Last update: March 10th, 1998
| 8. | Name: | Rick Ashley | |
| Organiz.: | Economics Department, V.P.I. | ||
| JELC: | C52 | ||
| Keys: | nonlinearity bispectrum bicovariance BDS | ||
| Title: | A DIRECT COMPARISON OF THE BDS, HINICH, AND OTHER TESTS FOR THE PRESENCE OF NONLINEAR DEPENDENCE IN TIME SERIES | ||
| We present a comprehensive comparison of the popular statistical tests used to detect nonlinear dependence in time series data. These tests include the BDS, Hinich bispectrum/bicovariance, Engle LM, and McLeod-Li tests. Although each of these tests is widely applied, its actual size and power under various combinations of sample length and alternative hypotheses is unknown. The results presented here are unique because of the wide variety of null and alternative processes considered and also because of the breadth of the study in terms of the tests considered. The tests are implemented, using the bootstrap where appropriate, in software available from the authors. Conceptually, these tests are quite distinct: the BDS test is a broad-based test for serial dependence, whereas the Hinich bicovariance test and the Engle/McLeod-Li tests focus on selected third and fourth moments, respectively. Our joint implementation of all four tests allows us to go beyond simple power comparisons and examine the potential for complementarities between these tests across the different alternative generating processes. | |||