Drawing Conclusions, Perhaps

One of Bill Benter's observations (see my previous post if you are unaware of whom he is) was that:

In the author's experience the minimum amount of data needed for adequate model development and testing samples is in the range of 500 to 1,000 races.

The English Premier League has been around for more years than there is betting data, but thanks to Joseph Buchdahl's Football Data web site at least we have data going back to the last few months of the second millennium.

This will come as no surprise to followers, but since that first 2000-01 season, there have been 7,498 Premier League matches, and after adjusting for the over-round, if you had backed the Draw in every game where no team had greater than a 50% implied probability of winning, you'd be up by around 2.49% from 3,525 matches.

If you narrowed this down to the 40% level, your ROI is up to 10.26% from 1,217 matches. There is a 1 in 99 chance that this outcome is from luck, so don't get carried away:

It's a promising sign though that the 40 selections so far this season heading into this weekend had an ROI of 34.3%.

The Draw appears to be most undervalued when the Away team is favourite, but not by a big margin:

Quite a contrast to when the Home team is a modest favourite:

When there is very little difference in win probability between the two sides, the Home / Away factor isn't apparent:

The 73 matches where the probabilities were identical had a 6% ROI. Games where the win probabilities of the two teams are close continue to offer an edge to Draw backers.  

Incidentally, has anyone noticed that the over-round on Pinnacle's closing prices which started the season (first 50 games) at 102.6% has since climbed over the last 50 games to 103%?