Our Algorithm

Our algorithm, Algol88, generates predictions on the outcome of global soccer matches and compares them to the odds generated by bookmakers. Its predictions have been back-tested on 5 years of historical data and shown strong performance in identifying value trades. This back-testing was subjected to an independent audit which examined the underlying assumptions and assessment of risk. 

Algol88, which is based upon artificial intelligence, makes value trades based on risk assessment and probability. The algorithm refines itself with each result so, by design, it continuously learns and develops.

Key to Algol88’s performance is the ability to place strategic trades based on value. For example, if FC Barcelona faces Malaga (top versus second from bottom in La Liga at the time of writing), then the model does not simply back the team it suggests will win. It uses statistics across all the major global leagues and the teams therein. It then calculates a probability and associated risk for the main betting markets (i.e. those with the greatest liquidity) and takes advantage of where Algol88’s assessment of the probability of a particular outcome is greater than those implied by the bookmakers’ odds.

The trading team compile in-depth news and statistical analysis on world soccer, complimented by information from extensive contacts within the sport itself and the media. This information, supported by the algorithm, allows the trader to place trades at the best prices available.

The algorithm then makes use of the liquidity in global soccer markets to assess where trades should be placed. Working closely with major European and Asian bookmakers, the trading team uses the results to maximise profits and reduce trading risk.

Along with the Wiener process (Brownian motion), a poisson is a central stochastic (random) process run in continuous time. The Markovian properties of Brownian motion have become interesting foundations for much research into modelling financial assets and portfolios as they allow for prices to evolve continuously with time. However, football also lends itself to an adapted form of modelling by Markov processes. Our back testing and real time results show great success when using this technique to estimate transition probabilities.

The design of the staking algorithm reaches long term optimality in terms of magnitude of bets. The basis is the Kelly criterion which has been widely researched in investment and betting. It is consistent with an alternative principle of maximising logarithmic utility. However, in the purest form, it has been criticised for effects on volatility and has been accused of creating non-deterministic errors (different behaviours in different runs). We build in an algorithmic parameter that has a strong dampening effect and this can be adjusted to meet strategic return objectives.

Although there are some great similarities between financial and sports trading, there are differences. A sports match does not run perpetually so losses and gains do not really run on a continuous function. The outcome results in a total (or half) loss of the bet, or a win. Analysis actually indicates a positive (profitable) skew to returns and there is less kurtosis than a standard normal distribution. However, the staking method further contains an adjustment for any possible thick tail in returns. Financial modelling using a normal distribution model of risk has been severely criticised in recent years. This approach estimates a finite variance on a finite sample size.

Some economists such a Nassim Taleb have become cult figures following the last recession. Taleb recognised that these models ‘underestimate the true degree of predictive difficulty (and of risk).’ Despite the analysis of football returns, to overcome any shortcoming in this regard, we model a balance between a Brownian motion and couchy distribution. It is a stable distribution that can handle traumatic real world events, and also any leptokurtic clustering possible in football results.

Accordingly, we present a highly profitable trading algorithm that is specifically designed to match the complexities of sports matches. It uses optimal staking and can be adjusted to fit the risk preferences.

This analysis is based on back testing not actual trading history over 5 years.
Past performance is not a reliable indicator of future performance. You should not rely on any past performance as a guarantee of future investment performance.