Abstract: The aim of this paper is to determine whether forward-looking option- implied returns forecasts lead to better out-of-sample portfolio performance than conventional time series models. We consider a simple two-asset setting with a risk-free asset and the S&P 500 index the risky asset with monthly allocation revisions. We carry out a comprehensive analysis with a wide range of time-series models, two risk-neutral density inference methods, two utility functions, and several performance metrics. Portfolios are compared over the period of January 1994 to April 2010. Our main contribution is to compare the merits of implied volatility smoothing and maximum entropy risk-neutral density estimation techniques. By using bid/ask quotes in place of the closing prices, we obtain smooth probability densities using the maximum entropy principle that outperform the probability densities obtained using the implied volatility smoothing method. We also identify which moments of the option-implied probability densities contribute most to portfolio performance.