Non-Parametric Data Dependent Bootstrap for Conditional Moment Model

A new non-parametric bootstrap is introduced for dependent data. The bootstrap is based on a weighted empirical-likelihood estimate of the one-step-ahead conditional distribution, imposing the conditional moment restrictions implied by the model. This is the first dependent-data bootstrap procedure which imposes conditional moment restrictions on a bootstrap distribution. The method can be applied to form confidence intervals and p-values from hypothesis tests in Generalized Method of Moments estimation The bootstrap method is illustrated with an application to autoregressive models with martingale difference errors.