Picking a CECL Methodology: Five Reasons Why Only One Method Makes Sense

Picking a CECL Methodology: Five Reasons Why Only One Method Makes Sense

By Adam Mustafa, Invictus Group CEO

FASB’s guidance for CECL is flexible when it comes to methodologies. In fact, many software-as-a-service (SaaS) providers and consultants make it a point to brag how their products can handle just about all of them. But let’s not confuse quantity with quality when it comes to CECL methods. The reality is that most CECL methodologies, from the too simple WARM method to overly complex regression modelling, are not up to the task. Worst of all, they produce wacky results that defy logic and will require management to rely on shaky qualitative factors to determine most or all the reserve.

However, there is one method that is head and shoulders above the rest: The PD/LGD Method. PD stands for Probability of Default. LGD stands for Loss Given Default. Yes, these terms sound wonky, but don’t let the quant jocks fool you. This methodology is actually simple to understand and use with the right approach. Let’s first explain this methodology in laymen’s terms, and then I will get into why it’s so superior.

What is PD?

The Probability of Default is exactly how it sounds. It’s the probability that a borrower defaults on a loan. When you make a loan, you hope that the probability is zero percent, but we all know it’s not. For purposes of this article, let’s not worry about how we estimate the PD for a given loan. But if we can reasonably estimate the probability that a given loan defaults, that is extremely valuable information.

What is LGD?

Loss Given Default is the loss of principal you expect to endure on a loan if it does default. That is usually equal to the difference between the outstanding principal balance and the expected recovery after liquidating the collateral following a default event.

Oversimplified Example

Let’s say you have one non-amortizing loan in your portfolio with a $100 balance. If this loan defaults at any time, the Exposure at Default (EAD) is $100. Assume that this loan is backed by $120 of collateral (a building). This collateral could potentially lose value in the future if real estate prices decline. So, if real estate prices drop by one-third unexpectedly, the building may only be worth $80.

Under this scenario, you would only recover $80 (ignoring expenses for now) of your $100 exposure, so your loss in this case is $20. This means that your LGD is 20 percent, which is equal to your $20 loss divided by the $100 exposure.

But, based on your underwriting, you estimate this loan has a 5 percent PD rate, meaning there is a 5 percent chance the borrower will default on that loan in the future. So, while you stand to potentially lose $20 if this loan defaults in a world where real estate values crater, you also only believe that there is only a 5 percent chance of this happening.

Using the PD/LGD Method for estimating the expected loss for this loan under CECL, your reserve in this case would be $1. This is equal to $100 total exposure x 20% LGD x 5% PD.

The Beauty of this Methodology

Let me give you five quick reasons why this method is so superior:

  1. It’s intuitive. It bifurcates the risk of a given loan into two simple concepts:

•  What are the chances this loan goes bad?

•  If it does go bad, how much money will we lose?

When a banker makes a loan, she primarily focuses on the borrower’s ability to service the loan. She will analyze a borrower’s cash flow and liquidity, which are the borrower’s resources in this regard. She will use ratios such as the debt-service coverage ratio (DSCR), debt-to-income ratio, and credit score to measure this capacity. What she may not realize is that she is really trying to directionally measure the probability of default — she just doesn’t think of it in those terms. The PD is just the inverse of the same concept.

The other important thing a banker does when she makes a loan is have a backup plan if the borrower can’t repay the loan because resources become insufficient. That backup plan is the collateral. She will analyze the collateral from many different angles and will calculate key ratios such as the Loan-to-Value (LTV) ratio to guide her. The goal is to make sure that she can rely on the collateral as much as possible to cover the loan if the borrower’s cash flow or liquidity resources cannot. She may not realize it, but what she is really doing is directionally estimating the Loss Given Default!

The point is that the PD/LGD Method is the only method that mirrors how bankers think about credit in the real world. This immediately makes it possible to provide value to a bank much greater than just checking a box for financial reporting. It also makes it easier for bankers to ‘square’ the results with their intuition about their loan portfolio.

  1. It uses loan-level information. It’s the only method that also makes use of loan-level information in a meaningful way. If we take these two concepts a step further, we can start to form relationships between loan-level data elements and either PD or LGD. For example, we can link a loan’s risk rating, DSCR, and the remaining time to maturity to PD. We can also link the collateral type, loan-to-value, and appraisal date to LGD.

The problem with the other methods, especially WARM and regression modeling, is that they are way too focused on what happened with loans from the past, while failing to utilize loan-level information in a significant way. Any method that doesn’t incorporate the risk characteristics of the loans that are currently in your loan portfolio is utterly useless, even if some bean-counter tells you it will check the box. You may win the short-term compliance battle. But in the long run, you will be unable to support and defend your CECL reserve with data and conviction, which will lead to excessive over-reserving and lack of credibility with auditors and regulators.

  1. It reduces the dependency on qualitative factors. Dependency on qualitative factors (about as reliable as throwing darts at numbers on a board) to drive 90 percent or more of the reserve was a problem before the pandemic and has only been exacerbated by it. The PD/LGD model makes it far easier to quantify the impact that changes in current or forecasted conditions will have on the reserve because it opens the door to utilize stress testing to drive your CECL result.
  2. It can be used for other strategic purposes. Banking 101 is all about risk and reward tradeoffs. The PD/LGD method allows us to translate the underwriting characteristics of a loan into a measurement of risk. If we can measure risk, we can then determine the yield and structure we need on that instrument to ensure we create an attractive risk-reward trade-off. This blows the door wide open to unlock massive insights with respect to loan pricing and strategic planning, allowing bankers to know what they should be doing more or less of relative to market conditions.
  3. It’s easy to deploy. Don’t let the quants scare you. This methodology can be deployed without massive amounts of loan-level data that go many years back. Most banks do not have vast amounts of historical loan-level data but have supplemented holes in their data with external source of loan-level information. Over time, as banks collect more internal loan-level information, they can rely less on external data.

Wrap Up

The PD/LGD method is the best method for CECL, hands down. The other methods may come with the allure of being cheaper and easier to implement. But they will cost you far more money in the long run because you run much greater risk of being excessively over-reserved and at an analytical disadvantage relative to your peers. These methods lack flexibility and are also just fundamentally second-rate. PD/LGD is intuitive, driven by loan-level information, minimizes the dependency on q-factors, and is the golden goose that lays many eggs that can be hatched across the organization to inform loan pricing, underwriting, and strategic planning.