Loophole seeking has become a standard tool in the world of distressed debt restructuring. The highest-profile example was the J. Crew restructuring in 2016. In that transaction, the company stripped $250 million in trademark collateral from the reach of its secured term lenders, and made that collateral available to refinance lower-priority debt. J. Crew’s advisers did this by finding a loophole. They combed through the 87,000-word loan agreement and found a “carve-out” provision intended to give the company freedom to make investments overseas in a tax-efficient way. But J. Crew claimed the provision also allowed its own collateral to be moved (“invested”) into one of its own subsidiaries that was “unrestricted”—i.e., free from the lenders’ reach.  The J. Crew transaction was repeated in several follow-on restructurings in companies like Neiman Marcus, PetSmart, and Cirque du Soleil.

In a new paper, we explore fundamental questions that the J. Crew transaction, and others like it, raise about the nature of contracts between sophisticated parties. Why would a loophole like this – an unanticipated exploitable mistake – come to exist? The example suggests that the contract’s complexity plays an important role. Indeed, the “trap door” carve-out was one of 21 different classes of carve-outs in the investments section of the loan document alone. One might expect that the more such opportunities arise, the greater the chance a clever borrower could find a loophole to exploit. A casual glance at the trap door term reveals several defined terms, such as “Investment,” “Loan Party,” and “Restricted Subsidiary,” and cross-references to several other sections of the document:

(t) Investments made by any Restricted Subsidiary that is not a Loan Party to the extent such Investments are financed with the proceeds received by such Restricted Subsidiary from an Investment in such Restricted Subsidiary made pursuant to Sections 7.02(c)(iv), (i)(B) or (n)

Additional interactions between these defined terms may also give rise to loopholes.

The possible connection between complexity and loopholes poses a modeling challenge for a contract theorist. Acknowledging the possibility of mistakes requires departing from a standard contract theory approach, which assumes omnisciently rational, forward-looking parties. It also requires a model that describes contract evolution over time and explains why contracts become complex in the first place.

We take on these challenges using a modelling approach that is new to contract theory: genetic algorithms. This tool, developed in the complexity science literature, has some attractive features for modeling the evolution of complex contracts. The main concept of the genetic algorithm is that “parents” from a current generation of individuals combine their attributes – i.e. their “DNA” – to form the “children” in the next generation. The probability of an individual being chosen to be a parent is increasing in their “fitness” – that is, their performance in the current generation.

As applied to contracts, we think these aspects of a genetic algorithm apply in real world commercial contracting. Lawyers rarely write contracts from scratch. Instead, they begin with contracts from previous deals or standardized forms and make adjustments from there. Also, since time is of the essence, lawyers cannot think ahead to fix every conceivable flaw. Instead, they tend to look backward, using contract terms that have worked well in the past and modifying ones that do not. The genetic algorithm incorporates these realistic features of contract evolution.

To model contract evolution using a genetic algorithm (GA), we start from a canonical GA created by computer scientist Melanie Mitchell. She designs a problem that an agent tries to solve by navigating a 10 square by 10 square board. There are rewards that are randomly placed on some of the squares. The agent’s fitness is a product of the number of rewards it finds. The agent’s “DNA” is a function that takes in what the agent observes in immediately adjacent squares (as an input, and outputs one of five possible actions (e.g. “move left”, “move up”, “pick up reward”). Mitchell’s game shows how agents that start with completely random DNA, and perform very poorly at first, can evolve complex, successful, and unexpected strategies over time through selection.

We modify this scenario to better represent the conflict between prinicipals and agents in financial transactions. Think of the rewards as investment opportunities and the DNA as the contract written by the principal to govern the agent’s behavior. We change the game to allow for green opportunities (value-creating) and red opportunities (value-destroying) to exist on the board. This represents a game in which the principal tries to manage an agent’s incentive to invest in more opportunities than the principal would prefer, as debt contracts do. The goal of the contract is to encourage the agent to explore around the board and find good opportunities, but also to avoid picking up bad ones. When the contract includes terms that give the agent more autonomy, he seeks out and finds more investment opportunities. But if it gives him too much autonomy, he will take bad opportunities. Restrictive terms act like covenants in debt contracts: They attempt to identify and block bad investments.

Much like Mitchell’s game, our GA generates some results and intuition that we did not expect. In early generations of our game, just as in Mitchell’s, contracts are very crude: They do a poor job of identifying and blocking bad investments. Because of this, contracts evolve to restrict the agent’s activity as a second-best solution. Analogizing to real world debt contracts, investment “baskets” limit the amount of money that the borrower can invest when investments are otherwise permitted. These terms limit the potential damage from a loophole; indeed, J. Crew’s ability to exploit the trap door was limited to “only” $250 million for exactly this reason. In our GA, we show that activity-increasing terms co-evolve with bad investment-blocking terms: As “bad blockers” get more accurate, activity-permitting terms become more permissive. But on the downside, this makes a loophole in a “bad blocker” term more severe. In this way, we highlight that interactivity between terms, one source of complexity, makes a contract less robust to mistakes as it evolves.

We also find that as more activity occurs, more states of the world will arise during the agent’s lifetime. This is another source of contract complexity. And this type of complexity, we find, increases the number of potential loopholes in the contract. Intuitively, the more terms in a contract come into play, the greater is the chance that a flawed term will materially affect the payoff to the contract. We hope that our paper will spur more theoretical and empirical research into the common causes and consequences of contract’s imperfections, even where sophisticated parties are on both sides.

This post comes to us from professors Kenneth Ayotte and Adam B. Badawi at the University of California, Berkeley – School of Law. It is based on their recent paper, “Loopholes in Complex Contracts,” available here.