Sovereign credit rating Assignment | Essay Help Services

Description

Write a critical research paper using secondary data, the paper has to show critical writing skills, it has to cover : . international business management • systematically review and evaluate • emerging trends • critique theories, models and frameworks • empirical basis • identify knowledge gaps Style (5000 words) : Title • Abstract (300) • Introduction (400) • Research Methodology (600) • Main Body (sections/sub-sections) ( 3000) • Conclusion (650) • References (30 academic references)

Emerging Markets Finance and Trade
ISSN: 1540-496X (Print) 1558-0938 (Online) Journal homepage: https://www.tandfonline.com/loi/mree20
Sovereign Risk Ratings: Biased Toward Developed
Countries?
Derya Gültekin-Karakaş , Mehtap Hisarcıklılar & Hüseyin Öztürk
To cite this article: Derya Gültekin-Karakaş , Mehtap Hisarcıklılar & Hüseyin Öztürk (2011)
Sovereign Risk Ratings: Biased Toward Developed Countries?, Emerging Markets Finance and
Trade, 47:sup2, 69-87, DOI: 10.2753/REE1540-496X4703S204
To link to this article: https://doi.org/10.2753/REE1540-496X4703S204
Published online: 07 Dec 2014.
Submit your article to this journal
Article views: 142
View related articles
Citing articles: 17 View citing articles
Emerging Markets Finance & Trade / May–June 2011, Vol. 47, Supplement 2, pp. 69–87.
Copyright © 2011 M.E. Sharpe, Inc. All rights reserved.
1540-496X/2011 $9.50 + 0.00.
DOI 10.2753/REE1540-496X4703S204
Sovereign Risk Ratings: Biased Toward Developed
Countries?
Derya Gültekin-Karakas*, Mehtap Hisarcıklılar, and Hüseyin Öztürk
Abstract: Sovereign credit ratings are widely used measurements for “country risk” in
international capital markets. However, they have been exposed to increasing criticism in
the aftermath of the recent global financial crises. Many international authorities proposed
new frameworks for the regulation and supervision of the credit rating sector, in which
many countries have taken various steps in this respect. Yet the reliability of sovereign
credit ratings has not been criticized in the literature. Using random effects ordered probit
modeling, this study explores the reliability of credit ratings. Separate analyses of developed
and developing countries suggest that the consistency of credit ratings differs by favoring
the developed country group.
KEYWORDS: credit-rating agencies, ordered logit, ordered probit, political risk, risk premium,
sovereign rating.
The internationalization of financial markets for the past few decades has dramatically expanded,
and diversified investment possibilities in these markets are accompanied by new
risks. Measuring these risks has been the core challenge prior to any financial investment
decision. Financing with minimum risk has brought about the necessity of a benchmark
risk measurement that the credit-rating agencies (CRAs) currently undertake.
Sovereign credit ratings (SCRs) have a core importance in the credit-rating industry.
Yet the CRAs have been intensely criticized due to misalignments in the SCRs. The
debates have intensified as the SCRs are also generally the benchmark for all other
debts in a national economy, so rating changes affect all the investment decisions in
a country regardless of their being financial or otherwise. While the SCRs allow for a
comparison among peer country groups, many think that the CRAs do not concentrate
on the financial or general economic performance of a country as indicated by the rate of
economic growth, the balance of foreign trade, and the public budget as well as the level
of foreign exchange reserves. Rather, the value judgments of the analysts or politics are
of core importance (Akıncılar 2008; Sandström 2008). Nevertheless, because the SCRs
are allegedly indicative of the performance of a national economy, changes in ratings,
especially downgrades, frequently trigger a negative market response.
Derya Gültekin-Karakas* (dkaraka@itu.edu.tr) is an assistant professor at the Faculty of Management,
Istanbul Technical University. Mehtap Hisarcıklılar (hisarciklilar@itu.edu.tr) is an assistant
professor at the Faculty of Management, Istanbul Technical University. Hüseyin Öztürk (huseyin.
ozturk@hazine.gov.tr), the corresponding author, is an expert in the International Capital Markets
Department, General Directorate of Foreign Economic Relations, Undersecretariat of Treasury,
Republic of Turkey. The opinions expressed in this paper are those of the authors and do not necessarily
reflect the opinions of the Undersecretariat of Turkish Treasury. A preliminary version of the
paper was presented at EuroConference 2010 of the Society for the Study of Emerging Markets,
Milas, Turkey, July 16–18, 2010, and at Eurasia Business and Economics Society 2010 Conference,
Istanbul, Turkey, May 26–28, 2010. The authors thank the participants in the conferences.
70 Emerging Markets Finance & Trade
Since the mid-1980s, the CRAs have become very influential in the process of accessing
funds in international capital markets by governments and corporations because
bond ratings highly affect international capital flows by supplying and confirming information
to investors. Of course, the role of these intermediaries has never been without
controversy. Among the many widely debated issues regarding the CRAs, the following
controversial ones deserve close attention: conflict of interest, lack of transparency, and
reliability. Many steps have been taken at the national and international levels to address
the conflict of interest and lack of transparency issues.1 Yet the question of reliability
merits serious discussion.
This fact has inspired researchers to revisit the issue. The determinants of sovereign
ratings have been a core subject in many studies (e.g., Alexe et al. 2003; Bennell et al.
2006; Bissoondoyal-Bheenick et al. 2005; Hu et al. 2002).2 Among these, Alexe et al.
(2003) discussed the issue of country risk and developed a nonrecursive regression model
to verify country grades and assign a rating for unrated ones. Their study has used Standard
and Poor’s and Moody’s ratings, and the authors concluded that the results of the
empirical analyses are in line with the actual ratings. Bennell et al. (2006) also discussed
the issue by comparing the artificial neural network (ANN) method with ordered probit
modeling. Their conclusion is that the ANN model has a higher predictive power than
the ordered probit modeling.
These and many other studies estimate all the SCRs using a single model. Our study, in
contrast to previous studies, examines the major determinants of credit ratings by estimating
two separate ordered probit models for developed and emerging market economies.
Such a two-modeled analysis thereby gives us an opportunity to contrast developed and
emerging market economies in order to find out if the CRAs are biased toward developed
countries. In the analysis, we use Moody’s data for the period 1999–2010. Given the fact
that the CRAs use roughly the same metrics, the use of Moody’s data will be fair enough
to capture the overall picture.
In this study, we will also shed light on the performance of the estimation results over
the two estimated models for developed and emerging economies. We will then compare
the predictions with the actual ratings. This comparison will be an interesting topic for
further studies because emerging market cases have been intensely discussed since the
late 1990s, for example, the Mexican, Malaysian, and Russian cases. In today’s global
economic conjuncture, macroeconomic indicators of the developed countries are getting
worse. Hence, by focusing on the credit ratings of the developed countries, we will be
able to evaluate the CRAs in terms of their timely reaction to changing conditions regardless
of their political support for a specific country or the classification of a country as a
developed one. So the emphasis will also be on a comparison of developed and emerging
country classifications in terms of their SCRs.3
Sovereign Rating
An SCR signals the creditworthiness of a sovereign or its debt issues. The rating is not a
guide for other aspects that are crucial for investment decisions, such as market liquidity
or price volatility. As a result, bonds with the same rating may have very different market
prices. Yet market participants have often treated similarly rated securities as generally
fungible, that is, the credit ratings are used as a calibration metric for market participants,
especially for portfolio management. According to the sovereign methodology, the credit
risk of a country is scaled with letter symbols and thereby the credit risk of a country is
May–June 2011 Supplement 71
consolidated. Governments generally seek higher credit ratings not only because these
can ease their access to international capital markets but also because these assessments
affect the ratings of other borrowers in a national economy (Larraín et al. 1997). That is
to say, a country’s risk is also a benchmark risk indicator for other borrowers that transact
in that country.4 Sovereign risk indicates not only the credit risk of government debts
denominated in foreign currency but also the debts of semigovernmental institutions and
private sector, that is, the whole domestic economy.
SCRs may reduce information asymmetries by giving information on the rated security.
They can also help to solve some principal-agent problems, such as capping the amount
of risk that the agent, that is, the investor, can take on behalf of the principal/issuer (Katz
et al. 2009). In addition, ratings can solve collective action problems of dispersed debt
investors by helping them to monitor performances via downgrades/upgrades that serve
as signals to take action. Beyond all these economic functions that SCRs play, their main
purpose is to compress various information about a country into a single parameter that
is easily comprehendible. In this way, credit ratings provide an efficient opportunity for
cross-country comparisons, that is to say, a credit rating is nothing but an aggregation of
various indicators into a single metric. According to Kunczik (2000), this process is similar
to taking a picture of a country at a specific time with respect to its financial, economic,
and political conditions. Indeed, a rating from a recognized CRA, while not intended to
do so, effectively reduces the burden on investors to examine the creditworthiness of a
security or issuer. SCRs are typically among the main tools used by portfolio managers
in their investment decisions and by lenders in their credit decisions.
Allegedly the SCRs are beneficial tools for measuring the credit risk of a country or a
corporation located within that country. However, there is discontentment with the basic
characteristics of the rating process applied by the CRAs. First, the SCRs published by
different CRAs appear as an output of a black box, the workings of which are unknown
to both issuers and investors. The processing of various and distinct variables into a single
metric seems complicated to issuers and investors. Moreover, the rules of compression
are not open to criticism. Although it is now evident that the CRAs allegedly use the
average weight of a variable in a unique model, they can apparently risk missing salient
improvements in that specific variable. The CRAs assert that beyond their quantitative
analyses, specific attention is given to qualitative factors, such as political factors and
projections. All in all, the process of assigning a credit rating is allegedly opaque to issuers
who are exposed to the direct effects of credit ratings. It seems strange to many to
use multiple variables in order to estimate the possibility and the capacity of a country
to pay back its debt liabilities in due course (Alexe et al. 2003; Brewer and Rivoli 1990;
Haque et al. 1997). In other words, the system of assigning an SCR seems to be reductionist
and oversimplistic.
The second critical issue that is debatable is the poor predictive power of the SCRs. The
CRAs were conspicuous among the many who failed to predict the sovereign-originated
crises in Mexico, Russia, and Asia. These crises have thus produced the sentiment that the
CRAs react to events rather than anticipate them. This procyclicality raises questions about
how seriously investors should take sovereign ratings on developing/emerging market
economies as an indication for sovereign credit risk (Larraín et al. 1997). According to
Kaminsky and Schmukler (2002), credit-rating decisions lag the market behavior, that
is, the CRAs behave procyclically when they downgrade or upgrade. Rating agencies
decide to downgrade (upgrade) a country when the prices of its financial instruments
go down (up). Many have concluded that the CRAs have abstained from downgrading
72 Emerging Markets Finance & Trade
SCRs in a timely manner because of the possible damage to their income streams that
they earn from sovereigns. Also, some have asserted that recent downgrades stem from
the fact that CRAs abstain from self-fulfilling crises in hard times by initiating timely
downgrades. There are also others who point out the lack of qualified staff in the CRAs
that can perform reliable analyses as the main cause of lagging behavior.
The CRAs have been blamed for the flawed credit ratings of developing/emerging
market economies that demonstrate a lack of foresight. Currently, the issue is reversed,
and the market has had doubts about the higher credit ratings of developed countries.5
Whether the concern is the credit ratings of developed countries or of the emerging market
economies, there is discomfort among market players regarding SCRs.
Last but not least, there are allegations that some CRAs have regional biases. Accordingly,
certain CRAs favor particular regions. Haque et al. (1997) note that Euromoney
usually gives higher credit ratings to Asian and European countries than to Latin or
Caribbean countries, while Institutional Investor is more generous to Asian and European
countries than to African countries.
Despite the pros and cons of the CRAs in the credit-rating market, credit ratings are, as
the CRAs contend, mere opinions and not recommendations to purchase, sell, or hold any
security. In the United States, the CRAs assert that they have the same status as financial
journalists and are, therefore, protected by the constitutional guarantee of freedom of
the press (Partnoy 1999, 2001). They claim that this protection precludes government
regulation of the content of a credit rating or the underlying methodology, that is, a credit
rating is a mere opinion and there is freedom of expression. While similar protection does
not exist in other countries, the CRAs have generally stated, in contrast to credit-rating
users, that their opinions do not represent financial advice. This has traditionally kept
them away from investor litigation and until recently prevented direct regulation of their
operations. All in all, the CRAs managed to insulate themselves from litigation or market
criticism by announcing that they were only expressing their views.
Even though the SCR is not tendered as financial advice to buy or sell, the flaws of the
ratings may produce extra cost by not showing the actual risk premium of the sovereign
and the whole national economy. Needless to say, the reliability of SCRs is a critical issue
in country risk measurement. Yet the problem is not limited to the national scale; the
international financial system is also exposed to risks stemming from flawed SCRs. Unless
a fair analysis is performed, the cost borne is not only by a national economy but also by
the entire international financial system, as seen in the previous financial crises.
Hence, the issue that makes the CRAs systemically important is the dependence of
the financial system on credit ratings for risk measurement. With an unfair downgrade/
upgrade, the dislocation of a credit risk will quickly prevail because of the herd behavior
of the market. The main reason for this quick response is that the regulatory framework
of the financial system dictates risk weights for different risk categories that are assigned
according to credit ratings.
Data and Econometric Method
Data
This study examines the SCRs of countries and the most influential factors acting on the
SCRs. It also attempts to find out whether the SCRs are biased upward toward the more
developed countries. Both the country risk rating and macroeconomic indicators data used
May–June 2011 Supplement 73
in the analysis are obtained from Moody’s. According to Moody’s rating classification,
there are twenty possible credit ratings for a country: Aaa, Aa1, Aa2, Aa3, A1, A2, A3,
Baa1, Baa2, Baa3, Ba1, Ba2, Ba3, B1, B2, B3, Caa1, Caa2, Caa3, and Ca; Aaa being the
highest rank that a country can get and Ca, the lowest. A rating between Aaa and Baa3
signals a good investment environment for a country, whereas any rating between Ba1
and Ca is speculative.
The data obtained from Moody’s originally had information on 106 countries for the
1999–2010 period. Because of the low number of observations for the Caa2, Caa3, and
Ca ratings, countries that have been in the “junk” category (i.e., countries with a credit
rating of Caa1 and lower) at least once over the period examined are excluded during
the estimations.6
To be able to assess the differences in the rating structures of countries with different
development levels, the data are divided into two subsamples of (1) high- and (2) low- and
middle-income countries.7 Therefore, besides estimating the models on a pooled sample
of all the countries in the data, separate models for high- and for low- and middle-income
countries are also estimated.
Econometric Method
Considering that the SCR of countries is a categorical variable measured with an ordinal
scale, ordered probit modeling is applied to estimate the models in this study. Let yit be
the propensity for the credit evaluation of country i at time t:
yit = β′xit + uit (1)
where β is a k × 1 parameter vector, xit is a k × 1 vector for the individual characteristics
measured at time t, and uit is the stochastic disturbance term. The observability criterion
for the sixteen possible outcomes in the model are given by:
Sit = s if μs–1 ≤ yit ≤ μs for s = 1, 2, 3, …, 16, (2)
where
s =
1
2
if the country’s credit rating is B3
if the country’s credit rating is B2
if the country’s credit rating is A
􀀁 􀀁
16 aa.

Note that μs are the threshold values where μ0 < μ1 < … < μ16, μ0 = –∞ and μ16 = +∞.
Assuming a standard normal distribution for the stochastic disturbance term (uit ∼ N(0, 1)),
the conditional probabilities could be written as:
Pr(Sit = s | xit ) = Φ(μs − β′xit) − Φ(μs–1 – β′xit), (3)
where Φ is the normal probability density function with Φ(–∞) = 0 and Φ(+∞) = 1. Using
the conditional probabilities, the log-likelihood function for the sample is written, and
the parameters maximizing this function are obtained.
A just rating for the countries could only be achieved if all the countries are evaluated
with a similar approach, where the same macroeconomic indicators are taken into
consideration. Introducing country-specific differences into the model as fixed or random
74 Emerging Markets Finance & Trade
effects would detract from the purposes of the study. Nevertheless, random effects ordered
probit models are also estimated in order to observe the changes that the estimation approach
makes on the results.
Let θi represent the unobserved country characteristics that are assumed to be common
in all years for the same country. The credit evaluation of a country i at time t then
takes the form:
yit = β′xit + θi + uit . (4)
It is assumed that θi is normally distributed with a variance of σθ
2. The correlation between
the two disturbance terms is
ρ = σθ
2 / (σθ
2 + σu
2 ) = σθ
2 / (σθ
2 + 1). (5)
The conditional probabilities for the rating outcomes of the countries can be rewritten
as
Pr (Sit = s | xit) = Φ(μs − β′xit − θi) − Φ(μs–1 − β′xit − θi). (6)
The conditional likelihood for the ith country is the product of all the conditional probabilities
for all the available years, T:
Li i P Sit
t
T
(θ ) = ( )
= Π
.
1
(7)
The probability functions in the conditional likelihood in Equation (7) include the
random country-specific component θi . The unconditional likelihood is thus written by
integrating the marginal likelihood over all the possible values of θi (Butler and Moffit
1982; Glick and Sahn 2000; Greene 2002). Defining q } = θi/σθ, the likelihood function
takes the form:
Li i P Sit d i
t
T
= ( ) ( )
= Π
∫ φ θ θ
θ
􀀁 􀀁
􀀁
,
1
(8)
where
Pr S sx x
x
it it s it i
s it i
( = ) = ( − ′ − ( [ − ]) )
− − − ′ −
Φ
Φ
μ β θ ρ ρ
μ β θ ρ
􀀁
􀀁
/
/
/ 1 1 2
1 1 1 2 ( ( [ − ρ]) ) / .
(9)
The log-likelihood function for the total sample is obtained by taking the natural logarithm
of the product of unconditional likelihood functions for all the countries in the
sample.8
Results
Regression Estimations
As mentioned above, the data set used in the estimations belongs to ninety-three countries
covering 1999–2010. The models are estimated for the whole sample as well as separately
for the high-income and low–middle-income countries. Disaggregating the data by the
income level will allow us to observe the differing effects of the explanatory variables
as well as to assess the “justness” of the ratings. All the indicators that are reported to
be considered for the SCRs of both industrialized and developing countries—with the
May–June 2011 Supplement 75
exception of percentage change in real gross domestic product (GDP), government
effectiveness index, and percentage change in real exchange rate—are included in the
estimations.9
Table 1 presents the ordered probit and random effects ordered probit results for the
pooled sample of countries. Model 1 includes the macroeconomic indicators that are
considered in the study while Model 2 has additional dummy variables for the income
level of the countries.10 The low-income country group is chosen to be the base category.
Because there are sixteen alternative rating categories for the dependent variable, fifteen
threshold values are estimated. These threshold values, however, are not reported in Table
1 for the sake of saving some space.11
The results for the country income dummy variables suggest that, holding everything
else constant, high-income countries are more likely to receive higher ratings than lowincome
countries; the increasing effect being higher for the Organization for Economic
Cooperation and Development (OECD) countries than the non-OECD countries. For
the ordered probit estimations, the parameter estimate for the lower–middle-income
countries is insignificant, suggesting that these countries are not significantly different
from the low-income countries in terms of their credit ratings. The upper–middle-income
countries, however, are likely to receive higher ratings than low-income countries;
the coefficient estimate being lower than that for the other two high-income dummy
variables. These results are in line with our expectations. As for the random effects
ordered probit model estimations, the results for the country income level dummies
change. The inclusion of the country-specific random effects reverses the sign of the
parameter estimate for the lower- and upper–middle-income categories. Holding unobserved
country-specific factors as well as other macroeconomic indicators constant,
upper–middle-income countries are likely to receive lower ratings than the low-income
and lower–middle-income countries.
Table 2 presents the regression estimates for the high- and low–middle-income
countries. Note that the observed credit rating ranges from B3 to A1 for the low–
middle-income and from Ba1 to Aaa for the high-income countries. The models for the
low–middle-income countries are, therefore, estimated by considering twelve alternative
rating categories, whereas the models for the high-income countries have eleven
alternative ratings.12
The regression results signal that current account balance, nominal exchange rate,
the ratio of government interest rate payments to government revenue, foreign exchange
reserves, government debt, and the ratio of government debt to government revenue are
the important macroeconomic indicators determining a country’s credit rating. For the
ordered probit estimations, the parameter estimate for GDP is highly significant for the
high-income countries while it is insignificant for the low–middle-income countries. After
controlling for the unobserved country specific random effects, the parameter estimate of
GDP for low–middle-income countries becomes highly significant, its value approaching
the estimate for the high-income countries.
Regarding the estimation results both for ordered probit and random effects ordered
probit models for the pooled sample of countries, parameter estimates for some macroeconomic
indicators become statistically significant after the inclusion of country
group dummy variables (Table 1). This is important since parameters that reflect a
sovereign’s ability to pay back its liabilities, for example, ratio of government expenditure
to GDP and ratio of government interest payment to revenue, become statistically
meaningful after the inclusion of dummies in Model 2 of the random effects ordered
76 Emerging Markets Finance & Trade
Table 1. Estimations for the pooled sample of countries
Ordered probit
Random effects
ordered probit
Variables Model I Model II Model I Model II
Current account balance/GDP −0.008 0.014** −0.004 −0.017**
(0.006) (0.006) (0.007) (0.008)
Current account balance 0.003 −0.000 0.004 0.005*
(0.002) (0.002) (0.003) (0.003)
Nominal exchange rate −0.119*** −0.095*** −0.300*** −0.298***
(0.021) (0.023) (0.023) (0.026)
Government expenditure/GDP −0.813*** −0.671*** −0.385 −0.481**
(0.211) (0.215) (0.235) (0.231)
Government financial balance/GDP −0.579*** −0.355* −0.301 0.081
(0.193) (0.196) (0.209) (0.210)
Government interest payment/revenue 0.047*** 0.045*** −0.008 0.097***
(0.017) (0.017) (0.021) (0.019)
Government primary balance/GDP −0.271*** −0.344*** −0.145* −0.537***
(0.072) (0.074) (0.087) (0.082)
Government revenue/GDP 0.869*** 0.707*** 0.528** 0.514**
(0.212) (0.215) (0.237) (0.232)
Exports, percentage change −0.001 −0.006 0.005 −0.003
(0.004) (0.004) (0.005) (0.004)
GDP 1.355*** 1.081*** 1.790*** 2.153***
(0.156) (0.166) (0.227) (0.237)
GDP per capita −0.004** −0.005** −0.017*** 0.008***
(0.002) (0.002) (0.002) (0.002)
GDP, percentage change −0.034*** −0.020*** −0.020*** −0.022***
(0.007) (0.007) (0.007) (0.007)
Imports, percentage change 0.005 0.010*** 0.001 0.008*
(0.004) (0.004) (0.004) (0.004)
Inflation −0.037*** −0.032*** −0.021** −0.028***
(0.009) (0.009) (0.010) (0.010)
May–June 2011 Supplement 77
Gross investment/GDP −0.006 0.017** 0.068*** 0.042***
(0.008) (0.008) (0.009) (0.009)
Domestic saving/GDP 0.048*** 0.016*** 0.042*** 0.027***
(0.005) (0.005) (0.005) (0.006)
Net foreign direct investment (FDI)/GDP −0.027*** −0.022*** −0.063*** −0.034***
(0.006) (0.007) (0.009) (0.009)
Foreign exchange reserves −0.979 0.130 1.727** 2.029**
(0.693) (0.703) (0.851) (0.873)
Government debt −0.554*** −0.556*** 0.565*** −0.788***
(0.120) (0.125) (0.158) (0.149)
Government debt/GDP 0.012** −0.004 −0.031*** −0.028***
(0.005) (0.006) (0.006) (0.006)
Government debt/revenue −0.004*** 0.001 0.002 −0.000
(0.001) (0.001) (0.002) (0.002)
High income OECD 3.925*** 8.571***
(0.292) (0.391)
High income non-OECD 2.477*** 4.173***
(0.281) (0.327)
Lower-middle income 0.229 −1.161***
(0.258) (0.286)
Upper-middle income 1.048*** −1.216***
(0.267) (0.304)
Rho 0.865*** 0.855***
(0.007) (0.008)
Number of observations 1,009 1,009 1,009 1,009
Log-likelihood −2,115.021 −1,838.452 −1,435.079 −1,266.024
Chi-squared 1,047.848 1,600.987 537.638 875.749
Notes: Standard errors are reported in parentheses. *, **, and *** significant at the 0.10, 0.05, and 0.01 levels, respectively.
78 Emerging Markets Finance & Trade
Table 2. Estimations for low–middle- and high-income countries
Ordered probit Random effects ordered probit
Variables
Low-middle
income
High
income
Low-middle
income
High
income
Current account balance/GDP −0.006 −0.007 −0.026** −0.011
(0.010) (0.009) (0.011) (0.014)
Current account balance −0.015** 0.007** −0.009 0.006
(0.006) (0.003) (0.006) (0.004)
Nominal exchange rate −0.124*** −2.558*** −0.237*** −2.392***
(0.022) (0.341) (0.024) (0.378)
Government expenditure/GDP −0.310 0.870 −0.078 1.068
(0.238) (1.011) (0.272) (1.075)
Government financial balance/GDP −0.310 0.663 0.102 1.196
(0.204) (1.027) (0.231) (1.098)
Government interest payment/revenue 0.042* −0.309*** 0.056** −0.148**
(0.022) (0.056) (0.026) (0.061)
Government primary balance/GDP −0.058 0.149 −0.213* −0.229
(0.103) (0.160) (0.119) (0.183)
Government revenue/GDP 0.329 −0.805 0.047 −0.942
(0.240) (1.010) (0.274) (1.074)
Exports, percentage change −0.002 −0.012* 0.007 −0.008
(0.006) (0.007) (0.006) (0.008)
GDP 0.520 3.362*** 3.172*** 3.901***
(0.451) (0.313) (0.494) (0.404)
GDP per capita −0.002 −0.010*** 0.012*** 0.005
(0.003) (0.004) (0.003) (0.004)
GDP, percent change −0.021** 0.002 −0.026*** 0.009
(0.009) (0.013) (0.010) (0.014)
May–June 2011 Supplement 79
Imports, percent change 0.008 0.014** −0.003 0.023***
(0.005) (0.006) (0.006) (0.006)
Inflation −0.052*** −0.019 −0.037*** −0.008
(0.011) (0.023) (0.012) (0.025)
Gross investment/GDP 0.018* 0.031** 0.079*** 0.062***
(0.010) (0.015) (0.012) (0.020)
Domestic saving/GDP 0.043*** 0.007 0.041*** −0.011
(0.007) (0.009) (0.007) (0.013)
Net FDI/GDP −0.021 −0.028*** −0.009 −0.046***
(0.016) (0.007) (0.018) (0.012)
Foreign exchange reserves 3.811** 3.873*** 1.091 7.658***
(1.914) (1.270) (2.034) (1.562)
Government debt −1.488** −2.613*** −1.883** −2.810***
(0.685) (0.246) (0.751) (0.307)
Government debt/GDP −0.014 0.004 −0.030*** −0.044***
(0.009) (0.010) (0.010) (0.012)
Government debt/revenue −0.003* 0.009*** −0.003 0.012***
(0.002) (0.003) (0.002) (0.003)
Rho 0.795*** 0.861***
(0.015) (0.014)
Number of observations 495.000 514.000 495.000 514.000
Log-likelihood −985.682 −756.177 −687.111 −555.175
Chi-squared 341.572 527.869 342.533 286.342
Notes: Standard errors are reported in parentheses.*, **, and *** significant at the 0.10, 0.05, and 0.01 levels, respectively.
80 Emerging Markets Finance & Trade
probit analysis. Yet for both ordered probit and random effects ordered probit models,
the ratio of government debt to revenue is found to be statistically insignificant. Government
debt is found to be an important indicator for high- and low–middle-income
countries (Table 2). Yet the same indicator is not statistically significant in the pooled
country sample. Another interesting point is that, after controlling for the unobserved
country-specific random effects, the parameter estimate for the ratio of government debt
to GDP becomes highly significant for both country groups with the same sign. The
same situation also holds for the ratio of current account balance to GDP and the ratio
of government primary balance to GDP. The parameter estimates for these indicators,
after controlling for the unobserved country-specific random effects, are significant
for the low–middle-income country group. Allegedly, the current account balance and
government primary balance dynamics in this country group are highly important in
determining their credit rating.
The marginal effects of the parameter estimates for the ordered probit regressions give
more detailed information about the effect of the explanatory variables on the probability
of observing each category.13 The marginal effects calculated for low–middle-income
countries are highly in line with real-life experience. Current account and fiscal balance
indicators—that is, ratio of current account balance to GDP, current account balance,
ratio of government expenditure to GDP, ratio of government financial balance to GDP,
ratio of government primary balance to GDP, ratio of government revenue to GDP—
generally are highest in the B1, Ba3, and Ba2 categories; a point increase in these indicators
increases the probability of a country being in B1, Ba3, and Ba2 categories. For
example, a one-percentage-point increase in the ratio of government expenditure to GDP
in low–middle-income countries increases the probability of that country being in the
B1 category by around five percentage points and of being in the B2 category by three
percentage points. A one-percentage-point increase in this ratio decreases the probability
of a low–middle-income country being in the Baa3 and Baa2 categories by around 3.3
and 3 percentage points, respectively. In the developing/emerging country sphere, that
is, B1, Ba3, and Ba2 categories, the current account and fiscal balance indicators are the
most sensitive macroeconomic dynamics. In the same way, inflation and government
debt to GDP ratio are generally highest in B1, Ba3, and Ba2 categories. Furthermore,
in investment-grade categories and the Ba1 category, the sign for the aforementioned
indicators changes and becomes more negative, in line with expectation, that is, as the
fiscal and current account dynamics of a country worsen, the country is highly likely to
be located in a speculative-grade rather than investment-grade designation.
As for the marginal effects in the high-income country group, the results suggest a
lack of a regular approach in CRAs toward the high-income country group. That is also
compatible with the original premise of the paper. CRAs’ rating behavior toward developing
countries is more or less captured by the model, while it is hard to say the same
for the developed country group. The findings verify that CRAs’ “rating methodology”
is much more benign toward advanced countries.
Credit-Rating Predictions
After estimating the models, a few approaches have been followed in order to validate
the goodness of fit of the estimated models and to compare the rating structures for two
country groups.
May–June 2011 Supplement 81
1. Using the model estimation results, the probabilities of being in each rating
category are calculated for all country and years. The sample means for
these predicted probabilities are then calculated and compared with the
actual rating distribution in the data. This approach originates from the frequency
interpretation of probability. The model has a perfect fit if the mean
predictions and the actual ratios are equal.
2. After calculating the predicted probabilities, a country’s credit rating for
that particular year is predicted as the category with the highest predicted
probability. One disadvantage with this approach is that the calculated
probabilities might be very close to each other for some categories. The approach
chooses one category over the others even if there is not a significant
difference between the predicted probabilities.
3. Calculating the proportion of correct predictions based on the calculations
in item 2 above.
Table 3 compares the actual rating distribution with the mean predicted probabilities
from the pooled sample as well as income-disaggregated estimations. The
mean predicted probabilities are very close to the actual ratios. The models for pooled
regression and high and low income are very encouraging in terms of robustness of
the models. In each group, the ratio of actual and predicted ratings is quite similar,
indicating that the models have favorable predictive powers.
The model estimations show that for developed countries, the models assign good
rating with higher probabilities to developed countries, such as Australia, the United
States, and many European Union member countries. For the European countries such
as Greece, the model predictions are in line with the actual facts. Yet the probabilities
assigned to categories are close to each other, especially after 2008. In other words, the
models’ certainty of giving good ratings to East European countries is not as robust as
that of West European countries, the United States, Australia, and so on. The predictions
are not as straightforward for developing countries. Ratings for some countries, such as
Chile, Egypt, and Mexico, are estimated to be lower than their actual ratings. However,
Colombia, for example, is estimated to be in line with the actual rating for almost the
entire period. An interesting case is Turkey. The models predict Turkey’s rating during
the 2000s to be lower than the actual rating. Yet in recent years, Turkey’s rating has been
forecast to be higher than its actual rating.14
Conclusion
In this study, we tried to shed light on the rating behavior of the CRAs. The basic
argument of the paper is that the CRAs give higher ratings to developed countries
regardless of their macroeconomic fundamentals. However, as a SCR indicates the
capacity and willingness of that country to pay its obligation, macroeconomic fundamentals
should be of core importance. In recent years, great importance has been
attached to the credit-rating phenomenon in the global regulatory fora. However, not
much attention has been paid to the SCR issue. As SCRs determine the availability
and cost of finance not only for sovereign countries itself but also for other borrowers
in that national economy, deficiencies in the process of rating assignment need
to be addressed.
82 Emerging Markets Finance & Trade
Table 3. Mean predicted probabilities
Pooled sample High income Low income
Rating Actual Prediction 11 Prediction 22 Actual Prediction Actual Prediction
B3 2.41 2.83 2.66 4.92 5.12
B2 4.25 3.90 3.88 8.66 8.58
B1 6.76 6.14 6.34 13.78 13.20
Ba3 3.38 3.04 3.30 6.89 6.53
Ba2 5.69 5.44 5.97 11.61 11.66
Ba1 8.01 8.00 8.58 1.33 1.80 14.96 15.34
Baa3 7.72 8.47 8.50 2.84 2.22 12.80 13.93
Baa2 6.27 6.70 6.31 4.55 4.14 8.07 8.70
Baa1 5.89 6.10 5.40 4.92 4.35 6.89 7.15
A3 4.73 4.98 4.10 6.06 5.84 3.35 3.10
A2 7.82 7.91 7.10 8.71 8.83 6.89 6.07
A1 5.98 5.77 5.78 10.61 11.17 1.18 0.61
Aa3 3.28 3.51 3.52 6.44 7.12
Aa2 5.12 5.29 5.69 10.04 10.82
Aa1 2.61 2.36 2.70 5.11 4.78
Aaa 20.08 19.56 20.17 39.39 38.92
Total 100 100 100 100 100 100 100
Notes: 1 Obtained from the pooled sample estimations of Model 1. 2 Obtained from the pooled sample estimations of Model 2, that is, Model 1 + income-level dummies.
May–June 2011 Supplement 83
Notes
1. The CRAs, as profit-oriented private institutions, are paid by the debt issuer in return for
rating bonds and giving advice. This conflict of interest may pressure these firms to assign higher
ratings that are out of line with reality. Furthermore, regarding transparency, the CRAs might fail
to adequately disclose information to their subscribers and unintentionally disclose confidential
information about the institutions they rate.
2. The studies that focused on estimation of SCRs are not solely the ones listed. Hoti and
McAleer (2002) identify fifty empirical studies for SCRs with various econometric methods.
3. In international capital markets, since the collapse of Lehman Brothers, many emerging
market economies have outperformed many developed countries in terms of financial, fiscal, and
economic terms.
4. Ratings of the foreign currency bonds of governments have generally served as ceilings for
ratings of foreign currency debt obligations of domestic issuers up to 2001. The analytic rationale
for that practice was that all domestic issuers are potentially subject to foreign currency transfer
risk. However, the country credit ceiling methodology has been relaxed since 2001, based on the
rationale that governments in default may choose to allow foreign currency payments on some
favored classes of obligors or obligations (Levey 2001).
5. After the collapse of Lehman Brothers, many developed countries announced fiscal measures
that damaged their fiscal balances and increased their debt stocks. Developed countries may in the
future need to finance their short-term debt liabilities with greater risk premiums, that is, in practical
terms with lower credit ratings. For instance, many investment-grade European Union countries,
such as Greece, paid higher risk premiums compared to countries that are deemed noninvestment
grade because their budgetary indicators began to deteriorate. However, this contradicts the very
basic definition of credit rating.
6. These countries are Argentina, Belize, Cuba, Ecuador, Jamaica, Nicaragua, Pakistan, Paraguay,
Ukraine, and Venezuela. Although they had higher ratings, Cayman Islands, Bermuda, and
Bahamas are also excluded from the regressions as these countries had many missing values for
the explanatory variables considered.
7. World Bank country classifications are used for this purpose. High income: OECD and
high income: non-OECD countries in the World Bank’s classification are grouped under the “highincome”
category while countries under the low-income, lower–middle-income, and upper–middleincome
categories are grouped under the “low- and middle-income” category.
8. Note that the log-likelihood function includes multiple integrals, which makes it computationally
very hard to obtain the parameter estimates for the model. The Stata code that is used during
the estimations in this study (Frechette 2001) makes use of the Gauss-Hermite quadrature suggested
by Butler and Moffit (1982), which reduces the T‑dimensional integrals to a single integral.
9. Summary statistics of the variables used in the estimations are reported in Appendix Table A1.
Percentage change in real GDP is excluded from the estimations because it measures the same
effect as the percentage change in nominal GDP, which is included in the models. There are some
other variables that are likely to be correlated with each other, such as the percentage changes in
imports and exports of goods and services. The variables that are likely to cause multicollinearity
problems have been excluded from the model estimation trials in order to observe the possible
effects that the multicollinearity might have on the results. The exclusion of these potentially correlated
variables, however, has not changed the results significantly.
Government effectiveness and percentage change in real exchange rate are not considered
because 409 and 387 observations, respectively, were missing for these variables.
There are some macroeconomic indicators that are reported either for the industrialized or the
developing countries. These variables are not included in the estimations to compare the rating
structures for these two groups of countries. Besides, most of these macroeconomic indicators are
likely to be correlated with each other; their simultaneous inclusion in one model is highly likely
to create problems of multicollinearity and to reduce the explanatory power. All in all, the variables
included in the models are the fundamental indicators for the ability to pay by sovereigns such as
current account, fiscal indicators, and GDP.
10. Time dummy variables are also experimented with but are not included in the final regressions
because they did not make a statistically significant contribution.
11. Almost all the estimated threshold values are significant at the 1 percent level, while some
are significant at the 5 percent level.
84 Emerging Markets Finance & Trade
12. If the same alternative categories were observed for the two group of countries, the “would
be” credit rating of the low–middle-income countries provided they had the same evaluation structure
as the high-income group and the “would be” credit rating of the high-income countries provided
they had the same evaluation structure as the low–middle-income ones would be calculated. The
differences in the observed and predicted ratings in this case would signal whether the rating for
these two groups are just or unjust.
13. Marginal effects are not reported in this study to save space. Figures can be obtained from
the author on request.
14. Here we comment on general country groups. For all the countries, detailed comments need
further information on their economies.
References
Akıncılar, M. 2008. “Kredi derecelendirme kurulus*ları” [Credit Rating Agencies]. In Ekonomik
kurumlar ve kavramlar sözlüg=ü [Dictionary of Economic Institutions and Concepts], ed. F.
Bas*kaya and A. Ördek, pp. 723–730. Ankara: Maki.
Alexe, S.; P. Hammer; A. Kogan; and M. Lejeune. 2003. “A Non-Recursive Regression Model
for Country Risk Rating.” Rutcor research report RRR 9-2003, Center for Operations Research,
Rutgers University, Piscataway.
Bennell, J.A.; D. Crabbe; S. Thomas; and O. Gwilym. 2006. “Modelling Sovereign Credit
Ratings: Neural Networks Versus Ordered Probit.” Expert Systems with Applications 30:
415–425.
Bissoondoyal-Bheenick, E.; R. Brooks; and A.Y. Yip. 2005. “Determinants of Sovereign Ratings:
A Comparison of Case-Based Reasoning and Ordered Probit Approaches.” Working
paper 9/05, Department of Econometrics and Business Statistics, Monash University.
Brewer, T.L., and P. Rivoli. 1990. “Politics and Perceived Country Creditworthiness in International
Banking.” Journal of Money, Credit and Banking 22 (August): 357–369.
Butler, J.S., and R. Moffit. 1982. “A Computationally Efficient Quadrature Procedure for the
One-Factor Multinomial Probit Model.” Econometrica 50, no. 3: 761–764.
Frechette, G.R. 2001. “Random-Effects Ordered Probit.” Stata Technical Bulletin 10, no. 59
(available at http://stata-press.com/journals/stbcontents/stb59.pdf).
Glick, P., and Sahn, D.E. 2000. “Schooling of Boys and Girls in a West African Country: The
Effects of Parental Education, Income and Household Structure.” Economics of Education
Review 19, no. 1: 63–87.
Greene, W.H. 2002. Econometric Analysis, 5th ed. London: Prentice Hall.
Haque, N.U.; N. Mark; and D. Mathieson. 1997. “Rating the Raters of Country Creditworthiness.”
Finance and Development 34 (March): 10–13.
Hoti, S., and M. McAleer. 2002. “Country Risk Ratings: An International Comparison.” Department
of Economics, University of Western Australia.
Hu, Y.-T.; R. Kiesel; and W. Perraudin. 2002. “The Estimation of Transition Matrices for Sovereign
Credit Ratings.” Journal of Banking and Finance 26, no. 7: 1383–1406.
Kaminsky, G., and S.L. Schmukler. 2002. “Emerging Market Instability: Do Sovereign Ratings
Affect Country Risk and Stock Returns?” World Bank Economic Review 16, no. 2: 171–195.
Katz, J.; E. Salinas; and C. Stephanou. 2009. “Credit Rating Agencies: No Easy Regulatory
Solutions.” Crisis Response Policy Brief 8, Financial and Private Sector Development Vice
Presidency, World Bank Group.
Kunczik, M. 2000. “Globalization: News Media, Images of Nations and the Flow of International
Capital with Special Reference to the Role of Rating Agencies.” Paper presented at the
International Association for Media and Communication Research (IAMCR) Conference,
Singapore, July 17–20.
Larraín, G.; H. Reisen; and J. von Maltzan. 1997. “Emerging Market Risk and Sovereign Credit
Ratings.” OECD Development Centre working paper 124, Paris.
Levey, D. 2001. “Revised Country Ceiling Policy-Rating Methodology.” Moody’s Global Credit
Research, Moody’s Investors Service, New York.
Partnoy, F. 1999. “The Siskel and Ebert of Financial Markets: Two Thumbs Down for the Credit
Rating Agencies.” Washington University Law Quarterly 77, no. 3: 619–712.
May–June 2011 Supplement 85
———. 2001. “The Paradox of Credit Ratings.” Law and Economics Research Paper 20, School
of Law, University of San Diego.
Sandström, A. 2008. “Political Risk in Credit Evaluation: Empirical Studies and Survey
Results.” Ph.D. dissertation, Swedish School of Economics and Business Administration,
Helsinki.
86 Emerging Markets Finance & Trade
Appendix Table A1. Summary statistics for the income disaggregated samples
Low- and middle-income countries
Variable Mean Median
Standard
deviation Minimum Maximum
Current account balance/GDP −2.4572 −2.7 7.8962 −36.2 35.6
Current account balance 0.7456 −0.67 15.0898 −41.81 160.82
Nominal exchange rate 0.6343 0.0077 2.379 0.0005 18.5
Government expenditure/GDP 28.8782 27.9 9.2653 12.1 55.5
Government financial balance/GDP −2.743 −2.6 3.5867 −23.1 11.2
Government interest payment/revenue 12.5703 97 12.858 0.6 92.8
Government primary balance/GDP 0.0747 0 2.9738 −8.4 11.6
Government revenue/GDP 26.1129 24 9.2423 11 50.9
Exports, percentage change 11.1729 11.4 15.7889 −42.7 74.3
GDP 0.1707 0.0365 0.3294 0.0006 2.2784
GDP per capita 14.8783 6.562 18.4056 0.37 118.566
GDP, percentage change 11.8232 10.3 10.1421 −29 83.4
Imports, percentage change 11.1255 12.1 16.4728 −44.6 57.2
Inflation 6.44 4.9 7.0854 −2.7 68.8
Gross investment/GDP 23.4612 22.1 6.4115 11.4 43.2
Domestic saving/GDP 21.1689 19.8 11.362 −12 56.9
Net FDI/GDP 3.8345 3.1 3.8418 −15.2 28.7
Foreign exchange reserves 0.0308 0.0075 0.0757 0.00001 0.8189
Government debt 0.0687 0.0111 0.1553 0.0003 1.2369
Government debt/GDP 43.3386 39 27.7598 4.1 170.9
Government debt/revenue 185.88 162.6 133.6118 13.4 851.6
May–June 2011 Supplement 87
High-income countries
Variable Mean

Research proposal