Sovereign Debt: Is to Forgive to Forget?

Jeremy Bulow and Kenneth Rogoff, American Economic Review 1989. Read here in the NBER Working Paper 2623 version (June 1988). No talk recording exists; written from the paper alone.

Eight years after Eaton and Gersovitz built sovereign lending on reputation, Bulow and Rogoff arrived with an eight-page proof that the foundation cannot hold anything. The abstract is one of the great flat statements in economics: “International lending to a less-developed country cannot be based on the debtor’s reputation for making repayments.” Not “is unlikely to be,” not “requires strong assumptions to be” — cannot. And the argument is not a model so much as an arbitrage, which is why it generalizes so alarmingly: it needs no assumptions on the borrower’s utility function beyond preferring more to less.

The arbitrage

Recall the reputational story: a sovereign repays because default means exclusion from credit markets, and credit access is valuable for smoothing consumption. Bulow and Rogoff ask the question a trader would ask: excluded from borrowing, fine — but can the defaulter still save? Their key institutional observation is that a country shut out of new loans can still walk into world capital markets as a customer paying cash up front. They formalize this as “cash-in-advance contracts”: pay a premium A today, receive a state-contingent, non-negative payout tomorrow — an insurance policy, a bank account, a portfolio of foreign assets, enforced not by the defaulter’s promise (worthless) but by the investor’s legal system, which works fine.

Now the trap. Any reputational debt contract specifies state-contingent repayments whose market value Dₜ can never exceed kW — some fraction of the value of the country’s entire future output. The authors show there must come a node in the game tree where the debt is high relative to that ceiling, and at that node the country does strictly better by defaulting and rolling the money it would have paid into cash-in-advance contracts indexed to exactly the same shocks. The insurance the reputation contract was providing, the defaulter can now buy retail, financed by its own suspended debt service. By backward induction the whole edifice unravels:

Reputation alone supports precisely zero sovereign debt. Worse for the reputational school, Theorem 2 adds that when lenders do have direct sanctions — the power to impose output losses κ on a defaulter — debt is supported only up to the present value of those sanctions, and a good repayment record adds nothing on top: “having a reputation for repayment in no way enhances a small LDC’s ability to borrow.” To forgive, in other words, is to forget — nothing of contractual value is destroyed by wiping the slate, because the slate was never load-bearing.

Where the theorem’s weight actually rests

The paper is admirably explicit about its pressure points, and the subsequent literature attacked all of them. The result needs (i) the defaulter to retain full access to saving instruments, richly indexed to the same shocks as its debt; (ii) competitive lenders, so no bilateral relationship has scarcity value; and (iii) no reputational spillovers beyond the lending relationship — and where such spillovers exist (their example: default triggering a trade war), Theorem 2 reprices the debt ceiling at the cost of the trade war, which is a sanctions theory wearing a reputation costume. The authors’ own candid discussion of “observable but not verifiable” shocks concedes the escape route that Grossman and Van Huyck took, and waves it off on the sensible ground that shocks visible to a huge pool of competitive lenders are hard to keep out of contracts — commodity prices and world demand can be hedged in ordinary asset markets even by a pariah.

What did the profession do with this? Empirically, it went looking for the direct sanctions the theorem demands and found them thin: gunboats are out of fashion, legal attachments recover cents on the dollar, and trade disruption around defaults is real but modest — which is awkward, because observed sovereign debt stocks are large. Theoretically, it spent thirty years rebuilding reputation with one of the theorem’s assumptions relaxed: defaulters who cannot save frictionlessly, lenders with information, richer punishment paths, political turnover, reputational spillovers into other arenas. The modern quantitative default literature — including the Hamann–Méndez-Vizcaíno–Mendoza–Restrepo-Echavarría paper two entries ahead — quietly sidesteps the theorem by assuming both exclusion and direct output costs of default, with the output costs doing the quantitative heavy lifting. That modeling convention is, in effect, the field’s settled verdict: Bulow and Rogoff won the argument about reputation in its pure form, and everyone now smuggles sanctions in through the output-cost parameter.

For this reading list the paper is the perfect dialectical partner to the entry before it. Eaton–Gersovitz: sovereign debt exists because default has future consequences in the credit market. Bulow–Rogoff: it cannot be only those consequences, because a saver’s market undoes them. Tomz and Wright, next, move the fight to the data — when do countries actually default, and does the pattern look like the models? — and the empirical answer turns out to embarrass the quantitative implementations of both camps, which is the kind of outcome that keeps a literature honest. (One housekeeping note for readers of the working paper: it is a 1988 document with genuinely rough typesetting — “NBER” appears as “NEER” on its own abstract page — and the published AER version is the citable text.)