The first article in this series focused on input and processing. This article will focus on the importance of assumptions and output.
Assumptions
Assumptions cover a wide range of items including prepayment speeds, rate sensitivities (or beta adjustments), key rates, interest rate scenarios, and projected balance sheet. Assumptions are a critical component of modeling and, as a result, garner substantial regulatory focus because they can be both quantitative and qualitative. They can influence all CAMELS rating assessments. For these reasons, understanding your assumptions needs to be a primary focus of A/L management.
Several of the critical types of assumptions are loan and security prepayments, and core deposit repricing sensitivities and repricing lags.
Prepayments - Which type of prepayment assumption does your institution use? Do you gather them from market sources? Has the bank specifically studied how their loan portfolio has performed or does management rely on general prepayment estimates (i.e. qualitative)? Are your assumptions updated for each model run?
Mortgage loans and MBS prepayment tables are the most readily available prepayment assumptions to obtain, which would be an example of a quantitative assumption. Sources for these prepayments can be found on Bloomberg and BondEdge subscription services. They are also available for free on websites such as the OTS, the Bond Market Association, and from us by request.
As each mortgage type has different prepayment speeds, segregating out each mortgage product and assigning a specific prepayment table to that product will have dramatic effects on your liquidity, income, and ultimately value simulations. As an example of the vastly different speeds, in March 2006 the constant prepayment rate (CPR) of a newly issued 6% 30-year MBS was 11; meaning the instrument prepaid at a rate of slightly less than 1% per month. As a comparison, a 5.5% 5-year balloon MBS had a CPR of 36; over three times as fast as the 30-year.
No matter which type of prepayment assumption a bank utilizes, documentation should exist to support your reasons for using market sources or in-house assumptions. Periodically the bank should compare these assumptions to actual portfolio behaviors. As a better practice, these comparisons should be presented to ALCO for approval.
Core Deposits - This low cost funding source is the main source of economic value creation in community banks. The need for bank specific assumptions is critical for both income simulations and value calculations as they represent the key-funding source supporting interest earning assets. A "better practices" approach is to apply bank-specific repricing sensitivities, lags, and decay rates utilizing an institution specific core deposit study. For more information on this topic please see our previous series of articles on Core Deposits published in March-May 2006.
How does your bank treat these assumptions? Two things to keep in mind are documentation and review of your assumptions. Documentation provides support for how you developed the assumptions and how they apply to your modeling process. Periodically reviewing your set of assumptions ensures these key components stay fresh and dynamic.
Output
Output quite simply is your results and how you report them. This step in the modeling process determines if you have "garbage out" or useable results. The way to do this is to perform a "sniff test". As a manager, wouldn't you want to be certain the reporting you are providing to the board is valid, consistent, and accurate?
Sometimes you have to step back and ask, "Does this make sense?"
One way to approach this concept is to compare your results versus past model runs and actual data. Trended results are one way to compare past model runs with current data. After compiling the results, look for deviations from the trend. If there are deviations, you need to determine if there were any actual events that would explain these variances such as leverage strategies, large loan charge-offs, and deposit mix shifts. If nothing seems to make sense, then you need to dig a little deeper into your modeling process.
Another method of comparing actual versus predicted results is back testing. This method is recommended by the FDIC in their document entitled Supervisory Insights and the OCC Bulletin 2000-16. We will be discussing the concept of back testing in future focus articles.
Another important question is how do you use modeling results? Is the resulting output used as a strategic management tool or are they simply compiled to satisfy a regulatory requirement? Do you have to blow the dust off them when you hand them over to your examiners?
A financial model can give a bank a powerful method for estimating interest rate risks, but can also demonstrate the impact of strategic initiatives such as mergers, portfolio restructurings, and budget goals. These initiatives should be readily incorporated in your ALCO reporting.
Looking out beyond one year using various simulations can help you understand how your balance sheet is predicted to behave! All too often, managers focus on the near term without consideration for longer-term IRR exposures.
Institutions should consider modeling a 24-month time horizon at a minimum. This will assist the bank in several ways. Selecting a time horizon greater than 12 months provides the ability to capture the additional impact of repricing mismatches, extended durations, and optionality in the balance sheet. Additionally, it provides management with a de-facto budget for the next fiscal year regardless of where you are in the calendar year.
Another consideration when designing your report set is the "understandability quotient." ALM topics and concepts can be quite complex.
Your report set should include an executive summary that clearly explains the model's results, along with the major assumptions, trended results, model limitations, and explanations for any substantial variances in results. This will help senior management understand the limitations of the model, interpret the results, and make better business decisions.