Research

This paper investigates how the agricultural cycle influences protest behaviour in Kenya from 2000 to 2019. Using satellite data to identify the crop cycle and local harvest periods at the lowest available administrative level (wards), combined with geocoded protest records, we find that protest activity does not decline during the harvest period but increases with a temporal lag. A dynamic difference-in-differences specification shows that protest activity gradually rises in the two to three months following harvest, peaking around the onset of the subsequent agricultural cycle. These patterns are consistent with theories of opportunity costs and seasonal scarcity: the opportunity costs of protest participation are highest during harvest and decline thereafter, coinciding with the lean season, when household food stocks become exhausted. The findings are robust across empirical specifications and contribute to a deeper understanding of the temporal dynamics of unrest in agrarian settings. They also highlight the importance of timing in the design of rural development and conflict-prevention policies.

Robiro, N., & Seelos, S. (2026). Predictable patterns of protest: The impact of agricultural cycles on social unrest in Kenya. Food Policy, 142, 103136.

This paper examines the microlevel effectiveness of irrigation-focused development aid in sub-Saharan Africa between 2005 and 2020 using georeferenced project data and high-resolution remote sensing. Employing a localized difference-in-differences strategy near country borders, the study finds evidence that irrigation aid raises the production value of cash crops, while staple crop yields remain largely unresponsive. This pattern indicates that the gains are concentrated in market-oriented production rather than extending broadly across crop categories. The study contributes to the literature on foreign aid and agricultural development by introducing a high-resolution empirical framework that leverages border matching and grid-level data to identify local treatment effects. This approach provides an alternative identification strategy to commonly used instrumental-variable and shift-share designs, expanding the scope for evaluating geographically targeted development interventions.

We study the effectiveness of financing protected areas in reducing forest cover loss. Specifically, we construct a unique georeferenced database comprising 398 PAs in Latin America and the Caribbean (LAC) (co-) financed by KFW Development Bank, one of the largest conservation financiers worldwide, and covering the time period 2000-2020. The empirical analysis identifies treatment effects using a geospatial statistical matching approach: first, we augment the database with key covariates granularily characterizing the PAs, obtained from several data sources including satellite imagery on forest coverage. Second, given that the selection of a given area to be protected is non-random, we fractionize each PA into cells of 500 ha size, then match non-treated cells using a CEM (Coarsened Exact Matching) algorithm conditional on covariates, country, and most importantly pre-treatment trends in forest cover during all years preceding the start of protection. Third, treatment effects are obtained at the project level using difference-in-differences for the matched treatment-control samples over the full 21 years sampling period, with flexible pre- and post-treatment periods. The findings indicate that financing PAs is effective: in more than half of the projects 17 out of 32 there is a statistically significant causal reduction of forest cover loss. The effect is sizeable, as indicated by an aggregate average effect of 8 % conserved tree cover compared to non-PAs for the average project period of ten years.

We analyse whether Air Pollution has a negative effect on Health Outcomes using the huge drop in Air Pollution in India during Covid-19 as a natural experiment