Last week marked the model’s best performance so far this season, with eight correct predictions out of ten and a profit of 64% if you followed its calls for betting. Although maybe we shouldn’t get too carried away with that one as the FT reports that a lot of people did very well on round 21 and the bookies suffered an awful weekend. Coral called it one of their worst weeks in 88 years.
Those bad results for the bookies came from the top seven teams bringing in a lot of accumulators, but we had upset predictions of wins for Newcastle vs Man City and West Brom vs Southampton, which were the two results that the model missed. Our predictions did well in a different area, picking a couple of fairly long odds games lower down the table and correctly forecasting away wins for West Ham at Cardiff and Sunderland at Fulham.
Performance over the four rounds since Christmas has been strong, with the new-and-improved model averaging over six out of ten predictions correct each week and a profit of over 40% per bet. Be warned that this level of accuracy is unlikely to last! But we can certainly enjoy it while it does.
As we move into the transfer window, the model encounters a tricky stage of the season. New players don’t have a data history and although it would be possible to obtain data for a player like Nemanja Matic, by raiding Squawka for his Benfica stats, would they be applicable to a model of the English Premier League? It’s not a theory that I’ve tested yet and so we’re forced to make some assumptions. When a new player signs who has no data, I switch them for one of their team-mates until they’ve built up a game or two’s worth of statistics.
If a team signs players who are a similar standard to current personnel, this works reasonably well. If they sign a superstar striker who’s going to change the second half of their season, then the model won’t see that coming. Thankfully from a prediction point of view, the transfer window has been fairly quiet so far.
In a similar vein, this week may well see the return of Andy Carroll to West Ham’s starting line-up, but he has no 2013/14 numbers yet so can’t be simulated by the model. For the moment, Carlton Cole will still be starting in the simulated games until Carroll has built up some minutes on the pitch. Other than these types of enforced changes, I’m taking the starting line-ups from Fantasy Football Scout as usual.
Here are this week’s percentages:
And passing and shooting stats:
The big call this week is Manchester United to win away at Chelsea. The model’s suggested that Chelsea would lose at home once before – to Liverpool on 29th December – which ended 2-1 in Chelsea’s favour. Could this be the week it happens? It is worth noting that our Man United line-up has Rooney in it, playing behind Welbeck. Switching Rooney for Kagawa, the chances see a huge 20% swing in Chelsea’s favour as you take a significant number of shots out of Manchester United’s play. The model doesn’t usually swing so far as a result of switching only one player, but for Manchester United, Rooney and van Persie make a huge difference because they represent such a large proportion of the team’s firepower.
Finally, here are the calls I’ll be backing this week (with a last minute check on whether Rooney plays).
Sunderland v Southampton – Away win
Arsenal v Fulham – Home win
Crystal Palace v Stoke City – Away win
Manchester City v Cardiff City – Home win
Norwich City v Hull City – Home win
West Ham United v Newcastle United – Away win
Liverpool v Aston Villa – Home win
Swansea City v Tottenham Hotspur – Close. Home win
Chelsea v Manchester United – Away win (if Rooney’s in)
West Bromwich Albion v Everton – Away win
A few of these are close to being called as draws using the rule of having a draw probability over 27%, but they don’t quite make it across that line, so I wouldn’t be surprised if a couple of draws trip up the predictions this week. Given that our simulated version of West Ham doesn’t contain an Andy Carroll, he should also be guaranteed to score against Newcastle! Good luck and I’ll see you next week.
