Nonlinearities in German unemployment rates: A nonparametric analysis

Within the class of seasonal nonlinear autoregressive models quarterly Ger-man unemployment rates are investigated with respect to the kind and structure of nonlinearities in the conditional mean function using recently developed non-parametric estimation and lag selection techniques. These methods include local constant and local linear estimators which are also used as building blocks for a nonparametric version of the FPE criterion. Furthermore, additive nonlinear autoregressive models are tted. There is substantial evidence of strong non-linearity in the seasonally unadjusted data which is found to be weakened and modiied if seasonal adjustment procedures are applied to the data rst. Independent of the speciication and treatment of seasonality, one observes that the impact of changes lagged one period on the current change in the unemployment rate turns from positive to negative once large positive changes occur.