Monthly Archives: April 2013

TPOEM 3pm Predictions: 27 April

I have made some widespread changes yet again to TPOEM and I am now in a better position to discuss how it works. Will give an overview and some stats from it in my next post.

In the meantime, here are the predictions for this afternoon’s EPL games:

TPOEM 27 Apr

You will notice that I have now added a most likely score (and probability) for each game – borrowing this idea from Neil Charles’ model predictions.


I have added in a prediction for Newcastle vs Liverpool and, knowing that I won’t be able to post predictions for the rest of the weekend games I have simulated those as well (using probable starting line-ups):

TPOEM 27 Apr 2

Today TPOEM is quite well aligned with Betfair, which is a bit boring but a good sign for the improvements I have made to it. A couple of outlier predictions with TPOEM respectively more and less confident about Arsenal and Chelsea’s chances than the bookies – something I will have to monitor. This will no doubt encourage a further tweak to the way the model handles each team.


TPOEM 3pm Predictions


Judging by the above, TPOEM may well be overestimating the probability of draws in general. The difference to Skybet odds suggests below par value in betting on Arsenal, Villa, Everton, Liverpool or Southampton (although might be slightly better with other bookies e.g. Betfair).

Staying true to the value-principles of the model, I’ll back draws for for Reading-Liverpool and Arsenal-Norwich even though neither result represents the most probable outcome.

In the next week or 2 I’ll post some more information on how TPOEM works, why I’m doing it and what I use it for. I actually use the model primarily for player appraisal, judging ex-post player performance (i.e. past results) rather than just to try and beat the bookies. Nevertheless, the prediction side is an interesting and amusing exercise (at least for me anyway!) – however this is still in its infancy and I am tweaking it every week at the moment.

The basics of TPOEM have striking similarities to Neil Charles’ model, which you can follow here:

I also use Excel and VBA (more than I ought to, it certainly slows me down) and EPL Index player data to arrive at game expectations, but we are coming out with very different results – that just goes to show how important the modellers’ input weightings and choice of variables are to any model.

Martin Eastwood’s EI index is another worth checking out, although on the face of it it does appear to be more of a top-down approach to predicting:

Top-down weighted models (focusing on macro- rather than micro-level data) are likely to be used more heavily by bookmakers which suggests that (assuming I’m right about the construction of the EI Index) Eastwood’s model may have a better chance of successfully predicting results than either TPOEM or Charles’ model. But unless any one of the models is spectacularly bad, we’ll need a lot more information before that becomes clear. Healthy competition in any case!