报告人：周前坤 教授（美国路易斯安那州立大学）

时间：2026年6月1日 15：00 -16：30

地点：文昌校区第二教学楼306室

报告摘要： This paper proposes a novel matrix completion framework for constructing valid prediction intervals for missing entries in high-dimensional panel data. In contrast to existing methods that primarily focus on mean imputation under strong factor assumptions, our approach exploits a quantile factor structure, offering robustness to both weak factors and heavy-tailed distributions. The proposed methodology proceeds by first extracting unobserved common factors from the fully observed data block via quantile factorysis (QFA), followed by estimating quantile-dependent factor loadings using smoothed quantile regression (SQR). A key theoretical contribution is the establishment of the asymptotic validity of the resulting prediction intervals under a weak factor structure, showing that inference remains reliable even when factor signals decay asymptotically. We also apply this general framework to program evaluation. By imputing unobserved counterfactual quantiles, the approach provides a unified and flexible framework for valid inference on individual treatment effects (ITE). Monte Carlo simulations and empirical applications demonstrate that the proposed method achieves reliable coverage accuracy and improved robustness compared to conventional approaches.

报告人简介：周前坤，现任美国路易斯安那州立大学经济学教授和Freeport- McMoran冠名教授，分别在北京大学和美国南加州大学获得硕士和博士学位。主要研究方向包括面板数据模型、非参数半参数计量模型、金融计量经济学、大数据分析和处理效应评估等。在Journal of Econometrics，Journal of Business and Economic Statistics, Journal of Applied Econometrics, Econometric Theory等国际权威期刊发表高水平论文30多篇，同时担任上述期刊的匿名审稿人。