Brain imaging and forecasting: Insights from judgmental model selection

In this article, we shed light on the differences between two judgmental forecasting approaches for model selection – forecast selection and pattern identification – with regard to their forecasting performance and underlying cognitive processes. We designed a laboratory experiment using real-life time series as stimuli to record subjects’ selections as well as their brain activity by means of electroencephalography (EEG). We found that their cognitive load, measured by the amplitude of parietal P300, can be effectively used as a neurological indicator of identification and forecast accuracy. As a result, judgmental forecasting based on pattern identification outperforms forecast selection. Time series with low trendiness and high noisiness have low forecasting accuracy because of the high cognitive load induced.