Measuring unknowable risk taking behaviour

On the website of Fair Isaac (history) one of the early events recorded in 1958 was when the company sends letter to the 50 biggest American credit grantors, asking for the opportunity to explain a new concept: Credit Scoring. Only one replies. Yet that one reply from American Investment Corporation provided the humble launch pad to prove the irresistable benefits of credit scoring (lewrence 1992:74).

While a discourse of risk may have eventually triumphed to become the pre-eminent means of conceptualizing consumers. The technologies through which risk itself is constituted are seen by experts to be subject to a permanent process of failure. Although certain regularities can be seen within the population, the future actions of any one individual are not known, but are inherently unknowable.

The effectiveness of a credit-scoring model can thus be judged only generally on how well it distinguishes at the level of the population of consumers the distinctive sub groupings of good and bad consumers. Each of credit scoring models are seem to be subject to numerous risks themselves. Methodological risk suffer from the assumption of equal covariance and normal distribution within the population sample, while procedural risks attach to the specific construction of a model which shows the problem of sample bias (Lewis 1992a). Temporal risks pose a threat tot he integrity of a scoring model's risk determination (Henley & Hand 1997:525). All these risks are perceived to affect the ability of a formulated credit scoring model to distinguish groups of good and bad borrowers, deplete the accuracy of the risk assessment made at an individual level and degrade the efficiency of the lender at producing profit. At any given threshold, more costly defaulters will be accepted and more profitable consumers will be refused credit: therefore within credit scoring the construction of the constitution of risk must be constantly evaluated, maintained and recreated to preserve the reliablity of such constitution. But not only have statistical models been problematized, but also challenged by alternatives. Each technology seeks to know better the risk adhering to an individual applicant within the context of a population. Each is concerned with the calculable effects of default, not cause. In every case, default is conceived as an inherent aspect of the group and individuals are persistently conceived as agglomerations of attributes that are historically associated with a repayment outcome.

None of these alternatives provides a clearly dominant paradigm for the construction of risk interms of showing superiority in the practice of making credit decisions and are bound to common conception of risk. Whereas risk technologies are presented as an advance upon traditional judgemental sactioning processes, assessable through a discourse of efficiency that measures its superiority in terms of greater calculability and accurecy, lower costs and higher revenues, competing risk technologies are locked into a discourse of relativism.

Credit scoring model based on the consumer credit history data held by creditors transformed credit scores into a commodity that could be bundled with individual credit reports sold to lenders who were unwilling or unable to formulate risk-scoring models of their own or wanted to incorporate the score ranking within their own cusotmiyed systems, thus, it became standardized continuous measure of risk constructed within the context of the sider national population an enduringly standardized measure of risk.

Just as credit scoring transfers the uncertainty of repayment into a calculable risk, statistical modelling transforms uncertainties in segmented marketing, debt collection and fraud avoidance into similarly numerical probabilities incorporable into a more efficient organisation of those domains. But market specific conditions and unaccounted population drift and the simple perils of chance all conspire to render levels of default imperfectly calculable and make uncertainty a seemingly irreclucible aspect of consumer credit.

Forms of credit such as credit cards are more profitable for customers who are a higher risk when interest charges, fees and penalities are taken into account. There has been a recent shift away towards optimizing profitability indepndant of the minimization of default risk. This increases the complexity of data mgt to target profitability simply turning the risk of default risk one variable to be included within a more diffuse actuarial form of decision making within the lender organisation.