Combined Forecasts from Linear and Nonlinear Time Series Models

Combined forecasts from a linear and a nonlinear model areinvestigated for timeseries with possibly nonlinear characteristics. The forecasts arecombined by aconstant coefficient regression method as well as a time varyingmethod. Thetime varying method allows for a locally (non)linear model. Themethods areapplied to data from two kinds of disciplines: the Canadian lynx andsunspotseries from the natural sciences, and Nelson-Plosser's U.S. seriesfromeconomics. It is shown that the combined forecasts perform well,especially withtime varying coefficients. This result holds for out of sampleperformance forthe sunspot and Canadian lynx number series, but it does notuniformly hold foreconomic time series.