Citation
Country risk analysis

Material Information

Title:
Country risk analysis directions for the 1990's
Creator:
Rosauer, Ruth
Publication Date:
Language:
English
Physical Description:
vii, 136 leaves : ; 29 cm

Thesis/Dissertation Information

Degree:
Master's ( Master of arts)
Degree Grantor:
University of Colorado Denver
Degree Divisions:
Department of Economics, CU Denver
Degree Disciplines:
Economics

Subjects

Subjects / Keywords:
Country risk ( lcsh )
Country risk ( fast )
Genre:
bibliography ( marcgt )
theses ( marcgt )
non-fiction ( marcgt )

Notes

Bibliography:
Includes bibliographical references (leaves 88-93).
General Note:
Submitted in partial fulfillment of the requirements for the degree, Master of Arts, Department of Economics.
Statement of Responsibility:
by Ruth Rosauer.

Record Information

Source Institution:
University of Colorado Denver
Holding Location:
Auraria Library
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
22839437 ( OCLC )
ocm22839437
Classification:
LD1190.L53 1990 .R67 ( lcc )

Full Text
COUNTRY RISK ANALYSIS
DIRECTIONS FOR THE 1990s
by
Ruth Rosauer
B.A., Trenton State College, 1975
B.A., Metropolitan State College, 1985
A thesis submitted to the
Faculty of the Graduate School of the
University of Colorado in partial fulfillment
of the requirements for the degree of
Master of Arts
Department of Economics
1990


This thesis for the Master of Arts degree by
Ruth Rosauer
has been approved for the
Department of
Economics
by
Steven Beckman
Date


o-y j 99a


Rosauer, Ruth (M.A., Economics)
Country Risk Analysis Directions for the 1990s
Thesis directed by Professor W. James Smith
This thesis concerns the topic of country
risk analysis. Country risk analysis Seeks to
determine whether or not a country represents a good
lending risk. The literature of country risk analysis
concerns both qualitative and quantitative issues.
This thesis, however, focuses on quantitative
approaches and seeks to uncover analytic approaches
likely to be useful in the 1990s.
Frank and Cline conducted the seminal study
on country risk analysis in 1971. Many other
researchers have modified and reapplied the Frank and
Cline independent variables in their own models.
Although the review of the literature in Appendix A
includes all the major studies done since then, the
central paradigm of the thesis is the Frank and Cline
model.
Country risk analysis is necessarily
concerned with the ability of a model to predict
future actions: in particular, will a country repay
its debt? The predictive ability of the Frank and
Cline model on its original Frank and Cline data set
was quite good. But since that data set ended in 1968


iv
it would be useful to test its post-sample predictive
ability. If its predictive ability were still high
there would be little or no need to develop an
alternative model for the 1990s.
In order to test the explanatory power of
Frank and Cline's independent variables, I created a
data set of 393 observations over the time period of
1975 to 1986. Using logit analysis it was found that
the originally significant variables were still
significant, but the overall explanatory power of the
model was quite weak.
In my critique of the Frank and Cline model I
discuss likely reasons for the poor out-of-sample
!
showing. In chapters four and five there is an
exploration of factors that should be taken into
consideration in the construction of country risk
analysis models for the next decade.
The form and content of this abstract are approved.
I recommend its publication.
Signed
ames Smi


CONTENTS
CHAPTER
I. INTRODUCTION
Rationale .............................. 2
Externalities .................... 4
Scope and Purpose ...................... 6
II. THE THEORETICAL FRAMEWORK
Independent Variables . ........... 10
Dependent Variables .................. 22
Econometric Techniques ................ 30
Discriminant Analysis ............ 31
Logit ............................ 34
III. QUANTITATIVE STUDIES
Frank and Cline ....................... 40
Feder and Just ....................... 46
Manfredi .......................... 48
Critique .............................. 50
IV. TOWARDS A NEW THEORETICAL FRAMEWORK
A Glance Backwards .................... 58
Unexpected Changes ............... 62
Political Risks ...................... 67
The Heuristic Element ................. 70
The Risk of Repudiation
75


vi
V. A MODEL FOR THE 1990s
Limitations of Modelling .............. 81
An Alternative Solution ............... 85
WORKS CITED ..................................... 88
APPENDIX
A. REVIEW OF THE LITERATURE ................. 94
B. PROBLEMS IN COUNTRY RISK ANALYSIS ....... 105
C. DEFINITIONS ............................. 109
D. DATA SET FOR TABLES 3-1 and 5 - 2 ______ 111
E. COMPUTER RUNS FOR POOLING TESTS ........ 122


vii
TABLES
Table
2- 1 Independent Variables ............ 12
3- 1 Logit Calculations ................ 52
3-2 Pseudo R2 Calculations ............ 55
5-1 Results of Pooling Test ........... 84


CHAPTER 1
INTRODUCTION
Country risk analysis is a term applied to
the process used by lenders to determine the
likelihood of a sovereign nation repaying the money
it seeks to borrow. Although lending can flow from a
poor nation to a richer one, or between countries who
are relatively equal; the threat of debt repudiation
is perceived to be greatest when occurring between an
agent of an industrialized nation as lender and
developing nation as borrower. Country risk analysis,
therefore, is primarily concerned with the third type
of arrangement.
Lending to nations by agencies outside that
nation (external debt) has been going on for
centuries. But the formal analysis of country risk is
a much more recent phenomenon. It wasn't until the
1970s that U.S. banks began incorporating systematic
country risk evaluation methods into their
operations. Although it has since become a "semi-
official" requirement of the Comptroller of the


2
Currency and the Federal Reserve Board, it is a
discipline still in its infancy. As recently as 1976
an Eximbank survey found that 23 out of 37 banks were
using strictly qualitative forms of analysis
(Burton and Inoue 42).
Frank and Cline's study, published in 1971,
is considered the pioneer effort in utilizing
quantitative methods to forecast loan defaults.
Indeed, it is often referred to as the "seminal
study" in this field (Mayo and Barrett 83). The wave
of threatened debt repudiations, interest payment
suspensions and loan re-structurings of the 1980s has
spurred additional investigation into quantitative
country risk analysis. Most of the pertinent
literature uses the term "country risk analysis"
specifically, a few refer to "early warning systems"
and others to "debt optimization."
Rationale
Why study country risk analysis? Because
lending to a sovereign nation is different from
lending to an individual. When a lender extends a
loan to an individual who later defaults on that


loan, the lender has several options to recover the
loss. Those sanctions are unpleasant enough for the
individual, and enforceable enough, to deter most
debt repudiation. With the cessation of "gunboat
diplomacy," however, creditors have less assurance
about recovering bad debts from nations. "Repayment
of international debt, however, is largely voluntary;
the penalties to be imposed on a country that does
not honor a contract are, at best, indirect" (Eaton,
Gersovitz and Stiglitz 484). Some observers
categorize those sanctions as "negligible" (Burton
and Inoue 41). Thus an accurate analysis of a
country's likelihood to repay a debt assumes a more
crucial role than the credit investigation of an
individual.
Another reason why country risk analysis has
assumed such importance is due to the large sums of
money involved. The Secretary General of the United
Nations, Javier Perez de Cuellar, estimated the size
of the collective external debt of the Third World
nations to be one thousand billion dollars in 1987.
By way of comparison, this was twice the value of
export earnings of the same group of countries in
1986 ("Developing Countries" 44).


4
Seventy percent of lending to Third World
nations has been done by commercial banks
("Developing Countries" 44). William Cline, writing
in the Columbia Journal of World Business points out
that, "Default by all non-oil developing countries
and Eastern Europe would obliterate US bank capital."
He estimates that default by either Mexico or Brazil
alone would wipe out 40 percent of our bank capital
(9). Banks have taken action in the past few years
to improve their capitalization ratios; the stakes
are not quite as high now as they were in 1983, but
they remain sizable. As the savings and loan
disasters of the 1980s have shown, the consequences
of unwise lending decisions accrue not only to the
banks and their stockholders.
Externalities
Donogh McDonald notes that the externalities
of bank failure are probably more important in the
long run than the profit sheet of an individual bank
(634). One such externality would affect U.S. bank
creditors directly:
If the big five Latin American borrowers were
to default on barely half of their debt, or if
investors were willing to value those debts at


5
only 50 cents on the dollar, the impact would
be sufficient to bankrupt the six largest banks
in the United States, institutions that in 1982
handled deposits of $340 billion, two-fifths of
the total demand, time, and saving deposits in
the U.S. banks . Without some major rescue
effort, neither the American financial system
nor that of the world could withstand a loss of
some two dollars out of every five dollars
worth of bank deposits without the massive
disruptions in economic activity that have
accompanied great financial failures in the
past (Makin 19).
In cases where bank losses exceed bank
capital, those externalities will ultimately be borne
by governments and taxpayers (Allsopp and Joshi
xvii).
Country risk analysis is important from the
debtor's point of view as well. It is the determining
factor in how much foreign capital the developing
nation can obtain. For the debtor nation external
debt is often its largest source of capital for
economic development projects (Feder "Economic
Growth" 352). It can also be used to soften the
adjustment to external shocks and even out the flow
of consumption due to fluctuations in income
(McDonald 604).
Caballeros notes that external debt is ". .
the decisive element in almost all economic policy
[in Latin America], it financed practically all


public investment and covered the fiscal deficit, as
well as providing international liquidity" (127).
Since loans are so vital to the developing
country's economy, and those loans are dependent on
the risk analysis conducted by the lending agent, it
behooves the potential borrower to be as well-
informed as possible about the system.
" . there may be scope for developing countries
to enhance their borrowing terms and, hence, their
debt capacity by recognizing how their own behavior
affects their creditworthiness," (McDonald 635)
Country risk analysis, then, is an endeavor
that ultimately affects governments and taxpayers as
well as international bankers.
Scope and Purpose
The primary purpose of this thesis is to
prepare the author to develop and test a new
econometric model for country risk analysis in the
future. Part of that preparation required a thorough
review of the country risk analysis literature.
Issues pertinent to such analysis are presented in
this paper.


7
Another part of that preparation is an
application of logit analysis to pooled cross-
sectional and time series data. The independent
variables are those used in Frank and Cline's 1971
but with a larger and more recent data set. This
experience provides me with a heightened awareness of
the difficulties in this type of econometric
modelling that reading of the literature alone could
not have provided.
Normative issues have been omitted as much as
possible. I recognize that one's own value system
acts as a filter through which one evaluates new
information, but care was taken to avoid areas posing
ethical dilemmas. Moral hazard,, the North/South
paradigm, the effects of loan decisions on the
poverty-stricken inhabitants of the Third World, and
the question of whether or not this activity should
be going on at all are important issues; but beyond
the scope of this paper. They have been mentioned
briefly or not at all.
Most important, the research and
experimentation undertaken to write this thesis have
generated the questions (posed in chapter 5) that
will set the direction for my further research in
country risk analysis.


CHAPTER 2
THE THEORETICAL FRAMEWORK
Has a strong theoretical framework been
established for country risk analysis work? According
to McDonald the answer would have to be "no."
. . as many of the authors admit, the
explanatory variables in the preceding
discriminant and logit analyses are introduced
in a very ad hoc way. Thus, the studies are
based on no underlying theory of why a country
follows this course of action or not. They
basically involved searching for statistical
relationships. Furthermore, the range of
variables tried and the widely different final
specifications arrived at do not add to one's
confidence. There has not been a consideration
here of the relevance of the different
explanatory variables chosen. However, it is
clear that absence of a theoretical structure
has led to difficulties in interpreting the
role of some variables (623).
Saini and Bates support this criticism.
Krishan Saini (Federal Reserve Bank of New York) and
Philip Bates (Standard and Poor's Corp.) are even
harsher in their appraisal of country risk
statistical studies:
A review of the major studies shows them to
contain serious shortcomings with respect to


9
their definition of the dependent variable, the
quality and availability of data, specification
of models, appropriateness of statistical
procedures, and the ability to adequately
forecast debt servicing difficulties based on
the analysis of past experience. (Saini and
Bates, "Survey of Quantitative" 341).
What do the creators of these country risk
models set forth as their own criteria for selecting
variables? Frank and Cline said the goal of their
1971 study was "to find an index or indicator of the
likelihood that a less developed country will
experience debt servicing difficulties. The indicator
should satisfy two criteria: (1) it should be
relatively simple, and (2) it should have a higher
degree of predictability" (329). Other studies'
theoretical frameworks were conspicuous by their
absence.
By and large, country risk analysis models
seem to have been developed by the researcher asking
himself, "What factors are most likely to make it
possible to repay this debt?" and then selecting his
explanatory variables accordingly. The vagueness of
this method is reflected in the large number of
independent variables that many of the researchers
initially include. Mayo and Barrett began with 50
basic variables in several groupings and ratios for


10
their country risk assessment model for the Eximbank.
Their final model retained only six of those variable
(84 and 85). Abassi and Taffler started with 42
variables and eventually reduced that number to the
nine which proved to be significant.
Independent Variables
What independent variables have country risk
analysts tried? For ease of discussion I have divided
them into some rather broad categories: balance of
trade, debt, domestic, and external (See Table 2-1).
There is some overlap between categories since many
of the factors are ratios. In some instances, an
indicator has been listed in two categories. For
example, "debt service/exports" appears both in trade
balance and debt.
The trade balance category is comprised of
those financial indicators that contain at least one
factor that is included in the balance of trade
statistics. Exports, imports, transfers, service
receipts and service payments make up the traditional
balance of trade statistics. These are particularly
pertinent to questions relating to country risk


11
because external debt is generally denominated in the
currency of the creditor and must be re-paid in that
currency. Foreign exchange earned through trade
represents a debtor nation's most likely way to
generate debt service payments (Calverley 84).
Poor showings in this area can be the result
of national trade policies such as tariffs, import
restrictions, or an unrealistically high exchange
rate. If a balance of trade deficit is unsustainable
it will also appear in debt service ratios (Krueger
171). Since the time lag for availability is longer
for debt statistics than for balance of trade ones,
trade statistics can serve as a good early warning
system for eventual debt problems.
In this survey of the literature, the two
most frequently mentioned country risk factors were
both in the trade balance area: debt service/exports
and imports/reserves. It is interesting to note that
each has been proven to be significant in at least
one study, and not significant in at least one study.
Debt service/exports (often referred to as the "debt
service ratio") is defined as principal + interest
due in year "t divided by exports of goods in year
"t." (Manfredi 26) Some researchers have modified the


12
Table 2-1 INDEPENDENT VARIABLES
BALANCE OF TRADE
Exports
Diversity of Exports ................. 3*
Fluctuations in Exports .............. 2
Growth rate of Exports................. 12 - S
Level of Exports...................... 8 S
Current account balance/exports......... 1 - S
Debt/exports............................ 9 - S
Debt service/exports........*......... 27 S NS
Exports/foreign capital inflow.......... 1
Exports/GNP. ......................... 3
Interest payments/exports............... 3
Foreign Exchange Generation................ 1
Imports
Level of Imports ....................... 3 - S
Percent non-compressible ............... 4 - NS
Current Account/Imports................. 1 - S
Debt service/imports.................... 1 - S
Foreign capital inflow/imports........ 1
Imports/GNP............................. 5 - S
Imports/reserves....................... 19 - S - NS
Reserve position at IMF/imports......... 1 - S
Transfer payments/imports............ 1 - S
* The number appearing after an independent variable
indicates how many of the studies suggested the use
of this variable. An "S" means that at least one
researcher tested the variable and found it
significant. An "NS" means that at least one
researcher tested the variable and found it to be not
significant.


13
Table 2-1 (continued)
DEBT
Credit History of Nation.................. 2 S
Amortization
Amortization/debt....................... 7 - S
Debt
Growth rate of Debt..................... 2
Level of Debt ......................... 12
Debt per capita......................... 1
Debt/exports............................ 9 - S
Debt/GNP.............................. 6 - S
Debt/reserves........................... 5
Supplier credit/total debt.............. 2 - S
Debt Service
Debt service/exports................... 27 - S - NS
Debt service/foreign exchange inflows. 5 - S
Debt service/GNP........................ 3 - S
Debt service/imports.................... 1 - S
Debt service this year/total debt. 2 - S
Interest
Interest/average outstanding debt.... 1
Interest/exports........................ 3
Interest/GNP............................ 2


14
Table 2-1 (continued)
DOMESTIC
Economic Growth Indicators
Capital formation..................... 4
Domestic Credit/GNP.................... 1
GNP (growth rate)...................... 8 - S
GNP (per capita)....................... 16 - S
Gross fixed capital formation/GNP....... 1 - S
Industrial Production (growth rate)... 1 - NS
Savings/GNP............................ 2
Total domestic credit.................. 3
Fiscal Policy
Financial structure.................... 1
Government budget deficits............. 5 - S
Military expenditures/GNP.............. 1 - S
Government expenditures/GNP............ 1
Monetary Policy
Capital Flight........................ 2
Consumer Price Index (rate of change). 11 - S
Exchange rate changes.................. 1 - S
Foreign assets of banks/money supply.. 1
GNP/money supply....................... 1
Money supply (growth rate)............. 3 - S
Money supply/government expenditure... 1 - NS
Overvalued exchange rates.............. 6
Reserves (growth rate)................. 2
Reserves (level of).................... 2


15
Table 2-1 (continued)
EXTERNAL
Changes in world financial markets
Interest rates.....................
Floating vs. fixed interest rates..
Private vs. public debt............
Exchange rate changes..............
Exports/foreign capital inflow.....
Foreign capital inflow/reserves....
Inflation in industrialized nations
Oil price shocks...................
Recession in industrialized nations
Tariff barriers....................
Terms of trade.....................
'JHP'OMHHMOHN


16
denominator to include service payments as well as
export of goods. Frank and Cline's rationale for the
importance of this ratio is:
... an increase in the debt service ratio
indicates increased vulnerability to foreign
exchange crises. Any shortfall in foreign
exchange earnings or capital imports which is
not covered by exchange reserves must be met by
reducing imports: since debt service is a fixed
obligation, the higher the debt service ratio,
the greater is the relative burden on import
reduction for a given shortfall in foreign
exchange (329).
Reserves to imports has been defined as: gold
reserves + holdings of dollars or sterling + reserve
position at the IMF (Frank and Cline 332) divided by
imports of goods and nonfactor services (Feder and Uy
144). Many researchers choose to omit the gold
reserves since the value of gold varies so widely.
Feder and Uy explain the significance of the
reserves/imports ratio:
The larger reserves are relative to imports,
the more reserves are available also to service
external debt, making it less likely that the
government would defer such payments.
Therefore, a higher reserve/import ratio is
expected to lead to higher creditworthiness
ranking (Feder and Uy 136).
Debt related indicators are a logical
inclusion in country risk analysis since a country
already over-extended will be a poor credit risk even
if its export earnings increase dramatically year



17
after year. Debt-related indicators are those that
pertain either to the total debt level, or components
of its debt service. To some extent debt service
ratios are also liquidity ratios as they represent
the minima of a government's need for liquidity to
maintain a positive credit rating. (Feder, Just, and
Ross 657) Debt service, comprised of interest +
amortization in year "t", is more pertinent for
short-term loans, rather than long-term.
Aside from the aforementioned debt
service/exports ratio, the level of debt itself is
the debt factor most frequently posed as a country
risk indicator. But how much debt can a nation incur
without risking its creditworthiness?
. . research has been unable to show
conclusively that there is a particular level
of debt measure against exports or another
variable which is too high for all countries
and below which none will be in difficulties:
the search for a critical level has proved
elusive (Calverly 93).
One of the major difficulties in studying
sustainable debt levels has been the lack of complete
debt data. Debt service statistics available today
for previous years show only that debt service
actually paid as opposed to what was owing for a
particular year (Peterson 96). In addition, private


18
and/or short-term lending figures have been non-
existent or sketchy at best. This problem is
particularly severe for the years before 1978 when
the World Bank began publishing short term debt
statistics. (See Appendix B for a further discussion
of limitations posed by data-related problems.)
Domestic factors are those which reflect
internal policies of the debtor nation. Analysis of
macroeconomic fundamentals in a pre-requisite for a
thorough understanding of a nations
creditworthiness. Such familiarity will make possible
the identification of economic vulnerabilities and
appropriate early warning signals of impending
financial distress. Although fiscal and monetary
policy are each important to a thorough analysis of a
countrys future prospects, analysts seem to rely
more on the outcomes of these policies, i.e., GNP and
Consumer Price Index (CPI), rather than focusing on
the underlying fundamentals of whether the
administration is one that promotes or exports or
import substitution, restricts money supply or
controls interest rates. It is unclear whether this
tendency to use outcomes to measure domestic
macroeconomic policies is because the outcomes are


19
easier to quantify, or because they are simply more
pertinent to debt repayment ability.
Per capita GNP is important because it is
assumed that "poor countries may be less able to cut
income to achieve foreign adjustments when
necessary." (Cline, "International Debt" 218) It is
also used as a summary of a variety of other factors.
The level of per capita income is a standard
summary of the wealth of the country and its
level of development, and, in a sense, an
indication of the government's ability to
muster additional resources to solve an
impending balance of payments crisis without
having to disrupt debt service payments. A
higher level of GNP per capita may imply a
higher level of nonessential consumption, which
provides the government with more flexibility
in releasing resources for debt service
payments (Feder and Uy 137).
The Consumer Price Index is often used
interchangeably for a nation's inflation rate. It is
another summary statistic since it reflects the
combined effects of fiscal deficits, exchange rate
policy, terms of trade, money supply and demand for
imports (Khan and Knight 4).
Capital flight is a domestic factor that is
difficult to quantify, so it has been omitted by
those creating quantitative country risk models.
However, Solberg suggests that the line "errors and
omissions" in the International Monetary Fund's


20
monthly International Financial Statistics be used as
a proxy for capital flight. He states that from 1973
to 1984, capital flight (as he defines it) totalled
US$ 96.5 billion for non-oil developing nations --
over 13 percent of the total disbursed external debt
to those countries in 1984 (Solberg 37). Krueger
argues that capital flight is especially useful as an
early-warning signal since, it reflects citizens'
expectations about CPI, exchange rates, etc. (55)
External factors have been difficult for
country risk researchers to incorporate into their
models for two reasons: 1) they are difficult to
quantify, and 2) they are difficult to identify. In
an effort to deal with the first difficulty, William
Cline's 1984 study included a variable for "global
borrowing" which is intended to "capture the changing
condition" in the supply curve of lending (Cline,
"International Debt" 228). As for the second,
Calverley wrote:
Predicting which external factors will be
important over the next few years is likely to
prove equally as difficult as in the 1970s. It
would be rash to suppose that they will be the
same. (55)
Despite these two difficulties, external
factors should not be omitted from country risk


21
models because of their influence on debt- and trade-
related indicators. External instability in money and
commodity markets will aggravate a developing
country's problems of economic management in general
(Khan and Knight 2). In particular, the trend
towards floating rates, higher real interest, and
shorter maturities have had a tremendous impact on
the developing country's debt service burden.
"According to the IMF, a one percentage increase in
the London interbank offered rate [LIBOR] would add
an extra $4 3/4 billion in the gross interest
payments of developing countries." (Gafar 39) Since
over 90 percent of the national external debt
contracted to private banks is done so at floating
rates, the impact of a rise in the six-month LIBOR
rate (from 9 1/4 percent in 1978 to 15 5/8 percent in
1981) is applicable to "old" debt as well as new
(Neuhaus 37 and 38).
There is no indicator within the trade
balance or debt categories that is not directly
influenced by external forces. [John Gafar's 1986
article in The Indian Economic Journal is an
excellent distillation of the effects of external
factors on developing nations' creditworthiness.] A


22
recession in the industrialized countries will lessen
demand for developing nations' exports, worsen terms
of trade and erect stiffer trade barriers that will
impact future, as well as present, prospects for
export development in the future (Gafar 41).
Sixty five independent variables have been
included in Table 2-1. With so many explanatory
variables proposed, the country risk analyst needs to
develop a strong theoretical framework to facilitate
choices for inclusion in a model.
Dependent Variables
There have been far fewer dependent variables
used than independent ones. The question that every
country risk analyst seeks to answer with his model,
"will this country repay this loan that we are
currently negotiating?" demands a "yes" or "no"
answer. Since the answer to this specific question
lies in the future and cannot be tested by the model,
economists are forced to devise other dependent
variables that they believe will yield comparable
information.


23
The most frequently substituted dependent
variable is the answer to a similar question, "Has
this country repaid all of its past debt obligations
in the time and manner specified in the loan
agreement?" Since, a "no" answer most likely means
that a country has rescheduled its debt (see
definitions in Appendix C), country risk analysts
have used rescheduling agreements as their dependent
variable. As Feder, Just and Ross explained,
If a country did have its external debt
rescheduled in a particular year, the value 1
is assigned to the observation. Ex post, the
"probability" of rescheduling equals 1 in such
a case. The value 0 is assigned to
nonrescheduling observations ("Projecting Debt"
654).
There are several sources that publish debt
reschedulings; the World Bank, OECD, and the IMF. But
determining a country's status as "rescheduling" or
"non-rescheduling" can be more complicated than
checking these published lists. One problem relates
to time. There is often a time lag of several months
between interruption of debt service payments and the
commencement of debt rescheduling discussions. After
discussions have begun, it may take as long as twenty
months before they are concluded. Length of
negotiations is usually a function of the number of


24
creditors involved in one case there were 1,200
(Ryser 111). Country risk analysts seeking to
develop an historical data set will only know that
year in which the rescheduling agreement was
concluded; not the year that debt interruptions began
or even the year when rescheduling negotiations
started. Yet this is important information if the
model is to be used to forecast debt servicing
difficulties.
It also needs to be taken into account that
countries may stave off formal reschedulings for
several months through debt service delays, balance
of payments support loans or the imposition of
emergency controls on foreign exchange and imports.
(Amelung and Mehltretter 269). Such manueverings
would certainly signal debt servicing difficulty to a
potential lender; but is difficult information to
obtain.
This is why some researchers increase the
time period that a country is considered as
"rescheduling" from year "t" (the year a rescheduling
agreement is reached) to years "t-2" or even "t-4"
through year "t." Mayo and Barrett's model classified
their countries as to whether "a rescheduling will


25
occur sometime within five years, meaning either in
the current year or anytime up to five years hence."
(Mayo and Barrett 85)
Payments interruptions may or may not lead to
immediate debt rescheduling. Information on these
interruptions is not generally available to anyone
but the debtor and creditors involved. Some
researchers, such as Saini and Bates, have been able
to access this additional information and have
incorporated it into their dependent variable.
Sargen notes that there is yet another
problem associated with this dependent variable.
A problem arises with countries which have
rescheduled debt more than once. . do those
reschedulings represent "new events" or
extensions of the original rescheduling?"
(Sargen 28)
Sargen chose to solve this problem by deleting such
countries from his observation pool.
A distortion of probability estimates can
also occur when countries have negotiated private,
unpublicized debt reschedulings. Since these
reschedulings will not show up on the formal lists
available to the public (and thus country risk
researchers) they will incorrectly given the
dependent variable value showing "not scheduled,"when


26
they have in fact rescheduled (Feder, Just and Ross
653).
A glance at Feder and Just's study would
suggest another variation of the binary dependent
variable: default/non-default. Although Feder and
Just said their study sought to estimate "default
probabilities," and consistently refer to their
dependent variable as "default", it should be noted
that the way in which they defined "default" is the
same as "debt rescheduling." If Abassi and Taffler
are correct in their statement that "no case of
outright default has occurred in the last 30 years,"
then one should assume that "rescheduling" is more
likely the case even if an author uses the term
"default."
When country risk analysis first began, the
major shortcoming of the binary dependent variable
based on restructuring or non-structuring was the
dearth of countries which would fall into the former
category. Some researchers, like Dhonte, found as few
as thirteen rescheduling events, Kugler had 19. Ingo
Walter points out the difficulties this poses for
econometric work.
. . because there are only a few observations
of "difficulties' in debt-service available in
any one year compared with the number of 'non-


27
problem' countries, the weights assigned to the
former are inordinately large and a single
country marginally falling into one or the
other category may significantly affect the
estimated parameters." (qtd. in Melvin and
Schlagenhauf s31)
As the pace of formal reschedulings has continued to
increase this will cease to be a problem. There were
26 formal debt restructurings in 1985 alone.
Rescheduling or non-rescheduling dependent
variables have one additional shortcoming. They can
provide a false positive. This has been recognized by
several country risk analysts, most notably Amelung
and Mehltretter who state, "... debt reschedulings
must not necessarily be the result of debt servicing
problems. A country may renegotiate its debt to
receive better terms, for instance." (Amelung and
Mehltretter 269) Again, there is no distinction made
in the publicly-available lists of rescheduling as to
voluntary vs. involuntary debt rescheduling. A few
models have tried to incorporate information
concerning voluntary debt reschedulings when
available through private channels.
Researchers with access to unpublished data
were able to modify their dependent variables
substantially in order to adjust for the errors
outlined above. For example, Saini and Bates'


28
modified dependent variable included involuntary debt
reschedulings and six balance of payments support
loans ("defined as foreign loans in the absence of
which a rescheduling would have been necessary or
arrears on external payments would have occurred")
("Survey of Quantitative" 346) while excluding
voluntary debt reschedulings. Feder, Just and Ross
augmented their dependent variable selection by
including
. . some instance of "serious arrears' taken
from the World Bank files. They also excluded
renegotiations that were identified as having
occurred "in circumstances of no great economic
stringency" and that were primarily a means of
giving aid (qtd. in McDonald 623).
A few researchers, like Kugler, altered the
basic binary dependent variable and created a three-
choice model: rescheduled that year, never
rescheduled, or rescheduled in a future year.
Zimmerman developed a five stage multinomial
logit model. Countries were assigned a dependent
variable number based on five stages of financial
difficulty. These were: 1. A country in formal loan
default. 2. A country in at least the second IMF
credit tranche and an IMF emergency standby position
as well. 3. A country is in at least the IMF second


29
credit tranche. 4. A country has an IMF emergency
standby arrangement only. 5. A country is not in
stages 1-4 (Zimmerman 46).
There were a few mavericks who chose
dependent variables completely different from the
prevailing rescheduling/non-rescheduling paradigm.
Hajivassiliou used three dependent variables in his
study of external debt repayment problems: presence
of IMF support and/or a request for a rescheduling,
net new loans/exports, level of debt service
obligations in arrears (Hajivassiliou 228). Feder
and Uy used creditworthiness and Heller and Frenkel
the "level of external indebtedness." Dhonte's
principal components analysis used no dependent
variable at all! (qtd. in Saini and Bates,
"Statistical Techniques" 10)
The binary rescheduling/non-rescheduling
classification, despite its shortcomings, remains the
most widely used dependent variable by country risk
analysts. It appears that its shortcomings are due
more to lack of "privileged" data (payments
interruptions, date of commencement of rescheduling
discussions, voluntary reschedulings, etc.) than to
its inherent usefulness in determining country risk.


30
Econometric Technique
The purpose of country risk analysis is to
provide a "yes" or "no" answer to the lending
institution asking the question, "Should we lend
money to this country now?" To answer that question
succinctly most researchers have chosen a binary
choice dependent variable. To accommodate these
givens, the econometric technique must be one that
permits a dichotomous dependent variable. This
eliminates the possibility of using a standard
regression model where y = a + bx + e because the
"yes/no" or "0/1" dependent variable has an error
term that cannot be normally distributed. Any
attempts to apply standard regression analysis to a
"0/1" dependent variable will fail.
In order to circumvent this problem, some
researchers have used less well-known techniques:
principal components analysis (Dhonte; Abassi and
Taffler), "state-space" form based on the Kalman
filter recursive algorithm (Melvin and Schlagenhauf),
cluster analysis (Schmidt), univariate analysis
(Schmidt), tobit (Hajivassilou) and economic block


31
decomposition analysis (Kim).
Despite these creative forays into
econometric technique, discriminant analysis and
logit have been the most widely used in country risk
analysis. Discriminant analysis uses that "yes/no" or
"0/1" decision as an exogenous variable; logit uses
it as the dependent variable.
Discriminant Analysis
Discriminant analysis begins with the assumption
that observations (one year for one country) can be
divided into one of two categories (rescheduled, non-
rescheduled). After that division, an attempt is made
to uncover the independent variables responsible for
those classifications. Saini and Bates provide a more
technical description of discriminant analysis:
The discriminant analysis is used to
discriminate between two groups of countries by
means of a set of explanatory variables XI, X2,
. .Xm. The discriminant function Z = sum of
WiXi, (i=l,2,...,m) represent a linear
combination of the explanatory variables. The
Z-value for each country is then computed by
substituting the values of the country/year
characteristics into the estimated discriminant
function. Performing this operation for each
rescheduling (and balance of payments support
loan) and non-rescheduling country yields a
frequency distribution of Z-values for each
group from which mean Z-values are computed.
The critical value (i.e., the value which
distinguishes rescheduling from non-


32
rescheduling cases) depends on the weights
assigned to the Type I and Type II errors
("Survey of Quantitative" 344).
Multicollinearity. Discriminant analysis for
country risk studies has been criticized for several
reasons. The most frequently voiced criticism of this
procedure is that of multicollinearity. Perfect
multicollinearity occurs when two independent
variables sum to a third. Although perfect
multicollinearity is extremely rare, imperfect
multicollinearity is frequently found in country risk
studies and occurs when two or more explanatory
variables are highly correlated. The higher the
correlation between variables, the less accurate is
the estimation of the coefficients (Studenmund and
Cassidy 179). According to Abassi and Taffler,
discriminant analysis is particularly prone to
multicollinearity errors because so many of the
ratios used as explanatory variables are calculated
from a small set of indicators (13). The net result
of studies with imperfect multicollinearity is that,
"both the signs and the stability of the coefficients
are suspect" (Abassi and Taffler 17).


33
Normality assumption. Mayo and Barrett feel
that the "principal flaw" in discriminant analysis is
that it violates the normality assumption (83).
Although Studenmund and Cassidy state that the
normality assumption is optional, they also hasten to
add that it is preferable since t-statistics are not
really applicable otherwise (66). Although there is
some debate as to whether or not t-statistics should
be applied to discriminant analysis coefficients, it
should be noted that Frank and Cline used them in
their 1971 study.
Serial correlation. Serial correlation is the
violation of the assumption that the error terms are
independent of each other (Studenmund 233). Sargen
points out that discriminant analysis studies have
pooled cross-sectional and time-series data which
leads to serial correlation. His argument is that a
country with a large deviation in one indicator will
usually continue to have that deviation in other
years and that this will affect the error rates
(Sargen 28). This criticism, however, can be equally
applied to logit analysis since these studies also
combine time-series and cross-sectional data.


34
The most serious charge against the use of
discriminant analysis for country risk studies comes
from Saini and Bates:
Another shortcoming of discriminant analysis is
the lack of any behavioral underpinning.
Although designed for analysis using a binary-
valued dependent variable, the discriminant
function assumes two distinct groups of
countries, implying that a country suddenly
becomes a member of the other group when it
reschedules. However, historical experience
indicates that a country reschedules its debt
after the combined effect of certain economic
variables reaches some critical threshold value
at which the government views rescheduling or
its equivalent as the most desirable course of
action ("Survey of Quantitative" 352).
Logit
In contrast to the discriminant analysis
model, logit analysis assumes classification of a
country as rescheduling or non-rescheduling as an
endogenous variable and by definition it becomes the
dependent variable.
Logit facilitates the transformation of the
standard regression analysis model into one where the
errors are normally distributed. It generates
coefficients, standard errors and t-statistics
similar to those used on Ordinary Least Squares
regression. In addition, logit generates a log
likelihood statistic that can be manipulated to test


35
for significance of the model.
Testing of the null hypothesis (i.e., testing to
see if the model is worthless) with a logit-tested
model begins with a formula to derive the maximum
score possible IF the model were useless (Judge et
al, 767). That maximum log-likelihood is equal to:
N [ (In N/T) ] + (T N) In [ (T N)/T]
(equation 2.1)
N = number of dependent variables = "1" (success)
T = total number of observations
The log likelihood score for the actual model
being tested is subtracted from the maximum log
likelihood and the difference is multiplied by 2.
This number is then compared to the critical value.
The critical value is obtained from a table of
critical values for chi squares. The degrees of
freedom number is obtained by subtracting one from
the total number of independent variables, including
the intercept. If the score for the log likelihood
is greater than the critical value, the null
hypothesis can be rejected. Rejection of the null
hypothesis means that the model does have predictive


36
value.
Unfortunately logit does not generate an R2
number to test for "goodness of fit." It does,
however, generate a pseudo R2 that gives a rough idea
of how much of the variation in the dependent
variable is due to the coefficients of the
independent variables. The formula for that pseudo R2
is: 1 (log likelihood of the model/maximum log
likelihood) (Judge et al 774). The closer this
number is to zero, the less explanatory the model is.
Feder and Just provide a persuasive argument
for the use of logit analysis in country risk
studies.
. . it makes more sense to claim that, in a
specific period, the country was pushed beyond
a critical level, leading to a rescheduling,
than to claim that the country suddenly became
a member of another species. In addition, more
appropriate statistical tests can be performed
to determine the relevance of various economic
indicators of debt servicing capacity (26).
Furthermore, they were quite satisfied with logit's
predictive performance. They obtained only 6-11
errors in predicting rescheduling or non-rescheduling
with 238 observations (26).
Melvin and Schlagenhauf are probably the most
outspoken opponents of using logit for country risk
analysis.


37
Economic models of country risk are generally
concerned with predicting the probability of
debt default or rescheduling due to country-
specific factors. Unfortunately, there are few
observations on countries that actually default
so that the resulting probabilities from LOGIT
analyses of the problem are at best mildly
suggestive, and at worst, dangerously
misleading ("A Country Risk Index" 601)).
Schmidt faults logit analysis models since
they "only reproduce the historic development because
they are not related to future events." (358) This is
a criticism that could have been applied to any of
the country risk modelling reviewed here.
Logit analysis does have some attributes in
its favor; it is not as prone to multicollinearity
errors as is discriminant analysis and it does not
violate the normality assumption. After critiquing
over a dozen country risk studies McDonald concludes
that logit analysis is "most suitable" for this type
of work (623).
Some researchers decided to submit the same
data set to both logit and discriminant analysis as a
way to determine which methodology was more accurate.
Saini and Bates used both techniques in their 1978
study. They found "no significant difference in the
error rates and the coefficient values generated by


38
the discriminant and logit functions." ("Survey of
Quantitative" 347)
Manfredi tested Frank and Cline's
discriminant analysis and Feder and Just's logit
model using the same data set. She found both models
about the same for Type I errors, but the logit model
superior for Type II errors. Type II errors are false
predictions of reschedulings (29).
Schmidt went so far as to test four
econometric techniques: univariate, multiple
discriminant analysis, cluster analysis and logit.
His conclusion was that logit was clearly better than
the others (369). Zimmerman tried to improve on the
binary logit model by devising a five-stage
multinomial logit model. He tested both on the same
data set and found, "the multinomial model is
approximately equal to earlier binary logit models in
both stability and "reasonableness" of country
specific probabilities. The multinomial model is
unfortunately quite costly to use and may in many
applications not be worthwhile." (3)
Although logit has been used in more studies
than discriminant analysis and it appears to be at
least the equal, if not the superior, to discriminant


39
analysis there is still room for improvement.
. .the results of logit analysis are still
difficult to interpret because of the lack of
any explicit procedure for selecting the
critical probability value (or, that value
which distinguished rescheduling from non-
rescheduling cases) . Feder and Just select
the critical value so as to minimize Type I and
II errors. While this method efficiently
classified past debt reschedulings, it may not
be helpful in forecasting a country's debt
servicing difficulties (Saini and Bates "Survey
of Quantitative 352).
It should be remembered that forecasting with
binary dependent variables will always be more
difficult than with a continuous dependent variable.
There is no exact equivalent of the R2 value for
these techniques to accurately provide the
researcher with the "goodness of fit" for his
estimated equation (Pindyck and Rubinfeld 300,301).
Until a new econometric technique is invented
that improves on logit, however, it will probably
remain the tool of choice with forecasters if for no
other reason than that it lends itself so readily to
computer applications. Several software programs for
logit are already available.


CHAPTER 3
QUANTITATIVE STUDIES
Frank and Cline conducted the first major
quantitative country risk analysis study. Their
effort has been followed by more than a score of
others. Some chose to re-test, with some
modifications, the same independent variables that
Frank and Cline employed. Since the Frank and Cline
model is so well-known it should be a useful starting
point to trace the progression of country risk
analytical work. This chapter will present the
results of Frank and Cline's study, as well as two
other studies that closely paralleled it, and will
close with a critique of this approach.
Frank and Cline
Frank and Cline's study, published in 1971,
used data for 26 countries over the time period 1960
to 1968. Since it has been hailed as the
"pioneering" and "seminal" work in this field, it


41
seems an appropriate starting point for an analysis
of country risk theoretical frameworks.
Frank and Cline used eight factors which they
"felt might have an influence on the capacity to
service debt." (330) Due to unavailability of data,
there were 145 observations drawn from this pool of
countries; only 13 instances of rescheduling were
found. Their stated purpose was "to find an index or
indicator of the likelihood that a less developed
country will experience debt servicing difficulties."
(329)
They used a one-year lag for their
independent variables since they assumed that debt
payment interruption in year "t" would only occur
after financial difficulties were encountered in year
"t-1." Those eight independent variables are:
1) Debt service ratio (debt service "t-1" /
exports "t-1"). This is by far the most widely used
financial indicator in country risk analysis. The
rationale for this ratio is that it shares a positive
relationship with vulnerability to unfavorable
foreign exchange fluctuations. [It should be noted
that Frank and Cline used what they called estimated
normal exports rather than actual exports "under the


42
assumption that authorities pay little attention to
temporary highs or lows in exports but base decisions
on what normal exports are expected to be." Exports
for t-1 were calculated as the "predicted exports in
year "t-l", based on a regression of the logarithm of
exports on time for the five-year period ending in
year "t-l." (331)]
2) Growth rate of exports (the average of the
exports for years t-4 to year t-l minus the average
of the exports for years t-8 to t-5; divided by the
latter. This is presumed to have an inverse
relationship to rescheduling because the higher the
growth rate of exports, the brighter were future
expectations and the less incentive to default.
3) Export fluctuation index (average absolute
percentage deviation from an eight-year trend from
year "t-7" to "t". Frank and Cline reasoned that
stability in exports would yield less vulnerability
for the debtor nation to exogenous factors.
4) Non-compressible imports (non-compressible
imports/imports). They hypothesized a positive
relationship between this ratio and an inability to
service debt in the event of a balance-of-payments
crisis.


43
5) Per capita income (Total GNP / ;
population). This ratio was included because, "It
would seem likely that the lower per capita income,
the less flexibility there would be for reducing
consumption and thus, the more likely debt
rescheduling." (331)
6) Debt amortization in year "t" /total
outstanding debt in year "t-1". Debt amortization is
pay-back of principal and a low value here would show
a debt structure heavily weighted to long-term debt.
Frank and Cline felt this would reduce a country's
flexibility in adjusting to short-term difficulties
and lead to more rescheduling.
7) Imports/GNP. A positive correlation was
assumed between this ratio and the need to reschedule
because they thought that a nation with a higher
import/GNP ratio would have its production capacity
curtailed if forced to restrict imports in order to
meet debt obligations.
8) Imports/reserves. "Reserves" is defined
here as international reserves, including foreign
exchange, gold holdings and net position at the IMF.
There should be an inverse relationship here because
the more reserves the less likelihood that there will


44
be difficulty in obtaining foreign exchange for debt
payments.
Frank and Cline applied a modified
discriminant analysis to this model and made "Z"
stand for the debt servicing capacity. They sought to
find a critical value of Z so that the Z reading for
a country would classify it as having "reached the
limit of their debt servicing capacity."
They found three of their eight explanatory
variables to be significant at the five per cent
level: debt service/exports, amortization/debt and
imports/reserves. It is interesting to note that they
wrote, "Although the assumptions of regression
analysis are not appropriate in discriminant
analysis, we found it useful to apply the usual
linear regression tests to obtain some notion of the
relative importance of the various variables. "
Additional iterations reduced their model to
only two independent variables, Xx (debt
service/exports) and X2 (amortization/total external
debt). Their final equation forecasted a rescheduling
if:
35.6 Xi2 342.8XxX2 42.4 X22 + 42.1 Xx + 73.1 X2
is greater than or equal to 9.643. (equation 3.1)


45
Interestingly enough, when Frank and Cline
published their study in 1971 they attempted to use
their model to predict rescheduling/defaults for 17
countries for 1968 to 1987. Their results were not
too encouraging. They grouped the 17 countries into
three categories of risk. They felt Group I countries
would face "serious debt servicing problems
regardless of aid assumption used or export growth
rate." Group III countries "should not face serious
problems under the various assumptions made." (341)
The countries in Group III are: Mexico, Argentina,
Bolivia, Iran and Nigeria. According to published
reschedulings by the International Monetary Fund for
1976 to 1987 four of the five in Group III have had a
combined total of 16 reschedulings. Iran is not
listed as a formal rescheduler, but its immersion in
a long and costly war with Iraq could not have been
predicted by Frank and Cline's model anyway.
The poor showing of their long-term
predictions could well be due to the incorrect
assumptions they made about export growth and terms
of foreign lending . but in at least one instance
(Mexico) their assumption (export growth rate of 8
per cent per year) was far below the mark of reality


46
(export growth achieved 43 per cent annually from
1971 1981). Even with this error in its favor,
Mexico wound up rescheduling six times since 1982
despite its Group III classification (Calverly 21).
Feder and Just
In 1976 Feder and Just published the results
of their country risk analysis study in the Journal
of Development Economics. They had used Frank and
Cline's earlier model as a departure point for their
own investigations. They began with nine explanatory
variables -- seven of them defined as Frank and Cline
had theirs. [Their export fluctuation index was
derived a little differently as they assigned equal
weights to all eight years.] They dropped the non-
compressible imports because
. . the data used for calculating it were not
comparable among countries for all of the years
used and because theoretical arguments have
been developed which qualify this indicator.
The usual argument is that imports of various
consumption goods, which are not vital
necessities, can be curtailed temporarily so as
to increase availability of foreign exchange
for debt servicing purposes. The assessment of
this factor thus requires detailed data on
import composition patterns. Moreover, there
may be raw materials and intermediate goods
that are imported for production of domestic


47
nonessential goods which can be reduced;: but
separation of these from other intermediate
goods is usually impossible (27).
Feder and Just added two indicators of their
own: total capital inflows/debt service payments and
growth of per capita income. [Capital inflow/debt
service was defined as net capital inflows (short and
long term), including direct investments and grants
for year t-1 divided by debt service on public or
publicly guaranteed debt for year t-1.]
They hypothesized that higher capital inflows
would lead to fewer cases of rescheduling due to
increased foreign exchange. They expected an increase
in per capita GNP to increase the savings available
to the debtor nation. They reasoned that increased
output, as evidenced by higher GNP, would provide
additional resources for investment, consumption and
debt repayment.
They opted for logit analysis rather than
discriminant and used "default" as their dependent
variable. Feder and Just defined default in such as
way that it was really a synonym for rescheduling.
Their data set contained 238 observations, 217 of
them non-defaulting, for 41 countries over the time
period of 1965 to 1972.


48
Feder and Just found six of their eight
explanatory variables to be significant at the five
percent level: debt service/exports,
imports/reserves, amortization/debt, income per
capita, capital inflow/debt service and export
growth. It is interesting that they found two of
Frank and Cline's variables to be significant (income
per capita and export growth) when Frank and Cline
did not.
The final form of the Feder and Just model
is:
In P = 59.2 X1 + 0.4 X2 39.6 X3 .01 X4
- 2.9 X5- 52.6 X6
Xx = debt service/exports
X2 = imports/reserves
X3 = amortization/debt
X4 = per capita income
X5 = capital inflow/debt service
X6 = growth rate of exports
Manfredi
Eileen Manfredi decided to apply both the
Frank and Cline and the Feder and Just models to a


49
data set she developed for the time period 1972 -
1977. She used 60 developing countries to obtain
approximately 300 observations; 13 of those were
instances of rescheduling.
Her experimentation found the Frank and Cline
variables still significant, but showed an
improvement with the inclusion of a capital flows
variable.
The capital flows variable, with its sharp
changes in magnitude and direction, gives the
Feder/Just model a greater degree of annual
variation in the probability of rescheduling
and fewer false predictions than does the
Frank/Cline model. Many countries with
unfavorable levels for other variables can
avoid rescheduling their debts if they can
maintain capital inflows at a high rate.
Continuous flows of grant aid to the poorest
countries and private capital flows to the
developing countries with higher incomes can
serve this purpose (28).
Because of the timing of her data set (1972-
1977) she was able to test for the effect of the
first major OPEC price increase on these explanatory
variables. Her conclusion was that "... there was
no significant difference between the accuracy of
predicting debt reschedulings in the years before the
first major oil price increase by the Organization of
Petroleum Exporting Countries (OPEC) and those for
subsequent years." (Manfredi 26)


50
Critique ~
Manfredi's is the last published application
of the Frank and Cline study. Her data set ended in
1977 so I decided to apply Frank and Cline's
significant explanatory variables to a more recent
data set. [Cline's 1984 study included two of his
three significant variables from the 1971 study to a
data set for 1968-1982, but he did not reapply his
original model to the new data set.] I obtained 393
observations over the time period 1975 to 1986; 142
of them were classified "rescheduling." These appear
on Table 3-1 as "0" in the dependent variable
category. I modified the time lag somewhat by
extending the definition of rescheduling to mean the
year of rescheduling as well as the two years
immediately preceding the rescheduling. I also
decided to use logit analysis for the reasons
outlined in the econometric technique section above.
The coefficients of the three variables Frank
and Cline found to be significant (debt
service/exports, imports/reserves and
amortization/debt) still proved to be significant at
the 5 percent level.


51
Frank and Cline found just two of the ;
variables (debt service/exports and
amortization/debt) to be explanatory enough to drop
the third variable from their model. For the time
period of their data set these two variables
generated an error rate of only 12.4 percent (Frank
and Cline 336). In my later application the model
containing those two variables alone had a lower
significance rating than the model with debt
service/exports, amortization/debt and
imports/reserves.
If today's country risk analyst simply used
Frank and Cline's two variables to predict
reschedulings their error rate would be greater than
the 12.4 percent Frank and Cline generated in 1971.
The chi square test shows that the logit model with
the two Frank and Cline variables (debt
service/exports and amortization/debt) is
significant. The pseudo R2 is often used in logit to
give some idea of the explanatory value of a model.
An R2 value of only .22 shows clearly that its
explanatory value for post-sample prediction is very
low. All of the other three variations tested on the
more recent data set had R2 values equal to, or


52
Table 3-1 LOGIT CALCULATIONS
Number of observations................... 393
Cases with dependent variable = 1 ....... 251
(non-rescheduling cases)
Cases with dependent variable = 0 ....... 142
(rescheduling cases)
Test #1 Debt Service and Amortization
Variable Coeffic. Std. Err. T-Stat. 2-Tail Sig.
INTERC. DSR AMDEBT 0.161435 -0.094511 29.60987 0.260201 0.013947 3.883467 0.620426 0.535 -6.776446 0.000 7.624595 0.000
Number of iterations Log likelihood for convergence 3 -199.55670
Test #2 Imports / Reserves And Amortization
Variable Coeffic. Std. Err. T-Stat. 2-Tail Sig.
INTERC. IMPRES AMDEBT 0.408408 -0.291851 19.82154 0.302226 0.049489 3.507681 1.351331 0.177 -5.897313 0.000 5.650895 0.000
Number of iterations Log likelihood for convergence 3 -193.01241
Test #3 Combined Variables
Variable Coeffic. Std. Err. T-Stat. 2-Tail Sig.
INTERC. DSR IMPRES AMDEBT 1.615455 -0.105630 -0.300591 29.63445 0.359472 0.015016 0.051223 4.191209 4.493971 0.000 -7.034382 0.000 -5.868251 0.000 4.493971 0.000
Number of iterations Log likelihood for convergence 3
INTERC
DSR
IMPRES
AMDEBT
= Intercept (constant)
= Debt Service / Exports
= Imports / Reserves
= Amortization / Debt


Table 3-1 LOGIT CALCULATIONS
Number of observations .................... 393
Cases with dependent variable = 1 ........... 251
(non-rescheduling)
Cases with dependent variable = 0 ........... 142
Test #4 Debt Service and Imports/Reserves
Variable Coeffic. Std. Err. T-Stat. 2-Tail Sg
INTERC. 2.95309 0.315252 0.36738 0.000
DSR -0.06250 0.011789 -5.30124 0.000
IMPRES -0.32311 0.047830 -6.75528 0.000
Number of iterations for convergence = 3
Log likelihood = -199.22990
INTERC.
DSR
IMPRES
Intercept (concept)
Debt Service/Exports
Imports/Reserves


54
higher than, the Frank and Cline pair. (See Table
3 -2 for values for the other models.)
Given that the highest R2 value for these
variations on the Frank and Cline model is rather
low, only .36, the use of these independent variables
to predict reschedulings must be seriously
questioned.
Why does Frank and Cline's model have such
little applicability today? To begin with, the
rationales for two out of their three significant
variables is poor. Even Frank and Cline stated that
they were uncomfortable about the debt
service/exports ratio.
The historical behavior of debt service ratios
and instances of default also indicates an
ability of some countries to tolerate high debt
service ratios. Mexico and Israel have not
defaulted nor requested debt rescheduling
despite debt service ratios of 39 and 26 per
cent respectively in recent years. Australia
managed to avoid defaults on public and private
debts with an investment service-exchange ratio
ranging from 43 to 44 percent during the period
1930-1934. Canada avoided defaults and the
imposition of exchange restrictions on current
transactions with an investment service-
exchange earnings ratio of 32 to 37 per cent
over the 1931-33 period. On the other hand,
Bolivia, Brazil, Colombia, Cuba, Peru, and
Uruguay defaulted in the period 1931 1933
with debt service ratios that were generally
lower, on the order of 16 to 28 per cent (330).


55
Table 3-2 PSEUDO R2 CALCULATIONS ;
Formula for calculating pseudo R2 values:
1 (log likehood of observed model/maximum log
likelihood if model is useless)
Debt Service and Amortization/Debt
1 199.56/255.1 = 0.22
Imports/Reserves and Amortization/Debt
1 193.01/255.1 = 0.24
Debt Service, Imports/Reserves and Amortization/Debt
1 163.50/255.1 = 0.36
Debt Service and Imports/Reserves
1 199.23/255.1
0.22


56
As for the amortization/debt ratio, Frank and
Cline's theory for its inclusion and direction may
have been sound in 1971, but developments in the
world financial markets since 1971 have served to
diminish, if not obliterate, its utility. Their
conclusion was that a country with predominantly long
term debt liabilities lacks shortrun flexibility in
reducing debt service commitments; which may in turn
force the country into rescheduling. But high world
interest rates of the late 1970s and 1980s made older
long term debt, contracted at lower rates, more
desirable than a series of short-term loans which
would have had to be renegotiated at ever increasing
interest rates. The present practice of using
floating interest rates also makes the long term
fixed interest rates look much more desirable in
hindsight. The passage of time may well alter the
rationale for this variable altogether as loan
repayment periods are becoming much shorter.
This only leaves the imports/reserves ratio
(the significant variable they chose to leave out of
their model) with its original rationale intact. Its
addition to the debt service/exports and


57
amortization/debt variables raises the R2 value to
0.36 from 0.22.
Frank and Cline, Feder and Just, and Manfredi
all worked with data sets that contained far fewer
instances of rescheduling than could be hoped for.
Frank and Cline's data set held 145 observations but
only 13 were instances of rescheduling. A fifty/fifty
split between rescheduling and non-rescheduling would
be ideal, but results obtained from an observation
pool where less than ten per cent of cases are
rescheduled are certainly suspect.


CHAPTER 4
TOWARDS A NEW THEORETICAL FRAMEWORK
With at least 65 independent variables to
choose from, the country risk analyst seeking to
develop a model to forecast for the 1990s will need a
strong theoretical framework to narrow the field. A
search for theoretical underpinnings could be
undertaken in at least three directions:
1) historically, in an attempt to identify indicators
that would have forecasted debt servicing
difficulties of the 1980s 2) politically, because not
all cases of debt interruption have been due to
inability to repay and 3) heuristically, to see if
the right questions have been asked in the past to
guide researchers in creating their country risk
models.
A Glance Backwards
National debt repudiation goes back to 1327
when King Edward III of England defaulted on debts to


59
Italian banks. By 1880 "54 percent of foreign -
government obligations were in default." Denmark,
Germany, Austria, the Netherlands, Spain, Greece,
Portugal, Russia, Turkey, and Egypt had all defaulted
at least once before the twentieth century dawned
most were repeat offenders. Maryland, Pennsylvania,
Mississippi and Louisiana defaulted on loan principal
in the 1800s (Suratgar 77).
In some instances, creditor nations would
actually take it upon themselves to administer part
of a debtor nation's affairs in order to collect
revenues for delinquent loans. Turkey and Egypt lived
through these foreign intrusions in the 19th century.
But when European powers attempted to physically
exert this authority in Venezuela, they were halted
by the Drago Principle and the Roosevelt Corollary.
The Drago Principle stated that
. . "the public debt of an American state
cannot occasion armed intervention, nor even
the actual occupation of the territory of
American nations by a European power." This
principle found general acceptance given the
determination of Theodore Roosevelt to enforce
it with the American Navy (Suratgar 80).
But when the Dominican Republic defaulted on
its external debts in 1904 and European creditors
threatened foreclosure


60
President Roosevelt announced . since the
United States could not permit European nations
forcibly to collect debts in the Americas, the
United States must itself assume the
responsibility of seeing that "backward" states
fulfilled their financial obligations. He
placed an American receiver-general in charge
of Dominican revenues, arranging to apply 55%
of customs receipts to the discharge of debts
and directing the balance to current expenses.
The Dominicans actually came out ahead. Under
the old system, the government officials had
only left 1Q% after their depredations. Similar
interventions by the United States occurred in
Haiti, Honduras and Nicaragua and the United
States did not abandon the Roosevelt Corollary
until 1930 (Suratgar 80).
During the international depression of the
1930s seventeen Latin American nations defaulted. By
1950, however, 14 had cleared up their arrears and
had their creditworthiness restored. As a consequence
of earlier default history private lenders were
reluctant to extend credit to sovereign nations for
several decades. Sovereign debt was largely
government to government lending from World War II
until the mid 1970s (Suratgar 82).
The Paris Club was formed in 1956 to
reschedule Argentina's debt (Krueger 168) but the
incidence of rescheduling was initially low. From
1956 to January 1978 the Paris Club had only
rescheduled debts for a total of eight countries
(Suratger 126). The arena of soveriegn debt was to


61
change dramatically by the late 1970s, however:, and
by the end of 1981 the Paris Club had held
renegotiations for 17 countries.(Suratger 112) In
1983 alone there were 17 Paris Club reschedulings
(Keller and Weerasinghe). 3) Arrearages were also
up; the IMF reports that only three nations were in
arrears in 1974, but 15 were in arrears in 1975. By
1981 32 nations were (Suratgar 83). Mexico's crisis
in 1982 was a signal to the international credit
market that sovereign debt was reaching global
crisis proportions.
As we look to the "past" to develop a
theoretical framework for forecasting future country
risk, the period prior to 1974 can be omitted.
Measures taken under the Roosevelt Corollary can no
longer be applied. Economic circumstances unique to
the Great Depression of the 1930s are unlikely to be
pertinent to country risk issues of the 1990s. The
exogenous factors of the two world wars and colonial
imperialism are not likely to affect credit
conditions of the 1990s either. Finally, a scarcity
of data for the period before 1970 would preclude any
attempts for researchers to quantitatively correlate
financial indicators with debt performance.


62
What were the factors contributing to the
increase in both magnitude and incidence of national
debt servicing difficulties? Anne Krueger says the
explanations can be divided into three groups: 1)
unexpected changes in worldwide conditions 2) the
unsustainability of the debt build-up in the 1970s
and 3) debtor nations' "unwillingness or inability to
adjust to the harsher economic realities of the
1980s." (Krueger 166)
Unexpected Changes in Worldwide Conditions
Oil prices were the most severe exogenous
shock during this period. The price of oil quadrupled
from 1973 to 1974. Saudi Arabian Ras Tanura went from
$1.90 per barrel in 1972 to $32.50 in 1981.
(Solberg 7) The first oil price shock had two types
of effects on international debt: 1) it increased the
trade balance deficits of the non-oil exporting
nations and 2) it fuelled the growth of the
Euromarkets. OPEC nations deposited US $100 billion
in the Eurocurrency markets from 1974 to 1982. Of
that, 49.5 percent had to be recycled to non-oil
developing nations just to cover their trade balance


63
deficits (Solberg 5). -
Higher oil prices from OPEC led OECD
countries to reduce their demand for imports, raise
interest rates, and some of them entered into a
period of recession. The sharp drop in commodity
prices was especially a burden to the non-oil
exporting developing nations, most of whom were
dependent on exports of primary commodities for
generation of foreign exchange. The real rate of
growth of world trade had averaged 8.5 percent per
year from 1963 to 1972; this was reduced to only 4
percent from 1973 to 1982 (Solberg 6). Heller
estimates that total exports from developing nations
dropped from $595 billion in 1981 to $508 billion in
1985; largely due to recessionary policies enacted by
OECD nations (Heller 174).
Unsustainability of debt. Debt levels of
developing nations were on the rise. "Outstanding
public debt of non-oil developing countries expanded
nearly fourfold in nominal terms between 1972 and
1979 but by only 56 per cent in real terms." (Kincaid
46) Total outstanding debt of non-oil developing
countries was 146.8 billion U.S.$ in 1975, by 1982 it


64
was 502.2 billion (Neuhaus 37). Debt levels relative
to GNP also rose: from 22% in 1973 to 25% in 1980 and
37% in 1985 (Allsopp v).
The composition of debt was also changing
from primarily public to primarily private creditors
(Neuhaus 37). In 1978 non-oil developing countries
borrowed 33 billion from private sources, in 1981
that figure was 78 billion (Allsopp vi) This is a
significant factor in servicing debt because private
debt terms tend to be of shorter duration than public
and interest rates to be floating rather than fixed.
The increase in short term loans has increased the
need for debtor nations to roll-over or reschedule
their debts. The debtor's costs for loan re-financing
can be significant (Solberg 15).
At the same time foreign direct investments
and foreign aid flows were reduced. Such non-debt-
creating monies "had financed over 40 percent of the
borrowng requirements of the non-oil developing
countries in 1973, by 1981 this proportion had fallen
to only 19.5 percent." (Solberg 9)
The dollar climbed 50% in value from 1981
until 1985. About 75% of developing nation debt is
denominated in U.S. dollars (Allsopp and Joshi xii)


65
so this would have resulted in an increase in their
"real" debt level even if they denied themselves
additional monies (Allsopp and Joshi iv).
Interest rates. Interest rates rose sharply
from 1978 to 1981. Six month LIBOR rates increased
from 9 1/4 percent to 16 5/8 percent. For non-oil
producing developing countries this increase in
interest rates resulted in a rise in long-term debt
levels to $37.5 billion in 1981 (or 8 1/2 percent of
their exports) from only 4 1/5 percent of their
export earnings in 1973 (Neuhaus 37). The swing in
real interest rates was even more dramatic. The real
six month LIBOR rate was -0.6 in 1980 and 21.6
percent in 1981 (Solberg 28).
A decline in export earnings, currency
revaluations and higher interest rates conspired to
raise the debt levels relative to means of repayment
even if a nation had not contracted any new debts.
A look at the historical record of the past decade or
so would indicate a need for the inclusion of debt
level and exogenous variables such as international
interest rates.


66
Debtor nation adjustment policies. In response
to all these exogenous shocks it would have been
prudent for developing countries to reduce their
import levels, lessen dependence on primary commodity
exports through liberalization of trade regimes,
curtail capital flight to bolster domestic capital
formation and devalue their currencies. The fact that
they did not do these immediately led to increasing
their borrowing requirements every year until 1981.
By 1985 developing nations' adjustment
programs resulted in a drop in imports of $82 billion
(or 14 percent) from their 1981 level (Heller 171).
The level of borrowing has decreased rapidly since
1982 and was only 10 billion in 1985 (Allsopp and
Joshi vi). Adjustment may not have occurred as
rapidly as would have been optimal, but it appears
that it did occur in a relatively brief period. A
model would need to include variables on compressible
imports, diversity of exports and domestic monetary
controls in order to predict how well, and how
rapidly, a nation could adjust to exogenous shocks.


67
Political Risks :
Most country risk models focus purely on
indicators of a nation's ability to repay its debts.
Accordingly they include factors that they believe
will help answer questions like: will a country earn
enough foreign exchange through exports to pay back
this loan in three years? If not, can its import
level be reduced enough to free up the funds
necessary for debt servicing? These types of
indicators are certainly critical to loan decisions.
You cannot "squeeze blood out of a turnip," or
extract a pound of flesh from a Prime Minister if a
nation simply doesn't have any liquid assets. It
would be foolish to lend $100 million to a country
with no hope of ever repaying that sum.
Some analysts point out that the chances of a
nation defaulting on its debt due to insolvency (or
"national bankruptcy") are practically nil. They
argue that there are always drastic, albeit
unpleasant, measures that a country could employ:
natural resources could be extracted more rapidly,
domestic consumption cold be reduced or additional
taxes levied to facilitate debt servicing . if a
nation were willing to do so (Allsop and Joshi xxi).


68
Willingness to pay could be even more purely
politically motivated. A national leader might choose
to interrupt debt payments (even though his nation
had the resources available for such payments) as
"... a popularity boost for the government because
a moratorium proves popular for ideological or
nationalistic reasons." (Calverly 19) Cases of
politically-inspired debt interruption or repudiation
are rare, but they have happened (for example, Cuba,
Peru and Nicaragua) (Cabalerros 130).
Seen in this light, the question of ability
to repay seems less important than the willingness of
nation to repay. This realization would seem to
demand the incorporation of political risk evaluation
into country risk models.
Defining political risk is difficult enough,
Quantifying it is well-nigh impossible. Feder and Uy
solved this problem by using a political turmoil

dummy variable to indicate:
. . countries where there is an adverse
external or internal political situation such
as military conflict (e.g. Iran-Iraq), internal
insurgency (e.g., El Salvador), violent
nondemocratic change of regime, etc. Countries
experiencing such phenomena are hypothesized to
be viewed as bad risks, ceteris paribus, and
the coefficient of the political risk dummy
variable should be negative (137).


69
Their study found that an adverse political rating
would lower a country's credit rating by
approximately five percentage points.
Other researchers have acknowledged the need
for consideration of political risk, but have chosen
to disregard the factor entirely. "The main reason
for excluding them [political risk factors] is that
they are very difficult to quantify and forecast."
(Feder, Just, and Ross 656)
Political risks leading to unwillingness to
repay are extremely difficult to quantify, but is
this sufficient reason to ignore them?
Graham Bird, writing in Lloyd's Bank Review,
sees the issue of willing to repay from a different
perspective:
. . the distinction between the ability to
service debt and the willingness to do so may
be largely illusory. Although economic
theorists have made much of the separation
between them, willingness and ability might be
more realistically seen as being positively
related. This, in turn, has implications for
the distinction between economic and political
factors. Is very much to be gained by examining
political factors separately? Not only is it
exceedingly difficult to evaluated them in any
scientific way, but also it may be argued that
political factors are in any case usually
proxied by economic variables (7).
This opinion is also held by John Calverly


70
who uses a case study of the Korean debt crisis of
1980 to conclude that political instability
frequently, if not universally, coincides with
economic difficulties (36). Based on that insight,
he shares his rationale for using five socioeconomic
indicators, that are quantitative, to "form an
essential background to political analysis." These
are: 1) GDP per capita, 2) percentage of labor force
in industry, 3) percentage of labor force in urban
areas, 4) population growth, 5) income distribution,
and 6) adult illiteracy (127).
Political risks should be integrated into
country risk analysis. Whether they have been, as
Bird and Calverly have suggested, has not been
proven.
The Heuristic Element
William Cline (of Frank and Cline) modified
his model of country risk analysis in his 1984 study.
Besides a switch from discriminant analysis to logit,
he also added to his explanatory variables and used a
larger and more recent data set (670 observations
from the time period 1968-82). He retained his three
original statistically significant variables (debt


71
service/exports, reserves/imports and :
amortization/debt) and found them still to be
significant. He tested some new explanatory variables
and found three of them significant at the 5 percent
level (current account deficit/exports, growth of per
capita GNP and global external borrowing/imports).
The last variable was included to correct
what Cline perceived as the major deficiency in
earlier country risk models: the omission of a
variable to incorporate the effects of changing
global financial markets on the supply of loans.
("International Debt" 206)
. . the variable [global borrowing] provides
a way of overcoming a major limitation of
previous models: their inability to capture the
changing conditions in the 1970s, when a sharp
increase in international lending prevented
reschedulings that would have been expected on
the basis of earlier model estimates. This
variable is also important in explaining the
wave of reschedulings in 1982, when capital
availability reversed sharply ("International
Debt" 228 and 229).
While the inclusion of this variable
represents a step in the right direction, it could
well be that there remains a flaw in the setting of
the dependent variable that will interfere with this
models ability to forecast difficulties in the 1990s
. . regardless of the explanatory variables used.


72
Cline provides a solid rationale for the continued
use of the basic dependent variable of
rescheduling/non-rescheduling.
For the purpose of assessing financial system
stability, a debt rescheduling represents the
appropriate threshold of severity for analysis.
IMF stabilization loans are far too common to
pose a threat to capital market stability, and
indeed they typically have acted as the signal
to the capital market that creditworthiness was
improving. Arrears in themselves, especially if
minor or temporary, pose no special systemic
problem. Debt reschdulings, however, mark a
major qualitative break in the spectrum of
erosion. Although in the past reschedulings
have usually avoided any direct loss to
creditors . Debt reschedulings are
therefore of sufficient concern that they serve
as a meaningful criterion for classification of
cases as having serious debt-servicing
difficulties. Of course, more extreme
disruptions-such as extended moratorium or
outright repudiation-are also appropriate
criteria for such a classification. For
purposes of the analysis here, the following
discussion will use the term "rescheduling" as
a summary term that also encompasses more
severe forms of debt disruption ("International
Debt" 207,209).
Let's take a closer look, however, at the
results of a rescheduling from a lender's point of
view. Rescheduling almost always involves a change in
the interest rate, a loan negotiation fee for the
lender, and a longer period of repayment for the
debtor. It almost always involves the extension of
new money to the debtor. Traditionally lenders have
benefitted from the increased interest rate and


73
front-end fee: :
On the rescheduling of its public debt Mexico
had to pay a spread of 2.25 per cent over
LIBOR, while a year before the spread was only
0.4 per cent. For Brazil the figures are an
increase of spread from an average of about 1
per cent in 1979 to 2 -2.5 per cent. In Poland
a spread of 1.875 per cent was applied, whereas
before the rescheduling Poland commanded a
spread of only 1 per cent. The front-end fee
paid in connection with arrangement of the
rescheduling even increases their margins
slightly. For the bank, therefore, the return
on the rescheduled asset is higher than the one
received before the exercise (Krayenbuehl 125).
Very recent occurrences have altered
Krayenbuehl's assessment somewhat. The profit margin
for bankers on debt rescheduling is no longer
guaranteed to increase. For example, Argentina's 1983
rescheduling resulted in a 1 1/2 point drop from its
previous spread over LIBOR. Loan fees also dipped.
Not all drops in spreads have been so great.
What hasn't changed about debt rescheduling
is that as long as the debtor nation is willing to
continue playing the game, making its debt service
payments and rescheduling when necessary, the banks
involved are still making a profit. Until 1987, a
profit greater than before the rescheduling. Yes,
creditors have incurred a greater risk, and that risk
is the possibility that Cline is so willing to lump
in with the rescheduling/nonrescheduling dependent


74
variable outright repudiation. In the event of an
outright repudiation the banks will face serious
difficulties that would not apply if the country
chose to reschedule instead. It is not appropriate,
therefore, to lump these two outcomes together.
As the stakes grow ever higher in terms of
magnitude of debt load, moratorias and even
repudiation become increasingly attractive for the
less developed country. This is the real threat to
the lending community, not rescheduling. A major
shift in emphasis needs to be made away from a
dependent variable that answers the question "Has
this country rescheduled debt in the past" in order
to forecast "Will this country reschedule in the
future?" to one that addresses the real risk inherent
in lending. Eaton, Gersovitz and Stiglitz say that
the critical question is "When will a country with
certain characteristics owing a certain amount of
debt under certain contractual arrangements pay or
receive funds from creditors with certain
characteristics?" (507)
In simpler English, the appropriate heuristic
for country risk analysis becomes: "How close is this
nation to reaching the point where the benefits of


75
repudiation are of a higher value than the effects of
negative sanctions?" A country risk model designed
for lending in the 1990s will need to answer this
question.
Risk of repudiation. Krugman and Obstfeld
make this question sound easy to answer. "The
decision to default is easy to analyze. A country
defaults whenever the benefit of default, B, is
greater than the cost of default, denoted by C"
(607).
They propose that the benefit of default is a
function of national output, total foreign debt,
principal due in the current period, the loan rate,
and level of new loans (if any). This function should
be extended to include intangible political benefits
(mentioned above) that may also accrue to the
defaulting nation.
In general, the penalties for default that
may accrue to the negligent borrower are:
1) decreased likelihood of additional loans, 2) loss
of trade credit, 3) unfavorable "demonstration
effects" on domestic contracts and property right"
(Allsopp and Joshi xv), 4) loss of international


76
respect, 5) "legal actions against deposits, property
and exports arriving abroad" (Calverly 19), 6) loss
of other sources of aid, 7) higher costs in future
borrowing.
Loss of additional loans may be quite
unpleasant for the debtor to contemplate. For
example, "Defaults in Latin America in the 1930s
locked the countries out of the capital markets for
three decades ..." (Cline, "International Debt" 10)
The loss of trade credit is probably even more of a
deterrent to default than loss of additional loans.
Trade disruptions costing less than 3 percent
of GNP, or 9 percent of the total value of
imports and exports would be more costly than
making payments of 5 percent of total external
debt . Such payments, made consistently,
would make commercial bank loans look very
solid (Bulow, and Rogoff 175).
As Jesus Silva Herzog, Mexican Finance Minister,
said:
"We asked ourselves the question what happens
if we say 'No dice. We just won't pay.' There
are some partisans to that. But it didn't make
any sense, We're part of the world. We import
thirty percent of our food." (Bulow and Rogoff
175)
The rights of creditors have been recently
reinforced by the Foreign Sovereign Immunities Act of
1976. This law permits sovereign nations to waive
their immunity for repudiation of debts. Enforcement


77
can still be difficult. -
One way these creditor rights can be enforced
is as follows:
Suppose that Brazil repudiates its debts to
Citicorp. If Citicorp's detective can track
down any bank accounts Brazil holds in the
United States, or even any computers purchased
by Brazil that have not yet been shipped, it
can attach the assets, arguing that they are
Brazilian property and are subject to
foreclosure (Bulow and Rogoff 174).
Despite these changes in creditor sanctions,
"Debtor assests that would be accessible to creditors
in the event of outright repudiation are worth only a
small fraction of outstanding debt." (Bulow and
Rogoff 156) The ruler of a sovereign nation with high
levels of debt, expectations for low growth and high
costs of obtaining additional loans may well see them
as negligible.
Generally, the presumption that there are large
costs to sovereign repudiation (whether
outright, or partial) has been weakened as the
legal and institutional framework has been
investigated . .It can be questioned whether
sanctions, such as sequestration of assests,
would or could be applied by the courts . .
The main penalty would appear to be the
exclusion from future borrowing. But it is
notable that is not much of a penalty ... if
new borrowing is unlikely to be available
anyway. . (Allsopp and Joshi xv).


78
The specific costs of default are difficult to
quantify as they will depend largely on the debt
contracts of each nation. Clever lenders may have
negotiated side agreements which will raise the cost
of default considerably (Kohler 751). Seniority
clauses inserted into loan agreements will easily
dissuade other lenders from extending credit.
(Eaton 494) Information on side agreements and
seniority clauses will be readily available to the
borrowing nation's rulers who seek to estimate the
costs of default. Country risk analysts might not be
privy to these additional specific costs, but
published figures will reveal a nation's dependence
on food imports, national output, the loan rate,
foreign assets and total foreign debt so these
should be candidates for independent variables.


79
CHAPTER 5
A MODEL FOR THE 1990s
Given the growth of real debt levels,
interest arrangements and lower expectations for
near-term export growth; debt repudiation by debtor
nations will become increasingly attractive in the
1990s. Since default is quite different from
rescheduling, country risk analysis should be done in
stages that are differentiated by categories of risk.
Country risk analysis models that only predict the
probability of rescheduling will not be as useful in
the 1990's as models that include each of three types
of risks inherent in country risk: default,
rescheduling and political/external factors that
influence willingness to repay.
The first stage would assess the
cost/benefits of default from the borrower's point of
view. This step would incorporate independent
variables that relate to this decision. If the nation
exhibits ambivalence between the two choices, then
loan negotiations should be terminated at that point.


If the nation demonstrates strong inclination o
repay, then the second stage of analaysis should be
undertaken: investigation of risks that are
associated with the need to restructure.
It would be too expensive and time-consuming
to include the entire universe of plausible financial
indicators in this predictive model. A manageable
number of significant explanatory variables would
have to be selected. Historical analysis suggests
measurements of the dependence on imported oil,
import compressibility, export diversification, terms
of trade, loan levels, foreign direct investment,
foreign aid flows, the direction of world interest
rates, capital flight and fluctuations in currency
exchange rates.
A potential borrower who has made a positive
showing in the first two quantitative phases would
then be subject to the final phase. This phase
depends on the qualitative judgment of the analyst as
he seeks to integrate information about exogenous
conditions and/or unique, country-specific
circumstances that could undermine the borrower's
willingness and/or ability to repay the loan.


81
A few additional political risk factors as
suggested by Calverly (GNP per capita, percentage of
labor force in industry, percentage of labor force in
urban areas, population growth, income distribution
and adult literacy) would give some additional
indications as to the political stability of a
country (Calverly 127).
Limitations of Modeling
As demonstrated in the critique section of
chapter three, Frank and Cline's model continues to
perform out of sample about as well as it did for
their original sample period. This is fairly strong
evidence that there is something right about the
model despite its low level of predictability. The
fact that Frank and Cline's three significant
variables still tested out to be significant more
than a decade later does suggest that the use of
historical relationships between variables to predict
future ones can be a useful starting point for
constructuing new models.
Efforts to increase the explanatory power of
these tests have been made by several people. Cline


82
found that the size of a nation's debt affected his
1984 model's ability to predict rescheduling. "An
examination of the detailed predictions shows a
systematic difference between the larger and smaller
debtor countries" ("International Debt" 231).
Angeloni and Short (1980) used dummy variables for
developed, centrally-planned and oil-producing
nations. The first two turned out to be significant
(McDonald 631).
A cross-pooling test for the significance of
per capita GNP in debt rescheduling was done for the
econometric model presented in Table 3-1. (See
Table 5-1) The test is based on a comparison of
log-likelihood ratios. The log-likelihood values are
negative exponents of e. So large values, in an
absolute sense, indicate a low likelihood.
Disaggregating the data into "poor" and "not poor"
should improve the results and reduce the sum of the
likelihood values. The test checks whether the
improvment is large enough to be significant. This
test yielded negative results. Per capita income was
not significant when the model included all three
independent variables (debt service ratio,
imports/reserves and amortization/debt) or any


83
pairings of them. This confirmed Frank and Cline's
finding that per capita GNP was not significant.
Although this one exploration of intercountry
differences proved fruitless, additional research
into such differences might turn up more significant
variables. Factors relating to geography, history,
climate, religion, etc. should be candidates for
study. These factors might adequately be represented
by dummy variables within a universal model, or they
may necessitate the development of unique models to
fit specific subsets of nations.
Heuristic issues that need additional
attention are default potential and the
identification of leading indicators. The former has
been neglected because no official defaults have
occured in the past 30 years, however, that doesn't
mean none will occur in the next thirty. This lack
can be easily remedied by instituting the first step
in the three-tiered model proposed here. An initial
sort based on probability of default would leave the
restructuring dependent variable in the second step
to its appropriate task predicting
reschedulings.


84
The second issue is underscored by the debt
crisis of this decade. The exogenous factors that so
negatively influenced the debt situatioin of the
1980s were completely unanticipated by economists and
bankers in the early 1970s (Calverly 55). Country
risk analysts should be looking for leading
indicators of debt servicing difficulties. Such
indicators are not easy to identify; but still need
to be sought. To be effective they need to go beyond
the "tendency merely to extrapolate past experience"
and go an extra step towards to forecasting "likely
structural changes in the future" (Cline, "Longer-
Term Forecasts" 125).
An Alternative Solution
Given these difficulties in creating a
quantitative model that can forecast future debt
problems, Madura and Veit suggest a well-diversified
loan portfolio as an alternative approach to country
risk management. They developed the portfolio concept
because:
Regardless of the sophistication of the country
risk and creditworthiness assessment techniques
employed, there will still be cases where some
foreign loan applicants that initially appeared
to be creditworthy, are later unable to repay


85
Table 5-1 RESULTS OF POOLING TEST i
Log Likelihood Values
Test 1 Test 2 Test 3 Test 4
All Countries -163.50 -199.23 -199.56 -193.01
Poor countries * - 54.46 - 57.96 - 67.22 - 57.54
Non-Poor Countries - 108.12 -140.27 -129.74 -133.32
Cross- Pooling 1.85 1.99 5.17 4.31
Chi Square 9.48 7.80 7.80 7.80
* Poor Country is defined here as one with per capita
income less than $500.
Independent Variables Used in Each Test
Test 1 = Intercept, Debt service Ratio,
Imports/Reserves, Amortization/Debt
Test 2 = Intercept, Debt Service Ratio,
Imports/Reserves
Test 3 = Intercept, Debt Service Ratio,
Amortization/Debt
Test 4 = Intercept Amoritization/Debt,
Imports/Reserves
Cross-Pooling Significance Formula:
-2 [ full sample log likelihood (non-poor sample
log likelihood + poor sample log likelihood) ]
Number of total observations: 393


86
their loans. If, however, banks compose well-
diversified foreign loan portfolios, the
probability of simultaneous default by a
significant percentage of foreign borrowers
would be low. Thus, overall foreign loan
portfolio risk would be minimized (Madura and
Veit 1095).
An ideal foreign loan portfolio would balance
loan levels to countries that have uncorrelated or
negatively correlated country risks. This approach
presents its own set of challenges but certainly
deserves additional research.
Where does all this leave country risk
analysis for the 1990s? First, empirical research
should not exclude historical investigation, but it
should not be as heavily weighted in that direction
as it is now. Second, more attention needs to be
given to future structural trends. This will guide
the search for leading indicators of debt
difficulties. Third, willingness to pay (i.e.,
proximity to default status) and political risks
cannot be ignored simply because they are more
difficult to quantify than financial indicators.
Fourth, the assumption about homogeneity of nations
should be relaxed until it can be thoroughly tested
as a hypothesis. Fifth, country risk analysts should
remain open to new statistical techniques as they


87
become available.
There is little that can be said with
certainty about country risk analysis as it enters
the next decade. It will continue to hold importance
as long as nations continue to engage in
international trade, developing nations seek external
sources for investment capital and bankers retain
hope of making profits on their currently outstanding
loans. The explanatory and dependent variables will
change as analysts seek to increase forecasting
accuracy. But the continuing significance of the
Frank and Cline variables teach us that historical
patterns of association may well provide the positive
heuristic necessary to answer the fundamental
question of country risk analysis: "Will this country
repay this loan that we are currently negotiating?"


WORKS CITED
Abassi, B., and R.J. Taffler. "Country Risk: A Model
of Economic Performance Related to Debt
Servicing Capacity." Working Paper No. 36 City
University Business School, London. 1982.
Allsopp, Christopher J., and Vijay R. Joshi. "The
Internationa/L Debt Crisis." Oxford Review of
Economic Policy 2, no. 1 (Spring 1986) i -
xxxiii.
Alun, Tawang. "An Economic Analysis of Developing
Countries' External Indebtedness." Ph.D. diss.,
Vanderbilt University, 1984.
Amelung, Torsten, and Thorsten Mehltretter. "Early
Warning Systems in Light of the International
Debt Crisis." Konjunkturpolitik 32, no. 5
(1986): 257 281.
Bird, Graham. "New Approaches to Country Risk."
Lloyds Bank Review October 1986: 1 16.
Bulow, Jeremy, and Kenneth Rogoff. "A Constant
Recontracting Model of Sovereign Debt." Journal
of Political Economy 97, no. 1 (February 1989):
155 176.
Burton, F.N., and Hisashi Inoue. "Country Risk
Evaluation Methods: A Survey of Systems in
Use." The Banker January 1983: 41 43.
Caballeros, Romulo. "External Debt in Centeral
America." Cepal Review no. 32 (August 1987):
123 148.
Calverley, John. Country Risk Analysis. London:
Butterworths, 1985.
Cline, William R. International Debt: Systemic Risk
and Policy Response. Cambridge, Massachusetts:
MIT Press, 1984.
i


89
__________"Longer-Term Forecasts in International
Economics." American Economic Review 75 (May
1985): 120 126.
__________"External Debt: System Vulnerability and
Development." Columbia Journal of World
Business Spring 1982: 4-14.
Davis, R.R. "Alternative Techniques for Country
Risk Evaluation." Business Economics 16 (May
1981): 34 41.
Denison, Daniel. "A Pragmatic Model for Country
Risk Analysis." Journal of Commercial Bank
Lending March 1984: 29 37.
"Developing Countries Owe More than $1,000 Billion;
Solutions Must Be Urgently Sought," U.N.
Chronicle, March 1988: 44.
Dhonte, Pierre. "Describing External Debt
Situations." IMF Staff Papers 22 (1975): 159 -
186.
Eaton, J., Mark Gersovitz, and Joseph E. Stiglitz.
"The Pure Theory of Country Risk." European
Economic Review 30 (June 1986):481 513.
Feder, Gershon. "Economic Growth, Foreign Loans and
Debt Servicing Capacity of Developing
Countries." The Journal of Development Studies
April 1980: 352 367.
Feder, Gershon, and Richard Just. "Study of Debt
Servicing Capacity Applying Logit Analysis."
Journal of Development Economics March 1977 :
25 38.
Feder, Gershon, Richard Just, and Knud Ross.
"Projecting Debt Servicing Capacity." Journal
of Financial and Quanitative Analysis 16, no. 5
(December 1981): 651 666.
Feder, Gershon, and L.V. Uy. "The Determinants of
International Creditworthiness and their Policy
Implications." Journal of Policy Modelling 7,
no. 1 (Spring 1985): 133 156.


90
Frank, Charles, and Cline, William R. "Measurement
of Debt Servicing Capacity: An Application of
Discriminant Analysis." Journal of
International Economics 1 (August 1971): 327 -
344.
Gafar, J.S. "The Impact of External Factors on the
Debt Burden and Import Capacity of Developing
Countries." The Indian Economic Journal 33, no.
3 (January/March 1986): 35 45.
Hajivassiliou, V.A. "The External Debt Payments
Problems of LDCs: An Econometrics Model Based
on Panel Data." Journal of Econometrics 36
(September/October 1987): 205 230.
Heller, Robert. "The Debt Crisis and the Future of
International Bank Lending." American Economic
Review 77, no. 2 (May 1987): 171 175.
Heller, Robert H., and Emanuel A. Frenkel.
"Determinants of LDC Indebtedness." Columbia
Journal of World Business (Spring 1982): 28 -
33.
Hope, Nicholas, and Thomas Klein. "Issues in
external debt management." Finance &
Development September 1983 : 23-25.
Kahn, Mohsin, and Malcolm Knight. "Sources of
payments problems in LDCs. Finance &
Development December 1983 : 2 5.
Keller, Peter M., and Nissanke E. Weerasinghe. The
IMF Yearbook Package. Washington, D.C.:
International Monetary Fund, 1988.
Kharas, Homi. "The Long-Run Creditworthiness of
Developing Countries: Theory and Practice."
Quarterly Journal of Economics 99, no. 3
(August 1984): 415 439.
Kharas, H.J. and J. Levinsohn. "LDC Savings Rates
and Debt Crises." World Development 16, no. 7
(July 1988): 779 786.


91
Kim,
Taeho. "Assessment of External Debt Servicing
Capacity: An Alternative Methodology." Journal
of Economic Development December 1985 : 35 -
51.
Kincaid, Russell, G. "Inflation and the External
Debt of Developing Countries." Finance &
Development December 1981 : 45 48.
Kohler, Daniel F. "To Pay or Not to Pay: A Model of
International Defaults." Journal of Policy
Analysis and Management 5, no. 4 (1986): 742 -
754.
Krayenbuehl, Thomas E. Country Risk, Assessment and
Monitoring. Cambridge: Woodhead-Faulkner, 1985.
Krueger, A.O. "Origins of the Developing Countries'
Debt Crisis 1970 to 1982." Journal of
Development Economics 27, no. 1 (October 1987):
165 187.
Krugman, Paul, and Maruice Obstfeld. International
Economics. Glenview, Illinois: Scott, Foresman
and Company, 1988.
Kugler, P. "Economic Indicators and Debt
Rescheduling 1977 1981: Empirical Results
from a Logit Model." Rivista Internazionale di
Scienze Economiche e Commerciale 31
(October/November 1984): 992 1005.
Makin, John H. The Global Debt Crisis: America's
Growing Involvement. New York: Basic Books,
1984.
Madura, J., and E.T. Veit. "Intertemporal Exposure
to Country Risk." Rivista Internazionale di
Scienze Economiche e Commerciale 34
(November/December 1987): 1095 1102.
Manfredi, Eileen M. "Predicting Debt Reschedulings
in Developing Countries." Agricultural
Economics Research 33, no. 2 (April 1981): 26 -
30.


92
Mayo, Alice L., and Anthony G. Barrett. "An £arly-
Warning Model for Assessing Developing-Country
Risk." Financing and Risk in Developing
Countries Ed. Stephen H. Goodman. New York:
Praeger Publishers, 1978.
McDonald, Donogh. "Debt Capacity and Developing
Country Borrowing." International Monetary Fund
Staff Papers December 1982 : 623 646.
Melvin, Michael, and Don Schlagenhauf. "A Country
Risk Index: Econometric Formulation and an
Application to Mexico." Economic Inquiry 23
(October 1985): 601 619.
Melvin, Michael, and Don Schlagenhauf. "Risk in
International Lending: A Dynamic Factor
Analysis Applied to France and Mexico." Journal
of International Money and Finance 5 (March
1986): 31 48.
Merrill, James. "Country Risk Analysis." Columbia
Journal of World Business Spring 1982 : 88 -
91.
Neuhaus, Paulo. "Floating Interest Rates and
Developing Country Debt." Finance & Development
(December 1982): 37 38.
Petersen, Hans J. "Debt Crises of Developing
Countries: A Pragmatic Approach to an Early
Warning System." Kon-junkturpolitik 23, no 2.
(1977): 94 100.
Pindyck, Robert and Daniel Rubinfeld. Econometric
Models and Economic Forecasts. New York: McGraw
Hill, 1981.
Ryser, Jeffery. "Deals that are Making a Dent in
the Third World Debt." Business Week, October
1988: 111-114
Saini, Krishan and Philip Bates. "A Survey of
Quantitative Approaches to Country Risk
Analysis." Journal of Money and Finance (June
1984): 341 356.


93
__________"Statistical Techniques for Determining
Debt Servicing Capacity for Developing
Countries: Analytical Review of the Literature
and Further Empirical Results." Federal Reserve
of New York Research Paper No. 7818, September
1978.
Sargen, Nicholas. "Economic Indicators and Country
Risk Appraisal." Federal Reserve Bank of San
Francisco Economic Review (Fall 1977): 17 35.
Schmidt, Reinhart. "Early Warning of Debt
Rescheduling." Journal of Banking and Finance
(June 1984): 357 370.
Solberg, Ronald L. Sovereign Rescheduling: Risk and
Portfolio Management. London: Sydney and
Wellington, 1988.
Studenmund, A.H. and Henry J. Cassidy. Using
Econometrics: A Practical Guide. Boston:
Little, Brown, 1987.
Suratgar, David, ed. Default and Rescheduling:
Corporate and Sovereign Borrowers. Washington,
D.C.: Euromoney Publications, 1984.
Zimmerman, Andrew Birge. "The Relationship between
Economic Indicators and Country Risk." Ph.D.
diss., University of California, 1985.