Information Cascade Simulator: Social Proof & Herd Behavior

simulator intermediate ~9 min
Loading simulation...
Cascade forms after ~5 agents — remaining agents follow the herd

With 50 agents, 70% signal quality, and default social weight, a cascade typically forms within 5 agents. Once established, all subsequent agents ignore their private information and copy the majority, even if the cascade is wrong.

Formula

Bayesian update: P(A|signal,history) = P(signal|A) * P(A|history) / P(signal)
Cascade condition: P(A|history) > threshold, ignore private signal
Signal quality: P(correct signal) = signal_quality (0.5 = random, 1 = perfect)

Following the Crowd

Imagine choosing between two restaurants. You have a slight preference for Restaurant A, but you see a long queue at Restaurant B. You reason: all those people probably know something I don't. So you join the queue at B. The next person sees an even longer queue at B and does the same. This is an information cascade — a chain of rational imitation that can lead everyone to the same choice, even if that choice is wrong. The queue at B might have started with just two people who happened to arrive first.

The Mechanics of Cascading

Each agent in the simulation receives a private signal about which option (A or B) is truly better. The signal is accurate with probability 'signal_quality' (0.7 = 70% correct). Agents also observe all previous choices. Using Bayesian reasoning, each agent combines their private signal with the social evidence. Once the social evidence becomes strong enough (exceeding the cascade threshold), agents ignore their private signals entirely and follow the crowd. This is individually rational but collectively fragile.

Correct vs Wrong Cascades

The simulation color-codes outcomes: cyan for correct cascades (the herd chose the truly better option) and red for wrong cascades (the herd converged on the inferior option). Run it many times to see that wrong cascades are disturbingly common — sometimes 20-30% of runs produce incorrect herds. The first 2-3 agents' random signals determine the fate of all subsequent agents. This is the fundamental fragility that Bikhchandani, Hirshleifer, and Welch identified in their seminal 1992 paper.

Breaking the Cascade

Information cascades are fragile precisely because they are based on observed behavior, not deep conviction. A single piece of high-quality contrary information can shatter a cascade, causing rapid reversal. In financial markets, this explains both bubbles (wrong cascades forming) and crashes (cascades breaking). The social weight parameter controls how much agents defer to social information versus their own signals. Reduce it to see more independent thinking — and more accurate aggregate outcomes. Increase it to see faster but more fragile cascades.

FAQ

What is an information cascade?

An information cascade occurs when people sequentially observe others' actions and rationally decide to follow the crowd rather than their own private information. Once enough early agents choose the same option, it becomes rational for subsequent agents to imitate — even if their private signal disagrees. This can lead entire populations to converge on wrong answers.

How is social proof related to cascades?

Social proof is the psychological tendency to assume others' behavior reflects correct information. Information cascades are the rational version: when you see 10 people choose Restaurant A, it's rational to infer they have information you don't. The problem is that those 10 people might all be following the same 2 early choosers.

Can cascades be wrong?

Yes, and this is the key danger. If the first 2-3 agents happen to receive misleading signals and choose incorrectly, everyone after them rationally follows — creating a 'wrong cascade.' Bikhchandani, Hirshleifer, and Welch (1992) showed that cascades are inherently fragile and can be based on very little actual information.

Where do information cascades occur in real life?

Cascades explain stock market bubbles and crashes, restaurant popularity, technology adoption (VHS vs Betamax), fashion trends, academic citation patterns, and even medical treatment fads. Anywhere sequential decision-making and observability of others' choices combine, cascades can form.

Sources

Embed

<iframe src="https://homo-deus.com/lab/behavioral-economics/social-proof-cascade/embed" width="100%" height="400" frameborder="0"></iframe>
View source on GitHub