Model-based bias correction for short AR(1) and AR(2) processes

Abstract

The class of autoregressive (AR) processes is extensively used to model temporal dependence in observed time series. Such models are easily available and routinely fitted using freely available statistical software like R. A potential caveat in analyzing short time series is that commonly applied estimators for the coefficients of AR processes are severely biased. This paper suggests a model-based approach for bias correction of well-known estimators for the coefficients of first and second-order stationary AR processes, taking the sampling distribution of the original estimator into account. This is achieved by modeling the relationship between the true and estimated AR coefficients using weighted orthogonal polynomial regression, fitted to a huge number of simulations. The finite-sample distributions of the new estimators are approximated using transformations of skew-normal densities and their properties are demonstrated by simulations and in the analysis of a real ecological data set. The new estimators are easily available in our accompanying R-package ARbiascorrect for time series of length n = 10, 11, ... , 50, where original estimates are found using exact or conditional maximum likelihood, Burg's method or the Yule-Walker equations.