Machine Learning in Bookmaking

Stephen Rothwell
FansUnite
Published in
5 min readOct 18, 2017

Machine Learning is becoming a standard tool of the sports betting industry. At fansunite.io we are keenly aware of this technology and actively incorporating it into our risk management strategy. This is integral in our platform being able to provide a truly winner’s welcome sportsbook.

Although many sports betting firms are beginning to build a machine learning stack, the reality is that the sharpest punters have been using machine learning techniques to make money since 2011, when deep learning networks first started to become popular. To combat ever sharper punters, bookies inevitably resort to copying Pinnacle Sports or Betfair opening lines. This is a great strategy because they utilize a model based on high volume which is well known to eliminate inefficiencies in the closing line.

Crowd Wisdom

The prevalent method to set odds is simply averaging the models of your sharpest customers on inefficient early lines. This is then honed by subsequent increases in volume from less sophisticated bettors until the closing line is reached. The very friendly PinnacleSports media team does an excellent job explaining the process here.

The key factor here is that a slow ratcheting up of limits and volume balance out inherent bias and variance in the bookmaker’s lines. This approach quickly reduces exposure (Fig.1). Because of this reduction in risk, they can offer very large closing line limits and welcome arbitrage bettors who are unfairly banned from most online betting sites.

Figure 1. Bias and Variance

Sometimes a bookmaker’s opening line is inefficient because of bias in the calculation. Smart bettors quickly take advantage and the bookmaker shifts the line to equivalate betting volume on either side of a matchup.

Similarly, high variance in opinion when the data between two teams is very similar can often lead to poor lines. By polling the crowd with low limits to start, Pinnacle can often limit exposure on early lines and avoid getting picked off on markets by sharp bettors.

This novel method of polling the crowd drives lines globally, and it’s no surprise that the default action for almost every sportsbook is to use Pinnacle or Betfair’s API to limit early exposure.

This approach sadly leads to a rise in limiting and arbitrage player banning because the vast majority of bookies don’t know how to create good lines. Smart punters know how to leverage Bayesian modelling and Deep Learning to create a 2–3% ROI on the dollar bet in their chosen sport. The Risk Management teams at most of these bookies simply don’t compile odds or trade because it’s more efficient to player profile and ban winners to increase profitability.

A New Approach

In 2009, the Netflix prize outright proved that the concept of combining two approaches to a probabilistic problem almost always provides superior results. Just like crowd wisdom shapes lines for Pinnacle Sports, an ensemble of weak machine learning algorithms shaped a million-dollar windfall for the BellKor Team. Most of the sharp syndicates caught on and use more sophisticated methods to get the tradeoff between bias and variance correct.

Machine Learning turns out to be a powerful approach to producing win probabilities which minimize bias and variance with respect to making profit. The idea is to produce a win probability that is “just right” and shape the line with data alongside the opinion of individual players (Fig. 2).

Figure 2. A simple example of trading off bias (underfitting) and variance (overfitting)

This is the approach that we utilize at fansunite.io. To produce lines we will use an ensemble of best in class Deep Learning networks, alongside other more common approaches to shape a line up to 24 hours before current markets take shape.

At Fansunite.io, the world’s preeminent social token betting platform, we have been actively shaping our risk management strategy to integrate machine learning into the setting of our lines. We offer an industry leading 1% margin and will maintain a winners welcome philosophy.

The Value to the Betting Customer

Our automated machine approach to setting lines offers the following core value to our customers.

  1. Savings we can pass on to our bettors. At fansunite.io we confidently automate the shaping of our lines using an ensemble of Deep Learning calculations. This means our risk management team is able to run super lean and efficiently, requiring only a few risk managers. Savings on overhead are passed directly to the customer in the form of industry leading margins.
  2. We welcome winners and arbitrage bettors. Most bookies don’t like arbers because they bet massive amounts, which can do damage if you are left holding the bag. With our machine learning stack we are extremely confident in the accuracy of our lines. We make sure the other side who is doing a sloppy job setting odds is holding the bag.
  3. 24/7 In-Play betting is in high demand but it is hard to set lines because there is no way to crowdsource odds quickly enough to limit exposure to sharps. By using Machine Learning, we can offer real time In-Play betting markets to our amazing customers.
  4. Stable Currency: Solid lines offer big rewards to currency and token holders by ensuring that the coin base is not drained by sophisticated traders and demand remains strong for our low margin lines. Any coin who is not using the same level of technology as the very best bettors will quickly be gutted.
  5. More Markets: Machine learning allows us to offer a traditional sportsbook environment, our competitors in the token space are peer to peer and won’t offer common bet types that we all know and love. Because we generate our own lines algorithmically we will ensure liquidity in popular markets.

Conclusion

Bettors want to maximize their available markets and bet at a low fee without any egregious limiting. FansUnite is leveraging machine learning to do just that and offer a competing opening line which is friendly to even the most advanced punters in the marketplace. To this end, Fansunite is looking to grow our exceptional data science team, please get in touch at info@fansunite.io to learn more.

--

--