Simulation analysis of the impact of volatility clustering upon the finite-sample distribution of threshold cointegration tests

Using Monte Carlo simulation, the finite-sample sizes of asymmetric cointegration tests are examined in the presence volatility clustering. The findings obtained show the asymmetric tests of Enders and Siklos (2001) to exhibit greater oversizing than the previously examined implicitly symmetric cointegration test of Engle and Granger (1987). Further, it is found that oversizing is driven by the size of the volatility parameter of the GARCH processes considered, rather than their degree of persistence. Interestingly, the application of consistent-threshold estimation is shown to increase the size distortion of the asymmetric tests, with the consistent-threshold MTAR test displaying the greatest size distortion of all tests considered.