Stock market phenomena such as the January and low price/earnings ratio effects entice investors with prospects of extraordinary returns. Most previous stock market anomaly research has focused on one or two return regularities at a time. Multivariate regression, however, can provide a unified framework for disentangling and analyzing numerous return effects. By simultaneously controlling for other attributes, it “purifies” the effect of each anomaly, affording a clearer picture of which anomalies are “real” and which are merely proxies for other effects.
While “pure” payoffs may be smaller than the naive payoffs (given the independent nature of the pure effects and the proxying behavior of the naive effects), their statistical significance is often greater. The residual reversal effect is an exception, emerging stronger in magnitude in its pure form than its naive form, primarily because the pure measure separates out related effects such as earnings surprise. Some effects, however, such as cash flow/price ratio disappear completely in their pure form. And both naive and pure returns to beta prove inconsequential in explaining cumulative returns.
The strength and persistence of returns to some of the anomaly measures, such as trends in analysts’ earnings estimates, represent evidence against semi-strong market efficiency. Furthermore, the significant payoffs to other measures, such as residual reversal, suggest that past prices alone do matter—that is, the market is not even weak-form efficient.
Controlling for tax-loss selling and other attributes in a multivariate framework mitigates the January seasonals exhibited by many of the naive anomaly measures. For instance, the small size effect’s January seasonal vanishes. The yield effect’s January seasonal remains strong, however. Also, because long-term tax-loss selling is more powerful than short-term, investor behavior appears suboptimal. A negative January seasonal in pure returns to the relative-strength measure appears to arise from profit-taking associated with tax-gain deferral.
Returns to many attributes appear to have market-related components. For example, naive returns to low P/E behave defensively, while pure returns to low P/E are not market-related at all. Apparently naive returns to low P/E are proxies for related defensive effects such as the yield effect. Returns to beta, however, are strongly procyclical in both their naive and pure forms.