Every banker dreams of being able to beat the market. Whoever can outguess the consensus of financial actors more than 50% of the time stands to reap huge rewards. But that trick requires better information than is available to the general marketplace and the capacity to make better use of that information.
Many financial markets are so efficient that it isn't possible to gain access to information that is not widely or even universally disseminated. Others, however, afford opportunities for the knowledgeable few to outperform the information-deprived many.
Mitchell Madison Group has developed an approach that takes both conventional and often-overlooked information and processes it in novel ways in order to generate projections of financial-product cash flows that are more accurate than those generally available. The products are those whose values tend to be uncertain, usually because of the existence of formal or informal puts or calls.
The potential exercise of these options, quite obviously, complicates the task of mapping cash flows and thus determining product and customer values.
Mortgage default as well as prepayment, the decision to pay off-rather than revolve-credit card outstandings, and the sudden withdrawal of a long- standing deposit are all examples of option exercise.
If the timing of these events can be better approximated than is now the case, a financial institution that is, say, long mortgages and credit card loans and short deposits will be able to more clearly assess the value of these accounts and the customers who hold them.
Knowing the expected worth of individual customers can yield incalculable benefits: more intelligent customer segmentation based on actual behaviors; new business initiatives, particularly in marketing and risk management, stemming from that segmentation; and improved intrabank communications grounded in a more relevant corporate language-the language of lifetime value, not short-term accounting profits and losses.
The ability to proactively analyze individual customer relationships from a sound, rather than fragmented, quantitative perspective will shortly come to be viewed as a key core competency of the best financial institutions.
Such a competency is to a large degree already in place at a number of credit card companies. Many have not applied the skill as universally as is possible, but they have employed it to achieve unusually rapid rates of customer growth-e.g., First USA.
More specifically, armed with the ability to predict lifetime customer values, an institution can, among other things:
1. Outstrip competitors in identifying and originating groups of customers with a high propensity to maintain large product balances for long periods.
2. Cement the loyalty of these customers by tailoring products and initiating focused re-wards programs. Thereby, it can further increase customer longevity and reduce turnover and account-acquisition costs. (Eventually, as this process unfolds, the institution's name or "brand" becomes indissolubly linked with quality products and service, predisposing high-value customers to deal with it almost regardless of price.)
3. Consistently arbitrage the market, selling paper that analysis indicates the market overvalues (e.g., mortgages that will likely be prepaid before the market expects) or buying paper that is undervalued (e.g., mortgages that likely will be prepaid later than is anticipated or not at all).
The process by which institutions reach these coveted goals is a two- stage one. Stage 1 involves understanding the technology of information sourcing and evaluation. Stage 2 involves understanding the attitudinal changes needed to fully utilize the new informational endowment.
Stage 1, in turn, can be subdivided into two elements: (a) gathering and integrating relevant data and (b) building stochastic models of that data that translate into probabilistic cash-flow projections and reasonably accurate net present value calculations at the customer or individual level.
Data capture and integration - Despite energetic efforts to improve information resources, many financial institutions still have problems mobilizing the data they need in the form they require. Bank production systems were never designed or maintained for the level of precision required to calculate customer lifetime values. So much scrubbing is essential.
High on the list of data inadequacies are out-of-bounds or missing variables-prime examples of "data base dirt."
Another problem is that much relevant information is discarded. Mortgage origination data, for example, tend to be thrown out after the underwriting process is over. Yet these data are singularly rich, containing the answer to such questions as, has there been a previous refinancing? (Previous refinancings are highly predictive of future refinancings for some, though not all, subpopulations.)
A third difficulty-one that concerns the inability to integrate information usefully-is the paucity of analytical data bases. Financial institutions are well-endowed with transactional data bases but not with those that permit quick analysis of the relationship among data inputs.
Stochastic modeling-The new approach to modeling customer valuation differs sharply from that accepted and practiced by many in financial services. Points of difference in the mortgage area include:
1. Whereas the mortgage industry typically uses only a few variables to estimate prepayment and default probabilities of standard blocks of mortgages, we employ many more variables-as many as 60-to predict the behavior of individual loans.
2. Whereas the mortgage industry typically models customer behaviors in isolation (default probabilities separately from prepayment ones) the new approach models these behaviors interactively (default together with prepayment, risks that often vary inversely but on occasion directly.)
3. Whereas industry practice assumes a constant probability of prepayment for a given block of loans, Mitchell Madison's approach reflects the fact that probabilities change over time. It predicts these probabilities per unit of time for the individual loans, using both time- invariant factors (borrower's age at loan origination) and time-varying ones (the more obvious, such as shifting loan-to-value ratios, as well as some less obvious, such as changes in the individual's usage of credit during the past year).
The end results of this exercise are individualized prepayment and default probabilities, expressed in percentages, which, when linked to a set of unbiased interest rate forecasts and internal bank costs, yield a cash-flow map for each loan month by month for the remainder of its life. Such cash flows are, of course, discounted to their present value in order to reveal the worth of the individual mortgager to the institution.
How accurate are the resulting values? The best evidence suggests that those churned out by the new technique are about 20% more accurate than values derived from standard Wall Street analyses.
Thus if the prevailing market price for the servicing rights to a block of mortgages came to 150 basis points, an analyst who used the model could probably identify subgroupings of loans within that block with likely prepayment speeds much above average and therefore servicing rights worth not 150 but 120 basis points. These rights could then be culled and sold, fetching the prevailing market price, or 30 basis points more than their real worth.
To be sure, it is one thing to have reliable customer valuations; it is quite another thing to use them to maximum advantage. Although a financial institution is at bottom just a collection of customer cash flows, many institutions, hitherto clueless about individual customer values, have chosen to organize themselves not on a customer but on a product basis.
As a result, they are not now structured to, for example, offer pricing concessions or the equivalent to those customers whose projected lifetime values are superior, thereby depriving competitors of their patronage.
Two conditions must be satisfied for this to occur: Managers have to start believing what well-constructed decision-support tools are telling them; and the output of these models must be delivered in real time to the point of customer contact.
Too many managers remain unconvinced of the value of models, although sometimes with good reason. Partly as a result, little priority has been assigned to delivering model outputs expeditiously to sales and service personnel.
Because it will take some time to arrange that delivery, competitive advantage in the use of good models of customer value is likely to reside in those institutions that are already in the process of building a centralized marketing arm staffed by a cadre of personnel who are perhaps easier to train than widely dispersed platform employees-in short, a direct bank with strong emerging outbound as well as inbound sales and servicing capabilities.