Barclays Premier League: Keep clear of one-sided matches for time being

Although Liverpool may look value at home to Hull, the model is not yet to be trusted in such lop-sided matches

Following Jack Houghton as he tries to perfect his new rating system for the Barclays Premier League, this month he warns that it might be worth avoiding lop-sided matches, which is why he's steering clear of Liverpool this weekend...

Punters - particularly ones who spend their time creating obscenely large and complicated spreadsheets - are strange creatures.

When I first started working in the industry back in 2002, the biggest surprise was that the most successful punters were not the high-rollers of betting lore, but were instead often obsessive characters who worked long hours in an attempt to gain marginal statistical advantages over the rest of the market.

Solely taking a numbers-based approach, though, was not always a guarantee of success. Many punters - including myself during some darker points in my punting past - would invest hours in a betting model, but were then reluctant to review its success, for fear of finding that the assumptions on which the model were based were flawed, and that the time invested was wasted. Instead, they would follow a losing formula too long, with sometimes fatal consequences for their betting bank.

Developing a new ELO-based rating system for the Barclays Premier League this year, then, I've taken a cautious approach. First, I've kept stakes small. Second, I've already started to see how the model I've created stands up to reality. Although a fiddly exercise, I've been able to look back over past results and see how predictive the ratings would have been had I been using them in previous seasons.

The answer is encouraging. On average, across all different percentage bands, the model is correct within a margin of error of +/-6%. In other words, if the model says that a team has a 30% chance of winning, they will typically go on to win 30% of the time, give or take.

What's interesting, though, is that the model is not uniformly accurate across all different percentage bands. Rather, it is incredibly accurate in situations where teams are closely matched, but loses its accuracy significantly in instances where there is a large rating discrepancy.

It's ongoing work, this ratings lark, and I now need to go back to my model and assumptions and try and figure out the problem.

In the meantime, though, it's good enough to have the weakness of the model in mind when considering the value bets for this weekend. The model churns out the following "pure" odds:

West Ham United 8.84; Draw 2.13; Manchester City 2.38.
Liverpool FC 1.36; Draw 6.25; Hull City 9.09.
Southampton 1.90; Draw 3.80; Stoke City 4.69.
Sunderland 6.21; Draw 2.24; Arsenal 2.53.
West Brom 2.62; Draw 2.98; Crystal Palace 3.51.
Swansea City 2.11; Draw 3.46; Leicester City 4.19.
Burnley 13.1; Draw 2.05; Everton FC 2.28.
Tottenham Hotspur 1.65; Draw 4.49; Newcastle United 5.79.
Manchester United 4.48; Draw 2.41; Chelsea 2.75.
Queens Park Rangers 3.17; Draw 2.72; Aston Villa 3.14.

And although Liverpool may look value at home to Hull - a home win is currently available at around 1.4740/85 - given what we know, it might be better to look elsewhere.

Recommended Bets

Back Stoke - 2 points at 6.411/2
Lay West Brom - 2 points at 1.9620/21
Lay QPR - 2 points at 2.56/4