We’re just about at the halfway mark in the season and the model’s been flying over the Christmas period. Up 35% at the bookies in the past two rounds, with six results out of ten called correctly on Boxing Day and seven out of ten last weekend.
When I first started putting these models together, I was surprised at how few results the bookies actually get right. That’s not to say the model’s better! But I had the impression that the favourite would win more often than is actually the case. If you closed your eyes and threw darts at the fixture list, completely random chance should average 3/10 picks correct every week. The bookies come in at a touch over 5/10 – based on their favourite winning – and Lawro for the BBC (whose predictions everybody loves to mock) has averaged a fraction above that.
If you run my model across a whole past season, it comes in at somewhere around 53% of results correct. Forcing yourself to bet on the result of every fixture, the margins for profitable betting are very fine… Averaging above six out of ten, you’d be seriously in business and could make a fortune. A couple of percent below five out of ten, you’ll bleed cash.
That 53% average from my model does have a bit of an unfair advantage in that at the end of the season, when all the fixtures have already run, we know for sure what the starting line-ups were in each game. One of the reasons why the model can be a bit volatile in its predictions is that we have to take predicted line-ups from Fantasy Football Scout and they’re not always spot on.
The reason for bringing up the topic of accuracy is that the model’s picked New Years Day to back some underdogs. It’s done that before (Fulham to beat Man City?!) but not for several fixtures all in the same week. I wouldn’t be at all surprised if there are some upsets to watch on Match of the Day, but will we have picked the right ones?
If you’re going to have a bet, do take a look at Fantasy Football Scout and see if you like their predicted starting line-ups, before you put your faith in the model.
Here are the percentages for each game:
Passing and shooting stats:
I did say there were upsets this week! The model thinks Swansea have a good chance against Man City, Southampton can beat Chelsea and Spurs can win at Old Trafford.
All of those together seems very unlikely, but any one out of the three would make for profitable gambling.
If you’re betting (and brave):
Swansea City v Manchester City – Home win
Arsenal v Cardiff City – Home win
Crystal Palace v Norwich City – Draw
Fulham v West Ham United – Home win
Liverpool v Hull City – Home win
Southampton v Chelsea – Home win
Stoke City v Everton – Home win
Sunderland v Aston Villa – Draw
West Bromwich Albion v Newcastle United – Away win
Manchester United v Tottenham hotspur – Away win
I’ve checked it three times already and that list is definitely what the model’s predicting. A lot of the games are very close, so it could go either way. Fingers crossed!