Since 2014, we have been producing regular six-month forecasts of the probability of irregular leadership changes – coups, rebellions, protests that result in state leader changes – for most countries in the world for the Political Instability Task Force (PITF). During 2015, we issued new forecasts each month, with a delay as short as five days and no longer than two weeks into each six-month forecasting window. This article describes the approach we use to generate our forecasts and presents several examples of how we present forecasts. The forecasts are derived from a statistical ensemble of seven thematic models, each based on a split-population duration model that aims to capture a specific argument or related set of covariates. This approach is modular in that thematic models can be swapped out or new models integrated, and it lessens the need for generalist ‘kitchen sink’ models. Together, the models achieve high out-of-sample accuracy. Based on our experience, we draw conclusions about the practical, policy, and scientific aspects of this and similar undertakings. These include thoughts on how to evaluate and present forecasts, the potential role of ensembles in model comparison, the role of ensembles and prediction in causal research, and considerations for future efforts in forecasting and predictive modeling.

We forecast Irregular Leadership Changes (ILC)-unexpected leadership changes in contravention of a state's established laws and conventions-for mid-2014 using predictions generated from an innovative ensemble model that is composed of several split-population duration regression models. This approach uses distinct thematic models, combining them into one aggregate forecast developed on the basis of their predictive accuracy and uniqueness. The data are based on 45 ILCs that occurred from

Forecasting of geopolitical events is a notoriously difficult task, with experts failing to significantly outperform a random baseline across many types of forecasting events. One successful way to increase the performance of forecasting tasks is to turn to crowdsourcing: leveraging many forecasts from non-expert users. Simultaneously, advances in machine learning have led to models that can produce reasonable, although not perfect, forecasts for many tasks. Recent efforts have shown that forecasts can be further improved by "hybridizing" human forecasters: pairing them with the machine models in an effort to combine the unique advantages of both. In this demonstration, we present Synergistic Anticipation of Geopolitical Events (SAGE), a platform for human/computer interaction that facilitates human reasoning with machine models.

Using updated Archigos Data, as well as structural and event data, we construct a split-population duration model of irregular leadership changes. These are leadership changes that occur outside of the normal, legal framework for leadership transitions. Our model was estimated in March 2014 and produced probability estimates of leadership changes in many countries in the world. We used a wisdom-of-the-crowds approach to combining estimates from various different models. Ukraine and Thailand are among those in which we had the highest predictions for irregular change of leaders.

We present an implementation of split-population duration regression in the spduration (Beger et al., 2017) package for R that allows for time-varying covariates. The statistical model accounts for units that are immune to a certain outcome and are not part of the duration process the researcher is primarily interested in. We provide insights for when immune units exist, that can significantly increase the predictive performance compared to standard duration models. The package includes estimation and several post-estimation methods for split-population Weibull and log-logistic models. We provide an empirical application to data on military coups.