The expected return and covariance matrix are commonly calculated on a calendar time scale (e.g. daily or monthly data). In this article, we assess the relevance of calculating them on a new time scale derived from traded volume. In particular, we evaluate portfolio optimizations where returns evolve on a data-based rather than calendar time scale. We empirically test the impact of this change of scale by comparing the performance of two well-known portfolio optimizations in an out-of-sample framework. We find that this change leads to gains in both risk-adjusted return and risk. We also find that the degree of deviation from the normal distribution (and independence) of returns is greater with returns calculated in calendar time than in data-based time, which explains the outperformance of this new approach.