Using UniCourt to Study 272 Argentina Bond Default Cases in the Southern District of New York
on Topics: Legal Data API
This article is a featured guest post from Gregory Makoff, Senior Fellow, Centre for International Governance Innovation, an expert in sovereign debt and an experienced debt capital markets professional, Mark Weidemaier, the Ralph M. Stockton, Jr. Distinguished Professor of Law at the University of North Carolina at Chapel Hill, and Ram Balasubramanian, Independent Consultant, Maruti Solutions, a data science consultant who helps small to medium size businesses make intelligent decisions using data.
Introduction
If you need a good lawyer to understand a lawsuit, you need a good lawyer and a good computer program to understand 272 of them. We show why in a recent paper analyzing litigation against Argentina following its December 2001 default on almost $100 billion of bonds. The paper (available here) is forthcoming in the U.C. Davis Law Review.
The topic of our study is the famous—many say infamous—lawsuits between Argentina and certain “holdout” investors between 2002 and 2016, when most such litigation was settled. The “holdout” investors had refused the country’s offers, made in 2005 and again in 2010, to settle at 34 cents-on-the-dollar in new bonds. Many of the holdouts sued in the Southern District of New York. The wave of lawsuits, all assigned to Judge Thomas P. Griesa, was unprecedented in the history of sovereign debt litigation.
Most accounts of the Argentina bond cases focus on one plaintiff and one legal issue: Elliott Capital Management (often referred to as Elliott Associates) and its October 2010 motion asking Judge Griesa enter an injunction (the pari passu injunction) blocking Argentina from paying various of its international bonds unless it also repaid the holdouts in full. Judge Griesa entered this injunction in February 2012, and this decision was, indeed, the most astounding ruling in sovereign debt law for at least 100 years. However, what’s missed in the literature is that Elliott was not acting alone. The litigation included hundreds of cases, many thousands of individual plaintiffs, and dozens of different issues. Judge Griesa imposed the injunction to put an end to the mass of litigation inundating his court. To understand what really happened, and to identify patterns, we needed to study all the cases, and that, in turn, required the use of a computer. Three of us endeavored to carry out this study, each of us bringing a complementary expertise to the table: Mark Weidemaier (law), Ram Balasubramanian (data science), and Greg Makoff (physics and finance and the author of a book on the Argentina bond cases due to be published in the fall of 2023).
Measuring the Activity of Various Plaintiffs
Our analysis used two data sets: (1) the dockets for all 272 cases brought in the Southern District Court downloaded from UniCourt; and (2) a set of 145 transcripts from hearings held in the cases. The objective was to develop quantitative measures of the activity of the various plaintiffs, including the intensity of each individual suit, and the similarity of activity across the universe of cases.
Among our headline results are that Elliott and other sophisticated hedge funds did not bring the majority of the litigation in the early period of the cases, between 2002 and 2010, when Judge Griesa grew increasingly frustrated with Argentina. Nor were they the dominant drivers of judicial activity. For example:
- Between 2002 and 2010, two-thirds of the cases were filed by small plaintiffs (cases with median judgment size of $2 million) rather than plaintiffs with large claims (cases for institutional investors whose claims were mostly worth over $50 million).
- Many of the small plaintiffs were quite active in court, including in lawsuits filed as class actions. To give a flavor of the intensity of the activity by the small plaintiffs, the class action cases, all by themselves, involved twenty-four hearings and seven appeals.
- Large plaintiffs undertook significant attachment efforts independently of Elliott, including Capital Ventures International and Aurelius Capital Management.
- Elliott and Kenneth Dart, its co-plaintiff in numerous actions in this period, spoke less than half of the words spoken in court in the 2003 – 2010 period.
As a rudimentary measure of case activity, we first looked at the number of docket entries in each case. As expected, activity was unevenly distributed. The maximum number of docket entries was 1,087, while the median number of entries was closer to 100. We did observe that the larger, more sophisticated investors tended to be associated with cases with a higher number of docket entries than smaller cases. However, some of the small plaintiff cases were quite active, one of the dockets containing 367 docket entries, still a very high number.
We enhanced this analysis by focusing on docket entries that represent meaningful activity by litigants, excluding administrative orders (i.e. scheduling orders, assessment of fees, notices) or judicial activities (i.e. opinions, orders). We calculated what we called a Core Docket Activity Score for each docket, which included only docket entries flagged as evidentiary (declarations or affidavits), motions, legal memoranda, and attachment-related entries (attachment, garnishment, or restraint)—activities of relevance in these particular cases.
Figure 1 graphs the maximum Core Docket Activity Score for lawsuits filed by the various plaintiffs, plotting only the value for the docket with the most activity if a plaintiff filed more than one case. We see the most activity in Elliott’s pari passu case, but other lawsuits also generated significant activity, including the class action cases. As one would expect, however, many cases involved a much lower level of activity, although most plaintiffs did seek a judgment and did take some form of enforcement action. Very few plaintiffs were entirely passive.
Measuring the Similarity of Activity in Different Dockets
One benefit of using digitized data is that advanced pattern recognition algorithms can be employed to uncover activity that is not be readily discernible through manual analysis methods. Using our database of dockets downloaded from UniCourt, we developed a powerful metric to identify which of the 272 Argentina bond cases were being litigated in a very similar fashion—either as a result of outright coordination between the plaintiffs or copycat behavior. The metric we developed was a similarity score using Natural Language Processing (NLP) techniques to measure the “distance” between every pair of cases in our dataset. This similarity score measured the degree to which the entries in one docket were similar to the entries in another docket. The score was designed so that if two dockets were identical the score would be zero, while if they were entirely dissimilar the score would be one.
Figure 2 above shows the similarity score of one particular class action case (Silvia Seijas et al. v. Republic of Argentina No. 2004 Civ. 0400) versus all 271 other cases in the universe. The graph shows a cluster of seven cases with similarity scores of less an 0.05 whereas every other case has a score over about 0.15. This makes sense because these other seven cases were brought by the same group of lawyers and the only difference between them was that each suit was brought with respect to a different series of defaulted Argentine global bonds. To calculate our metric the docket text for each entry in each case is converted into a vector and the similarity score between a pair of dockets is based on a weighted sum of the vector product of all pairs of entries. As you can see in this figure, the metric clearly distinguishes similar from dissimilar cases.
We calculated the similarity metric for all pairs of cases, generating a two-dimensional grid of scores between all pairs of cases. This grid showed a notable clustering of cases, multiple groups of cases forming groups that lined up with the authors’ understanding of which plaintiffs were working together and/or were employing a copycat strategy.
Figure 3 shows a graphical representation of the main clusters found in analyzing the cases. Further explanation of the different plaintiffs, their strategies, the relationships between them, and our methodology can be found in our recent study Mass Sovereign Debt Litigation: A Computer Assisted Analysis of the Argentina Bond Litigation in the U.S. Federal Courts 2002—2016, which is available at https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4157688. The article is forthcoming in the U.C. Davis Law Review.
Conclusion
The availability of digitized court records, including from UniCourt, opens up new avenues for legal research. This paper has shown two possible applications: counting activity in court and looking for coordination of plaintiffs. When applied to the Argentina bond cases, this quantitative methodology uncovered striking patterns of behavior that would not have been practical to study without the assistance of a computer.
We have no doubt that computer analysis of legal documents will be a vibrant and growing field over the next decade. Given costs of data and data processing are coming down, the only limiting factor should be the creativity of the researchers. To maximize the opportunity, we suggest the formation of multi-disciplinary teams; carrying out this analysis we found that legal subject matter expertise, data science training, and a deep familiarity with the cases were all important contributors to our ability to extract meaningful results from the data.