A fast subsampling method for nonlinear dynamic models

We highlight a fast subsampling method that can be used to provide valid inference in nonlinear dynamic econometric models. This method is based on the subsampling theory proposed by Politis and Romano[A general theory for large sample confidence regions based on subsamples under minimal assumptions, Technical Report 399, Dept of Statistics, Stanford University; Large sample confidence regions based on subsamples under minimal assumptions. Annals of Statistics 22, 2031–2050]. Fast subsampling directly exploits score functions computed on each subsample and avoids recomputing the estimators for each of them. This method is used to approximate the limit distribution of estimators, possibly simulation based, that admit an asymptotic linear representation with both known and unknown rates of convergence. Monte Carlo experiments demonstrate the desirable performance and vast improvement in the numerical speed of the fast subsampling method.