Professional Documents
Culture Documents
610
69 NO- 4 FRANKEL FLOATING EXCHANGE RATES 61}
which is proportional to the real interest considered to be the forward discount, defined
differential, that is, the nominal interest as the log of the forward rate minus the log of
differential minus the expected inflation the current spot rate, then (1) is a statement
differential. If the nominal interest differen- of covered (or closed) interest parity. Under
tial is high because money is tight, then the perfect capital mobility, that is, in the absence
exchange rate lies below its equilibrium value- of capital controls and transactions costs,
But if the nominal interest differential is high covered interest parity must hold exactly,
merely because of a high expected inflation sinee its failure would imply unexploited
differential, then the exchange rate is equal to opportunities for certain profits." However, d
its equilibrium value, which over time will be defined as the expected rate of depre-
increases at the rate of the inflation differen- ciation; then (1) represents the stronger
tial. condition of uncovered (or open) interest
The theory yields an equation of exchange parity. Of course if there is no uncertainty, as
rate determination in whieh the spot rate is in a perfect foresight economy, then the
expressed as a function of the relative money forward discount is equal to the expected rate
supply, relative income level, the nominal of depreciation, and (1) follows directly. If
interest differential (with the sign hypothe- there is uncertainty and market participants
sized negative), and the expected long-run are risk averse, then the assumption that there
inflation differential (with the sign hypothe- is no risk premium, though not precluded, is a
sized positive). The hypothesis is readily strong one.^
tested, using the mark/dollar rate, against the The second fundamental assumption is that
two alternative hypotheses: the Chicago the- the expected rate of depreciation is a function
ory, which implies a positive coefficient on the of the gap between the current spot rate and
nominal interest differential, and the Keynes- an equilibrium rate, and of the expected
ian theory, which implies a zero coefficient on long-run inflation differential between the
the expected long-run inflation differential. domestic and foreign countries:
(2) = -6{e -e) + TT - TT*
1. The Real Interest Differential Theory of
Exchange Rate Determination where e is the log of the spot rate; -K and IT*
are the current rates of expected long-run
The theory starts with two fundamental inflation at home and abroad, respectively.
assumptions. The first, interest rate parity, is (We can think of them as long-run rates of
associated with efficient markets in which the monetary growth that are known to the
bonds of different countries are perfect substi- public.)*' The log of the equilibrium exchange
tutes: rate e is defined to increase at the rate 7r - ir*
in the absence of new disturbances; a more
(1) d = r - r'
precise explanation will be given below. Equa-
where r is defined as the log of one plus the tion (2) says that in the short run the
domestic rate of interest (which is numeri- exchange rate is expected to return to its
cally very elose to the actual rate of interest
for normal values) and r* is defined as the log *For evidence that the deviations from covered interest
of one plus the foreign rate of interest.^ If d is parity are smaller than transactions costs, see Frenkel
and Richard Levich.
'My paper shows the conditions under which, despite
'The rates of interest referred to are instantaneous risk aversion, there is a zero risk premium in the forward
rates per unit of time, i.e., the forces of interest. In the exchange rate. Briefly, the conditions are no outside
absence of uncertainty {or, in the stochastic case, if the nominal assets and no correlation between the values of
term structures of the interest differential and the currencies and the values of real assets.
forward discount contain no risk premium), equation (1) *If there is no long-run growth of real income or
should also hold when the rates are deJined over any technical change in the demand for money, then the rates
finite interval r, since the left-hand and right-hand sides of monetary growth are T and ir*. If there is long-run real
would be equal to the expected future values of the growth or technical change, then we would adjust the
left-hand side and right-hand side, respectively, of (1) monetary growth rates before arriving at the expected
integrated from 0 to r. long-run inflation rates ir and v'.
612 THE AMERICAN ECONOMIC REVIEW SEPTEMBER 1979
equilibrium value at a rate which is propor- to rise above its long-run level, an incipient
tional to the current gap, and that in the long capital inflow causes the value of the currency
run, when e = ^, it is expected to change at the to rise proportionately above its equilibrium
long-run rate -rr - x*. For the present, the level.
justification for equation (2) will be simply For a complete equation of exchange rate
that it is a reasonable form for expectations to determination, it remains only to explain e.
take in an inflationary world. This claim will Assume that in the long run, purchasing
be substantiated in Appendix A, after a power parity holds:
price-adjustment equation has been specified
(4) e =p -p*
by a demonstration that (2), with a specific
value implied for 6, follows from the assump- where p and p* are defined as the logs of the
tions of perfect foresight (or rational expecta- equilibrium price levels at home and abroad,
tions in the stochastic case) and stability.^ respectively.'*^
The rational value of 6 will be seen to be Assume also a conventional money demand
closeiy related to the speed of adjustment in equation:
the goods market.
(5) m = p + <t>y - \r
Combining equations (1) and (2) gives
where m, p, and y are defined as the logs of
(3) e-e--^-[{r-T)^ (r* - x*)] the domestic money supply, price level, and
output. A similar equation holds abroad. Let
We might describe the expression in brackets us take the difference between the two equa-
as the real interest differential.** Alterna- tions:
tively, note that in the long run when e = e,vve
must have ? - r* = IT - ir*, where r and r* (6) m - m* = p - p* +
denote the long-run, short-term interest - \{r r*)
rates.^ Thus the expression in brackets is
equal to [{r - r*) - if - ?*)], and the Using bars to denote equilibrium values and
equation can be described intuitively as remembering that in the long run, when e = e.
follows. When a tight domestic monetary r - r* = IT - T*, we obtain
poiicy causes the nominal interest differential
(7) e-p^p*
^m -In* - ii>(y - y*) + \{-K - w*)
'Note, however, that the assumption of rational expec-
tations or perfect foresight is not required for equation This equation illustrates the monetary theory
(2) and thus is nol required for the model, of the exchange rate, according to which the
"This would nol be quite right because the nominal exchange rate is determined by the relative
interest rates are short term while the expected inflation
rates are long term. However. It does turn out, with a supply of and demand for the two currencies.
price-adjustment equation such as is adopted in Appen- It says that in full equilibrium a given
dix A, that e - e is proportional to the shorl-term real increase in the money supply inflates prices
interest differential [{r - Dp) - {r* - Dp*)\- and thus raises the exchange rate proportion-
'In words, long-run equality between the nominal ately, and that an increase in income or a fall
interest differential and the expected inflation differen-
tial follows from interest rate parity (equality between in the expected rate of inflation raises the
the interest differential and expected depreciation) and demand for money and thus lowers the
long-run relative purchasing power parity (equality exchange rate.
between depreciation and the inflation differential). An
alternative argument is that in the long run international
investment flows insure that real interest rates are equal '"This assumption rules out tbe possibility of perma-
across countries: J — ir = r* - w*. The investment nent shifts in the terms of trade and so gives essentially a
argument is not necessary, however, nor, if used, does it one-good model. It does allow the possibility of temporary
preclude the possibility of different real rates of interest shifts; the existence of large departures from the Law of
in the short run, since even the most perfectly classical of One Price for international trade even in closely matched
economies have fixed capital stocks that earn nonzero categories of goods has been shown in studies by Peter
profits in the short run. Isard and by Irving Kravis and Robert Lipsey.
VOL. 69 NO. 4 FRANKEL: FLOATING EXCHANGE RATES 613
Substituting (7) into (3), and assuming (5): m - p = <}>y - \r. Subtracting the
that the current equilibrium money supplies foreign version yields a relative money
and income levels are given by their current demand equation like (6): {m - m*) -
actual levels." we obtain a complete equation (p - p*) = (fiiy - y*) - X(r - r*). Bilson
of spot rate determination: then assumes that purchasing power parity
always holds:
(8) e = m - m* - <t>{y - y*)
(10) e = p - p* = {m - m*)
- y*) + Kr ~ r*)
This is the equation which is tested empiri- An increase in the domestic interest rate
cally for the deutsche mark in Section III. lowers the demand for domestic currency and
causes a depreciation. In terms of equation
II. Testable Alternative Hypotheses (9), a, the coefficient of the nominal interest
differential, is hypothesized to be positive
Equation (8) is reproduced here with an rather than negative.
error term:'^ The interest differential (r - r*) is viewed
as representing the relative expected inflation
(9) e = m - m* - (ftiy - y*) rate (TT - x*), either because international
+ a{r - /•*) + /3(7r - TT*) +u investment flows equate real rates of interest,
or because interest rate parity insures that the
where a (= —1/9) is hypothesized negative interest difl'erential equals expected deprecia-
and 13 (= 1/6 + \) is hypothesized positive tion, and purchasing power parity insures that
and greater than a in absolute value. Tests of depreciation equals relative inflation. Thus
a hypothesis are always more interesting if a the expected inflation difl"erential (were it
plausible alternative hypothesis is specified. directly observable) could be put into (10)
One obvious alternative hypothesis is Dorn- instead of the nominal interest difl'erential:
busch's incarnation of the Keynesian ap-
proach, in which secular inflation is not a (11)
factor. In fact the model developed in this
7r*)
paper is the same as the Dornbusch model in
the special case where 5r - TT* always equals In terms of equation (9), a is hypothesized to
zero.'^ The testable hypothesis is ^ = 0. be zero and /3 to be positive, if we use a good
Another—more conflicting—alternative proxy for (ir - 7r*). Or, more generally, the
hypothesis comes from the Chicago theory of hypothesis can be represented a -i- /3 = \ > 0,
the exchange rate attributable to Frenke! and a > 0. /i > 0. The relative size of a and )3
Bilson. The variant presented by Bilson would depend on how good a proxy we have
begins with a money demand equation like for the expected inflation difl'erential.
Indeed Frenkel begins his analysis with the
"Actual money supplies and actual income levels can assumption of a Cagan-type money demand
both easily be allowed to differ from their equilibrium function, which uses the expected inflation
levels, but these extensions of the analysis are omitted rate rather than the interest rate:
here in the interest of brevity.
'^It is not stated here where the errors come from.
(12) m - p = <t>y - Xw
Probably the most likely source of errors is the money
demand equation (5), which would bias the coefficient of
(m - m*) downward if it is not constrained to be one. The assumption of purchasing power parity
This possibility is discussed below. then gives equation (11) directly. Frenkel
'^See Dornbusch (1976c). The only other differences uses the expected rate of depreciation as
between the present model and the Dornbusch model are
that the former is a two-country model, while the latter is reflected in the forward discount in place of
a small-country model, and the latter uses a slightly the unobservable expected inflation difl'eren-
different price-adjustment equation. tial which, in well-functioning bond markets,
614 •M/C REVIEW SEPTEMBER 1979
'*In view of the result, (A4) in the Appendix, that the "Germany is viewed as the domestic country in the
purchasing power parity gap is proportional to the real econometric equations.
interest differential, it is not surprising that a theory '*The equality of current long-term real rates of
which assumes that the former is always zero should also interest follows from the result that in the long run the
assume that the latter is always zero. short-term real rates are equal, and the requirement of a
VOL. 69 NO. 4 FRANKEL: FLOATING EXCHANGE RATES 615
Number of
Technique Constant R- O.W. Observations
of the long-term interest differential is that it zero. This result is all the more striking when
is capable of reflecting instantly the impact of it is kept in mind that the null hypothesis of a
new information such as the announcement of zero or positive coefficient is a plausible and
monetary growth targets. The long-term seriously maintained hypothesis; the Chicago
government bond rate differential is the proxy (Frenkel-Bilson) hypothesis is rejected in this
used in the reported regressions, though other data sample. The coefficient on the expected
proxies are used as instrumental variables. long-run inflation differentia! is significantly
Details on the data are given in greater than zero. Thus the unmodified
Appendix B. Keynesian (Dornbusch) hypothesis is also
In each regression the signs of all coeffi- rejected. Furthermore, as predicted by the
cients are as hypothesized under the real real interest differential model the coefficient
interest differential model. When the single on the expected long-run inflation differential
equation estimation techniques are used, the is significantly greater than the absolute value
significance levels are weak, especially when of the coefficient on the nominal interest
iterated Cochrane-Orcutt is used to correct differential.
for high first-order autocorrelation. Several other points are also notably
But when instrumental variables are used supportive of the theory. {I concentrate on the
to correct for the shortcomings of the last regression in Table 1.) The coefficient of
expected inflation proxy, the results improve the relative money supply is not only signifi-
markedly. The coefficient on the nominal cantly positive, but is also insignificantly less
interest differential is significantly less than than 1.0. The coefficient of relative pro-
duction is significantly negative, and its point
estimate of approximately —.5 suits well its
rational term structure that the long-term real interest
differential be the average of the expected short-term real
interpretation as the elastieity of money
interest differentials. demand with respect to income. The sum of
616 THE AMERICAN ECONOMIC REVIEW SEPTEMBER 1979
/\
\fitted
\
the (negative) coefficient on the nominal As a final indication of the support Table 1
interest differential and the coefficient on the provides for the real interest differential
expected inflation differential is an estimate hypothesis, the Rh are high. Figure 1 shows a
of the semielasticity of money demand with plot of the equation's predicted values and the
respect to the interest rate; when converted to actual exchange rate values. The equation
a per annum basis, the estimate is 6.0, which tracks the mark's 1974 appreciation, 1975
provides another favorable cross-check.'^ depreciation, and 1976-77 appreciation."
The point estimate of a is - 5 . 4 . This To apply the estimated equation, let us
implies that when a disturbance creates a convert it to the form:
deviation from purchasing power parity,
(1 - 1/5.4 =) 81-5 percent of the deviation is - 1.39
expected to remain after one quarter, and - 1.35(r- 7.35(7r
{.^\5'^ =) 44.1 percent is expected to remain
after one year. The estimate of ^ on a per where a and /3 have been divided by four for
annum basis is {-log .441 =) .819. Previous use with per annum interest rates and the
work on the speed of adjustment to purchas- coefficient on the relative money supply has
ing power parity is less definitive than esti- been set to 1.0. The expression can be decom-
mates of money demand elasticities, but the posed into the equilibrium exchange rate
present estimates of the expected speed of
adjustment appear reasonable."* e = 1-39 + {m - m*) - .52iy - y*)
+ 6.00(7r - IT*)
"The semielasticity estimate and an average interest
rate of around 6 percent imply an interest elasticity of
around (6,0 x .06 =) .36, which is in the range of "The equation fails to track the continued sharp
estimates of the long-run elasticity made by Stephen depreciation of the dollar in January and February of
Goldfeld and others. 1978. The regressions that were reported in earlier
'*Hans Genberg estimates for Germany that 37 versions of this paper did not include this period, and
percent of an initial divergence from purchasing power consequently appeared more favorable to the real interest
parity disappears after one year. differential model.
VOL. 69 NO. 4 FRANKEL: FLOATING EXCHANGE RATES 617
and the size of the overshooting goods are cheaper than German goods; higher
e -e= - 1 . 3 5 [ ( r - 7r) - (r* - TT*)] demand will gradually drive up American
prices faster than the rate of monetary
As an illustration, let us conduct the hypo- growth, which in turn will drive up U.S.
thetical experiment of an unexpected I nominal interest rates, reduce the overshoot-
percent expansion in the U.S. relative money ing, and cause the spot rate to rise back
supply. If the monetary expansion is consid- towards its new equilibrium. After a year,
ered a once-and-for-all change, then the equi- approximately 44 percent of the initial real
librium mark/dollar rate decreases by 1.0 interest differential and purchasing power
percent. But in the short run the expansion parity deviation will have been closed. In the
also has liquidity effects; the interest semi- meantime, there should be an expansionary
elasticity of 6.00 implies a fall in the nominal effect on demand for U.S. output; lower real
interest rate of (1 percent/6.00 =) 17 basis U.S. prices will stimulate net exports and
points.^" This fall in the real interest differen- lower real U.S. interest rates will stimulate
tial induces an incipient capital outflow whicb investment. However, any efiTects on output
in turn causes the currency to depreciate have not been modelled in this paper.^^
further, until it overshoots its new equilibrium
by (1.35 X .17 percent =) .23 percent. The
IV. Econometric Extensions
total initial depreciation is 1.23 percent.
This calculation assumes no change in the It is possible that adjustment in capital
expected inflation rate. If tbe monetary markets to changes in the interest differential
expansion signals a new higher target for is not instantaneous, and that lagged interest
monetary growth, then the effect could be differentials should be included in the regres-
much greater.'' Suppose the annualized 12 sions. Formally, we could argue that due to
percent increase raises the expected inflation transactions costs, tbe forward discount
rate by, say, 1 percent per annum. Then there adjusts fully to the interest differential with a
will be an additional depreciation of 6.00 one-month lag:
percent on account of the lower demand for
money in long-run equilibrium plus 1.35 (13) d=h{r-r*) -H (1 - h ) { r - r * ) _ ,
percent more overshooting on account of the When (13) is used in place of (1), the spot
further reduced real interest differential. rate equation (8) is replaced by
Thus the total initial depreciation would be
8.58 percent, of whicb 7.00 percent represents (14) e = {m-m*)
long-run equilibrium and 1.58 percent repre- - {h/d){r - r*) - {]
sents short-run overshooting.
After the initial effects, the system moves
toward the new equilibrium as described in The results of regressions with a lagged inter-
Appendix A. provided capital is perfectly est differential are reported in Table 2. The
mobile and future money supplies do not coefficient on the lagged interest differential
deviate from their expected values. American is insignificantly less than zero. This evidence
supports tbe idea that capital is perfectly
mobile.
^Here we are assuming (for the first time) that the
estimated long-run interest semietasticity holds in the There are several reasons why one might
short run as well. If money demand is subject to lagged wish to constrain the coefficient on the rela-
adjustment, then the short-run effect of a monetary tive money supply to be 1.0 in these regres-
contraction on the interest rate and hence on the sions, in effect moving the relative money
exchange rate would be greater than that calculated here.
However the theory and econometrics behind equation
(8) would be completely unaffected. "The assumption of exogenous output can be relaxed
"On ihe other hand, if some of the expansion is by assuming that output is demand determined. Then a
expected lo be reversed in the following month, then the necessary condition for overshooting is that the elasticity
decrease in the equilibrium rate would be correspond- of demand with respect lo relative prices is less than 1.0.
ingly smaller. See the Appendix to Dornbusch {1976c).
6/5 THE AMERICAN ECONOMIC REVIEW SEPTEMBER 1979
Number of
Technique Constant m y - y r* {r - r»)_, -IT - TT' D.W. Observations
Note. Standard errors are shown in parentheses. Dependent Variable: (log of) Mark/Dollar Rate, See Table 1 for
definitions.
supply variable to the left-band side of the constraining the coefficient is to remove the
equation. First, our a priori faith in a unit simultaneity problem which otherwise occurs
coefficient is very higb. It is hard to believe if central banks vary their money supplies in
that the system could fail in the long run to be response to the exchange rate. The argument
homogeneous of degree zero in the exchange even extends to direct exchange market inter-
rate and relative money supply. Second, vention, which has been prevalent under
errors in the money demand equation are managed floating. The right-hand side varia-
known to have been large over tbe last few bles determine relative money demand;
years. Sucb errors, since tbey are correlated changes in money demand can be reflected in
with the money stock variable, would bias the either money supplies (the monetary ap-
coefficient downward; indeed, the coefficient proach to the balance of payments) or the
estimate in one regression in Table I appears exchange rate (the monetary approach to the
significantly less than 1.0. By constraining the exchange rate), depending on government
money supply coefficient to be 1.0, we make intervention policy.
sure that any possible errors in the money Table 3 reports the constrained regressions.
demand equations will go into the dependent The results are very similar to those in Table
variable, that is, will be uncorrelated with any 1. The /?^s indicate that over 90 percent of the
of the independent variables; thus they cannot variation in the dependent variable is
bias the coefficients on the interest and explained; the remaining 10 percent could be
expected inflation differentials, which are our attributed to errors in the two countries'
primary objects of concern. A third reason for money demand equations.
Noie\ Standard errors are shown in parentheses. Dependent Variable: (log of) Mark/Dol-
lar Rale H- German M,/U.S. M^. See Table 1 for definitions.
VOL. 69 NO. 4 FRANKEL FLOATING EXCHANGE RATES 619
, (1976b) "Foreign Exchange Market change Rates," / . Int. Econ., Aug. 1975, 5,
Speculation and Stabilization Policy Under 275-81.
Flexible Exchange Rates," paper presented M. V. N. Whitman, "Global Monetarism and
at Conference on the Political Economy of the Monetary Approach to the Balance of
Inflation and Unemployment in Open Payments," Brookings Papers, Washing-
Economies, Athens, Oct. 1976. ton 1975, i, 491-556.
I. Kravis and R. Lipsey, "Export Prices and the Board of Governors of the Federal Reserve System,
Transmission of Inflation," Amer. Econ. Statistical Releases, Washington, various
Rev. Proc, Feb. 1977, 67, 155-63. issues.
Robert Mundell, "Exchange Rate Margins and , Fed. Res. Bull., Washington, various
Economic Policy," in J. Carter Murphy, issues.
ed.. Money in the International Order, Business Week, various issues.
Dallas 1964. Deutsche Bundeshank, Statistical Supplements
, International Economics, New York to Monthly Reports, Bonn, various issues.
1968. International Monetary Fund, International Fi-
M. Mtissa, "The Exchange Rate, the Balance nancial Statistics, Washington, various
of Payments and Monetary and Fiscal issues.
Policy Under a Regime of Controlled Morgan Guaranty and Trust Co., World Finan-
Floating," Scand. J. Econ., June 1976, 78, cial Markets, New York, various issues.
229^8. U.S. Congress, Joint Economy Committee,
, "Real and Monetary Factors in a Economic Indicators, Washington, various
Dynamic Theory of Foreign Exchange," issues.
mimeo., Univ. Chicago, Apr. 1977. U.S. Council of Economic Advisors, Economic
J. Niehans, "Some Doubts About the Efficacy Report of the President, Washington, vari-
of Monetary Policy Under Flexible Ex- ous issues.