On July 10, about 150 statisticians / data scientists braved a typically cold Melbourne winter morning to hear Hassan Kamel of Crypton Capital and Darren O’Shaughnessy of the Hawthorn Football Club speak about the intersection of statistics (and data science more broadly) and sport. This event was held jointly with Data Science Melbourne, with the meeting venue and a light breakfast generously provided by KPMG.
Hassan spoke of his experience in betting (and winning!) on sporting events. He discussed two possible strategies for beating the bookmaker. The first strategy is to try to develop a predictive model that is overall better than the bookies’ predictions, and use that to place more accurate, and therefore more profitable, bets. His second strategy, which he has used to great success, involves developing a predictive model that is more accurate for a specific subset of games, by using variables that may be important to the outcome of a match but are not typically included in the bookies’ reckoning. With respect to this second approach, he spoke of data “weirdness”: these predictors need to be sufficiently obscure and unusual so as not to have been considered by bookies but also be relatively easy to access and describe phenomena that happen frequently enough to allow you to make a profit. An example Hassan gave was including barometric pressure, along with temperature, when predicting outcomes of AFL matches: it seems that some teams have an advantage for certain combinations of pressure and temperature (perhaps by how this affects the properties of the football?). This advantage can be exploited when placing bets on match outcomes, although the advantage is of course quite small!
Darren spoke about his experiences as an analyst for the Hawthorn AFL team. Often coaches want recommendations while a game is taking place: should a player change strategies, or perhaps be replaced on the field by another player? An important point that he made was that the randomness of games is often not understood by those making decisions about the game, and an important part of his role is to communicate uncertainty to people who have not had a great deal of experience with statistics. Darren spoke of the challenges in using GPS data to help determine what players have done, and should do, during games. It seems that although the future for the use of these data is promising, the current technology is not precise enough: in some instances, the GPS data indicates that some players spend most of the game in the stands!
Jessica Kasza, Damjan Vukcevic