Political institutions are coming under increasing pressure around the world, yet there is no continuously updated global dataset tracking major disruptions such as coups, term-limit evasions, or attacks on judicial independence.
In Barcelona School of Economics Working Paper 1555, “Semantic Similarity Measures in Newspaper Text for Detecting and Predicting Disruptive Institutional Events,” Laura Mayoral, Hannes Mueller, Margherita Philipp, Christopher Rauh, and Renato Vassallo use newspaper text to measure and code these events, building a new high-frequency database across more than 170 countries.
The authors then show that this information can be used not only to monitor institutional risk in near real time but also to predict future disruptions. As an illustration of why this matters, the paper finds that coups are followed by large and persistent declines in economic growth.
Why New Data Are Needed
Many of the most important threats to democracy do not appear in standard datasets quickly, if at all. Some are sudden, such as coups. Others are more gradual, such as leaders weakening courts or bypassing term limits. Because there is no continuously updated global database covering these different forms of institutional disruption, it is difficult both to study their economic consequences and to anticipate where they may happen next. The paper aims to fill this gap with a high-frequency and scalable monitoring tool.
How the Framework Works
The authors use a simple idea: instead of looking only for specific keywords, they compare the meaning of newspaper headlines to example headlines that describe each type of disruption (Figure 1). This allows the method to detect relevant news even when journalists use different words across countries and contexts. The article-level signals are then aggregated at the country-month level, combined with supervised classification and human verification, and used both to identify past events and to generate monthly forecasts of disruption risk over the following 12 months. Because the method focuses on meaning rather than exact wording, it performs especially well in cross-country settings where reporting styles differ widely.
Notes: Each vector corresponds to a headline, with proximity indicating similarity.
Model’s Performance
1. This new method outperforms dictionary-based methods to detect institutional disruptions.
The outperformance of the semantic-similarity approach is due to the fact that it captures meaning rather than relying on a fixed list of words. Its approach is especially useful for subtler forms of democratic erosion, such as attacks on the judiciary or term-limit manipulation, where there is often no standard vocabulary. Figure 2 shows receiver operating characteristic (ROC) curves across all events; a curve that sits closer to the top-left corner is better, because it catches more true events while making fewer false alarms. As evident in Figure 2, the nowcast consistently achieves better precision.
Notes: The nowcast was trained with data from 1989m01 to 2014m12 and the results above are for the true out of sample test set up to 2025m10.
2. The method can also help anticipate future risks.
The model is able to distinguish higher-risk from lower-risk months across different types of institutional disruptions. Overall, the forecasting exercise shows that news coverage contains useful signals about looming institutional stress. Even though these events are rare, the model can distinguish relatively well between higher-risk and lower-risk periods, making it a promising early-warning tool.
3. Coups have large and lasting economic effects.
Using their expanded coup database, the authors find that coups are followed by sharp and persistent declines in GDP per capita, with losses of around 10 to 15 log points within a decade (Figure 3).
Notes: Solid points represent the estimated average effects, with bars indicating 95% confidence intervals.
This shows that institutional breakdowns are not only political disruptions but also major economic shocks.
Key Findings
- Newspaper text can be turned into a practical tool for tracking and anticipating institutional breakdowns in real time
- By improving the measurement of coups, term-limit evasions, and attacks on the judiciary, the proposed model helps study democratic erosion more accurately
- The new model highlights the broader economic importance of these institutional disruptions
