Comparison of recursive parameter estimation and non-linear filtration

The contribution compares and contrasts the results of time-varying parameters estimation pursued by two methodologically different approaches, all in the context of a DSGE model. The original results [3] were obtained by the so-called recursive, rolling and "first observation" analysis. The novel methodology uses Unscented Particle Filter to filter trajectories of time-varying parameters. The comparisons are presented on the case of three parameters of a DSGE model, which demonstrate a match in the evolution of the parameter estimates. Recursive impulse response functions are compared on two cases, which also match. Although not all of the results obtained from two different methodologies completely match, the most important parameter changes are similarly captured by both approaches.