Reducing the cost of capital to finance the energy transition in developing countries

Finance is a critical enabler of climate action, yet access to affordable capital differs markedly across countries, creating barriers to mitigation and adaptation, particularly for renewable energy sources in developing economies where investment risks and cost of capital (CoC) are high. Because renewables are capital intensive, elevated financing costs disproportionately hinder their deployment relative to fossil generation. State-of-the-art integrated assessment and energy system models typically assume uniform financing conditions, biasing projections and limiting analysis of risk-sharing policies. This study asks how incorporating empirically derived, country- and technology-specific CoC into multiple climate-energy-economy models affects the effectiveness, cost, and equity of the energy transition, and what the implications are of policies that converge country risk premia (and thus CoC) between developing and developed countries. The purpose is to improve model realism regarding financing conditions and to evaluate whether fairer finance can simultaneously advance mitigation, energy access, and energy justice.

The paper builds on evidence from the IPCC AR6 highlighting finance as central to accelerating climate action and on research showing that unequal access to finance can impede low-carbon investment. Prior studies have demonstrated that uniform CoC assumptions bias energy system modeling outcomes and that differentiated or reduced CoC can substantially influence technology adoption, system costs, and emissions. Literature on energy justice stresses equitable access to modern energy and the importance of fair financing mechanisms for the transition. Empirical work has quantified drivers of financing costs for renewables, including country risk, technology maturity, and market learning, and has shown that policy design (e.g., auctions, guarantees) can lower CoC and levelized costs. This study advances the literature by combining empirically estimated, country- and technology-specific CoC with dynamic financial learning within an ensemble of IAMs to quantify climate, cost, and equity impacts of a stylized international CoC convergence policy.

Empirical estimation of cost of capital (CoC): The authors derived weighted average cost of capital (WACC) by country and technology using two approaches. For non-hydro renewables (project finance), they computed costs of debt and equity as sums of a global risk-free rate and premia reflecting country default and equity risk, infrastructure risk, and technology risk. Specifically, debt cost equals Global Risk-Free Rate + Country Default Spread + Infrastructure Premium + Technology Premium; equity cost equals Global Risk-Free Rate + Equity Risk Premium + Country Equity Premium + Technology Premium. The global risk-free rate is the nominal 10-year US Treasury yield (1.68% in March 2021). Country default and equity premia were sourced from NYU Stern datasets; infrastructure premia were based on hedonic approaches and industry reports. Technology maturity by country (solar PV, onshore wind, offshore wind) was defined via capacity share thresholds, informing debt shares (60% immature, 70% intermediate, 80% mature) and technology premia (1.5% mature, 2.375% intermediate, 3.25% immature), with an additional technology-specific risk wedge (e.g., onshore/offshore wind above PV: +0.1% debt, +0.6% equity) based on empirical comparisons. For fossil, hydro, and other non-project-financed technologies (utility balance sheet finance), firm-level WACC was estimated using company financial statement proxies: cost of debt as interest expense over total debt and cost of equity as total cash dividends over total equity. Country-level utility CoC was then computed by revenue-weighted aggregation across utilities. To isolate technology risk premia, they regressed the utility WACC net of risk-free and country risk components on country technology generation shares, weighted by GDP, deriving technology coefficients to reconstruct technology-specific WACC for all countries. The resulting CoC data for many country–technology pairs were compiled (Supplementary Tables) and aggregated to model regions by GDP weights. Financial learning (experience curves): The study implemented financing experience curves reducing technology risk over time as cumulative domestic deployment grows. Empirically derived learning implied a 5% reduction in capital cost for each doubling of domestic capacity (learning-by-financing factor b ≈ −0.074). In models with endogenous learning (IMACLIM, IMAGE, WITCH), CoC declines with deployment; GCAM and TIAM adopted exogenous CoC trajectories equal to the median from the endogenous-learning models. This yielded a reference scenario where renewables’ CoC falls by about 1–2 percentage points by 2100 in low-CoC regions and 1–4 points in high-CoC regions. Scenarios and climate policy settings: The study used five IAMs (GCAM, IMACLIM, IMAGE, TIAM, WITCH) and defined scenarios: DEF (model defaults), BASE (empirical, region- and technology-specific CoC, static over time), LRN (financial learning over time, used as CoC-reference), CONV (CoC-convergence: country risk component in developing countries linearly converges to EU/US levels by 2050, combined with learning; main scenario of interest), and SPILL (robustness: risk spillover raising EU/US CoC by +1 percentage point from 2020, shifting the convergence target upward). Two climate policies were evaluated: NDC (countries meet their NDCs by 2030 and maintain equivalent effort thereafter; ~2.6 °C by 2100) and 1.5D (global carbon budget 500–600 GtCO2 for 2020–2100 with cost-minimizing rising uniform carbon price post-2030). Model-specific CoC integration: Implementations were tailored to each IAM. In GCAM, CoC was incorporated via the fixed charge rate (FCR) to annualize capital costs, with FCR adjusted by CoC, lifetimes, and tax-related terms. IMACLIM used CoC as the discount factor in LCOE, feeding a multinomial logit for market shares. IMAGE used CoC to annualize investments into LCOE for logit-based allocation; default CoC was 10% in DEF and replaced with empirical values; carbon taxes were tuned per scenario to meet targets. TIAM applied technology-specific financial hurdle rates (WACC) to uplift capital recovery while leaving O&M unaffected. WITCH, which discounts via the endogenous marginal product of capital, adjusted plant installation costs using an additional hurdle factor removing the effect of the endogenous interest rate and replacing it with exogenous CoC. Mapping from country to model regions employed GDP-weighted averages; where empirical technology data were missing, regional averages (capacity-weighted) were used. Regional aggregation and empirical patterns: Countries were grouped into three macro-regions by CoC: high-CoC (Africa, Latin America, Middle East, non-EU Eastern European and transition countries), mid-CoC (China, India, rest of Asia), and low-CoC (Europe, North America, Pacific OECD). Empirically, high-CoC regions exhibit CoC about 4 percentage points higher, for a given technology, than low-CoC regions. Data and code: Empirical CoC inputs and model outputs are archived on Zenodo (doi:10.5281/zenodo.11545407). Model documentation and selected codes are publicly available (e.g., GCAM, WITCH); processing code for outputs is on Zenodo.

• Empirical CoC differentials: For a given technology, high-CoC regions have on average a ~4 percentage point higher CoC than low-CoC regions. Financial learning alone reduces renewables’ CoC by ~1–2 percentage points in low-CoC and ~1–4 points in high-CoC regions by 2100. • Energy mix and generation: Under CoC convergence, renewable electricity generation in high-CoC countries rises over time. In the NDC setting, renewables increase and fossil generation declines; in the 1.5D setting, renewables still rise while fossil generation is largely excluded by high carbon prices. Renewable electricity demand in high-CoC regions increases by roughly 10% (NDC) and 5% (1.5D) versus CoC-reference (model medians; Extended Data Fig. 4). • Emissions and carbon intensity: In NDC, CoC convergence reduces carbon intensity of electricity in high-CoC regions robustly across models (directionally consistent; Extended Data Fig. 6) and lowers cumulative energy-sector emissions through 2100 (Extended Data Fig. 5). In 1.5D, emissions are already constrained, so the main effect shifts to cost reductions. • Electricity prices: CoC convergence lowers electricity prices in high-CoC countries, stabilizing at a median −10% versus CoC-reference, with most models and regions showing reductions up to around −20% and a few showing small, temporary increases (Fig. 1b). • Bridging the 1.5 °C gap (in high-CoC regions): By 2050, CoC convergence fills on average ~30% of the renewable electricity gap between NDC and 1.5D and ~10% of the fossil phase-out gap, with ranges across models of ~10–69% (renewables) and ~4–38% (fossils) (Fig. 1d). • Policy costs in 1.5D: CoC convergence reduces policy costs (e.g., GDP loss, energy system costs, MAC area) across many regions, including some where financing costs are not the primary barrier (e.g., Middle East, Russia) (Fig. 2b; Extended Data Fig. 7). • Energy affordability and equity: Energy expenditures as a share of GDP fall by ~5% in Africa and Latin America and by ~2.5% in India+ (median across models) in 2100 (Fig. 3a). Inequality in per capita renewable generation (80:20 ratio) declines by ~4% (NDC) and ~2% (1.5D) (Fig. 3b). Renewable capacity Lorenz curves show improved equity with up to a 2-point Gini reduction (Fig. 3c).

The study demonstrates that equalizing access to affordable energy finance via convergence of country risk premia materially improves the effectiveness and fairness of the energy transition. Lowering CoC in developing regions makes capital-intensive renewables cheaper relative to fossil technologies, accelerating electrification—a central pillar of decarbonization—while reducing electricity prices and energy expenditures. In NDC-like worlds, convergence both boosts renewable deployment and lowers carbon intensity, while in stringent 1.5 °C pathways it reduces the overall cost of achieving targets. By shifting more renewable capacity to developing regions, CoC convergence also advances energy justice, reducing inequality in access to modern electricity and enhancing inter- and intragenerational equity. These benefits support a policy case for international risk pooling and fair finance mechanisms—consistent with Paris Agreement Article 6—that channel capital where renewable potential is high but financing costs are prohibitive. Practical levers include risk-sensitive policy design (e.g., auctions), multilateral guarantees, and enhanced roles for development finance institutions to de-risk investments and provide concessional capital. Ensemble modeling is key to establishing robustness of these conclusions across diverse modeling paradigms.

Incorporating empirically observed, country- and technology-specific CoC and dynamic financial learning into an ensemble of IAMs shows that policies driving convergence of financing costs between developing and developed regions can deliver a triple dividend: greener electricity systems, lower mitigation costs, and improved equity. CoC convergence increases renewable deployment in high-CoC regions, reduces electricity prices and energy expenditures, lowers carbon intensity in less stringent policy settings, and reduces policy costs under 1.5 °C constraints. The methodological framework—combining empirical CoC estimation, learning-by-financing, and scenario analysis across multiple IAMs—offers a template for better representing financial dynamics in transition modeling. Future research should examine policy-specific pathways to achieve convergence (e.g., guarantee funds, blended finance), assess potential risk spillovers more comprehensively, and expand coverage to enabling infrastructure (grids, storage, hydrogen) and other sectors beyond power to fully capture system-wide financing implications.

• Policy spillovers: The main convergence scenario assumes no adverse spillover to developed-country CoC; a robustness test with a +1 percentage point spillover to EU/US CoC reduces effect sizes but preserves directional results. A fuller treatment of spillovers could refine conclusions. • Sectoral scope: The analysis focuses on the power sector; enabling infrastructure (grids, storage, hydrogen) is only indirectly considered and often publicly financed, leaving their financing dynamics outside this study’s scope. • Modeling simplifications: IAMs necessarily simplify technology cost evolution, system integration of VRE, and storage representation; these choices can affect quantitative outcomes. • Data gaps and imputation: For some country–technology pairs, CoC components were imputed using decomposed risk elements; uncertainties remain in infrastructure and technology risk premia estimates. • Scenario stylization: The CoC convergence policy is stylized (linear convergence to EU/US by 2050) and agnostic about specific instruments; real-world implementation may be uneven across countries and technologies.