A Primer on Implementing and Validating Prepayment Models in an Interest Rate Risk Measurement Context
by Mike Arnold and Bruce Campbell, ALCO Partners
Mortgage assets either purchased or originated and containing embedded prepayment options have proliferated and become ever more complex over the past decade, making the product-related measurement of interest rate risk (IRR) simultaneously more important and complex. Whether viewed from the perspective of regulatory compliance or effective portfolio and management decision-making, modeling and measuring IRR associated with prepayment has become a banking core competency.
Banking regulations pertaining to the measurement of interest rate risk enforced by the big three bank regulatory agencies (OCC, OTS and FDIC) require banks to establish a risk measurement capability and to validate their interest rate risk measures periodically, which in most instances means at least annually. These regulations are particularly important for banks with significant holdings of mortgage backed investment securities and portfolio loans with embedded prepayment options. Proper prepayment model implementation and validation are high on the examination ‘hit list’. The table below summarizes the key decisions in selection and validation of a prepayment model for marketable securities and for mortgages held in portfolio.
Buy versus Build Issues
Industry best-practice prepayment models continue to improve with better data and understanding of factors contributing to prepayment. Resolution of the buy versus build decision begins, as always, with questions about the time and resources available within an institution. Greater internal resources are required to build models than to purchase them for securities. Developing prepayment models is harder for portfolio mortgages than securities because of the lack of data.
The question of who develops the prepayment model is of great interest to regulators and relevant to model validation. Regulators are less comfortable with home-grown proprietary models than they are with prepayment speed assumptions obtained through relatively open sources like Bloomberg Dealer Median Speeds or published by the OTS. The widespread acceptability of OTS assumptions is somewhat anomalous, in that the OTS published speeds are based on highly aggregated pools and may compare unfavorably with both internally developed and vendor-estimated prepayment models. Vendor models are a best bet on the regulatory front as they are widely used and tested. A rule of thumb to follow: proprietary models are both resource-intensive and invite more regulatory scrutiny. Best-practices and regulatory guidance are aligned on one conclusion; all prepayment models should be validated on a periodic basis by the users or an independent third party.
Data Used to Validate Prepayment Models Applicable to Securities
Finally, prepayment speed histories on issued securities are available, both from general and specialized data vendors, and through various prepayment model vendors who provide it as an adjunct to their core software development and distribution. Vendor data typically enjoys the advantage of being formatted so as to facilitate the validation tests described below. If a bank already licenses a Bloomberg, for example, obtaining prepayment history at the security level can be obtained directly and with little marginal cost. Investment banks provide data on prepayment speeds for their clients and data vendors offer similar services. Banks can track the history of securities in portfolio and, over time, develop an internal source for validation tests in which projected prepayment speeds are compared to actual prepayment speeds, as described later.
Data Used to Validate Prepayment Models Applicable to Portfolio Loans
Many institutions already maintain historical loan data that can support prepayment modeling of portfolio loans and model validation. Those that do not should implement a program to capture these valuable data going forward. Often, just a few years of data can support sophisticated modeling and analysis of prepayment behavior of the bank’s particular borrower population or be utilized to adjust models applicable to similar collateral issued in securities.
Success in mining the bank’s proprietary data requires a meeting of the minds within the bank about the priority of obtaining data to develop models or perform validations. This value is frequently not perceived until the regulator requires it. Most institutions punt the internal data development, but they shouldn’t. Larger institutions are mining their databases for this purpose and regulators are slowly but surely moving down-market to require more banks to perform at least some data mining.
The initial steps to obtain, format and scrub the data on portfolio mortgages can be expensive and time consuming. However, once the data is obtained and formatted for analytical purposes, maintenance going forward is relatively inexpensive and straight forward, provided that the IT department doesn’t change its approach to data capture and storage.
In the middle tier of banks, vendor solutions and outsourced modeling are both cost-effective and widely adopted. IT departments are quick to provide data if they don’t have to format it. Validation Tests Validation tests performed in the industry fall into three broad categories.
