EPL Betting Strategies Examined

This is part of a series of pages related to EPL Away Wins:

• Exploratory Data Analysis
• Data Modelling
• Betting Strategies (this page)

Using prediction data from a Logistic Regression model for Bets and Lays for Away wins, we will examine whether a betting strategy could be profitable. This study uses the prediction results from 1279 EPL matches from Sep 2018 to Mar 2020.

Change in Profit over Time

Away Win Bets

If we had placed an Away win Bet on all games where the ratio of my predicted probability / probability implied by average odds was > 1.8, we would have seen the following change in balance over time for one dollar bets.

The results for Bets demonstrate that the majority of Bets are lost, but there are a few highly profitable wins, with several occurring after 2019 that would have led to an overall profit. It may be difficult to implement a betting strategy that is so reliant on these infrequent high-value wins.

Away Win Lays

If we had Lay all games with a prediction to implied odds probability ratio greater than 1, we would have seen the following change in balance over time for one dollar wins.

Lay bets are mostly wins but any one loss has a significant impact on profit.

Combining Bets and Lays

If we combine these rules for Bets and Lays, we would have seen the following change in profit over time for this data.

We are still highly dependent on the infrequent successful Away Win Bets for our overall profit, however the successful Lays counteract the many losing Bets. Note that a significant starting balance would be required to fund these Lays. A strategy that takes into account the starting balance and risk is investigated in the next section.

Weekly Betting Strategy

What would a betting strategy look like? If we assume a $1000 starting amount, we can define a maximum risk for any particular week, e.g. 10% of our current balance. We can then make Bets and Lays based on the probability ratio limits discussed above, with the amount risked in any week based on the balance available in that week.

If we were to risk 10% of our current balance every week, we could have eventually turned our initial 1000 into 1949 dollars over 4 years. However, it would have taken about 2 years to show any profit at all.

Again, it should be stressed that these outcomes were reliant on choosing the optimum values for the applied probability ratio limits for this data. The profitability for future data using the same limits is unknown.

---

End