Working Papers
- “Network-Based Targeting with Heterogeneous Agents for Improving Technology Adoption,” May 2026. Working Paper, X Thread
Revised and Resubmitted (2nd round) to the Journal of Development Economics - Winner of the Young Scholars’ Best Paper Award at the International Conference hosted by the Centre for International Trade and Development, Jawaharlal Nehru University.
Abstract
Can we use social ties to improve information diffusion when agents differ in potential gains? This study investigates network‐based interventions that aim to improve the adoption of agricultural technologies when expected benefits are heterogeneous. I develop a two-stage learning model in which uninformed agents rely on their neighbors to decide whether to conduct costly experimentation to become informed and then whether to adopt once informed. Simulations reveal that when technology benefits vary widely, targeting highly central households can be less efficient than targeting randomly, particularly when experimentation costs are low. In contrast, targeting high-benefit households maximizes diffusion in highly heterogeneous environments. To validate these findings, I combine experimental and observational data from Malawi. By evaluating the marginal returns to continuous variations in targeted household characteristics, the empirical analysis supports the simulations' joint directional predictions. These findings underscore the need to consider both network structure and benefit heterogeneity when designing network‐based targeting policies to promote new technologies.
Abstract
The decision to adopt one technology versus another depends on how uncertain the decision maker views each technology. Attitudes towards risk (known probabilities) and ambiguity (unknown probabilities) have been shown to partially explain the observed sub-optimal adoption of agricultural technologies in developing countries. Although social learning can help resolve associated information frictions, and peer learning interventions are gaining traction, we know little about how these interventions work: is it access to information, or is it the active seeking of information that leads to effective knowledge transmission? To answer this question, we conducted an artefactual field experiment with potato farmers in Peru that focuses on their beliefs about the relative riskiness and ambiguity of different strategies of dealing with Late Blight. Our experiment allows us to understand the role of active discussion in improving coordination about common beliefs. We find that active discussion does not lead to better coordination about the common belief in our setting. Further analysis reveals the reason to be the inertia of majority, where a rigid majority of farmers holding structured (risky) beliefs refuse to coordinate on a belief different from their private beliefs, while the minority holding unstructured (ambiguous) beliefs lack the coherence to form an alternative bloc. These results suggest the need to complement bottom-up social learning interventions with top-down knowledge interventions to destabilize private beliefs and improve coordination.
- “Learning from Experience about Product Design: Evidence from Index Insurance in Kenya,” February 2026. Working Paper
Abstract
This paper studies how experience shapes learning and demand for index insurance products. I develop a theoretical framework that distinguishes between learning about the underlying risk environment and learning about the index insurance product's design quality, highlighting the potential for payouts to rationally reduce demand by serving as an information signal that corrects optimistic priors about the product's design. I test these predictions with data from the Index-Based Livestock Insurance (IBLI) pilot in Kenya. Using the random distribution of price subsidies and the exogenous timing of payouts, I causally identify that receiving a payout significantly reduces the responsiveness of demand to discounts on the extensive margin, with no differential effect of the disaster experience. Further analysis suggests that payouts serve as an information signal that corrects optimistic priors about the product's design quality, consistent with the theoretical framework's predictions. These findings challenge the predominant view that subsidies lead to long-term adoption of insurance products by providing evidence that subsidies can prompt rational market exit by accelerating the discovery of product design imperfections.
Abstract
Do people learn from experience how to cope with weather shocks? We use a unique four-wave panel household dataset from Uganda, merged with granular historical weather records, to understand the nexus between experience, weather shocks, and agricultural performance. Our identification strategy exploits cross-sectional variation in the climate experience of immigrant members of the households and the temporal variation in the realization of the weather shocks during the survey years. We show that although temperature shocks can be detrimental to agricultural performance, households with more experience perform relatively better. An additional 10 days of temperature shocks reduce the income of households with little experience by 8 percent, while the effects are negligible for those with higher-than-average experience. Our findings are robust to various robustness checks, including placebo tests on the timing of shocks and falsification tests. Suggestive evidence points towards the adoption of risk-reducing technologies as the driving factor behind the gains of the more experienced households. These findings highlight the relevance of initiatives that promote experiential learning.
Abstract
In settings with variable local geographic conditions, the impact of interventions can be confounded by heterogeneity in farmers’ ability to convert input into output. This paper introduces a novel plot-level measure of agricultural inefficiency that accounts for both input use and geographic endowments, enabling a more accurate assessment of intervention impacts compared to the conventional choice of actual yield as the outcome variable. We use this measure to evaluate a mobile phone-based agricultural extension program in rural Bangladesh. We observe that, in treated villages, after intervention, there is a 50 percent reduction in plot-level inefficiency, driven by plots that used rainfed water for cultivation. We found these effects to be driven by increased input usage by farmers doing rainfed farming. In addition, we document that the intervention benefits geographically remote farmers more, and find significant cross-community spillovers through geographic ties.
Work in Progress
- “Food Transfers in a Primitive Economy: Comparing Three Models of Risk-Sharing in Networks.”
with Francesco Amodio and Pau Milán
Details
We compare the empirical predictions of three models of risk-sharing — altruism, capacity constraints, and local information frictions — using food transfer data from an indigenous population in the Bolivian Amazon.
Book Chapter
- “Food Insecurity, Price Volatility and Trade: A Panel Data Analysis in Developing Countries.” Book Link
with Panchanan Das and Swayambhu Mukherjee
(published as Chapter 10 in Indian Agriculture under Multilateral and Regional Trade Agreements - Competitiveness and Food Security, Sharma and Bathla (eds.), CWS in association with Bookwell, Delhi, 2017, 177-194.)
Other Papers
- “Industrial Performance in West Bengal: Analysis of Technical Efficiency with ASI Data,” August 2016. [Available Upon Request]
with Panchanan Das and Swayambhu Mukherjee
Media/Posts
- “Information is power: ICT and agricultural productivity,” Ideas for India, April 2025. Link
with Digvijay Singh Negi and Rahul Rao - “Using Social Ties to Improve Technology Adoption: Does Heterogeneity Matter?” August 2023. Link
