The Aftermath of Appreciations

NBER WORKING PAPER SERIES

THE AFTERMATH

OF APPRECIATIONS

Ilan Goldfajn Rodrigo O. Vald6s

Working Paper 5650

NATIONAL BUREAU OF ECONOMIC RESEARCH 1050 Massachusetts Avenue Cambridge, MA 02138 July 1996

We would like to thank Rudi Dornbusch, Pierre-Olivier Gourinchas, Peter Henry, Nancy Marion, Jaume Ventura, and participants of the NBER conference on Determination of Exchange Rates, seminars at Clark University and M.I.T. for valuable comments and suggestions and Paulo Porto and Caroline Kollau for research assistance. Of course, any remaining errors are our own. Vald6s would like to acknowledge support in the form of a Finch Fund Fellowship. This paper was presented at the NBER’s “Universities Research Conference on the Determination of Exchange Rates” and is part of NBER’s research program in International Finance and Macroeconomics and NBER’s project on International Capital Flows. We are grateful to The Center for International Political Economy for the support of this project. Any opinions expressed are those of the authors and not those of the Central Bank of Chile or the National Bureau of Economic Research. @ 1996 by Ilan Goldfajn and Rodrigo O. Vald6s. All rights reserved. Short sections of text, not to exceed two paragraphs, may be quoted without explicit permission provided that full credit, including @ notice, is given to the source.

NBER Working Paper 5650 July 1996

THE ~ERMATH

OF APPRECIATIONS

ABSTRACT

This paper empirically analyzes a broad range of real exchange rate appreciation episodes. The cases are identified after compiling a large sample of monthly multilateral real exchange rates from 1960 to 1994. The objective is twofold. First, the paper studies the dynamics of appreciations, avoiding the sample selection of analyzing exclusively the crisis (or devaluation) cases. Second, the paper analyzes the mechanism by which overvaluations are corrected. In particular, we are interested in the proportion of the reversions that occur through nominal devaluations, rather than cumulative inflation

differentials.

We calculate the probability of undoing appreciations

depreciations for various degrees of misalignment.

without nominal

The overall conclusion is that it is very unlikely

to undo large and medium appreciations without nominal devaluations.

Ilan Goldfajn Department of Economics Brandeis University PO Box 9110 Waltham, MA 02254

Rodrigo O. Vald&s Research Department Central Bank of Chile Agustinas 1180 Santiago CHILE

1

Introduction

One of the leading explanations exchange rate was previously against the currencies

overvalued.

the discussion.

the United Kingdom

real devaluation.

of overvaluation

or on its empirical

reintroduced

This would explain the market speculation

and the subsequent

not agree on the concept appreciation)

behind almost all exchange rate crises is that the real

(sometimes

counterpart,

Although

called misalignment

the magnitude

In 1992, the exchange

or just

of two recent crises

of the European

System and cast doubts about the success of the future European magnitude

of the Mexican exchange rate crisis and its implications

instability

obliged the US treasury

Monetary

Union. In 1994, the for global financial

and the IMF to mobilize a rescue package.

on whether

exchange rate overvaluation

was the main

cause behind each of these crises. There has a]so been some effort in identifying mon factors to exchange rate crises and major devaluations. countries chosen in these studies is not adequate For example,

the question

ated by 25% in real terns answered

of crises or devaluations

Power Parity

knowledge,

episodes.

no studies that focus on characterizing

The importance

of describing

easier to understand rate as an instrument easy focus point.

as a practical to stabilize

appreciations matter.

largely on her ability to maintain

More generally

but their focus is

there are, to our

appreciations. and the likelihood

and coordinate

In several cases, the credibility

bias

has been given to the likelihood

Several countries

inflation

(PPP)

be

revert to its mean and not on how

little attention

in appreciation

cannot

or crises only. 2 This sample selection

on whether the real exchange rate will eventually Surprisingly,

questions.

that a currency which has appreci-

will face a crisis or need a large devaluation?

does not exist in studies that test Purchasing

occurs.

com-

1 However, the sample of

to answer some important

what is the probability

with a sample of devaluations

this reversion

do

rate crises in Italy, Spain, and

affected the perceived sustainability

There is a vast literature

economists

of devaluation

is

have used the exchange expectations

around

an

of the policy maker seems to depend

the exchange rate peg. There are several current

1See Dornbusch, Goldfajn and Vald6s (1995); Eichengreen, Rose, and W yplosz (1994) and (1995); and Edwards (1989) for some recent attempts to characterize exchange rate crises and devaluations. All these studies find that the RER is overvalued during the period previous to devaluations. 2Klein and Marion (1994) study the duration of peg regimes in Latin America avoiding this sample selection problem. However, they do not address the questions we try to answer here. Interestingly, they conclude that the level of the RER is the main determinant of the duration of pegs.

2

examples.

In the context of developing countries Argentina

cases.

Argentina’s

economic

policy and credibility

sustain

the peg. After four years of higher inflation

exchange regime, the Argentinean if one takes for granted

revert to its PPP value, the question A nominal devaluation economic

would probably

at home than abroad in a fixed considerably

undermine outflows

the credibility

d la Mexico.

understand

important.

perspective

the dynamics

rate devaluation.

crises are self-fulfilling, of reputation

based stabilizations, credibility

Krugman

when the authority

usually accompany

an overvaluation plausibility

in building

and how persistent

has stressed

This paper empirically

currency

on exchange-rate

the importance

in that literature,

of the peg. 3 Knowing whether

explanation.

of imperfect

is defined as the

it is possible to correct

is a key step in evaluating

inflation

differentials

the

rather than nominal

of how rigid nominal prices are

is. analyzes a broad range of real exchange rate appreciation

we define appreciations

as PPP departures

in the short and

medium run. The cases are identified after compiling a large sample of monthly tilateral

that

Finally, the analysis of how likely

sheds light to the question

inflation

whether

exchange

boom and real appreciation

Credibility,

episode to end through

cases. For that purpose,

a model to discuss

a (large) nominal devaluation

exchange rate movements

to

how they are corrected.

decides to devalue. The literature

of the imperfect credibility

is an appreciation

their

(1996) assumes that there are real costs in terms

such stabilizations.

without

of how likely it is

given how appreciated

and especially

for the consumption

likelihood of the abandonment

for Argentinean

there are several reasons why it is important

on the other hand,

as an explanation

Thus,

(and some derive) real costs of a nominal

For example,

Peso will

The same is true in the case of Brazil.

of appreciations,

In fact, several models assume

Even

of the government’s

investors the question

to have a smooth landing (avoiding a large devaluation),

From a theoretical

in real terms.

or long run the Argentinean

(and public) and international

currency is, becomes extremely

largely on its ability to

of how this reversion will occur is still relevant.

policy and induce capital

policymakers

depend

Peso appreciated

that in the medium

and Brazil are interesting

real exchange rates from 1960 to 1994. The objective

paper studies the dynamics

of appreciations,

is twofold.

mul-

First, the

avoiding the sample selection of analyz-

3See Rebelo and V6gh (1995) for an evaluation of competing explanations of the stylized facts of exchange rate-based stabilizations.

3

ing exclusively the crisis (or devaluation) of appreciation distribution

cases that exist under different definitions,

and exchange

are as follows: duration

cases. In particular,

First,

rate arrangement

the most striking

of the appreciation

build-up

our sample period (1980–94). when fundamentals

their duration,

characteristics.

and the return-to-normality

are more likely to suffer appreciations.

are considered.

are corrected.

In particular,

(or cumulative

inflation

rather than through

differentials).

We calculate

defines real exchange

section trates

nominal price adjustments

4 Figure 1 shows a typical result.

real depreciation

also calculates

exchange

episodes.

of appreciation

that

Section 3 characterizes

ap-

Section 4 decomposes

the

that takes

differentials,

respectively.

adjustment.

Section

5 concen-

episodes and calculates

transition

matrices.

the probability y of successful

on the dynamics

framework

into the fraction of the adjustment

rates and inflation

Note

reaches 35% or more.

episodes across time and exchange rate regimes.

nominal

of successful ap-

as follows. Section 2 sets the theoretical

rates and overvaluation

return-to-equilibrium

of the reversions

the probability

that there are no successful cases when an appreciation The paper is organized

by which the over-

we study what proportion

for various degrees of appreciation.

place through

Second,

Finally, we also show that episodes are notably shorter

nominal devaluations

preciation

the

episodes happen more often during the last part of

occurs through

preciations

between

phases.

The second objective of the paper is to analyze the mechanism valuations

temporal

The main conclusions

result is the large asymmetry

we present evidence that fixed arrangements Third, we show that appreciation

we analyze the number

This

Finally, section 6 concludes.

2

Methodology

and Data

In order to analyze and interpret appreciation

movements

of the real exchange rate (RER) as an

episode one needs to define an equilibrium

out of steady state.

This is not an easy task.

main reasons that prevented lack of a consensus

around

previous

attempts

a sound empirical

concept

and the dynamics

In fact, we speculate to characterize counterpart

that one of the

overvaluations

is the

to any definition

of the

4We formally define the term successful appreciation in section 4. For now, we mean appreciations that end without large nominal devaluations.

4

0.35{1

Appreciation [Percentage)

Figure 1: Probability equilibrium

of Successful Appreciation

RER. The RER between two countries

a common bmket of goods measured

in terms of a common numeraire:

Pi is the price of the basket in country The equilibrium

is defined as the relative cost of PI /P2, where

Z.

concept we use is Purchasing

Power Parity - PPP, probably

simplest and most powerful theory of real exchange determination.5 Law of One Price which states that, abstracting

It is based on the

from tariffs and transportation

costs,

free trade in goods should ensure identical prices of these goods across countries. implies that the same basket of goods in two different countries

the

This

must have the same

price, or P1/Pz = 1. This paper denotes by overvaluation partures

in the short or medium

valued RER) can be thought currency

generates

or appreciation

run .6 The correction

to occur through

unsustainable

current

the episodes of PPP deof PPP

deviations

the following channel.

account

deficits through

(or over-

An overvalued the loss of com-

5See Dornbusch (1987) for an historical perspective. 6We ignore undervalued episodes. The emphasis on overvaluations in the literature and policy discussions is probably because prices and wages are flexible upwards. Presumably, undervaluations are less costly to reverse.

5

petitiveness.

The latter

also leads to possible

of these effects will work to adjust international

domestic

recession and losses of reserves.

prices expressed

(P.).8

to

levels. 7

In the definition of RER we can theoretically categories:

in foreign currency

All

Price of exports

(Pz), price of imports

disaggregate

the price levels in three

(Pm) and price of nontraded

goods

The RER is then defined as follows:

(1)

Taking logs and rearranging

e = {~(Pm

–P:)

+8(P.

–P;)}

we have:

+ {7(P.

+ {(~–

–Pi)}

(P– P’)Pj+ (7–7’)P:},

~’)P2+

or equivalently, e=

Departures

from Law of One Price + Relative price of nontradables+

Terms of Trade effect. The idea is that when the law of one price holds, ceteris paribus, pressure on relative prices (current account deficits will be optimal, in equilibrium).

This amounts

wages and prices

to: Pz =P~

We can abstract

there will be no

and

Pm =P~,

from the direct Terms of Trade effect if we assume that the weights

are not so different between the baskets.

If a – a’ = O, ~ – ~’ = 0 and ~ – y’ = (),

then we have:

e=

{Q(P* – p;) + B(Pz–

where we remain with only two components,

P:)}

+

{’Y(P.

–

Pl)}j

(2)

namely departures

from the law of one

7A fundamental issue for the interpretation of our paper is whether the

RER is a trend-stationary

stochastic process —that is if it tends to revert towards its mean. Recent studies have shown that this is indeed the case. See Froot and Rogoff (1995), Isard (1995) and Breuer (1994). ‘Here the subscript m (or x) represents the import (export) good in the home country which is, also, the export (import ) good of the rest of the world.

6

price and nontradables

price differences.

If we assume that differences in nontradable sures (as in the case of haircuts), considered components

in the overvaluation

then only differences in tradable

measure.

of the RER above.

Therefore,

One approach

ments in the term Y(P. – p:) of equation

these movements. 9 A second approach

mentals

by regressing

proportion

in the paper.

without

move-

is to control for the effects of fundathat

are related

to nontradable

We first follow the approach of regressing

taking into account fundamentals.

this would be the optimal

of nontraded

is to assume that equilibrium

the two

from the law of one price.l”

We follow both approaches the RER on time trends,

prices should be

one needs to disentangle

the RER on several variables

prices but not to departures

pres-

(2) occur slowly and that time trends will

capture

simple procedure,

prices do not exert reverting

approach

Besides being a

if the price index had a small

goods, and their prices change smoothly. 11 For each country

we calculate

Epr = a + T’~, where EP~ is the predicted time trends

value from the regression

(linear and square) denoted

Using the predicted from equilibrium

(3)

by T.

value as our equilibrium

are calculated

of the log of the RER on two

real exchange

as follows (normalizing

rate, the departures

the series to 100 when the

RER is in equilibrium):

(4) where E is the original series. ‘An example of these movements is the Balassa-Samuelson effect. When there is a productivity growth differential between the traded and nontraded goods sectors and this differential is not homogeneous across countries, then the (cross country) relative prices of nontraded goods, and therefore, the RER, will change over time. 100ne could argue that some of the fundamentals chosen may also be related to the departures from the law of one price. In this case, this second approach will tend to underestimate the extent of overvaluation. Since the first approach does not control for fundamentals and may overestimate the extent of overvaluation, one can interpret the resulting two series as defining the boundaries of the true overvaluation episode. 11Since the RER is trend-stationary this is a perfectly valid procedure.

7

2.1

Controlling for Fundamentals

We also follow the second approach.

Here we assume that nontraded

with movements

Thus,

in fundamentals.

we want to clean RER movements

changes in the term Y(pn – p;) of equation Operationally,

we calculate

prices do change from

(2).

for each country:

where X is the set of fundamentals. The fundamentals traded

we use to isolate the RER movements

from movements

of non-

good prices are the following: 12

Terms of Trade

(TOT)

TOT shocks affect the relative price of nontradables

small open economies. 13 If there is a positive nontradables

will increase with the increase in permanent

the relative price of nontradables If the shock is temporary, is small,

the demand

nontradables

permanent

shock,

the demand

income.

will rise and we should observe a real appreciation.

for nontradables

will not react, provided

will not increase

the supply is unchanged.

Otherwise,

national

and the relative

even temporary

and smooth

long run trends will be captured In the case of large countries

run.

TOT shocks can have an

Then, the optimal

the resultant

predicted

RER through procedure

values,

is to

In this way

and very short effects smoothed. there is an endogeneity

problem

are defined simultaneously

to the relative price of nontraded

Government

An expansion

Spending

price of

sector and decrease

effect on the RER. Here we assume that TOT affect the equilibrium supply effects only in the long and medium

income

This will be the case

whenever there is a fixed cost to move resources out of the tradable

net out the effect of TOT

for

In equilibrium,

and therefore the effect on permanent

the supply in the short run.

in

RER if it increases the overall demand

in government

for nontradables.

because the TOT

goods.

spending will appreciate

the

This will be the case if the

12One may consider capital inflows as an additional fundamental. However, these flows are just the counterpart of the current account plus reserves, and therefore, are simultaneously determined with the RER. For that reason, we chose not to include them in the regressions. 13See Edwards (1989).

8

government

propensity

to consume

When thepropensities

arethe

debt the effect depends

on how permanent

As a general

more temporary

the government

spending

Openness

looking

prices increases

proportionally)

the the

and the less forward

equivalence will not hold). We measure government expendituresto

reflects how connected

and stands here for trade liberalization. imports

is the shock and how forward

consumption

are(Ricardian

is financed by

shocks are (when the shocks are temporary

as the ratio of government

Openness

inexpenditures

rule, the effect on nontradable

sector will not decrease

looking consumers

is larger than the private sector’s.

same and anincrease

are consumers.

private

nontradables

GDP.

the economy is to the rest ofthe

world

It is proxied here by the ratio ofexports

plus

toGDP.

A trade liberalization

generates

market general equilibrium of a crowding-in

an equilibrium

perspective.

RER depreciation

from a labor

The decrease in tariffs generates the necessity

to restore full emplyment,

This, in turn, requires a reduction

in the

price of nontradables.14 Some transitory rium RER’s.’5

shocks to the fundamentals

In this case, because the regression

long run relationship the latter

we consider have no effect on equilib-

between the RER and fundamentals,

may generate

These movements,

in equation

false short term movements

however, will be unrelated

the

short run movements

in our “equilibrium”

to movements

der to minimize this effect, we smooth the predicted

(3) will capture

in

estimate.

in the actual RER. In or-

RER’s with a 12-month centered

moving average.

2.2

Episode Definition and Phases

Figure

2 presents

an appreciation our estimate

an example

of an appreciation

episode.

We define the start

case as the time when the difference between

of “equilibrium”

is equal or higher than

RER (the predicted

a certain

threshold

the actual

value from equations

RER and (3) or (5))

(e. g., 1570 or 25Yo). The appreciation

ends when this difference hits a second threshold

associated

with the existence of no

14See Dornbusch (1974). 15An example is given by a transitory positive shock to the terms of trade.

9

of

appreciation.

We define this second threshold

blips, an episode has to be sustained

as 5%. In order to control for data

for more than 2 consecutive

months

to classify

as such.

A

Real Exchange Rate

PredictedRER

Time

History Figure2:

Appreciation

We define four notable points: (ii) End, when the appreciation

Start

Peak

Episode Definition

End and Phases

(i) Stafl, when theappreciation disappears

(iii) Peak, when theappreciation

–i.e.,

is the highest,

ciation first reached 5%. An appreciation

hits the threshold,

the RER hits the 5% benchmark, and (iv) History,

when theappre-

episode is then defined as the Start-End

period. There are also two phases: Peak-End,

representing

History-Peak,

the return

representing

to a “normal”

the build-up

problem

and

level.lG

16The phases add-up to more than an episode because the latter does not include the buildup to the appreciation threshold. In characterizing episodes, we are interested in what happens conditionally on being appreciated, not just in general.

10

2.3

Data Description

The initial sample is given by monthly 1994 (39,060 observations). is equivalent

data of 93 countries

Because of missing values the actual sample size of RER

to 86.170 of the potential

sample

(when we include fundamentals

actual sample size falls to 73.6% of the potential

people in 1985, with monthly

Statistics.

We construct of bilateral

trade data from the United

The list of countries the multilateral

or imports).

is presented

in appendix

RER for each country

RER’s with those trading

either exports

with more than 1 million

price data from the International

(IFS), and with origin-destination

partners

the

sample size). The initial sample is

composed of countries in the Summers and Heston database

~ade

during the period 1960-

Financial Nations’

Yearbook of

D.

as a trade-weighted

encompassing

average

4% or more of trade (in

The weights are fixed and represent

1985, or the closest year for which data is available.

Statistics

the trade flows of

They are presented

in appendix

E,17 In order to minimize the effect of movements our empirical may contain

measure

of RER using WPI when possible.

a large proportion

of nontraded

little effect on competitiveness. the random

in nontradables

walk hypothesis

prices, we construct

Consumer

price indices

final goods in their index that

It is not surprising,

have

then, that it is easier to reject

when WPI are used in PPP

tests.

When countries

do

not have a reliable WPI series we use CPI. This is the case with some developing countries

(see appendix

also a higher inflation a mean-reversion

D for a complete

tend to have

that even these cases have

process. our RER construction

of an imported

intermediate

implies that for some countries and competitiveness. may not detect

Since these countries

than the average, we are confident

One caveat regarding component

list).

Although

good that is not produced

at home.

This

the WPI may not be a good proxy for their price level we do not control for these cases and, therefore,

some appreciation

how the RER returns

is that some WPI’S may have a large

we

cases, this should not bias our results regarding

to equilibrium.

17We checked for data errors in the original data using graphic methods. The price series of El Salvador for 1977 was geometrically interpolated from December 1976 and January 1978 because it shows a break in 1978 (the IFS flags the series as having a break and it shows deflation of 21% in 1 month). Missing values of price data of Ghana (Apr. ‘81–Jan. ’82), Iran (Jul. ‘86–Mar. ’89), and Kuwait (Jan. ‘84–Dee. ’84) were also interpolated.

11

In order to analyze the role of the nominal exchange rate and inflation differentials in the return-to-normal for each month

and country.

ment description Restrictions. rangement

phase of the RER one needs a nominal We construct

of the IMF annual

The report status

presents

as of December

report

presents ments.

describing

a description

Exchange

the principal

when the arrangement

a summary

to construct features

of the exchange

a monthly

statistics

a nominal

arrange-

and Exchange ar-

of changes during exchange

of the arrangements.

of the coding and summary

When the arrangement

Arrangements

of each year and a chronology

With this data on hand we construct

country.

this index using the exchange

for each country

that same year. We use this information ments database

exchange rate index

arrange-

Appendix

describing

C

the arrange-

index for each month

and

is a peg we use the respective nominal exchange rate;

is an unknown

basket we usually

use the nominal

exchange

rate with respect to SDR (in some cases we use the last peg); when the arrangement is floating we use the currency used in the last peg that was in place. la The data

for the construction

sources are the following: pleted with unit import Openness percentage

and export

(the ratio of exports

This section presents

plus imports

their duration,

temporal

to GDP) and government

spending

completed

(as

with the

Appreciations

of appreciation distribution

conclusions

can be summarized

asymmetry

between phases.

and the

Report data for 1993-94 when possible.lg

several features

we analyze the number

frequency

prices from the IFS for 1960-64 and 1993-94,

of GDP) are from the Summers and Heston database,

Characterizing

normality

has annual

Terms of Trade are from the World Bank Tables com-

World Bank’s Development

3

of fundamentals

cases that and exchange

of the appreciation

we present

evidence

In particular,

exist under different definitions, rate characteristics.

as follows: First, the most striking

the duration Second,

of the episodes in our sample.

that

build-up

The main

result is the large and the return-to-

fixed arrangements

are more

18We classify target zones with a width less than 7% as pegs, We classify crawling pegs, managed floating, and periodic adjustable pegs aa flexible arrangements and use the underlying nominal exchange rate for our index. lgIn order to use these data in monthly regressions we interpolate the yearly data using June as the base month.

12

Third,

likely to suffer appreciations.

we show that

appreciation

more often during the last part of our sample period (1980-94). that episodes are notably

3.1

shorter when fundamentals

episodes

happen

Finally, we also show

are considered.

Number of Appreciations

The number

of appreciations

episodes that exist in our sample depends

cutoff that defines appreciations and the method

(the threshold

that defines the start date in figure 2)

we use in defining the “equilibrium”

(3) or (5)). Table 1 presents

RER (the RER~T in equations

these results.

Table 1: Number of Appreciation Apprec.

Cutoff

(percentage)

As expected, example,

Episodes

RER Estimate Trends Only

Fundamentals

15

173

158

20

111

91

25

71

56

30

52

34

35

36

20

the number

of episodes

declines with the appreciation

there are only 36 and 20 cases that had an appreciation

Also, there are less cases when we take into consideration in the equilibrium

RER estimation.

episodes that were previously considered

equilibrium

on both the

The methodology

detected

cutoff (for

larger than 35%).

the effect of fundamentals

disregards

some appreciations

because their actual RER movements

changes (given the movement

of fundamentals)

are now

.20

20We also get some new episodes because of movements in fundamentals. We smoothed the predicted RER in order to minimize the number of “false” appreciation ewes —the ones driven by excess movement of our equilibrium values. See section 2.1.

13

3.2

Duration

The average duration appreciations

and whether

different between

out controlling

incomplete

and Peak-End

defines

Moreover, duration

is very

In what follows we will

of15%and

25%, with and with-

Table 2 presents the statistics

of average duration

cases.

of Appreciations

Entire Episode

(Months)

History-Peak

Peak-End

Trends - 15%

22.2

19.5

11.1

Trends - 25%

22.8

26.8

11.1

Fundam.

- 15%

11.2

10.2

6.8

Fundam.

- 25%

8.5

12.3

4.6

The average duration is about

approximately

of appreciations

2 years.

also holds inthe

average duration

period.

produced

We also present

to estimate

the average duration

of shorter duration

History-Peak

of the Peak-End

of the History-End

using only time trends

Using fundamentals,

1 year. This pattern

fundamentals

equilibrium

that

phases.

thresholds

Table 2: Average Duration

equilibrium

on both the threshold

are considered.

cases: appreciation

for fundamentals.

including

depends

fundamentals

the History-Peak

focus on4 benchmark

in months,

of appreciations

and Peak-End

phases.

phase is approximately

Interestingly,

one half of the duration

Of course, behind this difference is the sudden return

the frequency

histograms

considering

hand, present the histogram

of duration

of the History-Peak

of our benchmark

to

is independent

of duration

of the threshold 14

threshold

phases duration

with

A present the cases for an appreci-

hold.

and the come-back phases.

cases.

Figures 5 and 6, on the other

and Peak-End

Figures 19 to 22 in appendix

Peak phase spreads over all categories last conclusion

fundamentals.

of 2570. The same conclusions

between the build-up

the

by nominal devaluations.

of 1570, with and without

ation threshold

drops by

when one takes into account

Figures 3 and 4 present the cases of entire episodes given an appreciation

the same threshold.

the

Duration

is highly asymmetric

The higher duration lwting

of the History-

more than 4 months.

and whether

fundamentals

This

are con-

sidered.

Also, including

fundamentals

reduces the duration

of the episodes (not only

the average duration). A final question regarding duration

is what happens with incomplete

cases, that is,

cases that remained being an episode when the data of the respective country ended. If these cmes had significantly

longer durations

than the complete episodes, there would

be evidence that they are of a different nature, namely equilibrium picked-up

by trends

and fundamentals)

appreciations

(not

that only in the long run would disappear.

Table 3 shows the average duration

(and number of episodes) of such cases. The main

conclusion

are almost

complete

is that these durations

Episodes

of Incomplete

(number)

Appreciations

History-Peak

Peak-Incomplete

Trends - 15%

15.1 (16)

16.6

7.8

Trends - 25%

11.4 (5)

20.4

6.8

Fundam.

- 15%

7.4 (8)

11,6

2.3

Fundam.

- 25%

9.0 (3)

17.7

1.7

Temporal Distribution

Several structural distribution capital

of

ewes.

Table 3: Average Duration

3.3

always smaller than the durations

changes

of appreciation

mobility,

in the world economy episodes.

and exchange

the temporal

Among other factors, changes in inflation levels,

arrangements

during some periods. ‘1 The presumption of appreciations

may have affected

may have produced

bunching

of cases

is that the first two have raised the likelihood

during the second part of our sample, while the movement

towards

more flexible exchange regimes may have decreased it. Because our panel data is unbalanced than others— the temporal

—some countries

the simple time path of number distribution

of cases. Instead,

have more observations

of cases is a misleading

indicator

we present the ratio of episodes to total

21See appendix C for a description of exchange arrangements during our sample period.

15

of

Appreciation Threshold = 15% 0,2-11

I............................. ... ... ... ,... ... — 1

. Duration (Months) Figure

3: Histogram

of Duration

- Trends

Only

Appreciation Threshold = 15% 0.35 0.3- ‘ 0.25- ‘ VY ~ $

0.2- ‘

z “E () 15. . g“ &

o.1- ‘ o.05- ‘ o- f Duration (Months)

Figure 4: Histogram

of Duration

16

- Fundamentals

Appreciation Threshold = 1590 0.45 0.4 0.35

0.3 0.25

0.1

0.05 0 Duration (Months) I=

History-Peak _

Figure 5: Phwes

Duration

Peak-End

I

- Trends Only

Appreciation Threshold = 15% 0.5 0.45 0.4

..................................................................................... I..............>, ...................................................................I

~.................................. .......m y+

~ 0.25

............................................................... “E 0“2 E 0.15 0.1 0.05 0

1-3

4-6

Im

7-12 ‘ 13-18”19-24”25-36”37-48 Duration (Months) Histow-Peak

_

Figure 6: Phases Duration

17

‘ 48+

Peak-End

- Fundamentals

countries

in the sample with data grouped

every 5 years. Cases are dated using the

date of Start ,22 The results for the benchmark 15% is presented

cases with an appreciation

in figures 7 and 8. The cases with a threshold

in figures 23 and 24 in appendix

threshold

of

of 25% are presented

A.

The graphs show that towards the second part of our sample the number of cases clearly increases. cases as during

In fact, during the period 1980–94 there are at least twice as much 1960–75 (controlling

cases with RER’s after controlling trend.

Interestingly,

of episodes around

3.4

episodes).

The

show an even more clear upward

when only trends are considered,

there is a notorious

bunching

1980–85.

of exchange

some countries

C describes statistics

for fundamentals

of potential

Exchange Arrangements

The overall trend though

for the number

arrangements

is towards

more flexible systems,

have changed their systems back to fixed regimes.

our characterization

of exchange

arrangements

al-

Appendix

and presents

summary

for our sample period.23

In order to evaluate whether cific exchange rate arrangements ment during

the episodes

appreciation

we compare the proportion

(more specifically

phases) with the proportion

episodes happen more often under speof each type of arrange-

during the History-Peak

and Peak-end

of each type observed in the total population.24

of the trends issues discussed above, a total average would be misleading ber of appreciations have declined.

has increased

Because

for the num-

over time and fixed exchange rate arrangements

In order to control for this problem we compare the proportion

type of arrangement

of episodes grouped every 5 years with the population

during those same 5 years. using the actual number

We then calculate

a weighted

of episodes that occurred

date of the episodes is assigned according

of each

proportion

average of this indicator

during those same 5 years,

to the Start date.

Table 4 presents

The these

22Notice that this ratio is not immune to composition effects. An example is given by developed countries having more data, and being less likely to suffer appreciations. 23Using a panel of annual data, Ghosh et al, (1995) study the impact of exchange arrangements on inflation and growth. They conclude that fixed regimes have less inflation and that the arrangement is unrelated to growth. 241ncases in which episodes have more than one arrangement we calculate the episode’s proportion of each arrangement according to the number of months each arrangement ww in place.

18

Appreciation Threshold = 15%

5 Yew Period

Figure 7: Temporal

Distribution

- Trends Only

Appreciation Threshold = 1590 0.45f’1

5 Year Period Figure

8: Temporal

Distribution

19

- Fundamentals

results.

Table 4: Exchange Arrangements of Appreciations Proportion of Each Arrangement Trends

Trends

Fundam.

Fundam.

15%

25%

15%

25%

History-Peak

Fixed

68.9

79.3

74.1

74.4

Phase

Flexible

24.5

17.9

21.6

22.1

Floating

6.6

2.8

4.4

3.4

Dual-Mult.

32.3

40.1

37.0

54.8

Peak-End

Fixed

63.1

71.9

68.6

65.7

Phase

Flexible

29.2

24.4

25.0

26.9

Floating

7,7

3.6

6.3

7.4

Dual-Mult.

35.8

48.2

36.2

50.9

Total

Fixed

62.0

60.8

61.5

59.7

Population

Flexible

31.3

32.4

31.9

33.3

Floating

6.7

6.8

6.7

7.0

Dual-Mult

16.9

17.1

17.5

17.5

Average calculated

every 5 years and weighted by the number of

episodes every 5 years. Number represents

The results preciations.

show that,

percentage

of time of each arrangement.

as expected,

fixed regimes are more likely to suffer ap-

This effect is higher when larger appreciations

regimes are less likely to suffer appreciations. adjustable

bands, adjustable

and multiple countries

exchange

are considered.

These regimes include crawling pegs,

pegs to baskets, and managed floating,

rates,

the results

have these arrangements

show that

during

at least twice as many as in normal

as implying

ity of appreciating.

However, in this case the reverse causality countries

In terms of dual

appreciations

could be interpreted

an episode starts

that dual-multiple

episodes, times.

This

regimes have a higher probabilalso exists.

are more likely to put in place dual markets

20

Flexible

When

in order to

improve the competitiveness

of certain sectors.25

There are clear asymmetries History-Peak

and Peak-End

rate arrangements happens

in the exchange arrangements

phases.

is notably

In particular,

arrangements.

notion that the return-to-equilibrium These results

fundamentals

The contrary

This fact gives support

of appreciation

Dual systems

during the

of fixed exchange

period,

is more easily accomplished

hold independently

are included.

the proportion

larger during the History-Peak

with flexible and floating

regimes,

prevailing

do not appear

to the

by flexible exchange

thresholds

and whether

more likely during

either

phase.

4

Nominal

Exchange Rate-Inflation

Decomposi-

tion in this paper is whether

One of the basic questions

there are appreciation

that do not end in exchange rate collapses or large devaluations,

episodes

More specifically,

one can ask how do appreciation

episodes end: Is the inflation differential —prompted

by the loss of competitiveness—

enough to return to the equilibrium

of the total work is done by the nominal questions

we constructed

a monthly

This index follows the movements changes in the currency baskets of currencies respect currency

exchange

nominal

rate?

a country

to which the peg is established. target

In order to answer these

exchange rate index for each country.

of the pegs that

are the nominal

RER? How much

In cases in which unknown

we use the nominal

the SDR. In cases of flexible and floating

may have, including

exchange rate with

regimes we use the price of the

last used as a peg.

In order to decompose equilibrium

the real depreciation

we calculate the total depreciation

that occurs during the return to the of the actual RER during the Peak-End

phase, and the total nominal actual depreciation appreciations

during that same period.

Successful

can then be defined as episodes that require less than a certain threshold

in order to return to the equilibrium. 26 Letting A denote percentage

change we have

25This effect also means that, in these cases, the nominal exchange rate used to calculate the real exchange rate loses its relevance. 26There is an important issue regarding appreciation cases that happen after a “structural” break in the equilibrium RER. Our methodology does not allow for such changes, so we count this break

21

the identity: AE = ANom + A (P – P“) where Nom is the nominal

exchange rate index and P and P* the price indices.

We

can then calculate s=l-A:;m as our successful index.

4.1

Detrended

4.1.1

Successful

RER

Index

Distribution

A first issue to analyze is the distribution

of our successful

indicator

S.

Knowing

this distribution

will allow us to measure how sensitive the definition

of successful is

to the threshold

for S. In particular,

successful,

threshold

if very few cases are partially

one chooses is not crucial.

Figures 9 and 10 present the histograms cases using the first methodology a large mass of cases that

(trends

of the S indicator as the equilibrium

inflation differential was partially Finally,

concept).

devaluation

does

successful cases —the

does all the work. There are few cases in which the appreciation

comparing

figures 9 and 10, we observe that

there is less probability

Searching For a Critical

Knowing the distribution

when larger appreciations

of success (for any S), There is less mass on

or close to S = 1 in figure 10, where the threshold

is 25Y0.

Cutoff

of S we can now search for the critical level of apprecia-

the level at which a successful episode is very unlikely to happen.

(arbitrarily)

We observe

successful.

are considered,

tion:

for our two benchmark

are not successful at all —the nominal

more than all the work.27 There is also some mass in totally

4.1.2

the

a successful appreciation

We define

when the nominal exchange rate does less than

as an episode (that has an end). The key is that if this is the case, then the RER during the whole episode is not under any pressure and nominal devaluations should not occur. This biases our results towards observing successful cases. ZTInflation differentials may have a negative contribution to the return. In this cases, nominal devaluations do more than all the work.

22


Work done by prices duringPeak-End

Figure 9: Histogram

of Success - Trends Only (15%)

Appreciation Threshold = 2590

..................................................................................... ~f $ . . .. . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. ~

------------------------------------------------------------------------------------

.,,

0.00- ‘ .00-.25 ‘ .25-.50 ‘ .50-.75 r .75-1.0 ‘ Work done by prices duringPeak-End

Figure 10: Histogram

1.0+

of Success - Trends Only (25%)

23

.

Cases with Success Index >.50 ““~

Figure 11: Probability


half of the work (S > 0.5). Since the mass of partially

- Trends Only

successful cases is small our

conclusions

do not critically depend on the successful definition.

probability

of success for different appreciation

as one case, regardless of its duration. between degree of appreciation, When appreciations successful between

—that

4.1.3

This probability

without

a

of success). only 10% of the cases are

5070 of the observed clearly decreases

devaluation;

real depreciation

with the appreciation

say 25% or more, it is unlikely

sooner or later a nominal

exchange

is required.

Successful

This subsection caused less than relatively

time, and the probability

is that for large appreciations,

to undo an appreciation rate correction

(In section 5 we explore in more detail the link

is they devalue less than

level. The conclusion

levels. Here each episode is considered

of 25% or more are considered

Peak and End.

Figure 11 shows the

Episodes:

Description

describes the appreciation half of the total

episodes in which the nominal devaluation

real depreciation,

so they can be considered

successful cmes. The initial sample includes appreciations

24

as

of 25% or more,

with respect to the trend RER.

Table 5: Successful Appreciation Country

Start-Date

Episodes

Duration

Actual

Actual

Fixed

Estimated

(months)

Build-up

Deprec.

X-Arr

Build-up

Paraguay

oct.’77

5

22.6

32.3

1.0

25.1

Nepal

oct.’72

2

22,1

36.5

1.0

27.0

Sri Lanka

Aug.’7O

86

-9.7

196.0

0.7

37.3

Sri Lanka

Feb.’94

9

14.0

5.4

0.0

25.1

Burundi

Feb.’85

16

23.4

40.9

1.0

27.5

Ethiopia

Aug.’84

22

35.0

55.1

1.0

37.7

Nigeria

Jan.’6O

45

-4.1

1.0

34.1

Success if S >0.50-

Appreciation

The list shows that these countries siders medium Notably,

all have fixed exchange

fixed arrangements

= 25%

are not typical appreciation

and large size countries

almost

arrangements

Threshold

the probability

of success is even smaller.

arrangements.

This does not mean that

should be kept in place, for the probability

is small.

cases; if one con-

The key policy recommendation

of success of these

is to avoid the appreciation

in the first place (or at least weight its benefits with the high probability

of future

devaluation). Finally,

notice that

appreciation

during

return-to-normality appreciations

4.2

RER

a couple of successful

the build-up phase.

mends

period

episodes

or an actual

do not suffer an actual

real

real depreciation

the

in the RER make these cases to be identified

and Fundamentals RER calculated

with

none of the conclusions change. Moreover, the conclusion regarding how

difficult it is to undo appreciations no experiences

as

under our definition.

If one repeats the exercise of the last section using the predicted fundamentals

during

without nominal devaluations

of successful episodes if appreciations 25

is stronger:

there are

of 3570 or more are considered.


....................................................................................

“,

0.00- ‘ .00-.25 ‘ .25-.50 ‘ .50-.75 ‘ .75-1.0 ‘

1.0+

‘

Work done by prices duringPeak-End


of Success -Fundamentals

(15%)

Figures 12 to 13 show these results.

4.3

Conditional Probabilities

This subsection

reports

ferent characteristics. of the sample,

probabilities

Table 6 presents

1980–1994, the probability

lower than in the first period. ful appreciations

of successful the results.

appreciations

of successful appreciations

of the episodes.

Third,

floating regimes are less prone to return to equilibrium

5

on dif-

First, during the second period

Second, there is no apparent

with the duration

conditional

pattern

is substantially relating success-

as expected,

through

flexible and

price changes.

Degree of Overvaluation and Transition Matrices

One of the objectives a certain

of this study is to identify the probability

period of time, for various levels of appreciation.

26

of RER reversion in

In particular,

we would

Appreciation Threshold = 25% 0.8 0.7 0.6 0.5 I

Work done by pricesduringPeak-End


of Success - Fundamentals

(2570)

Cases with Success Index >.50 0.35 o.~

0.25 J %

0.2

0.05 c

AppreciationThreshold(7.) Figure 14: Probability


27

- Fundamentals

Table6:

Probability

of Success-Different (Percentage)

Sampling

Trends Only Apprec.

Total

Float and

Start after

Long

Threshold

Sample

Flexible

1980

Duration

15

22.5

8.3

14.9

21.5

20

12.6

6.3

10.5

15.7

25

9.9

5.6

6.1

9.6

30

5.8

0.0

2.6

7.9

35

5.6

0.0

3.6

6.9

Fundamentals Apprec.

Total

Float and

Start after

Long

Threshold

Sample

Flexible

1980

Duration

15

32.3

2.2

16,8

38.0

20

24.2

3.8

13.6

33.3

25

10.7

0.0

5.3

16.0

30

2.9

0.0

3.8

5.6

35

0.0

0.0

0.0

0.0

Long duration

= End – Start

28

>6 months.

like to test the assertion correlated

that the probability

of returning

to equilibrium

to the degree of appreciation.

Reconstructed the probability

episodes,

transition

matrices forour

appreciation

cases. The matrices show

of reaching a specific exchange rate value conditional

of appreciation.

defined using the benchmark probability

Table 5 presents

cutoff of 15%. Therefore,

the results for the case of trend RER. There are two points to

from the table.

First, there is a high degree of inertia

diagonal

terms (shadowed

for contrast)

this is a consequence

although

show substantially

of the relatively

In fact, the transition

table for 24 months

the appreciation

but a high probability

of returning

As expected,

but also other transition

the longer the period considered,

With 48 months,

The more interesting

appreciation

degree

the whole

cases and get more unlikely

to an appreciation

for example,

the probability

96Y0. This confirms the latest PPP mean-reversion probability

of reversing

figures.

as

of return.

times as 1, 3, 24 and

of return ranges from 80 to

curve obtained

where increasing

for the

the level of

The reason for the nonlinearity

is the existence of a trade-off between distance and pressure factors. in figure 15 is plotted

matrices

results in the literature.

there is a threshold

implies a higher probability

5%

the higher the probability

and relevant result is the U-shaped

It shows that

of less than

It plots the last column of the transition

above (for 6 and 12 months),

of return.

30% in ta-

deepens.

15 plots the probability

48 months,

along the diagonal.

are achieved (for instance,

in large appreciation

for several levels of appreciation. described

A) shows

(O,24 to reach a value lower than 570). This result shows that smooth

are highly improbable

Figure

In

6 and 12

of moving to a slightly lower appreciation

(in this case 0.05 to reach 20-25%),

returns

All the

time shown:

(shown in appendix

we still observe higher probabilities

ble 5), there is a low probability

in RER’s.

higher probabilities.zg

short transition

Second, once high degrees of appreciation

appreciation

show

in the past.

highlight

lower inertia,

the matrices

of reaching a specific exchange rate value once a country

has surpassed 15% appreciation

months,

on a given degree

The overall sample is the exchange rate values of the appreciation

the conditional

part,

is positively

fixing the time period available to return,

Since each curve it is reasonable

to

28There is a substantial larger mass in the diagonal term of appreciations of equal or higher than 3070. However, there is also more support in this area.

29

Table 7: Transition

Matrices

of Appreciations

Detrended Appreciation

-6 and 12 Months

RER

Threshold

6 Months

= 15%

Matrix

RER Appreciation in t+6 months 30-25 ~ 25-20 0.07

! 0.05

20-15

15-10

10-5

5-

0.02

0.02

0.01

0.24

~ 30-25 ‘ 25-20

0.23

0.19

0.08

0.20

20-15

0.02

0.18

‘ 15-10

0.00

0.26

i =

0.00

0.50

10-5

12 Months

Matrix

RER Appreciation in t+12 months

I 30+ *. 30-25 25-20 20-15 15-10 {1 0.06 M

0.04

0.03

0.03

10-5 ~ 50.01 0.441

0.11

0.08

0.05

0.02

0.36

0.12

! 0.36

b. 20-15

0.04

0.07

15-10

0.01

0.01

0.04

0.01

0.00

0.03

10-5

i

~ 0.03

30

0.09

~]jlm

0.74

o

m

o \