Emerging Markets Instability: Do Sovereign Ratings Affect Country Risk and Stock Returns? Graciela Kaminsky George Washington University and Sergio Schmukler * World Bank February 28, 2001 Abstract Financial market instability has been the focus of attention of both academic and policy circles. Rating agencies have been under particular scrutiny lately as promoters of financial excesses, upgrading countries in good times and downgrading them in bad times. Using a panel of emerging economies, this paper examines whether sovereign ratings affect financial markets. We find that changes in sovereign ratings have an impact on country risk and stock returns. We also find that these changes are transmitted across countries, with neighbor-country effects being more significant. Rating upgrades (downgrades) tend to occur following market rallies (downturns). Countries with more vulnerable economies, as measured by low ratings, are more sensitive to changes in U.S. interest rates. JEL Classification Codes: F30, G12, G14, G15, G29 Keywords: credit ratings; emerging markets; country risk; stock returns; financial markets; spillover effects *We are grateful to Richard Levich, Rick Mishkin, Carmen Reinhart, and participants at the workshops held at NYU for helpful comments and feedback. We thank Gloria Alonso and Chris van Klaveren for excellent research assistance. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors and do not necessarily represent the views of the International Monetary Fund or the World Bank. Contact address: The World Bank, 1818 H Street NW, Washington, DC 20433. Phone (202) 458-4167. Fax: (202) 522-3518. Email addresses: graciela@gwu.edu, sschmukler@worldbank.org 1. Introduction Worldwide financial market instability has been the focus of attention of both academic and policy circles. Naturally, following the series of currency crashes in the 1990s, most of the discussion has centered on currency crises. The latest crisis in Turkey in February 2001 will certainly contribute to keeping an avid interest in the triggers of crises well into the new millennium. But currency collapses are not the only ones to have attracted attention. The daily volatility of stock and bond markets in non-crisis times have also stirred interest, with, for example, the vagaries of the NASDAQ index in the United States making the daily headlines of newspapers around the globe. Many have argued that globalization is at the heart of this volatility, with highly diversified investors not paying much attention to economic fundamentals and following the herd in the presence of asymmetric information. (See, for example, Calvo and Mendoza (2000)). Naturally, this argument has provided ammunition to those supporting the re-introduction of capital controls, as argued in Krugman (1998) and Stiglitz (2000). Policies that can lead to moral hazard, including bailouts by both international institutions and governments, have also been suggested as other culprits of financial volatility and financial excesses. (See, for example, McKinnon and Pill (1997) and Dooley (1998)). The list of culprits does not stop here. Rating agencies have also been under scrutiny lately as promoters of financial excesses. As discussed in Ferri, Liu, and Stiglitz (1999), their pro-cyclical behavior, upgrading countries in good times and downgrading them in bad times, may have contributed to magnifying the boom-bust pattern in stock markets. Even if rating agencies do not behave pro-cyclically, their announcements may still trigger market jitters. This is because most institutional investors can only hold 1 investment grade instruments (i.e. securities with ratings above a certain threshold). Thus, changes in ratings, downgrading (upgrading) sovereign debt below (above) investment grade, may have a drastic impact on prices, because these rating changes affect the pool of investors.1 Rating changes may also unveil new (private) information about a country and thus they may fuel rallies or downturns. This effect is likely to be stronger in emerging markets, where problems of asymmetric information and transparency are more severe. Finally, changes in ratings might act as a wake-up call, with rating changes for one country affecting other countries with similar economies. Research on the effects of changes in sovereign ratings has flourished in the 1990s. This work has mostly focused on the effects of ratings on the instruments being rated. For example, Cantor and Packer (1996) and Reisen and Von Maltzan (1997) and (1999) examine the effects of rating changes of sovereign debt and find a significant effect on bond yield spreads. Similarly, Hand, Holthausen, and Leftwich (1992) show that rating announcements directly affect corporate securities. Richards and Deddouche (1999), using emerging market bank-level data, examine the impact of rating changes on bank stock prices, but do not find statistically significant effects. Previous research has not examined, however, whether rating changes for one country trigger contagious fluctuations in asset markets in neighboring countries nor has it examined whether ratings for one type of security affect other asset markets. To our knowledge, the only exception is Kaminsky and Schmukler (1999), who examine spillover effects of rating changes, among different types of news, in neighboring 1These effects are not just confined to the pool of investors acquiring sovereign debt. When a credit rating agency downgrades sovereign debt of a country, all debt instruments from that country might have to be downgraded accordingly because of the sovereign ceiling doctrine. As a result, commercial banks in the country that turn out to be rated as sub-investment grade can no longer issue internationally recognized 2 countries and find that news regarding the creditworthiness of a sovereign borrower affects other countries' stock and bond markets. Cross-country contagion effects can be large, witness the spillover effects of the Russian default on developed and developing countries.2 Rating agencies may contribute to these comovements in financial markets around the world. Similarly, news for one particular market can affect yields of other securities. These effects can, in some episodes, become quite dramatic, as was the case of the default of the State of Minas Gerais on the Brazilian real. Again, rating agencies may contribute to heighten financial instability. Neither has previous research examined whether economic vulnerability may trigger a large reaction of domestic financial markets to international events. For example, hikes in world interest rates may affect more drastically countries with economies in distress (with banking fragilities, liquidity problems due to high concentration of short-term debt, or near insolvency) than countries with healthier economies.3 This "vulnerability" effect may, in fact, explain some conflicting results in the empirical literature that examines international transmission of shocks. For example, Eichengreen and Mody (1998) and Kamin and von Kleist (1999) find that U.S. interest rate shocks do not affect sovereign bond spreads, while Herrera and Perry (2000) find that they do. Interestingly, the Eichengreen and Mody (1998) and Kamin and von Kleist (1999) studies include data only up to 1997 (before the crises) while the Herrera and letters of credit for domestic exporters and importers, isolating the country from international capital markets. Similarly, corporations will not be able to issue debt in international capital markets. 2The word contagion here is used in a broad sense to denote cross-country spillover effects, regardless of the nature of the shock. For alternative definitions and related papers see www.worldbank.org/contagion. 3On a similar vein, Frankel, Schmukler, and Serven (2000) study the transmission of international interest rates to countries with different exchange rate regimes. 3 Perry (2000) sample includes observations on the crises in Asia, Russia, and Brazil, and thus comprises episodes with very fragile economies. This paper complements the previous research on rating agencies by also examining these possible cross-country and security-market spillover-effects of rating changes. It also contributes to the literature on contagion and international transmission of shocks by examining the effect of domestic vulnerability, as measured by the ratings of international agencies, on the extent of international spillovers. Our results can be summarized as follows. First, rating changes significantly affect bond and stock markets, with yield spreads increasing on average 3 percent and stock returns declining about 1 percent following a downgrade. Second, rating changes also contribute to contagion or spillover effects, with rating changes among emerging markets triggering changes in both yield spreads and stock returns in foreign countries. Still, the effect is smaller than that of rating changes of the domestic economy. Third, similar to the findings in the literature on contagion, the "contagion" effects of rating changes are of a regional nature.4 Fourth, fragile economies, as measured by the international ratings, are more severely affected by changes in U.S. interest rates. In fact, interest rates hikes in financial centers fuel increases in sovereign risk 50 percent larger in vulnerable circumstances, relative to the changes when countries have more healthy economies. 4See, for example, Kaminsky and Reinhart (2000a). 4 Lastly, domestic-country rating upgrades take place following market rallies, while downgrades occur after market downturns. Foreign changes in ratings have a sustained effect. The rest of the paper is organized as follows. Section 2 describes the methodology. Section 3 presents the data. Section 4 discusses the results. Section 5 concludes. 2. Methodology To study the effects of ratings and vulnerability, we follow two different methodologies. First, we estimate panel regressions. Second, we perform event studies. The two methodologies are complementary in the sense that they show different aspects of the data. A. Panel Regressions The panel estimations study the reaction of country risk and stock returns to changes in ratings and U.S. interest rates. The fact that we use daily data does not allow us to control for country fundamentals, which are typically reported on a monthly or quarterly basis. But we do control for past changes of the explanatory variables. We use only one lag since further lags appear to be insignificant. We estimate different specifications for both country risk and stock prices. The first specification is the following pooled panel: Yi = +' Yi,t + 'Rt + 'it +i , US (1) ,t -1 ,t such that i = 1,...,N and t = 1,...,T . 5 Yi,t represents alternatively the log change in spreads and the log change in stock market prices. The sub-indexes i and t stand for country and time, respectively. The error term can be characterized by an independently distributed random variable with i,t mean zero and variance i . We estimate equation (1) using least squares, allowing for 2 ,t heteroskedastic residuals. R stands for the change in ratings. The variable R is equal to 1 (-1) if there is t t an upgrade (downgrade) at time t by any agency on any type of debt (foreign or domestic currency) from any country in the sample. The variable is equal to zero otherwise. If changes in ratings convey new information to market participants, we expect < 0 in ^ the regression for country risk; namely, rating upgrades (downgrades) lead to decreases (increases) in country risk. Analogously, in the regression for stock returns, we expect ^ > 0. itUS stands for the change in U.S. interest rates; strictly speaking, the interest rate is 100×log +it (1 US). As argued in Kamin and von Kleist (1999), there are different channels through which changes in U.S. interest rates can affect country risk. First, if there is a positive probability that a government will not pay its debt, increases in U.S. rates will prompt a higher rise in the interest rate of the government's debt. The higher increase is to compensate the probability of no repayment. Second, rises in U.S. interest rates increase the burden of the debt, decreasing a country's repayment capacity. Third, increases in U.S. rates can decrease investors' "appetite for risk," reducing the demand for risky assets from emerging countries, thus increasing the country risk. In sum, if increases in U.S. rates lead to higher country risk, we expect ^ > 0 in the equation for 6 country risk. A similar explanation can be argued for stock returns. In fact, governments can levy taxes on corporations if they face higher debt payments. Therefore, we expect that U.S. interest rates negatively affect stock returns, or that ^ < 0 in the equation for stock returns. As a second specification, we estimate: Yi = +' Yi,t + i'Ri + 'Ri + 'it +i . i j j US ,t -1 ,t ,t ,t (2) The variable Ri is equal to 1 (-1) if there is an upgrade (downgrade) at time t i ,t by any agency on any type of debt (foreign or domestic currency) from country i. The variable is equal to zero otherwise. The variable Ri is similar to the latter but takes the j ,t value 1 (-1) when there is an upgrade (downgrade) in country j for ji. That is, this specification tries to examine whether there is a "contagious" effect of credit ratings. The third specification we estimate is: Yi = +' Yi,t + i 'Ri ,dc i,dc+ i, fc i, fc+ j,dc j,dc+ j, fc j, fc ,t -1 ,t 'Ri ,t 'Ri ,t 'Ri ,t + 'iUS +i .(3) t ,t The difference between this specification and the previous one is that we separate the ratings into ratings for domestic-currency debt (dc) and ratings for foreign-currency debt (fc), both for the domestic and foreign countries, i and j. If ratings are important, we expect the domestic country foreign-currency ratings (fc) to be significant in the equation for country risk, because this is the instrument that credit ratings are evaluating. In other words, we expect a statistically significant ^ i, fc> 0. A-priori, the estimated coefficient for domestic-currency debt, ^i,dc, is not expected to affect the country risk, after controlling for changes in foreign-currency ratings. Still, the coefficient for domestic- 7 currency debt captures an exchange rate risk and may provide further insights into the vulnerability of the economy. The fourth specification we estimate is: Yi = +' Yi,t + i ' Ri ,dc i,dc ,t -1 ,t + i 'Ri , fc i, fc ,t + r 'Ri ,dc r,dc ,t + r 'Ri , fc nr, fc ,t . (4) + nr ' Ri ,dc nr,dc ,t + nr 'Ri , fc nr, fc ,t + ' it +i US ,t The variable Ri r,dcis equal to 1 (-1) if there is an upgrade (downgrade) at time t ,t by any agency on domestic-currency debt from country r (for ri). r represents a country that belongs to the same geographic region (East Asia, Eastern Europe, and Latin America) as i. The variable is equal to zero otherwise. The variable Ri nr,dc is similar to ,t the latter but takes the value 1 (-1) for countries outside the geographic region. The variables with the superscript fc denote upgrades and downgrades on foreign-currency debt. In this specification we examine whether the "contagious" effect of credit ratings is similar within a region or across regions. The fifth specification we estimate is: Yi = +' Yi,t + i ' Ri ,dc i,dc ,t -1 ,t + i 'Ri , fc i, fc ,t + r 'Ri ,dc r,dc ,t + r 'Ri , fc nr, fc ,t . (5) + nr ' Ri ,dc nr,dc ,t + nr 'Ri , fc nr, fc ,t + 'iUS +i R t ,t This specification is similar to the previous one, but we allow for the vulnerability effect. That is, we use different coefficients, , for the sensitivity to changes in U.S. R rates. In particular, we divide the observations into two different groups, observations with low and high ratings. We expect that countries with high ratings should be less affected by changes in U.S. rates due to the three channels described above. (A similar argument can be made for stock returns.) First, given that higher ratings mean a lower probability of default, changes in U.S. interest rates will impact more spreads of countries 8 with lower ratings. Second, countries with higher ratings tend to have a lower level of debt, so the burden of the debt will increase less in countries with high ratings when U.S. rates increase. Third, if there is a flight to quality when the U.S. rates increase, spread from "riskier" countries (countries with lower ratings) should react more strongly. The specifications described assume a zero correlation between the error term and the explanatory variables. This correlation may arise if the explanatory variables are endogenously determined. We do not expect changes in U.S. interest rates or changes in ratings to respond to contemporaneous daily changes in emerging market spreads or stock prices. However, a correlation between the lagged dependent variable and the error term is possible. This correlation can arise if the error term is if, for example, the true original model were in levels. In that case, the error term in our equations would be in first differences and correlated with the lagged endogenous variable by construction. To correct for potential biased coefficients, we estimate the more complete specification, equation (5), using instrumental variables. As instruments, we use lagged values of the lagged dependent variable, as suggested by Anderson and Hsiao (1982). B. Event Studies The above specifications study the contemporaneous effect of ratings on spreads and stock returns. However, they do not examine any possible dynamic effects of upgrades and downgrades. To have a sense of any dynamic effects that might be taking place, we use event studies. Dynamics effects are interesting because market participants can anticipate changes in ratings. Therefore, the contemporaneous effect might be smaller than the total effect of rating changes. Moreover, credit ratings can act 9 procyclically, downgrading countries during bad times and upgrading them during good times. We will not be able to disentangle these two observationally equivalent hypotheses, but we are able to observe whether downturns and rallies take place before downgrades and upgrades. Dynamics effects are also interesting because the effect of upgrades and downgrades can dissipate over time. The event study looks at country risk and stock market spreads (domestic stock markets prices relative to the U.S. S&P500 index) in a 10-day window around an upgrade or downgrade. All spreads and prices are set to 100 at day ­10, in that way we can easily measure the cumulative effects over time and we can, at the same time, compare spreads across countries. To perform the event studies we work with "clean events," i.e. upgrades and downgrades that do not overlap in windows of +/- 10 days. This distinction is important when considering an event window, to be able to isolate the effect of each change in rating. Figure 1 plots the ratings over time for three major rating agencies for a sample of countries. The figure suggests that many upgrades and downgrades across rating agencies occur simultaneously across agencies. In particular, the East Asian countries are downgraded during the Asian crisis and upgraded afterwards. Only few changes take place before the crisis in the case of Malaysia and South Korea.5 3. Data Our data set contains daily series of EMBI spreads, stock returns, interest rates, and credit ratings. We work with 16 emerging markets including East Asian, Eastern 5For a detailed study on how ratings are changed, see Cruces (2001). 10 European, and Latin American economies. The countries are in the data set are: Argentina, Brazil, Chile, Colombia, Indonesia, Korea (South), Malaysia, Mexico, Peru, Philippines, Poland, Russia, Taiwan, Thailand, Turkey, and Venezuela. The data set covers the period January 1990-June 2000. Appendix Table 1 displays the available data for each country and variable. JP Morgan produces the EMBI and EMBI+ (henceforth EMBI) series for a group of emerging markets, but also on a country-by-country basis. The index by country is a total return index that tracks traded debt instruments denominated in foreign currency. The instruments used are Brady bonds, benchmark Eurobonds, loans, and Argentine domestic debt. The EMBI spreads mostly reflect the difference between each country's sovereign bond yields relative to yields of benchmark instruments issued from developed countries. The spreads are commonly used as measures of country risk or default risk. When the probability of a sovereign default increases vis-ŕ-vis the U.S., bond prices decrease and yield spreads increase. The other variables that we use in this paper, stock returns, interest rates, and credit ratings, were downloaded from Bloomberg. Stock market price indexes for each country are measured in U.S. dollars. We use ratings on sovereign debt issued in domestic and foreign currency. These ratings try to measure the ability of the issuer to pay back its debt. We work with ratings from three major international rating agencies: Moody's, Standard and Poor's, and Fitch-IBCA. Table 1 provides some measures of financial market instability in our sample. Daily changes (in absolute values) in both markets are large and oscillate around 2.5 percent for sovereign spreads and around 1.6 percent for stocks. Our number of observations is high (about 11 thousand for bond spreads and 22 thousand for stock 11 prices). Tables 1 and 2 examine the characteristics of the changes in rating in our sample. Table 2 reports the number of upgrades and downgrades per rating agency and Table 3 reports the number of upgrades and downgrades per country. This last table shows that countries with currency collapses during the 1990s, such as Korea, Malaysia, Brazil, and Indonesia, were frequently re-evaluated by rating agencies. Appendix Table 2 shows the scale and type of ratings used by each rating agency. 4. Results We examine first the impact effect of changes in ratings and then we concentrate on the dynamics aspects of market responses to rating changes. A. Panel Regressions The panel regression results for the country risk are reported in Table 4. The columns of the table display the alternative specifications. The first column shows that the coefficient for the lag dependent variable is positive and statistically significant. The coefficient for the changes in ratings (domestic and foreign) is negative and statistically significant, although small when compared to the average daily change in spreads. In days of rating changes, spreads only change by about 0.5 percent while the average absolute change of spreads in our sample is about 2 percent. The second column examines separately whether changes in domestic ratings have different effects from changes in ratings of foreign countries. Interestingly, we now find that changes in ratings of domestic debt not only have a statistically significant effect, but this effect is also economically important, with rating changes leading to 12 changes in the spreads of about 2.5 percent. Foreign ratings also matter, but their effect is substantially smaller averaging about 0.4 percent over the sample. Our sample on ratings includes ratings on foreign-currency debt and domestic-currency debt. The first rating captures sovereign risk while the second also makes an assessment of devaluation risk. Since we are examining sovereign yield spreads, ratings on domestic-currency debt should not affect yield spreads once controlled for ratings directly related to country risk. Thus, column 3 examines separately the effects of ratings on foreign- and domestic- currency debt. As expected, ratings on foreign-currency debt are not statistically significant. Moreover, ratings on sovereign debt, once estimated independently from those of domestic-currency debt, have stronger effects on sovereign risk, as captured by the yield spreads. On average, changes in the assessment of rating agencies about country risk lead to spread changes averaging about 3.2 percent. The crises of the 1990s and the speed at which a crisis in one country engulfed the region and even spread around the globe have spawned a still growing literature on contagion. Much of the research centers on the role of financial links versus trade links. While opinions about the channels of transmission diverge,6 almost everybody agrees that in several cases contagion has been mostly regional. The Tequila crisis was basically confined to Latin American countries and the crisis in Thailand spread only to Asian economies.7 We now examine whether these regional contagion effects are also present when we examine contagion effects of credit ratings. The results are reported in columns 4 and 5. Interestingly, regional effects seem to be stronger than those from countries 6For example, Kaminsky and Reinhart (2000a) and Kaminsky, Lyons, and Schmukler (2000) have pointed to the role of financial links and have focused on the behavior of international banks and mutual funds. Corsetti, Pesenti, and Roubini (2000) in contrast have focused on the role of trade links. 13 from other regions, with the within-the-region rating changes leading to an average increase in yields of 0.8 percent while the across-regions rating changes only triggering an average change in spreads of about 0.4 percent. It is the rating agencies' assessment of currency risk (ratings on domestic-currency denominated debt) the one that matters for regional contagion but it is the rating agencies' assessment of sovereign risk the one that matters when assessing across regions spillover effects of ratings. After Calvo, Leiderman, and Reinhart (1993) brought to the limelight the close relationship between the capital inflows episode to emerging markets during the early 1990s to monetary policy in the United States, the number of papers written on this topic has increased significantly. A large number of papers has focused on the relationship between capital flows or foreign exchange reserves and interest rates in financial centers, others have focused on the links between returns in emerging markets and returns in financial centers. Others, as described in the introduction, have focused on the effects of interest rate hikes on interest rates and country risk. Interestingly, while these links were quite strong in the early 1990s, these links diluted somewhat in the mid-1990s, but reappeared in the late 1990s. The changing relationship between financial markets in emerging economies and in financial centers is particularly clear in the research studying the determinants of country risk, as examined in the introduction (see, Kamin and von Keist (1999), on one hand, and Herrera and Perry (2000), on the other). While examining the determinants of this time-varying relationship is beyond the scope of this paper, we will now examine whether hikes in interest rates in financial centers are transmitted more strongly to 7 Kaminsky and Reinhart (2000b) analyze why some crises become systemic while some others are confined to the national borders or at most are of a regional nature. 14 vulnerable economies. We divide the sample into two equal parts according to the country ratings. The results indicate that vulnerable economies are more strongly affected by the vagaries of international financial markets than healthier economies. The effect is about 50 percent higher. Table 5 reports similar estimations for stock market returns. The results are less strong than in the case of sovereign debt. This is not unexpected since assessments on sovereign risk should affect more closely yields on sovereign debt rather than stock returns. Still, stock returns seem to react more strongly to fluctuations in interest rates in financial centers when the economy tends to be more fragile, as captured by low ratings from credit agencies. B. Event Studies In the panel estimations, we just focus the instantaneous response of bond and stock markets in emerging economies. To capture whether credit ratings have a persistent effect on the mood of investors, we rely on event-study methods commonly used in the finance literature. The event-study methodology also allows us to examine the claim that rating agencies behave procyclically, upgrading countries in good states and downgrading them in times of crises. Thus, we examine the behavior of asset markets around the time of the rating changes (+/- 10 day-windows). Standard event study methodology requires linking rating events to abnormal returns. That is why we base the event study on the yield spreads between sovereign government debt and the benchmark instruments from industrial countries. In the case of stocks, we use the dollar 15 "stock spreads" between emerging markets stock prices and the S&P500 U.S. stock market index. Figures 2 and 3 summarize the event-study results in some detail for the case of domestic upgrades and downgrades. The four plots in each figure show the cumulative abnormal returns over that window around the time of changes in ratings. The panels on the left examine the effects of upgrades while the panels on the right report the effects of downgrades. The top panels examine rating changes of both foreign- and domestic- currency denominated debt, the bottom panels do the same for changes in ratings of just foreign-currency denominated debt. Both figures only look at the responses in the days before and after ratings of the domestic debt. Day zero is the day of changes in ratings. With respect to the behavior of markets in the days leading to the rating changes, the evidence seem to support the hypothesis that rating agencies may have contributed to amplify the boom-bust pattern in emerging markets. Overall upgrades occur when markets are rallying and downgrades when emerging markets are collapsing. This effect seems to be stronger in the case of downgrades. For example, bond spreads increase up to 9 percent in the 10-days prior to downgrades. Similarly, the stock market spreads decline up to 7 percent. Naturally, these fluctuations could reflect an anticipation effect. Still, we are most inclined to interpret them as evidence of procyclical behavior of rating agencies. In fact, our results are consistent with the findings in Reinhart (2001). In that paper, the author examines whether rating agencies actions anticipated the crises of the 1990s. With a large sample of countries and crises, the author concludes that rating changes far from being leading indicators of crises have turned out to be lagging indicators of financial collapses. 16 With respect to the aftermath of the rating changes, the results are more ambiguous. We first examine the responses of bond yields. The results suggest an asymmetric response of bond spreads after upgrades and downgrades. In particular, according to this event study analysis, the effects of downgrades tend to be somewhat more sustained while the effects of upgrades are usually reversed within two days. Typically, after experiencing an upgrade, bond spreads decline about 2 percent but within ten days bond spreads increase by about 4 percent, relative to the value at day -10. A different picture emerges from the analysis of downturns. While the contemporaneous reaction is similar to that of an upgrade (the spread changes by about 2 percent), following downgrades, the bond market does not recover. On the contrary, spreads continue to increase by at least 2 percent. The effects are somewhat stronger when we examine downgrades of foreign-currency denominated bonds only. Spreads widened an extra 5 percent. In contrast, the effects of upgrades seem to be long lasting in the stock market, with domestic stock markets gaining an extra 2 percent return relative to that of the stock index in the United States. This is not the case for downgrades. Figure 4 displays event studies for foreign events. Instead of using as an event upgrades and downgrades on sovereign ratings from the domestic country, the figure displays the behavior of EMBI spreads and stock spreads around upgrades and downgrades of spreads from other emerging markets. The figures on the left display upgrades, while the figures on the right show downgrades. The top panel uses EMBI spreads, while the bottom one uses stock spreads. The results show that foreign-currency upgrades are followed by large decreased in EMBI spreads and large increases in stock 17 market prices. Foreign downgrades are followed by increases in EMBI spreads although the results are not statistically very important. 5. Conclusions This paper complements previous research on the effects of credit ratings on financial markets in emerging economies. Most of the previous research has focused on quantifying the effects of changes in ratings of a country on sovereign risk as measured by the yield spread of domestic instruments relative to developed country benchmark instruments. In this paper, not only did we expand this exercise with updated data, but also we tested new hypotheses to have a more complete characterization on the effects of sovereign rating changes. We found that rating changes have effects both on the instruments being rated and on other instruments within the same country. We found that sovereign ratings have a significant impact on stock returns. We also examined whether ratings of other countries' sovereign debt have the potential to trigger contagion in financial markets. We found that rating changes have spillover effects to other countries. The effects tend to be limited to the neighbor countries. This paper also complements the previous literature on financial market linkages. This literature has examined the effects of changes in interest rates in financial centers. The results in this literature have been mixed, with for example sovereign risk being affected positively by interest rate hikes in some episodes but not in others. One important restriction in all these studies is that country risk obeys a common linear specification. One possibility is that interest rate hikes may have more damaging effects in countries near insolvency or with very fragile economies. We investigated this 18 possibility and examined whether countries with lower ratings are affected more severely by changes in U.S. interest rates. We found that countries with more vulnerable economies are affected 50 percent more by fluctuations of interest rates in the rest of the world. While our results help to understand better the movements of financial markets in emerging economies, we are far from explaining daily volatility. While this is a hard task not only for developing countries by also for mature markets,8 there is still room for improvement. With respect to understanding better the effects of ratings, there are several potential extensions to this paper. We have not examined yet whether changes in ratings have more impact during crisis times than during tranquil times. Other extensions can be addressed with new data. For example, if ratings are informative, it will be instructive to analyze whether sovereign ratings are more informative for less transparent countries than for more transparent countries. Further extensions imply using other ratings, beyond sovereign debt ratings. It would be interesting to work with corporate ratings to investigate whether ratings convey different information for different groups of firms. For example, one can expect that firms issuing ADRs, with more transparent accounting standards and for which more information is available, to be less affected by ratings than firms trading in less transparent local markets. Also, since rating agencies also assess exchange rate risk, we could examine whether these ratings are informative by looking at whether they affect differently countries and companies with different with exchange rate exposure. Also, it would be interesting to examine whether firms producing traded-goods 8 R 2 in all studies explaining daily variations in stock prices or bond yields is very low. 19 are less affected by country-risk, that is whether collateral (valued in international markets) can act as a buffer to country-risk changes. 20 References Anderson, T., and Hsiao, C., 1982, "Formulation and Estimation of Dynamic Models Using Panel Data," Journal of Econometrics, 18:47-82. Calvo, G., Leiderman, L., and Reinhart, C., 1993, "Capital Inflows and Real Exchange Rate Appreciation in Latin America: The Role of External Factors," Staff Papers International Monetary Fund, 40:108-51. Calvo, G. and Mendoza, E., 2000, "Rational Contagion and the Globalization of Securities Markets," Journal of International Economics, June. Cantor, R. and Packer, F., 1996, "Determinants and Impact of Sovereign Credit Ratings," Federal Reserve Bank of New York, Economic Policy Review 2:2, 37-53. Corsetti, G., Pesenti, P., and N. Roubini, N., 2000, "Competitive Devaluations: A Welfare-Based Approach" Journal of International Economics, June. Cruces, J., 2001, "Statistical Properties of Sovereign Credit Ratings," mimeo, Washington University and Universidad de San Andres, Argentina. Dooley, M., 1998, "A Model of Crises in Emerging Markets," NBER Working Paper No. W6300. Eichengreen, Barry, and Mody, Ashoka, 1998, "What Explains Changing Spreads on Emerging-Market Debt: Fundamentals or Market Sentiment?" NBER working paper no. W6408. Frankel, J., Schmukler, S., and Serven, L., 2000, "Global Transmission of Interest Rates: Monetary Independence and the Currency Regime" World Bank working paper No. 2424. Ferri, G., Liu, G., and Stiglitz, J., 1999, "The Procyclical Role of Rating Agencies: Evidence from the East Asian Crisis," Economic Notes 3:335-355. Hand, J., Holthausen, R., and Leftwich, R., 1992, "The Effect of Bond Rating Agency Announcements on Bond and Stock Prices," The Journal of Finance, 157:2, 733-752. Herrera, Santiago and Perry, Guillermo, 2000, "Determinants of Latin Spreads in the New Economy Era: The Role of U.S. Interest Rates and Other External Variables," mimeo, World Bank. Holthausen, R., and Leftwich, R., 1986, "The Effect of Bond Rating Changes on Stock Prices," Journal of Financial Economics, 17, 57-89. 21 Kamin, Steven and von Kleist, Karsten, 1999, "The Evolution and Determinants of Emerging Market Credit Spreads in the 1990s," International Finance Discussion Paper 1999-653, Federal Reserve Board. Kaminsky, G., Lyons, R., and Schmukler, S., 2000, "Managers, Investors, and Crisis: Mutual Fund Strategies in Emerging Markets" NBER Working paper No. 7855. Kaminsky, G. and Reinhart, C., 2000a, "On Crises, Contagion, and Confusion," Journal of International Economics, June. Kaminsky, G. and Reinhart, C., 2000b, "The Center and the Periphery: Tales of Financial Turmoils," February 2000. Kaminsky, G. and Schmukler, S., 1999, "What Triggers Market Jitters?" Journal of International Money and Finance, June, 18:537-560. Krugman, P., 1998, "Saving Asia: It is Time to Get Radical," Fortune, September 7, pp. 74-80. Larrain, G., Reisen, H. and Von Maltzan, J., 1997, "Emerging Market Risk and Sovereign Credit Rating," OECD Development Centre, Technical Papers, No. 124. McKinnon, R. and H. Pill, 1997, "Credible Economic Liberalizations and Overborrowing," American Economic Review, 87, 2, 189-93. Reisen, H. and Von Maltzan, J., 1999, "Boom and Bust and Sovereign Ratings," OECD Development Centre, Technical Paper No. 148. Reinhart, C., 2001, "Do Sovereign Credit Ratings Anticipate Financial Crises? Evidence from Emerging Markets," mimeo, University of Maryland. Richards, A. and Deddouche, D., 1999, "Bank Rating Changes and Bank Stock Returns: Puzzling Evidence from Emerging Markets," IMF Working Paper. Stiglitz, Joseph, 2000, "Capital Market Liberalization, Economic Growth, and Instability," World Development, Vol. 28 N6 pp. 1075-1086. 22 of returns k 11,122 11,122 21,788 21,788 stoc Number d Observations an spreadsI 0.4652 0.4986 0.3171 0.3947 Maximum EMB the 0.0000 0.0000 in Minimum -0.4986 -0.3947 d use 1 0.0379 0.0291 0.0257 0.0203 Statistics Standard Deviation Table observations the Summary allr Median -0.0012 0.0160 0.0000 0.0095 fo Mean -0.0004 0.0243 -0.0001 0.0158 statistics y s s value value summar spread price EMBI absolute stock absolute displays in in in in spreads prices table The ressions.gre change change change change EMBI stock Log Log of Log Log of Table 2 Total Upgrades and Downgrades by Rating Agency The table displays the total changes in ratings for long-term sovereign debt in foreign and local currency. The sample used is the one available for stock returns. Total Agency changes Upgrades Downgrades Moody's 48 19 29 Foreign currency debt 37 14 23 Local currency debt 11 5 6 S&P's 75 28 47 Foreign currency debt 45 19 26 Local currency debt 30 9 21 Fitch 47 21 26 Foreign currency debt 30 15 15 Local currency debt 17 6 11 Total 170 68 102 Table 3 Total Rating Changes by Country The table displays the total changes in ratings for long-term sovereign debt in foreign and local currency. Agency Total changes Upgrades Downgrades Argentina 5 3 2 Brazil 10 7 3 Chile 4 3 1 Colombia 5 0 5 Indonesia 13 1 12 Korea (South) 18 9 9 Malaysia 11 3 8 Mexico 9 5 4 Peru 1 1 0 Phillipines 4 4 0 Poland 6 6 0 Russia 18 7 11 Taiwan 0 0 0 Thailand 10 2 8 Turkey 4 1 3 Venezuela 5 2 3 Total 123 54 69 *** ** * ** variables domestic respectively. and 6 IV -0.482 -0.029 (-1.156) (-3.73) 0.002 0.005 0.046 0.068 0.000 0.006 (0.138) -0.001 -0.014 -0.002 (-0.318) (-2.120) (-0.933) (1.005) (1.714) (2.490) (-0.764) 10,408 instrumental ificance sign "Foreign). The. of * *** * * ** level 5 0.041 (1.923) -0.032 0.008 0.000 (0.560) -0.008 -0.004 0.001 0.024 0.035 0.000 0.006 non-regional (-3.994) (-0.145) (-1.904) (-1.880) (0.278) (1.577) (2.348) (-0.940) 10,923 parenthesis in percent constructed. are are regional, * *** * * ** 10,5,1 4 foreign, variables 0.040 0.008 0.000 0.001 0.029 0.000 0.005 (1.894) -0.032 (-3.988) (0.558) -0.008 -0.004 (-0.144) (-1.911) (-1.921) (0.258) (2.747) (-1.029) 11,122 T-statistics indicate these Specifications Spreads *,**,*** domestic, how * *** * *** on EMBI (any, Alternative 3 0.040 (1.900) -0.032 0.008 0.029 0.000 0.005 (-3.943) (0.580) -0.003 -0.002 (-1.593) (-0.938) (2.761) (-1.020) 11,122 in heteroskedasticity. for instrument. countries information 4 an Change as more * *** ** *** Estimates Table Log different for correction 0.029 0.000 0.005 Panel variable text 0.040 from (1.909) -0.025 -0.004 (-3.079) (-2.574) (2.746) (-1.025) 11,122 White debt main Variable: the the dependent using See * *** *** sovereign lagged on debt. 12 0.040 0.029 0.000 0.004 errors, (1.906) -0.005 Dependent the (-3.415) (2.747) (-1.105) 11,122 of lag standard downgrades y robust seconda and domestic-currency currency currency currency y y with and currenc y y domestic ratings uses y y y currency ratings 6) upgrades domestic domestic domestic currenc currenc currenc and currenc andng currenc high* low* domestic and currenc and currenc estimates forei foreign domestic rates: rates rates rates describe and foreign foreign domestic panel (specification foreign-currency variable foreign foreign domestic foreign foreign domestic ratings interest interest interest interest Variables foreign countries, countries, countries, in reports denote ratings: country, country, country, countries, countries, countries, U.S. U.S. U.S. U.S. Observations dependent in countries, countries, countries, ionalg in in in in d of table estimation change country, The (IV) The currency" Explanatory Lagged Change Any Domestic Domestic Domestic Foreign Foreign Foreign Regional Regional Regional Non-re Non-regional Non-regional Change Change Change Change Constant R-square Number Th ** * variables domestic and 6 IV 0.337 0.003 0.000 0.005 0.000 0.000 0.002 0.000 0.010 respectively. (1.568) (0.629) (0.045) (2.243) (-0.204) (-0.351) (1.508) -0.007 -0.011 (-0.963) (-1.797) (0.283) 20,508 instrumental Foreign" The. significance *** *** * ** of 5 0.088 0.003 0.006 0.006 0.000 0.000 0.002 0.000 0.010 level (4.409) (0.865) (0.865) (3.467) (-0.080) (-0.164) (1.951) -0.007 -0.012 (-1.280) (-2.245) (-0.186) 21,247 parenthesis non-regional). in constructed. are percent1 are *** *** * ** regional, 10,5, 4 0.088 0.003 0.006 0.006 0.000 0.000 0.002 0.000 0.010 variables (4.416) (0.868) (0.800) (3.441) (-0.056) (-0.159) (1.952) -0.009 21,788 T-statistics foreign, (-2.564) (-0.071) indicate these Specifications Prices how *** ** ** domestic, Stock *,**,*** on (any, Alternative 3 0.088 0.004 0.006 0.002 0.001 0.000 0.009 in (4.430) (1.139) (0.812) (2.015) (1.219) -0.009 (-2.538) (-0.068) 21,788 heteroskedasticity. for information 5 instrument. Change countries an as more *** * *** ** Estimates Table Log for correction different 0.088 0.007 0.002 0.000 0.009 Panel text (4.435) (1.770) (3.186) -0.009 (-2.509) (-0.116) 21,788 variable White from main Variable: the debt the using dependent See *** *** ** Dependent errors, lagged sovereign debt. 12 0.089 0.002 0.000 0.009 (4.462) (3.496) -0.009 21,788 the on (-2.509) (-0.133) of standard lag downgrades y robust seconda domestic-currency currency currency currency y and y with and currenc y y domestic ratings y uses y y currency ratings 6) domestic domestic domestic upgrades currenc currenc currenc and currenc andng currenc high* low* domestic and currenc and currenc estimates forei foreign domestic rates: rates rates rates and panel describe foreign-currency variable foreign foreign domestic foreign foreign domestic foreign foreign domestic (specification interest interest interest interest Variables foreign countries, countries, countries, reports ratings denote ratings: country, country, country, countries, countries, countries, U.S. U.S. U.S. U.S. Observations in dependent in countries, countries, countries, ionalg in in in in d of table estimation country, The (IV) change currency" Explanatory Lagged Change Any Domestic Domestic Domestic Foreign Foreign Foreign Regional Regional Regional Non-re Non-regional Non-regional Change Change Change Change Constant R-square Number Table 6 Number Clean Events by Country Events are for 10-day windows, including foreign-currency and domestic-currency debt. The events are for domestic country events. The sample used is the one available for stock returns. Total Upgrades Downgrades events Latin America Argentina 3 1 2 Brazil 5 4 1 Chile 3 2 1 Colombia 5 0 5 Mexico 3 1 2 Peru 1 1 0 Venezuela 3 1 2 Total 23 10 13 East Asia Indonesia 5 1 4 Korea 8 7 1 Malaysia 7 3 4 Philippines 4 4 0 Taiwan 0 0 0 Thailand 7 1 6 Total 31 16 15 Eastern Europe Poland 5 5 0 Russia 12 7 5 Turkey 2 1 1 Total 19 13 6 Gran Total 73 39 34 Figure 1 Ratings of Foreign Currency Sovereign Debt for Selected Countries The figures report the sovereign ratings from three credit rating agencies for a selected group of countries. Sovereign letters are published in letters (AAA, Aaaa3SS,....). The scale is different for each agency. Appendix Table 2 gives a mapping between each rating letters and a numerical scale. Argentina Malaysia 5 8 4.5 7 Moody's 4 3.5 6 3Moody's Fitch-IBCA 5 2.5 S&P's 4 S&P's 2 3 1.5 Fitch-IBCA 1 2 0.5 1 0 0 09-n 19-n 29-n 39-n 49-n 59-n 69-n 79-n 89-n 99-n 00-n 90 91 92 93 94 95 96 97 98 99 00 Ja Ja Ja Ja Ja Ja Ja Ja Ja Ja Ja n-aJ n-aJ n-aJ n-aJ n-aJ n-aJ n-aJ n-aJ n-aJ n-aJ n-aJ Brazil South Korea Moody's 5 8 4.5 Moody's 7 4 S&P's 3.5 6 S&P's 3 5 2.5 4 2 Fitch-IBCA 3 Fitch-IBCA 1.5 1 2 0.5 1 0 0 09-n 19-n 29-n 39-n 49-n 59-n 69-n 79-n 89-n 99-n 00-n 09-n 19-n 29-n 39-n 49-n 59-n 69-n 79-n 89-n 99-n 00-n Ja Ja Ja Ja Ja Ja Ja Ja Ja Ja Ja Ja Ja Ja Ja Ja Ja Ja Ja Ja Ja Ja Venezuela 8 Thailand 6 Moody's Moody's 7 5 6 4 Fitch-IBCA 5 S&P's 4 3 S&P's Fitch-IBCA 3 2 2 1 1 0 0 90 91 92 93 94 95 96 97 98 99 00 90 91 92 93 94 95 96 97 98 99 00 an-J an-J an-J an-J an-J an-J an-J an-J an-J an-J an-J an-J an-J an-J an-J an-J an-J an-J an-J an-J an-J an-J Figure 2 Event Studies of EMBI Spreads The figures display the log of EMBI spreads (normalized to 100 at day -10), +/- one standard deviation. The events are only related to upgrades and downgrades in the domestic country. Upgrades of foreign- and domestic-currency debt Downgrades of foreign- and domestic-currency debt 110 120 108 118 106 116 114 104 112 110 102 108 100 106 104 98 102 96 100 98 94 96 94 92 92 90 90 -10 -8 -6 -4 -2 0 2 4 6 8 10 -10 -8 -6 -4 -2 0 2 4 6 8 10 Days relative to announcement Days relative to announcement Number of clean events: 28 Number of clean events: 17 Upgrades of foreign-currency debt Downgrades of foreign-currency debt 125 114 123 112 121 110 119 108 117 115 106 113 104 111 102 109 100 107 98 105 103 96 101 94 99 92 97 90 95 -10 -8 -6 -4 -2 0 2 4 6 8 10 -10 -8 -6 -4 -2 0 2 4 6 8 10 Days relative to announcement Days relative to announcement Number of clean events: 22 Number of clean events: 14 Figure 3 Event Studies of Stock Market Indexes The figures display the log of local stock market index relative to the U.S. S&P 500 (normalized to 100 at day -10), +/- one standard deviation. The events are only related to upgrades and downgrades in the domestic country. Upgrades of foreign- and domestic-currency debt Downgrades of foreign- and domestic-currency debt 104 100 103 98 102 96 101 94 100 99 92 98 90 -10 -8 -6 -4 -2 0 2 4 6 8 10 -10 -8 -6 -4 -2 0 2 4 6 8 10 Days relative to announcement Days relative to announcement Number of clean events: 39 Number of clean events: 34 Upgrades of foreign- and domestic-currency debt Downgrades of foreign- and domestic-currency debt 101 104 99 103 97 102 95 93 101 91 100 89 99 87 98 85 -10 -8 -6 -4 -2 0 2 4 6 8 10 -10 -8 -6 -4 -2 0 2 4 6 8 10 Days relative to announcement Days relative to announcement Number of clean events: 35 Number of clean events: 23 Figure 4 Event Studies -- Foreign-Country Events The top panel displays EMBI spreads, while the bottom panel diplays stock spreads , i.e. the log of local stock market index relative to the U.S. S&P 500 (normalized to 100 at day -10). Both panels also plot +/- one standard deviation. The events are only related to upgrades and downgrades in foreign countries, both on foreign-currency and domestic-currency debt. EMBI Spreads EMBI Spreads Upgrades of foreign- and domestic-currency debt Downgrades of foreign- and domestic-currency debt 102 110 101 108 100 106 99 104 98 102 97 100 96 98 95 96 94 94 -10 -8 -6 -4 -2 0 2 4 6 8 10 -10 -8 -6 -4 -2 0 2 4 6 8 10 Days relative to announcement Days relative to announcement Number of clean events: 99 Number of clean events: 63 Stock Spreads Stock Spreads Upgrades of foreign- and domestic-currency debt Downgrades of foreign- and domestic-currency debt 103 103 102 103 102 101 102 101 100 101 100 99 100 99 98 99 98 97 -10 -8 -6 -4 -2 0 2 4 6 8 10 -10 -8 -6 -4 -2 0 2 4 6 8 10 Days relative to announcement Days relative to announcement Number of clean events: 84 Number of clean events: 116 2000 2000 2000 2000 2000 2000 2000 2000 2000 2000 2000 2000 2000 2000 2000 2000 sg date 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, Ratinng End June June June June June June June June June June June June June June June June 1992 1992 1990 Soverei 1990 1990 7, 1990 7, 1990 1990 18, 1996 1990 1990 1990 date 1, 1, 1, 1, 1, 5, 1993 1994 1995 1, 1, 1992 1, 30, 1, 11, 5, Initial January January December January December January January December February June June April January January May January 1999 30, 2000 2000 2000 2000 2000 2000 2000 2000 2000 2000 2000 2000 2000 2000 2000 date 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, 30, Returns End June June June June June June June June June June June June June June December June Stock 1991 1 1992 1992 1993 1992 1996 5, 1995 1996 1993 1, 1996 1996 date 3, 23, 2, 2, 1995 1995 2, 2, 4, 1996 2, 2, 1995 1996 Table 30, 30, 3, 30, 23, Availability endix Initial January January January January November June June January January January April December January January June April ppA Data 1997 2000 2000 2000 2000 2000 30, 2000 2000 2000 date 30, 30, 30, 30, 30, 30, 30, 30, S readsp End June June June June June January June June June EMBI 1991 1991 1997 1991 1993 31, 1998 31, 1993 1995 31, 31, 1997 date 4, 17, 30, 30, 30, Initial April December April December May January January December December a Country Argentina Brazil Chile Colombi Indonesia Korea Malaysia Mexico Peru Philippines Poland Russia Taiwan Thailand Turkey Venezuela Appendix Table 2 Scale of Ratings for Sovereign Debt Moody's S&P FITCH- IBCA Rating Number Rating Number Rating Number Rating Number Aaa3SS 8.5 Ba2 5.1 AAA 8 AAA 8 Aaa3S 8.8 Ba1SS 5.3 AA+ 7.33 AA+ 7.33 Aaa3 8.7 Ba1S 5.5 AA 7 AA 7 Aaa2SS 8.9 Ba1 5.4 AA- 6.66 AA- 6.66 Aaa2S 9.2 Ba 5 A+ 6.33 A+ 6.33 Aaa2 9.1 B3SS 3.5 A 6 A 6 Aaa1SS 9.3 B3S 3.8 A- 5.66 A- 5.66 Aaa1S 9.5 B3 3.7 BBB+ 5.33 BBB+ 5.33 Aaa1 9.4 B2SS 3.9 BBB 5 BBB 5 Aaa 9 B2S 4.2 BBB- 4.66 BBB- 4.66 Aa3SS 7.5 B2 4.1 BB+ 4.33 BB+ 4.33 Aa3S 7.8 B1SS 4.3 BB 4 BB 4 Aa3 7.7 B1S 4.5 BB- 3.66 BB- 3.66 Aa2SS 7.9 B1 4.4 B+ 3.33 B+ 3.33 Aa2S 8.2 B 4 B 3 B 3 Aa2 8.1 Caa3SS 2.5 B- 2.66 B- 2.66 Aa1SS 8.3 Caa3S 2.8 CCC 2 CCC+ 2.33 Aa1S 8.5 Caa3 2.7 CC 1 CCC 2 Aa1 8.4 Caa2SS 2.9 CCC- 1.66 Aa 8 Caa2S 3.2 CC 1.33 A3SS 6.5 Caa2 3.1 C 1 A3S 6.8 Caa1SS 3.3 A3 6.7 Caa1S 3.5 A2SS 6.9 Caa1 3.4 A2S 7.2 Caa 3 A2 7.1 Ca3SS 1.5 A1SS 7.3 Ca3S 1.8 A1S 7.5 Ca3 1.7 A1 7.4 Ca2SS 1.9 A 7 Ca2S 2.2 Baa3SS 5.5 Ca2 2.1 Baa3S 5.8 Ca1SS 2.3 Baa3 5.7 Ca1S 2.5 Baa2SS 5.9 Ca1 2.4 Baa2S 6.2 Ca 2 Baa2 6.1 C3SS 0.5 Baa1SS 6.3 C3S 0.8 Baa1S 6.5 C3 0.7 Baa1 6.4 C2SS 0.9 Baa 6 C2S 1.2 Ba3SS 4.5 C2 1.1 Ba3S 4.8 C1SS 1.3 Ba3 4.7 C1S 1.5 Ba2SS 4.9 C1 1.4 Ba2S 5.2 C 1 Source: Bloomberg