Do Ceasefires Work? A Bayesian autoregressive hidden Markov model to explore how ceasefires shape the dynamics of violence in civil war

Abstract

Despite a growing body of literature focusing on ceasefires, it is unclear if most ceasefires achieve their primary objective of stopping violence. Motivated by this question and the new availability of the ETH-PRIO Civil Conflict Ceasefires Dataset, we develop a Bayesian hidden Markov modeling (HMM) framework for studying the dynamics of violence in civil wars. This ceasefires data set is the first systematic and globally comprehensive data on ceasefires, and our work is the first to analyze this new data and to explore the effect of ceasefires on conflict dynamics in a comprehensive and cross-country manner. We find that ceasefires do indeed seem to produce a significant decline in subsequent violence. However, the pre-ceasefire period (the period typically after a ceasefire agreement has been negotiated but before it is in effect) is shown to be prone to periods of intensified violence that are most likely a cause and effect of the subsequent ceasefire. This finding has significant implications for the research and practice community. Moreover, manifesting from the ubiquity of modern data repositories combined with a deficiency in meaningful labels, HMMbased semi-supervised data labeling strategies could pave the way for the next decade of conflict research progress.

Beyond time-homogeneity for continuous-time multistate Markov models

“…Markov models and HMMs play a fundamental role in many recent applications from modeling disease progression (Williams et al 2020) and seizure detection (Furui et al 2021), to assisting in volcanic eruption risk evaluation (Aspinall et al 2006), as well as studying the efficacy of ceasefires on conflict dynamics (Williams et al 2021). Applying numerical ODE solvers to fit continuous-time Markov models (HMMs or otherwise) in a truly time-inhomogeneous fashion has broad implications, particularly for studying longitudinal medical and biological data sets where the underlying phenomena exhibit continuously time-evolving rates of progression.…”