Forecast reconciliation with subset selection


Several reconciliation methods are developed to adjust base forecasts in a hierarchical framework to achieve coherence. In practice, it is common that the hierarchy of interest is large (i.e., high-dimensional), and the base forecasts of some series within the hierarchy are of poor quality due to intrinsic characteristics of the series (e.g., intermittence) or model misspecification. In this talk, I will formulate a reconciliation method with subset selection that eliminates the negative effects of some inferior base forecasts on the performance of the reconciled forecasts. Our method is expected to yield coherent forecasts that are comparable to or even better than those produced by state-of-the-art reconciliation methods, with fewer nodes being used.

Sep 7, 2023 2:15 PM — 3:00 PM
Monash University Prato Centre, Italy
Xiaoqian Wang
Research Fellow

My research interests include time series analysis and statistical computing.