Economic Review

FEDERAL RESERVE BANK OF SAN FRANCISCO ECONOMIC REVIEW

FALL 1877 Nicholas Sargen'

The current debate over commercial-bank dissatisfaction relates to their own procedures, lending to less-developed countries (LDCs) has and how much has to do with limitations in the primarily centered on the question of whether current state of the art in assessing country risks. private banks have extended too much credit to This paper is designed to facilitate appraisal the group of non-oil exporting developing coun of existing procedures by comparing techniques tries. Despite the considerable attention given to commonly used by commercial banks and offi the subject in the financial press, the major inter cial institutions, along with techniques that have national banks by and large dismiss the possibil been developed in the economic literature. The ity of widespread defaults or reschedulings on scope of the paper is limited to only one aspect of developing-country loans as being highly remote. country-risk appraisal-namely, the use of eco A more likely scenario, according to the banks, is nomic indicators to rank countries according to that individual countries occasionally may exper the probability ofdefault. The analysis addresses ience repayment difficulties requiring some refi the following questions: (i) What are the eco nancing or rescheduling. Most banks, therefore, nomic causes of debt reschedulings? (2) Which believe that the crucial problem is to be able to set of economic indicators does the best job of detect in advance which countries are likely to distinguishing between rescheduling countries experience repayment problems and when these and non-rescheduling countries? (3) How reli difficulties may arise. able are econometric techniques in predicting The banking community has recently shown debt reschedulings? great interest in the utilization of analytical tech Section I briefly reviews the experience with niques to detect potential default or rescheduling LDC debt reschedulings since the late 1950's, situations. Relatively little information is cur and describes techniques employed by commer rently available to appraise the various tech cial banks and official institutions for assessing niques now in use, and, as a result, it is often country risk. Section II compares two conceptual difficult for bankers to judge the adequacy of approaches used in the analysis of debt resched their own internal rating systems as compared ulings. The first approach views reschedulings as with those employed by other institutions, public resulting from fluctuations in prices of primary or private. products which then lead to a rapid accumula The difficulty is illustrated by a recent Ex tion of external debt relative to export earnings. port-Import Bank survey on bank practices in as The second approach treats debt reschedulings sessing country risk. 1 That study found that a as a monetary phenomenon, in which domestic large percentage of the 37 U.S. banks surveyed inflation and an overvalued exchange rate con are dissatisfied with their present country-ap tribute to increased demand for imports and to praisal methods and are actively seeking new export stagnation, and consequently to a rapid procedures. From the survey responses, though, build-up of external debts. Section III employs a it is not possible to determine how much of their statistical procedure--discriminant analysis-to identify the set ofeconomic indicators which best *Economist, Federal Reserve Bank ofSan Francisco. distinguish rescheduling countries from non-re scheduling countries. (A brief discussion of the

19 statistical procedure is included for interested ulings), and the second type of rescheduling is readers.) The final section assesses the relevance identified with long-term debt relief on official ofthe empirical findings to country-risk apprais credits (e.g. reschedulings for South Asian coun al and the desirability of using statistical proce tries and Ghana). In the "liquidity" cases, mone dures for this purpose. tary (and fiscal) factors appear to be at the root ofthe problem, and the inflation rate turns out to Our analysis suggests the importance of dis be the most imporant explanatory variable. tinguishing "liquidity" reschedulings from long Cases of chronic-debt relief, on the other hand, term debt reschedulings. The first type is associ appear less amenable to a monetary framework ated with a bunching of short-term commercial of analysis, and it is necessary to include the credits (typical of most Latin American resched- debt-service ratio to explain these reschedulings. I. Assessing CouniryRisk

Commercial banks encounter two types of re countries to postpone payments of their interest payment risk in international-lending operations or principal. The total amount of debt service which do not arise in domestic-banking oper rescheduled was on the order of $7.7 billion, of ations. The first type of risk, commonly referred Table 1 to as "sovereign risk," occurs when a national International Debt Reschedulings 1960-761 government refuses to permit foreign loans to be (Millions of U.S. $) repaid, or when a government seizes bank assets Total Amount Amount of U.S. without adequate compensation. The second type Country Year Rescheduled Debt Rescheduled of risk, often called "transfer risk," is associated Argentina* 1962 240 0 with foreign borrowers' problems in converting 1965 76 18 domestic currency into foreign exchange. Credits Brazil* 1961 300 0 extended to foreign borrowers by banks in the 1964 200 44.5 U.S. market or in the Euro-currency market are Chile* 1965 96 43 typically denominated in U.S. dollars (or in a key 1972 160 65 currency), and government foreign-exchange re 1974-75 597 231 strictions sometimes make it difficult for borrow Egypt 1966 N.A. N.A. ers to acquire sufficient foreign exchange to 1971 145 145 repay their loans. 2 Foreign-exchange controls are Ghana* 1966-70 295 0.7 particularly common in developing countries, 1~4 ~O 0 India* 1968-72 545 65 where fixed exchange-rate policies are still 1973-76 688 74 prevalent. Indonesia* 1965-68 427 96 Commercial banks assess both types of risk in 1970 2100 215 their country-risk appraisals. Cases of expropri Pakistan* 1971-74 987 270 ation or outright default on bank loans have been Peru* 1968-69 128 0 quite rare in the postwar period, however, and Philippines 1970 NA N.A. have been confined mostly to Communist take Turkey* 1965 220 15 overs in Cuba or Southeast Asia. The more com 1972 114 0 Uruguay 1965 N.A. N.A. mon case has been the formal restructuring or re Yugoslavia 1965 N.A. N.A. financing of external-debt obligations in the 1972 59 59 wake of foreign-exchange crises. Restructuring Zaire 1976 N.A. N.A. has usually involved a stretching of principal * Denotes countries which have experienced multilateral payments on a previous credit, while refinancing debt reschedulings. has involved new credits. 'Note: Information on debtreschedulings was compiled Close to 40 such instances have occurred since from a variety of sources including Bitterman 1956, involving about a dozen developing coun [6], Cohen [7], Feder-Just [l1J, Frank-Cline tries which formally negotiated with creditor [I2l, IMF [I7J [18], OECD [22J.

20 which roughly $1.3 billion constituted debt owed of country targets or limits, for the use of bank to the V.S. government or to U.S. nationals (Ta management in overseeing the bank's interna blet). Howiver, the economic cost of debt re tional portfolio. The latter process involves mak schedulings-measured as the difference be ing country comparisons about the risk of non tween the present discounted value of the repay repayment, and so subjective judgments play an ments stream before and after rescheduling important role. was considerably smaller.3 Most banks are reluctant to assign formal Most multilateral debt reschedulings have ei credit ratings to individual countries when set ther involved suppliers' credits (which frequently ting country guidelines. In the Eximbank survey, carry government guarantees) or official credits. for example, only about a fourth of the banks Many of the Latin American reschedulings, for surveyed (8 out of 37) translated their country example, have involved short-and medium-term evaluations into a country rating (usually with a commercial debt, so that negotiations were ar five-grade letter system A to E). Five of the ranged through ad hoc meetings of major private banks which rated countries utilized a weighted creditors (the so-called "Paris Club" or "Hague checklist system, with economic and political in Club" meetings). Debt-relief negotiations for dicators being used to measure a country's repay Ghana, India, Indonesia, Pakistan, and Turkey, ment prospects. The summary score, or country on the other hand, have been arranged through rating, in each case was obtained by assigning government consortia which were responsible for weights to individual indicators and then sum coordinating flows of financial assistance to ming the value ofindividual indicators. those countries. Private debts usually have not The checklist approach can be criticized for been rescheduled in these contexts, in part be failing to provide a conceptual framework for se cause the amounts involved were relatively small lecting individual indicators, and also for its arbi compared with official claims. trary selection of weights. However, statistical LDC workouts of debt to private bank credi procedures are currently being developed to cir tors have been much more infrequent and have cumvent some of these problems, by such agen tended to take the form of refinancing, rather cies as the V.S. Treasury Department and the than rescheduling ofexisting debt. The principal U.S. Export-Import Bank. Their statistical debt cases in earlier years involved Argentina and monitoring systems use a single predictive equa Brazil (early 1960's), Peru (1965), and the Phil tion, based on information about past debt re ippines (1970).4 Because of rapid expansion in schedulings, to screen "high risk" countries from international lending, however, banks since 1975 those with low probabilities of rescheduling. have become even more heavily engaged in nego (The methodology underlying the Treasury and tiations with developing countries, as in the re Eximbank systems is described in Section III.) cent negotiations with the governments of Chile Countries singled out as possible rescheduling and Zaire on debt-relief issues. In additions, they candidates are then subjected to in-depth eco have provided balance-of-payments financing for nomic and political analyses. Argentina and Peru to ease potential debt prob The econometric approach provides a means lems of these countries. In this situation, banks for identifying statistically significant variables and regulators alike have become concerned and for assigning weights which are not com about the need to improve methods for assessing pletely subjective. From a commercial-bank individual country risks. standpoint, though, the central issue is whether econometric techniques provide a more reliable Methods for assessing risk means of detecting defaults or debt reschedul Country appraisal can come into play at two ings than present procedures. A direct compari different stages. One phase involves the approval son of the two approaches is not possible, since of individual credits, and thus requires a report banks which make country ratings do not public by the bank's economics department on the bor ly test their rankings against experience. Pub rowing country's general political and economic lished studies which employ econometric situation. The second phase involves the setting techniques, on the other hand, report low error

21 rates in explaining past reschedulings, although framework used to explain debt reschedulings they have beenfar less successful in anticipating (Section H) and to methodological difficulties reschedulings than in explaining most encountered in applying statistical procedures to reschedulings.5 a small sample of rescheduling countries (Sec The problems can be traced to the conceptual tion III). II. Conceptual Approaches to Debt RescheduUngs

Part of the difficulty faced by commercial debt-filonitoring systems is based on the finan banks and regulatory agencies in assessing risks cial-ratioanalysispioneered by Avramovic and can be traced to the absence of a well-developed associates at the World Bank [3]. The approach conceptual framework for analyzing debt prob views reschedulings as a problem ofexternal debt lems of developing countries. Economic models management, and thus focuses attention on the of "optimal" foreign borrowing largely have been determinants of a country's "debt-service capac concerned with the effect of foreign borrowing ity". We are concerned here with that approach's on economic growth and with conditions neces underlying assumptions and their implications sary to ensure an efficient allocation of resources for the analysis of LDC debt problems. over time.6 These studies generally conclude that In the Avramovic study, one type of debt repayment ofexternal debt is not a problem, pro problem involves the near-term bunching of vided that the rate of return on domestic invest debt-service payments, while a second involves ment equals or exceeds the cost of foreign debt rescheduling over a longer time interval.8 borrowing.7 Such models, however, do not allow The Latin American reschedulings typified the for the fact that foreign borrowing must be re first type of problem: debt-service payments on paid in foreign exchange, and that foreign-ex short- and medium-term commercial debt were change receipts may be temporarily scarce. rescheduled over a fairly short time span-e.g., Second, they typically assume that domestic and one to five years. But in the case of the consortia international capital markets are perfectly com creditors to Ghana, India, Indonesia, and Paki petitive-assumptions which are highly unrealis stan, long-term official lending formed a signifi tic for most developing countries. cant portion of debt-service payments. In these The two approaches presented in this section cases, the reschedulings covered such a long explicitly deal with the foreign-exchange prob time-span-up to 30 years in the case of Indone lems which surround most debt reschedulings. sia- that they had a noticeable impact on debt The debt-service approach traces the LDC's for service burdens. eign-exchange problems to their heavy reliance Avramovic analyzes the short-run debt prob on exports of primary products and to the high lem as if the developing country were a firm fac volatility of these products on world markets. Fi ing a cash-flow or liquidity squeeze. The liquidity nancial ratios derived from individual balance problem in this case reflects a temporary short of-payments components hence are used to mea fall in foreign-exchange receipts, brought about sure a country's ability to service its external by an exogenous decline in the world price of the debt in the event of a shortfall of export receipts. LDC'sprincipal export product. Under these cir The monetary approach, on the other hand, is cumstances, the country can try to cover pay primarily concerned with the overall determina ments abroad by expanding its export volume, by tion of a country's balance of payments, and thus curtailing imports, by further borrowing or by focuses attention on that country's monetary-fis drawing down foreign-exchange reserves. Avra cal policy and exchange-rate policy. From this movic's analysis, however, assumes that most perspective, the underlying causes of debt re LDC's cannot expand export proceeds easily in schedulings are internally, rather than external the short run, and that they cannot easily "roll ly, generated. over" debt by borrowing from private capital Debt-service approach markets. Under these assumptions, a developing The analytic approach used in most statistical country has only two viable options available in

22 the short run-namely, to draw down reserves tence of assumed balance-of-payments rigidities (including drawings from the International and on the nature of foreign-exchange bottle Monetary Fund) or to reduce its import volume. necks. In Avramovic's analysis, the foreign-ex The Avramovic approach attempts to mea change constraint reflects two factors: (I) sure a country's ability to withstand an export limited possibilities for short-run expansion for shortfall (or a situation of capital flight) by con export production, and (2) inelastic demand for a structing financial ratios from individual bal country's major export product. It is assumed ance-of-payments components. The principal that if a country attempts to expand its export measure of "reserve adequacy," for example, is volume by increasing export production or by re the ratio of foreign-exchange reserves to annual ducing domestic consumption, the increased ex imports ofgoods and services. The higher the ra port volume will lead to a deterioration in the tio, the better equipped the country is to cover country's terms of trade; so that export receipts imports by temporarily drawing down foreign are not increased." But if the "small-country as exchange reserves. sumption" is applicable-if the country's share The traditional indicator of debt-service ca of the world market is so small as to leave the pacity, on the other hand, is the debt-service ra world price unaffected-the foreign-exchange tio--the proportion of foreign-exchange earnings bottleneck disappears. That is, the country can on current account (exports of goods and serv increase its export volume (and its export re ices) absorbed by interest payments and amorti ceipts) through increased domestic savings or zation on external debt. Those analysts using this through expanded production. indicator do so because debt-service payments The assumption of limited (or zero) capital represent contractually fixed obligations which mobility is also critical to the analysis. Ifa coun cannot be easily adjusted; hence, a higher ratio try is able to borrow from world capital markets implies a larger relative burden on import reduc (including commercial banks) to cover a tempo tion for a given shortfall in export receipts. The rary shortage of foreign exchange, the concepts reasoning behind this traditional indicator is that of "reserve adequacy" or "debt service capacity" there is a limit on a country's ability to tolerate a become much more difficult to define. Under reduction in its import volume. 9 these circumstances, it is not the country's lack One of the principal conclusions of the Avra of foreign-exchange reserves or the country's ex movic study is that the debt-service ratio is a rel port earnings per se which are important, but evant indicator of potential "cash squeeze" rather the country's ability to acquire foreign ex problems associated with foreign-exchange cri change. 12 In this case, the country must decide ses, but that it is less useful for analyzing debt whether the cost of foreign borrowing exceeds problems of a long-run nature. The reason is that the cost of adjusting to an export shortfall domestic savings rates normally rise during the through import reductions-i.e., profitability process of economic development, in which case considerations are relevant even in the short run. foreign-borrowing requirements needed to sus The main limitation of the approach, how tain a given target growth rate will diminish ever, is that it focuses on the events immediately through time. A country's debt-service ratio thus surrounding a rescheduling, rather than on the will tend to rise in the early stages of develop underlying causes. It provides few clues to ex ment, when domestic saving rates are low, but plain why countries borrow heavily, and it allows will tend to level off or decline with the later rise little scope for d6mestic policies to influence for in domestic savings. The ability to repay external eign borrowings or repayment prospects. Avra debt over the long run, therefore, hinges on the movie's analysis, for example, completely ignores difference between the marginal savings rate and the role which the domestic price level, the ex the initial savings rate, as well as on the relation change rate, and interest rates play in the process ship between the rate of return on investment of balance-of-payments adjustment. The key and the cost offoreign borrowing. 10 variables-the debt-service ratio, the reserves The usefulness of the debt-service approach as import ratio, the export growth rate, or the do an analytical tool hinges critically on the exis- mestic savings rate-are either exogenous or

23 structurally determined. As a result, the scope rowingforeign currency from abroad (r,): for balance-of-payments adjustment appears r' =if- Pd +e, (2) quite limited. where: e= expected appreciation of foreign currency. Monetary approach Real borrowing costs in the two markets (and The alternative approach uses a monetary real rates of return on capital in the two coun framework of analysis to study the problem of tries) will be equated- only if the expected ex debt-reschedulings. The monetary approach change-rate change is equal to the expected (like the debt-service approach) treats reschedul inflation-rate differential at home and abroad: ings as consequences of foreign-exchange short (3) ages. However, it is primarily concerned with the If investors believe authorities can maintain a overall determination of the balance of pay fixed exchange rate temporarily (despite a high ments, rather than with individual balance-of er domestic rate ofinflation), incentives will exist payments components. The scarcity of foreign to borrow more heavily from abroad, since real exchange in this case results from: (I) rapid borrowing costs are then perceived to be lower in money-supply expansion (associated with the fi the foreign market than in the domestic nancing of fiscal deficits) and consequent in market.13 crease in domestic inflationary pressures, and (2) Amore common situation, however, is one in maintenance of an overvalued fixed exchange which authorities impose interest-rate ceilings to rate. From this perspective, the underlying keep domestic borrowing costs low. Such a policy causes ofdebt reschedulings are rooted in domes tends to lower domestic saving and to ration po tic economic policies. tential borrowers out of the domestic market. An analysis of this monetary framework in The imposition of interest ceilings, therefore, volves: (1) the effects of domestic inflation and may also create incentives resulting in increased an overvalued exchange rate on the supply and demand for foreign funds. demand for foreign funds, and (2) the implica The amount of foreign borrowing, however, tions of exchange-rate flexibility for debt re also depends on lenders' expectations about re schedulings. Consider first the case of a develop payment prospects. In a highly competitive mar ing country which maintains a fixed exchange ket, such as the Eurocurrency market, loans to rate and which suffers from a higher inflation developing countries include an interest premi rate than the rest ofthe world. um-the spread over the London inter-bank of Inflation can influence the demand for foreign fer rate-which reflects the higher risk of funds in such a case through its adverse impact repayment. An increased demand for foreign on the trade accounts. That is, inflation would funds associated with an over-valued exchange tend to cause export demand to fall and import rate, therefore, need not result in an increased demand to rise, and the growing trade deficit, in volume of foreign borrowing-provided that turn, would increase trade-financing require there is a contraction (leftward shift) in the sup ments. A second type of inflation impact, noted ply schedule of foreign funds to offset that in by Friedrich Lutz [19], concerns the effect of an creased demand. over-valued exchange rate on the cost of borrow While economic theory provides no clear-cut ing funds from abroad. Lutz's analysis assumes reasons for expecting domestic inflation to lead that nominal interest rates in the domestic econo to an increased volume of foreign borrowing, the my (id) and abroad (if) reflect the real rate of effect may not be completely neutral, judging return on capital (r) and the expected inflation from the experience of those LDC's which have rate (p): rescheduled suppliers' credits. For instance, most id = rd + Pd, and (1) ofthe Latin American countries ofthis type were if= rf+ Pf . able to obtain ample suppliers' credits (usually In financing domestic investment, borrowers government-guaranteed) in the early stages of compare the real cost of borrowing in the domes inflation, but fewer such credits as the inflation tic capital market (rd) with the real cost of bor- progressed. In these cases, domestic inflation re-

24 suited in rapid growth of debt-service payments den of transferring real resources abroad to ser in the early stages ofinflation, but then in subse vice external debt. Rather, exchange-rate quentexport stagnation, which contributed to flexibility is relevant to debt reschedulings be rising debt-service ratios. cause exchange-rate movements are part of the overall adjustment process, whether the re Thus far, we have assumed that authorities in source-transfer problem is "real" or monetary. developing countries maintain fixed exchange Currency depreciation resulting from a price de rates. ActuaUy, most LDC's today continue to cline for some major export product, for exam peg their exchange rates to some key currency, ple, win create incentives towards increased although a growing number of them have experi export production. Similarly, depreciation re mented with some form of exchange-rate flexi sulting from domestic inflation will offset the ad bility in recent years. Under a freely-floating verse effects ofinflation on the trade accounts. In exchange rate, a country cannot experience a this sense, exchange-rate flexibility can help re shortage offoreign exchange, since there is no of duce the necessity for debt rescheduling. There ficial intervention in the foreign-exchange mar fore, one would probably expect fewer debt ket. The absence of a "foreign-exchange reschedulings under flexible exchange rates, al problem," however, does not imply a smaller bur- though not necessarily so in every case.

III. Empirical Evidence on Debt Reschedulings This section presents empirical evidence on proach is difficult to use in any "early-warning" the determinants of debt rescheduling, with em- system, at least pardy because World Bank data phasis on the characteristics distinguishing those on external debt are available only after a two- or countries which have rescheduled their debt from three-year lag for most countries. The U.S. Trea- those which have not-previous empirical stud- sury Department actuaUy discontinued use of its ies have largely concentrated on variables sug- debt-monitoring system because of the problem gested in the Avramovic study. The statistical ofobtaining up-to-date, accurate information on results confirm that reschedulings are associated LDC external debt. with a high debt-service ratio and a bunching of With respect to the monetary approach, how- external-debt obligations, but there is disagree- ever, inflation rates and exchange rates are gen- ment about the importance of other economic eraUy available with relatively short time lags. variables. Hence, an indicator system relying on the mone- Frank and Cline [12] used discriminant anal- tary approach is more likely than one based on ysis to investigate the importance ofeight indica- debt information to detect likely candidates for tors for the period 1960-68. They found only debt rescheduling. To date, however, there has three variables to be important: the debt-service been litde empirical work on the relationship be- ratio, the debt-amortization ratio, and the ratio tween monetary variables and debt reschedul- ofimports to reserves. Feder and Just [11], using ings, so this study attempts to establish whether a similar set of explanatory variables, applied 10- such a relationship exists. git analysis to explain reschedulings during the inflation and debt rescheduling 1965-72 period. Their results showed the impor tance of the three variables identified by Frank A clear relationship between inflation and and Cline, but three other indicators as well debt rescheduling is apparent for the 1960-76 pe the export growth rate, the level of per capita in riod (Table 2).15 Altogether, 70 percent of the come, and the ratio ofcapital inflows to debt-ser countries with long-term inflation rates above 10 vice payments. percent (measured by wholesale prices) resche Both studies report low error rates in identify duled their debts at some time during that peri ing past reschedulings. 14 Nonetheless, questions od. Moreover, aU six countries in the "high arise about the availability ofdata for testing the inflation" group had to reschedule at least once two basic (debt-service and monetary) ap between 1960 and 1976. proaches. For example, the debt-service ap- Other data suggest the important contribu-

25 tionof currencyovervatuation...... -as well. as. infla sion in the period preceding rescheduling. The tion...... -tp •. balance-of-payments difficulties prior depreciation, however, was insufficient in each todebtrescn.edulings(Table 3). In every case case to offset the adverse effects of sustain.ed cited, except EgyptandTurkey, a majorcurren~ high inflation on the trade account. cydevaluation was undertaken aroundtheperiod Application of discriminant analysis when

Table 2 Inflation and Debt Reschedulings: 1960-19761

Very High Inflation High-Inflation Middle-Inflation low-lnflatlon2 Group Group Group Group (above 20% p.a.) (10-20% p.a.) (5-10% p.a.) (less than 5% p.a.)

I. Argentina* (33%) 1. Bolivia (10.2%) I. Afghanistan (8.4%) I. Algeria (2.8%) 2. Brazil* (35%) 2. Colombia (14.6%) 2. Burma (7.9%) 2. Egypt* (3.9%) 3. Chile* (161 %) 3. Ghana* (11.3%) 3. Costa Rica (7.3%) 3. EI Salvador (3.2%) 4. Indonesia* (186%) 4. Israel (11.0%) 4. Dominican 4. Ethiopia (4.0%) 5. Uruguay* (53%) 5. Peru* (10.4%) Repub. (5.1%) 5. Guatemala (3.9%) 6. Zaire* (25%) 6. Philippines* (10.5%) 5. Ecuador (7.3%) 6. Guyana (4.0%) 7. South Korea (14.6%) 6. Greece (6.2%) 7. Honduras (3.6%) 8. Yugoslavia* (14.6%) 7. India* (7.7%) 8. Iran (3.6%) 8. Ivory Coast (5.1%) 9. Iraq (3.0%) 9. Jamaica (7.2%) 10. Jordan (4.4%) 10. Mexico (5.1%) II. Malaysia (3.0%) II. Pakistan* (7.0%) 12. Sri Lanka (4.4%) 12. Par~guay (9.1%) 13. Syria (4.5%) 13. Spain (5.8%) 14. Venezuela (4.4%) 14. Thailand (5.5%) 15. Tunisia (5.5%) 16. Turkey* (9.5%)

1 Figures in parentheses represent annual compound (WPI) inflation rates over the period 1960-1975. Asterisks denote debt reschedulings in the period 1960-76. (See Table I). Data from International Financial Statistics. 2 U.S. annual compound WPI inflation rate is 4.2% for 1960-75.

26 (e.g., countries) into two or more groups (e.g., this valueinthe rescheduling group, and to cate "rescheduling country" vs. "non-rescheduling gorizecountries with lower inflation rates in the country"). The rule is selected so as to minimize non-rescheduling grou.p. Applying the "10 per the expected cost ofmaking two types oferrors in cent cut-off rule"to the countries listed in Table classifying observations. In our analysis, Type I 2 yields the following setofresults: error occurs when a rescheduling country is clas Inftanoll Inftanoll sifieda.sanon-rescheduling country, and Type II Rate>10% Rate:510% error results when a non-rescheduling country is classified as a rescheduling country. Rescheduling group 71% 29% Suppose, for example, that the only difference (14 countries) (classified (Type 1errorrate) between rescheduling countries and non-re correctly) scheduling countries is that the inflation rate is Non- 13% 87% higher on average in the first group than in the Rescheduling group (Type 11 (Classified second group. Under these circumstances a sim (30 countries) error rate) correctly) ple way to classify countries would be to select some cut-off inflation rate, say 10 percent, and to categorize countries with inflation rates above

Table 3 Debt Reschedulings and Exchange Rate Devaluations1

3 year 3 year Debt CPllnflation Money-Supply (M,) Exchange Rate Reschedulings Rate(OAl) Growth Rate (0Al) Devaluation Very High Inflation Group

Argentina 1962 23.3 9.9 1962 1965 24.7 31.1 1964,1965 Brazil 1961 31.6 43.9 1961 1964 69.7 70.5 1962-65 Chile 1965 37.3 48.8 1962-65 1974 43.2 110.3 1972-76 Indonesia 1965 173.8 386.2 1966-68 1970 185.0 73.3 1970 Uruguay 1965 40.3 58.8 1965 Zaire 1976 24.6 25.5 1976 High Inflation Group

Ghana 1966 12.6 14.0 1967 1974 16.0 29.9 none Peru 1968 12.6 14.4 1967 Philippines 1970 6.2 10.9 1970 Yugoslavia 1965 16.3 16.1 1965 1972 11.4 15.8 1971

Middle or Low inflation Groups

India 1968 9.4 7.8 1966 1973 8.8 14.5 1972 Paldstan 1971 5.0 14.3 1971 Egypt 1966 9.2 10.1 none Turkey 1965 3.3 14.5 none 1972 11.4 21.6 none , Data from International Financial Statistics.

27 Thus, four of fourteen countries (29 percent) (which mayentail.a subjective judgment) and on which rescheduled their debt had long-term in the frequency of reschedulings relative to non flation rates less than 10 percent, so that use of a reschedulings. 17 10.percent cut-off value caused those four to be classified incorrectly (Type I error). Four of the Methodological Issues thirty countries (13 percent) which did not re l'hemain •• problem encountered in applying schedule, on the other hand, had a long-term in discriminant analysis (or other statistical proce flation rate above 10 percent, resulting in Type II dures)to debt-rescheduling data arises from the error. small number of obervations of this type. Pooled The same principle applies to a situation in time series.and cross-section data are typically which there are a number ofvariables which dif usedtoincrease the number of rescheduling ob ferentiate the two groups. In this case, a discrimi servations, and. this procedure is adopted here. nant "score" (or composite variable) is computed Each observation thus corresponds to a country as.a weighted average of the individual variables and a year. The Argentine multilateral resched for purposes of classifying individual observa ulingof 1962, for example, is treated as a sepa tions. The weights of the composite variable are rate observation from the Argentine reschedul selected so as to maximize the difference in mean ing of 1965. values for the two groups, given the specified set The procedure of pooling time series and ofvariables. cross-section data leads to further complications, The ability to classify countries correctly de however, which must be considered in interpret pends on how close the group means are relative ing our results (and those of other published to the group dispersions. This point is illustrated studies): in Figures la and 1b, which assume normal 1. The number of rescheduling cases (24) is "bell-shaped" distributions for the two-group still small in comparison with the non-reschedul case and a cutoff inflation value, c. The probabil ing cases (442).18 Plots of variables for the re ity of Type II error (Le., of misclassifying an ob scheduling group, moreover, suggest that the servation from the nonrescheduling group) with data are not normally distributed. Thus, one of a value of p > c ( = 10 percent), thus, is the the theoretical assumptions underlying discrimi shaded area under the bell-shaped function for nant analysis is violated. 19 group I to the right of c, while the probability of 2. The individual observations are "serially misclassifying an observation from group 2 correlated." A country which exhibits a high (or (Type I error) is the shaded area to the left of c low) inflation rate or debt-service ratio in one under the density function for group 2. Error year, for instance, tends to exhibit the same char rates in classifying observations (Le., the per acteristic in other years. This will affect the error centage of observations misclassified) will be rates, since a country which is misclassified (or much greater when there is considerable group correctly classified) in one year will tend to be overlap (Figure la) than if there is only a small misclassified (or correctly classified) in other degree of group overlap (Figure 1b). Differences years.20 in group means, therefore, do not always guaran 3. A problem arises with countries which tee that the rules will yield a useful classification have rescheduled debt more than once. Ghana scheme in which the errors are small. and India, for example, have had debt resched Finally, the proportion of Type I and Type II uled in a number of years since 1966 and 1968, errors depends on the particular cutoff point se respectively, in the process of coordinating aid lected for classifying countries. Moving the cut flows to those countries. Thus, do those resched off value, c, to the right in Figure 1, for example, ulings represent "new events" or extensions of increases the probability of Type I error and re the original reschedulings?21 A question also duces the probability of Type II error, while the arises regarding the treatment of observations of opposite is true ifthe cutoff value is moved to the rescheduling countries in non-rescheduling left. Selection of the cutoff point hinges on an as years. The results reported here delete such ob sessment of the cost ofmaking each type oferror servation, since we are primarily interested in

28 identifying characteristics which distinguish re ficult to test this proposition because ofthe limit scheduling from non-rescheduling countries, ednumber of reschedulings, although the rather than identifying the times of discriminant rule appears to explain recent cases rescheduling.22 as well as earlier cases. 4. The implicit assumption is that the fac Several further pitfalls are often encountered tors contributing to reschedulings are the same in in interpreting results from discriminant analy one period as in other periods-i.e., there are no sis.• One of the. most widely misunderstood as "structural" changes affecting reschedulings (or pects relates to the problem of determining the distributions) during the sample period. It is dif- importance of individual variables. Unlike the

Chart 1A NORMAL DISTRIBUTION WITH LARGE GROUP OVERLAP

Non- rescheduling Rescheduling Group Group

Probability density

0'-----"'----- Inflation Rate (p) Type 1 error Type 2 error

Chart 1B NORMAL DISTRIBUTION WITH SMALL GROUP OVERLAP

Non-rescheduling Reschedul ing Group Group

Probability density

0'--....:::;;.------..10-- Inflation Rate {p} Type 1 error Type 2 error

29 coefficients. in the linear-regression model, the The mea.ninfla.tion rate for the rescheduling discriminant-function coefficients are not gi"OUp wa.snea.l"ly Seven times larger than the unique. (However, the ratios of those coefficients nion-rescheduling group; the money-supply are unique.) Consequently, no test can be made growth rate was nearly four times larger, and the for< the absolute importance of a particular vari a.djusteddebhserviceratiowas about three times ai:l1e.(i.e., settinga p<.irticular coefficient equal to grea.ter. The standa.rd deviation of the inflation zeroorto some other value), although a number rate a.ndmoney-supply growth rate for the res ofrnethodshave been proposed to determine the cl'ledulinggroup, however, were also consider relative importance ofindividual variables. 23 abiylargerthan for the non-rescheduling group, Empirical results owing to the incidence of hyper-inflation and the Two sets ofexplanatory variables were used to sma.ll sample size. As a result, differences in the differentiate rescheduling and non-rescheduling coefficients ()fvariation (i.e., the standard devi cases in the 1960-75 period. The first set includ ation divided by the mean) for the two groups ed variables identified in previous empirical stud were much smaller than the differences in group ies: (1) the debt-service ratio; (2) the reserve means. import ratio; (3) the export growth rate (in U.S. Tests for equality of the multivariate group dollars); (4) the growth rate of real GNP and (5) means and variance-covariance matrices indicat the level of per capita GNP (in 1970 U.S. dol ed that group differences were statistically sig lars). The second set contained variables suggest nificant.26 Under these circumstances, the ed by the monetary approach, and also (6) the appropriate rule for classifying countries would (consumer-price) inflation rate; (7) the growth be a quadratic (rather than linear) function. In rate of the Ml money supply; and (8) a measure most cases tested, however, the linear function of relative purchasing-power parity (the differ yielded comparable results to the quadratic func ence between the domestic and U.S. inflation tion. The linear rule also had the advantage of rates, on a wholesale-price basis, less the rate of being easier to interpret, because of the smaller domestic currency depreciation vis-a-vis the U.S. number ofterms involved. dollar). All explanatory variables were expressed Two separate linear functions were obtained: as three-year annual averages, with the explana (I) Inflation RateIncluded tory variables lagging the dependent variable an 8.72· .21(CPI) - .01(MS) + .04(EX) - .35 (DSA) average ofone year-e.g., with the 1960-62 aver + .03(PP) + .07 (GNP) age inflation rate distinguishing rescheduling (2) Inflation RateExcluded and non-rescheduling cases in 1962. The debt 7.72 - .lO(MS) + .05(EX) - .36(DSA) .05(PP) service ratio was also adjusted to include sched +.lI(GNP) uled (rather than actual) debt-service where: payments. 24 CPI = average annual rate of consumer price inflation A forward step-wise regression procedure was over three-year period used to obtain a measure of the relative impor MS average annual rate of M, growth over three tance of each variable, prior to applying the dis year period criminant sub"routine.25 The results suggested EX averagleannual rate of growth of exports (in that the inflation rate and the adjusted debt-ser U.S. dollars) over three-year period vice ratio were the most important explanatory DSA = average debt-service ratio over three-year peri variables (Table 4). The inflation rate and the od (adjusted to include scheduled debt-service money-supply growth rate were highly correlat payments for rescheduling countries) ed, however, so that the relative importance of PP purchasing-power parity (i.e., a three-year aver the money-supply variable increased consider age of the difference between the domestic and ably when the inflation rate was excluded. Two the U.S. WPI inflation rates, less the rate of do of the variables, the reserve-import ratio and the mestic currency depreciation vis-a-vis the $). level ofper capita income, added little in the way GNP = average annual rate of growth of real output of explanatory power, and thus were omitted over three-year period. from the discriminant sub-routine. The functions were constructed so that the

30 more the negative value, the more likely the Iected. (The cutoff value for the results reported country would be classified in the· rescheduling assUl.nestheexpected cost of Type I error is three group. The prominence of the inflation· rate in times the expected cost of Type II error.) The equation 1 (or money-supply growth rate in over~n error rate is not very meaningful, howev- equation 2) and the adjusted debt-service ratio is er, in view of the large difference in sample size apparent from the weights of these variables in for the two groups ()fcountries. The percentage the discriminant functions, which corroborates ofrescheduling cases in the sample is roughly 5 the finding from the step-wise regression proce- percent; hence, a rule which classifies aU coun- dure. In addition, the negative signs of the coeffi- tries as non-rescheduling cases will have an over- dents of these variables are consistent with the aU error rate of 5 percent. For this reason, it is hypothesis that the probability of rescheduling important to examine the incidence of Type I and increases as their value increases. Type II errors and to see how they vary with the The percentage of c()untries classified incor- cutoffpoint. recdy with these functions ranges from 3 percent Type I error rates vary from 15 to 54 percent, to 11 percent, depending on the cutoff value se- while Type II error rates range from less than 1 Table 4 Sample Characteristics of Rescheduling and Non-Rescheduling Groups'

Variable Non-Rescheduling Group Rescheduling Group

Coefficient Coefficient Standard of Standard of Mean Deviation Variation' Mean Deviation Variation'

Inflation(CPI) Rate 5.6 5.7 1.02 36.7 48.5 1.32 (23.8)2 (21.5)2 (0.9)2 M, Growth Rate 13.9 8.2 0.59 49.6 78.0 1.57 (33.2)2 (31.2)2 (0.94)2 Export Growth Rate 16.3 18.1 1.11 9.7 13.3 1.37 Debt Service Ratio 7.6 5.8 0.76 21.1 8.5 0.40 Real GNP Per Capita Growth Rate 3.7 3.9 1.05 2.3 2.6 1.13 Purchasing Power Parity 4.3 7.2 1.67 8.1 15.0 1.85 Measure of Relative Importance (Percent of explanatory power accounted for by each variable)

Inflation rate Inflation rate Variable Included excluded Inflation(CPI) Rate 42.7% M, Growth Rate 2.0 33.0% ExportGrowth Rate 11.3 14.5 DebtService Ratio 35.5 37.9 Real GNP Per Capita 4.6 7.3 Growth Rate Purchasing Power 3.9 7.2 Parity , Standard deviation + mean

2 These figures are affected by the experience ofhyper-inflation surrounding the Indonesian reschedulings. Values excluding data for Indonesia are in parentheses. • Country data are from International Financial Statistics and from IBRD, World Tables.

31 TableS Discriminant Analysis Results: Classification of Rescheduling Countries 1

Results In<:luding Debt-Service Results Excluding Debt-Service Ratio in Discriminant Function Ratio in Discriminant Function

Countries Correctly Classified Countries Correctly Classified

Argentina (1965) Argentina (1965) Argentina (976) Argentina (1976) Brazil (961) Brazil (1961) Brazil (1964) Brazil (1964) Chile (1965) Chile (1965) Chile (972) Chile (1972) Ghana (1966) India (1968) India (1973) Indonesia (1966) Indonesia (1966) Indonesia (970) Indonesia (1970) Pakistan (971) Peru (1975) Turkey (1965) Turkey (1972) Uruguay (1965) Uruguay (1965)

Countries Incorrectly Classified Countries Incorrectly Classified

Argentina (1962) Argentina (1962) Egypt (1966) Egypt (1966) Ghana (1966) Ghana (1974) Ghana (1974) India (1968) India (1973) Pakistan (1971) Peru (1968) Peru (1968) Peru (1976) Philippines (1970) Philippines (1970) Turkey (1965) Turkey (1972) Yugoslavia (1965) Yugoslavia (1965) Yugoslavia (1971) Yugoslavia (1971) Zaire (1976) Zaire (1976)

1 Results based on two sets of linear discriminant functions; assuming expected costs of Type I error is three times the expected cost ofType II error: Debt Service Ratio Included 8.72 - .21 (CPI) - .01 (MS) + .04 (EX) - .35 (DSA) + .03 (PP) + .07 (GNP) Debt Service Ratio Excluded 4.07 - .22 (CPI) - .01 (MS) + .04 (EX) + .03 (PP) + .03 (GNP)

32 Chart 2 rectlyclassified(Table 5). Reschedulings in RELATION OF ERROR RATES TO VALUE OF CUT-OFF POINT these countries are associated with high inflation (Linear Discriminant Function) and rapid money-supply growth, and the dis criminant rule assigns a relatively large weight to Error Rate (%) these variables. These countries also tend to have high debt-service ratios, but that ratio need not be included to explain their reschedulings. Reschedulings in South Asian countries, on

40 the other hand, require some information on the Type 1 Error adjusted debt-service ratio. India and .Pakistan

30 experienced relatively low inflation rates for the group Qfrescheduling countries (partly owing to 20 the use of extensive price controls), but debt re lieUor these countries (and for Ghana) has be

10 Type 2 Error come a means of supplementing aid flows. The debt-service ratio, in particular, has been used as

0'--'----'----'---'----'------'-- Cut-off anjndicator of need for debt relief by the consor -1 0 1 2 Value tia ofaid donors. (50%) (75%) (90%) The results are somewhat paradoxical in the percent to 11 percent (Figure 2). The ability to light of the traditional approach taken by Avra classify non-rescheduling cases more precisely movie et al. On the one hand, the debt-service ra than rescheduling cases reflects the absence of a tio is found to be an accurate-but largely "well-behaved" distribution for the rescheduling redundant-indicator of those reschedulings as countries-i.e., the variables are highly skewed sociated with short-run balance of payments cri and exhibit large variances. ses. On the other hand, the debt-service ratio is The discriminant rules perform best in ex found to be a critical factor explaining those re plaining reschedulings in South American coun schedulings associated with long-run debt prob tries (Argentina, Brazil, Chile, Peru, Uruguay) lems. In the latter cases, the reasons are political and Indonesia, where 10 out of 12 cases are cor- as well as economic.

IV. Summary andConclusions

This paper has examined two sets of issues in port stagnation and over-importing-and gener volved in country-risk appraisal-the causes of ally to foreign-exchange crises. Cases of chronic past debt reschedulings, and the ability to antici debt relief, on the other hand, appear less amena pate future reschedulings. The evidence suggests, ble to a monetary framework ofanalysis. In par first, that there is a systematic pattern of debt ticular, it becomes difficult to measure the extent reschedulings which is amenable to economic of over-valuation on the basis of inflation-rate analysis.Reschedulings, in short, are not isolated differentials, because of the LDC's tendency to or random events, even though their underlying resort to price controls, capital controls, ex causes are not the same for all countries. change controls, and high tariff barriers. The analysis distinguishes between "liquidity" Knowledge of the causes ofpast reschedulings reschedulings, which are associated with the does not necessarily imply an ability to anticipate bunching of short-term commercial credits, and future reschedulings. The latter is affected by the other reschedulings, which are identified with difficulty ofcorrectly forecasting exogenous var long-term debt reliefon official credits. iables, by changes in structural parameters of es Monetary (and. fiscal) factors appear to be timating equations, and by problems caused by closely involved in the "liquidity" cases. Inflation the small samples used in analyses of previous andover-valued exchange rates lead to excessive resched:uJings. Even so, statistical procedures reliance on foreign borrowing and thence to ex- have an advantage over commercial-bank check-

33 list· systems because they provide a systematic important for country-risk appraisal. The analy m~thod foridentifying variables and for explicit sisin thispa~rsuggests that banks should focus ly considering trade-ofTs. on the inflation rate (and its determinants) and the debt-service ratio as the key economic varia An understanding of past reschedulings, bles afTecting a country's borrowings and its abil moreover, can be useful in delineating what is itytorepay.

FOOTNOTES 1. •See Goodman [131. The Federal Reserve has also recently con economic growth. The condition for equi-proportionate growth of duCtedaninformal survey of bank practices in defining, monitoring, debt and GNP is writtl:in: and controlling foreign lending exposure. 2. ·Sy denominating a loan to an LOC in a key currency, a commer i = r(so - s') cial bitnkcan avoid the risk of exchange rate depreciation of the LDCcurrency, but not the risk of non-repayment. (so - Kr) 3. Estimates of the cost of rescheduling are difficult to obtain since fairly detailed information on the repayment stream is re wherei average interest on foreign debt quired to compute the present discounted values. In case of res So, s' average and marginal savings rate chedulings of official credits it is customary to compute the "grant K incremental capital-output ratio element" of the rescheduling-i.e., the value of the repayment r = growth rate of GNP stream after rescheduling as a fraction of the value of the repay ment stream at commercial interest rates. See Avramovic [31, Malhematical Appendix, pp. 188-192. 4. For purposes of this study, refinancings of individual bank cred 11. "Hitherto, the discussion has been in terms of 'domestic' its are treated as a problem of credit risk, rather than as a problem growth variables, in particular the savings-investment balance. The of country risk. The distinction between refinancings and resche savings-investment gap is equal to the foreign exchange gap, by dulings in many cases is moot, although technically a refinancing definition. However,. this is no more than an ex-post accounting involvell an extension of new credit as compared to a "stretch-out" equality. Mon" interesting is the mechanism by which this equality of an existing credit. is bro!1ght about. The. capacity to transfer savings abroad may be 5. Forecasting precision is affected by the ability tc forecast ex· undermined bye deterioration in terms of trade. The foreign ex og&nous variables accurately and by changes in structural param change gap, allowing for the movement of export and import prices, eters, as well as by the standard error in the estimating equation. may be much larger than the savings-investment gap at constant 6. See Bade [41, Bardhan [51, and McCabe and Sibley [201. prices. The quality is restored ex-post, by a reduction in the 'inter 7. Aliber [11 discusses the analogy of the optimum indebtedness of national value' of domestic savings and, also, by an actual reduc the firm and that of developing countries. His paper examines tionin thedomeatic savings rate as income growth decelerates whether bank lending to developing countries constitutes an effi under the impact of the dl:iterioration of the terms of trade." Avra cient allocation of the world's resources and whether risk premi movic [31, p. 50. ums on LOC loans are too large relative to the cost of rescheduling. 12. The fact that countries such as Bra~i1, Mexico, and Israel have 8. Avramovic uses a separate analytic framework to examine each ready access to international capital markets helps to explain why type of problem. Our discussion is primarily concerned with debt theyareablet()llUCCeaafully sustain high debt-service ratios. in problems associated with a foreign exchange crisis, rather than thelle countries debt can be "rolled-over" much more easily than in with problems stemming from slow economic growth. most othllr developing countries. The theoretical underpinnings for separating the two types of 13. Thill situation existed in K()rea in the period immlldiately fol problems are the "two-gap" models of economic development, lowing the financial reforms of 1964-65. For further discussion of which assume that foreign exchange earnings are limited by inelas this point, see Sargen [231. tic export demand, and that technical substitution possibilities be 14. Feder-Just report overall error rates in classifying countries tween f()reign and domestically produced capital goods are fixed. (i.e., TYPe land Type II errors as a percent of the total number of Under these circumatances, the ex-snte condition for trsde bal observations) ranging from 2 to 5 percent, while Frank-Cline report anceand for equality of domestic savings and investment are writ errorrllltesbelween 8 and 18 percent oHhe sample. ten separately, rather than in the usual fashion, S-I = X-M. The 15. Countries listed in Table 2 coincide with those used in our sta foreign l:ixC/lange constraint is assumed to be binding in the short tillticlllllllnalysis dillcussed in Section III. Countries were selected run, \Vhilethesavinga conatraint is binding over the long run. For a using two criteria: (1) whether they had a debt-service ratio above critique of the two gapmodels, see Nelson [211. 5 percent; and (2) whether time series data on key series were 9. The popularity of the debt-service ratios as a default indicator available dating back 101980. The main group of developing coun datl:isback to the 1930's, when a number of Latin American coun tries omiUed from the llample are African nations. tri&a with high debt service ratios (15% or more) defaulted. See 16.For a dascription of the technique, see Eisenbeis and Avery Mrarnovic [31, p.194. Primary producing countries experienced [91. sharp declines in prices of their export products, increasing their 17. The discriminant technique attempts to minimi~e the following real dllbt burden; at the aame time, new credits were not forthcom "IOllll" function: ing.On the othl:ir hand, there are several examples of countries L = Cl .P (1/2) 11"2 + C2 P(211) 11"1, with high debt-service ratios which have not experienced debt diffi CUlties. These include Australia and Canada during the 1930's where P(1/2) is the probability of assigning an observation to (with investment service-export earnings ratio above 30 percent) group l,given it arose from group 2; C1 is the cost of misclassify and Mexico, Bra~iI, and Israel in recent years. ingan ()bservation to gr()up l,given it is from group 2; 11"1 and 11"2 10. Avramovic examines the properties of a model of foreign bor are thea priori probabilities of an observation being drawn from rowing which assumes a Harrod-Oomar (fixed coefficient) model of groups 1 and 2 respectively.

34 Cllrl continuing long·term debt relief to these countries. The cutoff value corresponds to In -_ . 22. This procedure is used by Feder and Just in their study. If one C2lr2 is Interested in identifying the year that a rescheduling occurs, one 18. The countries in the sample are listed in Table 2. Most of the can follow the procedure of treating the observations as "hold· data cover the period 1960-1975. However, three countries which outs" and seeing how they are classified by the discriminant rule. experienced debt difficulties in 1976 (Argentina, Peru, and Zaire) Alternatively, one may choose to assign observations to three were also included as rescheduling cases. Information on resche groups, instead of two. duling was obtained from Bitterman [6], IMF [17] [18] and OECD 23. See Eisenbeis [lO],pp. 13-14. [22]. 24. The differences in the adjusted debt service ratios and those 19. Non-normality does not necessarily imply that the results are reported by the IBRD (based on actual repayments) are especially invalid, but it may affect the error rate in ways that are not quantifi large for Chile (1974), Ghana (1966), and Turkey (1966), (1971). able. 'tVa are presently experimenting with transformations that Our revisions are based on information contained in Bitterman [6], more closely approximate a normal distribution. IMF [17] [18], OECD [22]. 20. The presence of serial correlation means that the number of 25. See Eisenbeis and Avery [9], pp. 70-75, for a discussion of the independent observations is considerably smaller than the total procedure. number of observations. At present, there are no procedures to 26. The test of quality of the dispersion matrix between. the reo correct for serial correlation using discriminant analysis as there scheduling and non'rescheduling groups yields an F21 .5646 statis are with regression analysis. To get around the problem, one can tic of 63.7, which is statistically significant (i.e. the variances for use each country in the 1960-76 period as one observation, but the the two groups are unequal). Similarly, the test for equality of group number of rescheduling cases is much smaller. means (based on the Mahalanobic D2 1 yields an F6,459 statistic 21. Reschedulings for India in 1973 and Ghana in 1974 have been of 43.5. The test, however, assures the dispersion matrices are treated as new events, because major decisions were reached on equal; hence, the results may not be fully accurate.

BIBLIOGRAPHY

1. Aliber, Robert Z., "Perspectives on LDC External Indebted 13. Goodman, Steven, "How the Big U.S. Banks Really Evalu· ness," manuscript, forthcoming in Lloyd's Bank Review 1977. ate Sovereign Risks," Euromoney, February 1977. 2. Anderson, T. W., An Introduction to Multivariate Statisti 14. Grinols, Earl, "International Debt Rescheduling and Dis cal Analysis (New York: John Wiley, 1958). crimination Using Financial Variables," U.S. Treasury Department 3. Avramovic, Dragoslav, et. aI., Economic Growth and Ex manuscript, 1976. ternal Debt (Baltimore: The Johns Hopkins Press, 1964). 15. iBRD, World 'fabies 1976, (Baltimore: Johns Hopkins Uni 4. Bade R., "Optimal Growth and Foreign Borrowing with Re versity Press). stricted Mobility of Foreign Capital," International Economic Re 16. IMF, International Financial Statistics, various issues. view, XIII (October, 1972),544-552. 17. , Multilateral Debt RlInegotlatlons: Exper- 5. Bardhan, P., "Optimum Foreign Borrowing," in Karl Shell, Ience of Fund Members, (Washington, D.C., August, 1971). ed., Essays In the Theory of Optimal Economic Growth (Cam 18. , Multilateral Debt Renegotiations: Exper- bridge, Mass.: MIT Press, 1967), 117-128. ience of Fund Members 1971-1974, (Washington, D.C., Septem 6. Bittermann, Henry, The Refunding of International Debt, ber 20, 1974). (Durham, N.C.: Duke University Press, 1973). 19. Lutz, Friecfrich, "Money Rates of Interest, Real Rates of 7. Cohen, Neil, "Econometric Debt Early Warning Systems," Interest, and Capital Movements," Chapter 11 in MaintainIng and U.S. Treasury Department, (Manuscript), August 1976. RlIstorlng Balance In International Payments (Princeton: 1966). 8. Dhonte, Pierre, "Describing External Debt Situations: A RolI·Over Approach," IMF Staff Papers, 1975. 20. McCabe, J. and D. S. Sibley, "Optimal Foreign Debt Accu 9. Eisenbeis, Robert and Robert Avery, Discriminant Analy mulation with Export Revenue Uncertainty," International Eco sis and Classification Procedures: Theory and Application (Lex nomIc Review, 17, October, 1976. ington, Mass.: D. C. Heath and Co., 1972). 21. Nelson, R., "The Effective Exchange Rate: Employment 10. Eisenbeis, Robert, "Discriminant Analysis: Application, and Growth in a Foreign Exchange·Constrained Economy", Journ,,1 Potential, and Pitfalls," FDIC Working Paper No. 75-2, forthcoming of Political Economy, 78, June, 1970,546·564. in Journal of Finance. 22. OECD, Development Assistance Committee, A Survey of 11. Feder, G. and R. Just, "A Study of Debt Servicing Capac Past Debt Relief Operations, 1956-1970, DACtFA(71)4, March ity Applying Logit Analysis," Journal of Development Studies 1971. (forthcoming). 23. Sargen, Nicholas, "Optimum Foraign Borrowing, Interest 12. Frsnk, C. R. and W. R. Cline, "Measurement o! Debt Ser Rates and Exchange Rates in Pacific Basin Countries," paper pre vicing Capacity: An Application of Discriminant Analysis," Journal sented at Western Economic Association Meetings, Anaheim, Cali of InternatIonal Economics I, 1971. fornia, June 1977.