Forecasting leadership transitions around the world

We use popular non-parametric (CART, TreeNet) and parametric (logit) techniques to identify robust economic, demographic and political conditions that lead to shifts in control in the executive branch of government in 162 countries during the period 1960-2004. We find that institutional aspects of the political system, executive characteristics, demographic variables, economic growth, and economic trade variables are all very important for predicting leadership turnover in the following year. Financial crises are not robustly useful for this purpose, but a vulnerability to currency crises in times of low economic growth implies very high conditional probabilities of job losses for democratic leaders in non-election years. In-sample, TreeNet predicts 78% of leadership transition events correctly, compared to CART's 70%, and TreeNet also generally achieves higher overall prediction accuracies than either CART or the logit model out-of-sample. (C) 2013 International Institute of Forecasters. Published by Elsevier B.V. All rights reserved.