View original document

The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.

Vol. 3, No. 6
JUNE 2008­­

Insights from the

F e d e r a l Rese r v e B a n k o f D a l l a s

Why Are Exchange Rates So Difficult to Predict?
by Jian Wang

A quarter-century

The U.S. dollar has been losing value against several major currencies this decade.

quest hasn’t found

Since 2001– 02, the U.S. currency has fallen about 50 percent against the euro, 40 percent

the elusive links

against the Canadian dollar and 30 percent against the British pound (Chart 1).

between economic

These steep, prolonged depreciations have brought a new urgency to understand-

fundamentals and

ing the factors that move exchange rates. Some way of forecasting them would allow

currency values.

businesses, investors and others to make better, more-informed decisions. Unfortunately,
exchange rates are very difficult, if not impossible, to predict — at least over short to
medium time horizons.
Economic differences between countries — in such areas as national income, money
growth, inflation and trade balances — have long been considered critical determinants
of currency values.1 However, there’s no definitive evidence that any economic variable
can forecast exchange rates for currencies of nations with similar inflation rates.

Chart 1

Dollar Drops Against Key Foreign Currencies
Index, Jan. 4, 1999 = 100


British pound
Canadian dollar











NOTES: Exchange rates are units of foreign currency per dollar.
SOURCE: Federal Reserve Board of Governors.

A “random walk”
model is just as good at
predicting exchange
rates as models based
on fundamentals.

Economists continue to seek the
keys to predicting currency values.
Some recent research supports the
idea that exchange rates behave like
financial assets, whose price movements are primarily driven by changes
in expectations about future economic
fundamentals, rather than by changes
in current ones. These studies suggest
that the real contribution of standard
exchange rate models may not lie in
their ability to forecast currency values.
Instead, the models imply predictability runs in the opposite direction:
Exchange rates can help forecast economic fundamentals.
The Disconnect Puzzle
Supply and demand hold sway
on currency exchanges, just as they
do in most other markets. Exchange
rates ebb and flow depending on the
shifting needs of the individuals, firms
and governments that buy foreign
goods and services, invest abroad,
and seek profit or protection through
The fundamentals that economists
link to exchange rates shape the forces

EconomicLetter 2

F edera l Reserve Bank of Dall as

of supply and demand. It seems logical, for example, that countries with
large trade deficits would see their
currencies decline, and countries with
strong growth and low inflation would
see their currencies rise in value.
But what may be logical in theory
hasn’t been easy to prove. In 1983,
Richard Meese and Kenneth Rogoff
challenged the long-held idea that economic fundamentals determine relative
currency values.2 The economists compared existing models to an alternative
in which fundamentals are excluded
and any exchange rate changes are
purely random. They found a “random
walk” model just as good at predicting
exchange rates as models based on
In short, their findings suggest
economic fundamentals — like trade
balances, money supply, national
income and other key variables —are
of little use in forecasting exchange
rates between countries with roughly
similar inflation rates.3 This result has
been labeled the “exchange rate disconnect puzzle.”
This disconnect can be illustrated
by comparing the one-month forward
exchange rate forecast and a random
walk, using data for the U.S. dollar
against the euro from March 2003 to
January 2008 (Chart 2). There’s no evidence that the forward rate follows the
spot rate more closely than a random
walk. The results are similar for three
other currencies —the British pound,
the Canadian dollar and the Japanese
yen. Indeed, for the euro, pound and
yen, the random walk has smaller prediction errors than forward rates (Table
Economists have offered several
reasons for the inability to find clear
links between exchange rates and
economic fundamentals, starting with
the inherent limitations of economic
models. A typical model relies on coefficients that specify the relationship
between exchange rates and fundamentals. Estimates of these parameters
are based on historical data, but their
predictive power stems from their abil-

Chart 2

Dollar, Euro Take a Random Walk
Log of dollars per euro

Economists have

Spot rate
Forward rate
Random walk


offered several


reasons for the inability


to find clear links


between exchange


rates and economic







NOTE: Spot and forward exchange rates are logarithms of spot and one-month forward exchange rates from March 2003 to
January 2008.
SOURCE: Financial Times data compiled by Haver Analytics.

Table 1

Forward Rate Forecast vs. Random Walk
(Mean Squared Prediction Errors)
British pound

Canadian dollar

Japanese yen

Forward rate






Random walk










NOTES: Ratios are calculated using the mean squared prediction errors of forward exchange rates and the random walk. If
the ratio is greater than 100 percent, the random walk is more accurate than forward rates in predicting exchange rates.

ity to determine currency values from
new data or projections (see box on
page 4).
In their disconnect-puzzle paper,
Meese and Rogoff conjecture that these
parameters may vary over time. They
note that monetary and other policies
in many countries have been in flux
since the early 1970s, when the fixed
exchange rate regimes of the Bretton
Woods system collapsed.
Model misspecification could
also be a factor in the exchange rate
disconnect puzzle.5 If the coefficient

values are skewed from their true values, forecasts based on these “wrong”
parameters can be more off base than
those generated by a random walk.
In addition to being difficult to
forecast, exchange rates are far more
volatile than the economic fundamentals that supposedly determine them.
Over a 30-year period, for example,
swings in the exchange rate between
the U.S. dollar and the British pound
have been far wider than the countries’ differences in output and inflation (Chart 3). The high volatility of

F ederal Reserve Bank of Dall as	

3 EconomicLetter

Modeling Exchange Rates
How do economists model exchange rates and economic fundamentals? Actual models
may be complex, but a relatively simple one might assume that the exchange rate at time
t (st ) is a linear function of some economic fundamental (ƒt ) and some error term (εt ):

st = α + βƒt + εt.
We have data for the exchange rate (st ) and the economic fundamental (ƒt ) up to time t. The
true values of coefficients α and β are unknown, but we can use historical data to estimate
^ and β
them. Let’s denote these estimates with α
If we have a forecast of the fundamental at time t + 1 (ƒ^t+1 ), we can project the exchange
rate at time t +1 (st+1 ):

With the longer horizons,
fundamentals can

^ƒ^ .
^st+1 = α

outperform a random

If α and β are constant over time and they capture the true relationship between exchange
rates and fundamentals, the model will predict future currency values. If the parameters
vary from time to time, or if the parameter estimates are seriously biased, the model may
yield incorrect results.

walk at forecasting
long-term changes
in exchange rates.

Chart 3

Exchange Rates Vary More Than Key Fundamentals


Exchange rate


Relative real GDP
Inflation differential


















NOTES: The chart shows percentage changes in the nominal exchange rate (the pound relative to the U.S. dollar), U.K.
GDP per capita relative to U.S. GDP per capita, and the inflation differential between the U.K. and U.S.
SOURCE: Haver Analytics.

exchange rates relative to economic
fundamentals is very difficult to replicate in a model without introducing
arbitrary disturbances.
The fact that standard, fundamentals-based models can’t outperform

EconomicLetter 4

F edera l Reserve Bank of Dall as

a random walk casts serious doubt
on their ability to explain exchange
rate fluctuations. The random walk
itself does a mediocre job predicting
exchange rate movements. It’s surprising we can’t find some economic

variables to help us beat this poor

improve predictability in standard
exchange rate models.7 Models
that rely on the money supply, real
incomes and other fundamentals do a
better job tracking the dollar’s movements against the deutsche mark over
eight or 12 quarters than one quarter
(Chart 4).
With the longer horizons, fundamentals can outperform a random
walk at forecasting long-term changes
in exchange rates. The practical value
of these results is limited, however,
because of the short-term nature of
many decisions affected by currency
Introducing the possibility of monetary policy feedback can also improve

Beyond the Disconnect
Researchers have been probing
the relationship between economic
fundamentals and exchange rates since
Meese and Rogoff first posed the disconnect puzzle.
Various combinations of economic
variables and econometric methods
have been tried in an attempt to predict exchange rates.6 These models
haven’t wholly disproved the idea of a
disconnect, but they’ve found evidence
that economic fundamentals matter— at
least under some conditions.
For example, longer time frames

Introducing the
possibility of
monetary policy
feedback can also
improve predictability.

Chart 4

Exchange Rate Predictability Improves with Time Horizon
One-Quarter Changes
















Actual changes
Model predictions











Twelve-Quarter Changes


Actual changes
Model predictions


Actual changes
Model predictions


Eight-Quarter Changes

Four-Quarter Changes

Actual changes
Model predictions


























NOTES: Actual changes are in the dollar–deutsche mark exchange rate. Model predictions are from long-horizon regressions:
st+k – st = αk + βk(ƒt–st ) + εt+k,k, where fundamental ƒt includes money supplies and real incomes in the U.S. and Germany.
SOURCE: “Exchange Rates and Fundamentals: Evidence on Long-Horizon Predictability,” by Nelson C. Mark, American Economic Review, vol. 85, March 1995, pp. 201–18.

F ederal Reserve Bank of Dall as	

5 EconomicLetter



There are
reasons economic
fundamentals aren’t
very helpful in
forecasting exchange
rates, even if currrency
values are determined by
these fundamentals.

predictability. Some central banks take
exchange rates into account when setting short-term interest rates. Models
that incorporate this feedback from
currency values to interest rates can
replicate exchange rate data relatively well. One of them looks at the
deutsche mark–dollar exchange rate
from 1979 through 1998 (Chart 5).8
Evidence suggests that models
incorporating central bank actions can
beat a random walk in forecasting
exchange rates.9
By extending time horizons and
introducing central bank actions,
researchers have shown links between
economic fundamentals and exchange
rates. They haven’t, however, overturned the pivotal Meese and Rogoff
finding that economic fundamentals
can’t predict exchange rates where
it really counts—in the short term.
A recent, comprehensive study concluded, “No model consistently outperforms a random walk…. Overall,
model/specification/currency combinations that work well in one period will

not necessarily work well in another
Given the empirical results,
should we decide that exchange rates
are not determined by economic fundamentals? Probably not. Admittedly,
currency values are difficult to predict,
and economic fundamentals offer
little help. But that doesn’t necessarily
mean that exchange rates are mainly
driven by irrational noise. There are
reasons economic fundamentals aren’t
very helpful in forecasting exchange
rates, even if currency values are
determined by these fundamentals.
Currencies as Assets
Expectations are at the heart of
recent explorations of the exchange
rate disconnect.
Economists incorporate expectations into an economic model using
an asset-pricing approach. In such
models, current data receive far less
weight than future factors in determining prices for long-lasting financial
assets. For instance, recent dividends

Chart 5

Taking Central Bank Policies into Account
Index of log real exchange rate (deutsche mark/U.S. dollar, October 1979 = 100)
Actual exchange rate


Exchange rate in model











NOTES: Exchange rates in the model are less volatile than actual exchange rates and have been scaled to have the same
mean and standard deviation as the actual exchange rate.
SOURCE: “Taylor Rules and the Deutschmark–Dollar Real Exchange Rate,” by Charles Engel and Kenneth West, Journal
of Money, Credit and Banking, vol. 38, August 2006, pp. 1175–94.

EconomicLetter 6

F edera l Reserve Bank of Dall as

are a minor consideration when people buy stocks. More weight is given
to expected future dividends and capital gains.
In asset-pricing exchange rate
models, currency values are determined not only by current fundamentals but also by expectations of
what the fundamentals will be in the
future.11 Current fundamentals receive
very little weight in determining the
exchange rate. Not surprisingly, they
aren’t useful in forecasting currency
values, and the exchange rate approximately follows a random walk.
Under certain conditions, the
asset-pricing approach can explain
the greater predictability of exchange
rates as time horizons lengthen.12 Some
fundamentals may behave like “noise”
that drives exchange rates away from
their long-run levels in the short term.
As time passes, exchange rates gradually move back to their long-run levels,
exhibiting long-horizon predictability.
The short-term noise may be related to fundamentals that aren’t observable—for instance, the risk premium
for holding a currency. Calculating the
premium from survey data reveals that
it has no long-run effect on exchange
rate movements. In other words, it is
stationary. In the short run, the premium can push exchange rates away
from their long-run levels. However,
they gradually move back over time.
In this case, exchange rates can be
predicted in the long run but not in
the short run.
While the asset-pricing approach
doesn’t allow us to predict short-term
exchange rates, it does lead to an
interesting implication: Exchange rates
should help forecast economic fundamentals. If the exchange rate is determined by expected future fundamentals, today’s currency values should
yield information about tomorrow’s
Empirical evidence supports
this prediction, although it’s not uniformly strong. One study, for example,
looks at exchange rates in Australia,
Canada, Chile, New Zealand and

South Africa — all countries where
commodities account for a large portion of exports.13 After allowing for
parameter instability, the study finds
that exchange rates help predict an
economic fundamental— in this case,
world commodity prices.
The asset-pricing approach
gains further support from research
that compares currency markets to
other financial markets. One study,
for example, examines opportunities
arising from the carry trade — a term
for borrowing in low interest rate currencies while lending in high interest
rate ones. The return is positive if
exchange rates don’t move to offset
gains from the rate differential. The
results show excess returns are only
compensation for the risks investors
undertake in the carry trade, suggesting that in some ways exchange rates
behave like other assets.14
The asset-pricing approach shows
promise, but empirical work hasn’t
yet solved the exchange rate puzzle
Rogoff and Meese introduced more
than a quarter century ago. Economic
models still do a poor job forecasting
short-term exchange rates.
This issue has become more
pressing. Globalization has made
economies more integrated than ever,
making exchange rates increasingly
important for both businesses and
policymakers. Making wise decisions
when conducting international business and economic policies requires a
better understanding and modeling of
exchange rates.

If the exchange rate is
determined by expected
future fundamentals,
today’s currency values
should yield information
about tomorrow’s

Wang is a senior economist in the Research
Department of the Federal Reserve Bank of Dallas.

For instance, see “A Monetary Approach to the

Exchange Rate: Doctrinal Aspects and Empirical
Evidence,” by Jacob A. Frenkel, Scandinavian
Journal of Economics 78 (2), 1976, pp. 200–24,
and “The Exchange Rate, the Balance of
Payments and Monetary and Fiscal Policy Under
a Regime of Controlled Floating,” by Michael
Mussa, Scandinavian Journal of Economics 78
(2), 1976, pp. 229–48.

F ederal Reserve Bank of Dall as	

7 EconomicLetter


“Empirical Exchange Rate Models of the


“Taylor Rules and the Deutschmark–Dollar Real

Seventies: Do They Fit Out of Sample?” by

Exchange Rate,” by Charles Engel and Kenneth

Richard A. Meese and Kenneth Rogoff, Journal

D. West, Journal of Money, Credit and Banking,

of International Economics 14, February 1983,

vol. 38, August 2006, pp. 1175–94.

pp. 3–24.



This caveat acknowledges that inflation can

“Out-of-Sample Exchange Rate Predictability

with Taylor Rule Fundamentals,” by Tanya

forecast the exchange rate for countries with

Molodsova and David H. Papell, Working Paper,

hyperinflation. Currencies experiencing hyper-

University of Houston, January 2008.

inflation will depreciate against currencies with


more stable prices.

Nineties: Are Any Fit to Survive?” by Yin-Wong


The prediction errors are measured with mean

squared prediction error

Finance, vol. 24, November 2005, pp. 1150–75.


∑ (st – ˆst )


Cheung, Menzie D. Chinn and Antonio Garcia
Pascual, Journal of International Money and



See “Empirical Exchange Rate Models of the




“Exchange Rates and Fundamentals,” by

Charles Engel and Kenneth D. West, Journal

where st is the logarithm of the exchange rate in

of Political Economy, vol. 113, June 2005, pp.

the data and s^t is st predicted by a model.



See “Testing Long-Horizon Predictive Ability


“Can Long Horizon Data Beat Random Walk

with High Persistence, and the Meese–Rogoff

Under Engel–West Explanation?” by Charles

Puzzle,” by Barbara Rossi, International

Engel, Jian Wang and Jason Wu, Working Paper,

Economic Review, vol. 46, February 2005, pp.

University of Wisconsin, the Federal Reserve


Bank of Dallas and the Federal Reserve Board,


For instance, see “Why Is It So Difficult to Beat

June 2008.

the Random Walk Forecast of Exchange Rates?”


by Lutz Kilian and Mark P. Taylor, Journal of

Prices?” by Yu-chin Chen, Kenneth Rogoff and

International Economics, vol. 60, May 2003, pp.

Barbara Rossi, February 2008, Working Paper,

85–107; “The Monetary Exchange Rate Model as

University of Washington, Harvard University

a Long-Run Phenomenon,” by Jan J. J. Groen,

and Duke University. In international commodity

Journal of International Economics, vol. 52,

markets, the exports from each of these coun-

December 2000, pp. 299–319; and “Nominal

tries are small compared with total world supply.

Exchange Rates and Monetary Fundamentals

So the value of currencies in these countries has

Evidence from a Small Post–Bretton Woods

negligible effects on international commodity

Panel,” by Nelson C. Mark and Donggyu Sul,


Journal of International Economics, vol. 53,


February 2001, pp. 29–52.

by Hanno Lustig, Nick Roussanov and Adrien


is published monthly
by the Federal Reserve Bank of Dallas. The views
expressed are those of the authors and should not be
attributed to the Federal Reserve Bank of Dallas or the
Federal Reserve System.
Articles may be reprinted on the condition that
the source is credited and a copy is provided to the
Research Department of the Federal Reserve Bank of
Economic Letter is available free of charge
by writing the Public Affairs Department, Federal
Reserve Bank of Dallas, P.O. Box 655906, Dallas, TX
75265-5906; by fax at 214-922-5268; or by telephone
at 214-922-5254. This publication is available on the
Dallas Fed website,

“Exchange Rates and Fundamentals: Evidence

“Can Exchange Rates Forecast Commodity

“Common Risk Factors in Currency Markets,”

Verdelhan, June 2008, Working Paper, University

on Long-Horizon Predictability,” by Nelson C.

of California, Los Angeles, Wharton School and

Mark, American Economic Review, vol. 85,

Boston University.

March 1995, pp. 201–18.

Richard W. Fisher
President and Chief Executive Officer
Helen E. Holcomb
First Vice President and Chief Operating Officer
Harvey Rosenblum
Executive Vice President and Director of Research
W. Michael Cox
Senior Vice President and Chief Economist
Robert D. Hankins
Senior Vice President, Banking Supervision
Executive Editor
W. Michael Cox
Richard Alm
Associate Editor
Monica Reeves
Graphic Designer
Ellah Piña

Federal Reserve Bank of Dallas
2200 N. Pearl St.
Dallas, TX 75201