We propose a Bayesian-averaging portfolio choice strategy with excellent out-of-sample performance. Every period a new model is born that assumes means and covariances are constant over time. Each period we estimate model parameters, update model probabilities, and compute robust portfolio choices by taking into account model uncertainty, parameter uncertainty, and non-stationarity. The portfolio choices achieve higher out-of-sample Sharpe ratios and certainty equivalents than rolling window schemes, the 1/N approach, and other leading strategies do on a majority of 24 datasets.
展开▼
