A chi-square statistic is constructed that compares variance ratios and mean simple returns from data with those implied by an asset pricing model. The statistic is applied to the Consumption based Capital Asset Pricing Model with time non-separable preferences. It favors habit persistence for annual data, time-separability for quarterly data, and durability for monthly data, respectively. Introduction of time non-separability yields only a marginal improvement. The power of the test is high when alternative hypotheses are formed by varying the relative risk aversion coefficient. It is lower for alternative hypotheses generated by varying the time non-separability parameter, especially for durability.