Abstract
Results are reported of a forecasting competition between linear autoregressive, GARCH, threshold autoregressive and neural network models of the UK monthly unemployment rate series. Evidence is uncovered suggesting the existence of nonlinearities in the series.