On the Evaluation of Skill in Binary Forecast

A good prediction is very important for scientific, economic, and administrative purposes. It is therefore necessary to know whether a predictor is skillful enough to predict the future. Given the increased reliance on predictions in various disciplines, prediction skill index (PSI) is devised. Twenty-four numerical examples are used to demonstrate how the PSI method works. The results show that the PSI awards not only the same score for random prediction and always predicting the same value but also nontrivial scores for correct prediction of rare or extreme events. Moreover, the PSI can distinguish the difference between the perfect forecast of rare or extreme events and that of random events by awarding different skill scores while other conventional methods cannot and award the same score. The data on growth of real gross domestic product forecast of the Bank of Thailand between 2000 and 2019 are also used to demonstrate how the PSI evaluates skill of the forecaster in practice.