International football tournaments are notoriously difficult to predict.

Not only do teams who many not have faced each other for years play each other, often at a neutral venue, but the turn over of players from one match to the next can be significantly higher to what is seen in club football.

Compounding this already difficult situation is the collation of accurate and reliable statistics about teams and their previous matches. While it’s certainly not the same as 50 years ago when players never seen outside of their home country would suddenly appear at a World Cup, but it is much easier to obtain in depth statistics about Brazil vs Germany than it is for Malaysia vs Fiji.

In this case study we will apply two different predictive models to the 2015 Asian Cup to see how accurately they can predict matches with limited data, and compare their accuracy in predicting the match winner to bookmakers.

Selection of sample Asian Cup teams

For the purposes of the case study two of the four groups involved in the Asian Cup were used, with each group stage match modeled and analysed. The groups selected are groups A and D.

These groups were chosen due to the difference in statistics available for the eight participating teams. Tournament heavyweights Japan, South Korea and hosts Australia all have extensive match and player statistics available, while minnows Palestine had as few as five matches to base predictions with.

The two groups also provide an interesting balance of teams who competed at the World Cup in the last calendar year (Japan, South Korea and Australia), through to a side who was required to come through a qualifying tournament (Palestine).

Match prediction models for Asian Cup case study

Two different models were applied to each group match for pools A and D of the tournament. The first, Total Shots Ratio (TSR), hoped to identify the “better” attacking team (and therefore the most likely to win), while the second will use a power ranking system known as the World Football Elo Ratings to calculate match probabilities.

Both TSR and the Elo Ratings will be compared to the bookmaker favourite to see which more accurately predicts the winner of each match. The Elo Rating’s match probabilities will also be used to identify potential value bets throughout the tournament.

Total Shots Ratio

To calculate the TSR of each of the eight participating teams, shots for, and shots against was obtained from Soccerway. For the case study data from up to a maximum of 20 matches was collected from the years 2013 and 2014.

With shot data not available for all matches these were not always sequential, and the total number of samples differed from team to team (for example only five samples were available for Palestine). This data will however be updated through out the tournament.

The application of TSR in a tournament setting compared to a league has one obvious problem, the goals scored and conceded by different nations were against teams of varying abilities.

For example within their 20 matches Australia were defeated by both Brazil and France 6-0, while fellow group A competitors Kuwait and Oman rarely played outside of Asia. TSR also does not take into account the standing of the respective matches, a friendly has the same weighting as a World Cup final in the rating calculation.

Inside a league setting where teams play the same teams multiple times this is not a problem and it would be expected that the modeling would be more accurate. Through this case study we hope to see if this also applies in a tournament setting.

There are also questions about TSR’s real assessment of a team’s attacking ability. Without a way to assess the quality and context of each shot TSR can potentially lead to a false assessment of a team’s true goal scoring potential.

World Football Elo Ratings

The Elo Ratings used in the case study will come directly from the World Football Elo Ratings website, with ratings updated between each match.

Where TSR’s achilles heel is its inability to distinguish between different levels of match and opponent, Elo Ratings are purely based on these factors. The stronger the opposition you beat, the higher your rating becomes. Similarly, losses against highly rated opponents are not punished as harshly as those against teams you are expected to beat.

It should also be noted that there is a ratings boost in probability calculations with Elo Ratings, this will be applied to Australia throughout the tournament.

Pre-tournament ratings

Asian Cup Group A

Team Elo Rating TSR Rating Bookmaker probability to win group*
Australia 1635 0.48 58%
Korean Republic 1619 0.56 44%
Kuwait 1513 0.44 6%
Oman 1543 0.44 7%

* Quoted odds to win Group A with Bet Easy

Pre-tournament group A was expected to be dominated by hosts Australia and the Korean Republic by bookmakers. This was mirrored in the Elo Ratings with both rated substantially higher than Kuwait and Oman. TSR suggests that Koran are the strongest attacking team in the group, while Oman are expected by both the bookmakers and Elo to finish third in the group.

Asian Cup Group D

Team Elo Rating TSR Rating Bookmaker probability  to win group*
Iraq 1498 0.45 13%
Japan 1712 0.55 83%
Jordan 1509 0.46 13%
Palestine 1322 0.45 2%

* Quoted odds to win Group D with Bet Easy

Japan are dominant favourites to win the group on all three models, and are the strongest team in the tournament according to the Elo Ratings. There is little to separate Iraq and Jordan on any of the measures as they look likely to battle out second placing in the group. Palestine are by far the weakest team according the the Elo Ratings and bookmakers, however interestingly have a similar TSR to Iraq and Jordan.

Match Results

At the conclusion of the two weeks of the group phase of the tournament all three models correctly predicted at least 75% of winners. This may have been aided by the lack of draws seen throughout the tournament as a whole, with all 24 matches during the group stage having a winner.

Predictive Model Selections & Results

Match TSR Prediction Elo Prediction Bookmaker Prediction Winner
Australia v Kuwait Australia Australia Australia Australia
Korean Rupublic v Oman Korean Republic Korean Republic Korean Republic Korean Republic
Japan v Palestine Japan Japan Japan Japan
Iraq v Jordan Jordan Jordan Iraq Iraq
Australia v Oman Australia Australia Australia Australia
Korean Rupublic v Kuwait Korean Republic Korean Republic Korean Republic Korean Republic
Palestine v Jordan Jordan Jordan Jordan Jordan
Iraq v Japan Japan Japan Japan Japan
Australia v Korean Republic Korean Republic Australia Australia Korean Republic
Oman v Kuwait Kuwait Oman Oman Oman
Japan v Jordan Japan Japan Japan Japan
Iraq v Palestine Iraq Iraq Iraq Iraq
Total Correct 9/12 10/12 11/12 x
Win Percentage 75% 83% 92% x
 x

Within the sample of 12 matches, only on three occasions did the models differ in their opinion of who the match winner would be. Interestingly, when all three models were unanimous in their choice of the strongest team the team saluted nine out of nine times.

Only on one occasion did both the Elo Rating and TSR disagree with the bookmaker favourite (Jordan over Iraq). While both Elo and TSR had the teams as almost exactly equal, they did both slightly favour Jordan however bookmaker favourite Iraq ultimately won the match 1-0.

On two occasions the TSR disagreed with both the Elo Ratings and bookmakers on which was the better team. In the first match, the Korea stunned Australia winning 1-0, while in the second example Kuwait were selected to defeat Oman (with Oman winning).

Overall the bookmaker’s favourite was the most successful in selecting the winner of matches with an amazing 92% strike rate. Blindly level stake betting the bookmaker favourite in each match would have netted a return of 44% over the tournament.

The second most accurate model was the Elo Ratings, selecting the winner in 83% of matches. Blindly level stake betting the top rated team using the Elo Ratings would have netted a return of 22%. Perhaps more interestingly though, betting only when an edge was identified through the expected win percentage an 9% return was achieved using level stakes.

The least accurate of the models was the Total Shot Ratio which predicted the winner in 75% of matches. Unsurprisingly betting blindly at level stakes also achieved the lowest return of 2%.

Conclusion

With so many international football tournaments each year at a range of different age groups there are many situations were punters may be required to handicap matches involving sides with little form, and with a high degree of mystery surrounding the participating teams and players.

This lack of data, or data which may be against a variety of different quality opposition can make modelling very difficult. Our pure team statistic based model, Total Shot Ratio was the lowest performing of the three models, however with such a small sample of matches it is hard to definitively rule it out completely as a tool to identify winners.

It was the only model to find the victory for Korea over the Socceroos, however this could be explained away as Australia were resting a number of players. It also incorrectly predicted Kuwait as a better attacking team to Oman which could be due to the differing levels of opposition the teams have faced in their previous 20 matches.

The Elo Ratings only major stumble was selecting Jordan to defeat Iraq in the opening round (albeit by only 1%). Its only other selection which lost, Australia to Korea, was easily explainable and was also incorrectly predicted by the bookmakers.

For me the Elo Ratings real party piece of the tournament was its ability to find six value bets (four wins and two losses), during the tournament. With level stake betting these returned a very reasonable 9% profit.

Yes this was less than simply betting both the Elo Ratings and bookmaker favourites blindly, however simply blind betting is not a sustainable strategy long term as its not known at any stage what the expected value of any bets are.

So if your looking for a football match prediction model for an upcoming tournament where stats are few and far between look to the bookmakers to give you the greatest lead on which team is more likely to win. However through the use of a power ranking system like Elo Ratings you might just find an exploitable edge in their prices.

Read each round’s prediction articles here:

2 Responses

  1. ทีเด็ดบอล

    It’s a shame you don’t have a donate button! I’d most certainly donate
    to this fantastic blog! I suppose for now i’ll settle for bookmarking and adding your RSS feed
    to my Google account. I look forward to new updates and will share this website with my Facebook group.

    Talk soon!

    Reply
    • Daniel Murphy

      Thank you for the kind words!

      The best way you can support us is to subscribe to our mailing list and do you best to learn from our articles and hopefully improve your betting over the long term.

      Reply

Leave a Reply

Your email address will not be published.