Two General Data Protection Regulation (GDPR) Compliant Approaches to Scoring Firm Financial Frailty in Business Litigation

A litany of data artifacts, including the possibility of source data drift, lack of generalizability, and imprecise risk categories, all weaken—and may even impugn—estimates of a firm's economic frailty when using the Altman Z-score as a formulaic measure of risk in business litigation. These limitations constitute potential veto points which may be exploited by opposing counsel in court proceedings. We offer two possibly complementary approaches to obtaining estimates of the probability of a firm’s likelihood of business failure. To illustrate these approaches, we use the case study data in O'Haver (1993) and Local Outlier Probabilities (Breunig et al., 2000; Kriegel et al., 2009) and PRIDIT (Brockett, et al., 2002; Lieberthal, 2008) to first order the outcomes in terms of a numeric score. Once ordered, the scores represent either probability-of-insolvency measure or an insolvency ranking. We then map the scores onto bivariate classes using Fisher-Jenks clustering. Each algorithm’s accuracy is obtained by comparing its predictions of either failure of viability to those of the labeled data in O'Haver (1993). Both procedures are sound and with equal accuracy to the original discriminant analysis featured in O'Haver (1993). We hold that these competing approaches are capable of navigating opposing counsel objections. Importantly, our approach also falls well within the interpretability criteria demanded by the EU General Data Protection Regulation (“GDPR”) and other regulations taking aim at black-box algorithms.