*Disclaimer – the following is for information purposes only. Any further action taken on the reader’s part using this information is at his/her own risk. This is article looks at past results and is not an indicator of future performance.


This season I decided to take a look at ELO in the Premier League. I started each team with 1500 points to see how well they improve/decline from that starting point as the season unfolds. I have written about it a few times on this blog and you can find those articles here.

Added to this, I thought it would be interesting to see how well one would do when using these ELO figures for sports betting. In this article, I will outline the results of this analysis.

This is very simple system and only requires up to date ELO ratings to implement it – how to calculate ELO ratings is another matter entirely and far beyond the scope of this article.

So, before we dive in, here is a look at the EPL’s currently ELO table:


The “System”

I am not a fan of systems. In my opinion, having a solid method for calculating odds and looking for discrepancies in the market is the way to make long-term profits in sports betting, but nevertheless here is a system for betting on the home team in the EPL.

I wanted to keep this as simple as possible and there are definitely areas where this can be improved upon, but the way the bet selection process works is simply by taking the away team’s figure away from the home team to find the rating difference. As an example let’s look at last week’s game between Stoke City and Crystal Palace:

  • Stoke had a rating of 1504 before kick-off
  • Palace had a rating of 1372 before kick-off
  • The difference in rating was 131 points

This fulfills the first criteria: the difference in rating is more than -(minus)60 points.

The second requirement looks at the odds available (for this analysis I used average odds taken from a selection of bookmakers which can be found at Joseph Buchdahl’s football-data.co.uk).

Looking at the data, the best results are found when using minimum odds of 2.0 (50% probabilty). If the home team has higher odds (lower probability), it is then an eligible bet.

Stoke were offered @2.1 (47.6%) so an imaginary bet was placed on them to beat Palace.

Stoke won the match 1-0, but it is important to stress that this system is NOT intended to predict winners or to calculate odds, just to determine possible betting opportunities.

The Results

At first I looked only at results from this season as this was all I had calculated for the EPL. This shows that there have been 86 eligible bets so far this season when using our criteria of higher than -60 rating difference and minimum odds of even money (2.0).


Starting with a bank of €100, a total of €694 would have been staked across 86 bets for a profit of +€258.

The bank total is updated every 10 games or each gameweek and 5% of the total is used to bet on each recommended selection. If we have a total bank of €100 in GW1, the stake for each bet that week is €5 (5% of 100). If our total bank in GW20 is €200, our stake for each bet that week is €10, etc…

The yield and return on investment for this season is very high. Too high. This is due to the very small sample size available.

RRR is the Risk Reward Ratio, and Profitability measures the success of the system by dividing Profit/Loss by the total amount lost.

Bank High and Bank Low are the highest and lowest points the bank has been this season when including our original €100.


This chart shows how the system has done so far this season. The x-axis shows the total games played this season, not the number of bets, the y-axis shows the profit and loss. The red line that climbs higher than the rest is included as a warning to those looking at how well their systems/betting does or would do. It is the profit/loss this system would make when using 5% of your bank as a stake after each result. I see this a lot online and it really grates me. One has to remember that games very often take place at the same time so you cannot use an updated bank balance after one game to decide on the stake for the next as the result is not yet decided. As these matches happen concurrently, this leads to misinformation which may have a serious impact on one’s betting strategy and bank.

The dashed grey line is there for comparison. It shows the P/L one would have made when using level stakes of €5 regardless of the size of the bank. The green/red area is the one we are most interested in. This shows the P/L for this season when using 5% of the total bank which is updated every gameweek.

For this season, the results were even better when using a lower odds threshold. Here are the results when using minimum odds of 1.57


This increases the bet count to 124, but again the sample size is too small to make any solid conclusions.

I couldn’t leave it at that, so last night I calculated ELO ratings for the EPL from 2012-2016 to get a more realistic picture of how this system performs. I was quite surprised by the results.



I calculated the ELO ratings from 2012-Present in two ways in order to compare them. The above graph shows the results when starting each team on 1500 points every season. Again, the red line is there to show results when updating the bank after every bet, which cannot be done. The dashed grey line shows the P/L when betting €5 per recommendation regardless of bank size. That would result in a profit of almost +€450. Our green area once again is what we are really interested in. If one had bet on each match this system has recommended since 2012, they would be in a very healthy profit – +€2,993 with a ROI of 17.48% and a yield of 17.58%.

The number of bets has greatly increased to 600 which is better for our sample size, but I would like to have more. Here are the results when the odds restriction is lifted:


When we allow for any price, the number of bets placed is more than doubled. The strike rate goes up, but our ROI, yield, and profitability all take a hit. In saying that, 8% yield is very good when looking long-term and shows that this system is capable of making a profit. With no odds restrictions, the time spent in a loss is increased and I think what may be more import for this system as a whole is patience and discipline. If you look at the first chart of this section again, before the profits skyrocket around match 1600, there is a period of heavy loss. That is where the 5% weekly adjust stakes has its advantage. The amount that is bet varies depending on the size of the bank. I chose 5% to keep it at a round number but for the more risk averse, a maximum of 2.5% is definitely recommended.

Out of curiosity, I decided to see how well teams do when using a continuous ELO – taking final points from season 1 and using them in season 2 etc.. The results were not good.


Using this method would have resulted in a loss of -€90, leaving just €10 in the bank. About enough for a pint and chips to drown your sorrows.

In saying that, I think using a continuous ELO may give a truer picture of team ability in the long term than simply relying on season to season figures. While I believe seasonal ELO is better for this system and for examining a football season in isolation, I think using a continous ELO has other advantages. It allows more accurate predictive analysis, and gives a long-term view of how teams are performing:


I chose Chelsea and Leicester to highlight an extreme. Last season was remarkable. Leicester won the league with the lowest ELO rating of a champion I have in my numbers, while Chelsea struggled right the way through. This season things appear to have righted themselves with Chelsea performing almost at a record level (highest ELO was 1812 back in the 2014/15 season. Leicester City on the other hand, have fallen past the default starting rating of 1500 points and are bang around the average mark.

Continuous ELO may make for more accurate odds derived from logistic regression too, which I will have an article on in the coming weeks.

Room for improvement

Is this Gambler’s Fallacy? Possibly. It’s always difficult to look at the past to make judgments for the future in football. In saying that, if we look at each season in isolation, only 2012 finished in a loss. This would be the Profit/Loss if each season began with a bank of €100 rather than rolling it over to the following season as in the examples above.


I feel that this system would be improved if one calculated their own odds and only bet on value selections using the Kelly Criterion.

I’m still very skeptical of this. But I feel there may be something here – albeit not to the huge profits seen from 2012-Present, but enough to pique my interest. I will test this out of sample next season looking at the Bundesliga and Premier League and post updates throughout the 2017/2018 season. It may well perform as it has in the past, it may do really really poorly. Only time will tell.

If you have any questions or anything to add/suggestions feel free to leave a comment below. You can also reach me on twitter @petermckeever


**NOTE: An earlier version of this article had incorrect figures for 2016-2017 data, and the first chart of the section 2012-Present. This has been rectified.