The 2022 ICC Men's T20 World Cup is just around the corner. It will be the 8th edition of the tournament, and will be hosted by defending champions Australia. At the time of writing, the hosts find themselves as tournament favourites, despite sitting in 6th in the world rankings. India and England are ranked first and second in the world respectively and most bookies have them just behind Australia in the running. Both will no doubt be determined to make up for and underwhelming showing in last year's edition of the tournament.
In this article we attempt to predict the outcome of the tournament by way of Monte Carlo simulation, which involves simulating the tournament many times (10,000 in our case) and aggregating the results. This then allows us to derive quantitative measurements for how each team can be expected to perform in the tournament.
From there we compare our results to the odds provided by the bookmakers, which then facilitates a discussion about the pros and cons of the model, and also gives some food for thought around the cricket itself.
For reference, we also applied the same methodology to last year's competition, with promising results. In those predictions, there were two main outliers when compared to what the bookmakers were predicting. One of these was West Indies who, whilst the bookmakers had them as 3rd favourite, our model only gave them a 1% chance of victory. This has to go down as a success for our model, as West Indies were incredibly underwhelming in the competition as they limped out at the group stage with just one victory to their name.
The other major outlier were Pakistan, who we gave a 20% chance of victory compared to roughly 10% as given by the bookies. Again, we'll count this as a success for our model, as Pakistan stormed into the semi-finals with a mightily impressive group stage showing. They couldn't quite see it through, but their performances were certainly strong enough to suggest that our model was onto something.
It was far from a perfect prediction, however. Neither of our top two favourites - India and England - made the final, and we also suggested that Australia, who won the competition, might struggle based on their recent results. All in all, though, it looks like our simulation did a decent job!
Assessing Team Strength
The starting point for simulating a competition such as this is to derive ratings for each team, in order to get a quantitative measure of the competitors' respective strengths. For this we use the Bradley-Terry model, which has been used extensively elsewhere on the site. Using a complete set of results from some pre-determined period of time, the output of the Bradley-Terry model gives a coefficient representing the quality of each team over this time period, and these coefficients can in turn be used to simulate an outcome of a contest between any pair of teams.
The raw abilities are not sufficient to provide a reliable simulation of the tournament, however, as there will always be external factors likely to have an impact on how any given sporting competition plays out. Fortunately, the Bradley-Terry model is flexible and allows us to control for such external factors. Typically, when we do something like this we will include a factor for home advantage, which allows us to apply a 'home advantage' coefficient to a home side when we are trying to predict a match.
In this year's edition of the tournament, this is a fairly straightforward process since there is a single host in Australia. The model will give a coefficient that represents an improved "ability" for a side when they are playing at home, which we can simply add on to Australia's ability rating when it comes to simulating the tournament.
The below table gives the coefficients as derived from our Bradley-Terry model, run on all T20I fixtures since October 2018. The model now also applies greater weighting to more recent matches, which is a refinement of the model we ran for last year's competition.
The 'Home Advantage' coefficient sits at 0.28, meaning that when we run the simulation, Australia will have an effective ability of 53.18 + 0.28 = 53.46. This means that with home advantage on their side they'd actually be favourites in match ups with Pakistan and South Africa despite our model computing them as a weaker team overall.
Simulating the Tournament
As we noted earlier, these coefficients can now be used to simulate an outcome of a contest between any pair of teams, which is done using the inverse logit function run on the difference between the ratings of the respective sides. This single match simulator can then provide the basis of a full tournament simulator.
The tournament commences with round 1, which is something of a preliminary round as it doesn't involve many of the major sides. It comprises of two groups, each of four teams: Group A, which contains Namibia, Netherlands, Sri Lanka and U.A.E., and Group B, which contains Ireland, Scotland, West Indies and Zimbabwe.
The winner of Group A (A1) and the runner-up of Group B (B2) will join Afghanistan, Australia, England, New Zealand in Group 1. Similarly the winner of Group B (B1) and the runner-up of Group A (A2) will join Bangladesh, India, Pakistan and South Africa in Group 2.
Our simulation captures all of the above, but there is one slight difference around how we deal with teams who are tied on points in the group phase. The tie-breaker that will actually be used is Net Run Rate, but we do not have this information available since we are only simulating a match winner, and not the winning margin. Instead, we simply randomise the order of any set of sides who finish on equal points in the same group. This is, admittedly, a limitation of the model, but we feel it unlikely to have any significant impact on the output.
Once everyone has played each other in each of Groups 1 and 2, the winners from each group play against the runners-up from the opposite group in the semi-final, and the winners of those semi-finals will, of course, meet each other in the final.
This tournament simulator was run 10,000 times, with the results aggregated below:
Our model gives England as the favourites to win the tournament, but there is not much to separate them from either India or Australia - this is in line with the general consensus that these three are the teams to beat going into the tournament. South Africa come out as our best of the rest, with a 16% chance of victory, narrowly ahead of Pakistan who they will face up against in the group stages.
It may be a little surprising to find New Zealand, who were runners-up in last year's edition, down as 6th favourites with just a 5% chance of victory. Their overall ranking as 6th favourites makes sense given that our model ranks them as the 6th best team in the competition, and the model does indicate a not insignificant gap in quality to the likes of South Africa and Pakistan. The gap in quality, however, is not sufficient to explain quite how large the difference is in the respective chances of winning the tournament. In order to explain this disparity, it makes most sense to contrast New Zealand with Pakistan who, as 3rd favourites in their group, can be seen as the Black Caps' counterparts on the other side of the draw.
As some of you may have spotted, it is the draw itself that is working against New Zealand. Most obviously, they will have to face off against two of the tournament's "big boys" in Australia and England. In most of our simulations, England and Australia qualify as the top 2, which means that the Black Caps will have to beat at least one of them if they are to replicate their performance from last year.
Whilst Pakistan do have to face up against India, the third of our "big boys", the next most difficult opponent they have is South Africa, with whom they are relatively equally matched. This means that there is a qualification spot very much "up for grabs" for the Men in Green, as they don't necessarily have to beat one of the tournament favourites to get through.
The difficulty for New Zealand is further compounded by the other sides they will have to face in the group as well. The 4th seeds in their group, Afghanistan, are rated as stronger than Bangladesh, the equivalent in Pakistan's group, which adds to the challenge. Additionally we have Sri Lanka, who are strong favourites to join the Black Caps' group as the winners of Group A. Sri Lanka, who the model has as the strongest of the potential qualifiers, have an ability rating that isn't actually that far off New Zealand, and our model expects them to be a much more challenging opponent than the equivalent in Pakistan's group, likely to be West Indies.
All of this adds up and means that New Zealand, despite their talented squad and fairly strong pre-tournament form, face a very difficult task to even progress to the knockout phases, let alone go on the win the tournament.
Below the aforementioned Sri Lanka and West Indies we start to get into the realm of major outsiders. Afghanistan just about manage to hit a 1% chance of winning, with the next best being Bangladesh at around 0.2%. Namibia and Zimbabwe are rated as the most likely to follow Sri Lanka and West Indies through the qualifying rounds, but the likes of Ireland, Netherlands, U.A.E and Scotland will all fancy themselves to sneak through instead.
None of these teams are given any real hope of making it into the knockout stages but a handful of our 10,000 simulations did throw up a few surprises. Only Netherlands failed to win the tournament at all, with Zimbabwe (9), Namibia (7) and Ireland (3) all managing to claim victory on multiple occasions. Potentially a cause for optimism if you love an underdog!
Assessing our Predictions
It's impossible to know - even after the event has happened - how accurate predictions of this type will be, even if in this case they do feel roughly in line with what we would expect. What we can do, however, is compare and contrast them to another set of predictions for the same event, which can often throw up some interesting discussion points.
In this case, we have compared our win % probability as calculated by our simulation model to the implied chances of winning based on the odds (as of Friday 14th October 2022) from a specific bookmaker. The teams under the line are those that our model believes that the bookies are 'over-rating' and the sides above the line are those that our model thinks the bookies are under-rating:
It's worth noting now that the purpose of this analysis is not to look for value in the betting market, and we don't recommend anyone use the above as a guide if they do want to gamble on the tournament. Our model is relatively simple and the computation that goes into calculating the bookmakers' odds is likely far more complex and uses many more variables. Comparing the two as we have done just gives us a yardstick by which to assess our predictions, and any discrepancies can then be used as a springboard for further discussion about the respective teams.
Straight away we can see that there is a strong agreement about who the front-runners are expected to be. Both the bookies and our model expect England, Australia and India to be there or thereabouts some the business end of the tournament. There is some disagreement, however, on the ranking between these three early favourites. The bookies have Australia as the favourite, which admittedly makes sense on an intuitive level given their status as both hosts and holders. Our model, however, ranks them as second favourites behind England.
The main factor at play here - and it's not exactly an especially technical one - is that our model simply does not rate Australia as a particularly strong team. Last year's triumph aside, results have been inconsistent over the past few years to the point that - as evidenced by their recent 2-0 defeat - they are not necessarily favourites even when playing at home. This is also reflected in the ICC World Rankings, which have them down in 6th, despite being the reigning World Champions.
As for England, the bookmakers actually have them down as third favourites, also behind India. Ultimately time will tell, but with the data available our model suggests that this could be a potential oversight, even if they still consider England as strong contenders.
The only other major outliers are New Zealand, who we also discussed in the previous section. It's possible that the bookies aren't accounting for the draw as much as we were there, which could explain some of the disparity. It's also possible that their recent tournament efforts are skewing things somewhat, with the bookies placing more significance on their final appearance last year as opposed to their relatively average wider performances over the past few years.
On the flip-side you have both South Africa and Pakistan, who are both given a sizeable boost in our model when compared to the bookies. Again, this is potentially due to the draw, with these sides being statistically in the weaker of the two main groups. There is also a relative lack of recent tournament pedigree between these two, with neither having reached a final since 2009, when Pakistan were crowned champions. Obviously, we don't know for sure what factors are at play when the bookmakers calculate their odds, but the model we've run here does suggest they might be overlooking something.
Thanks for reading! During the tournament we will be applying our player ratings model to assess who the top performers have been, so if it sounds of interest please keep an eye out for that! Otherwise, if you liked this please check out some of the other articles on the site or follow the Twitter!