Title: A Synthetic Business Cycle Approach to Counterfactual Analysis with Nonstationary Macroeconomic Data
Speaker: 奚晋（中国科学院数学与系统科学研究院）
Time: Thursday, October 15, 2026, 14:00-15:00
Venue: Gewu Building, Room 315
Abstract:
This paper develops a synthetic business cycle approach to counterfactual analysis with nonstationary macroeconomic data. We show that noncointegrated stochastic trends can generate large spurious treatment-effect estimates even when the true treatment effect is zero, despite constraints on the synthetic weights. To address this problem, we separate trends from cycles using the Hamilton filter, forecast the treated unit’s trend from its own history, and construct its cyclical counterfactual from donor cycles. The framework accommodates heterogeneous stochastic and deterministic trends without requiring prior knowledge of the nature of nonstationarity. With a fixed donor pool, we establish consistency and convergence rates for the synthetic-cycle weights and show that treatment-effect estimation error consists of a mean-zero stationary prediction error plus a vanishing estimation component. We establish asymptotically valid conformal tests for the treatment-effect path and for each post-treatment period. As an empirical illustration, we examine the effect of German reunification on West Germany’s GDP, demonstrating the advantages of the proposed approach for counterfactual analysis with nonstationary macroeconomic data.
About the Speaker：奚晋，现任中国科学院数学与系统科学研究院预测科学研究中心助理研究员。本科毕业于北卡罗来纳大学教堂山分校，获经济学与数学双学士学位。2024年获加利福尼亚大学圣迭戈分校经济学博士学位。主要研究方向包括机器学习、非平稳时间序列、人工智能与经济学交叉研究及宏观经济预测。研究成果发表于Journal of Econometrics、Journal of Business & Economic Statistics、Social Choice and Welfare。2026年获国家自然科学基金（C类）资助。曾任美国伊利诺伊大学厄巴纳香槟分校经济系访问学者。
