Authored by Ivan Cilik, Matt Nitka and Sean Statz
Even though the effective date isn’t until 2023 for the current expected credit loss (CECL) requirements, financial institutions are already working to implement necessary processes to adhere to the new accounting change. The standard is such a significant and extensive change that many institutions are having to start from scratch in order to realize their data requirements, multiple methodologies, forecast considerations and validation needs.
One of the most critical parts to complying with CECL is determining a methodology for measuring the allowance for expected credit losses. Institutions have several methodologies to choose from, and some are better for certain products than others, which means institutions may need to use a variety of methods because they have a number of different products. Methodologies range from “SCALE” and “WARM” methods, which are more popular with smaller institutions, to more complex ones, like the discounted cash flow (DCF) method, which tends to be favored by larger institutions. That being said, the amount of data and the resources an institution has will most likely be the determining factors in steering management toward one direction or another.
Fortunately, institutions should be able to leverage their existing processes or models when selecting their CECL model and building efficiencies into their implementation process. For instance, many asset liability management (ALM) models are already running a DCF approach on loans, bringing in the contractual information and running out the cash flows over the life of the loan. Through the ALM process, institutions are already extracting and vetting data that could then be used for their CECL methodology as well.
It is worth noting that the DCF method is considered more complicated than the others, but that doesn’t matter if it is the best method for your institution’s needs.
The standard for the DCF method essentially says that if an institution is using a cash flow approach, the discount should be at the financial asset’s effective interest rate. “When a discounted cash flow method is applied, the allowance for credit losses shall reflect the difference between the amortized cost basis and the present value of the expected cash flows,” it further states.
But what is DCF? It is often considered the most complicated method because it projects cash flows over the life of the loan, particularly longer-term loans, e.g., 30-year mortgages. Cash flows are behaviorally and credit adjusted before discounting at the loan’s effective interest rate. Its projections are taken from several sources. In fact, the DCF method draws heavily from institutional information, requiring access to current data at the instrument level — on the institution’s loans and investments — as well as historical data, to get an accurate read on trends and behaviors. This works best for longer-standing products that have many years of supporting documentation. Newer products that have only accumulated a few years of data may be better off using a simpler method. However, after data has been collected for several years on that product, the institution can merge the two methods.
Before diving deeper into DCF, it is helpful to understand the difference between contractual cash flows and discounted cash flows. Contractual cash flows are known factors: payment type (principal and interest, interest-only, etc.), payment amount, interest rate, maturity date, payment frequency and amortization. It is data that is typically tracked and found within the the institution’s core system. The data can be extracted for various processes and models, and used to project out cash flows of a loan.
From there, several assumptions are typically made to convert those contractual cash flows to expected cash flows, including:
The first one is the most common: the prepayment rate, which is the annual percentage of a loan’s outstanding balance expected to be paid off early. It’s also referred to as the conditional prepayment rate, or CPR. There are several different types of prepayment, such as partial, full or refinance. The higher a prepayment rate, the higher the likelihood for more prepayments, which means a shorter life for the loan and, thus, less interest for the lender in total. That is considered prepayment risk. While that doesn’t have a direct effect on credit, it impacts the timing of cash flows and, ultimately, the institution’s CECL estimate.
The next assumption is the probability of default, or conditional default rate (CDR). It is the odds that a borrower is going to default and is usually expressed as an annual percentage similar to the prepayment rate. The rate is then adjusted on a period-by-period basis. It is based on the credit quality of the borrower. The DCF approach can bring in a vectored default rate forecast, so an institution can change its approach with a borrower if they look like they may be fine for a few years, but may be challenged after that.
Another assumption is the loss given default, or how much the institution may lose if the borrower goes into default. It’s also known as the loss severity rate or the inverse of a recovery rate. It, too, is expressed as a percentage, and you actually need to combine this with the probability of default to calculate an expected loss. Collateral, principal paydowns, foreclosure or repossession costs, etc., influence the loss given default percentage. Unfortunately, it is oftentimes difficult to calculate because it requires hard-to-track historical information. To prevent that problem going forward, institutions should start documenting credit quality characteristics when a loan does go into a loss, so they can start to understand which loans are migrating and which ones are actually ending up in a loss status.
Recovery delay is the next assumption. When a loan goes into default, institutions usually experience a delay in recoveries. Timing is critical in DCF models because the period of delay will have varying effects on the CECL estimate. Often, this is made as a policy election and applied across the board based on loan classes. If that is material to an institution, the discounted cash flow works well as it is able to customize recoveries and detail how they would look to the institution.
Finally, the discount rate (effective interest rate) is the contractual interest rate adjusted for any net deferred fees or costs and/or premiums or discounts existing at origination.
Because CECL requires forward-looking projections within the methodology, the DCF method is most appropriately aligned with forecasting due to the timing capabilities. If you think of CECL like a big stress test and you run scenarios of different forecasts for, say, an economic downturn or the fluctuating unemployment rate, your institution can incorporate it in a discounted cash flow, which is helpful from a credit risk standpoint.
Projections would also take into account the standard historical experience, current conditions and reasonable and supportable forecasts. When we’re talking about longer-term loans, institutions should know they are not expected to predict 30 years into the future.
One of the most discussed benefits of the DCF method is that it automatically calculates contractual life by factoring in all of the cash flows and adjusting them for behavioral and credit-related aspects. Another plus is its ability to incorporate reasonable and supportable forecasts based on timing and volatility, allowing institutions to customize how they want to view these forecasts and giving them the power to stress test as needed. Additionally, the DCF method includes all different structures of loans, such as balloon payments, interest-only, lines of credit, etc., which is an advantage for an institution. One final pro is that it is defensible. Even though it is projecting actual cash flow using some subjective factors, the method relies less on the subjective and more on facts and figures.
As far as the cons to the DCF method, the biggest complaint is the method is highly complex. It requires a lot of data — historical, current, instrument-level. It’s just more data-intensive than most other methods. Another unpopular characteristic is that it typically requires software and the requisite resources to execute it. Finally, it is a lengthier process from start to finish. Data has to be scrubbed and inputted and then models have to be run and results analyzed. (That could be mitigated some by building off processes that are already being used for ALM or other models.)
Since institutions have varying levels of data as well as different needs for different products, they will need professional guidance to determine the best model implementation for their CECL requirements.