Does accounting for an artificial turf advantage in Dutch football increase predictive accuracy of probabilistic models?

Currently, one in three matches in Dutch professional football (“Eredivisie”) is played on an artificial turf surface. Recently, statistical evidence was reported that suggest an increased home advantage for Dutch teams playing on artificial turf against an away team that plays its home games on natural grass (van Ours, 2017). Here we investigate if accounting for this effect increases out-of-sample predictive accuracy of match outcomes. To do so, we implemented existing probabilistic models to make one-step-ahead forecasts, with and without the additional artificial turf predictor. The ranked probability score (RPS) is used to assess the accuracy of the forecasts and compare between models. We find that including the artificial turf home advantage as additional predictor does not improve the accuracy of the forecasts. We conclude that the evidence for a large artificial turf advantage in the Eredivisie is not strong.