Shrinkage Estimation of High- Dimensional Factor Models with Structura[ lnstabitities

ln large-scate pane[ data modets with latent factors the number of factors and their toadings may change over time This paper proposes an adaptive group-LASSO estimator that consistentty determines the numbers of pre- and post-break factors and the stabitity of factor loadings The data-dependent LASSO penatty is customized to account for unobserved factors and an unknown break date. A nove[ feature of our estimator is its robustness to unknown break dates Existing procedures either overestimate the number of factors by negtecting the breaks or require known break dates for a subsampte anatysis ln an empirical apptication, we study the change in factor [oadings and the emergence of new factors during the Great Recession.

This event is organised by the Econometric Institute.
Twitter: @MetricsSeminars