Managing a slippery asset.

Big hits from waves of prepayments in recent years have crushed the credibility of widely used prepayment model assumptions. This has left buyers of servicing looking for more foolproof ways to preserve capital investments in servicing. Here are some new ground rules to boost chances of getting good returns on purchased servicing.

FOR THE PAST DECADE, INVESTING IN purchased servicing has grown into a major business. Many investors have quietly enjoyed good returns. Others have experienced major failures, where losses of investment capital have been significant. The failures are well publicized, while the successes are not.

Regulators, reacting to the documented woes of failed investments in servicing, have been imposing limitations on the inclusion of servicing in banks' capital. Auditors and creditors, as well, have set more stringent requirements for measuring servicing value and calculating its risks.

Every investor in servicing knows that success in these ventures depends on superior performance from these assets in a few key areas, especially long life, low costs and low defaults.

Figure 1 shows how dependent a typical servicing investment is on these factors. Servicing is aggressively sought after, and today's servicing prices are high. Price multiples of five times annual service fee are common on new, 30-year, fixed rate conventional loans. Long life is essential if there is to be a return of the initial investment, much less any kind of return on that investment.

If servicing performs according to standard baseline assumptions, an investor recovers the initial cost of purchased servicing rights in the eighth year of investment life. A superior performing portfolio returns its purchase price in five years.

Learning from past mistakes

What can be learned from the failed investments? While any of the key performance elements can be miscalculated occasionally, the most common and most lethal deficiency has been in projecting and managing prepayments. There are several reasons this has been true:

* Cyclical interest rate declines--Interest rates have seen three important down cycles in 10 years. Declining rates are harmful to prepayment-sensitive investments, especially if they are purchased (as often happens) on the expectation of static rates. Servicing has been effectively hedged by only a few major owners.

* More-efficient refinance options--There has been a permanent increase in the efficiency of the primary and secondary mortgage markets. Collateralized mortgage obligations (CMOs) and other ingenious security structures have reduced absolute mortgage costs to borrowers. Product innovators also have found a ready market for a host of adjustable-rate mortgages (ARMs), intermediate term loans and balloons. These and similar types of mortgages now account for half the loans made. This sophisticated product menu allows today's knowledgeable borrowers to readily adjust to any change in the yield curve or satisfy any rate expectations. They also reduce friction among the borrower, the market and the ultimate lender. Even the time and cost involved in a mortgage refinance transaction has diminished continuously. The result is a new generation of faster-reacting mortgage customers who refinance more frequently, yielding a market of mortgages with shorter lives.

* Errors in basic techniques--Servicing investors who have encountered failure have consistently underestimated base mortgage prepayments through stable, rising and failing rate environments. The severity of the three major prepayment surges of the past decade has camouflaged this problem. The height of a dangerous wave at sea gets much attention. The depth of the water underneath it does not. Each time a prepayment wave has hit, a flurry of analysis has resulted to correlate prepays to random, interest rate-related phenomena. After each surge, investors expect a return to "normal," which doesn't materialize. Now, in retrospect, it's evident that factors other than interest rates were involved and that the "normal" prepayment rate is probably higher than once believed. Efforts to improve prepayment modeling have been blocked by the inability of observers to get the loan-level data that would reveal why prepayments have occurred.

Lack of relationship management--Prepayments are like the weather: everyone talks about it, but how many do anything about it? Many servicers take little or no effective action to identify and retain valuable borrower relationships. At times when refinancing is attractive to borrowers, smart servicers try to persuade them to recycle back into their portfolio.

To improve the performance of servicing investment capital, servicing investors must take three steps. First, they must forecast prepayments better by learning the discipline of behavioral prepayment modeling. Second, they must learn to buy or retain servicing that has the best possible chance to meet investment objectives. Finally, they must master the operation of servicing as a relationship business, taking active steps to improve the performance of this valuable goodwill asset.

Behavioral prepayment modeling basics

In spite of the apparent intricacy of mortgage calculations, a prepayment is just borrower behavior. Most borrowers respond to the world just like we do. They will move or sell a home, refinance, pay off or default on a mortgage for easily understood reasons.

A prepayment model is a mathematical expression that uses good statistical analysis to predict the behavior of individual borrowers. It is built by observing large numbers of prepayments, classifying the reasons for prepayment and linking those reasons to information in a loan's profile that would be predictive. The main output of such a model is a ranking of individual loans based on their probable tendency to prepay. In RF/Spectrum, (our PC-based portfolio management system containing a behavioral prepayment model), the relative prepayment likelihood of a loan is called its Prepayment Index (PPI), built from seven indicators from the loans data base. Each loan has characteristics that can (singly or in combination) be predictive of prepayment. A PPI of 1.00 means a loan will respond at generic national average prepayment rates, while a loan with a PPI of 1.50 means it is expected to prepay at 150 percent of national rates.

To be successful, a mortgage prepayment model must properly include all the principal behaviors that cause prepayment. If the model is based on observations of the wrong population, or if it uses faulty analysis, it will fail. Similarly, if it does not give weight to all the important influences on prepayment behavior, it will fail. No prepayment model can achieve perfection, but a good approach can greatly improve the odds of a winning investment in servicing.

The building blocks of prepayments

The American Housing Survey (AHS) of the U.S. Census Bureau reports recent mortgage prepayments that occurred during four years in successive two-year periods. This information helps us identify and quantify the building blocks that underlie actual prepayments. The AHS data showed:

* Sales of homes (that resulted in mortgage prepayment) averaged 4.1 percent per year.

* Loans paid-in-full without a new loan averaged 5 percent per year. Together, the prepayments that occurred before taking into account refinances averaged 9.1 percent per year.

* Prepayments from the refinancing of loans averaged 4.8 percent per year from 1985 to 1987 (a period of falling rates) and 2.2 percent per year from 1987 to 1989 (a period of stable rates), for a 3.5 percent annual rate overall.

* Another large group of homeowners, averaging 2.8 percent per year, moved between the time that these observations were made, but the fate of the loan was unknown. Some probably defaulted or prepaid as well.

Defaults are another important source of prepayment, but were not included in the AHS. The AHS results show that more than 28 percent of home loans outstanding in 1985 had prepaid by the end of 1987. Only one-third (9.6 percent) of those were due to refinance. A stable rate environment from 1987 to 1989 showed similar results: more than 22 percent of loans outstanding in 1987 had prepaid by 1989, 4.4 percent by refinance. The big surprise of the AHS data was the high level of non-sale, non-refinance payoffs in both survey periods. Apparently, many homeowners think paying off the mortgage is better than buying certificates of deposit (CDs). (Some of these payoffs may really be refinances reported inaccurately.) Like refinances, these prepays are interest rate sensitive, but have not been as volatile.

Findings counter the PSA model

The AHS model also includes vital data about the age at which loans prepay. Quite the opposite of the Public Securities Association (PSA) model, the results of the AHS studies of 1985-1987 and 1987-1989 both reveal that prepays in early years are higher than in later years.

The AHS 1987 survey shows that 15 percent of the people who bought a home in 1985 (and 10 percent of all 1985 owners) had moved by 1987. During the 1987-1989 period, the pattern repeats: 15 percent of 1987 buyers (and 10 percent of all 1987 owners) had moved by 1989. The likelihood of a move generally diminished the longer a family lived in its home.

The AHS findings imply that the PSA model may be incomplete or out-of-date as a baseline model in light of current behavioral analysis. At face value, the AHS findings suggest an average mortgage life of 6.7 years, less than the 14.6 years of the PSA model. Even allowing some offset for the uncertainties in' the study methods, it seems likely that what constitutes "normal" prepayment definitions must be revisited.

Revisiting current prepayment modeling practices

The current practice among most servicing investors is to predict prepayments using the PSA model as a base and then raise or lower prepayment levels based upon market interest rates relative to portfolio note rates. (A prepayment rate that equals the PSA model is called 100 percent PSA; twice that rate is described as 200 percent PSA.) This practice has been institutionalized, in part, by pressure from banking regulators, who push banks to use such "standard" approaches to valuation assumptions. Wall Street MBS dealers regularly publish prepayment data and forecasts denominated in both conditional prepayment rates (CPR) and PSA for use by investors. PSA is age-related for loans less than 30 months old. CPR is not.

Why does this practice miss the mark? The PSA model was developed in 1985 to simplify mortgage-backed securities' yield/price calculations. It was very useful in replacing a bewildering assortment of prior calculation practices. Even when it was introduced, the inventors never intended that this pricing convention would substitute for continuing factual analysis.

The basic premise of PSA is that prepays start off slow in early months and accelerate as loans age. But the AHS study now contradicts that. The PSA conclusion was based upon FHA mortgage prepayment observations from long ago (beginning in 1957) when the secondary market was almost non-existent. The interest rate environment at that time was also far different. Refinance alternatives then were more limited and comparatively very expensive.

Also, the PSA model is biased in a significant way because the statistical sample was limited to FHA prepayments. Those FHA loans were assumable, typically high loan-to-value (LTV) ratio mortgages. Their prepayment traits compare very poorly with today's conventional long-term, fixed-rate loans. They are especially questionable in predicting behavior of ARMs, balloons and the more exotic mortgage innovations of recent times. Dealers in interest-only mortgage-backed securities (that have characteristics similar to servicing rights) generally use carefully guarded proprietary behavioral models, based on a set of CPRs that change with time, to evaluate the likely prepayment behavior of those securities.

Improving behavioral prepayment forecasts

Behavioral prepayments can be better predicted based upon simple, easily observable factors:

* Interest rates-Changing interest rates provide the energy and motivation that drives a whole class of prepayments. These relationships have been thoroughly reported in many publications. What is not often analyzed is the impact that product innovation has had on this process. ARMs and intermediate-term loans are very new, but they achieved instant acceptance because they gave valuable alternatives to both lenders and borrowers. Before they existed, long-term rates had to move down materially before a borrower could economically justify refinancing. But with the new loan offerings available in today's market, a borrower can prudently refinance a long-term mortgage even when long-term rates have fallen little or not at all. If short-term rates drop relative to long-term rates (called a steepening yield curve), long-term debtors can reduce their rate and/or payments by refinancing "down" the yield curve, into shorter-term products. If the reverse occurs, and short-term rates rise above longer-term alternatives (an "inverted" yield curve), borrowers can be expected to refinance back up the curve to longer-term loans. This curve-shifting behavior is obvious from the pattern of prepayments. Furthermore, falling mortgage transaction costs mean lower refinance thresholds for borrowers.

* Housing activity--Housing activity is measured by sales of new and existing homes. Faster home sales mean that loans in that area will be prepaid faster. The National Association of Realtors (NAR) publishes sales data monthly, and Reserve Financial publishes quarterly a per capita measure called the "Housing Activity Index" for 140 state and local markets.

Housing activity moves in major, highly visible trends. It's easy to spot changes as they occur and incorporate them into prepayment forecasts, adjust valuations or take portfolio shifting actions.

* Homeowners' equity--Faster prepayments have been conclusively correlated to higher levels of home equity. Refinance requires that a borrower be both willing (incented) and able to act. The ability to refinance implies a borrower has sufficient equity and income to execute the refinance transaction. Home equity, in turn, is measurable, in the aggregate, from housing price performance in an area. If a market's prices have gone up substantially, so has homeowner equity. If equity is high, refinances will follow economic incentive much faster and home sales will increase, as well.

Conversely, a decline in home prices handcuffs homeowners. Borrowers can't refinance if equity is too greatly impaired, unless they have access to more cash to buy down the loan balance to make the deal work. Sale decisions also may be put off if the consequences of sale are emotionally or financially unacceptable. The primary indicator of homeowners' equity is what we refer to as the "adjusted loan-to-value ratio" (ALTV). ALTV adjusts original LTV for pay-downs of the mortgage balance and for changes in the value of the home. Borrowers with low ALTVs have access to more efficient refinance alternatives and have more mobility in making decisions to sell their homes.

The NAR also publishes home sales price data, which can be used together with loan-level portfolio data to calculate ALTVs. For instances where loan level information is not available, we publish the RF Home Equity Index(TM) an empirical measure of equity in 140 state and local housing markets. It is based on NAR home sales price data weighted by outstanding unpaid principal for each origination year. Similar to housing activity, home sales prices move in long, obvious trends.

When united with a good, interest rate sensitive model, housing activity and home equity knowledge will capture much more accurately the full range of homeowner prepayment expectations. Differentiation by region, ALTV, loan type and age is fundamental and easy. The information to do so is widely available at low cost.

The servicing relationship

Improving forecasts, of course, does nothing to prolong the life of servicing assets. Getting a return of, and a return on, an investment in servicing requires that the revenue-producing life of the asset be long. Yet the available evidence today suggests servicing investors face a highly informed, sophisticated borrower who is both willing and able to refinance. Marketers of mortgages are presenting borrowers with an endless stream of better, cheaper, faster refinance options. It seems likely that there is no reliable way to keep today's mortgage on the books for extended terms.

If that's true, then servicers must redefine their asset: it's not the mortgage, it's the borrower that matters. Buying a servicing right means getting an equity stake in a customer relationship. That relationship must be cemented and constantly enhanced, or it surely will be stolen by competitors. The industry must learn to keep the borrower relationship on the books, producing increasing revenue, no matter how often the form of that relationship twists and turns. For the smart mortgage banker, getting the loan on the books is not the end of the process, it's the beginning.

Unfortunately, the attitude of some servicers toward their clients too often ranges somewhere between neglect and hostility. Few continuously offer revenue-building products. If a trade-off between lower servicing costs and increased customer satisfaction is required, costs usually win. Ben Franklin said: "The borrower is servant to the lender." It may be true, but it's a bad marketing attitude.

Managing behavioral prepayments: an action plan

To improve investment performance, servicers should consider a four-point action plan:

* Make lower initial investments in servicing--Obviously, the lower the initial capital outlay, the shorter the time to recover the initial investment and the higher the yield. The highestpriced servicing assets today (five to six times the service fee) are Fannie Mac, Freddie Mac, conventional, conforming long-term, fixed-rate loans. Their quality and liquidity are unquestioned, but investors pay a steep price for those features. Virtually identical loans packaged as private investor servicing, for instance, are available at four to five times service fee. Adjustable-rate loan servicing is shunned by the market, and trades at two to two and one-half times service fee. If the liquidity or high standardization of the servicing isn't important, investment options such as those are probably more attractive.

* Make better prepayment decisions--Even if they intend to take no life-prolonging action, investors can seek to acquire loans that predictably will prepay slower. For instance, loans in areas with low housing activity will have longer lives, but their servicing rights will not cost a cent more than faster prepaying loans from areas with stronger markets. Housing markets with little or no home price growth will generate slow prepayments for years to come. (Watch out for steep declines in home prices that trigger higher defaults, however.) FHA loans, the primary ingredient in GNMA pools, have the high-LTV and slowprepay characteristics that flaw the PSA model. They not only stay around longer but they're less expensive, trading at around four and onehalf times service fee.

* Increase revenue and prolong servicing life by cross-selling more--It's self-evident that you can't have a productive selling relationship with clients unless you have things to sell and the dients know it. Also, you can't capture refinances for your own portfolio unless you have, or can get, refinance capability. Successful marketing of intangible products, such as first mortgages, home equity loans, investment services and insurance, requires special actions. Customers must perceive that you want their business and are capable of satisfying their needs. Being successful at this also requires a marketing mentality everywhere within the organization, one that measures and rewards crossselling performance.

Cross-sell smarter--Most cross-selling by servicers will be initiated by direct marketing (such as by mail or outbound telephone). Boosting direct marketing productivity needs a systematic, intelligent approach, formally called "propensity-based selling?" Servicing cross-sellers have the best shot at high yields, because they know so much about their prospects. Servicers know exactly which borrowers have note rates in excess of market rates--they don't have to buy lists. With investor agreement, their marketing can be extremely effective.

Borrowers with 50 percent ALTVs are great prospects for home equity products. Soliciting a 95 percent ATLV borrower would be an expensive and potentially embarrassing step. Servicers know payment histories and property types. Credit card and insurance product direct marketing yields can be doubled or tripled by preselecting the right target audience and mailing only to them. Top bank-owned mortgage servicers are especially good at this and regularly match their servicing data base to their other customer lists and to bank services profiles. A borrower tied to the lending institution by two or three well-delivered services makes for a very high-revenue, long-lived servicing asset.

Servicing is not just a financial asset, but a business that gets increasingly more challenging. Avoiding prepayment mistakes with better forecasts is essential. But the top performers are those that cukivate their borrowers as effectively after the loan is closed as before. These servicers can set the top of the market in price, if they choose, because they get the most return from each asset.

Hunter W. Wolcott is president and CEO of Reserve Financial Management Corporation, Miami.