EERI Economics and Economerics Research Insiue Wha is he Long Run Growh Rae of he Eas Asian Tigers B. Bhaskara Rao, Arur Tamazian and Rup Singh EERI Research Paper Series No 04/010 ISSN: 031-489 Copyrigh 010 by B. Bhaskara Rao, Arur Tamazian and Rup Singh EERI Economics and Economerics Research Insiue Avenue de Beaulieu 1160 Brussels Belgium Tel: +3 99 353 Fax: +3 99 353 www.eeri.eu
Wha is he Long Run Growh Rae of he Eas Asian Tigers? B. Bhaskara Rao School of Economics and Finance Universiy of Wesern Sydney, Ausralia raob13@bigpond.com Arur Tamazian Deparmen of Finance & Accouning Universiy of Saniago de Composela, Spain arur.amazian@usc.es Rup Singh School of Economics Universiy of he Souh Pacific, Fiji singh_r@usp.ac.fj Absrac New panel daa esimaes for he four Eas Asian Tigers show ha he conribuion of oal facor produciviy (TFP) o growh is much higher han pas esimaes. An exended producion funcion wih learning by doing implies ha TFP is abou 3.5% and hese counries will grow a his rae in he long run. JEL: O1, N1 Keywords: Asian Tigers, Sysems Dynamic GMM, Growh Accouning, Facor Accumulaion as Residual.
1. Inroducion This paper re-examines he decade old conroversy on he Eas Asian miracle. To conserve space we limi o Hong Kong, Korea, Singapore and Taiwan i.e., he Eas Asian Tigers (EATs). We refer o excellen surveys by Seral (000) and Jong-il Kim (00) for bibliography. When his conroversy sared in early 1990s, he average GDP growh rae (1965-1990) of EATs was 8.8% compared o UK s 3.7% and USA s 4.0%; see Table 1 in Jong-il Kim (00). The conroversy was on wheher EATs can coninue o grow a his high rae in he long run, which we may say due o he confusion beween he acual high growh rae and he long run or he seady sae growh rae (SSGR). Since heoreical growh models imply ha he SSGR equals growh of oal facor produciviy (TFP), a variey of growh accouning exercises was conduced o esimae TFP as he Solow residual. The general conclusion was much of he growh of EATs was due o facor accumulaion (85%) and no due o TFP. Therefore, heir growh raes will apper off evenually o equal TFP, which was esimaed beween 0% o 1.5%. A negleced issue in his conroversy is he effec of learning by doing (LBD) on TFP. Jong-il Kim (00) and Greiner, Semler and Gong (005) recognise is imporance by noing ha developing counries go hrough 3 hisorical sages. A firs LBD, simulaed by capial accumulaion and openness o rade, play an imporan role in improving TFP. In he second and hird sages human capial formaion and expendiure on R&D are imporan conribuors o TFP. In his paper we exend he sandard Cobb-Douglas producion funcion o ake ino accoun some LBD effecs due o capial formaion and rade openness (OT ). We also use an improved sysems GMM dynamic panel daa mehod of Arellano and Bover (1995) and Blundell and Bond (1998). Some of is advanages are: (1) i overcomes he weak insrumens problem generally causing finie sample biases and () minimises ime series problems, due o he persisence in he variables, because he model is esimaed as a sysem in boh he levels of he variables and heir firs differences.
This paper is organised as follows. Specificaion issues are in Secion. Empirical resuls are in Secion 3 and Secion 4 concludes.. Specificaion 1 Some exernaliies are manna from he heaven ype and do no need addiional invesmens by firms. In he early sages of developmen wo such imporan exernaliies are due o LBD of Arrow (196) and OT. These LBD effecs may be due o a rapid capial accumulaion hrough new invesmen in improved machines and exposure o inernaional compeiion simulaes efficien mehods of producion. I is hard o say a priori he relaive imporance of hese wo LBD effecs. The Cobb-Douglas producion funcion wih consan reurns can be exended o include such effecs as follows. Le a represenaive firm s producion funcion, where LBD effecs are hrough he economy wide capial sock, be: Y K A L (1 ) (1) i i i i i A B K i i () where Y is oupu, K is capial, L is employmen and i is an error erm such ha ln i N(0, ). facors. Therefore, B sands for he sock of knowledge which depends on auonomous ln B is he rae of growh of auonomous TFP. B may be consan g ln B 0 or grow a a consan auonomous rae of g ( B = Be 0 ) where B0 is he iniial sock of knowledge. ln B hus capures he effecs of rended variables affecing TFP. Subsiuions and aggregaion give he aggregae producion funcion: Y B0 e K L (3) 1 g(1 ) (1 ) (1 ) 1 To conserve space we shall be brief. See Rao (007) for deails.
n where 1/ and 1 i n ln N(0, ). Alernaive assumpions abou A in () are possible. If OT has an exernaliy hrough LBD effecs, which is imporan for he Eas Asian counries, A may be specified wih he wo LBD effecs as: i g1 got 0 (4) A B e K The aggregae producion funcion will be 1 (1 )( g1got) ( (1 )) (1 ) Y B0 e K L (5) This is he same as (3) excep ha g is now compued as (g 1 +g OT). The log-linear form of (5) where oupu and capial are measured in per worker erms is: ln y (1 )ln B (1 )( g g OT) 0 1 ( (1 ))ln k (1 ) L ln (6) where y ( Y / L) and k ( K / L). There is a seady sae soluion only when 1. If 1 here is no seady sae because here are no diminishing reurns o k and k does no become zero, which is he definiion of he seady sae. Therefore, in he following derivaion of SSGR i is assumed ha 1, and SSGR implied by (6) is: SSGR g1g OT n (7) (1 ) where n is he rae of growh of labour force. The SSGR implied by (3) is a special case of (7) where g 0. Noe ha in (7) SSGR increases wih respec o g and if 1.
3. Empirical Resuls The dynamic sysems GMM mehod in his paper essenially requires ha (6) should be esimaed in he levels and firs differences of he variables forms as a sysem. Esimaes of (6) for he period 197-006 are in Table-1. Table 1 GMM ESTIMATES OF PRODUCTION FUNCTION 197-006 Parameers (1) () (3) Consrains 0 0 0.3 Inercep 3.90E -03 0.38E -0 0.35E -0 0.30 [0.01]* 0.46 [0.01]* - g 0.034 0.03 0.031 1 [0.00] g 5.50E -04 0.77E -03 0.59E -03 [0.4] [0.05]* [0.1] 3.16E -03 [0.3] 1 (Level) 0.99 R Level Difference DW Levels Difference 0.996 0.094 1.684 [0.74] 1.683 [0.74] 6.418 IV [0.19] 0.99 0.996 0.109 1.683 [0.74] 1.68 [0.74] 6.37 [0.4] 0.99 0.996 0.09 1.674 [0.7] 1.674 [0.7] 7.767 [0.3] Noes: p-values are in square brackes. * = 5% and ** = 10% significance. IV is es for over-idenifying resricions of insrumens. See Daa Appendix for furher deails. The random effecs model (REM) is used in such dynamic panel models. However, we have conduced he Hausman es where he null is REM agains he alernaive of he fixed effecs model. The null could no be rejeced 3 a he 5% level. The compued es saisic, wih p-value in he square brackes is: 16.996 0.7 E. 3
In column 1 esimae of (6), wih a correcion for firs order serial correlaion in he levels equaion, is given. This serial correlaion coefficien is close o uniy highlighing he nonseady naure of he level variables. I is se a 0.99 o achieve convergence. 3 Esimaes in column (1) are good alhough he wo LBD coefficiens and g are insignifican. Esimae of (share of profis) a 0.3, highly significan, implying ha perhaps he conribuion of TFP o growh has been underesimaed in he previous growh accouning exercises. Esimae of g1implies ha auonomous TFP is 3.4%. To increase he degrees of freedom and see if one and/or boh LBD coefficiens become significan, we reesimaed he equaion in column (1) wih 3 alernaive resricions: (a) 0; (b) g 0; and (c) 0.5. The firs of hese resricions gave marginally beer resuls and is repored in column (). Esimae of and g have slighly increased. There is a marginal improvemen in R of he equaion in he differences. In column (3) esimaes of () wih he consrain ha is equal o is sylised value of one hird in many growh accouning exercises is repored. This has marginally decreased R of he equaion in he differences and g has now become insignifican even a he 10% level. The summary saisics of hese 3 esimaes are close and i is hard o selec he bes equaion. Esimaes of g1 remained virually sable in all he 3 equaions. Our subjecive preference is for he esimae in column () because is R s are marginally beer. Using () and he sample means for n (0.04) and OT (1.47), he esimae of SSGR is 3.35%. This has slighly increased o 3.43% when he averages for he recen 5 year period (00-006), wih n (0.014) and OT (.55), are used. Boh esimaes imply ha he SSGR of EATs is well over 3% and close o 3.5%. Using he sample average growh of oupu of 3 I may be hough ha here is no need for he levels equaion and he sysems esimaion. However, when only he equaion in firs differences is esimaed, a sandard pracice in many panel daa esimaes of growh equaions, none of he parameers were significan. This may be due o some limiaions noed in he ex.
6.7%, he conribuion of TFP is abou 50%. This implies ha facor accumulaion and TFP equally conribued o growh of he EATs. 4. Conclusions In his paper we have used a more efficien sysems GMM echnique o esimae he producion funcion wih LBD effecs. Our esimaes showed ha he share of profis is abou 0.5, which is less han he values used for ESTs in many pas growh accouning exercises. Nex, insead of esimaing TFP as he Solow residual, i can direcly be esimaed wih our exended producion funcion. Consequenly, i is possible o esimae he conribuion of facor accumulaion as a residual. This laer approach seems more convincing o us because TFP is esimaed direcly. Boh approaches show ha he conribuion of TFP o growh of EATs is higher han pas esimaes and inspiraion and precipiaion seem o have played equal roles in he growh process of hese counries. Needless o say here are limiaions in our paper. The wo LBD coefficiens in column (1) were no significan and heir significance is improved by imposing plausible consrains. Perhaps a larger sample may give beer esimaes. Our approach can be exended by idenifying addiional conribuors o TFP, besides capial accumulaion and OT. Alernaive measures of LBD due o capial accumulaion are also possible. However, hese are ouside he scope of his shor paper.
Daa Appendix Y is real oupu measured in 1990 US dollars. Daa are obained from he World Bank daabase: hp://unsas.un.org. K is real sock of capial compued using daa on gross fixed capial formaion wih he perpeual mehod. Depreciaion rae is 4% and he iniial sock of capial is 1.5 imes real oupu in 1969. These are close o he esimaes of Bosworh and Collins (003). L is labour force (or employmen, whichever is available) obained from he World Developmen Indicaors (007) and updaed using he Asian Developmen Bank (ABD) daabase: hp://www.adb.org/documens/books/key_indicaors/008. Where daa is unavailable, working age populaion (15-64 years, inclusive) are aken from World Bank Indicaors (007). OT is openness of rade compued as he raio expors plus impors o aggregae oupu. All series are in curren US dollars derived from he World Bank daabase.
References Arellano, M. and S. Bond. (1998). Dynamic panel daa esimaion using DPD98 for GAUSS: a guide for users. Mimeo, Insiue for Fiscal Sudies, London. Arellano, M. and O. Bover. (1995). Anoher look a he insrumenal variable esimaion of error-componens models. Journal of Economerics, 68(1): 9-5. Arrow, K. J., (196) ``The economic implicaions of learning by doing,'' Review of Economic Sudies, 9(3): 155-173. Bosworh, B. and Collins, S. M. (003). The Empirics of Growh: An Updae, hp://www.brookings.edu/~/media/files/rc/papers/003/09globaleconomics_bosworh/ 0030307.pdf Greiner, A., Semler, W., and Gong, G. (004) The Forces of Economic Growh: A Time Series Perspecive, Princeon, NJ: Princeon Universiy Press. Jong-il Kim (00). Toal Facor Produciviy Growh in Eas Asia: Implicaions for he Fuure, Asian Economic Papers, 1(): 50-70. Rao, B. B. (007). Esimaes of he Seady Sae Growh Raes for Seleced Asian Counries wih an Endogenous Growh Framework, Presened a he 36 h Conference of Ausralian Economiss, Hobar, Ausralia. hp://econpapers.repec.org/paper/pramprapa/389.hm Sarel, M., (1995). Growh in Eas Asia: Wha we can and wha we canno infer from i, in Produciviy and Growh, Proceedings of a Conference, Andersen, P., Dwyer, J. and Gruen, D. (eds.), Sydney: Reserve Bank of Ausralia, pp. 37-59.