Climate risk is a key driver of low agricultural productivity in poor countries. In this project, researchers use a cluster-randomized trial to evaluate a novel risk-mitigation approach: long-range forecasts that provide information about the onset of the Indian summer monsoon well in advance of its arrival. In contrast to traditional ex-post risk coping approaches, this novel ex-ante technology provides accurate information significantly before the monsoon’s arrival, enabling farmers to alter major upfront input decisions. Moreover, forecasts have the potential to be disseminated cheaply, even at scale. Researchers assign 250 villages to one of three groups: a control group; a group that receives an opportunity to purchase the forecast; and a group that is offered insurance. This design allows us to investigate farmers’ willingness to pay for forecasts; how forecasts affect farmer beliefs, up-front investments, and welfare; and benchmark these effects against the canonical ex-post loss mitigation tool: index insurance.


Approximately 65% of the world’s working poor depend on agriculture for their livelihoods (Castaneda et al. (2010)). Agricultural production is sensitive to highly variable climatic conditions. When faced with this type of risk, in the absence of full insurance, farmers make fewer profitable investments (Rosenzweig and Binswanger (1993)), exacerbating the gap in agricultural productivity between the developed and developing world (Donovan (2021)). Reducing the negative consequences of agricultural risk is therefore of first-order economic importance. Agricultural risk can be addressed in two ways: ex-ante — with interventions that reduce farmers’ exposure to risk — or ex-post — with interventions to reduce the consequences of negative shocks. Prior efforts have largely focused on ex-post coping strategies. While formal index insurance can improve outcomes substantially (Karlan et al. (2014)), demand is very low, even at actuarily fair rates (Cole and Xiong (2017)), and substantial subsidies are required to increase take-up (Mobarak and Rosenzweig (2014)). A small body of work has focused on ex ante interventions, documenting the potential of new agricultural production technologies to improve outcomes in the presence of climate risk (Emerick et al. (2016); Jones et al. (2022)). However, adoption of profitable technologies remains low among farmers in the developing world (Duflo, Kremer, and Robinson (2008); Jack (2011)), in part because introducing novel technologies can be costly.
In this project, researchers use a randomized controlled trial to evaluate a new ex-ante approach to improving farmer welfare: accurate long-range forecasts that provide information about the onset of the Indian summer monsoon well in advance of its arrival. Monsoon forecasts are promising for four main reasons. First, farmers have inaccurate beliefs about the monsoon’s onset (Gine, Townsend, and Vickery (2015)) and there is clear demand for accurate information, with farmers frequently turning to unvalidated traditional sources such as astrologers and ecological signals (Acharya (2011)). Second, forecasts can be delivered at low cost (e.g. via SMS) and farmers have been shown to respond to digital extension services. (Fabregas et al. (2019); Cole and Fernando (2020)) and information about market conditions (Aker (2010); Allen (2014)). Third, unlike short-run weather forecasts, which allow for marginal behavior adjustments only, this long-range monsoon forecast can lead to nonmarginal changes in agricultural investments, because it provides information that affects the entire growing season delivered well in advance of the monsoon’s arrival. In response, farmers can substantially adjust their production processes. Finally, in theory, a perfect forecast would completely eliminate weather risk.

Despite the large potential benefits of accurate monsoon forecasts, their usefulness has been limited by their limited accuracy. Though the monsoon’s onset is extremely important for the Indian economy (Rosenzweig and Udry (2019)), its climatology is complex, which has made modeling and skillful forecasting difficult (Webster (2006); Wang et al. (2015)). The Indian government’s own forecast, produced by the Indian Meteorological Department (IMD), generates predictions about monsoon onset over Kerala, where the monsoon first arrives in India. However, this is poorly correlated with onset in India’s agricultural regions, limiting this forecast’s usefulness for farmers (Moron, Robertson, and Pai (2017)). The IMD does produce more specialized regional forecasts, but they have remarkably low skill: in much of the country, they are negatively correlated with rainfall realizations (Rosenzweig and Udry (2019)).

In contrast to these existing forecasts, researchers employ a novel, extremely accurate forecast, developed in Stolbova et al. (2016) and maintained by the Potsdam Institute for Climate Impact Research (PIK). This forecast has two main benefits over previous approaches.
First, PIK provides an accurate forecast over the Eastern Ghats. The forecast has particular skill over Telangana, the site of our experiment. In addition to the forecast’s effectiveness in the state — the forecasted onset date has been accurate to within one week in each of the past 10 years — researchers selected Telangana because the state is home to 35 million people and is heavily dependent on the Indian summer monsoon: 55 percent of its workforce is employed in agriculture. The second advantage of the PIK forecast is that it can be delivered to farmers approximately 40 days in advance of monsoon onset, which is substantially earlier than the IMD’s forecast, allowing farmers to make early decisions about key inputs such as crops, planting time, labor supply, and fertilizer purchases.

Researchers use the novel long-range monsoon forecast to answer a series of key research questions, informed by a theoretical model of decision-making under risk – What is farmer willingness-to-pay (WTP) for a forecast of the monsoon’s onset date? How does such a forecast impact farmer beliefs? How does a forecast affect up-front investments such as crop choice, area planted, and fertilizer use? Do these impacts translate changes in key outcomes such as yields, off-farm labor, consumption expenditure, and migration? How do the impacts of
forecasts compare to a canonical benchmark: weather-based index insurance?