making sense of financed emissions – Bank Underground

Lewis Holden

Over 95% of banks’ emissions are ‘financed emissions’. These are indirect emissions from households and businesses who banks lend to or invest in (banks’ asset exposures). Banks disclose these in line with regulations designed to help markets understand their exposure to climate-related risks and their impact on the climate. But emissions disclosures vary drastically between different banks with similar business models. Data quality and availability is cited as the key reason for this. In this post, I demonstrate that variations in financed emissions estimates are explained by the extent of banking activities and asset exposures rather than data quality and availability. For example, whether estimates capture a subset of loan exposures or wider banking activities such as bond underwriting.

Comparing financed emissions between banks can be challenging because financed emissions scale with asset exposures. In Table A, I summarise financed emissions from a subsample of globally systemically important banks (G-SIBs) disclosures. For comparability, G-SIBs in Table A are of similar size.

Table A: G-SIB financed emissions

Sources: G-SIBs’ climate-related disclosures and annual reports for financial years ending 2024.

How can these G-SIBs, which all operate globally with similar business models and asset exposures, report financed emissions an order of magnitude different from one another? Data quality is usually cited as the key impediment to accuracy and comparability. For instance, emissions disclosures mention ‘data quality’ or ‘data gap’ an average of 10 times. But is data really the core challenge?

The data argument goes like this. Households and businesses which banks lend to and invest in must disclose emissions before banks can aggregate these to calculate financed emissions. But the majority of banks’ asset exposures are households, consumers and unlisted corporates that do not disclose their emissions. Because disclosure requirements only apply to large, listed corporates. Large, listed corporates predominantly access finance via capital markets rather than loans. Therefore, banks need to estimate the emissions of the households and businesses who make up their asset exposures in order to calculate financed emissions.

Is data quality and availability the source of variation?

I compare three different financed emissions estimates for a sample of UK banks:

1. Reported in banks’ climate disclosures.
2. My estimation model, with proxy emissions data supplied by data provider A.
3. My estimation model, with proxy emissions data supplied by data provider B.

The data providers I use are MSCI and LSEG. The estimate relating to each provider has been anonymised. Broadly, my estimates capture banks’ corporate and mortgage loan exposures, as recommended by the Partnership for Carbon Accounting Financials (PCAF). PCAF is the industry standard guidance for measuring financed emissions. Other exposures, such as consumer finance, and other banking activities, such as bond underwriting, are excluded.

In the absence of granular loan level data, my estimation model assumes banks’ borrowers can be proxied by an average. For example, loans to the UK transport sector are proxied by the mean carbon intensity for UK transport firms which disclose emissions data. This model has been developed by Bank staff and was applied in The Bank of England’s climate-related financial disclosure 2025.

Chart 1: Financed emissions disclosed by UK banks and estimated from my model

Sources: Banks’ climate-related disclosures and annual reports, MSCI and LSEG.

Despite the range of emissions data sources, proxies and aggregation methods, estimates fall within a range of around 10%. This implies the choice of emissions proxy data, and how estimation models aggregate this data, has a limited impact on aggregated financed emissions estimates.

Differences in financed emissions at the individual counterparty level may be more divergent. For example, the European Central Bank demonstrated that banks estimate a wide range of emissions for the same counterparty. My analysis does not dispute this. It simply demonstrates that when aggregated, financed emission estimates naturally converge towards the mean.

If data quality and availability don’t drive variations, what does?

The key driver of variance in financed emissions estimates is simply extent of business activities and asset exposures which banks estimate emissions for. I describe this as the ‘boundary’ of the estimate.

In Chart 1, I deliberately selected a subset of banks’ emissions reported on the basis of the same boundary as my model. This controlled for the boundary effect and isolated the effect of data quality and availability.

However, banks do not consistently disclose financed emissions on the basis of the same boundary. I identify three broad categories of boundary against which emissions can be estimated:

1. Minimal boundary – an estimate for a subset of loan exposures. Often those deemed high climate risk, such as to oil and gas companies.
2. PCAF boundary – an estimate covering most loan exposures. Excludes some loans with unknown use of proceeds, such as consumer finance.
3. All activities boundary – an estimate for all activities banks undertake and all asset exposures. In addition to loans, this may include ‘facilitated emissions’ – eg from bond underwriting, as well as assets managed on behalf of clients and not owned by the bank.

In Chart 2, instead of comparing estimates on the basis of the same ‘PCAF’ boundary, I deliberately compare financed emissions estimates across boundaries for the same sample of UK banks as in Chart 1. As I have already determined that data quality and availability has limited impact in Chart 1, this comparison isolates the extent to which the boundary impacts estimates.

Chart 2: Impact of boundary on UK banks’ financed emissions estimates

Sources: Banks’ climate-related disclosures and annual reports, MSCI and LSEG.

Expanding the boundary from ‘Minimal’ to ‘PCAF’ (A) increases the financed emissions estimate by almost 50%. This is because the ‘PCA’ boundary captures the majority of loan book emissions, while ‘Minimal’ boundary only captures emissions associated with a subset of high climate risk loans. This increase is material because while ‘high climate risk’ loans are banks’ most carbon intensive, they represent a relatively small proportion of total loans. This is particularly the case for UK banks whose largest exposures are residential mortgages.

Expanding the boundary from ‘PCAF’ to ‘All activities’ (B) increases the financed emissions estimate by almost another 50%. This is because the ‘All activities’ boundary captures emissions associated with the broadest range of banking activities, including assets under management. This effect is driven by the largest banks who undertake asset management and capital markets activities. The effect is more limited for banks which do not undertake these activities.

Interpreting emissions metrics across boundaries

Despite the variation in estimates of financed emissions across boundaries, there is no boundary which is superior. Instead, which boundary to rely on should depend on the use case.

In Table B, I propose a simple framework for how emissions metrics with different boundaries can proxy for two use cases – measuring climate-related financial risks and climate impact. ‘Financial risks’ means, for example, higher expected credit losses on loans. ‘Climate impact’ means banks’ contribution to climate change, such as the financing of carbon intensive activities.

Table B: Insights framework for financed emissions estimates

‘Minimal’ boundary estimates provide limited insights into banks’ financial risk exposure and impact. This is because they only capture a subset of banks’ activities.

‘PCAF’ boundary estimates are the most complete proxy for assessing banks’ exposure to climate financial risks. Loan exposures are the primary transmission channel through which financial risks will arise. This has been demonstrated in supervisory stress tests such as the 2021 Climate Biennial Exploratory Scenario. While other banking activities such as underwriting and asset management could expose banks to reputational and legal risks, the transmission of these risks into financial impacts is indirect.

‘All activities’ boundary estimates are the most complete proxy for climate impact. Banks’ impacts on climate change are not limited to direct loans and investments. The ‘PCAF’ boundary does not capture indirect impacts. For example, in managing investments in fossil fuel intensive companies, banks facilitate activity which will contribute to carbon emissions and subsequently climate impacts.

Conclusion

Differences in financed emissions estimates are caused by differences in the estimate boundary, not data quality. Transparency regarding estimate boundaries is therefore essential for interpretation of financed emissions metrics. No estimate boundary is best, with each offering insights into different use cases. The ‘PCAF’ boundary best proxies for banks’ exposure to financial risk, while the ‘All activities’ boundary best proxies for banks’ climate impact. The PCAF boundary should therefore be used by central banks in understanding climate financial risks, as well as in their own financial operations. Nonetheless, all emissions-based metrics are ultimately proxies. For financial risk purposes, they should be supplemented with more sophisticated tools such as scenario analysis.

Lewis Holden works in the Bank’s Financial Risk Management Division.

If you want to get in touch, please email us at bankunderground@bankofengland.co.uk or leave a comment below.

Comments will only appear once approved by a moderator, and are only published where a full name is supplied. Bank Underground is a blog for Bank of England staff to share views that challenge – or support – prevailing policy orthodoxies. The views expressed here are those of the authors, and are not necessarily those of the Bank of England, or its policy committees.