April 13, 2021

Measuring the performance of our biathlon model Part 2

In our previous article we started investigating the interesting topic of sports models performance, which was brought to our attention by one of our readers.

We established that it has previously been shown that even as FiveThirtyEight, has well calibrated forecasts producing high Brier scores, their predictions are far from capable of beating the betting markets. Some could find this surprising considering how large a site FiveThirtyEight truly is.

In this article the focus will not be on how well calibrated our biathlon predictions are, or what Brier score they would achieve. We are extremely bullish about how our model would perform if measured in such ways.

Instead we will try to focus on how our predictions are performing when measured against the betting market, as this is the benchmark brought to our attention by our reader.

In order to be able to test our model’s performance, in such a way, we need odds data. Therefore we reached out to a website focusing on making betting tips which has a biathlon tipster that has achieved astonishing results the past two season. As Godsofodds are measuring all of their tipsters against Pinnacle closing odds, we correctly assumed that they would have Pinnacle’s closing odds for all the biathlon markets. Luckily they agreed to provide us with all of Pinnacle’s opening and closing prices for Biathlon for the 2020-21 season.

We are particularly happy that we were able to obtain Pinnacle’s odds, as they are a bookmaker known for having particularly sharp odds. The reason why their odds is generally accepted to be sharper than most other bookmakers is because they, unlike most betting companies, do not close or limit the accounts of winning customers.

The opening odds are the first odds launched by Pinnacle and the closing odds are the odds as they were when Pinnacle stopped accepting wagers for the market. This normally happens when a race starts.

It is well known that the Pinnacle’s closing odds tends to be sharper, or to use different words more accurate, than their opening odds. There are several reasons for this. One reason is that more information becomes available as we are closing in on the starting time of the race. Another is a “wisdom of crowds” effect, that as more people have expressed an opinion, by placing bets, the odds will incorporate more opinions.

The most common wager offered on the sport of Biathlon by Pinnacle, is what is called a Head to head. It names two athletes and the punter is supposed to predict which will be highest placed on the final results.

Even if our model can easily produce estimates of what chances the various athletes have of finishing ahead of each other, this is something we cannot use to check the performance of our model for the 2020-21 season. This is because we never published any such predictions, and because of this it would be impossible for our readers to check if we are providing an honest representation of the forecasts of our model. Instead we will focus on the odds Pinnacle have for an athlete to win or not to win the race as we published information about this before each race of the season started.

In our race previews we have, among other things, indicated the chance our model estimated that the various athletes had of winning. These estimates together with Pinnacle’s to win/not to win odds we will use to check the performance of our Biathlon model.

In our next article in this series we will check how our model performed versus Pinnacle opening odds.

Mathis Brorstad

Mathis Brorstad

Mathis Brorstad is a Norwegian freelance writer. He is mainly covering Athletics, Biathlon and other Winter Sports. In the past he has done work on odds and probabilities.

View all posts by Mathis Brorstad →

Leave a Reply

Your email address will not be published. Required fields are marked *