Anmerkungen:
Nach Informationen von SSRN wurde die ursprüngliche Fassung des Dokuments June 6, 2021 erstellt
Beschreibung:
In this article, we link the realized accuracy of predictive panels to changes in distributions that occur between the training (in-sample) phase and the testing (out-of-sample) phase. We obtain polynomial upper bounds for the loss of accuracy between training and testing. We model covariate shift and concept drift as positive autoregressive processes and predict future variations in changes (improvement versus deterioration in accuracy). Based on two different datasets, our empirical results show that our indicators for distribution turbulence are contrarian, but have a strong explanatory power over realized portfolio returns