Robust Bayesian Model Selection for Autoregressive Processes with Additive Outliers

Abstract Autoregressive (AR) models of order k are often used for forecasting and control of time series, as well as for the estimation of functionals such as the spectrum. Here we propose a method that consists of calculating the posterior probabilities of the competing AR(k) models in a way that is robust to outliers, and then obtaining the predictive distributions of quantities of interest, such as future observations and the spectrum, as a weighted average of the predictive distributions conditional on each model. This method is based on the idea of robust Bayes factors, calculated by replacing the likelihood for the nominal model by a robust likelihood. It draws on and synthesizes several recent research advances, namely robust filtering and the Laplace method for integrals, modified to take account of the finite range of the parameters. The method performs well in simulation experiments and on real and artificial data. Software is available from StatLib.

[1]  L. Wasserman,et al.  Computing Bayes Factors by Combining Simulation and Asymptotic Approximations , 1997 .

[2]  R. Douglas Martin,et al.  ROBUST METHODS FOR TIME SERIES , 1981 .

[3]  G. Schwarz.  Estimating the Dimension of a Model , 1978 .

[4]  B. Stuck.  Minimum error dispersion linear filtering of scalar symmetric stable processes , 1978 .

[5]  R. Tsay.  Time Series Model Specification in the Presence of Outliers , 1986 .

[6]  R. Martin,et al.  Bias Robust Estimation of Scale , 1993 .

[7]  H. Akaike,et al.  Information Theory and an Extension of the Maximum Likelihood Principle , 1973 .

[8]  L. Wasserman,et al.  A Reference Bayesian Test for Nested Hypotheses and its Relationship to the Schwarz Criterion , 1995 .

[9]  G. Anandalingam,et al.  Using instrumental variables for selecting the order of arma models , 1988 .

[10]  R. Martin.  Approximate conditional-mean type smoothers and interpolators , 1979, 1979 18th IEEE Conference on Decision and Control including the Symposium on Adaptive Processes.

[11]  R. Shibata.  Selection of the order of an autoregressive model by Akaike's information criterion , 1976 .

[12]  J. Richard,et al.  Specification Searches: Ad Hoc Inference with Nonexperimental Data , 1980 .

[13]  A. Harvey.  Time series models , 1983 .

[14]  G. Kitagawa.  Non-Gaussian State—Space Modeling of Nonstationary Time Series , 1987 .

[15]  J. G. De Gooijer,et al.  Methods for Determining the Order of an Autoregressive-Moving Average Process: A , 1985 .

[16]  H. Jeffreys,et al.  Theory of probability , 1896 .

[17]  M. West,et al.  Bayesian forecasting and dynamic models , 1989 .

[18]  A. Raftery,et al.  Modeling flat stretches, bursts, and outliers in time series using mixture transition distribution models , 1996 .

[19]  G. Kitagawa,et al.  Non-Gaussian state-space modeling of nonstationary time series. Comment. Rejoinder , 1987 .

[20]  Adrian E. Raftery,et al.  Comment: Robustness, Computation, and Non-Euclidean Models , 1987 .

[21]  D. Madigan,et al.  Model Selection and Accounting for Model Uncertainty in Graphical Models Using Occam's Window , 1994 .

[22]  O. Barndorff-Nielsen,et al.  On the parametrization of autoregressive models by partial autocorrelations , 1973 .

[23]  Edward E. Leamer,et al.  Specification Searches: Ad Hoc Inference with Nonexperimental Data , 1980 .

[24]  C. Masreliez.  Approximate non-Gaussian filtering with linear state and observation relations , 1975 .

[25]  L. Tierney,et al.  Accurate Approximations for Posterior Moments and Marginal Densities , 1986 .

[26]  Gwilym M. Jenkins,et al.  Time series analysis, forecasting and control , 1972 .

[27]  G. Tunnicliffe Wilson,et al.  Interpolating Time Series with Application to the Estimation of Holiday Effects on Electricity Demand , 1976 .

[28]  Fred L. Ramsey,et al.  Characterization of the Partial Autocorrelation Function , 1974 .