The Impact of Macroeconomic Announcements on Emerging Market Bonds
Author: Mr. Jochen R. Andritzky, Mr. Geoffrey J Bannister, and Ms. Natalia T. Tamirisa

Contributor Notes

Author(s) E-Mail Address: ntamirisa@imf.org, gbannister@imf.org, jochen@andritzky.de
Publication Date:
01 Apr 2005
eISBN:
9781451861020
ISBN:
9781451861020
Language:
English
Keywords:
WP; global bond; Standard and Poor's; monetary policy; budget deficit; order flow; standard deviation
Download PDF (304.7 KB)

This paper examines how emerging bond markets react to macroeconomic announcements. Global bond spreads respond to rating actions and changes in global interest rates rather than domestic data and policy announcements. All announcements affect market volatility. Data and policy announcements reduce uncertainty and stabilize the trading environment, while rating actions cause greater volatility. Results are broadly robust to country-specific and panel analyses, assuming conditional variance and controlling for the surprise content of news. In subsamples, announcements are found to matter less for countries with more transparent policies and higher credit ratings. In a crisis, rating actions become less important, and investors focus more on simple and timely indicators, like CPI.

Abstract

This paper examines how emerging bond markets react to macroeconomic announcements. Global bond spreads respond to rating actions and changes in global interest rates rather than domestic data and policy announcements. All announcements affect market volatility. Data and policy announcements reduce uncertainty and stabilize the trading environment, while rating actions cause greater volatility. Results are broadly robust to country-specific and panel analyses, assuming conditional variance and controlling for the surprise content of news. In subsamples, announcements are found to matter less for countries with more transparent policies and higher credit ratings. In a crisis, rating actions become less important, and investors focus more on simple and timely indicators, like CPI.

“Brazil’s better than expected economic growth in the first quarter of the year helped emerging market bonds bounce back on Thursday. The South American giant’s gross domestic product grew at a pace of 2.7 percent between January and March, compared to the same period of 2003, the fastest expansion since President Luiz Inacio Lula da Silva took office in January 2003. The news was welcomed by investors who went back into the market to buy Brazilian bonds. The country’s benchmark global 40 bond rose 1.250 to bid 89.813 at a yield of 12.261 percent. The rise dragged J.P. Morgan’s Emerging Market Bond Index Plus (EMBI+) up 0.33 percent in total returns. Brazil’s debt comprises almost a quarter of the index.” Reuters News, May 27, 2004.

“Philippine sovereign dollar bond prices rallied across the board on Friday after the government said the budget deficit for the first eleven months was below the target. Sovereign bond prices have tightened across the spectrum by a quarter of a point. We expect the market to consolidate between now and the year-end,” said a Manila-based trader at a foreign investment bank. Spreads on Philippine sovereign dollar bonds due 2014 tightened by five basis points to 425 bps over Treasuries.” Reuters News, December 19, 2003.

I. Introduction

Asset prices are affected by the arrival of new information. This is one of the accepted cornerstones of modern finance theory, which has laid the foundation for a rich literature investigating the role of information in financial markets. One strain of this literature focuses on the effects of macroeconomic data announcements on various markets including equities, bonds, and currencies. Macroeconomic data announcements have been used to test the market efficiency and rational expectations hypotheses and, more recently, to investigate the microstructure of financial markets and the role of private information in the formation of asset prices. The results of these studies confirm that macroeconomic data announcements have a significant impact on financial markets, although this impact varies across markets and announcements. The direction and magnitude of the impact is generally ambiguous and depends, among other things, on the degree of uncertainty in the market and the surprise content of the announcements.

To date, most studies have concentrated on mature financial markets. This paper extends this literature by examining the effects of macroeconomic data releases on emerging markets. We examine the reaction of emerging bond markets to major domestic and international macroeconomic announcements for a number of countries. Given the imperfect nature of the emerging market bond asset class with relatively scarce information, the asymmetry of access to information between creditors and debtors, and the prevalence of default, we expect macroeconomic releases to have a direct bearing on investor sentiment regarding the creditworthiness of sovereign debtors, and therefore to have a significant impact on spreads of international bonds over benchmark U.S. Treasury bonds. Indeed, data releases are often cited in the press as factors explaining the widening or narrowing of spreads.

We use a gamut of econometric methods to estimate the effect of macroeconomic announcements on the daily change in emerging market bond spreads. The analysis covers announcements of GDP, industrial production, consumer prices, the trade balance, the fiscal balance, country rating actions, and domestic and U.S. policy interest rates for 12 emerging market countries from January 1998 to July 2004. Announcement effects are modeled in an event-study framework using dummy variables, and, as part of robustness checks, we use measures of surprise effects. We first estimate ordinary least squares (OLS) regressions assuming a constant variance of the residual, and then use a conditional variance GARCH model, with macroeconomic announcement variables entering both the level and variance equations, to gauge the effects of announcements on both the daily change in spreads and its volatility. The measure of volatility we focus on is the daily volatility of spreads on announcement days compared to non-announcement days.

The results are broadly consistent with, and in some respects stronger, than those for mature markets. While most domestic macroeconomic data releases do not appear to influence global spreads directly, they are found to have a significant effect on their volatility. In contrast, country rating actions and U.S. interest rate changes affect both the daily change in spreads and their volatility. These effects are observed even without controlling for the surprise content of announcements, suggesting that most announcements represent new information for emerging market investors. Results are broadly consistent across various country-specific and panel analyses, assuming conditional and unconditional variance. They are generally robust to controlling for the surprise content of news; country, region, and year effects; and to redefining the size of the event window. In subsamples, macroeconomic announcements appear to matter less for investment grade bonds and for more transparent countries. The importance of rating actions appears to decline in times of crisis.

The rest of the paper is organized as follows. In the next section we review the relevant empirical studies on the effects of macroeconomic data releases on financial markets. In Section III we describe our data sources and present some summary statistics that help characterize the data. Section IV describes our empirical approach and Section V presents results for tests of mean and variance equality, OLS and GARCH models and analyses over subsamples of the data. Section VI concludes.

II. Literature Review

The finance and monetary economics literature on the effects of macroeconomic announcements in mature financial markets is rich and extensive. Macroeconomic announcements are often used as a measure of public information to test the market efficiency hypothesis or, by correlating announcements with survey forecasts, to test the rational expectations theory.2 Examining how monetary policy affects financial markets has also been an important motivation for research on macroeconomic announcements. More recently, the literature on macroeconomic announcements has been spurred by the desire to understand the microstructure of financial markets, focusing on the role of private information, as revealed in order flow, as a complement to public information.

A. Announcement Effects in Mature Markets

There is a general consensus in the literature that macroeconomic announcements have significant effects on financial markets, both in terms of asset returns and their volatility. However, markets appear to react not to data releases per se but to the unexpected component of these releases. Consistent with theoretical predictions, market reaction to an announcement is stronger if its surprise content is higher.

Differences Across Markets and Types of Announcements

Effects of macroeconomic announcements tend to differ across financial markets and types of announcements. Trade balance news, for example, appears to have different effects on U.S. stock prices, exchange rates, and interest rates, depending on the size and sign of the surprise in the news (Aggarwal and Schirm, 1998). Announcements of retail sales, price indices and non-farm payroll data appear to have the greatest impact on the U.S. Treasury bond market (Kim and others, 2004; Fleming and Remolona, 1997; Balduzzi and others, 2001; Cook and Korn, 1991). To some extent, market participants’ reaction to macroeconomic news (particularly, employment announcements) could reflect their expectations of the Federal Reserve’s policy response to the news. They also reflect the cyclical nature of announcements: positive surprises in procyclical indicators (such as non-farm payroll data) generally have a negative effect on U.S. Treasury bond prices, while positive surprises in counter-cyclical indicators (such as initial jobless claims) have a positive effect (Balduzzi and others, 2001). Market reaction to announcements tends to be stronger if there is already a high degree of uncertainty in the market (Fleming and Remolona, 1997).

Foreign exchange markets are found to respond to various types of announcements (Chang and Taylor, 2003). However, trade balance, GDP and non-farm payroll employment data appear to affect foreign exchange markets most (Kim and others, 2004). According to some studies, exchange rates are most likely to react to news about the state of the economy rather than to inflationary pressures, but the magnitude of these effects is small (Edison, 1996). There is some evidence for spillover effects of news across major foreign exchange markets. In the U.S. stock market, CPI and PPI announcements, and changes in the federal funds rate appear to have the largest influence on returns and volatility (Kim and others, 2004; Bomfim, 2003).

Volatility Effects and the Adjustment Process

The impact of macroeconomic releases on volatility is generally ambiguous. Many studies find the so-called “calm-before-the storm” effect, with volatility declining before releases of major macroeconomic data (Jones and others, 1998). If an announcement reveals new information not incorporated in asset prices, volatility tends to rise on the announcement day, as markets adjust their positions in response. However, some announcements might lead to a decline in volatility. For example, an announcement of lower than expected inflation might reduce uncertainty in the market and create a more stable trading environment, because it allays the fears of an interest rate increase (Kim and others, 2004).

Macroeconomic data releases have an impact on financial markets that is similar to those associated with policy announcements. The reason for these apparent similarities is policy feedback effects, where policy is expected to be adjusted in response to statistical data releases. The calm-before-the-storm effects on volatility that characterize data releases also apply to policy announcements: volatility in the stock market, for example, tends to be lower on the days prior to meetings of the U.S. Federal Reserve Open Market Committee and higher on the days of the meetings (Bomfim, 2003).

Oral interventions by central bank officials in the United States, the euro area, and Japan have been found to influence both the level and volatility of exchange rates (Fratzscher, 2004). Oral communication by the central banks tends to reduce volatility in the foreign exchange markets and is found to be effective independent of the stance and direction of monetary policy and the occurrence of actual interventions, which suggests that markets perceive oral interventions as new information. The effectiveness of interventions increases if they deviate from the prevalent policy “mantra” and if they are coordinated across countries. Interventions also tend to be more effective when uncertainty in the market is higher. This suggests that oral interventions provide useful public information on the desired direction or level of the exchange rate, thereby helping reduce uncertainty.3

More recent literature based on high frequency transaction data has identified two typical stages in markets’ adjustment to public information (Fleming and Remolona, 1999; Chang and Taylor, 2003). A major macroeconomic announcement causes an immediate change in prices, while trading volume declines and bid-ask spreads widen. The widening of bid-ask spreads has two interpretations. On the one hand, it can be interpreted as the value of the “option to trade” offered by market makers for investors that want to move first on new information. On the other hand it can be interpreted as the effect of an information asymmetry where participants are unsure about the implications of the new announcement. In the second stage, trading volumes and price volatility rise, as investors reach a consensus about the implications of the new announcement. Volatility set off by public information is extended by traders’ private information, as trading order flow also plays an informational role (Green, 2004). All in all, the two-stage process takes place within a relatively short period of time, within minutes or hours following the announcement.

Notwithstanding their statistical significance, most announcement effects are shortlived. The analysis of intraday data shows that most indicators have a significant price effect on U.S. Treasury bonds only within the first minute after the announcement (Balduzzi and others, 2001). Similar effects have been found for prices and volatility in Australian Treasury bond markets, although trading volumes tend to be significantly higher for up to one hour after the announcement (Kim and Sheen, 2001). In the foreign exchange markets, for example, significant effects have been found only at less than 15 minute intervals after announcement (Chang and Taylor, 2003). The fast dissipation of announcement effects explains why such effects tend to be most pronounced in intraday data.

B. The Role of News in Emerging Markets During Financial Crises

There is a dearth of comprehensive research on the impact of macroeconomic announcements in emerging markets. The limited existing literature has focused on the role of policy announcements and other news during financial crises. Ganapolsky and Schmukler (1998) examine the reaction of Argentina’s stock market index, Brady bond prices, and peso-deposit interest rates to specific policy announcements and news reports received by markets during the Mexican crisis of 1994–1995. Announcements that were perceived as increasing the credibility of the currency board apparently had a positive impact on market returns. These announcements included the agreement with the International Monetary Fund, the dollarization of reserve deposits in the central bank, and changes in reserve requirements. The appointment of a new finance minister helped reduce volatility of stock and bond returns, while a lowering of the reserve requirements increased the volatility of interest rates.

Kaminsky and Schmukler (1999) investigate what types of local and neighboring-country news moved stock markets during the Asian crisis. Using Bloomberg, the authors classify news into several categories: agreements with international organizations or the financial community, credit ratings, economic news, including real, financial, and external sectors news; fiscal policy, political news, and capital controls. Using an event-study approach, they find that news about agreements with international organizations and credit rating changes were most important in explaining large movements in stock prices during the Asian crisis.4 However, they also point out that some price movements were not associated with any major news releases, which the authors interpret as suggesting markets were driven by herd instincts during the crisis. A question remains, however, as to which types of macroeconomic announcements matter for emerging markets in non-crisis times. In this paper, we conduct a more general analysis of announcement effects in emerging markets.

III. Data

We focus on the global bond market with foreign-denominated issues. Unlike local equity or bond markets, the global bond market has a well-diversified base of investors that can be expected to share a common understanding of the fundamental forces driving the market. We use the country sub-indices of the Emerging Market Bond Index–Global (EMBI Global) provided by JP Morgan to calculate the dependent variable.5 The dependent variable is the daily percentage change in country sub-index bond spreads. We focus on the relative change in spreads, rather than the absolute change, to provide a uniform scaling and allow comparison across time and countries.

The sample covers 12 emerging market countries for the period from January 5, 1998 to July 15, 2004. As of June 2003, the countries included in the sample accounted for about two thirds of the EMBI total capitalization and were reasonably diversified geographically. They were Brazil, Chile, China, Colombia, Korea, Malaysia, Mexico, Poland, South Africa, Thailand, Turkey, and Venezuela. Of the top ten sovereign borrowers, only Russia and the Philippines are not considered because of data limitations. Most series (except for Korea and Malaysia) exhibit some autocorrelation, albeit without any recognizable trends or unit roots. Fewer observations are available for Chile than for other countries, because until 2001 Chile did not have enough bonds outstanding to be eligible to enter the EMBI index (Table 1).

Table 1.

EMBI Spreads

article image
Source: Bloomberg and authors’ calculations.The table reports key statistics for the dependent variable, the daily percentage change in country EMBI spreads. The sample period runs from 01/05/1998 to 07/15/2004.

The main independent variables are dummy variables for various types of macroeconomic announcements, i.e., release dates, taken from Bloomberg’s calendars of events. Bloomberg is a consistent source of data on macroeconomic announcements for various countries, which is widely used by investors, analysts and traders. Bloomberg calendars of events are published on a regular basis and are available to a broad investor community.

The data cover statistical releases of the following macroeconomic indicators: real GDP; industrial production (IP); consumer prices (CPI); trade balance (TB); and fiscal balance (FB). We also take account of announcements of changes in the domestic policy interest rate (SR). To account for global, systemic factors, we include announcements of changes in the U.S. federal funds rate, denoted by SR(US). Being a major determinant of the risk-free benchmark curve, the U.S. short-term interest rate is an important factor influencing the cost of emerging market financing, and has been found in empirical work to be positively correlated with spreads.6

Since rating announcements have been proven to be important determinants of changes in spreads by earlier studies, we also include announcements of rating actions by Standard & Poor’s (SP), which cover changes in the actual rating, rating outlooks, and watch listings. We consider this a reasonably good proxy for all rating changes, given the high correlation between S&P, Moody’s and Fitch’s actions. Recent comparative studies of rating agencies have found that S&P tends to provide the earliest and the most thorough market assessment. According to Gande and Parsley (2005), S&P provides 36 percent more rating changes than Moody’s while preceding Moody’s rating actions in two-thirds of the time. Reisen and von Maltzan (1999) have found that S&P rating changes cannot be anticipated by markets.

In our setting, rating actions also serve to confirm the validity of our model specification. The rating announcement variable is coded on a –1/0/1 scale, where –1 indicates a negative rating action (e.g., downgrade) and +1 a positive rating action (e.g., upgrade). We expect spreads to widen (tighten) on negative (positive) announcements, resulting in a negative coefficient in regressions. When modeling volatility of returns, we use only the binary dummy of rating actions, as we expect the volatility to rise on days of rating announcements, no matter whether they were good or bad news.

The frequency of announcements varies across countries, types of announcements, and over time. While data on announcements of common indicators, such as GDP growth, industrial production, CPI, and the trade balance are widely available, budget announcements are reported only for Brazil, Poland, and Mexico. Changes in policy rates are available only for those countries that announce policy rates within their monetary policy framework. For all countries, however, the number of announcements has increased over time, reflecting in part the development of the IMF’s Special Data Dissemination Standard (to which all countries in our sample, except China, subscribe) and the general trend toward greater transparency in emerging market countries (Table 2).

Table 2.

Domestic and International Announcements

article image
Source: Bloomberg.

Period 1/1/2004 to 7/15/2004.

The table shows the total number of annoucements per indicator and year in all countries in the sample.

The number of overlaps in data releases is small, accounting for less than 10 percent of all announcements. Announcements generally are not clustered around certain days of the week, although several announcements may occur on the same day for a given country. An exception is Poland, where about 60 percent of CPI and budget deficit announcements occur on the same day. In Thailand, all but one announcement of industrial production and the trade balance occur on the same day in our sample. There are relatively few monetary policy announcements in our sample. For the U.S., they largely coincide with the cycle of monetary policy relaxation.

IV. Methodological Approach

The methodological approaches used in the literature have evolved over time. The literature started in the form of event studies, which examined the effect of announcements on asset prices, assuming constant variance of the residual in the return equation and estimating the return equation with ordinary least squares. Subsequently, the literature was broadened to examine the effect of announcements on the volatility of asset prices by explicitly modeling the variance. In terms of modeling announcements, the traditional approach of using dummy variables has progressed to incorporate the surprise content of news based on forecasts of macroeconomic indicators. Nonetheless, most studies suffer from a drawback that the robustness of results has not been examined extensively. In this paper, we seek to remedy this.

We examine how releases of macroeconomic information affect the level and volatility of spreads for sovereign bonds included in the EMBI index using a broad variety of approaches: (i) simple ANOVA equality tests of the mean and variance of the daily change in spreads to compare differences on announcement and non-announcement days (for non-overlapping announcements); (ii) country-by-country OLS and GARCH estimations to gain insight into individual country reactions of the daily change in spreads to announcements, and (iii) a dynamic panel regression and a panel GARCH model, which allow us to gain estimation efficiency and to try to generalize results for the sample as a whole. We then conduct extensive robustness and subsample tests.

The mean equality tests are the most general approach to examine the differences in mean and variance in subsamples. They represent simple F-tests, where F-statistics of the mean tests are calculated as the ratio of “between” and “within” sums of squares:

F=SSBSSW=n0(x¯0−x¯)2+n1(x¯1−x¯)2(∑i=1n0(xi,0−x¯0)2)+∑i=1n1(xi,1−x¯1)2)/(n0+n1−2),(1)

where x¯0 indicates the mean daily change in spreads on non-announcement days and x¯1 on announcement days respectively, and n is the number of observations in each subsample. This test is done for each type of announcement, excluding days of other announcements from the analysis.

The analogous F-statistic for the variance equality tests is given by the ratio of variances in the respective subgroups:

F=s02/s12.(2)

We also estimate OLS regressions for all countries, which assume an unconditional distribution of the variance of our dependent variable. We use macroeconomic announcement dummies as independent variables. All regressions control for day-of-the week effects. For country i, the model is specified as:

Ri,t=ci+aiRi,t−1+∑w=MonThuδi,wDi,w+∑k=GDPSR(US)δi,kDi,k,t+εi,t(3)E(εi,t)=0,

where Dw are days of the week dummies and Dk are announcement day dummies for different types of data and policy announcements (GDP, CPI, industrial production, trade balance, fiscal balance, domestic short-term policy rate, U.S. federal funds rate, and rating actions, as described above). The regression includes a constant ci and one lag of the dependent variable, Ri,t-1. Q-statistics of residual autocorrelation do not point to the need to include lags of higher order. The Lagrange multiplier test shows significant conditional heteroskedasticity effects in residuals (Engle, 1982)—a common effect in financial time series—suggesting that the least square estimates may be inefficient.

To improve efficiency and to examine the effect of announcements on the volatility of the daily change in spreads for individual countries, we employ a GARCH(1,1) model of Bollerslev (1986). The model allows us to take into consideration the long memory effects detectable in most financial time series by conditioning the variance of the daily change in spreads on past volatility:

Ri,t=βi+ϕiRi,t−1+∑k=GDPSR(US)φi,kDi,k,t+ui,tεE(ui,t)=0 and E(ui,t2)=hi,t(4)hi,t=λi,0+λi,1ui,t−12+λi,2hi,t−1+∑w=MonThuηi,wDi,w+∑k=GDPSR(US)μi,kDi,k,t+εi,t.

In this set up, the mean equation includes a constant βi and an autoregressive term Ri,t-1 in addition to the announcement dummies. The residual variance is modeled explicitly, using a constant λi,0, and the ARCH and GARCH coefficients, λi,1 and λi,2, for which λi,1, λi,2 >0 and λi,1, λi,2 <1 to ensure positive and stationary residual variance. GARCH equations are estimated using the maximum likelihood method. Announcement effects enter both the mean and the variance equations.

To examine the common impact of announcements on emerging bond markets as an asset class, we pool data and estimate a panel regression using the Generalized Method of Moments (GMM). The basic dynamic panel regression is given by:

Rt=k+θhRt−1+∑w=MonThuvwDw+∑k=GDPSR(UR)ϑkDk,t+εt(5)E(εt)=0.

We use the same country data as before, but implicitly restrict coefficients on announcement dummies to be the same for all countries. We also control for country-, region- and year-specific effects. Variance in this model is not modeled explicitly.

The panel GARCH model assumes that today’s variance of returns depends on past variance. Covariance between variables is assumed to be zero. This simplification reduces the number of coefficients to be estimated, increasing the degrees of freedom and making computation feasible. The following equations describe the model:

Rt=σ+ρRt−1+∑k=GDPSR(US)τkDk,t+utE(ut)=0 and E(ut2)=ht(6)ht=ξ0+ξ1ut−12+ξ2ht−1+∑w=MonThuψwDw+∑k=GDPSR(US)ωkDk,t+εt,

where ξ1, ξ2 > 0 and ξ1+ ξ2 < 1. Like the dynamic panel model, the GARCH model is estimated in different versions, controlling for country-, region-, and year-specific effects.

We also use a number of techniques to control for possible biases and errors. To ensure unbiasedness of GARCH estimations, we control for outliers using dummy variables. Outliers are defined as daily changes in spreads exceeding four standard deviations from the mean in our sample. Outliers constitute less than one percent of observations. They appear to correspond to isolated jumps in spreads, possibly reflecting episodes of temporarily low trading activity and hence low liquidity in the market. Some outliers are concentrated around idiosyncratic market events, such as the rebalancing of portfolios after Argentina was taken out of the EMBI Global index in 2001.

Besides estimating models with different sets of control variables, we test the robustness of the results to a different window during which announcements may have had an effect: three days (pre-announcement, announcement, and post-announcement) instead of one (announcement) day. Widening the window further would lead to a significant increase in the number of overlapping announcements. Overlaps would make it difficult to disentangle whether a certain effect on returns comes, for instance, from an announcement made five days ago or a pre-announcement reaction of an indicator release expected at the end of the week.

In addition to the robustness checks, we estimate the panel GARCH model for different subsamples, including: (i) investment grade and non-investment grade credits; (ii) countries with more and less transparent policies, based on various measures of transparency; (iii) crisis or near-crisis periods versus non-crisis periods, where crisis or near crisis periods include dates where EMBI Global spreads were more than two standard deviations above the detrended mean for the sample. These periods reflect financial crises in Asia and Russia in 1998, Turkey in 2000-2001, Argentina in 2001, and Brazil in 1998–1999 and 2002.

In all of the above models, we implicitly assume that the daily change in spreads for country i is the sum of its mean and a random error, which is analogous to the constant mean return model of Campbell, Lo, MacKinley (1997):7

Ri,t=μi+εi,tE(εi,t)=0(7)

When discussing variance, we refer to the daily volatility of relative changes in the EMBI country sub-index spread rather than intraday trading volatility. Our dataset of daily EMBI spreads does not allow a judgment about the latter. In the variance equality test, this means that the adjusted return variance on all non-overlapping announcement days is compared to those on non-announcement days. In the GARCH model, the expected squared residual from the mean equation is regarded as variance.

V. Results

We find that announcements generally do not have a systematic effect on the daily change in spreads (with the exception of rating announcements). However, they often have significant effects on the volatility of spreads. Data releases tend to lower volatility of emerging market bond spreads, while rating announcements have the opposite effect. These results are generally robust to using alternative model specifications and estimation methods.

A. Country-Specific Results

Our first set of tests compares the mean proportional change in spreads on announcement and non-announcement days (Table 3). Overall, the results foreshadow the results of OLS regressions and GARCH models, showing only a few cases where the mean return is significantly different on announcement days. The implication is that announcements have no systematic effect on daily changes in spreads. This is what we would expect, since we do not control for the surprise content of announcements or the direction of the surprise. The one notable exception is for the announcement of S&P rating actions, which are associated with a significant difference in the daily change in spreads for eight out of the twelve countries in our sample. For most countries, the sign of the effect is what we would expect, with a positive announcement leading to a decline in spreads and a negative announcement leading to a widening of spreads.

Table 3.

Mean Equality Tests

article image
Source: Authors’ estimates.The table presents p-values and means for the ANOVA test of mean equality announcement and non-announcement days. Overlapping announcements are by the type of announcement, by country, and for the entire panel. The test compares the mean daily percentage changes in EMBI spreads on excluded from the sample for the purposes of this test.

In contrast, tests of the equality of the variance (ANOVA) across announcement and non-announcement days show significant differences in many cases (Table 4). The effect tends to be significant for individual countries for GDP, fiscal balance and U.S. interest rate announcements. This is consistent with our findings from other models.

Table 4.

Variance Equality Tests

article image
Source: Authors’ estimates.

absolute value, i.e. 0/1 dummy.

The table reports p-values and variances for the ANOVA test of variance equality by the type of announcement, by spreads on announcement and non-announcement days. Overlapping announcements are excluded from the sample country and for the entire panel. The test compares the mean variances of the daily percentage changes in for the purposes of this test.

OLS regressions (Table 5) confirm the result from the mean equality tests that there are few significant effects of announcements on the daily change in spreads. Again the exception is rating announcements, which have a significant effect on the change in the level of spreads in six of nine countries where rating actions took place. In four out of six cases the coefficients are negative as expected, implying that an upgrade leads to a tightening of spreads.

Table 5.

Country-by-Country OLS Regressions

article image
Source: Authors’ estimates.***, **, and * indicates signifiance on the 1 percent, 5 percent, and 10 percent level.The table reports coefficients and robust standard errors of country-by-country OLS regressions of the daily percentage change in EMBI spreads on announcement dummies. For all indicators, except rating actions, independent variables are 0/1 dummies for announcement days. For rating actions, the independent variable is measured as -1, 0, and 1, with -1 indicating a negative change, 0 no change, and +1 a positive change in the rating or outlook on creditworthiness. All regressions control for days-of-the-week effects.

Estimations of GARCH equations at the country level (Table 6) confirm the two main results above: announcements do not have a systematic effect on the daily change in spreads (with the exception of rating announcements); announcements do seem to have an effect on the volatility of spreads on emerging market bonds. GARCH mean equations show only a few significant coefficients for announcements. An exception is rating announcements which are significant for five out of nine countries where rating changes took place. In four out of five cases the coefficient is negative, as we would expect. In contrast, the coefficients on the GARCH estimation for the variance equation show significant effects from announcements in many countries. The effects are consistent for GDP announcements (which appear to reduce the volatility of the daily change in spreads) and for the trade balance (increasing in the volatility of the daily change in spreads), possibly because the interpretation of what these announcements imply for bond pricing is consistent across countries. For other types of announcements, the effects vary across countries.

Table 6.

Country-by-Country GARCH Models

article image
Source: Authors’ estimates.***, **, and * indicates signifiance on the 1 percent, 5 percent, and 10 percent level.The table reports coefficients and robust standard errors of country-by-country GARCH models of the daily percentage change in EMBI spreads. For all indicators, except rating actions, independent variables are 0/1 dummies for announcement days. For rating actions, the independent variable is measured as -1, 0, and 1, with -1 indicating a negative change, 0 no change, and +1 a positive change in rating or outlook in the mean equation and as a 0/1 dummy variable in the variance equation. The variance equation controls for days-of-the-week effects.

B. Panel Results

Our next step was to run the GMM and GARCH estimations on the pooled data set to examine the common impact of macroeconomic announcements, rating announcements and U.S. interest rate announcements across countries. As mentioned above, we control for day-of-the-week effects, country and region effects, and year effects. In terms of effects on the daily change in spreads we can compare the results from the OLS regression and the estimation of the mean GARCH equation (Tables 7 and 8). Both estimations show significant effects on spreads for rating and U.S. interest rate announcements. Both types of announcements lead to tightening of spreads, which is in line with expectations. Positive rating announcements typically compress spreads. Likewise, U.S. interest rate announcements in our sample reflect a cycle of considerable monetary policy easing, and thus can be also expected to lead to lower spreads for emerging market bonds.8 Furthermore, fiscal balance announcements tend to lower spreads. These results are robust to controlling for country-, region- and year-specific effects. However, further robustness analysis, as described below, failed to confirm the finding of significant announcement effects for fiscal balance announcements, which might in part reflect the limited availability of data for this type of announcements.

Table 7.

Dynamic Panel Regressions

article image
Source: Authors’ estimates.***, **, and * indicates significance on the 1 percent, 5 percent, and 10 percent level.The table reports coefficients and robust standard errors for dynamic panel regressions of the daily percentage change in EMBI spreads on announcement dummies. The sample includes 19,900 observations. For all indicators, except rating actions, independent variables are 0/1 dummies for announcement days. For rating actions, the independent variable is measured as -1, 0, and 1, with -1 indicating a negative change, 0 no change, and +1 a positive change in rating or outlook. Country effects are dummy variables for individual countries. Region effects are dummies for three regions: Asia, Europe and Africa, and Latin America. Year effects are time-specific dummy variables.
Table 8.

Panel GARCH Models

article image
Source: Authors’ estimates.***, **, and * indicates signifiance on the 1 percent, 5 percent, and 10 percent level.The table reports coefficients and robust standard errors for panel GARCH models of the daily percentage change in EMBI spread. For all indicators, except rating actions, independent variables are 0/1 dummies for announcement days. For rating actions, the independent variable is measured as -1, 0, and 1, with -1 indicating a negative change, 0 no change, and +1 a positive change in rating or outlook in the mean equation and as a 0/1 dummy variable in the variance equation. Country effects are fixed effects for countries. Region effects are dummies for three regional groups: Asia, Europe and Africa, and Latin America. Year effects are time-specific dummy variables. The last column reports results from using a three-day-event window, which defines an announcement event from the day preceeding to the day following the announcement.

As regards the apparent importance of rating announcements, significant effects do not necessarily reflect the high information content of these announcements. They may simply reflect special institutional rules in the private sector. Many investors are restricted, either by client mandates or prudential regulations, from investing in sub-investment grade sovereign debt (defined by Standard and Poor’s as BB+ and below). This implies that if a country has a sub-investment grade credit rating, the investor base for its bonds may be significantly narrower than for a country with an investment-grade credit rating, including more prominently speculative investors as opposed to “buy and hold” long-term investors. An upgrade to investment grade, by widening the investor base, will confer a significant benefit to a country’s bonds, increasing demand for the bonds and making price movements less volatile. In our sample, this effect may be operating for three upgrades to investment grade: South Korea (1/25/99), Mexico (2/7/02), and South Africa (2/25/00).

With regard to the volatility of the daily change in spreads, as in the country equations, the panel GARCH variance equation shows significant effects for a number of different types of announcements (Table 8). GDP, fiscal balance, and U.S. interest rate announcements tend to reduce the volatility of spreads, while announcements of the trade balance and rating actions tend to increase the volatility of spreads. A possible explanation may be that investors generally have similar views about what a given GDP, fiscal balance, or U.S. interest rate announcement implies for the country risk (as reflected in international bond spreads), and convergence of their views reduces volatility. If investors’ views differ, as is likely to be the case with trade balance announcements, or if an announcement tends to lead to a rebalancing of investors’ portfolios, volatility is likely to rise. The above results are robust to controlling for country-, region-and year-specific effects. Alternatively, in the case of rating announcements, the tightening of spreads seen in the mean equation may involve rebalancing of investor portfolios that causes transitional volatility.

C. Further Robustness Checks

The above results are broadly robust to changes in the size of the event window and to controlling for the surprise content of announcements.

When we redefine the event window as covering three days―not only the announcement day but the preceding and the following days―and run the panel GARCH regression with the redefined announcement dummies, the results remain largely unchanged (Table 8). As in the baseline regressions, rating upgrades and U.S. interest rate changes are found to lead to tighter spreads. GDP and industrial production announcements no longer have significant effects on the level of spreads. In the volatility equation, announcement effects generally have the same sign as in regressions with the one-day window, but are smaller in magnitude, suggesting that they dissipate over time.

This finding is in line with Jones and others (1998) who find in a regime-switching GARCH model for the U.S. Treasury bond market that announcement shocks do not persist, and volatility declines on the day following the announcement. It is consistent with the interpretation that markets quickly incorporate new information into prices and that the trading or information-gathering process does not generate volatility on announcement or succeeding days. It is also noteworthy that with the three-day window all announcements are found to have an effect on volatility, which might imply that markets react to all types of macroeconomic announcements, albeit with different leads and lags.

Next we control for the surprise content of news. Our baseline regressions suggested that markets rarely react to announcements per se, and by isolating the unexpected component of announcement we can test if markets react to surprises. In addition, we can check if our baseline results are robust to measuring all types of announcements similarly. There is a risk that not controlling for surprises in our baseline regressions might bias the significance of results downward, if asymmetric effects of positive and negative announcements were to cancel each other.

We measure surprise as a standardized difference between the released indicator and the median expectation by market participants, based on Bloomberg surveys of market participants. Positive surprises are thus defined as a higher-than-expected value of the respective macroeconomic variable, while negative surprises correspond to lower-than-expected values. Bloomberg survey data are available for GDP, industrial production, CPI, trade balance, and fiscal balance announcements. We also include data on surprises in U.S. non-farm payroll data as a proxy for surprises relating to future U.S. interest rate announcements. Given the limited and non-systematic coverage of market participants by Bloomberg surveys, we cannot preclude that these data do not contain measurement errors. This requires caution in interpreting the results.

In GARCH regressions with measures of surprises in domestic announcements, we find that positive surprises in industrial production and CPI announcements lead to a tightening of spreads, possibly because they make market participants revise their growth forecasts of emerging market economies (Table 9). Negative and positive surprises in these and other announcements (GDP and trade balance) are found to have statistically significant effects on volatility. When we control for surprises in international macroeconomic announcements, proxied by the U.S. payroll data, we find that only surprises in CPI announcements (positive and negative) affect the level of spreads and that virtually all macroeconomic announcements have an effect on the volatility of spreads. Taken as a whole, these results are broadly consistent with the literature on mature markets which finds ambiguous effects of positive or negative surprises in announcements on volatility. The results can be also interpreted as suggesting that some announcements, for example, GDP generally do not represent “news”, as they are widely anticipated by markets based on releases of higher frequency data, for example, industrial production and CPI.

Table 9.

Controlling for Surprise Content of Announcements

article image
Source: Authors’ estimates.***, **, and * indicates signifiance on the 1 percent, 5 percent, and 10 percent level.The table reports coefficients and robust standard errors for a panel GARCH model of the daily percentage change in EMBI spread with the surprise contents of announcement as explanatory variables. Surprise is calculated as a standardized difference between the released indicator and the median expectation by market participants, based on Bloomberg surveys. The model in the first column uses domestic indicators only, while that in the second column includes announcements of U.S. non-farm payroll employment figures.

Regressions controlling for the surprise content of news also suggest that our baseline regressions are not biased by the fact that we do not differentiate announcements into positive and negative ones. Even after controlling for the surprise content of news, we still find that announcements do not have a systematic effect on the level of spreads. We confirm that announcements generally have statistically significant effects on the volatility of spreads. This implies that macroeconomic announcements generally represent new information for emerging markets. This finding is stronger than that in the literature on mature markets, which tends to detect announcement effects on volatility only after controlling for the surprise content of news. This might be the result of the generally less transparent and more uncertain environment for investors in emerging markets, which increases the information value of macroeconomic and policy announcements even when they confirm expectations.

D. Impact of Announcements in Different Country Subsamples

By way of extension, we consider whether the results obtained on the entire sample would hold in various subsamples. We find that they do, although―as expected―they appear more significant and pronounced in cases where information is scarce or limited.

First, we split the data into two subsamples distinguished by whether the underlying bond was of investment or non-investment grade credit rating (Table 10). Results show that for non-investment grade countries announcements of domestic interest rate changes and fiscal policy have more pronounced effects on the level and the volatility of spreads, while rating announcements tend to matter less. This could imply that macroeconomic announcements have a greater information content for non-investment grade countries than for investment-grade countries.

Table 10.

Differentiating Announcement Effects by Credit Ratings and Crisis Occurrence

article image
Source: Authors’ estimates.***, **, and * indicates signifiance on the 1 percent, 5 percent, and 10 percent level.

Asia and Russia 1998, Turkey 2000-2001, Argentina 2001, and Brazil 1998-1999 and 2002.

The table presents coefficients and robust standard errors for panel GARCH models of the daily percentage change in EMBI spreads, for various subsamples. For all indicators, except rating actions, independent variables are 0/1 dummies for announcement days. For rating actions, the independent variable is measured as -1, 0, and 1, with -1 indicating a negative change, 0 no change, and +1 a positive change in rating or outlook in the mean equation and as a 0/1 dummy variable in the variance equation. Columns (1) and (2) compare announcement effects for bonds with the investment grade rating to those with the non-investment grade rating. Columns (3) and (4) split the sample by crisis periods, defined as the detrended global EMBI spread being larger than its sample period mean plus two standard deviations. All estimations control for days-of-the-week effects in the variance equation.

We also examine if there are any systematic differences in the announcement effects during crisis periods compared to more tranquil periods. Our sample covers several crisis events: in Asia and Russia in 1998, Turkey in 2000–2001, Argentina in 2001, and Brazil in 1998–1999 and 2002. We find that the importance of CPI announcements tends to increase during crises, possibly reflecting their high frequency, accuracy, and clarity.

Rating actions tend to become less significant in times of crisis (Table 10). It might be that during a crisis, the information content of rating actions declines as markets perceive ratings as lagging indicators of creditworthiness. Uncertainty and volatility rise during a crisis, as investors tend to trade based on private information and focus on such types of news as cabinet changes and negotiations with the International Monetary Fund. The finding of a lower significance of rating actions in crisis is consistent with Kaminsky and Schmukler (2002): in their sample rating actions are also found to be less significant statistically during crisis periods. However, in their estimations, the coefficient on the rating variable is larger in crisis periods, which leads the authors to conclude that the importance of rating increases.

We also split the sample by the degree of transparency of countries’ policies (Table 11). For a measure of transparency we use the frequency of releases for the indicators covered by our study. We define the measure for each country as the number of macroeconomic releases in a given 200 day rolling period, normalized by the average number of releases over the period for our sample, which allows us to split the sample into two.9 We find stronger announcement effects, particularly of ratings, for less transparent countries. One possible interpretation of this finding is that in less transparent countries the information content of announcements is lower, there is greater uncertainty about economic developments in less transparent countries, and market participants attach a greater importance to rating changes which tend to be based on direct discussions between the rating agency and the authorities. The finding that ratings tend to be more important in less transparent countries is consistent with that in Kaminsky and Schmukler (2002).

Table 11.

Differentiating Announcement Effects by Degree of Transparency of Countries’ Policies

article image
Source: Authors’ estimates.***, **, and * indicates signifiance on the 1 percent, 5 percent, and 10 percent level.This table presents coefficients and robust standard errors for panel GARCH models of the daily percentage change in spreads, for different subsamples. For all indicators, except rating actions, independent variables are 0/1 dummies for announcement days. For rating actions, the independent variable is measured as -1, 0, and 1, with -1 indicating a negative change, 0 no change, and +1 a positive change in rating or outlook in the mean equation and as a 0/1 dummy variable in the variance equation. The first two columns compare announcement effects in countries which publish more indicators than the sample median and countries which publish less. Estimations in columns (3) and (4) use the Economic Opacity Index as a measure of transparency. The last pair of estimations compare countries which make smaller than average revisions to the initially released figures and countries which tend to make large revisions, following the initial announcement. All estimations control for days-of-the-week effects in the variance equation.

When we split the sample by the magnitude of data revision—another measure reflecting transparency of countries’ policies—we also find that ratings have stronger effects for countries with larger data revisions. Markets might be perceiving statistics of such countries as less reliable and are more uncertain about the economic situation there (Table 11).

VI. Conclusion

This study contributes to the nascent literature on the effects of macroeconomic announcements in emerging markets. We do not find evidence that individual domestic data and policy announcements have systematic effects on the level of international bond spreads for emerging market countries. Global bond markets appear to respond mainly to announcements of changes in international ratings, which are designed to serve as composite forward-looking indicators of domestic fundamentals and policy developments and a broad measure of country risk. Changes in global interest rates also tend to affect the level of spreads, possibly because they lower the cost of funding for international bond investors and the cost of financing for emerging market sovereigns. Although there are few statistically significant effects of announcements on the level of spreads, we find many on volatility.

The direction of the volatility effects is ambiguous. Domestic data announcements generally lead to lower volatility, suggesting that new statistical information helps reduce the relatively high uncertainty that characterizes emerging bond markets and helps stabilize the trading environment. Policy announcements appear to reduce volatility when markets are able to quickly reach a consensus as to what these announcements imply for country risk, as appears to be the case with announcements of changes in global interest rates. If there is scope for alternative interpretations, volatility tends to rise in response to a policy announcement, as tends to happen after changes in domestic policy rates are announced. International rating actions tend to raise volatility, possibly because they typically require investors to rebalance their portfolios. The above results are broadly robust across various estimation methods and specifications, including country-specific and panel analyses with conditional and unconditional variance.

Our subsample analysis shows that announcements tend to matter less for transparent countries and for countries whose bonds have investment-grade ratings. Most announcements, including rating actions, tend to become less consequential in times of crisis, possibly because investors turn their attention to other types of news. An exception to this is CPI announcements, which investors seem to react more strongly to in times of crisis, possibly because CPI can provide a rough indication of the likely duration and depth of the crisis and the prospective timing of the recovery.

Controlling for the surprise content of announcements does not materially change our results, suggesting that our baseline regressions are not biased by the fact that we do not differentiate between positive and negative announcements. This is a stronger result than that found in the literature on mature markets. We suggest that this might be the result of the generally less transparent information environment in emerging markets, which increases the information value of macroeconomic and policy announcements even when they confirm expectations.

References

  • Aggarwal, Raj, and David C. Schirm, 1998, “Asymmetric Impact of Trade Balance News on Asset Prices,” Journal of International Financial Markets, Institutions, and Money, Vol. 8, pp. 83–100.

    • Search Google Scholar
    • Export Citation

  • Arora, Vivek, and Martin Cerisola, 2001, “How Does U.S. Monetary Policy Influence Sovereign Spreads in Emerging Markets,” IMF Staff Papers, Vol. 48, No. 3, pp. 474–98.

    • Search Google Scholar
    • Export Citation

  • Bollerslev, Tim, 1986, “Generalized Autoregressive Conditional Heteroskedasticity,” Journal of Econometrics, Vol. 31, pp. 307–27.

  • Bomfim, Antulio, 2003, “Pre-Announcement Effects, News Effects, and Volatility: Monetary Policy and the Stock Market,” Journal of Banking and Finance, Vol. 27, pp. 133–51.

    • Search Google Scholar
    • Export Citation

  • Balduzzi, Pierluigi, Edwin J. Elton, and T. Clifton Green, 2001, “Economic News and the Yield Curve: Evidence from the U.S. Treasury Market,” Journal of Financial and Quantitative Analysis, Vol. 36, No. 4 (December), pp. 523–43.

    • Search Google Scholar
    • Export Citation

  • Campbell, John Y., Andrew W. Lo, and A. Craig MacKinlay, 1997, The Econometrics of Financial Markets (Princeton: Princeton University Press).

    • Search Google Scholar
    • Export Citation

  • Cantor, Richard, and Frank Packer, 1996, “Determinants and Impact of Sovereign Credit Ratings,” Federal Reserve Board of New York Economic Policy Review, No. 2 (Summer), pp. 37–54.

    • Search Google Scholar