The finite sample power of long-horizon predictive tests in models with financial bubbles

Using finite sample simulation methods, we assess the power of long-horizon predictive tests and compare them to their short-run counterparts, when the true underlying model contains financial asset bubbles. Our results indicate that long-run predictive tests using valuation predictors – specifically the dividend price ratio – do pick up the in-sample return predictability inherent in the asset bubbles. However, after size-adjustment, the long-run predictive framework has little advantage over its short-run counterpart when the predictor is highly persistent, but can provide non-trivial, yet still modest, power improvements when the predictor is moderately persistent. Finally, we provide a brief intuitive explanation for why a model with temporary collapsing bubbles may yield in-sample predictive power without implying the existence of profitable out-of-sample trading strategies.