I model the effect of associative memory on asset prices. The model includes mood-congruent memory, which predicts that the subjective goodness (or badness) of the agent’s affective state (e.g., mood) is a cue for positive (negative) information stored in long-term memory. Mood-congruence implies that events that raise affect (e.g., improving market conditions) cause market participants to recall a positively biased sample of information from memory, which makes the market participants optimistic about future market performance. My model also includes rehearsal, which implies that data recalled in the recent past are more likely to be recalled in the present. I prove that rehearsal generates autocorrelation in the bias across periods. The combination of the biased beliefs and autocorrelation provides a novel explanation for short-run continued overreaction to news and long-run correction of these effects. Because the sign of the memory bias in beliefs is driven by market movements, changes in market conditions have an outsized effect on asset prices, which generates excess volatility.
I also make a number of novel predictions. First, I predict that excess volatility is highest during downturns. This follows from “immune neglect,” which predicts that negative events have less influence on mood than positive ones. Positive markets have lower volatility for two reasons: first, there is less room for market affect to improve after positive shocks; and two, immune neglect blunts the impact of negative shocks on mood. I also predict that price biases are increasing in fundamental volatility since mood moves more for large shocks than for small ones. Finally, I argue that knowledge/experience may intensify these biases, as knowledgeable agents have a larger store of information that can be recalled in a biased fashion.