邹至庄青年学者交流工作坊第4期
论文题目：Discrete Belief Switching: A Non-Bayesian Theory of Inertia, Jumps, and Belief Biases
报告人：王泽宇
导师：蔡熙乾教授
摘要：
The paper provides an optimizing foundation for a class of non-Bayesian belief-updating rules in which agents hold a finite set of candidate hypotheses, stochastically recall one alternative each period, and commit to the recalled alternative only when its evidence gap exceeds a fixed threshold. This recall-then-commit architecture arises as the unique optimal policy under a hybrid information cost that pairs a Shannon flow cost on recall with a fixed cognitive cost on commitment. The resulting policy decomposes multiplicatively into a Luce-softmax recall stage and a hard-threshold commitment stage, with the log-likelihood-gap threshold admitting a closed-form expression in memory depth, discount factor, and the commitment cost. The framework delivers a two-wedge regret decomposition, a stability-adaptability trilemma analogous to the Lai-Robbins bandit bound, and six falsifiable predictions that discriminate the theory from Bayesian, rational-inattention, and adjustment-cost alternatives. It provides a unified optimizing account of belief inertia, discrete jumps, under- and over-reaction, recency bias, and hysteresis.
地点：经济楼D136
腾讯会议：586-6075-8211
时间：5月11日，周一中午，12:30-14:00
语言：中文
