A2043
Title: Bias-corrected method of moments for short-T dynamic panels with latent groups
Authors: Muhammad Afnan Arif - Asia-Europe Institute, Universiti Malaya, Kuala Lumpur, Malaysia (Malaysia) [presenting]
Abstract: Dynamic panel data models with short time dimension face the incidental parameters bias, rendering within-groups estimators inconsistent. Existing bias-corrected method of moments estimators address this problem but impose homogeneous slope coefficients, an assumption violated when units belong to latent groups with heterogeneous dynamics. A bias-corrected method of moments estimator is proposed for short-T dynamic panels where slopes are common within groups but heterogeneous across groups, with group membership unobserved. Within-group bias-corrected moment conditions are derived by evaluating the bias function at the group-specific autoregressive parameter, and pooling across groups yields misspecified conditions. A classification rule assigns units to the group whose bias-corrected moment function is closest to zero in a quadratic norm, and group-specific parameters are estimated by minimizing the within-group objective. Classification consistency, post-classification consistency, and asymptotic normality are established under fixed time dimension and large cross-section. A consistent information criterion for selecting the number of groups is developed. Simulation evidence shows that the proposed estimator recovers the latent group structure and outperforms pooled bias-corrected estimators.
