Old followers of my blog would be familiar with an article I’ve released back in 2017 about the profitability of backing the draw in close games in the Premier League. The idea was borrowed from David Sumpter’s book Soccermatics. Two and a half years have passed since then, but the strategy is still alive as could be seen in some of Cassini’s from the green-all-over blog recent articles. With still not that much sports to bet on and the Premier League scheduled to resume in less than a week (on June 17th), I thought the time is right to reevaluate backing the draw in the Premier League and beyond.
The null hypothesis
There are many trends one can notice within the data at any given sport. Once a trend is discovered and it looks profitable we would often build a narrative to support it. However, experienced punters know that in most cases those “trends” are nothing more than the result of random noise. That should be the starting point of any such research. Any trend is noise in the data until proven otherwise.
Furthermore, it is best practice to start from an idea based on some underlying truth about a sport and then look for a trend in the data that you expect to find. Finding some trend and then building a narrative about it is more likely to land you in the noise department.
So, what are the chances that the trend reported on in 2017 was just random noise?
What’s new?
2,5 years is enough time to gather quite a bit of additional data, so in case a “trend” was just noise you will expect an upward chart followed by a totally random movement shortly after it has been “discovered”. Where do we currently stand in this case?
To remind you, the original idea recommends backing the draw in the Premier League in games where the difference of the winning probabilities of the two teams as implied by the odds (after adjusting for overround) is less than 10%. In other words, the market implies the teams are somewhat evenly matched in strength. The data is as usual kindly provided by Joseph Buchdahl’s football-data.co.uk. Level staking is being used, the y axis represents the profit in units.
Here are the results of the strategy until now, with the release of my old article marked with a red line:
A quick note: there is a small difference to my initial article since I used maximum odds among all bookmakers back then. Now I am only using Pinnacle’s opening price. This will not always be the best price (though it often will), therefore the profit before the red line might be slightly smaller than what was reported in the initial article. If you are like me, you’d be banned from all bookmakers from the dataset but Pinnacle, so these are the only prices I care about.
It must also be said, that if this is the only strategy you are following withih those bookies (as opposed to arbitrage, value betting or bonus hunting) you are unlikely to be banned from anywhere. So if you have access to those feel free to make your own calculation, Pinnacle is what I am going to use.
What do we see?
At first sight, the picture does not look too well. We are having very close to a zero profit over those two and a half years. So it looks like an upward trend followed by a random walk after it’s discovery. Still, we must point out we should expect a small loss the size of the overround on average. Which is rather small for Pinnacle and the Premier League, but you would still expect to lose around 3.5 units, whereas you would have won around the same amount. Still, the profit compares unfavourably to previous years.
Is the value still there, is it gone or has it never been there in the first place?
There are different schools of thought in that regard. Tell a firm believer in the efficient market hypothesis that such a big market contains such an enduring inefficiency, and you would be laughed off in a second. However, one still has the opportunity to use the opening prices, as I have done in the backtest. While those would certainly still be pretty sharp, they will only be based on the estimation of the trader, one expert in the market, as opposed to all market participants.
More often than not, if the line movement towards closing does not agree with your hypothesis, you do not have a CLV (closing line value) and you are fooling yourself – or so the theory goes.
Well, what is the CLV of this strategy?
Backing the Draw in Close Games: Closing Line Value
The “CLV bank” measures the overround-adjusted closing line value per bet and adds it up. The “-Overround bank” simply adds up the overround per game, with a minus sign.
As you see, there is very little CLV to show for. Indeed, the price is drifting very slightly post opening, but not nearly enough to compensate for the entire overrround – it only seems to cover around 20% of it most of the time.
20% is something, but certainly not a lot. Whether it could be the result of something other than chance is up for further research. In any case, the strictest criterion for profitability – having a positive CLV bank from the strategy – isn’t satisfied here.
Is it more to this story than just the CLV?
Many punters don’t fully trust the efficient market hypothesis when it comes to betting and tend to question that the assumptions the theory relies on are all true in the case of betting (or even financial) markets. Personally, I am also no CLV fanatic.
Still, the Premier League is a huge betting market and therefore must be one of the most efficient ones. So I am really curious – how likely it is that a negative CLV covering only 20% of the overround over such series of bets (492 for this strategy) still indicates a small value in a market. How much would one league tend to deviate from the average in terms of CLV performance?
I decided to have a look at other leagues with the hope for an answer to those questions.
Backing the Draw in Close Games in La Liga
A great league to compare with would be Spanish La Liga. Back in 2017 I mentioned that La Liga was another league where the strategy seemed to have delivered great results in the past. How does it look now?
In terms of performance that is an entirely different picture. Ever since the previous article was posted, the strategy was up around 30 units over 167 unit bets. That’s a solid ROI of some 18%.
The case for the strategy offering value for both Premier League and La Liga would go along the following lines: These are probably the two biggest, most watched and most global football leagues. When a derby match comes, the majority of punters like to back team A or team B to win, while backing the draw is a relatively boring choice. Due to the leagues being this huge, all that square money might distort the prices. Market maker books like Pinnacle do not weigh bets from all punters the same. However, they still need to manage their risks. Therefore, they need to at least partly adjust their prices to account for that square action.
Of course, as in the case of the Premier League strategy, we shouldn’t rule out the possibility that the whole series was entirely the result of luck. So let’s have a look at the strategy CLV for La Liga:
La Liga Draw CLV
The CLV is in fact below the expected average (-overround) from backing random selections! In fact, in relative terms the CLV bank is underperforming with 14,33% for the whole period.
Comparing the CLV for La Liga with the results for the Premier League above, we see that such small deviations in the CLV as compared to the -Overround bank are not necessarily related to the inherent profitability of the strategy, but are rather a random process.
Small sample?
Of course, only looking at two leagues might not be sufficient to draw any conclusions. It is perhaps a good idea to compare with the other leagues, for which we have data. Doing so we not only get a feeling for what CLV deviation from the -Overround bank is normal. We can also see what other outliers are there in terms of ROI and compare with the Premier League (and La Liga).
Therefore, I have taken the time to calculate the same statistics presented above for (almost) all leagues in football-data.co.uk’s dataset since the year Pinnacle closing price has been provided.
Backing the Draw in Close Games: Summary results
Sum
For certain smaller leagues there were single games where Pinnacle opening and/or closing line was not available. In those cases I took the second best option available for this game. This would mostly mean odds from another bookmaker. Even in smaller leagues this occurs relatively rarely so it shouldn’t distort the results.
I have highlighted the results I have found most relevant and will go through some them in more detail.
P/L ROI
This metric simply measures how the strategy performed in all leagues, ROI-wise. The Premier League leads the pack (13.63%) with La Liga close second (10.6%). The narrative that punters don’t like backing the draw on big leagues has been mentioned earlier. However, that hardly explains the negative ROI for the German, Italian and French elite leagues. The Premier League and La Liga do have larger international audiences. They probably attract more punter money. However, that fact itself does not convince me that those figures are not a result of randomness.
Absolute ROI Over-/Underperformance
Here I measure how much the ROI deviates from the expected ROI over the long run. This is simply the absolute value of the difference between the ROI and CLV. The goal is to get a feeling of what deviation from the expected ROI could be expected in data series of certain length.
While the Premier League and La Liga stick out here as well, there is a more notable outlier. The Scottish third tier (League 1) has an extremely low ROI with this strategy. This translates to an even larger deviation from the CLV compared to Premier League and La Liga. Actually betting exactly the opposite of what the strategy suggests in the Scottish League 1 would have probably returned even higher profit than backing the strategy in Premier League and La Liga. Why? Is there a basic difference between the playstyle in this league and all others, including all other Scottisch ones? Or is this just noise…
Margin / CLV
The margin is as expected. Bigger leagues have slightly lower margins as it should be. League margins rank very similarly to league margin-adjusted CLVs for that strategy. Neither PL or La Liga, nore the other leagues have any margin-adjusted CLV with this strategy. The non-margin-adjusted CLV is still slightly positive for some leagues (or slightly negative for others). This is not the proper way to measure CLV in general. Still, whether it contains certain information is made more clear by the next metric.
Absolute CLV Margin Compensation
Even though no league has CLV with this strategy, we have seen that in the Premier League the line movement at least covers a small part of the margin. But is that significant in any way? The evidence from La Liga suggested that this is not the case. A look at the rest of the leagues reveals a clearer picture. Here you can see relative deviation of the CLV from the opening margin. I have called it the CLV Margin Compensation. If the CLV tends to be even worse than the negative margin, this figure would be negative. This is negated when looking at the absolute figures.
Over the long run these two figures would tend to converge, with the absolute compensation trending towards 0%. But for a shorter series of bets, the 22% deviation from the Premier League does not seem to indicate much. For example, there is 43,14% difference (in the opposite direction) for the Bundesliga. This shows again that only the margin-adjusted CLV could be a reasonable metric for a betting strategy. The non-margin adjusted CLV contains too much noise, does not seem to measure value in any way and in reality does not deserve the “V” in its description.
Backing the Draw in Close Games: Trend or Noise?
After having laid down all my numbers, I need to make a judgement on the soundness of this strategy for a second time.
I must say I was rather bullish on the strategy the first time I have discovered it. Re-assessing it two and a half years later, the PL was mostly flat while La Liga is still going strong. However, looking at the overall picture I see indications that those results might have been purely random. Such difference between observed and expected ROI is certainly not unseen in other leagues (see Scotland League 1). There is no meaningful CLV in none of the two leagues, so the broad market does not seem to buy it. If the PL ROI occurred in, let’s say, the French Ligue 2, nobody would bat an eye. Since it is the Premier League (and La Liga), we have all paid quite a bit of attention to it.
Conclusion
Therefore, after running the numbers again I now tend to move more towards the naysayers’ camp. I don’t see myself backing the draw in these games in the near future. I also don’t exclude the possibility that I (similarly to the market) am missing some underlying reason why the ROI of those two leagues are what they are. So I will certainly come back in a few years time and have a look again at how this trend has developed. If I have been on the wrong side of it, at least I would have learned something from the experience.
I hope you also learned something from today’s article. Next time it will be back to eSports, as my League of Legends models have delivered in the last few months, so it might make sense to share a bit more about that. Furthermore, the positive results there have encouraged me to look elsewhere such as CS:GO or Dota2. It is a wonderful world full of opportunities to win money and of course many more to lose some.
Partnership with RebelBetting
On an unrelated note, I have expanded my partnership with RebelBetting. I will be writing a few articles for the RebelBetting’s blog in the next months. The first one (comparing value betting and the stock market) is actually already submited and I hope to see it posted soon. In exchange, RebelBetting is offering all readers of the Church of Betting an exclusive promotion of 1 free month of their value betting product.
On top of that, due to the still ongoing COVID19 pandemic, the Rebelbetting team is currently adding an extra free month for a total of 2 free months! In those two months you can try the software, make some money with it and still have the chance to make up your mind when the 2 months are over. It might be a good idea to hurry up, as this one won’t last forever.
You won’t get these terms anywhere else. The football season is restarting in several countries already. Therefore, this is a good opportunity to jump into value betting if you have been considering it. Make sure to subscribe through some of the affiliate links on this blog in order to be eligable for the promotion.
In the meantime, I would appreciate you hitting the “Follow” button on my Twitter account or subscribing to my newsletter – in this way I can always keep you updated on new articles. Thank you for sticking around and stay safe out there!
