A1411
Title: Bias correction in factor-augmented regression models
Authors: Peiyun Jiang - Tokyo Metropolitan University (Japan) [presenting]
Yoshimasa Uematsu - Hitotsubashi University (Japan)
Takashi Yamagata - University of York (United Kingdom)
Abstract: A new estimation procedure is introduced for factor-augmented regression models to eliminate the bias of the ordinary least squares (OLS) estimators. The asymptotic properties of the OLS estimators are first derived, allowing for weak factors. It is demonstrated that the bias inherent in the common estimation approach is driven by the dependence between latent factors and observed regressors, the divergence rates of the signal eigenvalues, and the choice of the rotation matrix. An interesting result is that the commonly used rotation matrix generates the largest bias in the factor-augmented regression estimators, whereas an alternative rotation matrix can reduce the bias to some extent. To address these issues, a novel estimation procedure is proposed that projects out the observed regressors during the principal component (PC) estimation step, thereby mitigating the correlation between latent factors and regressors. The procedure induces exactly zero bias under weaker conditions than those required by existing methods. Monte Carlo simulations confirm that while common approaches suffer from significant bias and severe size distortion in testing, the proposed method demonstrates satisfactory accuracy in estimation and optimal testing performance in terms of size and power.
