A Nonlinear Time Series Workshop

The analysis ofwhat might be called "dynamic nonlinearity" in time series has its roots in the pioneering work ofBrillinger (1965) - who first pointed out how the bispectrum and higher order polyspectra could, in principle, be used to test for nonlinear serial dependence - and in Subba Rao and Gabr (1980) and Hinich (1982) who each showed how Brillinger's insight could be translated into a statistical test. Hinich's test, because ittakes advantage ofthe large sample statisticalpropertiesofthe bispectral estimates became the first usable statistical test for nonlinear serial dependence. We are forever grateful to Mel Hinich for getting us involved at that time in this fascinating and fruitful endeavor. With help from Mel (sometimes as amentor,sometimes as acollaborator) we developed and applied this bispectral test in the ensuing period. The first application ofthe test was to daily stock returns {Hinich and Patterson (1982, 1985)} yielding the important discovery of substantial nonlinear serial dependence in returns, over and above the weak linear serial dependence that had been previously observed. The original manuscript met with resistance from finance journals, no doubt because finance academics were reluctant to recognize the importance of distinguishing between serial correlation and nonlinear serial dependence. In Ashley, Patterson and Hinich (1986) we examined the power and sizeofthe test in finite samples.

Inhalt

1. Nonlinearity in Stochastic Processes: What it Is and Why it Matters.- 2. Detecting Nonlinear Serial Dependence.- 3. How to Run the Toolkit Program on a PC.- 4. Artificially Generated Data: Size Considerations.- 5. Artificially Generated Data: Power And Model Specification Considerations.- 6. Analysis of Stock Market Returns.- 7. Glint Tracking Errors in Radar.- 8. Seismic Data.- 9. Analysis of U.S. Real GNP.- 10. Dynamic Structure of Macroeconomic Technology Shocks.- 11. Climatological Data.