Dynamic Econometric Models Bayesian Analysis of Polish Inflation Rates Using Rca and Gll Models *

An extensive discussion of the empirical evidence of changes in the time series properties of inflation was provided in Cecchetti, Hooper, Kasman, Schoenholtz, and Watson (2007). In their paper they used an unobserved component model with stochastic volatility to characterize inflation and AR model with time varying coefficients and stochastic volatility to describe the growth of real GDP. These models were originally used by Stock and Watson (2007) and Nason (2006). Also Koop and Potter (2001) considered a time-varying parameter AR model where the coefficients evolve over time according to a random walk for quarterly change in the US CPI. All mention above authors found strong evidence of randomness of autoregressive parameters for inflation data. In our model-based analysis the mean of inflation is specified by a random coefficient autoregressive (RCA) or generalized linear (GLL) model. Unlike mentioned above papers, in our models the random parameters and the unobserved component follow stationary processes. Using monthly inflation data, our modelling framework and Bayesian estimation, we find remarkable changes in varying mean. The paper is organized as follows. Section 2 introduces the time-varying parameter (TVP) models and Bayesian estimation. Section 3 presents empirical results for Polish inflation. Section 4 concludes.