By Ishan Nath
The global share of employment in agriculture has been falling steadily for decades, declining from 43.4% in 1991 to 26.2% in 2019 (World Bank, ILO). However, Figure 1 shows that the movement out of agriculture seems to have paused, and even reversed, in low-income countries over the past half-decade. Given that agriculture in these countries remains especially unproductive relative to other parts of the economy (Lagakos and Waugh 2013), this pause raises concerns about a potential risk to the conveyor belt of structural transformation out of agriculture that has served as an engine of development for nearly two centuries.
This apparent interruption of long-run trends in sectoral labour shares has coincided with a stagnation of agricultural productivity growth, with real agricultural output per worker in both sub-Saharan Africa and South Asia in 2023 registering below its 2019 levels (World Bank WDI). While a few years of data on agricultural output may just be a blip, a range of recent evidence (e.g. Hultgren et al. 2025) suggests that anticipated warming in coming decades is likely to pose a major long-run risk to global agriculture.
These concerning recent trends in agricultural productivity and specialisation relate to an emerging literature that examines how climate change may slow the movement of labour out of farming. The key intuition is straightforward. In countries where agricultural productivity is low, food is relatively expensive and households must devote a large share of their income to meeting subsistence needs. That consumption has to come from somewhere, so countries with unproductive farms – and limited ability to rely on imports – tend to have high labour shares in agriculture to produce enough food to meet demand. This dynamic is at the heart of what development economists call the ‘food problem’ (Gollin et al. 2007), and it becomes more severe in a warming world with growing risks to agricultural productivity.
The food problem in a climate-stressed, low-trade world
In my recent research (Nath 2025), I formalise this mechanism in a multi-sector model of trade and sectoral specialisation. The paper begins by using global micro-data to empirically estimate the effects of exposure to extreme temperatures on productivity in different sectors, which provides information on the potential value of sectoral reallocation for adaptation. The effects on manufacturing and services are modest: in a high emissions scenario, global manufacturing productivity is projected to decline by about 1.7%, with losses of 10-14% in the hardest hit countries.
In contrast, previous work finds effects about ten times larger in agriculture, with projected losses approaching close to half of productivity in hot, low-income regions. Figure 2 shows the projected relative effects on agriculture vs. manufacturing in a high-emissions scenario considered in Nath (2025). This underscores the critical importance of sectoral specialisation for mediating climate damages – climate change is likely to be much more harmful if these economies stay specialised in the agricultural sector where damages are largest.
Despite the disproportionate vulnerability of agriculture to warming, the quantitative model simulations in Nath (2025) suggest that climate change is likely to keep more labour in agriculture in the very regions where its productivity suffers most. By the late 21st century, agriculture’s share of the economy in the poorest quartile of countries projects to be about 2.8 percentage points higher than it would be without global warming. The mechanism mirrors the traditional ‘food problem’ logic in a climate context: when agricultural productivity declines, food prices rise, and because food demand is inelastic, the economy must allocate more labour to agriculture simply to keep people fed.
In principle, countries facing climate damages in agriculture could also meet demand by importing more food, but in practice the results suggest that this mechanism has weak effects in developing countries. In the model, trade adjustment reduces climate damages by only 1-2% relative to a scenario with autarky, largely because those countries most vulnerable to global warming are also those that are least open to trade in agriculture. In the data, the poorest quartile of countries import only 9% of their consumption, compared with 45% in the richest quartile. The model calibration in Nath (2025) rationalises this pattern with very high trade costs for agricultural goods in poor countries, which mean that even large relative price changes do little to stimulate food imports in the climate change counterfactual.
The paper highlights two especially important and overlooked adaptation margins. The first is trade costs, particularly those affecting agricultural imports in developing economies. When the model counterfactuals in Nath (2025) are re-evaluated in a benchmark scenario in which low-income countries face trade costs more typical of high-income places, predicted climate damages in poor countries fall by roughly half. While the underlying causes of the model-inferred trade costs cannot be fully decomposed into policy levers, the paper shows that about one-third of these hypothetical adaptation gains can be achieved by reducing the proportion of trade costs directly attributable to observable barriers such as tariffs, trade agreements, and regulatory/customs processing delays.
The second adaptation mechanism is economic growth. Because food is a necessity and preferences change with income, rising incomes naturally move labour out of agriculture. Growth also increases the capacity of firms and households to adapt within both agricultural and non-agricultural sectors – for example through irrigation and improved seeds in agriculture or with air conditioning for indoor workers. In the model, end-of-century climate damages in low-income countries are three to four times smaller under scenarios with faster baseline growth than in low-growth scenarios, reflecting both within-sector adaptation and the reallocation of workers away from the especially climate-vulnerable agricultural sector.
Across all these results, a consistent lesson emerges: in a warming world, the share of workers who remain in agriculture will be a key determinant of the economic and human costs of climate change. Policies that reduce trade costs or accelerate growth can meaningfully reduce the severity of these costs by easing the ‘food problem’ constraint that may bind meaningfully as the planet warms.
A growing body of research is examining how global warming interacts with structural transformation and trade
The findings in Nath (2025) contribute to a rapidly emerging body of work linking climate change, structural transformation, and international trade – research that increasingly points toward a common set of mechanisms.