Evaluation of the quality of predictions in the case of a variable with strongly asymmetric distribution

Quantitative evaluation of the quality of predictions is investigated, with the case of short term solar flare predictions. Strong asymmetry of the predicted variable distribution is characteristic. Prediction evaluation is based on the deviation distribution f(v) (v=forecasted value minus occurred value). The method focuses on the degree of concentration of the deviations around the value v=0. For good predictions f(v) should be symmetric since its asymmetry indicates systematic bias (overestimated or underestimated predictions). Investigating by statistical tests whether the degree of concentration of f(v) is large enough, reliable criteria of good quality of predictions were obtained. Efficiency of this criterion, and of the whole method, is demonstrated by examples of flare predictions.