A0388
Title: Bias-corrections for correlations and heteroskedasticities in large linear panel models with interactive effects
Authors: Runyu Dai - Tohoku University (Japan) [presenting]
Yasumasa Matsuda - Tohoku University (Japan)
Takashi Yamagata - University of York (United Kingdom)
Abstract: An efficient iterative principal components (IPC) estimator of a large linear panel data model is considered with common factor type interactive effects. It is well-known that the original IPC estimator suffers from bias due to correlated and heteroskedastic idiosyncratic errors in cross-sectional and serial dimensions. The developed estimator corrects the bias by a residual sparse regression to correct correlations in both dimensions simultaneously, plus a conventional bias correction for heteroskedasticities. The asymptotic properties of the proposed estimator are rigorously established. Monte Carlo simulations show the approach works well in finite samples both in estimation and inference.
