Information cascade
An information cascade occurs when someone follows the choices of earlier people regardless of their own private information. In the basic model, following others can be individually rational, yet prevent the group from learning: once decisions cease to reveal new signals, further agreement need not provide independent evidence.[1]
How a cascade begins
Consider two equally likely urns. Urn A contains two a-balls and one b-ball; urn B contains one a-ball and two b-balls. People draw privately with replacement, then publicly choose which urn they think was selected. Each sees earlier choices but not earlier draws. Assume everyone updates correctly and follows their own draw when the evidence is tied.[2]
If the first two people choose A, their choices reveal two a-draws under these assumptions. The probability of urn A is then 4/5. A third person who draws b still assigns probability 2/3 to A, so chooses A. That third choice no longer reveals the private draw. Later people who understand this cannot count it as another independent a-signal.[2]
The mechanism depends on what people observe, the reliability of signals and the available choices. A crowd's agreement alone does not identify a cascade: people may share information or preferences, face sanctions, or benefit directly from choosing the same thing. New public information can also overturn a cascade.[1]
Laboratory evidence and decision rules
Anderson and Holt studied paid, sequential urn choices. Cascades formed in 87 of 122 eligible experimental periods, including 31 in the wrong direction. Their asymmetric treatment made simple counting and Bayesian inference give different recommendations; participants did not consistently follow the Bayesian rule.[2]
Huck and Oechssler examined hypothetical urn decisions presented as exam questions, with stipulated rational predecessors and incentives for correct answers and explanations. Their working paper found that following one's private signal described choices better than the Bayesian prediction. They also questioned whether aggregate agreement in the earlier experiment identified the reasoning used. The exam tasks differed from repeated, paid interaction with feedback; the results concern both behaviour and how its mechanism is inferred.[3]
See also
References
- ↑ 1.0 1.1 Sushil Bikhchandani, David Hirshleifer and Ivo Welch, “Information cascades”, author-hosted online version of the entry in The New Palgrave Dictionary of Economics, second edition, 2008, especially pp. 1–6 and 9–10.
- ↑ 2.0 2.1 2.2 Lisa R. Anderson and Charles A. Holt, “Information Cascades in the Laboratory”, university-hosted 27-page manuscript, especially pp. 3–6, 8–10 and 16–22. The final article appeared in American Economic Review 87(5), 1997, pp. 847–862.
- ↑ Steffen Huck and Jörg Oechssler, “Informational cascades in the laboratory: Do they occur for the right reasons?”, working paper, 23 January 1999, especially pp. 2–7 and Tables 1–3. A later version appeared in Journal of Economic Psychology 21(6), 2000, pp. 661–671.
