ECONOMETRICS: Robust Bias Adjustment in Potentially Misspecified Time Series Models; Professor Dennis Kristensen (University College London)

Abstract

We propose a class of bootstrap procedures for bias correction of a broad class of estimators. Compared to existing methods our proposal is computationally easier to implement. In particular, it does not require usage of any optimization algorithm. The procedure takes as input a resampling procedure as chosen by the researchers. The chosen procedure should reflect how much confidence the researchers have in the model being estimated. If the researcher suspects the model is misspecified, non- or semiparametric resampling methods should be used. If, on the other hand, the model is corretly specified, parametric bootstrap is preferred. We provide a theory of how each procedure comes with its own advantages in terms of biases and variances. A numerical study demonstrate these features in practice.