Introduction

Financial valuation is a cornerstone of modern commercial litigation, influencing outcomes across substantive areas of legal dispute, from bankruptcy to tax and corporate law. However, its ubiquity comes with substantial challenges for the judiciary. Conventional approaches to valuation, including discounted cash flow, comparable company, and comparable transactions analyses, leave open substantial areas of discretion to be exploited by economic experts. In our article, we argue that these “expert degrees of freedom” generate inconsistent and overly subjective valuations, with expert reports regularly diverging by orders of magnitude, to the frequent frustration of generalist judges. Using Monte Carlo simulations and a case example, our paper demonstrates the benefits of using contemporary statistical learning techniques to increase the precision of financial valuation while reducing this variability and expert bias.

Conventional Valuation Practices

The traditional valuation methods described above have been widely adopted in litigation. Each method involves a degree of subjectivity, allowing experts to materially influence the outcome of the valuation exercise. Discounted cash flow analysis, the notional “gold-standard” in litigation, requires an expert to estimate future cash flows, as well as the appropriate terminal value and discount rate. These decisions are more art than science, and they frequently rely on benchmark comparisons of discount rates from peer firms. The comparable transactions approach requires a finder of fact to determine appropriate reference transactions, without any guide for how such determinations can be credibly made.

We focus on the comparable companies approach to valuation, which generates a valuation estimate for a target company by taking the average (or median) of valuation multiples for selected comparable firms after adjusting for capital structure. We note in our paper that the comparable companies approach can be understood to be the functional equivalent of an existing statistical algorithm—matching based on k-nearest neighbors (k-NN). Both methods require an analyst to i) identify an outcome measure, ii) acquire a set of firm variables (or characteristics, in machine learning terminology), iii) generate measures of how similar the target is to the peer firms on these variables, iv) select the closest set of peers (this is the “k” in k-NN), and v) take the average (or median) of the outcome variable for the k-nearest matches.

Viewing comparable companies analysis as an example of k-NN matching reveals its inherent flexibility. Experts decide on the number of comparable companies, the distance metric, and the aggregation method, leading to a wide range of possible valuations. Our Monte Carlo simulation evidence demonstrates that this discretion can result in significant disparities between opposing expert valuations—even if experts on both sides follow textbook best practices. This undermines the credibility of the valuation process.

Our Approach: Data-Driven Valuation

In our paper we use straightforward and interpretable penalized regression methods to refine the valuation process. Using lasso, ridge, and elastic net penalization models we predict firm valuations based on historical market capitalization and earnings ratios. Through a large-scale simulation with firms’ publicly available data, we demonstrate that our approach offers more stable and reliable valuation estimates. We then show how our proposal could have been used in one particularly influential appraisal case— DFC Global Corp. v. Muirfield Value Partners, LP—to produce an estimate of firm value between the two provided by the litigants’ experts.

Moreover, our proposal allows for the use of daily, rather than quarterly, measures of firm valuation. We find that the more frequent measurement of firm valuation provides more precise estimates. Further, using daily data on firm value is already common via the use of event studies used to value the effect of firm events in securities litigation. Our approach thus unifies valuation methods used to value firms in securities fraud, appraisal actions, and other commercial disputes where company valuation is the focus. We believe that such a unification offers clear benefits to generalist judges who must adjudicate disputes in all these areas.

Conclusion

Our study challenges the conventional reliance on traditional valuation methods in litigation. By leveraging contemporary statistical learning techniques, we can achieve more consistent and empirically grounded valuation outcomes. This shift not only enhances the fairness and efficiency of the judicial process but also paves the way for more data-driven decision-making in corporate finance and litigation.