Accuracy of automatic forecasting methods for univariate time series data: A case study predicting the results of the 2018 Swedish general election using decades-long data series

Abstract

Abstract This study compared the accuracy of automatic time series forecasting methods in predicting the results of the 2018 Swedish general election using data from the Party Preference Survey opinion poll collected during the years 1984–2018. The general exponential smoothing state space (ETS) model performed best, outperforming even the exit poll collected at the time of the election, while the complex seasonal autoregressive integrated moving average (ARIMA) model was beaten by the simple exponential smoothing method. Holt’s linear trend method performed worse than even the naïve method. The results of this study show the usefulness of easily applied automatic forecasting methods.