Cargo Cult Climate Economics

Part 2 of 2 on a damning new paper that takes on the top-down climate-economics literature — “The empirically inscrutable climate-economy relationship”

“In the South Seas there is a Cargo Cult of people. During the war they saw airplanes land with lots of good materials, and they want the same thing to happen now. So they’ve arranged to make things like runways, to put fires along the sides of the runways, to make a wooden hut for a man to sit in, with two wooden pieces on his head like headphones and bars of bamboo sticking out like antennas—he’s the controller—and they wait for the airplanes to land. They’re doing everything right. The form is perfect. It looks exactly the way it looked before. But it doesn’t work. No airplanes land. So I call these things Cargo Cult Science, because they follow all the apparent precepts and forms of scientific investigation, but they’re missing something essential, because the planes don’t land.” — Richard Feynman 1974

Part 2 of 2.¹ Part 1 explains the Curtin and Burgess (CB26) argument why the climate-economy relationship is empirically inscrutable. Here I walk through what Curtin and Burgess found when they replicated and stress-tested three of the most influential top-down climate damage studies — and what follows from their results.

Note: The annual subscription for THB increases to $100 on June 1. Lock in your support at a lower level!

Last December, Nature retracted “The Economic Commitment of Climate Change,” by Kotz, Levermann, and Wenz (KLW24) — one of the most influential climate economics papers of the past decade. The paper claimed that climate change would cost the global economy $38 trillion a year by 2049 and projected an income reduction of 19 percent within 26 years regardless of future emissions.

KLW24 was the second most mentioned climate paper by the media in 2024, according to Carbon Brief. The paper was cited by central banks and governments to justify more aggressive climate policies.

I was among those who viewed the retraction as good news: science self-correcting, a bad paper removed, maybe things are getting back on track. It turns out there is more to the story — Much more.

The Kotz retraction was not a one-off case of flawed science belatedly retracted.

The new preprint by Finbar Curtin and Matt Burgess, of the University of Wyoming — “The empirically inscrutable climate-economy relationship” — makes undeniably clear that the KLW24 retraction was just a symptom of a much deeper problem in climate economics.

The methodological problems that result in KLW24 run through the entire top-down climate-economics literature. If it was appropriate to retract KLW24, then what now should happen to the hundreds of other papers with the same methodological shortfalls?

In Part 1, I explained the CB26 theoretical argument: the data structure economists use to estimate climate damages cannot, in principle, recover the thing it is trying to measure. Country-year panels pool observations across space and time where and when the relationship between temperature and economic output differs enormously — El Salvador is not Iraq, and India in 1970 is not India in 2020.

Fixed effects cannot solve that problem. The degrees of freedom run out. CB26 call that relationship empirically inscrutable. I go further and argue that seeking to connect two indicies — climate and GDP — can produce numbers, but cannot lead to meaningful results.

Today, in Part 2 I overview the CB26 replication of the three most influential papers in the top-down climate-economics literature. Each paper has similar weaknesses to those that resulted in the retraction of Kotz et al. 2024.

The table below summarizes what CB26 found with respect to those three papers: Burke, Hsiang, and Miguel 2015 (BHM15), Kahn et al. 2021 (KETAL21), and Bilal and Känzig 2025 (BK25).

Note: This table and the source notes below this post have been updated based on comments from Matt Burgess. Thanks Matt!

Across these three papers are the same structural problems as KLW24:

• Results driven by a small number of outlier observations that have nothing to do with weather;

• Sensitivity to specification choices;

• And, the pooling of data across places and times where the relationship of interest clearly differs.

These are not minor technical quibbles. They are the core identification failures that CB26 identified theoretically and also find in practice..

The retraction of KLW24 was due to data errors and an arbitrary lag structure. The data errors were the proximate cause — a decimal-vs.-percentage-point mistake that reviewers should have caught.

The lag structure was arguably worse: Kotz et al. estimated that a year-to-year change in temperature carried effects equally large seven years later as in the present year, then added those lagged effects across years to produce damage totals that dwarfed everything in the prior literature. Remove a single anomalous observation — the Uzbekistan outlier — and the damage estimates collapsed.

Now look at the table above.

• BHM15 nonlinear damage function attenuates when a handful of “growth miracles and growth disasters” leave the sample — events like Iraq’s post-2003 rebound, the Soviet collapse, Oman’s oil boom. These are real economic events with nothing to do with temperature, but they happened to fall in years of unusual weather in countries with volatile climates, and the regression assigned the weather as cause.

• KETAL21 long-run growth effects vanish with different model specifications.

• BK25 $1,200-per-tonne social cost of carbon disappears when local temperatures replace a global average index.

The same methodological issues appear across all four papers: results driven by influential outlier observations, damage estimates sensitive to choices the authors present as defaults, and a fundamental inability to distinguish a real climate-economy signal from statistical artifacts produced by pooling incompatible data.

The Kotz retraction removed one flawed paper, leaving behind a flawed literature.

CB26 focused on three papers, but their critique applies to the entire literature: The top-down methodology those papers rely on — panel regressions of GDP, agricultural output, labor productivity, or mortality on temperature and precipitation, with country or region fixed effects, across multi-decade samples — runs through a large fraction of the empirical climate-economics literature published over the past fifteen years.

The literature CB26 critique includes Dell, Jones and Olken (2012), Moore and Diaz (2015), Kalkuhl and Wenz (2020), Nath, Ramey and Klenow (2024) — each

“relating an aggregate climate index to an aggregate economic index and interpreting the resulting coefficient as a causal damage function.”

The three papers CB26 replicate are the most prominent examples of an approach shared across hundreds of published studies.

The top-down literature did not stay in academic journals. Top-down climate economics papers have been profoundly influential in regulatory frameworks, financial supervisory standards, and legal arguments that govern trillions of dollars in decisions. The table below documents some examples where those studies showed up in finance and policy.²

Long-time THB readers will know that I have been writing for years about implausible climate scenarios — Specifically, how the high-end RCP8.5/SSP5-8.5 emissions scenario was the basis for tens of thousands of research papers, producing the most alarming results, and achieving enormous institutional standing. We now know that those extreme scearnios are implausible — and we have known that for a while, thanks to my colleague Justin Ritchie’s work from a decade ago.

The top-down climate-economics literature has followed a similar path. An approach disconnected from the real world in specific, identifiable ways achieved enormous real-world influence.

In a 2021 paper, Ritchie and I argued that RCP8.5 came to have such a tight grip on climate research due to a confluence of factors — including incentives faced by researchers, a world that welcomes apocalyptic scenarios, and the rewards of publishing research deemed supportive of climate action. I would guess that some combination of those factors are at play in climate economics as well.

For climate economics, I’d add another factor to the mix — climate determinism.

The core idea is old and repeatedly discredited: that climate conditions explain a large share of variation in human outcomes including prosperity, social organization, and economic development.

In the nineteenth century, scholars attributed tropical poverty to the heat. In the early twentieth century, Ellsworth Huntington built a career arguing that temperate climates produced superior civilizations. These ideas fell from favor because the evidence did not support them and the ideological freight they carried was obvious.

Climate determinism has returned — Instead of historians speculating about civilizational effects of climate, economists ran panel regressions with fixed effects and reported precise numerical estimates of how a one-degree rise in a global index of temperature reduces GDP.

The math was new. The underlying claim was old: climate drives economic outcomes in ways that dominate human adaptation and institutional capacity.

The economists working in this literature did not set out to resurrect climate determinism — I have no doubt that they all understand perfectly well that institutions, technology, and governance matter enormously.

However, the top-down methodology they employed requires a simplistic assumption that an aggregated index of weather over a year at the global level can explain outcomes across the global economy. That is climate determinism.

Curtin and Burgess explain the significance of their paper, which in my view is seminal:

Our analysis should provoke a fundamental reevaluation of how climate-econometric studies are used and referenced in research and policymaking. Estimates of economic damages from climate change—and other related quantities like the SCC—affect trillions of dollars in public and private decisions. Often, decision-makers choose one or a small range of preferred damage estimates, which suit their purposes or political preferences (e.g., Democrat vs. Republican administrations’ choices of the SCC), and ignore others. We provide specific examples of such practices in the Section 1. Our analysis suggests that these practices are misguided, and they risk producing misleading or unwise decisions.

They conclude:

The climate–economy relationship therefore remains deeply uncertain. Recognizing this uncertainty is not a failure of economics, but a necessary step toward more honest analysis and more robust policy and practice. Future research should focus less on extracting ever more precise estimates from insufficient data, and more on understanding the mechanisms of adaptation, resilience, and institutional change that will shape economic outcomes in a warming world.

Amen.

Important: If you appreciate being kept up-to-date on the latest climate research, please let me know by clicking that “❤️ Like.” More likes means more readers of THB. Thanks!

Comments welcomed!

It is simple: THB exists because of its paid subscribers. Please consider joining them.

Technical details below.

SOURCE NOTES for TABLE 1 — CB26 Replication Findings

Burke, Hsiang & Miguel (2015), Nature 527, 235–239 — Claims unmitigated warming cuts global GDP per capita 23% by 2100 via an inverted-U damage function peaking at 13°C. CB26 find that removing 6 influential outlier observations attenuates effect sizes by more than 20% and weakens the quadratic relationship, though significance does not disappear entirely. The effect does vanish in the post-2000 sample — a temporal instability with no agreed explanation. Newell, Prest & Sexton (2021) found no statistically significant temperature effect on GDP across 800 model specifications using BHM’s own data. CB26’s conclusion: the magnitude of any climate-economy effect cannot be reliably estimated from this methodology — not that no effect exists.

Kahn, Mohaddes, Ng, Pesaran, Raissi & Yang (2021), Energy Economics 104 — Claims climate damages accumulate permanently into GDP growth rates, with 3°C of warming cutting global GDP by up to 18%. CB26 find that removing 9 influential observations attenuates but does not eliminate the effect. More consequentially, KMN calculate projected damages from the acceleration of climate change rather than its level — a specification choice that produces the counterintuitive result of positive projected effects under low-emissions RCP2.6 for sub-Saharan Africa. Effects prove unstable across moving time windows.

Bilal & Känzig (2026), Quarterly Journal of Economics — Claims 1°C of warming reduces world GDP by more than 20% and puts the Social Cost of Carbon above $1,200 per tonne. CB26 find BK26’s results sensitive to ENSO variation, a bandstop filter, decade fixed effects, and moving time windows in the PWT dataset underlying BK26’s main damage function. Note: BK26 themselves showed their country-level local temperature results carry no significance at 5% — CB26 did not need to demonstrate this. CB26’s core objection is that BK26’s exclusion restriction — that short-run global temperature shocks carry the same indirect effects as long-run temperature changes — is difficult to justify.

Source: Curtin & Burgess (2026), “The empirically inscrutable climate-economy relationship,” SocArXiv. CB26 identify specification fragility and deep structural uncertainty — they do not call for retraction of any paper. The Kotz et al. retraction followed a data error, which is qualitatively different from the specification sensitivities CB26 document.

SOURCE NOTES for TABLE 2 — Where the Top-Down Climate-Economics Literature Shows Up in Policy and Finance

U.S. IWG Social Cost of Carbon — $51/tonne interim value (2021) | DICE model augmented with BHM-era damage functions from the Howard & Sterner (2017) meta-regression of top-down panel GDP studies. | Set the regulatory carbon cost used across the entire U.S. federal government for fuel economy standards, EPA rules, and infrastructure cost-benefit analyses.

EPA Proposed SCC — $190/tonne (2022) | BHM and updated damage functions from the RFF/GIVE model, which explicitly incorporates BHM-era panel regression estimates. | Proposed near-quadrupling of the U.S. regulatory carbon price; drove EPA power plant and vehicle rules; contested in federal court.

CBO Climate-GDP Analysis (February 2025) | Meta-analysis incorporating BHM, Burke & Tanutama (2019), Kalkuhl & Wenz (2020), Kotz et al. (2024, since retracted), and Nath et al. (2024). | Official federal projections of macroeconomic climate damages; CBO noted that Kotz drove substantial upward pressure on its damage distribution.

Biden White House CEA/OMB Climate Macro White Paper (April 2024) | BHM, Dell, Jones & Olken (2012), Kahn et al. (2021), Acevedo et al. (2020), Casey et al. (2023) — all top-down panel GDP studies. | Official executive-branch assessment of climate macroeconomic impacts; used to justify regulatory ambition and characterize risks from inaction.

NGFS Central Bank Scenarios — Phase V (November 2024) | Kotz et al. (2024, retracted) embedded as the baseline damage function; NGFS warned users to “be aware” of the retraction only post-publication. See Roger Pielke Jr.’s analysis. | Adopted by 130+ central banks — Bank of England, ECB, Federal Reserve — as the standard framework for climate stress testing and capital adequacy assessment.

Bank of England Climate Biennial Exploratory Scenario (2021–2022) | NGFS scenarios drawing on BHM-lineage damage functions; Kotz methodology embedded in subsequent NGFS phases. | Directed UK banks to quantify climate-driven losses using top-down damage functions; informed PRA supervisory capital expectations.

SEC Climate Risk Disclosure Rules (2024) | Integrated assessment models incorporating BHM-era damage functions; NGFS and TCFD scenario frameworks tracing to the same top-down literature. | Required ~7,000 public companies to disclose climate financial risks using top-down-derived methodologies; partially stayed by courts.

TCFD Corporate Climate Reporting Framework | Top-down GDP damage literature underpins the “physical risk” scenario quantification in TCFD guidance; BHM-lineage results travel through NGFS into TCFD disclosures. | Adopted globally; mandatory in the UK, EU, Canada, and Australia; trillions in assets under management now subject to TCFD-aligned climate reporting.

IMF Fiscal and Financial Stability Climate Risk Analyses | Kahn et al. (2021) long-run growth effects; BHM; related top-down panel literature informing IMF staff country assessments. | IMF estimates of sovereign debt sustainability, fiscal space, and macroeconomic stability draw on the top-down framework; shape IMF surveillance and lending conditions.

Climate Litigation — Expert Testimony & Damages Claims | BHM, KMN, Bilal & Känzig; high-end SCC estimates derived from top-down damage functions; used in Rhode Island v. Chevron, Hawaii, and other cases. | Expert witnesses argue fossil fuel liability in the trillions; BK’s $1,200/tonne SCC amplifies claimed damages by orders of magnitude.

1. As with part 1, to state the obvious, this post reflects my interpretation of Curtin/Burgess.

2. Some might complain that this post is hard-hitting. Yes, it is.