Finally, in some of the examples given above, it is not just that one port-
folio outperforms another on average. In some cases, the outperformance is
present in almost every period of the sample. For example, in Bernard and
Thomas’s (1989) study, firms with surprisingly good earnings outperform
those with surprisingly poor earnings in 46 out of the 50 quarters studied.
It is not easy to see any risk here that might justify the outperformance.
5.1. Belief-based ModelsThere are a number of behavioral models which try to explain some of the
above phenomena. We classify them based on whether their mechanism
centers on beliefs or on preferences.
Barberis, Shleifer, and Vishny (1998), BSV henceforth, argue that much
of the above evidence is the result of systematic errors that investors make
when they use public information to form expectations of future cash
flows. They build a model that incorporates two of the updating biases from
section 3: conservatism, the tendency to underweight new information rela-
tive to priors; and representativeness, and in particular the version of repre-
sentativeness known as the law of small numbers, whereby people expect
even short samples to reflect the properties of the parent population.
When a company announces surprisingly good earnings, conservatism
means that investors react insufficiently, pushing the price up too little.
Since the price is too low, subsequent returns will be higher on average,
thereby generating both post-earnings announcement drift and momentum.
After a series of good earnings announcements, though, representativeness
causes people to overreact and push the price up too high. The reason is
that after many periods of good earnings, the law of small numbers leads
investors to believe that this is a firm with particularly high earnings
growth, and hence to forecast high earnings in the future. After all, the firm
cannot be “average.” If it were, then according to the law of small num-
bers, its earnings should appear average, even in short samples. Since the
price is now too high, subsequent returns are too low on average, thereby
generating long-term reversals and a scaled-price ratio effect.
To capture these ideas mathematically, BSV consider a model with a rep-
resentative risk-neutral investor in which the true earnings process for all
assets is a random walk. Investors, however, do not use the random-walk
model to forecast future earnings. They think that at any time, earnings are
being generated by one of two regimes: a “mean-reverting” regime, in
which earnings are more mean-reverting than in reality, and a “trending”
regime in which earnings trend more than in reality. The investor believes
that the regime generating earnings changes exogenously over time and sees
his task as trying to figure out which of the two regimes is currently gener-
ating earnings.
A SURVEY OF BEHAVIORAL FINANCE 41