Forecasting Regime Breakdown

• In Event: Poster Session: Democracy and Autocracy

Abstract

Political regimes breaking down can be momentous events, with ripple effects for societies and economies at home and abroad. Being able to forecast the occurrence of regime breakdown would therefore be of great interest and potential importance to organizations and other actors who want, e.g., to forestall coups, aid democratization processes, or safeguard civilians in the run-up to or aftermath of a regime breakdown. In this paper, we build and present high-performing predictive models to forecast breakdowns of political regimes in countries across the world. More specifically, we construct models that give monthly forecasts for aggregate regime breakdowns as well as for four particularly important sub-types of breakdowns, namely coups d'état, self-coups, popular uprisings, and incumbent-guided liberalization. Leveraging the Historical Regimes Data (HRD), we can base our predictions on more than 230 years of political history and over 2000 regimes changes, recorded with high temporal resolution and nuanced information on exactly how they broke down. As our predictive baseline, we train a set of machine-learning models using the predictors specified by Djuve et al (2020), which are a restrictive set of features selected for their theoretical relevance to the regime change literature. Thereafter, we compare the predictive performance of our baseline model to those of various thematic models trained with the same classifiers. Finally, we use ensemble methods to produce true forecasts of the likelihood of regime breakdown for all countries, globally, in the coming three years.

Authors

• Vilde Lunnan Djuve, University of Oslo

• David Randahl, Department of Peace and Conflict Research, Uppsala University

• Carl Henrik Knutsen, Department of Political Science, University of Oslo