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

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

Data

We used data from the 2021 EBA transparency exercise, which provides portfolio-level information of banks’ gross exposure and accumulated provisions (LLR) by NACE sector level 1 at the end of June 2021. We used the most recent data, but with additional robustness analysis, ensured that the results do not change using different years (the reader should note that due to the structure of this modelling, the provision coverage ratios oscillate with time in level but the relative difference across sectors is generally preserved). NACE is a standard classification of sectors in the European Union. It has various levels of granularity from 1 (least granular) to 4 (most granular), and the EBA transparency exercise relies on this classification. The exercise is an annual data collection to foster transparency and to complement banks’ own disclosures. The data published includes 111 EU banks across 25 countries and provides information regarding banks’ assets, liabilities, loan loss provisions and other financial information for each bank.

We used the legal entity identifier (LEI) code in the EBA dataset to complement this information with the historical net profit data from Bloomberg. The data identifiers were matched with each LEI code in our sample through manual research on the Bloomberg terminal. We started from the largest 60 banks in our sample representing 95% of the total banking exposure, but we excluded one bank because its name and LEI code were missing, which did not allow us to retrieve their income information. This bank represents ~2% of total EU banking assets. After this manipulation, our dataset covered more than 93% of total banking loans in the European Union and provided us with LLR, total lending amount for all NACE sectors (level 1) and cumulative net profits from 2016 to 2021 for the largest 59 banks in the EU. A summary of the sector-level statistics is reported in Supplementary Table 1.

High-carbon sectors classification

We added to this dataset the information necessary to classify sectors as high carbon (that is, sectors with high levels of emission intensity). Specifically, we complemented the data with the results of the EBA Risk Assessment exercise, which provides median values of CPRS as defined in ref. 5 within each NACE level 1. CPRS is a classification used to assess the exposure of investments to transition risks, including carbon taxation, and is a proxy for the level of carbon emissions associated with an investment. The exercise was carried out by the EBA and a sample of 29 volunteer banks from 10 countries representing 50% of the total EU banking assets, with the objective of obtaining a preliminary quantification of the exposure of banks to climate-related risks, particularly focusing on transition risk. The data annex provided (publicly available) discloses the share of CPRS sectors in each NACE level 1 section according to banks’ classification of their own clients in CPRS. This information is particularly useful because it allows us to have a more granular labelling of low-carbon and high-carbon sectors than the NACE level 1 (which would not be sufficient to address the heterogeneity of some sectors). The CPRS rely on NACE level 4, which provides a better discrimination between climate-sensitive sectors and others (additional information provided in Supplementary Information 2).

The bank-level information on total gross loan amount and LLR by NACE code were grouped into high-carbon and low-carbon sectors. We defined sectors as ‘high-carbon’ if they had a median share of CPRS higher than 95%, as reported by banks in the EBA Risk Assessment exercise. This gave us the following high-carbon sectors and their respective codes: 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. We acknowledge that our approach has limitations, but we extensively tested the robustness of our results to a change in the methodology used to classify low-carbon and high-carbon sectors (Supplementary Information 1). Moreover, we compared our classification to more granular data reporting emission intensity to provide transparency about their level of correlation. It should be noted that the banks participating in the climate risk exercise did not include Sweden, Denmark and Norway, but results do not change if those countries are excluded due to their relatively low materiality in the overall EU banking system.

Data availability prevented us from assigning carbon emissions to loans directly. However, the CPRS classification we used is highly correlated with GHG emissions intensity (Supplementary Fig. 1). The EBA Risk Assessment provides a breakdown of emission intensity by percentiles for CPRS and non-CPRS. They use individual firms’ GHG emissions from the data provider Trucost (representing 30% of total banks’ loan amount) and a proxy based on the average GHG emission intensity at NACE rev2 level 4 for the remaining loan amount. Each bank loan is classified in percentiles of emission intensity in a range from very low to very high (more details could be found in the EBA 2020 Risk Assessment Report, Table 19). We used these data to test the correlation between the share of loan amount in CPRS/non-CPRS and its emission intensity. There is a clear correlation between the share of loan amount of CPRS and the clusters of emission intensity (Supplementary Fig. 1). Around 85% of the loan amount classified as having ‘very high’ emission intensity are in CPRS. At the opposite end of the spectrum, only 8% of the loan amount of CPRS are in the ‘very low’ emission intensity bucket. The correlation between the share of loan amount in CPRS (non-CPRS) and its emission intensity is therefore strongly positive (negative) and around 90% (−90%). In Supplementary Information 1, we show that this correlation is very unlikely to change with different classifications using a set of robustness analyses.

Simulation of a divestment strategy

Using the data available, we could provide an estimate of the potential impact of a divestment from high-carbon assets on EU banks’ financials. The primary assumption in this estimation is that the total amount of loans of each bank is left unvaried. In other words, the simulation assumes that banks shift their lending portfolio directly from high-carbon to low-carbon investments. We also assume that sufficient low-carbon investments are available for these transactions. The labelling in our data allowed us to calculate the average risk estimate (PCR) of low-carbon and high-carbon sectors for all banks in our sample. We made use of the accounting relationship between provisions coverage ratio, LLP charges and net profits to assess the impact of a divestment from high-carbon assets on these metrics (all else being equal). Importantly, we did not rely on an explicit economic model, but on the accounting relationship among these metrics. In turn, our results were generated by the structure of the regulation as long as a bank divests from a low-PCR asset and re-invests in a high-PCR asset.

It should be noted that LLP changes are only the direct effect of this divestment on bank’s net profit changes at the time they make the investment. This is an expected loss, not necessarily a loss that will occur in the future. More specifically, three conditions need to be satisfied to generate an increase in costs from a divestment by banks:

  1. 1.

    Losses are costs that must be accounted for as ‘expected’ as opposed to ‘incurred’. That is, financial firms must account for any change in the portfolio expected losses, not the actual incurred losses;

  2. 2.

    Provision coverage ratios must be equal to model-based estimates of ‘expected losses’. That is, expected losses are proportional to measures of risk;

  3. 3.

    Risk estimates of the asset in which a bank is divesting are lower than the asset in which it is making a new investment;

Conditions (1) and (2) are provided by the structure of the regulation and replicated in the stylized analysis proposed in this paper (see Supplementary Information 4). Evidence supporting condition (3) is provided in our empirical analysis and further corroborated in the analyses described in the Discussion and Supplementary Information 2. In particular, from the three conditions above, it emerges that the results of the simulation are grounded on PCR differentials. For this reason, we paid particular attention to demonstrating a negative correlation between high-carbon sectors and risk measures.

More formally, we defined the PCR as the LLR (or accumulated provisions in EBA terminology) divided by the gross exposure for the high-carbon and low-carbon sectors i for each bank j. The PCR represents the expected credit loss (of non-default counterparties) and the corresponding loan loss provisions which banks must allocate to lending activities in each sector. This measure is assumed to be the model-based output from each institution risk model, in line with the accounting regulation:

$$\rmPCR_i,\,j=\frac\rmLoan\;\rmloss\;\rmreserves_i,\,j\rmGross\;exposure_i,\,j.$$

(1)

We then calculated the change in the level of LLR following a divestment from high-carbon assets. This was performed by assuming that all low-carbon loans replacing the high-carbon ones would require the average PCR of existing low-carbon assets. In other words, a divestment from low-PCR assets and re-investment in high-PCR assets would lead to an increase in the overall average PCR. More formally, the increase/decrease in provision for bank j is defined as follows:

$$\beginarrayl\rmLoan\,loss\,provision\,charges_j=\Delta \rmLoan\,loss\,reserves_j\\=\left(\rmPCR_\rmlow-carbon,\,j-\rmPCR_\rmhigh-carbon,\,j\right)\times \rmGross\,exposure_\rmhigh-carbon,\,j\endarray.$$

(2)

This result provides the expected increase or decrease in provisions if a bank had to shift the totality of its assets from high-carbon to low-carbon investments. This relationship is an accounting identity defined by the framework. The impact of additional loan loss provisions on a particular bank’s income statement is considered an LLP ‘charge’ (that is, additional cost) with direct effect on their net profit. In particular, the increase in provisions (that is, the LLP charges) is directly deducted from net profit, being an additional cost for the bank in the fiscal year of the divestment. This in turn provides a direct estimate of the change in net profits following a divestment from high-carbon assets. More formally:

$$\rmNet\,profit_j,t+1=\rmNet\,profit_j,t-\rmLoan\,loss\,provision\,\rmcharges_j,$$

(3)

where j refers to each bank in our sample, t is the starting point period and t + 1 is the period post divestments. Importantly, to simulate the effect of the divestment, we assumed it to occur entirely in one fiscal year. This divestment would probably be spread across multiple years, but frontloading the entire impact allows us to better investigate the implicit incentive structure created by the regulation. This simple approach allowed us to simulate what would be the impact of a divestment from high-carbon assets on banks’ balance sheets and income statements, testing the hypothesis that a potential divestment strategy might be costly, disincentivizing banks from taking such action.

link