An alleged weakness of heuristic optimisation methods is the stochastic character of their solutions. That is, instead of finding a truly optimal solution, they only provide a stochastic approximation of this optimum. In this paper we look into a particular application, portfolio optimisation. We demonstrate two points: firstly, the randomness of the ‘optimal’ solution obtained from the algorithm can be made so small that for all practical purposes it can be neglected. Secondly, and more importantly, we show that the remaining randomness is swamped by the uncertainty coming from the data. In particular, we show that as a result of the bad conditioning of the problem, minor changes in the solution lead to economically meaningful changes in the solution’s out-of-sample performance. The relationship between in-sample fit and out-of-sample performance is not monotonous, but still, we observe that up to a point better solutions in-sample lead to better solutions out-of-sample. Beyond this point, however, there is practically no more cause for improving the solution any further, since any improvement will only lead to unpredictable changes (noise) out-of-sample.
