Why we don't offer predictions

There are a few services out there that claim to give you actionable predictions with positive ROI. We have tested two of such services and their claims were not confirmed. Maybe we were unlucky as we didn’t test them for a long enough period of time.

However we think their promise is not attainable for one fundamental reason: the betting companies have better models. They employ a lot of mathematicians and statisticians because, if they are wrong, they stand to lose millions or tens of millions. One such service claims above 70% accuracy. This would ensure 30% ROI. Week after week, after week. I, personally, would not go public with this discovery.

Anybody familiar with the rule of compound interest would know this huuuuuge. Try calculating the ROI over one season in La Liga (38 games per team). You will be shocked!

The impact of such a performance would not go unnoticed by the betting companies (if you think they don't have people looking for ways that would improve their model's performance, think again!). If something out of the ordinary happens on the market (like large amounts of money would be placed on a specific, high-profit, outcome) they will adjust the rates. At some point, the prediction made by the mavericks would not be actionable anymore, because the newly adjusted payouts are no longer profitable.

Not to mention the fact that, for the price of a subscription, the big companies would have access to these “better” predictions. Given the fact that the betting companies’ profits take a cut of each bets, using an improved algorithm would reduce the chance of losing money with “surprise” results.

The betting companies are pretty accurate in their predictions too. Below is a table with the predictions from the 2015-2016 of La Liga

HomeTeam Win at home Draw Lose at home
  probability actual probability actual probability actual
Almeria 6.6 6 5 6 7.4 7
Ath Bilbao 9.9 13 4.6 4 4.4 2
Ath Madrid 13.2 15 3.5 4 2.3 0
Barcelona 15.5 16 2.2 2 1.2 1
Betis 7.2 5 5 3 6.8 11
Espanol 7.9 7 5.1 5 6 7
Getafe 7.2 6 5.1 6 6.7 7
Malaga 7.5 8 5 2 6.6 9
Osasuna 7.1 6 5.1 5 6.8 8
Real Madrid 15.1 16 2.5 1 1.5 2
Sevilla 10 11 4.6 4 4.5 4
Valencia 10 10 4.5 3 4.5 6
Valladolid 6.7 6 5.1 9 7.2 4
Villarreal 10 9 4.6 6 4.5 4
Levante 6.2 7 5.1 6 7.6 6
Sociedad 9.5 11 4.7 4 4.8 4
Granada 7 7 5.2 1 6.9 11
Vallecano 6.5 8 5 0 7.5 11
Celta 7.4 6 5.1 7 6.5 6
Elche 6.6 6 5.1 8 7.3 5

prob = cumulative probabilities according to the bookmakers
actual = the actual number of wins/draws/losses

Now, you may ask yourself: "If betting companies have such good models how can I make money through betting?". The bookmakers operate with large numbers. They can afford to loose on some games or some bets, because they win on others. You have to be smart about the bets you place. And you cannot do that by looking at the same numbers the machine learning algorithms are looking. You need smarter stats.