Topic: Contributions on computational methods for regression models Title: Least squares estimation of large dimensional threshold factor models Authors:  Daniele Massacci - Einaudi Institute for Economics and Finance (Italy) [presenting]
Abstract: Large dimensional factor models are estimated under the maintained assumption that the factor loadings do not change over time. The aim is to study least squares estimation of large dimensional factor models subject to regime shifts in the loadings parameterized according to the threshold principle. We propose to estimate the unknown threshold value by concentrated least squares, and factors and loadings by principal components. The estimator for the threshold value is superconsistent, with convergence rate that depends on both the times series and the cross-sectional dimensions of the available panel, and it does not affect the estimators for factors and loadings: these have the same convergence rate they would have if the threshold was known. We further propose model selection criteria robust to the threshold effect. Empirical application of the model documents an increase in connectedness in financial markets during periods of high economic policy uncertainty.