Fossil fuel lending has shown no systematic decline over the last decade
To analyse syndicated fossil fuel lending, this study uses data from Bloomberg which reports debt (loans and bonds) provided by banks to fossil fuel companies between 2010 and 2021. The data reports $7.1 trillion of bonds and loan deals extended by 709 banks, the majority of which are syndicated (Fig 1a).
Overall, fossil fuel lending has shown no systematic decline over the last decade (Fig. 1a). Banks provided $592 billion of bonds and loans for oil, gas and coal companies in 2021, 8% more than the yearly average between 2010 and 2016 (the year the Paris Agreement was signed). The distribution of banks’ lending volumes is highly skewed (Fig. S1); the top 30 banks provided 78% of total lending between 2010 and 2021. JP Morgan for example, the largest lender, had an 8% market share in 2021 while the median market share was just 0.03%.
The top tier of lenders will thus be decisive in setting the pace of the fossil finance phase-out. To assess the progress of the phase-out we compared the activity of the top 30 lenders between a pre-Paris (2010-2016) and post-Paris Agreement (2017-2021) period (Fig. 1b). A select cohort of European banks have led the way in the fossil finance phase-out; the four largest banks to decrease their lending by more than 40% from pre- to post-Paris are Swiss (UBS, Credit Suisse), German (Deutsche Bank) and Norwegian (DNB ASA) (Fig. 1b), partly reflecting the stronger climate commitments and policy stringency in the region prompting European banks to price transition risk higher27. However, these decreases appear to be offset by increased lending from large banks in other regions. In particular, two Canadian banks (Scotiabank and BMO Capital Markets) and three Japanese banks (Sumitomo, Mitsubishi UFJ and Mizuho) made substantial increases to their average annual fossil fuel lending (> 25% from pre- to post-Paris).
The ‘big-four’ US banks (JP Morgan, Citi, Wells Fargo and Bank of America) continue to dominate the market and made relatively small decreases in their lending activity from pre- to post-Paris (Fig. 1b). Bank of America decreased its lending by 25% on average from the pre- to post-Paris period, while the others decreased their lending by less than 10% (Fig. 1b). Of the 160 billion USD of annual lending provided by these four banks in the pre-Paris period, only 8 billion dollars has been phased out in the years following the Paris Agreement.
Syndicated lending networks determine the systemic importance of banks
As an individual bank phases out its fossil fuel lending, it not only pulls its own capital from the market but impacts all deals it was party to, and which must find substitute capital. We define a bank’s ‘syndication activity’ as the total value of deals in which a bank is a syndicate partner as a percentage of all deals made in a given year, to measure their influence and potential impact on the sector.
The historical relationships between banks form a lending network in which links are made between banks, weighted according to the number of syndicates in which they are co-participants. Figure 2a shows that the systemically important banks, with high syndication activity, are highly interconnected amongst themselves and sit at the core of the lending network (Fig. 1a, Methods). These central banks also act as bridges to banks that sit at the network periphery (Table S1, Methods). Through their connections, the core banks are thus able to draw in capital from peripheral banks as well as pooling their own capital.
The composition of the network core has changed over time (Fig. 2a). Several European banks have made a significant retreat from the core; Deutsche Bank, for example, goes from the 6th most central bank to the 19th. While Japanese and Canadian lenders moved into the core; Mizuho Financial, for example, increased its centrality ranking from 9th to 5th.
None of the big-4 US banks increased their individual lending between the pre- and post-Paris period, however they retained their positions in the network core and in fact increased their syndication activity (Fig. 2a). Citi’s annual lending, for example, decreased from 35 bn USD pre-Paris to 34 bn USD post-Paris period but its syndication activity increased from 32% to 37% (Fig. 2a). Banks can, in this way, increase their influence in the sector while at the same time lowering their direct investments and the position of banks within syndicated lending networks can reveal this critical aspect of their systemic importance.
Modelling banking phase out
The changing composition of the network core points to a substitution effect, whereby banks pulling out of lending syndicates are replaced by others. As an example, Figure 2b plots the share of Citi’s co-investments made with key syndicate partners over time. Deutsche Bank and UBS have pulled out of syndicates involving Citi to be replaced by other banks such as Sumitomo Mitsui and Mizuho Bank.
In an unregulated market capital substitution means capital flows are resilient to the phase-out of individual banks16. To investigate the role that capital requirements rules could play in counteracting this phenomenon, we introduce a network model to simulate capital substitution between banks. We investigate the role of setting an upper bound, the ‘capital limit percentage’, for the value of fossil fuel bonds and loans a bank can take on its books in a year. This parameter, defined as the maximum annual percentage increase in a banks’ fossil fuel lending, limits their ability to substitute capital when another bank exits the fossil fuel sector and serves as an analogue of capital requirements rules. A reasonable range for the capital limit percentage can be estimated from typical year-on-year changes in banks’ fossil fuel exposures (new fossil fuel assets as a percentage of total assets). While there has been no sector-wide decarbonisation of banks’ lending portfolios between the pre- and post-Paris period (Fig. 4a), year-on-year changes in bank’s fossil fall exposure fall in the range [-60%, +160%] (Fig. 4b, 95% CI). We therefore model the capital limit percentage up to a 200% increase in year-on-year fossil fuel investment. Results are derived using 2021 data (see SI Figure 1 for other years).
Figure 5a shows how a non-zero capital limit percentage leads to an efficiency gap between the value of deals impacted by banks’ phase-out (i.e., deals which have to find substitute capital) and the value of deals which actually fail at the end of the phase-out process in which banks sequentially exit the sector (see Methods for further details).
Initially, when only a few banks have exited the sector, many banks can provide substitute capital to the impacted deals and phase-out efficiency is near-zero. However, as the phase-out progresses, a tipping point is reached where candidate banks for substitution reach their capital limits, due to capital requirements rules, and phase-out efficiency increases sharply (Fig. 5a). The number of banks which must phase-out before this tipping point is reached increases as a function of the capital limit percentage (Fig. 5b) and the period of inefficient phase-out before this tipping point is responsible for the efficiency gap seen in Figure 5a.
Figure 5c shows the phase-out multiplier, defined as the ratio between the capital removed by a phased-out bank and the value of deals which fail due to its phase-out, for different capital limits. For non-zero capital limit percentages substitution results in a phase-out multiplier of zero before the tipping point, since all phased-out capital can be substituted. Beyond the tipping point we observe two types of substitution dynamics in the network. For sufficiently small capital limits (approximately <100%), the multiplier grows larger than 1 beyond the tipping point, meaning that $1 removed by the phasing-out bank results in more than $1 of debt lost to the sector. This is because the direct phase-out of a given bank leads to the indirect phase-out of syndicate partners from specific deals. However, for large capital limit percentages (approximately >100%) the phase-out multiplier never exceeds 1.
Random vs. Targeted phase out
We further model a targeted phase-out scenario in which the largest banks, with the greatest influence in the market (Fig. 1b), phase-out first. Figure 6a shows that for a fixed capital limit, an efficient phase-out requires significantly fewer banks to exit the sector in the targeted phase-out compared to the random phase-out scenario. This is a consequence of the systemic importance of a small number of banks in the fossil fuel sector: one hundred banks chosen at random lend the same total amount as around five of the most systemically important banks (Fig. 2). This targeted scenario results in a smaller efficiency gap than if banks are removed at random (Fig. 6b), and means that fewer banks must be mobilised to exit the sector, through voluntary action or regulation, before the tipping point is reached and phase-out becomes efficient (Fig. 6c).
Model variants and historical data
Results using 2021 data are broadly comparable with results using data from the years 2010-2020, with small variations in the efficiency gap over time (Figure S2 and SI Methods). However, a slowly growing tipping point for fixed capital limits is evidence that increased syndication activity by the top lenders (Fig 2a) is making fossil fuel lending more resilient to banking phase-out.
For completeness, we test alternative versions of the phase-out model. Model variants include changes to how new banking partners are assigned when a deal requires substitute capital (Fig. S3), limits to the number of possible partner banks who can act as the substituting partner (Fig. S4), and limits to the number of times a single deal can acquire substituted capital before failing (Fig. S5). These variants confirm our general result that restrictions to capital substitution increase phase-out efficiency, reduce the efficiency gap, and are necessary to enable meaningful phase-out from the fossil fuel sector.