Understanding your customer behaviors is a critical component to a variety of risk measurement and modeling processes. A great deal of focus is placed on Deposit Pricing (Beta) and Non-Maturity Deposits Life (Decay) assumptions because they significantly impact your income simulation and economic value calculations and results. One often-overlooked customer behavior is loan prepayment. 

Loan prepayment represents an important assumption for both ALM and Liquidity modelling. Loan prepayments modify the cash flow characteristics of a loan portfolio from their contractual terms into a realistic cash flow projection for a given scenario. Understanding past prepayment experience helps estimate future loan prepayments.  

Loan prepayments occur for a variety of reasons, ranging from systematic over-pays by the borrower, refinancing, sale of the collateral or business to death, divorce or forced liquidation of the collateral. For most modelling purposes, the reason for the prepayment is not as important as the result. An awareness of the possible reasons for loans to prepay, however, can guide future estimates of prepayments for certain stress scenarios. For example, in Liquidity Contingency Stress testing a scenario of a significant economic downturn in the local economy or a plant closure, you might expect increased prepayments from sale of collateral or even divorce, but a decrease in prepayments from systematic over-payment.

The biggest hurdle for community financial institutions in modelling historical prepayment behaviors is data. If the sample size is too small (generally less than 200 loans in a given pool), you end up with clumpy data that can provide unrealistic results. You can very easily produce a study with a zero prepay result one year and an extraordinarily high prepayment rate the next year given a small sample size. To overcome this, you can either limit the number of different loan pools to keep the loan count in each pool high, or you can average the loan prepayment results over multiple years. While it might be nice to segment your fixed rate commercial loans from your variable, if it lowers your loan count within either loan pool below 200, consider resisting the temptation. (You can always run it both ways if you want to see the difference.) The key is to segment your loans into pools that make sense logically, but also in a way that can be easily translated into your financial model. For example, you wouldn’t want to measure your prepayments by Collateral Type if your loans are grouped in your ALM model by Purpose Code.

The easiest way to measure your historical loan prepayments is to assemble a series of loan files that each list all active and paid off loans, with each file being a year apart. Loan prepayments are normally quoted in CPR or PSA, which are both annual measures, so the math is consistent. Each loan file should contain all information necessary to group the loans into appropriate pools, as well as an account number (for matching purposes) and all current financial terms necessary to compute its contractual future payment stream. Use those financial terms to calculate the contractual balance one year out into the future for each loan, and then compare that balance (by matching up account numbers) to the actual balance of that same loan one year out. Each loan will have one of the following three outcomes: 

1.       The actual loan balance at t + 1 year is the same as the projected contractual balance from time t. This loan shows no prepayment. 
2.       The actual loan balance is less than the projected contractual balance. The difference is the dollar amount of prepayment during that year. 
3.       The actual loan balance is greater than the projected contractual balance. This difference represents an underpayment. In many cases, this may   

          be just a single late payment if the difference is small. In other cases, it indicates a lack of performance by the borrower to satisfy the loan terms.

Prepayments for each loan pool are computed by totaling up prepayments for the year and dividing that by the average contractual balance for the 1-year period. This average balance can be estimated by (Actual Principal Balance(t) + Actual Principal Balance(t+1 year) + Prepayments(during t to t+1Yr))/2. While this will not produce an exact average loan balance, it is certainly close enough for prepayment estimation.

This process can then be repeated for each successive loan file and pool, starting with all loans with active balances at the beginning of that measurement period. This approach could overlook early loan payoffs that occur within the first year of a loan’s existence, but this should be a rare occurrence. This situation could happen on occasion for construction loans or other short-term loans, but if a loan with a 9-month maturity pays off a month or two early, that really doesn’t change the cash flow in a significant way.

Another data consideration is whether the loan data you are analyzing is impacted by charge-offs. If so, you will need to adjust for this as a change-off reduces a loan balance without prepayment. Lines of credit also present challenges as the balance is likely to go up during the draw period, before it begins paying back down.

Other, more sophisticated prepayment analyses are possible with large data sets and longer time frames. Given sufficient data, you can perform a separate analysis for various coupon ranges within a pool or during periods of rising and falling interest rates. The measurement methodology remains consistent, but the prepayment findings can be broken down further by coupon ranges or levels of seasoning, or the findings can be tracked over time through the peaks and troughs of the rate cycle.

NXTsoft offers a variety of
risk management solutions, including loan and deposit behavioral studies. For more information, please contact Tom Parsons at 303-320-1900 x706 or tparsons@NTXsoft.com.