Robust portfolio rules and detection-error probabilities for a mean-reverting risk premium

I analyze the optimal intertemporal portfolio problem of an investor who worries about model misspecification and insists on robust decision rules when facing a mean-reverting risk premium. The desire for robustness lowers the total equity share, but increases the proportion of the intertemporal hedging demand. I present a methodology for calculation of detection-error probabilities, which is based on Fourier inversion of the conditional characteristic functions of the Radon-Nikodym derivatives. The quantitative effect of robustness is more modest than in i.i.d. settings, because model discrimination between the benchmark and the worst-case alternative model is easier, as indicated by the detection-error probabilities.