I confess that, on its face, this did not strike me as the most exciting topic to read about (and that comes from someone who writes about the incredibly obscure world of sovereign debt contracts). After all, who even knows what EBITDA definitions are? Sounds like something from the tax or bankruptcy code. But don’t let the topic be off putting. This is a wonderfully interesting project; and elegantly executed (here). By the way, EBITDA stands for earnings before interest, taxes, depreciation blah blah. Turns out it is especially important for young companies, where potential investors want to know about the cash flow being generated (Matt Levine has been writing about it recently in the context of the WeWork debacle – here). It is also very important because it generally ties into the covenants in the debt instrument and can impact whether or not the covenants are violated.
Using machine learning techniques, Adam and Elisabeth look at the EBITDA definitions in thousands of supposedly boilerplate debt contracts. And they find a huge amount of variation in this supposedly boilerplate term; variation that can end up making a big difference to the parties involved. (For those interested, there is a nice prior study by Mark Weidemaier in the on how supposedly boilerplate dispute resolution terms in sovereign bonds are often not really all that close (here); and John Coyle’s recent work on choice-of-law provisions in corporate bonds is also along these lines (here))
The question that naturally arises here is whether the variation in these EBITDA definitions is the product of conscious and smart lawyering or just random variation that arises as contracts are copied and pasted over generations. (for more on this, see here (Anderson & Manns) and here (Anderson)). My understanding of the results is that these definitions are definitely not the product of random variation; instead, there seems to be a lot of sneaky lawyering to inflate the supposedly standard EBITDA measure.
