【Brownbag】Belief Updating Under an Ambiguous and Asymmetric Information Structure: An Experimental Study
2025/11/14
研討會日期 : 2025-11-14
時間 : 12:10
主講人 : Professor Joseph Tao-yi Wang (王道一)
地點 : Conference Room B110
演講者簡介 : Professor Joseph Tao-yi Wang received his Ph.D. in Economics from University of California, Los Angeles in 2005. He is currently a Distinguished Professor at National Taiwan University. His research fields are Economic Theory, Experimental Economics, and Neuroeconomics.
演講摘要 : The spread of fabricated information, or "fake news," and its impact have recently been a central issue in social sciences research. We conduct an experiment that studies individual belief-updating in the presence of ambiguity in information sources where a decision maker knows neither if an information is credible nor the likelihood of it being credible. In particular, the signal is potentially informative but has an unknown chance of coming from a fixed source irrespective of the true state. We introduce non-ambiguous tasks as control in which the chance of having a credible signal is known. In such tasks, subjects on average under-react to signals compare to the Bayesian benchmark. In addition, we find a slight difference between updating good and bad signals. By investigating individual behavior, we find that the majority of subjects follow a cursed-updating heuristic, instead of Bayesian updating. The cursed-updating heuristic first updates under each signal source separately, and then employs the prior probability of each source (instead of the correct posterior) to take the weighted average to form posterior beliefs. Such updating is "cursed" since Bayesian updating should employ the conditional probability of each source given the realized signal to form posterior beliefs. When subjects follow such heuristic, the posterior belief is influenced by a fixed amount irrespective of the characteristic of "fake news." Interestingly, the ambiguous tasks exhibit similar behavior as the non-ambiguous ones, hinting widespread cursed-updating.
