Journal of Development Economics

Vol. 68 (2002) 233 – 245

www.elsevier.com/locate/econbase

Exchange rate regimes, inf lation and output

volatility in developing countries
Michael Bleaney a,*, David Fielding b
a
School of Economics and CREDIT, University of Nottingham, Nottingham NG7 2RD, UK
b
Department of Economics, University of Leicester and CSAE, Leicester, UK
Received 30 September 1998; accepted 30 June 2001

Abstract

The median developing country has had significantly higher inflation than the median advanced
country since the early 1980s. We present a model in which a developing country may reduce
inflationary expectations by pegging its exchange rate to the currency of an advanced country, at the
expense of forgoing its ability to compensate for real exchange rate shocks. Different types of pegged
exchange rate offer varying degrees of anti-inflation credibility and of exposure to shocks. Tests on a
sample of 80 developing countries support the empirical predictions of the model. D 2002 Elsevier
Science B.V. All rights reserved.

JEL classification: F41; O11

Keywords: Exchange rate regimes; Inflation; Output volatility

1. Introduction

In recent years, median inflation rates in LDCs have been substantially higher than
in industrial countries. Even excluding cases of hyperinflation, poverty seems to be
associated with more rapid price growth.1 However, an LDC that can credibly commit
itself to an exchange rate pegged against an industrial currency might be able to do
better than the LDC average, and enjoy industrial country inflation rates. Recent work
suggests that LDCs which peg their exchange rates achieve lower inflation than those

*
Corresponding author. Tel.: +44-115-951-5464; fax: +44-115-951-4159.
E-mail address: michael.bleaney@nottingham.ac.uk (M. Bleaney).
1
A possible explanation for this difference is that monetary authorities have different preferences at different
levels of per capita GDP. Low inflation may entail costs that low income countries are less willing to accept.

0304-3878/02/$ - see front matter D 2002 Elsevier Science B.V. All rights reserved.
PII: S 0 3 0 4 - 3 8 7 8 ( 0 2 ) 0 0 0 0 2 - 0
234 M. Bleaney, D. Fielding / Journal of Development Economics 68 (2002) 233–245

whose exchange rates float (Edwards, 1993; Ghosh et al., 1995).2 We will present a
theoretical model to explore the costs and benefits of such a peg. In the model, LDC
monetary authorities face a trade-off when choosing an exchange rate regime. A float
allows greater freedom in responding to exogenous shocks, and so greater stability of
output (and inflation) than under pegged rates, at the expense of higher mean inflation.
Using a sample of LDCs with different exchange rate regimes, we will test whether,
after controlling for other factors, those with managed exchange rates enjoy lower
inflation and suffer higher output and inflation variability than those with floating
rates.

2. The model

Consider a model of the Barro and Gordon (1983) type, in which policy-makers have
two targets (for output and inflation), but their interest in the former tempts them to raise
output above the equilibrium level, creating an inflationary bias. With rational expect-
ations, this bias is built into the private sector’s inflationary expectations, and the expected
inflation occurs unless the government tries to establish a reputation for stable prices. One
way in which poorer countries could experience higher inflation in this model is that,
being poorer, they attach higher weight to the output objective. This causes their non-
reputational equilibrium inflation rate to be higher. So why do they not peg their exchange
rates to those of an advanced country? If this peg were credible, then inflation could be
reduced to advanced country levels without any loss in average output. The disadvantage
of pegging is the reduced capacity to adjust the nominal exchange rate to terms of trade
shocks. Rogoff (1985) first made the point that the authorities may wish to accept some
inflation in order to reduce output variability in the face of shocks. We will develop a
model in which the optimal choice of regime depends on the size of shocks to the
equilibrium real exchange rate.
In each time period, the government of a developing country maximises the utility
function

Z ¼ 0:5p 2  0:5bð y  y*  kÞ2 b > 0; k > 0 ð1Þ

where p denotes inflation, y is output and y* equilibrium output. Because k is positive, this
utility function is characterised by inflationary bias, and b determines the relative weight
given to output maximisation rather than price stability. A fundamental assumption of the
model is that in the advanced countries, b takes the lower value ba, implying greater
attachment to price stability. The government maximises Eq. (1) subject to an open-

2
One major difference between our paper and Ghosh et al. (1995) (and others such as Collins, 1996) is that
we are particularly interested in the distinction between countries that persist with a peg for an extended period of
time and those that do not. For industrial countries, the empirical debate on exchange rate regime effects has
focused on difference in inflation persistence rather than in level (Alogoskoufis, 1992; Bleaney, 2001; Burdekin
and Siklos, 1999; Obstfeld, 1995).
 M. Bleaney, D. Fielding / Journal of Development Economics 68 (2002) 233–245 235

economy expectations-augmented Phillips curve equation and an exchange rate regime.

The Phillips curve equation is
y ¼ y* þ aðp  pe Þ  cðq  q*Þ a > 0; c > 0 ð2Þ

where pe denotes expected inflation, q is the real exchange rate (an increase implying
appreciation) and q* is the equilibrium real exchange rate, which is assumed to follow a
random walk:

qt * ¼ qt *1 þ et efN ð0; re2 Þ: ð3Þ

The government may choose either a flexible exchange rate or a pegged exchange rate.
The distinction between the two lies in the information available to the government in
setting qt.3 Under flexible exchange rates, the government can choose qt after observing
the shock et, whereas under pegged exchange rates, it has to choose qt before observing et.
Under flexible rates, the government chooses qt = qt*, and Eq. (2) becomes
yf ¼ y* þ aðp  pe Þ ð2f Þ

where the f subscript denotes floating exchange rates. Under pegged rates (subscript p), the
government chooses q so that E( qt) = qt*, which implies that qt = q*t  1. Eq. (2) then
becomes
yp ¼ y* þ aðp  pe Þ þ ce: ð2pÞ

Substituting Eq. (2p) into Eq. (3), differentiating with respect to p and setting the
differential equal to zero reveals that with a pegged exchange rate, the government
chooses p such that

ð1 þ a2 bÞp ¼ a2 bpe þ abk  abce: ð4pÞ

Under rational expectations, the private sector chooses pe by taking expectations of Eq.
(4p), which yields
pe ¼ abk ð5pÞ

Substitution into Eq. (3) gives

 
p ¼ abk  abc=ð1 þ a2 bÞ e ð6pÞ

and substituting this into Eq. (2p) yields

 
y ¼ y* þ c=ð1 þ a2 bÞ e: ð7pÞ

The solution under floating rates is obtained by setting e = 0, but we also need to take

3
In reality, a government sets a nominal exchange rate peg rather than a real exchange rate peg. However, the
two are equivalent if the government is assumed to know expected inflation when the exchange rate is set.
236 M. Bleaney, D. Fielding / Journal of Development Economics 68 (2002) 233–245

account of the differences in the value of b under the two systems. Under floating rates, we
assume that b takes on its developing-country value, bd. Then we have

p ¼ abd k ð6f Þ

y ¼ y*: ð7f Þ

The value of b under pegged rates requires further discussion. Because each country is
allowed to reset the peg in each period, there is no guarantee that inflation will be reduced
to advanced-country levels. So there are multiple equilibria, depending on the degree to
which the LDC chooses to accommodate excess inflation by adjusting the peg. In the
simplest case, where the government is known to be unwilling to accommodate any excess
inflation, we have p = pe = abak. The government has to behave as if its true value of b is ba
rather than bd (otherwise Eq. (4p) cannot be made consistent with rational expectations). In
the general case, it is convenient to define a variable k, which is the degree of credibility of
the commitment to industrial country levels of inflation:

b ¼ kba þ ð1  kÞbd 0 < k < 1: ð8pÞ

If k = 0 then expected inflation is the same as if the exchange rate floated. In general,
however, expected inflation is less than this by an amount that depends on the degree of
credibility. Full credibility is equivalent to k = 1.
At this point, it is appropriate to consider different types of peg. We shall consider two:
a unilateral peg to an advanced country’s currency (‘‘unilateral peg’’) and a co-ordinated
peg to a single currency by a number of countries (‘‘co-ordinated peg’’ —the case which
we have in mind is the CFA zone, where the countries actually have a common currency).4
Since the real exchange rates of advanced countries follow something like a random walk,
we may write, for a unilateral peg,

q t ¼ qt * þ e t þ gt gfN ð0; rg2 Þ ð9puÞ

where g represents the shock to the anchor country’s real exchange rate. In the case of a
co-ordinated single-currency peg, the difference is that the nominal exchange rate cannot
be set solely by reference to the expected value of each individual country’s equilibrium
real effective exchange rate. Thus, we can no longer assume that the peg is selected such
that E( qt) = qt*. Instead, the peg is chosen such that E( qt) is equal to the mean value of qt*
over the participating countries ( Qt*). Denoting the difference between Q* and q* as x,
we may write, for the co-ordinated case:

qt ¼ qt * þ et þ gt þ xt : ð9pcÞ

4
In the working paper version (Bleaney and Fielding, 1999), we also analyse potential differences that may
arise between single-currency peg countries and basket peg countries. This part of the analysis is omitted here
because there was no evidence of any significant difference in the cross-country regression equations.
 M. Bleaney, D. Fielding / Journal of Development Economics 68 (2002) 233–245 237

Unlike e and g, x does not have a zero mean, since it will reflect the accumulated values of
e in the participating countries since the beginning of the co-ordinated peg.5
How will the choice of peg type affect its credibility? We assume that, relative to a
unilateral peg (up), co-ordination (cp) augments credibility, since all the governments have
to agree to change the exchange rate: an individual government cannot decide to devalue
unilaterally. So kcp >kup.
Using Eqs. (8p) and (9pc) and substituting into Eqs. (6p) and (7p), we get the
following:
 
p ¼ aðkba þ ð1  kÞbd Þk  aðkba þ ð1  kÞbd Þc
 1
 ð1 þ a2 ðkba þ ð1  kÞbd ÞÞ ðe þ g þ xÞ ð10pÞ

EðpÞ ¼ aðkba þ ð1  kÞbd Þk ð11pÞ

 2  2
varðpÞ ¼ aðkba þ ð1  kÞbd Þc ð1 þ a2 ðkba þ ð1  kÞbd ÞÞ
 ðr2e þ r2g þ r2x Þ ð12pÞ
 
y ¼ y* þ c=ð1 þ a2 ðkba þ ð1  kÞbd ÞÞ ðe þ g þ xÞ ð13pÞ
 
Eð yÞ ¼ y* þ c=ð1 þ a2 ðkba þ ð1  kÞbd ÞÞ X ð14pÞ
 2
varð yÞ ¼ c=ð1 þ a2 ðkba þ ð1  kÞbd ÞÞ ðr2e þ r2g þ r2x Þ ð15pÞ

where X is the expected mean value of x. These equations have to be evaluated as

follows:

UNILATERAL PEG: k = kup; x = X = rx2 = 0;

CO-ORDINATED PEG: k = kcp.

Comparing these equations with the certain outcomes for floating rates given by Eqs.
(6f) and (7f) yields the following predictions (F = floating rates):

mean inflation: F > UP >CP

inflation variance: CP > UP > F
output variance: CP > UP > F

These predictions hold for given values of the parameters (a, bd, c and re2). If these
parameters were identical across developing countries, and countries chose their exchange

5
For the jth country, the real exchange rate disequilibrium, xjt, is equal to ( Q *  q*)j,t  1, and xj,t + 1  xjt=
[(1/r)Aejt]  ejt,where r is the number of participating countries. To avoid increasing variance of member
countries’ equilibrium exchange rates over time, we would have to add a mean-reverting element to Eq. (3). In the
present context, however, this is an unnecessary complication to the model.
238 M. Bleaney, D. Fielding / Journal of Development Economics 68 (2002) 233–245

rate systems at random, then we could test these predictions directly using the raw data.
The model predicts, however, that the exchange rate system will be selected by taking the
expected value of Eq. (1), which implies that, with identical parameters, all countries
would make the same choice. The fact that not all countries in practice make the same
choice indicates either that the theoretical model is incomplete or that the parameters are
not identical across countries (in truth, probably both of these propositions are true). In
particular, more open economies (higher c) and those subject to larger shocks (higher re2)
would be more likely to choose floating rates. We deal with this by estimating a regression
model which includes factors that we believe to be correlated with these variables among
the regressors, as discussed in more detail below. In the next section, we report the results
of testing the predictions of the model on empirical data.

3. Empirical findings

Our classification of exchange rate regimes is based on Table 1 of Ghosh et al. (1995).
In order to focus on the recent period and to avoid eliminating too many countries that
have undergone shifts in regime, we use data for the 10 years —1980 to 1989. We also
omit several outlier countries with inflation rates averaging over 50% per annum. This
leaves us with a sample of 80 developing countries, which are listed in Appendix A.
Macroeconomic data come from the World Bank World Development Indicators (1999).
Each country’s regime can be classified into one of the three exchange rate regimes
discussed in the previous section: floating rates, an uncoordinated peg, and a co-ordinated
peg (the CFA).6
Table 1 gives the unconditional means of the inflation rate, the standard deviation of the
output growth rate and the standard deviation of inflation (all in logs) for the pegged and
flexible exchange rate countries. The 52 pegged exchange rate countries averaged inflation
rates far lower than those experienced by the 28 flexible-rate countries. The standard
deviation of output growth is on average a little higher under pegged rates. The standard
deviation of inflation is, however, much higher in the flexible-rate sample. As we shall see
later, this is entirely explained by the strong association between average inflation and its
volatility, which reflects factors such as infrequent adjustment of government-controlled
prices (e.g., electricity), varying degrees of wage indexation and oscillations in macro-
economic policy. Once we correct for this, the relationship between inflation variability
and exchange rate regime looks rather different.
In the regressions for output and inflation that follow, we assume that whether the
exchange rate is pegged is a weakly exogenous variable. We test this assumption using a

6
Ghosh et al. (1995) use a much larger number of categories, allowing (for example) for the frequency of
realignments of pegged rates within any given year (although this information was not always available and some
countries remain unclassified in this respect). Since there are relatively few countries operating pure floating
exchange rate regimes, we incorporate ‘‘intermediate’’ regimes into the floating-rate category, in order to
maximise the sample size. This gives us 28 floating-rate countries (U to Z in Ghosh et al.’s categorisation) and 52
countries, which have had pegged rates throughout the period (A to T).
 M. Bleaney, D. Fielding / Journal of Development Economics 68 (2002) 233–245 239

Table 1
Unconditional sample means of dependent variables
Regime Mean Inflation S.D. Inflation S.D. Output Growth Cases
Float 21.403 (1.663) 10.273 (0.090) 4.060 (0.393) 28
Basket peg 9.769 (1.921) 5.469 (1.037) 5.174 (0.453) 21
Single currency peg 8.670 (1.581) 5.782 (0.085) 5.133 (0.373) 31
Cross-country standard deviations in parenthesis.

Hausman test, adding the residuals from a Probit model of regime choice to the
regressions. This regression is reported in Table 2. The dependent variable equals unity
if a country adopts a peg and zero otherwise. The instruments in the regression are binary
variables for whether the country had a pegged exchange rate in 1979 (‘‘PEG79’’), and
whether it was free of exchange controls (‘‘NKR79’’, indicating an exchange rate regime
not under pressure) in 1979. Since the Hausman t-statistics reported in the output and
inflation regression Tables 3 –5 are never significant, we can conclude that there is no
evidence that the results are biased by endogeneity of the exchange rate regime. Aside
from the instruments, the other significant explanatory variables in the Probit regression
are openness proxies included in the inflation and output regressions below and discussed

Table 2
Probit regression for exchange rate regime
Dependent variable: probability of a pegged exchange rate
Variable Coefficient S.E. t-Ratio
Constant 4.27540 2.46600  1.734
agr 0.01488 0.02098 0.709
srv 0.04447 0.02016 2.206
imp 0.00478 0.01410 0.339
exp 0.02860 0.01738  1.644
sdt 1.96360 2.40780 0.816
ipc 0.14540 0.32964 0.441
pde 0.05309 0.16330 0.325
inc  0.04640 0.09645  0.482
rpd 0.04994 0.14356 0.348
siz  0.34960 0.38253  0.914
CFA 7.02830 0.00000
NKR79 1.70200 0.81819 2.080
PEG79 2.26820 0.67286 3.371
SSA  0.13290 0.64630  0.206
ASP 0.49219 0.60357 0.815
MDE 1.20490 0.74597 1.615

log-likelihood =  33.184
correlation (actual, fitted) = 0.62746
Notes: agr — agriculture share of GDP; srv — service share of GDP; imp — import share of GDP; exp — export
share of GDP; sdt — standard deviation of log terms of trade; inc — log real GDP; ipc — per capita log real GDP;
pde — log population density; rpd — log rural population density; siz — log area; CFA — CFA dummy;
NKR79 — freedom from exchange controls; PEG79 — pegged exchange rate in 1979; SSA — sub-Saharan
Africa dummy; ASP — Asia-Pacific dummy; MDE — Middle East dummy. For further details see text.
240 M. Bleaney, D. Fielding / Journal of Development Economics 68 (2002) 233–245

in the next paragraph. Greater openness makes a peg less likely, in accordance with the
theoretical discussion above.
Tables 3– 5 present our regression results for mean inflation, inflation variance and
output growth variance in each country over 1980 – 1989. All t-statistics are adjusted to
allow for heteroscedasticity. In the first regression, we allow not only for exchange rate
regime effects (i.e., a dummy variable FLX for floating rate countries and a dummy
CFA for CFA countries), but also for the possibility of systematic differences across
regions. (The regressions include the dummies SSA for Sub-Saharan Africa, ASP for the
Asia-Pacific region and MDE for the Mid-East; all other countries in the sample are in
the Americas). We also allow for a number of structural economic characteristics from
which we have abstracted in the theoretical model. These are the percentage shares in
GDP of agriculture (‘‘agr’’), services (‘‘srv’’), imports (‘‘imp’’) and exports (‘‘exp’’); the
standard deviation of the log terms of trade over the period (‘‘sdt’’), average log real
income in billions of US$, both in total (‘‘inc’’) and per capita (‘‘ipc’’); density of the
total population (‘‘pde’’) and rural population (‘‘rpd’’) in log persons per square
kilometer; and log country size in millions of square kilometers (‘‘siz’’). A country’s
economic structure could influence mean inflation, and inflation and output volatility, in
a number of ways. The government of a more open economy (higher imp and exp,

Table 3
OLS regression for mean inflation rates
Variable Coefficient S.E. t-Ratio Coefficient S.E. t-Ratio
Constant 0.10683 0.07821 1.3660 0.16700 0.01643 10.1674
agr 0.00060 0.00081 0.7348
srv 0.00038 0.00059 0.6444
imp  0.00010 0.00043  0.3158
exp  0.00110 0.00051  2.1253  0.00080 0.00036  2.1603
sdt  0.00980 0.09522  0.1033
ipc 0.02992 0.02010 1.4888
inc  0.00710 0.00581  1.2254
pde 0.00955 0.00911 1.0479
rpd  0.00500 0.00749  0.6694
siz 0.01738 0.00963 1.8038 0.00845 0.00448 1.8874
FLX 0.39513 0.05064 7.8023 0.36949 0.03548 10.4152
FLX.inc  0.01300 0.00235  5.5508  0.01170 0.00167  6.9750
CFA  0.09420 0.02699  3.4902  0.07930 0.01685  4.7060
SSA 0.03122 0.03269 0.9549
ASP  0.06490 0.02309  2.8114  0.07670 0.01726  4.4442
MDE  0.01150 0.04693  0.2441

R-squared 0.63180 0.60401

r 0.07667 0.07387
RSS 0.37037 0.39832
SC  4.4441  4.9191
RESET test F (1,62) = 2.1365 F (1,72) = 3.7496
Heteroscedasticity F (27,35) = 0.78447 F (9,63) = 2.1548
Exogeneity t (62) = 0.349 t (72) = 0.122
See notes to Table 2.
 M. Bleaney, D. Fielding / Journal of Development Economics 68 (2002) 233–245 241

lower srv) might find it easier to raise (trade) taxes, leading to less reliance on inflation
seigniorage and so a lower b parameter in the model above and lower inflation. It might
also lead to a greater vulnerability to external shocks, leading to higher inflation and
output volatility, as might a larger value of sdt. Economically larger and richer countries
(higher inc and ipc) may also find it easier to raise taxes, because of scale economies in
fiscal administration, and may be more diversified, leading to lower volatility. Geo-
graphically larger or less densely populated ones (higher siz, lower pde and rpd) may
find it more difficult to raise taxes.
Table 3 reports both an unrestricted version of the mean inflation regression, and a
restricted version that minimises the Schwartz – Bayesian Information Criterion (SC). All
variables omitted from the second regression are insignificant in the first; their omission
does not greatly alter the size or significance level of the other variables. A linear
regression fails a standard RESET test, because inc interacted with the FLX dummy is
highly significant. This interacted term is included in the regression equations in the table.
At the sample mean income level (21.4), floating the exchange rate is estimated to
increase the inflation rate by about 12 percentage points. The significant interacted term
means that floating rates increase inflation even more than this in countries with income
levels below the mean (but even in the highest income countries, the overall effect of the

Table 4
OLS regression for output growth standard deviations
Variable Coefficient S.E. t-Ratio Coefficient S.E. t-Ratio
Constant 0.032093 0.02591 1.2387 0.02977 0.01277 2.3317
agr  3.30E-05 0.00025  0.1346
srv  3.62E-05 0.00019  0.1917
imp  0.000120 0.00025  0.4829
exp 0.000123 0.00023 0.5323
sdt 0.062888 0.02443 2.5743 0.07592 0.02396 3.1685
ipc 0.005741 0.00541 1.0614 0.00317 0.00156 2.0316
inc  0.000823 0.00155  0.5295
pde  0.005569 0.00266  2.0915  0.00660 0.00158  4.1754
rpd  0.000477 0.00231  0.2063
siz  0.002200 0.00268  0.8207  0.00310 0.00107  2.9169
FLX 0.025899 0.01748 1.4816
FLX.inc  0.001069 0.00079  1.3482
CFA 0.013903 0.00775 1.7936 0.01417 0.00628 2.2565
SSA 0.005379 0.00604 0.8906
ASP 0.001383 0.00689 0.2007
MDE 0.008504 0.01018 0.8358

R-squared 0.44160 0.40095

r 0.01909 0.01824
RSS 0.02295 0.02463
SC  7.2251  7.7574
RESET test F (1,62) = 0.0316 F (1,73) = 0.0128
Heteroscedasticity F (27,35) = 1.5376 F (9,64) = 1.3370
Exogeneity t (62) = 0.282 t (73) = 0.492
See notes to Table 2.
242 M. Bleaney, D. Fielding / Journal of Development Economics 68 (2002) 233–245

float on inflation is still positive). Also in accordance with the predictions above is the fact
that CFA membership yields some additional gain in terms of lower inflation, with an
effect of about 8 percentage points. The other significant effects in the regression are a
negative relationship between openness and inflation, and a positive one between
geographical size and inflation. Both effects are consistent with the priors indicated above.
Table 4 presents the regression results for output volatility. The only significant regime
dummy here is CFA. CFA members have a level of volatility that is about 1.4 percentage
points higher than other countries. This accords with our theoretical prior, but there is no
significant difference between the other pegging countries and the floating ones, which we
might have expected. Unsurprisingly, countries facing larger external shocks (as measured
by sdt) have a higher output volatility. So also do smaller and less densely populated
countries, perhaps because they are less economically diversified.
Table 5 presents the regression results for inflation volatility, controlling for the
strong correlation between volatility and the average inflation rate, indicated in the table
as (‘‘mean p’’). Controlling for this effect, inflation volatility is significantly higher for
the CFA countries, the effect being worth about 3 percentage points. Again, it is
adherence to a co-ordinated peg rather than to a unilateral peg that seems to make a

Table 5
OLS regression for inflation standard deviations
Variable Coefficient S.E. t-Ratio Coefficient S.E. t-Ratio
Constant 0.03367 0.043054 0.78204 0.03929 0.016791 2.34006
Mean p 0.39022 0.053803 7.25276 0.42843 0.045395 9.43782
agr 0.00015 0.000589 0.26145
srv 0.00019 0.000424 0.44946
imp  0.00057 0.000222  2.54345  0.00044 0.00012  3.48699
exp 0.00055 0.000248 2.21132 0.00042 0.00013 3.31888
sdt 0.06046 0.044114 1.37045
ipc  0.00958 0.009179  1.04317  0.00529 0.00237  2.23645
inc 0.00208 0.002302 0.90531
pde  0.00420 0.004819  0.87287
rpd  0.00208 0.003229  0.64278
siz  0.00776 0.004362  1.77912  0.00433 0.00172  2.52200
FLX 0.05785 0.028012 2.06519
FLX.inc  0.00233 0.001303  1.79044
CFA 0.02881 0.012012 2.39843 0.03226 0.00941 3.42935
SSA  0.01119 0.010640  1.05197
ASP  0.00608 0.013269  0.45838
MDE  0.01066 0.017710  0.60209

R-squared 0.70888 0.67602

r 0.03431 0.03336
RSS 0.07298 0.08122
SC  6.0136  6.5092
RESET test F (1,61) = 1.7718 F (1,72) = 0.3320
Heteroscedasticity F (29,32) = 0.7200 F (11,61) = 2.1152
Exogeneity t (61) =  1.271 t (72) =  1.054
See notes to Table 2.
 M. Bleaney, D. Fielding / Journal of Development Economics 68 (2002) 233–245 243

difference to volatility figures. Like output volatility, inflation volatility is negatively

related to country size. The coefficients on imp and exp (negative and positive,
respectively) suggest a positive relationship between inflation volatility and a country’s
Balance of Trade.7
Overall, the results are broadly supportive of our model. Given that the floating-rate
sample also includes some ‘‘intermediate’’ cases, and all pegged-rate countries other
than the CFA zone devalued during the period, the categories do not conform rigorously
to the theoretical distinction between floating and fixed rates. The CFA is an exception,
since the CFA franc was not devalued at all, but there are no polar opposite cases of a
pure float. So it is not surprising that the model works best for the CFA zone, which, as
predicted, experienced greater instability of output and inflation but lower mean
inflation. Other pegged-rate countries have significantly lower mean inflation than
floating-rate countries, but differences in output and inflation volatility are statistically
insignificant. As a referee has pointed out, our equations probably overstate the expected
inflation benefits of pegging for a given country, since countries with less distaste for
inflation (and which are therefore more likely to float in our model) would probably
have higher inflation than the average pegged country even if they chose to peg their
exchange rate.

4. Conclusions

The divergence between the median inflation rates of developing and advanced
countries since the early 1980s has attracted virtually no research interest. We have
explored the hypothesis that this divergence can be attributed to the inability of LDCs to
import the anti-inflation credibility of the advanced countries in the way that they could
under the Bretton Woods system, when virtually every country pegged its exchange rate to
the US dollar with only infrequent adjustments.
Our empirical results, based on data from 80 developing countries over the period
1980 –1989, are generally consistent with the theoretical model. After allowing for effects
such as differing variances of terms of trade shocks across countries, the chief predic-
tion —that there is a trade-off in the choice of exchange rate regime between inflation
reduction and the stability of output (and inflation) —is supported by the data. The results
are most clear-cut for the polar cases of floating exchange rates and the CFA franc zone,
which experienced no devaluation during the period. The CFA countries had significantly
lower inflation and significantly greater output and inflation variance than the typical
floating-rate country. These differences were all significant at the 1% level. Countries with
other types of pegged exchange rates displayed a similar pattern, also achieving much
lower inflation, but the difference in output and inflation variance relative to countries with
floating exchange rates was less marked. Compared with the exchange rate regime,
structural factors appear to play a relatively minor role in explaining cross-country

7
The coefficients on FLX and FLX.inc in the unrestricted model suggest that floating exchange rates lead to
higher inflation volatility, even after controlling for the mean-variance effect. However, these two variables are
not jointly significant at the 5% level, and the SC indicates that they should be excluded from the regression.
244 M. Bleaney, D. Fielding / Journal of Development Economics 68 (2002) 233–245

differences in mean inflation rates, but are more important in explaining both output and
inflation volatility.
Reinhart (2000) has recently argued that developing countries which ‘‘float’’ their
exchange rate are to a large degree secret peggers, as evidenced by the volatility of their
reserves and interest rates, and the relative stability of their exchange rates. This could
explain why, in our sample, floating countries do not have significantly lower output and
inflation volatility than countries with unilateral pegs. What our results show, however, is
that there is a substantial inflation cost to floating. The puzzle that remains, therefore, is
what benefits accrue to developing countries which float their exchange rates in
compensation for this cost. We leave this puzzle to further research.

Appendix A. Appendix

The countries included in the sample are as follows. An asterisk denotes that the
country was classified as having a flexible exchange rate.
Algeria Chile * Ethiopia India * Malaysia Philippines *
Suriname Paraguay Bahamas Colombia * Fiji Indonesia *
Mauritius Western Syria Zaire * Bahrain Sierra Leone *
Samoa *
Gabon Iran Morocco * Sénégal Tanzania * Costa Rica *
Bangladesh Zambia * Gambia * Jamaica * Burma Madagascar *
Thailand Zimbabwe Barbados Ghana * Kenya Côte d’Ivoire
Nepal Congo Togo Lesotho Belize Hong Kong *
Guatemala South Korea * Niger Rwanda Tonga Singapore *
Liberia Egypt Guyana * Kuwait Nigeria * Solomon
Islands *
Trinidad St. Vincent Burundi Ecuador Haiti Burkina Faso
Pakistan Somalia * Turkey * Sudan Cameroon Dominican
Republic *
Honduras Seychelles Panama Swaziland Uruguay * South Africa *
Sri Lanka * El Salvador Dominica Malawi Venezuela Central African
Republic
Vanuatu Papua New
Guinea

References

Alogoskoufis, G.S., 1992. Monetary accommodation, exchange rate regimes and inflation persistence. Economic
Journal 102, 461 – 480.
Barro, R.J., Gordon, D.B., 1983. Rules, discretion, and reputation in a model of monetary policy. Journal of
Monetary Economics 12, 101 – 120.
Bleaney, M.F., 2001. Exchange rate regimes and inflation persistence. International Monetary Fund Staff Papers
47, 387 – 402.
 M. Bleaney, D. Fielding / Journal of Development Economics 68 (2002) 233–245 245

Bleaney, M.F., Fielding, D.J., 1999. Exchange rate regimes, inflation and output volatility in developing coun-
tries, CREDIT Discussion Paper No. 99/4, School of Economics, University of Nottingham.
Burdekin, R.K., Siklos, P., 1999. Exchange rate regimes and shifts in inflation persistence: does nothing else
matter? Journal of Money, Credit and Banking 31, 235 – 247.
Collins, S.M., 1996. On becoming more flexible: exchange rate regimes in Latin America and the Caribbean.
Journal of Development Economics 51, 117 – 138.
Edwards, S., 1993. Exchange rates as nominal anchors. Weltwirtschaftliches Archiv 129, 1 – 32.
Ghosh, A.R., Gulde, A.-M., Ostry, J.D., Wolf, H.C., 1995. Does the nominal exchange rate regime matter?
Working Paper no. 95/121, Washington, DC: International Monetary Fund.
Obstfeld, M., 1995. International currency experience: new lessons and lessons relearned. Brookings Papers on
Economic Activity 1, 119 – 220.
Reinhart, C.M., 2000. The mirage of floating exchange rates. American Economic Review 90, 65 – 70.
Rogoff, K., 1985. The optimal degree of commitment to an intermediate monetary target. Quarterly Journal of
Economics 100, 1169 – 1190.