This article was downloaded by: [Universite De Paris 1]On: 03 August 2013, At: 05:04Publisher: RoutledgeInforma Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House,37-41 Mortimer Street, London W1T 3JH, UK

Applied EconomicsPublication details, including instructions for authors and subscription information:http://www.tandfonline.com/loi/raec20

Revisiting purchasing power parity for Africancountries: with nonlinear panel unit-root testsChi-Wei Su a b , Tsangyao Chang c & Yu-Shao Liu d ea Department of International Business, Tamkang University, Taipei, Taiwan, ROCb School of Economics, University of Jinan, Jinan, Chinac Department of Finance, Feng Chia University, Taichung, Taiwan, ROCd Department of Finance, Xiamen University, Xiamen, Chinae PhD Program in Finance, Feng Chia University, Taichung, Taiwan, ROCPublished online: 14 Jun 2011.

To cite this article: Chi-Wei Su , Tsangyao Chang & Yu-Shao Liu (2012) Revisiting purchasing power parity for Africancountries: with nonlinear panel unit-root tests, Applied Economics, 44:25, 3263-3273, DOI: 10.1080/00036846.2011.570730

To link to this article: http://dx.doi.org/10.1080/00036846.2011.570730

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) containedin the publications on our platform. However, Taylor & Francis, our agents, and our licensors make norepresentations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of theContent. Any opinions and views expressed in this publication are the opinions and views of the authors, andare not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon andshould be independently verified with primary sources of information. Taylor and Francis shall not be liable forany losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoeveror howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use ofthe Content.

This article may be used for research, teaching, and private study purposes. Any substantial or systematicreproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in anyform to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://www.tandfonline.com/page/terms-and-conditions

Applied Economics, 2012, 44, 3263–3273

Revisiting purchasing power parity

for African countries: with nonlinear

panel unit-root tests

Chi-Wei Sua,b,*, Tsangyao Changc and Yu-Shao Liud,e

aDepartment of International Business, Tamkang University, Taipei,

Taiwan, ROCbSchool of Economics, University of Jinan, Jinan, ChinacDepartment of Finance, Feng Chia University, Taichung, Taiwan, ROCdDepartment of Finance, Xiamen University, Xiamen, ChinaePhD Program in Finance, Feng Chia University, Taichung, Taiwan, ROC

This study applies Panel Seemingly Unrelated Regressions (SUR)

Kapetanios et al. (Kapetanios–Shin–Snell (KSS), SURKSS) tests, pro-

posed by Wu and Lee (2009), to investigate the properties of long-run

Purchasing Power Parity (PPP) in 15 African countries. The empirical

results from the univariate unit root and panel based unit root tests

indicate that PPP does not hold for these 15 countries under study.

However, Panel SURKSS tests indicate that PPP is valid for four of these

15 countries. These results have important policy implications for these

15 African countries under study.

Keywords: purchasing power parity; panel SURKSS test

JEL Classification: C23; F31

I. Introduction

During much of the past few decades, a plethora of

studies has centred on the investigation of the

stationarity of the Real Exchange Rate (RER;

Papell, 1997; O’Connell, 1998; Sarno and Taylor,

2001). The results from such studies are not only

valuable for empirical researchers and policy makers,

but they have also unveiled extremely important

implications in international finance. To be more to

the point, a nonstationary RER indicates that any

long-run relationship between the nominal exchange

rate and domestic and foreign prices is virtually

nonexistent, therefore invalidating the theory of

Purchasing Power Parity (hereafter, PPP).1 In this

event, PPP cannot be used to determine the equilib-

rium exchange rate; what is more, the invalidation

of PPP disqualifies any monetary approach to

*Corresponding author. E-mail: cwsu@mail.tku.edu.tw1 PPP also states that the exchange rates between currencies are in equilibrium when their purchasing power is the same ineach of the two countries. This means that the exchange rate between any two countries should equal the ratio of twocurrencies’ price level of a fixed basket of goods and services. The basic idea behind the PPP hypothesis is that since anyinternational goods market arbitrage should be traded away over time, we should expect the real exchange rate to return to aconstant equilibrium value in the long-run. This means that the real exchange rate should be the I(0) stationary process, forwhich PPP to be hold. If the real exchange rate is I(1) process, this means that the real exchange rate will not return to aconstant equilibrium value in the long-run, which is a violation of the long-run PPP.

Applied Economics ISSN 0003–6846 print/ISSN 1466–4283 online � 2012 Taylor & Francis 3263http://www.informaworld.com

DOI: 10.1080/00036846.2011.570730

Dow

nloa

ded

by [

Uni

vers

ite D

e Pa

ris

1] a

t 05:

04 0

3 A

ugus

t 201

3

determining the exchange rate since that would

necessitate that PPP holds true.2

Empirical evidence regarding the order of integra-

tion of the RER is abundant, but consensus regarding

the validity of PPP has not yet been reached. For

information about the theoretical and empirical

aspects of PPP, as well as the RER, the interested

reader is referred to the work of MacDonald and

Taylor (1992), Taylor (1995), Rogoff (1996), Taylor

and Sarno (1998), Sarno and Taylor (2002), Taylor

and Taylor (2004) and Lothian and Taylor (2000,

2008). There is a growing consensus that RER

exhibits nonlinearities and, consequently, conven-

tional unit root tests, such as the Augmented Dickey–

Fuller (ADF) test, have low power in detecting mean

reversion of exchange rate. A number of studies have

provided solid empirical evidence for the nonlinear

and/or asymmetric adjustment of the exchange rate in

developed countries (Baum et al., 2001; Taylor et al.,

2001), in the Group of Seven (G7) countries (Kilian

and Taylor, 2003), in the Middle East (Sarno, 2000),

in Asian economies (Enders and Chumrusphonlert,

2004), in African countries (Chang et al., 2011), as

well as in 10 Latin American Integration Association

countries (Chang et al., 2010a). It is important to

note, nevertheless, that under no circumstance does

the finding of nonlinear adjustment necessarily sig-

nify the existence of nonlinear mean reversion or

stationarity. Thus, it is essential that stationary tests

based on a nonlinear framework be applied.3

More recently, it has been reported that conven-

tional unit root tests not only fail to consider

information across regions, thereby leading to less

efficient estimations, but also have lower power when

compared with near-unit-root but stationary

alternatives (Taylor and Sarno, 1998; Maddala and

Wu, 1999; Levin et al., 2002; Im et al., 2003; Choi and

Chue, 2007; Pesaran, 2007). It is not surprising that

these factors have induced considerable doubt on

many of the earlier findings, which were based on a

unit root in RER. In order to increase the power in

testing for a unit root, many researchers have

employed panel data. Levin et al. (2002) and Im

et al. (2003), for instance, have developed the

asymptotic theory and the finite-sample properties

of ADF tests for use with panel data. These two tests

have significantly improved power even in relatively

small panels, but both tests are unreliable in the

presence of cross-sectional dependencies (i.e. the

individual time series in the panel are cross-

sectionally independently distributed). Zellner

(1962) put forth a straightforward approach to

handle cross-sectional dependence across countries,

and this is to estimate equations using the Seemingly

Unrelated Regression (SUR) estimator. Furthermore,

O’Connell (1998) demonstrated that size distortions

can be avoided without a significant loss of power by

basing the panel-based test on SUR estimations

instead of Ordinary Least Squares (OLS) estimations.Taylor and Sarno (1998) and Breuer et al. (2001)

have shown that the ‘all-or-nothing’ nature of the

tests has not been fully addressed by recent method-

ological refinements to the Levin et al. (2002) test.

Although Taylor and Sarno (1998), Maddala and Wu

(1999) and Im et al. (2003) developed tests that permit

the autoregressive parameters to differ across panel

members under the stationary alternative, they are

not informative in terms of the number of series that

are stationary processes when the null hypothesis is

rejected. The reason is simple: they are joint tests of

2According to Holmes (2001) and Sarno (2005), PPP is important to policymakers for several reasons. First, it can be used topredict the exchange rate and determine whether a currency is over- or under-valued, which is particularly important for lessdeveloped countries and countries experiencing large differences between domestic and foreign inflation rates. Second, thenotion of PPP is used as the foundation on which many theories of exchange rate determination are built. Consequently, thevalidity is important to policymakers in developing countries who base their adjustments on PPP. Third, from a theoreticalperspective, if PPP is not a valid long-run international parity condition, this casts doubts on the predictions of open-economymacroeconomics, which are based on the assumption of long-run PPP. Indeed, the implications of open-economy dynamicmodels are sensitive to the presence or absence of a unit root in the real exchange rate. Finally, estimates of PPP exchangerates are often used for practical purposes, such as determining the degree of misalignment of the nominal exchange rate andthe appropriate policy response, the setting of exchange rate parities, and the international comparison of national incomelevels.3 Enders and Granger (1998) show that the standard tests for unit root and cointegration all have lower power in the presenceof misspecified dynamics. This is important since the linear relationship is inappropriate if prices are sticky in the downward,but not in the upward direction. Madsen and Yang (1998) have provided evidence that prices are sticky in the downwarddirection and that such stickiness means that real exchange rate adjustments are asymmetric. Other reasons for theasymmetric adjustment are the presence of transactions costs that inhibit international goods arbitrage and officialintervention in the foreign exchange market may be such that nominal exchange rate movements are asymmetric (Obstfeldand Taylor, 1997; Taylor and Peel, 2000; Taylor and Sarno, 2001; Sarno et al., 2004; Taylor, 2004; Juvenal and Taylor, 2008).Kilian and Taylor (2003) also suggest that nonlinearity may arise from the heterogeneity of opinion in the foreign exchangemarket concerning the equilibrium level of the nominal exchange rate: as the nominal rate takes on more extreme values, agreat degree of consensus develops concerning the appropriate direction of exchange rate movements, and traders act asaccordingly.

3264 C.-W. Su et al.

Dow

nloa

ded

by [

Uni

vers

ite D

e Pa

ris

1] a

t 05:

04 0

3 A

ugus

t 201

3

the null hypothesis. In this regard, Breuer et al. (2001)claim that, by analogy to a simple regression, whenan F-statistic rejects the null that a vector ofcoefficients is equal to zero, it is not necessarily truethat each coefficient is nonzero. Likewise, when theunit-root null hypothesis is rejected, it may very wellnot be justified to assume that all series in the panelare stationary.

In contrast to those panel-based unit root tests thatare joint tests of a unit root for all members of a paneland that are incapable of determining the mix of I(0)and I(1) series in a panel setting, Panel SURKapetanios–Shin–Snell (KSS), SURKSS tests inves-tigate a separate unit-root null hypothesis for eachand every individual panel member. In so doing, theyclearly identify how many and which series in thepanel are stationary processes. Hence, this empiricalstudy applies Panel SURKSS tests, which are theKapetanios et al. (2003) tests based on the panelestimation method of SUR, to test the validity of PPPfor 15 Common Markets for Eastern and SouthernAfrica (COMESA) and/or the Southern AfricanDevelopment Community (SADC) countries.Increasingly, African countries are adopting moreflexible exchange rate regimes. At the same time, wenotice huge volatility in exchange rates and price invarious African countries since mid-1980s. Besides,robust researches on PPP are very important forAfrican countries, since the early 1990s, these haveimplemented numerous exchange rate policies’ mod-ifications based on the assumption of PPP validity(Kargbo, 2003). Recently, several studies have exam-ined whether or not there is empirical support forlong-run PPP in African countries (Akinboade andMakina, 2006; Bahmanee-Oskooee and Gelan, 2006;Kargbo, 2006). But, once again, a strong consensuscould not be reached even if more results convergetowards PPP validity. For much of the past twodecades, African countries have been implementingstructural and macroeconomic adjustment pro-grammes designed to improve the external competi-tiveness and economic growth of these economies(Kargbo, 2004). In particular, exchange rate policyreforms were the focal point of the adjustmentprogrammes. Our study focuses on COMESA and/or SADC on account of history of attempts atintegration through their respective visions andreflected in their trade liberalization and tariffreduction programmes, and size. COMESA and/orSADC are strategically focused on promotion of

regional integration through trade development andinvestment promotion as the institution seeks toenable members make the adjustments necessary forthem to become part of the global economy withinthe World Trade Organization (WTO) frameworkand its regulations, and other global requirements.This empirical study contributes to field of empiricalresearch by determining whether PPP holds true forthese 15 COMESA and/or SADC countries. We findthat long-run PPP holds true for four of these 15COMESA and/or SADC countries, namelyBotswana, Burundi, Madagascar and the Seychelles.

The plan of this article is organized as follows.Section II presents the data used in our study. SectionIII first outlines the methodology we employ, thendiscusses the empirical findings and finally, presentssome economic and policy implications from ourempirical findings. Section IV reviews the conclusionsthat we draw in this article.

II. Data

In this study, we employ 15 monthly bilateralexchange rates (and consumer price indices are with2000¼ 100) from both the COMESA4 and/orSADC.5 The RER series of a country at time t isdefined as ðSt � PUS

t Þ=PHt , where St is the nominal

exchange rate of home country per dollar, PUSt and

PHt denote the consumer price indices of home

country and the USA, respectively. The choice ofthe USD benchmark currency is consistent withBahmani-Oskooee (1993), Mahdavi and Zhou(1994) and Holmes (2001) in their investigations ofPPP in Less Developed Countries (LDCs) and can beregarded as the floating rate with respect to the USdollar. The behaviour of the nominal exchange rateover the study period is characterized by a number ofinstances of major depreciations and devaluationsthat RER might imply nonlinear adjustment in thedata. Another reason for using the USD for the basecurrency is that internal foreign exchange markets aremostly dollar dominated. As in Table 1, only 15countries among COMESA and SADC membershave data available for the period from December1994 to July 2008. The source of the data is from theInternational Monetary Fund’s InternationalFinancial Statistics CD-ROM. Table 1 shows that,as of 2009, there are 19 COMESA member countries

4 The COMESA was formed in December 1994, replacing a Preferential Trade Area which had existed since 1981.5 The SADCC which was the forerunner of the socio-economic cooperation leg of today’s SADC was transformed into SADCon 17 August 1992, with the adoption by the founding members of SADCC and the newly independent Namibia of theWindhoek declaration and treaty establishing SADC.

Revisiting purchasing power parity for African countries 3265

Dow

nloa

ded

by [

Uni

vers

ite D

e Pa

ris

1] a

t 05:

04 0

3 A

ugus

t 201

3

and 15 SADC member countries, while eight coun-tries simultaneously belong to both COMESA and

SADC. In 2008, COMESA agreed to an expandedfree-trade zone including members of the SADC.Thus, the members of COMESA and/or SADC aretreated as one group (bloc) by this study and the

sample period is from December 1994 (the formationdate of COMESA).

Summary statistics are provided in Table 2. OurJarque–Bera test results indicate that, except for

Ethiopia/USD and Madagascar/USD, the bilateralRER data sets are approximately nonnormal for allother 13 country pairs. The log of the Zambia/USD,with values varying from 7.088 to 8.237 (SD of 0.267)

is the most volatile currency among these 15COMESA and/or SADC currencies, whereas thelog of the Seychelles/USD, with values varying from

1.542 to 2.023 (SD of 0.086) is the least volatilecurrency. As shown in Fig. 1, visual inspection of theRER series for these 15 country pairs reveals signif-

icant upward and/or downward trends in the RERseries for most of the countries against the US dollarduring this sample period. From these figures, for

most of the series, there seems to be some nonlinearadjustment patterns.

III. Methodology, Empirical Results andEconomic and Policy Implications

Panel SURKSS test proposed by Wu and Lee

The Panel SURKSS tests are based on Breuer et al.’s(2001) Panel SURADF test and Kapetanios et al.’s(2003) nonlinear unit test, in which (panel) SUR isused to run the KSS tests. This test was first proposedby Wu and Lee (2009) and proved very powerful intheir study. According to Wu and Lee (2009), thisnonlinear panel unit-root test is superior in power tothe Breuer et al. (2001) when the data generatingprocess is highly nonlinear. In contrast to thosepanel-based unit root tests that are joint tests of aunit root for all members of a panel and that areincapable of determining the mix of I(0) and I(1)series in a panel setting, Panel SURKSS tests inves-tigate a separate unit-root null hypothesis for eachand every individual panel member. In so doing, theyclearly identify how many and which series in thepanel are stationary processes.

In line with Kapetanios et al. (2003), the KSS test isbased on detecting the presence of nonstationarityagainst a nonlinear but globally stationaryExponential Smooth Transition Autoregressive(ESTAR) process. The model is given by

DXt ¼ �Xt�1f1� expð��X2t�1Þg þ �t ð1Þ

where Xt is the data series of interest, vt an indepen-dent and identically distributed (i.i.d.) error with zeromean and constant variance and � � 0 the transitionparameter of the ESTAR model that governs thespeed of transition. Under the null hypothesis Xt

follows a linear unit root process, but Xt follows anonlinear stationary ESTAR process under thealternative. One shortcoming of this framework isthat the parameter � is not identified under thenull hypothesis. Kapetanios et al. (2003) have used afirst-order Taylor series approximation for{1� expð��X2

t�1Þ} under the null hypothesis � ¼ 0and have then approximated Equation 1 by using thefollowing auxiliary regression:

DXt ¼ � þ �X3t�1 þ

Xk

i¼1

biDXt�i þ �t t ¼ 1, 2, . . . ,T

ð2Þ

In this framework the null hypothesis and alterna-tive hypotheses are expressed as � ¼ 0

Table 1. The members of COMESA and SADC as of 2009

Organization Country

Enough data(December 1994 toJuly 2008)

SADC Angola NOBotswana YESLesotho NOMozambique YESNamibia NOSouth Africa YESTanzania YES

COMESA Burundi YESComoros NODjibouti NOEgypt YESEritrea NOEthiopia YESKenya YESLibya NORwanda NOSudan YESUganda YES

COMESAand SADC

Democratic Republicof the Congo

NO

Madagascar YESMalawi YESMauritius YESSeychelles YESSwaziland NOZambia YESZimbabwe NO

3266 C.-W. Su et al.

Dow

nloa

ded

by [

Uni

vers

ite D

e Pa

ris

1] a

t 05:

04 0

3 A

ugus

t 201

3

(nonstationarity) against �5 0 (nonlinear ESTAR

stationarity). The system of the KSS equations that

we estimate here is6

DX1,t ¼ �1 þ �1X31,t�1

þXk1

j¼1

�1,jDX1,t�j þ "1,t t ¼ 1, 2, . . . ,T

DX2,t ¼ �2 þ �2X32,t�1

þXk2

j¼1

�2,jDX2,t�j þ "2,t t ¼ 1, 2, . . . ,T

DXN,t ¼ �N þ �NX3N,t�1

þXkN

j¼1

�N,jDXN,t�j þ "N,t t ¼ 1, 2, . . . ,T ð3Þ

We test the N null and alternative hypotheses

individually

H10 : �1 ¼ 0; H1

A : �1 5 0

H20 : �2 ¼ 0; H2

A : �2 5 0

HN0 : �N ¼ 0; HN

A : �N 5 0

where we compute the test statistics from the SURestimates of system (3). Breuer et al. (2001) havedemonstrated that the imposition of an identical lagstructure across panel members could bias test-statistics; thus, we select the lag structures for eachequation based on the approach adopted by Perron(1989).7 Wu and Lee (2009) have indicated that thistest has nonstandard distributions and the criticalvalues must be obtained by simulation.

Empirical results

Several univariate time series panel unit root tests arefirst employed to examine the null of a unit root inbilateral RERs for the 15 COMESA and/or SADCcountries that we study. Then, both first generationand second generation panel unit root tests areemployed. Based on the results in Table 3, there isno question that three univariate unit root tests – theADF (Dickey and Fuller, 1981), the Phillips andPerron (PP, 1988) and Elliot et al. (1996, GLS–DF)tests all lead us to conclude that the RER series of the15 COMESA and/or SADC countries contain unitroots. The Kwiatkowski, Phillips, Schmidt and Shin(KPSS, 1992) tests also yield the same results. Thisresult is consistent with that of existing literature and

Table 2. Summary statistics: ln(RER)

Country Mean Median Max. Min. SD Skewness Kurtosis Jarque–Bera Probability

Botswana 1.490 1.458 1.883 1.230 0.145 0.566 2.726 9.274 0.0097Burundi 6.578 6.656 7.244 6.071 0.244 �0.264 2.242 5.821 0.0544Egypt 1.427 1.362 1.788 1.213 0.186 0.448 1.794 15.420 0.0004Ethiopia 2.015 2.028 2.323 1.505 0.159 �0.265 2.956 1.931 0.3808Kenya 4.162 4.223 4.398 3.500 0.205 �1.340 4.093 57.209 0.0000Madagascar 7.120 7.143 7.531 6.776 0.133 �0.190 2.845 1.155 0.5613Malawi 4.052 4.126 4.365 3.525 0.229 �0.864 2.656 21.227 0.0000Mauritius 3.209 3.238 3.373 2.984 0.104 �0.656 2.357 14.591 0.0007Mozambique 2.715 2.698 3.056 2.476 0.149 0.488 2.192 10.974 0.0041Seychelles 1.714 1.699 2.023 1.542 0.086 1.183 5.022 66.191 0.0000South Africa 1.808 1.788 2.452 1.473 0.204 0.838 3.740 22.926 0.0000Sudan 0.775 0.834 1.097 0.259 0.211 �0.929 2.847 23.760 0.0000Tanzania 6.803 6.848 7.092 6.514 0.148 �0.251 1.824 11.174 0.0037Uganda 7.304 7.335 7.583 6.940 0.196 �0.451 1.886 14.037 0.0009Zambia 7.800 7.892 8.237 7.088 0.267 �1.035 2.871 29.386 0.0000

Notes: ln(RER)¼ ln (nominal exchange rate)þ ln(foreign price level)� ln(domestic price level). The USA as the base country.

6 In our study, we only consider a specification with a constant but without a time trend because time trend in real exchangerates is not consistent with the long-run PPP, therefore, we only estimate equations without the trend in our study. In fact, wehave found that the results are very sensitive to whether the time trend is incorporated into the model.7 The procedure poses several advantages. First, by exploiting the information from the error covariances and allowing for anautoregressive process, it produces more efficient estimators than the single equation methods. Second, the testing procedureallows for heterogeneity lag structure across the panel members. Third, the SURKSS panel integration test allows us toidentify which members of the panel contain a unit root. Put differently, the advantage of the test is that it is based on anindividual rather than a joint null hypothesis as in earlier versions of the panel unit root tests (Breuer et al., 2001). In our viewthis is very important in the present context as the COMESA and/or SADC countries under investigation have varyingdegrees of integration with global capital markets.

Revisiting purchasing power parity for African countries 3267

Dow

nloa

ded

by [

Uni

vers

ite D

e Pa

ris

1] a

t 05:

04 0

3 A

ugus

t 201

3

1.2

1.4

1.6

1.8

2.0

1996 1998 2000 2002 2004 2006 2008

BOTSWANA

6.0

6.4

6.8

7.2

7.6

1996 1998 2000 2002 2004 2006 2008

BURUNDI

1.2

1.3

1.4

1.5

1.6

1.7

1.8

1996 1998 2000 2002 2004 2006 2008

EGYPT

1.4

1.6

1.8

2.0

2.2

2.4

1996 1998 2000 2002 2004 2006 2008

ETHIOPIA

3.4

3.6

3.8

4.0

4.2

4.4

4.6

1996 1998 2000 2002 2004 2006 2008

KENYA

6.6

6.8

7.0

7.2

7.4

7.6

1996 1998 2000 2002 2004 2006 2008

MADAGASCAR

3.4

3.6

3.8

4.0

4.2

4.4

1996 1998 2000 2002 2004 2006 2008

MALAWI

2.9

3.0

3.1

3.2

3.3

3.4

1996 1998 2000 2002 2004 2006 2008

MAURITIUS

2.4

2.6

2.8

3.0

3.2

1996 1998 2000 2002 2004 2006 2008

MOZAMBIQUE

1.5

1.6

1.7

1.8

1.9

2.0

2.1

1996 1998 2000 2002 2004 2006 2008

SEYCHELLES

1.4

1.6

1.8

2.0

2.2

2.4

2.6

1996 1998 2000 2002 2004 2006 2008

SOUTH_AFRICA

0.2

0.4

0.6

0.8

1.0

1.2

1996 1998 2000 2002 2004 2006 2008

SUDAN

6.5

6.6

6.7

6.8

6.9

7.0

7.1

1996 1998 2000 2002 2004 2006 2008

TANZANIA

6.8

7.0

7.2

7.4

7.6

1996 1998 2000 2002 2004 2006 2008

UGANDA

6.8

7.2

7.6

8.0

8.4

1996 1998 2000 2002 2004 2006 2008

ZAMBIA

Fig. 1. The tendency of RERs given natural logarithms for 15 COMESA and SADC countries – The USD base (December

1994 to July 2008)

3268 C.-W. Su et al.

Dow

nloa

ded

by [

Uni

vers

ite D

e Pa

ris

1] a

t 05:

04 0

3 A

ugus

t 201

3

may be due to the low power of these three univariateunit root tests when the RERs are highly persistent.This result implies that PPP did not hold for these 15COMESA and/or SADC countries during the sampleperiod. Another reason for finding (possible spur-ious) unit-roots could be due to the recently for-warded argument that RER processes are likely to benonlinear due to the existence of transaction costsand hence the power of these tests might be poor insuch situations (Wu and Lee, 2009). Furthermore, weknow that univariate unit root tests might have lowpower when they are applied to a finite sample. Inthis situation, the panel-based unit tests are found tobe of great help, provided that they allow for anincrease in the power of the order of the integrationanalysis by allowing the cross-sectional and temporaldimensions to be combined. Tables 4 and 5 report theresults for the first generation and second generationpanel unit root tests. Three first generation panel-based unit root tests – the Im–Pesaran–Shin (Imet al., 2003), the MW (Maddala and Wu, 1999) andthe Hadri (2001) tests – all yield the same results,indicating that RERs are nonstationary in these 15COMESA and/or SADC countries. A serious draw-back of the first generation panel-based unit roottests is that they do not take (possible) cross-sectionaldependencies into account in the panel-based unitroot test procedure. O’Connell (1988) points out thatfailure to consider contemporaneous correlationsamong data will bias the panel-based unit roottest toward rejecting the joint unit root hypothesis.

Cross-sectional dependencies are taken into accountby the second generation panel unit root tests.Hence, these methods offer a superior way to studythe long run behaviour of the RERs. Both the error-component model (Choi, 2002) and the covariancerestriction model (Chang, 2002) are used in our study.Table 5 reports the results of these two secondgeneration panel-based unit root tests and results also

Table 3. Univariate unit root tests: ln(RER)

Levels First differences

Country ADF PP KPSS GLS-DF ADF PP KPSS GLS-DF

Botswana �1.935(2) �1.726[4] 0.259[10]*** �0.459[0] �7.112(1)*** �12.307[3]*** 0.074[2] �2.471[6]***Burundi �1.031(1) �0.834[5] 0.279[10]*** �1.237[1] �14.420(0)*** �14.451[5]*** 0.093[7] �25.851[0]***Egypt �1.100(3) �0.934[7] 0.198[10]** 0.111[0] �4.967(2)*** �11.703[7]*** 0.331[7] �2.029[10]**Ethiopia �0.273(4) 0.223[7] 0.384[10]*** �1.261[0] �3.147(3)** �7.748[5]*** 0.124[4]* �17.413[0]***Kenya 1.365(5) 0.705[7] 0.944[10]*** 1.045[1] �4.131(8)*** �9.407[6]*** 0.046[11] �14.872[0]***Madagascar �2.409(2) �2.137[4] 0.107[10] �1.060[0] �7.528(0)*** �12.541[2]*** 0.105[2] �16.177[0]***Malawi �2.280(12) �1.851[5] 1.066[10]*** �1.038[0] �3.492(11)*** �10.871[4]*** 0.100[5] �17.766[0]***Mauritius �1.074(0) �1.1631[4] 0.306[10]*** 0.222[0] �12.818(0)*** �12.821[3]*** 0.076[0] �1.637[8]*Mozambique �1.957(1) �1.789[5] 0.238[10]*** 0.192[0] �8.427(0)*** �8.424[2]*** 0.150[5] �15.505[0]***Seychelles �1.921(1) �1.523[2] 0.624[10]** �0.342[3] �7.777(2)*** �8.575[10]*** 0.067[3] �9.794 [2]***South Africa �1.770(0) �1.826[2] 0.294[10]*** �1.175[0] �11.770(0)*** �11.769[1]*** 0.230[2] �2.902[8]***Sudan 0.629(5) �0.220[6] 0.345[10]*** �1.129[0] �7.985(4)*** �19.803[16]*** 0.044[14] �17.297[0]***Tanzania �1.470(12) �1.389[1] 1.138[10]*** �0.576[0] �2.809(11)* �8.573[5]*** 0.122[2] �19.599[0]***Uganda �1.607(0) �1.607[6] 0.374[10]*** �0.902[0] �12.376(0)*** �12.379[6]*** 0.120[12]* �18.231[0]***Zambia �0.186(0) 0.103[7] 0.977[10]*** �0.985[0] �12.826(0)*** �12.894[7]*** 0.0494[10] �17.213[0]***

Notes: The numbers within parentheses indicate the lag order selected based on the recursive t-statistic, as suggested by Perron(1989). The numbers within square brackets indicate the truncation for the Bartlett Kernel, as suggested by the Newey andWest (1987) test.*, ** and *** indicate significance at the 0.1, 0.05 and 0.01 levels, respectively.

Table 4. First generation panel-based unit root tests:

ln(RER)

Model Test-statistic p-value

IPS (�t) 2.052 0.979IPS (�LM) �1.500 0.933MW 12.651 0.999Hadri (hom) 14.526*** 0.000Hadri (het) 11.500*** 0.000

Notes: Hadri (hom) and Hadri (het) denote the Hadri KPSStest assuming homogeneity and heterogeneity, respectively,in the estimation of the long-run variance.*** Indicates significance at the 0.01 probability level.

Table 5. Second generation panel-based unit root tests:

ln(RER)

Model Test-statistic p-value

Chang IV (2002) 1.212 0.779Choi (2002) 1.003 0.633

Revisiting purchasing power parity for African countries 3269

Dow

nloa

ded

by [

Uni

vers

ite D

e Pa

ris

1] a

t 05:

04 0

3 A

ugus

t 201

3

indicate that RERs are all nonstationary in these 15COMESA and/or SADC countries. Our resultssignify that the RER is a random process. In otherwords, based on these (second generation) panel unitroot tests, PPP does not hold for these 15 COMESAand/or SADC countries.

As we stated earlier that panel-based unit root testsare joint tests of a unit root for all members of a paneland that they are incapable of determining the mix ofI(0) and I(1) series in a panel setting, Panel SURKSStests investigate a separate unit-root null hypothesisfor each and every individual panel member. In doingso, they clearly identify how many and which series inthe panel are stationary processes. Wu and Lee’s(2009) Panel SUKSS test results indicate that thereare four stationary bilateral RERs among these 15COMESA and/or SADC countries – Botswana/USD,Burundi/USD, Madagascar/USD and Seychelles/USD, as shown in Table 6. These results indicatethat PPP is valid for only four of these 15 COMESAand/or SADC countries. To avoid the small-samplesize bias, we estimate the 1%, 5% and 10%

critical values, obtained from simulations based onobservations for each series and 10 000 replicationsusing the lag and covariance structure from the panelof RER data series for each of the 15 panelmembers.8 These are also presented in Table 6.

Recently, several papers (see, e.g. Akinboade andMakina, 2006; Zhou, 2008; Baharumshah et al., 2010)have raised much concern about the choice of thenumeraire currency in validating the parity condition.The available evidence on PPP depends to some extenton the base country. To demonstrate the robustness ofour empirical results, we also consider both the euroand the German Deutsche Mark (DM). We explorethe hypothesis using both the euro andDMas the basecurrency to the same set of data. According to theresults, both the euro-based and DM-based RERs(not reported here but are available from the authorsupon request) also provide weak evidence favouringthe long-run validity of PPP for these 15 COMESAand/or SADC countries under study.9 It is worthnoting that our empirical evidence appears to beinsensitive to the choice of numeraire currency in

Table 6. Results of nonlinear panel unit root test: ln(RER)

Critical values

Country � SURKSS(t�̂) 10% 5% 1%

Botswana �0.720 �3.467** �3.074 �3.401 �3.968Burundi �0.488 �3.593*** �2.747 �3.019 �3.544Egypt �0.064 �0.564 �2.731 �3.007 �3.575Ethiopia 0.134 0.949 �2.595 �2.889 �3.451Kenya �0.023 �0.467 �2.725 �3.022 �3.574Madagascar �1.534 �4.362*** �2.943 �3.235 �3.854Malawi �0.199 �1.895 �2.641 �2.916 �3.505Mauritius �1.220 �2.015 �2.808 �3.103 �3.640Mozambique �0.486 �1.931 �2.708 �2.996 �3.562Seychelles �0.936 �2.896* �2.726 �3.012 �3.600South Africa �0.195 �2.660 �3.024 �3.336 �3.894Sudan �0.163 �1.089 �2.669 �2.978 �3.517Tanzania �0.289 �1.055 �2.712 �2.995 �3.557Uganda �0.308 �2.288 �2.724 �3.032 �3.598Zambia �0.051 �0.811 �2.624 �2.904 �3.462

Notes: The estimated critical values are tailored by the simulation experiments based on 212 observations for each series and10 000 replications, following the work by Breuer et al. (2001). The error series were generated in such a manner to benormally distributed with the variance-covariance matrix given from the SUR estimation of the 15 countries’ panel structures.Each of the simulated RER-DOLLAR was then generated from the error series using the SUR estimated coefficients on thelagged differences.*, ** and *** indicate significance at the 0.1, 0.05 and 0.01 levels, respectively.

8As this test has nonstandard distributions, the Critical Values (CVs) of the SURKSS test must be obtained through MonteCarlo simulations. In the simulations, the intercepts, the coefficients on the lagged values for each series were set equal to zero.In what follows, we obtain the lagged differences and the covariances matrix from the SUR estimation on the actual exchangerate data. The SURKSS test-statistic for each of the 15 series was computed as the t-statistic calculated individually for thecoefficient on the lagged level. To obtain the CVs, the experiments were replicated 10 000 times and the CVs of 1%, 5% and10% are tailored to each of the 15 panel members.9A similar study done by Chang et al. (2010b) using Panel SURADF test of Breuer et al. (2001) also provide weak evidence offavouring PPP in these 15 African countries. Chang et al. (2010b) find that the long-run PPP only holds true for Botswana,Madagascar and Malawi.

3270 C.-W. Su et al.

Dow

nloa

ded

by [

Uni

vers

ite D

e Pa

ris

1] a

t 05:

04 0

3 A

ugus

t 201

3

the model. Our study is in line with the work of Li(2007) where he finds that PPP would be valid at thelow volatility state. Our results seem to be consistentwith this finding that PPP holds only in Botswana,Madagascar and the Seychelles (with the exception ofBurundi) with less volatile currency (Table 2).

Economic and policy implications

It should be noted in this regard that in the literature,PPP is reportedly more likely to hold in countrieswith higher inflation (Rogoff, 1996). Of particularinterest here is that compared with their main tradingpartners, African countries generally have higher(often double digit) inflation rates. For instance,countries like Botswana (11%), Burundi (14%),Madagascar (17%) and Malawi (31%) all recordeddouble-digit annual rates of inflation in the 1990s.10

Evidence on PPP in high-inflation countries (regimes)has led some scholars to believe that nominalexchange rate to follow the PPP path more closely(McNown and Wallace, 1989; Mahdavi and Zhou,1994; Holmes, 2001). Copeland (1989), for example,argued that high inflation penalizes agents for main-taining sticky prices and so attempts to fix thenominal exchange rates may be undermined. Holmes(2001) also argued that a regime of high inflation(possible as a result of monetary disturbance) sug-gests that the large size of price change dominates,and thus, the nominal exchange rate follows its PPPpath more closely. We would have expected that PPPshould hold true for most of the African countriesunder study, but our results are not consistent withthis expectation. It is worth noting that the resultshere are not consistent with those of Krichene (1998)and Nagayasu (2002), Holmes and Wang (2005),Chang et al. (2006), Nejib (2008) and Baharumshahet al. (2010) which support PPP for various groups ofAfrican countries.

Our empirical results might source from severalfactors such as differences in technology/productivityand preferences, different factor endowments, tradebarriers, transportation costs and differences in priceindex formations. The African countries experiencedreal shocks, such as droughts, reductions in the termsof trade, oil price shocks, civil wars and other formsof political instability during the past three decades.These problems could trigger destabilizing effects onthe PPP relationship in African countries. Periods ofpolitical instability are clearly associated with rapidrates of domestic currency depreciation, priceincreases and inflation in Africa. Another possible

explanation for the violation of the PPP is that theperiods of strong real appreciation which often implyinterventions in the exchange rate markets of thesecountries (see, e.g. the RER plots for Egypt,Ethiopia, Mauritius, Mozambique, South Africa,Sudan and Uganda, shown in Fig. 1). Brissimiset al. (2005) argue that long-run PPP is not likely tobe evidenced for economies in which the monetaryauthorities target the exchange rate and intervene inthe exchange rate market to support the pursuedexchange rate rule. They claim that policy behaviouraffects the short-run adjustment to PPP and theability to uncover long-run PPP empirically, evenwhen PPP holds. Thus, a possible empirical rejectionof PPP in these 15 COMESA and/or SADC countriescould be attributed to the policy makers’ interven-tions in the foreign exchange markets. Our finding isin line with the view that PPP (i.e. a stable equilib-rium level for the RER) is inadequate in the contextof RERs of African economies, which appear to belargely characterized by a strongly appreciating trend(Taylor and Sarno, 2001). The results also confirmthe view that real shocks are likely to be far moreimportant than nominal (e.g. monetary or financial)shocks in driving RER movements in these countriesover the sample examined. Therefore, it is possible toclaim that deviations in the short-run form the PPPare prolonged for most of these COMESA and/orSADC countries and there are no forces which arecapable of bringing the exchange rate back to its PPPvalues in the long-run.

One major policy implication of our study is thatthe validity of using PPP to determine the equilibriumexchange is clear and the governments of these fourcountries, such as Botswana, Burundi, Madagascarand the Seychelles, can use PPP to predict exchangerate that determine whether a currency is over orundervalued and experiencing difference betweendomestic and foreign inflation rates. Nevertheless,reaping unbounded gains from arbitrage in tradedgoods is possible for the rest of the 11 countries thatwe study here.

IV. Conclusions

Using monthly data for the period of December 1994to July 2008, this study empirically tests whether PPPholds among 15 COMESA and/or SADC countries.The results from the univariate unit root and twogenerations’ panel-based unit root tests all fail tosupport the PPP throughout all 15 countries.

10 Compared to other countries, Botswana (11%), Burundi (14%), Madagascar (17%) and Malawi (31%) all recordeddouble-digit annual rates of inflation and higher than average of the group of (9.82% during this sample period).

Revisiting purchasing power parity for African countries 3271

Dow

nloa

ded

by [

Uni

vers

ite D

e Pa

ris

1] a

t 05:

04 0

3 A

ugus

t 201

3

However, when we conduct Panel SURKSS tests, wefind PPP holds true for four of these 14 COMESAand/or SADC countries. Concerning policy, ourstudy implies that PPP can be used to determine theequilibrium exchange rate for Botswana, Burundi,Madagascar and the Seychelles four countries.A major policy implication of our study is that thevalidity of using PPP to determine the equilibriumexchange is clear – but for only four of these 15COMESA and/or SADC countries. Nevertheless,reaping unbounded gains from arbitrage in tradedgoods is not possible in these four countries.

Finally, as we know that exchange rates might beaffected by internal and external shocks generated bystructural changes may be subject to considerableshort-run variation. In fact, as we can see from Fig. 1,it seems that there exist some structure breaks in thedata series. Perron (1989) argued that if there is astructural break, the power to reject a unit rootdecreases when the stationary alternative is true andthe structural break is ignored. Meanwhile, structuralchanges present in the data generating process, buthave been neglected, sway the analysis towardsaccepting the null hypothesis of a unit root. Futurestudy will be in this direction.11

References

Akinboade, O. A. and Makina, D. (2006) Mean reversionand structural breaks in real exchange rates: SouthAfrica evidence, Applied Financial Economics, 16,347–58.

Baharumshah, A. Z., Lau, E. and Nziramasanga, M. T.(2010) Purchasing power parity in African countries:evidence from Panel SURADF test, South AfricanJournal of Economics, 78, 40–56.

Bahmani-Oskooee, M. (1993) Purchasing power paritybased on effective exchange rate and cointegration: 25LDCs experience with its absolute formulation, WorldDevelopment, 21, 1023–31.

Bahmanee-Oskooee, M. and Gelan, A. (2006) Testing thePPP in the non-linear STAR framework: evidencefrom Africa, Economics Bulletin, 6, 1–15.

Baum, C. F., Barkoulas, J. T. and Caglayan, M. (2001)Nonlinear adjustment to purchasing power parity inthe post-Bretton Woods era, Journal of InternationalMoney and Finance, 20, 379–99.

Breuer, J. B., McNown, R. and Wallace, M. S. (2001)Misleading inferences from panel unit root tests withan illustration from purchasing power parity, Reviewof International Economics, 9, 482–93.

Brissimis, S., Sideris, D. and Voumvaki, F. (2005)Testing long-run purchasing power parity underexchange rate targeting, Applied Financial Economics,24, 959–81.

Chang, Y. (2002) Non-linear unit root tests in panels withcross-sectional dependency, Journal of Econometrics,110, 261–92.

Chang, T., Chang, H. L., Chu, H. P. and Su, C. W. (2006)Does PPP hold in African countries? Further evidencebased on a highly dynamic nonlinear (logistic) unitroot test, Applied Economics, 38, 2453–9.

Chang, T., Liu, W. C., Yu, C. P. and Kang, S. (2010a)Purchasing power parity for ten Latin Americanintegration association countries: panel SURKSStests, Applied Economics Letters, 17, 1575–80.

Chang, T., Lu, Y. C., Tang, D. P. and Liu, W. C. (2011)Long-run purchasing power parity with asymmetricadjustment: further evidence from African countries,Applied Economics, 43, 231–42.

Chang, T., Tang, D. P., Liu, W. C. and Lee, C. H. (2010b)Purchasing Power Parity for 15 COMESA and SADCcountries: evidence based on panel SURADF tests,Applied Economics Letters, 17, 1721–7.

Choi, I. (2002) Combination unit root tests forcross-sectionally correlated panels, mimeo,Hong Kong University of Science and Technology.

Choi, I. and Chue, T. K. (2007) Subsampling hypothesistests for nonstationary panels with applications toexchange rates and stock prices, Journal of AppliedEconometrics, 22, 233–64.

Copeland, L. (1989) Exchange Rates and InternationalFinance, Addison-Wesley, New York.

Dickey, D. A. and Fuller, W. A. (1981) Likelihood ratiostatistics for autoregressive time series with a unit root,Econometrica, 49, 1057–72.

Elliott, G., Rothenberg, T. and Stock, J. (1996) Efficienttests for an autoregressive unit root, Econometrica, 64,813–36.

Enders, W. and Chumrusphonlert, K. (2004) Thresholdcointegration and purchasing power parity in thePacific nations, Applied Economics, 36, 889–96.

Enders, W. and Granger, C. W. F. (1998) Unit-root testsand asymmetric adjustment with an example using theterm structure of interest rates, Journal of BusinessEconomics and Statistics, 16, 304–11.

Hadri, K. (2001) Testing for stationarity in heterogeneouspanel data, Econometrics Journal, 3, 148–61.

Holmes, M. J. (2001) New evidence on real exchange ratestationarity and purchasing power parity in lessdeveloped countries, Journal of Macroeconomics, 23,601–14.

Holmes, M. J. and Wang, P. (2005) Do African countriesmove asymmetrically towards purchasing powerparity?, South African Journal of Economics, 73,292–301.

Im, K. S., Pesaran, M. H. and Shin, Y. (2003) Testing forunit roots in heterogeneous panels, Journal ofEconometrics, 115, 53–74.

Juvenal, L. and Taylor, M. P. (2008) Thresholdadjustment of deviations from the law of one price,Studies in Nonlinear Dynamics and Econometrics, 12,1–44.

Kapetanios, G., Shin, Y. and Snell, A. (2003) Testing for aunit root in the nonlinear STAR framework, Journal ofEconometrics, 112, 359–79.

11Recently, we have worked on a model called panel SURKSS with a Fourier function to capture the structural breaks. Themodels have been proved to be more powerful than that of panel SURKSS without a Fourier function. We have developedthe programme codes and the codes are available from the authors upon request.

3272 C.-W. Su et al.

Dow

nloa

ded

by [

Uni

vers

ite D

e Pa

ris

1] a

t 05:

04 0

3 A

ugus

t 201

3

Kargbo, J. M. (2003) Food prices and long-run purchasingpower parity in Africa, Development Southern AfricaJournal, 20, 321–36.

Kargbo, J. M. (2004) Purchasing power parity andexchange rate policy reforms in Africa, The SouthAfrican Journal of Economics, 72, 258–81.

Kargbo, J. M. (2006) Purchasing power parity and realexchange rate behavior in Africa, Applied FinancialEconomics, 16, 169–83.

Kilian, L. and Taylor, M. P. (2003) Why is it so difficult tobeat the random walk forecast of exchange rates?,Journal of International Economics, 60, 85–107.

Krichene, N. (1998) Purchasing power parities in five EastAfrican countries: Burundi, Kenya, Rwanda,Tanzania, and Uganda, IMF Working Paper,WP/98/148, African Department, InternationalMonetary Fund, Washington, DC.

Kwiatkowski, D., Phillips, P., Schmidt, P. and Shin, Y.(KPSS) (1992) Testing the null hypothesis of statio-narity against the alternative of a unit root: how sureare we that economic time series have a unit root?,Journal of Econometrics, 54, 159–78.

Levin, A., Lin, C. F. and Chu, C. S. (2002) Unit root testsin panel data: asymptotic and finite-sample properties,Journal of Econometrics, 108, 1–24.

Li, M. Y. L. (2007) Purchasing power parity under high andlow volatility regimes, Applied Economics Letters, 14,581–9.

Lothian, J. R. and Taylor, M. P. (2000) Purchasing powerparity over two centuries: strengthening the case forreal exchange rate stability: a reply to Cuddington andLiang, Journal of International Money and Finance, 19,759–64.

Lothian, J. R. and Taylor, M. P. (2008) Real exchange ratesover the past two centuries: how important is theHarrod–Balassa–Samuelson effect?, Economic Journal,118, 1742–63.

MacDonald, R. and Taylor, M. P. (1992) Exchange-rateeconomics: a survey, International Monetary FundStaff Papers, 39, 1–57.

Maddala, G. S. and Wu, S. (MW) (1999) A comparativestudy of unit root tests with panel data and a newsimple test, Oxford Bulletin of Economics andStatistics, 61, 6316–65.

Madsen, J. and Yang, B. (1998) Asymmetric priceadjustment in a menu cost model, Journal ofEconomics, 68, 295–309.

Mahdavi, S. and Zhou, S. (1994) Purchasing power parityin high inflation countries: further evidence, Journal ofMacroeconomics, 16, 403–22.

McNown, R. and Wallace, M. (1989) Cointegration testsfor long run equilibrium in the monetary exchange ratemodel, Economics Letters, 31, 263–7.

Nagayasu, J. (2002) Does the long-run PPP hypothesis holdfor Africa? Evidence from a panel cointegration study,Bulletin of Economic Research, 54, 181–7.

Nejib, H. (2008) The purchasing power parity and thesymmetry, proportionality conditions: panel cointe-gration evidence from some African countries,International Research Journal of Finance andEconomics, 16, 121–36.

Newey, W. K. and West, K. D. (1987) A simple positivedefinite heteroskedasticity and autocorrelation consis-tent covariance matrix, Econometrica, 55, 703–8.

Obstfeld, M. and Taylor, M. P. (1997) Nonlinear aspects ofgoods-market arbitrage and adjustment: Heckscher’s

commodity point revisited, Journal of the Japanese andInternational Economies, 11, 441–79.

O’Connell, P. G. J. (1998) The overvaluation of purchasingpower parity, Journal of International Economics, 44,1–20.

Papell, D. H. (1997) Searching for stationarity: purchasingpower parity under the current float, Journal ofInternational Economics, 43, 313–32.

Perron, P. (1989) The great crash, the oil price shock,and the unit root hypothesis, Econometrica, 57,1361–401.

Pesaran, M. H. (2007) A simple panel unit root test in thepresence of cross-section dependence, Journal ofApplied Econometrics, 22, 265–312.

Phillips, P. C. B. and Perron, P. (PP) (1988) Testing for aunit root in time series regression, Biometrika, 75,335–46.

Rogoff, K. (1996) The purchasing power parity puzzle,Journal of Economic Literature, 34, 647–68.

Sarno, L. (2000) Real exchange rate behavior in the MiddleEast: a re-examination, Economic Letters, 66, 127–36.

Sarno, L. (2005) Viewpoint: towards a solution to thepuzzle in exchange rate economics: where do westand?, Canadian Journal of Economics, 38, 673–708.

Sarno, L. and Taylor, M. P. (2001) Official intervention inthe foreign exchange market: is it effective and, if so,how does it work?, Journal of Economic Literature, 39,839–68.

Sarno, L. and Taylor, M. P. (2002) Purchasing power parityand the real exchange rate, IMF Staff Papers, 49,65–105.

Sarno, L., Taylor, M. P. and Chowdhury, I. (2004)Nonlinear dynamics in deviations from the law ofone price: a broad-based empirical study, Journal ofInternational Money and Finance, 23, 1–25.

Taylor, M. P. (1995) The economics of exchange rates,Journal of Economic Literature, 33, 13–47.

Taylor, M. P. (2004) Is official exchange rate interventioneffective?, Economica, 71, 1–11.

Taylor, M. P. and Peel, D. A. (2000) Nonlinear adjustment,long-run equilibrium and exchange rate fundamentals,Journal of International Money and Finance, 19, 33–53.

Taylor, M. P., Peel, D. A. and Sarno, L. (2001) Nonlinearmean-reversion in real exchange rates: toward asolution to the purchasing power parity puzzles,International Economic Review, 42, 1015–42.

Taylor, M. P. and Sarno, L. (1998) The behavior of realexchanges during the post-Bretton Woods period,Journal of International Economics, 46, 281–312.

Taylor, M. P. and Sarno, L. (2001) Real exchange ratedynamics in transition economies: a nonlinear analysis,Studies in Nonlinear Dynamics and Econometrics, 5,153–77.

Taylor, A. M. and Taylor, M. P. (2004) The purchasingpower parity debate, Journal of Economic Perspectives,18, 135–58.

Wu, J. L. and Lee, H. Y. (2009) A revisit to the non-linearmean reversion of real exchange rates: evidence from aseries-specific non-linear panel unit-root test, Journalof Macroeconomics, 31, 591–601.

Zellner, A. (1962) An efficient method of estimatingseemingly unrelated regressions and tests for aggrega-tion bias, Journal of the American StatisticalAssociation, 57, 348–68.

Zhou, S. (2008) Stationarity of Asian-Pacific real exchangerates, Economics Letters, 98, 16–22.

Revisiting purchasing power parity for African countries 3273

Dow

nloa

ded

by [

Uni

vers

ite D

e Pa

ris

1] a

t 05:

04 0

3 A

ugus

t 201

3