讲座题目：Embedding Large-Language-Model-Distilled Surrogates into Structural Macroeconomic Modeling
主讲人：包特 南洋理工大学经济系 长聘副教授，研究生院助理院长
讲座时间：11月17日（周二）14:00
讲座地点：MBA教育中心A15
主持人：刘小元 中央财经大学商学院 教授
主讲人介绍：
包特，新加坡南洋理工大学经济系长聘副教授、研究生院助理院长。主要研究领域为行为金融学、计算经济学和中国经济。在Economic Journal, Management Science、经济研究、管理世界等国内外期刊发表论文50余篇，担任Journal of Economic Dynamics and Control等五家国际期刊的副主编或编委。
摘要：
Structural macroeconomic models explain aggregate outcomes through explicit mechanisms but often rely on restrictive behavioural assumptions for tractability, whereas machine learning can represent richer behavioural relationships without establishing their equilibrium consequences. Here we develop a framework that embeds learned behavioural functions within structural macroeconomic models while retaining their economic constraints, market interactions and policy rules. It elicits responses from three large language models, distils them into compact neural networks and embeds the resulting functions within structural models, allowing economic outcomes to shape subsequent responses. We demonstrate it through household inflation expectations in a heterogeneous-agent New Keynesian model. History-dependent beliefs generate endogenous output and inflation cycles without shocks. Following inflationary histories, monetary easing produces delayed total output peaks closer to observed timing, while initial fiscal responses and output responses to forward guidance are attenuated. These model-generated findings show how learned behavioural functions can relax restrictive assumptions while preserving explicit economic mechanisms.
审核：林嵩
