We review economic and psychological models of expectation formation.
According to the rational model, expectations are optimal; according t
o the adaptive model, they minimize previous error; according to the e
xtrapolative model, they continue a trend. In an experiment, we asked
subjects to make forecasts from previous data in the same series or fr
om previous data in a different but related series. We also asked them
to estimate the probability that their forecasts would be correct. Th
e rational model failed because within-series forecasts were too poor,
cross-series forecasts were biased, and probability estimates showed
overconfidence. Regression analyses and a lack of performance improvem
ent showed that the adaptive model was inappropriate. Instead, within-
series forecasts were made by extrapolation and cross-series forecasts
appeared to be based on a faulty conditional rule.