A Non-Parametric Approach of Heteroskedasticity Robust Estimation of Vector-Autoregressive (VAR) Models

This contribution studies the application of heteroskedasticity robust estimation of Vector-Autoregressive (VAR) models. VAR models have become one of the most applied models for the analysis of multivariate time series. Econometric standard software usually provides parameter estimators that are not robust against unknown forms of heteroskedasticity. Different bootstrap methodologies are available which are able to generate heteroskedasticity robust parameter estimates. However, common literature is mostly focused on univariate time series models. This study applies a natural extension of the non-parametric pairs bootstrap methodology to different VAR models, taking into account empirical stock market data of the FTSE 100, DAX 30 and S&P 500. A comparison shows that the t-values of the bootstrap models’ parameters are considerably lower than the ordinary ones and that the determinants of the covariance matrices are clearly smaller.