The relationship between model complexity and forecasting performance for computer intelligence optimization in finance

The objective of this paper is to show that the ability of nature-inspired optimization routines to construct complex models does not necessarily imply any improvement in performance. In fact, the reverse may be the case. We demonstrate that under the dynamic conditions found in most financial markets, complex prediction models that seem, ex-ante, to be at least as good as more simple models, can underperform in out-of-sample tests. The correct application of these optimization methods requires a knowledge of how and when these techniques will yield beneficial outcomes. We highlight the need for future research to focus on appropriate protocols and a systematic approach to model selection when computer intelligence optimization methods are being utilized, particularly within the realm of financial forecasting.