This paper argues that inferring long-horizon asset-return predictability from the\ud properties of vector autoregressive (VAR) models on relatively short spans of data is potentially\ud unreliable. We illustrate the problems that can arise by re-examining the findings of Bekaert and\ud Hodrick (1992), who detected evidence of in-sample predictability in international equity and\ud foreign exchange markets using VAR methodology for a variety of countries over the period\ud 1981-1989. The VAR predictions are significantly biased in most out-of-sample forecasts and\ud are conclusively outperformed by a simple benchmark model at horizons of up to six months.\ud This remains true even after corrections for small sample bias and the introduction of Bayesian\ud parameter restrictions. A Monte Carlo analysis indicates that the data are unlikely to have been\ud generated by a stable VAR. This conclusion is supported by an examination of structural break\ud statistics. Implied long-horizon statistics calculated from the VAR parameter estimates are\ud shown to be very unreliable.