Chinese Zodiac – One of my favorite memories growing up as a child was going to a Chinese restaurant and reading about my Chinese Zodiac animal sign. My sign is the Rabbit, the fourth of all Zodiac animals. Simply based on what year I was born, I am able to tell whom I will be able to get along with and whom I should avoid. The Zodiac is only one of China’s fortune prediction methods.
What is easy to understand about the Zodiac is that it groups all peoples into twelve basic groups. The Zodiac has been around for many centuries and repeats itself every twelve years. Simply based on what year you and I were born, we should be able to know a lot about you before we even meet.
Vintage/Static Pool Approach to CECL – CECL is an approach to calculating the Current Expected Credit Loss of a loan based on certain characteristics. One of the most common characteristics used for the analysis is the year that the loan was originated. Does that sound like the Zodiac? The basic premise is that the total loss experience over the life of the loan pool with the specific characteristic (in Vintage/Static Pool it is year of origination) will be the expected loss for each and every loan that is remaining in the pool.
Below is a good example of the Vintage/Static Pool approach in that it takes the cumulative loss rates of each origination year and applies them to the associated balances to come up with the total expected loss of $7,400.
The main weakness of this approach is that the loan pools are still performing while you have balances. It is like telling someone what a branch of a tree will ultimately look like while it is still attached and growing on the tree. If you can guess at how much rainfall, sunshine, and nutrients the tree will get over the next few years, you be able to come up with a good estimate of how the branch will ultimately look. That is exactly the issue with this CECL approach is that you have to guess how other factors such as the economy, housing starts, and unemployment rates will affect your loss rates for each of the pools.
Another weakness of these models is that they offer very little actionable information. Looking again at our example, we see that loans originated in 2016 have the highest loss experience at 1.50%. While interesting, there is no action to be taken here. It is merely telling you yesterday’s results.
Chromosomes – A chromosome is a DNA molecule with part or all of the genetic material of an organism. In humans, each cell normally contains 23 pairs of chromosomes. Since we get 23 chromosomes from parent 1 and the 23 corresponding chromosomes from parent 2, the combinations and outcomes are limitless. However, when we do see certain groupings of these chromosomes in DNA, we can predict with deadly accuracy certain types of diseases and cancers.
Regression Analysis Approach to CECL – Regression Analysis is a set of statistical processes for estimating the relationships among variables. The main purpose of regression analysis is to return a single value based a number factors. In CECL’s case, we are calculate a loan loss value for a loan based on a number of factors associated with the loan such as Collateral Value, FICO Score, Interest Rate, etc.
The main challenge is to determine just what factors and how many are needed to do a decent job of coming up with the Current Expected Credit Loss for each loan. We already know that 23 characteristics will yield a limitless number of outcomes. Here is a simple formula of tracking 7 different loan characteristics with each characteristic having 10 different outcomes (such as an interest of 1.00%, 1.25%, 150%, etc.).
The beauty of using regression analysis for CECL purposes is that it yields a that is very objective answer. The whole point of regression analysis is that it provides a best fit based on the observed factors. There is no need for subjective guesses on the economy, housing starts, and so forth. The main struggle, however, is to figure out which factors to choose and how to manage the whole regression analysis process.
Another benefit of regression analysis is that it is not limited to your own institution’s data. Notice in our 7-factor model above, there are 10,000,000 possible combinations. To the extent that you can actually observe the actual 10,000,000 combinations in real life, the far more accurate the regression model becomes when reviewing your own institution’s individual loans.
Another benefit of regression analysis is that is yields vast amounts of actionable information. For example, the model may say that car loans with FICO scores below 550 and that have loan to value score of 80 or more, have a 50% higher chance than a typical loan of going bad within the first three years of origination. As a management team, you will update you underwriting to avoid this combination or, at a minimum, increase the interest rate you charge for that situation to cover your risk.
Zodiac Signs versus Chromosomes Recap – The Zodiac Signs (Popular Methodologies) approach to CECL essentially looks at a couple of factors to segment your loan portfolio into multiple pools that can be managed in Excel. Since these pools are still active in their loan loss generating capabilities, subjective factors need to be added to for future performance. Little actionable information is provided.
The Chromosome (Regression Analysis) approach to CECL, once built, is very easy to maintain. You simply track the 7 various characteristics for each loan to come up with a unique combination. That combination is then compared to the regression model’s 10,000,000 combinations to come up with a best fit for the loan loss. No other subjective factors are needed to modify the number. A side benefit is that vast amounts of actionable information will be provided.
Where to go for help
Popular Methodologies – There are many methodologies that are currently being explored by the market. Some of the approaches are Loss Rate, Discounted Cash Flow, Migration, and Probability of Default / Loss Given Default. A great resource for all these approaches is Google.
Regression Analysis – If you are looking for a regression model that has already been built off the loan loss experience multiple banks and credit unions, CECL Clearinghouse would be a good fit. This is a service that is being offered by Solver via its BI360 solution (Cloud or on-premise). All your institution needs to do is send a monthly loan file with seven characteristics to Solver and they will append the appropriate loan loss reserve to each loan based on the regression model.
Whatever you choose for CECL, you need to get started as the implementation date is fast approaching!