Further Evdence on Fnance-Growth Causaly: A Panel Data Analyss Chrysost BANGAKE Laboratore d Econome d Orléans (LEO), Unversé d Orléans. Faculté de Dro, d Econome et de Geston. Rue de Blos BP : 6739. 45067 Orléans Cedex 2. E.mal : chrysost.bangake@unv-orleans.fr Jude C. EGGOH Laboratore d Econome d Orléans (LEO), Unversé d Orléans. Faculté de Dro, d Econome et de Geston. Rue de Blos BP : 6739. 45067 Orléans Cedex 2. E.mal : comlanv-jude.eggoh@unv-orleans.fr Abstract (Frst draft June 2009) Ths paper reassesses long-run relatonshp between fnancal development and economc growth usng panel ntegraton and contegraton technques for a dynamc heterogeneous panel of 7 countres both developed and developng over the perod 960-2004. Our results provde clear support for the exstence of a sngle long-run equlbrum between fnancal development, economc growth and ancllary varables. Furthermore, we show that there s a b-drectonal causaly between fnance and growth as well at the level of the global sample as regards the varous groups of countres. Keywords: Fnancal development; economc growth; panel causaly; panel contegraton. J.E.L. Classfcaton: O, O6, O33.. Introducton Does fnancal development promote economc growth, or does economc growth cause fnancal development? Although ths queston dates back at least to Schumpeter (9), the semnal contrbuton of Kng and Levne (993a, 993b) revved the nterest of subject and gave a boost to ncreasng academc researches. There are numerous emprcal analyses whch addressed the relatonshp between fnancal development and economc growth. However, the emprcal attempts by cross-secton and panel data studes to address as accurately as possble the mpact of fnancal development on growth have faled to gve a consensus or a clear drecton. Furthermore, these results rely more on the correlaton than the causaly relatonshp between both varables. Indeed, several theoretcal arguments allow us to pont out that the causaly relatonshp between fnance and growth cannot be solved only by cross secton and panel data analyses. Quah (993) formally shows the lack of balanced paths across country whch volates the basc hypothess behnd the averagng and poolng of cross country data. Evan (995) ponts out the heterogeney of slope coeffcents across countres as a man lm of a cross secton analyss. An alternatve approach conssts n assessng the relatonshps between fnancal development and economc growth wh tme seres. Although tme seres favor to the dea of causaly than cross secton, theses studes rely more on anterory and posterory between both varables.
The drecton of causaly between fnancal development and economc growth s crucal because two oppose streams of research have held dfferent ponts of vews: the supplyleadng and followng demand hypothess (Patrck, 966). The supply-leadng hypothess poss a causal relatonshps runnng from fnancal development to economc growth. The dea s that delberate creaton of fnancal nstutons and markets ncreases the supply of fnancal servces and thus leads to real economc growth. However, the demand-followng hypothess postulates a causal relatonshp from economc growth to fnancal development. Indeed, an ncreasng demand for fnancal servces mght nduce an expanson n fnancal sector as the real economc growth. Nevertheless, Patrck (966) ponts out that the fnancegrowth causaly mght be accurately explored. Although many emprcal studes have nvestgated the causal relatonshp between fnancal development and economc growth, the results are stll ambguous. Two knds of studes are done: the frst generaton s studes used tme-seres whle the second generaton s reled on panel data causaly. Contrbutons by Demetrades and Hussen (996), Luntel and Khan (999), and Arests and Demetrades (200) are n lne of frst generaton studes. These studes have employed the Granger noncausaly and the Johansen contegraton tests between fnancal development and economc growth and report mxed results. Ths shows that a consensus on the role of fnancal development n the process of economc growth does not so far exst. Recent researches on long-run relatonshp between fnance and growth use panel causaly and contegraton technques. Works by Rousseau and Wachtel (2000) and Fler et al. (2003) provded an emprcal evdence of causaly from fnance to growth. Emprcal attempts by Chrstopoulos and Tsonas (2004) also support an undrectonal causaly from fnancal depth to growth. Contrary to prevous works, Hurln and Venet (2004) fnd that causal relatonshp from growth to fnance, whch s more occurred n developed than n developng countres. A smlar approach s used by Apergs et al. (2007) who examne whether a longrun relatonshp between fnance and growth exsts employng panel ntegraton and contegraton tests. Ther results support a b-drectonal causaly between fnancal deepenng and growth, whch remans robust to varous specfcaton of the sample. In the same ven, Dufrénot et al. (2007) show a strong long-run relatonshp between fnance and growth n the OECD countres; gven the heterogeney n non member OECD countres, s not possble to detect long-term relaton between the two varables. Ths paper ams to study contegraton and causaly relatonshp between fnancal development and economc growth. In the lne of the second generaton studes, our man contrbutons to the lerature are as follows. Frst, there s a scarcy of emprcal works on causaly relatonshp between fnancal development and economc growth usng panel data analyses. Ths study tres to fll ths gap. In so dong, we attempt to resolve the ssue of causaly,.e., whether better functonng fnancal markets exert a causal nfluence on growth or vce versa followng the methodology of Pesaran et al. (999). Second, our study departs from Apergs et al. (2007) n three aspects: we use new data, large sample and several tests of un root (Im, Pesaran and Shn, 999; Maddala and Wu, 2004; and Cho, 200). We also use contegraton test of Pedron (999), the Dynamc OLS and the Pool Mean Group estmator proposed by Pesaran et al. (999). Three dfferent measures of fnancal development are used to capture the dfferent channels through whch fnance can affect growth. Our results support the exstence of a sngle long-run relatonshp between fnancal development, economc Demetrades and Hussen (996) fnd a ltle systematc evdence n favor of the vew that fnance nvolves growth. Indeed, they found that for the majory of ther sample, causaly s b-drectonal, whle n some cases fnance follows growth. Luntel and Khan (999) also concluded that the causaly between fnance and growth s b-drectonal, usng a sample of ten developng countres. 2
growth and a set of control varable. Furthermore, our evdence ndcates clearly that there exsts a b-drectonal causaly between fnance and growth. These results are robust to varous specfcaton of our sample. The remander of the paper s organzed as follows: Secton 2 presents the emprcal approach used to study the relatonshp between fnancal development and economc growth. Secton 3 dscusses the man fndngs and robustness. Fnally, Secton 4 summares and concludes. 2. Econometrc methodology In ths secton, we successvely present the panel un root, contegraton tests and panel contegraton estmaton. 2. Testng for ntegraton Before proceedng to contegraton technques, we need to verfy that all varables are ntegrated to the same order. In dong so, we have used panel un root tests due to Im, Pesaran and Shn (2003), Maddala and Wu (999) 2 and Cho (200). These tests are less restrctve and more powerful compared to frst generaton tests developed by Levn and Ln (993) 3. Indeed, a drawback of these tests s that they do not allow for heterogeney n the autoregressve coeffcent. The tests proposed by IPS perm to solve Levn and Ln s seral correlaton problem by assumng heterogeney between uns n a dynamc panel framework. The basc equaton for the panel un root tests for IPS s as follows: y p y y ;,2,..., N; t,2,...,, (), t, t j, t j, t T j where y, t stands for each varable under consderaton n our model, s the ndvdual fxed effect and p s selected to make the resduals uncorrelated over tme. The null hypothess s that 0 for all versus the alternatve hypothess s that 0 for some,..., N and 0 for N,..., N. The IPS statstc s based on averagng ndvdual ADF statstcs and can be wrten as follows: N t ( tt ), (2) N where t T s the ADF t-statstc for country derved from the country-specfc ADF regresson as n Eq (). IPS show that under the null hypothess of non statonary, the t statstc follows the standard normal dstrbuton asymptotcally. The man queston s to know the asymptotc dstrbuton of the t statstc. Under the hypothess the resduals are not serally correlated and t T are ndependent and dentcally dstrbuted IPS propose the use of a group mean t bar statstc gven by: a N t _ barnt Et T Z tbar. (3) Var t T When the resduals are serally correlated, the standardzed follows: t bar statstc s modfed as 2 Henceforth, IPS for Im, Pesaran and Shn, and MW for Maddala and Wu. 3 For a useful survey on panel un root tests, see Hurln and Mgnon (2005) and Banerjee (999). 3
N N t _ barnt N E t p, 0 / 0 Wtbar p,. (4) N N Var t T p, 0 / 0 The W tbar p, statstc has a standard normal dstrbuton as N and T and N / T k, where k s a fne posve constant. Maddala and Wu (999) argue that whle the Im et al. tests relax the assumpton of homogeney of the root across the uns, several dffcultes stll reman. They propose the use of a test whch s based on combnng the p-values of the test-statstc for a un root n each cross-sectonal un. The MW statstc s gven by: N P 2 ln. (5) MW The P MW test s dstrbuted as Ch square wh 2N degrees of freedom under the hypothess of cross-sectonal ndependence. s the p-value of the un root tests. For the hgher values of N, Cho (200) has proposed a standardzed statstcs as follows: Z N N E 2ln. MW Var 2ln Under the null hypothess of a un root E 2ln 2 and 2ln 4 Var. Consequently, the Cho statstc s the centred and reduced average of the statstcs of MW ( N ). If the p-values are ndependent and dentcally dstrbuted, the Z MW statstcs N under null hypothess when N. converges to 0, Accordng to Breung (999), IPS test s not powerful when ndvdual trends are ncluded. Ths test s sensve to the specfcaton of determnstc trends. The MW test has the advantage over IPS because s value does not depend on dfferent lag lengths n the ndvdual ADF regressons (Chrstopoulos and Tsonas, 2004). Furthermore, Maddala and Wu (999) found that MW test s superor compared to IPS test. 2.2 Panel contegraton tests Once the order of statonary has been defned, we would apply Predron s contegraton test methodology. Indeed, lke the IPS and MW panel un root, the panel contegraton tests proposed by Pedron (999, 2004) also take n account heterogeney by usng specfc parameters whch are allowed to vary across ndvdual members of the sample. Takng account such heterogeney constutes an advantage because s unrealstc to assume that the vectors of contegraton are dentcal from an ndvdual to other for the panel. The mplementaton of Pedron s contegraton test requres estmatng frst the followng long run relatonshp: y x, 2 x2,... M xm, (7) for,..., N ; t,..., T ; m,..., M where N refers to the numbers of ndvdual members n the panel; T refers to the number of observaton over tme; M refers to the number of exogenous varables. The structure of estmated resduals s follows: ˆ ˆ uˆ. (8) ˆ (6) 4
Pedron has proposed seven dfferent statstcs to test panel data contegraton. Out of these seven statstcs, four are based on poolng, what s referred to as the Whn dmenson and the last three are based on the Between dmenson. Both knds of the tests focus on the null hypothess of no contegraton. However the dstncton comes from the specfcaton of the alternatve hypothess. For the tests based on Whn, the alternatve hypothess s for all, whle concernng the last three test statstcs whch are based on the Between dmenson, the alternatve hypothess s, for all. The fne sample dstrbuton for the seven statstcs has been tabulated by Pedron va Monte Carlo smulatons. The calculated statstc tests must be smaller than the tabulated crcal value to reject the null hypothess of absence of contegraton. 2.3 Panel contegraton Estmaton Although Pedron s contegraton methodology allows us to test the presence of the long run relatonshps, could not provde estmaton by error correcton model. To estmate a longrun relatonshp between the varables n panel framework n presence of contegraton, several estmators are proposed: dynamc OLS (DOLS), Pool Mean Group (PMG), OLS, Fully Modfed OLS (FMOLS). In ths paper, we consder two estmators wh error correcton: dynamc OLS (DOLS) and Pool Mean Group (PMG) because Kao and Chang (997, 2000) showed that both the OLS and Fully Modfed OLS (FMOLS) exhb small sample bas and that the DOLS estmator appears to outperform both estmators 4. 2.3. The Dynamc OLS (DOLS) Estmaton (Kao and Chang, 998, 200) The DOLS estmator proposed by Kao and Chang (998, 200) s an extenson of Stock and Watson s (993) estmator. In order to obtan an unbased estmator of the long-run parameters, DOLS estmator uses parametrc adjustment to the errors by augmentng the statc regresson wh the leads, lags, and contemporaneous values of the regressors n frst dfferences. Let us consder a panel model wh fxed effect: y x u,,..., N, t,..., T, (9) where y s a matrx,, s a vector of slopes k, dmenson, s ndvdual effect, u s an error term. We assume that x k, vector s autoregressve process of the frst order dfference: x (0). x The DOLS estmator s obtaned from the followng equaton: y jq x c x v. () 2 jq j j where c j s the coeffcent of a lead or lag of frst dfferenced explanatory varables. 2.3.2 Panel causaly: Pooled Mean Group (PMG) by Pesaran et al. (999) Our fnal step conssts of analysng the drecton of the panel data causal lnks among the varables. We apply the Pooled Mean Group (PMG) estmator due to Pesaran et al. (999). 4 See Kao and Chang (997, 2000) for more dscussons on the advantages of these estmators. 5
The Autoregressve Dstrbuted Lag ARDL ( p, q, q,..., q) model proposed by Pesaran et al. (999) s: y p j y, t j jx, t j, (0) j q j0 Where x k, s a vector of explanatory varables (regressors) for group, represent the fxed effect, j the coeffcent of the lagged dependent varables, and j are k coeffcent vectors. It s convenent to work wh the followng re-parameterzaton of (0). y y p q, t x j j0,2,...,n and t,2,..., T jy, t j j x, t j, () p, where q j,2,..., p and j m j m, j,2,..., q. q j j, j j p 0, j m j m, Pesaran et al. (999) assume that the ARDL ( p, q, q,..., q) model s stable f the roots of the p j followng equaton z 0 le outsde the un crcle. Ths assumpton ensures j j that 0, and hence there exsts a long-run relatonshp between y and x defned by y x where s a statonary process and the long run coeffcent are the same across the group. 3. Data and Emprcal results 3. Data The data set conssts of a panel of observatons for 7 countres both developed and developng 5 over the perod 960-2004. The Gross Domestc Product (GDP) per capa (constant prce 996, dollar US) s the real sector s ndcator. The fnancal development s measured through three varables n order to capture the varety of dfferent channels through whch fnance can effect growth: the rato of lqud lables to GDP (LL), the depos money bank assets to GDP (DEP) and the prvate domestc cred as rato to GDP (PRIV). We also use control varables as: government expendure as rato to GDP (GOV), the openness to trade as exports and mports dvded by GDP (OPEN) and the nflaton rate (INFL). 3. Panel Un root results The results from the panel un root test are presented n table. We tabulated two statstcs of IPS (wh constant and whout trend), and then we consder MW and Cho statstcs. All 5 The data are taken from the Penn World Table (PWT 6.2), and the fnancal data base realzed by Beck, Demrgüç-Kunt, and Levne (2005). The sample s the followng: 8 low ncome countres (Burkna Faso, Burund, Ivory Coast, Ethopa, Gamba, Ghana, Ha, Inda, Kenya, Madagascar, Nepal, Nger, Ngera, Pakstan, Rwanda, Senegal, Serra Leone, Togo); 30 mddle ncome countres (South Afrca, Argentna, Barbados, Bolva, Chle, Colomba, Costa Rca, Egypt, El Salvador, Ecuador, Gabon, Guatemala, Honduras, Iran, Jamaca, Malaysa, Morocco, Maurus, Panama, Domncan Republc, Paraguay, Peru, Phlppnes, Seychelles, Sr Lanka, Syra, Thaland, Trndad and Tobago, Uruguay, Venezuela); 23 hgh ncome countres (Australa, Austra, Belgum, Canada, Cyprus, Denmark, Fnland, France, England, Greece, Iceland, Ireland, Israel, Italy, Japan, Norway, New Zealand, Netherlands, Portugal, Sngapore, Sweden, Swzerland, USA). 6
varables are tested both n levels and frst dfference wh a constant and whout a trend. As can be nferred from table, the un-root hypothess cannot be rejected when the varables are taken n levels, except for the nflaton rate, whch s statonary n level. However, when frst dfferences are used the hypothess of un root non-statonary s rejected at the percent level of sgnfcance. These results lead us to conclude that our seres are characterzed as an I () process. The same results are obtaned n the sample of low ncome countres as n mddle and hgh ncome countres. Therefore, we can mplement a test for panel contegraton between fnancal development and economc growth. Table : Panel Un Root test Ztbar IPS Wtbar IPS P MW Z MW Level Dff Level Dff Level Dff Level Dff All countres GDP 9.374-37.0 *** 9.262-37.026 *** 94.52 638.49 *** -2.87 29.46 *** LIQ 4.54-29.726 *** 4.385-29.603 *** 00.3 582.2 *** -2.473 26.6 *** DEP 5.550-24.55 *** 5.399-24.037 *** 05.73 549.72 *** -2.5 24.93 *** PRIV 3.264-23.089 *** 3.093-22.982 *** 40.27 534.9 *** -0.02 23.272 *** OPEN 2.347-43.777 *** 2.304-43.574 *** 38.3 645.56 *** -0.693 29.88 *** GOV -2.04-44.929 *** -2.039-44.9 *** 72.59 653.93 ***.85 30.377 *** INFL -39.0 *** - -38.9 *** - 649.3 *** - 30. *** - Low ncome countres GDP.986-22.525 ***.909-22.37 *** 48.762 60.8 ***.504 4.709 *** LIQ 2.74-2.38 *** 2.53-2.090 *** 26.343 65.78 *** -.38 5.295 *** DEP.795-5.264 ***.780-5.229 *** 25.030 58.6 *** -.292 4.450 *** PRIV 2.826-6.63 *** 2.777-6.26 *** 24.495 64.39 *** -.355 5.32 *** OPEN -0.353-9.630 *** -0.365-9.487 *** 42.835 63.28 *** 0.805 5.000 *** GOV -0.359-26.070 *** -0.360-26.070 *** 35.702 65.78 *** -0.035 5.295 *** INFL -2.5 *** - -2.6 *** - 65.8 *** 5.2 *** Mddle ncome countres GDP 5.540-22.80 *** 5.490-22.78 *** 37.502 272.28 *** -2.053 9.378 *** LIQ 2.490-8.246 *** 2.43-8.94 *** 43.40 25.67 *** -.539 7.497 *** DEP.48-6.234 ***.40-6.57 *** 60.836 235.94 *** 0.076 6.06 *** PRIV -2.059-5.68 *** -2.05-5.082 *** 84.406 224.5 *** 2.227 4.985 *** OPEN.886-29.425 ***.849-29.36 *** 50.823 270.44 *** -0.837 9.2 *** GOV -2.6-29.723 *** -2.5-29.694 *** 80.235 276.3 ***.847 9.746 *** INFL -28.3 *** - -28.2 *** - 27.7 *** - 9.3 *** - Hgh ncome countres GDP 8.385-9.28 *** 8.326-9.28 *** 8.26 205.40 *** -3.934 6.68 *** LIQ 3.64-2.688 *** 3.09-2.60 *** 30.830 64.66 *** -.58 2.372 *** DEP 6.47-0.396 *** 6.279-0.339 *** 9.872 55.6 *** -2.724.380 *** PRIV 5.387-8.944 *** 5.586-8.98 *** 3.369 45.64 *** -.525 0.388 *** OPEN 2.283-25.942 *** 2.257-25.849 *** 36.656 2.83 *** -0.974 7.289 *** GOV -.729-2.930 *** -.728-2.930 *** 56.658 2.83 ***. 7.289 *** INFL -7. *** - -7.0 *** - 2.8 *** - 7.3 *** - *** Sgnfcant at percent. 3.2 Panel contegraton results The results of our contegraton analyss between fnance and growth are reported n table 2. We use four whn-group tests and three between-group tests to check whether the panel data are contegrated. The null hypothess of absence of contegraton at percent s rejected for all countres. Indeed, the dfferent statstcs to test panel data contegraton allow us to 7
conclude for the exstence of a long run relatonshps between economc growth (GDP) and the three ndcators of fnancal development (LIQ, PRIV and DEP). Moreover, ths result s robust to other specfcaton of our sample, because we fnd that fnance and growth are contegrated as well n low ncome countres as n mddle and hgh ncome countres. Table 2: Panel contegraton tests LIQ PRIV DEP All countres Statstc panel 20.02 *** 9.272 *** 9.832 *** Statstc panel -6.698 *** -6.85 *** -6.777 *** Statstc panel t (non para) -27.248 *** -26.799 *** -26.977 *** Statstc panel t (para) -25.789 *** -25.898 *** -25.40 *** Group panel -6.725 *** -6.578 *** -6.722 *** Group panel t (non para) -34.36 *** -32.980 *** -33.769 *** Group panel t (para) -32.50 *** -3.390 *** -3.064 *** Low ncome countres Statstc panel 0.835 *** 0.50 *** 0.595 *** Statstc panel -.986 *** -.724 *** -.75 *** Statstc panel t (non para) -8.883 *** -8.72 *** -8.374 *** Statstc panel t (para) -7.83 *** -6.860 *** -5.805 *** Group panel -2.06 *** -.607 *** -.855 *** Group panel t (non para) -22.79 *** -2.680 *** -22.597 *** Group panel t (para) -20.85 *** -9.323 *** -8.03 *** Mddle ncome countres Statstc panel 5.979 *** 5.535 *** 5.593 *** Statstc panel -.73 *** -.093 *** -.005 *** Statstc panel t (non para) -8.355 *** -7.794 *** -7.867 *** Statstc panel t (para) -6.973 *** -6.34 *** -5.954 *** Group panel -0.604 *** -0.444 *** -0.332 *** Group panel t (non para) -22.3 *** -2.73 *** -2.675 *** Group panel t (para) -9.905 *** -9.080 *** -8.504 *** Hgh ncome countres Statstc panel.09 *** 0.689 *** 0.753 *** Statstc panel -7.453 *** -7.545 *** -7.554 *** Statstc panel t (non para) -2.634 *** -2.84 *** -2.843 *** Statstc panel t (para) -.894 *** -2.804 *** -2.836 *** Group panel -7.339 *** -7.375 *** -7.369 *** Group panel t (non para) -6.370 *** -6.224 *** -6.283 *** Group panel t (para) -5.207 *** -5.992 *** -6.062 *** *** Rejecton of the null hypothess of no contegraton et percent. 3.3 Panel Estmaton Results As mentoned above, the strategy used to estmate a long run relatonshp between fnancal development and economc growth s based on two estmators wh error correcton: dynamc OLS (DOLS) and Pool Mean Group (PMG). The DOLS model used s: y F F X X u, (2) j3, t j j3, t j, t j j 8
where X Gov Open s a vector of exogenous varables comprsng the Government expendure n percent of GDP and openness to trade. Table 3 presents the results from dynamc OLS (DOLS) estmates of the contegratng relatonshp. The estmated coeffcents of the three ndcators of fnancal development (LIQ, DEP, and PRIV) are all posve and statstcally sgnfcant for all countres. The results of DOLS estmates confrm the exstence of a long run relatonshp between fnancal development and economc growth. The same concluson can be draw for varous specfcatons (low ncome countres, mddle ncome countres and hgh ncome countres) of our sample. However, the fnance-growth coeffcent can slghtly vary accordng to the sample: for nstance, the estmated coeffcent of the varable LIQ s 0.08 n low ncome countres, 0.398 for mddle ncome countres, versus 0.647 n hgh ncome countres. Ths can suggest a more ntense long-run relatonshp between fnance and growth n hgh ncome countres than n low ncome countres. In addon, the coeffcent of Government expendure and openness to trade exhb a statstcally sgnfcant posve coeffcent for all group of countres. Ths s consstent wh the fndngs of Apergs et al. (2007) and Dufrénot et al. (2007). Table 3: DOLS Estmates LIQ DEP PRIV All countres Fnance 0.348 (0.06) *** 0.295 (0.00) *** 0.220 (0.00) *** Gov 0.79 (0.02) *** 0.27 (0.020) *** 0.209 (0.02) *** Open 0.327 (0.020) *** 0.290 (0.09) *** 0.327 (0.09) *** F value 3.57 *** 66.08 *** 33.3 *** R2 0.352 0.407 0.355 Nbre dobs. 2982 2982 2982 Low ncome countres Fnance 0.08 (0.023) *** 0.07 (0.04) *** 0.5 (0.03) *** Gov 0.35 (0.025) *** 0.4 (0.024) *** 0.22 (0.023) *** Open 0.2 (0.027) *** 0.09 (0.026) *** 0.04 (0.026) *** F value 3.80 *** 6.72 *** 8.80 *** R2 0.85 0.26 0.237 Obs. 756 756 756 Mddle ncome countres Fnance 0.398 (0.023) *** 0.353 (0.07) *** 0.99 (0.07) *** Gov 0.46 (0.035) *** 0.02 (0.034) *** 0.23 (0.036) *** Open 0.369 (0.032) *** 0.335 (0.03) *** 0.429 (0.034) *** F value 65.97 *** 76.77 *** 47.45 *** R2 0.393 0.430 0.38 Obs. 260 260 260 Hgh ncome countres Fnance 0.647 (0.032) *** 0.397 (0.08) *** 0.357 (0.07) *** Gov 0.596 (0.046) *** 0.455 (0.046) *** 0.543 (0.045) *** Open 0.37 (0.033) *** 0.33 (0.033) *** 0.30 (0.033) *** F value 8.07 *** 33.06 *** 29.56 *** R2 0.6035 0.637 0.6255 Obs. 966 966 966 Standard error n parentheses; *** sgnfcant at percent. 9
After establshng that economc growth has long-run relatonshp wh fnancal development, we need to examne the causaly nexus between these two varables. The specfcaton we use to test for the causaly between fnance and growth s the followng: y F X u. (3) 0 2 where y s GDP per capa; F s a measure of fnancal development; X s a set of control varables, and u s the error term. The ARDL (,,) equaton assocated to equaton (3) s: y F F X X y. F and F 0 0 y and, t 2 2 y, t, t 3 X 3 4 X The error correcton equatons yeld: y y F X F 4, t, t F, t, t. F y X y X., t 0 0 2 2 2 2 X 4 4, The results of the error-correcton equatons are reported n table 4. We could notce that the non-causaly hypothess s rejected n all groups of countres. The error-correcton coeffcents s are negatve and statstcally sgnfcant mplyng that fnancal development does cause growth and the reverse relatonshp s true. Our results are n lne wh Apergs et al. (2007), Luntel and Khan (999), Demetrades and Hussen (996) whch show that the causal relatonshp between fnancal development and growth s b-drectonal. However, our results brng further evdence. The magnude of causal relatonshps depends upon the ncome level: the causal relatonshp runnng from fnance to growth s domnant n low ncome countres than n mddle and hgh ncome countres. When we consder the causal relatonshp from growth to fnance, the panel causaly results provde clear evdence n favour of mddle ncome countres. Table 4: Panel causaly tests LIQ DEP PRIV All countres DFY -0.04 (0.002) *** -0.048 (0.007) *** -0.049 (0.007) *** YDF -0.090 (0.0) *** -0.092 (0.00) *** -0.087 (0.009) *** Low ncome countres DFY -0.2 (0.030) *** -0.06 (0.028) *** -0.03 (0.030) *** YDF -0.0 (0.027) *** -0.095 (0.022) *** -0.080 (0.05) *** Mddle ncome countres DFY -0.030 (0.005) *** -0.035 (0.005) *** -0.053 (0.005) *** YDF -0.04 (0.04) *** -0.32 (0.020) *** -0.22 (0.07) *** Hgh ncome countres DFY -0.02 (0.003) *** -0.06 (0.002) *** -0.07 (0.002) *** YDF -0.089 (0.022) *** -0.038 (0.009) *** -0.036 (0.03) *** Standard error n parentheses; *** sgnfcant at percent. (4) (5) 0
4. Concluson and polcy mplcatons In ths paper we nvestgate the causal and contegraton relatonshps between fnancal development and economc growth n 7 countres, both developed and developng countres over the perod 960-2004. We have made use of panel un root tests, and panel contegraton analyss to conclude that there s strong evdence n favour of a long run relatonshp between fnancal development and economc growth for all groups of countres. Our results show that economc growth, fnancal development and the auxlary varables government expendure and openness are contegrated. Furthermore, our emprcal evdence ponts to a strong b-drectonal causaly between fnancal development and economc growth. However, the magnude of causal relatonshps depends upon the ncome level: the causal relatonshp runnng from fnance to growth seems to be stronger n low ncome countres than n mddle and hgh ncome countres. Ths suggests that fnancal reforms can have favourable mpact on economc growth n low ncome countres. These results have mportant polcy mplcatons. Indeed, polces amng at mprovng economc growth mght promote fnancal development and the fnancal markets polces must take account the no fnancal aspects of economc growth. So, the causaly relatonshp between fnancal development and economc growth could be seen through logc of recprocy. References Apergs N., Flppds I., and Economdou C., 2007. Fnancal deepenng and economc growth lnkages: a panel data analyss. Revew of World Economcs, 43, 79-98. Arests, P., Demetrades, P. and Luntel, K., 200. Fnancal development and economc growth: the role of stock markets. Journal of Money, Cred, and Bankng, 33, 6-4. Arests, P. and Demetrades, P., 997. Fnancal development and economc growth: assessng the evdence. Economc Journal, 07, 783-799. Banerjee, A., 999. Panel data un roots and contegraton: an overvew. Oxford Bulletn of Economcs and Statstcs, 6, pp. 607 629. Beck, T., Levne, R. and Loyaza N., 2000. Fnance and the sources of growth. Journal of Fnancal Economcs, 58, 26-300. Breung, J., 2000. The Local Power of Some Un Root Tests for Panel Data. In B. Baltag (ed.), Nonstatonary Panels, Panel Contegraton, and Dynamc Panels, Advances n Econometrcs, 5, JAI, Amsterdam, 6-78. Chrstopoulos, D. and Tsonas, E., 2004. Fnancal development and economc growth: evdence from panel un root and contegraton tests. Journal of Development Economcs, Vol. 73 (), 55-74. Demetrades, P. and Hussen, K., 996. Does fnancal development cause economc growth? Tme seres evdence from 6 countres. Journal of Development Economcs, 5, 387-4. Beck, T., Demrguç-Kunt, A. and Levne, R. 999. A New Database on Fnancal Development and Structure. World Bank Polcy Research Workng Paper N 246. Avalable at SSRN: http://ssrn.com/abstract=65009. Dufrenot, G., Mgnon, V. and Pegun-Fessolle, A., 2007. Testng the fnance-growth Lnk: Is there a dfference between developed and developng countres? CEPII, Workng Paper N 2007-24. Evans, P., 995. How to estmate growth equatons consstently. Paper presented at the 7th world congress of the Econometrc Socety, Tokyo, 995.
Fler, R., Hanousek, J. and Campos, N., 2003. Does stock markets promote economc growth?, CERGE-EI Workng Paper N 5, The Center for Economc Research and Graduate Educaton Economc Instute, Prague. Goldsmh, R., 969. Fnancal structure and development. New Haven, CT, Yale Unversy Press. Greenwood, J. and Smh, B., 997. Fnancal markets n development and the development of fnancal market. Journal of Economc Dynamc and Control, 2, 8-456. Hurln, C and Venet, B., 2004. Fnancal development and growth: A re-examnaton usng a panel Granger causaly test. Document de Recherche LEO, N 2004-8, Unversé dorléans. Hurln, C. and Mgnon, V., 2005. Une synthèse des tests de racne unare sur données de panel. Econome et Prévson, 3-4, (69), 253-294. Im, K., H. Pesaran, and Shn Y., 2003. Testng for un roots n heterogeneous panels. Journal of Econometrcs, 5, 53-74. Kao, C. and Chang, M. H., 2000. On the estmaton and nference of a contegrated regresson n panel data. In Baltag B. H. (ed.), Advances n Econometrcs: Nonstatonary Panels, Panel Contegraton and Dynamc Panels, 5, 79-222. Kng, R. and Levne, R., 993a. Fnance and growth: Schumpeter mght be rght. Quarterly Journal of Economcs, 08 (3), 77-737. Kng, R. et Levne, R., 993b. Fnance, entrepreneurshp and growth. Journal of Monetary Economcs, 32 (3), 53-542. Levne, R., Loyaza, N. and Beck, T., 2000. Fnancal ntermedaton and growth: causaly and causes. Journal of Monetary Economcs, 46, 3-7. Luntel, B. K. and Khan, M., 999. A quantatve re-assessment of the fnance-growth nexus: evdence from a multvarate VAR. Journal of Development Economcs, 60, 38-405. Maddala, G. and Wu, S., 999. A comparatve study of un root tests wh panel data and a new smple test. Oxford Bulletn of Economcs and Statstcs, 6, 63-652. Patrck, H. T., 966. Fnancal development and economc growth n underdeveloped countres. Economc Development and Cultural Change, 4, 74-89. Pedron, P., 999. Crcal values for contegraton tests n heterogeneous panels wh multple regressors. Oxford Bulletn of Economcs and Statstcs, 6, 653-670. Quah, D., 993. Emprcal cross-secton dynamcs n economc growth. European Economc Revew, 37, 426-434. Rousseau, P. and Wachtel, P., 2000. Equy markets and growth: cross country evdence on tmng and outcomes, 980-995. Journal of Bankng and Fnance, 24, 933-957. Schumpeter, J. A. 99. The theory of economc development, Cambrdge, Mass.: Harvard Unversy Press. Shaw, E. S., 973. Fnancal deepenng n economc development. New York: Oxford Unversy Press. Stock, J., and Watson, M. 993. A smple estmator of contegratng vectors n hgher order ntegrated systems. Econometrca, 6, (4), 783-820. 2