This paper was prepared for John B. Taylor and Michael Woodford, Editors

10

People tend to neglect “base rates,” the unconditional probabilities or frequencies of events, see

Meehl and Rosen (1955).

11

Rabin (1996) characterizes this judgment error as a tendency to over-infer the probability

distribution from short sequences.  Part of overconfidence may be nothing more than simple forgetting

of contrary evidence; a tendency to forget is by its very nature not something that one can learn to

prevent.

13

subjects for probability odds that they are right and very clearly explaining what such odds

mean, and even asking them to stake money on their answer.  The overconfidence

phenomenon persisted.  Moreover, in cases where the subjects said they were certain they

were right, they were in fact right only about 80% of the time:  there is no interpretation of

subjective probability that could reconcile this  result with correct judgments.

A tendency towards overconfidence among ordinary investors seems apparent when one

interviews them.  One quickly hears what seem to be overconfident statements.  But how can

it be that people systematically are so overconfident?  Why wouldn’t people learn from

life’s experiences to correct their overconfidence?

Obviously, people do learn substantially in circumstances when the consequences of

their errors are repeatedly presented to them, and sometimes they even overreact and show

too little confidence.  But still there seems to be a common bias towards overconfidence.

Overconfidence is apparently related to some deep-set psychological phenomena:  Ross

(1987) argues that much overconfidence is related to a broader difficulty  with “situational

construal,” a difficulty in making adequate allowance for the uncertainty in one’s own view

of the broad situation, a more global difficulty tied up with multiple mental processes.

Overconfidence may also be traced to the “representativeness heuristic,” Tversky and

Kahneman (1974), a tendency for people to try to categorize events as typical or repre-

sentative of a well-known class, and then, in making probability estimates, to overstress the

importance of such a categorization, disregarding evidence about the underlying

probabilities.

10

  One consequence of this heuristic is a tendency for people to see patterns

in data that is truly random, to feel confident, for example, that a series which is in fact a

random walk is not a random walk.

11

Overconfidence itself does not imply that people overreact (or underreact) to all news.

In fact, evidence on the extent of overreaction or underreaction of speculative asset prices

to news has been mixed.

There has indeed been evidence of overreaction.  The first substantial statistical

evidence for what might be called a general market overreaction can be found in the

literature on excess volatility of speculative asset prices, Shiller (1979, 1981a,b) and LeRoy

and Porter (1981).  We showed statistical evidence that speculative asset prices show

persistent deviations from the long-term trend implied by the present-value efficient markets

model, and then, over horizons of many years, to return to this trend.  This pattern of price

behavior, it was argued, made aggregate stock prices much more volatile than would be

implied by the efficient markets model.  It appears as if stock prices overreact to some news,

or to their own past values, before investors come to their senses and correct the prices.  Our

arguments led to a spirited debate about the validity of the efficient markets model in the

---

12

There has been some confusion about the sense in which the present-value efficient markets

model puts restrictions on the short-run (or high frequency) movements in speculative asset prices.

The issues are laid out in Shiller (1979), (appendix).  Kleidon (1986) rediscovered the same ideas

again, but gave a markedly different interpretation of the implications for tests of market efficiency.

13

An extensive summary of the literature on serial correlation of US stock index returns is in

Campbell, Lo and MacKinlay (1997).  Chapter 2 documents the positive serial correlation of returns

over short horizons, but concludes that the evidence for negative serial correlation of returns over long

horizons is weak.  Chapter 7, however, shows evidence that long-horizon returns are negatively

correlated with the price-earnings ratio and price-dividend ratio.  Recent critics of claims that long-

horizon returns can be forecasted include Goetzmann and Jorion (1992), Nelson and Kim (1993) and

Kirby (1997).  In my view, they succeed in reducing the force of the evidence, but not the conclusion

that long-horizon returns are quite probably forecastable.

14

finance literature, a literature that has too many facets to summarize here, except to say that

it confirms there are many potential interpretations of any statistical results based on limited

data.

12

 My own view of the outcome of this debate is that it is quite likely that speculative

asset prices tend to be excessively volatile.  Certainly, at the very least, one can say that no

one has been able to put forth any evidence that there is not excess volatility in speculative

asset prices.  For an evaluation of this literature, see Shiller (1989), Campbell and Shiller

(1988, 1989), West (1988), and Campbell, Lo and MacKinlay (1997, Ch. 7).

Since then, papers by De Bondt and Thaler (1985), Fama and French (1988), Poterba

and Summers (1988), and Cutler, Poterba and Summers (1991) have confirmed the excess

volatility claims by showing that returns tend to be negatively autocorrelated over horizons

of three to five years, that an initial overreaction is gradually corrected.  Moreover,

Campbell and Shiller (1988, 1989) show that aggregate stock market dividend yields or

earnings yields are positively correlated with subsequently observed returns over similar

intervals; see also Dreman and Berry (1995).

13

  Campbell and Shiller (1998) connect this

predictive power to the observed stationarity of these ratios.  Since the ratios have no

substantial trend over a century and appear mean reverting over much shorter time intervals,

the ratio must predict future changes in either the numerator (the dividend or earnings) or

the denominator (the price); we showed that it has been unequivocally the denominator, the

price, that has restored the ratios to their mean after they depart from it, and not the

numerator.  La Porta (1996) found that stocks for which analysts projected low earnings

growth tended to show upward price jumps on earnings announcement dates, and stocks for

which analysts projected high earnings growth tended to show downward price jumps on

earnings announcement dates.  He interprets this as consistent with a hypothesis that

analysts (and the market) excessively extrapolated past earnings movements and only

gradually correct their errors as earnings news comes in.  The behavior of initial public

offerings around announcement dates appears also to indicate some overreaction and later

rebound, see Ibbotson and Ritter (1988) and Ritter (1991).

On the other hand, there has also been evidence of what might be called underreaction.

Most days when big news breaks have been days of only modest stock market price

movements, the big movements tending to come on days when there is little news, see

Cutler, Poterba and Summers (1989).  Cutler, Poterba and Summers (1991) also found that

---