The correct measurement of economic policy uncertainty (EPU) plays a critical role in many policy settings - in particular where economic policy decisions need to be taken in response to large shocks. One such large shock is armed conflict. But, counterintuitively, the standard text-based EPU index systematically declines during armed conflict periods. Using a global news corpus covering 192 countries and over 5 million articles, we show that this decline is driven not by reduced uncertainty, but by a crowding out of reporting on economics and policy. We show that a combination of topic modeling and two-way fixed effects can be used to adjust the measurement of EPU, providing a new view on political risk during armed conflict. After adjustment, the EPU aligns more closely with firm perceptions, political risk insurance and investment during armed conflict.
The latest manuscript version can be downloaded here. CEPR Discussion paper version can be accessed here. The data we generate from our corpus from 1989 to 2024 for all countries can be downloaded here. Note, that not all countries are covered consistently in our news corpus. I recommend capping the sample to inlcude only countries with more than 10,000 articles.
This project was borne out of the puzzle that is shown in Figure 1 of the paper - the EPU index is falling in civil war. We were wondering why economic policy uncertainty would fall in armed conflict. It seemed unrealistic. We had noticed a similar pattern in the standard EPU data in our paper with Sandra García-Uribe and Carlos Sanz on the EPU in pre-civil war Spain (linked here). It turns out this is due to a fall in P and E terms NOT due to a fall in the uncertainty terms U. The idea for this paper was born.
