Sports forecasting: Current applications in sports science and moving towards Big Data

Publication: Book/ReportDissertations

423 Downloads (Pure)

Abstract

The desire to know what will happen, before it actually happens is at the heart of forecasting, which – in particular in the domain of sports – demands for interdisciplinary approaches. The present dissertation thus takes a holistic view on sports forecasting, including economic and mathematical aspects, aspects of sports science as well as aspects of data analysis and computer science.

While economically, the profitability of a forecasting model is commonly used to substantiate its predictive quality, the present dissertation uses theoretical considerations, as well as simulated and real-world data to prove that positive betting returns can be generated with inaccurate forecasting models and as such, profitability and accuracy need to be assessed separately in sports forecasting.

With regard to predictive accuracy, the high quality of collaborative forecasts in general and betting odds in particular is already well evidenced. In accordance with this, I have presented evidence for the high accuracy of betting odds based on football data in several further contexts: Pre-game betting odds provide more information about team strength than the actual match results, betting odds are a valuable predictor of success in penalty shootouts and first half goals are clearly outperformed by betting odds when forecasting second half goals in football.

The theories of market efficiency and crowd wisdom provide a theoretical framework for the forecasting accuracy of betting odds, which are also beneficial in the domain of sports science. Accordingly, the use of betting odds to obtain measures of relative or absolute team strength, an indicator for balanced or unbalanced matches and a measure of home advantage are highlighted methodologically.

Moreover, the present dissertation contributes to moving sports forecasting towards Big Data analysis. The characteristics of Twitter data as well as event and positional data as sources of Big Data in sports have been outlined and tested in the context of forecasts performed during the course of football matches (so-called in-play forecasting). For both data sources, no evidence for improvements on in-play forecasting were found. However, this can be considered to be only partly, if at all, driven by the data itself. In-play forecasting in general has been evidenced to be a highly difficult task, which supports the notion that goal scoring in football, if controlling for pre-game expectation, is a highly stable process.

The present results have theoretical implications for performance analysts as well as practical implications for bookmakers, professional gamblers and match analysts. Performance analysts in football should standardly use betting odds as a situational variable. The value of in-play information should not be overvalued by match analysts when drawing conclusions during a match or bookmakers when compiling in-play betting odds. Moreover, professional gamblers should be aware of the differences in profitability and accuracy when designing forecasting models.
Translated title of the contributionPrädiktive Modelle im Sport - Aktuelle Anwendungen in der Sportwissenschaft und der Weg zu Big Data Analysen
Original languageEnglish
Place of PublicationKöln
PublisherDeutsche Sporthochschule Köln
Number of pages134
Publication statusPublished - 18.03.2022

Fingerprint

Dive into the research topics of 'Sports forecasting: Current applications in sports science and moving towards Big Data'. Together they form a unique fingerprint.

Citation