Predicting Authoritarian Crackdowns: A Machine Learning Approach

Abstract

We have developed a quantitative indicator to predict if and when a series of protests in China, such as the one that began in Hong Kong in 2019, will be met with a Tiananmen-like crackdown. The indicator takes as input protest-related articles published in the People’s Daily—the official newspaper of the Communist Party of China. We use a set of machine learning techniques to detect the buildup in the articles of negative propaganda against the protesters, and the method generates a daily mapping between the current date in the Hong Kong protest timeline and the “as if” date in the Tiananmen protest timeline. We call this counterfactual date the Policy Change Index for Crackdown (PCI-Crackdown) for the 2019 Hong Kong protests, showing how close in time it is to the June 4, 1989, crackdown in Tiananmen Square.

Keywords: policy change, machine learning, protest, crackdown, propaganda

JEL Classification: C53, C63, D74, D83, K42, N45, P49