So with time on my hands I decided to join the bandwagon and produce some World Cup predictions. They will be hopelessly wrong.
The reason they will be wrong is that they are based on a statistical model that uses historical data to estimate the contributions of different factors to winning international games. That model is not very good- it can only account for about 30% of the variation in team performance. So most of what we’re interested in is unexplained.
So why bother? Because this model, as bad as it is, is probably not much worse than most of the other models out there. I hope someone like Roger Pielke Jr is going to do an analysis of forecasts after the event- these are usually illuminating.
The model might be worse than many others because I only used data up to 2014, so it’s a bit out of date. There are probably other factors beyond population, GDP per capita and experience and home advantage that one could include – but probably not many you could collect for all countries. You could include things like transfermarkt valuations of players. And you could manually adjust for exceptional countries like Uruguay.
But once you do those kind of things, the danger is that you have thrown in too many ad hoc judgments which reflect your biases. Generally speaking, forecasting models of sports events have poor records. Even the bookmakers don’t know. Which is a good thing, since we don’t want to know the outcome in advance.
So why make these forecasts? People like Goldman Sachs and lots of other worthies are doing it for the publicity. It is often said that the key thing is to forecast often, and randomly you are bound be right some of the time. Since few people remember your mistakes, and your successes bring you glory, what’s to lose?
So 28 of 48 in the group stage, without a draw as an option. Not bad at all…
Yes, that’s what I thought 🙂 74% success rate on decisive games, and 14 out of 16 qualifier correctly predicted. Probably going to get a lot worse from hereon in.