Can VAR models capture regime shifts in asset returns? A long-horizon strategic asset allocation perspective

We examine whether simple VARs can produce empirical portfolio rules similar to those obtained under a range of multivariate Markov switching models, by studying the effects of expanding both the order of the VAR and the number/selection of predictor variables included. In a typical stock-bond strategic asset allocation problem on US data, we compute the out-of-sample certainty equivalent returns for a wide range of VARs and compare these measures of performance with those typical of non-linear models that account for bull-bear dynamics and characterize the differences in the implied hedging demands for a long-horizon investor with constant relative risk aversion preferences. In a horse race in which models are not considered in their individuality but instead as an overall class, we find that a power utility investor with a constant coefficient of relative risk aversion of 5 and 5-year horizon, would be ready to pay as much as 8.1% in real terms to be allowed to select models from the MS class, while analogous calculation for the whole class of expanding window VAR leads to a disappointing 0.3% per annum. We conclude that most (if not all) VARs cannot produce portfolio rules, hedging demands, or out-of-sample performances that approximate those obtained from equally simple non-linear frameworks.