Exchange Rates and National Price Levels

The concept of purchasing power parity (PPP) has served as a building block for many exchange rate models during the 1970s and 1980s. The term is usually associated with Cassel (1918), but the PPP notion has been traced as far back as the sixteenth century (see Einzig (1970, pp. 145–46)).

The concept of PPP, which has several variants, is basically a notion that the exchange rate between the currencies of any pair of countries should equilibrate to a ratio of aggregate price indices for the two countries (multiplied by an appropriate constant scale factor), or that the percentage change in the exchange rate should equal the difference between the percentage rates of inflation in the two countries.⁴ As a parity or arbitrage condition that characterizes the equilibrium relationship between an exchange rate and a ratio of price indexes, PPP does not necessarily imply that causation simply runs in one direction, although in using the PPP assumption as a building block for models of exchange rate determination it has in general been assumed that causation runs from changes in exogenous variables through changes in the ratio of price indexes to changes in the exchange rate.

The assumption of continuous or short-run PPP has been a building block for the class of flexible-price monetary models, which were among the first empirical exchange rate models to emerge during the 1970s after the shift to a regime of floating but managed exchange rates. Examples of such models include Frenkel (1976), Bilson (1978, 1979), and Hodrick (1978). By the early 1980s, however, it was widely recognized that continuous or short-run PPP was rejected convincingly by the data.⁵

A weaker form of the PPP assumption is the hypothesis of a time-invariant value of the expected long-run PPP level, in which the PPP level is defined as the nominal exchange rate (currency A per unit of currency B) multiplied by the ratio of national price levels (expressed as the price level for country B divided by the price level for country A). The assumption of time-invariant expectations about the long-run PPP level has been a building block for the sticky-price monetary models, based on the analytic framework developed by Dornbusch (1976) and implemented empirically, for example, by Frankel (1979, 1981).

Assessments of this assumption of time-invariant expectations have looked for evidence of a tendency for PPP levels—or “real exchange rates”—to return toward equilibrium values over time. Some formal tests have focused on coefficients of serial correlation between deviations of a real exchange rate from an assumed equilibrium PPP level, typically represented by a sample mean or a trend line, and have interpreted the coefficient of serial correlation as one minus the speed of adjustment toward the equilibrium PPP level (for example, see Frankel (1985)). One limitation of such tests, however, is that a finding that the serial correlation coefficient was significantly less than unity would not, by itself, establish the time invariance of the expected long-run level of the real exchange rate. In particular, such a finding could not reject the “overshooting” hypothesis that “shocks” cause “jumps” in the level of the real exchange rate that is expected to prevail in the long run, but even greater jumps in the level of the real exchange rate that is observed in the short run.⁶

Although it seems impossible to devise a statistical test that could verify the hypothesis of time-invariant expectations about the long-run level of the real exchange rate, the hypothesis could be rejected if there were clear evidence that real exchange rates follow random walks, or that changes in real exchange rates exhibit no serial correlation. Roll (1979), Frenkel (1981b), and Cumby and Obstfeld (1984), among others, have found it difficult to reject the random-walk hypothesis. By contrast, Frankel (1985) succeeds in rejecting the random-walk hypothesis with a sample of 116 annual observations on a price-adjusted exchange rate between the U.S. dollar and the British pound. Frankel also finds that for shorter samples, in which he is unable to reject the random-walk hypothesis, his point estimates of the first-order serial correlation coefficient for the level of the real exchange rate are less than unity, as Frenkel (1981b) also found. Such point estimates are consistent, moreover, with evidence available from Boughton (1987); the real exchange rate equations that performed best in his postsample tests are equations that include a lagged dependent variable with a coefficient estimated to be somewhat less than unity. Thus, the hypothesis that real exchange rates follow random walks has not been established convincingly.

These considerations suggest three conclusions about the relationships between exchange rates and national price levels, or about the relationships between rates of change in exchange rates and inflation rate differentials. The first two conclusions are (1) that the assumption of continuous or short-run PPP can be rejected, and (2) that the assumption of long-run PPP—in particular, of time-invariant expectations about long-run real exchange rates—seems virtually impossible to support statistically (although it has not been rejected convincingly by statistical tests).⁷ The second point does not imply that empirical models should ignore the types of first-order homogeneity conditions that theoretical foundations suggest, but rather implies that models should provide scope for expected long-run real exchange rates to vary in response to unexpected changes in exogenous real variables. The elements of truth, however, also include a third and somewhat-related point: (3) that during episodes of hyperinflation, or over periods of time sufficiently long for ratios of national price levels to change radically, the PPP hypothesis may have considerable explanatory power.⁸

Exchange Rates and Interest Rates

Although rudiments appeared in the 1890s, Keynes (1923) is usually credited with developing the concept of interest rate parity, which focuses on relationships between exchange rates and interest rates (see Einzig (1970, pp. 214–15)).⁹ In essence, the concept of interest rate parity recognizes that investors have a choice between holding assets denominated in domestic currency, yielding the own rate of interest r_d, and holding assets denominated in foreign currency, yielding the own rate of interest r_f. Thus anyone with a unit of domestic currency to invest should compare the option of accumulating 1 + r_d units with the option of converting it at the spot rate into s units of foreign currency, investing this in foreign assets, and arranging to convert back his principal plus interest—at a forward exchange rate f (in foreign currency per unit of domestic currency)—into s (1 + r_f)/f units of domestic currency for delivery at the end of the interest-payment period. To the extent that investors can accumulate either (1 + r_d) or s (1 + r_f)/f units of domestic currency with certainty,¹⁰ arbitrageurs in pursuit of assured profit will move funds in whatever amounts are required to eliminate any discrepancies between these interest factors. Thus, asset market equilibrium requires the condition of covered interest parity (CIP):

or, for sufficiently small values of r_d,

By contrast, the notion of open or uncovered interest parity (UIP) is based on the argument that speculators who could arrange forward to deliver (or obtain) f units of foreign currency at some future date, in exchange for one unit of domestic currency, would be tempted to take an uncovered position in the forward exchange market if the spot exchange rate that they expected to prevail on that future date (s^e, in foreign currency per unit of domestic currency) offered them the chance to convert back into foreign currency (or domestic currency) with an expected profit s^e–f> 0 (or 1/s^e–1/f> 0). Under the extreme assumption of risk neutrality, such arguments translate into the UIP hypothesis:

and hence, for sufficiently small values of r_d,

Despite considerable confusion in some of the literature on the CIP condition, it is now in general accepted that observed deviations from CIP reflect transaction costs, the influence of capital controls, or the fact that the empirical data on interest rates do not refer to sufficiently comparable foreign and domestic assets.¹¹ It is also recognized that deviations from CIP are usually negligible when Eurocurrency data are used as measures of interest rates.¹² By the same token, however, observed deviations from CIP can play a central role in those empirical studies that emphasize that claims on the residents of different countries may not be comparable because of capital controls or different political or credit risks.

The evidence on UIP, by contrast, remains controversial, even though the hypothesis of equality between the forward exchange rate and the expected future value of the corresponding spot exchange rate has received considerable testing. To the extent that expectations about future spot rates are not observed directly, many tests have looked for indirect evidence by relying on the assumption that expectations are formed rationally, such that UIP would imply that forward rates were unbiased predictors of future spot rates. A discussion of econometric issues, and a survey and extension of test results, has recently been provided by Cumby and Obstfeld (1984) (see also Levich (1985)). Cumby and Obstfeld conclude (p. 139): “The test results are on the whole inconsistent with UIP, and they also suggest that forward premia contain little information regarding subsequent exchange rate fluctuations.” As Cumby and Obstfeld note, however, one caveat in rejecting UIP on the basis of indirect tests is the “peso problem”: in finite samples, UIP is not necessarily invalidated by the finding that forward rates are biased predictors of future spot rates, since bias can emerge whenever rational market participants had repeatedly expected exchange rates to change in response to a policy action or some other event that repeatedly failed to materialize over a considerable part of the sample period (see Rogoff (1979) or Krasker (1980)).

The qualified rejection of UIP has directed interest to the question of whether the magnitudes of deviations from UIP—in general referred to as exchange risk premiums (except under conditions in which CIP fails to hold)—have been large enough to “explain” a substantial part of the observed behavior of exchange rates. That possibility has emerged as one conceivable explanation for the observation of large differences between changes in spot exchange rates and the ex ante levels of forward premiums, since by definition (when CIP holds) such differences must equal sums of an exchange risk premium plus an unexpected change in the corresponding spot rate. Interest has also focused on whether exchange risk premiums have varied significantly over time.

Two new types of evidence have recently emerged: the first based on a statistical approach developed by Fama (1984), and the second consisting of survey data on exchange rate expectations, which have been analyzed by Frankel and Froot (1985). ¹³ Fama has provided indirect evidence (conditional on the hypothesis of market rationality) that, during the period from the end of August 1973 through the end of 1982, the variance of the risk premium exceeded the variance of expected changes (over one-month intervals) in the spot rate for exchange rates of the dollar against each of ten other major currencies (see also Hodrick and Srivastava (1986)). This ranking of variances is not supported by the direct evidence from the survey data analyzed by Frankel and Froot (which, however, provide a smaller sample size and measure changes over three- and six-month intervals), but the survey data do verify Fama’s implicit finding that both the magnitudes and the variances of exchange risk premiums have been large.¹⁴ Table 1 provides summary information on the magnitudes of exchange risk premiums during the 1980s, based on a subset of the data reported by Frankel and Froot. Neither the 13 Economist respondents nor the several hundred Amex respondents came close to expecting, on average, that future spot rates would equal the currently prevailing forward rates.¹⁵

Table 1.

Exchange Rate Data

(Percentage change over six-month horizon at annual rate)

Source: Adapted from Frankel and Froot (1985, Table 2)

Table 1.

Exchange Rate Data

(Percentage change over six-month horizon at annual rate)

	Expected
Non-Dollar	Depreciation	Forward Premium	Exchange Risk
Currency	of Dollar, Based	on Non-	Premium on Non-
Unit	on Survey Data	Dollar Currency	Dollar Currency
Averages from eight nonoverlapping Economist surveys, 1981–85
British pound	3.92	0.39	3.53
French franc	4.60	-5.44	10.04
Deutsche mark	12.81	4.28	8.53
Swiss franc	12.35	5.87	6.48
Japanese yen	12.71	5.16	7.55
Averages from four Amex surveys, 1981–84
British pound	6.11	2.56	3.55
French franc	2.25	-3.36	5.61
Deutsche mark	10.30	5.46	4.84
Swiss franc	6.31	7.56	-1.25
Japanese yen	8.73	6.77	1.96

Source: Adapted from Frankel and Froot (1985, Table 2)

View Table

Table 1.

Exchange Rate Data

(Percentage change over six-month horizon at annual rate)

	Expected
Non-Dollar	Depreciation	Forward Premium	Exchange Risk
Currency	of Dollar, Based	on Non-	Premium on Non-
Unit	on Survey Data	Dollar Currency	Dollar Currency
Averages from eight nonoverlapping Economist surveys, 1981–85
British pound	3.92	0.39	3.53
French franc	4.60	-5.44	10.04
Deutsche mark	12.81	4.28	8.53
Swiss franc	12.35	5.87	6.48
Japanese yen	12.71	5.16	7.55
Averages from four Amex surveys, 1981–84
British pound	6.11	2.56	3.55
French franc	2.25	-3.36	5.61
Deutsche mark	10.30	5.46	4.84
Swiss franc	6.31	7.56	-1.25
Japanese yen	8.73	6.77	1.96

Source: Adapted from Frankel and Froot (1985, Table 2)

The magnitudes of the exchange risk premiums shown in Table 1 make it difficult to defend the UIP assumption. Unlike indirect tests of the UIP assumption, comparisons of forward exchange rates with direct statements about exchange rate expectations are not distorted by peso problems.¹⁶ On the other hand, questions can always be raised about the quality of survey data, and it seems particularly relevant to extend or breakdown the survey evidence in order to check that the apparent aversion to exchange risk indeed applies to the most active participants in exchange markets. In the absence of survey evidence that major exchange market participants do not behave in a risk-averse manner, however, continuing reliance on the UIP assumption could be counterproductive.¹⁷

What, then, should be concluded about the relationships between exchange rates and interest rates? Two apparent elements of truth are (1) that observed deviations from CIP in general reflect transaction costs, the influence of capital controls, or the fact that the data under observation do not include interest rates on sufficiently comparable assets, and (2) that a growing body of evidence appears to reject the hypothesis of UIP. It should be added, however, that some economists remain skeptical of the apparent evidence against UIP and that, in any case, what should be kept in perspective is (3) that the important issue for purposes of modeling exchange rates is whether the UIP assumption introduces a relatively major or a relatively minor degree of inaccuracy. (The issue of whether continued reliance on the UIP assumption may be counterproductive will be discussed further in the second subsection of Section II, “Modeling Portfolio Preferences in a General Equilibrium Framework.”) What also should be kept in perspective is (4) that the interest parity conditions are simply relationships between endogenous variables, and that a satisfactory explanation of the co-movements of exchange rates and interest rates may require that such conditions be embedded in models that take account of a complete system of macro-economic relationships simultaneously (as will be discussed in the first subsection of Section II, “Complete Macroeconometric Models and the Treatment of Expectations”).

Exchange Rates and International Balances of Payments

The notion that exchange rates move to equilibrate supplies of and demands for currencies and, in that general sense, to bring balance to international payments has been traced as far back as the middle of the seventeenth century (see Einzig (1970, pp. 142-43)). Before the breakdown of the Bretton Woods regime in the early 1970s, it was popular in classroom models to simplify the analysis of flexible exchange rates by hypothesizing that the demand for foreign exchange varied mainly with import payments, whereas the supply of foreign exchange varied mainly with export receipts. Thus the classroom analysis was based on a single-period model of “flow” equilibrium in which the market-clearing exchange rate was driven primarily (if not entirely) by factors affecting the demands for and supplies of imports and exports in goods markets. Some variations of the analysis introduced changes in official foreign exchange reserves or “exogenous” private capital flows as factors that might accommodate or autonomously give rise to trade or current-account imbalances, but little if any consideration was given to the notion that net private capital flows might exhibit elastic responses to small changes in market incentives, such that trade and current account imbalances could exhibit large shifts with little if any change in market-clearing exchange rates. The failure to consider that possibility undoubtedly to some extent reflected the relatively low degree of capital mobility during the early years of the Bretton Woods regime. It also may have reflected an apparent consistency of the classroom model with the manner in which exchange rate expectations responded to current account imbalances during a regime in which official permission or pressure to adjust exchange rates was predicated on the occurrence of “fundamental disequilibrium,” which in practice had come to be interpreted as the occurrence of persistent current account imbalances.

Needless to say, experience during the first few years of generalized floating destroyed the credibility of exchange rate models that were based on the notion of single-period equilibrium in international flows of goods (or of goods and official reserve holdings). Attention shifted toward exchange rate models that highlighted equilibrium conditions for stocks of assets. Kouri (1976) made an important contribution to integrating the conditions for flow equilibrium in goods markets and stock equilibrium in asset markets, focusing attention on the dynamic inter-actions between the exchange rate and the current account. In particular, Kouri emphasized that the exchange rate must be consistent with conditions of asset stock equilibrium in the short run, that the exchange rate influences the current account balance and hence the change in the net foreign asset position, and that the change in the net foreign asset position in turn feeds back to influence the path of the exchange rate that maintains continuous asset equilibrium over time.

Models of the interactions between the exchange rate and the current account depend on the specification of goods and factor markets (see Bruce and Purvis (1985) for a recent survey). Two types of specification have received particular attention: models of specialized production in which each country produces a single and differentiated tradable good, and models in which each country produces both a nontradable good and a tradable good that is homogeneous across countries. In both types of model, the supply and demand conditions in goods and factor markets lead to a semi-reduced-form equilibrium condition between the current account (or trade balance) and a relative price variable, but the link between relative prices and the nominal exchange rate—hence, the link between the current account and the nominal exchange rate—depends on the time paths of money supplies and on whatever other variables influence the absolute levels of national price indices.

In addition to establishing that the association between the current account and the nominal exchange rate depends on the separate links of each of those two variables to relative price levels, the literature has clarified that the joint dynamics of the exchange rate and the current account in responding to a shock depends on which “forcing” variable is shocked, on the degree to which the shock is expected to be transitory or permanent, and on the time lag within which the change in the forcing variable follows the shock from which the change becomes anticipated. These points are illustrated, for example, by Dornbusch and Fischer (1980), and in more general terms by the dynamic framework provided in Mussa (1984) and Frenkel and Mussa (1985) (see also Obstfeld and Stockman (1985) for a survey of models of exchange rate dynamics).

Models of the simultaneous determination of the exchange rate, price levels, and the current account have also provided some understanding of the kinds of factors that determine and generate variability in the expected long-run level of the real exchange rate (that is, the nominal exchange rate adjusted by the ratio of national price levels). In the models developed by Kouri (1976), Dornbusch and Fischer (1980), and Mussa (1984), current account imbalances are “settled” with a single type of asset, and conditions are directly or indirectly imposed on the terminal net stock of that asset, to which current account imbalances must accumulate. In that context the solution for the long-run, stationary-state level of the real exchange rate is that constant level that is consistent with a stationary net foreign asset position—hence a balanced current account—under perfect foresight. Buiter (1981) has provided an overlapping-generations model that yields insights into the implications of moving from a stationary state to a steady-growth state and into the fact that technologies, rates of population growth, and rates of time preference can all influence the long-run equilibrium position. Obstfeld (1985) used a small two-country model that does not impose current account balance as a condition of long-run equilibrium; within such a framework, he infers from plausible parameter values that the expected level of the long-run real exchange rate may be quite sensitive to revisions in expectations about the relative long-run levels of national outputs.

These developments in the conceptual literature during the 1980s have been bringing the current account back into the analysis of exchange rate determination by emphasizing the interactions between the expected long-run value of the current account and the expected long-run level of the real exchange rate, and by stressing that the joint adjustment of those expectations can play a critical role in re-equilibrating exchange markets after exogenous shocks.¹⁸ In parallel with these developments in the conceptual literature, moreover, the corresponding issue of the joint sustainability of exchange rates and current account imbalances has re-emerged as a central consideration, both in policy discussions and in the formulation of exchange rate forecasts.

Thus, in attempting to characterize the elements of truth about exchange rates and international balances of payments, one finds that a certain symmetry has been established by the rejection of two extreme views. On the one hand, events during the 1970s and 1980s have discredited the classroom model of the 1960s, in which the exchange rate moved to equilibrate the current account (adjusted for changes in official reserves) within a single period; on the other hand, the re-emergence of the sustainability issue during the 1980s has re-established a strong presumption that exchange rate behavior is not completely independent of current or prospective developments in international balances of payments.

Complete Macroeconometric Models and the Treatment of Expectations

One of the changes in strategy that seems appropriate in the search for better ways to explain the behavior of exchange rates empirically is to move away from single-equation, semi-reduced-form approaches. The hope here is that models that simultaneously take account of a complete system of macroeconomic relationships will be able to improve on the single-equation, semi-reduced-form models in capturing the associations between exchange rates, interest rate differentials, and other variables.¹⁹ Undoubtedly, in studies such as Meese and Rogoff (1985), the failure to uncover a strongly significant, simple association between exchange rates and interest differentials in part reflects the fact that changes in different “exogenous” factors generate different types of covariation in exchange rates and interest differentials. An exogenous fiscal expansion, for example, may lead to an increase in domestic interest rates and to an appreciation of the domestic currency, whereas an “exogenous” decline in the demand for claims on the home-country government may lead to an increase in domestic interest rates and a depreciation of the domestic currency.

A second hope is that the ability of econometric models to explain exchange rate behavior may be improved through changes in the treatment of expectations. In reflecting on the poor out-of-sample performances of the models that have been developed to date, it has been emphasized that most of the empirical testing of these models has not been conducted in a “news” framework—or more precisely, in a framework within which changes in exchange rates could be explained in terms of revisions in expectations about, or unexpected outcomes for, the exogenous explanatory variables.²⁰ Moreover, on the basis of both a priori reasoning and anecdotal evidence, it has become widely perceived that much of the variability of exchange rates actually reflects responses to “news.” Thus one way to try to improve the ex post goodness-of-fit of exchange rate models would be to assemble data (or better data) on ex ante expectations about the variables that enter the models.

The prospect of assembling adequate time-series data on ex ante expectations, however, seems remote. Even so, the treatment of expectations can be improved by moving away from reliance on ad hoc assumptions, such as static expectations or perfect foresight, and toward hypotheses under which expectations are formed in ways that are consistent with the structural models or that are based on whatever information can be easily extracted from the time series of relevant variables.

In this context, it should be emphasized that reliance on the ad hoc assumptions of static expectations or perfect foresight could be an important part of the explanation for the poor empirical performances to date of portfolio-balance models of exchange rate determination, in which assets denominated in different currencies are perceived to be imperfect substitutes.²¹ A major source of interest in empirical estimates of portfolio-balance models has come from the policy issue of whether sterilized exchange market intervention—by changing the stocks of outside assets denominated in different currencies—can have a significant influence on the exchange rate.²² In addressing this issue it is important to distinguish between two channels of possible influence: through changes in the exchange risk premium or the deviation from UIP per se, and through changes in interest rates or revisions in expectations about future exchange rates as a result of changes in perceptions about the future levels of monetary policy variables or other relevant factors. It also seems crucial to recognize that the first channel may be impossible to isolate empirically unless the second channel is treated explicitly.

The basic point can be restated as a conjecture that the tests to date for systematic behavior of the exchange risk premium have modeled exchange rate expectations in a manner that may well be too inefficient to isolate the exchange risk premium. Specifically, almost all tests have relied either on the assumption that exchange rate expectations are static or on an assumption of perfect foresight, under which the expected future spot rate is represented as the realized value of the future spot rate plus an error term that behaves independently of any revisions in expectations about the future values of exogenous variables.²³ Reliance on the assumption of static expectations can be criticized for ignoring information about the structural model as well as possible inferences about the expected values of exogenous variables. Reliance on the assumption of perfect foresight can be criticized to the extent that no attempt is made to take account of the types of “news” that can be presumed to induce revisions in expectations about exogenous variables; the assumption of perfect foresight is difficult to reconcile with the notion that much of the variation in asset prices can be attributed to “news.” Implicitly, such approaches even assume that revisions in expectations about the future levels of asset stocks and wealth variables play no role in influencing the terms on which portfolio holders are assumed to remain willing to hold the existing stocks of assets under the existing distribution of wealth. Accordingly, that empirical work on portfolio-balance models has not found strongly significant evidence that the risk premium behaves systematically may in part reflect the reliance on inefficient approaches for capturing expectations about future exchange rates.²⁴

In recent years advances in computer solution algorithms have led to new developments in the treatment of expectations (see Taylor (1987)). In particular, several multicountry macroeconometric models have emerged in which expectations about exchange rates and interest rates are forward looking and consistent with the long-run solutions of the models.²⁵ In simulations of these models (and in some cases in the estimation as well), iterative techniques are employed to solve simultaneously for the current and expected future time paths of exchange rates and interest rates. Thus, to the extent that the solution paths for expected real exchange rates converge to long-run steady-state values, these models are able to rely on their own structures to capture the variability of expected long-run real exchange rates in response to shocks to the current or expected future values of exogenous variables.

In general, however, the ability of these forward-looking models to capture the variability of exchange rates may be limited by their menus of assets or their assumptions about portfolio behavior (as well as by their ability to measure changes in the expected future values of exogenous variables). In particular, the Liverpool model (Minford, Agenor, and Nowell (1984)), MINIMOD (Haas and Masson (1986)), and the Taylor (1987) model avoid distinguishing stocks of assets denominated in different currencies by imposing the UIP assumption. In contrast, the McKibbin-Sachs model (Sachs and McKibbin (1985)) avoids the UIP condition but may be limited (in the same way as traditional portfolio-balance models) by its assumption that exchange risk premiums vary only with changes in relative asset stocks and wealth variables. Thus, although the emergence of multicountry macroeconometric models with model-consistent expectations about exchange rates and interest rates appears to provide better ex post explanations of exchange rate behavior, it is debatable whether the existing examples of such models specify the menu of assets or the nature of portfolio preferences in a manner adequate or appropriate to capture the influences of portfolio behavior on exchange rates. This issue is the subject of the next subsection.

Modeling Portfolio Preferences in a General Equilibrium Framework

This subsection discusses two central but unresolved issues that are confronted in the specification of exchange rate models. The first is whether it is adequate or appropriate to treat assets as perfect substitutes or, equivalently, to employ the UIP hypothesis—an assumption that greatly simplifies the analysis by making it possible to describe international portfolio positions simply in terms of the net foreign asset position or, equivalently, the cumulative current account imbalance. To the extent that the UIP hypothesis is not regarded as adequate or appropriate, the second issue is how to specify a menu of assets and a basis for portfolio preferences that has solid microeconomic foundations within a general equilibrium framework.

In an important sense, the adequacy or appropriateness of relying on the UIP hypothesis can only be judged in terms of the particular objectives of each individual model. The major benefits that derive from the UIP hypothesis are that it extends considerably the set of forward-looking models for which it is possible to obtain analytic descriptions of exchange rate dynamics, and that it also appears to simplify considerably the process of computing an iterative solution to multicountry macroeconometric models with forward-looking expectations.

The other side of the issue, however, is whether reliance on the UIP hypothesis has major costs. One approach to evaluating those costs is to start with empirical tests of the UIP hypothesis, which have been providing new and mounting evidence that rejects the hypothesis. Moreover, some of the evidence has suggested that variations over time in the deviations from UIP may explain a major share of the observed variations in exchange rates (recall the discussion in Section I, under “Exchange Rates and Interest Rates”).

Another approach to evaluating the costs of relying on the UIP hypothesis is to consider whether there are serious weaknesses in the economic foundations of models in which international portfolio behavior is described simply in terms of net foreign asset positions, with no significance attached to the gross claims and gross liability positions that are associated with a given net foreign asset position. A basic consideration here is how rational investors should think about the sustainability of international portfolio positions or, more precisely, about the prospect that a country will fail to comply fully with the terms on its outstanding gross liabilities. In particular, the basic consideration in evaluating the prospect that unanticipated shocks may lead a country not to comply fully with the terms on its outstanding gross liabilities is whether the country’s ability and willingness to meet those terms depends on the stock of its gross liabilities per se (relative to such scale variables as its macroeconomic product and foreign exchange earnings) or simply on its net foreign asset or liability position. Certainly in the context of today’s debt-burdened developing countries, the gross foreign asset positions of the residents of the countries do not appear to be generating streams of income that are available for making payments on the gross external liabilities of those countries. This observation challenges the notion that rational portfolio managers should be indifferent to the magnitudes of the gross claims and liabilities that are associated with a given net foreign asset position.

To the extent that efforts are devoted to modeling exchange rates in general equilibrium frameworks that do not rely on the UIP assumption, how does one specify a menu of assets and a basis for portfolio preferences that has solid microeconomic grounding?²⁶ An important issue that arises here is whether it is satisfactory to rely on the types of optimizing models that have been developed in recent years, in which financial assets consist of balances of particular currencies that either enter the utility function as inputs to the production of consumption services in the corresponding countries (see Stulz (1984)) or are required to be obtained and delivered in advance of purchasing goods and services in the corresponding countries (see Obstfeld and Stockman and references therein (1985, pp. 964-72)). My own opinion is that these approaches are not satisfactory. No insight is provided by simply assuming that currency holdings enter the utility function directly; and the fact that different currency units are used for transaction purposes in different countries does not seem to be of much significance to investors in managing their international portfolios, particularly when many countries denominate large proportions of their international liabilities in foreign currency units.

An alternative approach is to supplement distinctions between the currency denominations of assets with an emphasis on the notion that the portfolio preferences of rational investors should depend fundamentally on differences in the prospective returns on capital in the countries obligated to make payments to the asset holders.²⁷ Thus investors should base their portfolio decisions on their perceptions of the extent to which income streams on assets located in different countries, or held as claims against the residents of different countries, are subject to different macroeconomic outlooks, political environments, or prospects of taxation. Under this approach, the currency preferences and country preferences of asset holders are determined simultaneously, and market-clearing exchange rates depend both on expectations (and uncertainties) about exogenous factors influencing the production of goods and the after-tax distribution of income in different countries, and on expectations (and uncertainties) about monetary policies and other factors influencing the absolute levels of prices in different countries.

One appeal of such an approach is its generality, which provides scope for exploring several allegedly important influences on exchange rates that have not yet received adequate empirical attention. With respect, in particular, to the behavior of dollar exchange rates and international payments imbalances during the 1980s, the approach provides scope for modeling the interrelated influences of the change in U.S. tax laws during 1981,²⁸ of the subsequent expansion of U.S. fiscal budget deficits,²⁹ of the relatively rapid expansion of the U.S. economy during 1983-84,³⁰ and of the passage in the United States of the Gramm-Rudman-Hollings bill in 1985.³¹ In addition, the approach provides scope for recognizing a type of “safe-haven” phenomenon whereby exchange rates between the currencies of industrial countries may have been affected during the 1980s by the implications of the international debt crisis for the stocks of net claims that investors desired to hold against the developing countries.³²

Of course some of these different influences on exchange rates, such as the effects of changes in fiscal budget deficits, can be analyzed without distinguishing assets either by currency or by country. Nevertheless, models in which assets are distinguished by country as well as by currency may provide new insights on the roles for monetary and fiscal policies as traditionally modeled, in addition to providing a framework for analyzing the effects of country-specific tax changes, regulatory policies, and macroeconomic developments.