Splitting Hairs and Risk Grades

One of the challenges lenders face is how to accurately assess the risk at origination for commercial or agricultural credits.  The next challenge is to continue this review of the risk through the life of the loan. As important as assessing the risk is, it is more important to reassess your risk rating models and scale.  If you have a scale or model that is in error, then the risk in the entire portfolio will be misread. 

In measuring risk, the starting point is to figure out what you are trying to measure.  When you book the loan, and through the loan’s life, you need to assess what is the probability that loan will default or PD.   Default may mean different things to different institutions.  Many times, when default is noted from a Fannie or Freddie perspective, it is a loan which is 90 days or more delinquent on a payment.  Of course, there are other types of default, such as payments that are only 1 or 2 months late or not following through on a loan covenant.  But for the focus on a risk model, it usually has a goal of predicting the probability that a loan will hit a serious payment default or worse status. 

Say you have a risk rating scale that generates one single resulting scale, then you should be able to apply a certain percentage for loan loss for all the loans in a rating category.  An example is your least risked loans may all be cash secured, so there is no risk of loss.  The next riskiest category, you may determine has a 1/100 of 1% loss and so on.  When looking at a traditional Allowance for Loan and Lease Losses (ALLL), you should add the loss percentages in each of these categories together with any special allocated losses you have on loans that are doubtful or worse loans, you should have a total for your allowance for the portfolio.

But other factors must be considered other than a PD to figure the allowance.  The next question is what your exposure at the time of default would be, or EAD.  What would be the projected balances on the loan when it defaults.  Loan balances will decrease with regular amortized payments or with prepayments of principal.  This is the probability of attrition, or PA.  The general condition of the economy and the interest rate cycle play a part in PA.  For example, if you have loans at a higher interest rate compared to current market rates, there is a higher chance loan balances will decrease with refinancing higher interest loans for lower, current rates.  Other prepayments may come when collateral is sold, and proceeds pay down the loan. 

The next factor in loan losses is what loss you have given default or LGD.  By this step you have figured the PD, which is a factor of the company performance, financial status, and to some extent, guarantor support.   The EAD is based upon projected balances at time of default, which looks at the PA.  LGD looks at the loan to value, presence of government guarantees, and involves some look at the guarantor strength.  The LGD would look forward based upon the economic conditions and other data points you study to figure your new loan loss under the Current Expected Credit Loss, CECL. 

Mixing all these factors together could be done in one comprehensive risk rating model, but it can be a little challenging.  A bad rating in the PD can be overshadowed by a good rating in the LGD and vice versa.  There is value in splitting the risk grades of the PD and the LGD into separate models to figure a better figure for loan loss reserves.  This way both sides of PD and LGD, can be captured more accurately.

I once had a loan to a company who experienced a severe drop in business.  Their financial status as a borrower would put this at a doubtful or loss status as the borrower cannot make regular payments anymore.  But the actual loss I had was nothing since we closed a loan at a 25% LTV and believed the property value to be like the date we closed the loan.  The LGD was $0 due to our collateral position. 

Consider a loan with a government guarantee. If the borrower goes to default status, your LGD is first based upon the difference between the loan and the net collateral value after selling, administrative, and legal expenses.  This LGD could be split into two with the enforceable government guarantee being set for a full recovery and the non-guaranteed portion as a total loss.

On the other hand, I once had a loan to a doctor clinic apart from the real estate.  The PD here was very low with a long history of strong financial performance of the clinic.  But if something happened, we would experience a substantial loss.  Our collateral of used furniture, medical equipment, and computers would not bring ten cents on the dollar.  The LGD would be very high.  We lowered our risk with key life insurance on the doctor that would pay off the loan. 

Understanding each of the factors of PD and LGD is essential to your risk rating models which will impact your loan loss calculations.  Your goal here is to find reality and not skew a model one way or the other.  These rating models need to be reviewed constantly and changes made when conditions dictate.