Mercurius Sports Trading Methodology

Mercurius focuses on finding and exploiting inefficient odds in the betting market, primarily betting exchanges. We use a combination of the following in our sports trading strategy:

Our automated sports trading strategies revolve around the process of value betting, the sports betting equivalent of value investing. We receive data from a partner firm (Wyscout) and use Artificial Intelligence (AI) to estimate the true odds of the possible outcomes of upcoming sporting events. Our algorithms then scan the market to identify overpriced and underpriced bets on the Betfair Exchange.

Once we find these value bets, where we have a predetermined minimum edge and enough confidence in our estimations, our software automatically places the wagers according to predetermined criteria for each league it monitors. We firmly believe that automated sports trading is a viable means of providing steady, long-term profits. Moreover, we believe that an alternative asset class could rise from our technology and create an investment opportunity akin to the stock exchange.

Our Specific Approach to automated sports trading is as follows:

• Use of Big Data: A complete emphasis on gathering, filtering, cleaning, and analysing granular sports data to estimate true probabilities of future sporting events.
• Automation: To ensure that the process, from data analysis to order execution, can take advantage of every opportunity and that it requires no effort for the end-user.
• Market Neutrality: We do not solely depend on the odds as set by bookmakers to make forecasts. Our software calculates fair odds, which reflects the actual gameplay of teams and their ratings.
• Scalability: We focus on the exchange market and do not require liquidity movements from sharp bettors to create an odds ‘benchmark.’ Also, our clients do not have to worry about bans or restrictions. Our ability to profit adapts and grows with the market.
• Long-Term Profit: Unlike strategies based on odds movements or market errors, Mercurius is designed to last as long as the sports betting market exists.

At present, we look at ‘back’ bets on the Betfair Exchange’s football market because it has the largest liquidity of any comparable market. However, we are currently developing and expanding our strategy. Potential future developments will include:

• The addition of ‘lay’ bets on the Exchange.
• Partnerships with additional exchanges.
• Other markets such as over/under goals.
• The Asian betting markets.

Fundamental value betting

In the betting market, the only scientifically viable method of attaining long-term profit is via spotting inefficiencies in the offered odds. First and foremost, we utilise the Betfair Exchange. It involves the participation of various decentralised participants in what is effectively a free market of odds.

Although on average, the odds offered in the exchange are more efficient than the ones provided by ‘Traditional’ bookmakers, Mercurius’ software finds it easier to operate on the exchange market due to the larger liquidity available and the lack of restrictions.

Using Fundamental Analysis to Find Value Bets

In the world of investing, fundamental analysis relates to the process of measuring a security’s intrinsic value to find its ‘fair market price.’ The practice of value investing involves seeking out undervalued stocks through a detailed analysis of factors such as the company’s Earnings per Share, Price to Earnings Ratio, and Dividend Payout Ratio.

The Mercurius value betting approach is the sports wagering equivalent of value investing. We also use fundamental analysis. In this instance, we implement it to uncover cases where the market has underpriced a specific bet.

Other sports trading strategies of this nature rely on price movements caused by sharps and copy the existing odds. In contrast, we use a combination of Big Data, AI, and sophisticated proprietary mathematical models to determine what the true probability of an outcome is. We do not rely on available odds. Instead, we determine whether a bet offers value or not from the actual sporting data.

There are several advantages associated with the use of fundamental analysis to find value bets:

1 - Not Dependent on ‘Soft’ Bookmakers & Bettors

So-called ‘soft’ bookmakers target casual players and offer more attractive odds as a marketing strategy to acquire this type of audience. However, they also routinely restrict the accounts of winners. ‘Sharp’ bookmakers set odds closer to the true probability. While they offer less of an ‘edge,’ they usually do not close accounts.

Fewer than 5% of market participants are ‘sharp’ bettors. They usually compare the odds provided by ‘sharp’ and ‘soft’ bookmakers and use the latter when they find a potential value bet. This opportunity arises because the odds offered by the ‘soft’ are higher than the ones offered by the ‘sharp’. While this is a profitable short to medium-term strategy, restrictions and bans mean a sharp bettor’s avenues become closed over time.

The odds our software provides are unique, and not dependent on the errors of ‘soft’ bookies, or the movement of ‘sharp’ bettors. We can also exploit the mistakes made by ‘sharp’ bookies who don’t ban profitable customers. As we operate on the exchanges, our clients are at no risk of account restriction or closure. As a result, our strategy has significant long-term potential.

2 - Market Neutrality

The Mercurius approach ensures that we don’t need a change in odds to estimate true probabilities. The algorithms we use identify whether there is a value betting opportunity long before the market reacts. As we identify the fair market odds for each event, we are in a strong position to profit from the overreactions of the other participants.

In sports betting, it is common for bettors to ‘follow the money’ and place bets on events where the odds are rapidly falling. We do not get involved in this process as we have already determined the fair price. The nature of our strategy means we do not get sucked in by market movements.

On occasion, these rapid market movements will open up a value betting opportunity that did not exist shortly beforehand. As a consequence, our approach is sustainable in the long-term.

3 - Scalability

The process of using fundamental analysis means we can permanently exploit market inefficiencies on betting exchanges. The practice of finding value betting does not change. It always involves finding the real probability of an outcome and finding occasions when the market underestimated it.

Without further improvements in our sports trading strategy, we can handle £3 million on the Betfair Exchange alone, a fact that ensures we can grow the liquidity managed by our bot exponentially. As we focus on major events and competitions at present, there is no shortage of liquidity in the market.

Finding value: our process

Mercurius adopts a tried and trusted 3-step approach to determining value bets. One of the significant strengths of our process is its emphasis on data, rather than sentiment or bias. Our AI algorithms ingest and analyse data 24/7.

Next, the software spots market inefficiencies and automatically executes the orders. Furthermore, the process of machine learning ensures that our AI constantly updates its projections for the future. As a result, it makes the necessary alterations to improve its accuracy.

1 - Use of Quantitative Analysis to Compile Odds

We are in a partnership with Wyscout, a professional football platform that provides data from over 200,000 football matches around the globe. Our AI gathers and analyses approximately 20,000 data points for each one of these historical football matches. It weighs the strengths of individual players and football teams.

Examples of the data we analyse include:

• The starting and finishing position of each ball movement.
• The types of ball movements, whether it is a cross, shot, or header, for example.
• The outcome of each movement.
• The players involved in each ball movement.
• Which body parts the players used to pass the ball.

Our software continually monitors the data received to determine anomalies. An example of an outlier is if a player scores a goal from an extraordinary position such as the halfway line. We also check for glaring errors by watching the data provider’s video. If we detect any mistakes, we alert the data provider immediately and ensure it does not negatively impact our information.

Team Ratings

We use Monte Carlo simulation to establish team ratings and also to find fair odds. Monte Carlo involves running data through hundreds of thousands of simulations to see all the possible outcomes of a decision.

Our software assesses every shot available and sequences of passing in the historical record to determine the likelihood of them resulting in a goal. The result is an Expected Goals (xG) rating, which we use as the basis of establishing a team rating. Next, we utilise a modified Dixon and Coles model to generate an offensive and defensive rating for each team.

Mercurius uses a Bayesian hierarchical model to update ratings after every match. As AI continues to learn, it retrospectively changes xG as new data becomes available.

Generating Fair Odds

After the AI gathers ample information about both football teams in a contest and has appropriately adjusted their respective ratings, it begins to generate fair odds by calculating the actual probability of an outcome. Mercurius simulates a match between the teams 100,000 times. The more times the teams play, the more precise the prediction.

2 - Finding a Positive Expected Value (EV)

Our AI scans the Betfair Exchange’s odds every 20 seconds in the pursuit of a positive expected value (EV). EV is a measure of what you can expect to win or lose on each bet placed at the same odds over time. A negative EV means you will likely lose money in the long run; a positive EV means you will likely earn a profit.

You can only obtain a positive EV if the odds are better than the true probability of an event. For example, the chance of correctly calling heads on a coin flip is 50%, assuming the coin is not biased. In betting terms, a 50% probability equates to odds of 2.00 or Even Money.

If you bet €10 on a coin toss with the chance of winning €11, you have a positive EV. If the return is €9, you have a negative EV, while a €10 return has neither a positive nor negative expected value.

Likewise, if Team A has a 40% chance of winning a football match, its true odds should be 2.50. If you get odds of 2.70, for example, you have a positive EV. As the exchange consists of alternative investors and offers different odds and varying levels of liquidity, there is a higher chance of finding bets with a positive EV, also known as value bets.

3 - Execution

Once the software discovers a value bet, it executes the process automatically according to a specific staking plan. There is a minimal delay between the uncovering of a bet with a positive EV and the placing of the wager.

In summation, the AI will decide:

• How much to stake.
• The right match and outcome on which to place the wager.
• The odds weighted according to the liquidity in the market.

The staking plan

Although each bet is estimated to have positive EV, there is an inherent risk in the process. The only thing you have under control in this situation is how you expose yourself to this risk. As a result, you should design a sensible staking plan based on your preferred level of exposure.

Mercurius has decided that its trading strategies should be exposed to this inherent risk in a way to minimize the risk of ruin function (ROR). Mercurius has tested a wide variety of options and settled on Kelly’s Criterion capped at 1% of a client’s bankroll.

In an unpublished paper titled: "Gambler’s Ruin, Large as a Basis for Precautionary Strategies," Nassim Nicholas Taleb and Aaron Brown said that "if you bet more than Kelly amount, you will eventually be ruined."

The main risk attached to following Kelly’s Criterion, therefore, is the tendency to overbet. By sticking with 1%, you significantly reduce the likelihood of betting excessively, thus preserving your bankroll.

Also, by capping it, bankrolls are more protected from unforeseeable risks (black swan events). This staking plan also implements ‘compounding,’ which ensures maximum profitability in the long-term.

All bets are placed according to a specific ‘confidence interval,’ which varies depending on the league. The AI does not act until it has gathered adequate information during a new season. For example, it did not bet on the English Premier League in the 2018/19 season until 100 matches were played. Improvements in the AI meant that it had sufficient confidence to bet after 60 games in the 2019/20 season.

Mercurius’ AI only makes a bet when it discovers a minimum ‘edge’ over the market. For reference, that edge is 5% in the English Premier League and 10% in France’s Ligue 1.

What the future of sports trading holds for Mercurius

The Mercurius R&D team consists of professional sports traders, PhDs in Mathematics and Statistics, and data scientists. We continuously monitor our strategies and work tirelessly to improve upon them, and the technology we use.

Mercurius’ reach includes constant dialogue with academics in many of Italy’s top universities:

• Polytechnic University of Turin
• Polytechnic University of Milan
• Bocconi University
• University of Padua
• University of Bologna
• Catholic University of Milan

Although the Mercurius process relates only to ‘back’ bets on the 11 leagues mentioned above, we plan to expand shortly with the following innovations:

• Lay betting on the aforementioned leagues to double liquidity, increase investment options, and provide hedging opportunities.
• Opta data feed integration. We will achieve this by cross-checking Wyscout, and Opta data feeds. The result will be an even higher rate of accuracy for the AI.
• A better reading of market changes to improve execution and reduce the slippage involved in trades.
• Establishing partnerships with other exchanges and involvement in the Asian markets to further improve our access to liquidity.