Information Cascades

Imagine choosing between two restaurants: one empty, one full. You might have private information suggesting the empty one is better. But the crowd at the other place is also information — perhaps they know something you don’t. Rationally, you weight their apparent judgment against your private signal. If you join the crowd, the next person sees an even stronger signal. Soon everyone is in the same restaurant, regardless of quality.

This is an information cascade: a sequence of decisions where people rationally ignore their private information to follow the apparent consensus. Each person thinks: “All those people can’t be wrong.” But all those people made the same calculation. The cascade can be completely detached from underlying reality.

Cascades are fragile and self-reinforcing simultaneously. Once started, they gather momentum — each new follower strengthens the signal for the next. But they can shatter instantly when new public information arrives. The restaurant empties when someone shouts “rats!” The stock crashes when earnings disappoint. The cascade reversed is just as sudden as the cascade built.

The mechanism differs from mimetic desire. Mimesis is wanting what others want because they want it. Cascades are believing what others apparently believe because their behavior seems informative. One is about preferences; the other is about inferences. Both produce herding; the psychology differs.

The vulnerability: cascades built on inferred signals are as good as those signals. If early movers had weak information, or if their visible behavior masked private doubts, the cascade has no foundation. The full restaurant might be full because someone famous once ate there — once — and everyone since has been following everyone else’s following.

Protecting against cascades means valuing your private information even when the crowd disagrees. It means asking not just “what does the crowd know?” but “what does each individual in the crowd actually know versus merely infer from others?” The distinction between genuine consensus and self-reinforcing inference is invisible from outside.