Sovereign Bonds Since Waterloo

It asks how creditors have fared, not how borrowers behave. The framing is historical: the authors treat the Battle of Waterloo in 1815 as “the birthday of modern sovereign debt markets - and of their recurring boom-bust cycles,” noting that the Napoleonic Wars accelerated the independence of a dozen Latin American republics that quickly sought financing in London, and that the resulting first emerging-market debt boom “ended abruptly in the financial panic of 1825.” The question that follows is stated directly: “Given the frequent defaults and limited enforcement of external sovereign debt, why are investors attracted to this asset class?” The paper lists three departures from the existing literature. First, perspective: “we take a different perspective - that of an investor,” where “the bulk of the existing work takes the borrowing countries’ perspective, often focusing on the determinants and costs of default.” Second, coverage in time and geography, against earlier work on “short samples or a limited number of countries.” Third, granularity: they trace “the financial history of more than 1,500 individual bonds at a monthly level,” combining prices, coupon payments, and default haircuts.

A 200-year monthly bond price panel and a 200-year default/haircut archive, merged at the bond level. The pricing sample covers 1,552 foreign-currency sovereign bonds issued by 91 countries, with 266,134 monthly price observations; the sample is restricted to bonds issued by central governments in USD or GBP, traded on the London or New York exchanges, with at least one year of maturity and a fixed coupon. External debt here is “defined… by currency rather than jurisdiction of issuance or residency of ownership,” and the authors stress that “sample selection for our pricing data is not dictated by any priors other than location” – they start with today’s emerging markets and work backwards, adding sovereigns that tapped London and New York in the past, “including many of today’s advanced countries such as Australia, Canada, Germany, Greece, Italy, Japan, Portugal, and Spain.” Sources differ by era: the Money Market Review, The Economist, Circular to Bankers, Course of the Exchange and Banker’s Magazine before 1870; Goetzmann and Rouwenhorst’s digitised Investor’s Monthly Manual data for 1870-1930; The Economist and Financial Times for London 1930-1980; and, newly coded here, NYSE quotations from the Commercial Financial Chronicle and the Bank and Quotation Record, since New York “became the main trading platform for foreign sovereigns after 1914.” For the modern period they use the bond-level microdata underlying JP Morgan’s EMBI Global rather than the off-the-shelf country indices, dropping public-company and sub-sovereign issues and all non-USD/GBP instruments, leaving more than 600 bonds. The default archive is “a census of all distressed sovereign debt restructurings with foreign commercial creditors from 1815 to 1980,” spliced to the Cruces-Trebesch (1978-2013) and Fang-Schumacher-Trebesch samples, built primarily from the annual reports of the British Corporation of Foreign Bondholders, the US Foreign Bondholders Protective Council, and the French Association Nationale des Porteurs Français de Valeurs Mobilières, cross-checked against Fenn’s Compendium, Fortune’s Epitome, Kimber’s Records, Moody’s Manuals and the London Stock Exchange Yearbooks. A third, supplementary dataset gives monthly bond-level bid-ask spreads for 62% of the pricing sample (165,638 observations).

313 events in 91 countries between 1815 and 2015, selected on four criteria. The criteria follow Cruces and Trebesch: only distressed restructurings, defined as exchanges of debt at a loss; only external sovereign debt, meaning central-government bonds or loans owed to private foreign creditors, excluding sub-sovereign bonds (“as these are a separate asset class”), private-to-private debt, official bilateral and multilateral debt, and domestic-currency sovereign debt; only medium- and long-term debt, excluding short-term rollovers and bridge financing; and only finalised deals, excluding agreements never actually implemented. The authors note the count is a lower bound because they collapse multiple restructurings arising from the same default into one event – 358 individual restructurings become 313 cases, weighted by restructuring amounts in USD – “so that each default receives just one haircut estimate.” Of the 91 defaulters, 68 also have monthly bond price series; price data were also collected for 23 further countries that never defaulted.

As one minus the ratio of the present value of the new debt to the present value of the old defaulted debt, discounted at the same exit yield. The formula follows Sturzenegger and Zettelmeyer and is applied identically across the whole 200-year span. What makes the same discount rate on both legs the right choice is the holdout thought experiment the authors spell out: the measure “compares the present value of the new and the old debt in a hypothetical scenario in which the sovereign keeps servicing any remaining outstanding old debts on an equal basis as the newly issued debt” – imagine “a small holdout creditor who avoided a haircut and whose old, non-exchanged bonds continue to be repaid as if no default happened (akin to what happened to the €6 billion holdouts on English law bonds in Greece in 2012).” Because “both old and new bonds face the risk of another default in the future and they both benefit from the debt relief effect of the restructuring,” a common rate is required for the comparison to be meaningful. Haircuts are computed bond by bond – using information on 1,134 defaulted bonds – then aggregated to an event-level weighted average using amounts outstanding, because “because of the heterogeneous nature of the debt renegotiations, it is not possible to simplify the calculations by relying on a ‘representative bond.’” The authors report robustness to alternative discount rates (a flat 10%, and a risk-free lower bound using UK/US long-term yields) and alternative haircut definitions (face-value and “market” haircuts), concluding that “the haircut formula and the choice of the discount rate matter, in particular for the estimated means, but the overall picture and the dispersion of haircuts across space and time is similar, irrespective of the method used.”

Losses are large, highly dispersed, remarkably stable across two centuries, and almost never total. Average haircut in the full 313-event sample is 44%, median 39%, standard deviation 30%, range -14% to 100%. Splitting by era: the historical sample (defaults pre-1970, all bond restructurings) averages 51% with a median of 48%; the modern sample (post-1970, including 152 bank-debt and 23 bond defaults) averages 39% with a median of 34%; the 23 recent bond restructurings average 37%. Weighting by amount restructured pulls the full-sample mean down to 39%, because “for the poorest countries, where haircuts tend to be deeper, the amounts of debt involved are usually much lower.” Face-value haircuts average only 24% with a median of 0. Across time the authors find “strong recurring features”: “the level of creditor losses averaged between 40% and 50% - with no visible time trend or outlier spells,” and every decade since 1815 featured a few restructurings, the sole major exception being the Bretton Woods period from WW2 to the 1970s, when closed capital accounts meant “barely any new defaults or restructurings on privately held sovereign debt occurred.” Ten events have negative haircuts, mostly early-stage crisis deals that lengthened maturities at higher interest rates than before – deals that “do not imply debt relief, but may nevertheless be beneficial for the government, at least in the short term.”

In revolutions, regime repudiations, and the dissolution of countries or empires. The authors name three full cancellations: Lenin’s cancellation of all external debts after the 1917 Communist revolution, the Communist takeover of China in 1949 (China’s bonds had been in default since 1939, but were declared cancelled and void only after Mao came to power), and Cuba in 1960 after the Castro revolution – and note that “explicit debt cancelations only occurred in China, Cuba, and Russia” among Communist takeovers. They identify five selective cancellations by new governments refusing predecessors’ debts: Spain 1824 (bonds of the Cortes of Cádiz), Greece after 1826 (bonds raised by the independence militias), Portugal 1834 (bonds of Dom Miguel), Mexico 1865 (Benito Juárez, on bonds issued by Maximilian I), and the Dominican Republic 1872. “In most repudiation cases, the debts remain in default until today or were in default for more than a generation”; Spain and the Dominican Republic are the exceptions, settling after 10 and 16 years at haircuts of 40% and 95%. Break-ups are the other source of near-total losses: Austria-Hungary’s defaulted debt “was only settled in the 1970s with an average haircut of 98%,” and the Baltic states’ bonds “were fully canceled after the Soviet occupation in August 1940.” In the modern period, 100% haircuts appear only for a small number of HIPCs that defaulted in the 1980s and took nearly thirty years to settle. The conclusion the authors draw is methodological as much as historical: “Default is not a binary (0,1) process, as usually modeled in the related literature.”

Monthly total return = price change plus accrued coupon, with missed coupons and restructuring outcomes coded from the authors’ own archive. Coupons are treated as accrued interest spread evenly over the payment period. In restructuring months, the return combines the old defaulted bond with the new instrument received, “with the implicit assumption that creditors keep the newly restructured bond in the portfolio,” then accounts for any face-value write-off and any cash payments including arrears settlement; a pure maturity reprofiling has zero write-off and shows up in the new bond’s secondary-market price instead. This is why the NPV haircut methodology is not decisive for the return series – “the haircut estimate is a snapshot at one point in time, while returns are measured continuously on a monthly level” – and why “the key default-related variable that matters for the returns is missed coupons.” Real returns deflate GBP bonds by the Bank of England’s historical inflation index and USD bonds by the BLS series. Global portfolio returns are value-weighted monthly averages across all actively traded bonds, compounded to annual, then averaged arithmetically (the benchmark in prior long-run-returns work) with geometric means also reported. In years when the portfolio mixes GBP and USD bonds, returns “enter without converting them into a common currency to avoid bias,” which the authors are explicit about interpreting as the perspective of “an investor who holds all outstanding foreign-currency bonds… and who is hedged against currency fluctuations between these two currencies.” Excess returns are computed bond by bond against total return series on long-term UK gilts or US Treasuries in the matching currency; transaction fees and taxes are not deducted.

6.85% real arithmetic average for 1815-2016, with an excess return of 4.29 percentage points over UK/US government bonds and a Sharpe ratio of 0.32. Nominal returns average 7.99%, geometric real return 5.78%, standard deviation 15.03%. Dropping the two world wars raises the real mean slightly to 6.99%. By era: 1815-1869, 7.10% real but volatile (SD 17.28); 1870-1914, 6.28% real with strikingly low volatility (SD 8.01) and a Sharpe ratio of 0.48; 1915-1945, 6.01% arithmetic but only 4.29% geometric with SD 19.91, “the worst risk-return ratio”; 1946-1973, 5.85%, the lowest, “mostly due to the fact that many bonds that went into default in the 1930s continued to be nonperforming for decades,” with 41 restructurings needed to settle the 1930s and 1940s defaults and with the averages “biased downward due to selection effects”; and 1995-2016, 9.89% real, above the historical average. The authors drop 1974-1994 entirely “due to a lack of representative bond pricing data,” because the syndicated-bank-loan era left “fewer than 10” countries with actively traded bonds in the 1980s – which means the sample “omits the encompassing debt crisis in emerging and developing countries in the 1980s,” a period when the fragmentary evidence suggests returns “were well below the historical average.” A survivorship check restricting to the 15 countries with more than 100 years of data each gives returns “similar to our baseline numbers.”

Coupons, not capital gains. “Around 70%, or 5.6 percentage points, of the nominal yearly return of 8.0% over the past 200 years is due to coupon payments,” and coupons are the dominant driver in every decade, contributing “roughly… between 4 and 8 percentage points to the ex-post nominal returns.” The average nominal coupon across the whole pricing sample is 5.8%, rising to 7.0% in the modern 1989-2016 sub-sample. This is what allows returns to hold up through default spells, since “sovereigns in default on principal payments often continue to service coupons in full or in part, which pushes up investor returns.” The authors also note, with a light touch, that the mechanism is visible in current markets: “In the current environment of near zero interest rates in advanced economies, coupons on many of the serial defaulters are in the 6-10 percent range.”

Yes, and the ordering is consistent across eras. Splitting the 91 countries into 51 serial defaulters (two or more external defaults since 1815 or independence, or a share of years in default above the 20% sample median) and 40 others, serial defaulters earn 7.1% real against 5.6% for other periphery sovereigns and 2.9% for spliced UK/US government bonds, with excess returns of 4.6 and 3.4 percentage points respectively. The same ordering holds for volatility: serial defaulters are most volatile in every era (SD 16.8 in the full sample against 10.4 and 9.3). The one exception to the return ordering is the interwar period, “when bonds of center countries perform better” – serial defaulters return 6.8% against 10.5% for other periphery issuers and 6.5% for US Treasuries, with a negative excess return of -2.8 points. Country by country, among issuers with at least ten years of price data, “not a single country in Table 6 shows a negative arithmetic return, on average, and only two countries have negative excess returns (Bolivia and China),” while repeat defaulters like Argentina, Brazil, Ecuador, Greece, Mexico, Ukraine and Venezuela show “long-run excess returns between 4% and 12%.” Returns also rise with the raw count of past defaults, though the authors report that relationship with a t-statistic of 1.87 and p-value of 0.07, i.e. marginal significance; the country-level mean-return-versus-volatility relation is stronger (t = 5.37, p = 0.00). A CAPM exercise in the appendix finds serial defaulters have higher betas because their returns are more sensitive to US/UK equity market conditions, while the asset class’s beta overall is low – “well below 0.5” historically – “suggesting that external bonds provided a diversification service with respect to US/UK equities.”

A drop of about 15%, a multi-year flat stretch, and on average a break-even four years after the default – with a wide spread. Using the 92 default episodes (out of 161 sovereign bond defaults) with sufficient price data on both sides, and indexing cumulative real total returns to 24 months before default, “the total cumulative return drops by about 15% initially and then stagnates for a few years.” Investors who entered two years pre-default “break even four years after the initial default date, on average, thus recouping the losses suffered with delay (this can be described as a U-shaped recovery)”; on geometric averaging the break-even comes about five years out. But “the variation is large”: the upper quartile of episodes sees “barely any drop in total returns” and investors are up 50% five years after default, while the lower quartile shows an L-shape in which “six years after the first default, investors are still in negative territory, far from breaking even.” Almost all bottom-quartile episodes are pre-WW2. Since the 1990s only Argentina’s 2001 and Ecuador’s 2008 defaults produced long-lasting losses: “it took investors until 2016 to break even in Argentina (15 years) and about five years after Ecuador’s 2008 default.” Splitting the 92 cases at the median haircut of 42%, low-haircut defaults recoup losses “within three years after the initial default” while high-haircut defaults take “more than six years, on average.” The paper also reports the two-sided nature of the summary statistics: in default years, bond-level real returns average 2.37% (median 1.98%) against 6.05% (median 4.85%) in non-default years, and the average bond price in default is 35 against 89 outside default.

Losses are an order of magnitude smaller than returns, per year. Averaging across years rather than across episodes: in the historical bond period (1815-1973) the global portfolio returned 6.5% real per year against “an average investor loss due to bond restructuring events (haircuts) of just 1.2% across years”; in the modern period (1995-2016) the figures are 9.9% and 0.8%. The authors’ summary is that “over the past 200 years, returns on bonds in normal times offset losses during debt restructurings by a wide margin.” (This calculation counts only bond restructurings, dropping the many 1980s and 1990s bank-loan restructurings.)

It beats almost everything except US and advanced-economy equities, and matches US equities on risk-adjusted return. Over the full 1815-2016 sample the external sovereign bond portfolio returns 6.85% real against 8.49% for US equities, 8.25% for the 16-country advanced-economy equity portfolio, 5.47% for UK equities, 4.25% for US Treasuries, 2.94% for UK government bonds, 3.15% for a 16-country domestic sovereign bond portfolio, and -1.09% real for US AAA corporate bonds (from 1900). “In the full sample, the return on external sovereign bonds is significantly higher than that of US corporate bonds, UK equities, US or UK government bonds and that of the portfolio of domestic sovereign bonds from 16 countries.” Its Sharpe ratio against bills, 0.35, is “on the same level as US equities, and exceeding that of the other asset classes.” In the modern 1995-2016 sample it does better still – 9.89% real return and the highest Sharpe ratio (0.66) of any asset in the comparison, beating US equities (8.40%), UK equities (6.38%), US corporate bonds (3.29%), the advanced-economy equity portfolio (9.17%) and domestic sovereign bonds (5.23%). The authors attribute this to “high average coupon rates during the modern period coupled with the paucity of serious credit events,” and then immediately qualify it: “this does not imply that this benign combination is set to become the ’new normal’ for the sovereign bond market, as rising debt difficulties in developing countries can morph into defaults abruptly.”

Liquidity collapses in global shocks and never fully recovered after WW2, and less liquid bonds earn more. Using bid-ask spreads for 62% of the pricing sample – proxied for historical London bonds by the gap between the “business done” and “closing price,” following Alquist and Chavaz and Flandreau – the authors report two findings. First, “average bond liquidity declines markedly during global wars and financial crises, with bid-ask spreads spiking in the initial months of the shock,” and “WW2 had a particularly lasting impact, as market liquidity never fully recovered after 1945, with average bid-ask spreads remaining at a level far above those in the 1920s”; idiosyncratic crises (the Ottoman and Egyptian defaults of the 1870s, Russia 1998, Argentina in the 1890s and early 2000s) also moved aggregate liquidity. Second, “total returns tend to be higher when bid-ask spreads are high, except for the first months of major shocks, when returns collapse while bid-ask spreads spike,” and at the bond level “average returns are significantly higher for less liquid bonds,” consistent with the sovereign bond liquidity premium found by Alquist and by Chavaz and Flandreau. The authors leave “an in-depth analysis of sovereign bond market liquidity over the full 200 years… for future research.”

That the workhorse risk-neutral-lender models cannot produce these numbers, and that the puzzle of post-default market re-entry is largely resolved. On theory: “Our results are hard to reconcile with seminal quantitative models of sovereign default,” which assumed risk-neutral investors and spreads reflecting only expected default losses, because “when investors are risk-neutral, excess return above ‘risk-free’ bonds should be zero in expectation” – yet real ex-post excess returns run 2% to 4% for the full global sample. “Thus, investors typically receive a compensatory premium for holding sovereign risk that exceeds historical credit losses,” which the authors offer as “support to a growing body of quantitative work that assumes risk-averse (or uncertainty averse) creditors.” On re-entry: the results “go a long way in solving a puzzle that has preoccupied the literature for decades - namely, why sovereigns can borrow again despite a history of default,” the illustrations being Argentina re-accessing markets “only months after exiting its seventh default” in 2016, including a 100-year bond, and the African issuance boom by formerly HIPC countries such as Ghana and Zambia. The mechanism is the high return-to-risk ratio: it “helps our understanding of why sovereign debtors can undergo repeated cycles of over-borrowing, often followed by default and a subsequent market re-entry (i.e., serial default).” The paper also parallels the equity premium literature explicitly, noting that its own finding of “high excess returns coupled with a relatively low return volatility” is the same shape of puzzle Mehra and Prescott posed for stocks, but over far more countries and far further back.

Expectations formation, and the borrower’s side of the ledger. The conclusion points to “a growing literature emphasiz[ing] expectations errors on the part of global investors,” under which “during periods of optimism and financial stability, creditors may become dismissive of the possibility of default and expect full repayment of the rich coupons offered in this asset class, resulting in lending booms” – the “market exuberance” pattern central to Reinhart and Rogoff and formalised by Gennaioli and Shleifer – and suggests the new data can test it, along with long-run-risk and disaster-risk asset pricing models. The second agenda is explicitly the mirror image of the paper’s own perspective: “we see the need to reexamine our results from the vantage point of debtor countries and with a view to debt sustainability,” posing three questions – what the appropriate scope of debt relief is, how serial restructurings and resolution delays can be reduced, and “importantly, what motivates countries to pay the large observed premia to their foreign creditors?”