Reported Trading Volume on the NYSE and NASDAQ

Phillip R. Daves The University of Tennessee Department of Finance, 435 SMC Knoxville, TN 37996 (865) 974-1724 (Voice) (865) 974-1716 (Fax) [email protected]

James W. Wansley The University of Tennessee Department of Finance, 428 SMC Knoxville, TN 37996 (865) 974-1724 (Voice) (865) 974-1716 (Fax) [email protected] Rongrong Zhang1 The University of Tennessee Department of Financ e, 439 SMC Knoxville, TN 37996 (865) 974-1730 (Voice) (865) 974-1716 (Fax) [email protected]

May 2003 An earlier version of this paper was presented at the 2002 Eastern Finance Association meetings in Baltimore, MD. The authors are grateful for helpful comments from Jonathan Clarke, Tim McCormick (NASD) and Jeffrey Smith (NASD).

1

Corresponding author.

Abstract Reported Volume on the NYSE and NASDAQ

The literature on asset markets has devoted surprisingly little attention to trading volume. [see Harris and Raviv (1993)]. We seek to fill a portion of this gap by examining the determinants of trading volume for individual securities in two distinct markets: the New York Exchange and the NASDAQ market. We are particularly interested in using this information about the determinants of trading volume to examine how double counting on NASDAQ has changed as a result of the implementation of order handling rules. Because of this double counting, recent literature has treated volume in these two markets separately. We find that the median weekly turnover for NYSE securities (during 1996-2000) is significantly related to (a) the firm’s beta estimated from the OLS market model (b) the standard deviation of residuals from the OLS market model regression, (c) the natural log of its average price, (d) the natural log of its average market capitalization, (e) its first-order autocovariance of returns, (f) whether or not it is a member of the S&P 500, (g) whether or not options trade on this stock, and (h) institutional ownership. Turnover on NASDAQ is related in the same way to these independent variables, although the values of the coefficients are different. We find that the overall level of double counting on NASDAQ was relatively constant through 1999, and then declined in 2000.

2

Reported Volume on the NYSE and NASDAQ

1. Introduction The literature on asset markets has devoted surprisingly little attention to trading volume. [see Harris and Raviv, (1993)]. We seek to fill a portion of this gap by first developing a cross-sectional regression model to explain the determinants of trading volume for individual securities in two distinct markets: the New York Exchange and the NASDAQ market. We next use this model in the two markets to explore whether there have been changes in the way trading volume is reported on the NYSE and NASDAQ as NASDAQ order-handling rules have been implemented. In conventional models of asset pricing where the asset market is complete and there exists a representative agent, asset allocation is optimal and asset prices are determined by aggregate risk only. Trading volume plays only a minor role. However, in an incomplete asset market, both aggregate and individual risk affect equilibrium prices, and their behavior depends on the nature of investor heterogeneity. In this case, trading volume plays a more important role and conveys information about the pricing of assets [see Wang (1994)]. In addition, trading volume and measures of information and liquidity will be related. Bessembinder et al. (1996) find substantial cross-sectional variation in the relation between trading volume and volatility measures. The trading volume for individual stocks is more closely related to a proxy for firm-specific information while the trading volume for the S&P500 futures contract is more closely related to a proxy for market information. Chordia, Roll and Subrahmanyam (2001) examine the time-series behavior of liquidity. They find that trading activity increases prior to scheduled announcements of GDP and the unemployment rate and falls back to normal levels on the announcement day. Trading activity also responds

3

to short-term interest rates, the term spread and equity market returns. Chordia, Roll and Subrahmanyam (2000) investigate the common determinants of liquidity and find that trading volume is negatively related to spread. If trading volume is economically important, then differences in its determinants from market to market may reflect segmentation among the markets. This paper further explores the relation between trading volume, information, and firm and market characteristics by examining the cross-sectional determinants of trading volume for individual firms in two distinct markets: the NYSE, which is primarily an auction market and NASDAQ, which is primarily a dealer market. Volume comparisons between these two markets are made difficult because the trade-reporting conventions are different in auction and dealer market; reported volume on the NYSE is not directly comparable to reported volume on NASDAQ. In a dealer market, market makers post bid and offer prices and buy shares from or sell shares to public investors at these prices. In this way, a dealer is on one side of each trade. In an auction market, most transactions occur between an actual buyer and seller. Thus, on the NASDAQ market, when 100 shares of stock move from one public investor to another, these shares usually pass through a dealer who buys and resells them. Two transactions result and 200 shares of volume are recorded. Dyl and Anderson (2002) indicate that the historical rule of thumb is that NASDAQ trading volume is roughly double-counted compared to NYSE volume. [See also Atkins and Dyl (1997) and Wansley, Daves and Stewart (2002)]. 2 Previous literature has investigated portfolio theory’s implications for trading volume. For example, Tkac (1999) provides a rebalancing benchmark model for trading volume that relates individual stocks’ trading activity to market-wide volume. Two- fund separation

2

Some double counting exists on NYSE also. Atkins and Anderson (2002) report that from 1997 to 2000 specialists participated in approximately 25% of all transactions on the NYSE.

4

suggests that trading volume, as measured by turnover ratio, should be equal across stocks and equal to the market turnover ratio. Tkac (1999) finds systematic deviations of firms’ turnover from the markets’ turnover that are negatively related to size and positively related to institutional ownership and option availability. Lo and Wang (2000) investigate the implication of portfolio theory for the cross-sectional trading volume for NYSE and AMEX ordinary shares from 1962 to 1996. They find turnover ratios vary considerably in the cross section of NYSE and AMEX stocks. However, Tkac (1999) only includes NYSE firms in CRSP deciles seven through ten (large capitalization stocks) 3 . Both Tkac (1999) and Lo and Wang (2000) omit NASDAQ stocks altogether. Considering the following comment from Lo and Wang (2000): We also omit NASDAQ stocks altogether since the difference between NASDAQ and the NYSE (market structure, market capitalization, etc.) have important implications for the measurement of trading volume [see, e.g., Atkins & Dyl (1997)], and this should be investigated separately. The double counting of trading volume for NASDAQ stocks and the difficulty that this induces in comparisons of trading volume between exchanges has also been documented in several studies. Gould and Kleidon (1994) use Time and Sales Reports (TMTR) intraday transaction records to measure investor-to-investor trading volume for firms on NASDAQ. They find that for five randomly selected days between April and June 1994 for all NASDAQ stocks, only 42 percent of reported volume represented investor-to- investor trades. Atkins & Dyl (1997) examine a sample of firms switching from NASDAQ to NYSE.

3

Her sample accounts for an average of 83% of aggregate NYSE/AMEX market capitalization and 87% of aggregate NYSE/AMEX monthly trading volume.

5

They document that the firms’ trading volume drops to about 50 percent of the previously reported NASDAQ trading volume. However, any simple rule of thumb will be inadequate to fully describe NASDAQ reported volume over time because there has been an explosion in the use of electronic order crossing services, commonly referred to as electronic communications networks (ECNs), which allow investors to cross orders electronically without the intervention of a dealer. And to further complicate matters, the reporting conventions for dealer-mediated transactions, along with the requirements for routing orders, have been changing since 1997 as order handling rules have come into force. Specifically, public limit orders now compete directly with market makers’ quotes, and public traders may obtain pricing and quotes from electronic communications networks (ECNs) and thereby bypass market makers entirely. In addition, starting in 2001, market makers who engage in a riskless principal transaction whereby they purchase (or sell) on the open market securities needed to satisfy a customer buy (or sell) order will report the transactions as a single trade, which effectively eliminates the double counting for those trades. As a result of these changes, Dyl and Anderson (2000) hypothesize that the systematic differences in transactions reporting between NYSE and NASDAQ have declined over time. If this is the case, then the meaningful comparison of volume on the two different markets will require year-by- year adjustments to account for the time- varying effects of the reporting conventions. Double-counting is not the only difference in reported volume between the exchanges and NASDAQ. Freund and Webb (1999) observe a difference in the time series of average daily volatility measures between NASDAQ and the NYSE. On the NYSE, volume is positively correlated with market wide sources of risk while on NASDAQ, volume is

6

positively correlated with the stock-specific variance measures. They suggest that the size of NASDAQ volume relative to NYSE volume cannot be attributed solely to differences in the methods of counting volume in the two market environments. The type and quantity of information driving trading may be different across these two distinct markets, and so we develop a model that allows for a variety of determinants of volume. In so doing, we adopt the suggestion of Lo and Wang (2000) and investigate separately the determinants of trading volume, as measured by turnover, for NYSE and NASDAQ firms from 1996-2000. Following Lo and Wang (2000), we use weekly turnover, as measured by cumulative daily turnover 4 , to measure trading volume. Our samples 5 consist of 1647 individual securities on the NYSE and 1816 individual securities on NASDAQ from 1996 through 2000. We use the median weekly turnover over the entire five-year period for each individual stock as the dependent variable and a series of firm-specific characteristics as independent variables. We run cross-sectional regressions on NYSE securities and NASDAQ securities separately to gauge the differences in the determinants of trading in the two distinct markets. We then run a cross-sectional regression on the combined NYSE and NASDAQ samples that corrects for a fixed multiple-counting factor on NASDAQ, and estimate this factor from year to year. The remainder of the paper is organized as follows. Section 2 discusses our model and hypotheses. Section 3 describes the data. Section 4 reports the empirical results of testing the model and discusses the implications of the results for future empirical studies. Section 5 concludes.

4

Daily turnover is calculated as shares traded divided by shares outstanding. In an earlier version of our paper, we ran regression without sp500 membership, institutional ownership and options availability variables. Our sample for Nasdaq firms consisted of 5109 firms. Our sample for NYSE/AMEX consisted of 2295 firms. The regression results are similar to the reported results here. 5

7

2. Methodology 2.1 Turnover (Tov) Previous studies have used turnover, defined as the ratio of the number of shares traded during a day to the number of shares outstanding at the end of a day [see Campbell, Grossman, and Wang (1993), Lo and Wang (2000)], to measure trading volume. Lo and Wang (2000) demonstrate that various measures of trading activity, such as share turnover (share volume divided by shares outstanding), dollar turnover (dollar trading volume divided by market capitalization), equal- weighted turnover, value-weighted turnover, and shareweighted turnover for a two-asset portfolio, are identical. They conclude that turnover ratio is the most natural measure of trading activity. In this paper, turnover (Tovit ) is defined as follows: Tovit =

Vit Sit

(1)

Where Vit is the share volume for stock i on day t, and Sit is the outstanding shares for stock i on day t. We obtain the cumulative weekly turnover for individual stocks by summing daily turnover in each week. Then we use median weekly turnover over 260 weeks over the period from 1996 to 2000 as our measure of turnover. 2.2 Determinants of Trading Volume We rely on finance theory and prior empirical results to suggest the determinants of median weekly turnover. We include ten firm- specific characteristics as regressors as described below. Tables 1 and 2 provide summary statistics for turnover and the regressors for both the NYSE sample and the NASDAQ sample.

8

2.2.1 Excess Expected Return (alpha) Alpha is the intercept from the OLS market regression of stock i’s return on the value-weighted market return. Alpha captures the liquidity premium and any information heterogeneity [see Wang (1994)]. Turnover should be negatively related to the liquidity premium, i.e., low liquidity stocks should have lower turnover. In the model of competitive stock trading developed by Wang (1994), information heterogeneity may result in either high trading volume or low trading volume. Therefore turnover may be positively or negatively correlated with the premium for information heterogeneity. Lo and Wang (2000) find mixed results on the impact of excess expected return on turnover. During some periods the association is positive, during other periods it is negative. 2.2.2

Systematic Risk (beta) and Residual Risk (residual) Beta measures systematic risk, and residual is the standard deviation of residuals

from the OLS market model. Lo and Wang (2000) find that beta and residual are positively related to turnover. Freund and Webb (1999) document that NYSE trading volume is more closely related to market wide volatility as measured by the variance of the market index. In contrast, NASDAQ volume is more closely correlated with stock-specific volatility as measured by residual variance. They conclude that volume is related to different components of risk in these two markets. This suggests that the type and quantity of information that drives trading may be different between NASDAQ and NYSE markets. Estimates of alpha, beta and residual are developed from OLS regressions of each individual stock on the CRSP value-weighted index from 1996 to 2000. .

9

2.2.3

Average Price (price) Price is measured as the average of the natural log of stock i’s price over the period

of 1996 through 2000. Since trading costs are inversely related to price, turnover should be positively related to price. Lo and Wang (2000) find that price and turnover are positively related. 2.2.4

Average Market Capitalization (capitalization) Capitalization is defined as the average of the natural log of stock i’s market

capitalization over the period from 1996 to 2000. Lo and Wang (1998) develop the intuition that if size and price drive expected return, they should drive turnover as well. Tkac (1999) finds that firm size has a negative impact on trading volume. However, her sample consists of NYSE stocks from CRSP deciles seven through ten (large capitalization stocks), so her finding may not apply to the universe of stocks. Lo and Wang (2000) find size is negatively related to turnover during the period from 1962 to 1971 and is positively related to turnover after 1971. They suggest that with the growth of the mutual fund industry and other large institutional investors, it has become more difficult to invest in small-capitalization companies because of liquidity and corporate control issues. This has led to more active trading in large-capitalization stocks than in small-capitalization stocks. 2.2.5

Average Dividend Yield (divyield) Divyield is the average dividend yield of stock i over the period from 1996 to 2000.

Due to differential taxation of dividends and capital gains, dividend-capture trading occurs-traders purchase a stock before its ex-dividend date and sell it shortly after. A number of studies have documented dividend-capture trading. For example, Green (1980) examines the trading volume around ex-dividend days and documents evidence for tax- induced clientele

10

effects. Koski and Scruggs (1998) also analyze trading volume around ex-dividend days. They find evidence of significant abnormal volume by securities dealers that is positively related to dividend yield and negatively related to transaction costs. If dividend-capture trading is prevalent, this suggests a positive relation beteen divyield and turnover. 2.2.6

First-order Autocovariance of Returns (autocov) Lo and Wang (2000) find that the first-order autocovariance of returns is positively

correlated with turnover. This is consistent with the trading cost interpretation since firstorder autocovariance is used as a proxy for trading costs [see Roll (1984)]. A large negative autocovariance implies a large bid/ask spread which should imply lower turnover. Our firstorder autocovariance is also obtained from the OLS market model regression. 2.2.7

S&P 500 Membership (sp500) A number of studies have examined the effects of S&P 500 membership on price or

volume [see Lamoureux and Wansley (1987), Pruitt and Wei (1989), Tkac (1999), Lo and Wang (2000)]. Pruitt and Wei (1989) find a positive price impact that is related to changes in institutional demand in response to inclusion in the S&P 500 index. A large portion of S&P 500 stocks are held by passive indexers, these stocks may experience lower turnover. Tkac (1999) finds that S&P 500 inclusion does not significantly increase the trading of firms that are already trading above benchmark levels, but does result in additional trading for firms that undertrade the benchmark prior to inclusion. Lo and Wang (2000) document a positive relationship between S&P 500 membership and individual stocks’ turnover ratios. Firms listed on the NYSE generally have larger market capitalization than those on NASDAQ firms, and are more likely to be included in indexes such as the S&P 500. As such, trading in index derivatives is more likely to affect the trading activity of NYSE firms

11

than NASDAQ firms. Given the growth of indexation and index arbitrage activities, we include S&P 500 membership as an independent variable to explore its effect on trading volume in the two distinct markets. We use a binary variable, which takes on a value of 1 if the stock is included in the S&P 500 index during any quarter between 1996 and 2000, otherwise it has a value of zero. The S&P 500 membership information is obtained from the COMPUSTAT files. 2.2.8

Option Availability (option) Kumar, Sarin and Shastri (1995) find that trading volume, volatility, and bid-ask

spreads decline for the stocks in the Nikkei 225 Index after the listing of index options. They suggest that the advent of options trading causes a migration of speculative and market-wide information-oriented trading activity from the underlying market to the options market. On the other hand, Tkac (1999) finds that option availability is positively correlated with trading volume. Options complement stock market trading activity and provide greater liquidity to stocks since their presence attracts information traders and increases the trading volume of underlying stocks. We use the Standard & Poor’s Common Stock Guide (1996 to 2000) to identify stocks with options traded for each individual year. Then we set the dummy variable option equal to one if there is option traded on this stock for at least three out of the five-year period, otherwise option takes on a value of 0. 2.2.9

Institutional Ownership (institution) Institution is defined as the percentage of common shares held by institutional

investors and is taken from the monthly Standard & Poor’s Common Stock Guide from 1996 to 2000. Many researchers have examined the trading behavior of institutional investors and individual investors [see Chan and Lakonishok (1995), Cready and Utama (1996), Tkac

12

(1999) and Nofsinger (2001)]. Cready and Utama (1996) find trading volume around earning announcement dates responds as a quadratic function of institutional ownership and reaches a maximum at around 50% institutional ownership. Tkac (1999) shows that institutional ownership and turnover ratios are positively correlated, which suggests that institutional investors engage in more active trading strategies. Nofsinger (2001) finds that institutional investors and individual investors have different trading behaviors in their response to firmspecific news releases and macro-economic announcements. These studies demonstrate that institutional investors actively manage their portfolios. This suggests that the percentage of institutional ownership may be positively related to turnover. To assess the determinants of turnover, we run the following cross-sectional regression model on our NYSE sample and NASDAQ sample separately:

Tovi = β 0 + β1 *α lphai + β 2 * betai + β3 * residual i + β 4 * pricei + β5 * capitalizationi (2) + β 6 *divyieldi + β7 * auto covi + β8 * sp 500 i + β9 * optioni + β10 * institutioni + νi 3. Data We use the CRSP Daily Master Files to construct a weekly turnover measure for individual NYSE and NASDAQ stocks from January 1996 to December 2000 (260 weeks). We exclude ADRs, SBIs, REITS, and closed-end funds. We require that each stock in our sample have turnover data for at least one-third of the five- year period, i.e., at least 87 weekly turnovers. Summary statistics for the 1647 firms in our NYSE sample and the 1816 firms in our NASDAQ sample appear in Tables 1 and 2. __________________________________ Tables 1 and 2 here __________________________________

13

The mean value for Tov (median weekly turnover) for firms trading on the NYSE is 1.39 percent compared to 3.12 percent for NASDAQ firms. This difference in mean weekly turnover is roughly consistent with the decline in trading volume when firms move from NASDAQ to either the NYSE or AMEX of approximately 50 percent reported by Atkins and Dyl (1997) and Wansley et al. (2002)6 . As expected, the average price, market capitalization and dividend yield are all higher in the NYSE sample. The beta coefficient and the standard deviation of residuals are higher in the NASDAQ sample, and both samples exhibit negative first-order autocovariance. Although two- fund separation suggests that trading volume, as measured by the turnover ratio, will be equal across stocks and equal to the market turnover ratio, we find considerable variation in weekly turnover within our samples. For the NYSE sample, median turnover ranges from a low of 0.005 percent to 10.19 percent. For the NASDAQ sample turnover ranges from 0.007 percent to 28.12 percent. Thus, on the NYSE (NASDAQ) the most actively traded stocks trade 2000 (4000) times as often as the least actively traded stocks, relative to their shares outstanding. Clearly, there are cross-sectional differences in firm turnover within the two markets that we examine.

4. Empirical Results 4.1 Regression Results for NYSE and NASDAQ

6

In an earlier version of our paper, we included AMEX firms with our NYSE sample. In that version we have a larger sample of 2295 NYSE/AMEX firms and 5109 Nasdaq firms. Our final sample in this paper is substantially reduced when we impose the data requirement of information availability on institutional ownership and option availability from the S&P Stock Guide. Higher turnover on Nasdaq is consistent with the empirical findings of Atkins and Dyl (1997) and Wansley et al. (2002). Atkins and Dyl report that volume drops to about 50 percent when a firm moves from Nasdaq to NYSE.

14

Our initial regression results for the NYSE sample are presented in Table 3. Table 4 presents the results for the NASDAQ sample.

___________________________________ Tables 3 and 4 go here ___________________________________ Our cross-sectional regressions of weekly median turnover have considerable explanatory power. For the NYSE sample the adjusted R2 is 43 percent, and is 54 percent for the NASDAQ sample. In Table 3 we see that an NYSE firm’s weekly median turnover is positively related to (a) the firm’s beta estimated from the OLS market model, (b) the standard deviation of residuals from the OLS market model regression, (c) the natural log of its average price, (d) the natural log of its average market capitalization, (e) its first-order autocovariance of returns, (f) whether or not it is a member of the S&P 500, (g) whether or not options trade on this stock, and (h) institutional ownership (the percentage shares held by institutional investors). Turnover for NYSE firms is not significantly related to the firm’s average dividend yield (divyield) or its intercept from the market model OLS regression (alpha ). Table 4 shows that turnover for NASDAQ stocks is also significantly related to the same factors as in the NYSE sample. And, as with the NYSE sample, divyield and alpha are not significant. The signs and magnitudes of the coefficients in the NYSE and NASDAQ regressions are similar, and the two models have similar explanatory power, suggesting that turno ver is related to the independent variables in the same way, with one notable difference being the double counting of volume (and hence turnover) on NASDAQ relative to the

15

NYSE. Next, we explicitly account for double counting on NASDAQ and use the same model on both NASDAQ and NYSE stocks. 4.2 Pooled regression results for the NYSE and NASDAQ samples Equation (2) cannot be run on the pooled NYSE and NASDAQ samples because the double counting of trades on NASDAQ 7 causes the left hand side variables to be measured differently in the two markets. If the level of double-counting of trades for each firm trading on NASDAQ also depends arbitrarily on the independent variables in equation (2), then the estimated coefficients will include the effects of the cross sectional variation in double counting as well as the effects of the cross sectional variation in turnover. In the absence of a model of the cross sectional variation in double counting [see Wansley et al. (2002) and Dyl and Anderson (2002)], it would be then impossible to identify separately the coefficients that determine turnover for NASDAQ stocks. However, if the overall level of double counting for all firms on NASDAQ is the same, or is at least unrelated to the chosen independent variables, then it is possible to estimate this overall level. Under this assumption, if the double-counting factor for volume on NASDAQ is λ, then this is the same as assuming: Reported volume on NASDAQ = λ * NYSE-equivalent volume

(3)

We can rewrite equation (2) for NASDAQ stocks:

(Tovi ) = β0 + β1 *α lphai + β 2 * betai + β 3 * residual i + β 4 * pricei + β5 * capitalizationi λ + β 6 *divyieldi + β7 * auto covi + β 8 * sp500 i + β 9 * optioni + β10 * institutioni +ν i

(4)

Equation (4) can be more conveniently written in terms of the calculated turnover as: 7

As explained in Section 1, double counting of trades also exists on the NYSE, albeit at a lower level. For a more complete discussion, see Atkins and Anderson (2002), Atkins and Dyl (1997), Smith et al. (1997), and Wansley et al. (2002).

16

Tovi = λ( β0 + β1 *α lphai + β2 * betai + β 3 * residuali + β4 * pricei + β5 * capitalizationi + β 6 *divyieldi + β7 * auto covi + β8 * sp 500 i + β9 * optioni + β10 * institutioni ) +ν i

(5)

For NYSE stocks, equation (2) is used without modification. Equation (2) and equation (5) can now be combined and estimated using nonlinear least squares in the following model:

Tovi = ((λ − 1)* nas + 1)( β0 + β1 *α lphai + β 2 * betai + β 3 * residuali + β 4 * pricei + β5 * capitalizationi + β 6 *divyieldi + β 7 * auto cov i + β 8 * sp500 i + β9 * option i + β10 * institutioni ) +ν i (6)

Where nas = 1 for NASDAQ stocks, and nas = 0 for NYSE stocks. When nas = 1, equation (6) estimates the linear coefficients β 0 through β 10 and the multiplicative coefficient λ. When nas = 0, for NYSE stocks, equation (6) estimates only β 0 through β 10 . The estimate of λ is the overall effect of double counting on NASDAQ, which is assumed to be the same for all firms on NASDAQ. Table 5 shows the results of the non- linear least squares estimation of equation (6). Notice that the coefficients are generally of the same sign and significance as the coefficients in Tables 3 and 4. The estimated value of λ is 2.19 implying that on average there are 2.19 trades on NASDAQ per trade on NYSE—1.19 extra inter-dealer trades occur for each investor-to-investor trade during this time period. This is roughly consistent with the results of Atkins and Dyl (1997), and very similar to the results obtained by Wansley et al. (2002). 4.4 S&P 500 Effect The results in Tables 3,4 and 5 show that S&P 500 membership is consistently positively related to turnover both on NASDAQ and on the NYSE. Tkac (1999) found that the positive impact of S&P membership on turnover only occurred for firms that had 17

turnover less than the market turnover prior to inclusion in the S&P 500. We found similar results for our two samples. We ran the regressions in equation (2) separately for firms with turnover less than the sample median and for firms with turnover greater than the sample median, both for NYSE and NASDAQ stocks. For NYSE stocks, S&P 500 membership is positively related to turnover for low turnover stocks but not significantly related to the high turnover stocks. In contrast, although S&P 500 inclusion is significant for the entire NASDAQ sample, it is not significant in either the high trading or low trading sub-sample. 4.5 Potential Bias Since we obtain information on option availability and institutional ownership information from Standard and Poor’s Common Stock Guide, we may introduce a bias by including only larger firms with higher turnover ratios in our sample. Larger firms are more actively followed by financial analysts and are more likely to be included in Standard and Poor’s Common Stock Guide. Therefore we estimate nonlinear least squares regression equation (6) for 2295 NYSE firms and 5109 firms without the option and institution variables. The significance level and magnitude of coefficients (not reported) for the other eight regressors are very similar to the results we reported here. So our reduced samples do not appear to bias our results. 4.6 Changes in Market Structure Atkins and Anderson (2002) discuss several changes in market structure that occurred during the time period of this study (1996-2000) that have the potential to affect the extent of dealer participation in trades and thus the overcounting of volume on NASDAQ relative to that on the NYSE. These changes include (a) public limit orders now compete directly with market makers’ quotes, (b) pricing and quotes from electronic communications networks

18

(ECNs) are now available to the public, (c) changes were made in quote procedures beginning in 1997 and the market moved toward decimilization by 2001. 8 Atkins and Anderson find some decrease in overcounting of volume that occurs on NASDAQ compared to the NYSE, although they conclude the overcounting is still substantial. In order to determine whether the results reported in Table 5 are affected by changes in market structure over time, we re-estimate on an annual basis equation (6) for the pooled sample of NYSE and NASDAQ firms. 9 If changes in the NASDAQ market structure that occurred between 1996 and 2000 reduced the level of overcounting on NASDAQ, relative to that on NYSE, we would expect to see a decline in λ. Our results are presented in Table 6 where we report the estimated λ, and the percent volume difference implied by each lambda. For example, a λ of 2.0 implies that reported volume on the NASDAQ is twice the level of reported volume on NYSE, holding constant other determinants. By construction, the percent of volume double-counting implied by lambda is [1 – 1/λ], and Table 6 shows that the estimated percent of volume reduction for NYSE firms relative to NASDAQ firms decreases from 59 percent in 1996 through 1999 to 47 percent in 2000. Thus there is some evidence suggesting that the overcounting of NASDAQ trades has been reduced, at least by 2000. 5. Conclusions and Future Research

8

Another significant change in market structure occurred in 2001 when Nasdaq adopted a Riskless Principal Trade-Reporting rule to minimize the degree of double counting of customer limit orders. This change, of course, occurred after the time period of this study. 9 The regression results reported in Table 6 are estimated using weekly observations over each of the years reported in the table. Thus, rather than a maximum of 260 observations in the pooled regressions of Table 5, we have a maximum of 52 observations for each firm.

19

Our results support the NYSE findings of Tkac (1999) and Lo and Wang (2000) indicating that turnover is not the same from stock to stock. Two fund separation does not hold for firms that trade either on the NYSE or on NASDAQ. In investigation of the cross-sectional determinants of turnover, we find two important results: first, raw turnover on NYSE and NASDAQ are determined by the same set of independent variables and firm characteristics, and the signs of the coefficients and their relative magnitudes are similar between the two different market structures. Second, we find that when we account for the double counting inherent in the dealer market we find some evidence that the overcounting of trading volume reported in prior work has declined somewhat, likely due to structural changes in the NASDAQ market. Our estimate of this overcounting in 2000 is 47 percent. This estimate is somewhat higher than the 2001 estimate of 38 percent found by Atkins and Anderson (2002). One plausible reason for the difference in these two estimates (besides the difference in years) is that our estimate is taken from a pooled cross-sectional regression that accounts for other firm-specific determinants of volume.

20

Reference: Andersen, Torben G., 1996, Return Volatility and Trading Volume: An Information Flow Interpretation of Stochastic Volatility, Journal of Finance 51 (March). Atkins, Allen B. and Edward Dyl, 1997, Market Structure and Reported Trading Volume: vs. the NYSE, Journal of Financial Research, Vol. XX, no. 3, 291-304. Bali, R. and G. L. Hite, 1998, Ex-dividend day stock price behavior: Discreteness or tax-induced clienteles? Journal of Financial Economics 47, 127-159. Barclay, Michael J., Eugene Kandel and Leslie M. Marx, 1998, The Effects of Transaction Costs on Stock Prices and Trading Volume, Journal of Financial Intermediation 7. Bessembinder, H. K. Chan and P. Seguin, 1996, An Empirical Investigation of Information, Differences of Opinion, and Trading Activity, Journal of Financial Economics 40 (January), 105-134. Campbell, John Y., Sanford J. Grossman, Jiang Wang, 1993, Trading Volume and Serial Correlation in Stock Returns, Quarterly Journal of Economics 108, (November). Chan, L., and J. Lakonishok, 1995, The Behavior of Stock Prices around Institutional Trades, Journal of Finance, 50, 1147-1174. Chordia, Tarun, Richard Roll and Avanidhar Subrahmanyam, 2000, Commonality in Liquidity, Journal of Financial Economics, Volume 56, (April). Chordia, Tarun, Richard Roll and Avanidhar Subrahmanyam, 2001, Market Liquidity and Trading Activity, Journal of Finance, Volume 56, (April). Cready, William and Siddharta Utama, 1996 Institutional Ownership, Differential Predisclosure Precision and Trading Volume at Announcement Dates, Louisiana State University, University of Indonesia, Working Paper Series. Dyl, Edward and Anne Anderson, 2002, Market Structure and Trading Volume, University of Arizona , February. Fama, Eugene F. and Kenneth R. French, 1992), The Cross-Section of Expected Stock Returns, Journa l of Finance 47, (June). Freund, Steven, and Gwendolyn P. Webb, 1999, Recent Growth in Trading Volume and Its Relation to Market Volatility, The Journal of Financial Research 22, (Winter), 489-501. Gould, John F. and Allan W. Kleidon, 1994, Market Maker Activity on : Implications for Trading Volume, Stanford Journal of Law, Business and Finance 1,

21

(Fall). Green, Jerry, 2001, Taxation and the Ex-Dividend Day Behavior of Common Stock Prices, Harvard University Working Paper Series. Harris, M. and A. Raviv, 1993, Differences of Opinion Make a Horse Race, Review of Financial Studies 6, 473-506. Lamoureux, C., and J. Wansley, 1987, Market Effects of Changes in the Standard & Poor’s 500 Index, The Financial Review 22, 53-69 Liljeblom Eva, Löflund Anders, Hedvall Kaj, 2001, Foreign and Domestic Investors and Tax- induced ex-dividend Day Trading, Journal of Banking and Finance 25, 16871716 Lim,T., A. Lo, J. Wang, and P. Adamek, 1998, Trading Volume and the MiniCRSP Database: An Introduction and User’s Guide, Working paper LFE-1038-98, MIT Laboratory for Financial Engineering. Lo, Andrew, and Jiang Wang, 1998a, Trading Volume: Implication of an Intertemporal Capital Asset-Pricing Model, work in progress, MIT. Lo, Andrew, and Jiang Wang, 2000, Trading Volume: Definition, Data Analysis and Implication of Portfolio Theory, Review of Financial Studies 13, (Summer), 257-300. Lo, Andrew, and Jiang Wang, 2000, Foundations of Technical Analysis: Computational Algorithms, Statistical Inference, and Empirical Implementation, forthcoming in Journal of Finance. Nofsinger, John R., 2001, The Impact of Public Information on Investors, Journal of Banking and Finance 25, 1339-1366. Raman, Kumar, Sarin Atulya, and Shastri Kuldeep, 1995, The impact of index options on the underlying stocks: The evidence from the listing of Nikkei Stock Average Options, Pacific-Basin Finance Journal 3, 303-318. Roll, Richard, 1984, A Simple Implicit Measure of the Effective Bid-Ask Spread in an Efficient Market, Journal of Finance 39, 1127-1140. Scruggs, John and Jennifer Lynch Koski, 1998, Who Trades Around the Ex-Dividend Day? Evidence from NYSE Audit File Data Financial Management, Vol. 27, No. 3, (Autumn). Smith, Jeffrey, James Selway, Lorraine Reilly and Timothy McCormick, 1997, NASD Data Relating to the NASDAQ Stock, Market, Inc. and its Listed Companies, NASD Working Paper 97-01, June 16, 1997.

22

Smith, Jeffrey, James Selway and Timothy McCormick, 1998, The NASDAQ Stock Market: Historical Background and Current Operations, NASD Working Paper, 98-01, August 24, 1998. Tkac, P., 1999, “A Trading Volume Benchmark: Theory and Evidence,” Journal of Financial and Quantitative Analysis 34 (March), 89-114. Wang, Jiang, 1994, A Model of Competitive Stock Trading Volume, The Journal of Political Economy 102, (February), 127-168. Wansley, James W., P. Daves, and D. Stewart, 2002, What Determines Reported Volume: A Comparison of NYSE and Trading, The University of Tennessee working paper.

23

Table 1 Summary Statistics of NYSE-listed companies with CRSP share codes 10 and 11

tov alpha beta residual price capitalization divyield autocov sp500 option institution

Mean Standard Deviation Minimum 0.0139 0.0098 0.00005 0.0003 0.0047 -0.0283 0.6294 0.3807 -0.8409 0.0580 0.0265 0.0175 3.0871 0.8053 0.2032 13.7019 1.7112 9.0235 0.0008 0.0011 0.0000 -0.0001 0.0009 -0.0186 0.2799 0.4491 0 0.5981 0.4904 0 0.5037 0.2093 0.0082

Maximum 0.1019 0.0151 2.2805 0.2799 10.8326 19.4251 0.0151 0.0115 1 1 0.9628

Number of observations is 1647 stocks. 1. tov is the median weekly turnover. 2. alpha, beta, and residual are the intercept, slope coefficients and the standard deviation of residuals of the time -series regression of an individual stock’s return on the market return. 3. price is the average of natural log of an individual stock’s price. 4. capitalization is the average of natural log of an individual stock’s market capitalization. 5. divyield is the average of an individual stock’s dividend yield. 6. autocov is the first-order autocovariance. 7. sp500 is a dummy variable equal to 1 if the stock is included in the S&P 500 index during any quarter between 1996 and 2000; otherwise sp500 is equal to zero. S&P500 memberships information is obtained from COMPUSTAT tape. 8. option is a dummy variable equal to 1 if options trade for this stock. 9. institution is the percentage of institutional ownership which is equal to the number of shares held by institutions divided by the total number of shares outstanding. 10. The information on option and institution is collected from Standard & Poor’s Common Stock Guide (1996-2000).

24

Table 2 Summary Statistics of NASDAQ-listed companies with CRSP share codes 10 and 11

tov alpha beta residual price capitalization divyield autocov sp500 option institution

Mean Standard Deviation Minimum 0.0312 0.0293 0.00007 0.0003 0.0073 -0.0438 0.9617 0.6560 -4.2587 0.0989 0.0502 0.0196 2.4541 0.8891 0.0093 12.0525 1.4956 8.0870 0.0005 0.0011 0.0000 -0.0006 0.0022 -0.0282 0.0352 0.1844 0 0.3761 0.4845 0 0.3409 0.2196 0.0006

Maximum 0.2812 0.0360 3.2589 0.7314 5.8183 19.1706 0.0173 0.0291 1 1 0.9507

Number of observations is 1816 stocks. 1. tov is the median weekly turnover. 2. alpha, beta, and residual are the intercept, slope coefficients and the standard deviation of residuals of the time -series regression of an individual stock’s return on the market return. 3. price is the average of natural log of an individual stock’s price. 4. capitalization is the average of natural log of an individual stock’s market capitalization. 5. divyield is the average of an individual stock’s dividend yield. 6. autocov is the first-order autocovariance. 7. sp500 is a dummy variable equal to 1 if the stock is included in the S&P 500 index during any quarter between 1996 and 2000; otherwise sp500 is equal to zero. S&P500 memberships information is obtained from COMPUSTAT tape. 8. option is a dummy variable equal to 1 if options trade for this stock. 9. institution is the percentage of institutional ownership which is equal to the number of shares held by institutions divided by the total numb er of shares outstanding. 10. The information on option and institution is collected from Standard & Poor’s Common Stock Guide (1996-2000).

25

Table 3:Cross-sectional regressions of median weekly turnover (1996-2000) (NYSE ordinary common shares with CRSP share codees 10 and 11)

Tov i = β 0 + β1 * alpha i + β 2 * betai + β 3 * residual i + β 4 * price i + β 5 * capitaliza tioni + + β 6 * divyield i + β 7 * auto cov i + β 8 * sp500 i + β 9 * optioni + β 10 * institutioni + ν i Explanatory variables Intercept alpha beta residual price capitalization divyield autocov sp500 option institution Adjusted R2

Model -0.0048* (-1.78) 0.0741 (1.62) 0.007*** (12.15) 0.1148*** (10.75) 0.0011** (2.5) -0.0004* (-1.69) -0.1164 (0.71) 0.6815*** (3.01) 0.0018*** (3.10) 0.0044*** (8.76) 0.0135*** (13.35) 0.43

* Indicates significance at the 10% level. ** Indicates significance at the 5% level. *** Indicates significance at the 1% level. t-statistics are in parentheses.

26

Table 4:Cross-sectional regressions of median weekly turnover (1996-2000) (NASDAQ ordinary common shares with CRSP share codes 10 and 11)

Tov i = β 0 + β1 * alpha i + β 2 * betai + β 3 * residual i + β 4 * price i + β 5 * capitaliza tioni + + β 6 * divyield i + β 7 * auto cov i + β 8 * sp500 i + β 9 * optioni + β 10 * institutioni + ν i

Explanatory variables Intercept alpha beta residual price capitalization divyield autocov sp500 option institution Adjusted R2

Model -0.0152** (-2.41) 0.0084 (0.12) 0.0184*** (20.00) 0.1426*** (10.87) 0.0083** (7.7) -0.0013* (-1.94) 0.0894 (0.20) 0.5705** (2.42) 0.0090*** (3.14) 0.0146*** (12.13) 0.0126*** (4.79) 0.54

* Indicates significance at the 10% level. ** Indicates significance at the 5% level. *** Indicates significance at the 1% level. t-statistics are in parentheses.

27

Table 5:Cross-sectional regressions of median weekly turnover (1996-2000) (NASDAQ ordinary common shares with CRSP share codes 10 and 11)

Tov i = (( λ − 1) * nas + 1) β 0 + β1 * alpha i + β 2 * betai + β 3 * residual i + β 4 * price i + β 5 * capitaliza tioni + β 6 * divyield i + β 7 * autocov i + β 8 * sp500 i + β 9 * optioni + β 10 * institutioni + ν i Explanatory variables Intercept alpha beta residual price capitalization divyield autocov sp500 option institution

? F-value

NLIN (NYSE/NASDAQ) -0.0067** (0.002) 0.0095 (0.0234) 0.0084** (0.0004) 0.0653** (0.005) 0.0033** (0.0004) -0.0005** (0.0002) -0.0178 (0.1409) 0.288** (0.083) 0.002** (0.0007) 0.0063** (0.0004) 0.0078** (0.0009) 2.188** (0.0715) 1053.23

** Indicates significance at the 5% le vel. Approximate standard errors are in parentheses. The regression is run on 3463 observations from the pooled sample of NYSE and NASDAQ firms.

28

Table 6 Chow Test and λ Results for Annual Pooled NYSE and NASDAQ Regressions: 1996-2000

Estimated λ Percent volume reduction implied by λ

1996 2.457 59.3 %

1997 2.333 57.4 %

Year 1998 2.441 59.0 %

1999 2.541 60.6 %

2000 1.889 47.1 %