For the Sophisticated Predictions Contest for Euro 2024 I have had an attempt to build a prediction.
There is information in bookie odds and often the most accurate probability of a match result between two teams is in the Closing Match odds at time of kick off. At least if you know a better prediction you will be rich. But for the rest of us non-professional bettors and even most professional bettors the closing line will be the best prediction for the probability of a football match that you can get.
Given the probability we can easily deconstruct the relative ratings of the teams. On a basic level the favourite is better team than the underdog - and the shorter their odds the bigger the gap in the ratings.
So I have a basic ELO-like system, but it is based on the closing match odds rather than the result. Using odds over result I think the ratings converge on something more realistic and do it much quicker.
For example England played Germany in the Nations League. My ratings for each before the match were 1285 + 45 for England + home advantage and 1306 for Germany. This is a difference of 24. Now England by the closing odds were 41% chance to win and 27% to draw and 32% to lose. By the ELO equations this would mean that the market was rating England 35 ratings points more. So what I do is split the difference and adjust my current rating for both teams.
Method explained in more detail here.
To that end I have loaded my spreadsheet with the Euro 2024 Qualifying odds, the Nations League 22/23 odds and the odds that are already released for the Euro 2024 Group Stages. I seeded the countries with realistic ratings and then match by match updated the ratings based on the difference between the match odds implied rating and the current ratings. It is not alot of data points for each country (eg England 16 matches and Germany who had no qualifying it is only 9 matches). But the below ratings I am happy enough with.
Match day odds implied team ratings for teams in Euro 24.
The ratings above are just inputted into a Monte Carlo sim.
Odds to reach various stages for countries in Euro 24 based on market implied rating.
So these are the predictions I submitted for the contest.
Some notes:
1. The market also has outright predictions. They wont be as robust as the match odds ratings but do add some information. If I had I more time to develop my predictions I would like to adjust the above predictions towards the outrights markets. Although the match odds ratings will be king, there will be some teams that have more strength in depth or have players injured going into the tournament. Information like this will be captured in the outright odds but not the match day odds. So it is worth using. But in general the outright odds probably have more biases and less liquidity.
2. The match day odds imply a rating for the team but it is based on the line up on that day. If the latest match for example was a dead rubber in the qualifiers and a weak line up was put out then the team rating may have been adjusted incorrectly. I am mitigating this kind of thing I have used the odds too already available for the Group matches - so they should be very representative of the expected lineups etc.
3. There are parameters that I could have tweaked and adjusted more. For example the home field advantage. I have read that Croatia will have strong support, England generally have strong support wherever they go too. Perhaps HFA isnt a binary +50 for the home team but more a sliding scale. The predictions dont consider this.
4. The Sim requires running a simulation of the tournament and after each match updating the team ratings based on the result of the match. This is important especially for a small team. If Georgia travel far in the tournament they will be getting a ratings upgrade as they go along. This parameter is usually the k factor in an elo based rating. This I could adjust and looks like I should. I used a fairly standard value of 20 for the k-factor. I also think that the k-factor could be different for each team. For example a very well known team like England maybe you don't want to update their ratings as much based on a few random losses, but an unknown team like Georgia could need to be tweaked more if they start going on an unexpected run.
For comparison - the market odds:
1. I am longer on England than the market. This might just mean that England will rotate more in the Group games, or that they will have another level they can rise to in the knock out games compared to the level they need for Group and Qualifiers which my ratings were derived from. Or it could mean a bias for England in the market - but would be very wary to jump straight to this conclusion when there could be other reasons.
2. I am a bit bigger on Germany too. With Germany not playing any Qualifiers and I decided to not use friendly data there is somewhat a lack of data points for the Germans. So I wouldnt be as confident on their ratings.
3. The mid range teams and underdogs I seem to be a little shorter on than the market is. As mentioned I think that this is related to the k-factor and how much adjustment the sims make based on match results (a bit like momentum). It looks like I should dampen this a little. I am reluctant to. International football is a lot more unknown than club football for example. I think the ratings at the start of the tournament can be taken with a pinch of salt and should be quickly adjusted if a team seem to be getting many unexpected wins. The markets seems to think the opposite.
4. I have been really lazy with the sim especially for the 3rd placed group teams. Its probably no big deal but if you can spot a bias because of that then be interesting how it could manifest. I believe that the way I assigned the 3rd place teams in the R16 (just basically randomly rather than applying the matrix based on which groups they come from) wont make a difference and will balance out.
Comments
Post a Comment