Gsdeemer Third Edition: Alberto Ades, Rumi Masih, Daniel Tenengauzer September 1999

Download as pdf or txt
Download as pdf or txt
You are on page 1of 16

Economics

Research from the


GS Financial WorkbenchS M
at https://www.gs.com

GSDEEMER
Third Edition

Alberto Ades, Rumi Masih,


Daniel Tenengauzer
September 1999
Important disclosures appear at the end of this document.

GSDEEMER Third Edition

INTRODUCTION

Finally, we have extended GSDEEMER to allow


speeds of adjustment to equilibrium to differ between
crisis and tranquil periods.

GSDEEMER serves as our principal valuation tool


and forms the core-modelling framework that we use
in forecasting a set of 27 emerging market currencies.
After introducing it in June 1996 as our preferred
methodology for estimating long-run sustainable
equilibrium exchange rates for emerging-market
currencies, the model has gone through a number of
refinements. These revisions are outlined in previous
versions of this publication as well as in separate
1
releases . In light of important changes in emerging
market exchange rate behaviour over the past two
years, as well as continued academic progress on
econometric theory and its applications to exchange
rates, we have updated and refined our GSDEEMER
methodology further. Our objective, as always, has
been to establish a superior model that delivers more
reliable and stable estimates of long-run exchange
rates.

We hope that these innovations and the expanded set


of analytical tools will provide clients with a better
understanding of the factors driving exchange rate
behaviour in emerging markets in the long run.
Furthermore, these tools should improve the quality
of our valuation estimates, and better supplement our
risk assessment model, GS-WATCH.
The following section, after reviewing the basic
GSDEEMER framework, provides an intuitive
appraisal of the methodological innovations
introduced in this edition of GSDEEMER and
presents a new set of fitted values, misalignments
associated with the new estimated regressions. We
then assess the forecasting ability of GSDEEMER as
a long-run valuation tool.

In this chapter, we have introduced several


extensions and modifications to our Second Edition
GSDEEMER, released in the 1997 edition of this
publication. These modifications have sprouted
from work conducted over the past 12 months and
aim to improve the quality of GSDEEMER in a
number of ways. Firstly, we have treated global real
interest rates as strictly exogenous in the model. This
means that the real exchange rate and any other
variables used to drive the long-run equilibrium
exchange rate no longer have any influence on global
real interest rates. This assumption is justified
theoretically, since global real interest rates are
typically taken as given by small emerging-market
economies. The results of treating this variable as
exogenous are numerous, one of these being that we
now obtain slightly different estimates for both the
estimated equilibrium values as well as the
elasticities of other variables explaining the real
exchange rate.
The second innovation is a
methodological one: whilst we have allowed for
short-run dynamics to influence the estimates of the
elasticities of equilibrium exchange rates to long-run
fundamentals, we no longer take those lags into
account in calculating the estimates of long-run
equilibria. This has enabled us to fine-tune our
estimates better, as well as reduce their volatility.
1

Recap on Theoretical Model and Methodology


GSDEEMER relies on a standard theoretical model
of a small open economy. Various economic
fundamentals govern the determination of the
long-run equilibrium real exchange rate in such an
economy. The typical model postulates that, over the
long run, the real exchange rate is determined as
follows:
RERt = f(TOTt,OPENt,FISCALt,RPRODt,RLIBt)
where (excluding time subscripts) RER is the trade
weighted real effective exchange rate, TOT is terms
of trade, OPEN is the openness of the economy to
trade, usually measured by the sum of exports and
imports as a proportion of nominal GDP, FISCAL
refers to variables such as government consumption,
or public investment or a combination of both as a
share in GDP, RPROD is relative trade-weighted
productivity, and RLIB is 3-month US LIBOR
deflated by the US consumer price index.
The econometric techniques that we employed were
based on co-integration analysis, which we used to
estimate the functional relationship specified above,
and involved many separate steps. Firstly, we

For example, see GSDEEMER and STMPIs: New Tools for Forecasting Exchange Rates in emerging markets (Oct. 1996); GSDEEMER : Second
Edition (Sept. 1997); GSDEEMER: New Estimates on Revised Measures of Relative Productivity (Sept. 1998).

S.02

September 1999

analysed the time series characteristics of each and


every variable given by the functional relationship,
and classified them into stationary (mean-reverting)
and non-stationary series. Secondly, we tested
whether the set of variables co-integrated in the sense
of Johansen (1991), which allowed us to question
whether a long-run relationship existed between
these variables. Thirdly, to answer the question of
how to know which variables enter the long-run
relationship determining the real exchange rate, we
used a procedure known as long-run structural
modellingor LRSM, developed by Pesaran and Shin
(1997). LRSM uses economic theory rather than a
purely mechanical and statistical basis to precisely
isolate those variables relevant in explaining the real
exchange rate over the long run. This is in contrast to
short-term speculative factors or what we refer to as
dynamics arising from transitory shocks to
fundamentals which may cause the real exchange
rate to deviate from its long-run path to equilibrium.
Having done this, we finally estimated the
relationship to obtain parameter estimates. The
estimator we used is dynamic OLS or DOLS
developed by Stock and Watson (1993). We chose
this estimator because it simultaneously deals with a
number of deficiencies common to estimators used in
the previous literature. Amongst a number of
favourable statistical properties, the DOLS estimator
is robust to biases arising from small samples, which
is very typical of emerging markets databases.

Theoretical Innovation: Importance of Treating


LIBOR as Exogenous
While it is important to classify variables into those
that influence the real exchange rate in the long run
and those that only play a significant short-run role, it
is equally important to classify variables into
exogenous and endogenous. Unlike most of the
explanatory variables, it can be generally assumed
that the influence of US interest rates is truly
2
exogenous.
Small, emerging-market economies are unlikely to
influence this variable since they are all price takers
in the global capital market. By contrast, other
variables such as the terms of trade, openness, fiscal
and relative productivity do influence each other and
are also influenced by the real exchange rate. In this
sense, they may share feedback effects.
What difference does treating LIBOR as exogenous
make to GSDEEMER? First, it has direct relevance
to whether LIBOR is included in the GSDEEMER
long-run specification. For many countries, LIBOR
was included in the specification, but treating it
exogenously could mean it should no longer enter,
and vice versa. Second, including irrelevant
variables in the long-run relationship leads to
coefficient inefficiencies and severe small sample
bias this is particularly pertinent in the case of
GSDEEMER, where data for many countries are
limited to relatively short samples.
Finally,
excluding a relevant variable may lead to bias due to
omitted variables, which could lead to violations of
assumptions governing the underlying model and
bias in the forecasts.

Using the long-run set of estimated coefficients, we


constructed a series of predicted equilibrium real
exchange rates and estimated the misalignment
series as the difference between the actual and
predicted real exchange rate. In doing so, we arrived
at country-specific misalignments based on long-run
fundamentals drawn from a potential set
pre-specified by economic theory.

Methodological Innovation: Reducing Volatility


of GSDEEMER Estimates
One fundamental aim in using GSDEEMERs is to
gain a reliable assessment of the true long-run
equilibrium exchange rate. It is important to
periodically assess whether movements in the data
series are indicative of changes in long-run
fundamentals. Put another way, we would ideally
want to ensure that short-run or cyclical influences
are filtered out as much as possible in our estimates of

This was not such a trivial issue to contend with, partly because LIBOR is also non-mean reverting. The presence of exogenous non mean-reverting
variables complicates estimation of cointegrating relationships quite substantially. The appropriate estimation techniques have only recently been
developed in the literature. For example, see Pesaran, Shin and Smith (1999).

S.03

September 1999

long-run equilibrium exchange rates. In GSDEEMER,


Second Edition we tried a number of different
procedures to achieve this result. While each
procedure had its merits, the end result was not
deemed very satisfactory.

calculation of the fitted values delivers a smoother set


of predicted values and more accurate estimates of
where fair values lie. At the same time, the
coefficients in the GSDEEMER regressions are
estimated using one of the most robust methods of
estimating long-run relationships.

In a recent assessment of GSDEEMER values, we


found substantial volatility arising over periods of
less than two years. Such changes in estimated
GSDEEMERs were sometimes hard to associate
with changes in long-term fundamentals. One
potential reason for this volatility is that in the
estimation procedure, we allowed for the inclusion
of leads and lags of dynamic terms in the estimated
regressions. While this is part of the DOLS
estimation procedure, these dynamic variables are
essentially seen as nuisance terms, which are there
only to correct for potential feedback between the
3
right-hand side variables and the RER. This point
can be illustrated by the following DOLS regression:

New GSDEEMER Estimation Results


To provide a sense of how these innovations have
refined our GSDEEMER estimates in Table 7.1, we
compared the new estimates to those based on the
model we had previously been using. We used July
1999 as a reference period for comparison. Column 2
in Table 7.1 shows the spot level of the currency as of
July 30, 1999. Columns 3 and 4 report the
GSDEEMER point estimate as of the same date and
the associated over- (+) or under-valuation (-) of the
currency. The final two columns report the same
information as for columns 3 and 4 but based on new
GSDEEMER estimates.
We also present a
scatter-plot in the chart below to provide a sense of
where each of the misalignments resulting from the
new GSDEEMER estimates lie in comparison to the
older estimates. Finally, we present time charts of
old versus new GSDEEMERs for each currency over
the recent past. These time charts are provided in the
Appendix to the chapter.

RER t = const + 1 TOTt + 2 OPEN t + 3 FISCALt


+ 4 RPROD t + 5 RLIB t +
j=+K

j=K

t - j +
j

FISCALt j +

j= +K

j=K

OPEN t j +
j

j=+K

j=K

j=+K

j=+K

j=K

j=K

j RPROD t j +

The following section, after reviewing the basic


GSDEEMER framework, provides an intuitive
appraisal of the methodological innovations
introduced in this edition of GSDEEMER and
presents a new set of fitted values and misalignments

RLIB t j + error
where the long-run elasticities are given by the
terms associated with the levels, and is a difference
operator xt = xt xt-1, associated with the lagged and
led dynamic terms whose coefficients are given by
the s, s, s, s and s. The variables in differences
should enter the estimation procedure. However,
there is no reason why they should influence the
predicted values of the long-run RER. While the
impact over each quarter is admittedly marginal, this
may contribute to swings in the forecasts over the
immediate quarters ahead which may lead to having
an accumulated and pronounced effect on the
end-of-period GSDEEMER estimate.

Old and New GSDEEMER Values


60.0%
Old
New
50.0%

$/V EB

40.0%

$/V EB

30.0%

20.0%
$/MXN
$/IDR
$/IDR $/INR
$/TRL
$/HKD
EUR/CZ K$/HKD
Bas ket/PLN
$/PHP
Bas ket/PLN
$/CNY
$/THB
$/PHP
$/HUF
$/CLP $/CNY
$/SGD
$/CLP
$/ILS $/KRW$/MY R $/PEN
$/COP $/ECS
$/ILS
$/Z
$/TW D
$/KRW
$/Z A
AR
R
$/ECS
$/MY R
DEM/BLG
DEM/BLG
$/EGP
EUR/CZ
$/EGP
K

$/A RS

10.0%

0.0%

-10.0%

$/BRL

-20.0%

$/BRL

Allowing dynamic terms to enter only during the


estimation phase, but excluding them from the
3

-30.0%
0

10

15

20

25

30

Excluding them from the estimation would not necessarily lead to less volatile estimates, but would most certainly result in biased and inefficient
ones.

S.04

September 1999

Table 7.1 Old and New GSDEEMER Values


New-Gsdeemer

Old-Gsdeemer

Currency
$/ARS
DEM/BLG
$/BRL

Spot

July

Overvaluation (+) or

July

Overvaluation (+) or

July 30, 1999

Undervaluation (-)

GSDEEMER

Undervaluation (-)

1.00

GSDEEMER
1.13

12.9%

1.12

12.4%

995.10

901.08

-9.4%

931.26

-6.4%

1.80

1.39

-23.2%

1.47

-18.6%

$/CLP

512.70

516.35

0.7%

523.66

2.1%

$/COP

1817.00

1781.71

-1.9%

1783.03

-1.9%

$/CNY

8.28

8.49

2.6%

8.63

4.3%

EUR/CZK

36.44

39.14

7.4%

41.28

13.3%

$/ECS

11480.00

10914.01

-4.9%

11356.08

-1.1%

$/EGP

3.40

3.96

16.3%

3.85

13.3%

$/HKD

7.76

8.27

6.5%

8.44

8.8%

$/HUF

237.64

244.61

2.9%

245.07

3.1%

$/IDR

6890.00

7855.75

14.0%

7792.14

13.1%

$/ILS

4.17

4.05

-2.9%

4.09

-1.9%

$/INR

43.29

48.59

12.2%

48.80

12.7%

$/KRW

1204.00

1174.67

-2.4%

1159.44

-3.7%

$/MXN

9.40

11.02

17.2%

10.99

16.8%
-3.0%

$/MYR

3.80

3.57

-6.1%

3.68

$/PHP

38.86

40.13

3.3%

41.01

5.5%

Basket/PLN

0.96

1.02

5.5%

1.03

7.0%

$/PEN

3.34

3.30

-1.0%

3.32

-0.6%

$/ZAR

6.17

5.98

-3.0%

5.94

-3.7%

$/SGD

1.68

1.70

1.3%

1.71

1.3%

$/THB

37.20

38.46

3.4%

38.32

3.0%

$/TRL

429500.00

474270

10.4%

475065

10.6%

$/TWD

32.19

31.20

-3.1%

31.08

-3.4%

$/VEB

612.80

915.13

49.3%

862.59

40.8%

associated with the new estimated regressions. We


then assess the forecasting ability of GSDEEMER as
a long-run valuation tool.

To compare the forecasting performance of these 3


models, we estimated each model up to the second
quarter of 1989 and then predicted real exchange
rates for up to 10 years ahead. To compare the ability
of this forecast to predict the realised exchange rate,
we computed two standard measures of forecast
performance. The first measure is the root mean
square forecast error (RMSFE, a standard statistical
tool used for forecast assessment) that we compute
for each and every year up to 10 years ahead. The
second measure is the Diebold-Mariano statistic
based on Diebold and Mariano (1995). This provides
a broader assessment over the entire evaluation
period, and helps to choose which is a superior model
4
for long-term forecasting.

AN ASSESSMENT of GSDEEMER
PERFORMANCE
Clients, especially corporations, often inquire about
exchange rate projections over long horizons of up to
10 years. We have found in previous empirical
analysis that GSDEEMER outperforms other models
in its ability to forecast exchange rates over the long
term. To demonstrate, we have compared
long-horizon
forecasting
performance
of
GSDEEMER against two other models: (i) the OLD
GSDEEMER model, and (ii) a PPP-based model
based on the idea that the long-run equilibrium real
exchange rate is a constant, or simply reverts back to
its own mean over time.
4

Our sample of countries was limited to those with


long enough spans of historical data to allow
sufficient estimation up to the June quarter of 1989.

For an application of this test in an assessment of alternative models used in predicting financial crises see GS-WATCH: A New Framework for
Predicting Financial Crises in Emerging Markets, by A. Ades, R. Masih and D. Tenengauzer (Dec. 1998).

S.05

September 1999

Country
Argentina

Brazil

Mexico

Hong Kong

Korea

Years
Ahead
1
2
3
4
5
6
7
8
9
10
1
2
3
4
5
6
7
8
9
10
1
2
3
4
5
6
7
8
9
10
1
2
3
4
5
6
7
8
9
10
1
2
3
4
5
6
7
8
9
10

Table 7.2. Comparison of Forecast Performance over Long-Horizons


Root Mean
Predictive Accuracy
Square Forecast Error
Diebold-Mariano
PPP
OLDGSD
NEWGSD
PPP
OLDGSD
0.544
0.044
0.047
5.130
0.112
0.587
10.169
0.322
0.186
14.317
0.560
0.119
11.409
0.690
0.033
8.087
0.392
0.006
11.233
0.441
0.002
16.975
0.583
0.002
23.892
0.754
0.008
32.093
0.820
0.085
-5.908***
-3.244***
0.258
0.287
0.114
0.179
0.023
0.418
0.186
0.002
0.023
0.468
0.001
0.001
0.780
0.000
0.025
1.711
0.074
0.000
2.182
0.183
0.025
2.326
0.250
0.053
2.368
0.252
0.042
0.681
0.000
0.026
-4.855***
-3.115***
0.015
0.024
0.010
0.000
0.052
0.001
0.008
0.099
0.003
0.030
0.136
0.019
0.029
0.082
0.025
0.049
0.048
0.062
0.016
0.038
0.029
0.001
0.016
0.003
0.017
0.008
0.000
0.042
0.005
0.001
-2.866**
-1.541
0.061
0.016
0.000
0.026
0.043
0.008
0.013
0.046
0.005
0.005
0.059
0.003
0.004
0.191
0.023
0.000
0.160
0.003
0.008
0.281
0.016
0.034
0.432
0.027
0.098
0.653
0.037
0.063
0.487
0.008
-1.958
-3.674***
0.015
0.002
0.009
0.006
0.005
0.009
0.000
0.000
0.000
0.013
0.000
0.000
0.168
0.003
0.003
0.410
0.001
0.002
0.857
0.015
0.023
0.770
0.007
0.018
0.216
0.041
0.022
0.859
0.006
0.000
-4.102***
-1.024

Notes: *, ** and *** denote significance at the 10, 5 and 1 pr cent levels, respectively. If the figures are significant, this implies that we cannot
reject the hypothesis of equal mean-square error i.e. GSDEEMER is not an inferior predictor of the real exchange rate compared to the alternative model, at conventional levels of significance. If negatively significant, this favours GSDEEMER outright against the alternative model.
If significant and positive, this favours the alternative model over GSDEEMER. The Diebold-Mariano (1995) test is associated with the null
hypothesis of equal accuracy in forecast performance.

S.06

September 1999

Table 7.3. Speeds of Convergence in Crisis and Tranquil Periods


Adjustment Coeff
Convergence
Adjustment Coeff Convergence
Crisis
in Quarters
Tranquil
in Quarters
Latin America
Argent
-0.81
1.2
-0.04
25.0
Brazil
-0.47
2.1
-0.21
4.7
Chile
-0.80
1.2
-0.03
35.7
Colombia
-0.25
4.0
-0.03
33.3
Ecuador
-0.86
1.2
-0.12
8.5
Mexico
-1.28
0.8
-0.13
7.8
Peru
-0.91
1.1
-0.62
1.6
Venezuela
-0.73
1.4
-0.06
17.0
Regional Average
-0.76
1.6
-0.15
16.7
Asia
China
-0.67
1.5
-0.26
3.8
HK
-0.29
3.5
-0.06
15.6
India
-0.44
2.3
-0.28
3.6
Indonesia
-1.42
0.7
-0.20
5.1
Korea
-0.95
1.1
-0.26
4.0
Malaysia
-0.40
2.5
-0.25
4.0
Philippines
-0.84
1.2
-0.16
6.2
Singapore
-0.37
2.7
-0.05
20.0
Thailand
-0.97
1.0
-0.45
2.2
Taiwan
-0.32
3.1
-0.15
6.8
Regional Average
-0.67
1.8
-0.21
7.1
Emerging Europe
Bulgaria
-0.40
2.5
-0.15
6.7
Czech
-0.13
7.8
-0.03
3.3
Hungary
-0.69
1.4
-0.20
5.0
Poland
-0.84
1.2
-0.10
10.3
Romania
-0.45
2.2
-0.12
8.2
Regional Average
-0.50
2.9
-0.12
12.7
Africa & Middle-East
Egypt
-0.22
4.5
-0.13
8.0
Israel
-0.39
2.6
-0.26
3.8
SA
-0.74
1.4
-0.17
5.8
Turkey
-1.46
0.7
-0.25
4.0
Regional Average
-0.70
2.3
-0.20
5.4
All Average
-0.67
2.1
-0.18
10.7
Notes: Reported are error-correction or speed-of-adjustment coefficients based
on estimation of an assymetric ECM described in the text. All coefficients are statistically
significant at the 10 per cent level or higher.

S.07

September 1999

The results of this exercise are presented in Table 7.2.


The first three columns of statistics refer to the
RMSFEs for each of the models in forecasting ahead
with the horizon shown in number of years. The last
two columns present the Diebold-Mariano statistics
that result from a statistical comparison of forecast
performance using new GSDEEMERs as the
benchmark model. A negative and significant
statistic implies that the new GSDEEMER model is a
superior exchange rate to the alternative model
indicated in the column header.

convergence of the exchange rate to equilibrium


going forward could lead to wrong forecasts in
almost every state of the world. If the exchange rate
collapses during the transition, the estimated speed
would prove an under-estimate, and conversely if the
exchange rate were not to collapse, because crisis
periods are so frequent in emerging markets, our
estimated speeds of adjustment tend to be biased
towards the exchange rate converging too rapidly.
This could be particularly problematic for our
long-term exchange rate forecasts, which as argued
above, rely quite heavily on GSDEEMER estimates,
and the assumption that over the long run (at the
estimated speed of adjustment), spot exchange rates
converge to GSDEEMER.

Overall results, both in terms of yearly RMSFEs and


Diebold-Mariano statistics point to new GSDEEMER
models outperforming old GSDEEMERs up to 10
years ahead. The PPP model performs the worst of
all. In some of the countries such as Korea and
Mexico, old GSDEEMERs do not perform
significantly worse than new GSDEEMERs for short
term horizons. However, it is noteworthy that after 5
years, no model challenges the new GSDEEMERs in
terms of forecasting ability.

We have therefore now incorporated a re-estimation


of such speeds of adjustment. The technique that we
employed allowed for periods of fast and slow speed
of convergence. The exchange rate converges fast
when it experiences a crisis. Conversely, the
exchange rate converges slowly during periods when
it does not experience a crisis, which we call tranquil
periods.

Why does PPP perform so poorly even after such a


long horizon? Various factors ignored by PPP can
play a very substantial role in explaining exchange
rate behavior over 5 to 10-year horizons. James
Lothian and Mark Taylor (1996) have in fact recently
argued that mean reversion (or convergence to PPP)
may take several decades, and hence a 10-year
analysis just may not be enough time to establish this
type of behavior.

To capture these two speeds of adjustment, we


estimated a special form of an asymmetric error
correction model (ASECM) given by the following
specification:
K

RER t = const + a j RER t j + j TOTt j +


j =1

CRISIS and TRANQUIL SPEEDS of


ADJUSTMENT to LONG-RUN
EQUILIBRIUM

j =1

j =1

j =1

j =1

+ j FISCALS t j + j
j OPEN t j
K

RPROD t j + jRLIB t j + 1 i CRECTt 1 + 2 i


j =1

Investors and policy-makers are often interested to


know not only where fair value lies, but also how long
the exchange rate is likely to take to reach its fair
value given an estimated misalignment. Up until
recently, we have reported adjustment to long-run
equilibrium making the (sometimes heroic)
assumption that exchange rates would not collapse
during the transition. However, the samples we used
to estimate such speeds of adjustment did contain
periods of sharp exchange rate correction. In other
words, the estimated speed of adjustment is basically
an average of scenarios involving both crises and
tranquil periods. Using such speed to project

TRECTt 1 + error
where is a difference operator xt = xt xt-1, and the
error correction term (ECT) is separated into those
values observed over crisis periods (CRECT(t)) and
those observed over tranquil (TRECT(t)) periods by:
CRECT(t) = ECT(t)*CRISIS(t) where CRISIS(t) = 1
if the percentage change in the real exchange rate
over the previous quarter was 10% and CRISIS(t) =
0 otherwise,

S.08

September 1999

and

speed just over 4 years. Comparatively, Asian


currencies converge substantially faster back to
equilibrium. Overall, using an average for all the
emerging markets, we find that exchange rates
converge back to equilibrium about 6 times faster
during crises. Hence, using average speeds over both
types of scenarios to project exchange rates forwards
would provide upward biased estimates of such
speed of convergence.

TRECT(t)
=
ECT(t)*TRANQUIL(t)
with
TRANQUIL(t) = 1 if the percentage change of the real
exchange rate over the previous quarter was <10%
and TRANQUIL(t) = 0 otherwise.
With this simple specification, ASECM allows
estimation of the two error correction terms using one
single equation. The coefficients associated with the
crisis and tranquil error correction terms (1 and 2)
provide the speeds of adjustment under crisis and
5
tranquil scenarios for the exchange rate.

CONCLUSION
In this chapter, we have presented three important
improvements to our GSDEEMER models. The first
two modifications aim at providing more reliable and
stable GSDEEMER estimates. The third aims at
providing better estimates of the speed at which
exchange rates converge to those GSDEEMER
values. The results suggest that GSDEEMER
models that impose exogeneity of global real interest
rates and exclude leads and lags from the calculation
of misalignments produce overall better long-term
exchange rate forecasts, and incorporating these new
techniques into our regular framework should allow
for better judgement going forward.

We estimated separate ASECMs for each of the


GSDEEMER exchange rates. Table 7.3 reports the
estimates for the two coefficients, as well as the
number of quarters it would take the exchange rate to
converge back to equilibrium after a shock. The first
two columns report the estimates corresponding to
the crisis scenario. The third and fourth columns
report the same estimates but for the tranquil
scenario. All coefficients have the right (negative)
sign, and are statistically significant. Furthermore,
all regressions were checked to pass for model
adequacy by way of standard diagnostics.

Alberto Ades/Rumi Masih

Results indicate that exchange rates in Latin America


are associated with the fastest speeds of adjustment
following a crisis. In fact, all the currencies in this
region, with the exception of Brazil and Colombia,
adjust back to long-run equilibrium in less than 2
quarters, assuming a crisis scenario.
Asian
currencies take slightly longer to adjust, with Hong
Kong and Taiwan taking the longest in the region.
The remaining countries of Emerging Europe, the
Middle East, and Africa are the slowest to adjust,
even in a crisis scenario. The regional average of
almost 3 quarters for Emerging Europe compares to
only 1.6 and 1.8 quarters for Latin America and Asia
respectively.
The speeds of adjustment under the tranquil scenario
illustrate the point made earlier about upward bias
much more poignantly. Currencies in Latin America
are associated with the slowest adjustment back to
equilibrium during tranquil periods, with the average
5

Asymmetric adjustments are now familiar in several contexts, ranging from macroeconomics to producer price dynamics in inventories and sales.
In terms of econometric implementation, the popular ECM, wherein dependent variables change in response to a measure of disequilibrium, means
that it is natural to represent adjustment of this kind in such a framework. For more applications see Scholnick (1996) and Frost and Bowden (1999).

S.09

September 1999

References
Diebold, F.X. and R.S. Mariano (1995), Comparing
predictive accuracy, Journal of Business &
Economic Statistics, 13, 253-263.
Frost, D. and R. Bowden (1999), An asymmetry
generator for error-correction mechanisms, with
application to bank mortgage dynamics, Journal of
Business & Economic Statistics, 17, 253-263.
Johansen, S. (1991), Estimating and hypothesis
testing of cointegrating and vector autoregressive
models, Econometrica, 59, 1551-1589.
Lothian, J.R. and M.P. Taylor (1996), Real exchange
rate behavior: The recent float from the perspective
of the past two centuries, Journal of Political
Economy, 104(3), 488-509.
Pesaran, M.H. and Y. Shin, (1999), Long-run
structural modelling, University of Cambridge DAE,
Working
Paper
No.9419.
(http://www.econ.cam.ac.uk/faculty/pesaran/lrs.pdf).
Pesaran, M.H., Y. Shin and R.J. Smith (1999),
Structural Analysis of Vector Error Correction
Models With Exogenous I(1) Variables, University of
Cambridge DAE Working Paper No.9706. Revised
November
1998
(http://www.econ.cam.ac.uk/faculty/pesaran/pss2re
v.pdf).
Scholnick, B. (1996), Asymmetric adjustment of
commercial bank interest rates: Evidence from
Malaysia and Singapore, Journal of International
Money and Finance, 15, 485-496.
Stock, J.H. and M.W. Watson (1993), A simple
estimator of cointegrating vectors in higher order
integrated systems, Econometrica, 61, 783-820.

S.10

September 1999

Appendix: OLD and NEW


GSDEEMER Estimates

Brazil

Argentina
425

180

OLD
375

NEW

160

325

RER

140

275

120

225

100

175

80

125

60

75

40

25

20
Jul-87

Jul-84

Jul-90

Jul-93

Jul-96

Jul-99

Jul-84

Jul-87

Jul-90

Jul-93

Jul-96

Jul-99

China

Chile

180

150

160
125
140
100

120
100

75
80
50

60
40

25
20
0

0
Jul-87

Jul-84

Jul-90

Jul-93

Jul-96

Jul-99

Apr-81

Apr-84

Apr-87

Apr-90

Apr-93

Apr-96

Apr-99

Apr-87

Apr-90

Apr-93

Apr-96

Apr-99

Czech Republic

Colombia
120

135

110
115

100
90

95

80
70

75

60
55

50
40

35

30
20
Apr-81

15
Apr-84

Apr-87

Apr-90

Apr-93

Apr-96

Apr-99

Apr-81

S.11

Apr-84

September 1999

Ecuador

Hong Kong

150

OLD

190

NEW
125

RER

165
140

100

115
75
90
50

65

25

40

Apr-81 Apr-84 Apr-87 Apr-90 Apr-93 Apr-96 Apr-99

Oct-83

Hungary

Oct-86

Oct-89

Oct-92

Oct-95

Oct-98

India

125

New-GSDEEMER

145
TWI

115
120
105
95

95

85

70

Old-GSDEEMER

75
45
65
55
Apr-84

20
Apr-87

Apr-90

Apr-93

Apr-96

Jan-81 Jan-84 Jan-87 Jan-90 Jan-93 Jan-96 Jan-99

Apr-99

Indonesia

Israel
120

250
225

115

200

110

175

105

150

100

125

95

100

90

75

85

50

80

25

75

Apr-81 Apr-84 Apr-87 Apr-90 Apr-93 Apr-96 Apr-99

Apr-84

S.12

Apr-87

Apr-90

Apr-93

Apr-96

Apr-99

September 1999

Malaysia

Korea

142.20

150
OLD
NEW

122.20

RER
102.20

125

82.20
62.20
100

42.20
22.20

75
Oct-80

2.20
Jun-84

Feb-88

Oct-91

Jun-95

Feb-99

Apr-82

Jun-85

Aug-88

Oct-91

Dec-94

Feb-98

Peru

Mexico

200

140

US$/PEN

180
120

160

100

140
120

80

100
60

80

40

60
40

20

20

0
Apr-82

0
Apr-85

Apr-88

Apr-91

Apr-94

Apr-97

Apr-81

Apr-84

Apr-87

Apr-90

Apr-93

Apr-96

Apr-99

Poland

Philippines
124

150

114
130

104
94

110

84
74

90

64
70

54
44

50

34
24
Apr-81

30
Apr-84

Apr-87

Apr-90

Apr-93

Apr-96

Apr-99

Apr-87

S.13

Apr-90

Apr-93

Apr-96

Apr-99

September 1999

South Africa

Singapore

140.0

121.2

130.0

101.2

120.0
81.2

110.0

61.2

100.0
OLD
90.0

NEW

41.2

RER

80.0

21.2

70.0

1.2
Apr-81

60.0
Apr-84

Apr-87

Apr-90

Apr-93

Apr-96

Apr-99

Apr-84

Apr-81

Apr-87

Apr-90

Apr-93

Apr-96

Apr-99

Apr-96

Apr-99

Thailand

Taiwan

150

145

140
130

120

120
110

95

100
90

70

80
45
Apr-81

70
Apr-84

Apr-87

Apr-90

Apr-93

Apr-96

Apr-99

Apr-81

Apr-84

Apr-87

Apr-90

Apr-93

Venezuela

Turkey

140
120

135

100
80

110

60
40

85

20
60
Jul-87

0
Jul-90

Jul-93

Jul-96

Jul-99

Jul-83

S.14

Jul-86

Jul-89

Jul-92

Jul-95

Jul-98

September 1999

Goldman Sachs Economic Research Group


In London
(0171) 774 1000

Gavyn Davies, Managing Director and Chief International Economist


Jim ONeill, Managing Director and Chief Currency Economist
David Walton, Managing Director & Co-Director of European Economic Research
Andrew Bevan, Director of International Bond Economic Research
Erik Nielsen, Director of New European Markets Economic Research
Martin Brookes, Executive Director and Senior Economist
Stephen Potter, Consultant, Global Economics
Linda Britten, Executive Director and Global Economics Manager, Support & Systems
Binit Patel, Executive Director and International Economist
Al Breach, Consultant, New European Markets Economic Research
Stphane Do, International Economist
Francesco Garzarelli, International Economist
Stephen Hull, International Economist
Sandra Lawson, Associate
Joshua Rauh, Associate Economist
Carlos Teixeira, Associate Economist
Philippa Knight, Research Assistant
Javier Prez de Azpillaga, Research Assistant
Alexander Perjssy, Research Assistant
Thomas Stolper, Research Assistant
AnnMarie Terry, Research Assistant

In Paris
(331) 4212 1341
In Frankfurt
(069) 7532 1200
In New York
(212) 902 6807

Philippe Gudin, Executive Director & Director of French Economic Research

In Toronto
(416) 343 8793
In Tokyo
(813) 3589 8911

Mark Chandler, V.P. and Senior Economist


Marcel Kasumovich, V.P. and International Economist
Tetsufumi Yamakawa, Managing Director and Director of Japan Economic Research
Yoshito Sakakibara, V.P. and Senior Economist
Yuriko Tanaka, V.P. and Associate Economist
Takuji Okubo, Associate Economist
Tomohiro Ohta, Research Assistant

In Hong Kong
(852) 2978 1941

Sun Bae Kim, Managing Director and Co-Director of Asia Economic Research
Donald Hanna, Co-Director of Asia Economic Research
Fred Hu, V.P. and International Economist
Dick Li, Associate Economist
Wilbur Maximo, Associate Economist
Alice Po, Research Assistant

Thomas Mayer, Managing Director and Co-Director of European Economic Research


William Dudley, Managing Director and Director of US Economic Research
Paulo Leme, Managing Director of Emerging Markets Economic Research
Alberto Ades, V.P. and Emerging Markets Currency Economist
Edward McKelvey, V.P. and Senior Economist
John Youngdahl, V.P. and Senior Money Market Economist
Federico Kaune, V.P. and Senior Economist
Jan Hatzius, V.P. and International Economist
Susana Hernandez, Associate Economist
Rumi Masih, Associate Economist
Jess Viejo, Associate Economist
Rajib Pal, Research Assistant
Elizabeth Peters, Research Assistant
Daniel Tenengauzer, Research Assistant

Goldman Sachs Research personnel may be contacted by electronic mail through the Internet at [email protected]

S.15

September 1999

Goldman Sachs Global Research Centres


New York
Goldman, Sachs & Co.
1 New York Plaza, 47th Floor
New York, New York 1000, USA
Tel: (1) 212-902-1000
Fax: (1) 212-902-2145

Paris
Goldman Sachs Inc et Cie
2, rue de Thann
75017 Paris, France
Tel: (33) 1 4212 1341
Fax: (33) 1 4212 1499

Frankfurt
Goldman, Sachs & Co. oHG
MesseTurm
D-60308 Frankfurt am Main,
Germany
Tel: (49) 69-7532-1000
Fax: (49) 69-7532-2800

Singapore
Goldman Sachs (Singapore) Pte.
50 Raffles Place,29-01 Shell Tower
Singapore 048623
Tel: (65) 228-8128
Fax: (65) 228-8474

London
Goldman Sachs International
Peterborough Court
133 Fleet Street
London, EC4A 2BB, England
Tel: (44) 171-774-1000
Fax: (44) 171-774-1181

Hong Kong
Goldman Sachs (Asia) L.L.C.
Cheung Kong Center,
68th Floor
2 Queens Road Central
Hong Kong
Tel: (852) 2978-0300
Fax: (852) 2978-0479

Tokyo
Goldman Sachs (Japan) Ltd.
ARK Mori Building, 10th Floor
12-32, Akasaka 1-chome
Minato-ku, Tokyo 107, Japan
Tel: (81) 3-3589-7000
Fax: (81) 3-3587-9263

Korea
Goldman, Sachs & Co.
Dong Ah Life Insurance Building
33 Da-Dong, Chung-Ku
Seoul, South Korea
Tel: (822) 3788-1000
Fax: (822) 3788-1001

Goldman Sachs Research personnel may be contacted by electronic mail through the Internet at [email protected]
1999 Goldman Sachs International. All rights reserved.
This material is for your private information, and we are not soliciting any action based upon it. This report is not to be construed as an offer
to sell or the solicitation of an offer to buy any security in any jurisdiction where such an offer or solicitation would be illegal. The material
is based upon information that we consider reliable, but we do not represent that it is accurate or complete, and it should not be relied upon
as such. Opinions expressed are our current opinions as of the date appearing on this material only. While we endeavour to update on a
reasonable basis the information discussed in this material, there may be regulatory, compliance, or other reasons that prevent us from
doing so. We and our affiliates, officers and employees, including persons involved in the preparation or issuance of this material may,
from time to time, have long or short positions in, and buy or sell, the securities, or derivatives (including options) thereof, of companies
mentioned herein. No part of this material may be (i) copied, photocopied, or duplicated in any form, by any means, or (ii) redistributed
without Goldman, Sachs & Cos prior written consent.
This material has been issued by Goldman, Sachs and Co. and/or one of its affiliates and has been approved by Goldman Sachs
International, which is regulated by The Securities and Futures Authority, in connection with its distribution in the United Kingdom and
by Goldman, Sachs Canada in connection with its distribution in Canada. This material is distributed in Hong Kong by Goldman Sachs
(Asia) L.L.C.; and in Japan by Goldman Sachs (Japan) Ltd., and in Singapore through Goldman Sachs (Singapore) Pte. This material is
not for distribution in the United Kingdom to private customers, as that term is defined under the rules of The Securities and Futures
Authority; and any investments, including any convertible bonds or derivatives, mentioned in this material will not be made available by
us to any such private customer. Neither Goldman, Sachs and Co. nor its representative in Seoul, Korea, is licensed to engage in the
securities business in the Republic of Korea. Goldman Sachs International and its non-US. affiliates may, to the extent permitted under
applicable law, have acted upon or used this research, to the extent it relates to non-US. issuers, prior to or immediately following its
publication. Foreign-currency-denominated securities are subject to fluctuations in exchange rates that could have an adverse effect on the
value or price of, or income derived from, the investment. In addition, investors in securities such as ADRs, the values of which are
influenced by foreign currencies, effectively assume currency risk.
Further information on any of the securities mentioned in this material may be obtained upon request, and for this purpose
persons in Italy should contact Goldman Sachs S.I.M. S.p.A. in Milan, or at its London branch office at 133 Fleet Street, and
persons in Hong Kong should contact Goldman Sachs (Asia) L.L.C. at 3 Garden Road. Unless governing law permits otherwise you
must contact a Goldman Sachs entity in your home jurisdiction if you want to use our services in effecting a transaction in the securities
mentioned in this material.

You might also like