A0315
Title: Time-varying coefficient regression models: Estimation and prediction by local linear smoothers using asymmetric kernels
Authors: Masayuki Hirukawa - Ryukoku University (Japan) [presenting]
Abstract: Trending time-varying coefficient regression models are investigated. Time-varying coefficients are estimated nonparametrically by local linear ("LL") regression smoothing. Because the domain of varying coefficients is [0,1], nonstandard, asymmetric kernels ("AKs") that are free of boundary bias are employed. Among all such kernels, a particular focus is on the beta and gamma kernels due to their popularity in empirical studies in economics and finance. Convergence properties of AK-LL estimators for varying coefficients at a fixed design point and in the vicinity of 1, including their bias and variance approximations and asymptotic normality, are explored. Their finite-sample properties, as well as the effectiveness of an implementation method, are examined via Monte Carlo simulations. Implications for prediction are also considered, in combination with long-run variance estimation for asymptotic variances of AK-LL estimators.
