Stock Splits Information or Liquidity?

Transcription

1 Stock Splits Information or Liquidity? Alon Kalay University of Chicago Booth School of Business Mathias Kronlund University of Chicago Booth School of Business Original version: November 4, 2007 Current version: April 5, 2009 Abstract We empirically test liquidity- and information-related theories that seek to explain the abnormal returns around stock split announcements. While much of the recent literature on stock splits has focused on liquidity-related theories, our evidence is mostly consistent with an information hypothesis, whereby a stock split is interpreted by the market as good news about the firm s future performance. In particular, we find that analysts on average increase earnings estimates for the next fiscal year by 2-3% around stock split announcements, suggesting that analysts expectation of the fundamental performance of the firm increases following the split announcement. Further, we find that abnormal returns around stock splits are related to the information environment of the firm, as with fewer analysts are associated with higher abnormal returns. While splitting typically have experienced high levels of past earnings growth, we find that splitting nevertheless experience less mean reversion of earnings growth compared to a matched sample of. We also examine abnormal returns, stock splitting activity and prices following the tick size changes on the NYSE in 1997 and 2001 as a natural experiment for changes to the tick-to-price ratio, which is an important element in many liquidity-related theories. Our findings are inconsistent with the prediction of these liquidityrelated theories on several accounts: there is no systematic change in splitting activity or splitting prices following tick size changes as predicted by effective tick size theories, and abnormal returns do not appear to be related to prices in a way that is consistent with predictions of an optimal trading range. In the analysis, we control for earnings announcements and dividend changes that often coincide with stock splits. The authors thank Ray Ball, Phil Berger, Eugene Fama, Avner Kalay, Christian Leuz, Doug Skinner, and participants and the University of Chicago Accounting and Finance brown bags, for helpful comments and suggestions. Kronlund gratefully acknowledges financial support from Liikesivistysrahasto, Osk. Huttusen Säätiö, and Suomen Kulttuurirahasto. Alon Kalay: akalay@chicagobooth.edu, Mathias Kronlund: mkronlun@chicagobooth.edu

2 1 Introduction Many previous studies have documented abnormal returns surrounding stock split announcements. However, the underlying reason why we observe such abnormal returns has not been resolved. While a stock split appears to be a merely superficial change to a security, two broad theories have emerged in the literature to explain the abnormal returns around stock splits. The first theory broadly relates to increased liquidity that stocks may achieve via a split, while the second theory relates to managers private information about fundamentals that is released to the market via stock splits. In this paper, we perform several tests to distinguish between the current theories that seek to explain the abnormal returns around stock split announcements. We analyze the change in analyst earnings forecasts as a new measure of the change in the market s expectations of fundamentals surrounding the stock split. We also use the tick size changes on NYSE in 1997 (from 1/8 th to 1/16 th of a dollar) and 2001 (from 1/16 th to pennies) as a natural experiment for an exogenous shock to the tick-to-price ratio, which is an important element in several variants of the liquidity-related theories. Finally, we analyze the relationship between the prices at which stocks split and abnormal returns, since lower stock price levels achieved via splits have been hypothesized to increase stock liquidity. These tests combined, constitute a comprehensive empirical comparison of the main theories of stock splits to date. In the liquidity-related theories, the underlying mechanism by which a stock split can increase the liquidity of the stock is either by changing the price level itself or by changing the tick-toprice ratio. The first of these theories implies that stock splits, by lowering the price level, can increase liquidity by bringing the stock price to an optimal trading range, which induces more investors to enter the market for the security. This could result either from a constraint that small investors can face when the stock price is high, as small investors may not be able to afford round lots when the stock price is too high (Baker and Gallagher (1980)), or from a behavioral preference for certain stock prices, whereby investors prefer to buy stocks with a lower price and managers therefore split their stock to keep the nominal price constant (as in Benartzi et al. (2005)). In the other liquidity-related theories, changing the tick-to-price ratio may either increase broker-mediated trading by increasing the profits brokers make per trade (Brennan and Hughes (1991), Schultz (2000)) in a commission-induced sponsorship hypothesis, or by lowering bid-ask spreads by reducing information asymmetry between informed and uninformed 2

3 traders (Angel (1997), Anshuman and Kalay (1998, 2002)) in an optimal effective tick size hypothesis. We discuss these theories in more detail and lay out their testable predictions in Section 2 of the paper. In an information-related theory, a stock split conveys managers private information related to the fundamental performance of the firm (Fama, Fisher, Jensen and Roll (1969), henceforth FFJR, and Grinblatt, Masulis and Titman (1984), henceforth GMT). On average, the market interprets the stock split as positive news of future firm performance resulting in abnormal returns on the stock split announcement. To examine this hypothesis, previous studies have analyzed the path of actual dividends and earnings after stock splits (for example FFJR, Lakonishok and Lev (1987)). However, such measures of ex-post performance are at best noisy measures of expected performance, and it is also difficult to form a null hypothesis of what the market s expectations of future performance was before the split. In contrast, we use changes in analysts earnings forecasts around the split as a direct measure of the change in the market s expectations of future firm performance before and after the stock split. To distinguish between the liquidity and information-related hypotheses, we first analyze abnormal returns, splitting activity and the association between abnormal returns and the presplit price level as well as the information environment of the firm. Our sample consists of all stock split announcements on the NYSE with a split factor of 2-for-1 and above that were announced between 1988 and 2007, and we split the sample into three regimes based on the prevailing tick size on the NYSE (1/8 th, 1/16 th, and 1/100 th of a dollar). We control for simultaneous announcements of earnings and dividend changes that often coincide with stock split announcements. We confirm that abnormal returns are significant across the full time period in our sample, and we also find that abnormal returns are significant in each of the three tick size regimes. Abnormal returns in the full sample are 1.76% in a three-day window around the split declaration date, and 1.46% when controlling for announcements of earnings and dividend changes. We do not find that splitting activity systematically increases or that splits occur at lower prices following the tick size changes on NYSE, which would be predicted by both an effective tick size hypothesis and a commission-induced sponsorship hypothesis. We also find evidence that abnormal returns are related to the price at which stocks split, as that split at lower stock prices are associated with higher abnormal returns around the split. This result does not appear to 3

4 be consistent with the prediction of optimal trading range theories if higher priced stocks are further away from their optimal range. Both of these two latter results are inconsistent with what we would expect under the hypothesized mechanisms that increase liquidity in the liquidityrelated theories. We further find that abnormal returns around stock splits are related to variables that proxy for the information environment of the firm, such as market capitalization, the number of analysts that follow the firm, and trading volume. This correlation suggests that with more publicly available information are associated with lower abnormal returns around stock split announcements. When all three proxies for the information environment are included in a multiple regression, the number of analysts is the only significant explanatory information variable for abnormal returns. These results suggests that a stock split is interpreted by the market as more positive news for where there is less available information, and most notably when there are fewer analysts that cover the firm. Finally, we find that analysts increase their earnings forecasts for the next full fiscal year by 2.3% around stock split announcements (the average revision is even higher, 2.7%, when we exclude observations with coinciding dividend changes and earnings announcements). To the extent that analysts forecasts proxy for the market s expectations of future performance, this result suggests that stock splits convey positive private information related to the fundamental performance of the firm, which is consistent with an information hypothesis. Splitting have often experienced very high levels of past earnings growth, which makes it unlikely that such earnings growth could continue in the future. Nevertheless, consistent with prior hypotheses (Asquith, Healy and Palepu (1989) and Ikenberry and Ramnath (2002)), we find that splitting experience a lower level of mean-reversion of actual earnings growth after the split relative to a matched set of with similar past earnings growth. Therefore, it is possible that the change in analysts earnings forecasts around stock splits represents an increase in analysts weights on permanent versus transitory components of earnings. That is, as the relevant null (i.e. expected future earnings growth without the split) for with high past earnings growth would be a reduction in future earnings growth, the analysts reaction to the split may be an upward revision from an otherwise mean-reverting earnings growth forecast. This finding is also consistent with prior research and helps address the conclusions reached by GMT and Nayak and Prabhala (2001), who find that there is also information in the split which is not 4

5 related to changes in regular dividends. Such a hypothesis also has the potential to address the finding (first documented by GMT and further developed by Nayak and Prabhala (2001)), that non-dividend paying are associated with higher abnormal returns around splits, since nondividend paying on average have a higher transitory share of earnings (Skinner and Soltes (2008)) and may elicit stronger analyst forecast revisions as a result. On balance, our results suggest that the abnormal returns around stock splits are most consistent with an information hypothesis, and not consistent with several variants of liquidity-related hypotheses. An information story thus warrants renewed attention as an explanation for the market s reaction to stock splits. 2 Stock split theories and empirical predictions Ever since Fama, Fisher, Jensen and Roll s seminal paper (1969), financial economists have sought to understand why markets react to stocks splits, as a stock split appears to be a merely cosmetic transaction that increases the number of shares outstanding that the firm has while reducing its share price by the split factor. Multiple theories have since been proposed to explain the role and purpose (if any) of stock splits beyond this cosmetic effect and thus to explain the observed market reaction. In our review of the literature, we identify four main categories of hypotheses for stock splits that seek to explain this phenomenon, namely: i) Achieving an optimal stock price, or trading range ii) Achieving an optimal effective tick size iii) Motivating spread - or commission induced sponsorship iv) Providing new information about fundamentals to the market. The first three of these hypotheses broadly relate to the liquidity or the demand for trading the stock, while the fourth hypothesis is distinct in that it relates to the fundamental performance of the firm. These categories are similar to the ones discussed in Benartzi et al. (2005). In this paper, we develop predictions for each of the categories of theories above as they relate to: abnormal returns, the splitting price levels in each of the three tick size regimes on the NYSE (1/8 th, 1/16 th, and 1/100 th of a dollar), the aggregate splitting activity (number of splits) around the tick size regime changes in 1997 and 2001, the relationship between abnormal returns and pre-split prices, and analyst earnings forecast revisions around the announcement of a split. We then test these predictions across the tick size regimes to analyze to what extent each category of hypotheses is supported by this empirical evidence. In this section, we first briefly discuss each 5

6 of the four categories of theories in the literature, and outline their empirical predictions as they relate to our tests. 2.1 Achieving an optimal trading range One common hypothesis is that split their stock to keep stock prices in an optimal range. Baker and Gallagher (1980) hypothesize that small investors cannot afford to buy round lots when share prices are too high and that lowering the stock price therefore attracts more small investors. This view is has often also been used by management to explain why their split their stock. Baker and Gallagher (1980) and Baker and Powell (1992) conduct surveys of CFOs and find that CFOs cite keeping their stock price in an optimal trading range as one of the main reasons why they split the stock. In support of the hypothesis that this is important for small investors, Schultz (2000) documents empirical evidence of a higher presence of small buy orders around the time of the split. Lakonishok and Lev (1987) also argue that splits help return stock prices to their normal trading range and show that post split prices tend to converge to historic levels. Additionally, McNichols and Dravid (1989) show that the stock split factor is larger the further away the price is from the norm. More recently, Benartzi et al. (2005) show that the average nominal price for stocks in the U.S. has remained relatively constant at approximately $30 - $40 since the Great Depression, while other countries have not followed the same price obsession. However, Benartzi et al. (2005) argue that the underlying reason for the price maintenance is still a puzzle since they suggest that the optimal trading range should only be valid for real rather than nominal prices. The arguments above suggest that a stock s liquidity can be increased by achieving an optimal trading range and stock splits thereby should result in a short term price increase as future expected returns decline due to increased liquidity (Amihud and Mendelson (1986)). However, the theories are somewhat vague with regard to the underlying mechanism by which stock prices in an optimal trading range achieves increased liquidity and the original arguments (e.g., small investors not affording round lots when prices are high) seem less likely to apply today. In addition, we cannot rule out that cheaper stocks simply enjoy higher liquidity for some behavioral reason. 6

7 2.2 Achieving an optimal effective tick size Another group of theories for stock splits relate liquidity to the tick-size-to-price ratio or the effective tick size. Angel (1997) reviews theoretical arguments in this spirit and presents some empirical evidence in support of this hypothesis. Anshuman and Kalay (1998, 2002) develop a market micro structure model based on a rational expectations model of trade, that explains the benefit of an optimal effective tick size by distinguishing between two elements of the bid-ask spread: i) an asymmetric information component that results from the ratio of informed vs. uninformed traders in the market and ii) a discreteness related spread (due to the tick size relative to the price). In their model, in equilibrium, equate the marginal cost and marginal benefit of these two spread components to find the optimal effective tick size. They also present empirical evidence in support of their model. Nevertheless, a similar model by Easley, O Hara and Saar (2001) argues that stock splits can change clientele to include more uninformed traders without leading to ultimately lower costs. A related argument by Harris (1994) is that a nontrivial tick size enforces time and price priority in the limit order book and thus encourages market depth. That is, a larger tick size can create incentives for liquidity traders to enter their orders in a timely manner due to the significant cost associated with waiting and being forced to move a tick up or down 1. In sum, these mechanisms can cause the relationship between the tick-size-to-price ratio and liquidity to be non-monotonic, and can thus increase liquidity by targeting an optimal tickto-price ratio using stock splits. 2.3 Motivating spread or commission induced sponsorship Another group of hypotheses related to stock splits in the prior literature focuses on the commissions that brokers receive from trades, whereby stocks with a lower price can achieve increased analyst following and marketing by brokers. Schultz (2000) documents that the effective commission (effective spread) is higher for lower priced stocks and argues that brokers 1 Supporting this view, Lipson (University of Georgia Research Magazine; documents a significant increase in institutional trading costs on the NYSE around 2001 when the minimal tick size was reduced to a penny as a result of line jumping and fewer limit order trades being fulfilled. According to Lipson, pension and mutual funds have reacted to the tick change by reducing the number of orders placed in the limit order book. 7

8 as a result promote stocks that have lower prices. Therefore, companies can get brokers to promote their stock by splitting their shares, which in turn can increase demand for the security. Brennan and Hughes (1991) further hypothesize that analyst coverage increases for stocks that are more lucrative for brokers. Therefore, managers with good inside information may split their stocks to increase analyst following and the amount of information related to their stock in the market. However, Ikenberry and Ramnath (2002) find that the increase in the number of analysts following a stock after a split announcement is similar to that experienced by a matched set of, implying that a relationship between prices and analyst following is not likely to be of substance. 2.4 Providing new information about fundamentals to the market The final category of hypotheses suggests that managers split their stock to inform the market that they expect improved future performance, for example, in the form of more sustainable increases in earnings and dividends. The market thus interprets the stock split as good news. FFJR originated this idea in their original event study on splits, where they document unusually high abnormal returns preceding the splits (they use the split exercise date and not the declaration date as the event date) that in turn predict increases in future dividends. Past studies have also found a correlation between split announcements and earnings performance (Lakonishok and Lev (1987) Asquith, Healy and Palepu (1989), Ikenberry and Ramnath (2002)). In one of the most extensive stock split studies conducted since FFJR, GMT proceed to document abnormal returns around the announcement dates as well as the ex-dates in their sample, even after controlling for a wide range of news announcements that tend to accompany stock splits. They further show that the abnormal returns for stock splits are higher for nondividend paying relative to dividend paying. They conclude that the results are consistent with an information based hypothesis since who pay dividends have an alternative mechanism with which to communicate their positive expectations to the market. They further suggest that a split could either be a way for managers to signal their private information to the market (signaling hypothesis) or a way to draw attention to the firm (attention hypothesis) which is an action that mangers of undervalued would seek, thus resulting in a positive announcement effect. Nayak and Prabhala (2001) attempt to explicitly measure the portion of the announcement effect that arises from changes in dividend expectations and 8

9 conclude that 46% of the split valuation effect is unrelated to changes in dividend payment expectations. Nevertheless, there are several empirical results that seem to contradict the information hypothesis. Muscarella and Vetsuypens (1996) document that American Depositary Receipts (ADRs) tend to split in the U.S. market even when the firm s stock does not split in the home country, and that such solo-splits are accompanied by abnormal returns. If we believe that the role of splits is to reduce asymmetric information, it would be hard to argue that the depositary banks have inside information related to the firm that the managers in the firm s home country do not have 2. Furthermore, while the idea that stock splits could convey private information is simple, there are a few theoretical issues that remain unresolved in the literature. First, the potential cost of a false signal (i.e. for stocks splitting without news) is not clear. Some potential explanations could include reputation risk or a future negative market reaction as a result of a false signal 3. One additional cost may be that if a firm s stock price falls below a certain threshold some institutional investors may be excluded from holding it or the firm may fail to meet the exchange s minimum price requirement. 2.5 Empirical predictions We develop empirical predictions for each of the four categories of hypotheses across five key dimensions: a) The presence of abnormal returns around the split announcement (in a three day trading day window), b) The price levels at which stocks are expected to split, c) The predicted aggregate number of splits after the tick size change, d) The relationship between pre-split prices and abnormal returns, and e) The predicted change in analysts EPS forecast around the split announcement (based on the change between the consensus forecasts before and after the split announcement). We summarize the predictions for each of the four categories of hypotheses in Table 1. For the predictions related to the optimal effective tick size hypothesis, we assume that there has been no 2 Benartzi et al. (2005) also highlight the splitting activity of ETFs and Mutual Funds as evidence that splitting is not solely driven by the desire to reduce asymmetric information. However, it is plausible that splits may serve a different role for common stocks than they do for mutual and exchange traded funds. 3 Ikenberry et al. (1996) find negative excess returns three years following the split for a small sample of that that have a negative price run up prior to the split. 9

10 underlying structural changes in the market micro structure throughout the period in question that could affect the optimal effective tick size other than the tick size regime changes in 1997 and 2001 (examples of such confounding changes could be a reduction in the cost of acquiring trading related information, or a change in the level of asymmetric information in the market). Second, although the spread or commission induced sponsorship hypothesis predicts an increase in the number of analysts following the stock, our paper focuses only on the change in EPS forecast and not in the total analyst following which prior research has been found to be insignificant. 3 Research design and descriptive statistics The empirical analysis consists of four parts. First, we conduct an event study of abnormal returns around split announcement dates to document whether splits are associated with significant abnormal returns in each of the tick size regimes. Second, to test if such abnormal returns are consistent with the liquidity-based theories discussed above, we analyze whether splitting activity and pre-split price levels change in response to the changes in tick size. We also examine the relationship between pre-split stock prices and abnormal returns around the split announcement. Third, we analyze changes in analyst EPS forecasts around the split announcement dates to see whether analysts significantly revise their earnings expectations after a stock split. Finally, we examine the pre-versus post-split earnings growth of splitting relative to a matched sample of to characterize what kind of earnings information (if any) might be conveyed via the split. We obtain a list of all stock splits on NYSE with a declaration date between 1 January 1988 and 31 December 2007 from CRSP. We only include stocks that are listed on NYSE as these stocks were affected by the tick size regime changes that is employed as a natural experiment in our tests. We start our sample in 1988 since I/B/E/S coverage (which we require for our study on analyst estimates) is very limited prior to this period. We include splits for common stock (CRSP event code 5523), for which the split factor is at least 2:1. The resulting sample consists of 1306 splits. Limiting our sample to splits with a factor of at least 2:1 makes the empirical tests more applicable to the effective tick size hypothesis, as maintaining an optimal effective tick size would require relatively large price factor adjustments in response to tick size changes. Second, limiting the split factor to a minimum of 2:1 limits the number of stock dividends in our sample, which may behave differently from stock splits. Another commonly used threshold in the 10

11 literature is 5:4, used for example by GMT, which is the accounting definition for the maximum payout in a stock dividend. We assign each stock split to a tick size regime as follows: the first regime (1/8 th tick size) is from 1 January 1988 to 23 June 1997, the second regime (1/16 th ) is from 24 June 1997 to 28 January 2001, and the third regime (1/100 th ) is from 29 January 2001 to 31 December Table 2 (Panel A) reports descriptive statistics for the that split. The total number of unique in our sample is 879. The mean number of splits per firm is 1.49, the median is 1 and the maximum number of splits by any firm is 6. The mean pre-split price is $70.29 (we use the price of the last day of the month prior to the split). We also see that some split even with negative earnings and negative book equity. Table 2 (Panel B) reports statistics by calendar year for the market capitalization and earnings per share for the splits in our sample. The mean and median market cap of that split this stock increased over the 90 s and peaked in 1999, but has declined somewhat since. The median EPS has been declining over time reaching a low during 2000, during the end of regime 2, at $0.64/share and then reverting back to 1990 levels of above $1.00/share in The distribution of split factors is presented in Table 3. An overwhelming majority (92%) of splits in our sample have a factor of 2:1. To measure abnormal returns around the announcement dates we obtain stock returns from CRSP and calculate the cumulative return net of the value-weighted market return over three trading days (-1 to +1 days) around the split announcement date. As our split announcement date, we take the declaration date from CRSP. In some cases, it is possible that news about the split has already leaked prior to the official declaration date, but this should only bias against finding any significant abnormal returns. Splits are often announced in conjunction with earnings or dividends. To measure abnormal returns for splits where the returns are less likely to be contaminated by information other than the split itself, we control for stock splits that coincide with quarterly earnings announcements and announcements of dividend changes. To do this, we flag observations where there is either an earnings announcement or a dividend change announcement in the 3-day window around the split announcement. We obtain dividend announcement dates and dividend amounts from CRSP and define a dividend change announcement as an announcement of any cash dividend (CRSP 11

12 distribution code 12xx or 13xx) for which the previous announcement of the same type was not of the same amount. We obtain the earnings announcement dates from Compustat. We are able to link 1277 of our sample of 1306 splits to in Compustat. If a split cannot be linked to a firm in Compustat, we take a conservative approach and assume that there has been an earnings announcement in the window. Table 4 reports the number of splits that coincide with earnings and dividend change announcements. We see that ~24% of the splits coincide with earnings announcements, and ~38% coincide with dividend changes in the 3-day window. Since some of the earnings and dividend change announcements overlap, ~50% of the splits in our sample coincide with at least one of these announcements in the three-day window. To measure changes in market expectations of earnings surrounding the split announcement we measure the revision in analyst earnings estimates from before and after the split. Using the changes in analyst forecasts as a proxy for changes in market expectations of earnings has an advantage over other methods used to measure the relationship between split and the fundamental performance of the firm, since we do not need to make any separate assumptions about what the market had thought before the split about the firm s future earnings path (that is, it is hard to define a null with which to compare the actual future earnings). We use the I/B/E/S detailed file to compute the outstanding analyst EPS consensus forecast (EPS forecast) before and after the split announcement. Details on how we compute the analyst consensus is provided in the appendix. We analyze changes in the EPS forecast for the next full fiscal year after the split announcement to ensure that the forecasts are made at least a year before the announcement of that fiscal year earnings. For example, if a split is announced in March 2005, we compare the EPS forecast for the full fiscal year 2006 (in this example assuming that the fiscal year ends on December 31). We use this forecast horizon as longer horizons are more likely to be concerned with fundamental long term firm performance as opposed to temporary changes in earnings expectations (e.g., resulting from accruals, seasonality, and one-off charges or revenues). We also focus on annual forecasts as opposed to quarterly forecasts for the same reason. We are able to construct pre- and post-split consensus estimates in I/B/E/S for 530 of the splits in our sample. As in the analysis of abnormal returns, we control for earnings and dividend change announcements made between the consensus estimates. To do this, we flag individual analyst estimates before and after the declaration date that are not comparable as result of an earnings or dividend change announcement taking place between them. We then retain all consensus 12

13 estimates that consist of at least three comparable individual forecasts before and after the declaration date. The final sample consists of 174 splits with EPS forecasts that are uncontaminated by earnings and dividend announcements (Table 4). In our analysis, we scale the change in EPS consensus by the firm s stock price prevailing at the end of the month prior to the split announcement. To analyze the earnings growth of that split their stock we compare the earnings growth of the in our sample to those of a matched sample of based on pre-split earnings growth. We match on past earnings growth in the four years preceding the split in order to create a sample of matched that experience a similar earnings growth path prior to the split announcement but do not split. The main matching criterion that we use is minimizing the sum of squared differences in annual earnings growth between the splitting firm and the potential matches (potential matches are in the same size quintile and industry). Details on the computation of earnings growth and the matching criteria are discussed in the appendix. Our final sample includes earnings data for 1150, 925 and 850 splitting, for operating income before depreciation, income before extraordinary items and net income, respectively. We find individual matches for 1021, 799 and 726 across the same earnings categories 4. 4 Results This section is organized as follows. We first report results on abnormal returns around split announcements. Next we present our findings related to splitting activity and prices, and finally we report results on analyst EPS forecast revisions and earnings growth. We conclude with a discussion of our main findings. 4.1 Abnormal returns In Table 5, we see that stock splits are associated with abnormal returns. When we do not exclude observations for which there are coinciding earnings or dividend change announcements, the mean abnormal return over 3 days is 1.76% [t-stat 15.06] and highly statistically significant. The mean abnormal return when we exclude observations that coincide with earnings or dividend change announcements is slightly lower, 1.46%, but still highly significant [t-stat 8.96]. 4 Note that the number of firm-year observations reported in Table 12 for each year is not identical to the number of firm observations mentioned above because are not required to have data available in Compustat across all years 13

14 Further, when analyze the returns by tick size regime we see that returns are significant across all regimes both when including and excluding observations with earnings and dividend changes. The average abnormal returns around stock splits increase across the regimes in the full sample (abnormal returns are lowest in the earliest 1/8 th regime and highest in the 1/100 th regime), however, this pattern changes when we exclude observations that coincide with earnings and dividend change announcements. Specifically, the abnormal returns decline significantly in the last regime when exclude observations that have coinciding earnings and dividend change announcements, suggesting that in the most recent regime bundle positive earnings and dividend information with split announcements to a greater extent. To further examine the characteristics of the abnormal returns we test whether the abnormal returns around stock splits are associated with proxies of how much public information there is about the firm. We use the following measures as proxies for information availability: the number of analysts following the firm, the firm s market capitalization, and the stock s monthly trading volume. The number of analysts is the number of estimates in the I/B/E/S summary file on the date closest to the split declaration date but within a maximum of 60 days of the split announcement. Market cap and trading volume figures are from the CRSP monthly file for the month prior to the split declaration date. In the regression, we take logs of all of the independent variables. We also exclude all observations with earnings and dividend change announcements in this analysis. The results are reported in Table 6. We see that the abnormal returns are negatively correlated with the number of analysts, the firm s market cap, and trading volume, which indicates that stock splits are associated with higher abnormal returns for for which there is less available information. While all of these explanatory variables are highly correlated (correlations not tabulated), when we include both the number of analysts and the firm s market cap or all three of the variables together in the regression (specifications 5 and 6 respectively), we see that the number of analysts captures almost all of this effect. The number of analysts increases from 6 to 20 when moving from the 25 th to 75 th percentile in our sample (see descriptive statistics in Table 2), and thus the coefficient on the (log of) number of analysts (-0.72) corresponds to a difference of -0.87% in abnormal returns between the stocks with relatively few versus relatively many analysts. 14

15 4.2 Splitting activity and prices To further investigate whether mechanisms for increased liquidity may be driving the abnormal returns, we examine the splitting activity and splitting prices around the tick size changes on the NYSE. The optimal effective tick size hypothesis suggests that splitting activity should increase after a change in tick size (ceteris paribus) as need to adjust their stock price to the new optimal price-to-tick ratio. The other theories do not predict changes in the splitting activity following a tick size regime change. Table 7 and Figure 1 reports that there indeed appears to be an increase in splitting activity in 1997 and 1998 after the first regime change (118 splits in 1997), but the splitting activity is the lowest in 2001 (31 splits) after the second regime change. Taken together, these results do not seem to support the fact that splitting activity increases following a regime change. Rather, although we do not explicitly test this, it merely appears that splitting activity to some extent tends to follow the general market development with high activity in the end of the 90 s and during the last stock market run-up, and low activity following the bust in 2001 and Our results here are quite similar to those found by Lipson and Mortal (2006). Table 8 (Panel A) further reports the pre-split price levels by tick size regime. Again, the relative tick size hypothesis predicts that pre-split prices should be lower following the tick size regime changes. However, this prediction is not confirmed in the data as the average prices at which split their stock has not monotonically decreased following the two decreases in tick size. Rather, prices appear largely to have stayed constant or slightly increased over time in our sample. Panel B reports a difference in means test of pre-split stock prices across the three regimes. We see that the mean pre-split price is the lowest in the 1/8 th regime ($65), the highest in the 1/16 th regime ($81), and $75 in the final 1/100 th regime. These differences are statistically significant, but it is uncertain whether we can read any economic significance into these differences 5. In sum, the lack of a trend in splitting prices and the lack of evidence of increased splitting activity following the tick size regime changes provide evidence against an effective tick size hypothesis as an explanation of the abnormal returns around stock splits. To further test whether the abnormal returns surrounding the stock split result from the arrival of new information as opposed to the achievement of an optimal trading range or a commission 5 We have not clustered the standard errors in this test as it is unclear over what period (if any) we could consider price levels to be independent. 15

16 induced sponsorship hypothesis we examine how the abnormal returns vary by pre-splitting price. The descriptive statistics in Table 7 show that there is significant variation in pre-split prices across our sample and that there is considerable variation in prices during all years. We test whether the pre-split price is associated with the abnormal returns around stock splits by including price as an explanatory variable in our information based regressions and by analyzing abnormal returns after splitting our sample into three tiers based on their pre-split price 6. Our results are reported in Table 9. First, we can see that price has no linear relationship with abnormal returns across our full sample (specifications (1) (3)). Once we control for other announcements (specifications (4) (6)) price has a small negative (linear) relationship that goes away when the other information proxies are included in the regression. However, when we split our sample into three tiers (Panel B) we see that abnormal returns are highest for stock with lowest pre-split prices levels, 2.22% vs. 1.52% and 1.54% in the medium and high tier respectively. This difference is significant at the 5% level (results not tabulated). We find similar results when we control for other announcements in Panel C, except that now the returns for high priced stocks are the lowest, and the difference in abnormal returns between the low and medium price groups is no longer significant. These results appear inconsistent with the argument that a stock split increases liquidity mainly by making stocks cheaper and thus resulting in abnormal returns, as it is the stocks that are already the cheapest that are associated with the highest abnormal returns around stock splits. That is, it is not clear under such a hypothesis why lower priced stocks would experience larger abnormal returns when they were less expensive to begin with. One potential explanation for this result, which we do not test explicitly, is that lower priced stocks may face a higher cost for providing a false signal arising from the concern that when a firm s stock price falls below a certain threshold it may not meet the exchange requirement or lose some of its institutional investor base. Lipson and Mortal (2006) also compare the predictions of the relative tick size, sponsorship, and trading range stories in response to the tick size regime changes on NYSE. They conclude that the evidence goes against the effective tick size hypothesis, since post-split price levels do not change and there is no temporary increase in splitting activity after regime changes. However, in 6 We define the pre-split price as the price available on the CRSP daily file two trading days prior to the split. Note also that since a majority of the splits in our sample are 2:1 splits this is equivalent to sorting on post split prices. 16

17 contrast to our paper, they focus mainly on the liquidity aspects of splits and its affect on clientele, and do not consider information-based theories. In summary, our results on abnormal returns along with the results related to the splitting activity across regimes appear to be inconsistent with the hypotheses that the abnormal returns surrounding split announcements is caused by an increase in liquidity achieved via an effective tick size or an optimal trading range. We cannot however rule out other mechanisms that may increase liquidity surrounding the stock split that is distinct from the theories tested in this paper, as for example an increase in the information environment of the firm following a stock split. 4.3 Analyst revisions Next we analyze whether analysts revise their earnings estimates around the stock split. While the results on abnormal returns above may also be explained by some mechanism that links lower post-split prices to increased liquidity, only an information story would predict that analysts revise their EPS forecasts around the split announcement 7. In Panel A of Table 10, we see that indeed the consensus estimate based on the outstanding mean EPS forecast increases following the declaration of a split announcement. Across all observations, the magnitude of the change in the consensus forecast scaled by price ( ΔEPS / P ) is 0.13 percent, and highly significant. The coefficient is somewhat higher, 0.15 percent (and still significant at the 1% level), when we compute the change in the EPS consensus after excluding observations for which there are any earnings or dividend change announcements between the pre- and post-split analyst estimates that may otherwise confound our results. The results using the consensus based on median EPS forecasts (so that a consensus estimate is not driven by outlier analysts) are very similar to that when using the mean estimate as the consensus; we observe an increase of 0.12% across all estimates and 0.13% when controlling for earnings or dividend change announcements. To provide some economic intuition for these figures, we can scale the coefficients by the mean forward P/E value for our full sample (the mean forward P/E is 18) in order to translate these units into approximately a percent change in forecasted earnings, since 7 Ikenberry et al. (2002) also look at analyst forecasts in relation to stock splits, but they focus on explaining the post split abnormal return drift via the potential under-reaction or downward bias of analyst forecasts for that split their stock relative to a matched sample. If analysts under-react, our measure of changes in the expectation of future earnings is biased downwards. 17

18 ( ΔE / P) * (average P/ E) Δ E/ E. By this logic, the coefficient above of 0.13 percent in ΔEPS / P corresponds to approximately a 2.3% increase in the average earnings forecast for the pooled sample, while the coefficient after controlling for earnings and dividend change announcement corresponds to a 2.7% increase in the earnings forecast. We can thus see that the revision in analyst forecasts is similar in magnitude to the level of abnormal returns around the split announcement (for comparison, the abnormal returns across the full sample was 1.76%). We also examine how the change in consensus EPS forecast following a split differs by tick size regime. Table 11 shows that for our full sample (Panel A) and for the sample that excludes observations with earnings and dividend change announcements (Panel B) the change in EPS forecasts is significant across all regimes and but is significantly higher in the last regime compared to the first two regimes. One potential alternative explanation for this result is that analysts are revising their earnings estimates due to the past earnings growth experienced by splitting as opposed to the split announcement per se. To test this alternative hypothesis we re-run the same test on a matched set of (the matching is as discussed in section 3) to analyze whether analysts significantly revise their earnings estimates over the same time period also for that have not split their stocks but have had similar past earnings increases. Since our matched experience earnings growth that is similar to that of the that split their stock by construction, if analysts are just sluggish in revising their estimates, we should be able to observe some level of forecast revision for the matched set of as well. We report our results in panel B of Table 10. While the mean (median) revision is positive for the matched set of, it is smaller than for the splitting and not statistically significant. The average analyst revision of EPS/Price is 0.10% for the matched sample compared to 0.13% for the full sample of splitting. The change is even smaller for the matched sample that corresponds to the subset of splitting where we control for earnings and dividend change announcements; 0.06% for the matched compared to 0.15% for the splitting. However, since the standard errors for the analyst revision in the matched sample are quite large, it is impossible to statistically reject the hypothesis that the means are the same. So while this evidence is encouraging it is only anecdotal. Additionally, it is worth noting that many analysts do not update or even reiterate (hold constant) an earnings estimate for many of the matched, so that we have fewer observations of analyst estimates for the matched set of than for the splitting before and after the split date. Indeed we 18

19 might expect there to be fewer such observations for the matched, as the split date is merely a random date for the matches. Nevertheless, we compute the change for the matches only based on for which estimates were updated or reiterated, i.e. we do not assign a value of zero to estimates that were not updated (i.e. we exclude stale estimates). This also implies that the consensus change that we calculate for the matched is conditional on there being a sufficient number of analyst updates around this date. In total, we find 243 pre-and post split date consensus estimates for the full matched sample and 155 for the matched sample that corresponds to the subset of splitting where we control for earnings and dividend change announcements. In sum, these results help alleviate the concern that the increase observed in the mean and median analyst forecast is purely a reaction to past earnings growth, as the matched sample of with similar earnings growth see significantly smaller updates than that of the splitting, even conditional on there being an update for the matched. 4.4 Analyst revisions and earnings growth Finally we analyze what type of earnings-related information might be conveyed via the stock split that analysts can be reacting to. While splitting on average have experienced high levels of past earnings growth, some of the previous papers in the literature (Asquith et al. (1989), Ikenberry et al. (2002)) have suggested and found evidence consistent with the idea that that split their stock experience a decreased level of reversion in their earnings growth. If that is indeed the case, a stock split may cause analysts to infer that the past earnings growth experienced by that split their stock is not transitory in nature, resulting in an upwards revision of analysts expectations about future earnings. This logic is consistent with several past findings in the literature. First, it provides one explanation for why GMT conclude that: it appears that some of the informational impact of splits (and stock dividends) is not dividend related. While there is a strong relationship between dividend payments and earnings (all earnings eventually have to be paid out through either dividends or repurchases), it is possible that the information released via a split announcement concerns earnings growth, which is related to future regular dividends, but that the reaction to the split is not driven solely by a change in expectations of regular dividend payments. More recently, Skinner (2008) documents that the role of dividends in payout has decreased significantly over time which makes it more likely that the information content in splits, if it 19