Bootstrapping Nonlinear Least Squares Estimates in the Kalman Filter Model.
PITTSBURGH UNIV PA CENTER FOR MULTIVARIATE ANALYSIS
Pagination or Media Count:
The bootstrap is proposed as a method for estimating the precision of forecasts and maximum likelihood estimates of the transition parameters of the Kalman filter model when the estimates are obtained via Newton-Raphson. It is shown that when the system and the filter are in steady state, the bootstrap applied to the Gaussian innovations yields asymptotically consistant standard errors. That the boot strap works well with moderate sample sizes and supplies robustness against departures from normality is substantiated by emperical evidence. Keywords Parameter estimation. Author
- Statistics and Probability