A0780
Title: Time-series GMM estimation with missing observations: A comparison of two alternative models
Authors: Masayuki Hirukawa - Ryukoku University (Japan) [presenting]
Abstract: The focus is on the two-step, efficient generalized method of moments (GMM) estimation of over-identified moment condition models using time-series data in which some observations are missing. The missing data problem arises for both high- (e.g., weekends and holidays for daily data) and low-frequency series (e.g., durations of two world wars for long-term annual or quarterly data). We investigate two alternative moment-condition models that can accommodate missingness, namely, the amplitude modulated and equal spacing models. The former assigns zeros in the place of missing observations, whereas the latter simply ignores missing observations and treats the observed data as if they were equally spaced in chronological order. We explore both large- and finite-sample properties of their corresponding efficient GMM estimators when the inverse of a kernel-smoothed heteroskedasticity and autocorrelation consistent (HAC) estimator is employed as the optimal weighting matrix in the second step. Moreover, a bandwidth selection method for kernel HAC estimation is proposed from the viewpoint of estimation optimality.
