SEMINAR OF THE THEMATIC PROJECT

Seminar of the Thematic project (2023/01728-0) Econometric Modelling and Forecasting in High-Dimensional Models (HDEN&For)

Speaker:  Danielle Bianchi (Queen Mary College, UK)

Title: Dynamic Variable selection in high-dimensional predictive regressions.

Abstract: We develop an approximate inference method for dynamic variable selection in high-dimensional regression models with time-varying parameters. Specifically, we propose a variational Bayes algorithm that features dynamic sparsity-inducing properties to identify subsets of “active” predictors over time. An extensive simulation study shows that our approach produces more accurate variable selection than established static and dynamic sparse regression methods. The simulation results also highlight that our approach has a significant computational advantage compared to an equivalent MCMC algorithm while retaining a similar variable selection accuracy. We empirically test the performance of our approach within the context of an important problem for policymakers: inflation forecasting based on aggregate economic conditions. The in-sample analysis uncovers patterns in the dynamic relationship between macroeconomic conditions and future inflation rates broadly consistent with economic theory. This leads to significant gains in out-of-sample point and density forecasting compared to popular sparse regression approaches. Our results highlight the importance of a dynamic approach towards variable selection for time-series modelling and forecasting.

Type: Online(via Zoom)