Blog — 16 Jul, 2021

Taking Loss Given Default Estimation to the Next Level: An Aspiration for All Creditors, Not Just Banks

Credit forms the foundation of national and international commerce. Credits range in sophistication from the ultra-complex (e.g. asset-backed securitization) through to the very simple (e.g. unregulated credit unions). Regardless of type, size, sophistication and geography, creditors can be sure of only one thing: There is no such thing as absolutely risk-free and therefore not all debtors will make repayments on time and in full.

Whilst much focus has been placed on the risk of payment failure (i.e. default risk), relatively fewer efforts have been made on predicting the proportion of debt that will never be recovered (i.e. loss given default or LGD). Reasons for this asymmetric focus of efforts include:

  1. Lack of historical loss data (particularly for low-default portfolios) and limited internal expertise on modelling LGD
  2. Difficulty in building quantitative models due to non-binary nature of losses, often leading to overly simplified assumptions
  3. LGD analysis can be resource-intensive (the number of exposures are often much greater than the number of counterparties – i.e. multiple exposures to the same counterparty)

The lack of attention to LGD is worrying as LGD is mathematically “equal” in importance to probability of default (PD) in estimating the probability-weighted loss (or expected loss). Take the example of an investment manager's dilemma of selecting between two sovereign bonds, which have the same credit rating and similar implied returns. All things being equal, the bond with lower LGD would lead to a greater risk-adjusted return. Perhaps what is equally worrying is the extensive use of the analytically limited approach of using average historical losses to predict future losses.

Besides cyclical and idiosyncratic default events, history has recently reminded us that unexpected events will occur leading to a greater number of defaults and higher related losses. Attention should therefore be placed on not only predicting when default rates will rise, but how much will not be recovered once the dust settles. These considerations should not only be the concern of bank risk functions but for all counterparties that are subject to credit risks including corporates, counterparties in supply chains, asset managers, insurance companies, pension funds, hedge funds and government departments.

Deeper analysis and differentiated risk indicators

Lack of quality, cross-asset class, historical loss data (e.g. project finance, real estate finance, financial institutions, funds, sovereigns, etc.) reduces the ability for deeper analysis and the identification of asset class or sector-specific risk factors. This problem is further exacerbated by a lack of internal expertise in modelling LGD. This is not only a problem for non-banks, with typically smaller credit risk functions, but also for large banks. Whilst the latter may well have some historical loss data, it can prove to be a double-edged sword. Historical data alone may lead to models with little predictive power, especially when historical data is biased by certain time periods and/or lack of sufficient representation from negative market periods.

Differentiated LGD risk factors cannot be identified by simply using historical loss data.  The use of data must be augmented with experienced-based economic intuitiveness. For example, there is little loss data available outside of the U.S. and Western European markets. However, it is universally accepted that country specifics – creditor-debtor dynamics, legal environment, insolvency regimes and rule-of-law – will impact ultimate losses.

Cutting-edge solutions to help elevate your analysis

The S&P Global Market Intelligence LGD Scorecards are a family of cutting-edge LGD models. The aim of framework is to elevate the LGD analysis well above the simple use of averages or misspecified quantitative models, reflecting the precise specifics of each exposure and related debtor. For example, we do not just look at “labels” of debt such as senior or junior, but rather look at precisely how many liabilities are legally senior and junior to the exposure under consideration.

The output can be either a point-in-time (PIT) or downturn LGD. This allows a multitude of use cases including banking and non-banking risk management, investment management, IFRS 9 expected credit loss (ECL) estimation and regulatory capital calculations. As these models are not specified or built on historical data alone, their application is truly global, equally applicable to all counterparties both in emerging and developed markets.

The underlying LGD framework (“LGD engine”) is common for all models, across all low-default asset classes, with differentiated risk factors for each sub-asset class.

The LGD engine is composed of six primary factors, each playing a differentiated role in the estimate of loss:

  • Pre- and post-default quantity and riskiness of cash flow/assets/economic value
  • Seniority of exposure (e.g. senior bond)
  • Jurisdiction
  • Economic expectations
  • Collateral and guarantees/insurance
  • Recovery costs and workout policy (restructuring versus disposal)

These factors are further refined to cater for differing characteristics across asset classes (e.g. project finance economic value (EV) is based on a net present value of future cash flows).

The combination of the above factors is both supported by data and on-the-ground experience from observed resolutions. The current (or pre-default) EV is stressed based on economic forecasts, credit strength of the debtor, current EV and expected volatility of EV. This, stressed EV, is adjusted for jurisdictional influences (operations in higher-risk countries lead to greater haircuts) and recovery costs. Secured creditors are paid first with the pro-rata stressed EV related to their collaterals, with the remaining stressed EV paid down the creditor waterfall (senior creditors get paid before junior creditors). At no point are averages used in the estimation process, leading to a robust estimate of loss for each individual exposure.

The LGD Scorecards have been tested on over 2000 defaulted exposures from North America, Latin America, Asia, Africa, and Europe[1]. Perhaps the greatest benefit of the LGD Scorecards is not only the ability to predict losses but to offer a framework that serves as a foundation for thinking, providing valuable insight into what will impact potential future losses.

Sustained analysis and surveillance

As with all risk measures, the need for near-continuous analysis and surveillance is paramount. This is particularly the case in emerging markets, where changes can and do occur abruptly. Furthermore, certain requirements, such as those under IFRS 9, are also ongoing throughout the year. Counterparty credit risk in supply chains is another example where analysis must be full and timely. LGD models should therefore be easy and efficient to use on large and small portfolios and sensitive to changes in underlying risk factors. The latter is not possible when using “fixed” LGD estimates (based on average historical losses).

Stress testing and scenario analysis are also important considerations to contend with. Scenario analysis is quickly becoming key for most institutions, particularly concerning environmental, social and governance (ESG) issues. LGD models must be able to forecast losses under different ESG related scenarios – a key requirement from shareholders and governments alike.

The S&P Global Market Intelligence LGD Scorecards are designed to perform frequent analysis on large and small portfolios of exposures. For listed or publicly rated counterparties, the LGD Scorecards are fully automated and can source all required risk factor inputs from S&P Global Market Intelligence databases. This link allows for regular updates and quick estimation of losses for large portfolios, allowing for greater focus on the results and the subsequent steps required to manage identified risks.

The framework is also designed to assess multiple forecasts (e.g. different financial scenarios based on varying climate change scenarios) and blend them to estimate a probability-weighted LGD. This ability is also used for reporting under IFRS9.

Understandably, not all creditors will look to devote scarce resources to in-depth LGD estimation. However, this should not stop them all from aspiring to raise their LGD methodologies to the next level. As discussed above, with the S&P Global Market Intelligence LGD Scorecards, it is entirely possible to have it all: Automation, efficiency and the ability to undertake sustained in-depth analysis.

Learn more about our LGD Scorecards here >



[1] Testing results available on request.

Learn more about Loss Given Default Scorecards