Bayesian agents observe private signals + public history → herding behavior
Cascade formed: —
Correct decisions: —
True state: —
0.70
30
6
How it works: True state is A or B (equal prior). Each agent gets a private signal: correct with probability p.
Agent updates Bayesian posterior given all previous public decisions, then acts on majority.
Once a cascade forms, private information is ignored — all subsequent agents copy the herd regardless of their signal.
Early mistakes propagate indefinitely, making cascades fragile.