Random Walk Perception and Information Acquisition in an Interactive Prediction Task Experiment

Abstract

We investigate experimentally individual random walk perception biases and the existence of decision clustering in a simple interactive prediction task. Our design is quite general and presents a series of sequential choice problems in which the subjects are asked to forecast the subsequent outcome of a discrete binary random process. The data is generated in such a way that observation of other participants' cumulated choices makes it possible to obtain a more precise estimate of the probability distribution governing the outcomes. We are mostly interested in the timing of subjects' decisions, the decision being a binary choice of a single purchase or sale of a security within a finite time sequence based on acquired information. Our data points to some compelling insights into rationality of Bayesian updating. Majority of our subjects display a type of irrational impatience: in tasks where they should optimally learn as much information as possible and wait until the last period to decide, they make their decisions too quickly, incurring excessive decision costs. This happens even when subjects can observe others' choices at no (explicit) cost whatsoever. This finding contrasts with a setting where explicit delay costs are incorporated.