On the change point test for SVAR-GARCH models via ICA
Sangjo Lee, Sangyeol Lee · Communications for Statistical Applications and Methods · 2025
This paper addresses the issue of testing for parameter changes in structural vector autoregressive (SVAR) models with generalized autoregressive conditional heteroskedasticity (GARCH) errors, where the errors follow a linear combination of independent univariate GARCH processes.For change point detection, we first transform the SVAR model to a vector autoregressive (VAR) model and estimate its parameters to calculate the VAR residuals.Next, we apply independent component analysis (ICA) to decompose the VAR residuals into a mixing matrix and independent GARCH components.These components are then fitted with univariate GARCH models to obtain the GARCH residuals.Using these VAR and GARCH residuals, we formulate the locationscale based cumulative sum (LSCUSUM) test to detect parameter changes.We confirm the robust performance of the LSCUSUM test through Monte Carlo simulations and demonstrate the practicality of our method with a real-world data analysis using the exchange rates of Asian currencies.