A non-Bayesian, generalized least squares (GLS)-based approach is formally proposed to estimate a class of time-varying AR parameter models. This approach has partly been used by Ito et al. (2014, 2016a,b), and is proven very efficient because, unlike conventional methods, it does not require the Kalman filtering and smoothing procedures, but yields a smoothed estimate that is identical to the Kalman-smoothed estimate. Unlike the maximum likelihood estimator, the possibility of the pile-up problem is shown to be small. In addition, this approach enables us to possibly deal with stochastic volatility models and models with a time-dependent variance-covariance matrix.
↧