Identification of a time-varying model using the wavelet approach and the AR process

The aim of the paper is to give recommendation for working with time frequency modeling of macroeconomic time series on the basis of a comparative study. We investigated the wavelet analysis and the time-varying autoregressive process. We focused on two main areas - sample size for an available data set and its shortening and optimization of parameters of mentioned methods. In the case of the time-varying autoregressive process, we investigated optimization of parameters such as lag length, windowing and overlap. In the wavelet analysis approach, we investigated the type of wave and scale. Performance of the methods was presented on the gross domestic product data of USA, United Kingdom and Korea. These representatives were chosen from the perspective of available sample size and for the reason that the countries represent economy in different geographic areas. An advantage of the wavelet analysis is better time resolution. An autoregressive process provides better frequency resolution, but it is quite sensitive to sample size.