2025 SIMIS-CDEF Seminar Series: No. 4

时间： October 27^th, 2025 (Monday), 2.00-4.00 pm (Beijing Time)

Venue: R910, Level 9, Block A, International Innovation Plaza, No. 657 Songhu Road, Yangpu District, Shanghai

Organizer: SIMIS Center for Digital Economy and Finance, SIMIS-CDEF

Chair: Qingfu Liu (刘庆富), Fudan University & SIMIS

Zoom会议号： 861 9872 2645 Passcode: 上海数学与交叉学科研究院

---

Talk One: Beyond First-Order Bias in Predictive Regressions: Estimation, Inference, and Return Predictability

Authors: Liang Jiang, Ron Kaniel, Yayi Yan, Jun Yu

Speaker: Yayi Yan (严雅毅), Shanghai University of Finance and Economics

摘要： This paper exposes significant higher-order bias in predictive regressions and its critical distortion of inferences about return predictability. While conventional methods address first-order bias, we show that higher-order biases, especially for persistent predictors and long predictive horizons in small samples, severely distort both statistical and economic conclusions. To resolve this, we introduce iBoot, a unified framework that combines indirect inference and bootstrap techniques to correct first- and higher-order biases simultaneously in short- and long-horizon regressions. Simulations demonstrate iBoot’s advantages: maximal bias reduction, accurate confidence interval coverage, exact size control, and high power. Applied to canonical stock return predictors, iBoot rejects perceived significance for the dividend-price ratio, book-to-market ratio, and term spread, while confirming the cross-sectional premium as a robust monthly predictor. These findings necessitate addressing higher-order bias in predictive modeling and equipping researchers with a reliable toolkit for disentangling statistical artifacts from genuine economic relationships.

---

---

Talk Two: Quantile Treatment Effects under Local Interference

Authors: Zequn Jin, Gaoqian Xu, Zixin Yang

Speaker: Zequn Jin (金泽群), Shanghai University of Finance and Economics

摘要： We study quantile treatment effects (QTE) under network interference. For units sharing the same degree, our estimands compare outcome quantiles across exposure configurations that differ in own treatment and the number of treated neighbors. Under a local-interference condition and mild restrictions on the degree distribution, these QTEs are point identified. We propose two estimators: (i) a degree-stratified nonparametric CQTE estimator, and (ii) a linear quantile-regression estimator. We establish consistency and weak convergence for both procedures and develop dependence-robust inference based on weighted and multiplier bootstrap. Monte Carlo experiments document small finite-sample bias and near-nominal coverage across a range of network designs. In an application to a randomized input-subsidy program in Mozambique, we uncover substantial distributional heterogeneity in both direct and spillover effects.

---

---