Empirical Results 1. Post-split return drift

Table 1 reports monthly abnormal returns based on the calendar-time approach in the year following split announcements. We measure abnormal returns by controlling for Fama–

French (1993) three factors and Carhart (1997) four factors, respectively. We also control for five factors by including a liquidity factor in the Carhart model. Panel A shows the results for our main sample.

[TABLE 1 ABOUT HERE]

In the OLS regressions, stock splits generate at least 40 basis points each month in the post-split period for equal-weighted portfolios. Value-weighted portfolios exhibit lower and

insignificant returns when controlling for momentum and liquidity. However, OLS regressions may suffer the low power to detect abnormal returns (Loughran and Ritter, 2000). WLS regressions correct this problem and yield significant abnormal returns for both equal-weighted and value-weighted portfolios.

To determine whether our results are robust, we examine two sample periods by including stock splits before 1984. Panels B and C of Table 1 report the results for the period from 1963 to 2011 and 1926 to 2011, respectively.² The results are very similar across Panels A to C. Our results suggest a positive return drift in the year following the split announcement. By including more recent stock split cases, we confirm the findings of Ikenberry and Ramnath (2002) that post-split abnormal returns are positive and significant.

3.2. Post-split return drift by event months

Table 2 reports the post-split drift by different horizons based on the calendar-time approach. Panel A shows the monthly abnormal returns over 3, 6, and 12 months following split announcements, and Panel B shows the abnormal returns of each event month in the first post-split year. We report the results based on the five-factor model only, but the results are qualitatively similar when the three-factor model and the four-factor model are used.

[TABLE 2 ABOUT HERE]

2 We do not report results for the five-factor model because the liquidity factor is not available prior to 1963.

Although we confirm the positive return drift in the first post-split year by including more recent stock splits (see Table 1), the results in Table 2 show that the abnormal returns are not uniformly distributed among the 12 months after the splits. Most of the abnormal returns occur in the first few months after split announcements. For example, in the WLS regressions, the equal-weighted and value-equal-weighted three-month return drifts are 1.16% and 0.47%, respectively, per month (Panel A), and no significant positive abnormal return is observed after event month 7 (Panel B). In Figure 1, we plot the equal-weighted abnormal returns over 12 months after splits.

The abnormal return declines sharply in the third month after the splits and approaches to zero from month 8 (except for month 12). These results suggest that the post-split return drift is a short-term phenomenon. The duration of return drift matters for the holding period when forming the trading strategy; that is, the more precisely an investor can identify the duration of return drift, the better his or her investment performance may be.

[FIGURE 1 ABOUT HERE]

3.3. Relation between post-split drift and SUE

We have shown that the post-split drift is a short-term anomaly, which may last for three to seven months. As PEAD is also a three- to six-month phenomenon, we examine whether the post-split drift is related to the SUE effect. We first test whether earnings surprises help explain the abnormal returns following stock splits.

Table 3 reports the regressions of post-split three-month BHARs on earnings surprises (SUEs). The BHAR is the split firm return minus the return of the corresponding size, B/M, and momentum-matched control portfolio. We define SUE as the earnings surprise based on analysts’

forecasts and actual earnings, both reported in I/B/E/S, as in Livnat and Mendenhall (2006).³ We control for size, B/M ratio, prior return, pre-split price level, and change in liquidity around stock splits in the regressions. We find that SUEs are positively related to post-split returns (Model 1).

The earnings surprises prior to stock splits do not affect post-split returns (Model 2). These results are robust when we define SUE as the decile ranking instead of a continuous variable.

This finding is consistent with the notion that the market underreacts to the signal in split announcements about future positive earnings surprises, leading to predictable returns in the post-split period. We also find that the returns following stock splits are positively related to improvement in liquidity (i.e., negative coefficient of change in ILLIQ), a result consistent with Lin, Singh, and Yu (2009). The finding that changes in liquidity are significantly related to abnormal returns following splits suggests that earnings surprises may not be the only factor that drives the post-split drift.

[TABLE 3 ABOUT HERE]

Because the post-split drift is related to earnings surprises, we examine whether the SUE effect can account for the post-split drift. We pool all firms covered in both CRSP and

3 Our results are similar when we use the time-series models based on Compustat data to estimate SUEs.

Compustat with available I/B/E/S data and run cross-sectional regressions of monthly returns on a split dummy, which indicates whether a firm made a split announcement in the past three months, and SUE in the most recent quarter. If the post-split drift is driven by the same piece of information contained in SUEs, the split dummy will not be significant in explaining cross-section stock returns once the SUE is controlled. We control for important anomaly variables in the regressions that have been shown to be related to stock returns, such as size, B/M, past returns, asset growth, net share issue, accruals, returns on assets, and liquidity. Table 4 reports the empirical results. For robustness, we use both the monthly Fama–MacBeth regressions in Models 1 and 2 and panel regressions in Models 3 and 4 with double-clustered standard errors (Petersen, 2009).

[TABLE 4 ABOUT HERE]

Model 1 in Table 4 shows that the coefficient of split dummy is 1.16, significant at 1%

level, after controlling a variety of anomaly variables. This result suggests that split firms, on average, generate 1.16% abnormal returns over the three-month post-split period. This finding is consistent with the abnormal return in the equal-weighted cases reported in Panel A of Table 2.

SUE is also significant, suggesting that the SUE effect and the post-split drift do not subsume each other.⁴

4 Our results, although weaker, hold when SUE is defined based on Compustat data, known as the rolling seasonal random walk model.

Although both stock splits and earnings surprises are related to future earnings

changes, they appear to contain different pieces of information that helps explain cross-section returns.

In Model 2 of Table 4, we separate split firms into high, medium, and low groups of splits if their split factor is greater than or equal to 1, less than 1 but greater than or equal to 0.5, and less than 0.5, respectively. We find a significantly positive abnormal return associated with firms with high split factors, amounting to 1.35% per month in the post-split three-month period.

Firms with medium split factors also generate abnormal returns, but firms with low split factors generate no significant returns.

In panel regressions in Table 4, the coefficient for the split dummy is almost unchanged (Model 3), compared with the Fama–MacBeth regression (Model 1). This finding proves that stock splits are an important factor in predicting returns in the short run. The high split dummy continues to be strongly significant (Model 4). SUE becomes insignificant, suggesting that the premium for earnings surprises may vary over time.