|
主讲人 |
Haiqi Li |
简介 |
<p>This study develops a novel framework for estimating common parameters and latent group heterogeneity in panel quantile regression models, where a subset of coefficients share common values across unknown groups. To address challenges from non-differentiable objective functions, we propose a Smoothed Panel Quantile Regression with a Mixed Group Structure (SPQR-MGS) and introduce an innovative penalized K-means method for simultaneous group identification and parameter estimation. Theoretically, this approach consistently recovers the true group structure. While the post-selection estimator exhibits asymptotic normality, we uncover substantial asymptotic bias and consequently develop two bias-correction approaches---analytical and half-panel jackknife---that yield asymptotically unbiased, zero-mean normal distributions. We further extend the methodology to accommodate multi-dimensional group structures. Monte Carlo simulations confirm superior finite-sample performance against benchmark methods. An empirical application to stock return predictability validates the practical utility of our SPQR-MGS framework and penalized K-means in financial econometrics.</p> |
