Evaluating measures of dependence for linearly generated nonlinear time series along with spurious correlation

The issue of determining dependence between two series is typically one of the most important aspects in any quantitative analysis. This study, using a Monte Carlo analysis, investigates the performance of several dependence measures for linearly generated nonlinear time series based on the family of AR(1) - ARCH(1) in variable models presented by Bera et al. (1992 and 1996) and it finds that copulas capture the concept of dependence better than the correlation coefficient. In addition, this study examines the performance of the test for zero association and it discovers that the spurious behavior can be eliminated asymptotically for this type on nonlinear processes, although the power of the test remains relatively low.