A0702
Title: Nonpivotal Granger causality tests based on vector autoregressive models
Authors: Ying-Chao Hung - National Taiwan University (Taiwan) [presenting]
Abstract: Granger causality is a classical and important technique for measuring predictability from one group of time series to another by incorporating information on variables described by a vector autoregressive (VAR) model. Traditional methods for validating Granger causality are based on the Wald type tests, which may encounter the following implementation problems: (i) test statistic inflation due to singularity or near-singularity of the underlying covariance matrix; and (ii) infeasibility or huge computational cost for tuning parameter selection. An alternative procedure for testing Granger causality based on non-pivotal statistics is considered. The proposed testing method has a strong theoretical basis and does not require any calibration of tuning parameters. Further, an initial numerical investigation yields very competitive power values compared to the Wald-type tests. Finally, the purpose is to extend the proposed method and establish associated asymptotic theories for large orders (or infinite order) of vector autoregressive models.