excerpt: 'This paper investigates the impact of structural breaks in the factor structure on factor-augmented forecasting. We decompose the break in the factor loading matrix into rotational and shift components. To effectively utilize the pre-break data and maintain robustness against shift breaks, we propose a novel factor estimator that minimizes the L2 distance between pre- and post-break loading matrices through the rotation of factor estimates. We call this estimator the “rotated factors” and analyze its the asymptotic properties, along with two competing factor estimators, in the presence of different types of breaks. To leverage the respective advantages of each factor estimator in an automatic data driven way, we introduce a method that averages over sets of factor estimates using a leave-h-out cross- validation criterion. Simulations demonstrate that combining different factor estimates through the proposed cross-validation averaging approach leads to improved forecasting performance compared to existing methods. Furthermore, we evaluate the effectiveness of our methods in an empirical application with US macroeconomic data and emphasize the importance of incorporating structural breaks into factor-augmented forecasting models.'
0 commit comments