Forecasting Federal Elections: New Data From 2010–2019 and a Discussion of Alternative and Emerging Methods

Abstract

I produce and compare election forecasts for the period 2010-19. The dominant methods of polls, odds and economic models have mixed success, each predicting three out of four federal elections. Consistent with prior research, polls then prediction markets are found to produce the most volatile forecasts, while economic models produce the least. Prediction markets are found to efficiently price available information. Economic models performed well at the 2019 election, however, they were limited by outdated dummy variables for 'honeymoon effects'. I conclude by discussing some alternative and emerging forecasting methods, including demographic explanations of voting and online sentiment analysis.

Erratum

“…In [1], the author has provided the below corrections in Tables 1 and 2, and a change of per cent in the text (page 29): Table 1 Under the header “ Economic models ”, all the values in the column “ Mean absolute error (%) ” have been corrected. …”