A sensitivity analysis of threshold determination for asymmetric error correction models

The properties of the Granger-Lee (Journal of Applied Econometrics, 4, S145-59, 1989) asymmetric error correction model under consistent-threshold estimation are considered, with the relationship between the threshold range and rejection of the symmetry hypothesis examined. The results of Monte Carlo experimentation show rejection of symmetry to depend positively upon the size of the threshold range employed. However, even if consistent-threshold estimation is employed with a relatively small threshold range, the resulting test exhibits a substantial increase in power relative to the use of the Granger-Lee test is its original form using an imposed deterministic threshold. This increase in power and its dependence upon the size of the threshold range are further illustrated by application to the data of the seminal study of Granger and Lee.