Generalized variance ratio tests in the presence of statistical dependence

We develop extensions of the variance-ratio statistic for testing the hypothesis a time series is uncorrelated and investigate their finite-sample performance. The tests employ an estimator of the asymptotic covariance matrix of the sample autocorrelations that is consistent under the null for general classes of innovations including EGARCH and non-MDS processes. Monte Carlo experiments show that our tests have better finite-sample size and power properties than the standard variance-ratio tests in experiments using time series generated by EGARCH and non-MDS processes