Abstract

In financial time series analysis, serial correlations and the volatility clustering effects of asset returns are commonly checked by Ljung-Box and MCleod-Li Q test and filtered by ARMA models. However, it is known that the both tests are not robust to heavily tailed data. We propose a new diagnostic test for linear time series model in the presence of additive outliers using a generalized serial correlations statistics. This test (TRA test) is analogous of Portmanteau test and is based on the truncated autocovariance estimate (TRA-estimate). It is designed to increase the resistance to aberrant residuals by contaminated observations. The asymptotic of the proposed statistic is derived. A simulation study shows that the TRA test have good robustness properties over the Ljung-Box and Li tests.