Forecasting sport practice with timeseries

Abstract

The concern in assessing sport participation rates in Europe (Rodgers, 1977) and to identify the methodologies for measuring and comparing (Hovemann & Wicker, 2009), allow the definition of public policies encouraging the participation in physical activity, anticipate development of the market for equipment and materials for sport (Tuyckom, Bracke & Scheerder, 2011), and guide the application of resources in sports federations. The identification of the potential number of practitioners in sports federations is a management problem (Camps & Pappous, 2016), manly from a strategic point of view to support decision making on the allocation of their resources.

Time series have often been used to predict team results or situations that can improve their chances of winning a competition (Yiannakis, Selby, Douvis and Han, 2006), based on previous results, in soccer (Stern, 1991), basketball (Lopez & Matthews, 2015, Ruiz & Perez-Cruz 2015) or ice hockey (Brockwell & Davis, 1991). Considering the importance in predicting the number of practitioners enrolled in sports federations, we developed this study to support the decision making for the planning of resources using the time series.

To carry out the prediction of the practitioners in the federations, a data sample with 22080 registers and 19 variables was used, representing the practice of sport in all Portuguese districts from 1999 to 2014, in 69 modalities at various levels. The data obtained was consolidated calculating the total of practitioners of all modalities. Data processing was performed with Anaconda and IPython (Continuum Analytics, 2016), using Pandas (McKinney & others, 2010). The predictions were developed with 16 observations (16 years) - which is considered feasible for small time series (Makridakis & Hibon, 2000) - using R (R Development Core Team, 2008) with the library forecast (Hyndman & Khandakar, 2008), employing the Exponential smoothing state space model (ETS) and Autoregressive Integrated Moving Average (ARIMA) to determine the number of practitioners for the next four years. Two models were used to assess what is more accurate comparing to the actual number of practitioners when the data is available. Was obtained the expected number of practitioners, the upper and lower interval with 80% and 90% confidence for each one of the used models.

The results obtained can be used as guidelines for sports federations, public entities financing sports federations and the market for sport products and services to anticipate resource needs.