ECB’s Shifting Perspective on Downturn LGD: Addressing the Timing Lag

What was the European Central Bank thinking when they updated the ECB Guide to Internal Models in 2025 (EGIM2025)? At least for the downturn LGD part, we can build a plausible reconstruction.

The ECB probably wished to preserve the clarity and regulatory consistency of defining downturn periods by economic criteria, yet simultaneously needed to address the omission of timing lags, which could undermine the conservatism of downturn LGD estimates.

Marco Folpmers

This (reconstructed) thought process may have motivated the decision in the 2025 update to highlight explicitly a complementary mechanism – the “LGD reference value.” Interestingly, this reference loss given default concept, described by the European Banking Authority in 2019 (EBA2019), is repurposed and elevated by the ECB almost as a practical “backstop.”

This backstop ensures that downturn LGD estimates cannot fall below levels suggested by actual realized losses, even if those losses peak in years distinct from the identified economic downturn, unless the bank has a very good substantiation.

Let’s now unpack this regulatory evolution in more detail, providing background on downturn LGD, analyzing the regulatory texts, and discussing the implications for risk practitioners.

LGD and the Challenge of Downturn Calibration

LGD measures the expected loss severity given a borrower’s default and plays a crucial role in credit risk modeling and capital calculations under the Internal Ratings-Based (IRB) approach. Because recovery rates tend to worsen during economic stress, regulators require banks to estimate an appropriate downturn LGD that reflects this increased loss severity.

Traditionally, downturn LGD is calibrated by first identifying economic downturn periods based on macroeconomic indicators, and then analyzing historical loss data within those periods. This approach, while methodologically sound, assumes that economic stress and peak credit losses of the specific bank coincide. But is that the case?

Let’s  take a step back and introduce LGD pools, the LGD quantification and the reference value.

Downturn LGD and the Reference Value

For the EU, the story starts with CRR art. 181 (1)(b), where it is stated that banks should use LGD estimates “that are appropriate for an economic downturn.” This has been further clarified in lower level, EBA guidance.

The most important guidance is found in EBA2019. In this paper, Figure 1 illustrates that first a risk differentiation model is needed to allocate loans to LGD pools, and then each pool is assigned two LGD values, the latter of which is the downturn LGD.

Table 1: LGD risk differentiation and quantification (calibration) (source: EBA2019)

The overall downturn LGD and, more specifically, the downturn LGD per pool, has to be quantified based on historical data. More precisely, this quantification needs to take place by first selecting a historical time window that qualifies as “a downturn.” (Several downturn windows may also be selected.)

In a second step, for this specific “downturn” window, the downturn LGDs are determined with the help of the realized losses during that window. The bank needs to establish that during the downturn, losses were indeed at an elevated level.

Once EBA2019 has explained how the downturn LGDs have to be quantified per LGD pool, the EBA introduces a “reference value” as a “non-binding challenger.” The “reference value” is calculated as the mean of two aggregated realized loss amounts for the two years in which the losses were the highest during the historical window. So this reference value is calculated without any consideration to the identified downturn years and serves as a “non-binding” check to see if the previously calculated downturn LGD is sufficiently conservative.

The EBA Example: Retail Mortgages

In EBA2019 we see how this works. First, an economic downturn period is selected for a specific portfolio; in the EBA example this is a retail mortgage portfolio. Based on two macro indicators (house price index and GDP growth), two downturn periods are identified, 1990-1991 (trough in house price index) and 2008-2010 (trough in GDP growth). Subsequently, banks are supposed to calculate the realized LGD for each of these periods and then settle upon the worst (highest) outcome, per LGD pool. This will deliver the “DT LGD” values per pool as illustrated in Table 1 above.

The issue is that there is an assumption that the most relevant losses arise during the identified time windows. Is that always the case? It makes more sense to assume that there is typically a time lag involved. And although the EBA does ask for an impact assessment of the downturn identification, it does not explicitly ask to take lagged dependence into account, and, where relevant, to shift the LGD quantification to a more relevant, lagged observation period, immediately after the downturn window identified by the macro indicators.

This focus on the identified window per se is strange. A retail customer does not immediately default on his mortgage loan once he gets unemployed. In the same way that an SME company does not immediately stop servicing its loan once the business cycle (as measured by GDP growth) declines.

This could have been repaired by introducing the reference value as a binding measure, but the EBA stops short of that in EBA2019 and makes it very clear that the reference value is neither binding nor fit for a regulatory compliant LGD quantification. The reference value is also not connected by the EBA to the identification with the help of macro indicators, with or without a lag.

Reconstructing the Timing Gap and Regulatory Response

Fast forward to Frankfurt, 2025. The ECB is about to launch their update of the ECB Guide to Internal Models (EGIM2025). On reviewing the previously mentioned EBA2019 “final” guidelines and their own previous version of the ECB Guide (from 2024), they hit upon the downturn LGD quantification. The ECB may have picked up the lag issue as explained above. They also noticed that the previous version of the ECB guide (2024) does not explicitly mention the reference value for LGD. How to account for the relevance of the lag without introducing additional instruments?

Faced with this challenge, the ECB appears to have opted for a pragmatic solution. They maintain the macroeconomic downturn definition as the cornerstone for downturn LGD but introduce a supplementary “LGD reference value” (EGIM2025, art. 304-308). This LGD reference value is based upon EBA2019, but, in their text, the ECB does not present the statistic as “non-binding” (as EBA2019 does). On the contrary, it asks institutions to “thoroughly analyze cases where the reference value is materially larger than the final LGD estimates” and to “appropriately consider the existence of any lag between a downturn period and its potential impact on the relevant loss data.”

In summary: The existing EBA2019 guidelines and the previous EGIM version (2024) did not sufficiently acknowledge that macro-identified downturns have a lagged impact on realized losses. At the same time, a “reference value” was introduced in EBA2019, as an extra check, that was not related to the macro downturn.

In my reconstruction, in 2025 the ECB discovered that this lagged impact was not properly acknowledged so far. In the 2025 update EGIM2025, they therefore explicitly included the reference value in the EGIM’s text and “upgraded” the status of this reference value to pick up elevated loss levels due to time lags. In this way, they could repair the omission without the introduction of new instruments.

Parting Thoughts

The 2025 update to the ECB Guide on Internal Models represents a thoughtful evolution, balancing regulatory clarity and practical risk sensitivity. By maintaining economic downturn definitions while incorporating a realized-loss-based reference value now (almost) as a supervisory floor, the ECB enhances the robustness and credibility of downturn LGD modeling.

At the same time, this cannot be the end state. FRM practitioners deserve a proper LGD measurement framework with a meaningful treatment of macroeconomic identification and lagged peak-levels of realized losses.

Dr. Marco Folpmers (FRM) is a partner for Financial Risk Management at Deloitte the Netherlands.