A Framework for Risk Management
in Diversified Financial Companies

by

David P. Greene, Ph.D.
General Electric Company

and

Ross M. Miller, Ph.D.
NatWest Markets

February 1996

This paper develops a high-level risk management framework for evaluating the risk modeling and evaluation process in a financial enterprise that markets several heterogeneous products lines. Given the current impracticality of a uniform metric for risk in such an environment, the focus of this framework is the process by which risk (and its relationship to return) is modeled. It is necessary that this process be on line from the time of a product's inception if an appropriate measure of risk is to be applied to the product when it matures.

A key design feature of this framework is to create a sensitivity to the numerous, sometimes extraordinarily complex, options that are embedded within ordinary financial products in the minds of key decision-makers. This framework also promotes the development of an early-warning system to detect problems associated with embedded options and other shortcomings of informally developed models in order to prevent or contain losses as the market evolves.

This paper begins with an overview of the framework, examines the five elements of the framework including examples of their application, and concludes with some comments about the framework.

A guide to the risk management framework is given in Figure 1. The five elements are numbered to reflect their relative importance. In practice, this framework has captured all of the salient risk-related features for a variety of financial services businesses that included insurance, leasing, consumer lending, and commercial lending. Furthermore, because this approach considers how businesses develop and refine their models as well as the models themselves, it can evaluate how a business adapts to the risks inherent in its ever-changing environment.

The first element, embedded options, concerns the options implicitly embedded in a financial product. These options are important because the failure to recognize them is a common cause of unanticipated variances in earnings. The second element, feedback, concerns the formal mechanism for assessing the outcomes of a business’s actions or customer products—-how it shapes and refines its models especially when they go "off target." The third element, structural dynamics considers the perspective used by the business to identify and evaluate its marketplace--how the business frames its models relative to its competitors and the rest of the enterprise. In Figure 1, structural dynamics is drawn on the fringe of the flow diagram to represent its meta-level role. The fourth element, model depth evaluates the detail and accuracy of the business's models. The fifth and final element, model consistency, concerns the degree to which the models used share a common framework, assumptions or "world view".

This framework can be conveniently summarized in the form of a one-page business "report card." The top half of the scorecard summarizes the five elements of the framework, with the structural dynamics included in the overview. The bottom half of the scorecard gives three or four major recommendations generated from the examination of the business within this framework. Detailed supporting material can also be used to back up the report card.

A fundamental premise of financial economics, that financial commitments can be decomposed into an underlying set of options, is a completely alien notion to the vast majority of practitioners in the financial services industry, with the possible exception of investment-related arenas such as mortgages and annuities. (The capital budgeting view of embedded options, despite its appearance in business publications, e.g., Sharp, 1991, has also had limited influence on practice.) The single most important element in developing an enterprise-level risk management capability is to promote the awareness of this financial fact of life and to promote its incorporation into both the models and mindset of the firm. This section will focus on two specific instances where the option-based view provides critical insights into how risk should be modeled and priced.

The first example of an embedded option is implicit default option that is contained in all consumer debt and is most important in unsecured debt. This option gives the borrower statutory means, such as bankruptcy, to limit recourse by the lender. The traditional approach to modeling default at the individual (and small firm) level is to view these events as being statistically related to observable attributes of the borrower and the economic condition of his locale (avoiding discrimination and red-lining either explicitly or implicitly). Such "scoring" models can be implemented in a variety of ways using techniques that range from classical statistical inference to neural networks. Within an options framework, however, default is the result of a conscious (and rational) decision by the borrower that is made in light of all associated costs and benefits so that the value of the default option depends on the projected volatility of these costs and benefits.

Although the rigorous incorporation of optionality into consumer lending is easier to preach than to practice, a true understanding of the risk of consumer lending is impossible without taking into account all the options that are given to the consumer in the context of the lending relationship. Defaults by consumers are not just random occurrences with observable statistical properties, they are consumer choices. (Fraud, which lenders tend not to view as a random occurrence, is a limiting case of default by choice where the borrower disguises his identity in an attempt to shield himself from the cost of default.) It is of more than passing interest that the increasing use of neural networks to enhance consumer credit scoring models, may be tightening the indirect link between the traditional observable attributes used to estimate default and the option-based view of default. (Cf. Hutchinson, Lo, and Poggio, 1994).

The second example that relates to consumer lending is the early termination option embedded in almost every loan product, which allows the borrower to "put" the loan back to the lender at a cost that varies anywhere from a slightly negative cost, when perverse incentives exist to roll the loan over, to a prohibitively high positive cost, when the profitability of the loan is linked to the (depressed) market value of the underlying collateral. The difficulty with these options is that they can be entirely invisible to the lender. Indeed, such options in the environment where the loan was initiated may have near-zero value because as options they were then very much out-of-the-money. The problem is that changes in borrower attributes, unbeknownst to the lender, can turn such options into in-the-money options. (The process by which such change can be observed at the enterprise level is covered in the discussion of structural dynamics later in the paper.)

The sudden emergence of invisible options like the one for early termination has the potential for a severely negative impact on the firm. Obvious damage occurs when entire portfolios of seasoned consumer debt are acquired and the option is not figured into the acquisition price. More subtle damage occurs, however, when options are embedded in loans that are already on the books, which financial theory tells us should already be reflected in the value of the firm. In this case, the sudden, unexpected inflow of cash and booking of losses associated with early termination can disrupt the financial processes of the firm, even leading to a downgrading of its debt in cases where the rating agencies require observable market feedback to value the firm properly.

The problem of hidden embedded options is often exacerbated by the tendency of lenders to ignore the status of borrowers once a loan is made as long as the payments are received in a timely manner. Despite the cost savings that can result, the potential for disaster from failing to reevaluate certain loans periodically becomes more apparent when the optionality of the loans is acknowledged explicitly.

In the previous section we evaluated one aspect of efficiency, that all options embedded in a product were properly valued. In this section we examine another aspect, that all the information related to the past performance of a product is used effectively in the decision-making process. Although most financial services companies collect enormous amounts of raw data, the decision of what data to collect and how to process it is often flawed. Once data collection has been organized in a certain way, it tends to become institutionalized and, therefore, exceptionally resistant to change. Not only can generating the resulting reports be irrelevant and time consuming, but the sheer excess desensitizes decision makers to useful signals. For reviewing risk, the issue is whether the information the business is collecting can be used and actually is being used to help them better understand and model their market.

Useful feedback lies on three dimensions: timing which considers the duration of actions, effects and observations, encoding, which considers which data are captured and how they are reported, and evaluation which looks at the incentives that influence interpretation of the reported results.

Timing relates to how quickly we know of a customer's choice, such as the exercise of an option, after it occurs. Timing is often effected by how readily the business's information systems can distinguish different states of nature and compile them in a usable form. Timely and accurate feedback can be especially difficult from the corporate perspective if a variety of disparate businesses have been acquired over time, because different "legacy" systems can be based on incompatible partitions of the states of nature.

More problematic for feedback is the duration of an event. As was pointed out in the previous section on embedded options, the duration of payment or benefit terms in a financial product such as mortgages or leases can allow for changes in the underling preferences and alternatives available to the customer. If those underlying conditions are not considered or monitored, the negative consequences can be rapid and severe. While due diligence and seasoning are standard practice in some markets, they are often overlooked by many businesses or products which have fundamentally similar structure. For new products or markets, the desire for rapid growth can create potentially serious exposure which goes unrecognized.

A practical dilemma is the contradiction of growth and consistency--shareholders want to see perpetual, sometimes unrealistic growth, but they want it to be stable. In practice, growth tends to be a volatile mix of success and failures. Using entirely acceptable accounting procedures, such as the timing and booking of sales or the allocation of costs, the business can report smooth results. Unfortunately, those same reports are often used to evaluate the marketplace and do not clearly reflect the underlying conditions needed to assess cause and effect (Bushman and Indjejikian, 1993). Also, without the addition of contextual information, such as prevailing economic conditions at the inception of a market action relative to current economic conditions, a business would lack the necessary feedback to refine its models, even though from an accounting standpoint it would appear to be profitable.

This raises the difficulty in how feedback generates guidance and reinforcement. Even if systems are set up to provide useful feedback about risk, unless the incentive structure rewards an appropriate risk-adjusted measure of return, the feedback will have no effect on the process. Indeed, with insufficiently sophisticated systems, it may be impossible to generate an incentive-compatible measure of return.

Compounding the problem is the perception that if things are going well financially relative to an arbitrary benchmark, what we are doing is all right; hence, we know what we are doing and should continue doing it. As we noted earlier, with new products or markets large firms bear the risk of informal models by employing large tolerances. Over time, with the appropriate use of feedback, the models should become more accurate and the tolerances more precise. In practice, there can be a tendency not to refine models after a successful production introduction because the initial success can be misperceived as model accuracy which reinforces the initial informal model (Einhorn and Hogarth, 1978).

For example, the idea of being conservative about risk by underwriting to target zero loss levels, which is unexplored in the academic literature but occurs in practice, can result in setting thresholds too high. The business may be successful at screening out bad credit risks initially but is leaving opportunities for competitors to get a toehold in the market. By not developing an accurate model of the market, the business can end up "conservatively" defending a share of a diminishing market when, in fact, the market is not diminishing but has simply transitioned into a different form.

A third dimension in assessing enterprise-level risk, structural dynamics, requires understanding the context and environment in which the models are evaluated, both from the perspective of the parent company as well as the individual business. Most businesses will categorize themselves by their industry, sub-industry, product, customers and competitors. In some cases those designations are critical for regulatory purposes and constrain the firm's behavior; however, in most cases the categories are arbitrary, overlapping and self-selected. While such classifications are useful for providing comparisons or referent standards they can led to distortion if they provide too narrow a perspective for modeling.

Problems arise when a business imputes some absolute association to its self-perception and its products; for example, it views itself as a bank because it has "bank" in its name and offers banking products. As long as the market accepts those definitions the perspective is appropriate. But if customers start viewing the business merely as the utility it provides (e.g. interest and liquidity) a perceptual shift can occur as new information and alternatives become available. Perceptual shifts can be especially acute if legal or regulatory barriers exist in one market but not the other, as was the case when Merrill Lynch’s original Cash Management Account was able to sidestep banking regulations to offer better value for similar services (Farkas, 1991). Conversely, some businesses face the risk of vanishing overnight though the passage of a single piece of legislation, e.g., The Tax Reform Act of 1986.

A major competitive advantage that diversified financial services firms possess, at least in theory, over more specialized firms is that aggregate shifts in consumer behavior detected in one business can serve as an early warning system for related businesses. For example, the structural shifts that led to unprecedented mortgage refinancing in the late 1980s and early 1990s were a harbinger of things to come in other consumer financial businesses with embedded early termination options.

Implicit in all models is the partitioning of the possible states of nature. Model depth is the fineness and appropriateness of this partition. Like feedback, the depth available to a model can be highly dependent on the information systems that provide the data for the model.

Financial service firms apply a wide range of models to estimate credit risk, residual value (for leased assets), and profitability (for credit cards). From a statistical standpoint, assessing depth is relatively straightforward where clear objectives can be specified for quantitative models, unfortunately this is rarely the case. As noted in the previous section, the choice of what to model and how to model it is usually a reflection of standard practice in the industry. Although such practices are perceived as conservative since they represent the prevailing standard, they can reinforce a restricted perspective making it harder to detect structural shifts and to differentiate the businesses product.

While a top-down perspective of the business should motivate model construction with bottom-up validation through empirical data, in practice, the models usually evolve in a piecemeal fashion to address specific concerns or requirements. In many cases the inclusion of certain features are historical artifacts which may no longer be appropriate but which are institutionalized in the systems and accounting.

The final element examined in assessing enterprise-level risk was the consistency of the quantitative models employed in the decision-making and accounting processes. The level of consistency that is implicit in the theoretical financial literature is impossible to achieve in a diversified financial enterprise. The decentralization of decision-making required for the enterprise to function guarantees inconsistency in the models and their underlying assumptions in both their construction and implementation. It is financial folklore, for example, that even Wall Street's most prestigious firms with unlimited access to state-of-the-art modeling techniques can have a dozen or more different, mutually-inconsistent models of the term structure of interest rates in operation simultaneously.

Although such inconsistency can have a rational basis; in reality, it serves as evidence of potentially serious problems when the inconsistencies cannot be rationalized. The gratuitous failure to create a consistent framework for the models within a business is strongly symptomatic of problems within the models themselves beyond those directly attributable to the inconsistency. Although Ralph Waldo Emerson is well-known for saying that "a foolish consistency is the hobgoblin of little minds," in this case consistency is not at all foolish but rather a superb signal of overall model quality.

Before concluding, there is an interesting rationalization of model inconsistency that needs to be acknowledged because it reflects an interesting divergence between the worlds of theory and practice. In more than one instance the authors have encountered practitioners who claim to consciously use inconsistent models as a means of risk diversification, so that if one model turns out be flawed, the entire holdings of the business are not affected. The suggestion that such diversification can also be intentionally achieved within a consistent meta-model is not well-received. In part, this poor reception appears to be related to a subtle cognitive bias that is very difficult to remedy; however, it is also related to practical difficulties in achieving consistency across models with fundamental different functionality. For example, calibrating models that project regional business conditions for different types of consumer loans made within a single business so that they match on a region-by-region basis can be an impossible task when the models rely on underlying states of nature that are orthogonal across models.

The risk management framework described in this paper has proved in practice to be an excellent mechanism for understanding how risk manifests itself within individual businesses and how risks are related throughout the enterprise. In many instances adoption of this framework has led to immediate and substantial changes in the risk management modeling procedures within a business. Not surprisingly, the notion that "everything is an option," which is second nature to financial economists, is difficult to incorporate into the mindset of many practitioners; however, this framework has made some progress in that area. The subtle ways in which optionality can manifest itself in financial products requires a deeper education than that which can be provided by this framework.

Bushman, Robert M. and Raffi J. Indjejikian, "Stewardship value of 'distorted' accounting disclosures," Accounting Review, Vol. 68, No. 4, 765-782, October 1993.

Einhorn, H. and Hogarth, R., "Confidence in judgment: persistence of the illusion of validity," Psychological Review, Vol. 35, No. 5, pp. 395-416, 1978.

Farkas, Mitchell S., "The account that transformed a brokerage into a bank," Financial & Accounting Systems, Vol. 6, No. 4, pp. 8-12, Winter 1991.

Hutchinson, James M., Andrew W. Lo, and Tomaso Poggio, "A nonparametric approach to pricing and hedging derivative securities via learning networks," Journal of Finance, Vol. 49, No. 3, pp. 852-889, July 1994.

Sharp, David J., "Uncovering the hidden value in high-risk investments," Sloan Management Review, Vol. 32, No. 4, pp. 69-74, Summer 1991.