Understanding Loss Given Default Review of Three Approaches
Loss Given Default (LGD), often the term used to refer to the “severity of loss” of an investment, estimates the portion of an exposure (bond or loan equivalent) that is unlikely to be recovered in the event of default. When it comes to estimating the LGD of financial transactions, various techniques can be applied. In this blog, we discuss three methods: (1) a qualitative scorecard approach, (2) a quantitative statistical approach, and (3) an innovative market-based approach.
It is widely accepted that data on loss (and recovery) rates are scarce and difficult to find. This is particularly the case for what is generally referred to as a low-default portfolio (LDP), which includes non-financial corporates (NFCs), financial institutions, sovereigns and project finance. In this blog, we focus on NFCs where data scarcity can be attributed to many factors. NFC is not very often by default. When a default occurs, the event and the resulting loss and recovery must be recorded, which usually takes years.
A Look at Default and Recovery Data
At S&P Global Market Intelligence, we have built a database of defaults and loss rates dating back to 1981, which is called CreditPro® We used this data for recovery statistics on 7,836 transactions. Figure 1 below shows the data grouped into six different loss severity bands. The first bucket (0%-10%) is for stocks that have suffered minimal losses, while the last bucket (90%-100%) is for stocks that have suffered maximum losses. Figure 1 shows that the loss distribution exhibits a clear bimodal characteristic, which means that the most likely loss from a defaulted transaction will be either very low or very high. It is therefore not surprising that the regulatory LGD is assumed to be 45% by various financial institutions around the world, since it is based on the average of such a bimodal distribution. Additionally, losses could fall into any category, meaning that an average of 45% could lead to under-provisioning or over-provisioning against losses, making it difficult to optimize the allocation of the capital.
Figure 1: Observed Losses from Defaulted Transactions
Source: S&P Global Market Intelligence CreditPro, as of March 31, 2022.
Evaluation of three methods to discover LGD
Dashboards and statistical approaches for LGD calculations are offered by S&P Global Market Intelligence. To compare them, we identified data on more than 6,500 broadcasts by more than 2,000 issuers. Issuers were based in emerging markets in the Middle East, Africa, India, South America, APAC and Europe. They also came from various industries, although the financial industry was excluded to maintain consistency of results and comparability of inputs, given the unique characteristics of the industry. Each trade was valued with these two LGD approaches, as well as the third method to find the market’s implied LGD.
Method 1: Dashboard approach
S&P Global Market Intelligence has built a LGD Dashboard which produces loss estimates based on a fundamentally driven methodology. This dashboard is designed to estimate LGD, which is a point estimate (PiT) reflecting current economic conditions and capturing the fact that a default during an economic downturn is usually accompanied by lower recoveries. The LGD Dashboard is expert-driven, as it requires an analyst’s opinion on certain key inputs, while outputs are produced using a transparent, rules-based approach. The LGD engine is made up of six main factors, each of which plays a different role in estimating the loss. This includes:
- Quantity and risk of cash flows/assets/economic value before and after default
- Age of the exposure (e.g. senior bond or not)
- Economic expectations
- Securities and guarantees/insurance
- Recovery costs and restructuring policy (restructuring versus disposal)
Figure 2: LGD dashboard frame
Source: S&P Global Market Intelligence. For illustrative purposes only.
The importance of the factors listed above has been validated by existing data and field experience of observed post-fault resolutions. Pre-default Expected Value at Default (EV) is accentuated based on economic forecasts, obligor creditworthiness, current EV and expected EV volatility. This stressed EV is adjusted for jurisdictional influences (eg, operations in high-risk countries) and collection costs. Secured creditors are paid first with the pro rata stressed EV tied to their security, while the remaining stressed EV is paid according to the creditor cascade (i.e. senior creditors are paid before junior creditors). At no time are averages used in the estimation process, leading to a robust loss estimate for each individual exposure.
Using the LCG scorecard, we projected the LGD for the over 6,500 transactions in the sample and obtained an average LGD of 46.02% from the 2020 year-end financial statements. The average LGD was almost the same as the 45% rule used by various financial institutions around the world, as mentioned above. Figure 3 below shows the distribution of LGD results across the six LGD buckets, which shows a normalized shape with buckets three and four having the highest frequency.
Figure 3: Expected PCD using the scorecard
Source: S&P Global Market Intelligence Scorecards, as of March 31, 2022
Method 2: Statistical approach
S&P Global Market Intelligence has also built an LGD statistical model, called LossStats™ Model (LSM). LSM is designed to predict losses using a statistical technique and assumes a continuous conditional probability distribution which belongs to the exponential model family. LSM estimates the LGD distribution of bonds and loans issued by corporations, taking into account industry and instrument specific characteristics and leveraging an extensive collection database.
Using LSM, we projected LGD for over 6,500 transactions and achieved an average LGD of 44.48% compared to the 2020 year-end financial statements. The average LGD per LSM is, again, almost the same as the 45% rule. Figure 4 below shows that the vast majority of transactions are in the third bucket, unlike the dashboard method which showed a normalized type distribution. This is no surprise as statistical models tend to offer less granularity and precision than expert-driven approaches, but compensate for this more efficiently as they require less human intervention. The LSM statistical model also required fewer inputs. For example, the LSM model asks entities to be flagged as secure or insecure without indicating the amount of such seniority/subordination. The LGD dashboard, on the other hand, examines precisely the number of liabilities that are legally above and below the exposure considered.
Figure 4: Expected LGD using LSM statistical model
Source: S&P Global Market Intelligence Credit Analytics, as of March 31, 2022
Method 3: Market-induced LGD approach
Finally, we applied a third innovative method to estimate the losses of a financial transaction. We found the yield at worst (YTW) of over 700 bonds issued by entities in our representative universe. YTW is a measure of the return investors expect to receive based on an issue’s key characteristics, the most important being credit risk. By removing the YTW risk-free rate from each trade, we found the spread which is assumed to be an approximation of the credit risk premium (i.e. the premium investors demand to offset the risk loss they expect to incur in holding the investment). This credit risk premium is equivalent to an expected loss (EL), which is a combination of the probability of default (PD) and the LGD. We have calculated the PD of each issuer using our DP Model Fundamentals (PDFN), a statistical model that combines business and financial risks to estimate an entity’s probability of failure. By extracting the EL from the YTW of each trade and calculating the issuer PD, we were able to find the market implied LGD of each trade.
The average LGD using the YTW method was 44%, which again agrees well with the 45% rule and the averages found through the dashboard and statistical methods. Figure 5 below shows an even distribution across the different loss buckets, demonstrating that the market does indeed use different techniques to calculate LGD and does not necessarily rely on the 45% rule.
Source: S&P Global Market Intelligence CapitalIQ & Credit Analytics, as of March 31, 2022
In Summary: Evaluate the Tradeoffs
We have proposed a description of alternative methods for estimating LGD for financial transactions. While the average LGD is almost the same for both scorecard and statistical approaches, the distribution of LGD LGD results tells a very different story. The LGD Dashboard is fundamentally driven and applies an intuitive credit and rule-based approach with no data training needed to design the model. It achieves a high level of differentiation by looking at loss slices. The statistical approach is quantity-oriented, which means that it can be applied with little human intervention, but allows less differentiation with respect to loss categories. The trade-off between accuracy and efficiency is often seen when comparing statistical and expert-driven models. The choice of which approach to use depends on the specific preferences of an institution. It is important to understand the differences, however.
Taking Loss Given Default Estimation to the Next Level: An Aspiration for All Creditors, Not Just Banks | S&P Global Market Intelligence (spglobal.com)