International bank lending: Political risk and portfolio diversification

International bank lending: Political risk and portfolio diversification

International Bank Lending: Political Risk and Portfolio Diversification John A. Conybeare, ColumbiaUniversity The extent to which a large U.S. bank...

741KB Sizes 0 Downloads 26 Views

International Bank Lending: Political Risk and Portfolio Diversification John A. Conybeare,

ColumbiaUniversity

The extent to which a large U.S. bank, all U.S. banks, and banks in the Group of Ten took account ofpolitical risk in their international country exposures in 1976 is tested using a simple porrfolio diversification model. Assuming that political risks are important relative to economic risks, and that political risks are uncorrelated across countries, these banks’ exposures should be negatively related to political risk indices. However, theporrfolios of these banks appear to be related to political risk only insofar as political risk is roughly approximated by GNPper capita. International banks were not yet able to systematically vary their international porrfolios with respect to political risk.

Since the early 1970s a great deal of attention has been focused on the risks incurred by multinational banks in lending to a large number of international borrowers. The oil crisis of 1973 led to a large increase in international commercial bank lending as part of the process of recycling OPEC oil revenues. By 1975-1976 this lending pattern had led to fears that a significantly large default by any one borrower could result in a collapse of the international banking system. The failure of the West German Herstatt bank in 1974 and the default of Zaire in 1975 appeared to provide some justification for these fears. Largely in response to claims that continually rising oil prices would eventually trigger the default of a major sovereign borrower and the failure of an important international bank, U.S. bank regulatory authorities have, over the past few years, moved to monitor banks’ international portfolios more closely. An alternative and more optimistic view would be that high risk investments in, say, non-oil-developing countries are simply a result of the rational portfolio diversification predicted by the capital asset pricing model. Much of the alarm about international lending has been directed at the problem of political rather than economic risks. One of the best known studies of political risk has defined it as ‘ ‘ . . . the risk or probability of occurrence of some political event(s) that will change the prospects for Address correspondence to John A. C. Conybeare. Columbia University, !~PW York. NY 1002 7.

Journal of Business Research 12,363-375

of Political

Science,

( 1984)

@ Elsevier Science Publishing Co., Inc. 1984 52 Vanderbdt Ave., New York, NY 10017

Department

363 0148-2%3/84/%3.00

364

John A. C. Conybenre

the profitability of a given investment” [6, p. xi]. If political risks are important and are being adequately taken into account by international banks, this should be reflected in the banks’ choices of an international portfolio. The purpose of this paper is to construct a test which would show whether political risks are taken into account in banks’ actual ex post portfolios, and hence whether the degree of alarm frequently expressed about international bank loans is justifiable. Theory: Political Risk and Efficient Portfolios The modem theory of finance [9] suggests that the rational investor may improve a portfolio of assets through diversification. Consider the case of an investor with two assets available to him: xl with an expected return rl = 5% and standard deviation in return, or risk, sI = 20%; and x2 with return r2 = 15% and risk s1 = 40%. These assets may be combined in a portfolio (P), xl having weight wI , and x2 having weight w2: P(x*,x*)=xlwl

+xzw*.

(1)

The expected return (L’) from this portfolio will be the sum of the weighted returns, where wI and w2 [Eq. (2)] are the proportions of each asset in the portfolio E(x1,X*)=wlrl

+w,r,.

(2)

If the assets are combined in the proportions w, = 5 and w, = 5 , the overall portfolio return will be 8.3%. The superiority of this portfolio may be seen in the standard deviation (5) of the new portfolio [Eq. (3)], where p is the covariance between the rates of return of the two assets

s(x,,x2)=(w1*s12

+wz2sz2 +2ps,s2w,w2)"2.

(3)

If the returns of the two assets do not covary, the portfolio will have an overall standard deviation or risk of 18.7% for a return of 8.3%. Varying the weight of the two assets in the portfolio will produce an efficiency frontier of portfolios which will normally dominate either of the individual assets. The shape of this locus will depend upon the covariance of the returns of each asset (p). The optimal portfolio will be some combination of the two assets in this locus, depending upon the patticular investor’s risk preference. The optimal portfolio may also be determined with respect to an additional risk free asset, leading to a set of

International

365

Bank Lending

portfolios along the “capital market line,” such portfolios being either lending portfolios (containing some of the risk free asset) or leveraged portfolios (the investor borrows to buy more of the selected portfolio on the efficiency locus). A point on the capital market line is selected to reflect the investor’s risk preference. What is of interest here is not the composition of an optimal portfolio, but how the composition will change if the riskiness of a particular asset changes. If the sensitivity of the total portfolio to a change in the riskiness of an asset is constant (k) this may be represented by the rate of change in the portfolio variance [V in Eq. (4)] with respect to the risk (standard deviation sl ) of asset xl [Eq. (5)]. Equation (5) may then be rearranged to produce a function for the relationship of wI (the amount of xl) to the risk (3,) of xl [Eqs. (6) and (7)]. In the case where the covariance of the returns is zero, Eq. (7) reduces to Eq. (8), which shows that the weighting of the asset should have a curvilinear negative relationship to its risk V(x1,Xz)=w12S,2

+w,%22

av

-=~w,~,Q

+2ps,s2w,w,,

+2ps2w,w2

(4)

=k,

as, w12

+ps2w,w2/s,

w1 = [-ps2w2 w1 =

[email protected]

(k/2s1)“2

-kk/2s,

=O,

+2s,k)“21/2s,, if p = 0.

(6) (7) (8)

Portfolio theory has been applied to international investment by, among others, Levy and Sarnat [9] and Grubel [5], who calculate optimal portfolios under various assumptions. The theory should also be applicable to international banks. A recent issue of Euromoney [ 121 provided data on Euromarket lending to 75 countries, ranking borrowing nations according to the market’s revealed overall (i.e., political and economic) risk rating. The simple correlation between the risk ranking and the country exposure of the Euromarket in 1979 was -0.232 (-0.427 in curvilinear form), suggesting that exposure and market estimated risk are indeed related in the manner shown in Eq. (8). This same theory may also be applied to the question posed at the beginning of this paper: do international banks adjust their portfolios to take political risks into account? If would be impossible to calculate a bank’s optimal international loan portfolio without information which has not been

366

John A. C. Conybeare

available (viz., returns on loans to each country, the standard deviations of those returns, and the covariance between the returns of each pair of countries). What can be done, however, is to make certain assumptions which will still produce a testable hypothesis about the correct relationship of political risk to a bank’s portfolio. First, political risks are an important element in the overall country risk, or standard deviation in returns from that country, or else that political risk has a positive correlation with the other determinants of total risk. The first assumption is the important one, since if political risks are not an important determinant of overall country risk there is no reason to be concerned about political risks at all. The voluminous literature cited in Kobrin [8] suggests that most analysts in the field believe political risk to be important. This being the case, indices of political risk may be substituted for the standard deviations of returns from each country in Eqs. (3)-(8). Secondly, the political risks which affect rates of return are not correlated between countries (i.e., returns do not covary across pairs of countries), so that the relationship between a bank’s exposure in a country and that country’s political risk index is that shown by Eq. (8). This assumption [that p in Eq. (7) equals zero] is justified mainly on the grounds of excluding the alternatives. If political risks are positively correlated across countries (i.e., p tends to the value one) the gains from portfolio diversification are reduced-assuming, as before, that political risks are important. In the extreme case, where political risks are perfectly correlated, there would be no reason to have an internationally diversified portfolio at all, and the bank should simply pick that country which reflected its preferred combination of risk and return. However, the fact that banks do diversify their international loan portfolios suggests that they do not believe this to be the case. Nevertheless, there may well be individual cases of intercountry political risks being correlated (e.g., regional warfare in the Middle East), though in the global aggregate the correlation is probably low. At the other extreme, there is no theoretical rationale for believing that political risks should be negatively correlated, since it is hard to imagine what kinds of factors could, at least in the global aggregate, increase one country’s political riskiness at the same time that it lowered another country’s risk. Thus, the assumption that the political risks are, in the global aggregate, uncorrelated seems to be the only reasonable assumption upon which to base the following test. The economic studies of internationally diversified portfolios have suggested that this is also true for purely economic risks, so that nonsystematic risks in international bank lending can be eliminated through diversification [4].

International

Bank Lending

367

Finally, any large scale test of diversification must rely on ex post data of a bank’s actual exposure in a country. If one is testing whether the bank ex ante took political risk into account in making portfolio decisions, one must assume that the bank was able to realize its ex ante plans. A bank may be ex post locked into an exposure it would not ex ante have voluntarily entered. High risk countries have, on occasion, forced banks to extend further loans which the banks would not otherwise have made, through threats of default in the absence of further finance. However, such cases are unusual, and notable more for the publicity they attract than for their frequency. Nevertheless, specifying the objective of this test more accurately, it is an attempt to determine whether the relationship between bank exposure and political risk is that which it ought to be ex post given the assumptions that political risk is an important determinant of overall risk and that, in the global aggregate, political risks are not correlated across countries. The relationship between exposure and risk ought to be that shown in E!q. (8), which implies that exposure should be negatively related to a country’s political risk. The following sections of the paper attempt such a test on three examples of international bank exposure: case A (a single bank, one of the largest commercial banks in the US), case B (all US banks) and case C (banks in the Group of Ten and Switzerland). Case A: A Single U.S. Bank

Data were obtained on various categories of the global exposure of one of the largest U.S. commercial banks at June 30, 1976. Five variables were constructed as measures of the bank’s global exposure: total assets (TAS); total assets less liabilities (ASL); and total assets less foreign liabilities (AFL) by customer address, reported by foreign branches; loans plus investments (LIN) and loans with maturities over one year (LOA), both reported by overseas branches, wholly owned banking subsidiaries, domestic international and Edge Act subsidiaries. The variable ASL was included as a measure of net country exposure of the bank’s foreign branches, though AFL is probably a better measure of net exposure since only foreign held liabilities (e.g., deposits by residents of the country in a foreign branch not located in their country of residence) could be used to offset the bank’s asset exposure in the country. The bank in question did not, at that time, attempt to quantify political risk. Its quantitative (cardinal) risk assessment was limited to the construction of two economic risk indices, an “adaptability” index (ADN) and a “debt servicing” index (DSN). The adaptability index was

368

John A. C. Conybeare

Table 1: Case A: Economic Risk and Country Exposure of a Single U.S. Banka Dependent Variable

Independent Variables DSN ADN

TAS

46.873 (3.454) 47.373 (2.829)

TAS ASL

12.012 (3.576) 11.049 (2.567)

ASL AFL

27.290 (3.921) 27.702 (3.178)

AFL LIN

15.409 (4.930) 13.207 (2.956)

LIN LOA LOA

6.026 (0.599) -3.539 (-.302)

R2

F

0.362

11.928

0.276

8.002

0.378

12.787

0.239

6.588

0.423

15.372

0.325

10.100

0.5 36

24.297

0.294

8.739

0.017

0.359

0.004

0.091

e n = 23; intercepts omitted; f statistics in parentheses beneath regression coefficients. Countries included in the sample are West Germany, Japan, Taiwan, Indonesia, Phillipines, Paraguay, India, Korea, Finland, Argentina, Mexico, Egypt, Spain, Brazil, U.K., Chile, Peru, Zaire, Sri Lanka, Italy, Uruguay, Jamaica, and Portugal.

a mixture of general economic indices (e.g., GDP per capita, export concentration and earnings, consumer price movements, savings, food and fuel imports, exchange rate adjustments, and IMF position). The debt servicing index was composite measure of assessments of the external debt servicing capacity (net external earnings and transfers relative to current public debt), months of imports covered, public external debt as a proportion of GNP and a “compressability” ratio (proportion of earnings from exports spent to finance essential imports and debt payments). The bank’s two economic risk indices, ADN and DSN, are highly collinear, and were therefore regressed separately against the dependent variables. The results in Table 1 are, with exception of LOA, consistent with the theory outlined above. Economic risk is negatively associated

International

Bank Lending

369

with the bank’s exposure in the 22 countries (a mixture of developed and developing countries) for which risk assessments were available. Clearly, the bank’s expost exposure did take into account its own ex ante economic risk evaluations, in a manner consistent with the relationship shown in Eq. (8). Was the same true for political risks? The extent to which political risk variables would produce results consistent with diversification theory was tested using six variables created by Jackman [7]. These variables, with the predicted relationship to bank exposure in parentheses, are democratic performance, DEM (positive); institutional longevity, INS (positive); coups d’ etat, CDE (negative); collective protest, PRO (negative); internal war, WAR (negative) and socialist strength, SOC (negative). The variables are explained and justified as risk indices in the Appendix. Table 2 shows the results of regressing the bank’s five asset categories on these risk indices. There is little collinearity among the risk indices, except between CDE and WAR (r = 0.41) and between DEM and SOC (r = 0.35). Only collective protest (PRO) and socialist strength (SOC) are significant predictors for the whole range of dependent variables, yet they have coefficient signs opposite to that predicted by the theory; viz., country exposure is positively related to these risk indices. Only one of the equations (that explaining the variable LIN) has an F statistic significant at the 5% level. Logarithmic equations produced only slightly higher R2 and F statistics for those asset categories having no negative values (TAS, LIN, and LOA). In general, the risk indices explain very little of the distribution of the bank’s country exposures. It has been suggested that firms may use GNP per capita as a much simpler measure of political risk [ 111. The global exposure of the bank had no significant linear relationship to GNP per capita, in U.S. dollars for 1975 (data from Ref. 12). However, a strong curvilinear relationship was apparent (see Table 2) for the assets TAS, LIN, and LOA. This suggests that political risks were systematically incorporated into the bank’s portfolio policy only insofar as political risk is measured by per capita income. Case B: U.S. Bank Claims on Foreigners The same six risk variables were used to explain short-term claims on foreigners reported by banks in the U.S. at October 1976 (data from Ref. 3). Table 3 shows the results for this variable (USB). The resulting equations do not produce any significant results. In logarithmic form the equation produced a slightly higher R2 statistic (0.223), but no significant t statistics. GNP per capita had no linear or nonlinear relationship to

8.564 (1.087) 0.895 (0.404) 3.885 (0.883) 0.987 (0.458) -3.459 (-0.641)

0.355 (0.557) 0.052 (0.288) 0.713 (0.486) 0.220 (1.263) 0.186 (0.428)

CDE

Variables

(5.765)

In (GNP per capita) 2.037 (6.460) 1.753 (6.158) 1.857

9.570 (1.060) -0.462 (-0.182) 3.010 (0.597) 1.797 (0.728) -2.014 (-0.326)

INS

Independent

4.480 (2.438) 0.718 (1.390) 2.031 (1.981) 1.488 (2.963) 1.633 (1.300)

PRO

Exposure

sot 9.778 (1.512) 3.733 (2.053) 6.501 (1.801) 5.631 (3.186) 6.236 (1.410)

WAR -0.047 (-0.874) -0.007 (-0.48 1) -0.022 (-0.722) -0.014 (-0.942) -0.033 (-0.910)

of a Single U.S. Banka

0.204

0.226

0.243

0.069

0.281

33.238

37.920

41.725

0.629

3.315

1.654

1.227

0.126 0.163

2.042

F

0.194

R2

Q n = 58 (for DEM, INS, CDE, PRO, WAR, SOC); n = 132 (for GNP per capita); intercepts are omitted; t statistics are in parentheses beneath regression coefficients.

In (LOA)

In (LIN)

In (TAS)

LOA

LIN

AFL

ASL

TAS

Dependent Variable DEM

Table 2: Case A: Political Risk and Country

$ a 2

k3”

n h

P

3 3”

(0.178) 1.832

(0.175) -0.429 (-0.079)

In (INS) 0.184 (0.527) -1.164 (-3.464)

(0.503) 99.589

(1.350) 9.729 (-0.258)

GPAc

In (DEM) 0.206 (0.2 14) In (CPA)= -0.048 (-0.253)

a Intercepts bn = 36. cn=56.

In (GPA)c

In (USB)b

In (USB)b

are omitted;

I statistics

beneath

In (PRO) 0.148 (0.84 1) 0.000 (-0.236)

(3.158) 6.314 (0.698)

(0.840) 55.763

3.614

regression

coefficients.

In (WAR) 0.074 (0.496) 0.749 (-2.605)

(-1.177) -0.163 (-0.634)

(-0.627) -0.591

-0.094

WAR

In (SOC) 0.102 (1.207) -0.811 (-4.428)

(1.800) 32.964 (1.057)

(1.081) 109.600

17.360

sot

0.754

165.064

0.227

17.998

0.688

0.007

1.385

0.290

2.941

0.545

F

0.223

0.034

0.265

0.101

R2

of U.S. Banks and All Banks in the Group of Ten”

PRO

Variables

are in parentheses

ln (CDE) 0.026 (0.239) -1.104 (-7.420)

(0.583) 0.952 (0.316)

(0.494) 3.149

0.700

CDE

Independent

In (GNP per capita) 0.000 (0.476) 1.458 (12.848)

4.216

10.077

USBb

GPLC

INS

Dependent Variable DEM

Table 3: Cases B and C: Foreign Exposure

z 2 ?Y z

ii 2

312

John A. C. Conybeare

USB. Thus US bank claims on foreigners show no clear relationship to any of the possible risk indices. Case C: Foreign Claims of Banks in the Group of Ten Finally, the six risk variables were tested on two dependent variables measuring assets (GPA) and assets less liabilities (GPL) in foreign countries of banks in the Group of Ten, Switzerland and of the foreign branches of U.S. banks in the Caribbean area and the Far East at December 1976 (data from Ref. 1). These results are also presented in Table 3. No independent variables are significant in both linear equations, and only DEM has a significant and correct sign. The logarithmic equation for GPA fares much better, with four significant variables, although two (INS and WAR) have the wrong signs. The best prediction is obtained in a curvilinear regression on GNP per capita (R2 = .75). Again, the results show little significant relationship between country exposure and political risk, though more than was evident in cases A and B. Conclusion The country exposures of the banks examined do not appear to have any consistently negative relationship to a variety of measures of political risk, with the partial exception of GNP per capita in cases A (the single U.S. bank) and C (banks in the Group of Ten). This generalization should be qualified with two comments on the specification of risk. First, banks may be using different political risk indices than those used in this paper. This could be a serious problem if the banks were able to measure some aspects of political risk which are better estimates of “true” political risk than those used here. This is unlikely since, as is noted in the Appendix, all of the available measures of political risk cited in the literature measure the same kinds of phenomena, and it is hard to imagine what dimensions of political risk have not already been incorporated into the existing indices, except for those relatively ad hoc, unquantifiable aspects of risk which could not be incorporated into any index. If banks are simply using different, but inferior, measures of political risk, the conclusions of this paper still hold. Second, it may be that political risk is simply not important relative to other risks in the determination of the overall variance in the returns from an individual country. A casual survey of the political events affecting the international banking market (e.g., military govemments in South America, religous revolutions in the Middle East and wars in Africa and

International

Bank Lending

373

South East Asia) makes such a judgment hard to accept. It is more likely that banks are either not making systematic assessments of political risks or are unable to incorporate these ex ante judgments into their ex post portfolios. Kobrin [8, p. 741 suggests that the former reason is most prevalent: With very few exceptions, managers rate political instability (or political risk) as one of the major influences on the foreign investment decision process. Yet, again with very few exceptions, the same surveys report the absence of any formal or even rigorous and systematic assessment of political environments and their potential impact on the firm.

This paper has provided some evidence for inferring that international bankers have had no more success at systematically incorporating political risk into portfolio decisions than have foreign investors in general. Banks with large international exposures should therefore be able to significantly improve their performance by devising quantitative indices of political risk, or by using almost any of the many indices already available. They may then be able to deal with political risk in a systematic and global context, rather than as an ad hoc response to specific crises. Appendix: Choice of Political Risk Indices Many objective indices of political risk are available, mostly aiming to measure the same kind of phenomenon: ‘ ‘ . . . an event occurring either in the environment (for example, instability) or at the junction of environment and enterprise (for example, a nationalization), typically associated with an act of government that has unfavorable consequences for the firm” [8, p. 68-691. The specific types of events typically measured are those such as violence, changes of institutional structure and degree of democracy or popular support. The problems associated with measuring such events are discussed at some length in Haendel et al. [7, pp. 53 -72 1, who devise their own composite political stability index for a sample of developing countries. It would be hard to justify one set on political risk indices over another on a priori grounds unless one wished to measure specific types of political risk better captured by some indices than others. Since they all measure the same kinds of events there does not appear to be any compelling reason to choose one index or set of indices over another for the measurement of general political risk. Jackman’s indices were chosen because they appeared to cover the broad range of political risk factors, they were available for a sample of countries roughly the same as the sample for which bank exposure data were available, and they also

John A. C. Conybeare

374

covered a long time period (post World War Two to the 1970s). Most of the other indices available are too time specific to be usable in conjunction with 1976 bank exposure data. Haendel’s stability indices, for example, are for the period 1961-1966, and show the effects of choosing a narrow time period (e.g., the Dominican Republic is ranked as the most unstable developing country). The five indices selected from Jackman are briefly explained below. Democratic performance (DEM) was constructed from four components: number of adults voting as a promotion of the voting age population, party competitiveness, electoral irregularity and freedom of the press [7, p. 641. This variable was included in deference to the general belief, probably originating in Tocqueville’s analysis of American democracy, that democracies are less likely to produce large scale unanticipated changes in the political structure, because democratic culture and voting procedure force changes into an incrementalist framework. Institutional longevity (INS) has three components: age in decades of the current constitutional form by 1968, age of the largest political party divided by the number of parties winning more than 5% of the lower house seats [7, pp. 94-951. Longevity should lower political risk, since the longer an institutional structure remains unchanged, the less chance that foreign investment will face sudden changes in the rules governing its operating environment. Corps d’ etat (CDE) is the number of attempts at irregular power transfer (successful or otherwise) from 1948 to 1957 [7, p. 971. Collective protest (PRO) has three components: riots, antigovernment demonstrations and political strikes [7, p. 981. Internal war (WAR) includes armed attack events, assassinations and deaths from political violence [7, p. 981. Socialist strength (SOC) is the proportion of seats in the lower house held by parties of noncommunist left [7, p. 1271. The strength of socialist forces in a country is usually taken to be an index of the probability of expropriation or at least some loss for which compensation may not be forthcoming.

References 1.

Bank for International

2.

The Country

3.

Federal

4.

Goodman,

Settlements,

Risk League

Reserve

Bulletin

Report,

Basel, 1977.

(February

1980):

40-48.

(May 1977).

L. S., Bank Lending

Federal Reserve Bank NYMonthly 5.

Annual

Table, Euromoney

to Non-OPEC LDCs: Are the Risks Diversifiable, Rev. (Summer 1981): I-20.

Grubel, H. G., internationally Diversified Portfolios: Welfare Flows,Am. Ecorz. Rev. 58 (December 1968): 1299-1314.

Gains

and Capital

International 6.

375

Bank Lending

Haendel,

D., G. T. West, and

D.C. Heath

& Co., Lexington,

Overseas Investment and Political Monograph No. 21, Lexington Books-

R. G. Meadow,

Risk, Foreign Policy Research institute MA 1975.

I.

Jackman, R. W., Politics and Social Equality, A Comparative Analysis. Wiley, New York, 1975.

8.

Kobrin, (1979):

9.

Levy,

S. J., Policital 67-80.

Risk,

H., and M. Sarnat,

A Review

International 1970):

Am. Econ. Rev. 60 (September 10.

Markowitz,

11.

Rummel,

R. J., and D. A. Heenan, Harvard Bus. Rev. (January-February

H. M., Portfolio

Selection,

12.

World Bank, World Bank Washington, D.C., 1978.

Atlas:

and

Reconsideration,

Diversification 668-675.

of Investment

J. Finance 6 (March How Multinationals 1978): 67-76.

Population,

J. Int. Bus. 10(l)

Per Capita

1952): Analyse

Product

Portfolios,

77-91. Political

Risk,

and Growth,