Warming with Borders: Forced Climate Migration and Carbon Pricing

Layer 1: Overview

This paper asks how the threat of forced climate migration — international displacement driven by climate-induced natural disasters — should alter optimal carbon taxation. The motivation is twofold. First, climate change is intensifying natural disasters that disproportionately afflict developing nations, generating large cross-border population flows that existing integrated assessment models (IAMs) ignore. Second, migration and climate policy are simultaneously among the most contested political issues, yet their interaction has received almost no joint economic analysis.

The paper proceeds in two stages. First, it documents empirically that natural disasters cause international migration. Using a global annual panel (165 countries, 1980–2013) from EM-DAT and UN migration flow tables, the paper estimates a fixed-effects regression of log-migration flows from developing (origin) to developed (host) countries on disaster frequency, controlling for GDP per capita and population. The key coefficient implies a semi-elasticity of approximately 2.3%: a unit increase in natural-disaster occurrence is associated with a 2.3% rise in migration to host regions. To link disaster frequency to carbon concentrations, a time-series cointegration analysis yields an elasticity of 13.49 for climatological and hydrological disasters (6.74 when meteorological disasters are added), implying an overall elasticity of climate refugees to CO2 concentrations of 11.87 (5.93 with meteorological events).

Second, these empirical estimates calibrate a quantitative multi-region integrated assessment model (IAM) in which energy-related emissions generate two externalities simultaneously: output damage through temperature, and population reallocation from origin to host regions. The model features a North–South structure (Kyoto Annex I countries as host; rest of world as origin), Cobb-Douglas production with capital, labor, and energy (coal-proxy), region-specific climate damage parameters drawn from Hassler et al. (2019), and a climate module following Golosov et al. (2014). Social welfare in host regions can optionally include a direct disutility from immigration (parameterized using data on European Pay-to-Go programs and the 2016 EU–Turkey Agreement). The model is simulated over 300 years starting from 2015, with 10-year periods.

The paper then analytically characterizes and quantitatively estimates optimal carbon prices under three policy regimes: (1) unilateral host-only action, (2) globally cooperative (first-best), and (3) a Nash equilibrium with all regions active.

The central quantitative finding is an asymmetry across policy regimes. Under unilateral host-region action, accounting for forced climate migration raises the optimal carbon price by approximately 22% (from $44.72 to $54.73 per ton of carbon when calibrated to climatological and hydrological disasters only; to $49.77, an 11% increase, when meteorological events are included). The dominant mechanism is the “Labor Effect”: migrants move without capital and dilute per capita income in host regions because environmental resources and capital are finite, making the negative welfare consequences exceed the positive labor-supply benefit under a Cobb-Douglas technology with climate damages. The social cost of immigration (disutility of anti-immigration sentiment) adds only marginally to the carbon price ($54.99 vs. $54.73 per ton under the Pay-to-Go calibration). When border control is modeled explicitly, a planner facing US-calibrated deportation costs ($4.6 × 10^5 per immigrant) prefers tightening the carbon tax over using border control, validating the main finding. Only when border control is costless does the optimal strategy switch to low carbon taxes and restricted immigration.

In contrast, the globally optimal SCC is nearly unchanged by forced climate migration ($118.62 without FCM vs. $123.03 with FCM), because the Global Labor Effect balances out: costs of population growth in the host are offset by the adaptation benefit of relocating people to less climate-vulnerable areas. Under Nash equilibrium, host SCCs rise modestly ($44.72 to $49.89 under C&H disasters), while origin SCCs fall slightly ($73.81 to $72.51) as migrants, once relocated, face lower climate damages. The welfare cost to host-region natives from applying the no-FCM policy when FCM is in fact present amounts to a 0.193% permanent consumption equivalent.

Policy implication: in the absence of a global climate agreement (the prevalent situation), developed countries have substantially stronger unilateral incentives to price carbon than existing IAMs suggest, because they indirectly bear the economic costs of climate-induced immigration. The global SCC, however, is not materially affected, so the case for international coordination rests on the same foundation as before.

Layer 2: Deep Dive

What is the empirical identification strategy and what are the main threats to it?

The empirical strategy exploits the quasi-random timing of natural disasters within an origin country using a two-way fixed-effects (country and year) panel regression. The dependent variable is the log of annual unilateral migration flows from each origin country to the pooled group of host countries (43 OECD-type destinations). The independent variable is the frequency (or log frequency) of climate-related natural disasters in the origin country in the same year. Country fixed effects absorb time-invariant push/pull factors; year fixed effects absorb common global shocks. Main threats discussed: (1) Endogeneity of contemporaneous GDP and population, addressed by using first lags of controls. (2) Reporting bias in EM-DAT (disasters in early years may be under-recorded), addressed by computing the ratio of warming-related to geophysical disasters (reporting bias should be type-orthogonal) and by restricting to large disasters (>=1,000 affected or >=100 deaths). (3) The paper focuses exclusively on the contemporaneous (same-year) migration response, treating lagged effects as lower bounds. (4) The semi-elasticity estimates are used as calibration inputs, not as causal estimates of structural parameters — the author acknowledges the causal chain from concentrations to disasters is not fully established.

What are the four theoretical components of the unilateral host SCC and how do they combine?

The unilateral host SCC (equation 12) is the sum of: (1) Standard Output Damages — the present discounted value of climate damage to final output, the only component in standard IAMs; (2) Emissions Reallocation — the reduction in origin-region emissions as migrants move to the host, which lowers global concentrations and benefits the host, making this component negative (it reduces the carbon price); (3) Immigration Social Cost — the direct disutility of newly arrived immigrants borne by host natives (parameterized by gamma), which adds to the carbon price when gamma > 0; and (4) Labor Effect — the net welfare consequence of a larger host labor force, which comprises a positive externality (higher output) and a negative externality (dilution of per capita consumption due to finite environmental resources and capital). Under Cobb-Douglas production with climate damages and capital (Result 1), the net Labor Effect is always a negative externality that raises the carbon price. In the quantitative exercise, the Labor Effect dominates all other FCM-related components and accounts for essentially the entire 22% increase in the unilateral SCC.

Why does the global SCC remain nearly unchanged when forced climate migration is included?

The global planner internalizes the welfare of both host and origin regions. The ‘Global Labor Effect’ contains two offsetting terms: costs to host natives from capital dilution and per capita income reduction, and benefits to origin-region emigrants who move to a less climate-vulnerable, more economically developed area. These effects largely cancel. In addition, migration reallocates economic activity away from high-damage origin regions, lowering expected global climate damages. Migration costs calibrated to equalize consumption per capita across regions (absent climate change) prevent the global planner from strategically using pollution to trigger welfare-improving migration. Quantitatively, the global SCC rises only slightly, from $118.62 to $123.03 per ton of carbon (less than 4%), and may even fall after roughly four decades as the adaptation benefit grows.

How is the social cost of immigration (anti-immigrant sentiment) parameterized and calibrated?

The parameter gamma represents the marginal social cost of immigration to native households — their willingness to pay to prevent a marginal unit of immigration. Two calibration approaches are used: (A) Pay-to-Go programs: using data on European Assisted Voluntary Return programs in 2015, the paper derives gamma = 7.1 × 10^3 (in terms of final good per billion migrants). (B) EU-Turkey Agreement: using costs from the 2016 deal managing the Syrian refugee influx, the paper derives gamma = 7.3 × 10^3. The similarity of the two estimates provides cross-validation. The baseline quantitative exercise disables this feature (gamma = 0), treating it as a sensitivity; a UK Brexit-era survey value implies a four-fold increase in the unilateral SCC but is judged unrepresentative of permanent preferences. The paper is explicit that these are positive descriptions of political preferences, not normative endorsements.

What heterogeneity in the migration response is documented empirically?

Three dimensions of heterogeneity are explored: (1) Income: Unlike for slow-onset climate migration (where middle-income countries drive the response), poorer countries show a stronger migration response to disasters (positive and significant interaction between disaster frequency and a poor-country dummy, column 4 of Table B.1). This is interpreted as evidence that migration costs are less binding when disaster severity forces departure. (2) Disaster type: Climatological and hydrological disasters have higher and statistically significant migration-response coefficients than meteorological disasters (Table B.5). This differential is why the paper presents results under two calibrations (C&H disasters vs. C&H&M disasters). (3) Disaster severity: Restricting to large disasters (>=1,000 affected or >=100 deaths) yields an even larger migration response (column 5 of Table B.1).

What robustness checks are run on the empirical results?

The paper runs an extensive set of checks reported in Online Appendix B: (1) Zero-inflated negative binomial (ZINB) model to handle zeros in the dependent variable. (2) Bilateral migration flows with origin-destination fixed effects. (3) Three-year non-overlapping windows (to reduce zero mass in independent variable), which more than doubles the estimated coefficients. (4) Per capita migration as the dependent variable. (5) Disaster frequency weighted by share of affected population. (6) Inverse hyperbolic sine (IHS) transformation. (7) Excluding China and India. (8) Excluding Singapore and South Korea. (9) Controlling for conflict (battle-related deaths). (10) Controlling for a climate vulnerability index. (11) Controlling for the second lag of disasters. (12) Polynomial regression to check for acceleration. (13) Poisson specification. (14) Checking that an upward trend in disaster ratios relative to geophysical events is not attributable to reporting bias. Results are consistent across all specifications.

What is the Nash equilibrium result, and how does it differ from both the unilateral and first-best settings?

In the Nash equilibrium, each region implements its own best-response carbon policy. Host regions’ NE SCC resembles the unilateral SCC (Section 4) except that the ‘Emissions Reallocation’ component drops out, because when all regions are strategically active, the host cannot treat origin emissions as exogenously reduced by migration. Quantitatively, host NE SCC rises from $44.72 (no FCM) to $49.89 (with FCM, C&H disasters) — a roughly 11.5% increase. Origin region NE SCC falls slightly from $73.81 to $72.51, because origin planners care about the welfare of their emigrants who now live in lower-damage host regions. Without FCM, the origin SCC is 1.6 times higher than the host SCC (reflecting greater vulnerability and larger population in origin). With FCM, this gap narrows. The NE global SCC is lower than the first-best because each region only partially internalizes the global externality.

How does the border control extension interact with the optimal carbon tax?

When the host planner can choose both a carbon tax and a border control stringency (share of migrants admitted), the optimal carbon tax with FCM is lower than in the no-border-control case, because restricting migration inflows reduces both the Labor Effect cost and the Immigration Social Cost. At the same time, restricting inflows reduces the Emissions Reallocation benefit. In equilibrium, the marginal cost of deportation equals the net benefit of keeping an additional immigrant out. Quantitatively, when border control costs are calibrated to US Department of Homeland Security data ($4.6 × 10^5 per detained immigrant), the carbon tax remains essentially equal to the no-border-control case and migration inflows are also nearly unchanged — the planner finds it optimal to abate emissions rather than pay deportation costs. Only when border control is costless does the planner switch to a low carbon tax and high migration restriction. This sensitivity analysis validates the main finding under realistic border enforcement costs.

How does this paper relate to, and differ from, Cruz and Rossi-Hansberg (2024)?

Cruz and Rossi-Hansberg (2024) use a highly spatially disaggregated model with endogenous migration to quantify welfare costs of climate change under an exogenous global carbon tax. The key differences are: (1) This paper derives optimal carbon taxes — both globally and regionally — rather than taking them as exogenous. (2) This paper provides closed-form analytical characterizations of the SCC under multiple policy regimes, enabling clear decomposition of mechanisms. (3) Migration in this paper is exclusively ‘forced’ (disaster-driven), not microfounded by economic incentives (though Appendix F relaxes this); Cruz and Rossi-Hansberg treat migration as fully endogenous to economic conditions. (4) This paper explicitly analyzes strategic interactions (Nash equilibrium) between regions. (5) This paper can account for anti-immigration sentiment (gamma) and border control policies. The approaches are thus complementary: Cruz and Rossi-Hansberg offer richer spatial geography and fully endogenous migration; this paper offers analytical tractability and policy-regime analysis.

What are the policy implications and their scope conditions?

The principal implication is that developed countries (host regions) have approximately 22% stronger unilateral incentives to impose a carbon tax than existing IAMs indicate, once climate-induced international displacement is accounted for. This result holds under climatological and hydrological disasters calibration and US-level border enforcement costs; it is smaller (~11%) when meteorological events are added and even smaller when border control is assumed freely available. The global SCC is barely affected, so the normative case for a global agreement is not strengthened or weakened in magnitude, but the analytical structure of the globally optimal tax is qualitatively different. Scope conditions: the model abstracts from internal migration, micro-founded voluntary migration, endogenous TFP growth, and capital mobility across regions. Results are robust to Stern discounting, more catastrophic damage functions, and Negishi weights. The welfare cost of ignoring FCM in policy design is modest in magnitude (0.193% consumption equivalent) but positive and policy-relevant as a systematic downward bias in host-country incentives.

What does the microfounded migration extension show?

Online Appendix F relaxes the forced-migration-only assumption by introducing economically motivated migration: individuals in the origin choose migration based on consumption differentials across regions, subject to migration costs calibrated to eliminate non-climate migration at steady state. The host unilateral SCC rises to $79.52 per ton of carbon under microfounded migration, compared to $54.73 under forced-only climate migration and $44.72 with no migration (Table F.1). This indicates the 22% increase in the main analysis is a lower bound: broader climate-related migration (including voluntary economic responses to climate shocks) would generate even larger incentives for host regions to tighten carbon pricing. However, this extension sacrifices analytical tractability and closed-form solutions.

What is the welfare cost of ignoring FCM?

Table 6 reports the welfare cost of applying the sub-optimal ’no FCM’ carbon tax to a world in which FCM is actually occurring. The cost is measured as the percentage increase in consumption in every period that would be needed to make host-region natives as well-off as they would be under the correctly calibrated FCM-inclusive policy. Without immigration disutility, the cost is 0.193%. With the Pay-to-Go disutility calibration, it is 0.195%. These figures are small but positive and increasing in the social cost of immigration. They represent the aggregate efficiency loss to host-region natives from the systematic underestimation of the unilateral SCC in existing IAMs.

How is the migration–concentrations link empirically constructed for model calibration?

The paper uses an elasticity decomposition: the elasticity of climate refugees to CO2 concentrations is the product of two elasticities. The first — the elasticity of migration to disaster frequency — is estimated from the panel regression and equals 0.88 after pooling countries into two regions. The second — the elasticity of disaster frequency to carbon concentrations — is estimated from a time-series cointegration analysis following Thomas and Lopez (2015), yielding 13.49 for climatological and hydrological disasters alone and 6.74 when meteorological events are included. The product gives overall elasticities of 11.87 and 5.93 respectively. These are then used to calibrate the linear migration function B (the flow of migrants per unit change in carbon concentrations), using historical average concentration increases, average migration flows relative to host population, and the elasticities. B = 5.03 × 10^-5 (C&H disasters) or 2.52 × 10^-5 (C&H&M disasters).

Key Concepts

Forced Climate Migration (FCM): In the paper’s usage, the specific subset of climate migrants who are forced to move internationally because of climate change-induced natural disasters (rapid-onset events such as floods, storms, and heatwaves), as distinct from voluntary economic migration or migration driven by slow-onset climate variables such as temperature trends.

Social Cost of Carbon (SCC): The monetary value of the present and future economic damage caused by a marginal one-unit increase in carbon emissions today, which under the Pigouvian framework equals the optimal carbon tax. The paper distinguishes three variants: the unilateral host-region SCC, the globally optimal (first-best) SCC, and the Nash-equilibrium SCCs for host and origin regions.

Labor Effect: A novel component of the unilateral SCC in the model, capturing the net welfare consequence of a larger host-region labor force due to FCM. It contains a positive sub-term (higher labor raises output) and a negative sub-term (capital dilution and reduction in per capita consumption because environmental goods are finite). Under Cobb-Douglas production with climate damages and capital, the net Labor Effect is always negative (raises the carbon price), as shown in Result 1.

Emissions Reallocation: The reduction in origin-region emissions that mechanically follows when population — and therefore emission-generating activity — moves from the high-emission-intensity origin region to the host region. This component enters the unilateral SCC with a negative sign (it reduces the carbon price), because the host planner benefits from lower global concentrations induced by fewer emitters in the origin.

Social Cost of Immigration: The direct disutility experienced by host-country natives from the arrival of immigrants in the current period, parameterized by gamma, representing the native household’s marginal willingness to pay to prevent an additional unit of immigration. It is calibrated using data on European Pay-to-Go programs and the EU–Turkey Agreement. It adds to both the unilateral and Nash-equilibrium host SCCs, but quantitatively contributes only a small increment above the Labor Effect.

North-South Calibration: The paper’s two-region parameterization in which ‘host’ corresponds to Kyoto Annex I countries (most European nations, the United States, Canada, Australia, New Zealand) and ‘origin’ corresponds to the rest of the world. Host regions have higher GDP per capita, lower climate vulnerability parameters (theta), and higher emissions per capita; origin regions are more exposed to climate damages and more densely populated.

Nash Equilibrium (non-cooperative) SCC: The carbon price chosen by a local planner as the best response to other regions’ optimal strategies, without the Emissions Reallocation component (since other regions’ emissions are now also strategically set). In this setting, host SCCs rise relative to the no-FCM benchmark but less than under unilateral action; origin SCCs fall slightly because origin planners account for the welfare of emigrants residing in host regions.

Integrated Assessment Model (IAM) with FCM: The paper’s quantitative framework that combines a neoclassical multi-region growth model, a climate module following GHKT (Golosov et al. 2014), region-specific damage functions, and an endogenous migration flow driven by carbon concentrations. The model is solved by direct optimization over savings rates and energy-labor shares, simulated for 300 years, with each period representing 10 years.