Title: Picking scores: Forecasting low probability events
Authors: James Reade - University of Reading (United Kingdom) [presenting]
Carl Singleton - University of Reading (United Kingdom)
Alasdair Brown - University of East Anglia (United Kingdom)
Abstract: In sport, and football in particular, there is a great level of satisfaction taken in correctly guessing what the score will be in advance of a game taking place. Part of this satisfaction must be drawn from just how difficult it is to predict exact scores; these are low-probability events. Furthermore, it is believed that agents are unable to distinguish between low and tiny probability events, and this forms one explanation for observed departures from price efficiency in gambling markets. We investigate scoreline forecasts from a range of sources; bookmaker prices, user tipsters, experts, and statistical models. We rank the various forecasts by a range of scoring rules both relative and absolute, and quantify the extent to which known biases are existent in these different types of forecast.