The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
Working Paper Series Changes in Trading Activity Following Stock Splits and Their Impact on Volatility and the Adverse Information Component of the Bid-Ask Spread A.S. Desai, M. Nimalendran and S. Venkataraman Working Papers Series Issues in Financial Regulation Research Department Federal Reserve Bank of Chicago December 1996 (WP-96-21) 1/ ■■BHMEHRn ....x''' LIBRARY FEDERAL RESERVE B A N K OF CHICAGO DEC 1 7 1996 FtUtKAL KtitKVt BANK OF CHICAGO C h a n g e s in T r a d i n g Activity Following Stock Splits a n d Their I m p a c t o n Volatility a n d the A d v e r s e Information C o m p o n e n t of the B i d - A s k S p r e a d A. S. Desal D e p a rtm e n t o f F in a n c e C o lle g e o f B u s in e s s A d m in is t r a t io n U 7 D C a lv in H a ll K a n s a s S ta te U n iv e r s it y M a n h a tta n , K S 6 6 5 0 6 -0 5 0 3 (9 1 3 ) 5 3 2 -6 8 2 0 M. Nimalendran D e p a rtm e n t o f F in a n c e , I n s u ra n c e , a n d R e a l E s ta te C o lle g e o f B u s in e s s A d m in is t r a t io n U n iv e r s it y o f F lo r id a P .O . B o x 1 1 7 1 6 0 G a in e s v ille , F L 3 2 6 1 1 -7 1 6 0 S. Venkataraman E c o n o m ic R e s e a r c h F e d e r a l R e s e rv e B a n k o f C h ic a g o 2 3 0 , S o u th L a S a lle S tre e t C h ic a g o , I L 6 0 6 0 4 -1 4 1 3 R e v is e d , A u g u s t, 1 9 9 6 W e would liketothank Jim Angel, David Brown, David Ellis,Mark Flannery, Joel Hasbrouck, Joel Houston, Chris James, Gautam Kaul, and theparticipants at the Rutgers University conference on Recent Developments in Asset Pricing and Optimal Trading Strategies, the 1994 Financial Management Association Meetings and the Northwestem/JFI Conference on Market Microstructure and the Design o f Financial Systems fortheiruseful comments. The views expressed in thispaper are not necessarily those of the Federal Reserve Bank ofChicago, or the Federal Reserve system. All errors, ofcourse, are our own. C h a n g e s in T r a d i n g Activity Following Stock Splits a n d Their I m p a c t o n Volatility a n d the A d v e r s e Information C o m p o n e n t of the B i d - A s k S p r e a d ABSTRACT This paper examines the changes in trading activity around stock splits, and its impact on both the volatility and the bid-ask spread. After a stock split, there is a significant increase in the volatility and the spread, even after controlling for the effects of microstructure biases like price discreteness and bid-ask bounce. The change in the number of trades is positively related to the change in total volatility, as well as to the temporary and permanent components of volatility. This suggests that the change in trading activity is associated with both informed and noise traders. The change in the number of trades is also negatively related to the change in the total spread, as well as the adverse information content of the spread. Firms that are successful in attracting a large number of additional trades to their stock experience a smaller increase in spreads. These results suggest that a crucial determinant of the liquidity changes experienced by a firm after a stock splitisthe success of the split in attracting new trades to the security. C h a n g e s in T r a d in g A c tiv ity F o llo w in g S to c k S p lits a n d th e ir I m p a c t o n V o la tility a n d th e A d v e r s e I n fo r m a tio n C o m p o n e n t o f th e B id -A s k S p r e a d . 1. Introduction W e examine the impact of trading activity and market microstructure on the volatility and the adverse information component of the bid-ask spread around stock splits. Previous studies of stock splits document an increase in both the volatility and the proportional bid-ask spread aftera split However, they do not provide a satisfactory explanation for changes in these characteristics. In this study, we firstexamine the relationship between changes in volatility and changes in trading activity. This analysis is motivated by Jones, Kaul and Lipson (1994), who find that volatility isprimarily and positively related to the number of trades. W e also examine the relationship between changes in trading activity and the components of the bid-ask spread. Our analyses also provides insights about the managerial objective of enhancing the liquidity of the stock by splitting it.Splits per se do not altereither the cash flows of the firm or the claims of the security holders. Yet, in any given year, about 10% of the firms splittheir stock. Surveys of corporate managers by Baker and Gallagher (1980) and Baker and Powell (1993) reveal that the two most important reasons given by managers for undertaking a split are to bring the stock price into a better trading range and to improve itsliquidity. Managers believe that the lower stock price makes itpossible for wealth constrained “small” traders to purchase round lots. Baker and Powell argue that the managerial view of enhanced liquidity isthis increase in the diversity and number of shareholders. Lamoureux and Poon (1987) and Maloney and Mulherin (1992) document an increase in the number of shareholders after the stock split, and their evidence is therefore consistent with the managerial motivations for stock splits. On the other hand, studies have also found that after a split, there isalso an increase in the proportional bid ask spread (Copeland (1979) and Conroy, Harris, and Benet (1990)), a decrease in the split-adjusted trading volume (Copeland and Lamoureux and Poon), an increase in brokerage fees (Copeland), and an increase in the volatility of the stock’s returns (Ohlson and Penman (1985) and Dubofsky (1991)). Based on these measures, liquidity appears to decrease. Taken together, the evidence suggests that by splitting the stock, managers achieve the objective of increasing shareholder diversity but the splitdoes not, on average, lead to an improvement in traditional measures of liquidity. However, these studies do not examine the implications of the change in shareholder diversity for trading activity, and the consequent changes in volatility and the bid-ask spread. W e investigate two potential explanations for the increase in volatility following a split First Ohlson and Penman (1985), Dravid (1988), and Dubofsky (1991) argue that part of the increase in volatility could be attributed to microstructure biases. Particularly, both bid-ask bounce and price discreteness induce an upward bias in volatility estimates based on transaction prices, and this bias is exacerbated after the splitdue to the lower share prices. W e avoid die bias due to bid-ask bounce by using returns based on bid-bid prices (see Kaul and Nimalendran (1990)). The correction for price discreteness follows the model in Ball (1988). W e document that while bid-ask bounce and price discreteness do inflate volatility estimates, the stock’s volatility increases afterthe spliteven after we correct for these biases. The second explanation for the increase in volatility is an increase in trading activity. Jones, Kaul, and Upson (1994) find that volatility isprimarily and positively related to the number 2 of trades. Thus, ifthe number of trades increases after the split,then this would increase the volatility. There are several theoretical reasons why the number of trades might increase after a split. Black (1986) argues that noise traders prefer low priced stocks to high priced stocks. Ifthey do, then the lower per share price after a split would attract noise traders and the resulting increase in the number of trades would increase the volatility. The increase in noise trading would be consistent with the managerial objective of increasing shareholder diversity. Brennan and Hughes (1991) argue that the lower per share price afterthe splitmight give analysts the incentive to collect more information on firms. They provide evidence that the number of analysts following a firm increases after the firm announces a stock split. This suggests the presence of a larger number of informed traders in the security afterthe split, and again, the resulting increase in the number of trades would lead to an increase in the volatility. Itisalso possible that the levels of both noise and informed traders increase after a split. In Admati and Pfleiderer’s (1988) model of strategic trading with costly information acquisition, the number of informed traders isdetermined endogenously. A higher number of noise traders would result in a higher number of informed traders as well. Thus, iflower share prices aftera splitattract noise traders, the level of informed traders would increase endogenously. Note that all three models above imply an increase in the number of trades but differ in theirimplications about the mix of traders (noise versus informed). Our analysis shows that the number of trades does increase afterthe split. Further, we find a significant positive relationship between the change in the number of trades and the change in the volatility. However, since we cannot directly observe the mix of traders, we have to rely on indirect methods to provide relative 3 impacts of noise and information on volatility changes. W e employ two approaches. The first approach uses multi-period volatility and variance ratios (the ratio of multi-period to single-period volatilities). In the second approach, we use insider trading intensity as a proxy for informed trading in a cross-sectional regression model. Our analysis of multi-day volatility and variance ratios indicates that a significant part of the increase in the volatility isin the component that is due to pricing errors, and that there is also an increase in the permanent component of the volatility. In the cross-sectional regression framework, we find that changes in the total volatility, as well as changes in the transient and permanent components of the volatility are positively related to changes in the number of trades. This relationship isconsistent with an increase in both noise and informed trading after the split. The models of Black (1986), Brennan and Hughes (1991), and Admati and Pfleiderer (1988) have implications for the adverse information component of the bid-ask spread as well. An increase in noise trading after a split(as in Black) will decrease the adverse information component, while an increase in informed trading (as in Brennan and Hughes) will increase this component of the spread. Admati and Pfleiderer argue that an increase in the number of noise traders would also endogenously increase the number of informed traders. The impact of this change in trading activity would depend on whether the signals received by the informed traders are close substitutes or complements.1When informed traders get identical signals, Admati and Pfleiderer show that competition among traders would decrease the adverse information in the 1 Evidence to support the hypothesis thatsplitsconvey private information to the market ispresented in Brennan and Copeland (1988) and in McNichols and Dravid (1990). 4 market However, ifinformed traders get diverse signals, then the effect on the adverse information depends on the precision of the signal. Ifthe signal issufficiently imprecise, then adverse information would decrease due to competition between informed traders, even ifthe number of informed traders increases. Thus, by relating the change in the adverse information component of the spread to measures of trading activity, we can draw inferences about the impact of splits on liquidity. While we find that, on average, both the proportional spread and the adverse information component of the spread increase after the stock split,these changes are less pronounced the larger the increase in the number of trades. This negative relationship implies that spread changes are either driven by noise traders or by informed traders with substantially similar signals (or some combination thereof). The evidence sheds lighton several aspects puzzle regarding stock splits. First, we establish that the observed increase in volatility and spreads cannot be attributed solely to statisticalproblems (like price discreteness and bid-ask bounce). Consistent with the findings of Jones, Kaul and Lipson, we also establish a positive relationship between the change in volatility and the change in the number of trades. The fact that this relationship holds true for both the transitory and permanent components of volatility suggests thatboth informed and noise traders are contributing to the observed increase in the number of trades. Finally, we establish that while spreads (total, as well as the adverse information component) increase from before to after a split, firms thatexperience a substantial increase in the number of trades have a smaller increase in spreads relative to other firms. This suggests that, all else kept the same, attracting additional traders to the security enhances the liquidity of the stock. The puzzle that stillremains, of course, 5 is (i)what distinguishes firms in terms of their ability to attract additional traders to their stock, and (ii)why a splitunconditionally worsens liquidity. These remain interesting areas for future research. The rest of the paper is organized as follows. In section 2, we describe the data and the sample characteristics. In Section 3, we present our analysis of volatility changes afterstock splits. Changes in the adverse information of the spread are discussed in Section 4. Our conclusions and a summary are presented in the last section. 2. 2 .1 Data and Sample Characteristics D a ta The initial sample consists of N A S D A Q - N M S firms that announced stock splits and are listed in the CRSP 1990 data base. W e confine our sample to N A S D A Q - N M S firms because CRSP provides bid-ask spreads (inside quotes), daily trading volume, and number of trades presently only for these firms. W e further restrictour sample to announcements of splitsduring the period January, 1983 to December, 1990. This restriction isimposed because transaction prices and bid-ask spreads for N M S securities are available in the CRSP data base on a regular basis only afterNovember, 1982. There are 980 splits announced by 739 firms meeting the above screens. Pre-split microstructure variables for the sample (such as volatility, spreads, trading intensity, etc.) are estimated over the 180 day period ending 21 days before the announcement of the split. The post-split characteristics are estimated over the 180 day period beginning from 21 6 trading days after the stock firsttrades ex-split. W e exclude the period from 20 days before the announcement to 20 days afterthe stock trades ex-split to avoid any contamination due to information effects around the announcement day and the transient microstructure effects around the ex-split date. W e require that allrelevant data items be available during the pre-split and the post-split estimation periods. Finally, we exclude from our initial sample all observations that have either a stock split or a stock dividend within 400 days of each other. This screen ensures that our estimation period data are not contaminated by events similar to the ones examined in this study. The final sample consists of 366 stock splits announced by 341 firms. In our sample of 366 split announcements, 147 are 3-for-2 splits (a splitfactor of 0.5) and 138 are 2-for-l (a splitfactor of 1.0). Sixty three announcements involve a split factor of less than 3-for-2, while eighteen are splits greater than 2-for-l. Previous studies have suggested that there is a difference between the motives of firms issuing small versus large stock splits.Elgers and Murray (1985) document that small splits (with splitfactor < 25%) are associated with smaller pre-splitprices and smaller market value firms than large splits. Further, they also suggest that small splitfactors may be motivated by a desire to signal optimistic expectations, while larger split factors are motivated by liquidity reasons. Baker and Powell (1992) find a significant difference between the preferred trading ranges for the small (< 2-for-l) versus large (k 2-for-l) splits. Since there appears to be differences between the motives for small versus large splits,we partition our sample into two groups: a small splitfactor group of 3-for-2 splits and smaller, and a large split factor group of 2-for-l splits and greater. 2.2 S a m p le C h a r a c t e r is t ic s 7 In Table 1A we report sample characteristics for some selected variables (market value, number of shares, price, and number of market makers) for the entire sample, and for the two sub samples based on the split-factor. W e find that there is a significant difference in the median presplitmarket values of the equities of the firms in the two sub-samples. The median value of equity in the small splitfactor sub-sample is $85 million compared to $182 million for the large split factor sub-sample. This difference in firm size isdriven by the higher pre-splitshare price for the large splitfactor sub-sample ($37.7) compared to the small splitfactor sample ($21.8), because the difference in the median pre-split outstanding shares isnot statistically significant. W e also find that there isno statistical difference in the number of market makers between the two groups in the pre- and post-splitperiods. Since we find significant differences in some firm characteristics between the two groups, we analyze the market microstructure variables also by groups. [Insert Tables 1A and IB here] In Table IB, we report statistics for microstructure variables and measures of trading activity. The table reports pre- and post-split means and medians for the proportional bid-ask spread, the daily volatility based on transaction prices, the average daily number of trades, the average volume turnover, the average volume turnover per trade, and the average number of insider trades. The statistics are also presented for the ratio of post-splitto pre-split values for these variables. Our objective isto confirm for our sample what existing studies have documented. First, the median values of both the pre- and post-split proportional spreads are higher for the small splitfactor sub-sample than for the large split factor sub-sample. This is consistent with a smaller market value and lower pre-split share prices for the small split factor 8 sub-sample compared to the large splitfactor sub-sample. Next, we find a significant increase in the total spread after the stock splitfor the total sample as well as for each sub-sample. This increase in the post-split spreads isconsistent with the findings of Copeland (1979) and Conroy, Harris, and Benet (1990). The median ratio of the post-split to the pre-split proportional spread is 1.08 for the small split factor sub-sample, and 1.45 for the large split factor sub-sample. While both these values are significantly greater than 1.0 ,the median change in the spread isobviously much greater for the large splitfactor sub-sample. W e also document a significant increase in the post-split volatility relative to the pre-split volatility for the total sample and also for each sub-sample. For the total sample the median increase in volatility based on closing transaction prices is70%. For the small splitfactor sub sample, the median increase in volatility is42%, while for the large splitfactor sub-sample, itis 119%. Ohlson and Penman (1985), who also use closing transaction prices for their volatility estimates, report an increase of 63% to 83% in the volatility after the split, and their results are comparable to ours. In addition to spreads and volatility, we also examine changes in trading activity. W e find a significant increase in the number of trades afterthe split for both groups. The median increase for the small split factor sub-sample is 12% compared to 37% for the large splitfactor sub sample. However, the median volume turnover (defined as the ratio of the transaction volume to the number of outstanding shares) increases by only 7 % for the small split factor group and decreases by 7% for the large split factor group, with the decrease for the latter group being statistically insignificant. The increase in number of trades isconsistent with the findings by 9 Lamoureux and Poon (1987), and the effect on volume isconsistent with Murray (1985) and Lakonishok and Lev (1987), who find that splits do not appear to exert a permanent effect on volume. More interestingly, we find that the turnover volume per trade is significantly smaller in the post-split period compared to the pre-split period for both groups. The median reduction in the order size is 18% for the small split factor stocks and 35% for the large split factor stocks. The smaller order size isconsistent with an increase in noise traders who are likely to trade smaller quantities. Finally, we report statistics on the average daily number of insider trades for our sample. Seyhun (1988) finds evidence consistent with the argument that corporate insiders generally trade on private information. This suggests that measures of trading activity by insiders can be used as a reasonable proxy for informed trading. However, insiders may also trade for liquidity reasons. But itis not obvious why their liquidity motivations for trading in theirfirms’stock would be altered during the 180 day intervals before the announcement and after the ex-day of the split. Ifinsiders withhold such trading prior to the splitin anticipation of the announcement, then such trades are likely to be executed soon after the splitis announced. In our study, we measure the post-split insider trading activity beginning twenty days after the ex-splitdate. Thus, itisunlikely that the execution of any such backlog of trades will have a significanteffect on our measures of informed trading activity by this group of investors. For the sample of 366 firm events, we obtained insidertransactions from the Securities and Exchange Commission’s Ownership Reporting System (ORS) cumulative file (which covers the period 1980-1991). For this study, we classify allofficers and directors and others who have 10 substantial ownership in the firm as insiders. In addition, any transaction that was classified as an open market purchase/sale, private purchase/sale, and exercise of options was considered as valid transactions. The time period used for measuring insider trading activity in the pre-split and post split is identical to that used for allother variables. In our sample, while the mean ratio of the post-split insider transactions to the corresponding pre-split value issignificantly greater than 1.0 , the median is not. This difference arises due to skewness of the data. The observed changes in the microstructure variables are consistent with previous empirical findings. In the next section, we test the effect of microstructure biases and trading activity on changes in the daily volatility. 3. 3 .1 Changes in Volatility Following Stock Splits E ffe c t s o f M ic r o s t r u c t u r e B ia s e s o n V o la t ilit y E s tim a te s Ohlson and Penman (1985), and Dravid (1988) suggest that the increased bid-ask spread and the larger effect of price discreteness on lower priced stocks may account for part of the increase in volatility that is observed afterthe split. Dubofsky (1991) finds that, for a sample of A M E X listed stocks, there isno statistically significant increase in the weekly return volatility afterthe split. Since volatility estimates based on weekly returns are less affected by bid-ask errors and price discreteness, Dubofsky argues that measurement errors created by the bid-ask errors and price discreteness could partially explain the increase in the return volatility of daily returns. In this section, we firsttest this microstructure explanation for the observed increase in the volatility using our sample of stock splits. 11 3. l . a B ia s D u e to B id - A s k B o u n c e Roll (1984) shows that in an efficient market, ifthe probability of the transaction price being at the bid or the ask isequally likely, then, using transactions prices to estimate the true volatility of the stock returns would induce spurious volatility equal to s2/2,where s is the percentage bid-ask spread. Kaul and Nimalendran (1990) show that for a portfolio of small market value N A 5 D A Q - N M S firms, this spurious volatility could be as high as 50% of the underlying true volatility. Even for the largest firms, this proportion could, on average, be as high as 23%. For our study, this bias could be particularly significant ifthe bid-ask spread increases aftera split. However, given the availability of bid and ask prices for the N A S D A Q - N M S firms, we can avoid the bid-ask spread bias by estimating volatilities based on returns computed using bid-bid prices.2 In Table 2 we report statistics for sample variances estimated using daily returns in the pre- and post-split periods. The estimate based on transaction price returns isdenoted by ,and that based on bid-to-bid returns isdenoted by o \ .Since the sampling distribution of the estimated variances are highly skewed and kurtotic, inferences based on the sample mean under the assumption that the underlying distribution is normal could be misleading. Therefore, we report only the sample medians before and after the split, and robust non-parametric test statistics for the differences in the sample medians (note that sample means also give the same results). [Insert Table 2 here] 2 The CRSP data base gives closing bid and closing ask prices in addition to the closing transactions price.W e construct return series based on bid prices by adjusting fordividends and distributions on ex-days. 12 For the total sample, the median ratio of the post-split volatility to the pre-split volatility, based on transaction price returns, is 1.70, indicating a 70% increase in the volatility. However, there isan 80% increase in the volatility based on bid-to-bid returns. Both of these increases are significant at the 1% level. Since the lattermeasure avoids the bias due to the bid-ask bounce, our results suggest that bid-ask bounce alone cannot explain the increase in volatility. An increase in the bid-ask bounce corrected volatility is also observed for the two sub-samples based on the split factor. For the small split factor sub-sample, this volatility increases by 41%, while for the large splitfactor sub-sample, the increase is 124%. To estimate the bias in the transaction return volatility, we compare the median values of C j with those of for each sample, in both the pre- and post-split periods. For the total sample, the pre-split transaction return volatility is about 95% higher than the corresponding bid-bid return volatility, indicating a significantly large bias due to the bid-ask bounce. Likewise, in the post-splitperiod, the bias due to the bid-ask bounce is about 84%. These biases are also observed for each sub-sample. For the small splitfactor sub-sample, the biases are 90% and 89% in the two periods respectively, while those for the large splitfactor sub-sample are 91% and 71% in the preand post-splitperiods respectively. Since previous studies have also documented an increase in the transaction return volatility after stock splits, itwould be informative to estimate how much of this increase isdue to the bidask bounce. Note that the increase in the spread after the split will exacerbate the bias due to the bid-ask bounce. For the total sample, the median o \ increases by 3.17x10"*. On the other hand, the 13 bid-bid return volatility increases by 1.91x10^, which is about 60% of the increase in . Therefore, about 40% of the increase in the transaction return volatility can be attributed to the bid-ask bounce. Similarly, about 47% and 36% of the increase ina 2 for the small and large split factor sub-samples respectively can be attributed to the bid-ask bounce. 3. l.b B ia s D u e to P r ic e D is c r e te n e s s Gottlieb and Kalay (1985) and Ball (1988) examine the effect of price discreteness on the inflation in the volatility estimates. Ball shows that ifstock prices follow a Geometric Brownian motion with an instantaneous true underlying variance o2,and price P, then the bias induced by price discreteness can be approximated by dVbP2,where d isequal to the minimum price change (typically, $0,125). W e apply this correction to the volatility measure to obtain an unbiased estimator. The estimator,a \ D ,iscomputed using Equation A-l in Appendix A, and the results are presented in the lastrow of Table 2. The median ratio of the post-split volatility corrected for both the bid-ask bounce and price discreteness is 1.81 in the total sample, and this estimate issignificantly greater than 1.0 at the 1% level. Thus, even aftercorrecting for price discreteness, there isa significant increase in the volatility afterthe split. A significant increase in the bias corrected volatility isalso observed for each of tne two sub-samples, although we observe a much larger increase for the firms with a large split factor (118% versus 42%). To estimate the bias due to price discreteness, we compare the volatility corrected for 14 both the bid-ask bounce and price discreteness ( D) with the volatility corrected for the bid-ask bounce alone ( ) . For the total sample, price discreteness inflates this volatilityestimate by 4.7% in the pre-split period, and by 3.9% in the post-splitperiod. This bias isrelatively small compared to the bid-ask bias, and further, ithas a negligible effect on the change in volatility. For the small splitfactor sub-sample, the biases in the pre- and post-split periods are 5.4% and 5.7% respectively, while those for the large splitfactor sub-sample are 0.4% and 7.8% in the two periods. The preceding analysis suggests that the bid-ask bounce introduces a substantial bias in estimates of volatility based on transaction prices, while the bias due to price discreteness is negligible. More importantly, even aftercorrecting for these biases, we find a significant increase in the volatility after the split. Thus, microstructure biases alone cannot account for the previously documented increase in the volatility after stock splits. Further, firms that execute large splits experience a much larger increase in the bias-corrected volatility than do firms that execute small stock splits. 3 .2 E f fe c t o f C h a n g e s in T r a d in g A c t iv it y o n C h a n g e s in V o la t ilit y As argued earlier, an increase in either noise traders or informed traders (or both) would lead to an increase in the volatility of the stock. To examine the effect of the change in the trader mix after the spliton the change in the volatility estimates, we need to estimate changes in trader types after stock splits. Since itisnot possible to directly identify the type of traders, we use two approaches. First, we analyze stock return dynamics and market microstructure variables to infer 15 the effect of changes in the types of traders. This allows us to decompose the change in the volatility into changes in the permanent (information driven) component and the transient (noise driven) component. In the second approach, we relate changes in volatility (and itscomponents) to changes in trading activity using proxy variables for noise and informed trading. 3 .2 .a C h a n g e s in P e r m a n e n t V o la t ilit y French and Roll (1986) argue that, ifthe effects due to noise trading (i.e.pricing errors) are subsequently corrected, then the volatility based on longer period returns would reflectthe permanent component Thus we estimate the volatility based on multi-day returns. Due to the limited number of observations in each estimation period, we use overlapping data and the estimator in Lo and MacKinlay (1988). This estimator isgiven by Equation A-2 in Appendix A, and corrects for the effects of both the bid-ask bounce and price-discreteness on volatility. In Table 3, we report these multi-period volatility estimates for cumulating intervals up to 30 days for the total sample and for sub-samples based on the splitfactor. For the total sample, we find that the median 30-day return volatility increases by 57% afterthe split,and the increase isstatistically significant atthe 1% level. This indicates that a significant component of the increase in volatility ispermanent The one day return volatility however, increases by 81% for this sample (see Table 2). The larger increase in the one-day volatility relative to the increase in the 30-day volatility suggests that there is also a large component of the increase in volatilitythat istransient and attributable to noise. Similar results are obtained for the sub-samples based on the split factor. However, for the small split factor group, the increase in the permanent component of the volatility (based on 30-day returns) is only 32% compared to 81% for the large split factor 16 group. [InsertTable 3 here] These results indicate that for our sample, there is a substantial increase in volatilitythat is permanent, in addition to a significant increase that istransient. Further, our finding that there is a significant increase in the multi-day volatility are in contrast to those of Dubofsky (1991). He finds that, for a sample of firms listed on A M E X that executed stock splits, there is no significant change in the volatilitybased on weekly returns. 3 .2 .b C h a n g e s in V o la t ilit y d u e to N o is e T r a d in g An alternative metric to determine the relative contribution of noise trading to the total volatility of a security’s returns isthe variance ratio, defined as the ratio of the variance based on k-period returns to k times the variance based on one-period returns. The presence of noise trading would induce negative autocorrelation in the returns, thereby reducing the variance based on multi-period returns. The variance based on one period returns would be unaffected by this negative autocorrelation ifittakes more than one period for the mis-pricing to be corrected. Consequently, French and Roll (1986) argue that one minus the variance ratio reflects the fraction of the one period volatility that can be attributed to noise. Lo and MacKinlay (1988) show that the variance ratio can be written as a weighted sum of the autocorrelations: VR(k) = l + j=i K where Pj denotes the estimate of thej* order autocorrelation of daily returns. Ifall the 17 (1) autocorrelations are due only to mis-pricing errors that are subsequently corrected, then, as suggested by French and Roll, (1-VR) would estimate the effect of noise trading on volatility. However, Kaul and N imalendran (1990) document that for N A S D A Q - N M S firms, the average autocorrelation of returns based on bid-bid prices at lag one is0 .IS, and this is much larger than the negative autocorrelations athigher lags. In Table 4 we report the median autocorrelations up to lag 10 for our sample. W e find positive autocorrelations at lags one and two that are similar to the numbers reported by previous researchers. These positive autocorrelations at low lags lead to variance ratios that are greater than one, and gives infeasible estimates for the effects of noise trading. To mitigate the effects due to large positive autocorrelations at short lags, we define one period of time as being three days. Our long run measure of variance iscomputed over 30 days, i.e. 10 periods. Estimates of the variance ratios based on these returns provide feasible estimates for the effects of noise. [Insert Table 4 here] Sample statistics for the variance ratios are reported in Table S. These variance ratios are computed using the estimator given by Equation A -3 in Appendix A. The median differences in the ratios (computed as the median of the matched difference in the post-split and pre-split variance ratios) are significantly negative for the entire sample and also for the two sub-samples based on the split factor. These ratios suggest that the fraction of volatility that can be attributed to noise trading is higher after the splitrelative to the pre-split level for all three samples. Further, since the total one-day volatility is also increasing for these groups, itsuggests that the volatility due to noise trading afterthe splitis substantially higher relative to the pre-split level. 18 [InsertTable 5 here] These estimates of the variance ratios, along with the bias-corrected estimates of the preand post-split one-day volatilities (reported in Table 2) can be used to estimate the fraction of the increase in the volatility that is due to an increase in noise trading. Since both the variance ratio estimates and the volatility estimates have been purged of microstructure biases, we can assume that any remaining mis-pricing errors are caused by noise traders. For the total sample, the median variance ratio is 0.94 before the split, and, from Table 2, the median pre-split one-day volatility is 2.75x1O'4.Thus, the volatility due to noise is0.165x1O'4 before the split. Similarly, after the split, the median variance ratio is0.79, the total volatility is4.61xl0'4,and the volatility due to noise is 0.968X10"4.Thus, about 43% of the increase in the total volatility of 1.86x10"* is due to the increase in noise volatility, with the remaining being attributable to an increase in the permanent component Similar results are obtained for the sub-samples based on the splitfactor. For the small splitfactor sub-sample, about 41% of the increase in the total volatility is due to noise, and for the large split factor sub-sample, approximately about 40% of the increase in the total volatility can be attributed as such. The above results suggest thatthere is a significant increase in b o th the noise and the permanent components of the volatility afterthe split. Further, as indicated in Table IB, the total number of trades increases, on average, after the split. Note that the total number of trades consists of trades executed by both noise and informed traders. Thus, an increase in both components of the volatility, coupled with an increase in the total number of trades, is consistent with the argument that after a stock split, there is an increase in the level of both noise and 19 informed trading. In the next section, we investigate this further using a cross-sectional regression framework. 3 .2 .c C r o s s - s e c t io n a l A n a ly s is o f V o la t ilit y C h a n g e s Trading activity can be measured by either the number of transactions or the size of the trade (i.e.the turnover volume). Earlier studies have documented a positive relationship between volatility and trading volume (see Karpoff (1987) for a review). However, Jones, Kaul, and Lipson (1994) conclude that itis the number of transactions per se, and not their size, that generates volatility. That is,the effect of trade size is subsumed in the number of transactions. Given their conclusions, we use changes in the number of trades as our measure of changes in trading activity.3 Ifwe assume that insider transactions are information driven, then we can use the number of trades executed by insiders as a proxy for informed trading. The change in the number of insider transactions after the split (from the pre-splitvalues) would then proxy for the change in informed trading activity. By netting out the insidertrades from the total number of trades, we can use the change in the net number of trades as a proxy for the change in noise trading.4 Using these proxy variables, we investigate the relationship between changes in trading 3 Jones, Kaul, and Lipson include both the number of trades and volume in theircross-sectional analysis of volatility. In our study, we are interested in the change in trading activity. W e find that inour sample, the change in the number of trades ishighly correlated with the change in turnover volume: the correlation coefficientbetween these two variables is0.79. Thus, including both measures of changes in trading activity results in theusual problems associated with multi-collinearity in the independent variables. 4 However, ifinsider transactions are motivated by liquidity concerns, then thechange in trades would not allow us to estimate the change in informed trading. Moreover, since the number of insider trades is a small fractionof the total number of trades (see Table IB), the effectof theirtrades on volatility would be subsumed in our proxy for noise trading. 20 activity and changes in the bias-corrected total volatility,the change in the long term volatility, and the noise component of the volatility. In our firstmodel, the dependent variable is the change in the bias-corrected total volatility of the stock. Specifically, the model is: _2 > ° B ,D ,2 _2 i f n n t 2) ( + a.L In = an u + a. 1 In In n t J i w NTT,"! J + 0X3(1 + SFACj)+ cc4 ln(MVALj)+ e, where Og Djisthe bias corrected volatility in periodj (jequals 1 for the pre-split period and 2 for the post split period), NNTj is the number of trades in periodj,net of insider trades, NTT ,is the number of insider trades in periodj,SFAC is the announced splitfactor, and M V A L isthe market value of the splitting firm’s equity, measured two days before the announcement of the split. The data presented in Table 2 indicates that the change in the bias-corrected volatility is much higher for the sub-sample with the large split factors (SFAC k 1.0). To the extent that the magnitude of the splitfactor conveys information to the market (see McNichols and Dravid (1990)), this would affect the volatility as well, through itseffect on the permanent component of the volatility. To control for the effect of this signal on volatility changes, we include the split factor as an additional independent variable in our regressions. Finally, inclusion of M V A L allows us to control for omitted variables which are correlated with the size of the firm. The estimates of the model parameters in Equation (2) are presented in the firstrow of Table 6.Our estimate of the coefficient on the change in the net number of trades (cti) is significantly positive. Thus, an increase in the number of noise transactions increases the volatility 21 of the stock afterthe split. The coefficient on our measure ofinformed trading (a2)is insignificantly different from zero at allconventional levels. There are two possible explanations for this observation. First, this coefficient would be insignificant ifthere isno significant change in the number of insider transactions after the split. The statisticsin Table IB indicate that the median change isindeed zero. Alternatively, ifinsider transactions are a poor proxy for informed trading, and in fact are motivated by liquidity concerns, then theireffect will be subsumed in the measure of the change in net trades. [Insert Table 6 here] The coefficienton the splitfactor isalso significantly positive. Firms which employ a larger splitfactorexperience a greater is the change in the total volatility. While this finding is consistent with the statisticspresented in Table 2,additional insights about the role of the size of the splitfactor can be obtained by examining the change in the components of the volatility. If splits convey private information to the market, as suggested by the signaling hypothesis of splits, then the change in the permanent component of the volatility would be positively related to the magnitude of the splitfactor, and the change in the noise component of the volatility would be unrelated to the split factor. W e further investigate this below. Finally, the coefficient on M V A L issignificantly positive. Iflarge firms have low volatility to begin with, even a small increase in the volatility would result in a large percentage change. This would manifest in a positive relationship between firm size and volatility changes.5,65 5 When we partition our sample into two sub-samples based on thepre-splitfirm size (MVAL), we find thatthe mean pre-splitbias-corrected volatilities for the small and large firm sub-samples are0.000461 and 0.000342 respectively. The mean ratios of the post-splitto pre-split volatilityforthese two sub-samples are 2.35 22 In our second model, the dependent variable is the change in the long term volatility of the stock. Specifically, we estimate the following model: { NIT^ 'n n t 2] + a,ln = a„u+ a.1In 2, iNrrJ I n n t ;J U 2Ji + oc3(1+ SFACj) + <x4 ln(MVALj) + e, where Zj isthe volatility based on 30-day returns in periodj (jequals 1 for the pre-split period and 2 for the post splitperiod) and the other variables are as defined earlier. Estimates of the parameters of this model are presented in the second row of Table 6.There is a significant positive relationship between the change in the long term volatility and the change in the net number of trades. The change in the long term volatility is again unrelated to the change in insider trades. A possible reason for this is that the change in insider trades isa poor proxy for informed trading. Further, the coefficient on the splitfactor (a3) is significant and positive. Moreover, the estimate of this coefficient (and the associated t-statistic) is greater than that in Equation 2. This is consistent with the hypothesis that splits are signals of private information about the firm. The effect of thisrelease of information on the total volatility isprimarily driven by itseffect on the permanent component of the volatility. Further evidence to support this argument isprovided in the model relating the change in the noise component of the volatility to trading activity.6 and 3.43 respectively. 6 We also estimate the model in Equation (2), as well as allsubsequent cross-sectional regression models, without the term involving our measure of firm size (MVAL). In allcases, our parameter estimates are virtually identical to those obtained with the inclusion of the M V A L term as an independent variable. Thus, our conclusions do not depend on the inclusion or omission of this variable, and we report the results for the estimates obtained by inclusion of this term. 23 The lastrow of Table 6 reports our estimates of the following model: (4) + a 4 ln(MVAL-,)+ a 5In -§• + £, V2 i )\ In this model, since we control for the change in the permanent component of the volatility on the right hand side, the dependent variable then measures the change in the noise component of the volatility. Once again, the estimate of a \ issignificantly positive, while the estimate of a 2 is insignificantly different from zero. Thus, the change in the noise component of the volatility is significantly positively related to the change in the net number of trades, but not to the change in insider trades. Interestingly, the coefficient on the splitfactor variable (a3)isalso insignificant and provides supporting evidence for the argument presented earlier, that splits convey private information to the market. Since the noise component of the volatility would be unrelated to private information released through the split, the change in this component of the volatility would be unrelated to the size of the announced split factor. Taken together, our results suggest the following. Stock splits, on average, result in an increase in trading activity, and this in turn leads to an increase in the volatility even after we control for microstructure biases in the estimated volatility. There is an increase in both the permanent as well as the transient component of the volatility, and the increase in trading activity positively affects both of these components. These results allow us to extend the conclusions drawn by Jones, Kaul, and Lipson (1994) about the relationship between the level of the volatility and the level of the number of transactions. Changes in these variables are also positively 24 correlated in the case of stock splits. Finally, the change in the permanent component of volatility ispositively related to the split factor, but the change in the noise component is not. This is consistent with the signaling hypothesis for stock splits. 4. Changes in Bid-Ask Spreads Following Stock Splits In this section, we investigate changes in bid-ask spreads following stock splits. The descriptive statisticspresented in Table IB indicate that for the total sample, the proportional spread increases after the splitby an average of 32% and thatthis increase iseven higher for firms that employ a large split factor. W e examine two possible reasons for this increase. First, an upward bias in post-split absolute spreads, caused by price discreteness, could result in an increase in the post-split proportional spread. Discreteness in stock prices leads to minimum absolute spreads of 12.5 cents, and increments in the spreads are also forced to be in steps of 12.5 cents.7 Second, the increase in the proportional spread could be due to an increase in one or more components of the bid-ask spread. Specifically, an increase in the information asymmetry in the market would lead to an increase in the adverse information component of the spread, thereby leading to an increase in both the absolute spread as well as the proportional spread. 4 .1 E f fe c t o f P r ic e D is c r e te n e s s o n S p re a d s Suppose a stock has a pre-split share price of $50 and a pre-split bid-ask spread of $1,375 7 Unlike the NYSE, NASD does not have a minimum spread policy. However, N A S D A Q isdesigned to process spreads of l/32ndfor stocks priced under $10 and l/8thfor stocks priced above $10. On the consolidated tapes (CTA), trades in N A S D A Q stocks under $10 are rounded to l/16th(Source: Market 2000 Study by the SEC). 25 (i.e. a proportional spread of 2.75%). After a 2:1 stock split,the share price would drop to $25. In order to keep the proportional spread constant, the absolute spread after the split would have to be $0.6875. Ifthe spread isconstrained to be in multiple of eighths, the post-split spread would be adjusted upwards to the nearest eighths. This results in a ‘target’post-split absolute spread of $0.75. Ifthe post-split spread is indeed $0.75, the proportional spread would be $0.75/$25 or 3% after the split,indicating a 9.1% increase in the proportional spread. In this case, all of the increase in the proportional spread can be attributed to price discreteness. Ifthe post-split spread is set at $0,875 (i.e. a proportional spread of 3.5%), the observed increase in the proportional spread would be 27.27%. Ifwe control for price discreteness, the spread increases from the target spread of $0.75 to the actual spread of $0,875, representing a 16.67% increase. The preceding example illustrates how discreteness in absolute spreads can inflatethe observed increase in proportional spreads. W e employ the adjustment procedure illustrated in this example to our sample of stock splits. Specifically, letABSj represent the absolute spread in period j (j= 1 for the pre-split period and j=2 for the post split period). Let AS represent the ratio of A B S 2 to [ABSi/(l + SFAC)], where SFAC isthe announced split factor. Thus, AS measures the uncorrected change in the spread. The target post-split spread, TS2,is computed as [ABSi/(l + SFAC)] rounded up (ifnecessary) to the next highest multiple of $0,125. Then the ratio [ABS2/TS2],denoted by AS* represents the change in the absolute spread after correcting for price discreteness. Ifprice discreteness is the only reason for the observed increase in spreads, then AS* should equal 1.0. On the other hand, ifAS* is greater than 1,then price discreteness alone cannot explain the observed increase in spreads. 26 To estimate AS*, we use the median absolute spread in the pre and post-splitperiods for each stock. In Table 7, we report the cross-sectional mean and median values of AS and AS* for the total sample as well as for the sub-samples grouped by the splitfactor. For the total sample, we find that the median increase in the absolute spread isa significant 50% before correcting for price discreteness, and 33% afterthe correction. Thus, price discreteness alone cannot account for the observed increase in spreads following stock splits.The statistics for the sub-samples based on the splitfactor lead to similar conclusions for each sub-sample. W e also note that the increase in spreads is greater for the large split factor sub-sample, both before and aftercorrecting for price discreteness. [InsertTable 7 here] 4 .2 C h a n g e s in th e A d v e r s e S e le c tio n C o m p o n e n t o f th e S p r e a d Since price discreteness alone cannot account for the observed increase in spreads, we focus on the components of the proportional spread to gain insights about the reasons for the observed increase. In particular, we decompose the proportional spread into itsorder processing and adverse information components, using the methodology in George, Kaul, and Nimalendran (1991). The total spread in general consists of three components: order processing, adverse information, and inventory cost. In George, Kaul, and Nimalendran, the part of the inventory cost component that decays within a day across a number of transactions is included in the order processing component, while that which does not decay within a day is included in the adverse selection component [see Jegadeesh and Subrahmanyam (1993)]. However, Stoll (1989) has 27 found that the inventory cost component is a small fraction of the total spread (less than 10%). Madhavan and Smidt (1991) also find that inventory effects are economically and statistically insignificant. Given these results, we focus only on the order processing and adverse information components in our study. The methodology in George, Kaul, and Nimalendran depends on using the difference in returns based on transaction prices and returns based on bid-to-bid prices to purge the bias due to changing expected returns and partial adjustment. In addition, by taking the difference between the two returns, the effects due to the unanticipated component of returns (which are a large fraction of the error) are eliminated. This substantially increases the efficiency of the estimates. Let R*tand R® represent the returns based on the closing transaction price and the closing bid price of firm iat time trespectively. Define R?t= R?t- R®, as the difference in these returns. George, Kaul, and Nimalendran show that ifSj isthe quoted spread, and jq is the fraction of the quoted spread due to order processing costs (and 1-7tjis the fraction due to adverse selection costs, then Cj = 2^-[Cov(RPt,RPt_,)] = jtjSj (5) Ifwe assume that for a group of stocks the fraction of the quoted spread that isdue to order processing costs (JtO is constant and equal to it , then we can use the following crosssectional model to estimate k : Q = ito+ JcSj + Ei,i= 1,.... N (6) W e firstestimate Equation (6)separately for the pre-split and the post-split periods for the 28 total sample, as well as for the sub-samples based on the splitfactor. In thisestimation, Si is the average proportional spread. For each sample, this procedure provides us with estimates of it in the pre-split and the post-split periods. These estimates of n are then used to compute the order processing costs as a percentage of the share price (= itSj) and the adverse information costs (= (l-7t)S0. Table 8 reports the average values of the proportional spread, as well as of the order processing and adverse information components of the spread. These values are reported for the pre- and the post-split periods, and for the total sample as well as for the sub-samples based on the magnitude of the split factor. The last column in Table 8 reports the difference between the pre- and post-split values. [InsertTable 8 here] In the total sample, the proportional spread increase by 0.513, and this increase is roughly evenly divided into the increases in the order processing and the adverse information component of the spread. For the small splitfactor sub-sample, virtually allof the increase in the proportional spread isdue to the increase in the adverse information component, with the order processing component remaining statistically unchanged from the pre-split level. By contrast, there isa relatively large increase in the proportional spread, and about 69% of this increase is due to an increase in the order processing component. The increase in both the total proportional spread and the adverse information component of the spread suggest that unconditionally, liquidity worsens after a stock split. Moreover, the increase in the adverse information component also suggests an increase in the information 29 asymmetry in the market. This isconsistent with an increase in informed trading activity (as in Brennan and Hughes (1991)). However, Admati and Pfleiderer (1988) argue that an increase in noise trading would endogenously increase the number of informed traders, and the net impact of this change in trading activity would depend on whether the signals received by the informed traders are close substitutes or complements. Ifthe signals received are diverse and sufficiently precise, information asymmetry would increase even ifthe number of noise traders in the market increases. Thus, a net increase in the adverse information component isnot inconsistent with an increase in the level of noise trading afterthe split. W e examine below the relationship between changes in trading activity and changes in both the total proportional spread and the adverse information component. 4 .3 E f f e c t o f C h a n g e s in T r a d in g A c t iv it y o n C h a n g e s in S p re a d s To examine the relationship between changes in the proportional spread and changes in trading activity, we estimate the following cross-sectional model using ordinary least squares regression: = a 0 + a,ln "n n t 2" + a 2ln I nnt.1 where Sj isthe proportional spread in period j (j= 1 for the pre-split period and 2 for the post splitperiod) and the other variables are as defined earlier. Estimates of the parameters of this model are presented in Table 9. In the firstrow of Table 9, we present our estimates using both measures of changes in trading activity. In the second and third rows, we use the change in the net 30 number of trades and the change in insidertrades as the measure of change in trading activity, respectively. [Insert Table 9 here] The results presented in Table 9 indicate that changes in proportional spreads are inversely related to changes in the net number of trades, regardless of whether we control for the change in insidertrades. Our estimates of at are significantly less than zero in the firsttwo rows of Table 9. Thus, while unconditionally, the proportional spread increases after the split, thiseffect is less pronounced the larger the increase in the net number of trades. Curiously, the change in the proportional spread is unrelated to the change in insider trades when the change in net trades is also used as an explanatory variable, but inversely related to itwhen the change in insider trades isthe only measure of the change in trading activity. The estimate of eta isinsignificant in the firstrow of Table 9, but significantly less than zero in the last row. This suggests that the effect of the change in insider trades on proportional spreads is subsumed in the effect of the change in non-insider trades. There are two possible reasons why this might be the case. First,insider trades are relatively small in number when compared to non insider trades (see Table IB). Second, changes in insider trades may be motivated by liquidity reasons rather than being information driven. Ifso, an increase in insider trades afterthe split would tend to attenuate the increase in spreads. This inverse relationship is observed in the negative estimate of <Xa in the last row of Table 9. Ifchanges in insider trades were information driven, then we would expect this estimate to be positive. W e investigate this furtherby examining the relationship between changes in the adverse information component of the spread 31 and our measures of changes in trading activity. Finally, we note that the estimates of the coefficients on the splitfactor, as well as those on the firm size variable are significantly positive in all three versions of Equation (7) in Table 9. Firms that employ a larger split factorexperience a greater increase in the proportional spread, as do large firms. Changes in trading activity would also have an impact on the adverse selection component of the spread. An increase in noise trading would reduce this component of the spread, while an increase in informed trading would tend to increase this component. To examine the relationship between the change in the adverse information component of the spread and our measures of trading activity, we estimate the following model: (8) where COVj isdefined as the auto-covariance of R® in periodj, and the other variables are as defined earlier. From Equation (6),this allows us to control for the change in the order processing component of the spread. Since the dependent variable in Equation (8) isthe change in the total proportional spread, and we control for the change in the order processing costs on the right hand side, Equation (8) allows us to estimate the change in the adverse information component of the spread. Table 10 reports our estimates of the parameters of Equation (8).As in Table 9, we estimate three versions of the model. The change in the adverse information component of the 32 spread isinversely related to the change in the net number of trades (the estimate of cti is significantly negative in the firsttwo rows of Table 9). This isconsistent with an increase in noise trading after the split. Interestingly, allof our estimates of the coefficient on the change in insider trades (ai) are significantly negative. Ifchanges in insider trades are motivated by information, a large increase in these trades would result in a proportionately large increase in the adverse information component of the spread. Consequently, we would expect these coefficients to be positive. Our results suggest that this is not the case. [Insert Table 10 here] Taken together, our analysis of the changes in spreads following stock splits suggest the following. The observed increase in the proportional spread isnot solely due to price discreteness. Rather, we observe an increase in both the order processing as and the adverse information component of the spread. While this indicates that liquidity worsens after the split,the increase in spreads isinversely related to changes in trades. Since we find that on average, the number of trades increases afterthe split, the decline in liquidity isless pronounced the larger the increase in the number of trades. 5. Conclusion This paper has examined the impact of the change in trading activity surrounding a stock spliton the liquidity of the stock. The results establish that the increase in volatility and spreads cannot be attributed solely to microstructure biases like price discreteness and bid-ask bounce. There isan increase in the number of trades around the split, and the larger the increase in the 33 number of trades, the larger isthe change in volatility and the smaller isthe change in spreads (total, as well as adverse information component). Since the change in the number of trades affects both the transient and permanent components of volatility, the increased trading activity cannot be attributed solely to either an increase in informed trading or an increase in noise trading. The finding that increased trades lead to lower spreads isconsistent either with these trades being predominantly noise motivated, or with the informed traders having very similar information (so thatcompetition between them reduces spreads). These findings suggest that any analysis of the impact of stock splits on traditional measures of liquidity (like volatility and spreads) must firstexamine why differentfirms seem to be more or less successful in attracting additional trades to their security. The subsequent consequences for liquidity then seem to be consistent with existing theories on the way in which a change in trading activity affects liquidity. An analysis of these causes for a change in the number of trades represents an interesting direction for future research. 34 APPENDIX A A.1 E s t im a t o r o f th e v o la t ilit y c o r r e c t e d f o r p r ic e d is c re te n e s s Ball (1988) shows that ifstock prices follow a Geometric Brownian motion with an instantaneous true underlying variance o2 and price P, then the bias induced by price discreteness can be approximated as dVbP2,where d isequal to 1/8. This approximation is valid for values of d/aP less than 2.50 (see Ball (1988), Table HI). To correct for this bias, we need to estimate 1/P2. Since there is a price trend in the pre-split period, we use the average of I/P3 from the estimation periods instead of one over the average of price squared. From Jensen’s inequality, since E(l/P2) > l/EtP2),the estimated bias would bias ittowards an upper value. For the sample of firms in this study, the average value of d/aP is 0.34 and 99.9% of the estimates are less than 2.50 (based on an estimate of s using bid-to-bid prices and an estimate of P using the average bid price in the estimation period). Hence, Ball’s approximation should be valid for our sample. W e correct for the bias due to price discreteness by deflating the volatility estimate using bid-to-bid prices as follows: 2 ^B.D = a5 **B ~ Y- 6(T2 -T,)t ^ P B2t where a BD is the volatility corrected for price discreteness, (A.1) isthe volatility estimated using bid- to-bid prices, (T2 - TO is the range of the estimation period, and PB,tis the bid price at time t(Ti^ t £ T 2). 35 B. 2 E s t im a t o r o f m u lt i- p e r io d v o la t ilit y W e use the following estimator of the k-period volatility,based on Lo and MacKinlay (1988): -k+l (A.2) ° ‘*k) S ^R , J ^ 6r|;p,2, where d*(k) = estimate of k-period volatility based on bid-to-bid prices and with correction for price discreteness (the second term corrects for discreteness), m = actual number of overlapping k-period observations, n = number of one-period (daily) observations, Rg = k-period return using overlapping one period returns based on bid-bid prices, jj.k= the sample mean of overlapping k-period returns, T = number of daily observations (= 180), and PB,,= bid price on day t. B .3 E s t im a t o r o f th e v a r ia n c e r a t io The variance ratio isdefined as the ratio of the k-period volatility to k times the oneperiod volatility. The volatility estimates are corrected for the biases due to the bid-ask bounce and price discreteness. For each firm i,the variance ratio isgiven by: 36 Var(R^) — VR(k) = k* GEt Var(R'B)- 2 yfk-j^l -lM (A~3) k J 6P2 where Var(Rg) is the k-period variance based on bid-to-bid returns. The quantity [d^P2] in the RHS of the above equation isthe adjustment for the bias die to price discreteness. The final term in the RHS of above corrects for the small sample bias in the expected value of the autocorrelation. Even ifthe returns are uncorrelated, the expected value of the autocorrelation is biased by -1/(T-1) [See Kendall ar.d Stuart (1977)]. 37 TA BLE1A Sample Characteristics for 366 Stock Split Announcements made by N A S D A Q - N M S Firms between January 1983 and December 1990, for the Total Sample, and for Sub-samples Classified by the Split Factor.* TotalSample Variable Mean N LargeSplitFactor Sub-Sample (SFAC £1.0) Mean Median 21L0 156 366 p-value'H’ SFAC 0.72 0.50 0.41 0.50 1.14 MVAL 259 123 176 85 371 182 NSHR pAV rl P“ pAV 7.36 4.38 6.52 4.10 8.50 4.93 0.110 25.80 22.80 19.65 17.51 34.14 31.61 <0.001 30.50 27.50 23.30 21.80 40.30 37.70 <0.001 17.70 16.50 16.50 15.50 19.30 18.0 0.001 PB r2 NMMKi 18.20 17.70 16.90 15.90 19.90 19.0 <0.001 7.80 6.00 7.30 6.00 8.40 7.0 0.12 nmmk2 7.00 6.00 6.70 6.00 7.50 6.0 0.32 I r2 t Median SmallSplitFactor Sub-Sample (SFAC 2 0.5) Mean Median 1.00 <0.001 N isthenumberofobservationsineachsub-sample; SFAC istheannouncedsplitfactor; MVAL isthe marketvalueofequity(in$ millions),measuredtwodaysbeforetheannouncementofthesplit; NSHR isthe numberofoutstandingshares,inmillions,asoftwodaysbeforetheannouncementofthesplit;PAVand PAVaretheaveragebidpricesinthepre-splitandthepost-splitestimationperiods,respectively;P*andP*are theclosingbidpricestwodaysbeforetheannouncementofthesplitandtwodaysaftertheex-splitday, respectively;andNMMKi andNMMK-i aretheaveragenumber ofmarketmakers inthepre-splitandpost splitperiods. ** p-ValueisfortheWilcoxon SignRank Sum testofdifferencesinmediansbetweenthetwo sub-samples. 38 T A B L E IB Changes in Samples Estimates of the Proportional Spread, Transaction Price Based Daily Volatility, and Measures of Trading Activity for 366 Stock Split Announcements Made by N A S D A Q - N M S firms between January 1983 and December 1990. 1 Variable, Subscript 1= Pre-Split Subscript2 = Post-Split SampleSize(N) Proportional s, Spread s2 S2/S, DailyReturn Volatility BasedonClosing TransactionPrices (x 104) of 2 of'O? AverageNumber ofDailyTrades AverageVolume Turnover (xlO3) | AverageVolume Turnoverper Trade(x 104) AverageNumber ofInsiderTrades (x 101) TotalSample Mean SFACS0.5 SFAC^l.O Median Mean 3.11 2.32 3.36 2.68 3.62 2.79 3.09 1.32** 1.23t+ 3.64 ♦* 1.20 7.73 5.63 12.42 8.80 366 ** 2.25 Median 210 1.70n Mean Median 1 156 II 2.78 1.77 | 2.48 1.08" 3.60 ♦* 1.49 8.29 5.95 6.97 4.56 11.38 8.37 13.83 9.67 ** 1.98 1.42" ** 2.61 1.45" 2.19" NT, 21.56 9.37 15.88 7.76 29.21 10.96 nt 2 32.33 12.00 20.77 9.95 47.90 ** 1.71 17.56 nt 2/n t , 1.53“ 1.28" 1.39” 1.12" 1.37" VT, 3.40 2.50 3.24 2.52 3.69 2.49 vt 2 3.80 2.30 2.41 3.90 2.13 v t 2/v t . 1.19“ 1.00 3.81 ** 1.24 1.07" 1.11 0.93 I (VT/NT), 3.19 2.64 3.45 3.00 2.83 2.09 (VT/NT)2 2.53 1.92 2.81 2.44 1.89 1.41 I | (VT/NT)2 /(VT/NT), ** 0.83 0.77" 0.93* 0.82" 0.68“ 0.65" 1 NIT, 11.98 6.11 10.21 5.56 14.37 6.67 nit2 11.76 ** 6.67 10.26 5.83 13.77 7.22 nit2/nit1 2.12 1.00 2.21" 1.00 2.01" 1.00 +t(*) The median ratioissignificantlydifferentfrom 1atthe 1%(5%)levelbasedon theWilcoxonSignRank Test. ( ) Hie mean ratioisstatisticallydifferentfrom 1atthe 1% (5%) levelusingasimplet-test. 39 I I | T A B LE2 Median Estimates of Daily Return Volatilities Before and After a Split, Based on Transaction Prices, Bid Prices, and Bid Prices Adjusted for Price Discreteness for the Total Sample and for the Sub-Samples by Split Factor. Foreachfirm,thepre-andpost-splitvolatilityestimatesarebasedon 180dailyreturnsinthetwoestimationperiodsrespectively. Medianestimatesreported inthetablearebasedon thecross-sectionalestimatesforeachsample.Allreportedmedianvaluesarescaledby afactorof104.d* isthevolatilityestimate basedon transactionpricereturns, d* isthevolatilityestimatebasedon bid-bidreturns,and d* Disthevolatilityestimatecorrectedforboththebid-ask bounceandpricediscreteness. Itiscomputedasfollows: d2 S.2 ^B.D a i (Ta-T,)£p2T where(T2 -TO istherangeoftheestimationperiod,andPB,isthebidpriceondayt. TotalSample,N = 366 SFAC £ 0.5,N = 210 SFAC £ 1.0,N= 156 PreSplit PostSplit Difference (p-value/ Ratio (p)tt PreSplit PostSplit Difference (p/ Ratio (P)ft PreSplit PostSplit Difference (P)f Ratio (P)n $ 5.63 8.80 2.90 (<0.001) 1.70 (<0.001) 5.95 8.37 1.74 (<0.001) 1.42 (<0.001) 4.56 9.67 4.66 (<0.001) 2.19 (<0.001) K 2.88 4.79 1.72 (<0.001) 1.80 (<0.001) 3 14 4.43 0.99 (<0.001) 1.41 (<0.001) 2.39 5.66 2.65 (<0.001) 2.24 (<0.001) 2.75 4.61 1.63 (<0.001) 1.81 (<0.001) 2.98 4.19 1.00 1.42 (<0.001) 2.38 5.25 2.61 (<0.001) 2.18 (<0.001) (<.001) t p-valuefortwo tailtestbasedon theWilcoxonmatchedpairsignranktestforHo:Median Difference= 0. tt p-valuefortwo tailtestbasedontheWilcoxonmatchedpairsignranktestforHq:Median Ratio= 1. 40 T A B LE3 Median Estimates of Multi-Period Return Volatilities in the 180 Day Pre-Split and Post-Split Periods, for the Total Sample and for Sub-Samples Classified by the Split Factor. The estimatorusedis n-k-4 Oc 2(k) = d^ m(l-k/n) X W - A J 6T where,m istheactualnumberofoverlappingk-periodobservations,nisthenumberofoneperiod(daily) observations, R*is thek-periodreturnusingoverlappingoneperiodreturnsbasedonbid-bidprices,|ikisthe samplemean oftheoverlappingk-periodreturns,andPBtisthebidpriceon day t. 1 Numberof Days Pre-Split Volatility (x 104) Post-Split Volatility (x104) Median Difference (x 104) Median Post-Split Volatility/Pre-Split Volatility TotalSample,N = 366 5 18.38 28.43 9.26” 1.62” 10 37.29 57.36 18.11” 1.58” 20 68.43 102.74 28.20” 1.54” 30 95.76 139.17 34.33” 1.57” SFAC £ 0.5,N = 210 5 19.06 27.19 5.69” 1.37” | 10 38.09 53.17 10.65” 1.32” I 20 75.79 97.96 12.11” 1.29” I 30 102.10 126.52 11.87” 1.32” I SFAC ^ 1.0,N = 156 * (**) f (n) I 5 16.76 32.33 12.85” 2.00” 10 36.29 64.71 25.04” 2.08” 20 61.51 111.56 47.74” 1.96” 30 77.24 146.57 55.65” ---- -- 1.81” — . ..n 1 Indicates a rejection of the null hypothesis that the median difference is zero based on the Wilcoxon Matched Pair Sign Rank Test, at the 5 % (1 % ) level for a two-tail test. Indicates a rejection of the null hypothesis that the median ratio is one based on the Wilcoxon Sign Rank Test, at the 5% (1%) level for a two-tail test. 41 TA BLE 4 Median Auto-Correlations ofReturns (xlO2)Based on Bid Prices fortheTotal Sample, and forthe Sub-Samples Classifiedby the SplitFactor. TotalSample,N=366 SI-AC £ 0.5, N= 210 SAFC 2:1.0,N=156 Lag Pre-Split Post-Split Difference (p-value)t Pre-Split Post-Split Difference (p-value)* Pre-Split Post-Split Difference (p-value)* 1 15.40 10.98 -3.19 (c.001) 14.57 11.45 -2.30 (0.094) 16.60 10.77 -4.85 (<0.001) 2 3.38 1.95 -1.16 (0.011) 2.46 3.81 0.60 (0.670) 4.52 0.24 -3.31 (<0.001) 3 1.21 -1.00 -1.89 (<0.001) 1.25 -0.45 -1.06 (0.111) 1.10 -1.31 -3.95 «0.001) 4 0.07 1.13 1.05 (0.12) -0.33 1.19 1.28 (0.277) 0.45 1.11 1.01 (0.286) 5 0.30 0.72 0.55 (0.62) 0.65 0.15 0.189 (0.716) -0.97 1.04 1.08 (0.239) 6 -0.30 0.05 0.56 (0.40) -0.01 -0.11 -0.55 (0.552) -1.21 0.77 2.23 (0.048) 7 -1.10 -0.97 -0.26 (0.62) -1.08 -0.92 -0.18 (0.950) -1.28 -1.12 -0.26 (0-410) 8 -1.15 -1.26 -0.21 (0.35) -1.14 -0.62 0.11 -1.16 -2.54 -0.66 (0.407) -0.73 (0.11) -2.18 -1.56 0.24 (0.866) -0.09 -2.24 -2.41 (0.008) 0.22 -1.10 -0.14 0.40 (0.394) -0.48 -1.29 -0.53 (0.384) 9 -1.38 -2.08 10 -0.85 -0.51 (0.602) (0.99) t p-value based on the Wilcoxon matched pairs sign rank test. H o :Median difference = 0. 42 TA BLES Median Estimates of Variance Ratios During the 180 Day Pre-Split Period and the 180 Day Post-Split Period, For the Total Sample and for Sub-Samples Classified by the Split Factor. The estimatorforthevarianceratiois VR = where,Var(R‘ ,)isthe 1-period(3-day)dayreturnvariancebasedonbid-to-bidreturns,andVar(R‘)isthek-period (30-day)returnvariancebasedon bid-bidprices,andk isequalto 10.Thequantity[(1/6P2]istheadjustmentfor thebiasduetopricediscreetness,whered isequalto1/8th.,andP '.sthepriceofthestock. The finalterminthe equationcorrectsforthesmallsamplebiasintheexpectedvalueoftheautocorrelation,andT isthenumberof one-periodreturnsused. r r I Pre-Split Post-Split Difference ™ (N = 366) 0.94 0.79 -0.13** SFAC <i 0.5 (N=210) 0.95 0.88 -0.12** SFAC £1.0 (N=156) 0.93 0.75 -0.15** Indicatesarejectionofthenullhypothesisthatthemediandifferenceiszerobased on theWilcoxonMatched PairSignRank Test,atthe 1percentlevel. 43 TA B LE6 Ordinary Least Squares Estim ates of Model of the Determinants of the Change in the B ias Corrected V olatility Follow ing Stock Sp lits/ 1 Dependent Variable Intercept ,( nnt^ In n t J -0.892 (-1.68)* 0.393 (4.62) /V 2> i.1 2 X 2J V^l -1.136 (-2.07)** 0.428 (4.25)*** '2 ' In °B,D,2 .CTB.D.l, -0.196 (-0.53) 0.117 (1.73)* (1+SFAC) I nit J ln(MVAL) | ' i ' B,D,2 In °_2 B.D.l, | lnf ( 0.028 (0.53) -0.22 (-0.38) 0.042 (1.11) y 2 "N Adjusted R 2 0.173 (1.75)* 0.089 (2 .02)*** 0.08 0.198 (1.66)*** 0.086 (1.84)** 0.08 0.051 (0.65) 0.035 (1.11) 0.633 (18.25)*** 0.52 o*D,isthedailyvolatilityinperiodj,correctedforthebiasduetoboththebid-askbounceandpricediscreteness(j= 1forthepre-splitperiodand 2forthe post-splitperiod);E*isthevolatilitybasedon30-dayreturnsinperiodj;NNTj isthenumberoftradesinperiodj,netofinsidertrades;NITjisthenumberof insidertradesinperiodj;SFAC istheannouncedsplitfactor;andMVAL isthemarketvalueofthesplittingfirm’sequity,measuredtwodaysbeforethe announcementofthesplit. t-statisticsareinparenthesis.*indicatessignificanceatthe10% level,** atthe5% leveland *** atthe 1% level. 44 T A B LE7 Effect of Price Discreteness on Absolute Spreads Around Stock Splits for the Total Sample and for Sub-Samples Classified by the Split Factor. The table reports the cross-sectional mean and median values for the ratio of the median post-split spread to the median pre-split spread (AS), and also for the ratio of the median post-split spread to the median target spread in the post-split period (AS*). For each firm i, . ABS,. and AS = • '' TSW ABSU ABSW where ABSjj isthe median absolute spread in periodj (j= 1 for the pre-splitperiod and 2 for the post-split period), TS2,i= {ABSi,i/(l+SFAQ)} rounded up to the next highest eighths, and S F A Q isthe splitfactor. Total Sample (N = 366) r AS AS* Mean (p-value)f Median (p-value)n Mean (p-value)t Median (p-value)t+ 1.55 (< 0.001) 1.50 « 0 .001) 1.53 (< 0.001) 1.33 (< 0.001) SFAC < 0.5 (N = 210) 1.36 (< 0.001) 1.50 (< 0.001) 1.42 (< 0.001) 1.33 (< 0.001) SFAC > 1.0 (N = 156) | 1.80 (< 0 .001) 2.00 (< 0.001) 1.69 0 .001) 2.00 « 0.001) 1 [ I I (< I I I 1 f p-values are for the t-test that the sample mean isequal to one. t+p-values are for the Wilcoxon Signed Rank test that the sample median isequal to one. 45 TA B LE8 Changes in the Average Proportional Spread and its Order Processing and Adverse Information Components Following Stock Splits. 1Total Sample | Small Split IFactor SubSample (SFAC < 0.5) Large Split Factor SubSample (SFAC >1.0) Pre-Split 3.111 Post-Split 3.624 Difference 0.513*” Order Processing 1.683 1.947 0.264*” Adverse Information 1.428 1.677 0.249*** Spread 3.360 3.638 0.278*** Order Processing 1.783 1.765 Adverse Information 1.577 1.873 0.296*** Spread 2.777 3.605 0.828*” Order Processing 1.498 2.067 0.569*** Adverse Information 1.279 1.538 0.259**’ Spread 46 -0.018 TA BLE 9 Ordinary Least Squares Estimates of M odels of the Determinants of the Change in the Proportional Spread Follow ing Stock Sp lits/ Dependent Variable Intercept , f NNT,1 I nnt J Infs 0 kA j -1.338*** (-7.49) -0.429 (-13.06)*** Infs 0 j -1.355 *** (-7.64) -0.435*** (-13.57) H Is , J -0.923*♦* (-4.59) , f Nrr,^ I m J -0.017 (-0.88) -0.070*** (-2.82) (1 + SFAC) ln(MVAL) Adjusted R2 0.239 (6.16) 0.105 (6.90)” * 0.39 0.240#** (6.20) 0.107 *** (7.01) 0.:9 0.139 (2.19)** 0.076*** (4.03) 0.10 Sjistheproportionalspreadinperiodj(j= 1forthepre-splitperiodand2forthepost-splitperiod);NNTj isthenumberoftradesinperiodj,netofinsider trades;NITjisthenumberofinsidertradesinperiodj;SFAC istheannouncedsplitfactor;andMVAL isthemarketvalueofthesplittingfirm’sequity, measuredtwodaysbeforetheannouncementofthesplit * t-statisticsareinparenthesis.*indicatessignificanceatthe10% level,**atthe5% leveland ***atthe1% level. 47 T A B L E 10 Ordinary Least Squares Estimates of Models of the Determinants o f the Change in the Adverse Information Component of the B id -A sk Spread Follow ing Stock Splits. Dependent Variable In IsJ In ( s 2 ] j Infs 0 ^s t J , NNT.'i Intercept . fc o v ,^ Ico v J -0.486*** (-5.26) 0.800 (33.21)*” -0.171*** (-9.47) -0.511*** (-5 53) 0.799 (32.92)*” -0.180*** (-10.12) -0.247*** (-2.49) 0.898 (36.99)*” { I , nnt J f Nrr,^ N J -0.024*** (-2.49) -0.042*** (-4.04) Adjusted R2 I (1 + SFAC) ln(MVAL) -0.015 (-0.73) 0.049 (6.32) 0.85 -0.013 (-0.62) 0.051 (6.54)*** 0.85 -0.079 (-3.60)*** 0.033 (3.84)*** 0.81 Sjistheproportionalspreadinperiodj(j= 1forthepre-splitperiodand2 forthepost-splitperiod);COVj isthecovariancebetweenRf andR°,,whereRj5is thedifferencebetweenthetransaction-pricebasedreturnandthereturnbasedonbid-to-bidprices; NNTj isthenumberoftradesinperiodj,netofinsider trades;Nil)isthenumberofinsidertradesinperiodj;SFAC istheannouncedsplitfactor;andMVAL isthemarketvalueofthesplittingfirm’sequity, measuredtwodaysbeforetheannouncementofthesplit. f t-statisticsareinparenthesis.*indicatessignificanceatthe10% level,**atthe5% leveland *** atthe1% level. 48 R eferen ces Admati, Anat R. and Paul Pfleiderer. "A Theory of Intraday Patterns: Volume and Price Variability," R e v ie w o f F in a n c ia l S tu d ie s , 1988-1989, vl(l), 3-40. Baker, H. Kent, and PatriciaL. Gallagher, “Management’s View of Stock Splits,” F in a n c ia l M a n a g e m e n t, 1980, v9(2), 73-77. Baker, H. Kent, and Gary E. Powell, "Why Companies Issue Stock Splits," M a n a g e m e n t, 1992, v21(2), 11. Ball, Clifford A. "Estimation Bias Induced by Discrete Security Prices," 1988, vl8,841-865. Black, Fischer. "Presidential Address: Noise," J o u r n a l o f F in a n c e , F in a n c ia l J o u r n a l o f F in a n c e , 1986, v41C3), 529-544. Brennan, Michael J. and Thomas E. Copeland, "Stock Splits, Stock Prices and Transaction Costs," J o u r n a l o f F in a n c ia l E c o n o m ic s , 1988, v22(l), 83-101. Brennan, Michael J. and Patricia J.Hughes, "Stock Prices and the Supply of Information," J o u r n a l o f F in a n c e , 1991, Vol. 46(5), 1665-1691. Conroy, Robert M., Robert S. Harris and Bruce A. Benet, "The Effect of stock Splits on Bid-Ask Spreads," J o u r n a l o f F in a n c e , 1990, v45(4), 1285-1296. Copeland, Thomas E. "Liquidity Changes Following Stock Splits," v34(l), 115-141. J o u r n a l o f F in a n c e , Copeland, Thomas E. and Dan Galai. "Information Effects of the Bid-Ask Spread," F in a n c e , 1983, v38(5), 1457-1469. 1979, Jo u rn a l o f Dravid, Ajay R., “A Note on the Behavior of Stock Returns around Ex-Dates of Stock Distributions,” J o u r n a l o f F in a n c e , 1987, v42(l), 163-168. Dubofsky, David A. "Volatility Increases Subsequent to N Y S E and A M E X Stock Splits," J o u r n a l o f F in a n c e , 1991, v46(l), 421-431. Elgers, Pieter T. and Dennis Murray, “Financial Characteristics Related to Management’s Stock Split and Stock Dividend Decisions,” J o u r n a l o f B u s in e s s F in a n c e a n d A c c o u n t in g , 1985, vl2(4), 543-551. French, Kenneth R. and Richard Roll. "Stock Return Variances: The Arrival of Information and the Reaction of Traders," J o u r n a l o f F in a n c ia l E c o n o m ic s , 1986, vl7(l), 5-26. George, Thom as J., Gautam K au l, and M . Nim alendran. "Estim ation o f the B id -A sk Spread and Its Com ponents: A New Approach," Review of Financial Studies, 1991, v 4 ,623-656. Gottlieb, Gary and Avner Kalay. "Implication of the Discreteness of Observed Stock Prices," J o u r n a l o f F in a n c e , 1985 v40(l), 135-153. Jegadeesh, Narsimhan, and Avanidhar Subrahmanyam, “Liquidity Effects of the Introduction of the S&P 500 Index Futures Contract on the Underlying Stocks,” J o u r n a l o f B u s in e s s , 1993, v66(2), 171-188. Jones, Charles M., Gautam Kaul, and Marc C. Lipson, “Information, Trading, and Volatility,” J o u r n a l o f F in a n c ia l E c o n o m ic s , 1994, v36(l), 127-154. Kaul, Gautam and M. Nimalendran. "Price Reversals: Bid-Ask Errors or Market Overreaction?" J o u r n a l o f F in a n c ia l E c o n o m ic s , 1990, v28,67-93. Kendall, Maurice and Alan Stuart, T h e 1977. A d v a n c e d T h e o r y o f S t a tis tic s , New York, Macmillan, Karpoff, Jonathan M., “The Relationship Between Price and Trading Volume: A Survey,” J o u r n a l o f F in a n c ia l a n d Q u a n tita tiv e A n a ly s is , 1987, v22(l), 109-126. Lakonishok, Josef and Baruch Lev. "Stock Splits and Stock Dividends: Why, W h o and When," J o u r n a l o f F in a n c e , 1987, v42(4) 913-932. Lamoureux, Christopher G. and Percy Poon. "The Market Reaction to Stock Splits," F in a n c e , 1987 v42(5), 1347-1370. Jo u rn a l o f Lo, A. and C. MacKinlay. "Stock Market Prices Do Not Follow Random Walks: Evidence form a Simple Specification Test," T h e R e v ie w o f F in a n c ia l S tu d ie s , 1988, vl, 41-66. Madhavan, Ananth, and Seymour Smidt, “A Bayesian Model of Intraday Specialist Pricing,” J o u r n a l o f F in a n c ia l E c o n o m ic s , 1991, v30(l), 99-134. Maloney, Michael T. and J.Harold Mulherin, “The Effect of Splitting on the Ex: A Microstructure Reconciliation,” F in a n c ia l M a n a g e m e n t, 1992, v21(4), 44-59. McNichols, Maureen and Ajay David. "Stock Dividends, Stock Splits, and Signaling," F in a n c e , 1990, v45(3), 857-880. Jo u rn a l o f Murray, Dennis. "Further Evidence on the Liquidity Effects of Stock Splits and Stock Dividends," J o u r n a l o f F in a n c ia l R e s e a r c h , 1985, v8(l), 59-67. O hlson, James A . and Stephen H. Penman. "Volatility Increases Subsequent to a Stock Split: A n Em p irical Aberration," Journal of Financial Economics, 1985, v l4 (2 ), 251-266. Roll, R. "A Simple Measure of the Effective Bid-Ask Spread in an Efficient Market," F in a n c e , 1984, v39,1127-1139. Jo u rn a l o f Seyhun, H. Nejat,“The Information Content of Aggregate Insider Trading,” J o u r n a l o f B u s in e s s , 1988, v61(l), 1-24. Stoll, H.R. "Inferring the Components of the Bid-Ask Spread: Theory and Empirical Tests," J o u r n a l o f F in a n c e , 1989, v44, 115-134.