Do National Adaptation Plans Attract Adaptation Finance?

Evidence from the Composition of Aid

By Pierre Beaucoral, Michaël Goujon, and Sébastien Marchand

Université Clermont Auvergne, CNRS, IRD, CERDI

Summary

Do National Adaptation Plans (NAPs) mobilize additional development finance, or redirect existing finance toward adaptation? Drawing on the aid-allocation literature and the IPCC risk framework, we treat a NAP as a signal of adaptation-planning capacity, the merit side of the classic need–merit dilemma. Adoption leaves a country's climate hazard and exposure unchanged, so it lets us ask whether donors respond to capacity rather than to need.

We combine NAP submission dates (UNFCCC NAP Central) with project-level OECD Creditor Reporting System commitments for 145 recipient countries and 107 donors over 2009–2024, and exploit staggered adoption with the Callaway & Sant'Anna (2021) difference-in-differences estimator. For the 40 countries in the 2021–2024 adoption cohorts, NAP adoption raises the share of commitments marked for adaptation by 2.65 percentage points, from a pre-adoption mean of 6.03%. Adaptation-marked commitments also rise by about 35%, while we detect no change in total commitments or in the probability of receiving any adaptation finance.

These patterns point to NAPs changing the composition rather than the size of recipients' aid envelopes. The unchanged-envelope reading remains suggestive: pre-trends reject for total commitments, and the adaptation-commitment estimate is sensitive to cohort composition. We read the results as a reduced-form allocation response to NAP adoption, not as evidence that NAPs generate additional development finance.

Key Insights

🧩 Composition, not size

The firmest result: the adaptation share of an adopter's commitments rises by 2.65 pp (from 6.03%, p < 0.01), and adopters gain 0.34 pp of the global adaptation pool (p < 0.05). Total and non-adaptation commitments show no detectable change.

💰 How much more adaptation aid?

Adaptation-marked commitments rise by ≈ 35% (ATT 0.304 log points, p < 0.05), mostly in the adoption year. We treat this as an upper bound: much of it comes from the 2021 cohort's unusually low 2020 base year.

🌍 Who responds?

Official donors raise adaptation commitments: DAC bilaterals ≈ 38% and multilaterals ≈ 74% (both p < 0.05). Other donors, mostly private foundations, do not respond, but they gave these countries almost nothing before adoption. The gap is significant only at 10%.

⚖️ Need vs. merit

The signal is not confined to the better-governed: low-governance adopters show a significant effect of their own. But the splits cannot say where the response is larger, and the LDC and income splits point toward the better-off. Gains in global share are other recipients' losses in share, even as every group's adaptation finance kept growing in absolute terms.

Methods & Results Summary

Theory. A simple recipient–donor allocation model in the tradition of Collier & Dollar (2002) and Berthélemy (2006): a donor splits a fixed global budget across recipients and, within each recipient, between adaptation and other aid. NAP adoption raises donor-perceived capacity and generates two empirically separable channels:

  • Composition (H1): the adaptation share of the adopter's envelope rises; adaptation commitments rise unless donors substitute strongly across recipients.
  • Envelope (H1): total commitments fall if donors treat recipients as substitutes and rise if they treat them as complements; the sign is open.
  • Global reallocation (H2): the adopter's share of the global adaptation pool rises.
  • Heterogeneity (H3, H4): a stronger response among donors that screen on capacity, and where baseline capacity is low.

Data and design.

  • Adaptation finance: OECD CRS commitments with the Rio adaptation marker (principal or significant), aggregated to recipient–year, 145 recipients (59 adopters, 86 never-adopters), 2009–2024.
  • NAP adoption: year of first NAP submission to the UNFCCC (NAP Central), an absorbing treatment.
  • Estimation: Callaway & Sant'Anna (2021), doubly robust, never-treated comparison, controls for WGI Government Effectiveness and log population; adoption cohorts with at least five countries (2021–2024; 40 adopters, 126 countries); multiplier-bootstrap inference with simultaneous confidence bands.

Average effects. The within-country adaptation share rises by 2.65 pp (SE = 0.86) from a pre-adoption mean of 6.03%. Adaptation-marked commitments rise by 0.304 log points (≈ 35%) (SE = 0.126), and the adopter's share of the global adaptation pool by 0.34 pp (SE = 0.139). Total (0.072) and non-adaptation commitments (0.040) are statistically indistinguishable from zero, but their pre-trend tests reject, so we read them as descriptive. Disbursements move in the same direction but not significantly (0.131), and adoption does not change whether a country receives any adaptation finance at all: adopters already received some in more than nine in ten pre-adoption years.

Who responds. DAC bilateral (≈ 38%) and multilateral donors (≈ 74%) raise adaptation commitments; other donors do not. Part of the multilateral response runs through the climate funds that also finance NAP preparation: excluding them lowers the multilateral estimate by more than 40%. On the recipient side, every subgroup estimate is positive and of similar size; no difference between subgroups is significant, and the splits lack the power to detect differences of this size.

Is it relabeling? Following projects across years in the CRS microdata, the increase comes mainly from new activities (0.374 log points, p < 0.01), while previously unmarked projects re-marked as adaptation show no detectable change. Excluding NAP-support, readiness and climate-fund flows does not materially reduce the estimate. This cannot tell whether the new projects are additional programming or already-budgeted money placed in new adaptation-labeled projects.

Robustness & caveats. The estimates survive a not-yet-treated comparison group, the de Chaisemartin & D'Haultfœuille estimator, excluding India, a balanced panel and a Poisson model in levels; the exception is measuring finance in raw US dollars, which makes the estimate nearly 30% smaller and not significant. Mitigation commitments do not rise significantly (0.156), which fits a response specific to adaptation, although the gap with adaptation is not significant. We are explicit about the limits:

  • Moving the 2021 cohort's base year off 2020 cuts the adaptation-commitment estimate to between a quarter and two-thirds of its size and removes its significance; the two shares hold up better.
  • Randomization inference favors the two shares; reshuffling adoption years among adopters still gives positive estimates, so part of the effect may be common to all 2021–2024 adopters rather than tied to their submission year.
  • HonestDiD bounds: all three effects lose significance at or before a parallel-trends violation of a quarter of the largest pre-adoption change.
  • Adding the thin 2015–2020 cohorts weakens the adaptation estimate, while total commitments rise for early adopters.
  • The country-level design drops the 37.6% of adaptation-marked commitments booked to regional or unspecified recipients.

Bottom line. A NAP is a way to steer the finance a country already receives, not a key to additional development finance. Because the global pool is shared, adopters' gains in share are other recipients' losses in share, even though every group's adaptation finance kept growing in absolute terms. If donors read plans as readiness signals, countries that cannot yet produce one lose ground. Supporting plan preparation where capacity is weakest is the natural complement to the signal documented here.