INTRA-INDUSTRY REACTIONS TO STOCK SPLIT ANNOUNCEMENTS. Abstract. I. Introduction

Transcription

1 The Journal of Financial Research Vol. XXV, No. 1 Pages Spring 2002 INTRA-INDUSTRY REACTIONS TO STOCK SPLIT ANNOUNCEMENTS Oranee Tawatnuntachai Penn State Harrisburg Ranjan D Mello Wayne State University Abstract We examine whether favorable information conveyed by stock split announcements transfers to nonsplitting firms within the same industry. On average, nonsplitting firms shareholders experience positive and significant abnormal returns at the stock split announcements of their industry counterparts. In addition, industrywide and firm-specific characteristics are important determinants in explaining nonsplitting firms stock returns. These firms earnings increase significantly, and the earnings changes are positively related to the stock price reactions. Finally, we find no evidence that investors revise the value of nonsplitting firms because they anticipate a decline in earnings volatility. JEL Classifications: G0, G3 I. Introduction Several studies report an association between the returns of firms releasing information and those of nonannouncing firms within the same industry. This relation, known as intra-industry information transfer, is documented in different contexts such as bankruptcy announcements (Lang and Stulz 1992), bond rating adjustments (Akhigbe, Madura, and Whyte 1997), dividend change announcements (Firth 1996; Laux, Starks, and Yoon 1998), and securities offerings (Szewczyk 1992). In this article we extend prior studies by examining whether stock split announcements affect equity value of nonsplitting firms in the same industry. All of the events examined in the existing literature have direct cash-flow implications for the announcing firms, and possibly for nonannouncers. Therefore, We have benefited from the comments of Thomas George, Jeffrey Harris, Sudha Krishnaswami, Neal Maroney, Tarun Mukherjee, Oscar Varela, Gerald Whitney, two anonymous referees, the editor, and seminar participants at Brock University, Chaoyang University of Technology, National Central University, Penn State Harrisburg, University of New Orleans, University of Northern British Columbia, University of Saskatchewan, Yuan Ze University, and the 1999 Financial Association meeting. All errors are our responsibility. 39

2 40 The Journal of Financial Research they convey information about past and current cash flow that was not already public, as well as information about the firms future prospects. However, stock splits are cosmetic accounting changes with no direct effect on either the announcer s future cash flow or that of nonannouncers. Thus, the stock split announcement-period reactions for both splitting and nonspitting firms only reflect the market s inference about the future prospects of these firms conveyed by the event. By analyzing the stock price reactions of nonsplitting firms, we provide an ideal setting to test hypotheses relating to the transfer of information about firm value within an industry. The prior literature documents that the market reacts favorably to the announcement of stock splits and presents two major hypotheses to explain the results (Grinblatt, Masulis, and Titman 1984; Desai and Jain 1997; Conroy and Harris 1999). The information content hypothesis suggests investors react positively to stock splits because these events reveal information about increases in future cash dividends or earnings, or both. The trading range hypothesis posits firms that experience a run-up in stock prices split their stocks to improve trading liquidity by moving the firms share price into an optimum trading range. 1 The information content and trading range hypotheses are not mutually exclusive. Given that managers have superior information about their firm values and that it is costly to trade shares below the optimum trading range, managers will split their stocks to move the share price, conditional on the favorable future information being revealed, into the trading range. Ikenberry, Rankine, and Stice (1996) refer to this explanation as the self-selection hypothesis. The findings of most prior studies that examine intra-industry information transfers are consistent with a net contagion effect at the announcement of a corporate event; that is, the stock price reaction for other firms in the industry moves in the same direction as that of the announcing firm. However, Lang and Stulz (1992) and Laux, Starks, and Yoon (1998) argue that in highly concentrated industries where competition for market share is high, certain events may trigger a change in the competitive balance within the industry. Thus, announcements of these events will result in stock price movement for other firms in the industry in a direction opposite that of the announcing firm, that is, a competitive effect. We test whether split announcements have a net contagion or competitive effect on other firms in the same industry. Furthermore, we analyze the stock price reactions of nonsplitting firms to examine the effect of industry and firm-specific factors on intra-industry transfers of information. Finally, we investigate the sources of the change in nonsplitting firms equity value by examining the level and volatility of their earnings performance. 1 Lakonishok and Lev (1987) and Ikenberry, Rankine, and Stice (1996) suggest the optimum trading range is the average price of firms in the market and the industry, whereas Conroy and Harris (1999) suggest the target price is the price level after the preceding split.

3 Intra-Industry Reactions 41 We use a sample of 327 clean stock splits announced between 1986 and 1995 and find that shareholders of nonsplitting firms experience a significant abnormal increase in stock prices during the announcement period. This significant reaction suggests that there is an intra-industry reaction to stock split announcements and that the information conveyed by the announcement has a net contagion effect on the equity value of nonsplitting firms. Cross-sectional regression results indicate intra-industry information transfers are influenced by the degree of concentration within the industry; firms in industries with a low level of concentration experience significantly higher returns than those in industries with a high level of concentration. In addition, we find that firm-specific characteristics explain some of the differential industrywide reaction to stock split announcements. Consistent with Lang and Stulz (1992), we find that when the degree of homogeneity between splitting and nonsplitting firms, as measured by their earnings correlation, is high, nonsplitting firms experience significant net contagion effects. Furthermore, nonsplitting firms with a high degree of asymmetric information captured by return variance have a more positive stock price reaction than firms with a low asymmetric information level. Finally, we document that the degree of mispricing, measured by book-to-market ratio, matters; firms that are more likely to be underpriced experience significantly greater announcement-period returns than those that are overpriced. We also find that subsequent short- and long-run earnings performance for nonsplitting firms improves significantly and that the change in earnings is positively related to the announcement-period stock returns. However, we find little evidence of a decline in earnings volatility after the split announcement. These findings suggest stock split announcements reveal industrywide information about future earnings increases but not about a decline in risk. Our results are generally consistent with stock splits revealing favorable information and support the information content and self-selection hypotheses proposed by most prior studies. II. Intra-Industry Reactions to Stock Split Announcements Contagion and Competitive Effects Asquith, Healy, and Palepu (1989), McNichols and Dravid (1990), and Desai and Jain (1997) conclude that stock splits reveal favorable future information. They find that stock splits are followed by abnormal increases in dividends or earnings, or both. A stock split results in an increase in stock price for the splitting firm, but it might also reveal information about the industry in general. The direction of the industry stock price movement depends on whether the information revealed has a net contagion or competitive effect on nonsplitting firms.

4 42 The Journal of Financial Research Szewczyk (1992) argues that because of homogeneity of firms within an industry, information released by a firm causes the market to revise the value of both announcing and nonannouncing firms in the same direction. Asquith, Healy, and Palepu (1989) document that firms that announce stock splits are in industries that experience abnormally high earnings growth during the announcement year. They suggest that the similarity of firms within the industry results in the unusual earnings increase across all firms in the industry. The findings of these studies imply that the positive information revealed by split announcements results in a significant increase in share prices for nonsplitting firms. This positive reaction for nonsplitting firms is called the contagion effect. However, stock splits may produce a negative effect on nonsplitting firms in the industry by conveying unfavorable information about these firms. This effect is more pronounced in industries with imperfect competition where the announcement of an event reveals comparative information about other firms in the industry. For example, the performance of nonsplitting firms could be perceived as poor relative to the superior performance of splitting firms. At the extreme, wealth could be redistributed from nonsplitting firms to the splitting firm. Hence, a positive split signal by a firm may result in a decline in stock prices for other firms in the industry. This negative reaction is called the competitive effect. The contagion and competitive effects are not mutually exclusive, and thus the observed stock price reaction is the sum of these two effects. A significant change in nonsplitting firms stock prices indicates that stock split announcements are not only firm-specific events but affect the industry as well. On the other hand, an insignificant stock price reaction is consistent with one of the following two explanations. First, stock split announcements may reveal only firm-specific information and thus do not have an industrywide effect. Second, stock splits have an intra-industry effect, but the positive (contagion) effects for some firms offset the negative (competitive) effects for other firms in the industry. Industry Characteristics Influencing Intra-Industry Reactions Lang and Stulz (1992) argue that in highly concentrated industries where the competition among firms for market share is high, the competitive effect is the dominant effect. Thus, stock split announcements by firms in highly concentrated industries are more likely to reveal unfavorable information about their competitors, causing a shift in the competitive balance within the industry. Therefore, relative to nonsplitting firms in industries where competition is low, firms in highly competitive industries are expected to be more negatively influenced by the stock split announcements of their industry counterparts. We follow the Antitrust Guidelines of the Justice Department and use the Herfindahl-Hirschman Index (HHI) as a measure of industry concentration level (see Parkin 1996, pp ). HHI is calculated by summing squared market share

5 Intra-Industry Reactions 43 over the largest fifty firms in the four-digit Standard Industrial Classification (SIC) code industry. 2 We define market share as a firm s annual sales at the fiscal year-end preceding the stock split announcement as a percentage of industry sales for that year. A low (high) HHI indicates a low (high) level of concentration and, hence, a low (high) level of competition among firms. Given the findings documented in prior studies, we anticipate a negative relation between HHI and announcementperiod returns of nonsplitting firms. Firm-Specific Characteristics Influencing Intra-Industry Reactions Degree of Similarity. Lang and Stulz (1992) and Firth (1996) find that abnormal returns of nonannouncing firms whose characteristics are closely related to those of announcing firms are greater than the abnormal returns of nonannouncing firms whose characteristics are dissimilar. They argue that this result is consistent with greater intra-industry information transfer for firms with similar characteristics because they are affected by common factors. Their results imply that stock split announcements have a greater effect on nonsplitting firms with a high degree of similarity to splitting firms than on firms with a low degree of similarity. Because Firth (1996) suggests that firms that are similar to each other have a high correlation of earnings, we use earnings correlation (CORR) as a measure of the degree of similarity between firms. We use annual earnings before interest and taxes to eliminate the effects of capital structure and taxes on earnings. The correlation coefficient between splitting and nonsplitting firms is computed over the ten-year period preceding the stock split announcement. We require a minimum of four years of pairwise data to estimate CORR. A nonsplitting firm is considered highly similar (highly dissimilar) to a splitting firm if its earnings correlation is high (low). We anticipate a positive relation between CORR and the abnormal returns of nonsplitting firms. Level of Asymmetric Information for Nonsplitting Firms. Grinblatt, Masulis, and Titman (1984) and Ikenberry, Rankine, and Stice (1996) find significantly higher stock split announcement-period returns for firms classified as having high levels of information asymmetry. They conclude that stock splits reveal more information for firms with little or no news available in the market. If stock splits reveal industrywide information, their findings imply that nonsplitting firms that have a high level of asymmetric information will experience a greater effect from the announcement than will other firms in the industry. Dierkens (1991) and Krishnaswami and Subramaniam (1999) suggest that volatility in residual stock returns captures the degree of information asymmetry. 2 Similar to Lang and Stulz (1992) and Laux, Starks, and Yoon (1998), we also calculate HHI by including all firms in the industry and obtain similar results.

6 44 The Journal of Financial Research Therefore, we use return variance (RVAR), defined as the dispersion of Center for Research in Security Prices (CRSP) value-weighted and market-adjusted returns in the year preceding stock split announcements, as a proxy for the level of asymmetric information. 3 To obtain a reasonable estimate of RVAR, we require return data for a minimum of thirty trading days. Firms with high (low) RVAR are assumed to have high (low) levels of asymmetric information. The hypothesis predicts a positive relation between RVAR and abnormal returns for nonsplitting firms. Degree of Mispricing for Nonsplitting Firms. Fama and French (1992) find that stocks with high book value to market value (i.e., value stocks) outperform firms with low book value to market value (i.e., glamour stocks). Lakonishok, Shleifer, and Vishny (1994) find evidence consistent with Fama and French. Furthermore, they show that earnings growth for value stocks is significantly higher than that for glamour stocks and argue that high (low) book-to-market (B/M) ratio stocks are underpriced (overpriced). If stock splits reveal positive information about the industry, we expect a positive (negative) relation between the returns of nonsplitting firms and the degree of undervaluation (overvaluation) of these firms. Consistent with Lakonishok, Shleifer, and Vishny (1994) and Ikenberry, Rankine, and Stice (1996), we use the B/M ratio to proxy for the degree of mispricing. B/M is defined as the book value of assets divided by the book value of assets plus the difference between market value and book value of equity as of the year-end before the split announcement. 4 Firms are more likely to be undervalued (overvalued) if their B/M ratio is high (low). Therefore, the hypothesis predicts a positive relation between the nonsplitting firms returns and the B/M ratio. III. Data Selection and Matching Process Data Selection The sample used in this study is obtained from the CRSP file and crossreferenced with the Wall Street Journal Index. There are 4,497 stock distributions (i.e., stock splits and stock dividends) announced from 1986 to Following Grinblatt, Masulis, and Titman (1984), we delete stock distributions with a split factor (defined as the number of additional shares per existing share) less than or equal to 0.25 (600 observations). Similar to Lang and Stulz (1992), we assign firms to an industry group based on the primary four-digit SIC code obtained from Compustat. We exclude firms whose (a) four-digit SIC codes are not available 3 We also use variance of the three-day, market-adjusted returns around the quarterly earnings announcement calculated over twenty quarters before the stock split announcement as another proxy for information asymmetry, and we obtain similar results. 4 We obtain similar results when B/M is defined as book value to market value of equity.

7 Intra-Industry Reactions 45 (452 observations); (b) SIC codes begin with 49 and 6 (992 observations), representing public utilities and financial institutions, because these firms are regulated and announcements by these firms convey little incremental information (Szewczyk 1992); and (c) shares are not traded on the American Stock Exchange (AMEX), Nasdaq, or New York Stock Exchange (NYSE) (65 observations). Foster (1980) argues that the amount of information is inversely related to the sequence of information releases. Consistent with Foster, Pilotte and Manuel (1996) find that the market reaction to the first stock split is greater than that to subsequent announcements. Therefore, to focus on stock splits that are expected to convey the greatest amount of industrywide information, we exclude splits conducted less than one year from the previous stock split announced by a firm in the same industry (1,607 observations). 5 This is consistent with Lang and Stulz (1992) and Firth (1996), who also eliminate observations having low industrywide information in their studies of cross-firm information transfer. Similar to Firth, we eliminate 436 observations with contemporaneous announcements over the announcement period. Finally, we delete firms with no return information during the announcement-period window and firms with no industry matches. The final sample consists of 327 stock splits announced during 1986 to Matching Process The nonsplitting firms include all other firms listed on Compustat that satisfy the following criteria: 1. They have the same four-digit SIC code as splitting firms. 2. Their shares are traded on AMEX, Nasdaq, or NYSE. 3. They have five-day announcement-period returns available on CRSP. 4. They do not announce stock splits during an eleven-day period centered on the splitting firm announcement day. The last criterion ensures that stock splits announced by matching firms during the event window are not a source of the significant return results for the nonsplitting firms. There are 3,684 nonsplitting firms in 199 four-digit SIC codes. The average (median) number of nonsplitting firms for each splitting firm is (7), with a minimum of 1 and a maximum of 66. Three observations have more than 50 matching firms. Because the announcement period for a given split is the same for all nonsplitting firms in an industry, there may be a significant correlation in stock returns among industry firms that could bias the test statistics. We control for this 5 We observe that the announcement-period returns for nonsplitting firms at the announcement of the second split within an industry are lower relative to the first split.

8 46 The Journal of Financial Research clustering problem by forming an equally weighted portfolio of all nonsplitting firms in the same industry for each split. The portfolio approach for evaluating the nonsplitting firm sample eliminates the random noise of individual stocks, resulting in a more powerful test of the predictions detailed in section II. However, we also replicate all of the tests at the firm level and obtain similar results. IV. Empirical Results Abnormal Returns of Nonsplitting Firms Similar to Ikenberry, Rankine, and Stice (1996), we use five-day cumulative abnormal returns calculated from two days before to two days after the announcement day to measure the announcement-period returns for both splitting and nonsplitting firms. We define abnormal returns as returns in excess of the CRSP value-weighted market returns. The announcement-period abnormal return results for both splitting and nonsplitting firms are presented in Table 1. For splitting firms, we document a significant average cumulative abnormal return of 3.82%. This result is similar to the finding of Ikenberry, Rankine, and Stice and confirms the conclusion of prior studies that the market views the announcements of stock splits as good news. For nonsplitting firms, we find that the mean cumulative abnormal return is 0.34%, significant at the 10% level. As stated earlier, low announcement-period TABLE 1. Abnormal Returns for Splitting and Nonsplitting Firms Around the Announcements of Stock Splits. Splitting Firms Nonsplitting Firms Mean Mean Period N (% Positive) t-statistic N (% Positive) t-statistic ( 20, 3) (72%) (60%) ( 2, +2) (72%) (52%) (+3, +20) (60%) (51%) Note: Mean abnormal returns are reported for 327 firms that announced stock splits and for portfolios of 3,684 nonsplitting firms. Abnormal returns are value-weighted, market-adjusted returns. The nonsplitting firms are firms listed on Compustat that have the same four-digit SIC code as splitting firms, whose shares are traded on major exchanges, and that do not announce stock splits during an eleven-day period centered around the stock split announcement day (day 0). The portfolios of nonsplitting firms are equally weighted, grouped for each split announcement. (% Positive) represents the percentage of portfolios that experience positive returns. Significant at the 1% level. Significant at the 10% level.

9 Intra-Industry Reactions 47 returns for nonsplitting firms might indicate either that stock splits do not reveal information about other firms in the industry or that the positive (contagion) effect of some firms are offset by the negative (competitive) effect of other firms. We distinguish between these two explanations by examining the mean absolute returns of nonsplitting firms at the announcement period (results not reported). We find that the mean absolute return for these firms is 5.20% and significant (at the 1% level). This suggests an intra-industry reaction exists at the announcement of stock splits. Furthermore, the mean absolute return is significantly higher than the average return. This implies the announcement-period returns for nonsplitting firms are affected by both the positive contagion effect and the negative competitive effect. However, the contagion effect dominates the competitive effect on average, resulting in positive announcement-period abnormal returns. Thus, the favorable information conveyed by stock split announcements has a net positive effect on the equity value of other firms in the industry. Overall, we conclude that the findings of previous studies that the market revises the announcing and nonannouncing firms values in the same direction in the context of different corporate events also extend to stock splits. We further examine abnormal returns of splitting and nonsplitting firms during the month before and after the split announcement. The results are also presented in Table 1. Consistent with the findings of previous studies that investors anticipate the split announcement, we find a significant increase of 7.08% in splitting firms stock prices in the pre-announcement period. For nonsplitting firms, the increase in stock price is 1.29%. This result suggests investors revise their beliefs of firm value for both splitting and nonsplitting firms in the same direction and provides further support for the intra-industry net contagion hypothesis. 6 After the split announcement, splitting firms continue to experience a significant abnormal run-up in prices. However, the abnormal returns for nonsplitting firms in the post-split period are statistically insignificant, consistent with Firth (1996). Industry Reactions and Industrywide and Firm-Specific Characteristics In this section, we estimate a multivariate cross-sectional ordinary least squares (OLS) model to test the significance of the industrywide and firm-specific factors detailed in section II in explaining the variation in nonsplitting firms 6 An analysis of the ratio of abnormal returns for nonsplitting firms to splitting firms indicates a decline from the pre-announcement period (0.18) to the announcement period (0.09). When we examine the ratio of mean absolute returns, we find the ratio for the two periods to be similar. This suggests that the decline in the ratio of abnormal returns at the announcement period occurs because the contagion effect and competitive effect offset each other and lower the announcement-period returns for nonsplitting firms. This provides additional support for the information transfer hypothesis.

10 48 The Journal of Financial Research announcement-period returns. We specify the model as follows: where AR = α + β 1 ln(hhi) + β 2 CORR + β 3 RVAR + β 4 B/M + β 5 ARS + β 6 SPF + ε, (1) AR = five-day, value-weighted abnormal returns of nonsplitting firms; 7 ln(hhi) = natural log of HHI; CORR = earnings correlation; RVAR = return variance; B/M = book-to-market ratio; ARS = five-day, value-weighted abnormal returns of splitting firms; and SPF = split factor. In addition to industrywide and firm-specific factors, we include announcement-period abnormal returns of splitting firms (ARS) because Firth (1996) finds a positive and significant association between returns of announcing and nonannouncing firms. Similarly, McNichols and Dravid (1990) document a positive relation between the split factor (SPF) and the level of information revealed by splitting firms; therefore, we add SPF as an additional independent variable. Akhigbe, Madura, and Whyte (1997) suggest that announcements of large firms that are viewed as dominant have a stronger effect on nonannouncing firms than those of small firms. Thus, we estimate an alternative model where ARS is replaced by an interactive variable between ARS and dominance level of splitting firms (DOM). DOM is a dummy variable that equals 1 if splitting firms market value of common equity is greater than the industry median and 0 otherwise. The coefficient of the industry concentration measure (ln(hhi)) is expected to be negative, whereas the coefficients of the degree of similarity (CORR), the asymmetric information level (RVAR), and the degree of underpricing (B/M) are expected to be positive. Finally, the coefficients for ARS, SPF, and the interaction term are expected to be positive. The regression estimates are presented in Table 2. 8 We find that the coefficients of ln(hhi) are negative and weakly significant in models (1) and (2). This result indicates that for nonsplitting firms in industries with a low degree of competition, the benefit from a stock split announcement is greater than that for nonsplitting firms in industries with a high degree of competition. This is consistent with the finding of Lang and Stulz (1992). 7 The regression results are robust to the abnormal returns estimated using the market model. 8 The sample size declines from 327 to 281 observations because insufficient data are available to estimate RVAR (40 observations) and CORR (21 observations).

11 Intra-Industry Reactions 49 TABLE 2. Multivariate Regression Results Between Cumulative Abnormal Returns of Nonsplitting Firms Portfolios and Industry and Firm-Specific Variables. Model Model Model Model Model Model Variables (1) (2) (3) (4) (5) (6) Intercept (0.714) (0.782) (2.629) ( 0.079) ( 0.674) ( ) ln(hhi) ( 1.753) ( 1.864) ( 2.612) CORR (2.612) (2.802) (1.696) RVAR (2.620) (2.456) (2.376) B/M (2.067) (2.146) (2.176) ARS (0.225) DOM ARS (1.766) (1.558) (1.677) (1.351) (1.510) SPF ( 0.191) ( 0.407) ( 0.141) ( 0.177) ( 0.466) ( 0.298) Adjusted R p-value of F-statistic N Note: The model is defined as follows: where AR = α + β 1 ln(hhi) + β 2 CORR + β 3 RVAR + β 4 B/M + β 5 ARS + β 6 (DOM ARS) + β 7 SPF + ε AR = five-day, value-weighted, market-adjusted returns of nonsplitting firms; ln(hhi) = natural log of the Herfindahl-Hirschman Index, which is the sum of squared market share over the largest fifty firms in the same four-digit SIC code; CORR = earnings correlation measured by the correlation of earnings before interest and taxes for a ten-year period preceding the announcement; RVAR = return variance defined as variance of CRSP daily market-adjusted returns in the year preceding the stock split announcement day; B/M = book-to-market ratio measured as the ratio of the book value of assets to (book value of assets book value of equity + market value of equity); ARS = five-day, value-weighted, market-adjusted returns of splitting firms; DOM = a dummy variable that equals 1 if splitting firm s market value of equity is greater than the industry median and 0 otherwise; and SPF = split factor obtained from CRSP and defined as the number of additional shares per existing share. All values except returns and split factor are obtained from Compustat and are numbers at the fiscal year-end preceding the stock split announcement. The t-statistics are in parentheses. Significant at the 1% level. Significant at the 5% level. Significant at the 10% level.

12 50 The Journal of Financial Research After controlling for industry concentration level, we find that the coefficients for CORR are positive and significant in both models. This implies that stock split announcements reveal more information for nonsplitting firms with a high degree of similarity to the splitting firm than for those that are dissimilar. The estimated coefficients of RVAR are also positive and significant (t = in model (1) and t = in model (2)). Thus, shareholders of nonsplitting firms with a high level of information asymmetry gain more than shareholders of firms that have low asymmetric information, because the split announcement reveals more information about these firms. Consistent with the predictions detailed earlier, we observe positive and significant coefficients for B/M. This result, coupled with the finding that stock splits reveal a significant amount of intra-industry information, supports the hypothesis that nonsplitting firms that are likely to be undervalued (high B/M) experience a greater increase in stock prices at the stock split announcement than firms that are likely to be overvalued. In general, we conclude that firm-specific factors are important in explaining the cross-sectional variations of intra-industry information transfer in the context of stock split announcements. Furthermore, we find that the estimated coefficient of ARS is positive, as predicted, but not statistically significant. However, in model (2), we find the coefficient of the interaction term between ARS and DOM to be positive and marginally significant. This implies that stock split announcements by large firms have a greater effect on intra-industry reactions than announcements by small firms, and it is consistent with the findings of Akhigbe, Madura, and Whyte (1997). Finally, we find SPF to be insignificant, suggesting that, after controlling for other factors, the split factor is not important in explaining announcement-period returns for nonsplitting firms. In Table 2 we also present results for individual subsets (models (3) (6)) to confirm that the significance results in models (1) and (2) are not caused by multicollinearity. 9 We regress nonsplitting firms announcement-period returns on industrywide and firm-specific factors individually, controlling for the dominance level and split factor. The coefficients of ln(hhi), CORR, RVAR, and B/M are of the predicted sign and continue to be statistically significant. We further examine whether the industrywide and firm-specific factors are significant determinants of the pre-announcement-period returns. We estimate regression models similar to those described earlier but substitute announcementperiod returns with pre-announcement-period returns (results not reported). We find a negative and significant coefficient for ln(hhi) and a positive and significant coefficient for RVAR. For CORR and B/M, we find the coefficients to be positive but statistically weak. Thus, we conclude that the factors that affect 9 We find that the condition indexes are and for models (1) and (2), respectively. To confirm that multicollinearity does not distort our conclusion, we also test several other models and find similar results. We thank the editor for suggesting this point.

13 Intra-Industry Reactions 51 announcement-period returns also play an important role in explaining preannouncement-period returns for nonsplitting firms. Earnings Performance and Change in Earnings Volatility After Stock Split Announcements Change in Earnings. Asquith, Healy, and Palepu (1989) find that split announcements convey favorable information about future earnings changes. McNichols and Dravid (1990) find a positive relation between announcementperiod returns and subsequent earnings increases for splitting firms. These findings, coupled with the evidence presented earlier that stock splits reveal favorable industrywide information, predict an increase in earnings performance for nonsplitting firms after the announcement. Furthermore, if investors react to nonsplitting firms anticipating an increase in future earnings, the announcement-period abnormal returns of nonsplitting firms should be positively associated with the firms subsequent earnings changes. We test this hypothesis by examining standardized raw changes in earnings ( E) in both the short run (S) and the long run (L). Similar to Healey and Palepu (1988), we define these variables as: E S = E +1 E 1 P 1 (2) E L = E +3 E 1 P 1, (3) where E is earnings measure and P is closing stock price. We use two earnings measures: earnings per share (EPS) and earnings before interest and taxes per share (EBITPS). The subscripts 1, +1, and +3 represent the year-end before, the year-end after, and the three year-ends after the split announcement, respectively. In addition, we compute standardized abnormal earnings change ( AE), defined as the difference in earnings change between post- and pre-split announcement periods. Short- and long-run changes in abnormal earnings are expressed as: AE S = (E +1 E 1 ) (E 1 E 2 ) P 1 (4) AE L = (E +3 E 1 ) 3 (E 1 E 2 ) P 1 (5) An implicit assumption underlying the use of the abnormal earnings measures is that earnings change follows a random walk; hence, the best prediction of future earnings change is the past earnings change. Because we define the long run as a three-year period after the split announcement, the pre-announcement prediction

14 52 The Journal of Financial Research of long-run earnings change is three times the most recent annual earnings change (E 1 E 2 ). 10 As a robustness check, we also calculate E and AE using net income and earnings before interest and taxes standardized by total assets and obtain similar results. Given that stock splits are associated with earnings increases, we eliminate 443 firms that split the stocks within three years after the announcement to avoid any contamination of the nonsplitting firms earnings results. 11 Similar to the abnormal return analysis, we group nonsplitting firms into equally weighted portfolios for each split announcement. The earnings results are presented in Table 3. Nonsplitting firms experience a significant increase in earnings after stock split announcements. The mean short- (long-) run increase in raw EPS is 3.15% (6.14%) and highly significant. Similarly, raw EBITPS increases 4.10% (6.31%) in the short (long) run. An examination of abnormal earnings changes also indicates a significant increase in short- and long-run earnings levels. These findings support the hypothesis that stock split announcements reveal favorable information about industrywide earning changes for both the short and long run. Table 4 presents regression results relating investors revision of nonsplitting firms value to improvement in these firms future earnings performance. The dependent variable is the nonsplitting firms five-day abnormal returns, and the independent variables are the short- and long-run changes in earnings per share (EPS). We also include the split factor (SPF) and the interaction between splitting firm dominance (DOM) and its abnormal returns (ARS) as additional independent variables. We predict the coefficients for all these variables to be positive. Consistent with the predictions, we find that the estimated coefficients of both short- and long-run EPS change are positive (0.028) and significant at the 10% level. The coefficient of the short-run abnormal EPS change is (significant at the 5% level), whereas that of long-run abnormal EPS change is insignificant. 12 In general, we conclude there is a positive relation between returns of nonsplitting firms and earnings increases after stock split announcements. This suggests that investors of nonsplitting firms react positively to the announcement because they anticipate an improvement in the firms operating performance. We also investigate whether investors of nonsplitting firms in industries with a low degree of competition or with a high degree of similarity, information 10 We also control for outliers by trimming 1% on either tail of the earnings measure. We obtain similar results. 11 The results for the full sample are similar to those reported here. 12 We also regress abnormal returns of the independent variables at the firm level and find that the coefficient for the long-run abnormal EPS change is significant at the 5% level. Similar results are obtained when alternative earnings measures are used.

15 Intra-Industry Reactions 53 TABLE 3. Earnings Performance of Nonsplitting Firms Portfolios After Stock Split Announcements. Earnings Measures E AE Short run ( 10 2 ) EPS (4.87) (3.33) [63%, 318] [56%, 315] EBITPS (5.03) (4.54) [62%, 314] [58%, 313] Long run ( 10 2 ) EPS (6.02) (1.85) [64%, 265] [53%, 263] EBITPS (5.41) (2.92) [65%, 264] [58%, 263] Note: Earnings performance is measured by changes in earnings ( E) and abnormal earnings ( AE) for both short-run (S) and long run (L) and is defined as follows: E S = E +1 E 1 P 1 E L = E +3 E 1 P 1 AE S = (E +1 E 1 ) (E 1 E 2 ) P 1 AE L = (E +3 E 1 ) 3 (E 1 E 2 ) P 1 where E is earnings per share (EPS) and earnings before interest and taxes per share (EBITPS), and P 1 is closing stock price at the year-end before the split announcement. Subscripts 1, +1, and +3 represent the year-end before, the year-end of, and the three year-ends after the split announcement, respectively. Nonsplitting firms are grouped into equally weighted portfolios for each split announcement. The numbers in parentheses are t-statistics. The numbers in brackets are the percentages of portfolios that experience positive earnings changes and total number of portfolios, respectively. Significant at the 1% level. Significant at the 10% level. asymmetry, and underpricing anticipate superior subsequent earnings performance relative to other firms. We test this hypothesis by including an interaction term between earnings measures and dummy variables for these firms in the models (results not reported). Dummy variables equal 1 for nonsplitting firms in industries with a low degree of competition (HHI less than or equal to 1,800) or for firms with a high degree of similarity, information asymmetry and underpricing (earnings correlation, return variance, and book-to-market ratio greater than the median firm in the same industry), and 0 otherwise. 13 The coefficients of these interactive 13 The Justice Department classifies an industry with HHI greater than 1,800 as a highly concentrated industry (Parkin 1996, p. 299).

16 54 The Journal of Financial Research TABLE 4. Regression Results Between Cumulative Abnormal Returns of Nonsplitting Firms Portfolios and Post-Split Earnings Performance. Model Model Model Model Variables (1) (2) (3) (4) Intercept (1.248) (0.595) (0.437) (0.548) EPS S (1.650) EPS L AEPS S (1.817) (2.135) AEPS L (1.300) DOM ARS (1.833) (1.773) (1.540) (1.279) SPF ( 1.066) ( 0.816) ( 0.596) ( 0.733) Adjusted R p-value for F-statistic N Note: The model is defined as follows: AR = α + β 1 EPS + β 2 (DOM ARS) + β 3 SPF + ε where AR = five day, value-weighted, market-adjusted returns of nonsplitting firms; EPS = short- and long-run change in earnings per share ( EPS S and EPS L ) and short- and long-run change in abnormal earnings per share ( AEPS S and AEPS L ) for models (1) (4), respectively; DOM = a dummy variable that equals 1 if splitting firm s market value is greater than the industry median and 0 otherwise; ARS = five day, value-weighted, market-adjusted returns of splitting firms; and SPF = split factor. Nonsplitting firms are grouped into equally weighted portfolios for each split announcement. The t-statistics are in parentheses. Significant at the 5% level. Significant at the 10% level. variables are significant. Therefore, we conclude there is a stronger relation between announcement-period returns and future earnings changes for these firms relative to other nonsplitting firms. Nevertheless, the F-statistics of these models are relatively low, suggesting the relation between announcement-period returns and subsequent earnings increase is, at best, weak. Decline in Earnings Volatility. We test whether stock split announcements also convey information about earnings volatility, that is, whether the increase in the value of nonsplitting firms can also be explained by a decline in earnings volatility

17 Intra-Industry Reactions 55 after the announcement. 14 Following Sant and Cowan (1994), we define pre- (post-) period annual earnings volatility as variance of annual earnings per share (unscaled) and earnings per share standardized by share price (scaled), calculated over a ten-year period before (after) the announcement. We also calculate pre- and postperiod quarterly earnings volatility computed over the twenty quarters before and after the announcement. Change in earnings volatility is defined as the difference in earnings variance between the post- and pre-split period. Similar to prior results, we group the nonsplitting firms into equally weighted portfolios for each split announcement. We find that although the mean change in earnings volatility is negative for all four measures, only the change in earnings volatility measured by annual earnings per share is statistically significant (results not reported). We further test whether investors reaction to nonsplitting firms is influenced by their anticipation of a decline in earnings volatility. We regress announcement-period abnormal returns of nonsplitting firms on changes in earnings volatility. We find that the estimated coefficients of all change in earnings volatility measures are insignificant. 15 Thus, we conclude that, overall, a decline in earnings volatility after the announcement cannot explain the increase in firm value for nonsplitting firms. V. Summary and Conclusion We examine whether stock split announcements affect stocks prices of nonsplitting firms in the same industry. The results indicate that shareholders of nonsplitting firms experience a small but significant increase in equity value during the stock split announcement period of their industry counterparts. This finding, coupled with the positive abnormal returns of splitting firms, suggests the favorable information conveyed by stock split announcements transfers to nonsplitting firms within the same industry. The cross-sectional regression results indicate that the industry reaction is negatively related to the HHI and positively related to earnings correlation, return variance, and book-to-market ratio, after controlling for split factor, abnormal returns, and dominance level of splitting firms. Thus, we conclude that (a) nonsplitting firms in industries with a low degree of competition gain more from the split announcement than nonsplitting firms with a high degree of competition, (b) nonsplitting firms that are similar to splitting firms earn significantly higher abnormal returns than firms that are dissimilar, (c) firms with high asymmetric 14 We thank the referee for suggesting this test. 15 The regression results of firm level are similar to those reported earlier except that the coefficient of change in scaled quarterly earnings variance is significant at the 10% level.

18 56 The Journal of Financial Research information obtain higher returns than firms with low asymmetric information, and (d) nonsplitting firms that are underpriced benefit more from the announcement than firms that are overpriced. Finally, we find an improvement in earnings performance for nonsplitting firms, and the announcement-period returns for these firms are positively related to the earnings increase and abnormal earnings change. We find little evidence that the value of nonsplitting firms is enhanced because of a decline in earnings volatility. These findings suggest stock splits convey favorable industrywide information about subsequent earnings improvement. Another possible explanation for the positive intra-industry reaction is that stock splits also increase the likelihood of a split for nonsplitting firms, causing investors in these firms to react positively. 16 References Akhigbe, A., J. Madura, and A. Whyte, 1997, Intra-industry effects of bond rating adjustments, Journal of Financial Research 20, Asquith, P., P. Healy, and K. Palepu, 1989, Earnings and stock splits, Accounting Review 64, Conroy, R. M. and R. S. Harris, 1999, Stock splits and information: The role of share price, Financial Management 28, Desai, H. and P. C. Jain, 1997, Long-run common stock returns following stock splits and reverse splits, Journal of Business 70, Dierkens, N., 1991, Information asymmetry and equity issues, Journal of Financial and Quantitative Analysis 26, Fama, E. F. and K. R. French, 1992, The cross-section of expected stock returns, Journal of Finance 47, Firth, M., 1996, Dividend changes, abnormal returns, and intra-industry firm valuations, Journal of Financial and Quantitative Analysis 31, Foster, G., 1980, Externalities and financial reporting, Journal of Finance 35, Grinblatt, M. S., R. W. Masulis, and S. Titman, 1984, The valuation effects of stock splits and stock dividends, Journal of Financial Economics 13, Healy, P. and K. Palepu, 1988, Earnings information conveyed by dividend initiations and omissions, Journal of Financial Economics 21, Ikenberry, D. L., G. Rankine, and E. K. Stice, 1996, What do stock splits really signal, Journal of Financial and Quantitative Analysis 31, Krishnaswami, S. and V. Subramaniam, 1999, Information asymmetry, valuation, the corporate spin-off decision, Journal of Financial Economics 51, Lakonishok, J. and B. Lev, 1987, Stock splits and stock dividends: Why, who, and when, Journal of Finance 42, Lakonishok, J., A. Shleifer, and R. W. Vishny, 1994, Contrarian investment, extrapolation, and risk, Journal of Finance 49, Lang, L. H. P. and R. M. Stulz, 1992, Contagion and competitive intra-industry effects of bankruptcy announcements: An empirical analysis, Journal of Financial Economics 32, Laux, P., L. T. Starks, and P. S. Yoon, 1998, The relative importance of competition and contagion in intra-industry information transfers: An investigation of dividend announcements, Financial Management 27, We thank the referee for suggesting this possibility.

19 Intra-Industry Reactions 57 McNichols, M. and A. Dravid, 1990, Stock dividends, stock splits, and signaling, Journal of Finance 45, Parkin, M., 1996, Microeconomics (3d ed.) (Addison-Wesley, Reading, MA). Pilotte, E. and T. Manuel, 1996, The market s response to recurring events: The case of stock splits, Journal of Financial Economics 41, Sant, R. and A. Cowan, 1994, Do dividends signal earnings?: The case of omitted dividends, Journal of Banking and Finance 18, Szewczyk, S. H., 1992, The intra-industry transfer of information inferred from announcements of corporate security offerings, Journal of Finance 47,