Model-based financial regulations impair the transition to net-zero carbon emissions

The paper investigates whether widely used model-based risk regulations in finance, particularly accounting rules that rely on statistical, backward-looking risk estimates, unintentionally disincentivize banks from reallocating capital away from high-carbon sectors toward low-carbon ones. The context is growing concern over climate-related financial risks and the need for policies that mobilize private finance for the net-zero transition. The authors posit that because regulations such as IFRS 9 (expected credit loss accounting) and capital requirements often depend on historical risk models, they may undervalue transition risk in high-carbon sectors and overstate risk in low-carbon sectors. The study aims to quantify the resulting bias in banks’ expected loss provisioning across sectors and to evaluate the implications of divestment from high-carbon assets for banks’ balance sheets, profitability, and incentives, thereby assessing a potentially important barrier to financing the green transition.

The study builds on a body of work highlighting climate-related financial risks, including transition and physical risks, stranded assets, and systemic risk concerns (e.g., Semieniuk et al., Battiston et al., Campiglio et al.). It references policy initiatives like climate stress testing and disclosure and proposals for green-tilted prudential tools (e.g., differentiated capital requirements). Prior research has documented potential macro-financial impacts of disruptive transitions, the exposure of banking systems to climate risks, and limitations of relying solely on transparency and disclosure. The authors situate their contribution in the debate on whether current financial regulation adequately captures emerging climate risks and whether regulatory frameworks might themselves unintentionally hinder capital reallocation needed for decarbonization.

Data and classification: The authors use the European Banking Authority (EBA) 2021 transparency exercise, which reports bank-level gross loan exposure and accumulated provisions (loan loss reserves, LLR) by NACE Rev.2 level 1 sector for EU banks. They combine this with the EBA Risk Assessment exercise that reports the share of Climate Policy Relevant Sectors (CPRS) within each NACE level 1 to classify sectors as high carbon or low carbon. Sectors with a median CPRS share >95% are classified as high carbon: A (Agriculture, forestry and fishing), B (Mining and quarrying), D (Electricity, gas, steam and air conditioning supply), E (Water supply, sewerage, waste management), H (Transport and storage), and L (Real estate activities). Robustness checks include alternative sector classifications, partial allocations based on CPRS shares at more granular levels, and different time windows. Key measure: Provision Coverage Ratio (PCR) = LLR / Gross exposure, used as a proxy for expected credit losses (ECL) under IFRS 9 for each bank and sector cluster (high vs low carbon). Sample: 59 of the largest EU banks (after excluding one due to missing identifiers), covering >93% of total EU banking loans. Banks’ cumulative net profits (2016–2020) are matched via LEI to Bloomberg financials to benchmark income statement impacts. Divestment simulation: Assuming total loan volume per bank is constant and sufficient low-carbon investment opportunities exist, the authors simulate a full reallocation of high-carbon loan exposures to low-carbon sectors. The change in provisions is computed via accounting identities: (1) PCR_ij = LLR_ij / GrossExposure_ij (2) Loan loss provision charges_j = ΔLLR_j = (PCR_low-carbon,j − PCR_high-carbon,j) × GrossExposure_high-carbon,j (3) NetProfit_j,t+1 = NetProfit_j,t − Loan loss provision charges_j This isolates the direct immediate accounting impact (expected loss recognition) of rebalancing from low-PCR (high carbon) to high-PCR (low carbon) assets, holding other factors constant and frontloading the impact into one fiscal year. Additional evidence: To explore why backward-looking models rate high-carbon firms as lower risk historically, the authors analyze financial ratios and Bloomberg 5-year probability of default (PD) estimates for 228 oil and gas firms and 235 renewable energy firms (2010–2021). They compute interest expenses/EBITDA and total borrowing/total assets, and simulate forward-looking stresses: a US$100 carbon tax on Scope 1–2 emissions and a US$20 per barrel write-down of oil reserves, to show how relative risk metrics could reverse under plausible transition policies. Robustness: Results are tested across bank size quartiles, countries, alternative sector labels, partial allocations by CPRS shares, and different time periods (quarterly averages March 2020–June 2022). Data availability prevents direct assignment of emissions to loans; CPRS classification is used as a proxy and is shown to be highly correlated with emission intensity.

• In 2021, exposure-weighted average PCRs are substantially lower for high-carbon sectors (1.8%) than for low-carbon sectors (3.4%) across 59 major EU banks, indicating nearly double expected loss provisioning for low-carbon lending. This pattern holds across bank sizes and most countries, with Italy as a notable exception. Smaller institutions show larger absolute effects, though relative PCR differentials are uncorrelated with size. • Divestment simulation: Reallocating lending from high-carbon to low-carbon sectors would increase portfolio-average PCR by more than 100 basis points for most banks. Smaller banks (lowest size quartile) would see about a 2.35 percentage point increase versus a simple average of ~0.9 percentage points across banks. • Loan loss reserves (LLR): Such a shift would require a gross-exposure-weighted average increase of about 35% in LLR across EU banks. Some banks could see LLR more than double, with impacts increasing in proportion to their initial share of high-carbon loans. • Profitability impact: The associated loan loss provision (LLP) charges could amount to as much as 5 years of profits for some banks, and, on a loan-weighted average basis, about 15% of cumulative 2016–2020 profits across the sample. Aggregate lost profits from the provisioning increase could be on the order of €28 billion for the 59 banks, highlighting materiality. For context, ECB short-term scenario estimates for 41 large EU banks suggest physical risk impacts around €17 billion and transition risk around €53 billion. • Robustness over time and classification: Using March 2020–June 2022 averages yields similar results (around a 100% increase in PCR, 33% rise in provisions, and a 14% impact on prior 5-year profits), and findings persist under alternative sector groupings or partial CPRS-based allocations. • Origins of the bias: Historical financial ratios and model-based PDs favor high-carbon firms. From 2010–2021, oil and gas firms had lower average interest expenses/EBITDA (16% vs 32%) and lower total borrowing/total assets (31% vs 42%) than renewable firms; Bloomberg 5-year PD estimates were consistently higher for renewables. Under plausible transition scenarios, these relationships could invert (e.g., a US$100 carbon tax could raise oil and gas interest expenses/EBITDA from 16% to 46%, and a US$20/bbl reserves write-off could raise their debt/assets from 16% to 86%, far above renewables’ 32%). • Implication: Model-based, backward-looking risk frameworks can create perverse incentives that favor high-carbon lending by making divestment appear costlier in accounting terms, potentially slowing capital reallocation to low-carbon investments.

The findings directly address whether existing financial regulations, by relying on backward-looking risk models, deter banks from divesting high-carbon assets. Empirically, high-carbon sectors have lower model-estimated expected losses (lower PCRs) than low-carbon sectors, so reallocating portfolios to low-carbon investments mechanically increases expected provisions and hits profits, given IFRS 9 expected loss accounting. This creates a misalignment between regulatory incentives and societal goals of decarbonization, potentially contributing to systemic transition risk by delaying the shift of financial resources to low-carbon activities. The analysis suggests that the bias stems from historical financial performance metrics that favored high-carbon firms, which may not be informative under structural change. Incorporating forward-looking scenario analyses of climate and transition risks could better capture emerging risk and help remove inadvertent biases. The results are relevant for policymakers and regulators considering whether current frameworks (including accounting and prudential rules) adequately internalize transition-related systemic risks and whether adjustments are needed to avoid hindering the green transition.

Model-based financial regulations—especially expected credit loss accounting—can unintentionally discourage banks from divesting high-carbon exposures by increasing provisioning and reducing profits upon reallocation to low-carbon assets. Using EU bank data, the study documents substantially lower PCRs for high-carbon sectors and shows that a full divestment could materially increase PCRs and LLRs, erode profits, and thus create perverse incentives that slow the net-zero transition. Evidence from firm-level financial ratios and PDs illustrates how backward-looking models likely understate future transition risks for high-carbon sectors. The paper contributes by quantifying the accounting-driven incentive problem, demonstrating robustness across classifications and time, and highlighting the need for forward-looking risk assessments. Future research should examine whether similar biases are embedded in other model-based regulatory frameworks (e.g., capital requirements), refine methodologies to link emissions and credit portfolios more directly, and design or evaluate forward-looking risk models and scenarios that better capture climate transition dynamics and systemic risk implications.

• Emissions linkage: Direct assignment of carbon emissions to individual loans was not possible; CPRS-based sector classification is used as a proxy, albeit one shown to be highly correlated with emission intensity. • Simulation assumptions: Assumes constant total loan volumes and sufficient availability of low-carbon lending opportunities; frontloads the entire portfolio reallocation into one fiscal year, which may exaggerate timing effects versus gradual transitions. • Scope of impacts: Focuses on direct accounting effects (LLP charges) and does not model indirect or general equilibrium effects (e.g., network spillovers across banks, changes in asset prices or project availability) or broader strategic responses. • Measurement constraints: PCRs are model-based, backward-looking estimates that vary over time; results rely on available EBA data periods and bank self-reported classifications. Country coverage of the CPRS exercise excludes some Nordics, though sensitivity tests suggest low materiality for overall conclusions. • External data: Profit data and firm-level financials (for oil and gas vs renewables) come from Bloomberg and illustrative scenario assumptions ($100 carbon tax, $20/bbl write-down) are stylized stress tests rather than full macro-modelled scenarios.