PRESENTATION: Ranking earthquake forecasts: On the use of proper scoring rules to discriminate forecasts
Here is the presentation from PhD Student Francesco Serafini at the Edinburgh School of GeoSciences annual Postgraduate Research Conference. He is funded by the Real-time earthquake rIsk reduction for a reSilient Europe project (H2020 RISE).
Please find a pdf copy full paper on the arXiv preprint server.
ABSTRACT: This work is about ranking earthquake forecasts with the use of positively oriented (the higher the better) scoring rules, the results can be easily extended to any context in which we aim to compare forecasts for a collection of binary events. In the case of earthquakes, it may be earthquake activity (probability of observing at least one earthquake in each space-time-magnitude bin). In this context, it is essential that the score is proper. We show the consequences of using improper scores in terms of the probability of expressing a preference. Furthermore, we use the probability of expressing a preference to retrieve information about the amount of data and number of bins needed to achieve a high probability of discriminate between forecasts.