Abstract

The purpose of this article is to propose a method to minimize the difference between electoral predictions and electoral results. It builds on findings that stem from established democracies, where most of the research has been carried out, but it focuses on filling the gap for developing nations, which have thus far been neglected by the literature. It proposes a two-stage model in which data are first collected, filtered and weighed according to biases, and then output using Bayesian algorithms and Markov chains. It tests the specification using data stemming from 11 Latin American countries. It shows that the model is remarkably accurate. In comparison to polls, not only does it produce more precise estimates for every election, but it also produces a more accurate forecast for nine out of every ten candidates. The article closes with a discussion on the limitations of the model and a proposal for future research. (C) 2020 International Institute of Forecasters. Published by Elsevier B.V. All rights reserved.