Skip to content

Latest commit

 

History

History
49 lines (29 loc) · 4.28 KB

README.md

File metadata and controls

49 lines (29 loc) · 4.28 KB

Final Project Findings

This project was undertaken to assess the profitability of the UK football betting market using the accumulator betting method. Unfortunately, due to my university's regulations, I cannot publish the code here, although I can explain the method and the findings briefly.


Contents


Method

Accumulator betting is the betting format on which you combine single-way bets into one large bet which pays off only if all the component bets also succeed. If one fails, the whole accumulator fails. This was investigated due to the severe lack of research into accumulator betting, and so I thought it had potential to provide large profits.

There exists a plethora of existing strategies for single-way bets, so my idea was to combine different strategies for single-way bets into making an accumulator. For this, I selected 2 different strategies, which I labelled ST1 and ST2. For these strategies, you can visit the papers by Lisandro Kauntiz et al. and Edward Wheatcroft. So, I would use these strategies to select bets from a gameweek, and then further strategies (below) would compile certain bets from these selected bets into an accumulator.

I developed four different strategies to form an accumulator from the bets selected by the strategies above. They were used with a range of different accumulator sizes to accurately assess the profitability. The different strategies are encompassed in the grid below.

image

  • Confidence : Uses confidence parameter to order bets by confidence
  • No Confidence : Choose randomly without confidence parameter
  • Priority ST1 : Prioritise Strategy 1 bets
  • Priority ST2 : Prioritise Strategy 2 bets

These used a confidence parameter which I derived using a Neural Network, which was trained on previous match odds data and outputted a single float value representing confidence in the selected bet. E.g., if ST1 selected a bet, its data would be run through the ST1 confidence neural network and output a confidence value in the bet suggested. These confidence values were essential to then forming an accumulator from these bets.

Results


betting without kelly


The above graph demonstrates how profitable this idea turned out to be (NOTE: this is all in a simulated environment, i.e., run on past game data. However, there is no reason it could not be implemented quite easily in real life!). Using just £10 stakes on each accumulator, we see betting on low size accumulators seems to provide optimal returns, although if you get lucky, a higher size accumulator can also be very profitable. For consistent profit however, accumulator sizes of 2 or 3 is recommended.


betting with kelly


This graph shows the same but instead of £10 stakes per accumulator, I implemented a bankroll strategy known as the Kelly bankroll management strategy. For more info on this, see here. It uses the fact that odds set by bookies are a relatively accurate measure of probability of event occurring. Using this bankroll management strategy, less profits were realised, but the profit was much more consistent, and all sizes of accumulator profited. This shows that this method of accumulator betting could potentially provide profit with little to no risk of a loss.

Conclusion

This project of mine is the one I am most proud of, mainly because it has the biggest implications in real-life. Although it is untested, the evidence is there to suggest accumulator betting, when done right, can be incredibly profitable and consistent. This goes against literature which would suggest accumulator betting is risky and not as profitable as single-way betting, but my research suggests this may be untrue. Either way, it proves this area is drastically under researched, and definitely warrants further investigation.