Jane Harvill

Department of Mathematics and Statistics

Mississippi State University

Starkville Mississippi


Multivariate Nonlinear Time Series Modeling

A multivariate extension of the univariate nonlinearity test of Tsay (1986) is presented. Simulation results show that the multivariate test is more powerful than its univariate counterpart, especially for series having nonlinear structure involving several components of the vector process and weakly or moderately cross-correlated process error terms.

Next, exploratory methods for determining appropriate lagged variables in a vector nonlinear time series model are investigated. The first is a multivariate extension of the R statistic from Granger and Lin (1994), which is based on an estimate of the mutual information criterion. The second method uses Kendall's tau and partial tau statistics for lag determination. These methods provide nonlinear analogues of the autocorrelation and partial autocorrelation matrices for a vector time series. Simulation results indicate that the methods reliably identify appropriate lags.

Finally, a brief discussion of work in progress model estimation techniq ues in the vector nonlinear time series case follows. Some non-parametric methods which avoid the "curse of dimensionality" are suggested as possible solutions to this problem.


Back to Colloquium Series