For the second weekend in a row, the ELO model has done well with its predictions albeit without being very accurate. Round 20 saw the ELO model tip a total of 7 out of 9 with a mean absolute prediction error (MAPE) of 34.7.
The two main misses for tipping were considered upset wins. The first one was the Friday night clash between Richmond and Collingwood. Our model gave the Pies the slight edge at 52% however Richmond were able to overcome the Deledio effect, winning against expectation without the star. The other big miss was Melbourne’s upset win over Hawthorn, which will have some big effects on our expectations for the remainder of the season. I’ll release our simulations tomorrow and discuss these effects. A few other games were quite a lot closer (see GWS v Gold Coast) or much bigger blowouts (see St Kilda v Carlton) than we expected, driving up our MAPE.
These results lift our season total to 124 correct tips out of 171 games (72.5%), with a MAPE of 29.4
Our results plots show that we’ve been tracking along nicely with our tipping percentage over the last period, although our accuracy in the margin has been slipping above our goal.
At the end of the season I’ll do a bit more work on this plot as a potential source of improvement for our model. For now, I’ll keep posting it for reference.