In this section we are going to overview in simple terms how Mercurius finds potential value bets through statistical learning, which is a recently developed area in statistics that blends with parallel developments in computer science and, in particular, machine learning.
The sports betting market is usually associated with emotion and irrationality, although the opposite is true: it is indeed heavily math-powered. Thus, adopting a scientific approach is the only way not to be fooled by the sneaky concept of bad luck.
The (Big) data revolution has enabled predictions in every field. In layman's terms, a predictive model is a matter of estimating useful quantities from past data. As far as Mercurius is concerned, the useful quantities are outcome probabilities, estimated on the basis of historical results and statistics.
A theoretical result of statistical learning is that simplicity wins over (useless) complexity. In football, this means that trying to use as much information as we can possibly think of (injuries, news, line-ups, player details, transfers, manager changes, number of supporters at the venue,etc...) often turns out to be an effort that leads to non-statistically significant results. What happens is that we end up not being able to distinguish the signal from the noise. And this is bad.
That's why Mercurius statistical models focus mainly on the most important metric of the Beautiful Game: goals scored. By exploiting established findings of well known scientific papers and by tuning the parameters so that we can adapt the model to different football leagues, we are able to assess:
- the different abilities of the involved teams;
- the home effect, i.e. the tendency (different for each team) to score more frequently and generally to perform better at home;
- the time effect, whose intuitive meaning is that more recent results impact more on the evaluation process;
- the separation between defensive and offensive skills for each team;
- the tough relationship between goal scored in tight/low scoring matches.
At the end of this fully data-driven analysis, Mercurius obtains an estimate of the probability of a match ending as a home win, a draw or an away win. In turn, this projections are filtered according to a confidence criterion, so that we avoid to consider the events where our model is more prone to performing poorly.
The next step is to decide whether it is convenient to bet on a particular event and, if so, how much to wager. As mentioned here, we need to find odds that are long in our favor and therefore we download market odds every half an hour, comparing different online bookmakers to spot convenient odds. Once a value bet is found, before it can be labeled as a great value bet, we check if it satisfies a further set of criteria, which may be presented as follows:
- is there enough value, given how (un)probable is the specified event?
- is the value large enough to cover the uncertainty associated with our estimate?
- would a slight decrease in the odds make the bet lose its value?
Only then, after all these steps, you will find a value bet in your Market page!
So, do not worry if you find none, Mercurius is always working for you.