Emerging Market Spread Compression: Is it Real or is it Liquidity?

Kenichiro Kashiwase and Laura E. Kodres

 2004 International Monetary Fund

WP/04/xxx

IMF Working Paper Research Emerging Market Spread Compression: Is it Real or is it Liquidity? Prepared by Kenichiro Kashiwase and Laura E. Kodres [PRELIMINARY NOT YET Authorized for distribution by Mr. Rajan] October 2005

Abstract This Working Paper should not be reported as representing the views of the IMF.
The views expressed in this Working Paper are those of the author(s) and do not necessarily represent those of the IMF or IMF policy. Working Papers describe research in progress by the author(s) and are published to elicit comments and to further debate.

Spreads on emerging market countries' sovereign bonds have fallen dramatically since mid2002. Some have attributed the fall to improved economic fundamentals while others to ample global liquidity. The paper models spreads and attempts to empirically distinguish between the two factors. The results indicate that fundamentals, as embedded in credit ratings, are very important, but that expectations of future U.S. interest rates and deviations from those expectations are also a key determinant of emerging market spreads. JEL Classification Numbers: Keywords: emerging market bonds; Authors E-Mail Address: lkodres@imf.org

-3Contents

-4I. INTRODUCTION The determinants of the difference between the yield on a countrys emerging market debt securities and the U.S. 10-year Treasury note, denoted the spread, has been subject of academic inquiry for some time. The dramatic decline in spreads to levels only seen just prior to the 1997 Asian crisis has renewed attention in the subject. This decline occurred alongside a dramatic decline in policy interest rates in industrial countries and a marked increase in money growth as these countries loosened monetary policy in an effort to revive their economies following the bursting of the equity bubble in 2000, leading some commentators to claim that the fall in spreads was based on excess global liquidity. The spread compression was also accompanied by a seeming improvement in the real fundamentals of many emerging market countriesdebt ratios declined in some countries, current account deficits improved, and other structural policies were adopted. Reflecting these improvements, the average credit rating of the countries included in the EMBI Global index improved from BB to BBB- between early 2000 and 2004.1 As interest rates begin to rise, attempting to discern how much of the improvement in emerging market spreads is due to improvements in country fundamentals and how much is due to excess liquidity has important ramifications. If much of the compression in spreads is attributable to solid reforms undertaken in recent years then tighter monetary conditions in industrial countries should have only a small effect on spreads. However, if much of the spread compression was due to liquidity, then tighter monetary policy and a drying up of liquidity could mean reversals in spreads, which could be especially pronounced if excessive liquidity had also led to leveraged positions. This paper attempts to distinguish between the two factorsfundamentals and liquidityby constructing a model that takes into account several features of emerging market spreads and how they adjust to domestic fundamentals and interest rates.2 We first note that to the extent that rates paid by emerging market borrowers follow industrial country interest rates, a decline in interest rates, ceteris paribus, implies lower debt burdens of emerging market countries and an improvement in fundamentals as measured by debt service ratios, debt-toGDP ratios, and the likelihood of default. Thus, in observing an improvement in fundamentals the model should control for the interaction between industrial country interest rates and the domestic fundamentalsimprovements in fundamentals need to be in addition to the effects of lower interest rates in order to separately identify the effects of liquidity. We employ a 2-stage model to purge the country fundamentals of the indirect effects of interest rates. Improvement in the ratings is based on the unweighted average and accounts for about a half of the countries included in the EMBI Global universe. The rating of overall EMBI Global universe remained BB during the reference period. The model represents a type of forecasting model for spreads and, as such, does not distinguish between supply and demand factors for debt securities and their influence on spreads.
2 1

-5-

A common feature of previous examinations of emerging market spreads on which we try to improve is the use of credit ratings as a proxy for domestic country fundamentals. Many studies use credit ratings because it encapsulates a host of economic variables.3 While a handy and efficient measure, the measure is relatively coarse, that is, there are a fixed number of categories (e.g. AAA, AA+, A, ... C-, and SD, referring to default), and alterations among them are not associated with a fixed (or linear) response in spreads. The model below attempts to enrich the informational content of ratings by: (1) using the indications for future up- or down-grades represented by the rating outlook; (2) scaling the ratings variable using logarithms to account for their non-linear relation with spreads; and (3) computing predicted values of ratings depending on three types of fundamentals. One of the motivations for adding the credit rating outlook to the series of actual ratings is an observation that markets react first and foremost to hints of future ratings changes rather than the actual event when it occurs. Sy (2002) provides formal evidence of the early reaction by market participants. He observes that when a countrys spreads are excessively high a rating downgrade frequently follows, similarly excessively low spreads are often followed by upgrades, suggesting markets anticipate future ratings changes (or alternatively that credit rating agencies are late to alter ratings).4 Below we look more carefully at spreads and changes in the outlook and show a stronger and more timely response to the outlooks than the actual rating changes. Previous studies that have examined the influence of global interest rates on emerging market spreads have variously used a short-term interest rate, such as a 3-month eurodollar deposit rate or the Fed Funds rate, a long-term interest rate, such as the U.S. 10-year Treasury note or some weighted average of 10-year government securities from industry countries, or the slope of the yield curve, as measured by, say the difference between the U.S. 10-year note and 3-month Treasury bill. The use of this latter variable is prompted by the use of carry trades in which participants borrow funds at the Fed Funds rate or similar rate and then invest in longer-term higher yielding securities. In many cases, the results have been less than satisfactory, with some authors finding support for the global liquidity story in a positive relation between the given interest rate and spreads and other finding either no relation or a negative relation.5 The results may be confounded by the potential impact of interest rates on fundamentals or for other reasons. We chose to use a forward-looking short-term interest ratethe Fed Funds 3-month ahead futures rate in part to
3 4

See for example, GFSR (April 2004) and Sy (2002).

Rating agencies, however, may in turn argue that higher debt spreads increase borrowing costs and implying a greater risk of default. Thus, changes to ratings are not behind the curve but accurately reflect the increased risk of default as spreads widen. Eichengreen and Mody (1998), Kamin and von Kleist (1999), Slok and Kennedy (2003), and McGuire and Schrijvers (2003) all find a negative or inconclusive relationship.

-6try to further separate any influence of interest rate with fundamentals, but as well, to better capture the view that financial markets are forward looking and therefore incorporate expectations of future interest rates into their current trading decisions. In the next section we describe the data in more detail and then move to the results. We then use our model to forecast emerging market bond spreads under two different scenarios. Finally, we sum up our results and suggest some areas of further research. II. DATA A. Variables We collected a number of daily and monthly series during the period of January 1991 and May 2005 for 33 major emerging market economies as well as financial data from the U.S. economy. A.1. Emerging Market Bond Spreads To examine the factors that determine emerging market spreads, we collected daily observations of the Emerging Market Bond Index (EMBI), the EMBI Plus, and the EMBI Global for individual countries, as well as the broad index. Due to a set of specific criteria for country admission and liquidity, these three bond indices consist of varied number of member countries and financial instruments. Hence, starting points of the series differ across countries and the indices (Table 1).6 There are some instances where a countrys index has missing observations in the middle of its series.7 For this reason, when necessary, the spreads based on these indices are spliced to obtain a longer time series in the sample period of Because the EMBI only includes Brady bonds, countries bonds must meet strict liquidity criteria for inclusion to the index. For this reason, there were only five countries in the original EMBI. Their series start at the end of 1991. The EMBI Plus was introduced in January 1998, included 14 countries at its inception, and introduced broader liquidity criteria for inclusion of additional instruments: (1) loans, (2) U.S. dollar local market bonds, and (3) Eurobonds. The EMBI Global, introduced in January 1998, uses more relaxed liquidity criteria. Countrys admission requirements under the EMBI Global are different from those under the EMBI and the EMBI Plus. Countries to be included in the EMBI or EMBI Plus must be rated BBB+ or lower by Standard & Poors. On the other hand, countries under the EMBI Global only need to satisfy one of the following criteria: (1) classified as having low or middle per capita income by the World Bank; (2) has restructured external or local debt in past 10 years; or (3) currently has restructured external or local debt outstanding.
7 6

Nigeria, for example, has missing observations during the period between April 1998 and April 1999. Likewise, Pakistan has missing observations during the period of February 2003 through March 2004. In case of the Philippines, its EMBI series terminated in January 1997 (Table 1).

-7-

Table 1. Availability of EMBI, EMBI Plus, and EMBI Global

EMBI Aggregate Index Algeria Argentina Brazil Bulgaria Chile China Colombia Cote d'Ivoire Croatia Dominican Republic Ecuador Egypt El Salvador Hungary Lebanon Malaysia Mexico Morocco Nigeria Pakistan Panama Peru Philippines Poland Russia South Africa South Korea Thailand Tunisia Turkey Ukraine Uruguay Venezuela 91M1~02M6 n.a. 93M4~02M6 91M12~02M6 94M11~02M6 n.a. n.a. n.a. n.a. n.a. n.a. 95M6~02M6 n.a. n.a. n.a. n.a. n.a. 91M12~02M6 n.a. 92M1~02M6 n.a. 97M2~02M6 97M5~02M6 91M12~97M1 94M11~02M6 98M12~02M6 n.a. n.a. n.a. n.a. n.a. n.a. n.a. 91M12~02M6 1/ EMBI Plus 97M12~05M5 n.a. 93M12~05M5 94M4~05M5 94M8~05M5 n.a. n.a. 99M5~05M5 n.a. n.a. n.a. 95M2~05M5 02M5~05M5 n.a. n.a. n.a. 02M1~04M12 93M12~05M5 93M12~05M5 93M12~05M5 n.a. 96M7~05M5 97M3~05M5 99M4~05M5 94M10~05M5 97M8~05M5 94M12~05M5 n.a. n.a. n.a. 99M7~05M5 01M7~05M5 n.a. 93M12~05M5 EMBI Global 97M12~05M5 99M3~03M2 93M12~05M5 94M4~05M5 94M7~05M5 99M5~05M5 94M3~05M5 97M2~05M5 98M4~05M5 96M8~04M6 01M11~05M5 95M2~05M5 01M7~05M5 02M4~05M5 99M1~05M5 98M4~05M5 96M10~05M5 93M12~05M5 97M12~05M5 93M12~05M5 01M6~05M5 96M7~05M5 97M3~05M5 97M12~05M5 94M10~05M5 97M12~05M5 94M12~05M5 93M12~04M4 97M5~05M5 02M5~05M5 96M6~05M5 00M5~05M5 01M5~05M5 93M12~05M5 1/

1/

1/

2/

2/

1/

1/

1/ The series are terminated as countries failed to maintain a set of required criteria for inclusion. 2/ There are some missing observations during the period. For Nigeria, the EMBI series does not have observations during April 1998 through April 1999. For Pakistan, the EMBI Global series has missing observations during March 2003 through March 2004. Source: J.P. Morgan Chase.

-8January 19919 and May 2005. It is important to note that a sovereigns market weight changes over time as inclusions or exclusions of a country or various debt issuances occur.10 Despite such complexities, the EMBI, the EMBI Plus, and the EMBI Global have become market benchmarks and provide us with the most comprehensive, readily available, and easyto-analyze data.11 A.2. Credit Ratings and Outlooks Monthly sovereign credit ratings and their outlooks are collected from the Ratings Direct, provided by the Standard & Poors. We used long-term credit ratings for assessing the sovereigns foreign currency default risk since debt instruments included in the EMBI family (EMBI, EMBI Plus, and EMBI Global) are mostly long-term and denominated in U.S. dollars. In general, most rating agencies issue both long- and short-term credit ratings in domestic and foreign currencies. In addition, they assign credit outlooks for each country in both currencies, though the rating agencies note that their outlook does not guarantee any future changes in the ratings themselves. Changes in credit outlooks are associated with emerging market bond spreads in during our sample period. We will demonstrate this point in section III. As you can see from Figure 1, variation in the credit outlook appears to add information for examining country risk. There have been 200 and 259 monthly observations over the sample period that recorded changes in long-term credit ratings and outlooks, respectively.12 Moreover, there have been 146 monthly observations for which changes in outlooks occurred when long-term credit ratings remained the same.13 Thus, we can enrich our variable on countries risk by using the outlook.14 Short-term ratings, however, do not Although aggregate EMBI series starts from January 1991, country specific indices are available from December 1991 at the earliest.
10 9

Similarly, when a sovereign retires a debt instrument or amortizes principal of the debt, and if the issues current outstanding face value falls short of a required level under the criteria, the issue is removed from the countrys index. Market weight data are made available by J.P. Morgan Chase. This monthly series starts in December 1993. Sy (2002) uses EMBI Plus sovereign spreads, noting the additional advantage that they control for floating coupons, unusual features, and principal collateral and rolling interest guarantees.

11

There have been seven monthly observations which countrys long-term credit rating and/or outlook changed twice during the same month, though they are counted as one monthly change.
13 14

12

See the areas represented by (C) and (E).

Kaminsky, et. al. (2003) and Sy (2002) also refer to the importance of outlooks in their analysis of the spreads.

-9appear to add very much fundamental information since the timing of their changes generally coincides with that of their long-term ratings and outlooks changes. There were only 6 monthly observations for which there was a change in short-term ratings but no simultaneous change in long-term credit ratings or outlooks.15
Figure 1. Changes in Sovereign Credit Ratings and Outlook: January 1991 ~ May 2005 1/ (Number of months)
Changes in: Long-term Credit Rating - 200 Short-term Credit Rating - 106 Outlook - 259 (E) (G) (F) (C) (A) = 61 [17%] (B) = 6 [ 2%] (C) =136 [39%] (D) = 26 [ 7%] (E) = 10 [ 3%] (F) = 49 [14%] (G) = 64 [18%] (A) + (B) + (C) + (D) + (E) + (F) + (G) =352 [100%]

(B) (D) (A)

1/ Based on 33 emerging market economies included in J.P. Morgan Emerging Markets Bond Index (EMBI), EMBI Plus, and EMBI Global. The term in [ ] indicates the number of observations as a percent of total number (352) of months. Source: Standard & Poors, Moody's Investors Service, and Authors calculations.

A.3. Fundamentals Many studies use sovereign long-term credit ratings as a direct proxy for the economic fundamentals of the emerging market economies.16 The ratings as determined by the rating agenciesalthough encompassing critical economic fundamental variablesare in essence a measure of ability and willingness of sovereigns to meet their financial obligations in a timely manner. They are determined by quantitative as well as qualitative information.17

15 16 17

See the area represented by (B). See for example, the IMFs Global Financial Stability Report (April 2004) and Sy (2002).

See Standard & Poors (2002a). Standard & Poors lists a sovereign ratings methodology profile, in which they discloses a number of quantitative indicators in the area of economic, financial, and political risks. Willingness to pay plays an important role as a qualitative (continued)

- 10 Eichengreen and Mody (1999) identified a set of economic variables that most closely explain ratings. To be more consistent with the way that the ratings are actually constructed, this paper takes a holistic approach and uses aggregate political fundamentals as well as economic and financial fundamentals to construct a predicted rating. Specifically, we use monthly data of aggregate political, economic, and financial risk indicators from the PRS Group.18,19 Aggregate indicators are particularly useful because, due to their construction, they provide relatively high variation even during an early period of the sample where macroeconomic information is generally scarce. A.4. Fed Funds Futures The Fed uses a target rate for Fed Funds to transmit of its monetary policy objectives and this rate has become a market-wide benchmark for various financial activities. Leveraged carry traders who borrow at the short-end of the yield curve to invest in emerging market bonds keenly watch short-term benchmark rates, such as the Fed Funds rate. For this reason, we look at the implied yield on the 3-month Fed Funds futures and evaluate how market expectations of future U.S. monetary policy affect the emerging market bonds. The Fed Funds futures rate has the advantage that it influences interest rates all along the U.S. yield curve.20 It is noteworthy that the 3-month ahead Fed Funds futures rate has a 0.99 correlation with 3-month LIBOR and U.S. Treasury bill rate, and 0.69 with 10-year U.S. Treasury notes at a daily frequency over the entire sample period. A.5. Dummy Variables of Expected Tightening and Loosening We also created dummy variables to reflect expected rate tightenings and loosenings by analyzing the difference between 3-month ahead Fed Funds futures rate (FF3M) and the target rate (FFTR) at a daily frequency. The dummy variable of expected tightening (E_UP) takes the value of 1 in a given month t if the futures price would round to the next 25 basis point increment for more than half of the trading days over the month, and takes the value of 0 if otherwise. Thus, E_UP is defined as:

measure in evaluating sovereigns credit ratings. A government can default selectively on its obligation even when it possesses the financial capacity for the timely debt service. Erb et al. (2000) use a composite risk index, consisting of economic, financial, and political risk indicators to analyze the importance of country risk in the pricing and returns of emerging market bonds. See Appendix I.A. for a discussion of what variables are included in the risk indicators and how they are constructed. Using Fed funds futures rates, Kuttner (2001) disentangled expected from unexpected policy actions, and concluded that the impact of unexpected rate decisions on the interest rates of both short- and long-term maturities were significantly positive.
20 19 18

- 11 E _ UPt = 1 if 0.5

Xt Nt E _ UPt = 0 if otherwise,

where X t is the number of days that satisfy the condition 12.5 100*(FF3Mt - FFTRt) in a given month, and N t is the number of all trading days in the same month. The dummy variable of expected loosening (E_DOWN) is constructed in a corresponding way. By using these dummy variables, we can estimate how much the expectation of rate increase or cut influences the bond spreads. In addition, these dummy variables help capture turning points in market expectations of U.S. monetary policy.
A.6. Volatility of Fed Funds Futures

Uncertainty of future U.S. monetary policy is perceived to have a large impact on the financial markets, making decisions about financial risk allocation more difficult. To measure this uncertainty, we again used the difference between the implied yield of threemonth ahead Fed Funds futures contract and the target rate at a daily frequency. In a rolling 90-calendar-day window, we calculated standard deviation of the difference. The daily series of standard deviation was then averaged over each month. This measure of uncertainty of future U.S. monetary policy, expressed in standard deviation, was closely linked to the widening of bond spreads during the tightening cycle of 1994-1995 (Figure 2). Heightened anxiety at the beginning of the current tightening cycle (mid-2004) concerning the speed and

22

See Global Financial Stability Report (2004).

- 12 Figure 2. Volatility of Fed Funds Futures Market and Emerging Market Bond Spread
Emerging market bond spread (log scale; basis points; left scale) Volatility of 3-month ahead Fed funds futures minus target rate 1/ (standard deviation times 100; right scale)

2,400 2,000 1,600 1,200 1,000 800 600 500 400 300

50 40 30 20 10

91 92 93 94 95 96 97 98 99 00 01 02 03 04 05

1/ This monthly volatility measure is based on a 90-day rolling standard deviation. Sources: Bloomberg L.P.; J.P. Morgan Chase; and Authors' calculation.

magnitude of the tightening to come was also reflected in rising spreads. More recently, such tension has been alleviated by Feds rhetoric to remove a policy accommodation at a measured pace.
A.7. Interactive Terms: Dummy Variables Plus Volatility

Because impact of the volatility of the Fed Funds futures market on the spreads appears to be more pronounced under the expectations of rising relative to declining rates, interactive terms, defined as dummy variables of expected tightening/loosening and volatility of Fed Funds futures market, are constructed. These can capture this asymmetric response in the emerging bond market spreads. We conjecture that when the change in expectations is accompanied by less uncertainty, the interactive term might end up insignificant.
A.8. Volatility Index of S&P 500 (VIX)

The Chicago Board Options Exchange (CBOE) Volatility Index, denoted VIX, is based on the S&P 500 options prices. The VIX is often used as a proxy for investors attitude toward risk and appears to explain movements of the emerging market bond spread in recent years. The spread compression seems to coincide with the reduction of the VIX, which is generally interpreted as an increased investors risk appetite.22 However, during the Fed tightening cycle of 1994-1995, the relation between the fluctuations of the VIX and the emerging market bond spread appears to have broken down (Figure 3). In the absence of large price

- 13 -

Figure 3. VIX and Emerging Market Bond Spread

Emerging market bond spread (log scale; basis points; left scale) VIX 1/ (percent; right scale)

2,400 2,000 1,600 1,200 1,000 800 600 500 400 300

40 35 30 25 20 15

91 92 93 94 95 96 97 98 99 00 01 02 03 04 05

10

1/ The SPX Volatility Index, measuring the implied volatility in the prices of a basket of options on the S&P 500.. Sources: Bloomberg L.P.; J.P. Morgan Chase; and Authors' calculation.

fluctuations of the S&P 500 Index, the values of the VIX swung within a narrow range and stayed subdued during the period. In December 2004, the VIX reached a historical low and has been hovering there in recent months, mirroring the decline in bond spreads.
B. Total Credit Rating-Outlook Index (CROI)

One of the main contributions of this paper is to introduce a Total Credit Rating-Outlook Index (CROI) which encompasses both the information of sovereign long-term credit ratings and outlooks to better proxy for underlying fundamentals that influence emerging market bond spreads.
B.1. Log Linearity Between the Spreads and Ratings

Although many studies take a log of the spreads and look at the relationship between the spreads and an index of cardinal numbers assigned to sovereign long-term credit ratings, there is little supporting evidence provided for this specification. Often the index is constructed such that its value increases by some increment as the rating deteriorates, or vice versa. If the spread in log is regressed against an explanatory variable of the credit rating index measured in this way (in levels), the relationship should be interpreted as representing a translog function. In this paper, we begin our exercise by observing a linear relationship in log between the spreads and an index of ratings increasing from 1 to 22 in log scale by an

- 14 Figure 4. Average vis--vis Estimated Bond Spreads on Long-Term Sovereign Credit Ratings 1/
(log scale: basis points)
Average Estimated 2/

6,000 4,000 3,000 2,000 1,200 800 600 400 200 100 50 30 20

AAA + AA -

+ BBB -

+ BB -

+ CCC - CC C SD

1/ Based on the monthly observations during December 1991 through May 2005 for the countries included in the EMBI, EMBI Plus, and EMBI Global. 2/ Estimated spreads are from the univariate regression of the spreads in log on a constant term and a dummy variable of long-term sovereign credit ratings in log. The dummy variable increases by an increment of 1 in log, representing a one-notch decline in the rating. Sources: Bloomberg L.P.; J.P. Morgan Chase; and Authors' calculation.

increment of 1 as the ratings deteriorate from AAA, through AA+, AA, and AA, and all the way down to SD. With this logarithmic scaling, a one-notch decline in the ratings, (i.e. AA+ AA), is represented by an increase of 100 percent in the index level. Figure 4 indicates their relationship in log, which appears linear, implying that there is a nonlinear relationship in level between the spreads and the rating index.
B.2. Construction of the Total Credit Rating-Outlook Index (CROI)

To define the Total Credit Rating-Outlook Index (CROI), we use information on both the long-term credit rating and outlook for a sovereigns foreign currency denominated instruments. We use regression results to estimate the individual response to both the rating and the outlook24 The index value of the CROI in logs is found on Table 2. At each time t for

24

A detail discussion of how to construct the CROI is found in the Appendix 1.B.

- 15 Table 2. Total Credit Rating-Outlook Index (CROI)

(Index value in log) Category: Sovereign Long-term Credit Ratings (R i ) Stable (STB ) 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0 9.0 10.0 11.0 12.0 13.0 14.0 15.0 16.0 17.0 18.0 19.0 20.0 21.0 22.0 Credit Outlook (O ) Positive (POS ) Negative (NEG ) 0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0 9.0 10.1 11.1 12.1 13.1 14.1 15.1 16.1 18.0 19.0 20.0 21.0 22.0 2.7 3.7 4.7 5.7 6.7 7.7 8.7 9.7 10.7 11.7 12.7 13.7 14.7 15.7 16.7 17.7 18.7 18.0 19.0 20.0 21.0 22.0

Investment Grade (G 1 ) AAA AA+ AA AAA+ A ABBB+ BBB BBBNoninvestment Grade: Tier 1 (G 2 ) BB+ BB BBB+ B BCCC+ Noninvestment Grade: Tier 2 (G 3 ) CCC CCCCC C SD Source: Authors' calculations.

ach country i CROI takes on one of the values in the table depending on the configuration of the current rating and the outlook. It is important to recognize that the CROI is an index that, by construction, attempts to calibrate how ratings and outlooks relate to spreadsthat is, it uses (past) information about how spreads react to changes in ratings and outlooks to make this calibration. Table 2 indicates that there are four unique properties embedded into the CROI that reflect various non-linear responses of spreads to changes in ratings or outlooks. For instance, given log linearity, a spread of a country with a credit rating above the 1st tier of non-investment grade category and an outlook of positive is lower than that of other country with a one-notch higher credit rating and an outlook of negative. The CROI does not distinguish between positive and stable, nor negative and stable outlooks for the countries in the 2nd tier of noninvestment grade category. The index values depend entirely on the credit ratings since the addition of outlooks were statistically insignificant in this category. As well, when a countrys outlook changes from stable to positive with its rating remained the same, this improvement is reflected in a larger reduction of the CROI for countries in the investment grade category (-0.97) relative to those in the 1st tier of noninvestment grade category (-0.90).

- 16 On the other hand, deterioration in ones outlook increases the value of the CROI by a larger degree for countries in the 1st tier of noninvestment grade category (2.08) than those in the investment grade category (1.72). And, given a credit rating above the 1st tier of noninvestment grade category, an increase in the CROI responding to a negative outlook is higher in an absolute magnitude than the decline in the index responding to a positive outlook. Using market capitalization of EMBI Global for each member, we aggregated the newly constructed CROI and the Long-term Credit Rating (LTCR). Figure 6 compares the CROI and the LTCR in log level plotted against the EMBI Global in basis points. One of the most striking findings is that CROI appears to be more closely linked to spreads. In fact, during the period of January 1994 through May 2005 the correlation of spreads with the CROI is 84 percent compared to only a 67 percent with the LTCR.
III. RESULTS

We analyze both effects of economic fundamentals and liquidity on the bond spreads. The liquidity effects result from the examination of the following variables; (1) the level of U.S. interest rates, (2) market expectation of near-term tightening or easing, and (3) volatility of the Fed Funds futures market. In this analysis, we make comparisons of the results from regressions based on the newly constructed Total Credit Rating-Outlook Index (CROI) and the standard Long-term Credit Rating (LTCR). We obtain our main results using a fixedeffect panel regression model that is estimated with two-stage least squares (2SLS). The 2SLS method allows use of the newly constructed Total Credit Rating-Outlook Index (CROI) as the dependent variable in the first stage to produce a refined (predicted) ratings variable that captures not only the information within economic, financial, and political fundamentals but also the influence of U.S. interest rates. Because we believe that U.S. interest rates play a

- 17 Figure 6. Aggregate "Fundamentals": Total Credit Rating-Outlook Index (CROI) vis-a-vis Long-term Credit-Rating Index (LTCR)
(EMBI Global: left scale in log,basis points and CROI and LTCR: right scale, index 1/)
Brazilian real devaluation (Jan..1999) Russian ruble devaluation (Aug.1998), LTCM crisis (Sep. 1998)

2,000

Mexican peso devaluation (Dec.1994)

Turkish lira devaluation (Mar.2001)

Argentina's debt moratorium (Dec.2001)

14 13.5 13 12.5

1,000

Thai bath devaluation (Jul.1997)

500

12 11.5

300 94 95 96 97 98 99 00 01 02 03 04 05 11

Long-term Credit Rating (LTCR) Total Credit-Outlook Index (CROI EMBI Global

1/ The higher the index goes the worse the "fundamentals" become. Aggregate "fundamentals" were calculated based on the countries' market capitalizations in EMBI Global. Sources: Bloomberg L.P.; J.P. Morgan Chase; Standard & Poor's, Ratings Direct; and Authors' calculations.

role in determining both countrys default risk (hence the credit ratings and outlooks) and the bond spreads, we can use this technique to control for the effect of interest rates on the predicted values of the fundamentals. By doing so, we can better isolate the liquidity effect from U.S. interest rates on the bond spreads. Moreover, by not using ratings alone as a proxy for fundamentals we are, in principle, able to capture more of the information embedded in ratings in analyzing the sovereign bond spreads. As a starting point, we examine spreads by applying a standard OLS regression using the set of variables outlined above, including the newly constructed Total Credit Rating-Outlook

- 18 index (CROI) as one of the independent variable. For comparison, we also estimate the same OLS model using the long-term credit rating (LTCR) as an independent variable.
A.1. Basic Model

We first estimate the spreads by using a fixed effect (unbalanced) panel regression model,26 using only two explanatory variables: (1) the fundamentals of the CROI (or LTCR) and (2) 3-month ahead U.S. Fed Funds futures rate (FF3M). Table 3 indicates the regression results based on this basic model: yit = Xit + vi + eit where yit is the spread in log (SPRD) and Xit includes the CROI (or LTCR) and FF3M, vi is a country specific factor, and eit is a random error. Although we collected data for the spreads for 33 countries, due to limited data availability on the credit ratings and outlooks, 30 countries are included in the estimation.27
Table 3. OLS Regression Results: CROI vs. LTCR, December 1991 ~ May 2005 1/
Dependent variable: Log of bond spreads (SPRD )
Explanatory variables CROI FF3M Constant R2 within between overall 0.505 0.824 0.715 Coefficient 0.213 0.064 3.085 Standard error 0.004 0.004 0.053 p-value 0.000 0.000 0.000 R2 within between overall 0.445 0.824 0.688 Explanatory variables LTCR FF3M Constant Coefficient 0.212 0.060 3.147 Standard error 0.005 0.004 0.058 p-value 0.000 0.000 0.000

1/ The results are based on the panel regression of fixed-effect model which uses 2766 monthly observations. Sources: Bloomberg, L.P.; J.P. Morgan Chase; The PRS Group, International Country Risk Guide; Standard & Poor's, Ratings Direct; and Authors calculations.

In an aggregate level, the correlation between the SPRD and CROI was significantly higher than that between SPRD and LTCR, however, the regression of the CROI is only 3 percent higher in explaining the variation of the spread (0.715 as oppose to 0.688 for overall R2). Although the coefficient values and R2 statistics are similar, the improvement in the Within R2 reflects the closer time-series association obtainable using the CROI where a countrys

A Hausman test indicates that we cannot reject the null hypothesis of systematic difference between the estimates of the random effect and those of the fixed-effect models. However, because we suspect that a country specific factor is not completely independent of the CROI or the LTCR, we apply the fixed-effect model of panel regression. Algeria, Cote dIvoire, and Nigeria are excluded due to lack of the data on sovereign credit ratings and outlook. A total of 2766 monthly observations are included for estimation.
27

26

- 19 outlook changes more frequently due to financial/economic/political developments, while the credit ratings remain unchanged.
A.2. A Model with Market Expectation and Volatility

Next, we estimate the following fixed-effect panel regression model by the OLS.29
SPRDi,t = 1 + 1FF3Mi + 2E_UPi + 3E_DOWNi + 4V_FFi + 5 (V_FFi *E_UPi) + 6(V_FFi *E_DOWNi) + 7VIXi + 8FUNDAMENTALSi,t + ui + eit

The explanatory variables included in this regression are: 3-month ahead Fed Funds futures rate (FF3M); dummy variables for expectations of a rate rise (E_UP) and of a rate cut (E_DOWN) in the next three months; a volatility of Fed Funds futures market (V_FF) represented by the 90-day rolling standard deviation of the difference in FF3M and FFTR; interactive terms of V_FF and market expectations; the CBOE Volatility Index (VIX) for the S&P 500; and the fundamentals. For the measure of the fundamentals, we again apply both the CROI and the LTCR. The use of the CROI improves the overall fit, yielding a higher R2 of 0.750 relative to 0.734 under the model with LTCR. And again the embedded information in the outlooks better explains the variation of the spreads over time, but does not add much value to the explanation of spreads across countries. The additional explanatory variables improve the R2 We conducted a Hausman test to judge the appropriateness of a fixed effect model. By comparing estimated coefficients with a random and a fixed effect models, our null hypothesis that these estimated coefficients are statistically the same was rejected at 1 percent level of statistical significance. Thus, we conclude that a fixed effect panel regression is appropriate. R2 of Within is based on the panel regression of Within estimator which entails the question about the effect of Xit when Xit changes within each country (i) over time. On the other hand, R2 of Between is based on the Between estimator which entails the question about the effect Xit when Xit changes across country (i).
34 29

- 20 of Within by 13 percent and 16 percent for the regression using the CROI and the LTCR, respectively. Since the additional variables are the same across countries, the R2 of Between deteriorates slightly for both regressions.34

Table 4. Regression Results: CROI vs. LTCR, January 1991 ~ May 2005 1/
Dependent variable: Log of bond spreads (SPRD )
Explanatory variables Fundamentals CROI Other explanatory variables: FF3M E_UP E_DOWN V_FF V_FF*E_UP V_FF*E_DOWN VIX Constant R2 Within Between Overall 0.634 0.817 0.750 Coefficient Standard tp-value error statistics 0.004 54.4 0.000 Explanatory variables Fundamentals LTCR Other explanatory variables: FF3M E_UP E_DOWN V_FF V_FF*E_UP V_FF*E_DOWN VIX Constant R2 Within Between Overall 0.608 0.814 0.734 Coefficient Standard tp-value error statistics 0.004 50.9 0.000

0.197

0.201

0.074 -0.248 0.181 1.581 0.699 -1.140 0.021 2.679

0.004 0.031 0.040 0.226 0.288 0.326 0.001 0.052

19.3 -8.1 4.6 7.0 2.4 -3.5 16.8 51.4

0.000 0.000 0.000 0.000 0.015 0.000 0.000 0.000

0.070 -0.241 0.176 1.768 0.529 -1.131 0.025 2.576

0.004 0.032 0.041 0.233 0.298 0.337 0.001 0.057

17.8 -7.6 4.3 7.6 1.8 -3.4 19.4 45.6

0.000 0.000 0.000 0.000 0.076 0.001 0.000 0.000

1/ The results are based on the panel regression of fixed-effect model which uses 2766 monthly observations. Sources: Bloomberg, L.P.; J.P. Morgan Chase; The PRS Group, International Country Risk Guide; Standard & Poor's, Ratings Direct; and Authors' calculation.

The p-values of explanatory variables suggest that they all be included in the models. The estimated coefficient for the CROI signifies that the spreads will increase by 0.197 in logs when the index value of the CROI increases by 1 (or 100%) in logs. Given a country with a rating of B and outlook of stable, a 3-month ahead Fed Funds futures of 3.5 percent, other explanatory variables remaining at their most recent 12-month averages, and a constant represents a countrys fixed-effect, when its rating falls by a one-notch to B-, ceteris paribus, the countrys spread will go up from an initial spread of 500 basis points to 610 basis points, an increase of 22 percent (0.197 in logs). Similar exercises can be performed but will yield differential effects given the construction of the CROI. When the long-term credit rating index (LTCR) is included instead as a measure of fundamentals, results are in the same direction, but by construction we can only evaluate changes in the ratings.37
37

Changes in ratings are accompanied with changes in outlooks at 50 percent of the time as show in areas (F) and (G) on Figure 1.

- 21 -

The three-month ahead Fed Funds futures rate (FF3M) has estimated coefficients of 0.074 and 0.070 in the regressions using the CROI and the LTCR for fundamentals, respectively. They are both statistically significant at 1 percent level, and their explanatory power in their respective regression is equivalent in magnitude. When participants of the Fed Funds futures market expect there will be a one percent (100 bp) increase in the U.S. policy rate over the next three month window, the spreads today will increase by about 40 bp (7 percent) from an initial spread of 500 bp. Relative to the fundamentals, the level in the expected interest rate contains less explanatory power for spreads. The dummy variables (E_UP and E_DOWN) capture a one-time shift in the level of the spreads based on a change in the interest rate expectation itself, irrespective of the magnitude of the likely changes in the U.S. policy rate. When market participants expect that there will be a rate rise (E_UP) in next three months, in absence of market volatility in Fed Funds futures market, the coefficient of 0.248 for the regression with the CROI implies that the spread will decrease by roughly 25 percent. (A smaller but opposite effect is present for E_DOWN.) While these are counter-intuitive results, it is possible that when an expectation of a rate increase is present, the volatility of the Fed Funds futures market is exacerbated, yielding a much larger positive effect on the spread. Similarly, VIX, the implied volatility of S&P 500, increases the spread as it goes up. Uncertainty about future U.S. monetary policy can be captured by the volatility of the Fed Funds futures rates minus its target rate and can have a significant effect on emerging market bond spreads. In fact, we find that the volatility of the Fed Funds futures market influences spreads to a greater degree than the level of interest rate does. The estimated coefficients of V_FF are 1.581 and 1.768 for the regression of CROI and that of LTCR, respectively. They are both statistically significant at 1 percent level. If volatility of Fed Funds futures market doubles, from 0.1239 to 0.24 in absence of market expectation (E_UP and E_DOWN), the 100 percent increase in V_FF measured in standard deviation will widen the spread by about 19.3 percent (21.5 percent) for the model with the CROI (LTCR). The interactive terms capture a potentially asymmetric response by market participants and differentiate between an expected rate tightening from that of rate easing in the presence of high volatility. Because a high volatility in the Fed Funds futures market during a monetary easing may have less influence on the spread than during a monetary tightening, these interactive terms can usefully distinguish between these scenarios. Table 4 indicates that the estimated coefficients of V_FF*E_UP and V_FF*E_DOWN are both statistically significant. This result implies that the volatility (V_FF*E_UP) in the presence of a rate tightening expectation pushes the spreads higher. The increase partly offsets the impact attributed to a one-time decline of the spread from the expectation itself. Hence the interactive term
39

Twelve percent is an average over the last twelve months of V_FF.

- 22 becomes quite significant in explaining the dynamics of the spread when market expectation comes into play and uncertainty is amplified. Lastly, the CBOE Volatility Index (VIX) captures volatility on the S&P 500 options prices and is viewed as a proxy for investors attitude toward risk. The disconnect between the VIX and the spreads during the U.S. monetary tightening of 1994 through 1995 suggests there may be another such disconnect in the tightening cycle that began at the end of June 2004. Over the entire sample period, however, regardless of the choice of fundamental variables, the estimated coefficient of the VIX displays statistical significance at 1 percent level. An increase of the VIX by one standard deviation will increase the spreads by 0.128 (0.152) in logs in the regression of CROI (LTCR).40 If the VIX surges to the level observed in September 199841, the regression of CROI (LTCR) estimates the spread will leap by 0.495 ( 0.588) in logs while other factors remain constant. In this section, we presented the results from fixed-effect panel regressions run by OLS. There are, however, apparent shortcomings in this model. First, the model fails to precisely estimate how much the fundamentals actually influence the bond spreads. Information represented by credit ratings and outlooks contain qualitative information such as the willingness to pay as well as the ability to pay, with this latter information potentially related to the effect of U.S. interest rates and emerging countries political developments on the probability of default. The estimated coefficient of the fundamentals in this model thus includes effects coming from the level of the interest rate and from developments of political economy, which may lessen the explanatory power of the three-month ahead Fed Funds futures rate. In order to better address these issues, we apply the two stage least squares (2SLS) in the next section.
C. Estimation by Two Stage Lease Squares (2SLS)

To apply the 2SLS model, we proceed first by running a regression on the Total Credit Rating-Outlook Index (CROI) against three measures of fundamentals represented by economic, financial, and political risk rating indicators. These aggregate indicators prove

One standard deviation of the VIX is roughly 6 percent during the sample period of 1991 through 2005. At the dawn of the LTCM crisis on August 31, 1998, S&P 500 Index fell 6.8 percent from the previous day, and the VIX surged to 44.3. The VIX was jittery through the end of October 1998. It reached a historical high of 45.7 October 8, 1998. It is not until recently that many emerging market economies have made higher frequency data available for key macroeconomic time-series.
44 41

40

- 23 particularly useful in that they provide relatively high variation even during the early period of our sample.44 We also include the contemporaneous level of the U.S. Fed Funds target rate as an exogenous variable in the first-stage regression to control for a possible omitted variable bias noted above. Recall that to the extent that rates paid by emerging market borrowers follow industrial country interest rates, a decline in interest rates, ceteris paribus, implies lower debt burdens of emerging market countries and an improvement in fundamentals as measured by debt service ratios, debt-to-GDP ratios, and the likelihood of default. Thus, in observing an improvement in fundamentals the model should control for the interaction between industrial country interest rates and the domestic fundamentals. For the second-stage regression, we used the predicted values of the fundamentals as one of the explanatory variables. As well, to judge the usefulness of the CROI variable we alternatively replace it with the Long-term Credit Rating (LTCR) in the 2SLS estimation.
C.1. First-Stage Regression of 2SLS

In order to effectively utilize our sample of cross-sectional time-series data, we run the following fixed-effect panel regressions for the first stage of the 2SLS.45
CROI i ,t = const + a1FFTRt + a2 ERRi ,t + a3 FRRi ,t + a4 PRRi ,t + ui + ei ,t LTCRi ,t = const + b1 FFTRt + b2 ERRi ,t + b3 FRRi ,t + b4 PRRi ,t + ui + ei ,t

We apply the long-term credit ratings (LTCR) as an alternative choice for the dependent variable and compare the results. The estimated coefficients are found on Table 5. Regardless of the choice of dependent variable, all explanatory variables are statistically significant at 1% level. Financial risk rating indicator (FRR) exhibits the largest coefficient, which is as large as the sum of the estimated coefficients on economic (ERR) and political risk rating (PRR) indicators. When the FRR goes up by one unit (i.e. 30 to 31) and thus fundamentals improve, ceteris paribus, the CROI and the LTCR will come down by 20.0 and 17.8 percent, respectively. Since the dependent variables measure the level of countrys default risk, it appears sensible that the economic fundamentals are somewhat less significant relative to the financial fundamentals. As with the spread OLS regressions, the explanatory power of the economic and financial fundamentals is somewhat larger using the CROI than the LTCR suggesting that information of outlooks embedded in the CROI are more closely linked with the fluctuations of the economic fundamentals and financial soundness. As in the first regression exercise, we conducted a Hausman test to examine the appropriateness of a fixed effect model. The test statistic was significant at the 1 percent level, suggesting a fixed effect panel regression was an appropriate characterization of the data.
45

- 24 -

Table 5. First-Stage Regression: January 1990 ~ May 2005 1/

Total Credit Rating-Outlook Index (CROI ) FFTR Coefficient Standard error t-statistics R2: Within Between Overall -0.121 0.013 -8.98 ERR -0.092 0.008 -10.93 FRR -0.199 0.007 -26.94 0.353 0.741 0.616 PRR -0.131 0.006 -22.93 Const. 31.385 0.455 68.96 Long-term Credit Ratings (LTCR ) FFTR -0.085 0.013 -6.65 ERR -0.061 0.008 -7.65 FRR -0.178 0.007 -25.43 0.311 0.701 0.580 PRR -0.121 0.005 -22.26 28.670 0.433 66.23

1/ Fixed effect panel regression based on 3911 monthly observations. Longer time-series data are available for most countries on economic risk (ERR), financial risk (FRR), and political risk (PRR) indicators. Algeria, Cote d'Ivoire, and Nigeria are excluded due to unavailability of data. Sources: Bloomberg, L.P.; The PRS Group; and Authors' calculation.

The estimated coefficient of the Fed Funds target rate has a negative sign. Our a priori expectation was that higher interest rates would lead to a worsening, not an improvement, in ratings. We speculate that the three fundamental variables may already be incorporating any affect from interest ratesthat is, higher U.S. interest rates worsen the domestic budget deficit and the coefficient on the ERR is statistically capturing the response and thus we could be over-fitting the interest rate variable (obtaining a negative sign). Nevertheless, to prevent the coefficients of the fundamentals from being potentially biased when omitting the variable of FFTR, given its significance, we keep this contemporaneous interest rate in our regression. Overall, three fundamentals and Fed Funds target rate explain 4 percent more of the variation of the CROI than that of the LTCR, 62 versus 58 percent of the overall variation. It is evident that both the ratings and outlooks still depend on a set of qualitative variables including willingness to repay debt.46
C.2. Second-Stage Regression (2SLS)

In the second-stage regression of the 2SLS, we use the predicted values of the Total Credit Rating-Outlook Index (pCROIi.t) and the Long-term Credit Rating (pLTCRi,t) to evaluate how well these predicted values as well as other variables explain fluctuations of the sovereign bond spreads. The second-stage of the 2SLS is also estimated with a fixed-effect panel regression.
46

Willingness to pay is a qualitative issue that distinguishes sovereigns from most other types of issuers. Partly because creditors have only limited legal redress, a government can (and sometimes does) default selectively on its obligations, even when it possesses the financial capacity for timely debt service. (see Standard & Poors (2004)). Venezuelas decision not to repay in January 2005 is a case in point.

- 25 -

SPRDi ,t = const + c1 FF 3M t + c2 E _ UPt + c3 E _ DOWN t + c4V _ FFt + c5 (V _ FFt * E _ UPt ) + c6 (V _ FFt * E _ DOWN t ) + c7VIX t + c8 FUNDAMENTALSi ,t + ui + ei ,t

where FUNDAMENTALSi ,t take one of the following predicted values.

FUNDAMENTALSi,t= pCROIi,t = pLTCRi,t

The explanatory variables included in this regression are the same as the ones included for the OLS estimations. Table 6 shows the results from the second regressions of the 2SLS. We find the larger coefficients on pCROIi.t and the pLTCRi,t.in the 2SLS model, by 37 percent and 52 percent, respectively, than those on CROIi.t and the LTCRi,t in the OLS model. Thus our estimation of the fundamentals in first stage of the 2SLS model better explains spreads than using the raw CROI and LTCR variables directly The other estimated coefficients as well as their significance levels display very similar results in both regressions under 2SLS. When comparing the results of 2SLS and those of OLS, explanatory power of the volatility in the Fed Funds futures market (V_FF) increases considerably, and that of 3-month ahead Fed Funds futures rate (FF3M) goes up slightly under the 2SLS. Explanatory powers of other right hand side variables are somewhat lower in 2SLS. To some extent, this increase in V_FF reduced the explanatory powers of the interactive terms. The variable of V_FF*E_UP becomes not significant even at 10 percent level and V_FF*E_DOWN, on the other hand, remains significant but only at 5 percent level. When market participants expect one percent (100 bp) increase in the U.S. policy rate over the next three-month period, the spreads increase today by 7.7 percent, controlling for interest rates through pCROI. The estimated coefficients of the CBOE Volatility Index (VIX) are the same between the model with pCROI and with pLTCR. An increase of VIX by 1 percent raises the spreads by 1.8 percent. The explanatory powers of the expected tightening (E_UP) and of easing (E_DOWN) are broadly the same in an absolute magnitude regardless of the choice of predicted fundamentals. When an expectation of tightening/easing kicks in there will be a one-time decrease/increase of the spreads with a magnitude of 18 to 20 percent. The 2SLS model would predict higher overall spreads than the OLS model. When using the median values of the underlying explanatory variables 47 to estimate the spread in the second regression of the 2SLS, the estimated spread increases. The estimated spread was around 500 basis points under the OLS when the raw CROI is used as fundamentals. When the

These parameters include a predicted value of the fundamentals that is equivalent to a long-term credit rating of B with a stable outlook, 3-month ahead Fed funds futures of 3.5%, and the last twelve-month averages of VIX and V_FF with a tightening expectation.

47

- 26 Table 6. Second-Stage Regressions that Incorporate Predicted Values of Default Risk, January 1991 ~ May 2005 1/
Dependent variable: Log of bond spreads (SPRD )
Explanatory variables Fundamentals CROI(hat) 2/ Other explanatory variables: FF3M E_UP E_DOWN V_FF V_FF*E_UP V_FF*E_DOWN VIX Constant R
2

Coefficient

Standard tp-value error statistics 0.008 35.4 0.000

Explanatory variables Fundamentals LTCR(hat) 2/

Coefficient

Standard tp-value error statistics 0.009 34.9 0.000

0.270

0.307

0.077 -0.211 0.188 1.897 0.525 -0.853 0.018 1.829

0.005 0.037 0.048 0.270 0.345 0.391 0.001 0.095 0.476 0.769 0.675 -0.025 -0.054 -0.035

16.7 -5.7 3.9 7.0 1.5 -2.2 12.1 19.4

0.000 0.000 0.000 0.000 0.129 0.029 0.000 0.000

Other explanatory variables: FF3M E_UP E_DOWN V_FF V_FF*E_UP V_FF*E_DOWN VIX Constant R
2

0.072 -0.215 0.194 1.916 0.537 -0.915 0.018 1.450

0.005 0.037 0.048 0.271 0.347 0.392 0.002 0.106 0.472 0.767 0.671 -0.019 -0.055 -0.037

15.6 -5.8 4.1 7.1 1.6 -2.3 11.9 13.7

0.000 0.000 0.000 0.000 0.121 0.020 0.000 0.000

Within Between Overall Memorandum (Fundamentals) 3/ Economic Financial Political

Within Between Overall

1/ Results are based on the fixed effect panel regression model which uses 2743 monthly observations. Fewer observations are included as the risk indicators for Croatia during January 1997 through November 1998 are unavailable. 2/ Predicted values of CROI and LTCR from the first-stage regressions are used as "fundamentals". 3/ These coefficients are calculated based on the estimated coefficients of the predicted "fundamentals" multiplied by the estimated coefficients of the economic (ERR), financial (FRR), and political (PRR) risk indicators in the first regression of 2SLS. Sources: Bloomberg, L.P.; J.P. Morgan Chase; The PRS Group, International Country Risk Guide; Standard & Poor's, Ratings Direct; and Authors' calculation.

predicted CROI is used under the 2SLS estimation, the estimated spread increased to 650 basis points. When the LTCR is used, the estimated spread increased from 500 bp under the OLS to 760 bp under the 2SLS. When there is a large and persistent gap existing between the

- 27 estimated spreads under the OLS and the 2SLS, the fundamentals implied by ratings and outlooks are diverging from the true fundamentals. This gap indicates ratings agencies are factoring in an additional or a reduced default risk that appears at odds with what the true fundamentals would otherwise suggest. Such qualitative indicators include the ratings agencies evaluation of the countries willingness to pay and ability to pay which is implicitly in the error term of the first stage regression. The gap can also be created by revisions and alterations in the quality of data released by the national authorities after the contemporaneous rating or outlook has been established.
D. Graphical Interpretation of the Models

To find out more directly how the three models perform, we aggregate the estimated EMBI global spreads, weighted by market capitalizations of the countries in the sample, and compare them against the actual spread. The aggregation is based on 30 countries,50 and an actual spread is based on two indices of EMBI and EMBI Global.51 Panel (A) in Figure 8 seems to suggest that the estimated spread based on the regression using the CROI picks up more than the one with LTCR during periods spread widening. However, even after controlling for the U.S. interest rate, this basic model yet provides a scant explanation of spread widening prior to and during the Tequila crisis of 1995 as well as during major episodes of financial crises and euphoria.
50

Unbalanced panel regression was run by the OLS with a fixed-effect model. Although there are 30 countries included in the regression, the beginning of their spread series varies across countries. For this reason, their market capitalization series start at different periods. The earliest starting point of the series is December 1993. Thus, the OLS as well as 2SLS estimated spreads begin from the end of 1993. Aggregate EMBI series through November 1997, and aggregate EMBI Global series from January 1998 through May 2005, and taking average of the two for December 1997 are spliced as to create an actual series. They are Argentina, Brazil, Mexico, the Philippines, Poland and Venezuela.

51

57

- 28 -

Panel (B) shows that adding the other interest rate related variables, notably the volatility of the Fed Funds rate, and a proxy for risk aversion greatly enhances the overall time-series fit. Nonetheless, the fit is less accurate during the Tequila Crisis and during the Russian financial crisis and the ensuing crisis at Long-Term Capitol Management (LTCM) in the fall of 1998. Panel (C) reflects the result from the 2SLS model. Comparing the estimated spread of 2SLS with that of OLS from panel (B), we find that the predicted values of the CROI and the LTCR provide yet a better time-series fit of the data. It is worth highlighting that the estimated spread in 2SLS reached the historical low in recent months, below actual spreads, while the estimated spread from the simplest OLS framework is above actual spreads. This largely reflects very low volatility of Fed Funds rates and the other interest rate and risk variable (see discussion below). The level of spread compression seen in the recent period resembles the size of compression last seen in 1997. Although our models displayed on panel (B) and (C) seem to explain the actual fluctuations relatively well in recent years, we are still left with its poor fit during 1995, the period following the Mexican peso devaluation in December 1994. When looking at individual countries estimated spreads, we find that several countries experienced significant spreads

- 29 Figure 8. Actual vs. OLS Estimated Spreads

(log scale; basis points) Panel (A) Basic Model ("Fundamentals" and 3-month Ahead Fed Funds futures (FF3M))

1,600 1,400 1,200 1,000 800 600 500 400 300

CROI as "fundamentals" Actual

1,600 1,400 1,200 1,000 800 600 500

LTCR as "fundamentals"

400 300

91 92 93 94 95 96 97 98 99 00 01 02 03 04 05
Panel (B) Model with Market Expectation and Volatility

1,600 1,400 1,200 1,000 800 600 500 400 300

CROI as "fundamentals" Actual

1,600 1,400 1,200 1,000 800 600 500 400

LTCR as "fundamentals"

300

91 92 93 94 95 96 97 98 99 00 01 02 03 04 05
Panel (C) 2SLS

1,600 1,400 1,200 1,000 800 600 500 400 300

CROI as "fundamentals" Actual

1,600 1,400 1,200 1,000 800 600 500 400

LTCR as "fundamentals"

300

91 92 93 94 95 96 97 98 99 00 01 02 03 04 05
Sources: Bloomberg L.P.; J.P. Morgan Chase; PRS Group, International Country Risk Guide; Standard & Poor's, Ratings Direct; and Authors' calculations.

- 30 widening during this time frame.57 In case of Mexico, its spread increased by a factor of six. Our model does not explain widening of their spreads in this magnitude. There are several reasons the model failed to capture the 1995 episode. First, ratings assigned by Standard & Poors (and other rating agencies) tended to lag the event and even then the downgrades were relatively small. In Mexico, for instance, its spread rose to 1900 bp in March 1995 from 300 bp in January 1994, but it wasnt until two days after the devaluation of Mexican peso on December 21, 1994 that Mexicos outlook was finally lowered to credit watch negative from positive while its credit rating remained BB+.58 Mexicos downgrade to BB finally occurred on February 10, 1995. At that time, they revised Mexicos outlook to stable. Second, there were a number of countries whose ratings and outlooks had not been assigned during the 1994-95 episode, yet their Brady bonds were actively traded in the secondary market. The OLS regression models are only run for the relatively small set of countries and so the estimated spread during this period does not include some countries where actual spreads widened substantially. The excluded countries are Bulgaria, Ecuador, Morocco, and Nigeria, with Brazil and the Philippines being added in December and November 1994, respectively. The spotty ratings data and its lack of movement during the period partly explain why the model performed poorly. In any event, the link between actual spreads and the assigned ratings and outlooks during this period appear weak.
D.1. Predicted Fundamentals and Economic, Financial and Political Risk Indicators

It is important to note that the gap of the actual and the estimated spreads in the 2SLS during 1995 is not very different from the gap estimated in the OLS. This suggests that the differences between the evaluations of rating agencies and the predicted fundamentals is insufficient to explain the poor performance of the model during this time period. It is useful to make a comparison among the newly constructed Total Credit Rating-Outlook Index (CROI), the Long-term Credit Rating (LTCR) and the predicted values of the CROI based on economic, financial, and political risk rating indicators and U.S. interest rates. The fundamentals predicted by the CROI depicts a marked deterioration from the end of 1994 through early 1995 as U.S. interest rates rose and the Mexican crisis unfolded, while changes in both the CROI and the LTCR were more benign (Figure 9). Nonetheless, bond spreads widened greatly during the earlier period, far outstripping the deterioration of fundamentalsregardless of how measured. Some have attributed the widening of the spreads in 1994-95 by herding behavior of market participants and/or a common lender who rebalanced their portfolios by precipitously, scaling back the positions from affected
58

An outlook of positive was assigned to Mexico in November 1993, and the credit rating of BB+ had been assigned on July 30, 1992.

See Appendix 1.A. Data Description and Sources for information of what factors make up the political, as well as economic and financial risk indicators.

61

- 31 countries (Kaminsky et. al., 2003). In addition, some claimed credit ratings underpinning countries default probabilities were guided by outdated crisis models and thus their assigned ratings failed to provide early warning signals ahead of a currency crisis (Reisen, 2002). It appears that, indeed, the ratings and outlooks both fail to predict the widening of the spreads during the Mexican crisis period. Figure 9. "Fundamentals" and EMBI Global Spread
Predicted "Fundamentals" (left scale: basis points, right scale: index 1/) Predicted "Fundamentals" (left scale: basis points, right scale: index 1/)

(right scale) 1,600 13.5 13.5 1,400 Predicted 1,200 CROI 2/ 1,000 13 13 800 12.5 12.5 600 500 12 12 EMBI Global 400 Total credit (left scale) Long-term credit 11.5 11.5 rating-outlook ratings (LTCR) 300 index (CROI) 11 11 94 95 96 97 98 99 00 01 02 03 04 05 94 95 96 97 98 99 00 01 02 03 04 05

14

14

Predicted CROI 2/

14 13.5 13 12.5 12 11.5 11

1/ The higher the index goes the worse the "fundamentals" become. Aggregate "fundamentals" were calculated based on the market capitalization of countries in EMBI Global. 2/ Predicted CROI is based on the first-stage regression of the 2SLS. Sources: Bloomberg L.P.; J.P. Morgan Chase; PRS Group, International Country Risk Guide; Standard & Poor's, Ratings Direct; and Authors' calculations.

A considerable improvement in fundamentals has taken place only since the beginning of 2003. The predicted CROI reached an all-time low in April 2005. The EMBI Global spread also reached a record low of 345 bp in February 2005. The right panel in Figure 9 shows that the improvement of the fundamentals significantly lagged behind the compression of the spread. The CROI, as compared to the LTCR, continued to come down through mid 2004 as U.S. monetary accommodation took place. This appears to suggest that the spread compression which had taken place since the 4th quarter of 2002 through the beginning 2004 was driven mainly by the decline of default risks linked to the U.S. interest rates and its liquidity effect in the financial market. Predicted fundamentals of the CROI is based on the economic, financial, and political risk indicators as well as U.S. Fed Funds target rate. Figure 10 exhibits a considerable improvement in all three risk indicators since the beginning 2003. In particular, the financial risk indicator (FRR) shows a marked increase, reaching over a decade high (thus an improvement in financial environment). Similarly, the economic risk indicator (ERR) is at a 10-year high. Despite a recent improvement in political risk rating (PRR), the level we see today is about the same as the level during the Tequila crisis.61

- 32 Figure 10. Aggregate Risk Rating Indicators and the CROI: EMBI Countries 1/
(Index 2/)
Political Risk Rating (PRR)
CROI 3/ (right scale)

Economic (ERR) and Financial Risk Ratings (FRR)

74 72 70 68 66 64 62

14 13.5 13 12.5 12

40 38 36 34 32 30

CROI 3/ (right scale)

Financial Risk Rating (left scale)

14 13.5 13 12.5 12

Political Risk Rating (left scale)

11.5 11

Economic Risk Rating (right scale)

11.5 11

94 95 96 97 98 99 00 01 02 03 04 05

94 95 96 97 98 99 00 01 02 03 04 05

1/ Aggregation is calculated based on market capitalization of EMBI Global Index. 2/ The higher the index goes the lower the risk bocomes. Thus a higher index value corresps to an improvement of fundamentals. Political risk rating goes from 0 to 100. Both economic and financial risk ratings go from 0 to 50. 3/ Total Credit Rating-Outlook Index (CROI) which captures information of ratings and outlooks goes up when a default risk increases. Sources: Bloomberg L.P.; J.P. Morgan Chase; PRS Group, International Country Risk Guide; and Authors' calculations.

D.2. Model Excluding Argentina

During the period between 2004 to early 2005, the 2SLS model overestimates the tightening of the actual spreads by 70.5 basis points on average and as much as 150 basis points in May 2004 when the market started pricing in an initial monetary tightening and the volatility of the Fed Funds futures market picked up largely. Since the predicted CROI reflects the

- 33 fundamentals measured by the three risk indicators, while controlling U.S. interest rates, the estimated spreads based on 2SLS yields a significant gap against the actual spreads when the actual spreads reflect a higher degree of default risks. This entails a case of Argentina whose credit rating has been selective default with a not meaningful outlook since November 2001. Over this period, however, Argentinas fundamentals measured by the economic (ERR), financial (FRR), and political (PRR) risk ratings have improved significantly. Thus, it is meaningful to analyze how our models perform without Argentina. When Argentina is excluded from the estimation the aggregation we obtain has a much better fit. A mean-absolute-error (MAE) during January 2004 through May 2005 declined to 25 basis points from 70.5 bp when the predicted CROI is used in the 2SLS for estimating the spreads. (Table 7) A reduction in the similar magnitude is also obtained when predicted LTCR is used in the 2SLS estimation.

Table 7. Mean-Absolute-Error (MAE) in Aggregated Estimates

OLS 1/ LTCR (1) January 1994 ~ May 2005 (A) All countries (B) Excluding Argentina (2) January 1998 ~ May 2005 (A) All countries (B) Excluding Argentina (3) January 2004 ~ May 2005 (A) All countries (B) Excluding Argentina 156.3 149.8 134.9 119.3 42.6 28.8 CROI 136.2 120.4 112.6 87.5 52.2 31.8 2SLS 2/ LTCR 3/ CROI 3/ 143.7 120.0 112.5 65.8 66.0 23.2 139.1 117.1 109.7 63.6 70.5 25.0

1/ The OLS is applied to the fixed-effect panel regression model that includes explanatory variables of "fundamentals" (LTCR or CROI), 3-month ahead Fed Funds futures rate, dummy variables of market expectations, volatility of Fed Funds futures market, its interactive terms with market expectations, VIX, and a constant term. 2/ The 2SLS is applied to the fixed-effect panel regression model. The first regressions of LTCR and of CROI are estimated on the Fed Funds target rate, economic (ERR), financial (FRR), and political (PRR) risk ratings and a constant. Their predicted values are then used to estimate the spreads with the same set of explanatory variables as in the OLS. 3/ Predicted values are used to estimate the spreads in the second regression of the 2SLS. Source: Authors' calculations.

The better fit obtained by excluding Argentina is attributed not just to the improvement of the fundamentals in recent periods but also to the fact that Argentina effectively decoupled from other emerging financial markets.64 Its weight in the broad index fell to a one-tenth of what it Argentina defaulted on December 23, 2001. Its spread soared to 4500 bp from 2500 bp in November 2001, continued to climb well into 2002 until it finally reached 6800 bp. Although the spread of broad index increased, there was a financial decoupling of other emerging (continued)
64

- 34 was prior to its crisis. However, the very high Argentinean spreads of between 4500 bp and 6000 bp continued to influence the broad index reducing the fit. In most of the remainder of the paper, we estimate the spreads excluding Argentina. When keeping the same set of explanatory variables, Figure 11 displays an improvement of the 2SLS estimates relative to the OLS, especially during the period of 2000. Since Argentina accounted for 20 percent of the EMBI Global Index, when it is excluded, its weight is added proportionately to the weights of other countries in the broad index.

Figure 11. Actual, OLS, and 2SLS Estimated Spreads: Excluding Argentina
(log scale; basis points)
CROI as "fundamentals" (OLS)

1,600 1,400 1,200 1,000 800 600 500 400 300

Actual 1/

1,600 1,400 1,200 1,000 800 600 500

CROI as predicted "fundamentals" (2SLS)

400 300

94 95 96 97 98 99 00 01 02 03 04 05
1/ Actual spread excluding Argentina is calculated based on the market weights Sources: Bloomberg L.P.; J.P. Morgan Chase; PRS Group, International Country Risk Guide; Standard & Poor's, Ratings Direct; and Authors' calculations.

D.3. Goodness of Fit for Individual Country

Appendix Figure 1 displays individual countrys actual as well as estimated spreads. These estimated spreads are based on the models which include the Total Credit Rating-Outlook Index (CROI) as a proxy representing fundamentals. In the application of the OLS, the CROI is directly used as one of the explanatory variables. On the other hand, 2SLS uses predicted value of the CROI which was regressed against economic, financial and political risk rating indicators as well as the U.S. policy rate. markets from Argentina. The Brazilian real, for example, indicates this decoupling as it had recovered and held its value during this time. J.P. Morgan & Chase reduced the weight of Argentina in the broad index to less than 2 percent in 2002 from a 23 percent weight on average between 1994 through 2001. The drop in weight has allowed index-based investors to cut their Argentine exposure dramatically.

- 35 -

Based on goodness of fit for the estimated spreads, we can make several interesting observations. First, those countries where the estimated spreads are below the actual spreads (e.g. Argentina, Brazil, Colombia, Dominican Republic, Ecuador, Lebanon, Panama, the Philippines, and Turkey) are all countries that have had recent balance of payment difficulties and have received IMF assistance. Most of the countries for which the estimated spreads are above the actual spreads (e.g. Bulgaria, Croatia65, Hungary, Poland, and the Ukraine) are emerging East European countries. Among them, there are EU accession countries that have been characterized by heavy capital inflows in recent years as convergence to the euro area is expected. The large number of countries where the estimated spreads track actual spreads closely (e.g. Chile, China, El Salvador, Malaysia, Peru, Russia, South Korea, Thailand, and Uruguay) have not experienced economic difficulties in recent years, suggesting in normal times the model performs relatively well.
Table 8. History of Rating and Outlook: Argentina
Date: 2-Apr-97 22-Jul-99 10-Feb-00 31-Oct-00 14-Nov-00 19-Mar-01 26-Mar-01 8-May-01 6-Jun-01 12-Jul-01 Sovereign Credit Rating Long-term BB BB BB BB BBBBB+ B B BCredit Outlook Stable Negative Stable CW-Neg Stable CW-Neg CW-Neg CW-Neg Negative Negative

Source: Standard & Poor's, Ratings Direct.

Our models do not specifically include any crisis or contagion dummies. Yet, they do explain the actual spreads prior to and during financial crises in some individual countries. For instance, Kaminsky et al. (2003) argue that markets foreshadowed turbulence in Argentina in 2001. Indeed, they simply follow signals of default risk indicated by the ratings and outlooks leading up to the debt moratorium by Argentina on December 23, 2001. The 2SLS estimates show the deterioration of the predicted fundamentals starting in mid-1999, long before the widening of the Argentine spread (Appendix Figure 1). Standard & Poors downgraded Argentines credit ratings four times while reviewing its outlook six times during the period of July 1999 through July 2001 (Table 8). On September 19, 1998, within a month after Russia declared its default, Brazils credit outlook was downgraded to negative from stable (Table 9). The estimated spread by the OLS does a better job than the 2SLS approach tracking the actual spread well for Brazil
65

J.P. Morgan Chase terminated the series for Croatia after June 2004.

- 36 (Appendix Figure 1). It appears the marked widening of the spread in Brazil came from the revised default probability measured by the rating agency, with the outlook downgrade foreshadowing a rating revision on January 14, 1999. At the beginning of Brazils devaluation and its subsequent decision to float its currency, markets again anticipated the rating agencys move as Brazils credit rating was lowered by a one-notch to B+ while holding the Brazils outlook at negative.
Table 9. History of Rating and Outlook: Brazil
Date: 2-Apr-97 10-Sep-98 14-Jan-99 9-Nov-99 Sovereign Credit Rating Long-term BBBBB+ B+ Credit Outlook Stable Negative Negative Stable

Source: Standard & Poor's, Ratings Direct.

For Russia, the model performs less well prior to the 1998 crisis in that predicted fundamentals improve in the pre-crisis months, failing to signal the impending crisis. The earliest signal sent by Standard & Poors was on June 9, 1998 when they lowered Russias credit rating by a one-notch to B+ with a stable outlook assigned (Table 10). Within the week before the Russian crisis began on August 18, 1998, the rating agency lowered Russias credit rating by two notches while also downgrading the outlook to negative. The estimate from 2SLS explains the large pick-up of the spread in October 199866. Yet, it fails to explain a prolonged period of extremely high spread. The difficulties of Long-Term Capitol Management (LTCM) at the end of September 1998 complicate the interpretation of Russias experience. Russia officially went to default on January 27, 1999. However, even before this announcement of default, the spread for Russia had reached the level being equivalent of countries in a default and stayed high for a period of nearly two years. There were seven countries whose spreads surged to the level which neither their default risks measured by the rating agency nor their predicted fundamentals could explain.67 The average risk premium that the markets priced in for these countries was 250 bp. in August, 340 bp. in the month after, and 190 bp in October 1998.68 The contagion triggered by the Russian financial turmoil was not well-captured by the model.
66

As noted earlier, this pick-up can be mainly explained by a large spike in the VIX which is connected to the LTCM crisis.

67

These countries include Croatia, Malaysia, Morocco, South Africa, South Korea, Thailand, and Venezuela.

This risk premium is a difference between the actual spread and the estimated spread on a countrys default risk measured by the rating agency.

68

- 37 -

Table 10. History of Rating and Outlook: Russia

Date: 27-May-98 9-Jun-98 13-Aug-98 17-Aug-98 16-Sep-98 27-Jan-99 Sovereign Credit Rating Long-term BBB+ BCCC CCCSD Credit Outlook CW-Neg Stable Negative Negative Negative NM

Source: Standard & Poor's, Ratings Direct.

E. Forecasting the Spreads

As the monetary tightening cycle continues and industrial countries interest rates rise, calibrating how much of the compression in emerging market spreads was due to improvements in real fundamentals and how much was due to excess liquidity could have important ramifications. The impact of an interest rate rise on spreads may be fairly benign if the lower spreads have been primarily the result of improved fundamentals, but a reversal could be quite abrupt if excessive liquidity were to blame, and could be even more pronounced if the excessive liquidity also led to leveraged positions. To examine this issue, we use the model to forecast spreads based on six sets of assumptions (Table 11) while applying a fixed-effect panel regression model under the 2SLS including 29 countries (excluding Argentina) for the period of January 1994 through May 2005. Scenario 1 includes: (1) no change in fundamentals; (2) Fed funds futures rates predict a tightening of 25 bp on every FOMC meeting through December 2005 and no further tightening beyond, which implies that the policy rate will stay at 4.25 percent during 2006. In addition, futures markets are assumed to predict policy interest rates as accurately as they were predicted in 1999 when markets got it right; and (3) volatility remains at the average of last twelve months ending May 2005. Scenario 2 includes the same set of assumptions (2) and (3) as in

On February 3, 1994, the day before the first tightening of the episode 1, market expected only 50 bp rise by the end of May, 1994. However, the Fed actually tightened the policy rate by 125 bp. Market participants underestimated the tightening by 100 bp. During the second episode, market expected a 38 bp tightening by the end of September 1999. The actual tightening by the Fed was only 50 bp.

70

Table 11. Simulation: Assumptions under Five Scenarios

Fed Funds futures rate 1/ VIX Market Expectation 2/ Volatility in the futures market

Scenarios

Fundamentals

Scenario 1 No changes

Fed raises the target rate by 25 basis points at every FOMC meeting until it reaches 4.25% in December 2005.

Market expects a rate increase through December 2005. There is no further expectation of rate increase/cut through the end of 2006.

Stays at the average of last Stays at the average of last 6 months. 6 months.

Three risk indicators (ERR,FRR,PRR) decline, thus Scenario 2 fundamentals deteriorate, by one Same as above. standard deviation over 18 months Same as above. Same as above. Same as above. Same as above. Same as above.

Same as above.

Scenario 3 Same as above.

Up by one standard deviation over the next 18 months.

- 38 -

Scenario 4 Same as above.

Fed raises the target rate by 25 basis points at every FOMC meeting until it reaches 4.25% in December 2005. Starting on August 8, 2006, Fed will start cutting the rate by a total of 100 bp which implies a rate cut of 25 bp at every FOMC meeting after August. Same as above. Same as above.

Market expects a rate increase Up by one standard through the end of 2005, and then a deviation over the next 18 rate cut starting mid 2006 through months. the end of the year.

Up by one standard deviation over the next 18 months.

Scenario 5

Ddecline by two standard deviations over 18 months Same as above.

Scenario 6

Ddecline by three standard deviations over 18 months

Same as above.

Up by two standard deviations over the next 18 months. Up by three standard deviations over the next 18 months.

Up by two standard deviations over the next 18 months. Up by three standard deviations over the next 18 months.

1/ FOMC meetings will be held on August 9, September 20, November 1, and December 13 in 2005. Their schedule in 2006 is to meet on February 1, March 28, May 10, June 29, August 8, September 20, October 24, and December 12. 2/ When the futures market expects a rate increase through the end of 2005, for example, the dummy variable (E_RAISE ) takes the value of 1 through September 2005. This is because a "market expectation" is based on the three-month ahead Fed Funds futures rate.

- 39 scenario 1, except that deterioration of the fundamentals is allowed by one standard deviation. Scenario 3 includes a volatility of U.S. stock market running up through the end of 2006 by one standard deviation. Other assumptions remain the same as in scenario 2. In scenario 4~6, we alter the assumption of monetary policy and include an expectation of monetary easing from mid 2006. Hence the key interest rate will go up to 4.25 percent by the end of the year, and come down to 3.25 percent by the end of 2006 with a 25 bp cut on each FOMC meeting starting August 2006. Fundamentals are assumed to deteriorate, and volatilities of Fed Funds futures market and U.S. stock market increase by the same unit of one, two, and three standard deviations for scenario 4, 5, and 6, respectively. Based on these six scenarios, the model forecasts the spreads for the next 18 months through December 2006. Comparing the past two tightening cycles, the episode of 1994-1995 had a 300 bp tightening over the 12 month period, compared to a 175 bp tightening over the same period during 1999 through 2000. Moreover, the speed of tightening was much faster during the first episode. In addition, the Fed funds futures market indicates that the participants initially underestimated the magnitude of the tightening over the first three months by a much larger degree during the first cycle.70
E.1. Scenarios 1-3: Tightening of Monetary Policy Through 2005

When we assume there are no changes in the fundamentals of the emerging market economies under scenario 1, yet assume there will be a 100 basis point tightening by the end of 2005, these assumptions imply a marginal improvement of the predicted fundamentals in the 2SLS estimates since the U.S. interest rate enters negatively in the first regression of the 2SLS (Figure 13).
Figure 13. Assumptions of Aggregate "Fundamentals": Excluding Argentina
(Index 1/)

13.5 13 12.5 12 11.5 11

Scenario 1 Actual

14 13.5 13
Scenario 2, 3 Actual Scenario 5 Scenario 6

12.5 12 11.5 94 96 98 00 02 04 06 11
Scenario 4

94

96

98

00

02

04

06

1/ The higher the index goes the worse the "fundamentals" become. Aggregate "fundamentals" were calculated based on the market capitalization of countries in Emerging Market Bond Index Global (EMBIG). Sources: Bloomberg L.P.; J.P. Morgan Chase; PRS Group, International Country Risk Guide; Standard & Poor's, Ratings Direct; and Authors' calculations.

- 40 The predicted fundamentals with other assumptions embedded in the scenario 1 give the aggregate spreads (excluding Argentina) shown in Figure 14. In comparison with the EMBI Global (ex Argentina), the aggregate spread points to an increase of another 70 bp or so by June 2006, three-quarters of which come from the market expectations that the Fed will stop raising rates in December 2005. When the current tightening cycle is expected to end, market participants must re-evaluate their portfolio and make necessary adjustments. At this time, the emerging bond market will likely be put into a first test. The level of interest rate itself has a very limited impact on the spread. What is clear from the model is that even without a change in fundamentals, the accuracy of markets predictions of future interest rates is far important. Repeated failures in predicting future interest rates can build up fairly quickly and easily create an excess volatility in the Fed Funds futures market as seen in the tightening cycle of 1994-1995.

Figure 14. Projected Spreads: Excluding Argentina 1/

(log scale; basis points)
1,600 1,400 1,200 1,000 800 600 500 400 300 94 96 98 00 02 04
Scenario 1 Scenario 3 Scenario 2 Estimated by 2SLS (CROI)

Actual

1,600 1,400 1,200 1,000 800 600 500 400 300 94

Actual

Estimated by 2SLS (CROI)

Scenario 6

Scenario 5

Scenario 4

06

96

98

00

02

04

06

Sources: Bloomberg L.P.; J.P. Morgan Chase; PRS Group, International Country Risk Guide; Standard & Poor's, Ratings Direct; and Authors' calculations.

Scenario 2 includes an assumption that the fundamentals of the emerging market economies deteriorate over the course of next 18 months starting June 2005. This assumption is based on a cumulative deduction of one standard deviation by the end of 2006 in the underlying economic, financial and political risk rating indicators. The deduction contributes to deterioration of the predicted fundamentals, and to bring them to the level being equivalent to that in early 2004 (see Figure 10). The spread is projected to go up by 110 bp to 380 bp by mid 2006 and by additional 20 bp by year-end. Thus, an additional 60 bp comes from deteriorations of the three fundamentals. When these assumptions are accompanied by an uncertainty in the U.S. stock market, scenario 3 displays that the spread will rise by 160 bp to 430 bp by mid 2006. This uncertainty measured by the VIX is assumed to go up by one standard deviation, or equivalently by six percent, by end 2006. This increase will push

- 41 the VIX up to the long-run average71 of 20 percent. As a result, the spread will widen by additional 40 bp by the end of the year. The tension in the U.S. stock market alone will add a total of 70 bp to the aggregated spread. Suppose volatility of the Fed Funds futures market goes up to the level of uncertainty experienced at the beginning of the current tightening cycle and other assumptions remain as in scenario 2. The aggregate spread will go up by 180 bp to 450 bp by mid 2006. By year end, it will go up by an additional 65 bp to 515 bp.72 Uncertainty involved with U.S. monetary policy in this magnitude creates a 110 bp spread widening.73
E.2. Scenario 4~6: Accommodation Following Tightening of Monetary Policy

Scenario 4, 5, and 6 assumes a monetary tightening through the end 2005 by 100 bp to 4.25 percent and a subsequent accommodation starting in August 2006 for a total of 100 bp which brings the Fed Funds target rate back to 3.25 percent in December 2006. The three measures of fundamentals are assumed to deteriorate gradually over time until they reach a total decline of one, two, and three standard deviations under scenarios 4, 5, and 6, respectively. To see how this translates to values of the predicted CROI consider that the CROI reached 11.5 in May 2005, the aggregate emerging market has a long-term credit rating of halfway between BB+ and BB with a stable outlook. At the end of 2006, the predicted CROI goes up to 12.1 under scenario 4, which translates to two notch declines of BB- with a positive outlook or to a one-notch decline of BB with a stable outlook. Similar deteriorations occur in scenario 5 and 6. Not that since the CROI takes information of both credit ratings and outlooks, a value in the CROI is associated with more than one combination of a credit rating and an outlook. Scenario 4 assumes that the VIX as well as volatility in the Fed Funds futures market to increase by one standard deviation by end 2006. The magnitude of uncertainty goes up to two and three standard deviations in scenario 5 and 6, respectively. In scenario 4, the spread goes up by 205 bp to 475 bp by mid 2006. Uncertainty in both the Fed Funds futures market and the U.S. stock market pushes the spread further up by an additional 50 bp. The aggregate spread will reach 525 bp by end 2006. As noted earlier, an increase measured by one standard deviation implies for the VIX to go up only to the long-run average of last 15 years.

This average is based on the monthly average over the period of January 1991 through May 2005.
72 73

71

This scenario is not shown on the figures.

Volatility in the Fed Funds futures market is based on the monthly average of rolling 90day standard deviation of the gap between 3-month ahead Fed Funds futures and the target rates. In this measure, the volatility is 0.08 in May 2005 (Figure 2). When it goes up by two standard deviations (0.13), the volatility reaches the level seen in May 2004.

- 42 In scenario 5, the spread rises to 780 bp by end 2006, a total increase of 510 bp. An increase of two standard deviations in the volatility of the Fed Funds futures market approximates a market anxiety seen at the beginning of the current tightening cycle in mid 2004. An increase in the VIX by the same measure can be compared to a similar increase in September 2001. A three standard deviation increase from the current record low level of the VIX is still far less from the peak seen during the LTCM crisis. However, the rise in this magnitude approximates the event took place in September 1998. Last time we observed a three standard deviation increase in the volatility of the Fed Funds futures market was in January 2001 when the IT bubble burst and the Fed started cutting the interest rate. Scenario 6 projects that the spread will jump by 890 basis points to reach 1160 bp by end 2006. (Figure 14) These models, however, do not account for a situation in which the deterioration of the fundamentals accompanying downgrading of ratings and/or outlooks also changes market perception of default risks. When we include an assumption of outlooks that get downgraded to negative for all countries in December 2004, the aggregate spread is projected to reach 730 basis points, an increase of another 300 basis points by June 2005. If market perceptions were to change, the aggregate spread would likely increase by more than 300 basis points.
IV. CONCLUSIONS

While it difficult to parse out a pure liquidity effect from the model, the use of several interest rate variables clearly have an effect on spreadsand cumulatively this effect is greater than for fundamentals alone. This implies that the Fed can take play a role in reducing the risk of any disruptions in the emerging bonds market. A clear communication strategy by the Fed that helps guide market expectations can promote financial stability by keeping the volatility of the expected U.S. monetary policy low, thus contributing a more modest widening in emerging market spreads when fundamentals start to worsen. While the Fed plays an important role, emerging market economies also have a role. To avoid abrupt increases in spreads they must put policies in place during good times to help insure that their overall fundamentals will not deteriorate. Even when the U.S. interest rate increases, they can still offset any negative impact by implementing good economic policies more generally. The model attempts to control for interest rate effects occurring through fundamentals by using a two-stage-least-squares approach along with a more refined variable for fundamentals. The new variable utilizes not only rating changes (a variable used in previous studies to proxy for country economic fundamentals) but incorporates the outlooks for ratings as well. This adds some, but not a great deal, of explanatory power. The model is explicitly designed as a descriptive model for the determinants of emerging market bond spreads and does not account for the supply-side of the sovereign emerging market bond market. Future research could attempt to model both the demand and supply side of the market to better hone the effects of U.S. interest rates on emerging market bond spreads.

- 43 APPENDIX 1
Appendix 1.A: Data Description and Sources

The data on economic, financial, and political risk variables come from International Country Risk Guide provided by The PRS Group, Inc. They assess risks and assign numerical values for the underlying indicators which make up the aggregate economic, financial, and political risk indicators. These underlying indicators are found on Table 1.A.1. Each variable receives a point based on the level of its risk. While the minimum number of points that can be assigned to each variable is zero, the maximum number of points depends on the weight given to the variable in the overall economic, financial, or political risk assessment. The points assigned to each variable are taken from a fixed scale which the PRS Group predetermined. The higher the rating gets, the lower the risk becomes for a country. While the Composite Risk Rating Index and Political Risk Rating (PRR) takes a numerical value between 0 and 100, both Economic (ERR) and Financial Risk Rating (FRR) take a numerical value between 0 and 50.
Appendix Table 1.A.1: Assessment of Country Risks
Risk rating indices Economic Risk Rating (ERR) GDP Per Head Real GDP Growth Annual Inflation Rate Budget Balance Current Account Balance Financial Risk Rating (FRR) Foreign Debt Debt Service Current Account International Liquidity Exchange Rate Stability Political Risk Rating (PRR) Government Stability Socioeconomic Conditions Investment Profile Internal Conflict External Conflict Corruption Military in Politics Religion Tensions Law and Order Ethnic Tensions Democratic Accountability Bureaucratic Quality Index Index Index Index Index Index Index Index Index Index Index Index % of GDP % of Exports % of Exports Months of Imports % of appreciation or depreciation US$ Percent Percent % of GDP % of GDP Components Unit measured Total points allocated 50 5 10 10 10 15 50 10 10 15 5 10 100 12 12 12 12 12 6 6 6 6 6 6 4

Source: The PRS Group, Inc.; International Country Risk Guide.

- 44 Appendix 1.B: A Procedure of Constructing the CROI

There are four steps required to construct the Total Credit Rating-Outlook Index (CROI). The CROI, by its construction, includes a factor that differentiates prices of bonds by investment/noninvestment grade categories as well as by the negative, stable, and positive outlooks. Thus, it includes critical information of outlooks. Since the CROI integrates these unique properties, the first step is to explain how we embed such features in the index.
Step 1: Instead of applying a unified log-linear relationship between the spreads and the ratings (Ri)74 across the entire rating spectrum, we conducted regression analyses and assessed if there are more than one log-linear relationship across the rating spectrum. We divided the spectrum into three categories: (1) investment grades, denoted G1 ; (2) 1st tier of noninvestment grades, denoted G2 ; and (3) 2nd tier of noninvestment grades, denoted G3 . By creating a set of intercept and slope dummy variables including ones for outlooks, we ran a variance-constrained OLS regression based on a pooled sample to evaluate levels of statistical significance for these three grade categories as well as for the rating and outlook variables.75
SPRD = b1 + b2 ( R ) + b3 (G1 * POS ) + b4 ( R * G1 * POS ) + b5 (G1 * NEG ) + b6 ( R * G1 * NEG ) +b7 (G2 * STB ) + b8 ( R * G2 * STB ) + b9 (G2 * POS ) + b10 ( R * G2 * POS ) + b11 (G2 * NEG ) + b12 ( R * G2 * NEG ) +b13 (G3 * STB ) + b14 ( R * G3 * STB ) + b15 (G3 * POS ) + b16 ( R * G3 * POS ) + b17 (G3 * NEG ) + b18 ( R * G3 * NEG )

When defining investment grades as ratings of AAA through BBB, 1st tier of noninvestment grades as BB+ through CCC+, and the 2nd tier as CCC through SD, F-statistics of joint hypothesis suggest both the investment and the 1st tier of noninvestment grade categories are statistically significant at the 1% level.76 However, the 2nd tier of noninvestment grade category was not. The adjusted R2 reported by this regression was 0.715, compared to 0.676 reported under the univariate regression77. Appendix Table 1.B.1 reports also estimated coefficients as well as standard errors of right-hand side variables. The level of significance for the estimated coefficients suggests that both an intercept and slope dummies for negative outlook under the investment grade category are not statistically significant at the The rating variable (Ri) takes a cardinal number in log such that AAA=1, AA+=2, , and SD=22. Where STB is a dummy variable for a stable outlook, POS for a positive outlook, NEG for a negative outlook, R is for log-term credit ratings in log, and SPRD is log of spreads in basis points. We suppressed (n) for indicating the n-th observation of the sample. We divided the rating spectrum in a number of ways to see what combinations of rating groups provide the best fit and plausible significance levels for estimated coefficients. This univariate regression is based on the long-term credit ratings (R) only, without the three categories.
77 76 75 74

- 45 Appendix Table 1.B.1: Selecting an Underlying Model for the Total Credit RatingOutlook Index (CROI)
(1) Break between G 1 and G 2 : (2) Break between G 2 and G 3 : (1) BBB-/BB+ (2) CCC+/CCC Feasible GLS Variance-Constrained OLS based on the pooling sample Without U.S. Policy Rate Coef. Std. Err. FFTR (1) Investment grades: AAA ~ BBB- (G 1 ) Constant R G1*POS R* G1* POS G1*NEG R* G1* NEG 2.794 0.256 0.716 -0.103 0.135 0.043 0.109 0.013 0.225 0.026 0.271 0.031 0.000 0.000 0.001 0.000 0.618 0.168 p-value With U.S. Policy Rate Coef. Std. Err. 0.082 2.532 0.256 0.692 -0.107 0.072 0.039 0.005 0.105 0.012 0.214 0.025 0.258 0.029 p-value 0.000 0.000 0.000 0.001 0.000 0.781 0.189 With U.S. Policy Rate (Heteroskedasticity corrected) Coef. Std. Err. 0.083 2.783 0.233 0.439 -0.079 0.090 0.035 0.004 0.081 0.009 0.164 0.019 0.192 0.023 p-value 0.000 0.000 0.000 0.007 0.000 0.639 0.117

(2) 1st tier of noninvestment grades: BB+ ~ CCC+ (G 2 ) 1.740 G2*STB -0.128 R* G2* STB 0.499 G2*POS -0.041 R* G2* POS G2*NEG 0.703 -0.029 R* G2* NEG

0.161 0.015 0.251 0.021 0.233 0.020

0.000 0.000 0.047 0.051 0.003 0.146

1.499 -0.111 0.058 -0.015 0.561 -0.020

0.154 0.015 0.240 0.020 0.222 0.019

0.000 0.000 0.810 0.475 0.011 0.295

1.017 -0.068 -0.297 0.021 0.423 0.003

0.119 0.012 0.207 0.017 0.188 0.016

0.000 0.000 0.152 0.230 0.024 0.839

(3) 2nd tier of noninvestment grades: CCC ~ SD (G 3 ) G3*STB n.a. 2/ -0.070 0.024 R* G3* STB n.a. 2/ G3*POS R* G3* POS n.a. 2/ G3*NEG 0.417 0.944 -0.035 0.046 R* G3* NEG R2(Adj.) / Log likelihood RMSE F-stat / Chi^2-stat 1/ Overall "Positive" "Negative" Group 1 (G1) "Positive" and "Negative" "Positive" "Negative" Group 2 & Group 3 Group 2 (G2) "Stable", "Positive", and "Negative" "Stable" "Positive" "Negative" Group 3 (G3) "Stable", "Positive", and "Negative" "Stable" "Positive" "Negative" Chi^2-stat (H0: constant variance) 3/ 0.715 0.504

0.004

0.658 0.448

n.a. 2/ 0.023 n.a. 2/ n.a. 2/ 0.786 0.898 -0.050 0.043 -0.062 0.742 0.480

0.007

0.382 0.246

n.a. 2/ 0.014 n.a. 2/ n.a. 2/ 0.465 0.830 -0.022 0.040 -0.051 -1601.2 n.a. 2/

0.000

0.575 0.589

11.0(***) 38.4(***) 39.4(***) 18.1(***) 46.7(***) 38.1(***) 53.8(***) 91.8(***) 1.98 24.1(***) 3.1(**) 8.3(***) n.a. 2/ 1.81 1.06

15.5(***) 32.8(***) 40.2(***) 30.5(***) 33.1(***) 36.7(***) 52.3(***) 73.1(***) 2.2(*) 20.9(***) 2.9(**) 7.3(***) n.a. 2/ 1.8 25.1(***)

83.9(***) 295.0(***) 179.7(***) 71.7(***) 70.5(***) 468.7(***) 454.5(***) 157.9(***) 2.83 119.4(***) 0.32 n.a. 2/ n.a. 2/ 0.32 n.a. 2/

1/ The number of asterisks indicates the level of significance: *** at 1% level; ** at 5% level, and * at 10% level. 2/ Estimated coefficient is not available. 3/ Breusch-Pagan / Cook-Weisberg test for heteroskedasticity. The null hypothesis is based on a constant variance for the fitted value of the spreads. The level of significance is expressed by the number of asterisks where (***) indicates 1%, (**) 5%, and (*) 10%. Source: Authors' calculations.

- 46 10% level. However, a F-statistics for a joint hypothesis of these two variables to be zero indicates that they are significant at 1% level. Similarly, both an intercept and a slope dummies for negative outlook under the 2nd tier of noninvestment grade category are not significant even at 10% level, in contrast, a F-statistics indicates that they are jointly insignificant, agreeing the individual level of significance. For intercept and slope dummies for a positive outlook under the 1st tier of noninvestment grade category, a joint test statistics suggest they are not statistically significant at 10 percent level. Their individual teststatistics, however, indicate that they are significant at 5 percent level. A chi-square statistics reported under this regression was 1.06, accepting the null hypothesis of no heteroskedasticity.78 Under the hypothesis that liquidity influences the spreads, to avoid any omitted variable biases on the estimated coefficients, we re-ran the regression while controlling for Federal Funds target rate (FFTR). The result from Table 1.B.1 indicates that the regression with FFTR gives a better fit with an adjusted R2 of 0.75. In addition FFTR is statistically significant at 1% level. Both the intercept and the slope dummies for positive outlook under the 1st tier of noninvestment grade category became more highly insignificant, and so did the intercept dummy of negative outlook under investment grade category. Furthermore, a chi-statistics indicates that we reject a null hypothesis of constant variance in favor of heteroskedasticity. Thus, by running a Feasible Generalized Least Squares (FGLS), we corrected for heteroskedasticity. Table 1.B.1 indicates that, contrary to the earlier results, the Group 3, the 2nd tier of noninvestment grade category, became no longer significant at 5 percent level. After excluding the variables that are not statistically significant to explain the spreads, we ran another FGLS and obtained the following estimated result for the next step.
SPRD = 2.78 + 0.08(FFTR) + 0.23(R) + 0.46(G1*POS) 0.08(R*G1*POS) +

0.11(G1*NEG) + 0.03(R*G1*NEG) + 1.00(G2*STB) 0.06(R*G2*STB) + 0.44(G2*NEG)

Step 2: Given the coefficients from the regression result in step 1, we estimated the spreads ( SPRD ) in log based on the combination of all possible credit ratings (R) and outlook (O) on average. Based on the estimated spreads, we calculated unweighted difference of the spreads (DSPRDU) between positive (POS) and stable (STB) outlook and that between negative (NEG) and stable (STB) outlook for each credit rating as follow.
U DSPRDRi ,O = SPRD Ri ,O SPRD Ri , STB

where O takes either positive or negative outlook, and

Based on Breusch-Pagan / Cook-Weisberg test for heteroskedasticity, the null hypothesis is defined as a constant variance for the fitted value of SPRD.

78

- 47 Ri [ AAA, AA+, AA,..., SD] , and R1 = AAA; R2 = AA+;..., R22 = SD

W Step3: We then calculated weighted differences of the spreads ( DSPRDGs ,O ) for each grade

category (Gs), where G1 is an investment grade category, G2 is the 1st tier, and G3 is the 2nd tier of noninvestment grade categories. Our weight (WGTRG,sO ) is based on the number of i observations (OBS) sorted by credit ratings (Ri) and outlook (O) under the category of Gs. For example, for the investment grade category,
WGT
G1 Ri ,O

G G OBS Ri1,O + OBS Ri1, STB

OBS
i =1

10

G1 Ri ,O

G + OBS Ri1, STB i =1

10

Weight of Ri when the rating (Ri) is in G1 of the investment grade category

where

R1 , R2 ,K , R10 G1 = investment grade R11 , R12 ,K , R17 G2 = noninvestment grade, tier 1 R18 , R19 ,K , R22 G3 = noninvestment grade, teir 2

W The weighted differences of the spreads ( DSPRDGs ,O ) are calculated as:

W U DSPRDG1 ,O = DSPRDRi ,O *WGTRG1,O = i i =1

10

Weighted difference of the s spreads for the investment grade category (G1)

R1 , R2 ,K , R10 G1 = investment grade R11 , R12 ,K , R17 G2 = noninvestment grade, tier 1 R18 , R19 ,K , R22 G3 = noninvestment grade, teir 2

For the investment grade (G1) category, we obtained the following estimates in log.
W DSPRDG1 , POS = 0.226 W DSPRDG1 , NEG = 0.398

For the 1st tier of noninvestment grade (G2) category, the estimates are:
W DSPRDG2 , POS = 0.150 W DSPRDG2 , NEG = 0.346

For the 2nd tier of noninvestment grade (G3) category, the estimates are:

- 48 W DSPRDG3 , POS = 0 W DSPRDG3 , NEG = 0

These estimates signify discount/penalty factors based on a given outlook under the three grade categories. A discount of 0.226 given to countries with a positive outlook in the investment grade (G1) category implies that their spreads are lower by 20.6 percent relative to other countries with a stable outlook in the same grade category. Similarly, a penalty of 0.346 given to countries with a negative outlook in the 1st tier of noninvestment grade (G2) category implies that their spreads are higher by 34.6 percent relative to other countries with a stable outlook in the same grade category. In contrast, a positive outlook in the 2nd tier of noninvestment grade category does not bear a discount factor, and neither does a negative outlook a penalty factor. In addition, the discount/penalty factors account for asymmetric responses by the investors. The widening of the spreads in response to a negative outlook is larger in an absolute magnitude than the spread tightening in response to a positive outlook. Investors, instead of handsomely rewarding countries with positive outlook, penalize them harshly with a negative outlook.
Step 4: These discount/penalty factors feed into a calculation of adjusted spreads for each rating (Ri) with outlooks (O) of both positive and negative. While we use the estimated
Adj spreads ( SPRD Ri , STB ) for a stable outlook from step 1, the adjusted spreads ( SPRDRi ,O ) are defined as follow.

Adj W SPRDRi ,O = SPRD Ri , STB + DSPRDGs ,O where R1 , R2 ,K , R10 G1 = investment grade R11 , R12 ,K , R17 G2 = noninvestment grade, tier 1 R18 , R19 ,K , R22 G3 = noninvestment grade, teir 2

Using these adjusted spreads and the estimated coefficients of the regression from step 1, we calculate the values of ratings (Ri). These values are, in essence, the index values of the Total Credit Rating-Outlook Index (CROI). Thus, the CROI takes the following form when an outlook is stable.
CROI Ri , STB = i

(i = 1, 2,..., 22)
When an outlook (O) is either positive or negative, the CROI takes the following form.
CROI Ri ,O = i + FGs ,O

where

- 49 FG1 , POS = 0.97, FG1 , NEG = 1.72 FG2 , POS = 0.90, FG2 , NEG = 2.08 FG3 , POS = 0, FG3 , NEG = 0

Table 2 shows values of the CROI based on the formulae above. The initial CROI is altered to incorporate the four unique properties that are attributed to the discount/penalty factors calculated earlier. These properties are discussed in the section B.1.

- 50 REFERENCES

Eichengreen, B., and A. Mody, 1998, What Explains Changing Spreads on Emerging Debt: Fundamentals or Market Sentiment? NBER Working Paper 6408 (Cambridge, Mass.: National Bureau of Economic Research). Erb, C., Harvey, C.R., Viskanta, T., 2000, Understanding Emerging Market Bonds, Emerging Markets Quarterly. Spring 2000, pp. 7-23. IMF, 2004, Appendix 1: Determinants of the Rally in Emerging Market DebtLiquidity and Fundamentals, Global Financial Stability Report; World Economic and Financial Surveys (Washington: International Monetary Fund, April), pp. 60-70. Kamin, S.B., and K. von Kliest, 1999, The Evolution and Determinants of Emerging Market Credit Spreads in the 1990s, BIS Working Paper 68 (Basle: Bank for International Settlements). Kaminsky, G., C. Reinhart, and C. Vegh, 2003, The Unholy Trinity of Financial Contagion, Journal of Economic Perspectives. Fall 2003, vol. 17, no. 4, pp. 51-74. Kaminsky, G., C. Reinhart, and C. Vegh, 2004, When It Rains, It Pours: Procyclical Capital Flows and Macroeconomic Policies, NBER Working Paper 10780 (Cambridge, Mass.: National Bureau of Economic Research). Kuttner, Kenneth, N. 2001. Monetary Policy Surprises and Interest Rates: Evidence from the Fed Funds Futures Market. Journal of Monetary Economics, 47 (2001) 523-544 McGuire, P., and M. Schrijvers, 2003, Common Factors in Emerging Market Spreads, BIS Quarterly Review, (Basel: Bank of International Settlements). Reisen, H., 2002, Ratings since the Asian Crisis, World Institute for Development Economics Research (WIDER), United Nations University, January 2002, Discussion Paper No. 2002/2. Slok, T. and M. Kennedy, 2004, Factors Driving Risk Premia OECD Working Paper No. 385, (Paris: Organization for Economic Co-operation and Development). Standard & Poors, 2004, Sovereign Credit Rating: A Primer, RatingsDirect, The McGraw-Hill Companies, March 15, 2004, New York. Standard & Poors, 2002, Standard & Poors Ratings Definitions, RatingsDirect, The McGraw-Hill Companies, New York.

- 51 Sy, A., 2002, Emerging Market Bond Spreads and Sovereign Credit Ratings: Reconciling Market Views with Economic Fundamentals, Emerging Market Review, Vol. 3, Issue 4, pp 380-408. The PRS Group, Inc., 2004, The International Country Risk Guide Database, New York. Wilson, Peter Venezuela Debt Rating Cut to Selective Default by S&P, Caracas, Venezuela, January 18, 2005.

Appendix Figure 1. Actual, OLS, and 2SLS Estimated Spreads:

Argentina
3,000 3,000 2,000 5,000 2,000 3,000 2,000 1,000 1,000 500 500 300 200 300 91 92 93 94 95 96 97 98 99 00 01 02 03 04 05 300 1,000

(log scale; basis points)

Brazil

5,000

3,000

2SLS (CROI)

OLS (CROI)

2,000

1,000

500

Actual

500

300

200

91 92 93 94 95 96 97 98 99 00 01 02 03 04 05

Bulgaria
5,000 500 2,000 300 200 1,000 500 200 100 50 50 100

Chile
500

5,000

2,000

300 200

1,000

500

200

100

100

50

91 92 93 94 95 96 97 98 99 00 01 02 03 04 05

91 92 93 94 95 96 97 98 99 00 01 02 03 04 05

50

Sources: Bloomberg L.P.; J.P. Morgan Chase; PRS Group, International Country Risk Guide; Standard & Poor's, Ratings Direct; and Authors' calculations.

-3-

Appendix Figure 1. (continued) Actual, OLS, and 2SLS Estimated Spreads:

China 2SLS (CROI) OLS (CROI)
300 1,000 200 500 100 300 200 500 300 200 1,000 500 2,000 2,000

(log scale; basis points)

Colombia

500

300

200

100

50

Actual
30 100

50

30

91 92 93 94 95 96 97 98 99 00 01 02 03 04 05

91 92 93 94 95 96 97 98 99 00 01 02 03 04 05

100

Croatia 1/
1,000 2,000

Dominican Republic
2,000

1,000

500 1,000 300 200 500

500 1,000

300

200

500

100

100

300 50 200

300 200 91 92 93 94 95 96 97 98 99 00 01 02 03 04 05

50

91 92 93 94 95 96 97 98 99 00 01 02 03 04 05

1/ The actual srpead is terminated after June 2004.

Sources: Bloomberg L.P.; J.P. Morgan Chase; PRS Group, International Country Risk Guide; Standard & Poor's, Ratings Direct; and Authors' calculations.

-4-

Appendix Figure 1. (continued) Actual, OLS, and 2SLS Estimated Spreads:

Ecuador 2SLS (CROI)
5,000 3,000 500 300 2,000 300 200 1,000 1,000 500

(log scale; basis points)

Egypt

5,000

3,000

2,000

Actual
1,000 100 200

1,000

500 500 300 50

OLS (CROI)

100

300

91 92 93 94 95 96 97 98 99 00 01 02 03 04 05

91 92 93 94 95 96 97 98 99 00 01 02 03 04 05

50

El Salvador
700 500 300 200 100 50 30 20 100 10 500

Hungary
500 300 200 100 50 30 20 10 91 92 93 94 95 96 97 98 99 00 01 02 03 04 05

700

500

300 200

300

200

100

91 92 93 94 95 96 97 98 99 00 01 02 03 04 05

Sources: Bloomberg L.P.; J.P. Morgan Chase; PRS Group, International Country Risk Guide; Standard & Poor's, Ratings Direct; and Authors' calculations.

-5-

Appendix Figure 1. (continued) Actual, OLS, and 2SLS Estimated Spreads:

Lebanon
2,000 2,000 1,000 1,000 500 500 200 100 50 20 300 500 200 100 50 20 1,000 2,000

(log scale; basis points)

Malaysia

2,000

1,000

2SLS (CROI) Actual

500

300

200

OLS (CROI)
100

200

100

91 92 93 94 95 96 97 98 99 00 01 02 03 04 05

91 92 93 94 95 96 97 98 99 00 01 02 03 04 05

Mexico
3,000 2,000 1,000 1,000 500 300 200 100 500 300 200 2,000

Morocco
2,000

3,000

2,000

1,000

1,000

500

500 300 200

300

200

100

91 92 93 94 95 96 97 98 99 00 01 02 03 04 05

100

91 92 93 94 95 96 97 98 99 00 01 02 03 04 05

100

Sources: Bloomberg L.P.; J.P. Morgan Chase; PRS Group, International Country Risk Guide; Standard & Poor's, Ratings Direct; and Authors' calculations.

-6-

Appendix Figure 1. (continued) Actual, OLS, and 2SLS Estimated Spreads:

Pakistan 1/
5,000 700 600 500 400 500 400 600 3,000 700

(log scale; basis points)

Panama

5,000

3,000

OLS (CROI) Actual

2,000 1,000 500

2,000

1,000

500

300 200 100 200

200

2SLS (CROI)
300

300

300

100

91 92 93 94 95 96 97 98 99 00 01 02 03 04 05

91 92 93 94 95 96 97 98 99 00 01 02 03 04 05

200

Peru
2,000 2,000

Philippines 1/
2,000

2,000

1,000 1,000 500 500 300 200

1,000

1,000

500 300 200

500

300

300 200

200

91 92 93 94 95 96 97 98 99 00 01 02 03 04 05

100

91 92 93 94 95 96 97 98 99 00 01 02 03 04 05

100

1/ The data is missing between March 2002 and March 2004 for Pakistan and February 1997 and October 2001 for the Philippines.

Sources: Bloomberg L.P.; J.P. Morgan Chase; PRS Group, International Country Risk Guide; Standard & Poor's, Ratings Direct; and Authors' calculations.

-7-

Appendix Figure 1. (continued) Actual, OLS, and 2SLS Estimated Spreads:

Poland
2,000 10,000 5,000 3,000 2,000 1,000 500 300 200 100 91 92 93 94 95 96 97 98 99 00 01 02 03 04 05 3,000 2,000 1,000 500 300 200 100 5,000 10,000

(log scale; basis points)

Russia

2,000

1,000

Actual 2SLS (CROI)

500 200 100 1,000

500

200

100

50 20

OLS (CROI)
50

20

91 92 93 94 95 96 97 98 99 00 01 02 03 04 05

South Africa
1,000 500 300 200 100 100 50 50 30 1,000

South Korea 1/
1,000 500 300 200 100 50 91 92 93 94 95 96 97 98 99 00 01 02 03 04 05 30

1,000

500 300 200

500

300

200

100

50

91 92 93 94 95 96 97 98 99 00 01 02 03 04 05

1/ The actual spread is terminated after April 2004.

Sources: Bloomberg L.P.; J.P. Morgan Chase; PRS Group, International Country Risk Guide; Standard & Poor's, Ratings Direct; and Authors' calculations.

-8-

Appendix Figure 1. (continued) Actual, OLS, and 2SLS Estimated Spreads:

Thailand Actual 2SLS (CROI)
300 200 300 200 100 100 50 30 50 91 92 93 94 95 96 97 98 99 00 01 02 03 04 05 50 1,000 500 500 500

(log scale; basis points)

Tunisia

1,000

500

300

OLS (CROI)

300 200

200

100

100

50

30

91 92 93 94 95 96 97 98 99 00 01 02 03 04 05

Turkey
2,000 3,000 2,000 1,000 500 500 200 100 50 300 200 100 1,000

Ukraine
3,000 2,000 1,000 500 300 200 100 91 92 93 94 95 96 97 98 99 00 01 02 03 04 05

2,000

1,000

500

200

100

50

91 92 93 94 95 96 97 98 99 00 01 02 03 04 05

Sources: Bloomberg L.P.; J.P. Morgan Chase; PRS Group, International Country Risk Guide; Standard & Poor's, Ratings Direct; and Authors' calculations.

-9-

Appendix Figure 1. (continued) Actual, OLS, and 2SLS Estimated Spreads:

Uruguay Actual
2,000 2,000 2,000 1,000 1,000 500 300 200 300 200 91 92 93 94 95 96 97 98 99 00 01 02 03 04 05 100 500 500 300 200 1,000 3,000 3,000 3,000

(log scale; basis points)

Venezuela

3,000

2,000

1,000

2SLS (CROI)

500

300

200

OLS (CROI)

100

91 92 93 94 95 96 97 98 99 00 01 02 03 04 05

Sources: Bloomberg L.P.; J.P. Morgan Chase; PRS Group, International Country Risk Guide; Standard & Poor's, Ratings Direct; and Authors' calculations.