Information Cascade & Bayesian Herding

Bikhchandani-Hirshleifer-Welch (1992): rational agents follow the crowd, ignoring private signals

Information Cascade Theory

In a Bayesian model, agents act sequentially, observing prior choices but only their own private signal about the true state. A cascade starts when it becomes rational to ignore your private signal and follow the crowd.

P(A | actions) > P(B | actions)
→ choose A regardless of signal
(cascade: private info lost!)

Agent n observes log-odds update:

L_n = L_{n-1} + log(P(s|A)/P(s|B))
Cascade when |L_n| > log(p/(1-p))

Cascades are fragile: they rest on thin information. A small shock can break them. This explains fashion, technology adoption, bank runs, and herding in financial markets.

Even with correct signals, wrong cascades happen: if early agents (by chance) all choose wrong, all subsequent agents follow — a rational cascade to the wrong outcome.

Run a simulation to begin