Momentum of return predictability

We find the momentum of predictability (MoP) that the forecasting performance of some univariate regressions is persistent. A univariate model which outperforms the benchmark during recent past period can also beat the benchmark in the near future out-of-sample. Accordingly, we propose a forecasting strategy that involves switching between a model of interest and the benchmark model, based on observations of their recent past performance. We obtain significant stock return predictability both in statistical and economic terms. Predictability is found to be stronger for longer forecasting horizons. Success of the MoP strategy is also seen in forecasting exchange rates.