–
by John Kamaras
02/02/2021
Summary
After almost three and a half month of football activity, I want to evaluate my equations’ performance in real life. I remind you that I have two sets of equations; the results equations are easier to apply and predict only match result (1, X, 2). The expected goals equations, more complicated but with the ability to predict more outcomes.
In this article, I will evaluate the equations’ results as a classification method and as a value betting method. The classification method considers the more probable results according to equations when the value betting method uses the estimations with lower odds than the bookmakers’.
You can find the optimal equations for each division in my book at Amazon bookstore.
From Oct 18 till Jan 31, we examined 2143 games; you can download the predictions of the equations along with the game info and odds from here.
The columns with names eq1,eq2,eqX are the predicted probabilities for each outcome from the results equations. The column eqC is the most probable outcome. The columns wa1, waX, wa2 are the predicted probabilities for each outcome from the expected goals equations, and the column waC is the most probable outcome. The columns waO, waU, waOU referring to Over/Under 2.5 goals predictions.
I’ll remind you that the bookmakers’ odds Home, Draw, Away, Over, Under are the average ones; you can find better odds for sure. Finally, you can find the Covid and VAR effect to football outcomes here.
Section one – Results equations as a classifier.
outcome | games | 1 | per1 | X | perX | 2 | per2 | ret | prof% |
1 | 1572 | 698 | 44.4% | 409 | 26.0% | 465 | 29.6% | 1448 | -8.6% |
2 | 367 | 94 | 25.6% | 88 | 24.0% | 185 | 50.4% | 395 | 7.2% |
X | 204 | 72 | 35.3% | 61 | 29.9% | 71 | 34.8% | 210 | 2.9% |
As you can see on the table, we had excellent results, even for blind betting, when the results equations suggest as the most probable outcome, the draw or the away win.
We had 571 games with a positive expected value, a pool from which we can “safely” pick our bets. A major problem with the home win prediction which, for blind betting, returns 8.6% loss.
Section two – Results equations as value bets.
outcome | games | 1 | per1 | X | perX | 2 | per2 | ret | prof% |
1 | 1479 | 656 | 44.4% | 385 | 26.0% | 438 | 29.6% | 1359 | -8.9% |
2 | 353 | 88 | 24.9% | 84 | 23.8% | 181 | 51.3% | 385 | 8.4% |
X | 204 | 72 | 35.3% | 61 | 29.9% | 71 | 34.8% | 210 | 2.9% |
The picture is almost the same, great results for the draw and away win predictions and very poor results for the home win. We had 557 games with a positive expected value, a pool from which we can “safely” pick our bets but a very large number of home win predictions we must avoid; I’ll try to address this issue later.
Section three – Expected goals equations as a classifier.
outcome | games | 1 | per1 | X | perX | 2 | per2 | ret | prof% |
1 | 1727 | 749 | 43.4% | 451 | 26.1% | 527 | 30.5% | 1565 | -10.3% |
2 | 405 | 111 | 27.4% | 101 | 24.9% | 193 | 47.7% | 410 | 1.3% |
The expected goals equations never (or rarely) predict a draw as a most probable result because they follow the bookmakers’ approach to the game more. We have slightly worse results from the results equations, although the away win prediction still has a positive expected value. The problem with the home win gets bigger.
Section four – Expected goals equations as value bets.
outcome | games | 1 | per1 | X | perX | 2 | per2 | ret | prof% |
1 | 1555 | 686 | 44.1% | 392 | 25.2% | 477 | 30.7% | 1434 | -8.4% |
2 | 375 | 104 | 27.7% | 91 | 24.3% | 180 | 48.0% | 380 | 1.3% |
We had a slight improvement for home win predictions.
Section five – Combining the two methods.
This section presents the games’ results that two methods predict the same outcome as most probable.
outcome | games | 1 | per1 | X | perX | 2 | per2 | ret | prof% |
1 | 1468 | 666 | 45.4% | 382 | 26.0% | 420 | 28.6% | 1351 | -8.7% |
2 | 242 | 62 | 25.6% | 55 | 22.7% | 125 | 51.7% | 261 | 7.2% |
We observe a minor decrease of away win predictions regarding the capital return percentage and a slight increase to home win returns. The problem with the home wins has troubled me enough; although the equations are not suitable for blind betting, the negative return is too big. Of course, because of the closed stadiums, we might have a very smaller home team advantage or even a zero one.
A slight increase in home team advantage (about 3-4 percent) is more than enough to reverse the picture, but I decided to do a little more digging. Below is the table for home win predictions when the home team is a clear favorite, when it is a better team, in other words, when the home team advantage doesn’t play so much role in the game’s outcome. I choose the odds from one to 2.10 ,and I’ll use the results equations, as a classifier, because they seem to do a little better than expected goals this year.
outcome | games | 1 | per1 | X | perX | 2 | per2 | ret | prof% |
1 | 692 | 405 | 58.5% | 149 | 21.5% | 138 | 19.9% | 669 | -3.4% |
Much better picture for the home teams but still under expectations. Now let’s see the games’ results with home team odds bigger than 2.10 when the home team advantage starts to kick in the game’s reality.
outcome | games | 1 | per1 | X | perX | 2 | per2 | ret | prof% |
1 | 880 | 293 | 33.3% | 260 | 29.5% | 327 | 37.2% | 778 | -13.1% |
The loses for the home team are so big that I am sure that there is value on draw and away win, but I’m not going to look it up because this is not the right way to do things!
You can find an article about the Covid and VAR effect to football outcomes here.
Final thoughts
We can establish a pool of predicted outcomes with positive or near positive expected returns from our analysis. During these three and a half months of football activity, the method gave us over 550 games from which we could select our bets more safely. Applying your style to game selection is essential; in my book, I present 40 real-life examples that describe my betting style. You can use them as it is or to give you an idea.