AI 中文总结
研究定价代理与购买代理对匹配价值估计不同时的动态定价，考虑交易结果更新信念及归因提取消耗关系资本等情况，分析公共学习、卖家利润、提取率等影响，指出交易有信息外部性，还区分多种要素并确定测试所需记录。
AI 中文摘要
交易不仅能分配产品，还能创造消费前不可得的信息。我们研究在定价代理和购买代理对匹配价值估计不同、交易结果更新公共信念且归因提取会消耗关系资本的情况下的动态定价。公共学习使交换两极分化：更高的精度促使兼容的委托政策倾向于交易，不兼容的政策倾向于拒绝。卖家的精确预期利润必须考虑到接受会选择其后验边际这一事实。其最优政策因此跟踪公共边际、信任距离和后验精度。更高的提取率会降低当前接受率和未来结果信号的到来，产生一个尊重参考的区域和一个提取学习陷阱。信号精度不相等时，交易规则还会选择共同质量；一个对选择不敏感的学习者可能会变得更自信但更不准确。交易为后来的参与者创造了信息外部性，而共享的错误指定模型会使代理在未正确的情况下达成一致。分析区分了数据生成、贝叶斯知识、关系资本和经过验证的计算，并确定了测试这些机制所需的状态和协议记录。
英文摘要
A transaction can allocate a product and create information unavailable before consumption. We study dynamic pricing when a pricing agent and a buying agent hold different estimates of match value, traded outcomes update a public belief, and attributed extraction depreciates relationship capital. Public learning polarizes exchange: greater precision drives compatible delegated policies toward trade and incompatible policies toward rejection. The seller's exact expected profit must account for the fact that acceptance selects its posterior margin. Its optimal policy therefore tracks the public margin, trust distance, and posterior precision. Higher extraction reduces both current acceptance and the arrival of future outcome signals, producing a reference-respecting region and an extraction-learning trap. With unequal signal precision, the trading rule additionally selects common quality; a selection-naive learner can become more confident and less accurate. Transactions create an information externality for later participants, while a shared misspecified model can make the agents agree without becoming correct. The analysis distinguishes data production, Bayesian knowledge, relationship capital, and verified computation, and identifies the state and protocol records required to test the mechanisms.
Comments37 pages, 4 figures; 23-page online appendix included
