPOs
there could be more than one lead underwriter), and has discretion on how the IPO shares are
allocated. We will use bookrunner and lead underwriter interchangeably. Boehmer, Boehmer,
and Fishe (2002) report that the bookrunner allocates about 75% of the shares in an average
IPO, with other syndicate members (mainly the co-managers) allocating the rest. Many
factors affect how the lead underwriter exercises its discretionary power.
Theoretical and empirical evidence suggests that the lead underwriter can use its
discretionary power to acquire information from investors and to meanwhile minimize the
costs of such information acquisition through the bundling of IPOs (Benveniste and Spindt
(1989), Sherman (2000), Sherman and Titman (2002), Hanley and Wilhelm (1995), Cornelli
and Goldreich (2001), Aggarwal, Prabhala and Puri (2002), and Ljungqvist and Wilhelm
(2002)). The lead underwriter and the issuing firm can use the discretionary share allocation
to achieve corporate governance goals (Brennan and Franks (1997), Stoughton and Zechner
(1998), and Field and Karpoff (2002)), or to improve aftermarket liquidity (Booth and Chua
(1996)). The lead underwriter can also adjust the number of shares allocated to increase the
IPO valuation and support the aftermarket price (Aggarwal (2000), Fishe (2002), and Zhang
However, the lead underwriter may not always use its discretion to act in the best
interest of the issuing firm. Loughran and Ritter (2002, 2004) suggest that the lead
underwriter may underprice the issue and allocate underpriced shares to favored clients for
quid pro quos. Ritter and Welch (2002) argue that such agency issues may play a large role in
explaining IPO underpricing, especially during the Internet bubble period (1999 – 2000).
Recent evidence revealed through investigations by securities regulators indicates that
allocating IPOs in return for commission revenue was an important business practice of at
least some underwriters during the Internet bubble period.
The lead underwriter may allocate hot IPOs to favored clients and ask these clients to
directly kick back some of their profits through trading commissions in explicit profit-sharing
agreements. This violates NASD Rule 2330f 1(a), which states “...no member or person
associated with a member should share directly or indirectly in the profits or losses in any
account of a customer...” Alternatively, investors may merely infer that if they generate
commissions, they will receive IPO allocations. Recent investigations reveal the existence of
IPO-related stock churning. The aforementioned CSFB and Robertson Stephens settlements
are examples. Most importantly, our discussions with both underwriters and institutional
investors have found unanimous agreement that commission business was a very important
determinant of IPO allocations, especially during the bubble period.
IPO-related stock churning was not necessarily limited to clients of CSFB and
Robertson Stephens. Other big investment banks may have also been involved in similar
practices.4 However, stock churning is not the only way to inflate commissions. In its Letter
of Acceptance, Waiver and Consent (NASD No. CAF030001, AWC hereafter), Robertson
Stephens Inc. discloses that a “Syndicate Rank” based on commissions paid by clients is used
to allocate IPO shares. “The Syndicate Department generally allocated shares by using a
formula that was weighted over the course of 18 months in favor of those accounts that
generated commissions closer in time to the IPO. This formula was used to calculate a
customer’s ‘Syndicate Rank’. …The Syndicate Department also had discretion to allocate
some IPO shares independent of the Syndicate Rank on a case-by-case basis.” (AWC, page 5)
Robertson Stephens Inc. is apparently not the only investment bank that used the
syndicate rank or a similar measure in allocating IPOs. Reuter (2003) reports that for IPOs
underwritten by different investment banks from 1996 to 1999, a fund manager’s IPO
allocation is positively related to its commission payments to the lead underwriter of the IPO.
IPO-related trading is the focus of this paper. The reason for IPO-related trading is to
generate commissions to increase allocations of hot IPOs. It may not be a phenomenon only
during the Internet bubble period that investment banks used commission revenue to
determine IPO allocations.5 However, as suggested by Wall Street traders, the game was
different during the Internet bubble period.6 Because of the large amounts of money left on
See, for example, “SEC intensifies inquiry into commissions for hot IPOs: Goldman, Bear Stearns and Morgan
Stanley get requests for data” by Susan Pulliam, Randall Smith and Charles Gasparino in the Wall Street
Journal, December 13, 2000.
See, for example, Aaron Lucchetti, “SEC probes rates funds pay for commissions,” Wall Street Journal,
September 16, 1999, p. C1.
See “U.S. probes inflated commissions for hot IPOs” by Randall Smith and Susan Pulliam in the Wall Street
Journal, December 7, 2000.
the table, investors were willing to directly pay back part of the profits through inflated
commissions around the IPO dates, and they also engaged in round-trip trades just to generate
commissions. The more active role of hedge funds in later years contributed to this trend,
because hedge funds are lightly regulated and have no need to explain the payment of inflated
IPO-related trading includes two different types of trading that affect aggregate
volume near the time of an IPO: stock churning and the timing of excessive portfolio
rebalancing and regular trading. Stock churning is closely associated with large liquid stocks.
Investors who engage in the churning would want to reduce price risk and other costs other
than trading commissions. It would also be in the best interest of all parties that such trading
does not catch attention. Even if there are simultaneous sales and purchases of a stock, traders
presumably would want to avoid less liquid stocks where this might be noticed. For these
reasons, large liquid stocks are obviously preferred. On the other hand, the timing of regular
trading or excessive portfolio rebalancing is not necessarily limited to large liquid stocks.
However, if such trading is at least partially to boost commission generation, it is more likely
to be associated with large liquid stocks because large liquid stocks are more likely to be
associated with big volume transactions.
So, we hypothesize that the trading volume of large liquid stocks is linked to the
money left on the table by IPOs. If commission generation is a factor in IPO allocation, we
will observe a positive effect of money left on the table on trading volumes around IPO dates.
II. Data, the Metric for IPO-Related Trading, and Summary Statistics
We use the Securities Data Company’s (SDC) new issues database to identify IPOs
from 1993 to 2001. All unit offerings, American Depository Receipts (ADRs), Real Estate
Investment Trusts (REITs), and closed-end funds are excluded. We also exclude banks and
savings and loans (SIC 6020 – 6120 and 6712) and all IPOs with an offer price of less than $5.
We use the Center for Research in Security Prices (CRSP) database to check the IPO offer
date and first-day closing price. We exclude all IPOs that are not included in the CRSP
database. The IPO offer dates reported in the SDC database are often one-day earlier than the
actual trading date (the pricing date usually is the day before the actual trading date but after
the market close). We use the date when the IPO first appears in the CRSP database as the
offer date when there is only a one-day difference between the SDC database and the CRSP
database. When the difference is more than one day apart, the New York Stock Exchange’s
Trade and Quote (TAQ) database and Yahoo! are used to verify the dates.
The transaction data used to calculate trading volume come from the TAQ database.
For each stock, transactions that were executed after the market close and/or on regional
exchanges are excluded because the liquidity conditions are usually worse in aftermarket
trading and/or on regional exchanges.
B. The Metric for Capturing IPO-Related Trading
IPO-related trading is likely to be associated with large liquid stocks. We use the
trading volume of the top 50 liquid stocks to capture IPO-related trading. This metric is
constructed as follows. First, for each trading day, we rank all stocks based on the average of
the past twenty days’ intra-day quote-to-quote return standard deviation, with the lowest
volatility ranked as one. Second, we exclude stocks with a price below $5 or with a volatility
rank higher than 2500.7 Stocks with a low price or high volatility are unlikely to be good
candidates for IPO-related trading because of their high risk and limits on commissions paid
on low priced stocks. Third, we calculate the share volume for each remaining stock, during
which transactions that were executed after the market close or on regional exchanges are
excluded. For stocks listed on Nasdaq, we divide the volume by two to reflect the different
conventions of reporting volume on Nasdaq versus the American and New York Stock
Exchanges. We then calculate the mean and standard deviation of the daily volume for the
We exclude roughly two thirds of the 7000 plus stocks traded every trading day by applying the volatility filter.
past twenty trading days for each stock. For any trading day, if the difference between a
stock’s current trading volume and the past twenty-day’s average is more than four times
greater than its past twenty-day standard deviation of volume, the stock is excluded. This
excludes about 3% of the sample. The reason for doing so is because IPO-related trading is
unlikely to increase a liquid stock’s daily volume by this magnitude.
Some non-IPO related reasons, such as stock splits or significant news, may cause
such dramatic increases in daily trading volumes. To minimize the noise in our metric, we
exclude those stocks for those specific days. Finally, for each trading day (Day t), all
remaining stocks are ranked based on the past twenty-day’s average daily volume. The 50
stocks with the highest volume are identified. We use past volumes, instead of current
volumes, to rank stocks to avoid any potential look-ahead bias. The total trading volume of
these 50 ranked stocks, denoted as TVOL50, is then used as the metric to capture IPO-related
trading on Day t.
Although IPO-related trading is more likely to be associated with large liquid stocks,
there are no specific rules from which we can tell what stocks investors would exactly choose
to trade. It is plausible that stocks ranked among the top 200 or 500 could be good candidates,
especially when fund managers merely try to time trading motivated by portfolio rebalancing
to around IPO dates. Each top stock ranked by trading volume would capture the IPO-related
trading with a positive probability. But it is plausible that the higher the rank, the higher the
probability. Meanwhile, stock trading volumes are very volatile, and the daily volume of any
given stock has a lot of noise. So, if we view the IPO-related trading as a signal we want to
capture, we need to include a certain number of stocks in our measure in order to increase the
signal-to-noise ratio. The addition of each stock along the rank of trading volume would
potentially capture more of the IPO-related trading, but at a decreasing rate. There are no clear
rules on which we can determine the maximum signal-to-noise ratio. We use TVOL50 to try
to achieve the close-to-optimum signal-to-noise ratio, but it is relatively arbitrary to just use
the top 50 stocks.8
C. Summary Statistics
We partition our sample period into three sub-periods: the pre-Internet bubble period
(1993 – 1998), the Internet bubble period (1999 – 2000) and the post-Internet bubble period
(2001). We report the summary statistics of daily trading volumes of all stocks and the 50
most liquid stocks in Table 1. For the three sub-periods, the number of stocks traded per day
does not change much – 7,390 stocks for the pre-Internet bubble period, 7,745 stocks for the
Internet bubble period and 7,126 stocks for the post-Internet bubble period. However, trading
volume has increased dramatically, partly due to stock splits and partly due to higher turnover
ratios. For example, the mean daily total trading volumes of all stocks for the three
sub-periods are, respectively, 615, 1,496, and 1,993 million shares. For the top 50 stocks
(TVOL50), the mean daily total trading volumes are 102, 322, and 531 million shares,
respectively. Note that the top 50 stocks are selected based on the past twenty-day’s trading
volume, and they do not necessarily capture the most heavily traded 50 stocks of the current
The summary statistics for IPOs are reported in Table 2. We have 3,499 IPOs in the
nine-year sample period, of which 2,620 IPOs went public during the pre-Internet bubble
period, 803 IPOs went public during the Internet bubble period, and only 76 IPOs went public
during 2001. The summary statistics about the whole sample and for different sub-periods
reported in Panel A of Table 2 are consistent with what has been reported in the literature. We
also report the summary daily statistics of our IPO sample in Panel B of Table 2. These
statistics indicate that IPOs are clustered. Except for the post-Internet bubble period, we have
IPOs on more than half of the trading days, and have more than 1.5 IPOs per day. But IPO
activity also varies from day to day. For example, even for the Internet bubble period during
To make sure better metrics are not excluded, we also repeated our analysis using the top 30 stocks, or ranking
stocks based on both volume and the bid-ask spread. The results (not reported) are similar but weaker.
which IPOs are highly clustered, the standard deviation of the amount of money left on the
table by IPOs is $285 million per day, while the mean is only $126 million per day. Note that
all of the money left on the table figures assume no existence of overallotment options, and
the number of shares offered is measured as the domestic tranche.
Another noticeable feature about IPOs is the day of the week pattern. The lead
underwriter and the issuing firm usually finalize the offer price and allocate shares to
investors the day before the trading of the new issue starts. To avoid weekend uncertainties,
firms rarely go public on Mondays. For the other four weekdays, slightly more firms start
trading on Thursday and Friday (Figure 1.1). This is a departure from practice in the 1980s,
when most IPOs started to trade on Tuesday through Thursday. As shown in Figure 1.2 the
first-day returns also demonstrate a day of the week pattern. However, in Figure 1.2 the
first-day returns on Monday are dominated by a few outliers because of the small number of
IPOs that went public on Monday.
III. The Effect of Money Left on the Table on Trading Volume
A. The Model
The summary statistics in Table 1 indicate that stock-trading volume is very volatile,
and the metric that we use to capture IPO-related trading contains a lot of noise. Therefore, it
is important to control for the impact on trading volume of factors that are unrelated to IPO
activity. The literature on trading volume has focused on the contemporaneous relations
between volume and price movements (see e.g. Karpoff (1987), Campbell, Grossman, and
Wang (1992), Gallent, Rossi, and Tauchen (1992), Blume, Easley, and O’Hara (1994),
Hiemstra and Jones (1994), Jones, Kaul, and Lipson (1994), He and Wang (1995), Andersen
(1996), and Llorente, Michaely, Saar, and Wang (2001)). The time series of trading volumes
used in this paper is a stationary process after de-trending, which is consistent with the
literature (Gallant, Rossi, and Tauchen (1992), Andersen (1996) and Llorente, Michaely, Saar,
and Wang (2001)). We use Date _ Indext to de-trend our time series of TVOL50, where
Date _ Indext is the sequential number of each trading day, t , divided by 2268, the total
number of trading days in our sample period. (For the rest of this paper, we use subscript t
to indicate the value of a variable on day t .) Thus, Date _ Indext starts at a value of 0 in
January 1993 and ends at a value of 1 in December 2001.
The time series of TVOL50 also demonstrates a correlation with general market
conditions in addition to the time trend. We use S & P500 t , the nominal level of the S&P 500
index, as a control for it. The literature and the analysis of our data also suggest that trading
volumes demonstrate calendar-related patterns. IPO activities, as indicated in Figure 1, also
show a weekly pattern. We use four weekday dummies (Tuesday through Friday) and eleven
month dummies (January through November) as controls for calendar-related patterns. The
volume literature suggests a strong relationship between trading volume and price volatility.
IPOs on the first day also have asymmetric betas in up and down markets (e.g., Loughran and
Ritter (2002)). We use the S&P 500 Index return and its absolute value, Market t and
Market t , to control for market-wide price movements.
In sum, we use the following autoregressive model:
Volumet = α + η1 * Date _ Indext + η 2 * S & P500 t + λi * Weekday i
+ mi * Monthi + β 1 * Market t + β 2 * Market t + δ i * Volumet −i (1)
i =1 i =1
+ γ i * Money t +i + ε t
i = −1
In the model, the dependent variable, Volumet , is the natural log of TVOL50 for Day t
multiplied by 100. This enables us to interpret any change of Volumet in percentage terms.
Four lagged variables of Volumet are used to remove the autoregressive part of the TVOL50
metric.9 We use the daily amount of money left on the table by IPOs for the day before the
current day ( Moneyt −1 ), the current day ( Moneyt ), and the next five days ( Moneyt +1 through
Moneyt + 5 ) to test our hypothesis that trading volume increases to affect hot IPO allocations. If
our hypothesis is correct, we would expect that the coefficients, γ −1 through γ 5 , are positive
and significant. We include one lagged money left on the table variable. Regulatory
investigations indicate that sales people at CSFB asked for kickbacks one day after some hot
IPOs started trading. We include the money left on the table on Day t − 1 to test whether it
was a common practice for there to be ex post profit-sharing after the IPO. The use of money
on the table on Day t + 1 to Day t + 5 is based on the assumption that institutional
investors are able to forecast which IPOs will be hot deals several days in advance. The
“partial adjustment” literature as well as discussions with industry participants suggests that
this is a reasonable assumption.
B. Regression Results and Analysis
We run regressions as specified in equation (1) for the three sub-periods. The
regression results are reported in Panel A of Table 3. The coefficients for the control variables
are consistent with our expectations. The positive coefficients for Date _ Indext during the
pre-and the Internet bubble periods capture the increase in trading volume over time.
S & P500 t also helps to capture the time variation in trading volume, and the negative
coefficient of this variable for the Internet bubble period reflects the downward trend in the
latter part of 2000. The time variation in volume in 2001 is captured by Date _ Indext ,
S & P500 t and the eleven month dummies. Volume increases when the market is volatile,
and this is captured by the significant coefficient for Market t in all three sub-periods. For
This is the same approach that Naranjo and Nimalendran (2000) use to measure the unexpected trading volume
in the U.S. dollar – German Deutsche Mark foreign exchange market.
example, the coefficients on Market t and Market t for the pre-Internet bubble imply that a
1% daily market increase was associated with a 7.83% increase in volume relative to volume
in a flat market. The coefficient for the variable Market t is negative during the pre- and the
Internet bubble periods, reflecting the asymmetric relationship between volume and price
movements in up and down markets. The first two lags of the past volumes capture most of
the autocorrelation in trading volume, which is consistent with expectations.
The money left on the table on the day before the current day has no significant impact
on the current day volume for all three sub-periods. This indicates that the ex post profit
sharing setup after the IPO offer day, although used by some CSFB sales people, was not a
widespread practice. The money left on the table variables for the rest of the six days from t
to t + 5 show generally positive and sometimes significant coefficients during the pre- and
the Internet bubble periods. For the pre-Internet bubble period, only the coefficient for
Moneyt + 2 is significant. For the Internet bubble period, the coefficients for Moneyt ,
Moneyt +1 and Moneyt +5 are statistically significant at the 1% or 5% levels. This suggests
that hot IPOs on the current day or a few days down the road are associated with higher
current day trading volume if they leave money on the table. This is consistent with our
hypothesis. None of the coefficients during the post-Internet bubble period are significant,
reflecting changes in the regulatory environment and relatively little money being left on the
table by relatively few IPOs.
Besides statistical significance, the coefficients for the money left on the table
variables also suggest that IPO-related trading captured by liquid stocks is economically
important. For example, for the pre-Internet bubble period, the coefficient for Moneyt + 2
indicates that, if IPOs are scheduled to go public on Day t + 2 and these IPOs are average
IPOs (which would leave $28.41 million on the table), the current day trading volume of
those 50 liquid stocks would be increased by 0.68 percent (2.38*28.41/100, since we measure
the money left on the table in $100 millions in the regressions). This would translate into
690,737 shares (0.68%*101,579,000), and $69,074 in trading commissions (based on an
average 10¢ per share commission). For the Internet bubble period, the numbers are more
striking. If there are IPOs that start trading on the current day and they are only average IPOs,
the current day volume of those 50 liquid stocks would be increased by 1.58%, or 5.1 million
shares, resulting in $506,868 of additional commissions.
The coefficients for the money left on the table variables during the Internet bubble
period suggest that IPOs do not only affect the trading volume on the offer dates. However, it
would exaggerate the impacts if we simply add up the implied additional commissions for the
six-day window of [t , t + 5] because we do not have IPOs on every day (we do not include
the money left on the table on day t − 1 because the ex post profit sharing setup is apparently
not a common practice). To get a more realistic picture and still stay conservative, we
aggregate the money left on the table by IPOs for a rolling six-day window of [t , t + 5] , and
re-run the regressions as in equation (1) by replacing the six money variables with the
aggregated variable. The regression results are reported in Panel B of Table 3. The coefficients
for the control variables are similar to those in Panel A, and are not reported. The coefficient
for the aggregated money left on the table variable is positive for all three subperiods but is
only statistically significant for the Internet bubble period. For this sub-period, this aggregated
variable has a t-statistic of greater than 4. This is not surprising in the sense that the link
between IPO allocation and commission generation was more dramatic during the bubble
period. And it is consistent with our hypothesis.
We focus on the Internet bubble period for the assessment of economic importance.
Assuming that new issues appear with an equal pace and all days that have IPOs are average
days using the numbers reported in Table 2, we would have 278 (the number of days that have
IPOs) / 504 (the total number of trading days) ∗ $228 million (the conditional mean amount of
money left on the table) ∗ 6 (the number of days in the window period) = $756 million left on
the table for the six-day window during the Internet bubble period. This indicates that
IPO-related trading could cause a 0.27 ∗ 756 / 100 (again in the regressions the money left on
the table is measured in $100 millions) = 2.04 percent increase in the current day trading
volume. On an average trading day, this would translate into 6.6 million shares, and $656,410
additional commissions if the average per share commission is 10¢. Another exercise would
also help illustrate this. Assuming that only one day has IPOs during the six-day window and
that day happens to be the biggest day with $2,609 million left on the table, the regression
coefficient would suggest an increase of 7 percent, or 23 million shares, in the current day
trading volume of those 50 liquid stocks. On a 10¢ per share commission, this would mean
$2.3 million in additional commissions.
The above calculations are conservative. Our metric does not capture all IPO-related
trading, IPOs are often clustered, and the aggregated amount of money left on the table during
a six-day window does not only affect the current day trading volume. But we believe that this
is sufficient to show that these numbers are economically not trivial. We want to point out that
these numbers are lower bound estimates. For the Internet bubble period, it is only a 0.09%
payback of money left on the table – we divide the commissions generated on a 10¢ per share
basis by the estimated amount of money left on the table. This is far lower than the 30% or
even 65% payback ratios as revealed in the CSFB regulatory settlement.
Discussions with practitioners and the regulatory settlement suggest that only hedge
funds, which were allocated approximately 7% of the shares in several of the IPOs featured in
regulatory settlements, regularly paid high per share commissions with explicit profit-sharing
arrangements. Mutual funds, which are regulated, are not alleged to have paid extremely high
per share commissions. And it is also lower than the 10% increase in trading volume that
Ritter and Welch (2002) conjectured. This is partly due to the fact that we purposely stay
conservative in our above calculations. Meanwhile, there are several other reasons that make
it a lower bound estimate. Wash sales and excessive trading of less liquid stocks are not
included. The average commission is assumed to be 10¢ per share, which may be an
underestimate. Furthermore, Ritter and Welch may overestimate the effect on trading volume
to the degree that trades that would have occurred anyway are merely redirected from ECNs,
or other venues where there are no IPOs to hand out, to integrated securities firms that have
hot IPOs to allocate.
Another reason to think that our estimates are way too low is that if underwriters are
capturing only 0.09% of the money left on the table, they have little reason to underprice,
since underwriters capture 7% of the proceeds on the median IPO through the gross spread.
Also, knowledgeable practitioners have supplied “guesstimates” of the payback rate to us
ranging from a low of 5% to a high of 30% during the Internet bubble.
Almost all IPOs in recent years have used bookbuilding to sell their shares.
Bookbuilding gives the lead underwriter great discretion in allocating IPOs. How the lead
underwriter uses its power is of great importance to the issuing firm, investors, and securities
regulators. Academic researchers have generally assumed that the lead underwriter uses its
allocation discretion in the best interest of the issuing firm to mitigate information asymmetry
and market the IPO. However, recent regulatory investigations indicate that the IPO allocation
process also involves quid pro quos in the form of commission generation.
This paper examines how the allocation of IPOs, as well as the money left on the table
by these IPOs, is linked to commission generation via IPO-related trading, which includes
stock churning and the timing of excessive portfolio rebalancing and regular trading around
IPO dates. IPO allocation data are confidential, and we use an indirect measure – the trading
volume of liquid stocks – to capture IPO-related trading and commission generation. Instead
of using allocation data, which could potentially describe how the pie is sliced, we use the
total amount of money left on the table by all IPOs. Our results are consistent with the
hypothesis that commission generation is an important factor in IPO allocation. Such a
connection is not limited to a few IPOs underwritten by a few investment banks as revealed in
the regulatory investigations – it is widespread, and could have been existing well before the
Internet bubble period, though it was much less aggressive. Our results also provide
quantitative measures about how commission generation and IPO-related trading are linked to
the money left on the table by IPOs. Each $1 billion left on the table during the six trading
days beginning on day t generates abnormal volume in the 50 liquid stocks of, in different
subperiods, between 2.7% and 4.0% on day t, although only during the Internet bubble period
is this point estimate reliably different from zero.
The evidence in this paper shows that there is a short-term connection between trading
commissions and IPO allocation. The evidence in this paper suggests that the gross spread is
not the only way through which IPO underwriters are compensated. This complements the
evidence shown in Reuter (2003) and Goldstein, Irvine, Kandel, and Weiner (2004) that
institutional investors and investment banks often have a long-term relation.
Aggarwal, Reena, 2000, Stabilization activities by underwriters after initial public offerings,
Journal of Finance 55, 1075-1103.
Aggarwal, Reena, 2003, Allocation of initial public offerings and flipping activity, Journal of
Financial Economics 68, 111-135.
Aggarwal, Reena, Nagpurnanand R. Prabhala, and Manju Puri, 2002, Institutional allocation
in initial public offerings: Empirical evidence, Journal of Finance 57, 1421-1442.
Andersen, Torben G.., 1996, Return volatility and trading volume: An information flow
interpretation of stochastic volatility, Journal of Finance 51, 169-204.
Benveniste, Lawrence M., and Paul A. Spindt, 1989, How investment bankers determine the
offer price and allocation of new issues, Journal of Financial Economics 24, 343-361.
Blume, Lawrence, David Easley, and Maureen O’Hara, 1994, Market statistics and technical
analysis: The role of volume, Journal of Finance 49, 153-181.
Boehmer, Ekkehart, and Raymond P. H. Fishe, 2001, Do underwriters encourage stock
flipping? The link between trading profits and pricing in IPOs, Working paper, Texas
A&M University and University of Richmond.
Boehmer, Beatrice, Ekkehart Boehmer, and Raymond P. H. Fishe, 2002, Do institutions
receive favorable allocations in IPOs with better long run returns?, Working paper,
University of Georgia, Texas A&M University, and University of Richmond.
Booth, James, and Lena Chua, 1996, Ownership dispersion, costly information, and IPO
underpricing, Journal of Financial Economics 41, 291-310.
Brennan, Michael, and Julian Franks, 1997, Underpricing, ownership and control in initial
public offerings of equity securities in the UK, Journal of Financial Economics 45,
Campbell, John Y., Sanford J. Grossman, and Jiang Wang, 1992, Trading volume and serial
correlation in stock returns, Quarterly Journal of Economics 108, 905-939.
Chen, Hsuan-Chi, and Jay R. Ritter, 2000, The seven percent solution, Journal of Finance 55,
Cornelli, Francesca, and David Goldreich, 2001, Bookbuilding and strategic allocation,
Journal of Finance 56, 2337-2369.
Ellis, Katrina, Roni Michaely, and Maureen O’Hara, 2000, When the underwriter is the
market maker: an examination of trading in the IPO aftermarket, Journal of Finance
Field, Laura Casares, and Jonathan M. Karpoff, 2002, Takeover defenses of IPO firms,
Journal of Finance 57, 1857-1889.
Fishe, Raymond P. H., 2002, How stock flippers affect IPO pricing and stabilization, Journal
of Financial and Quantitative Analysis 37, 319-340.
Gallant, A. Ronald, Peter E. Rossi, and George Tauchen, 1992, Stock price and volume,
Review of Financial Studies 5, 199-242.
Goldstein, Michael, Paul Irvine, Eugene Kandel, and Zvi Wiener, 2004, Brokerage
Commissions and Institutional Trading Patterns, Babson College working paper.
Hanley, Kathleen Weiss, and William J. Wilhelm, Jr., 1995, Evidence on the strategic
allocation of initial public offerings, Journal of Financial Economics 37, 239-257.
Hansen, Robert S., 2001, Do investment banks compete in IPOs?: the advent of the "7% plus
contract", Journal of Financial Economics 59, 313-346.
He, Hua, and Jiang Wang, 1995, Differential information and dynamic behavior of stock
trading volume, Review of Financial Studies 8, 919-972.
Hiemstra, Craig, and Jonathan D. Jones, 1994, Testing for linear and nonlinear Granger
causality in the stock price-volume relation, Journal of Finance 49, 1639-1664.
Jenkinson, Tim and Howard Jones, 2003, Bids and Allocations in European IPO
Bookbuilding, Journal of Finance, forthcoming.
Jones, Charles M., Gautam Kaul, and Marc L. Lipson, 1994, Transactions, volume, and
volatility, Review of Financial Studies 7, 631-651.
Karpoff, Jonathan M., 1987, The relation between price changes and trading volume: A
survey, Journal of Financial and Quantitative Analysis 22, 109-126.
Ljungqvist, Alexander P., and William J. Wilhelm, Jr., 2002, IPO allocations: discriminatory
or discretionary? Journal of Financial Economics 65, 167-201.
Llorente, Guillermo, Roni Michaely, Gideon Saar, and Jiang Wang, 2001, Dynamic
volume-return relation of individual stocks, NBER working paper.
Loughran, Tim, and Jay R. Ritter, 2002, Why don’t issuers get upset about leaving money on
the table in IPOs? Review of Financial Studies 15, 413-443.
Loughran, Tim, and Jay R. Ritter, 2004, Why has IPO underpricing changed over time?
Financial Management 33 (3), 5-37.
Mathewson, Judy, 2004, U.S. SEC proposes new initial public offering rules, Bloomberg
News, October 13.
Naranjo, Andy, and M. Nimalendran, 2000, Government intervention and adverse selection
costs in foreign exchange markets, Review of Financial Studies 13, 453-477.
Reuter, Jonathan, 2003, Are IPO allocation for sale? Evidence from the mutual fund industry,
MIT working paper.
Ritter, Jay R., and Ivo Welch, 2002, A review of IPO activity, pricing, and allocations, Journal
of Finance 57, 1795-1828.
Sherman, Ann E., 2000, IPOs and long-term relationships: An advantage of book building,
Review of Financial Studies 13, 697-714.
Sherman, Ann E., and Sheridan Titman, 2002, Building the IPO order book: Underpricing and
participation limits with costly information, Journal of Financial Economics 65, 3-29.
Stoughton, Neal M., and Josef Zechner, 1998, IPO-mechanisms, monitoring and ownership
structure, Journal of Financial Economics 49, 45-77.
Zhang, Donghang, 2004, Why do underwriters allocate extra shares that they have to buy
back? Journal of Financial and Quantitative Analysis 39, 571-594.
Summary Daily Statistics for Trading Volumes
This table reports summary daily statistics for trading volumes for all CRSP-listed stocks and stocks in the metric that is used to capture IPO-related stock
trading – the top 50 stocks based on lagged trading volume. The sample period is from 1993 to 2001. The TAQ database is used to calculate the daily
trading volume of each stock. We exclude transactions that are executed on regional exchanges and/or after the market close. For stocks listed on Nasdaq,
we divide the volume by two. We report the statistics separately for three sub-periods.
Mean daily Mean total daily Minimum total Maximum total Std of total daily Mean number of
volume per volume (in daily volume (in daily volume volume (in stocks per Day
stock (in 1000s) 1000s) 1000s) ( in 1000s) 1000s)
Panel A: 1993 - 1998
All Stocks 81 615,246 125,019 1,698,260 242,698 7,390
TVOL_50 2,032 101,579 17,773 341,661 50,574 50
Panel B: 1999 - 2000
All Stocks 193 1,495,511 626,049 2,763,978 361,702 7,745
TVOL_50 6,435 321,766 103,713 758,875 109,067 50
Panel C: 2001
All Stocks 280 1,992,530 659,668 3,290,132 394,134 7,126
TVOL_50 10,628 531,393 160,954 930,199 122,167 50
Summary Statistics on IPOs
This table reports summary statistics for the IPO sample for the sample period from 1993 to 2001 and for
three sub-periods. Panel A of the table reports descriptive statistics of the IPO sample, and Panel B reports
daily statistics. The first-day return is defined as the return from the offer price to the first-day close price.
Money left on the table is defined as the difference between the offer price and the first-day close price
multiplied by the number of shares offered (domestic tranche only), assuming no exercise of overallotment
options. Both the proceeds and the money left on the table are adjusted using the CPI to year 2000 dollars
(year 2001 is not adjusted). Positive IPOs refer to the IPOs that have positive first-day returns. In Panel B,
all daily statistics are unconditional, except for the amount of money left on the table in the last row, which
is conditioning on at least one IPO on that day.
Overall 93-98 99-00 2001
Number of IPOs 3,499 2,620 803 76
Mean offer price, $ 12.86 12.19 14.86 14.66
Mean number of shares offered (‘000) 5,282 3,869 7,981 25,459
Mean proceeds (million dollars) 99.18 71.73 155.80 446.92
Mean first-day return (%) 27.11 15.87 64.96 14.44
Mean money left on the table (million 26.73 10.33 79.04 39.04
Number of Trading Days 2,268 1,516 504 248
Number of Days with IPOs 1,288 953 278 57
Mean Number of IPOs per Day 1.54 1.73 1.59 0.31
Mean Number of Positive IPOs per Day 1.21 1.34 1.31 0.25
Mean Daily Money Left on the Table 41.23 17.86 125.93 11.97
Standard Deviation of Daily Money Left on 146.95 43.20 284.86 51.49
the Table (million dollars)
Minimum Daily Money Left on the Table -276.52 -7.67 -276.52 -36.11
Maximum Daily Money Left on the Table 2608.72 518.49 2608.72 616.31
Mean Daily Money Left on the Table 72.60 28.41 228.30 52.06
Conditioning on at least 1 IPO (million
Regression of Trading Volume on Money Left on the Table by IPOs
This table reports regression results for all IPOs for three sub-periods. In Panel A, the model specification is
Volumet = α + η1 * Date _ Indext + η 2 * S & P500 t + λi * Weekday i
+ mi * Monthi + β 1 * Market t + β 2 * Market t + δ i * Volumet −i
i =1 i =1
+ γ i * Money t +i + ε t
i = −1
In the regression, Volumet is the natural log of the daily total volume of the top 50 stocks on NYSE,
Amex and Nasdaq ranked by lagged volume, multiplied by 100. We number all the trading days from
January 1993 to December 2001 sequentially, and Date _ Indext is the sequential number of Day t
divided by the total number of trading days (2,268), and S & P500 t is the S&P500 index. The weekday
dummies, Weekday1 through Weekday 4 , correspond to Tuesday to Friday. The eleven monthly
dummies, Month1 through Month11 , refer to January to November. The market return on Day t ,
Market t is the return on the S&P500 Index, and Market t is the absolute value of the market return.
The Money t +i variable is the summation of the total amount of money left on the table of all IPOs on
Day t + i , where i = -1, 0, ..., 5 . The amount of money left on the table is measured as the difference
between the first-day close price and the offer price multiplied by the number of shares offered. For
presentation purpose, the money left on the table is measured in $100 million (this variable is measured in
millions in previous tables). The significance of the money left on the tables variables is indicated using *
(significant at the 10% level), ** (significant at the 5% level) and *** (significant at the 1% level).
T-statistics, corrected for heteroscedascity, are in parentheses. In Panel B, we use the aggregated amount of
money left on the table from Day t to Day t + 5 , Aggregate _ Money t , to replace the amounts of
money left on the table on each day of the six-day window (note that the day before the current day is not
included). The coefficients for all the control variables are similar to the ones reported in Panel A and are
omitted. In both Panel A and B, the numbers of observations, which are the number of trading days, for the
three sub-periods are respectively 1516, 504, and 248.
Variable 1993 - 1998 1999 - 2000 2001
Intercept 790.92 977.37 1,792.73
(13.93) (6.40) (4.98)
Date _ Indext 54.03 313.90 -113.96
(4.29) (6.57) (-0.35)
S & P500 t 0.03 -0.06 -0.08
(5.41) (-3.35) (-2.09)
Tuesday 15.91 7.05 14.27
(10.89) (3.33) (4.50)
Wednesday 17.29 8.09 18.08
(12.98) (4.01) (6.25)
Thursday 11.68 4.85 15.62
(8.64) (2.35) (6.11)
Friday 6.46 -1.83 7.84
(3.68) (-0.68) (1.96)
January 12.97 18.61 6.51
(4.99) (4.35) (0.20)
February 9.82 6.15 0.45
(4.04) (1.84) (0.02)
March 9.06 11.41 1.23
(3.84) (3.20) (0.06)
April 9.41 16.14 1.78
(4.01) (3.80) (0.07)
May 5.17 1.72 -8.35
(2.21) (0.43) (-0.41)
June 3.19 1.99 -10.67
(1.39) (0.49) (-0.63)
July 5.05 0.51 -19.53
(1.96) (0.11) (-1.46)
August 3.66 -3.39 -27.25
(1.54) (-0.78) (-2.49)
September 5.84 0.68 -3.89
(2.36) (0.18) (-0.38)
October 8.59 5.63 -8.46
(3.56) (1.72) (-1.15)
November 3.68 1.04 -7.19
(1.21) (0.23) (-0.98)
Market t -0.82 -0.57 1.04
(-1.77) (-1.22) (1.80)
Market t 8.65 4.34 4.99
(11.33) (6.58) (4.71)
Volumet −1 0.41 0.34 0.34
(9.07) (5.25) (3.46)
Volumet −2 0.07 0.08 -0.10
(2.45) (1.76) (-1.55)
Volumet −3 0.03 0.01 -0.06
(1.28) (0.27) (-0.96)
Volumet −4 0.02 -0.02 0.02
(0.56) (-0.52) (0.45)
Moneyt −1 -0.71 -0.11 1.92
(-0.74) (-0.57) (1.12)
Moneyt 0.05 0.69*** -0.90
(0.06) (3.83) (-0.75)
Moneyt +1 -0.42 0.43*** 0.76
(-0.56) (2.87) (0.65)
Moneyt + 2 2.38*** -0.04 -0.57
(2.97) (-0.28) (-0.34)
Moneyt +3 -0.10 0.27 0.61
(-0.12) (1.59) (0.46)
Moneyt + 4 1.40 -0.02 -1.80
(1.41) (-0.10) (-1.45)
Moneyt +5 -0.72 0.39** -1.66
(-0.62) (2.14) (-0.90)
Adjusted R 2 (%) 88 83 58
Aggregate _ Money t 0.40 0.27*** 0.29
(1.27) (4.43) (0.34)
Adjusted R 2 (%) 88 82 56
Day of the Week Pattern of IPOs
This figure demonstrates the existence of a day of the week pattern in both the number of deals and the
average first-day return for IPOs. We group IPOs by year. Figures 1.1 and 1.2 report the number of IPOs
and the mean first-day return on different weekdays, respectively. For each year in each figure, one bar
represents one weekday from Monday through Friday from left to right. Please note that relatively few
IPOs went public on Monday, and this makes the mean first-day return of Monday in Figure 1.2 sensitive to
outliers (if they are present). Only two companies went public on Monday in 2000 and one company went
public on Monday in 2001. The two IPOs that went public on Monday in 2000 (Diversa Corp. and
NewForma.com Inc.) have an average first-day return of 258%. For expositional purposes, we remove
these two IPOs in Figure 1.2. The one company that went public on Monday in 2001 has a 0% first-day
Figure 1.1 Day of the Week Pattern for IPOs
Number of IPOs
1993 1994 1995 1996 1997 1998 1999 2000 2001
Figure 1.2 Day of the Week Pattern for IPOs
Mean First-Day Returns (%)
1993 1994 1995 1996 1997 1998 1999 2000 2001