Predicting the European Championship and spare parts demand

Last week, albeit a year late, the European Championship finally kicked-off. A great tournament with some of the best footballing nations of the world held in eleven different countries. A time-honoured tradition amongst many groups of friends and colleagues is to enter so-called European Championship prediction pools. A friendly bet to see who has the most football-knowledge, or luck, by predicting the match scores, future European champion, and top goal scorers. At Gordian this is no different, so game on!

 

My first instinct, as a consultant at Gordian and as an Industrial Engineer, is to look for data that can support me in my predictions. However, this becomes quite hard for events that only occur once every four years, such as the European Championship. For many countries, it is the first time in many years that they are facing each other, and since then, a lot has changed; new players joining, old players leaving, and even the debut of the Video Assistant Referees at the European Championship. All-in-all, previous results might not necessarily be a good predictor of the future because of a lack of sufficient data.

 

Similarities

In this way, predicting football matches is not too dissimilar to forecasting spare parts. As, for spare parts, we encounter the same problems regarding a lack of historical data. After all, slow moving spare parts just don’t move often enough to provide a proper statistical basis for traditional forecasting methods.

 

However, looking beyond the traditional forecasting methods, there are interesting possibilities using more innovative methods, such as in the field of machine learning. Intuitively, one would think this would be even more challenging, as machine learning is known to require large amounts of data to make accurate predictions. However, machine learning has shown potential in both the fields of football and spare parts forecasting.

 

In predicting football outcomes, researchers are looking at other explanatory variables to use for their predictions to overcome this challenge. Market value of players, club match performance, a countries population, its gross domestic product. These are all variables used by an international team of researchers that have developed an algorithm that has had significant success in predicting previous World Cups and European Championships. The successor of this algorithm has predicted that France will be crowned the new European Champion with a probability of 15% [1] (take one guess who I picked in my prediction this year).

 

Researching machine learning

Currently, Gordian is also researching the use of machine learning in the forecasting of spare parts demand. The first study [2] showed that using machine learning can result in subtle to significant improvements in forecast accuracy. However, it is also stated that further research is still required to prove the robustness of such methods, especially for extremely slow moving parts. Therefore, at least for the time being, we still have to rely on, and utilize, practical knowledge from the technical experts as some of the most important input for our decisions.

 

This gives us three alternatives when insufficient data is available: finding alternative explanatory variables, using expert knowledge, or ideally: a combination of both. This last one is the approach I’ve taken in my predictions for the EC pool. We will see if my ‘expert’-knowledge is enough to help me win the price pool, or if it ruined any chances the data-driven approach has given me.

Source 1: https://www.eurekalert.org/pub_releases/2021-06/uoi-uef060721.php

Source 2: Nick Roggeveen thesis (Forecasting Demand for Spare Parts: A Machine Learning Approach)

Get in touch

Wouter Heijnen
Junior Consultant
E-mail us
info@gordian.nl