Social Learning — Bayesian Herding & Information Cascades
Bikhchandani-Hirshleifer-Welch 1992 · Sequential decisions · Cascade onset
Signal & Agent Parameters
Signal precision p(correct):
0.65
Number of agents N:
50
True state (0=A, 1=B):
B
Prior Pr(B):
0.50
Run New Trial
100 Trials
Cascade started at:
—
Cascade direction:
—
Final belief Pr(B):
—
Correct decision:
—
Cascade rate (100 trials):
—
Model:
agents act sequentially, each observing all prior actions + private signal.
P(B|actions,signal) ∝ P(actions|B)·P(signal|B)·P(B)
Cascade:
once log-odds |L| exceeds signal strength, rational agents ignore private signal and herd on prior actions.
Cascade onset: |Σ logit(aᵢ)| > logit(p)
Cascades can be wrong — early movers dominate; fragile to new information.