A0234
Title: Noise cancelling observation-driven models
Authors: Jannik Steenbergen - Aarhus University (Denmark) [presenting]
Leopoldo Catania - Aarhus BBS (Denmark)
Abstract: A new class of observation-driven models are proposed with a filter that downweighs the forcing variable when the model error term is below a certain threshold in absolute magnitude. The new updating mechanism relies on the premise that error terms of low absolute magnitude generally constitute noise and do not indicate a change in the time-varying parameter. The asymptotic properties of the maximum likelihood estimator of static model parameters are established, and the usefulness of the new noise-cancelling updating mechanism is demonstrated in a US inflation forecasting study. Results suggest that the new updating mechanism can improve the accuracy of forecasts compared to standard observation-driven counterparts.
