Money Demand Function: Heterogeneous Panel Application 1

Transcription

1 Money Demand Funcion: Heerogeneous Panel Applicaion 1 Nasri Harb UAE Universiy Mailing address: Economics Deparmen UAE Universiy P.O. Box Al-Ain, Unied Arab Emiraes Tel: (971) Fax: (971) nasri.harb@uaeu.ac.ae May 2003 Keywords: Money Demand, Time series, Panel Coinegraion, FMOLS. JEL Classificaions: C22, C23, E41, F41. 1 I hank M. Alawad and H. Sherif a UAE Universiy for helpful commens. 1

2 Absrac We consruc an aggregae daa panel for GCC's six counries and verify he coinegraion hypohesis among he variables of he money demand funcion using Pedroni's heerogeneous panel coinegraion ess (1995, 1997, and 2001). Then, we esimae he idiosyncraic, panel and group mean coinegraing vecors using FMOLS and a modified version of FMOLS developed by Pedroni (2000). We find srong evidence of coinegraion amongs our variables. The esimaed elasiciies have he expeced signs in general bu are significan only in he case of he scale variable. The group mean esimaor produce sensible resuls and shows resuls conformed o he heory. Real M1 shows a beer performance han real M2. 2

3 1 Inroducion Money Demand Funcion (MDF) has been a leading subjec in applied economics in developed counries because i provides he moneary auhoriy wih a clear view of he reacion of differen macroeconomic aggregaes o a change in money supply. Laer, i has become a subjec of ineres in many developing counries as well in order o faciliae he implemenaion of an accurae moneary policy. Sriram (1999, 2001) presened a horough review of recen sudies covering his opic. This sudy aims o esimae he elasiciies of he MDF for he six counries of he Gulf Cooperaion Council, GCC: Bahrain, Kuwa Oman, Qaar, Saudi Arabia, and he Unied Arab Emiraes. This work of research is imporan in his period because he six counries are moving owards a single currency by The six counries peg heir naional currencies o he US dollar 2. They have esablished a cusom union in early 2003 in a move owards a closer economic union. In many counries, he esimaion of he MDF could no be performed because of shor daa span, which lowers he power of uni roo ess required before proceeding o he esimaion. Many counries, for insance, have only yearly available daa wih a maximum span of 20 o 30 years. To overcome his problem, economericians usually use one of wo opions. They eiher inerpolae daa o increase frequency and herefore, expand he number of observaions, or build a panel. For he GCC counries, he firs opion has been used Darra and Muawaa (1996) and Khaib and Towaijari (1999). Darra and Muawaa (1996) esimaed he Unied Arab Emiraes' MDF. They inerpolaed non-oil GDP o obain quarerly daa and esimaed an error correcion model by OLS. They regressed he firs difference of he log of UAE real M1 on he firs difference of each of he log of non-oil GDP, local and foreign ineres raes, log of inflaion rae and nominal exchange rae plus an error correcion erm. They showed ha M1 demonsraes an empirical superioriy. The parameers are sable and have he expeced signs even if some of hem are saisically no significan (local and foreign ineres rae). However, Phillips (1992) showed ha despie is superconsisency, he OLS esimaor's disribuion is no normal. Moreover, 2 In January 2003, Kuwai has pegged is dinar o he US dollar in preparaion for he single currency. 3

4 Banerjee e al (1986) demonsraed ha he number of observaions in small samples may be oo lile for he superconsisency o overcome he effec of endogeneiy of he regressors. Khaib and Towaijari (1999) have also used OLS o esimae Saudi Arabia's money demand funcion. They regressed he log of real M1 on he log of non-oil GDP, local ineres rae, expeced inflaion rae and real exchange rae. They used he residuals o esimae an error correcion model. They concluded ha he effec of he ineres rae is low and saisically non significan. They explained his by he islamic social and culural values in Saudi Arabia. In he presen aricle, we ake a differen direcion o overcome he shor span daa and he bias of he OLS esimaor. Precisely, his aricle exends he lieraure in hree ways: (1) i uses a panel o es for non saionariy and coinegraion of he daa. Indeed, Perron (1991) and ohers suggesed ha he span of he daa is more imporan o he power of coinegraion es han he frequency, (2) moreover, our mehodology allows for heerogeneiy across he members of he panel. Pedroni (1995, 1999, and 2001) demonsraed ha imposing homogeneiy falsely across a panel has some dramaic effecs as we will see below. He proposes a se of saisics, which are used in he presen aricle, o es he null of no coinegraion wih heir asympoic disribuions, and (3) his aricle uses Fully Modified Leas Squares (FMOLS) o esimae he idiosyncraic elasiciies of MDF. FMOLS has been proposed by Phillips (1992) who showed ha i dominaes OLS and Maximum Likelihood esimaion even in small samples in he presence of coinegraion. I has been used frequenly in recen coinegraion lieraure. Besides, Pedroni (2000) showed ha he FMOLS can be adjused o capure he panel's poenial heerogeneiy. This paper uses a modified version of FMOLS as well o esimae he panel's elasiciies. The major resuls of his aricle can be summarized as follows: The differen variables used in our esimaion are nonsaionary and coinegraed as suggesed by he heory. We do no find srong evidence of he non saionariy of M2. Our ess show evidence of coinegraion amongs he variables. Elasiciies wih respec o he scale variable are in general conformed o he heory in mos cases and a he panel level. The oher idiosyncraic elasiciies have he correc signs in many insances bu are generally no significan. However, our group mean esimaor shows some gain of power. The 4

5 elasiciies wih respec o income and ineres rae are conformed o he heory. Our sudy shows ha, in line wih previous sudies, M1 ouperforms M2. The remaining of his aricle is organized as follows: secion ( 2) inroduces he heoreical model and discusses he mehodology, secion ( 3) displays major resuls, and secion ( 4) concludes. 2 The Model and he mehodology 2.1 The Model In he lieraure, he money demand is formulaed as he following: ' ' ( S, O ) f > 0, f < 0 M (1) = f S O where M is he quaniy demanded of money, S is a scale variable and O is he opporuniy cos of holding money. The variables used for he esimaion of he money demand funcion depend on he heoreical funcion of money. Two major approaches have been used: 1) ransacions approach, and 2) asses approach. In he ransacions approach, households choose beween wo sores of value: liquid money and ineres bearing asses. Thus, in deciding how o allocae heir porfolios, hey balance beween he marginal benefi of liquid money and is opporuniy cos, i.e. he ineres rae. The asses approach is based on he consumer demand heory approach, which assumes ha households derive uiliy by holding real and financial asses, which include money. The quaniy demanded of money is decided by solving for he opimal porfolio allocaion. The major facors affecing he money demand in his case are: oal wealh and opporuniy cos, which consiss mainly on expeced rae of inflaion 3. Since he firs approach reas money as a means of ransacions, i emphasizes he liquid aspec of money, which suggess using M1 as he moneary aggregae. On he oher hand, he asse approach emphasizes he role of money as a financial asse. Therefore, a broader money aggregae, M2, is used o include subsiues for liquid money such as erm deposis. Two major argumens were cied in favor of using a broader moneary aggregae. The firs is ha a broader definiion pus he researcher in he safe side when she conducs an 3 Oher approaches like precauionary demand approach and cash in advance approach have been used as well. See Sriram (1999) for deails. 5

6 empirical sudy because i is less sensiive o major financial innovaions. The oher argumen is he sable relaionship beween money, broadly defined, and he nominal income. We conduc his sudy using alernaively boh money definiions M1 and M2 o compare i wih previous lieraure. The choice of he scale variable depends also on he role of money. When money is seen as a means of ransacions, GNP or GDP is used as scale variable. Bu, when i is seen as an asse, he righ scale variable should be oal wealh for which daa scarcely exis. Proxies have been used insead: permanen income, which can be consruced from presen and expeced fuure income, privae consumpion (C) and GDP. We use he las wo variables alernaively in his sudy. In erms of opporuniy cos of holding money, many measures have been used in he lieraure. Among hose measures is he expeced inflaion rae because by holding one dollar oday, he real value of money decreases while ha of real asses does no. Anoher measure is he price of alernaive asses. Since he economic agens choose beween he reurn on alernaive asses and reurn on money, he would prefer he higher reurn. Wheher o use inflaion rae or ineres rae is a pracical maer. Indeed, boh variables have been included simulaneously in many sudies especially in developing counries wih high inflaion raes. Few auhors used inflaion rae alone like Eken e al (1995) and Yashiv (1994) who uses inflaion and ineres raes alernaively. We use only he nominal ineres rae in wha follows. The main reason is ha he concerned counries have no experienced periods of excessive inflaion. In an open economy model, wo more addiional opporuniy coss have been used: he expeced exchange rae of he naional currency and he foreign ineres rae. If he local currency is expeced o depreciae, agens prefer o subsiue heir local currency by a foreign and sable currency. Since five members of he panel have had a fixed peg o he US dollar (he only excepion is Kuwai which used o fix is currency o SDR unil January 2003), we use Official Exchange Rae series which is he amoun of he local currency per one uni of SDR because his variable represens an average of he exchange rae versus a se of oher currencies. The definiion of he exchange rae is he number of local currency per one uni of SDR. Therefore, an appreciaion of he local currency is seen as a decrease in he exchange rae. Because no series on expeced exchange rae is 6

7 available, we use he acual series as a proxy. Foreign money and capial markes are subsiues for he local ones, hey may have a subsanial effec on money demand in developing counries because heir financial markes are no well developed. We use LIBOR as an indicaor for he foreign ineres rae variable. Thus, we esimae he following money demand equaion m = β + β y + β r + β x + β l + ε, (2) where he i index refers o he given member of he panel. m is he naural logarihm of he real moneary aggregae for counry i. y is he logarihm of he real scale variable, r is local nominal ineres rae, x is he nominal exchange rae and l is he foreign ineres rae which is he same for all panel's members (LIBOR). β z, z = 0 4, are he slopes o be esimaed. β 0 is he idiosyncraic fixed effec of each member of he panel. β 1 is he long run elasiciy of real money demand wih respec o he scale variable. This elasiciy is expeced o be posiive. The sign of β 2 depends on he moneary aggregae. Since curren and checking accouns in he GCC region bear no ineres rae, we would expec a negaive β 2, since people swich o ineres bearing accouns (saving and erm deposis) when ineres rae increases. On he oher hand, when M2 is used as moneary aggregae, β 2 is expeced o be posiive because an increasing ineres rae, ceeris paribus, makes money deposis more aracive. β 3 and β 4 are expeced o be negaive since depreciaion of local currency and increase in foreign ineres raes make naional money a less favorable sore of value. The five series in he model are expeced o be non-saionary. However, rewriing he previous equaion in he following way ε = m β β y β r β x β l, (3) suggess ha he five variables mus be coinegraed, wih he following coinegraing vecor {1, - β 1, -β 2, - β 3, -β 4 }. Since here is no reason o expec a homogeneous coinegraing vecor across members, and since imposing a homogeneous vecor across he panel may lead o dramaic consequences, as we will see in deails below, we use he heerogeneous panel ess and esimaions echniques as described in Pedroni (1995, 1997, 2000 and 2001). Those echniques are deailed in he following subsecion. 7

8 2.2 The Mehodology We sar by esing for he exisence of uni roos in he daa series. Many ess have been proposed o es for he null of no saionariy in panels. Quah (1992, 1994) ess assume homogeneiy across members and do no, herefore, allow for any heerogeneiy across he panel. Levin and Lin (1993) ess exend Quah's (1992, 1994) work. They derived he asympoic disribuion for panel uni roo ess ha allow for member-specific effecs and ime rend. The ess impose dynamic homogeneiy on he auoregressive coefficien ha indicaes he presence or he absence of a uni roo. Im, Pesaran and Shin (1997) (IPS, hereafer) proposed a uni roo es ha allow for heerogeneiy across members and residual serial correlaion. I consiss of esing he null ha λ i =1 (where i indicaes he cross secional member) agains he alernaive ha λ i =1 < 1 in some or all i in x = i + θ 1 λi ) pi 1 + ρ j x j (4) j= 1 µ ( x + υ where x is he ime series o be esed, x i = x x 1 is he firs difference and µ i is, he fixed effec. θ i allows for an idiosyncraic linear rend for each group while ν is i.i.d 4. The resuling LM-bar saisic is Γ LM = N 1 N [ LM NT N E{ η }] i= 1 Ti 1 N N Var{ η } i= 1 Ti where LM is he average of he N LM saisics, N being he number of he panel's members. E{ η Ti } and { Ti } Var η are respecively he average and he variance of ηti which is a pooled log-likelihood funcion defined in IPS (1997). An alernaive es presened in IPS (1997) is he -bar, which is based on he Augmened Dickey-Fuller es, i.e. 1 N [ NT N E{ (,0)}] i 1 T p = i 1 N N Var ( p,0) i= 1 { } T i (5) N Γ = (6) { } where E i ( p,0) and i ( p,0), T i { } Var are respecively he average and variance of he, T i Augmened Dickey-Fuller disribuion of order p i Mone Carlo experimens show ha 4 IPS (1997) presened a modified es o allow for serially correlaed disurbances as well. 8

9 IPS (1997) ess ouperform Levin and Lin's (1993) es. They have greaer power and beer small-sample properies. Moreover, IPS (1997) showed ha -bar es has beer performance over LM-bar es when N and T are small. In addiion, hey proposed a cross-secionally demeaned version of boh ess o be used in he case where he errors in differen regressions conain a common ime-specific componen. The nex sep is o es for coinegraion among he variables. In convenional ime series, he same uni roo ess can be applied for boh raw daa and residuals wih proper adjusmens o he criical values when applied o he laer. Bu, Pedroni (1995, 1997, and 2001) showed ha esing for coinegraion in panel daa is no so sraighforward. The only case in which raw daa and residuals have equivalen disribuion is when he regressors are sricly exogenous and when he pooled OLS slope is consrained o be homogeneous. This is due o he fac ha in his case, he OLS esimaor converges o a non-random value. However, if he regressors are no sricly exogenous in panel daa, he observed ha: In he case where he alernae hypohesis is ha he coinegraing relaionship is no consrained o be homogeneous across members and he parameers esimaion is allowed o vary across individual members hen, proper adjusmens should be made o he es saisics hemselves. If no, his may have he effec of ransforming a convergen es saisic ino a divergen one asympoically. In pracice, his means ha as he sample size grows large, one is cerain o rejec he null of no coinegraion regardless of he rue relaionship. Imposing homogeneiy falsely across members when he rue relaionship is heerogeneous generaes an inegraed componen in he residuals making hem non-saionary leading an economerician o conclude ha her variables are no coinegraed even if hey really are. For hese reasons, he developed a se of saisics o es he null of no coinegraion for he case of heerogeneous panels and derived heir asympoic disribuions. In paricular, he considered he following ime series panel regression y = i. + δ i + X β i + e α (7) 9

10 where y and X are he observable variable wih dimension of (N*T) 1 and (N*T) m respecively. N is he number of members in he panel and T is he ime span of each panel member. α i and δ i are he idiosyncraic fixed effec and deerminisic rend, β i is an m 1 parameer vecor for each panel and e is a vecor of residuals. The variables y and X are inegraed of order 1. He defines wo ses of saisics. The firs se of hree saisics Z ˆ ν NT, Z ˆ ν NT 1 and Z dimension of he panel. The saisics are as follows NT is based on pooling he residuals along he wihin ˆ ˆ N T Z L, 1 e (8) i ν NT i= 1 = 1 N T 1 Z ˆ NT eˆ 1 λ 1 i= 1 = 1 i= 1 = 1 N T ( eˆ e i ) i, 1 ˆ i, ρ (9) 1 N T N T Z ~ 2 σ n eˆ i ( eˆ eˆ i ) = = λ NT, 1 i, 1 i, (10) i 1 1 i= 1 = 1 where e ˆ i, 1 is he residual vecor of he OLS esimaion of (7) and where he oher erms are properly defined in Pedroni (2001). The hree saisics are respecively analogous o he panel variance raio, panel rho, and panel saisics in Phillips and Ouliaris (1990). The second se of saisics is based on pooling he residuals along he beween dimension of he panel. I allows for a heerogeneous auocorrelaion parameer across members. The saisics are as follows: N T 1 ~ Z ˆ NT eˆ 1 λ 1 i= 1 = 1 = 1 T ( eˆ e i ) i, 1 ˆ i, ρ (11) N T 1 / 2 ~ Z NT n eˆ 1 i= 1 = 1 = 1 T ( eˆ eˆ i ) i, 1 i, ~ σ λ (12) These saisics compue he group mean of he individual convenional ime series saisics 5. The asympoic disribuion of each of hose five saisics can be expressed in he following form: 5 A group mean variance raio saisics is no presened because i is dominaed by he wo oher saisics. 10

11 X N, T µ ν N N(0,1) (13) where X N,T is he corresponding form of he es saisic, while µ and ν are he mean and variance of each es respecively. They are given in Table 2 in Pedroni (1999). Under he alernaive hypohesis, Panel ν saisic diverges o posiive infiniy. Therefore, i is a one sided es were large posiive values rejec he null of no coinegraion. The remaining saisics diverge o negaive infiniy, which means ha large negaive values rejec he null. As we saed previously, in he presence of I (1) variables, he effec of superconsisency may no dominae he endogeneiy effec of he regressors if OLS is used. This resuls in a biased and a non normal disribuion of he residuals. This disribuion depends also on he nuisance parameers associaed wih he serial correlaion of he daa. The problem is amplified in a panel seing by he poenial dynamic heerogeneiy over he cross secional dimension. Specifically, as his dimension increases, second order biases could be expeced o occur by he poor performance of he esimaors designed for large samples as hey are averaged over he panel's members. For his reason, Pedroni (2000) showed how FMOLS can be modified o make inferences in coinegraed panels wih heerogeneous dynamics as he cross secional dimension becomes large even wih relaively shor ime series. 3 Resuls The daa used in he presen aricle are annual. They cover he six counries members of he GCC and run from 1979 o Therefore, we have 22 annual observaions for each member. We use non-oil GDP in he case of he 5 oil producers: Kuwa Oman, Qaar, Arabia and UAE. Real variables are obained by dividing he nominal variable by he corresponding Consumer Price Index. Deails of all he daa sources and ypes are presened in he appendix. The following subsecion (3.1) discusses he exisence of he uni roo in he daa series while subsecions ( 3.2) and ( 3.3) discuss he major resuls of he paper. 11

12 3.1 Uni Roo Tes As menioned above, we use IPS (1997) -bar es o verify he exisence of uni roo in he panel series. The LM-bar es resuls will be presened as well for he sake of comparison only. IPS (1997) demonsraed ha -bar es has more power han he LMbar es. The resuls are shown in able (1). In general, our variables are non saionary. However, wo cases should be discussed. M2 rejecs he null of non saionariy in mos cases. This is ypical for developing counries as menioned above where M1 shows a beer performance. We ried o verify wheher nominal M2 behaves beer han real M2, bu he resuls were no encouraging. We will, herefore use real M2 in he group panel only as suggesed by IPS ess. On he oher hand and wha is surprising is ha real consumpion does no rejec he null in all four cases while real GDP does rejec in he case of demeaned daa wih a consan only. Theoreically, his should no be he case because boh of hem normally share he same rend and are herefore coinegraed. This may be explained by he low qualiy of daa. Hence, we'll assume ha our GDP is non saionary and proceed cauiously and rely more on he case of M1 and consumpion. The las wo columns show ha he firs order differences of our variables rejec easily he non-saionariy hypohesis. Since LIBOR series is he same for all members, we use he convenional ess of ADF and Phillips-Perron (PPS). These ess are presened in able (2). Boh ess suppor he hypohesis of non-saionariy of LIBOR series, while hey boh rejec he null of nonsaionariy of he firs order difference. The differen ess on our variables demonsrae ha hey are inegraed of order one. Nex sep would be o es for coinegraion among our se of variables. 3.2 Coinegraion Analysis The resuls of he coinegraion analysis ess are presened in able (3). Those ess are developed by (Pedroni 1995, 1997, 2001). The resuls of he ADF es are presened in he same able for he sake of comparison only. Since M2 is an I(1) a he group mean level, we presen is resuls a his level only. Whi M1 and real GDP, he only es ha rejecs he null of no coinegraion is he es, using eiher panel daa or group demeaned daa. When consumpion is used as he scale 12

13 variable, he evidence of coinegraion is weaker, where he null is rejeced only a he group level. Those resuls are no encouraging especially ha he es ends o over rejec he null of no coinegraion while panel-ν and panel-ρ ess end o under rejec i in case of small N and T (Pedroni 1995, 1997, 2001) which is our case. For insance, a he 95% significance level, he rae of rejecion is 10% insead of 5% for he es and is beween 1% and 3% for panel-ν and panel-ρ ess. This may explain he non-rejecion of he null using he las wo ess. Using M2 as he moneary aggregae shows some weak evidence of coinegraion a he group level. By he heory, hose variable should be coinegraed. Bu, our resuls do no suppor he heory. To improve our resuls, we invesigae he relaionship beween local ineres rae and LIBOR. Since five of he concerned counries peg heir currencies o he US dollar, we would expec ha heir ineres raes are highly correlaed wih american ineress. To make our poin, we es he coinegraion relaionship beween boh ineres raes. The resuls are shown in able (5) which shows a clear evidence of coinegraion beween boh variables. Since boh variables are I(1) and share he same rend, here is no need o include boh of hem in our equaion. Therefore, we can remove he LIBOR variable and obain he following money demand funcion m = β + β y + β r + β x + ε. (14) Table (5) shows he coinegraion ess amongs he four remaining variables: moneary aggregaes, scale variables, local ineres rae and exchange rae. There is a much sronger evidence of coinegraion amongs (m1, gdp, r, x) even wih he Panel-ν which ends o under rejec he null of no coinegraion. Also, he null is rejeced when we use aggregae consumpion, wih weaker significance hough. A he group levels, we find some evidence of coinegraion also. We conclude ha our variables are coinegraed using eiher moneary aggregaes or eiher scale variables. 3.3 FMOLS Esimaion As menioned above, since he OLS esimaor yields a biased disribuion of he residuals, we use FMOLS mehodology proposed by Phillips (1992) o esimae he idiosyncraic 13

14 coinegraion vecors and he modified FMOLS mehodology proposed by Pedroni (2000) o esimae he panel's coinegraion vecor. FMOLS is superior o OLS when applied o heerogeneous panel wih I (1) variables. Is disribuion is sandard and is asympoically unbiased and free of nuisance parameers. Moreover, while he OLS esimaor does no yield consisen sandard errors in he presence of endogenous regressors, FMOLS resuls in consisen sandard errors and herefore consisen -saisics. The modified FMOLS proposed by Pedroni (200) allows for heerogeneiy by allowing he associaed serial correlaion properies of he errors processes o vary across members of he panel. The resuls of he regressions esimaions are shown in able (6). Real M1 (and M2) and real GDP (real non-oil GDP) were used as moneary aggregae and scale variable respecively. The individual elasiciies wih respec o GDP, β 1 are posiive and significan in four cases ou of six as expeced by he heory and are close o each oher. The income elasiciy in case of Kuwai is no significan bu posiive. When he parameer is negaive, i is non significan. However, regarding he elasiciy wih respec o ineres rae, we observe ha i is no significan, for all counries while he elasiciy wih respec o he exchange rae is significan in hree cases ou of six.. In he case of aggregae consumpion, idiosyncraic esimaions do no differ much. These resuls are in line wih he sudy of Khaib and Towaijari (1999) who obained negaive bu non significan semi elasiciy. The differen in he sign may be resuling from 1) heir use of an inerpolaed daa covering he 1977 o 1997 period, while our daa cover he 1979 o 2000 period and 2) while hey use OLS for heir esimaion, we use FMOLS. This non significance of he semi elasiciy wih respec o local ineres rae may be explained by he Islamic culure which prohibis paying and receiving ineress. Those idiosyncraic elasiciies esimaion wih FMOLS should be inerpreed wih grea cauion because of he shor span of he daa (22 observaions). No longer series were available publicly. Bu, since our panel has more power wih longer span daa series (132 observaions), is resuls are more reliable. The panel esimaions are shown in he boom rows of able (6). We provide wo esimaions: pooled panel esimaion and group mean esimaion. The panel esimaion pools he daa of all members, bu allows for heerogeneous serial correlaion properies across members. The pooled panel -saisic 14

15 can be used o es H 0 : β i =β 0 for all i versus H 1 : β i =β a β 0 where β 0 is he hypohesized common value for β under he null and β a is an alernaive common value. Bu, he group mean esimaor allows for heerogeneous elasiciies and allows o es H 0 : β i =β 0 for all i versus H 1 : β i β 0 for all so ha he value of β is no necessarily consrained o be he same across he members under H 1. Two more advanages are cied in favor of group mean esimaor: 1) when he rue coinegraing vecors are heerogeneous, he group mean esimaor provides he mean value of he coinegraing vecors and 2) is -saisic exhibis lile disorion in small sample, (Pedron 2000). The pooled esimaor does no show much difference compared o individual FMOLS esimaion using eiher scale variable. However, we obain differen resuls group mean esimaor. Even if he elasiciy wih respec o GDP and C is higher and significan, he semi elasiciy wih respec o ineres rae is significan, while i is no wih respec o he exchange rae. Since no heory predics a common value for he elasiciies (or semi elasiciies) wih respec o our variables, we are emped o accep he group mean esimaor as more accurae han he pooled panel. Therefore, wih GDP used as scale variable, we may conclude ha he average semi elasiciy wih respec o ineres rae is negaive as expeced by he heory while ha of he exchange rae is no. Does his mean ha he effec of he islamic culure has disappeared? No. The socieies in our concerned counries are a mix of muliculural ehniciies, which are no all islamic. Islamic banking is no he dominan form of banking in hose counries. For insance, he oal liabiliies of he whole banking sysem were $99,448.6million whereas he oal liabiliies of Islamic banking insiuions in Bahrain were $1,348 million, a iny 1.3% of he oal banking sysem. Finally, in he las row of able (6), we presen he esimaion resuls when M2 is he moneary aggregae. I shows a poorer behavior han M1 in line wih previous money demand sudies in developing counries. 4 Conclusion Despie bad qualiy of daa, we were able o esimae sensible elasiciies of money demand funcion using heerogeneous panel daa for he GCC counries. Even if he individual esimaion shows no effec of ineres rae on money demand, he group mean 15

16 esimaion demonsraes ha i does. Whereas M1 shows a good performance in our research, M2 does no. A major exension o his work is he parameers sabiliy es. To conduc a sound moneary policy, he cenral banker needs o make sure ha he elasiciies are sable hroughou ime. However, such es for a heerogeneous panel has no been developed ye. I is a new challenge for he economericians. 16

17 Appendix: Daa informaion Official exchange rae (aa) has been used as he exchange rae variable he six member of he panel. Daa was obained from IFS CD-ROM. 1) Bahrain a) ineres rae: average ime deposi rae, b) Inflaion: Bahrain's CPI, c) Real GDP: (Nominal GDP)/CPI, d) Real consumpion: Nominal Consumpion/CPI. Daa Sources: Inernaional Financial Saisics CD-Rom, Minisry of Finance and Naional Economy of he Sae of Bahrain' web sie, Foreign rade of he Arab Counries, published by he Arab Moneary Fund. 2) Kuwai: a) ineres rae: deposi rae, b) Inflaion: Kuwai's CPI, c) Real GDP: (Nominal GDP-oil expors)/cpi, d) Real consumpion: Nominal Consumpion/CPI. Daa Sources: Inernaional Financial Saisics CD-Rom, Minisry of Finance and Naional Economy of he Sae of Bahrain' web sie, Foreign rade of he Arab Counries (various issues), published by he Arab Moneary Fund. 3) Oman: a) ineres rae: average ime deposi rae, b) Inflaion: Oman's CPI, c) Real GDP: (Nominal GDP-oil expors)/cpi, d) Real consumpion: Nominal Consumpion/CPI. Daa Sources: Inernaional Financial Saisics CD-Rom, Prices and Financial Saisics in he ESCWA Region (various issues), Unied Naions, Economic and Social Commission for Wesern Asia. 4) Qaar: a) ineres rae: Deposi Rae, b) Inflaion: Qaar's CPI, c) Real GDP: (Nominal GDP-oil expors)/cpi, d) Real consumpion: Nominal Consumpion/CPI. Daa Source: 17

18 Inernaional Financial Saisics CD-Rom, Foreign rade of he Arab Counries, published by he Arab Moneary Fund, The Arabic Economic Unied Repor (various issues), published by he Arab Moneary Fund, Minisry of Finance and Naional Economy of he Sae of Bahrain' web sie. 5) Saudi Arabia: a) ineres rae: Deposi Rae (Bahrain's ineres rae used form missing daa), b) Inflaion: Saudi's CPI, c) Real GDP: (Nominal GDP-oil expors)/cpi, d) Real consumpion: Nominal Consumpion/CPI. Daa Sources: Inernaional Financial Saisics CD-Rom, Arab Counries Economic Indicaors (various issues), published by he Arab Moneary Fund. 6) Unied Arab Emiraes: a) ineres rae: average ime deposi rae, b) Inflaion: UAE's CPI, (Some observaions were missing. We esimae he missing observaions by aking he averages of he previous and he following year.) c) Real GDP: (Nominal GDP-oil expors)/cpi, d) Real consumpion: Nominal Consumpion/CPI. Daa Sources: Inernaional Financial Saisics CD-Rom, Foreign rade of he Arab Counries, published by he Arab Moneary Fund, Economic Bullein (various issues), Cenral Bank of he UAE. 18

19 Bibliography [1] Alyousha, A. (1997): Invesigaing Bahrain business cycles, Applied Economics, vol. 29, Issue 1, pp [2] Banerjee, A., Dolado, J.J., Hendry, D. F., and Smih, G.W. (1986), Exploring Equilibrium Relaionships in Economerics hrough Saic Models: some Mone Carlo Evidence, Oxford Bullein of Economics and Saisics, vol. 48, [3] Darra, A. and Muawa, A. (1996): Modeling Money Demand in he Unied Arab Emiraes, The Quarerly Review of Economics and Finance, vol. 36, No. 1. [4] Eken, S., Cashin, P., Erbas Nur S., Marelkino, J., and Mazara A. (1995): Economic Dislocaion and Recovery in Lebanon, IMF Occasional Paper, No [5] Hossain, Akhar (1994): The Search for a Sable Money Demand Funcion for Pakisan: An Applicaion of he Mehodology of Coinegraion, The Pakisan Developmen Review, vol. 33, No. 4, pp [6] Im, K., Pesaran, H. and Shin, Y. (1997): Tesing for Uni Roos in Heerogeneous Panels, Universiy of Cambridge, DAE Working Paper, No [7] Khaib-Kswan M and Towaijar H. A. (1999): Coinegraion Error Correcion and he Demand for money in Saudi Arabia, Economia-Inernazionale, vol. 52, No. 3, pp [8] Laidler, D. (1993): he Demand for Money: Theories, Evidence and Problems, Harper Collins College Publishers, 4 h ediion, New York. [9] Levin, A. and Lin,C. F. (1993): Uni Roo Tess in Panel Daa: New Resuls, Universiy of California, San Diego, Discussion Paper, No [10] Moosa, Imad A. (1992): The Demand for Money in India: A Coinegraion Approach, The Indian Economic Journal, vol. 40, No. 1, pp [11] Pedron P. (1997): Panel Coinegraion: Asympoic and Finie Sample Properies of Pooled Time Series Tess wih an Applicaion o he PPP Hypohesis: New Resuls, unpublished manuscrip, Indiana Universiy. [12] Pedron P. (1999): Criical Values for Coinegraion Tess in Heerogeneous Panels wih Muliple Regressors, Oxford Bullein of Economics and Saisics, vol. 61, No. 4, pp [13] Pedron P. (2000): Fully Modified OLS for Heerogeneous Coinegraed Panels, Advances in Economerics, vol. 15, pp. 19

20 [14] Pedron P. (2001): Panel Coinegraion: Asympoic and Finie Sample Properies of Pooled Time Series Tess wih an Applicaion o he PPP Hypohesis, unpublished manuscrip, Indiana Universiy. [15] Perron, P. (1991): Tes Consisency wih Varying Sampling Frequency, Economeric Theory, vol.7, pp [16] Phillips, P. (1992): Hyper-Consisen Esimaion of a Uni Roo in Time Series Regression, Cowles Foundaion Discussion Papers, No [17] Phillips, P. and Ouliaris, S. (1990): Asympoic Properies of Residual Based Tess for Coinegraion, Economerica, No. 58, pp [18] Quah, D. (1992): Inernaional Paerns of Growh: Persisence in Cross-Counry Dispariies, unpublished manuscrip, London School of Economics. [19] Quah, D. (1992): Exploiing Cross-Secion Variaions for Uni Roo Inference in Dynamic Daa, Economic Leers, 44, pp [20] Sriram, S. (1999): Survey of Lieraure on Demand for Money: Theoreical and Empirical Work wih Special Reference o Error-correcion Models, IMF Working Paper, WP/99/64. [21] Sriram, S. (2001): A Survey of Recen Empirical Money Demand Sudies, IMF Saff Papers, Vo. 47, No 3, pp [22] Yashiv, E. (1994): Money Demand in a High Inflaion Economy: he Case of Israel, he Review of Economics and Saisics, V. 76, pp

21 Table 1: IPS ess Firs order difference Variable -bar LM-bar -bar LM-bar M1 Raw daa Consan 0.21*** 0.86*** Consan+ rend -0.84*** 1.21*** Demeaned daa Consan 0.81*** -0.52*** Consan+ rend -0.36*** 0.57*** M2 Raw daa Consan Consan+ rend Demeaned daa Consan -1.63** Consan+ rend GDP Raw daa Consan 0.59*** 0.47*** Consan+ rend -2.07* Demeaned daa Consan Consan+ rend -2.19* Consumpion Raw daa Consan -1.66* 2.17* Consan+ rend -2.05* Demeaned daa Consan -0.30*** 0.25*** Consan+ rend -1.90* 2.31* Consan -0.49*** Raw daa Nominal Consan+ rend 0.49*** ineres rae Consan -1.61** Demeaned daa Consan+ rend -1.17*** Consan -0.51*** Raw daa Consan+ rend -0.25*** Exchange rae Consan -0.68*** Demeaned daa Consan+ rend -0.29*** ***canno rejec he null of non-saionariy a he 10% level. **canno rejec he null of non-saionariy a he 5% level. *Canno rejec he null of non-saionariy a he 1% level. 0.32*** -0.40*** 1.86* 1.53** 0.47*** 0.65*** 0.69*** 0.71*** Table 2: ADF and Phillips & Perron Tess Firs order difference Variable ADF PP ADF PP Consan -1.92* -1.34* LIBOR Consan+ rend -2.32* -2.06* *canno rejec he null of non-saionariy a he 5% level

22 Table 3: Coinegraion Analysis Tess (equaion 2) Tes Consan wihou rend Consan + rend Panel-ν Panel-ρ Panel- Panel-adf Group-ρ Group- Group-adf Panel-ν Panel-ρ Panel- Panel-adf Group-ρ Group- Group-adf Moneary aggregae=m1, Scale variable=gdp ** -2.87*** *** -4.12*** Moneary aggregae=m1, Scale variable=consumpion ** -3.02*** Moneary aggregae=m2, Scale variable=gdp ** -2.39*** ** -3.05*** * -2.53*** Group-ρ Group- Group-adf *** -4.40*** Moneary aggregae=m2, Scale variable=consumpion ** -2.36*** Group-ρ Group- Group-adf *** -2.36*** *rejecs he null of no coinegraion a he 10% level. **rejecs he null of no coinegraion a he 5% level. ***rejecs he null of no coinegraion a he 1% level *** -1.87*** 22

23 Table 4: Ineres Raes Coinegraion Tess Tes Consan wihou rend Consan + rend Panel-ν Panel-ρ Panel- Panel-adf Local ineres rae and LIBOR 3.59*** -2.75*** -2.86*** -3.93*** **rejecs he null of no coinegraion a he 5% level. ***rejecs he null of no coinegraion a he 1% level. 2.25*** -1.42** -2.66** -2.82*** Table 5: Coinegraion Analysis Tess (equaion 14) Tes Consan wihou rend Consan + rend Panel-ν Panel-ρ Panel- Panel-adf Group-ρ Group- Group-adf Panel-ν Panel-ρ Panel- Panel-adf Group-ρ Group- Group-adf Moneary aggregae=m1, Scale variable=gdp 1.61* *** -4.38*** *** -7.03*** Moneary aggregae=m1, Scale variable=consumpion 1.51* * -1.58* *** -3.33*** Moneary aggregae=m2, Scale variable=gdp ** -3.01*** ** -3.64*** ** Group-ρ Group- Group-adf *** -5.08*** Moneary aggregae=m2, Scale variable=consumpion ** -1.66** Group-ρ Group- Group-adf ** -2.08** *rejecs he null of no coinegraion a he 10% level. **rejecs he null of no coinegraion a he 5% level. ***rejecs he null of no coinegraion a he 1% level ** -1.70** 23

24 Table 6: FMOLS and Modified FMOLS regressions GDP panel Consumpion panel counry GDP ineres Ex. rae C Ineres Ex. rae Bahrain 0.77 *** (6.47) (-0.87) 0.66 (1.20) 0.51 *** (6.33) 0.00 (0.22) (-0.04) Kuwai 0.26 (1.37) 0.02 (1.09) *** (-3.43) 0.17 (1.01) 0.02 (0.80) *** (-3.12) Oman 0.85 *** (15.68) (-1.05) *** (-2.51) 0.63 *** (19.10) (-0.31) (-1.19) M1 Qaar (-1.28) (-0.46) *** (-2.65) (-1.36) (-0.84) *** (-3.45) Arabia 0.86 *** (6.84) 0.01 (0.72) 0.07 (1.79) 1.49 *** (9.96) 0.03 *** (2.26) 0.13 *** (4.61) UAE 1.00 *** (12.21) (-1.51) 0.03 (0.61) 0.94 *** (6.80) 0.00 (0.00) ** (-1.96) Panel 0.59 *** (16.85) (-0.85) -0.37*** (-2.04) 0.56 *** (17.08) 0.00 (0.87) *** (-2.11) Group 0.78*** (11.48) -0.05*** (-2.36) 0.04 (1.85) 0.65*** (11.17) (0.13) (0.87) M2 Group 0.42 *** (5.52) 0.01 (0.33) (0.58) 0.32 *** (6.54) *** Significan coefficien wih 95% confidence level (1.67) 0.03 (1.52) 24