Behavioral Biases and Investor Behavior : Predicting the Next Step of a Random Walk (Revisited)

A recent paper by Bloomfield and Hales (2002) reports results of laboratory experiments with MBA-student participants that support the existence of regime-shifting beliefs of the type theorized by Barberis, Shleifer, and Vishny (1998). Specifically, they find a strong tendency for subjects to expect reversals in performance after seeing historical sequences with many reversals and to expect trending after seeing historical sequences with fewer reversals. We argue that the Bloomfield and Hales experiment does not provide a useful test of BSV because the set of sequences shown to subjects are not from a random walk process, but instead are more consistent with what would be expected if the true underlying process was of a regime-shifting type. A simple chi-square goodness of fit test based on the frequency of reversals strongly rejects that the set of sequences used in the experiment were generated by a random walk process. Furthermore, consistent with an underlying regime shifting process, the Bloomfield and Hales sequences have far too many observations in the tails of the distribution of reversal rates. We contend that because of the particular set of performance sequences used, the Bloomfield and Hales experiment cannot distinguish whether subjects rationally conclude that the underlying process is of a regime-shifting type or whether the subjects' belief in regime-shifting arises from behavioral biases as suggested by BSV. When we modify the Bloomfield and Hales experiment to be consistent with the crucial BSV assumption that the underlying data-generating process is random, we find no evidence supportive of investor belief in regime-shifting. Subjects in our experiment are more likely to expect a reversal in performance, rather than trending, after observing sequences with fewer reversals. This finding is consistent with the well-documented gambler's fallacy effect, which is a form of representativeness bias that arises when individuals know the underlying return-generating process