Emerging Market Spreads
Then Versus Now
Author: Yishay Yafeh1, Mr. Paolo Mauro, and Nathan Sussman2
  • 1 0000000404811396https://isni.org/isni/0000000404811396International Monetary Fund
  • | 2 0000000404811396https://isni.org/isni/0000000404811396International Monetary Fund

Contributor Notes

Authors’ E-Mail Addresses: pmauro@imf.org, msussman@mscc.huji.ac.il, msyafeh@mscc.huji.ac.il
Publication Date:
01 Nov 2000
eISBN:
9781451859652
Language:
English
Keywords:
WP; emerging market; emerging market country; gold standard; vis-à-vis yield; number of month; emerging market crisis; primary market yield; Brady issue
Download PDF (1.1 MB)

This paper analyzes yield spreads on sovereign debt issued by emerging markets using modern data from the 1990s and newly-collected historical data on debt traded in London during 1870–1913, a previous “golden era” for international capital market integration. Applying several empirical approaches, we show that the co-movement of spreads across emerging markets is higher today than it was in the historical sample. We also show that sharp changes in spreads today tend to be mostly related to global events, whereas country-specific events played a bigger role in 1870–1913. Although we find some evidence that economic fundamentals, too, co-move more strongly today than at that earlier time, our interpretation of the results is that today’s investors pay less attention to country-specific events than their predecessors did in 1870–1913.

Abstract

This paper analyzes yield spreads on sovereign debt issued by emerging markets using modern data from the 1990s and newly-collected historical data on debt traded in London during 1870–1913, a previous “golden era” for international capital market integration. Applying several empirical approaches, we show that the co-movement of spreads across emerging markets is higher today than it was in the historical sample. We also show that sharp changes in spreads today tend to be mostly related to global events, whereas country-specific events played a bigger role in 1870–1913. Although we find some evidence that economic fundamentals, too, co-move more strongly today than at that earlier time, our interpretation of the results is that today’s investors pay less attention to country-specific events than their predecessors did in 1870–1913.

I. Introduction

The frequency and virulence of financial crises that affected emerging markets in the second half of the 1990s have led to calls for reform of the current international financial architecture. Many observers have also wondered whether globalization in international financial markets, perhaps owing to informational and technological advances, has gone too far. The 1990s were characterized by large and volatile private international capital flows toward emerging market countries and, for the first time after several decades, large amounts of sovereign bonds were issued by emerging market countries and actively traded on secondary markets. This paper seeks to shed light on today’s international financial environment by comparing it with that of 1870-1913, a previous “golden age” for emerging market bonds and international capital flows toward “emerging markets.” Our focus is on sovereign bond yield spreads and on comparing the nature of financial crises and the degree of financial integration in emerging markets, “then” (1870-1913) versus “now” (1992-2000).

There is a growing consensus that global economic integration reached a peak in the late nineteenth and early twentieth century, collapsed with the world wars and the intervening great depression, and gradually increased again after the collapse of the Bretton Woods system to attain levels similar to pre-1914 in the 1990s (Sachs and Warner, 1995). O’Rourke and Williamson (1998) show that capital outflows from Britain to contemporary developing economies were extremely high, and barriers to movement of capital (and labor) were virtually absent. Bordo et al. (1998) describe the period between 1870 and World War I as an era of global finance in which large amounts of foreign securities were actively traded in England. Obstfeld and Taylor (1998) argue that only in the 1990s did financial integration return to the levels experienced in the era of the classical gold standard.2

Our main contribution is to analyze a newly-collected data set on monthly observations of secondary market yields on sovereign bonds denominated in British pounds and traded in London during 1870-1913, issued by the “emerging markets” of the day, namely Argentina, Brazil, China, Egypt, Japan, Portugal, Queensland,3 Russia, Sweden, and Turkey.4 We compare the characteristics of those data with a variety of similar data sets on emerging market spreads today. For this purpose, the best modern data set on secondary market sovereign bond yield spreads is that on Brady bonds—the largest and most liquid emerging debt market during 1992-2000.5 Brady bonds are denominated in U.S. dollars and spreads are computed vis-à-vis yields on U.S. government long-term bonds—just as spreads are computed vis-à-vis British consols for the historical data.

A number of institutional differences in the markets for which data are available “then” and “now” imply that the comparison cannot be perfect. In particular, all countries issuing Brady bonds have previously defaulted on (or restructured) loans from foreign commercial banks, whereas only some countries in the “emerging markets then” group had defaulted prior to 1870. Conversely, some of the historical emerging market countries defaulted on the bonds we are considering during 1870-1913, whereas no country defaulted on (or restructured) Brady bonds, with the exception of Ecuador in the aftermath of its late 1999 crisis. These differences need to be kept in mind in interpreting the results.

Although it might be argued that the 1990s have been “special” in some sense, perhaps because we all tend to remember more recent crises better, we feel that the 1990s are indeed representative of the “today” that we are interested in, for three reasons. First, it is events in the 1990s that have generated calls for reform of the international financial architecture. Second, the 1990s are the first time since the first world war to see the return of large private capital flows toward emerging markets in the form of bonds; under this strict definition of “emerging markets,” there were no emerging markets between the first world war and the 1990s. Third, and perhaps most important, crises have not occurred more frequently in the 1990s than in the 1980s or 1970s, as shown by Bordo and Eichengreen (2000) with respect to currency and banking crises.6

In comparing 1870–1913 to 1992–2000, we address the following questions. How frequent are crises and sudden improvements in emerging market spreads? To what extent do sharp changes in spreads tend to affect more than one emerging market at a time? How large is the common component in the variation of all emerging market spreads? To what extent do investors benefit from holding a portfolio of bonds issued by several emerging markets rather than by only one emerging market? What kinds of events trigger changes in spreads? Do crises reflect news about macroeconomic developments, political events, or reforms?

To analyze these issues, we adopt a variety of approaches. We consider the number of sharp changes in spreads defined in a number of ways. We then compute the proportion of these changes that affect more than one country at a time. To assess the extent to which the variation in emerging markets’ sovereign bond yield spreads is accounted for by a common component, we use principal components analysis. This approach is similar to that adopted by other studies (for example, Nellis, 1982), which use the extent of interest rate variation that is explained by the first principal component to gauge the extent of international financial integration. To analyze the nature of events that cause major changes in spreads, we search the spread series for “structural breaks,” and systematically relate the breaks to significant events. We supplement these results by briefly considering two case studies, one relating Japanese spreads to events in China using monthly data for 1870–1913, and one relating Korea’s spreads to events in other Asian countries using daily information for 1996–98.

The main conclusions of our empirical analysis are the following:

  • For the typical emerging market, financial crises with associated sharp increases in sovereign spreads were common in 1870–1913, though far less common than in the 1990s. Even less common in 1870–1913 were truly “global” crises with increases in sovereign spreads in almost all of the emerging markets, whereas these seem to have been the norm since the early 1990s.

  • The proportion of the variance in emerging market spreads accounted for by the first principal component was about ½ in the historical sample and about ¾ in the 1990s.

  • The diversification benefits from investing in several emerging markets rather than in only one emerging market are lower today than they were in the past.

  • In the historical sample period, most structural breaks in the spread series are related to country-specific (mainly political) events, whereas in the modern sample period most of the breaks are related to global crises.

Increased co-movement of spreads “now” compared with “then” may be due to higher co-movement of economic fundamentals, or different patterns of investor behavior. While providing new evidence that, indeed, economic fundamentals co-vary to a greater extent today than they did in the past, we argue that changes in investor behavior are also an important factor underlying our results.

The remainder of the paper is organized as follows. Section II describes the data sets used for this study. Section III presents the main empirical results on emerging market spreads. Section IV discusses possible interpretations of the main results, and reports additional evidence on fundamentals. Section V concludes.

II. Data Description

A. Historical Data on Spreads (1870-1913)

Our data set consists of monthly observations on sovereign bond yields for 1870–1913. The data were collected by hand from the London Times and The Economist’s Investor’s Monthly Manual. The data are available daily, but only end-of-the-month observations were collected, owing to resource constraints. All bond coupons were payable in pounds in London. For each emerging market, yields are calculated as the ratio of interest payments to market price7 and spreads are computed as the absolute (percentage point) difference between the yields on bonds issued by the emerging market and the yields on British consols. Further detail on the bonds is provided in Appendix I

Our historical sample includes ten contemporary “emerging markets:” Argentina, Brazil, China, Egypt, Japan, Portugal, Queensland, Russia, Sweden, and Turkey. These countries represent some of the major borrowers on the London market and several types of “emerging markets” in terms of their geography, macroeconomic policies, and economic and institutional structure.8

Our definition of “emerging markets” is similar to that adopted by Bordo and Eichengreen (2000, henceforth BE), who classify countries as “emerging markets”—following modern parlance—using relative per capita incomes and especially on the basis of whether they were net recipients of capital inflows. We apply that definition somewhat more stringently, in the sense that we do not include in our sample some countries, e.g., the United States, that BE classify as emerging markets. By contrast, all countries in our sample that are also considered by BE are classified by BE as emerging markets rather than industrial countries. Countries included in our sample that are not considered by BE (China, Egypt, Russia, and Turkey) are clearly emerging markets using BE’s criteria.

The London market for foreign government bonds during this period was very active and liquid. The total market value of government bonds traded in London was £3 billion in 1875 and £4 billion in 1905, of which £0.5 billion in 1875 and £1 billion in 1905 (or 45 percent of Britain’s GDP in 1875 and 55 percent of Britain’s GDP in 1905) issued by emerging markets in our sample.9 Table 1 (which we compiled from The Economist’s Investor’s Monthly Manual) reports the total market value of bonds traded in London by issuing country. Our sample includes the larger emerging markets of the day; it excludes the advanced countries in the industrial core of Europe and the smaller emerging markets.10 Complementary evidence from an alternative source (Suzuki, 1994) on bond issuance activity on the London market by the emerging markets in our sample is presented in Appendix II.

Table 1.

Market Value of all Government Bonds Traded in London, 1875 and 1905.

article image
Data Source: The Economist’s Investor Monthly Manual

Asterisks denote countries included in our sample of “emerging markets” for 1870-1913.

“Other” includes Antigua. Barbados, Bolivia, British Columbia. British Guyana, Ceylon, Colombia, Costa Rica, Danubian Principalities, Gold Coasi, Grenada, Guatemala, Honduras, Hong Kong, Jamaica, Liberia, Mauritius, Moorish territories, Nicaragua, Paraguay, San Domingo, Sardinia, Serbia, Sierra Leone. St. Lucia and Trinidad.

B. Modern Data on Spreads (1992-2000)

The data on emerging market spreads on sovereign bonds denominated in U.S. dollars are drawn from J.P. Morgan and consist of the EMBI (Emerging Markets Bond Index) and EMBI+ bond yield spreads (vis-á-vis yields on U.S. long-term government bonds). EMBI and EMBI+ spreads are the most closely watched indicators of emerging market spreads by market participants, and have been widely used by researchers in previous work.

The EMBI and EMBI+ spreads are available at a daily frequency and—being secondary market spreads—at all times, including times of crisis. By contrast, primary market yields are observed with erratic frequency and are often not available in times of crisis (arguably, the most interesting times), when many countries are just unable to launch new issues.

The EMBI spreads are a weighted average of the spreads on a variety of Brady bonds issued by the country being considered. Yields on those Brady bonds that are collateralized are “stripped” yields, that is, yields after the value of the collateral has been subtracted from the value of the bond. The bonds typically have a long maturity.

Brady bonds were by far the most widespread and actively traded form of emerging market sovereign bonds in the 1990s. Although their relative importance has been declining in recent years, they still accounted for more than half of sovereign debt in the emerging markets surveyed by the Emerging Market Traders Association in 1999 (Table 2). They also accounted for a large portion of the sovereign debt issued by each of the countries considered in our sample (Table 3). The EMBI+ spreads include also a number of non-Brady issues (both sovereign and corporate bonds), but still consist mainly of Brady bonds, reflecting their relative importance in overall market capitalization and trading activity.

In modem times, there was no significant active secondary market for emerging market bonds prior to the introduction of Brady bonds in the early 1990s. Most foreign borrowing by emerging market countries took the form of bank loans. It was only following the payments difficulties experienced by a number of emerging market countries (beginning with Mexico) that bank loans were repackaged as Brady bonds and secondary market trading began on a large scale.

Table 2.

Secondary Market Transactions in Debt Instruments, Emerging Markets, 1993-19991/

article image
Source: Emerging Markets Traders Association

All emerging markets surveyed by the Emerging Markets Traders Association

Table 3.

Secondary Market Transactions in Debt Instruments, Emerging Markets, 1993-1999

article image
Source: Emerging Markets Traders Association

All emerging markets surveyed by the Emerging Markets Traders Association

Including loans, options and warrants, and local market instruments in both domestic and foreign currencies.

Breakdown by instrument not available.

As the EMBI yields are based upon Brady bonds, the sample consists of countries that have defaulted, or experienced payments difficulties, on foreign commercial bank loans in the past. Another anomaly observed on Brady bond spreads is a sizable differential between (stripped) yields on Brady bonds and other bonds issued by the same country, controlling for differences in maturity and other factors. Although a number of potential explanations have been put forward for this anomaly, none is fully satisfactory (International Monetary Fund, pp. 75-76).

Despite these drawbacks, the EMBI and EMBI+ spreads are the best available data for our purpose and permit a meaningful comparison between today’s emerging markets and those in historical times. To obtain a similar number of countries as that in the historical data set, we analyze the period since November 1994; this gives us daily (or end-of-the-month) EMBI spreads for the following eight emerging markets: Argentina, Brazil, Bulgaria, Mexico, Nigeria, the Philippines, Poland, and Venezuela. To obtain a larger sample (14 countries—Ecuador, Korea,11 Morocco, Panama, Peru, and Russia in addition to those listed above) we also analyze daily data on EMBI+ yields since April 30,1998.

C. Data on Exports in Common Currency

The data on exports in common currency used in Section IV are drawn from the following sources. For the historical sample, the data on exports in local currency are drawn from Mitchell (1998) and the data on exchange rates vis-á-vis the British pound are drawn from Schneider et al. (1991). For the modern sample, data on exports in U.S. dollars are drawn from the IMF’s International Financial Statistics. Further detail on the data sources is provided in Appendix I.

III. Emerging Market Spreads, Then and Now

A. An Informal Look at the Data

Simple inspection of the spreads over 1870–1913 and 1992–2000 (Figures 1 through 4) reveals the following:

  • The spreads are substantially higher (in basis points) in modern times than in historical times.

  • There is a great deal more common variation across countries in modern times than in historical times. Around the time of the Mexican crisis (December 1994) and the Russian crisis (August 1998) spreads rise in near-unison in most emerging markets. By contrast, in historical times there seem to be many more country-specific developments in the spreads. There are several instances in which the spread drops in a particular country around the time of an identified country-specific event. For example, the spread on Japanese bonds dropped significantly around the time when the gold standard was introduced (Sussman and Yafeh, 2000).

  • The period 1870–1913 saw times of turbulence and sharp changes in spreads, but also tranquil times. By contrast, during 1992–2000 there seems to have been considerable volatility in most countries almost all the time.

Figure 1.
Figure 1.

Historical Spreads, 1877-1913

(in basis points)

Citation: IMF Working Papers 2000, 190; 10.5089/9781451859652.001.A001

Data Source: The EconomistSpreads are yields on each emerging market’s government bonds issued in British pounds minus yields on British government bonds.
Figure 2.
Figure 2.

Historical Spreads

(in basis points)

Citation: IMF Working Papers 2000, 190; 10.5089/9781451859652.001.A001

Data Source: The EconomistSpreads are yields on each emerging market’s government bonds issued in British pounds minus yields on British government bonds.
Figure 3.
Figure 3.

Modern Spreads, 1992-2000

(in basis points)

Citation: IMF Working Papers 2000, 190; 10.5089/9781451859652.001.A001

Data Source: J.P. MorganEMRI spreads (yields on government bonds issued in US dollars by each emerging market, minus US bond yields)
Figure 4.
Figure 4.

Modern Spreads, 1992-2000

(in basis points)

Citation: IMF Working Papers 2000, 190; 10.5089/9781451859652.001.A001

Data Source: J.P. MorganEMBI spreads (yields on government bonds issued in US dollars by each emerging market, minus US bond yields)

The next sections confirm these informal observations using a variety of techniques and testing for robustness of the results.

B. Estimation and Results

Sample statistics

In absolute terms, spreads are higher today than in the historical sample. The cross-country mean of the period average spread over the modern sample is around 800 basis points, compared with less than 300 basis points in 1870-1913 (Table 4). This comparison might however be influenced by the fact that interest rates in Britain in the historical sample were much lower than interest rates in the United States in the 1990s (on average, 3.0 percent, compared with 6.8 percent, respectively). The cross-country means of each country’s mean spread over the sample period divided by the interest rate in Britain (in historical times) or in the United States (in modern times) amount to 0.815 and 1.24, respectively (omitting Turkey from the historical sample because of data problems, see Appendix I). The difference is statistically significant. Similar results hold using medians instead of means, both over the sample period in one country and cross-sectionally.

Table 4.

Spreads-Sample Statistics

(In basis points)

article image
Data Sources: The Economist and J.P. Morgan website

The standard deviation of the spreads is typically higher in the modern sample than in the historical sample. However, the cross-sectional average of the coefficient of correlation (the standard deviation divided by the mean) is roughly the same in the modern sample as it is in the historical sample.12

As already mentioned in the context of the mean spreads, an important issue is whether the spreads are affected by interest rates in the base country (Britain in the historical sample, the United States in the modern sample). The implications of this issue are pervasive: interest rates in the base country are not only higher but also more volatile in the modern sample than in the historical sample, and it might be argued that higher volatility of interest rates in the base country tends to increase both the volatility and the co-variation of spreads in emerging markets.13 This issue is unresolved from both a theoretical and, especially, an empirical standpoint. From the theoretical point of view, Kamin and von Kleist (1999) suggest that an increase in interest rates in the base country would tend to raise the absolute (percentage point) spreads.14 Increases in the base country’s interest rates may also reduce the emerging countries’ creditworthiness, reinforcing the positive effect of base country interest rates on absolute spreads, and mitigating (and possibly overturning) their negative effect on relative spreads.

From the empirical point of view, existing studies (including Kamin and von Kleist, 1999) do not find significant and robust effects of U.S. interest rates on emerging market spreads. Nevertheless, some of the estimates in the following sections control for the effect of interest rates in the base country.

Correlations

Correlation coefficients for the spreads across pairs of emerging markets are considerably higher in modern times than in historical times. The average correlation coefficient is 0.77 in modern times compared with 0.41 in historical times (Table 5). All of the correlation coefficients are positive and significant in modern times, whereas there are a number of coefficients that are close to zero (or even negative) in historical times.15

Table 5.

Correlation Matrices

article image
Data Sources: The Economist and J.P. Morgan website.

Correlations between British (or US) yields and emerging market spreads.

Common component

To gauge the extent of co-movement of these spreads among countries in different sample periods, this section computes the percentage of variation accounted for by the first principal component in the sovereign bond yield spread series for the various emerging market countries considered. The overall result is that the proportion of variance in emerging market spreads accounted for by the first principal component was about ½ in 1877–1913, and about ¾ in the 1990s. Therefore, that proportion was high in historical times, but it is significantly higher in modern times, in both a statistical and an economic sense.

In historical times (1877–1913), and omitting Turkey from the sample,16 the percentage of variation in the nine series accounted for by the first principal component is 52.0 percent—the standard error of that percentage is 2.0 percentage points.17

In modern times, the main sample considered is that of the eight emerging markets (Argentina, Brazil, Bulgaria, Mexico, Nigeria, the Philippines, Poland, and Venezuela) for which the EMBI spread data are available since November 1994. Monthly data are used for consistency with the estimates based upon historical data. The percentage of variation accounted for by the first principal component in the sovereign bond yield spread series for these eight series is 80.0 percent. (The standard error of that percentage is 3.1 percentage points.)18

Using daily EMBI+ spread series for a larger sample of fourteen countries (Argentina, Brazil, Bulgaria, Ecuador, Korea, Mexico, Morocco, Nigeria, Panama, Peru, the Philippines, Poland, Russia, and Venezuela) since April 30, 1998, the percentage of variation accounted for by the first principal component is 72.2 percent.

Appendix III reports the percentage of variation accounted for by the first principal component when one or two countries at a time are dropped from any of the samples considered above. It shows that this does not alter the result that the percentage of variation accounted for by the first principal component is higher in modern times than in historical times.

The larger common component in emerging market spreads in modern times than in historical times does not seem to be accounted for by greater variation in interest rates in the base country in modern times than in historical times. To show this, we adopt two approaches. First, we estimate the amount of variation that is accounted for by the first principal component using, for each emerging market—instead of the spreads—the logarithm of the ratio of the interest rate in the emerging market to the interest rates in the base country. (For small spreads, this is approximately equal to the ratio of the spread to the interest rate in the base country.) This method ensures that the estimation ignores all instances in which the same multiplicative change affects both the interest rate in the base country and the interest rate in the emerging market. Using this method, the amount of variation that is accounted for by the first principal component is 57.5 percent (standard error: 2.0 percentage points) in 1877–1913 excluding Turkey; 49.8 percent (standard error: 2.0 percentage points) in 1875–1913 including Turkey; and 81.0 percent (standard error: 3.0 percentage points) in 1994–2000.

Second, for each emerging market, we run a univariate regression of the emerging market yield on the interest rate in the base country (Britain in the historical sample and the United States in the modern sample), save the residuals, and then run the principal components estimation on the residuals for all of the emerging market countries. The rationale is to conduct the principal components analysis on that portion of emerging market yields that is orthogonal to the interest rate in the base country. Using this approach, the amount of variation accounted for by the first principal component is 55.8 percent (standard error: 1.9 percentage points) in 1877–1913 excluding Turkey; and 85.0 percent (standard error: 2.5 percentage points) in 1994–2000.

Beta coefficients and the benefits of portfolio diversification

Consistent with the higher common component in emerging market yields today compared with the past, we find that the benefits of holding a portfolio of bonds issued by a variety of emerging market countries rather than by only one country are smaller today than in the past. To show this, for each emerging market country we estimate a univariate regression with the ex-post return (capital gain plus coupon payments) on that country’s bonds on the left-hand side and the ex-post return on a market-weighted19 portfolio of bonds issued by all emerging market countries on the right-hand side. We find that the beta coefficients tend to be considerably closer to one in the modern sample than in the historical sample (Table 6). On average (across countries), the absolute difference between one and a country’s beta coefficient is 0.43 in the historical sample and 0.26 in the modern sample. (The standard errors for the various countries’ beta coefficients range from 0.04 to 0.14). The R2 coefficients are far larger in the modern sample than in the historical sample.

Table 6.

Beta Coefficients on Returns in Modern and Historical Samples

article image
Source: The Economist’s Investor Monthly Manual and J.P. Morgan websiteThe beta coefficients are estimated by regressing each country’s monthly bond returns on the returns on a market-weighted portfolio of bonds issued by all emerging market countries.

C. Emerging Market Crises, Then and Now

Financial crises are certainly not a new phenomenon. They occurred frequently and with severe consequences in the late 1800s and early 1900s. “Emerging market” countries often defaulted on their debts: Turkey’s default on its foreign debt in the mid-1870s was associated with an increase in sovereign spreads of a magnitude not seen since then.20 The crash of 1890 in Argentina (and Uruguay, not in our sample) led to the insolvency of Baring’s, the famous London merchant bank. Lindert and Morton (1989) provide a detailed chronology of debt defaults and reschedulings in a very large sample of countries since 1820. As for the ten countries in our sample, they show that several defaults took place—Argentina in 1830 and 1888-1893, Brazil in 1898 and 1914, Egypt in 1876, Russia in 1839, and Turkey in 1876–1881.

Nevertheless, a systematic analysis of sharp changes in sovereign spreads suggests that crises (and sudden improvements in a country’s spreads) were less frequent in 1870-1913 than in the 1990s. Specifically, in this section we compute the number of instances in which spreads changed sharply, in 1870-1913 versus 1992-2000, according to the following definitions:

  1. Proportional change in the spread: the spread rises or falls by more than 10 percent (20 percent, 30 percent, or 40 percent) of its initial value.

  2. Absolute change in the spread: the spread rises or falls by more than 100 basis points (200 basis points, or 300 basis points).

Each of these definitions has advantages and disadvantages. The “proportional change” definition is less sensitive to the fact that the absolute magnitude of spreads (in basis points) is higher during some periods than others; however, it will identify many episodes as “sharp” changes when the spread is close to zero. Conversely, the “absolute change” definition will tend to identify more episodes as “sharp changes” during times of large absolute spreads, but will not do so when the spreads are close to zero.

Using all the data available and a cutoff of 200 basis points, there are 79 sharp changes in the long historical sample and 151 sharp changes in the (much shorter) modern sample. Similarly, using a cutoff of 20 percent, there are 36 sharp changes in the historical sample and 165 in the modern sample (Tables 7 and 8). The difference becomes even more pronounced considering that more than half of the sharp changes in the historical sample took place in Turkey. This result is confirmed when different cutoffs are used. It is also interesting to note that, especially using cutoffs of 30 percent or above, the number of sharp increases (crises) is much larger than that of sharp decreases, in both the historical and the modern sample. This is consistent with the notion that crises start abruptly but dissipate slowly and gradually, or that instances of panic take place more frequently than sudden improvements in investors’ views regarding a given country.

Table 7.

Sharp Changes in Spreads, by Country

article image
Source: The Economist’s Investor Monthly Manual and J.P. Morgan web site.
Table 8.

Sharp Changes in Spreads, for Different Cutoffs

article image
Source: The Economist’s Investor Monthly Manual and J.P. Morgan web site.

The presence of far more sharp changes in our modern sample than in our historical sample suggests that emerging market crises (and sharp improvements in spreads) have been more common in the 1990s than at the time of the classical gold standard. This result is consistent with the findings of Bordo and Eichengreen (2000) who, using a different definition of crises as currency and banking crises, show that crises were less prevalent in 1880–1913 than in the post-Bretton Woods period (and at least as frequent in the 1970s–1980s as in the 1990s).

In addition, there is evidence that, today compared with the past, there are relatively more instances in which sharp changes (crises or sudden improvements in spreads) take place in several countries rather than in only one country. Specifically, the number of months when sharp changes take place in more than one country as a share of the number of months when sharp changes take place in at least one country is higher in the modern sample than in the historical sample (Table 9). This is the case using any of the cutoff points considered in this paper, although it is most clearly illustrated by focusing on the 10 percent cutoff for the historical sample and the 30 percent cutoff for the modern sample. Using these cutoffs, the proportion of months with sharp changes in only one country is roughly the same in the historical and the modern samples—10.2 percent and 10.4 percent, respectively. Nevertheless, in the historical sample the proportion of months with sharp changes in more than one country is 1.1 percent, compared with 7.5 percent in the modern sample.

Table 9.

Number and Type of Crises, 1887-1913 and 1994-2000.

article image
Sources: The Economist’s Investor Monthly Manual and J.P.Morgan web site.The historical sample includes 9 countries: Argentina, Brazil, China, Egypt, Japan, Portugal, Queensland, Russia, Sweden.The modern sample includes 8 countries: Argentina Brazil, Bulgaria, Mexico, Nigeria, Philippines, Poland, Venezuela.The sample periods were chosen to ensure that there are no missing observations.

Search for breaks

Before attempting to identify the nature of events that cause sharp changes in spreads, we search for “structural breaks” in the spread series, using various techniques based on Perron (1989).

An iterative search for breaks

This method assumes no a priori knowledge of potential break dates. Instead, it is based on using all the available data for repeated estimations of the following equation:

log (Spread)t=β0+β1log (Spread)t−1+β2Δlog (Spread)t−1+β3Δlog (Spread)t−2+β4TREND+β5EVENTlong+β6EVENTshort, (1)

where EVENTlong is a dummy variable that takes the value zero at all times prior to the proposed break and the value one from the time of the break onwards, and EVENTshort takes the value one at the time of the event, and zero at all other times. If an event has a long-term impact on yields, then the “long” dummy variable will be different from zero (assuming the series is not unit root). A significant “short” dummy implies that an event creates only a short-term “blip.”21 The method involves repeated estimation of Equation (1) while moving the break date and the corresponding EVENT dummy variables one observation at a time and recording its statistical significance. The sample is then split in two at the point where the statistical significance of the EVENTlong dummy is highest, and the process is repeated within each half of the sample until no statistically significant break points are detected in any sub-sample.

The “Moving Windows” approach

An alternative method for searching for breaks at unknown dates in the spread series is based on the construction of a two-year window, which is then shifted by one month at a time. A modified version of Equation (1) is estimated within each “window,” and the dates yielding EVENTlong dummies with the highest statistical significance are recorded.22 This method can identify “shorter” breaks more easily than the iterative search based on the whole sample period described in the paragraph above.

The determinants of changes in spreads, “then” versus “now”

Consistent with our earlier results on the co-movement of spreads “then” versus “now,” this section establishes that breaks in the spread series were determined by country-specific events in the historical sample, whereas they are largely associated with global events in the modern sample. Table 10 reports significant breaks in the historical spread series and describe the corresponding events. Evidently, most of the breaks took place at the time of important country-specific events that might be related to the country’s ability to repay its external debt. Major political events such as news about the beginning or end of wars and rebellions feature very prominently, as do economic news. For example, a military campaign against indigenous rebels, the end of a civil war, and a domestic revolt were all associated with breaks in Argentina’s spread series; similarly, an armed rising and the war against Sudan affected Egypt’s spread. Banking crises affected Queensland’s spreads. In several cases, changes in domestic monetary policy and regime were also associated with breaks in the spread series. The case of Japan is described in detail in Sussman and Yafeh (2000), who show that the adoption of the Gold Standard (1897) and Japan’s victory over Russia (1905) improved Japan’s “credit rating” significantly. Similarly, a break in the spread series is observed in Portugal at the time when that country left the Gold Standard. It is important to note that there are no historical instances in which breaks occur simultaneously in a large number of countries. Even the best-documented crisis in the nineteenth century, Baring’s Crisis of 1890, did not result in significant breaks in more than two countries, and certainly not in distant countries.23

Table 10.

Breaks in Historical Sample

article image
article image
Sources: The Economist Investor’s Monthly ManualThe breaks are identified through an iterative procedure as described in the text.