Long-term Performance after Stock Splits: A Closer Look
K.C. Chen, Sangphill Kim, and Peter H. Xu
* California State University-Fresno, University of Massachusetts-Lowell, and Prudential Investments,
Criag School of Business
California State University, Fresno
Fresno, CA 93740-0007
(559) 278-5646 (O) (559) 278-4911 (F)
Long-term Performance after Stock Splits: A Closer Look
Firms that split their shares from 1931 through 1990 outperform size-matched
firms by 5.10 percent and beta-matched firms by 4.62 percent in the first year following
the split. Most of this drift occurred over the period from 1976 to 1990; for the earlier
time periods, the beta-adjusted excess returns in the year after the split are insignificant.
Firms splitting their shares under-perform both size-matched and beta-matched firms in
the second and third years after the split. Furthermore, firms that did not experience
positive excess returns in the six months before the split do not exhibit drift either. The
evidence suggests that the behavior of stock returns following stock splits is similar to the
short-term momentum and long-term reversals documented for stock returns in general,
and may not be attributable to the split announcement per se.
Long-term Performance after Stock Splits: A Closer Look
Several studies on stock splits find conflicting evidence regarding whether there is a long-term drift in stock returns following the split announcement. Fama, Fisher, Jensen,
and Roll (1969) examine stock splits from 1927 to 1959, and, consistent with market
efficiency, they find no abnormal returns on the splitting firms in 30 months after the split.
Ikenberry, Rankine, and Stice(1996) study stock splits from 1975 to 1990 and observe
abnormal returns of 7.93 percent in one year after the split announcement
1. The results of
these two studies, however, are not readily comparable since they examine stock splits
over different time periods and use different methods to adjust for risk. Fama, et al. use
the market model while Ikenberry, et al. use size and book-to-market ratio to compute
In this study, we re-examine stock splits over a 60-year period, spanning from
1931 through 1990. The sample period is also divided into four 15-year subperiods:
1931-1945, 1946-1960, 1961-1975 and 1976-1990. The purpose of doing so is to find out
whether there are post-split excess returns, using either beta or size to proxy for risk,
across all subperiods. The much longer sample period also includes the period from 1960
to 1974, which has not been investigated previously.
We find two interesting results. First, although both size- and beta-adjusted excess
returns in the first year after the split are positive for all subperiods, they are the largest
1 Desai and Jain (1996) use a more inclusive sample of stock splits over roughly the same time period
(1976-1991) and find an abnormal return of 7.05 percent for split firms in one year after the split.
2for the subperiod from 1976 to 1990. For splits over this most recent subperiod, the size-
adjusted and beta-adjusted excess returns in one year after the split are 6.67 percent and 6.64 percent, respectively, both highly significant at the one percent level. For the other three subperiods, the drift is smaller and insignificant when beta is used to proxy for risk. Secondly, the drift or the excess performance tends to be short-lived. Both size-adjusted and beta-adjusted excess returns beyond one year after the split are mostly negative. Since stocks that split their shares often have experienced large price runups before the split, the short-lived drift and subsequent negative excess returns are consistent with the short-term momentum and long-term mean reversals documented by recent studies for stock returns in general
3. By separating the splitting firms into deciles based on presplit
performance, we find that firms that did not experience positive excess returns prior to the split exhibit no drift either. The results suggest that the drift after stock splits and momentum in stock returns in general may be one and the same puzzle. This has
implications for other corporate announcements such as earnings and dividend surprises that also have been identified with post-announcement drift.
The rest of the paper is organized as follows. Section II discusses data and
methodology. Section III presents empirical results on both size- and beta-adjusted excess returns in each of the first three years after the split. Finally, Section IV concludes the paper.
2 This is similar to the finding of Michaely, Thaler, and Womack (1995) who examine long-term drift after dividend initiations and omissions. They find that the drift is more pronounced in the period from 1975 to 1988 than in the period from 1966 to 1974. 3 Jegadeesh and Titman (1993) and Chan, Jegadeesh, and Lakonishok (1996) find that stock returns exhibit momentum over 3- to 12-month holding periods. Fama and French (1988) and DeBondt and Thaler (1985)
find that stock returns are negatively correlated over long horizons of 3 to 5 years. Over horizons shorter than a month, stock returns also exhibit reversal patterns (See Jegadeesh 1990).
II. Data and methodology
The CRSP monthly master file is used to identify stock splits by all NYSE and
ASE firms in the period from 1931 through 1990. To be included in the analysis, the split
has to be at least 5 new shares for 4 old shares and has a valid announcement date. For
the purity of the sample, stock dividends and combinations of stock dividends and splits
are excluded. The selection procedure results in 5036 stock splits4.
The size ranks used in this study are directly obtained from the CRSP tape.
CRSP computes size ranks using firms’ market capitalization for the previous year-end.
If market capitalization for the previous year-end is not available, the capitalization on
the earliest date in the current year with available price is used for ranking. Size rank of 1
(10) corresponds to securities with the smallest (largest) capitalization. The beta ranks are
obtained by ranking all stocks recorded on the CRSP tape. Betas are estimated once a
year, using 60 monthly returns before a relevant year. The market index used in the
estimation is the CRSP value-weighted index. All 60 monthly returns have to be available
for a firm to have a valid beta rank. Stocks with the smallest (largest) betas are assigned a
rank of 1 (10).
Table 1 presents summary statistics of the sample. The distribution of stock splits
is heavily skewed towards the more recent periods. The number of stock splits over the
four subperiods 1931-1945, 1946-1960, 1961-1975 and 1976-1990 is 47, 329, 1619 and
5. Of the 5036 total stock splits, 4995 have valid size ranks and 3453
have valid beta ranks. The average size rank for all firms is 6.64, which implies that
4 If stock dividends are included, the total sample size would be 6077. Desai and Jain (1996) find that post-
announcement excess returns are similar for stock dividends and for stock splits.
splitting firms are above average in market capitalization. The average beta rank for all firms is 4.52, which implies that splitting firms have lower systematic risk on average. As Table 1 shows, stock splits are often accompanied with significant presplit price runups. Of all stock splits, the average six-month excess return over the CRSP equal-weighted index is 18.99 percent.
INSERT TABLE 1 HERE
Post-split excess returns are calculated assuming an equal-weighted buy-and-hold strategy similar to that used by Ikenberry, Rankine, and Stice (1996), Desai and Jain (1996) and Ikenberry, Lakonishok, and Vermaelen (1995). Both excess returns and significance levels are obtained using a bootstrapping procedure. Specifically, for each firm in the sample, we randomly select another firm that is not the actual splitting firm but has the same size or beta rank as the splitting firm at the time of the split. This pseudo firm is assumed to be bought and held for three years after the split. If the pseudo firm stops trading in the middle of the three-year period, another pseudo firm is picked randomly from the original list to replace the vanished pseudo firm for the remaining
months. If an actual splitting firm stops trading in the middle of a year, the monthly returns on the size- or beta-matched pseudo firm in the remaining portion of the year are used to compute the annual return on the splitting firm. For example, if a sample split occurred in February of 1986 and the firm vanished in August 1986, then the annual return on this firm in the second year after the split will be the compounded return using its monthly returns from March 1986 through August 1986 and monthly returns on the pseudo firm from September 1986 through February 1987. Portfolios are assumed to be
5 If stock dividends are included, the numbers of observations for the four subperiods are 130, 820, 2010 and 3117, respectively. Fama, Fisher, Jensen, and Roll (1969) examined stock splits during 1927-1959.
rebalanced annually; therefore, the splitting firm in this example will not be included in computing portfolio returns in the third year after the split.
We repeat this process 1,000 times. The difference between average returns on the portfolio of splitting firms and on the portfolio of pseudo firms is taken as the excess return on the splitting firms. The significance level is determined by a p-value which is equal to the proportion of 1000 repetitions when the return on the pseudo portfolio exceeds the return on the portfolio of splitting firms.
III. Empirical results
A. Size-adjusted excess returns
Average annual returns on the splitting firms and on the size-matched firms in each of the three years after the split are reported in Table 2. During the entire sample period from 1931 to 1990, the average return on the splitting firms in the first year after the split is 16.42 percent, which is 5.10 percent higher than the average return on the non-splitting firms matched on size. The p-value of 0.000 indicates that none of the 1000 portfolios of pseudo firms earned higher return than the portfolio of splitting firms.
The drift following the split announcement seems to be short-lived. Beyond one year after the split, the portfolio of splitting firms underperforms the portfolio of size-matched firms in both years. Specifically, the average size-adjusted excess returns in the second and third years after the split are -1.63 percent and -1.15 percent, respectively, both statistically significant. The p-values of 0.998 and 0.982 indicate that 998 and 982 portfolios of pseudo firms have higher return in the second and third year, respectively, than the portfolio of splitting firms.
Their sample size was 940.
INSERT TABLE 2 HERE
The results for the four 15-year subperiods indicate that the positive drift in the first year after the split is mainly attributable to splits that occurred between 1976 and 1990. The average size-adjusted excess return during this most recent period is 6.67% percent, compared to 3.09%, 2.11% and 2.83% in the other three subperiods. As is true with the aggregate sample, the excess returns on the splitting firms in the second and third years after the split are mostly negative. This pattern of reversal beyond one year after the split is particularly strong and statistically significant for the subperiods 1931-1945 and 1961-1975.
B. The momentum effect hypothesis
The positive drift in the first year and subsequent reversals in the second and third years after the split are similar to the pattern documented by some recent studies for stock returns in general. This suggests that the positive drift in the first year may not be market underreaction to the split announcement per se, but the momentum effect from the good performance before the split. In order to investigate this possibility, we rank the sample based on firms’ excess returns in the six months prior to the split month and place them in deciles. Decile 1 (10) consists of firms that have the smallest (largest) presplit excess returns. We calculate presplit excess returns by subtracting the returns on the CRSP equal-weighted index from the returns on the splitting firms. The requirement of availability of stock returns in the six months before the split reduces the sample size to 4,765 splits.
INSERT TABLE 3 HERE
The post-split annual returns in each of the deciles are shown in Table 3. The
firms placed in the lowest three deciles have average presplit six-month excess returns of
-21.31%, -7.02% and -0.52%, respectively. This indicates that a lot of firms splitting their
shares did not experience positive excess returns before the split. Consistent with the
momentum effect hypothesis, the firms with their presplit excess returns placed in these
three lowest deciles do not exhibit post-split drift6. Furthermore, for the firms in the other
seven deciles that exhibit significant positive drift in the year following the split, the
magnitude of the drift seems to be positively correlated with the presplit excess returns.
For example, the firms in decile 4 with an average presplit excess return of 4.71 percent
show a drift of 2.45 percent in the first year after the split, while the firms in decile 10
with an average presplit excess return of 94.22 percent show a drift of 10.41 percent.
Table 3 shows that, in general, the negative excess returns beyond one year after
the split are also correlated with presplit excess returns. For example, the average excess
return in the second year following the split is -2.49 percent for the firms placed in decile
4, and -6.50 percent for firms placed in decile 10. This is consistent with the long-term
reversal pattern documented for stock returns in general.
C. Beta-adjusted excess returns
One of the primary motivations for this study is to test the robustness of results
using alternative proxies for risk consistently throughout different subperiods. In this
section, we replicate the bootstrapping procedures using beta-matched firms to compute
6 The momentum effect predicts that the firms in the lowest deciles with very negative prior excess returns should continue to exhibit negative excess returns after the split. As Table 3 shows, for the firms placed in the lowest two deciles, the post-split excess returns in the first year following the split are indeed negative. However, these negative excess returns are not statistically significant.
excess returns and associated p-values. Betas are estimated once a year and from the market model using monthly returns in five calendar years before a relevant year. For example, the beta of any firm listed on the CRSP tape for any of the twelve months in 1985 is estimated using the 60 monthly returns from 1980 through 1984. In order for a firm to have a valid beta and to be included in the ranking, we require that all 60 monthly returns are available on the CRSP tape. Although this requirement reduces the number of splitting firms with valid beta ranks and the number of potential stocks to be included in the bootstrapping, it ensures that the measurement errors of betas are properly controlled.
The average annual returns on the splitting firms and on the beta-matched firms are reported in Table 4. Because of the reduction in the sample size, the annual returns on the portfolio of splitting firms are slightly different from those reported in Table 2. For the new aggregate sample, the average returns on the portfolio of splitting firms and the portfolio of beta-matched firms are 17.23 percent and 12.61 percent, respectively, in the first year following the split. The difference of 4.62 percent is highly significant with a p-value of 0.000. As before, the splitting firms exhibit statistically significant negative excess returns over the second and third years after the split.
INSERT TABLE 4 HERE
Table 4 shows that most of the post-split drift occurred in the subperiod 1976-1990 during which the average beta-adjusted excess return is 6.64 percent in the first year following the split. For the subperiods 1946-1960 and 1961-1975, the beta-adjusted excess returns in the first year following the split are only 1.60 percent and 0.82 percent, respectively. Moreover, the p-values associated with these excess returns indicate that