Exchange Rate Regimes: Classification and Consequences

Atish Ghosh
1

PDR
IMF
Aghosh@imf.org

Anne-Marie Gulde
MAE
IMF
Aguldewolf@imf.org

Holger Wolf
BMW Center for German and European Studies
Georgetown University
Wolfhc@georgetown.edu

1
The paper is based on our forthcoming book, Exchange Rate Regime, Choices and
Consequences, MIT Press. The views expressed do not reflect the official views of the IMF.

- - 2
Introduction

Does the choice of exchange rate regime matter? Few questions in international
economics have sparked as much debate yielding as little consensus. Over the thirty years
since the breakdown of Bretton Woods countries have adopted a wide variety of regimes,
ranging from dollarization and currency boards to simple pegs and basket pegs, crawling
pegs and target zones to clean floats and dirty floats. This very proliferation of exchange rate
regimes suggests that they must matter for somethingbut quite what, remains an open
question. In this paper, we try to answer that question.

Our approach is unabashedly empirical. Theory certainly offers many insightstoo
many, in fact. There is such an abundance of possible linkages between the exchange rate
regime and macroeconomic performancesome offsetting, others reinforcingthat, at a
theoretical level, it is difficult to establish any unambiguous relationships at all. Accordingly,
in this book, we draw on the experience of some 150-member countries of the International
Monetary Fund (IMF) over the past thirty years to address some simple questions. We begin
with the (deceptively) simple question of what exactly is meant by a fixed (or pegged,
we use the terms interchangeably) versus a floating exchange rate regime, and how might
one go about classifying countries as belonging to one category or the other. We then present
evidence on the link between exchange rate regimes, inflation and output.

Classification

How should a countrys exchange rate regime be classified? The textbook answer is
simple: either the exchange rate is fixed or it floats. The richness of real world regimes
belies this elegant dichotomy as most governments try to reach some (often uneasy)
compromise between the different elements of the impossible trinityindependent
monetary policy, rigidly fixed exchange rates and complete capital mobility (Frankel (1999)).

Popular regimes run the gamut from currency boards and traditional pegs to crawling
pegs, target zones and floats with varying degrees of intervention (Table 1.1, see also
Edwards and Savastano (1999)). Before we can undertake any empirical work, therefore, we
must first decide upon the appropriate level of aggregation, and on a methodology for
classifying regimes. In this chapter, we take up both these issues.

- - 3

Table 1.1: Major Characteristics of Different Exchange Rate Regimes

Regime Main Characteristics and Principal Issues
Dollarization Key feature: A foreign currency acts as legal tender. Monetary policy is delegated to the
anchor country. Potential benefits: Dollarization reduces the time-inconsistency problem
(subject to the perceived probability of a re-introduction of domestic money) and real
exchange rate volatility). Potential drawbacks: Under dollarization external shocks cannot
be buffered by exchange rate movements, imposing costs if business cycles are
asynchronous; while seignorage revenues decline. Issues: The lender-of-last-resort function
must be shifted to the fiscal authority.
Currency
Boards
Key feature: A fixed exchange rate regime (mostly enshrined in law) is complemented by a
minimum backing requirement for domestic money in foreign currency. Potential benefits:
The time-inconsistency problem is reduced (subject to the perceived probability that the
regime is abandoned) and real exchange rate volatility is diminished. Potential drawbacks:
External shocks cannot be buffered by exchange rate movements, imposing costs if
business cycles are asynchronous. The scope for lender of last resort activity is restricted to
excess reserve holdings and fiscal mechanisms. Requires high reserve holdings. Issues:
Lender of last resort limits, exit strategy if used as a transitory regime.
Monetary
Union
Key feature: A group of countries using a common currency issued by a common regional
central bank. Potential benefit: A monetary union reduces the time inconsistency problem
by requiring multinational agreement on policy, and reduces real exchange rate volatility.
Potential drawbacks: Member countries suffering asymmetric shocks lose a stabilization
tool. The cost depends on the extent of asymmetric costs and the availability and
effectiveness of alternative adjustment tools. Issues: Unknown responsiveness of
wage/price setting behavior and migration/investment pattern to the altered regime.
Potential sensitivity of voting equilibria to distribution of shocks.
Traditional Peg Key feature: Fixed rate against a single currency or a currency basket. Potential benefits:
The time inconsistency problem is reduced through commitment to a verifiable target.
Devaluation option provides potentially valuable policy tool in response to large shocks.
Reduces real exchange rate volatility. Potential drawbacks: Provides a target for
speculative attacks. Avoids real exchange rate volatility but not necessarily persistent
misalignments. Does not by itself place hard constraints on monetary and fiscal policy, and
thus provides only a partial solution against time inconsistency problem; the credibility
effect depends on accompanying institutional measures and record of accomplishment.
Issues: Doubts about sustainability in the presence of full capital mobility.
Crawling Peg Key feature: A rule-based system for altering the par value, typically at a predetermined
rate or as a function of inflation differentials. Potential benefits: An attempt to combine
flexibility and stability. Often used by (initially) high inflation countries pegging to low
inflation countries in an attempt to avoid trend real appreciation. Potential costs: At the
margins, a crawling peg provides a target for speculative attacks. Among variants of fixed
exchange rates, it imposes the least restrictions, and may hence yield the smallest credibility
benefits. The credibility effect depends on accompanying institutional measures and record
of accomplishment. Issues: Exit strategy, either to harder peg, or greater flexibility.
Bands Key feature: Exchange rate is flexible within a preset band; endpoints defended through
intervention, typically with some intra-band intervention. An attempt to mix market-
determined rates with exchange rate stabilizing intervention in a rule based system.
Potential benefits: Provides a limited role for exchange rate movements to counteract
external shocks and partial expectations anchor, retains exchange rate uncertainty and thus
motivates development of exchange rate risk management tools. Potential drawbacks: On
the margin, a band is subject to speculative attacks. Does not by itself place hard constraints
on monetary and fiscal policy, and thus provides only partial solution against the time
inconsistency problem. The credibility effect depends on accompanying institutional

- - 4
measures, record of accomplishment, and the characteristics of the band (firm or adjustable,
secret or public, width, strength of intervention requirement).
Float with
discretionary
intervention
Key feature: Exchange rates are determined in the foreign exchange market. Authorities can
and do intervene, but are not bound by any intervention rule. Often accompanied by a
separate nominal anchor, such as an inflation target. Potential benefits: The arrangement
provides a way to mix market determined rates with exchange rate stabilizing intervention
in a non-rule-based system. Potential drawbacks: Does not place hard constraints on
monetary and fiscal policy. Absence of rule conditions credibility gain on credibility of
monetary authorities. Limited transparency.
Pure Float Key feature: The exchange rate is determined in the market without public sector
intervention. Potential benefits: Adjustments to shocks can take place through exchange
rate movements. Eliminates the requirement to hold large reserves. Potential drawbacks:
Does not provide an expectations anchor. Exchange rate regime places no restrictions on
monetary and fiscal policy; time inconsistency problem arises unless addressed by other
institutional measures.



Classification Approaches

Intuitively, a pegged exchange rate is one whose value, in terms of some reference
currency or commodity (usually gold), does not vary, or varies only within narrow, pre-
defined limits. But an exchange rate peg is much more than that since it implies a formal
commitment by the central bank to maintain the parity through foreign exchange intervention
and, ultimately, through the subordination of its monetary policy to the exchange rate
objective if necessary. In a floating regime, by contrast, the central bank undertakes no such
commitment.

This suggests that the exchange rate regime might be best defined by the stated
intentions of the central bank (which every IMF member country is required to report and
publish each year), yielding a de jure classification. The classification emphasizes the
importance of public pronouncements as a signal for the private sectors expectations.
Pushed to its extreme, the implication is that, in comparing two countries with identical
histories, the announced exchange rate regime still matters because it conveys information
about future policy intentions, thus influencing expectations and outcomes.

But what is one to make of a putative fixed exchange rate that is devalued each year
(if not more frequently)? Recalcitrant governments have been known to abuse the credibility
benefits of fixing the exchange rate, pursuing expansionary policies that are inconsistent with
maintaining the peg (Tornell and Velasco (1995,2000)). Clearly, if the central bank does not
take its commitment to defend the parity very seriously, it is not much of a fixed exchange
rate regime. Conversely, if the central bankwhile abjuring any formal commitment
nevertheless intervenes heavily in the foreign exchange markets, then the exchange rate can
hardly be described as a freely floating.
1


To the extent that there is a sizable number of such soft pegs and hard floats,
classifying countries solely according to their declared regime may give misleading results.

- - 5
An alternative de facto classification scheme, therefore, uses the observed behavior of the
nominal exchange rate (and perhaps indicators of monetary policy) to define the exchange
rate regime.
2


De facto classifications are not without their own drawbacks, however. Foremost
among these is their essentially backward-looking nature. While the stated regime in
principle conveys information about future policy intentions, observed actions necessarily
pertain to the past. Again taken to its extreme, the de facto approach implies that a country
announcing an exchange-rate based disinflation programeven if fully crediblewould
derive no credibility benefit in terms of lowering inflationary expectations. While ultimately
an empirical question, the assumption seems at odds with much of modern macroeconomic
theory, which emphasizes the importance of expectations.

Beyond this fundamental concern, de facto measures must contend with a number of
conceptual difficulties and practical problems. Stability of the nominal exchange rate
typically the most significant component of de facto measuresmay reflect either an
absence of shocks or an active policy offsetting shocks; only the latter warrants inferences
about policy. More generally, since countries have different structures and are subject to
different shocks, it is difficult to infer the underlying exchange rate policy from the observed
exchange rate movement. For example, a small open economy with a narrow export base
operating under a pure float will likely experience greater exchange rate volatility than a
larger, more diversified economy also operating under a pure float. Since de facto measures
are inherently relative-countries have more or less fixed exchange rates
3
-the small
country might be (correctly) classified as floating, while the larger country might be
(incorrectly) assigned to the fixed category.

In principle, identification can be achieved by controlling for country characteristics
(Calvo and Reinhart (2000), Levy-Yeyati and Sturzenegger (1999)) and by incorporating
foreign exchange intervention or interest rate movements in the de facto classification.
4
In
practice, neither is straightforward. Interest rates in many developing countries are set
administratively; often bearing little relation to what would be the market clearing rates.
More importantly, central banks typically treat intervention data as confidential; information
is not available on a consistent basis for large, cross-country data sets.
5


Some studies use the change in gross reserves as a proxy for intervention, which has
serious drawbacks. Just as intentions and actions may differ for the exchange rate regime, so
statistics and reality might diverge for data on foreign exchange reserves. As the use of
forward markets, swaps, non-deliverable forwards and a variety of other off-balance sheet
instruments by central banks have became more commonplace, gross reserves even if
reported accuratelybecome ever less revealing.
6
Furthermore, movements in central bank
reserves, particularly in low income countries, are also influenced by a plethora of other
factors. These might include servicing of foreign debt or payments for bulky purchases such
as oil imports or aircraft, which have little to do with intentional foreign exchange
intervention but result in large movements in reported reserves.
7



- - 6
A more fundamental identification problem concerns the distinction between
intervention undertaken to meet an explicit exchange rate target, and intervention motivated
by other policy objectives. Inflation-targeting regimes are a case in point. In a small open
economy that is subject to significant exchange rate pass-through to domestic prices, an
inflation-targeting framework might place considerable weight on exchange rate stability in
the face of nominal shocks
8
thus yielding a low de facto score of exchange rate flexibility.
However, the same framework may dictate considerable exchange rate adjustment in the face
or real shocks. Depending on the relative incidence of real and nominal shocks, a de-facto
scheme may thus classify an inflation-targeting regime as either fixed or flexible. Moreover,
as the incidence of shocks varies across time or countries, so will the de facto classification,
even though the underlying policy regime remains the same.

Ultimately, neither the de jure nor the de facto method is ideal. De jure classifications
focus on the stated policy intentions of the monetary authorities; difficulties arise when
policy practices diverge from promises. De facto classifications are based on actual
movements of the exchange rate (and sometimes, other variables), but are backward- looking
and may capture exchange rate policy very imperfectly.

In our judgment, the drawbacks of the de jure classification are less severe. Most
countries claiming to operate under fixed exchange rate regimes indeed maintain stable
exchange rates over prolonged periods, while most countries claiming to have floating
regimes experience substantial exchange rate variability, albeit occasionally tempered by
large foreign exchange interventions. Accordingly, we prefer to use the de jure classification
as our main method for categorizing regimes.

Notwithstanding this preference, several studies suggest that there are at least some
cases of soft pegs and hard floats (particularly among emerging market countries). To
examine the robustness of our findings to classification problems, we complement our de jure
classification with a consensus classification. This sample consists only of those countries
that are classified in the same category (fixed, intermediate, or floating) by both the de jure
measure and by a de facto measure that we construct. In essence, the consensus sample drops
hard floats and soft pegs. To preview the results, eliminating these ambiguous cases does
not materially affect our findings on inflation performance, but it is of greater importance
when we get to the results on economic growth.
9

The De Jure Classification
Our de jure classification is based on the stated intentions of the monetary authorities,
as reported in the International Monetary Funds Annual Report on Exchange Arrangements
and Exchange Restrictions.
10
Table 1.2 lists the reported exchange rate regimes, along with
their relative frequencies for virtually all IMF member countries from 1970 to 1999, some
4,300 observations covering 167 countries.
11


Table 1.2: De Jure Classification of Exchange Rate Regimes 1970-99
(in percent of total observations)

- - 7


Full Sample: 1970-
1999
1970-1979 1980-1989 1990-1999


Pegged Regimes 65.4 84.8 68.4 46.6
(1) Hard Pegs 13.2 10.0 13.8 15.4
Dollarized 0.6 0.5 0.7 0.6
Currency Board 5.2 4.6 4.7 6.1
Monetary Union 7.4 4.9 8.4 8.7
(2) Single Currency Pegs 32.8 32.8 61.2 61.2 27.4 27.4 13.9 13.9
(3) Basket Pegs 19.4 13.6 27.1 17.2
Published Basket Pegs 9.8 7.3 13.9 8.1
Secret Basket Pegs 9.6 6.2 13.3 9.1

Intermediate Regimes 20.4 11.0 22.5 26.4
(4) Floats with Rule-based Intervention 8.4 5.9 9.5 9.6
Co-operative Regimes (EMS) 5.2 3.8 5.7 6.0
Crawling Pegs 1.4 1.1 1.3 1.9
Target Zones and Bands 1.0 0.5 1.8 0.8
Unclassified Rule-based Systems 0.7 0.5 0.8 0.9
(5) Floats with Discretionary Intervention 12.0 5.1 13.0 16.8
Managed Floating w. Heavy Intervention 1.4 0.5 3.2 0.6
Unclassified Managed Floating 8.1 0.8 6.1 16.1
Other Floats 2.4 3.8 3.7 0.1

Floating Regimes 14.2 4.3 9.1 27.0
(6) Floats 14.2 4.3 9.1 27.0
Floating with Light Intervention 1.5 1.1 2.6 0.8
Floating with no Intervention 12.7 3.2 6.5 26.2


Total observations 100 100 100 100 100 100 100 100 100 100 100 100
Total observations in proportion of full sample 100 100 100 30.4 30.4 30.4 33.1 33.1 33.1 36.5 36.5 36.5

- - 8
At its most detailed level, the classification comprises fifteen regimes ranging from
hard pegs, such as currency boards and dollarization, to intermediate regimes such as
crawling pegs and target zones, and finally floats with varying degrees of intervention.
12

Single currency pegs are the largest group; accounting for almost a third of the observations,
followed by basket pegs and managed floats. Over the entire sample, a little more than a
quarter of the observations, divided roughly equally, fall into the two extreme regimes of
hard pegs and pure floats.

Fifteen regimes is too fine a classification for most questions; for the bulk of our
empirical analysis, we condense regimes into three groups: pegged, intermediate and
floating. Their respective compositions are indicated in bold type in Table 1.2. The right-
most columns report the frequency distributions by decade. Over the sample period, the
prevalence of pegged regimes has declined sharply, from eighty-five percent of all
observations during the 1970s to less than fifty percent by the 1990s. The sharpest gains were
recorded by the floating group, increasing from a mere four percent in the 1970s to more than
twenty-five percent during the 1990s (Figure 1.1).
13


On occasion, we also make use of a more detailed, six-way classification subdividing
the pegged regimes into hard pegs, single currency pegs and basket pegs; intermediate
systems into those with publicly known or rule-based intervention (crawling pegs and target
zones) versus those with more discretionary intervention; with floats forming the final
category. Table 1.2 reports the composition of this finer classification, in italic type.

The time pattern for the more detailed classification (Figure 1.2) shows that the
traditional single currency peg has lost market share, shrinking from sixty percent of
observations to less than fifteen percent over the sample period. Winners include the pure
floats and floats with discretionary intervention, increasing from nine percent to almost forty-
four percent, while hard pegs display a more modest increase from ten to fifteen percent.

Consensus Classification

Since policy actions sometimes differ from stated intentionsmost notably in the
case of soft pegs and hard floatsit is useful to complement the de jure classification with
one based on observed behavior. To this end, we construct a consensus classification
which essentially drops soft pegs and hard floats by using the intersection of the de jure
classification and a de facto classification that we construct.

We first compute a continuous de facto measure based on observed exchange rate
behavior, then convert it into a discrete three-way classification of pegged, intermediate, and
floating regimes using the relative frequency distribution of regimes in the de jure
classification. The consensus sample then simply consists of all observations for which the
two classification methods agree.


- - 9
The most obvious variable on which to base the de facto measure is the nominal
exchange rate itself. As discussed above, stability of the nominal exchange rate may however
reflect either a deliberate policy of keeping the exchange rate fixed, or an absence of shocks.
One option is to augment the measure by incorporating variables such as interest rates and
the change in reserves as proxies for intervention (Calvo and Reinhart (2000), Kaminsky and
Schmukler (2001), Levy-Yeyati and Sturzenegger (1999)).
We pursue the simpler approach of using the variability of the nominal exchange rate
as the de facto measure, for two reasons. First, our interest lies not in the de facto
classification as such, but rather in constructing a robustness check for our empirical results
obtained using the de jure classification. Second, the methodological problems identified
above are, in our view, too severe to justify the loss of sample size implied by trying to add
reserve and interest rate movements to the de facto measure.
Even for our relatively simple de facto measure, several issues must be addressed.
First, how should exchange rate volatility be measured? Second, against which reference
currency should the volatility be assessed? Figure 1.3 plots examples of two idealized types
of nominal exchange rate movementa crawling peg, and a float without trend
together with their combination (a float with a trend). While the float with trend is clearly the
most volatile, it is less clear how to rank the crawling peg relative to the float without trend.
The crawling peg is more predictable, yet the average movement during the year (end point
to end point) may well exceed that of a pure float without trend.

For our de facto measure, we attach equal weight to both features, creating an annual
score based on the mean and variance of the monthly depreciation rates. The continuous de
facto measure is given by
2
e
z
2
e


= + , where
e

is the average monthly rate of change

of the nominal exchange rate during the year, and is the variance of those monthly
changes. For each country in the sample, we compute this measure separately against each of
the G-7 currencies, the ECU and the SDR, and then select the reference currency yielding the
smallest Z score.
2
e
14


To construct a discrete de facto measure comparable to the de jure classification, we
drop the (very small) number of observations for which either the de jure or the de facto
measure is unavailable. For each sample year, we then map the continuous score into three de
facto regimes (pegged, intermediate, and float) by imposing the relative frequency
distribution of the de jure classification for that year.
15


A Comparison

The de jure and the de facto classifications are supposed to capture the same concept
- the nominal exchange rate regime. If most governments do what they say and say what
they do, one would thus expect substantial overlap. Table 1.3 compares the de jure and de
facto classifications. Along the diagonal of the matrix, both classifications coincide; off-

- - 10
diagonal elements represent divergences between the two classifications. The overlap is
greatest for the pegs (where almost seventy percent are classified in the same way), whereas
the distinction between floats and intermediate regimes is hazier. Overall, the overlap is
substantial: almost 65 percent of the observations are classified identically under the two
schemes. These observations constitute the consensus subsample that we will use below
for our robustness tests.

Table 1.3: Distribution of De Jure and De Facto Classifications


De Facto Classification
Pegged Intermediate Floating
De Jure Classification
Pegged Regimes 2265 378 178
Intermediate Regimes 432 270 179
Floating Regimes 124 233 254

Total 2821 881 611
Percentage Consensus 80.3 30.6 41.6




The off-diagonal entries provide an indication of the importance of soft pegs and hard
floats. Conditional on the de facto classification being correct, soft pegs are observations
classified as de jure pegs but de facto floats. There are 178 observations in this category (6
percent of all de jure pegs), while a further 378 (13 percent of all de jure pegs) are classified
as intermediate regimes under the de facto classification. At the other end of the spectrum, de
jure floats that behave like de facto pegs hard floats or fear of floating cases account
for 124 observations (about 20 percent of all de jure floats). A further 233 observations (or
38 percent of de jure floats) would be classified as an intermediate regime under the de facto
classification.

Classifications: Conclusions
Most theoretical analyses rely on a simple dichotomy between fixed and floating
exchange rates. Empirical work does not enjoy that luxury. The rich diversity of regimes
poses two challenges for empirical work: how fine a classification scheme to adopt, and how
to classify individual countries.

Trying to squeeze the various regimes into a rigid dichotomization of fixed versus
flexible regimes risks losing important distinctions. On the other hand, too detailed a
classification risks loosing sight of the forest among the trees. We strike a compromise, using
both a broad three-way classification (corresponding to the intuitive notion that countries

- - 11
might have fixed or floating exchange rate regimes, or an intermediate arrangement) as
well as a more detailed, six-way classification that allows us to distinguish, for instance,
between hard pegs and traditional, adjustable pegs, or between pegs against secret- versus
published baskets.

The second issue concerns the identification of regimes. Are regimes best classified
by words -- the stated commitment of the central bank-- or by actions and behavior, most
notably of the nominal exchange rate? There is no clear answer; both approaches have their
advantages and drawbacks. We prefer the de jure approach, both because of the central role
of signaling and expectations formation in modern macroeconomics, and because of the
significant conceptual and practical problems associated with de facto classifications.
Nevertheless, as a robustness check, we also create a consensus classification that aims to
drop the ambiguous cases -- soft pegs and hard floats -- to examine the sensitivity of our
main results to the classification methodology.


Consequences

Based on the classification reported above, do exchange rate regimes matter for
performance? We examine this issue for prices and output, beginning with a baseline
specification before turning to a range of robustness tests.

Inflation

To examine the conditional link between inflation and the exchange rate regime we regress
the scaled inflation rate /(1+ ) (which we shall simply refer to as inflation, , in what
follows) on two exchange rate regime dummies for pegged (Peg) and intermediate (Int)
regimes respectively. Floating regimes are the excluded category (so that the coefficients on
Peg and Int should be interpreted as the inflation differential relative to a floating exchange
rate regime). Of course, inflation is likely to depend on much more than just the exchange
rate regime. Higher real GDP growth, y, by raising money demand, should reduce inflation.
Conversely, faster growth of the money supply, m, should be associated with higher
inflation. Romer (1993) argues that greater trade openness (the ratio of exports plus imports
to GDP, Open) raises the costs of a monetary expansion, which, by the logic of the policy
credibility models, should imply lower inflation in more open economies.

16


The policy credibility literature suggests that an independent, conservative central
banker can help solve the time consistency problem (Rogoff, 1985). Although it is difficult to
measure the conservatism of the central bank governor, Cukierman (1992) uses the turnover
rate (CBTurn) of the central bank governor as an (inverse) proxy for central bank
independence, on the grounds that less independent central bank governors can be fired more
easily. A higher turnover rate of the central bank governor should therefore be associated
with higher inflation. Other factors include inflationary terms-of-trade shocks (TofT) [Fischer
(1993)] and the fiscal balance (Gov. Bal.), either because of direct money financing, or
simply by contributing to aggregate demand pressures.

- - 12

In addition, we include annual dummies in the regression to capture shocks that are
common across countries, but vary over time, such as oil price shocks. The baseline equation
therefore becomes:
17

0 4 5 6 7
. .
Peg Int Mon
Peg Int m y OPEN CBTurn TofT Gov Bal
8
= + + + + + + + + +


To interpret the results, we need to take account of the possibility that money growth
itself is endogenous to the nominal exchange rate regime. Since money growth is included as
one of the conditioning variables in the regression, any indirect effects of the exchange rate
regime operating through money growth would not be picked up by the coefficients on the
regime dummies. To illustrate, suppose that the growth rate of the money supply depends
upon the exchange rate regime:

(1)
0 Peg Int
m Peg In = + + + t

where Peg and Int are dummy variables for pegged and intermediate regimes, and is a
random shock. Suppose further that the inflation rate depends upon money growth, the
exchange rate regime, and a vector of other conditioning variables, z:


0 Peg Int Mon j j
Peg Int m z = + + + + + (2)

The direct effect of the exchange rate regime on inflation, conditional on money
growth-what we termed the confidence effect above-is given by the coefficient
Peg

(
Int
for intermediate regimes). From (1), however, there is also an indirect effect on inflation
through the money growth channel, given by
Mon Peg
. Since Peg is a dummy variable, an
estimate of
Peg
is given by the difference in the average growth rates of the money supply
under pegged and floating regimes: Peg
Peg
m = Flt m (and likewise for
Int
). The full effects of
the exchange rate regime on inflation, combining the confidence and discipline effects can
therefore be obtained from (2) as ( ) Peg Flt
Peg Peg
m m = +
Mon
, below we report both the
confidence effect,
Peg
and this combined effect,
Peg
.
19


Table 2.1 reports the coefficients for alternative samples. Panel 1 refers to the de jure
classification of exchange rate regimes, and Panel 2 to the consensus classification. Within
each panel, the coefficients on the regime dummies in the first column capture the sum of the
confidence and the discipline effects [ ( ) Peg Flt
Peg Peg Mon
m m = + ]. The coefficient in the
second column reports the confidence effect alone [
Peg
]. The difference between the two
columns thus captures the discipline effect.

The various independent variables enter the regression with the expected signs and are
statistically significant (with the exception of the terms of trade). Faster money growth and

- - 13
lower GDP growth are associated with higher inflation, while more open economies,
countries with larger fiscal balances, and those with more independent central banks exhibit
lower inflation.

Turning to the exchange rate regime dummies, inflation under pegged regimes is 10
percentage points per year lower than under floating regimes, of which about 4 percentage
points are associated with lower money growth, and the remaining 5 percentage points
represent the confidence effect.
20
At 13 percentage points, the inflation differential vis--
vis intermediate regimes is larger.

From Panel 2 of Table 2.1, under the consensus classification, the difference in inflation
rates between pegged and floating regimes is twice as large--almost 22 percentage points per
year (of which 7 percentage points are associated with lower money growth, and 14
percentage points with greater confidence).
21
Taken together, the results suggest that, by
pegging their exchange rate, countries can improve their inflation performance substantially,
with the full anti-inflationary benefits--particularly through the confidence channel--accruing
to those countries that maintain a hard peg and avoid frequent devaluations.

Differentiating across the three income groups, the results are weakest for the upper
income (industrialized) countries, for which the differences in inflation rates across regimes
are negligible and statistically insignificant. One possible explanation is that these countries
have strong institutional frameworks and enjoy low inflation rates anyway, so that the
exchange rate regime makes little difference.

This is borne out by splitting the full sample into low inflation observations (those below
10 percent per year) and high inflation observations (those above 10 percent per year), and
re-estimating the regression for the two subsamples. As Table 2.2 indicates, the difference in
inflation between pegged and floating regimes holds at inflation rates above 10 percent per
year, but not below. (The cut-off of 10 percent per year is just about right: if the low inflation
sample is defined to include inflation rates of 12 percent per year or below, the coefficient on
the Peg dummy becomes negative and statistically significant).

Table 2.2 also reports regressions for a variety of other sub-samples each capturing
alternative ways (other than the exchange rate regime) through which countries can gain
policy credibility. These samples include countries with no current account restrictions (IMF
Article VIII countries), countries with no capital controls (as defined by the IMFs Annual
Report on Exchange Arrangements and Restrictions), countries with especially low rates of
turnover of the central bank governor, and countries that are particularly open to international
trade.

For countries without current account restrictions, high trade openness ratios, or low
central bank turnover rates, the difference in inflation performance between pegged and
floating regimes remains. Abjuring capital controls, on the other hand, does seem to provide
an alternative means of gaining policy credibility. Inflation in these countries averages less
than 10 percent per year, and the de jure exchange rate regime makes only a marginal

- - 14
difference to their inflation performance (in the consensus classification, however, the
difference remains pronounced, at about 17 percentage points per year).

Inflation: Extensions and Robustness Tests

The results suggest a rather strong link between the exchange rate regime and inflation
performance. In this section, we check whether the results are robust to a variety of different
econometric specifications, and to controlling for the potential endogeneity of the regime
itself.

A Finer Classification

We begin by disaggregating the three de jure exchange rate regimes into their more
detailed six-way classification: hard pegs, single currency pegs, basket pegs, rule-based
flexible systems, floats with discretionary intervention, and pure floats. We include dummies
for the first five regimes, leaving pure floats as the omitted category. The coefficients on the
regime dummies, reported in Table 2.3, should thus be interpreted as inflation differentials
relative to a pure float.

Hard pegs (currency boards and dollarization) perform best, with a 15 percentage point
inflation differential vis--vis pure floats, split roughly equally between the discipline and
confidence effects. Basket pegs do somewhat better than single currency pegs. The
confidence effect, however, is slightly greater for the latter, consistent with the view that,
compared to basket pegs, the transparency and simplicity of a single currency peg should
foster greater confidence (Frankel, Schmukler, and Servn, 2000b).

Notwithstanding the general tendency for inflation to be higher under more flexible
regimes, average inflation under pure floats is some 8 percent per year lower than under
floating regimes with discretionary intervention. This gives some support to the contention
that extreme regimesstrict pegs or pure floatsperform particularly well. Alternatively,
the finding may be an indication that countries with generally good macroeconomic
fundamentals can better afford to allow their exchange rate to float freely (whether
endogeneity of regime choice can account for the good inflation performance of pegged
regimes is taken up below). Otherwise, the results for this more detailed classification are
very much consistent with those for the three-way categorization into pegged, intermediate,
and floating regimes.

Contamination

Are the results driven by contamination across regimes? For example, fixing the
exchange rate may allow the government to cheat, delivering apparently good inflation

- - 15
performance during its tenure while leaving a legacy of a collapsing peg and high inflation to
its successors. (At least in the de jure classification, fiscal deficits have on average been
larger under pegged regimes.) If so, using the contemporaneous exchange rate classification
would inappropriately attribute inflation that had developed under the pegged regime to the
subsequent floating period.

The problem can be addressed either by dropping the first few years following any
regime change from the dataset or, more stringently, by reclassifying the first few years of
any floating regime under the previous peg. An alternative is to drop the fiscal balance from
the regression, on grounds that pegged regimes might be associated with larger deficits, and
that by conditioning on the deficit, this channel for pegged exchange rates to adversely affect
inflation is being purged from the regression. As specifications [1]-[4] in Table 2.4 indicate,
none of the modifications changes the qualitative results--pegged regimes continue to exhibit
both confidence and discipline benefits.

Fixed Effects
Does the association between pegged regimes and low inflation stem from the samples
cross-section dimension, from its time dimension, or both? Results coming from only one of
these dimensions would be less compelling.

The panel results suggest that countries operating under pegged rates have lower inflation
on average. This does not imply, however, that a country switching from a floating to a
pegged regime will enjoy a lower inflation rate (even ignoring transitional dynamics) as the
results could be driven entirely by country-specific factors. To address this possibility, we re-
estimate the panel regression, including country-specific fixed effects.

As the fifth panel in Table 2.4 shows, the imputed effect of the exchange rate regime does
indeed become somewhat smaller. The total inflation differential between pegged and
floating regimes declines from 10 to 7 percentage points per year in the de jure
classification, and from 21 to 15 percentage points per year in the consensus classification.
In both cases, however, the coefficients remain highly significant.

Endogeneity of Regime Choice

The above results all suggest that countries operating under pegged exchange rate
regimes enjoy lower inflation. However, whether a country adopts, or maintains, a pegged
exchange rate might itself depend upon its inflation performance. If countries that have low
inflation are also more likely to maintain a pegged exchange rate, then the coefficient on the
pegged regime dummy is subject to simultaneity bias.

To address this issue, we develop a simultaneous equation framework that explicitly
allows for the endogeneity of regime choice. Inflation is assumed to depend upon whether

- - 16
the country has an exchange rate peg, Peg, and a vector of other characteristics, ,
comprising the independent variables included in the regressions reported above:
1
z


1 1 1 1
z Peg = + + (3)

The decision to peg is assumed to depend upon the inflation rate, as well as other country
characteristics, z
2
:
22


*
2 2 2 2
Peg z = + + (4)

Peg
*
is an unobserved desire to peg the exchange rate and captures non-systematic
factors. Peg = 1 if, and only if, Peg* > 0. If
2
< 0 (countries with high inflation are less
likely to maintain an exchange rate peg), then the anti-inflationary benefit of pegging, as
estimated above, is overstated.

Using the simultaneous equation framework, we can estimate the residual effect of a
pegged exchange rate regime on inflation, controlling for the endogeneity of the choice of
regime. We first estimate the fully reduced form of the regime choice [obtained by
substituting (3) into (4)]:

Peg z = + (5)

where includes all of the independent variables, and
1 2
[ , ] z z z = is a composite error term. The
predicted values for from this probit are then substituted into the semi-reduced form of
the inflation regression (3) with the appropriate correction for the standard error (Maddala
(1989)).
Peg

In principle, the non-linearity of the probit function is sufficient for identifying the
inflation regression; in practice, an explicit exclusion restriction makes a more compelling
case for identification. We use two restrictions, on country size and on export concentration.
The literature on exchange rate regime choice suggests that smaller countries and countries
with more geographically concentrated exports tend to have a pegged exchange rate. There
is, however, little reason to believe that either country size or export concentration influences
inflation.

Since it is difficult to find plausible instruments that distinguish between pegged and
intermediate regimes, we use a dichotomous classification (for this section only), grouping
intermediate and floating regimes together. The inflation regression is thus not comparable to
the results reported above, where intermediate and floating regimes were separated. To
provide a benchmark, we first re-estimate the simple OLS regression for this dichotomous
classification. The results, given in the bottom panel of Table 2.5, indicate a differential of 12
percentage points per year in favor of pegged regimes (and 13 percentage points per year
under the consensus classification).


- - 17
Turning to the simultaneous equation framework, it is reassuring to note that, as
hypothesized, smaller countries and countries with greater geographic concentration of
exports are more likely to have an exchange rate peg (Table 2.5, top panel).
23
In all, the
probit correctly predicts some 70-80 percent of the observations.

The second-stage inflation regression suggests that the OLS estimates may indeed be
subject to some simultaneity bias. In the two-stage simultaneous equation framework, the
inflation differential in favor of pegged regimes falls from 11.7 to 8.1 percentage points per
year for the de jure, and from 13.5 to 9.8 percentage points per year for the consensus
classifications. Nonetheless, the benefit of pegged regimes for lowering inflation remains
both economically and statistically significant.
24

INFLATION: CONCLUSIONS

Is the exchange rate regime linked to inflation? The results are compelling. Inflation is
lower under pegged exchange rates, reflecting both lower money growth (the discipline
effect) and greater confidence in the currency (the credibility effect). The results are robust
across a range of sub-samples. They are also robust to using different definitions of exchange
rate regimes, allowing for cross-regime contamination, and controlling for the potential
dependence of the regime choice on inflation.

- - 18

References
Ball, Laurence, 1999, Policy Rules for Open Economies, in John B. Taylor (ed.), Monetary Policy Rules,
Chicago, University of Chicago Press.

Ball, Laurence 1990, Why Does High Inflation Raise Inflation Uncertainty?' NBER Working Paper No.
3224, (Cambridge, Massachusetts).

Berger, Helge, Jan-Egbert Sturm, and Jakob de Haan, 2000, An Empirical Investigation Into Exchange Rate
Regime Choice and Exchange Rate Volatility, CESifo Working Paper No. 263 (Munich).

Bleaney, Michael, 1999, The Disappearing Openness-Inflation Relationship, IMF Working Paper No. 99/161,
Washington, DC.

Calvo, Guillermo, and Carmen Reinhart, 2000a, Fear of Floating, NBER Working Paper No. 7993,
(Cambridge, Massachusetts).

Calvo, Guillermo, and Carmen Reinhart, 2000b,Fixing for Your Life, NBER Working Paper No. 8006,
November (Cambridge, Massachusetts).

Clarida, Richard, Jordi Gali, and Mark Gertler, (2001), Optimal Monetary Policy in Open versus Closed
Economies, Mimeo, New York University.

Corsetti, Giancarlo and Paolo Pesenti, (2001), Welfare and Macroeconomic Interdependence; forthcoming,
Quarterly Journal of Economics.

Cukierman, Alex, 1992, Central Bank Strategy, Credibility, and Independence (Cambridge: MIT Press).

Devereux, Michael, and Charles Engel, 1998, Fixed Versus Floating Exchange Rates: How Price Setting
Affects the Optimal Choice of Exchange Rate Regime, NBER Working Paper Series No. 6867,
(Cambridge, Massachusetts).

Devereux, Michael, and Charles Engel, 1999, The Optimal Choice of Exchange Rate Regime, NBER
Working Paper Series No. 6992, (Cambridge, Massachusetts).

Edwards, Sebastian, 1989, Real Exchange Rates, Devaluation, and Adjustment (Cambridge: MIT Press).

Edwards, Sebastian, and Miguel Savastano, 1999, Exchange Rates in Emerging Economies: What do we
Know? What do we Need to Know? NBER Working Paper No. 7228, (Cambridge, Massachusetts).

Fischer, Stanley, 1993, Role of Macroeconomic Factors in Growth, Journal of Monetary Economics, Vol. 32,
(December), pp. 485-512.

Flood, Robert, and Nancy Marion, 1982, The Transmission Of Disturbances Under Alternative Exchange Rate
Systems With Optimal Indexation, Quarterly Journal of Economics, Vol. 97, pp. 43-68.

Flood, Robert, and Nancy Marion 1991, Exchange Rate Regime Choice, IMF Working Paper No. 91/90
(Washington: International Monetary Fund).

Frankel, Jeffrey, 1999, No Single Currency Regime is Right for all Countries at all Times, Essays in
International Finance, No. 215. (Princeton, N.J.: Princeton University International Finance Section).

- - 19

Garber, Peter, and Lars Svensson, 1995, The Operation and Collapse of Fixed Exchange Rate Regimes, in
Handbook of International Economics, ed. by Gene Grossman and Kenneth Rogoff, Vol.3.
(Amsterdam: North-Holland), pp. 1865-1912.

Hausmann, Ricardo, Michael Gavin, Carmen Pages, and Ernesto Stein, 1999, Financial Turmoil and the
Choice of Exchange Rate Regime, Inter-American Development Bank, Research Department
Working Paper No. 400.

[Haussmann, Ricardo, Ugo Panizza, and Ernesto Stein, 2000, Why Do Countries Float The Way They Float?,
Mimeo, Inter-American Development Bank.

Kenen, Peter, 2000, Currency Areas, Policy Domains, and the Institutionalization of Fixed Exchange Rates,
Mimeo, Princeton University.

Klein, Michael and Nancy Marion, 1994, Explaining the Duration of Exchange Rate Pegs, Journal of
Development Economics (December), Vol. 54: 2: pp. 387404.

Lahiri, Amartya and Carlos Vgh, 2001, Living with the Fear of Floating, NBER Working Paper No. 8391,
July (Cambridge, Massachusetts).

Lane, Philip, 1997, Inflation in Open Economies, Journal of International Economics, Vol. 42, May, pp. 327-
347.

Larrain, Felipe, and Andres Velasco, 1999, Exchange Rate Policy for Emerging Markets, Occasional Paper
No. 60, (Washington, Group of Thirty).

Levy Yeyati, Eduardo and Federico Sturzenegger, 1999, Classifying Exchange Rate Regimes, mimeo,
Universidad Torcuato Di Tella, Buenos Aires.

Levy Yeyati, Eduardo, and Federico Sturzenegger, 2001a, To Float or to Trail; mimeo, Universidad Torcuato
Di Tella, Buenos Aires.

Levy Yeyati, Eduardo, and Federico Sturzenegger, 2001b, Exchange Rate Regimes and Economic
Performance, Staff Papers, International Monetary Fund, Vol. 47, Special Issue, 62-98.

Maddala, G.S., 1989, Limited Dependent and Qualitative Variables in Econometrics, (Cambridge: Cambridge
University Press).

Masson, Paul, Miguel Savastano, and Sunil Sharma, 1997, The Scope for Inflation Targeting in Developing
Countries, IMF Working Paper No. 97/130 (Washington: International Monetary Fund).

Obstfeld, Maurice and Kenneth Rogoff 1995a, Exchange Rate Dynamics Redux, Journal of Political
Economy, 103: 624-660.

Obstfeld, Maurice, and Kenneth Rogoff, 1995b, Mirage of Fixed Exchange Rates, Journal of Economic
Perspectives (Fall) Vol. 9, pp. 7396.


Obstfeld, Maurice, and Kenneth Rogoff, 2000, New Directions for Stochastic Economy Models, Journal of
International Economics, (February), pp. 117154.

Poirson, Hlne, 2001, How Do Countries Choose Their Exchange Rate Regime?, IMF Working Paper No.
01/46 (Washington, DC, International Monetary Fund).

- - 20

Rogoff, Kenneth, 1985, The Optimal Degree of Commitment to an Intermediate Target, Quarterly Journal of
Economics, Vol. 100, pp. 1169-1190.

Romer, David, 1993, Openness and Inflation: Theory and Evidence, Quarterly Journal of Economics, Vol.
CVIII, pp. 869-903.

Svensson, Lars (2000), Open-Economy Inflation Targeting, Journal of International Economics, Vol. 50, pp.
155-183.

Tornell, Aaron, and Andrs Velasco, 1995, Fiscal Discipline and the Choice of Exchange Rate Regime,
European Economic Review, Vol. 39 (April), pp. 759770.

Tornell,Aaron, and Andrs Velasco, 2000, Fixed Versus Flexible Exchange Rates: Which Provides More
Fiscal Discipline? Journal of Monetary Economics, Vol. 45, (April), pp. 399436.

- - 21

Endnotes

1. See Calvo and Reinhart (2000a,b), Haussmann, Panizza and Stein (2000).
2. See Calvo and Reinhart (2000a,b), Kaminsky and Schmukler (2001), Levy-Yeyati and Sturzenegger (1999)
and Poirson (2001).
3. Levy-Yeyati and Sturzenegger (1999) use an endogenous clustering approach combining similar
observations into groups. Other approaches impose a priori thresholds on a single aggregated de facto
score. Neither approach corresponds closely to theoretical concepts (nor does our de facto classification
rule, described below).
4. While direct intervention is the more common tool, changes in interest rates or in domestic credit expansion
can also be used to affect the exchange rate in the face of shocks, and may thus deserve a role in de facto
classifications. See Calvo and Reinhart (2000a) for a careful analysis along these lines.
5. See Ghosh (2002) for an analysis of central bank incentives to intervene secretly.
6. In a number of recent exchange rate crises, assumptions of the private sector (and even the IMF) about the
true level of usable reserves were later revealed to be spectacularly wrong as the monetary authorities in
these countries had engaged in large, off-balance sheet transactions (or had otherwise encumbered their
reserves) in the run-up to the crises.
7. While official payments and receipts can be excluded from reserves (Levy-Yeyati and Sturzenegger
(1999)), other outliers, including bulky trade transactions, are more difficult to allow for.
8. See Ball (1999), Clarida, Gali and Gertler (2001), Corsetti and Pesenti (2001), Hausmann, Panizza and
Stein (2000), Lahiri and Vgh (2001), Masson, Savastano and Sharma (1997), Obstfeld and Rogoff (2000)
and Svensson (2000).
9. It is interesting to note in this context that Levy-Yeyati and Sturzenegger (2001a,b), using a de facto
measure, find different growth results compared to earlier results of Ghosh, Gulde, Ostry, Wolf (1995) who
use the de jure classification, while their results for inflation are more similar.
10. Under the Second Amendment of the IMFs Articles of Agreement, member countries are free to choose
their exchange rate regime, but are required to inform the IMF of their choice, and to promptly report any
changes to their exchange rate arrangements.
11. The Appendix provides a more detailed description of the data set.
12. Intervention frequency is based on assessments by IMF country desk officers and on summaries of the
implementation of exchange rate regimes contained in annual IMF country documents.
13. The uptick in pegged regimes in 1999 reflects the shift from the EMS system (classified as intermediate) to
EMU (classified as a peg).
14. For the G5 countries, we drop their own currency. For the small set of countries reporting a de jure peg vis
vis a currency not in this group, we added the currency in question (the South African Rand, the Indian
Rupee, the Spanish Peseta, the Australian Dollar and the Portuguese Escudo) to the search for that
particular country. The Z scores are clustered around zero with a long declining right-hand tail peaking at
about 1 percent per month.
15. As discussed above, there is no fully satisfactory way of mapping the continuous score into a discrete
classification. Our identification is based on the assumption that the overall frequency distribution of the de
jure regimes (given the partial offset between hard floats and soft pegs) is reasonably accurate.
16. See Romer (1993) and Lane (1997) for empirical evidence. Bleaney (1999) finds the relationship to be
unstable in the 1990s.
17. To control for potential endogeneity, money growth, real GDP growth, and the fiscal balance are
instrumented using their lagged values; t-statistics are computed using White heteroscedastic consistent
standard errors.
18. The coefficient on money growth is constrained to be equal across regimes. An alternative formulation
would allow for a differential impact of money growth on inflation, depending upon the exchange rate
regime:
0
Peg Int Flt
Peg Int m m m
Peg Int Peg m Int m Flt m = + + + + + + +

- - 22
19. We assume that any other determinants of money growth in (1) are uncorrelated with Peg and Int. The
standard error of
Peg
, SE (
Peg
), is calculated from the variance-covariance matrix of
peg
and
money,

treating Peg m Flt m as known, and the reported t-statistic is simply the ratio
Peg
/SE (
Peg
).
20. Conditional on money growth, the effect of the pegged exchange rate regime is given by
Peg
, as before.
The unconditional coefficient, however, now becomes
Peg Flt
Peg Flt
Peg Peg Mon Mon
m = + m . Estimating this
alternative regression yields
Peg
Mon
=0.125 (t-stat.: 2.89
***
) and
Flt
Mon
= 1.00 (t-stat.: 9.57
***
); since
Peg
Mon
<
Flt
Mon

the inflation differential in favor of pegged regimes (unconditional on money growth) becomes larger,
while the conditional effect,
Peg
, remains roughly the same.
21. The consensus sample drops de jure pegs with high exchange rate volatility and de jure floats with low
exchange rate volatility. Inasmuch as greater exchange rate volatility is associated with higher average
inflation, the consensus sample tends to drop the de jure pegs with the highest and the de jure floats with
the lowest inflation ratesthereby widening the estimate inflation differential. Levy-Yeyati and
Sturzenegger (2001b), using a de facto classification, find that the inflation advantage of fixed exchange
rates derives primarily from durable pegs in low- and moderate-income countries.
22. Optimal regime choice as a function of country characteristics is the subject of a substantial literature of its
own. Recent contributions include Berger, de Haan and Sturm (2000), Edwards (1996), Flood and Marion
(1991), Garber and Svensson (1995), Klein and Marion (1994), Obstfeld and Rogoff (1995a,b), Larrain
and Velasco (1999), and Poirson (2001), among others.
23. Note that only the coefficients on country size and export concentration are identified, since these variables
do not enter the inflation regression. To identify the other coefficients of the semi-reduced form (4) from
the estimated fully-reduced form (5) would require some additional identifying restrictions.
24. An alternative approach to the issue of regime endogeneity is to compare the performance of countries that
switched from floating to pegged regimes: in the three years following the adoption of the peg, median
inflation was some 15 percentage points per year lower in countries that switched regimes.