Page 1

UNCORRECTED PROOF

Following the rapid growth in the international debt of less developed countries

in the 1970s and the increasing incidence of debt rescheduling in the early 1980s,

country risk, which reflects the ability and willingness of a country to service its

financial obligations, has become a topic of major concern for the international

financial community (Cosset and Roy, 1991). Political changes resulting from the

AN EMPIRICAL ASSESSMENT OF

COUNTRY RISK RATINGS AND

ASSOCIATED MODELS

Suhejla Hoti and Michael McAleer

University of Western Australia

Abstract. Country risk has become a topic of major concern for the international

financial community over the last two decades. The importance of country ratings

is underscored by the existence of several major country risk rating agencies,

namely the Economist Intelligence Unit, Euromoney, Institutional Investor, Inter-

national Country Risk Guide, Moody’s, Political Risk Services, and Standard and

Poor’s. These risk rating agencies employ different methods to determine country

risk ratings, combining a range of qualitative and quantitative information regard-

ing alternative measures of economic, financial and political risk into associated

composite risk ratings. However, the accuracy of any risk rating agency with regard

to any or all of these measures is open to question. For this reason, it is necessary

to review the literature relating to empirical country risk models according to

established statistical and econometric criteria used in estimation, evaluation and

forecasting. Such an evaluation permits a critical assessment of the relevance and

practicality of the country risk literature. The paper also provides an international

comparison of risk ratings for twelve countries from six geographic regions. These

ratings are compiled by the International Country Risk Guide, which is the only

rating agency to provide detailed and consistent monthly data over an extended

period for a large number of countries. The time series data permit a comparative

assessment of the international country risk ratings, and highlight the importance

of economic, financial and political risk ratings as components of a composite

risk rating.

Keywords. Country risk; Economic risk; Financial risk; Political risk; Composite

risk; Risk ratings; Risk returns; Volatilities; Component analysis; International

comparison.

1. Introduction

1.1. Country risk

0950-0804/04/00 0001–50

#Blackwell Publishing Ltd. 2004, 9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main St.,

Malden, MA 02148, USA.

JOURNAL OF ECONOMIC SURVEYS Vol. 18, No. 0

CSE: ARJOES 001

Page 2

UNCORRECTED PROOF

ators sometimes threaten to repudiate their ‘borrowings’ (Bourke, 1990).

Political risk is generally viewed as a non-business risk introduced strictly

by political forces. Banks and other multinational corporations have identified

political risk as a factor that could seriously affect the profitability of their

international ventures (Shanmugam, 1990). Ghose (1988) argues that political

risk is analogous to sovereign risk and lies within the broader framework of

country risk. Political risk emerges from events such as wars, internal and external

fall of communism, and the implementation of market-oriented economic and

financial reforms, have resulted in an enormous amount of external capital

flowing into the emerging markets of Eastern Europe, Latin America, Asia, and

Africa (Ramcharran, 1999). These events have alerted international investors to

the fact that the globalisation of world trade and open capital markets are risky

elements that can cause financial crises with rapid contagion effects, which

threaten the stability of the international financial sector (Hayes, 1998). In light

of the tumultuous events flowing from September 11, 2001, the risks associated

with engaging in international relationships have increased substantially, and

become more difficult to analyse and predict for decision makers in the economic,

financial and political sectors.

Given these new developments, the need for a detailed assessment of country

risk and its impact on international business operations is crucial. Country risk

refers broadly to the likelihood that a sovereign state or borrower from a

particular country may be unable and/or unwilling to fulfil their obligations

towards one or more foreign lenders and/or investors (Krayenbuehl, 1985). A

primary function of country risk assessment is to anticipate the possibility of debt

repudiation, default or delays in payment by sovereign borrowers (Burton and

Inoue, 1985). Country risk assessment evaluates economic, financial, and political

factors, and their interactions in determining the risk associated with a particular

country. Perceptions of the determinants of country risk are important because

they affect both the supply and cost of international capital flows (Brewer and

Rivoli, 1990).

Country risk may be prompted by a number of country-specific factors or

events. There are three major components of country risk, namely economic,

financial and political risk. The country risk literature holds that economic,

financial and political risks affect each other. As Overholt (1982) argues, inter-

national business scenarios are generally political-economic as businesses and

individuals are interested in the economic consequences of political decisions.

The lending risk exposure vis-a ` -vis a sovereign government is known as sovereign

risk (Juttner, 1995). According to Ghose (1988), sovereign risk emerges when a

sovereign government repudiates its overseas obligations, and when a sovereign

government prevents its subject corporations and/or individuals from fulfilling

such obligations. In particular, sovereign risk carries the connotation that the

repudiation occurs in situations where the country is in a financial position to

meet its obligations. However, sovereign risk also emerges where countries are

experiencing genuine difficulties in meeting their obligations. In an attempt to

extract concessions from their lenders and to improve rescheduling terms, negoti-

2

HOTI AND MCALEER

# Blackwell Publishing Ltd. 2004

Page 3

UNCORRECTED PROOF

Derivative assets, such as futures and options, are used to hedge against price

risk in commodity markets. In particular, country risk ratings are used to hedge

against issued bonds. Optimal hedging strategies and an evaluation of the risk

associated with risk ratings require knowledge of the volatility of the underlying

process. As volatility is generally unknown, it must be estimated. These estimated

and predicted volatilities are fundamental to risk management in financial models

that describe the risk-return trade-off. Although there does not yet seem to be an

conflicts, territorial disputes, revolutions leading to changes of government, and

terrorist attacks around the world. Social factors include civil unrests due to

ideological differences, unequal income distribution, and religious clashes.

Shanmugam (1990) introduces external reasons as a further political aspect of

country risk. For instance, if the borrowing nation is situated alongside a country

that is at war, the country risk level of the prospective borrower will be higher

than if its neighbour were at peace. Although the borrowing nation may not be

directly involved in the conflict, the chances of a spillover effects may exist.

Additionally, the inflow of refugees from the war would affect the economic

conditions in the borrowing nation. In practical terms, political risk relates to

the possibility that the sovereign government may impose foreign exchange and

capital controls, additional taxes, and asset freezes or expropriations. Delays in

the transfer of funds can have serious consequences for investment returns,

import payments and export receipts, all of which may lead to a removal of the

forward cover (Juttner, 1995).

Economic and financial risks are also major components of country risk. They

include factors such as sudden deterioration in the country’s terms of trade, rapid

increases in production costs and/or energy prices, unproductively invested

foreign funds, and unwise lending by foreign banks (Nagy, 1988). Changes in

the economic and financial management of the country are also important

factors. These risk factors interfere with the free flow of capital or arbitrarily

alter the expected risk-return features for investment. Foreign direct investors are

also concerned about disruptions to production, damage to installations, and

threats to personnel (Juttner, 1995).

1.2. Country risk ratings

Since the Third World debt crisis in the early 1980s, commercial agencies such as

Moody’s, Standard and Poor’s, Euromoney, Institutional Investor, Economist

Intelligence Unit, International Country Risk Guide, and Political Risk Services,

have compiled sovereign indexes or ratings as measures of credit risk associated

with sovereign countries. Risk rating agencies provide qualitative and quantita-

tive country risk ratings, combining information about alternative measures of

economic, financial and political risk ratings to obtain a composite risk rating.

This paper provides an international comparison of country risk ratings and

returns compiled by the International Country Risk Guide (ICRG), which is

the only risk rating agency to provide detailed and consistent monthly data

over an extended period for a large number of countries.

COUNTRY RISK RATINGS

3

# Blackwell Publishing Ltd. 2004

Page 4

UNCORRECTED PROOF

at least 50 published empirical papers over the last three decades in refereed

journals. Although the first two papers were published by Frank and Cline

(1971) (Journal of International Economics) and Feder and Just (1977) (Journal

of Development Economics), there were 16 papers published in the 1980s, a

further 30 papers published in the 1990s, and with the 2 most recent papers

having been published in 2001. Thus, the literature is essentially two decades

old. There is no leading journal in the literature on country risk, with the Journal

established market for pricing risk ratings as a primary or derivative asset,

estimating and testing the volatility associated with risk ratings would seem to

be a first step in this direction. In the finance and financial econometrics litera-

ture, conditional volatility has been used to evaluate risk, asymmetric shocks, and

leverage effects. The volatility present in risk ratings also reflects risk considera-

tions in risk ratings. As risk ratings are effectively indexes, their rate of change (or

returns) merits attention in the same manner as financial returns (for further

details, see Hoti et al., 2002).

The plan of the paper is as follows. Section 2 provides a quantitative classi-

fication of empirical country risk models, which forms the database, and also

classifies and describes the data. Various theoretical and empirical model

specifications used in the literature are reviewed analytically and empirically in

Section 3. Section 4 discusses the empirical findings of the published studies. A

comparison of ICRG country risk ratings, risk returns, and their associated

volatilities for twelve representative developing countries is given in Section 5.

Concluding remarks and some suggestions for future research are presented in

Section 6.

2. Classification of Country Risk Models and the Data

For purposes of evaluating the significance of empirical models of country risk, it

is necessary to analyse such models according to established statistical and

econometric criteria. The primary purpose of each of these empirical papers is

to evaluate the practicality and relevance of the economic, financial and political

theories pertaining to country risk. An examination of the empirical impact and

statistical significance of the results of the country risk models will be based on an

evaluation of the descriptive statistics relating to the models, as well as the

econometric procedures used in estimation, testing and forecasting.

This paper reviews 50 published empirical studies on country risk (the papers

are listed in the Appendix). A classification of the 50 empirical studies is given

according to the model specifications examined, the choice of dependent and

explanatory variables considered, the number of explanatory variables used,

econometric issues concerning the recognition, type and number of omitted

explanatory variables, the number and type of proxy variables used when vari-

ables are omitted, the method of estimation, and the use of diagnostic tests of the

auxiliary assumptions of the models.

Scrutiny of the ECONLIT software package and the Social Science Citation

Index for the most widely cited articles in the Country Risk literature yields

4

HOTI AND MCALEER

# Blackwell Publishing Ltd. 2004

Page 5

UNCORRECTED PROOF

Table 1. Classification by Type of Data Used.

of Development Economics publishing 6 papers, the Journal of International

Business Studies publishing 4 papers, the Journal of Banking and Finance,

Economics Letters and Applied Economics each publishing 3 papers, Applied

Economics Letters and Global Finance Journal each publishing 2 papers, and 27

other journals each publishing one paper on the topic.

Country risk has been surveyed previously by Saini and Bates (1984) and Eaton

and Taylor (1986). Saini and Bates (1984) provide a survey of the quantitative

approaches to risk analysis by reviewing the problems in the statistical

approaches of published empirical papers. In particular, they examine the short-

comings with regard to definitions of dependent variables, the quality and avail-

ability of data, model specifications, appropriateness of statistical methods, and

the ability to forecast debt servicing difficulties adequately. Eaton and Taylor

(1986) review the theoretical aspects of numerous papers relating to LDC debt

and financial crises, with an emphasis on the policy implications to be drawn.

Although they do not analyse the empirical aspects of the various papers, they

examine the three main issues in empirical applications, namely the determinants

of rescheduling, how credit terms are fixed, and the factors determining the

amounts borrowed. While the primary purpose of Rockerbie (1993) is to explain

the interest spread on sovereign Eurodollar loans on the basis of various indica-

tors of default risk in lesser developed countries and developed countries, he

provides a useful summary of risk indicators in the empirical papers examined.

Thus, the present paper may be seen as a continuation of these surveys using more

recently published contributions to the literature on country risk.

In Table 1, the 50 studies are classified according to the type of data used,

namely cross-section or pooled, which combines time series and cross-section

samples. Common sources of data are the International Monetary Fund, Bank

for International Settlements, various sources of the World Bank, Euromoney,

Institutional Investor, Moody’s, Standard and Poor’s, and various country-specific

statistical bureaux. Almost three-quarters of the studies are based on pooled data,

with the remaining one-quarter based on cross-section data.

Table 2 classifies the 34 studies using pooled data according to the number of

countries, which varies from 5 to 95 countries, with mean 48 and median 47, with

the frequency of occurrence of each number generally being 1. The same 34

studies using pooled data are classified according to the number of annual and

semi-annualobservationsinTables3and4,respectively.Fortheannualobservations,

the range of the 19 data sets is 5–24 years, with the mean, median and mode of the

number of observations being 12, 11 and 5, respectively, with the frequency of

Type of dataFrequency

Pooled

Cross-section

Total

34

16

50

COUNTRY RISK RATINGS

5

# Blackwell Publishing Ltd. 2004

Page 6

UNCORRECTED PROOF

8, 10, 12, 19

9

11

13, 14, 15, 16, 17, 18, 22, 24

Total

occurrenceofeachnumbervaryingbetween1and5.Therangeofthe8datasetsusing

semi-annualobservationsis8to38half-years,withthemeanandmedianandmodeof

the number of observations being 18.5 and 17, respectively.

Tables 5 and 6 classify the studies using cross-section data according to the

number of countries and the number of time series observations, respectively. In

Table 5, 1 study did not report the number of countries used, while another study

used data on 892 municipalities. Of the remaining 16 studies, the range is 18 to

143 countries, with mean 55.3 and median 50.5. There are 29 data sets using time

series observations in Table 6, with range 1 to 23, mean 5.3, median 3, and mode

1. Indeed, the most commonly used number of time series observations is 1, with a

frequency of 10 in the 29 data sets, so that more than one-third of the cross-

section data sets used are based on a single year.

3. Theoretical and Empirical Model Specifications

The general country risk model typically estimated, tested and evaluated is given as:

f Yt;Xt;ut;?

ð Þ ¼ 0

ð1Þ

in which f(.) is an unspecified functional form, Y is the designated (vector of)

endogenous variables, X is the (vector of) exogenous variables, u is the (vector of)

errors, ? is the vector of unknown parameters, and t¼1,...,n observations. As

Table 2. Classification of Pooled Data by Number of Countries.

Numbers of countries

Number of

studiesFrequency

5, 16, 17, 19, 24, 25, 26, 30, 32, 39, 41, 43, 48, 54, 55, 56, 60,

65, 68, 74, 75, 80, 85, 90, 95

27, 33, 40, 47, 59, 79

Total

25

6

1

2

37

Note: Three studies used two data sets.

Table 3. Classification of Pooled Data by Number of Annual Observations.

Numbers of observations Frequency

55

3

2

4

1

31

Note: One study used two annual data sets, two studies used one annual data set and one semi-annual

data set, and another study used one annual data set, one semi-annual data set, and one monthly data set.

6

HOTI AND MCALEER

# Blackwell Publishing Ltd. 2004

Page 7

UNCORRECTED PROOF

Numbers of countries

will be discussed below, equation (1) is typically given as a linear or log-linear

regression model, or as a logit, probit or discriminant model. The elements of Y

and X will also be discussed below. Defining the information set at the end of

period t?1 as It?1¼[Yt?1,Yt?2,Yt?3,...;Xt,Xt?1,Xt?2,Xt?3,...], the assumptions

of the classical model are typically given as follows:

(A1) E(ut)¼0 for all t;

(A2) constant unconditional and conditional variances of ut;

(A3) serial independence (namely, no covariation between utand usfor t6¼s);

(A4) X is weakly exogenous (that is, there is no covariation between Xtand us

for all t and s);

(A5) u is normally distributed;

(A6) parameters are constant;

(A7) Y and X are both stationary processes, or are cointegrated if both are

non-stationary.

Diagnostic tests play an important role in modern empirical econometrics, and

are used to check the adequacy of a model through testing the underlying

assumptions. The standard diagnostic checks which are used to test assumptions

(A1) through (A7) are various tests of functional form misspecification, hetero-

scedasticity, serial correlation, exogeneity, third- and higher-order moments of

the distribution for non-normality, constancy of parameters and structural

change, unit root tests, and tests of cointegration. There is, in general, little or

no theoretical basis in the literature for selecting a particular model of country

risk. In empirical analysis, however, computational convenience and the ease of

Table 4. Classification of Pooled Data by Number of Semi-Annual Observations.

Numbers of observations Frequency

8, 17, 22

16, 38

Total

2

1

8

Note: One study used two semi-annual data sets, two studies used one annual data set and one semi-

annual data set, and another study used one annual data set, one semi-annual data set, and one

monthly data set.

Table 5. Classification of Cross-section Data by Number of Countries.

Frequency

18, 20, 27, 29, 30, 35, 49, 52, 71, 88, 93, 143, 892, unstated

45, 70

Total

1

2

18

Note: One study used three data sets. The sample with 892 observations refers to municipalities rather

than countries.

COUNTRY RISK RATINGS

7

# Blackwell Publishing Ltd. 2004

Page 8

UNCORRECTED PROOF

Others

interpretation of models are primary considerations for purposes of model

selection.

Of the 70 models used in the 50 studies, which are reported in Table 7, all but

six are univariate models. The most popular model in the literature is the logit

model, which is used 23 times, followed by the probit, discriminant, and Tobit

models, which are used 10, 7, and 3 times, respectively. Thus, more than half

of the models used in the literature are probability-based models. Given the

popularity of the linear and log-linear regression models in empirical economic

research, it is surprising to see that the linear regression model is used four times,

the log-linear regression model is used only twice, and both regression models

are used in the same study only twice. The artificial neural network model is

also used twice. Of the remainder, the multi-group hierarchical discrimination

model, two-way error components model, random-effect error component equa-

tions, naive model, combination model, G-logit model, nested trinomial logit,

Table 6. Classification of Cross-Section Data by Number of Time Series Observations.

Numbers of observationsFrequency

1

2

3, 7, 8, 10, 11, 23

4, 20

5

Total

10

4

1

2

5

29

Note: More than one time series data set was used in some studies.

Table 7. Classification by Type of Model.

ModelFrequency

Only linear single equations

Only log-linear single equations

Both linear and log-linear single equations

Logit

Probit

Discriminant model

Tobit

System of equations

Artificial neural network model

4

2

2

23

10

7

3

6

2

11

Total70

Note: More than one model was used in some studies and two studies used no model. The ‘Others’

category includes one entry for each of multi-group hierarchical discrimination model, two-way error

components model, random-effect error component equations, naı¨ve model, combination model,

G-Logit model, nested trinomial logit, sequential-response logit, unordered-response logit, classification

and regression trees, and cluster analysis.

8

HOTI AND MCALEER

# Blackwell Publishing Ltd. 2004

Page 9

UNCORRECTED PROOF

probability of partial reneging when a borrower has decided to reschedule, trichotomous variable of

debt rescheduling, the probability of general, commercial, official, and band debt rescheduling (in

the current year or in the future), the probability of debt default, and discriminant score of whether

a country belongs to a rescheduling or non-rescheduling group.

3. Refers to Institutional Investor, Euromoney, Standard and Poor’s, Moody’s, and Economist

Intelligence Unit country or municipality credit risk ratings, and average agency country risk

ratings.

4. Includes one entry for each of limit on debt arrears, dummy for significant debt arrears, probability

of experiencing significant debt arrears, and probability of emerging debt-servicing arrears.

sequential-response logit, unordered-response logit, classification and regression

trees, and cluster analysis, are used once each.

The dependent variable for purposes of analysing country risk is broadly

classified as the ability to repay debt. Of the different types of dependent variables

given in Table 8, with more than one dependent variable being used in some

studies, the most frequently used variable is debt rescheduling, which is used 36

times. This dependent variable is defined as the probability of general, commer-

cial, and official debt rescheduling or debt default (in the current year or in the

future), and discriminant score of whether a country belongs to a rescheduling or

non-rescheduling group. The second most frequently used variable is agency

country risk ratings, which is used 18 times. In the empirical analyses, this

Table 8. Classification by Type of Dependent Variable Used1.

TypeFrequency

Debt rescheduling2

Agency country risk ratings3

Debt arrears4

(Average) value of debt rescheduling

Exchange rate movements

Fundamental valuation ratios

Demand for debt

Supply of debt

Propensity to obtain agency municipality credit risk ratings

Public debt to private creditors

Total reserves

(Relative) bond spreads

Weighted average loan spread

Spread over LIBOR

Yield spreads of international bonds

Payment interruption likelihood index

Sovereign loan default

Credit risk rating

Income classification

Stock returns

Secondary market price of foreign debt

Dummy for debt crisis

Total

36

18

4

3

3

3

3

3

2

2

2

2

1

1

1

1

1

1

1

1

1

1

91

Notes:

1. More than one dependent variable was used in some studies.

2. Includes variables defined as the probability of debt rescheduling (as proxy for debt default), the

COUNTRY RISK RATINGS

9

# Blackwell Publishing Ltd. 2004

Page 10

UNCORRECTED PROOF

5, 9, 10

6

8

11, 14, 15, 18, 20, 23, 32

12

Total

dependent variable is defined as Institutional Investor, Euromoney, Standard and

Poor’s, Moody’s, and Economist Intelligence Unit country or municipality risk

ratings, and the average of agency country risk ratings. Ten types of dependent

variable are used more than once, with debt arrears (defined as the limit on

debt arrears), dummy for significant debt arrears, probability of experiencing

significant debt arrears, and probability of emerging debt-servicing arrears

being used 4 times each, and average value of debt rescheduling, exchange rate

movements, fundamental valuation ratios, demand for debt, and supply of debt

being used 3 times each. Dependent variables, such as the propensity to obtain

agency municipality credit risk ratings, public debt to private creditors, total

reserves, and total or relative bond spread, are used twice each, with the remain-

ing 10 types of dependent variable, which are used once each, including weighted

average loan spread, spread over LIBOR, yield spreads of international bonds,

payment interruption likelihood index, sovereign loan default, credit risk rating,

income classification, stock returns, secondary market price of foreign debt, and

dummy for debt crisis.

There are three types of explanatory variables used in the various empirical

studies, namely economic, financial and political. Treating country risk variables

as economic and/or financial, and regional differences as political, Tables 9 and

10 present the numbers of each type of variable and their frequency. In Table 9,

the number of economic and financial variables ranges from 2 to 32, with mean

11.5, median 8 and mode 6. Seven of the 19 sets of economic and financial

variables have a frequency of one, with a frequency of 2 occurring 3 times, a

frequency of 3 occurring 5 times, and frequencies of 4, 5, and 6 occurring once

each. In Table 10, the number of political variables ranges from 0 to 13, with

mean 1.86, median 0 and mode 0. The absence of any political variable occurs 30

times in the 50 studies.

Of the remaining 10 sets of political variables, 2 have a frequency of 4, one has

a frequency of 3, 2 have a frequency of 2, and five have a frequency of one.

Hundreds of different economic, financial and political explanatory variables

have been used in the 50 separate studies. The set of economic and financial

variables includes indicators for country risk ratings, debt service, domestic and

Table 9. Classification by Number of Economic and Financial Explanatory Variables.

NumberFrequency

2, 3, 7, 13, 16

4

3

4

2

7

5

1

6

50

Note: Country risk indicators are treated as economic and/or financial variables.

10

HOTI AND MCALEER

# Blackwell Publishing Ltd. 2004

Page 11

UNCORRECTED PROOF

2, 3, 4

8

international economic performance, domestic and international financial perform-

ance, monetary reserves, and structural differences. Indicators for country

political risk ratings, domestic and international armed conflict, political events,

and regional differences are used in the set of political variables.

The unavailability of the required data means that proxy variables have fre-

quently been used for the unobserved variables. Tables 11 and 12 are concerned

with the important issue of omitted explanatory variables in each of the 50

studies. It is well known that, in general, omission of relevant explanatory vari-

ables from a linear regression model yields biased estimates of the coefficients of

the included variables, unless the omitted variables are uncorrelated with each of

the included explanatory variables. For non-linear models, consistency replaces

unbiasedness as a desirable statistical characteristic of an estimation method. In

some studies, there is an indication of the various types of variables that are

recognised as being important. Nevertheless, some of these variables have been

omitted because they are simply unavailable. The classification in Table 11 is by

recognition of omitted explanatory variables, where the recognition is explicitly

stated in the study. Such an explicit recognition of omitted explanatory variables

is used primarily as a check of consistency against the number of proxy variables

used.

Of the 50 studies in Table 11, exactly three-fifths did not explicitly recognise

that any variables had knowingly been omitted, with the remaining 20 studies

Table 10. Classification by Number of Political Explanatory Variables.

NumberFrequency

0

1, 2

3, 8, 10, 11, 13

4, 5

6

Total

30

4

1

2

3

50

Note: Regional differences are treated as political variables.

Table 11. Classification by Recognition of Omitted Explanatory Variables.

Number omitted Frequency

0

1

30

13

2

1

Total 50

Note: The classification is based on explicit recognition of omitted explanatory variables, and is used

primarily as a check of consistency against the number of proxy variables used in the corresponding

studies.

COUNTRY RISK RATINGS

11

# Blackwell Publishing Ltd. 2004

Page 12

UNCORRECTED PROOF

correctly measured variable is less than in the case where the proxy variable is

excluded; (3) a reduction in measurement error is beneficial; and (4) it is prefer-

able to include the proxy variable than to exclude it. When two or more proxy

variables are used to replace two or more variables, which are unavailable, it is

not necessarily the case that the four basic results stated above actually hold.

Thus, among other outcomes, the absolute bias in the estimated coefficients of

both the correctly measured and incorrectly measured variables may be higher if

recognising that 39 explanatory variables had been omitted. The number of

explanatory variables explicitly recognised as having been omitted varies from 1

to 8. Including and excluding the 30 zero entries for omitted explanatory variables

give mean numbers omitted of 0.78 and 1.95, respectively, medians of 0 and 1,

and modes of 0 and 1. Thirteen of the 20 studies, which explicitly recognised the

omission of explanatory variables, noted that a single variable had been omitted.

The classification in Table 12 is given according to the type of omitted explan-

atory variable, which is interpreted as predominantly economic and financial or

political. More than two-thirds of the omitted explanatory variables are pre-

dominantly economic and financial in nature, and the remaining one-third is

predominantly political. Somewhat surprisingly, very few studies stated dynamics

as having been omitted from the analysis, even though most did not explicitly

incorporate dynamics into the empirical specifications.

As important economic, financial and political explanatory variables have been

recognised as having been omitted from two-fifths of the 50 studies (see Table 11),

proxy variables have been used in most of these studies. Tables 13 and 14 are

concerned with the issues of the number and type of proxy variables used. The

problems associated with the use of ordinary least squares (OLS) to estimate the

parameters of linear models in the presence of one or more proxy variables are

generally well known in the econometrics literature, but extensions to non-linear

models, which dominate the literature on country risk, are not yet available.

Nevertheless, as a guide for analysis, the basic results are outlined below. These

results are of special concern as one-half of the studies explicitly recognises the

omission of at least one explanatory variable.

In the case where only one proxy variable is used to replace a variable which is

unavailable, the well-known results are as follows: (1) the absolute bias in the

estimated coefficient of the proxy variable is less than the case where the proxy

variable is excluded; (2) the absolute bias in the estimated coefficient of the

Table 12. Classification by Type of Omitted Explanatory Variables.

Omitted variableFrequency

Economic and financial factors

Political factors

Total

28

11

39

Notes: The various omitted variables are classified according to whether they are predominantly

economic and financial or political in nature.

12

HOTI AND MCALEER

# Blackwell Publishing Ltd. 2004

Page 13

UNCORRECTED PROOF

two or more proxy variables are not used than when they are used, a reduction in

measurement error may not be beneficial, and it may not be preferable to include

two or more proxy variables than to exclude them. The reason for the different

outcomes is that the covariation in two or more measurement errors may exacer-

bate the problem of measurement error rather than containing it.

Table 13 classifies the 20 studies by the use of proxy variables, which ranges

from 1 to 7. Including and excluding the 2 zero entries for the number of proxy

variables used give mean numbers omitted of 2.45 and 2.72, respectively, a

median of 2 in each case, and a mode of 1 in each case. By comparison with

Table 11, in which 13 of the 20 studies explicitly recognised the omission of a

single explanatory variable, Table 13 shows that only 7 studies used a single proxy

variable. Otherwise, the results in Tables 11 and 13 are reasonably similar.

The classification in Table 14 is given according to the type of proxy variable

used, which is interpreted as comprising predominantly economic and financial or

political factors. More than two-thirds of the proxy variables are predominantly

economic and financial in nature, and the remaining one-third is predominantly

political, which is very similar to the results given in Table 12.

In Table 15 the classification is by method of estimation, in which more than one

estimation method is used in some studies. Five categories are listed, namely OLS,

maximum likelihood (ML), Heckman’s two-step procedure, discriminant methods,

and Others, which includes entries for, among others, propagation algorithm,

regression-based techniques, approximation, minimax, Bayesian, optimal minimum

distance, stepwise, optimisation, binary splits, jack-knife methods andOLS and WLS.

Even though logit, probit, and Tobit models in Table 7 are used 40 times in total, ML

Table 14. Classification by Type of Proxy Variables Used.

Proxy variables Frequency

Economic and financial factors

Political factors

Total

34

15

49

Note: Some studies used economic, financial and political proxy variables.

Table 13. Classification by Number of Proxy Variables Used.

Number Frequency

0, 3, 6

1

2

4, 5, 7

Total

2

7

4

1

20

Note: Two studies explicitly recognized the omission of explanatory variables but used no proxy

variables.

COUNTRY RISK RATINGS

13

# Blackwell Publishing Ltd. 2004

Page 14

UNCORRECTED PROOF

Type of diagnostics

is used for estimation purposes only 35 times. Moreover, while linear and log-linear

models are used only 7 times in total in Table 7, OLS is used 14 times in Table 15 (15

timesifbothOLSandWLSareincluded).Finally,whilediscriminantmodelsareused

7 times in Table 7, discriminant estimation is used only three times in Table 15.

Finally, the classification in Table 16 is by use of diagnostics to test one or more

auxiliary assumptions of the models. The role of diagnostic tests has become well

established in the econometrics literature in recent years, and plays an increasingly

prominent role in modern applied econometrics (see McAleer (1994) for further

details). Most diagnostic tests of the auxiliary assumptions are standard, and are

available in widely used econometric software packages. Unbelievably, 42 of the 50

studies did not report any diagnostic tests whatsoever. Of the eight which did report

any diagnostic tests at all, there were two entries for White’s standard errors for

heteroscedasticity, and one entry for each of WLS and heteroscedasticity, transfor-

mation for non-normality, White’s covariance matrix for heteroscedasticity, Chow

test, Hajivassiliou’s test for exogeneity, and serial correlation. This is of serious

concern, especially as the ML method is known to lack robustness to departures

from the stated assumptions, but is nevertheless used 35 times. Models such as the

logit and probit are also sensitive to departures from the underlying logistic and

normal densities, respectively, so that the underlying assumptions should be checked

rigorously. Asthe use of diagnosticshasbeen ignored inthe countryrisk literature, in

general, the empiricalresults should be interpreted with some caution and scepticism.

Table 15. Classification by Method of Estimation.

MethodFrequency

OLS

ML

Heckman’s two-step procedure

Discriminant methods

Others

Total

14

35

2

3

17

71

Note: More than one estimation method was used in some studies. The ‘Others’ category includes

entries for, among others, propagation algorithm, regression-based technique, approximation, mini-

max, Bayesian, optimal minimum distance, stepwise optimisation, binary splits, jack-knife methods,

and OLS and WLS.

Table 16. Classification by Use of Diagnostics.

Frequencies

None

Others

Total

42

8

50

Note: The ‘Others’ category includes entries for WLS and heteroscedasticity, White’s standard errors

for heteroscedasticity, White’s covariance matrix for heteroscedasticity, Chow test, transformation for

non-normality, Hajivassiliou test for exogeneity, and serial correlation.

14

HOTI AND MCALEER

# Blackwell Publishing Ltd. 2004

Page 15

UNCORRECTED PROOF

Variables

4. Empirical Findings for Country Risk Ratings

Of the 91 types of dependent variables used in the 50 studies, 27 studies examined

debt rescheduling on 36 occasions and 17 considered country risk ratings on 18

occasions (see Table 8 for definitions of these two types of variables). Table 17

reports four types of risk component variables used in the 17 country risk ratings

studies, namely economic, financial, political, and composite. Composite risk

variables are ratings or aggregates that comprise economic, financial and political

risk component variables, and were used in all 17 studies. Of these studies, only

two did not use economic variables and only one did not use financial variables.

Political variables have been used less frequently, namely in 10 studies. Table 18

presents the number of country risk components used, as well as their frequency.

All four country risk components have been used in 10 studies, 4 studies used

variables representing three risk components, 3 studies used variables represent-

ing two risk components, and no study used variables representing only one risk

component.

In Table 19, the 17 are classified according to the risk rating agency they used,

namely Institutional Investor, Euromoney, Moody’s Standard and Poor’s, Inter-

national Country Risk Guide, Economist Intelligence Unit, and Political Risk

Services. These agencies are leading commercial analysts of country risk. While

the rating system for the International Country Risk Guide will be analysed in the

next section, the rating systems for the other agencies are briefly discussed below.

Unless otherwise stated, the information regarding the agency rating systems has

been obtained from the website of Foreign Investment Advisory Service Program,

which is a joint service of two leading multilateral development institutions,

namely the International Finance Corporation and World Bank (http://www.fias.

net/investment_ climate.html).

Institutional Investor compiles semi-annual country risk surveys, which are

based on responses provided by leading international banks. Bankers from 75

to 100 banks rate more than 135 countries on a scale of 0 to 100, with 100

representing the lowest risk. The individual ratings are weighted using the Institu-

tional Investor formula, with greater weights assigned to responses based on the

extent of a bank’s worldwide exposure and the degree of sophistication of a

bank’s country risk model. The names of the participating banks are kept

strictly confidential (Howell, 2001). Institutional Investor country risk surveys

are published in the March and September issues of the monthly magazine. In the

Table 17. Risk Component Variables Used in Country Risk Ratings.

Frequency

Economic

Financial

Political

Composite

Number of studies

15

16

10

17

17

COUNTRY RISK RATINGS

15

# Blackwell Publishing Ltd. 2004