Why it’s so hard to predict the Melbourne Cup winner

by Michael Elith
They say the Melbourne Cup race is the hardest race to pick, but with it just around the corner, I have started thinking about the best way to pick a winner.
You would assume that the odds should take the guess work out of picking the winner, with the horse with the lowest odds having the best chance of winning the race. But when it comes to the Melbourne Cup only around 20% of the winners have been the favourite leading into the race.

In previous roles we have built our own predictors to pick the big race’s winner, considering past horse performance, previous winners characteristics and jockey and trainer traits. It is only on the rare occasion that our model predicts the odds-on favourite. Even more rare was the victory for that favourite, netting us some healthy winnings.

Only around 20% of past winners have been the favourite leading into the race”

The Melbourne Cup is a bit of a unique horse race in the sense that it is the “race that stops a nation”, where every Australian becomes a punter for the day and backs their favourite more on gut feel, the name of the horse or the the Jockey’s colours, than on horse racing knowledge.

This swings the odds dramatically and makes it extremely hard to determine the winner from the odds alone. With many people taking a guess, the bookkeepers must continually fluctuate their odds to balance their books and ensure they go home with money in their bag at the end of the day. Therefore, the favourite in the ring may not have the best past performance but is the favourite among the masses.
This does keep me thinking though about predictive modelling and regression analysis and just how reliable it is in forecasting future outcomes. How do we predict the outcome of an event that has never occurred before? I.e. the Melbourne Cup this year will never of be run in this exact weather, with the track in this exact condition, with these exact horses with the exact same crowd.
For example, regression analysis was useless in trying to predict the 2015 Melbourne Cup winner where Michelle Payne rode Prince of Penzance, a 100-1 outsider, to victory. Never had a female jockey won, and Prince of Penzance became only the fourth horse in history to cross the line in first position as a triple figure roughie.

How do we predict the outcome of an event that has never occurred before?

We do a lot of attribution modelling and business forecasting based on regression analysis and we do experience this problem where we are asked to forecast the success of a campaign that is being run for the first time, new creative, new messaging, to a new audience. This becomes a best guess game. Which is way we are strong believers in a test and learn approach.
Roll out the new campaign in small bursts mimicking the desired conditions, collect the data, forecast and optimise, to ensure that the modelling can be based on comparable data points. It is the only way you can ensure that you predictive modelling isn’t going the way of the Melbourne Cup odds.
Good luck picking a winner next week.


Let's use data to make confident decisions