The Costs of Sovereign Default: Evidence from Argentina

That default is endogenous to the state of the economy, so the observed association between default and bad outcomes cannot be read as a cost. The authors set the question up through the classic exchange: “The seminal paper of Eaton and Gersovitz (1981) argues that reputational concerns are sufficient to ensure that sovereigns repay their debt. In a famous critique, Bulow and Rogoff (1989b) demonstrate that reputation alone cannot sustain sovereign borrowing in equilibrium, without some other type of default cost or punishment.” The size and shape of that cost is then “one of the key parameters that determines the equilibrium quantity of debt issued and the incidence of default” in the quantitative literature. Numerous papers have tried to measure it, but “the fundamental identification challenge is that governments usually default in response to deteriorating economic conditions, which makes it hard to determine if the default itself caused further harm to the economy.” The contribution is methodological as much as substantive: the paper supplies a shock to default probability that is plausibly uninformative about fundamentals.

A holdout creditor used the pari passu clause to have US courts block Argentina’s payments to its exchange bondholders, and Argentina chose default over settlement. The background: in 2001 Argentina entered a deep recession with fourth-quarter unemployment at 14.7 percent, and in December, after heavy IMF borrowing, “defaulted on over $100 billion in external sovereign debt and devalued the exchange rate by 75 percent.” A January 2005 unilateral offer was accepted by holders of $62.3 billion of defaulted debt; a December 2010 exchange took in holdouts owed $12.4 billion of principal, after which “the remaining holdout creditors were owed an estimated $11.2 billion, split between $6.8 billion in principal and $4.4 billion in accumulated interest,” and “Argentina had restructured over 90 percent of the original face value of its debt.” NML, having bought defaulted bonds issued under New York law, argued Argentina had “breached the pari passu clause, which requires equal treatment of all bondholders, by paying the restructured bondholders and refusing to honor the claims of the holdouts.” Judge Griesa of the Southern District first ruled for the holdouts on pari passu in December 2011, then clarified that repayment of holdouts was required under a “ratable” formula so long as the exchange bondholders were paid, and when Argentina kept paying only the latter, “ordered the financial intermediaries facilitating Argentina’s payments to stop forwarding payments to the restructured bondholders.” After the Supreme Court declined Argentina’s second appeal on 16 June 2014, Argentina sent a coupon to Bank of New York Mellon against court orders and BNYM did not forward it; the 30 June coupon went unpaid, a 30-day grace period ran, ISDA declared a credit event on 1 August 2014, and the 3 September CDS auction “resulted in a recovery rate of 39.5 cents on the dollar.”

Because they eliminated the status quo option. “The cumulative effect of these legal rulings was to change the menu of options available to Argentina. The status quo option, in which Argentina continued to pay its restructured bondholders without paying the holdouts, became infeasible. Instead, Argentina could attempt to settle with the holdouts and avoid defaulting on its restructured bondholders, or it could default on the restructured bondholders. Argentina effectively chose to default.” The authors flag the reading this creates for their own design: “In the simplest interpretation of these events, making the required payments was not possible, and Argentina was forced to default by the U.S. court system,” but “if a settlement was possible, the rulings might have also raised the probability of a settlement,” and if a settlement would have affected Argentine firms through some channel other than avoiding default, the exclusion restriction fails. Their answer is a magnitude argument developed in Section V: “the amount of money required for a settlement was small.”

Daily equity, CDS and exchange-rate data from 3 January 2011 to 29 July 2014. Equities are the American Depository Receipts of Argentine firms trading on the NYSE and NASDAQ – “relatively liquid, and… traded by a wide range of market participants” – which restricts the ADR sample to twelve firms, supplemented for cross-sectional work by peso-denominated stocks traded only in Argentina. The authors use the six-ADR MSCI Argentina Index and also build their own value-weighted indices for the whole market, banks, and non-financials, excluding YPF (nationalised in 2012) from the value-weighted indices and omitting a real-estate index because there are only two closely related firms. Default risk comes from CDS: Markit’s imputation of the risk-neutral default probability from the term structure of spreads via the ISDA Standard Model, with the five-year cumulative probability as the focus. Controls – VIX, the S&P 500, the MSCI Emerging Markets Asia ETF (chosen “to ensure that movements in the index are not directly caused by fluctuations in Argentine markets”), the Markit CDX High Yield and Investment Grade indices, and WTI crude – are included throughout, though the authors note “controlling for these factors is not necessary, under our identification assumptions, but can reduce the magnitude of our standard errors.”

Fifteen rulings by judges, deliberately excluding anything that could carry Argentine government or counsel information. The ruling list was built from the Wall Street Journal, Bloomberg News, the Financial Times, LexisNexis searches, and the website of the law firm Shearman. Timing is genuinely hard – “for many of the events, we are unable to determine precisely when the ruling was issued” – so the authors triangulate from contemporaneous news coverage, the date on the ruling, and PDF creation/modification metadata, which they describe as “suggestive but not definitive evidence.” They “include as events only orders by a judge or judges” and “exclude orders that were issued during oral arguments, because those events also include opportunities for lawyers representing Argentina to reveal information.” Several rulings are dropped but also removed from the non-event sample: three with no contemporaneous media coverage, one where the ruling itself could not be found, one issued the Friday before Superstorm Sandy, one the night before Thanksgiving, one issued at the start of an oral argument, and the 28 July 2014 ruling, excluded “because this ruling was made very close to the formal default date, and news articles on that day focused on the last-minute negotiations, not the ruling.” Non-events are non-overlapping two-day windows at least two days from any event or excluded event.

OLS is biased two ways, and the event study requires a stronger assumption than the authors want to make. The simultaneous-equations framework makes both biases concrete: simultaneity, because “poor earnings by large Argentine firms might harm the fiscal position of the Argentine government,” and omitted variables, because “a shock to the market price of risk… could cause both CDS spreads and stock returns to change.” “In order for the OLS regression to be unbiased, equity market returns must not affect default probabilities and there must be no omitted common shocks. These assumptions are implausible in our context, but we present OLS results for comparison purposes.” A conventional event study would require “that changes to Argentina’s risk-neutral probability of default during the event windows… are driven exclusively by those legal rulings, or other idiosyncratic default probability shocks” – the authors run it in the appendix and report similar results, but prefer the weaker condition below.

It identifies the effect from a change in variances rather than from a claim that nothing else happened. “This strategy instead relies on the identifying assumption that the variances of the common shocks F_t and equity return shocks η_t are the same on non-event days and event days, whereas the variance of the shock to the probability of default ε_t is higher on event days than non-event days.” Under that assumption, differencing the event and non-event covariance matrices of returns and default-probability changes gives a matrix proportional to [[α², α],[α, 1]], and the estimator is the ratio of the off-diagonal to the lower-right element: “covE(ΔD,r) − covN(ΔD,r) divided by varE(ΔD) − varN(ΔD).” The authors report that the instrument’s relevance is testable and tested – they “reject the hypothesis that λ = 0 using a test for equality of variances,” with the Stock-Yogo weak-identification F-test also reported – and that standard errors and confidence intervals come from a bootstrap. A useful visual check on the design appears in the summary data: on non-event days both the Argentine and the Mexican equity indices co-move with Argentina’s default probability, but “on the event days, only the Argentine equity index co-moves with the Argentine default probability measure.”

The Supreme Court’s 16 June 2014 denials, which can be timed to the minute from SCOTUSBlog’s live blog. The Court denied two appeals and a petition; the denial meant Griesa could stop Bank of New York Mellon paying coupons on the restructured bonds unless the holdouts were also paid, and “because Argentina had previously expressed its unwillingness to pay the holdouts, this news meant that Argentina was more likely to default.” SCOTUSBlog reported the denials at 9:33am EST, the petition denial at 10:09am, and a link to the ruling at 10:11am. “From 9:30am to 10:30am, the MSCI Argentina Index fell 6 percent and five-year cumulative risk-neutral default probability rose by 9.8 percent.” When Buenos Aires opened, “the local stocks associated with the MSCI Argentina Index opened 6.2 percent lower than it closed the previous night, implying virtually no change in the ADR-based blue rate.”

A 10 percent rise in the five-year cumulative risk-neutral default probability causes a 6.043 percent negative log return on the value-weighted ADR index. The effect is also “statistically and economically significant” for the MSCI Argentina Index, the value-weighted bank ADR index, the value-weighted non-financial ADR index, and YPF; the MSCI index falls by more than the authors’ own value-weighted index “because the MSCI Argentina Index consists mostly of YPF and bank ADRs, which fall by more than the ADRs of non-financial and real estate firms.” Extrapolation is done explicitly and labelled as such: “We linearly extrapolate the log-return to find that an increase in the risk-neutral default probability from 40 percent to 100 percent, which is roughly what Argentina experienced, would cause around a 30 percent fall in the value-weighted index, by our estimates. This increase also caused a 39 percent fall in the value of banks, a 30 percent decline in non-financials, and a 43 percent fall for YPF, by our estimates,” while “a completely unexpected default (a change from 0 percent to 100 percent in the risk-neutral default probability) would cause a 45 percent fall in the value index.” The stated conclusion is calibrated to that: “Our results are consistent with the hypothesis that Argentina’s default caused significant harm to the value of Argentine firms.”

The official rate does not move; all three parallel rates depreciate about 1 percent per 10 percent of default probability, significantly only for the black-market rate. “Unsurprisingly, given the Argentine government’s exchange rate policy, we find no contemporaneous effect of increases in the probability of default on the official exchange rate. We find that a 10 percent increase in the risk-neutral probability of default causes a 1 percent depreciation in all three measures of the parallel exchange rates. The results for the Dolar Blue (black market exchange rate) are statistically significant; the results for the ADR blue rate and blue-chip swap rate are not.” The authors also disclose that the latter two “are both substantially influenced by a single outlier event (the Supreme Court announcement),” while noting that across the other events both rates “exhibit a consistent pattern of depreciation during events in which the risk-neutral probability of default increases, and appreciation during events in which that default probability decreases.” Their interpretation is offered with an explicit caveat: the pattern is “consistent with the empirical coincidence of devaluation and default documented by Reinhart (2002), and with models in which a government finds it optimal to simultaneously default and devalue… However, we must emphasize that the exchange rates we measure are not the freely convertible exchange rates studied and modeled by those authors.”

Larger than the settlement, once locally traded firms are included. The paper multiplies 2011 market values by the estimated effect (converted from log to arithmetic returns) for the non-YPF index firms and for YPF separately, then extends to locally traded firms on the assumption “that these firms experience losses at the same rate as the firms with ADRs.” The result: “The losses for firms with ADRs are comparable to the ultimate cost of the 2016 settlement,” and “when considering the losses on these two broader classes of firms, the direct reduction in the market value of these firms as a result of default significantly exceeds the face value of all holdout claims.” The comparison matters for identification as well as for interpretation, because Section V uses it to rule out the settlement-payment channel: even assuming Argentina eventually pays the full $15 billion of holdout debt, “that represented only 3 percent of GDP, and 45 percent of foreign currency reserves, as of 2013,” and “because only a small fraction of Argentine firms are publicly traded, and there is no reason to expect the tax burden to repay the debt to fall exclusively on these firms, the direct repayment costs could only account for a tiny fraction of the loss of market value experienced by these firms.” Supporting that, the authors note that past settlements and the eventual 2016 one “were paid out of general government revenues, funded in part by issuing new government debt, and not borne by the particular firms we study.”

It is far too large for the usual calibration, which the authors flag as a puzzle rather than resolve. Average yearly earnings for the ADR firms including YPF, from the first quarter of 2009 through the second quarter of 2011, were “roughly $2.4 billion,” and “we estimate that, in response to a completely unanticipated default, the market value of these firms would decline by roughly eight years of annual earnings.” The benchmark: “in the quantitative sovereign debt model of Aguiar and Gopinath (2006), a country loses 2 percent of its output upon default, and is ‘redeemed’ with a 10 percent chance each quarter. If a firm’s ADR dividends followed the same process after a default, the ‘cashflow news’ (Campbell (1991)) associated with default would represent roughly 4-5 percent of the firm’s annual earnings. Accounting for leverage explains part but not all of the difference.” The candidate reconciliation is marked as tentative: “One (speculative) explanation for our results is that default causes very persistent or permanent output losses, or equivalently a decline in growth rates, and this is reflected in firm earnings,” consistent with Gornemann and with Aguiar-Gopinath’s trend-growth shocks – “However, we must emphasize that this is not the only possible explanation for our results, and that further research on this topic is necessary.” Even the earnings window is caveated: it was chosen for CRSP data availability and “that time period coincides with a recession in many developed countries.”

Exporters, foreign subsidiaries and large firms underperform; banks look badly hit but imprecisely so; importers do not show up. The four theories tested and their sources: Bulow and Rogoff on creditors interfering with exports; Mendoza and Yue on firms losing the financing needed to import intermediate goods; a bank-balance-sheet literature (Bolton-Jeanne, Acharya-Drechsler-Schnabl, Gennaioli-Martin-Rossi, Perez, Bocola); and Cole and Kehoe on “general reputation,” under which default signals a higher risk of disreputable behaviour in other arenas such as investment protection, so foreign-owned firms should suffer. The test is on zero-cost long-short portfolios of locally traded stocks, split at the median, with the estimated object being excess sensitivity to the default shock “above and beyond what would be expected from the Argentine equity market’s and exchange rate’s exposures to the default shock, and the sensitivity of the portfolio to the Argentine equity market and exchange rate” – a generalisation of Bernanke and Kuttner’s CAPM-inspired approach. Results: export-intensive firms underperform; “the long-short importer portfolio underperforms by a statistically insignificant amount”; the nine foreign subsidiaries underperform non-financials that are not subsidiaries, “consistent with the general reputation theory of Cole and Kehoe (1998)”; larger firms (above-median 2011 market capitalisation) “significantly underperform relative to smaller firms; however, this may reflect the relative illiquidity of smaller firms’ stocks, rather than a difference in real outcomes”; and ADR and non-ADR firms do not differ substantially. For banks the estimate is “economically large, but not statistically significant,” and there is a complication: “a ‘de-levered’ portfolio of bank stocks… outperforms a de-levered portfolio of non-financial firms, which suggests that the assets held by these Argentine banks are not substantially impaired by the sovereign default. This result is not necessarily surprising – Argentina did not default on its local law, locally owned debt.”

Deliberately little, and they say why in both directions. “The over- or underperformance of the portfolios is not an ideal test of the theories. For example, if we do not observe that importing firms underperform, it may be because the firms we observe are not the ones who would have difficulties, or because our import-intensive and non-import-intensive firms also differ on some other characteristic that predicts over- or underperformance (essentially an omitted variables problem). The reverse is also true; a significant result does not necessarily validate the theory, but might instead be found because of a correlation across firms between importing and some other firm characteristic.” Their summary verdict: “We interpret this cross-sectional analysis as lending modest support to several of the theories in the existing literature that try to understand the costs of sovereign default.” The abstract-level claim in the conclusion is likewise hedged – “We provide suggestive evidence that exporters and foreign-owned firms are particularly hurt by sovereign default.”

Judicial endogeneity, the settlement channel, RUFO, a general change in sovereign debt law, and a coordination-device channel that cannot be tested. On endogeneity: “Formally, the interpretation of the laws in question does not depend on the state of the Argentine economy. Substantively, because the amount required to repay the litigating holdouts in full was small relative to the Argentine economy… news about the Argentine economy’s prospects would not materially change the Argentine government’s ability to pay. Moreover, even if the judges were responding to economic fundamentals, under the null hypothesis that default does not affect fundamentals, the judges would have no information advantage over market participants.” On the settlement channel: meeting the courts’ demands required only $1.5 billion, “only around 10 percent of the estimated $15 billion holdout debt outstanding,” and “‘me too’ claims caused the size of the 2016 settlement with litigating creditors to grow to $9.3 billion” – yet even $15 billion is 3 percent of GDP. On RUFO, the argument turns on the word “voluntarily,” and the authors present the evidence rather than settle it: Argentina’s counsel told the Second Circuit that Argentina “would not voluntarily obey” orders to pay the holdouts in full; commentators saw loopholes; exchange bondholders could have waived, since 25 percent are needed to trigger; and “when the RUFO clause expired at the end of 2014, no progress in settlement talks between the holdouts and Argentina was reported.” Their fallback is that the binding case is the favourable one: “suppose the RUFO clause was binding, and settlement with the holdouts was not possible. In this case, the legal rulings caused Argentina to default, and our identification assumption holds” – and if RUFO were binding while settlement remained possible, rulings for NML should have raised restructured bond prices, whereas “we observe that restructured bond prices decline along with the stock returns.” On a general change in sovereign debt law: the appendix shows “the stock markets of Brazil and Mexico and the risk-neutral default probabilities of more than 30 countries did not respond to these legal rulings (our estimates are close to zero, and relatively precise),” in contrast to the OLS estimates, which do correlate – “presumably due to common shocks affecting Latin America or emerging markets more generally.” The channel they concede they cannot rule out is that the rulings acted “as a sort of coordination device,” provoking unrelated government action or changing the Peronist government’s hold on power – though “our costs of default are inclusive of the effects on government policy changes and political fortunes, if these changes occur because of the default. Our exclusion restriction is only violated if these changes are unrelated to sovereign default.”

The authors cannot separate them, and say the test they ran lacks power. They first argue the relevant pricing kernel is that of US investors, citing Edison and Warnock that these ADRs are held at float-adjusted market weights without home bias, documenting that they are held by “large, diversified financial institutions (who are also the type of institutions that trade CDS),” and noting ADR turnover well above that of the underlying local equities. They then argue the shocks are Argentina-specific: “The legal shocks are an almost canonical example of idiosyncratic risk, and it is very unlikely that U.S. investors’ stochastic discount factor is meaningfully affected by these legal rulings,” supported by the absence of any effect on other emerging market CDS and equity indices and by the near-irrelevance of the risk-price controls. Against a Gabaix-Maggiori-style shortage of Argentina-specific expert capital, they note that “Argentine-listed multinationals, such as Tenaris and Petrobras Brazil are unaffected by the default shocks,” so any such shortage “would have to defined more narrowly than firms trading on the Buenos Aires Stock Exchange.” That leaves two live readings: expected cash flows fell, or the dividends’ exposure to priced risk rose. The discriminating test is return predictability, and it is inconclusive: “We have run our heteroskedasticity-based estimator using two-day-ahead returns, rather than contemporaneous returns, as the outcome variable. We found no significant effects, but our standard errors are too large to rule out economically plausible return predictability.” They add that “for some purposes, it may not matter whether the value of Argentine firms fell because they were expected to be less profitable as a result of the default or because they became riskier as a result of the default.” The symmetric caveat applies to the CDS side: the changes are most straightforwardly read as changes in the physical default probability, but “it is theoretically possible that the legal rulings induced changes in the covariance between Argentine default or recovery rates and priced risk factors.”

The authors concede they are a small fraction of it, and address one specific worry. “Ideally, we would study the returns of assets whose cashflows exactly matched Argentine GDP” – Argentina’s real GDP warrants exist but “are very illiquid, have option-like features, may have been ‘pari passu’ with the holdout’s debt claims, and are affected by the measurement of Argentine inflation.” The concern they can address is that the ADR decline reflected an expected tightening of capital controls on dividend payments rather than real damage: “Ex-post, we know that this did not occur, and we are not aware of any evidence suggesting it was ever likely,” and had investors feared it, one would expect a wedge between local peso returns and ADR returns – but all three market exchange rate measures would themselves be affected by a change in the control regime, and “we do not find any evidence that there is a different effect across these three exchange rates. This suggests that, if changes in capital controls conditional on default were anticipated, the anticipated changes would have applied equally to bonds, stocks, and other means by which Argentines can acquire dollars.” They also pre-empt the charge that Argentine CDS are thinly traded: from Q1 2011 to Q2 2014, “the Argentine sovereign was the 15th most commonly traded sovereign CDS and the 48th most commonly single-name CDS overall.”

That its estimate is of the cost of default as Argentina would experience it, and that this could cut either way. “Our estimates of the cost of default include the consequences of whatever policies the government is expected to employ, conditional on default,” including renegotiation, finding other ways to borrow, balancing budgets through taxes or spending cuts, and affecting currency convertibility, plus the reactions of firms and households – so “the external validity of our results depends on the extent to which other defaulting countries would behave similarly to Argentina in the aftermath of a default.” Argentina’s standing was ambiguous: although the 2005 and 2010 exchanges reached 91.3 percent participation, “above the level generally needed by a sovereign to resolve a default and reenter capital markets,” an attachment order from ongoing litigation left the government unable to issue international-law bonds, while it could and did issue local-law dollar bonds, some bought by foreigners and some affected by the rulings. Hence the two-sided conclusion: “If the costs of default for Argentina were lower than that of a typical sovereign debtor, because Argentina was already unable to borrow in international markets, then our estimates understate the costs for the typical sovereign. On the other hand, because Argentina chose to default despite an ability to pay, the costs might be higher than is typical. These complications emphasize the uniqueness of Argentina’s circumstances.”