A comparative study of the finite-sample performance of some portmanteau tests for randomness of a time series

Testing for the randomness of a time series has been one of the most widely researched topics in time-series analysis. The present paper carries out a comparative study of the finite-sample performance of some well-known portmanteau tests in this area. Using Monte Carlo simulation experiments, we find that (i) the empirical sizes of some oft-used parametric portmanteau tests are severely undersized when the data generating process is skewed, (ii) the non-parametric portmanteau test possesses proper sizes only when the number of rank autocorrelations is chosen to be small relative to the sample size, (iii) the non-parametric portmanteau test is more powerful than the parametric portmanteau tests in the case of skewed distributions, and (iv) the choice of the number of sample autocorrelations (or rank autocorrelations) can significantly affect the size as well as the power of the tests considered.