If you had followed the betting advice in Soccermatics you would now be very rich
While writing Soccermatics I did something I had never done before: I bet on football. I did it as an experiment. I wanted to work out what paid off and what didn’t, but I also put “my money where my mouth is” and used some of my book advance to back the games my model predicted as winners. I wanted to truely test my skills as a mathematician.
I’ve written about the details of my betting strategy elsewhere: extensively in the book itself, then afterwards for the Economist 1843 magazine and more recently for Pinnacle Sports. In short, I used a statistical model to show that the odds for the Premier League have a weak longshot bias (which means it pays off to back the favourite) and a slightly stronger bias against draws between evenly matched teams, in particular matches between the big six. It was by exploiting these biases in the odds that my strategy made money.
I made a small profit during my experiment, about a 25% return over half a season. I was pretty pleased with myself and it made a good story for the book, but I haven’t gambled (much) since. I felt there were other things I could apply my skills to. As I wrote in the book:
I just made the plot below of the profits I would have made if I had invested £100 in my strategy at the time I made the model (at the start of the 2015–16 season) and bet according to the rules I described in the book.
My £100 would now be over £1800! In fact, in Soccermatics my bankroll was £500, so I would just now be sitting on ten grand if I’d just trusted my model! What a fucking idiot I am.
OK. Calm down, David. There are a few caveats here. The first thing to note is that from around the start of last season until the start of this season (matches 400 to 800) I would have seen my profits stagnate and slowly dwindle. It is only since September this year that the real profits have been made. In fact, when I last reanalysed this data for the Pinnacle article in October, I was far less bullish about the model.
Secondly, the big profits since September 2017 came during a period in which the big six in the Premier League have won alot of their matches. This helped my model, as did some high profile draws. The alternative universe, where I followed my own betting tips, would have seen me place £734 on a draw between Arsenal and Chelsea on Wednesday the 3rd of January (the last match shown in the figure). As Bellerin’s strike hit the back of the net for Arsenal to equlaise for 2–2 I would have been dancing around the room, singing the praises of the bookmaker which had given me pre-match odds of 3.51, providing me a profit of £1,842. This result, along with the Gunners 3–3 against Liverpool before Christmas, is extraordinary, and can definitely be attributed to luck. The current surge of my betting strategy is unsustainable.
So let me emphasise (and not just to make myself feel better) that the really big, recent rise in value of the Soccermatics strategy is due to a bit of very helpful randomness created by Arsenal and other big six teams.
There is, however, reason to believe that the Soccermatics strategy works in the long term, albeit with a lower rate of return than it is currently experiencing. The bookmakers edge in the data I used to assess the model is 1.1% on average, so the model significantly outperforms random bets throughout the whole period, not just the last few months. I reassessed my strategy after reading an article by Church of Betting showing a long-term profitablity of backing draws between equally strong teams. This analysis includes more historical data than I used in the book and also looks at other leagues (which Church of Betting shows don’t tend to have the same bias).
It is always possible to find statistical patterns in historical data that appear to give profits, only to find that they don’t hold in the future. We are fooled by randomness. Joseph Buchdahl’s excellent book — Squares & Sharps, Suckers & Sharks — carefully analyses a whole range of different betting strategies and finds that they usually fail to beat the market in the long term. He argues that a true winning strategy must have (1) a proven record on future performance and (2) have some sort of ‘explanation’ of why it works. His own strategy, based on using Pinnacle odds to beat other bookmakers, passes both these tests.
My model can make some claim to pass these two tests, albeit on a smaller data set. I first tweeted about my model at match 50 (in the graph above), I submitted my model to the publishers around match 100 and the book was published around match 300. So the profits shown are based on predictions of the future, not an historical fit to the data. I also have some form of ‘explanation’ for my model: in the Premier League punters bet against the big six, probably becuase they support other big-6 teams. When two big-six teams meet then all the focus is on one of the two teams winning and the draw is forgotten by the punters and the bookies. These circumstances are somewhat unique to the Premier League, because there are so many ‘just for fun’ gamblers involved in the market and six teams with a worldwide following.
It is easy to be wise after the fact and to talk about the big bets I could have made. I did put £30 at 3.51 on a draw in the London derby (and on the Arsenal vs. Liverpool game, where I got odds of 3.61, before Christmas) and I was pretty pleased when the equalisers went in at the end of both matches. But the truth is that I’m not the sort of person with the nerve to put £734 on a single match (even if the bookies would let me). And I don’t think I ever will be. Profits change much more quickly than people do.
I’m sorry about this, but I can’t help being sanctimonious again. If you do want to be a statistical gambler then you need a set of skills. I would encourage everyone to get those skills, because they are both useful in your everyday life and greatly increase your possibilities on the job market. My betting model is one of many examples in Soccermatics of how maths gives us an edge in the game and in the rest of life. I would never encourage anyone to take up gambling full time, but I would encourage everyone to learn more maths and statistics. It is the safest bet of all.
Acknowledgement: Thanks for Joseph Buchdahl (https://twitter.com/12Xpert) for double-checking my calculations. Using a slightly different way of calculating ‘fair odds’, but reimplenting my model, Joseph found a similar growth curve to the one I show, but climbing to £1400 instead of £1800.
OK. Calm down, David. There are a few caveats here. The first thing to note is that from around the start of last season until the start of this season (matches 400 to 800) I would have seen my profits stagnate and slowly dwindle. It is only since September this year that the real profits have been made. In fact, when I last reanalysed this data for the Pinnacle article in October, I was far less bullish about the model.
Secondly, the big profits since September 2017 came during a period in which the big six in the Premier League have won alot of their matches. This helped my model, as did some high profile draws. The alternative universe, where I followed my own betting tips, would have seen me place £734 on a draw between Arsenal and Chelsea on Wednesday the 3rd of January (the last match shown in the figure). As Bellerin’s strike hit the back of the net for Arsenal to equlaise for 2–2 I would have been dancing around the room, singing the praises of the bookmaker which had given me pre-match odds of 3.51, providing me a profit of £1,842. This result, along with the Gunners 3–3 against Liverpool before Christmas, is extraordinary, and can definitely be attributed to luck. The current surge of my betting strategy is unsustainable.
So let me emphasise (and not just to make myself feel better) that the really big, recent rise in value of the Soccermatics strategy is due to a bit of very helpful randomness created by Arsenal and other big six teams.
There is, however, reason to believe that the Soccermatics strategy works in the long term, albeit with a lower rate of return than it is currently experiencing. The bookmakers edge in the data I used to assess the model is 1.1% on average, so the model significantly outperforms random bets throughout the whole period, not just the last few months. I reassessed my strategy after reading an article by Church of Betting showing a long-term profitablity of backing draws between equally strong teams. This analysis includes more historical data than I used in the book and also looks at other leagues (which Church of Betting shows don’t tend to have the same bias).
It is always possible to find statistical patterns in historical data that appear to give profits, only to find that they don’t hold in the future. We are fooled by randomness. Joseph Buchdahl’s excellent book — Squares & Sharps, Suckers & Sharks — carefully analyses a whole range of different betting strategies and finds that they usually fail to beat the market in the long term. He argues that a true winning strategy must have (1) a proven record on future performance and (2) have some sort of ‘explanation’ of why it works. His own strategy, based on using Pinnacle odds to beat other bookmakers, passes both these tests.
My model can make some claim to pass these two tests, albeit on a smaller data set. I first tweeted about my model at match 50 (in the graph above), I submitted my model to the publishers around match 100 and the book was published around match 300. So the profits shown are based on predictions of the future, not an historical fit to the data. I also have some form of ‘explanation’ for my model: in the Premier League punters bet against the big six, probably becuase they support other big-6 teams. When two big-six teams meet then all the focus is on one of the two teams winning and the draw is forgotten by the punters and the bookies. These circumstances are somewhat unique to the Premier League, because there are so many ‘just for fun’ gamblers involved in the market and six teams with a worldwide following.
It is easy to be wise after the fact and to talk about the big bets I could have made. I did put £30 at 3.51 on a draw in the London derby (and on the Arsenal vs. Liverpool game, where I got odds of 3.61, before Christmas) and I was pretty pleased when the equalisers went in at the end of both matches. But the truth is that I’m not the sort of person with the nerve to put £734 on a single match (even if the bookies would let me). And I don’t think I ever will be. Profits change much more quickly than people do.
I’m sorry about this, but I can’t help being sanctimonious again. If you do want to be a statistical gambler then you need a set of skills. I would encourage everyone to get those skills, because they are both useful in your everyday life and greatly increase your possibilities on the job market. My betting model is one of many examples in Soccermatics of how maths gives us an edge in the game and in the rest of life. I would never encourage anyone to take up gambling full time, but I would encourage everyone to learn more maths and statistics. It is the safest bet of all.
Acknowledgement: Thanks for Joseph Buchdahl (https://twitter.com/12Xpert) for double-checking my calculations. Using a slightly different way of calculating ‘fair odds’, but reimplenting my model, Joseph found a similar growth curve to the one I show, but climbing to £1400 instead of £1800.
No comments:
Post a Comment