Local Currency Sovereign Risk

It asks whether governments default on debt denominated in their own currency, and it could not previously be answered because no empirical measure of the credit risk on local currency sovereign debt existed (Introduction, p. 1). The authors note that “explicit LC sovereign risk measures are absent from the academic literature,” even though local currency debt had become the dominant form of emerging-market sovereign financing: total LC debt outstanding in 2011 averaged 5-6 times FC debt outstanding across the sample countries (citing BIS, 2013), foreign holdings of LC debt rose from roughly 5 percent of LC debt outstanding in 2004 to over 20 percent in 2011, and the LC share of total offshore emerging-market debt trading volume rose from 35 percent in 2000 to 71 percent in 2011, reaching $4.64 trillion (pp. 2-3). Meanwhile the conventional measures were becoming unusable: J.P. Morgan’s EMBI+ had to discontinue country indices for Egypt, Thailand, Malaysia, Morocco, South Korea and Qatar as their FC debt shrank, and sovereign CDS are an incomplete substitute because in emerging markets “defaults on local currency bonds governed under domestic law do not constitute credit events that trigger CDS contracts” (pp. 3-4).

It is the LC bond’s nominal yield minus a synthetic LC risk-free rate formed by adding the swap-implied long-term forward premium to the U.S. Treasury yield (Section 2.2, pp. 11-12). Formally the FC credit spread is the FC yield minus the Treasury yield at the same tenor, while the LC credit spread is the LC yield minus the Treasury yield minus the zero-coupon cross-currency swap rate. The authors emphasize that a domestic government bond yield cannot serve as the risk-free rate in an emerging market “where the risk of sovereign default and capital controls are non-negligible” (p. 1), so the risk-free benchmark must be imported from the dollar curve and converted through the swap market. The measure admits two readings: as the spread of the LC bond over a synthetic LC risk-free bond (a pure credit spread in local currency), and as the synthetic dollar spread over Treasuries on a swapped LC bond, which a dollar investor “can lock in … even if the value of the currency plummets as long as explicit default is avoided” (pp. 1-2). Section 2.1 and Proposition 1 show that a fixed-for-fixed LC/dollar cross-currency swap can be decomposed into two interest-rate swaps, making it “completely analogous to a standard forward contract” and letting the authors extract a zero-coupon forward premium from the term structure of par swaps rather than wrestling with coupon and payment-date mismatches (pp. 9-10).

Daily zero-coupon LC and FC sovereign yield curves plus swap curves for 10 emerging-market governments, January 2005 to December 2011, at a 5-year benchmark tenor (Section 2.2, p. 10). The country set “is mainly constrained by the lack of sufficient numbers of FC bonds outstanding,” and all ten belong to the J.P. Morgan EM-GBI investable index for emerging-market local bonds; the sample’s length “is constrained by the availability of long-term currency swap data” (p. 10). Swap data come from Bloomberg, with extremely illiquid days (bid-ask over 400 basis points on the swaps, mainly Indonesia during the 2008 crisis) excluded, though the authors report the main results are unaffected by including them (fn. 4, p. 10). A validation check finds a mean 99 percent correlation between one-year forward premia implied by the swap market and by the forward market.

It averages 128 basis points across the ten countries, ranges from roughly 60 basis points in Mexico and Peru to 313 basis points in Brazil, and is positive and statistically significant for every country – including after netting out swap transaction costs (Section 3.1, p. 13). Significance is assessed with Newey-West standard errors allowing for heteroskedasticity and serial correlation. Half the bid-ask spread on cross-currency swaps – the relevant one-way transaction cost – averages 19 basis points, and the authors report that “LC credit spread remains significantly positive for every country after subtracting one half of the bid-ask spread on the CCS” (p. 13). They read this as evidence that “emerging market nominal LC sovereign bonds are not free from credit risk from the investor’s perspective,” and stress the magnitude is what distinguishes it from the small, transient covered-parity deviations previously documented in developed markets: the failure “is so large as to make clear the importance of credit risk on LC debt, rather than only pointing to a temporary deviation from an arbitrage relationship” (Section 3.6, pp. 20-21).

Less expensive in promised synthetic dollar terms: FC credit spreads average 195 basis points against 128 for LC, a 67 basis point gap that grows to 86 basis points once swap transaction costs are included, and the LC-minus-FC differential is significantly negative for every sample country except Brazil (Section 3.2, pp. 14-15). The authors point out this comparison understates the gap for nine of ten countries, because LC bonds are typically subject to local taxation while FC international bonds are exempt from interest withholding taxes, so “swapped LC over FC spreads would become more negative after taking into account positive taxes on LC bonds” (p. 14). Brazil is the exception and the authors treat it as informative rather than anomalous: its Imposto sobre Operaçoes Financieras (IOF) tax on financial transactions, introduced in October 2009 and abandoned in June 2013, varied between 2 and 6 percent on foreign investment in fixed-income instruments; LC and FC spreads diverged sharply after the rate went to 6 percent in October 2010, reaching a record 400 basis points in November 2011, and converged to around 50 basis points after the tax was removed on 5 June 2013 (p. 14). Brazil’s real-denominated eurobonds traded on the Luxembourg Stock Exchange – payable in dollars and so free of convertibility risk – trade at significantly lower spreads than onshore bonds, which the authors read as direct evidence that “taxes and convertibility risk are important components of the LC credit spread from the offshore investors’ perspective” (p. 15).

Foreign-currency spreads move almost as one; local-currency spreads are far more country-specific (Section 3.4, pp. 16-17). Principal component analysis gives the first component less than 54 percent of the cross-country variation in LC credit spreads but over 81 percent for FC spreads; the first three components reach slightly under 80 percent for LC against about 97 percent for FC. Average pairwise cross-country correlation is 42 percent for LC spreads versus 78 percent for FC. Sovereign CDS behave like FC spreads, not LC: the first principal component explains 80 percent of CDS variation with pairwise correlations averaging 77 percent, consistent with Longstaff et al. (2011). Within a country the two spreads are nonetheless positively correlated – mean 54 percent, from 91 percent in Hungary down to 18 percent in Indonesia – and the authors read that cross-country heterogeneity itself as a hint that “incomplete market integration is potentially important in the relative pricing of the two types of debt” (p. 17).

Foreign-currency spreads track global risk factors closely; local-currency spreads much less so (Section 3.5.1, pp. 17-18). The first principal component of FC credit spreads correlates 93 percent with the VIX, 88 percent with the Merrill Lynch U.S. BBB corporate spread over Treasuries, and 76 percent with the Chicago Fed National Activity Index; for LC spreads the same correlations are 76, 71 and 57 percent. Because the first component explains so much more of FC than LC variation, the gap is wider in raw spreads: the VIX correlates 70 percent with FC credit spreads on average across countries but only 41 percent with LC credit spreads (Panel B, p. 18). CDS spreads again behave like FC spreads. The authors’ conclusion is calibrated to what these are – correlations, not identified causal effects: “the observed global factors are more important in driving spreads on FC debt than on swapped LC debt” (p. 18).

Yes – once currency risk is swapped away, LC bond excess returns carry no significant loading on global equity returns, while FC excess returns do (Section 3.5.2, pp. 18-20). The authors define three dollar excess-return variants on an LC bond depending on the hedge – unhedged, holding-period hedged with an FX forward, and swapped – and regress quarterly excess returns on global equity excess returns (S&P 500 over 3-month T-bills) and on local equity excess returns hedged to the same degree. FC excess returns have significantly positive betas on both global and hedged local equity returns, with the S&P loading the larger of the two. Hedged and swapped LC excess returns “do not load on the S&P, but have a significantly positive beta on local equity returns,” while FX-unhedged LC excess returns load on both (p. 20). The interpretation the authors draw is precise about where the global risk sits: “for foreign investors, the main risk of LC bonds is that emerging market currencies depreciate when returns on global equities are low,” so that “once currency risk is hedged using cross currency swaps, LC debt appears to be much less risky than FC debt in the sense that it has significantly lower loadings on global equity returns than FC debt” (p. 20).

Partly – the paper’s calibration shows a 34 percent expected depreciation upon default would reconcile the average gap with equal credit risk, but it argues this “quanto adjustment” cannot explain the gap’s variation over time or across countries (Sections 4.1.1-4.1.3, pp. 22-28). The LC credit spread measures expected default loss as a fraction of local-currency face value under the local investor’s numeraire; converting to the dollar numeraire introduces a covariance term between the exchange rate and default. If the currency depreciates upon default, the dollar investor holding a swapped LC bond is over-hedged and profits from unwinding the swap against unmatched bond cash flows, and that expected profit shows up ex ante as a lower LC credit spread. With a mean LC-over-FC differential of -67 basis points and a mean FC credit spread of about 1.97 percentage points, “if the level difference between the two credit spreads were entirely driven by the quanto adjustment … then our data would imply 34 percent (0.67/1.97) expected depreciation upon default” – close to the estimate the authors back out independently from Turkey’s local-currency quanto CDS (p. 26). Crucially, the authors then argue against over-reading this: the implied expected depreciation would have to have a standard deviation of about 56 percent to match the observed time variation, against about 2 percent in the Turkish quanto CDS data; and across countries the implied depreciation is negatively rather than positively related to the observed correlation between changes in credit spreads and the exchange rate, the opposite of what the mechanism predicts (Section 4.1.3, pp. 27-28). And because the calibration only establishes what depreciation would equalize expected losses, the authors explicitly refuse the stronger inference: “even if the LC credit spreads are consistently lower than FC credit spreads, we cannot conclude that a lower default probability on LC debt than on FC debt from a dollar investor’s perspective” (p. 26).

No – it states explicitly that the direction is ambiguous a priori (Section 4.1.4, pp. 28). Sovereigns can always print the currency the LC bond is denominated in, but inflation is costly and “a government might find it preferable to explicitly default rather than tolerate the necessary money-printing.” Incentives also cut both ways on the holder side: FC debt is mainly held by global investors and LC bonds mainly by local pension funds and commercial banks, so a government might favour its citizens and default on foreign obligations, or might protect its reputation with international creditors and access to global capital markets. “These opposing forces make it unclear whether we should expect higher default probabilities on LC or FC debt” (p. 28). Section 4.1.5 adds risks that are specific to LC debt and unhedgeable with offshore swaps – changing taxation, regulation and custody, a more uncertain domestic bankruptcy procedure, and convertibility risk where a government blocks repatriation without technically defaulting – but notes these push the wrong way to be the main explanation, since “they would make the LC credit spreads higher than FC credit spreads across our sample countries” (p. 29).

The swapped LC bond, because it is long the more liquid LC cash bond and short the less liquid long-dated swap (Section 4.2, pp. 29-30). Mean bid-ask spreads are 28.6 basis points on LC bonds, 45.3 on FC bonds, and 39.4 on the currency swaps. From the EMTA quarterly Debt Trading Volume Survey of roughly 60 offshore institutions, mean quarterly volume is $48 billion for LC bonds against $19 billion for FC bonds; because the LC market is about five times larger, offshore turnover is lower for LC debt (30 percent versus 61 percent), but the survey misses onshore trading by local investors, and the authors note foreign holdings average about 15 percent of LC debt outstanding in the sample, so “if local investors traded 18 percent as frequently as foreigners, the total turnover ratios for LC and FC debt would be same” (p. 30). The authors are candid that swap volumes are not directly observed, resting on “conversations with traders and anecdotal evidence” that trading in long-dated currency hedges is relatively low.

A no-arbitrage model of partially segmented markets in which global risk aversion passes fully into the FC credit spread but only incompletely into the LC credit spread (Section 4.3, pp. 30-32, with the model in Appendix C). The model combines a reduced-form Duffie-Singleton specification for credit-event arrival rates driven by a local and a global factor with a preferred-habitat structure in the tradition of Vayanos and Vila (2009) and Greenwood and Vayanos (2010), and has three ingredients: FC bonds are priced by risk-averse diversified global investors whose stochastic discount factor depends only on the global factor; LC bonds face a distinct, downward-sloping local clientele demand, since domestic pension funds, insurers and banks “are often required by law to hold a large fraction of their portfolios in LC treasury bonds”; and a single risk-averse credit arbitrageur partially integrates the two markets by equalizing the price of risk across them net of the onshore-offshore wedge. The equilibrium LC credit spread is then an outcome of the arbitrageur’s optimal position size, which depends on the arbitrageur’s risk aversion, the correlation of returns across the two assets, and the size and elasticity of local demand. The paper’s framing here is explanatory rather than tested-structural: it is offered as a mechanism consistent with the documented correlations, and its testable prediction is taken up in Section 5.3.

Global risk aversion and liquidity, not macroeconomic fundamentals (Section 5.1, pp. 32-35). The benchmark panel regression has country fixed effects, the VIX as the proxy for global risk aversion, bid-ask spreads on LC bonds, FC bonds and swaps as liquidity proxies, and a battery of controls for cash-flow fundamentals: the Chicago Fed National Activity Index for global conditions, 30-day rolling realized volatility of local MSCI equity returns, and country-specific FC and LC debt/GDP ratios, inflation level and volatility, changes in terms of trade, and monthly changes in FX reserves. Estimation is monthly, following Driscoll and Kraay (1998), with Newey-West standard errors at 12 lags and clustering by month. A one-percentage-point rise in expected 30-day S&P 500 volatility is associated with an 8 basis point rise in the LC credit spread, a 19 basis point rise in the FC credit spread, and hence an 11 basis point fall in the differential – the coefficient on the FC spread is three times that on the LC spread. A one standard deviation rise in the VIX above its mean cuts the differential by 33 basis points, and the post-Lehman VIX spike, a 3.5 standard deviation move, corresponds to roughly a 115 basis point move in the differential, controlling for bond and swap illiquidity and for deteriorating local and global fundamentals. On explained variation, the VIX and liquidity alone deliver a within-R-squared of 29.4 percent for the LC spread, 71 percent for the FC spread, and 46.7 percent for the differential; adding them on top of the macro controls raises the differential’s R-squared from 29.3 to 56.1 percent, so the paper reports that the VIX and liquidity “account for 86 percent of total explained variations in the credit spread differential” (p. 35; the Introduction gives this share as 83 percent). The authors are careful about what this does and does not say: “This is not to say that macroeconomic fundamentals do not explain a significant portion of the level of the respective credit spreads. Rather, our results emphasizes their weaker ability to explain the within-country time-variation of credit spread measures” (p. 35).

Yes – the VIX and liquidity forecast FC over swapped-LC excess returns while macro fundamentals do not (Section 5.2, pp. 35-37). A high VIX predicts higher FC excess returns than swapped LC excess returns, and so a positive return on the strategy of going long FC bonds and short swapped LC bonds; conditional on fundamentals, a one standard deviation rise in the VIX above its mean forecasts a positive 3.7 percent annualized excess return on that trade. Global fundamentals proxied by the CFNAI do not forecast swapped LC or LC excess returns once the VIX is controlled, which the authors take as evidence that “it is unlikely that the predictive power of VIX is due to its correlation with unobserved macroeconomic fundamentals” (p. 36). The liquidity proxies are the strongest predictors of the FC-over-swapped-LC return: the predictable R-squared rises from 4 percent with the VIX and CFNAI alone to 20 percent once the three liquidity factors are added, and adding all remaining macro fundamentals lifts it only a further 3 percentage points. The signs line up with the position’s construction – illiquidity of LC bonds positively forecasts swapped LC excess returns while illiquidity of FC bonds and swaps forecasts them negatively, “consistent with the fact that investors holding swapped LC bonds go long in the liquidity risk in the cash market and go short offshore liquidity risk in the swap market” (p. 37).

Yes, and it is the paper’s sharpest supporting evidence: countries where the two bonds’ returns are more correlated are exactly the countries where global risk aversion passes through more fully into the LC spread (Section 5.3, pp. 37-38). Relaxing the assumption of a common pass-through coefficient, the authors estimate country-specific VIX loadings for LC and FC spreads from interactions of country dummies with the VIX, then compare the country-by-country ratio of the two loadings against the correlation of swapped-LC and FC excess returns. The model predicts the ratio should rise with the return correlation, and the two columns correlate at 84 percent – “Differential sensitivities to VIX explain the bulk of the cross-sectional variations in excess return correlations,” a result the authors report is robust to excluding the crisis period (p. 38). This is the mechanism’s distinguishing prediction because arbitrage that equalizes pricing across the two markets is only effective to the extent the two positions co-move; where they do not, the arbitrageur takes a smaller position and the local clientele’s demand keeps the LC spread insulated from global risk shocks.

That the standard assumption of a government borrowing only from foreigners in real or foreign-currency debt has become the wrong default (Introduction §1.1, pp. 7-8; Conclusion, pp. 38-39). The authors place their facts against the quantitative sovereign default literature building on Aguiar and Gopinath (2006) and Arellano (2008), and against a newer theoretical literature asking when a sovereign would explicitly default on own-currency debt (Aguiar et al. 2013, Araujo et al. 2013, Corsetti and Dedola 2013, Da-Rocha et al. 2012, Jeanne 2012), which they say “are missing an empirical measure of the credit risk on these local currency bonds.” They also frame the contrast that opens the conclusion: “While European countries find themselves borrowing in a currency they cannot print, major emerging markets sovereigns are increasingly borrowing in their own currency” (p. 38). The summary claim is carefully bounded to the measurement: LC nominal bonds “are not default free, despite of the sovereign’s option to inflate away the debt,” and once currency risk is hedged to first order, LC bonds are safer than FC bonds specifically “in terms of the correlations between asset returns and global risk factors” (p. 39) – not safer in some unconditional sense.