Nonparametric lag selection for time series models (Q2744945)

The paper is focused on applications of a nonparametric version of the final prediction error (FPE) for the lag selection in nonlinear autoregressive time series under general conditions including heteroskedasticity. First of all the authors prove consistency of FPE based lag selection in the presence of heteroskedasticity and derive the probabilities of incorrect lag selection for nonparametric FPE criteria. Based on these probabilities they conclude that overfitting is more likely than underfitting and suggest a correction factor to reduce overfitting. The other important contribution is an asymptotic formula for the nonparametric FPE as a function of the bandwidth and closed formula for the optimal bandwidth which minimizes the FPE.NEWLINENEWLINENEWLINEFor FPE calculation a local linear estimator is introduced in addition to the Nadaraya-Watson estimator in oder to cover a broad class of processes. To achieve faster computation, a plug-in bandwidth is proposed for the local linear estimator. The report of a Monte-Carlo study demonstrated that the correction factor generally improves the probability of correct lag selection for both linear and nonlinear processes and that the plug-in bandwidth works at least as well as its commonly used competitors. The proposed methods were applied to the analysis of two real data examples: Canadian lynx data and daily returns of DM/US-Dollar exchange rates.