Purpose We will demonstrate the importance of taking into account \mean reversion in asset prices and will show that this type of modeling leads to a high share of equities in pension funds' asset allocations. Design/methodology/approach Firstly, we will study the long-run statistical characteristics of selected financial assets during the 1895-2011 period. Such an analysis corroborates the fact that, for long holding periods, equities exhibit lower risk than other asset classes. Moreover, we will provide empirical evidence that stock market returns are negatively skewed in the short term, and show that this negative skewness vanishes over longer time horizons. Both these characteristics favor the use of a semi-parametric methodology. Findings Our empirical study led us to two major findings. Firstly, we noticed that the distribution of stock returns is negatively skewed over short time horizons. Secondly, we observed that the fat-tailed shape of the returns distribution disappears for time periods longer than five years. Finally, we demonstrated that stock returns exhibit \mean-reversion\. Consequently the optimization program should not only take into account the non-Gaussian nature of returns in the short run, but also incorporate the speed at which volatility \mean reverts to its long-run mean. Originality/value To simulate portfolio allocation, we used a CF VaR criterion with the advantage of providing an allocation that is independent of the saver’s preferences parameters. A backtesting analysis including a calculation of replacement rates shows a clear dominance of the “non-Gaussian” strategy, as the retirement outcomes under such a strategy would be positively affected.