10 Downing Street: who’s next? Seemingly unrelated regressions to forecast UK election results

ABSTRACT

In recent years, the British polling industry has encountered difficulties in its attempts to measure voting intentions in important popular consultations, notably the 2014 referendum on Scottish independence, the 2015 UK general election, and the 2016 EU membership referendum. In such a context, it is extremely valuable to explore how different forecasting models that rely on political and economic variables can be used to predict the outcome of elections. In this paper, we propose such a model by introducing a political economy equation to estimate the vote share obtained by the incumbent party in UK general elections three months before the vote as well as a set of seemingly unrelated regressions (SUR) to forecast the results of the Official Opposition, the Liberal Democrats, and the remaining parties.

Acknowledgements

I am particularly grateful to Ruth Dassonneville and Jean-François Godbout for their suggestions and helpful comments. I also wish to thank Matthew Lebo, Helmut Norpoth, Michael Lewis-Beck, Ruth Dassonneville, Richard Nadeau, and Éric Bélanger who kindly accepted to share their data with me. Finally, I thank the editor and the three anonymous reviewers for their comments on the manuscript of this paper.

Disclosure statement

No potential conflict of interest was reported by the author.

Notes

1 GDP is preferred to other measures since it constitutes the most global economic indicator there is (see Dassonneville and Lewis-Beck 2014).

2 The necessary quarterly GDP data were not available for the elections prior to that of 1959.

3 Contrary to our expectation, the previous vote share variable produced a negative (and statistically significant) coefficient for the remaining smaller parties. Consequently, this variable was not used for the remaining parties' equation because there is no logical explanation for why this should happen.

4 With a total of 16 cases and a maximum of five parameters to be estimated (for the incumbent equation of the SUR model), we were able to make step-ahead forecasts for the 10 general elections held between 1979 and 2017.

5 Additional comparisons were made with four existing models as well as with a SUR model using vote intentions for each party collected during the last week of the campaign. The main finding from these supplementary analyses is that existing models – with the exception of Bélanger, Lewis-Beck, and Nadeau’s (2005) – and our own SUR model show substantially similar levels of accuracy both within- and out-of-sample as well as step-ahead. A SUR model using the vote intentions of the last week of the campaign is more accurate than our own model (see appendices C to H in the supplementary materials).

6 Appendix I in the supplementary materials proposes more information on the subject as well as an application of Nadeau, Lewis-Beck, and Bélanger’s (2009) swing ratio.