Quote-based competition, market share, and execution quality in NASDAQ-listed securities

Quote-based competition, market share, and execution quality in NASDAQ-listed securities

Quote-based competition, market share, and execution quality in NASDAQ-listed securities Kee H. Chung a,*, Chairat Chuwonganant b a State University of New York (SUNY) at Buffalo, Department of Finance and Managerial Economics, Buffalo, NY 14260, United States b Department of Finance, College of Business Administration, Kansas State University, Manhattan, KS 66506-0503, United States Received 22 June 2006; accepted 2 January 2007 Available online 29 April 2007 Abstract We show that competitive quotes help increase dealer market share on NASDAQ, despite the fact that a large proportion of order flow is preferenced.

We find that decimal pricing and the introduc- tion of new trading platforms such as SuperSOES and SuperMontage have significantly changed the effect of quote aggressiveness on dealer market share. In particular, decimal pricing reduces (increases) the price (size) elasticity, SuperSOES increases the size elasticity, and SuperMontage increases both the size and price elasticity of dealer market share. We also show that market centers provide greater price improvements and faster executions when they post competitive quotes.  2007 Elsevier B.V. All rights reserved.

JEL classification: G18; G19 Keywords: Dealer market share; Quote aggressiveness; Order preferencing; Decimalization; SuperSOES; SuperMontage 1. Introduction In this study we examine the extent to which dealer market share in NASDAQ securi- ties is related to quote aggressiveness and how the effect of quote aggressiveness on market 0378-4266/$ - see front matter  2007 Elsevier B.V. All rights reserved. doi:10.1016/j.jbankfin.2007.01.023 * Corresponding author. Tel.: +1 716 645 3262; fax: +1 716 645 3823. E-mail addresses: keechung@buffalo.edu (K.H. Chung), cchuwong@yahoo.com (C. Chuwonganant).

Journal of Banking & Finance 31 (2007) 2770–2795 www.elsevier.com/locate/jbf

share has been affected by changes in trading environments such as decimalization, SuperSOES, and SuperMontage. Whether the elasticity of market share with respect to quote aggressiveness varies with trading environments is an important question because the answer to the question can help devise better trading environments and platforms. Prior research offers limited evidence concerning the relation between quote aggressive- ness and market share. Only recently have studies begun to shed some light on the direct empirical link between quote aggressiveness and market share. Blume and Goldstein (1997) find that non-NYSE market makers attract more order flow for NYSE stocks when they post the best available quotes.

Bessembinder (2003a) finds substantial quote-based competition for order flow in NYSE-listed stocks. Bessembinder also shows that off-NYSE liquidity providers post aggressive quotes when they are willing to give better-than-normal executions. However, the results of NYSE-listed securities may not be directly relevant for NASDAQ-listed securities because the nature of order flow competition between NAS- DAQ dealers is different from that between trading venues for NYSE-listed stocks. Klock and McCormick (2002) examine the effect of quote aggressiveness on dealer market share and find a positive relation between the two variables.

However, their results are based on limited data (i.e., seven months in 1996) before the introduction of the new order handling rules in 1997. Consequently, whether their results hold in the post market reform environ- ment is unclear. Chung et al. (2006) analyze the relation between dealer market share and quote aggressiveness using the cross-sectional regression method that is similar to the one employed in the present study. Their study also uses limited data (i.e., November 2000 and June 2001) and it does not examine how the effect of quote aggressiveness on market share has been affected by changes in trading environments.

Goldstein et al. (2005) analyze inter-market competition in NASDAQ-listed securities during the second quarter of 2003 using a sample of 100 stocks. They show that ECNs are capable of competing with NASDAQ on quotes, while AMEX and Chicago use non-price methods to attract trades. The authors also find that trading venues attract more orders when they quote aggressively on both sides. In this study, we analyze how quote aggressiveness affects dealer market share during the five-year period from 1999 through 2003 using a large sample of NASDAQ stocks. In particular, we investigate whether changes in trading environments such as decimal pricing, SuperSOES, and SuperMontage can explain changes in the price and size elasticity of dealer market share over time.

Considering the academic and regulatory debates on how these new protocols and trading platforms can affect market quality and investor wel- fare, the results of our study would be of great interest to both regulatory authorities and the general public. To the extent that the incentive to quote aggressively is determined by whether aggressive quotes attract more order flow and this incentive ultimately determines the inside spreads (i.e., execution costs), it is important to understand how different trad- ing protocols and platforms have changed the price and size elasticity of market share. Our results show that aggressive quotes help increase dealer market share, despite the fact that a large proportion of NASDAQ order flow is preferenced.1 We find that decimal 1 The rule of best execution is also likely to reduce both the effect of quote aggressiveness on market share and the dealer’s incentive to quote aggressively.

In deciding how to execute orders, brokers and dealers have a duty to seek the best execution that is reasonably available for customers’ orders. Best execution requires dealers to execute customer orders at prices that are equal to or better than the National Best Bid or Offer (NBBO), regardless of their own quotes. Under this rule, execution quality may be similar between dealers who quote aggressively and dealers who do not. Consequently, brokers may not have a strong incentive to send orders only to those dealers who post competitive quotes.

K.H. Chung, C. Chuwonganant / Journal of Banking & Finance 31 (2007) 2770–2795 2771

pricing and the introduction of new trading platforms such as SuperSOES and SuperMon- tage have significantly changed the effectiveness of aggressive quotes in raising dealer mar- ket share. In particular, decimal pricing reduces (increases) the price (size) elasticity, SuperSOES increases the size elasticity, and SuperMontage increases both the size and price elasticity of dealer market share. We also show that market centers provide greater price improvements and faster executions when they post competitive quotes.

The paper is organized as follows. Section 2 describes data sources and presents descrip- tive statistics. Section 3 examines whether quote aggressiveness varies with the dealer Herfin- dahl index and dealer type. Section 4 investigates how decimal pricing and the introduction of new trading platforms, such as SuperMontage and SuperSOES, have affected the price and size elasticity of dealer market share. Section 5 examines the relation between quote aggres- siveness and execution quality. Section 6 provides a brief summary and concluding remarks. 2. Data sources and sample characteristics We obtain trade, inside quote, and dealer quote data for this study from the NAS- TRAQ Trade and Quote Data.

We obtain data on the number of shares outstanding from the CRSP database. Our study sample consists of 2004 stocks listed on NASDAQ from January 1999 to December 2003. We omit the following quotes and trades to minimize data error: quotes if either the ask or bid price is non-positive; quotes if either the ask or bid size is non-positive; quotes if the bid-ask spread is greater than $5 or less than or equal to zero2 ; before-the-open and after-the-close trades and quotes; trades if the price or volume is non-positive; trade price, pt, if j(pt  pt  1)/pt  1j > 0.5; ask quote, at, if j(at  at  1)/at  1j > 0.5; and bid quote, bt, if j(bt  bt  1)/bt  1j > 0.5.

We calculate monthly values of the following variables for each stock (stock i in month t): share price as measured by the mean value of quote midpoints, PRICE(i,t); the average daily number of trades, NTRADE(i,t); average dollar trade size, TSIZE(i,t); return volatility as measured by the standard deviation of daily quote midpoint returns, VOLATILITY(i,t); and market capitalization as measured by the market value of equity, MVE(i,t). For each stock, we then calculate the mean value of each variable using monthly time-series observations. Finally, we calculate the mean, standard deviation, and select per- centile values of each variable using cross-sectional observations.

Panel A of Table 1 shows the results. The average share price and market capitalization are $16.43 and $1188 million. The average daily number of trades and trade size are 718.7 and $9418, respectively. The average standard deviation of daily quote midpoint returns is 0.0068. The total number of market makers in our study sample is 378 and, of those, 13 are institutional brokers, five are wirehouses, and five are wholesalers.3 There are two 2 Shkilko et al. (forthcoming) show that zero and negative spreads are not rare on NASDAQ. 3 In this study we do not treat electronic communications networks (ECNs) as market makers for two reasons.

First, there are significant differences between ECNs and dealers in their quote and trade behavior. Second, Rule 605 market quality reports of two ECNs (i.e., INCA and INET) were alleged to contain material errors. On October 18, 2005, the Securities and Exchange Commission (SEC) alleged that from June 2001 through May 2004, INCA and INET repeatedly published Rule 605 reports that contained inaccurate order execution quality information. Pursuant to their settlement with the SEC, INCA and INET paid $700000 in civil penalties ($350000 each) and agreed to institute remedial undertakings. In addition, INCA and INET agreed to retain a third party regulatory auditor to conduct a comprehensive regulatory audit of the Rule 605 compliance program of INCA and INET by the end of 2006.

2772 K.H. Chung, C. Chuwonganant / Journal of Banking & Finance 31 (2007) 2770–2795

Table 1 Descriptive statistics Variable Mean Standard deviation Percentile 1 5 25 50 75 95 99 Panel A: Descriptive statistics for stock characteristics Share price ($) 16.43 18.62 1.41 2.30 5.88 12.99 21.59 41.68 62.84 Number of trades 718.70 3,129.33 3.50 4.82 15.17 66.85 323.34 2505.37 13927.10 Trade size ($) 9418 7270 1221 1829 4195 7833 12489 22740 32453 Return volatility 0.0068 0.0055 0.0007 0.0012 0.0030 0.0053 0.0090 0.0175 0.0245 Market value of equity ( in thousands) 1188043 11060000 4843 10321 37824 113640 386847 2739400 12793200 Panel B: Descriptive statistics for dealer quote aggressiveness and market share PTINS(i,j) 0.2232 0.1745 0.0550 0.0560 0.0868 0.1642 0.3083 0.5896 0.7823 PTINSA(i,j) 0.0591 0.0628 0.0300 0.0302 0.0305 0.0345 0.0578 0.1728 0.3412 RELQS(i,j) 0.7459 0.6524 0.0870 0.1556 0.3450 0.5959 0.9680 1.7756 2.9464 MS(i,j) 0.1175 0.1221 0.0301 0.0308 0.0414 0.0728 0.1398 0.3665 0.6300 Panel A shows the descriptive statistics for our study sample of 2004 NASDAQ stocks and Panel B shows the descriptive statistics for dealer quote aggressiveness and market share.

We calculate monthly values of the following variables for each stock (stock i in month t): share price as measured by the mean value of quote midpoints, the average daily number of trades, average dollar trade size, return volatility as measured by the standard deviation of daily quote midpoint returns, and market capitalization as measured by the market value of equity. For each stock, we then calculate the mean value of each variable using monthly time-series observations. Finally, we calculate each variable’s mean, standard deviation, and select percentile values using cross-sectional observations.

We measure dealer J’s price aggressiveness for stock i in month t by the percentage of dealer time at the inside, PTINS(i,j,t) and the percentage of dealer time at the inside alone, PTINSA(i,j,t). We measure size aggressiveness by the relative magnitude of a dealer’s quoted depth in relation to the average quoted depth of all dealers at the inside for a given stock, RELQS(i,j,t). Dealer J’s market share in stock i during month t, MS(i,j,t), is measured by V(i,j,t)/RjV(i,j,t), where V(i,j,t) is stock i’s volume accounted for by dealer j during month t. To calculate the descriptive statistics of these variables, we first calculate the mean value of PTINS(i,j,t), PTINSA(i,j,t), and RELQS(i,j,t) using monthly time-series observations of each variable.

We then calculate the mean, standard deviation, and select percentile values of each variable using dealer-stock observations.

K.H. Chung, C. Chuwonganant / Journal of Banking & Finance 31 (2007) 2770–2795 2773

dimensions of quote aggressiveness: price and size. We measure price aggressiveness (of dealer j in stock i during month t) by both the percentage of dealer time at the inside market, PTINS(i,j,t) and the percentage of dealer time at the inside market alone, PTINSA(i,j,t). We obtain PTINS(i,j,t) by dividing dealer j’s time at the inside ask, inside bid, or both for stock i by the total trading hours in month t. Likewise, we obtain PTINSA(i,j,t) by dividing dealer j’s time at the inside ask alone, inside bid alone, or both for stock i by the total trading hours in month t.

We measure size aggressiveness by the relative magnitude of a dealer’s quoted depth in relation to the average quoted depth of all dealers at the inside for a given stock, RELQS(i,j,t).

Dealer J’s market share in stock i during month t, MS(i,j,t), is measured by V(i,j,t)/ RjV(i,j,t), where V(i,j,t) is stock i’s volume accounted for by dealer j during month t. Because NASDAQTrader.com maintains Market Participant (MP) Monthly Share Vol- ume data only for the most recent 12 months, we obtain the data that are required to cal- culate MS(i,j,t) directly from NASDAQ.4 NASDAQ obtains and compiles the share volume data from trade reports that are directly entered into NASDAQ Market Center Trade Reporting Service (ACT).5 Although there have been few changes in data construc- tion and format in recent years, they may not materially affect our results because we measure the price and size elasticity of dealer market share using monthly cross-sectional data.

To calculate the descriptive statistics of these variables, we first calculate the mean value of PTINS(i,j,t), PTINSA(i,j,t), RELQS(i,j,t), and MS(i,j,t) using monthly time-series observations of each variable. We then calculate the mean, standard deviation, and select percentile values of each variable using dealer-stock observations. Panel B of Table 1 shows the results. Dealers are at the inside market 22.3% of time on average, and they are alone at the inside market only 5.9% of time. These results are in line with the finding of previous studies that although dealers are required to quote on both sides, they tend to post competitive quotes (i.e., inside market quotes) on only one side.

Chan et al. (1995) show that NASDAQ dealers rarely (0.7% of the total trading time) post quotes that lie at both the inside bid and inside ask, except when the width of the inside spread is unusu- ally large. Chung and Zhao (2004) show that the percentage of time during which the mar- ket maker’s quotes are at either the inside bid or the inside ask is only 24.5%. Chung and 4 We thank Tim McCormick for providing the data. Although we exclude ECNs from the list of market makers, we measure dealer market share using all trades, including trades executed on ECNs. See Fink et al. (2006) for a detailed description of difficulties associated with measuring ECN market shares.

5 NASDAQ Market Center for NASDAQ Trading automatically reports trades to ACT and identifies the member that provides the liquidity as the reporting member. ECN trade reporting varies because each ECN has different policies on reporting obligation. There are two general policies that ECN employs: (1) ECN requires its NASD member subscribers to report trades executed in the ECN’s system, which means the subscriber is identified as the reporting member and credited with the trade; (2) ECN reports the trades executed in its system, identifies itself as the reporting member, and therefore, is credited with the volume.

These varying policies result in different volume credited to the market maker versus the ECN in the daily volume reports and Monthly Volume Summaries reports. Beginning with April 1, 2003 data, volume reported from all MPIDs assigned to a single NASDAQ broker/dealer organization is aggregated under a single MPID in the NASDAQ Broker/Dealer Data volume reports and monthly share volume reports. For example, if a NASDAQ broker/dealer firm is using a second or third MPID to facilitate SuperMontage order management, the volume activity reported for these related MPIDs is aggregated under the firm’s primary MPID.

Visit http://www.nasdaqtrader.com/trader/ tradingdata/generalvolumecalc.stm for a detailed description of NASDAQ trading volume data. 2774 K.H. Chung, C. Chuwonganant / Journal of Banking & Finance 31 (2007) 2770–2795

Zhao also show that market makers rarely (2.4% of time) post competitive prices on both sides of the quote. 3. Dealer Herfindahl-index and dealer quote aggressiveness Although prior research sheds light on whether the distribution of market share across dealers in a given stock (i.e., the Herfindahl-index of a stock) has an impact on dealer com- petition, there has been no prior research concerning whether the distribution of trading volume across stocks for a given dealer has any impact on dealer quotation behavior. We believe this is an interesting issue because dealers who focus on fewer stocks may not only post aggressive prices and sizes, but also be more effective in raising their market shares through those quotes.

To examine whether dealers who make markets in a small number of stocks tend to quote more aggressively than those who make markets in a large number of stocks, we calculate the Herfindahl-index of each dealer. The Herfindahl-index of dealer j is defined as: H-INDEX(j) = Ri[100V(i,j)/RiV(i,j)]2 , where V(i,j) is stock i’s dol- lar volume executed by dealer j. The dealer Herfindahl-index measures whether the deal- er’s trading is concentrated on a few stocks or dispersed across many stocks. A dealer with a high Herfindahl-index makes markets in a small number of stocks, whereas a dealer with a low Herfindahl-index makes markets in many stocks.

To shed light on whether dealers exhibit different quotation behaviors depending on whether they have a concentrated or dispersed market making business, we estimate the following regression model: QAðj; tÞ ¼ a0 þ a1H-INDEXðj; tÞ þ a2DUMIBðjÞ þ a3DUMWHðjÞ þ a4DUMWSðjÞ þ eðj; tÞ; ð1Þ where QA(j,t) denotes dealer J’s average quote aggressiveness (across stocks) measured by PTINS(j,t), PTINSA(j,t), and RELQS(j,t). H-INDEX(j,t) denotes the Herfindahl-index of dealer j. To control for dealer types, we include three indicator variables (DUMIB(j), DUMWH(j), and DUMWS(j)), which equal one for institutional brokers, wirehouses, and wholesalers, respectively, and zero otherwise.

We estimate the above model for each month and calculate the mean a coefficients across months and the z-statistics. We obtain the z-statistic by adding individual regres- sion t-statistics across months and then dividing the sum by the square root of the number of regression coefficients. The results (see Table 2) show that all three measures of quote aggressiveness (i.e., PTINS, PTINSA, and RELQS) are positively and significantly related the dealer Herfindahl-index after controlling for dealer types. This indicates that dealers who have a concentrated market making business quote more aggressively.

The results also show that wholesalers, wirehouses, and institutional brokers post more aggressive price (PTINS) and size quotes (RELQS) than regional firms. Among wholesalers, wire- houses, and institutional brokers, wholesalers quote most aggressively. 4. Quote aggressiveness and dealer market share In this section we first estimate the price and size elasticity of dealer market share with respect to quote aggressiveness using cross-sectional data in each month. We then examine whether aggressive price quotes are less effective in raising dealer market share when the tick size is smaller by comparing the price elasticity of dealer market share before and after K.H.

Chung, C. Chuwonganant / Journal of Banking & Finance 31 (2007) 2770–2795 2775

decimalization. Similarly, we examine whether the introduction of new trading platforms, such as SuperMontage and SuperSOES, has affected the price and size elasticity of dealer market share. Boehmer et al. (2007) analyze whether market share is related to execution quality mea- sures reported under the Securities and Exchange Commission (SEC)’s Rule 605 (formerly Rule 11Ac1-5). While Boehmer, Jennings, and Wei examine whether market share is related to the post-trade execution quality measures (i.e., effective spreads and execution speeds), our study focuses on whether market share is related to the pre-trade market qual- ity measure (i.e., quote aggressiveness).

This distinction has important implications for research design because most market participants can easily observe pre-trade market quality when they make order routing decisions, whereas they can only find out post-trade execution quality with some time delay. Hence, a reasonable conjecture would be that there is a contemporaneous relation between market share and quote aggressiveness (as proposed in the present study), but a lagged one between market share and execution qual- ity (as shown in Boehmer et al., 2007).

4.1. Regression results We note that dealer quotes frequently reflect limit orders from customers rather than the dealer’s trading interest. Because our main focus is to examine whether aggressive quotes (regardless of their origins) lead to larger market share, it is unimportant to distin- guish whose interests are reflected in dealer quotes. Because our variable measurement interval (i.e., a month) is a reasonably long period, we conjecture that a dealer’s market Table 2 Effects of H-index and dealer types on quote aggressiveness PTINS(j,t) PTINSA(j,t) RELQS(j,t) Intercept 0.1139** 0.0192** 0.6971** (132.90) (69.90) (149.99) H-INDEX(j,t) 0.0255** 0.0046** 0.4699** (8.09) (4.53) (27.52) DUMIB(j) 0.0195** 0.0068** 0.0081** (5.15) (5.23) (3.35) DUMWH(j) 0.0352** 0.0071** 0.0932** (6.01) (3.64) (4.19) DUMWS(j) 0.1351** 0.0311** 0.2991** (23.87) (18.85) (5.17) Mean R2 0.052 0.040 0.058 Mean sample size 295 295 295 This table reports the results of the following regression model: QAðj; tÞ ¼ a0 þ a1H-INDEXðj; tÞ þ a2DUMIBðjÞ þ a3DUMWHðjÞ þ a4DUMWSðjÞ þ eðj; tÞ; where QA(j,t) denotes dealer J’s average quote aggressiveness (across stocks) measured by PTINS(j,t), PTIN- SA(j,t), and RELQS(j,t).

H-INDEX(j,t) denotes the Herfindahl-index of dealer j. To control for dealer types, we include three indicator variables (DUMIB(j), DUMWH(j), and DUMWS(j)), which equal one for institutional brokers, wirehouses, and wholesalers, respectively, and zero otherwise. We estimate the above model for each month and calculate the mean a coefficients across months and the z-statistics. We obtain the z-statistic by adding individual regression t-statistics across months and then dividing the sum by the square root of the number of regression coefficients. Numbers in parenthesis are z-statistics.

** Significant at the 1% level. 2776 K.H. Chung, C. Chuwonganant / Journal of Banking & Finance 31 (2007) 2770–2795

share in month t is determined largely by his quotes in month t, rather than by quotes dur- ing the previous months. We estimate the following regression models for each stock using the monthly cross-sec- tional (across dealers) data: logðMSði; j; tÞÞ ¼ k0it þ k1it logðPTINSði; j; tÞÞ þ k2it logðRELQSði; j; tÞÞ þ e1ði; j; tÞ ð2Þ and logðMSði; j; tÞÞ ¼ w0it þ w1it logðPTINSAði; j; tÞÞ þ w2it logðRELQSði; j; tÞÞ þ e2ði; j; tÞ: ð3Þ Our focus here is to look at how inter-dealer differences in market share for a given stock can be explained by inter-dealer differences in quote aggressiveness during each month.

Panel A of Table 3 reports the standard error-weighted mean values of regression coefficients and the t-statistics. The method assigns smaller weights to the regression coef- ficients that are less meaningful. Panel B presents the equal-weighted mean values of regression coefficients and the z-statistics.

Table 3 shows that a 1% increase in PTINS results in a 0.59–0.62% increase in dealer mar- ket share, depending on averaging methods. Similarly, a 1% increase in PTINSA results in a 0.56% increase in dealer market share. The table also shows that a 1% increase in RELQS results in a 0.41–0.56% increase in dealer market share, depending on model specifications or averaging methods. The t- and z-statistics indicate that the results are all statistically sig- nificant at the 1% level. These results indicate that dealers who quote aggressively have a lar- ger market share than dealers who quote less aggressively during a given month.

Overall, our results indicate that aggressive quotes help increase dealer market share, despite the fact that a large proportion of NASDAQ volume is either internalized or rou- ted through payment for order flow agreements.6 Our results are consistent with the find- ing of Bessembinder (2003a) that there is substantial quote-based competition for order flow in NYSE-listed stocks. Although orders are expected to execute at best available prices whether or not the order-receiving dealer is at the inside market, brokers send more orders to those dealers who post competitive quotes.

A possible explanation for this result is that dealers may provide greater price and/or size improvements as well as speedier executions when they post competitive quotes. 6 Order preferencing reduces the effect of quote aggressiveness on dealer market share. To the extent that preferenced orders are captive orders, they are less likely to be affected by quote aggressiveness than unpreferenced orders. Godek (1996) holds that order preferencing weakens the linkage between quote aggressiveness and order flow. Bloomfield and O’Hara (1998) show that order preferencing can significantly degrade market performance if preferenced orders are a large share of the market or are received by virtually all market makers.

Chung et al. (2004) show that a large proportion of order flow on NASDAQ is preferenced. They find wide variation in the percentage of internalized volumes across stocks, ranging from zero to almost 100%, with a mean value of around 25%. They also show that the non-inside volume accounts for more than 50% of the total volume during both the pre- and post-decimalization periods. Chung, Chuwonganant, and McCormick estimate that over 75% of NASDAQ trading volume is either internalized or routed via payment for order flow arrangements. Therefore, it is unclear whether aggressive quotes result in a larger market share on NASDAQ.

Benveniste et al. (1992), Easley et al. (1996), Battalio et al. (1997, 2001a,b), Securities and Exchange Commission (1997), Battalio and Holden (2001) and Peterson and Sirri (2003) also examine the effect of order preferencing on market quality.

K.H. Chung, C. Chuwonganant / Journal of Banking & Finance 31 (2007) 2770–2795 2777

For instance, Chung et al. (2004) show that orders receive greater price improvements when they are executed by dealers who are at the inside market. Bessembinder (2003a) shows that off-NYSE liquidity providers post aggressive quotes when they are prepared to give better-than-normal trade executions, and order routing responds systematically to these quotes. In a similar vein, dealers on NASDAQ may also post competitive quotes when they are willing to give better-than-normal trade executions.

We perform empirical investigation of this issue later in the paper.

4.2. Decimal pricing and the price and size elasticity of dealer market share The NASDAQ Stock Market began its decimal test phase with 14 securities on March 12, 2001, followed by another 197 securities on March 26, 2001. All remaining NASDAQ securities converted to decimal trading on April 9, 2001. We conjecture that decimal pric- ing reduces the effect of price aggressiveness on dealer market share for at least two rea- sons. In general, the price priority rule becomes less consequential when tick size is smaller because traders can easily step in front of other traders without incurring a large cost (Harris, 1994, 1997).

As a result, the fact that the dealer is at the inside market is less likely to result in an increase in his market share during the post-decimal period (since Table 3 Elasticity of dealer market share with respect to price and size aggressiveness log(MS(i,j,t)) t-Statistic log(MS(i,j,t)) t-Statistic Panel A: Results based on the standard error-weighted mean value of regression coefficients log(PTINS(i,j,t)) 0.5939** 292.40 log(PTINSA(i,j,t)) 0.5581** 291.65 log(RELQS(i,j,t)) 0.5496** 156.58 0.4093** 113.66 log(MS(i,j,t)) z-Statistic log(MS(i,j,t)) z-Statistic Panel B: Results based on the equal-weighted mean value of regression coefficients log(PTINS(i,j,t)) 0.6223** 866.81 log(PTINSA(i,j,t)) 0.5605** 946.93 log(RELQS(i,j,t)) 0.5569** 282.89 0.4142** 213.02 This table reports the results of the following regression models: logðMSði; j; tÞÞ ¼ k0it þ k1it logðPTINSði; j; tÞÞ þ k2it logðRELQSði; j; tÞÞ þ e1ði; j; tÞ and logðMSði; j; tÞÞ ¼ w0it þ w1it logðPTINSAði; j; tÞÞ þ w2it logðRELQSði; j; tÞÞ þ e2ði; j; tÞ; where MS(i,j,t) is dealer J’s market share in stock i [i.e., V(i,j)/RjV(i,j), where V(i,j) is stock i’s volume accounted for by dealer j] for time t, PTINS(i,j,t) is the percentage of time during which dealer j is at the inside for stock i for time t, RELQS(i,j,t) is the ratio of the average depth of dealer J’s inside market quotes to the average depth of all dealers’ inside market quotes for stock i for time t, PTINSA(i,j,t) is the percentage of time during which dealer j is at the inside alone for stock i for time t, and e1(i,j,t) and e2(i,j,t) are the error terms.

We estimate the above regres- sion models for each stock using the monthly cross-sectional (across dealers) data. Panel A reports the standard error-weighted mean values of regression coefficients and the t-statistics, where the reciprocal of the standard error of the estimated coefficient is used as weight. The method assigns smaller weights to the regression coeffi- cients that are less meaningful (i.e., smaller t-statistics). Panel B shows the equal-weighted mean values of regres- sion coefficients and the z-statistics. We obtain the z-statistic by adding individual regression t-statistics and then dividing the sum by the square root of the number of regression coefficients.

** Significant at the 1% level. 2778 K.H. Chung, C. Chuwonganant / Journal of Banking & Finance 31 (2007) 2770–2795

other traders can more easily step in front of him), relative to the pre-decimal period. In addition, price elasticity is likely to be smaller after decimalization because there are fewer shares available at each price point under decimal pricing than under $1/16 pricing.7 To test this conjecture, we compare the price elasticity of dealer market share in the periods before and after decimalization. We consider the period from January 2001 to February 2001 as the pre-decimal period, and the period from May 2001 to June 2001 as the post-decimal period.8 We report the standard error-weighted and equal-weighted mean values of the price elasticity of market share in Panel A of Table 4.

The results show that the mean values of price elasticity during the post-decimal period are significantly smaller than the corre- sponding values during the pre-decimal period, regardless of whether we use PTINS or PTINSA as the measure of price aggressiveness. For example, the mean value of k1it esti- mates for the pre-decimal period is 0.7416 (0.7629), while the corresponding value for the post-decimal period is 0.6607 (0.6801) according to the standard error-weighted (equal- weighted) results. Similarly, the mean value of w1it estimates for the pre-decimal period is 0.6643 (0.6658), while the corresponding value for the post-decimal period is 0.6136 (0.611) according to the standard error-weighted (equal-weighted) results.

Overall, our results suggest that the effect of competitive price quotes on dealer market share is weaker under the smaller tick size scheme because such scheme reduces the importance of price priority and results in smaller guaranteed depths.9 Although decimal pricing reduced the price elasticity of market share, it increased the size elasticity of market share. Panel B of Table 4 shows that the size elasticity of market share after decimal pricing is significantly larger than the corresponding figure during the pre-decimal period. Bessembinder (2003b) finds that quotation sizes decreased dramati- cally after decimalization for stocks in all capitalization groups.

He finds that the mean NBBO quote size decreased by 26.1% (from 1031 shares to 762 shares), and the greatest percentage reductions in quote sizes are for large capitalization stocks. Hence, decimal pricing led to smaller displayed depths on average, and aggressive size quotes became more effective in raising market shares than under fractional pricing. This result should not be surprising because the extent to which traders wanted to buy or sell larger sizes than quoted sizes is probably greater under decimal pricing because of smaller quoted depths. Although the above results show that both the price and size elasticity of dealer market share differs significantly between the pre- and post-decimal periods, the different elasticity 7 See, e.g., Goldstein and Kavajecz (2000), Bacidore et al.

(2003), Bessembinder (2003b) and Chakravarty et al. (2004). In particular, Goldstein and Kavajecz (2000) investigate the impact of reducing the minimum tick size on the liquidity of the market. They analyze both spreads and depths for periods before and after the NYSE’s change from eighths to sixteenths. The authors find that depth declined throughout the entire limit order book after the change.

8 The SEC initially mandated Rule 605 from April 2001, which coincides with the timing of decimal pricing. Our results are not likely to be materially contaminated by this concurrent event because our study concerns the relation between market share and pre-trade market quality (i.e., quote aggressiveness), not post-trade execution quality. 9 There may be alternative interpretations of this result. For example, to the extent that large tick sizes can be viewed as restriction that is imposed on the marketplace, reducing tick sizes can be viewed as eliminating that restriction. When a restriction is eliminated, then the market should be more responsive to both prices and quantities.

Hence, decimalization could be expected to increase both price elasticity and size elasticity. From this perspective, our price elasticity result may be viewed as an empirical anomaly that runs counter to theory. We thank the referee for this point.

K.H. Chung, C. Chuwonganant / Journal of Banking & Finance 31 (2007) 2770–2795 2779

Table 4 Comparison of the price and size elasticity before and after decimal pricing Standard error-weighted mean values Equal-weighted mean values Before After Difference (after  before) t-Statistica Before After Difference (after  before) t-Statistica Panel A: Comparison of the price elasticity of dealer market share before and after decimalization Price elasticity estimated from model (2): k1 0.7416 0.6607 0.0809** 10.83 0.7629 0.6801 0.0828** 9.54 Price elasticity estimated from model (3): w1 0.6643 0.6136 0.0507** 7.82 0.6658 0.6110 0.0548** 7.18 Panel B: Comparison of the size elasticity of dealer market share before and after decimalization Size elasticity estimated from model (2): k2 0.5786 0.6421 0.0635** 5.26 0.5780 0.6505 0.0720** 5.81 Size elasticity estimated from model (3): w2 0.4417 0.5575 0.1158** 8.18 0.4359 0.5609 0.1250** 7.66 Standard error-weighted regression results Non-weighted regression results Price elasticity Size elasticity Price elasticity Size elasticity k1 w1 k2 w2 k1 w1 k2 w2 Panel C: Regression coefficients on the indicator variable representing the post-decimal period Regression coefficient on POSTDECIMAL: b1 0.0890** 0.0688** 0.1064** 0.1187** 0.0815** 0.0709** 0.1093** 0.1317** (7.29) (8.77) (5.95) (9.50) (7.14) (8.64) (5.75) (9.37) R2 0.041 0.143 0.050 0.042 0.023 0.053 0.033 0.021 Number of observations 6424 6424 6424 6424 6424 6424 6424 6424 In this table, we compare the price and size elasticity of dealer market share before and after decimal pricing.

Panel A compares the price elasticity before and after decimalization. Panel B compares the size elasticity before and after decimalization. We consider the two-month period from January 2001 to February 2001 as the pre-decimal period and May 2001 to June 2001 as the post-decimal period. We report the standard error-weighted and equal-weighted mean values. We calculate the standard error-weighted mean values of elasticity estimates from the first- pass regression using the reciprocal of the standard error of the estimated coefficient as weight. The method assigns smaller weights to the first-pass regression coefficients that are less meaningful (i.e., smaller t-statistics).

Panel C shows the estimate of b1 from the following regression model: k1it; k2it; w1it; or w2it ¼ b0 þ b1POSTDECIMAL þ b2 logðPRICEði; tÞÞ þ b3 logðNTRADEði; tÞÞ þ b4 logðTSIZEði; tÞÞ þ b5VOLATILITYði; tÞ þ b6 logðMVEði; tÞÞ þ b7H-INDEXði; tÞ þ eði; tÞ; where k1it, k2it, w1it, and w2it are the estimated coefficients from the first-pass regressions (i.e., regression models (2) and (3)), POSTDECIMAL is an indicator variable which equals one for the post-decimal period and zero otherwise, PRICE(i,t) is the average quote midpoint of stock i in month t, NTRADE(i,t) is the average daily number of transactions of stock i in month t, TSIZE(i,t) is the average dollar trade size of stock i in month t, VOLATILITY(i,t) is the standard deviation of stock i’s daily quote midpoint returns in month t, MVE(i,t) is the market value of equity of stock i in month t, H-INDEX(i,t) is the Herfindahl-index of stock i in month t, and e(i,t) is the error term.

We show the results of the weighted and non-weighted regres- sions. In the weighted regression, we use the reciprocal of standard error from the first-pass regression as weight. Numbers in parenthesis are the t-statistics. a The t-statistic testing the equality of mean values between before and after periods. ** Significant at the 1% level.

2780 K.H. Chung, C. Chuwonganant / Journal of Banking & Finance 31 (2007) 2770–2795

estimates between the two periods may result from changes in stock attributes rather than decimal pricing per se. To examine this possibility, we estimate the following regression model: k1it; k2it; w1it; or w2it ¼ b0 þ b1 POSTDECIMAL þ b2 logðPRICEði; tÞÞ þ b3 logðNTRADEði; tÞÞ þ b4 logðTSIZEði; tÞÞ þ b5VOLATILITYði; tÞ þ b6 logðMVEði; tÞÞ þ b7H-INDEXði; tÞ þ eði; tÞ; ð4Þ where k1it, k2it, w1it, and w2it are the estimated coefficients from the first-pass regressions (i.e., regression models (2) and (3)), POSTDECIMAL is an indicator variable which equals one for the post-decimal period and zero otherwise, PRICE(i,t) is the average quote midpoint of stock i in month t, NTRADE(i,t) is the average daily number of transactions of stock i in month t, TSIZE(i,t) is the average dollar trade size of stock i in month t, VOLATILITY(i,t) is the standard deviation of stock i’s daily quote midpoint returns in month t, MVE(i,t) is the market value of equity of stock i in month t, H-INDEX(i,t) is the Herfindahl-index of stock i in month t, and e(i,t) is the error term.

Panel C of Table 4 shows the estimates of b1 from the weighted and non-weighted regres- sions. In the weighted regression, we use the reciprocal of standard error from the first-pass regression as weight. This method assigns smaller weights to the less meaningful first-pass regression coefficients (i.e., smaller t-statistics). In contrast, in the non-weighted regression, the estimated coefficients from the first-pass regressions have the same weight, regardless of their statistical significance. We employ both approaches to assess the robustness of our results to different estimation methods. Numbers in parenthesis are the t-statistics.

The first four columns show the results of the weighted regressions and the next four columns show the results of the non-weighted regressions. The results show that the esti- mated coefficients on POSTDECIMAL are significantly negative in the price elasticity regression (i.e., when the dependent variable is k1it or w1it) and significantly positive in the size elasticity regression (i.e., when the dependent variable is k2it or w2it) from both the weighted and non-weighted regressions. These results indicate that the observed changes in both the price and size elasticity shown in Panel A and B cannot be attributed to concurrent changes in stock attributes.

To shed additional light on the effect of decimalization, consider a dealer firm that increases its size quote aggressiveness (RELQS) from 0.3450 to 0.9680, which corresponds to a move from the 25th percentile position to the 75th percentile position (see Panel B of Table 1) among dealers in our sample. Let us also assume that the dealer’s initial market share was 0.05. Panel B of Table 4 shows that the mean size elasticity estimated from regression model (3) is 0.4417 and 0.5575, respectively, during the pre- and post-decimal period. These figures indicate that the dealer’s market share during the pre-decimal period would increase by 0.0399 [=0.4417 · {(0.9680  0.3450)/0.3450} · 0.05] with more aggres- sive size quotes.

This is equivalent to a 79.8% (=0.0399/0.05) increase in market share and the dealer’s new market share would be 0.0999 (=0.05 + 0.0399). In contrast, the dealer’s market share during the post-decimal period would increase by 0.0503 [=0.5575 · {(0.9680  0.3450)/0.3450} · 0.05] to 0.1003 (=0.05 + 0.0503) with more aggressive size quotes. In relative term, this is equivalent to a 100.6% (=0.0503/0.05) increase in market share, which is 20.8% (=100.6%  79.8%) greater than the corresponding figure for the pre-decimal period.

K.H. Chung, C. Chuwonganant / Journal of Banking & Finance 31 (2007) 2770–2795 2781

4.3. SuperSOES and the size elasticity of dealer market share The NASDAQ Stock Market implemented the final phase of its automated order exe- cution system, SuperSOES, on July 30, 2001. From this date forward, orders for all 3500+ NASDAQ National Market (NNM) securities were automatically executed using Super- SOES.10 SuperSOES is a revised version of the Small Order Execution System (SOES), NASDAQ’s auto-execution system for small agency orders. SuperSOES accepted orders of up to 999999 shares at a time, allowing traders to execute large orders in a single trans- action.

SuperSOES also provided a more robust trading platform that could handle increased trading volumes and reduced dual liability for market makers, encouraging the display and execution of larger orders.

SuperSOES was a significant change in how market participants interact with NAS- DAQ to obtain quotes and execute trades. SuperSOES made accessing liquidity more effi- cient, allowing participants to access large blocks of shares automatically in one order. Previously, numerous trades had been required to achieve the same result. SuperSOES could access several quotes with one order as well as access more size (up to 999999 shares), whereas SelectNet accessed one participant per order and the prior SOES share limit was 1000 shares for automatic execution. Thus, SuperSOES significantly reduced the number of orders for a given volume of activity.

Based on these considerations, we conjecture that thesize elasticity of market share after the implementation of SuperSOES is greater than the corresponding figure before the implementation.11 Panel A of Table 5 shows the size elasticity of dealer market share in the periods before and after SuperSOES implementation. We consider the two-month period from May 2001 to June 2001 as the pre-SuperSOES period and August 2001 to September 2001 as the post-SuperSOES period. According to both the standard error- weighted and equal-weighted results, the mean values of size elasticity during the post- SuperSOES period are significantly greater than the corresponding values during the pre-SuperSOES period, regardless of different model specifications.

These results are con- sistent with our expectation that the introduction of SuperSOES increased the impact of size aggressiveness on dealer market share.

To assess the economic significance of SuperSOES, consider a dealer firm that increases its size quote aggressiveness (RELQS) from 0.3450 to 0.9680. Let us also assume that the dealer’s initial market share was 0.05. Panel A of Table 5 shows that the mean size elas- ticity estimated from regression model (2) is 0.6421 and 0.9834, respectively, during the pre- and post-SuperSOES period. These figures indicate that the dealer’s market share during the pre-SuperSOES period would increase by 0.058 [=0.6421 · {(0.9680  0.3450)/0.3450} · 0.05] with more aggressive size quotes. This is equivalent to a 116% (=0.058/0.05) increase in market share and the dealer’s new market share would be 0.108 (=0.05 + 0.058).

In contrast, the dealer’s market share during the post-SuperSOES 10 The first pilot in the conversion to the SuperSOES trading system was launched on Monday, July 9, 2001, when NASDAQ implemented 18 pilot SuperSOES securities and two test stocks. The second pilot conversion of additional 80 NASDAQ stocks was launched on July 16, 2001.

11 The main difference between SOES and SuperSOES is that the latter allowed market participants to access large blocks of shares automatically in one order, compared with the pre-SuperSOES environment. Hence, we do not examine the effect of SuperSOES on price elasticity. 2782 K.H. Chung, C. Chuwonganant / Journal of Banking & Finance 31 (2007) 2770–2795

Table 5 Comparison of the size elasticity before and after SuperSOES implementation Standard error-weighted mean values Equal-weighted mean values Before After Difference (after  before) t-Statistica Before After Difference (after  before) t-Statistica Panel A.

Comparison of the size elasticity of dealer market share before and after the implementation of SuperSOES Size elasticity estimated from model (2): k2 0.6421 0.9834 0.3413** 13.97 0.6505 0.9822 0.3317** 14.06 Size elasticity estimated from model (3): w2 0.5575 0.7551 0.1976** 7.58 0.5609 0.7774 0.2165** 8.68 Panel B. Comparison of the size elasticity of dealer market share before and after Island’s departure from NASDAQ Size elasticity estimated from model (2): k2 0.5716 0.4057 0.1659** 7.47 0.5761 0.4119 0.1642** 9.02 Size elasticity estimated from model (3): w2 0.3785 0.2059 0.1726** 13.21 0.3961 0.2222 0.1739** 11.54 Standard error-weighted regression results Non-weighted regression results Size elasticity Size elasticity k2 w2 k2 w2 Panel C: Regression coefficients on the indicator variable representing the post-SuperSOES period Regression coefficient on POSTSUPERSOES: b1 0.3483** 0.2050** 0.3388** 0.2246** (15.26) (9.51) (13.63) (9.60) R2 0.099 0.069 0.061 0.044 Number of observations 6424 6424 6424 6424 Panel D: Regression coefficients on the indicator variable representing Island’s departure from NASDAQ Regression coefficient on POSTISLAND: b1 0.1684** 0.1687 0.1646** 0.1698 (9.28) (10.32) (8.29) (9.43) R2 0.009 0.021 0.006 0.013 Number of observations 6424 6424 6424 6424 Panel A compares the size elasticity of dealer market share in the periods before and after SuperSOES imple- mentation.

We consider the two-month period from May 2001 to June 2001 as the period before SuperSOES and August 2001 to September 2001 as the period after SuperSOES. Panel B compares the size elasticity of dealer market share before and after Island ECN’s departure from NASDAQ in August 2002. We consider the two- month period from June 2002 to July 2002 as the pre-Island departure period and August 2002 to September 2002 as the post-Island departure period. We report the standard error-weighted and equal-weighted mean values. We calculate the standard error-weighted mean values of elasticity estimates from the first-pass regression using the reciprocal of the standard error of the estimated coefficient as weight.

The method assigns smaller weights to the first-pass regression coefficients that are less meaningful (i.e., smaller t-statistics). Panels C and D show the estimates of b1 from the following regression model: k2it or w2it ¼ b0 þ b1POSTSUPERSOESðor POSTISLANDÞ þ b2 logðPRICEði; tÞÞ þ b3 logðNTRADEði; tÞÞ þ b4 logðTSIZEði; tÞÞ þ b5VOLATILITYði; tÞ þ b6 logðMVEði; tÞÞ þ b7H-INDEXði; tÞ þ eði; tÞ; where k2it and w2it are the estimated coefficients from the first-pass regressions (i.e., regression models (2) and (3)), POSTSUPERSOES is an indicator variable representing the post-SuperSOES period and POSTISLAND is an (continued on next page) K.H.

Chung, C. Chuwonganant / Journal of Banking & Finance 31 (2007) 2770–2795 2783