The present paper aims to model, predict, and explain presidential election results using selected quarterly macroeconomic indicators, i.e., gross national product, consumer price index, unemployment rate and gross national product from 1994-2017. We also seek to provide predictions of presidential winner prior to the elections based on the beta distribution and the support vector regression (SVR) as prediction models. Two models are primarily built based on beta distribution and SVR. Due to the forecasting aspect, model performance focuses on one goodness-of-fit measure, i.e., the prediction error rather than the squared correlation coefficient R2 as it makes little sense in a practical regression perspective. The best model is the one with the least mean square error (MSE). In this effect it turns out that the SVR with kernel type encapsulated postscript eps radial has a mean square error of 0.006 on the test set and is a better model compared to the beta distribution model with a mean square error of 1.216. Thu, an accurate solution to prediction of presidential vote elections via SVR analysis is proposed.
Bibliographical noteFunding Information:
This research has been partly supported by the Human Sciences Research Council (HSRC) of South Africa, Faculty of Health and Applied Sciences (Namibia University of Science and Technology), Department of Economics and Statistics, and Faculty of Economics and Management Sciences, Kabale University. We extend our sincere thanks to the two anonymous reviewers who provided insightful comments that presented a distinctive paper.
© 2020 NSP.
- Support vector machine
- Time series data