Preventing Misleading Aggregation Results
While assumptions such as prepayment speeds, deposit pricing and withdrawal speeds get much examiner attention, an equally important modeling issue is the aggregation of account data. The NCUA’s Interest Rate Risk Questionnaire defines data aggregation as the act of combining general ledger accounts when inputting data into the model. Basically, it involves grouping “like” products. Grouping “unlike” products could result in unreliable results. Whether you are building your model from scratch or adding new products, consideration should be given to which accounts are aggregated.
One of the more common data aggregation “errors” we see in our independent model validation reviews is the grouping together of collateralized mortgage obligations (CMO) with pass-through mortgage-backed securities (MBS). Yes, they are both investments, they are both even MBS, but CMOs can behave very differently than a straight pass-through MBS. In a pass-through, each investor shares equally in the risk-return trade off. If rates increase, principal payments slow down for all investors the same. If rates fall, cash flows increase the same for all investors. In a CMO, investors purchase different pieces (known as tranches) of the CMO and not every investor shares the same risk-return trade off. If rates move up or down, purchasers of one tranche may not see their cash flows change very much while purchasers of another tranche may see material increases or decreases in cash flows. These latter investors have assumed more risk and are, therefore, compensated with a higher return relative to the investors who purchased the more stable cash flow tranches.
In the end, simply aggregating CMOs and MBS into a single account can provide very misleading results. It may be acceptable to take all of your FNMA 30-year fixed rate MBS pass-throughs and aggregate them, but CMOs should be separated out and each one modeled individually to capture its unique cash flow expectations.
