We produce a social unrest risk index for 125 countries covering a period of 1996 to 2020. The risk of social unrest is based on the probability of unrest in the following year derived from a machine learning model drawing on over 340 indicators covering a wide range of macro-financial, socioeconomic, development and political variables. The prediction model correctly forecasts unrest in the following year approximately two-thirds of the time. Shapley values indicate that the key drivers of the predictions include high levels of unrest, food price inflation and mobile phone penetration, which accord with previous findings in the literature.
In this paper I assess the ability of econometric and machine learning techniques to predict fiscal crises out of sample. I show that the econometric approaches used in many policy applications cannot outperform a simple heuristic rule of thumb. Machine learning techniques (elastic net, random forest, gradient boosted trees) deliver significant improvements in accuracy. Performance of machine learning techniques improves further, particularly for developing countries, when I expand the set of potential predictors and make use of algorithmic selection techniques instead of relying on a small set of variables deemed important by the literature. There is considerable agreement across learning algorithms in the set of selected predictors: Results confirm the importance of external sector stock and flow variables found in the literature but also point to demographics and the quality of governance as important predictors of fiscal crises. Fiscal variables appear to have less predictive value, and public debt matters only to the extent that it is owed to external creditors.
greater regional spillovers ( Arezki et al. (2020) ). However, it is likely that the drivers of unrest are disparate and interact with socioeconomic conditions in complex and non-linear ways that are difficult to enumerate.
We employ a flexible machinelearningapproach to gauge the importance of a large set of predictors and capture non-linearities. Our preferred model has a balanced accuracy level of 66% and, in that sense, is correct in predicting unrest approximately two-thirds of the time. We find a relatively modest role for predictors in the literature, with
empirically that the early warning systems in use at the time were not able to predict the Asian currency crises of the 1990s or the Global Financial Crisis, respectively. In a similar vein, my first result in this paper is that the out-of-sample predictions for fiscal crises obtained from common econometric approaches, on average, cannot outperform an uninformed heuristic rule of thumb and are considerably less accurate than what their in-sample performance suggests.
By using some of the most popular machinelearningapproaches, I obtain improvements in predictive