Title: Testing for changes in systematic correlation with approximate threshold group-factor models
Authors: Mirco Rubin - EDHEC - Nice (France) [presenting]
Daniele Massacci - Kings College London (United Kingdom)
Dario Ruzzi - University of Bristol (United Kingdom)
Abstract: The Approximate Linear Factor Model is extended in order to allow for different groups of individuals to be affected by both common and group-specific latent factors, with loadings switching over time as function of an exogenous variable. The model is inspired by the recent extensions of the classical approximate factor model allowing for either a) time-varying loadings with threshold-type regime switches or b) specific factors affecting only a finite number of groups. We propose inference procedures to detect i) the presence of switches in the loading structure, ii) the level of the exogenous variable determining the switches, iii) the number of factors which are across groups in each regime, as opposed to group-specific factors and iv) changes across regimes in the (average) systematic correlation among individuals in different groups due to the pervasive factors. Our measure of systematic correlation is related to the canonical correlations and the R-squared - via the loadings - of the estimated pervasive factors, and is used to test for changes in comovement between different groups across regimes.