A higher-order correct fast moving-average bootstrap for dependent data

We develop the theory of a novel fast bootstrap for dependent data. Our scheme deploys i.i.d. resampling smoothed moment indicators. characterize class parametric and semiparametric estimation problems which method is valid. show asymptotic refinements new procedure, proving that it higher-order correct under mild assumptions on time series, estimating functions, smoothing kernel. illustrate applicability advantages our procedure M-estimation, generalized moments, empirical likelihood estimation. In Monte Carlo study, we consider an autoregressive conditional duration model compare with other extant, routinely-applied first- methods. The results provide numerical evidence yields accurate confidence intervals, while remaining computationally lighter than its competitors. A real-data example dynamics trading volume US stocks illustrates relevance method.