Information Cascades

Bayesian herding: agents follow the crowd even when private signals say otherwise

Parameters

Cascade State

True stateA
Agents decided0
Correct fraction-
Cascade started at-
Public belief P(A)0.50

Agent Decisions

Green=correct, Red=wrong, Blue=cascade follower
Bayesian information cascade: Each agent observes a private binary signal (correct with prob p) and all previous public choices, then picks the option with higher posterior probability. Cascade: once 2 more agents have chosen A than B (or vice versa), all subsequent agents rationally ignore private signals and herd — even if their signal says otherwise. This creates fragile unanimous-looking agreement that can suddenly reverse. Signal reliability p controls cascade depth and probability of correct herding.