Yves here. Perhaps readers will beg to differ, but I found this study to be maddening. It makes two assumptions that I find vitiate its validity. Its express purpose is to challenge long-standing studies, such as those done by researchers at the US War College in the early 2000s, that climate change would produce mass migration, for instance, via flooding in regions particularly exposed to sea/ocean level rises such as Bangladesh and Sri Lanka. From NASA:
The analysis is not on point. First, it looks at gradual changes in temperature, the sort of thing that might lead those who had moved to the US Sunbelt to head back north, as opposed to climate change effects that are housing or livelihood-threatening. Even though it was internal displacement, the US Dust Bowl of the 1930s led to 3.5 million moving from the Great Plains. Aside from climate change directly producing out-migration via severe flooding or droughts, climate change also produces resource conflicts, such as over water, or feeds the intensity of ethnic strife.
Second, and I am not making this up, they are assuming legal migration. Were they asleep during migrations that resulted from war, like out of Syria in 2014 and 2015, or the ones in Africa where significant numbers try to get out of the way of upheaval? Crisis-driven mass or low-level migration is often not legal, witness the exodus from Myanmar.
Third, the fact that the well off are mobile and can thus get out of the way of the bad effects of climate change early (when impacts are modest) does not establish that when the consequences become more severe, that lower-income groups will not move out of necessity.
By Céline Azémar, Rodolphe Desbordes, Markus Eberhardt, and Eric Neumayer. Originally published by VoxEU
Climate-driven displacement is widely expected to push millions across international borders. Drawing on monthly bilateral flows for 127 developing-country origins, this column finds the reverse: a temperature shock reduces emigration rather than raising it, because climate stress cuts origin incomes and poor households cannot finance a costly international move. On this evidence, the ‘climate refugee’ is the exception, and hardship more often traps poor populations in place than sets them moving.
Editors’ note: This column is based on CEPR Discussion Paper 21634 “Trapped at home: Weather shocks, income, and international migration”.
The ‘climate refugee’ has become a fixture of policy planning. The Institute for Economics and Peace (2020) estimates that ecological threats place 1.2 billion people at risk of displacement by 2050, and security and migration agencies increasingly treat large cross-border flows as a working assumption. The research record is more cautious. In its dedicated assessment, the IPCC (2022) judged the evidence on climate and international migration to be mixed, with flows that may rise or fall depending on context.
Part of the disagreement is a data problem. The bilateral migration series that is long enough to study slow-moving climate trends, built from the census stocks of Özden et al. (2011), is recorded at five- or ten-year intervals, too coarse to separate a transitory weather anomaly from everything else moving alongside it. With observations that sparse, the confounders correlated with a warming climate – among them origin income, emigration policy, diaspora networks, and the business cycle at both ends of a migration corridor – cannot be held fixed, a limitation that is long familiar from the related literature on climate and income (Dell et al. 2014, Desbordes and Eberhardt 2024).
In a recent paper (Azémar et al. 2026), we revisit the question using monthly data. The migration flows are taken from Chi et al. (2025), who convert the location histories of roughly three billion Meta/Facebook users into a monthly panel of cross-border moves. We restrict the sample to 127 developing-country origins and 180 destinations, from 2019 to 2022. The monthly frequency is what makes our exercise possible: we compare each corridor with itself, stripping out everything specific to an origin-destination pair within a given year, everything tied to a country’s own seasonal calendar, and every global month-to-month shock common to all migration corridors, including the pandemic mobility collapse. What remains is the within-corridor weather anomaly: the deviation of a country’s temperature (or precipitation) in a given month from its own seasonal norm. And it is this residual variation that identifies the migration response.
Warmer-Than-Usual Weather Links to Less Cross-Border Movement
Our results run against the climate refugee narrative. A warmer-than-usual year at the (developing-country) origin is followed by less cross-border movement, not more. A one-degree origin-temperature anomaly reduces the bilateral flow by about 5.7% over the following 12 months, while origin rainfall has no detectable effect.
The estimates survive many robustness checks: alternative fixed effects, dropping the largest migration corridors, re-clustering for spatially correlated weather, alternative specifications for the pandemic controls or removing them entirely, a leave-one-destination-out exercise, and a placebo test that reshuffles the calendar timing of each origin’s weather and finds the effect disappears once the true timing is scrambled. The same effect, and the same null effect on rainfall, reappears in an entirely separate data source: annual arrival flows for 1990–2019 built from three decades of IPUMS census microdata covering 177 origins.
The average effect across countries conceals a sharp asymmetry (Figure 1). Splitting origins at their long-run baseline temperature, the response is close to zero in the cooler half of origins and large and negative in the already-hot half, where a one-degree anomaly lowers emigration by around 16%. The effect is concentrated precisely where the temperature-income relationship is steepest, and precisely where further warming is expected to fall hardest. Splitting the sample instead by income leaves the response unchanged, so this is not a story about the very poorest origins alone: the effect holds across the developing-country income range.
Figure 1 Effect of a one-degree origin-temperature anomaly on bilateral emigration
Notes: Effect of a one-degree origin-temperature anomaly on bilateral emigration, cumulative over twelve months, for all developing-country origins and for origins split at the median of long-run (1990–2018) baseline temperature. Estimated by Poisson pseudo-maximum-likelihood on monthly bilateral flows, 2019–2022.Source: Azémar et al. (2026).
Temperature Shocks Depress Income, Making Moves Unaffordable
Why would heat keep people at home? Moving across a border is expensive. It takes visa fees, travel, recruitment costs, and increasingly the financial reserves that richer-country immigration regimes demand up front. A household that loses income cannot simply substitute towards a costly international move; it may lose the means to finance one at all. This is the resource-constrained immobility trap of Black et al. (2013) and Benveniste et al. (2022), and it carries a clear empirical implication: the temperature shocks that depress emigration should also depress origin income.
We demonstrate that they do so using two separate datasets. On a sub-national grid of GDP per capita (Rossi-Hansberg and Zhang 2025), a one-degree annual temperature anomaly lowers income by about 2%, in line with the macro-climate literature (Burke et al. 2015). A monthly panel of satellite night-time lights, built for the same countries, returns the same negative sign for a weather anomaly on income at the monthly frequency on which the migration response is identified. Read together, the two elasticities imply that each 1% climate-driven fall in origin income is associated with a fall of roughly 2.6% in emigration. When the budget tightens, the move becomes unaffordable.
If the binding force is the financing of the move rather than temperature as such, any shock that erodes origin income and the capacity to pay should immobilise migration in the same way. We demonstrate that it does. Entering a monthly index of internal conflict risk (Aizenman et al. 2026) into our regression models, a rise in origin conflict risk also lowers cross-border out-migration and origin night-time lights, while leaving the weather coefficients untouched. The weather response is not a disguised conflict effect, and the constraint is general rather than specific to climate.
The bilateral structure of the migration data allows for one further test. A wealthier destination costs more to reach, so a budget constraint should bite harder on routes to rich countries. We confirm this by finding a markedly stronger migration response in corridors to higher-income destinations. A larger pre-existing diaspora, which lowers the cost of moving and on certain US-bound routes was found to turn a weather shock into an emigration push (Mahajan and Yang 2020), produces no such amplification here once destination income is accounted for. This points to a financing constraint rather than a network effect shaping climate-induced international migration.
The ‘Climate Refugee’ Policy Assumption Is Wrong
These findings bear directly on how climate displacement is projected. A common forecasting shortcut multiplies a global population at climate risk by a positive emigration elasticity. For the poor, hot origin countries where the constraint binds, that calculation has the wrong sign, and hence overstates near-term cross-border displacement. The populations most exposed to climate damage are usually those least able to finance an international move, so in the near-term climate stress is more likely to dampen orderly emigration from poor origins than to deliver a wave of refugees to distant, wealthy countries.
These findings are troubling. Where the principal welfare gain from emigration accrues to those who stay behind, through remittances (di Giovanni et al. 2015), an income shock that forecloses departure also forecloses the remittances the origin economy relies on. Cross-border movement is itself a form of adaptation, and a constraint that tightens as conditions worsen withholds that adaptation from the people least responsible for the warming that drives it. Whether those who cannot leave the country instead move within it is a margin our cross-border data cannot see, and it is the obvious next question.
On present evidence, the climate refugee is the exception rather than the rule. Hardship of this kind tends to trap poor populations in place rather than to set them moving, and a policy debate built on the opposite assumption is bracing for the wrong problem.
See original post for references
