The challenge of phasing-out fossil fuel finance in the banking sector

The Paris Agreement goal is at risk without a rapid decline in fossil fuel consumption, as projected emissions from existing fossil fuel assets could push warming past 2 °C within decades. Investment in new fossil fuel assets is incompatible with agreed climate targets, implying that financing for existing assets must be phased out along just and orderly decarbonisation pathways. Banks, as the primary source of financing for fossil fuel companies, play a central role in sustaining the fossil fuel economy. Voluntary initiatives (e.g., the UN-convened Net Zero Banking Alliance) have emerged but face accusations of greenwashing and insufficient action relative to net-zero rhetoric. While stranded asset risk is often seen as long term, banks retain financial incentives for short- to medium-term fossil investments due to ongoing exposures. Syndicated lending—where multiple banks jointly finance large deals—dominates fossil fuel financing and creates networks that amplify the impact of individual banks and reduce informational frictions. These networks also facilitate substitution of finance across borders, particularly where climate policies differ, allowing foreign banks to replace domestic banks that withdraw (e.g., Australian coal lending replaced by Chinese and Japanese lenders). A timely, orderly phase-out requires understanding the syndicated fossil fuel debt market and its network structure, which governs who lends with whom and how markets respond to policy and regulation. This study focuses on the syndicated lending network for fossil fuel deals, distinct from interbank liability networks, to illuminate mechanisms of finance substitution and to explore how prudential regulation might counteract them. We describe the network structure, track critical lenders since the Paris Agreement, analyze substitution dynamics, and test phase-out scenarios under different regulatory constraints, including the roles of systemically important banks and regional coordination.

The study contributes to literatures on financial networks and complex systems, emphasizing network topology (core-periphery, centrality) as key to understanding systemic influence in finance. It contrasts with research on interbank liability networks and contagion by isolating syndicated lending relationships in the fossil fuel sector. It connects to climate-finance work on transition risks, stranded assets, stress-testing, and macroprudential policy, noting how market failures and limited international coordination undermine carbon pricing. Prior evidence shows syndicated loans’ importance to fossil finance, informational frictions reduced by syndication, and cross-border substitution in response to uneven climate policies. The paper’s network focus is positioned to enhance climate stress-testing tools and macroeconomic models by providing a more granular representation of banking sector transition dynamics.

Data: Proprietary Bloomberg dataset on syndicated debt (loans and bonds) to fossil fuel companies from 2010–2021, covering 14,391 deals totaling $7.1 trillion. Majority of debt is syndicated (81% by volume). Data include deal sizes and participating banks but lack borrower location and asset type (coal, oil, gas). Comparative checks indicate Bloomberg’s superior global coverage. Syndicated lending network: Constructed from co-investments among banks in fossil fuel syndicates only (other sectors excluded). Edges between banks are weighted by the frequency of co-participation. Centrality metrics used include eigenvector, betweenness, closeness, and degree centrality to identify systemic importance and network core-periphery structure. Phase-out model: Represents the system as a bipartite network between N banks and M fossil fuel deals in a given year (primarily 2021 for simplicity; robustness checked across years). Each bank–deal edge is weighted by the bank’s assumed equal share of the deal’s value (syndicate-level allocations are not observed). Banks phase out sequentially according to scenarios: random order, targeted (largest lenders first), or regional (only banks from a specified region). When a bank phases out, deals it financed become at risk and seek substitute finance. Two substitution regimes are modeled: (1) any substitute—any active bank not already in the deal can replace financing, subject to limits; (2) syndicate substitute—only banks with historical co-activity with the remaining syndicate members are eligible, ranked by frequency of past collaboration. Additional constraints include limiting the candidate pool to the top N ranked banks per deal and limiting the number of substitution attempts per deal before failure. Regulatory constraint: A bank-level cap on fossil lending growth (“finance limit percentage”) sets the maximum annual increase in a bank’s fossil fuel assets (e.g., 10% cap: $1.0bn in 2021 implies $1.1bn maximum in 2022). Empirically, year-on-year changes range roughly from −60% to +160%, so caps from 0% to 200% are tested. An alternative cap based on exposure as share of total assets yields similar qualitative results but affects mainly highly exposed banks unless very strict. If a deal fails, finance withdrawn indirectly from remaining partners is not reassigned (robustness with reassignment shows similar qualitative behavior but weaker efficacy of caps and no indirect phase-out). Metrics:

• Total relative efficiency: cumulative value of failed deals (normalized) divided by cumulative value of impacted deals (normalized) up to each step; approaches 1 for maximally efficient phase-out (all at-risk deals fail) and 0 if all at-risk deals succeed via substitution.
• Efficiency gap: difference between total impacted value and failed value over the process, expressed as a multiple of total sector deal value; larger gaps indicate more inefficient substitution.
• Phase-out multiplier: ratio of cumulative failed deal value to cumulative directly phased-out finance by exiting banks; values >1 indicate that an exiting bank’s withdrawal induces additional indirect phase-out of partners’ finance via failed deals. Assumptions: (i) Phase-out is complete exit from fossil finance without distinguishing coal/oil/gas; (ii) sequential bank exits; (iii) substitutes must be banks not already in the deal; (iv) prudential limits strictly bind; (v) no behavioral adaptations (e.g., recapitalization) considered. Robustness checks across years (2010–2021) show qualitative stability with some temporal changes in ease of reaching tipping points.

• Scale and trend: $7.1 trillion of fossil fuel syndicated debt (2010–2021), with 81% syndicated. No systematic decline in fossil lending over the last decade; 2021 financing was $592bn versus a 2010–2016 average of $584bn.
• Concentration: Highly skewed distribution; top 30 banks provided 78% of total lending (2010–2021). JP Morgan had 8% market share in 2021; median lender share ~0.03%.
• Regional and bank-level shifts: Several European banks substantially reduced lending post-Paris (>40% declines for UBS, Credit Suisse, Deutsche Bank, DNB ASA). Increases observed for Canadian (Scotiabank, BMO) and Japanese (Sumitomo Mitsui, MUFG, Mizuho) banks (>25% increases). US big-4 decreased modestly post-Paris (BoA −25%; JP Morgan, Citi, Wells Fargo each <10% decrease).
• Network structure and influence: Core-periphery network with systemically important banks central and highly interconnected, bridging to peripheral banks. European core presence receded (e.g., Deutsche Bank centrality rank from 6th to 19th), while Japanese and Canadian banks moved into the core (e.g., Mizuho from 9th to 5th). US big-4 retained core positions and increased syndication activity despite modest declines in direct lending (e.g., Citi lending ~ $35bn to $34bn but syndication activity rose from 32% to 37%).
• Finance substitution: Evidence of substitution in syndicates (e.g., Citi’s co-investment shares shifted from UBS/Deutsche towards Sumitomo Mitsui/Mizuho). Substitution makes markets resilient to the exit of individual banks.
• Regulatory tipping points: With prudential caps on fossil lending growth, a tipping point emerges where substitution capacity is exhausted and phase-out efficiency rises sharply. Stricter caps lead to earlier tipping (fewer banks need to exit). Without caps, efficiency remains near zero as deals find substitutes.
• Phase-out multiplier: For caps <100%, multiplier can exceed 1, meaning $1 directly removed triggers >$1 total finance withdrawn via induced deal failures among partners. For caps >100%, multiplier stays <1; at 0% cap (no substitution), efficiency is 1 and multiplier is highest early, then declines as fewer deals remain.
• Targeted exits: Phasing out largest lenders first achieves non-zero efficiency with fewer bank exits than random order and reduces the efficiency gap at given caps; tipping point in terms of finance withdrawn is similar but reached with fewer banks.
• Network-constrained substitution: Under the “syndicate substitute” rule, early efficiency is slightly higher (some isolated deals fail) but tipping is delayed because large core banks, with higher absolute caps, facilitate substitution longer. Further restricting candidate substitutes (top-N partners) or limiting the number of substitution attempts per deal brings the tipping point earlier.
• Regional scenarios: When only a region phases out (syndicate substitute rule), substitution by external regions generally makes phase-out inefficient, with strong asymmetries. Estimated successfully phased-out shares: UK, Japan, EU each <20% of regional volume; US ~25%; Canada ~50%; China ~75% due to relative isolation of Chinese banks. US banks commonly substitute for other regions, with strong interdependencies among US, Japanese, and Canadian banks; EU banks play a minor substituting role for US/Japan/Canada.

The analysis shows that networked syndication enables substitution that undermines bank-by-bank or regional exits, explaining the lack of sector-wide decline post-Paris despite some banks reducing exposure. Prudential regulation that limits banks’ fossil lending growth can induce a tipping point from inefficient to efficient phase-out, especially if rules are stringent. Targeting systemically important banks optimizes the transition by reducing the number of exits required and shrinking the efficiency gap. Regional dynamics highlight the need for coordination, particularly involving US, Canadian, and Japanese banks, given their central roles in substituting finance for other regions. Findings address the research question by demonstrating how network topology and regulatory limits interact to govern phase-out efficacy, offering actionable insights for central banks and supervisors to design macroprudential interventions and inform climate stress-testing with realistic syndication dynamics.

The paper demonstrates that the syndicated fossil fuel lending network is resilient to uncoordinated, voluntary exits but becomes susceptible to efficient phase-out when prudential limits on fossil lending growth are imposed, revealing a regulation-dependent tipping point. Concentration among top lenders and their central network positions make targeted phase-out of systemically important banks especially effective. Regional coordination is crucial; absent such coordination, US (and to a lesser extent Japanese and Canadian) banks are poised to substitute phased-out finance, delaying sectoral decline, whereas China’s relative isolation enables more effective regional phase-out. The study contributes a tractable network model and new metrics (efficiency gap, phase-out multiplier) to guide policy design. Future research should integrate supervisory data, allow for behavioral responses (e.g., recapitalization), extend to asset-type heterogeneity (coal/oil/gas), and embed these network mechanisms into climate stress-testing and agent-based macro-financial models to assess broader economic spillovers and support just, orderly transition planning.

Data limitations include absence of borrower location and asset-type (coal/oil/gas) granularity; reliance on a proprietary dataset with restricted access; and assumption of equal within-syndicate bank allocations due to lack of bank-level deal shares. The network considers only fossil fuel syndication, excluding relationships in other sectors. The model assumes complete phase-out upon exit, sequential bank exits, no re-use of indirectly freed capital (primary specification), and no behavioral adaptations such as recapitalization. Results for figures largely use 2021 data for clarity, though qualitative robustness is shown across years. Regional analysis applies the syndicate substitute constraint and excludes global simultaneous regulation. Alternative exposure-based caps affect primarily highly exposed banks unless very strict.