Estimation of volatility causality in structural autoregressions with heteroskedasticity using independent component analysis

Abstract

In this study, we developed a new statistical procedure to identify and estimate both the causal effects and the volatility structure in multivariate time series using structural autoregressions with heteroskedasticity (SVAR–GARCH) model. Specifically, using a recent method based on independent component analysis (ICA), we could simultaneously identify the causality and decompose the error terms into non-Gaussian independent components. We then used univariate GARCH to model the volatility of each independent component. The structure of volatility transmission along the SVAR model then was modeled using volatility impulse response function (VIRF) to independent component shocks. Simulation results were given to verify the effectiveness of the procedure. Empirically, we investigated the intraday volatility transmission effect along the US Treasury curve, based on high-frequency CME bond futures data. The result validated the general belief that intraday volatility transmission is a function of liquidity and maturity along the yield curve.