SportMonks Football Predictions API: Be smarter than bookmakers

In August 2019, we launched the SportMonks Football Predictions API. Our predictions API is based on our own data and uses cutting edge Machine Learning and statistical models.

Our team of data scientists has genuinely delivered an excellent and innovative product. It shows because more than 500 customers have already experienced the wonders and profitability of the Predictions API.

We are thrilled with the number of reactions and feedback we have received so far. This blog will give more insight into how the Predictions API works, what it will provide you, and how you can get started to make profits right away.

We spent many months creating the perfect Predictions API and will continue to do so in order to improve the model, meaning you will get better and more predictions.  For example, we just updated our algorithm with some massive improvements. We expect substantial performance upgrades to many of the League Hit ratios.

Although we are pleased with the model’s performance so far, there is room for improvement for some leagues. But the good thing is, you can easily track which Leagues have a good Prediction score.

Use this endpoint: https://soccer.sportmonks.com/api/v2.0/predictions/leagues?api_token=__TOKEN__

Note that you need to have an active subscription to use our API. Luckily you can get started free of charge by utilizing our 14-day trial.

The rest of the blog will give you a more in-depth explanation of how you can use the Football Prediction API and Value Bet API for your own application.

How to use the Predictions

The Prediction API comes with several features that help you understand and monitor the Prediction model’s performance. The Value Bet API is available as a complementary feature of the Prediction API. They are two different tools that you can use together. I will now explain how to use and understand these APIs.

The Value Bet API

First, the Value Bet API is independent and thus not related to the Prediction API. While the Prediction API evaluates game events’ probabilities, the Value Bet API processes thousands of historical odds data and market trends to find Value opportunities, compared against bookmaker odds.

Once the opening odds are available, the value detection algorithm runs every 10 minutes up to the match’s beginning. Each Value detected by the model comes with a set of features described below.

The stake feature

Each of the Value Bet detected by the model has a stake. The stake helps to manage the risk that the model would take in the bet. It is somewhere between the Kelly (1) bet and the Markowitz (2) theory.

The Value Bet API measures the risk with the volatility of the value bet strategy’s profit and loss. It also calculates the stake to have an average risk of one unit (it could be one euro, one dollar, etc.)

To illustrate, let’s assume one always bets on an event with a 20% success rate. The fair odd in this case is 5. If your stake is 3 units, 20% of the time, your profit is 12 units, and 80% of the time, your loss is -3 units. The volatility and thus, the risk of the profit and loss is 6 units in this case.

The Value Bet API gives for each Value detected the stake that will provide you with a volatility unit of 1 if the match was played multiple times.

This way you own the same risk for every Value in the API found. The maximum allowed stake is 5 units. But it is crucial to keep this scale the same. You can also take the risk of 2 (or any number) units on average. In that case, you have to multiply all the stakes by 2 (or any number). In this example, the maximum stake will be 10 units.

Introducing the fair odd feature

Our algorithm also allows you to find the fair odd of a value. The fair odd is useful to play against bookmakers that are not listed in SportMonks. You can consider any odd above the fair odd as Value. In the previous example (fair odd = 5), random bookmakers who offer odds above 5 are worth taking.

The Prediction API

The predictions API’s goal is to evaluate the probabilities of a given match using machine learning and statistical methods on our data. All Predictions are available 3 weeks forward. This allows you to bet against opening odds for example.

To help you track the performance, the API comes with a league predictability feature. It is a set of metrics that measures the quality and predictive power of the model per league.

– hit ratio: number of times the model predicts the right outcome (1, X, 2) over the last 100 matches of the league. The closer to 1,00, the better. The random prediction has a hit ratio of 0.33 (33%).
– log loss: average log loss over the last 100 matches of the league. This is measuring the quality of the probabilities—the closer to 0, the better. We consider the quality good when it is below -1.02 and high when it is below -0.98. The random prediction has a log loss of -1.09. So, we consider any log loss below -1.07 as poor predictability.
– predictability: not everybody is comfortable with numbers. The predictability tells you in a word if the league has a poor, medium, good or high predictability.
– predictive power: this tells you if the log loss has increased in the last 50 matches or not. It can be up, down or unchanged. When the predictive power is up, the league becomes more predictable.

Conclusion

Please let us know when you have any more feedback or questions for us. Please check here if you need more detailed information about the Football Predictions API. We are putting massive efforts into the Football Prediction API, and we need your input to improve our products day by day. Be aware that results from the past are no guarantees for the future.

>>> I want to see how an API response of the Predictions API looks

>>> I want to start testing the Value Bet and Prediction API

Are you interested in getting more Football Data like Livescores, (live)Odds, (live)Stats, Fixtures, Standings, or any other type of Football Data, for 900+ Leagues? Check here for all our Football offerings.

Good luck!

See the Kelly_criterion(1956).
See the mean variance theory(1952).