How Has Specialization in Turkish Exports Evolved Over Time? A Study Based on Galtonian Regressions

Transcription

1 Loyola Unversty Chcago Loyola ecommons Topcs n Mddle Eastern and North Afrcan Economes Qunlan School of Busness How Has Specalzaton n Turksh Exports Evolved Over Tme? A Study Based on Galtonan Regressons Guzn Erlat Mddle East Techncal Unversty Haluk Erlat Mddle East Techncal Unversty Recommended Ctaton Erlat, Guzn and Erlat, Haluk, "How Has Specalzaton n Turksh Exports Evolved Over Tme? A Study Based on Galtonan Regressons". Topcs n Mddle Eastern and North Afrcan Economes, electronc journal, 10, Mddle East Economc Assocaton and Loyola Unversty Chcago, 008, Ths Artcle s brought to you for free and open access by the Qunlan School of Busness at Loyola ecommons. It has been accepted for ncluson n Topcs n Mddle Eastern and North Afrcan Economes by an authorzed admnstrator of Loyola ecommons. For more nformaton, please contact ecommons@luc.edu. 008 the authors

2 HOW HAS SPECIALIZATION IN TURKISH EXPORTS EVOLVED OVER TIME? A STUDY BASED ON GALTONIAN REGRESSIONS GUZIN ERLAT AND HALUK ERLAT Department of Economcs Mddle East Techncal Unversty Ankara, Turkey Fax: Emal: gerlat@metu.edu.tr Emal: herlat@metu.edu.tr Prepared for presentaton at the 8 th Annual Meetng of the Mddle East Economc Assocaton, January 3-6, 008, New Orleans, USA.

3 1. Introducton In two earler papers we nvestgated the comparatve advantage of Turksh exports vs-à-vs the European Unon for the perod (Erlat and Erlat, 005) and the pattern of Turksh trade for the perod (Erlat and Erlat, 006). The frst paper was a descrptve study that focused on dentfyng 3-dgt SITC classfed sectors that showed ncrease n comparatve advantage over tme, usng the Balassa (1995) ndex of revealed comparatve advantage (RCA). The second paper, on the other hand, consdered both exports and mports and tred to establsh f the pattern of Turksh trade, for the longer perod of , remaned essentally the same (.e., f the pattern was persstent) or f t showed apprecable change (.e., f the pattern was dynamc), usng a contngency-table approach due, ntally, to Gagnon and Rose (1995). The present paper attempts to combne the notons underlyng these two papers to nvestgate how persstent the specalzaton n Turksh exports s over tme. We use two measures for specalzaton; one s the RCA ndex and the other s the Lafay ndex (Lafay, 199). A study smlar to ours, usng Markov Chan methods and the RCA ndex, was done recently by Flztekn (006). Ours dffers from ths paper n usng () n addton to the RCA ndex, the Lafay ndex that takes mports also nto account n measurng specalzaton and s argued, by Casell and Zaghn (005), to be superor to the RCA ndex snce t allows to control for ntra-ndustry trade and for dstortons nduced by macroeconomc fluctuatons, and () n consderng the mean and the varance of the dstrbutons of the measures over tme nstead of the whole dstrbuton. Ths s done by mplementng Galtonan regressons (see, e.g., Hart and Pras, 1956) that nvolve regressng the cross-secton of ndex values at perod on the cross-secton at the begnnng perod, t. The sgn of the slope coeffcent and t 1 ts relaton wth the correlaton coeffcent provde us wth the nformaton as to how the pattern and degree of specalzaton of Turksh exports evolve. Hence, n the next secton we shall descrbe our two measures of specalzaton and n the followng secton explan how Galtonan regressons work. We descrbe our data and the technologcal classfcaton of the sectors that we use n ths paper n Secton 4. Secton 5 has our emprcal results and our conclusons are n Secton 6.. Measures of Comparatve Advantage 1

4 We shall be usng two measures of comparatve advantage. The frst one s due to Balassa (1965) and s entrely based on the performance of exports. It s named the ndex of Revealed Comparatve Advantage (RCA) and s obtaned as the rato of the share of exports of a partcular good n the total exports of a gven country to that of a group of countres that also ncludes the country n queston. Hence, f we denote Turkey s exports of good by X and the exports of good for the group of countres by X w, then the RCA ndex may be wrtten as, N X / = 1 X (1) RCA =, = 1, K, N N X / X w = 1 w When ths ndex s greater than unty, t wll ndcate that Turkey has comparatve advantage n the exports of good. As we wll be usng the RCA ndex values n regresson analyss, t would be desrable f ther dstrbuton, at a gven t, be normal. To ensure that ths dstrbuton s, at least, symmetrc, we shall transform RCA as RCA 1 () RSCA =, = 1, K, N RCA + 1 where RSCA stands for the Revealed Symmetrc Comparatve Advantage of good. Ths measure takes on values between -1 and 1. However, smply usng exports to measure comparatve advantage, n vew of the ncreasng mportance of ntra-ndustry trade, not only n developed countres but also n developng countres lke Turkey (see Erlat and Erlat, 003) may lead to erroneous conclusons. Ths, of course, depends on the aggregaton level used. At the 3-dgt level, ntrandustry trade may very well ndcate the mports of ntermedate goods used n the manufacture of fnal goods n that category. Hence, n addton to the RCA ndex, we also calculated a measure, due to Lafay (199) as slghtly modfed by Corell and Zaghn (005), where mports, M, are also taken nto account: N X M = 1( X M ) X + M (3) LFI =, = 1, K, N N N X M 1( X M ) + = + = 1( X + M )

5 In other words, LFI measures the comparatve advantage of Turkey n good by comparng the (normalzed) trade balance ( balance ( ( X M N = 1 ) X M ) of the th good to the overall (normalzed) trade ) and weghtng t by the share of trade n the th good n total trade. Hence, LFI does not measure specalzaton wth respect to other countres, as RSCA does, but wth respect to the total structure of trade of a gven country. Hence, postve values of LFI ndcate comparatve advantage n the sense that they pont to the level of specalzaton n the good n queston. Negatve values wll, of course, ndcate despecalzaton. Also note that N LFI = 0. A further advantage of LFI s that, by consderng the dfference between X M and N = 1 X M, t controls for the effects of cyclcal factors on trade flows n the short run, mplctly assumng, of course, that... cyclcal factors nfluence aggregate and dsaggregate trade flows n the same way. (Casell and Zaghn, 005: 11). 3. Galtonan Regresson t1 Suppose we have N cross-secton observatons on a varable, Y, n two ponts n tme, and t. Further assume that and have a bvarate normal dstrbuton. Then, the Y t 1 Y t regresson of y = Y Y on, t, t, t y = Y Y s lnear:, t1, t1, t1 (4) y, t = β y, t + u t, = 1,..., N 1, where u, s N (0, ) and s ndependent of y. Now, f β = 1, then the dstrbuton of the t σ,t1 Y t 1 at wll have remaned the same at t. If β > 1, then the nequalty n the dstrbuton of the observed at would have ncreased n t. If 0 < β < 1, then the nequalty observed Y t 1 at would dmnsh n t n the sense that there would be a movement towards the center of t1 the dstrbuton. Fnally, f dstrbuton of the n has been reversed n t. β < 0, then ths would ndcate that the structure of the unequal Y t 1 t y, t t y 1, t 1 Lettng σ = E ( ) and σ = E ( ) and notng that E y u ) 0, we obtan (, t1, t = (5) t = β σ σ t1 σ + 3

6 Followng Hart and Pras (1976), we fnd the correlaton between the and, to be σ t = t ρ ρ = 1 ( σ / ) from whch we obtan σ σ (1 ). Substtutng n (5) yelds, y,t1 y,t (6) t t1 σ β = σ ρ In other words, the change n the dsperson of the Y between the two perods depends on the relatonshp between β and ρ. If β > ρ, then the dsperson, and, of course, nequalty wll ncrease. Ths, of course, may take place even when β les between 0 and 1, so that even when there s a movement towards the centre of the dstrbuton, there may be an ncrease n dsperson. Hence β > 1 s not a necessary condton for the dsperson to ncrease. Now, f Y refers to the comparatve advantage measures descrbed above, then, followng Laursen (000), we may obtan the followng characterstzatons of specalzaton: t t 1. If, β = 1, the specalzaton pattern has remaned the same n as t was n. If β > 1, then the specalzaton pattern n s strengthened n t. Ths may be t1 called β-specalzaton. In other words, those sectors that had exhbted hgh levels of comparatve advantage at wll show even hgher levels of comparatve advantage at t, t1 whle those sectors wth low or no comparatve advantage wll become more so.. If 0 < β < 1, then hgh comparatve advantage ndex values wll be reduced whle low ndex values wll ncrease. Ths may be called β-de-specalzaton. v. If β < 0, then the specalzaton pattern n would be reversed n t. v. If β = ρ, then the dsperson n wll be th same as n t. t 1 t 1 v. If β > ρ, the dsperson would ncrease, ndcatng that the degree of specalzaton has ncreased. Ths may be named σ-specalzaton. v. If β < ρ, the dsperson would decrease and the degree of specalzaton would decrease. Hence, we would have σ-de-specalzaton. 4. The Data and The Classfcaton The data on Turksh exports and mports cover the perod and have been obtaned from the Turksh Statstcal Insttute. The data on the exports by Turkey s tradng 4

7 partners were obtaned from the UNCTAD-ITC and OECD databases. These seres cover the perod. Hence, the RSCA computatons only cover ths perod. Turkey s tradng partners consdered n ths paper are the pre-expanson ffteen 1 European Unon countres, namely, Austra, Belgum, Denmark, Fnland, France, Germany, Greece, Ireland, Italy, Netherlands, Portugal, Span, Sweden and the UK. All data are n $US terms and nclude 56, 3-dgt sectors classfed accordng to SITC Rev.3. The technologcal classfcaton of the sectors, whch we have also used n prevous work (e.g., Erlat and Erlat, 005, 006) classfes SITC 0, (ex. 6), 3 (ex.35), 4, 56 as Raw materal-ntensve goods (RMIG) SITC 6, 6 (ex. 6, 67, 68), 8 (ex. 87, 88) as Labour ntensve goods (LIG) SITC 1, 35, 53, 55, 6, 67, 68, 78 as Captal-ntensve goods (CIG) SITC 51, 5, 54, 58, 59, 75, 76 as Easy-to-mtate research-ntensve goods (EIRG). SITC 57, 7(ex.75,76,78),87,88 as Dffcult-to-mtate research-ntensve goods (DIRG). The detals of ths classfcaton are gven n the Appendx. 5. Emprcal Results In most applcatons of Galtonan regressons to the comparatve advantage ndexes dscussed above, the begnnng and endng perods, and t, are chosen to contan several t1 perods n between and, even when sub perods are consdered, ths fact does not change (see, e.g., Dalum, Laursen and Vllumsen, 1998 and Laursen, 000). What we shall do n ths study s to take the frst perod for whch data are avalable as and perform regressons for t +1, t + 1 1, t, to obtan a sequence of estmates for β and ρ, and ther ratos. We shall be able, thereby, to follow the pattern of export specalzaton over tme wth respect to t 1. t 1 In our case we have two choces for t 1, 1969 and Snce the LFI ndex uses only the trade flows for Turkey, t can be calculated for the perod, whle the RSCA ndex requres data on the trade flows of the EU15 countres and that s only avalable for the perod. Hence, we were able to run regressons usng the LFI ndex for both 1969 and 1990 as the base years, whle we were able to do ths for RSCA wth only 1990 as the base year. 1 The actual number of countres consdered s fourteen snce the data for Belgum also contans the data for Luxemburg. 5

8 Fgure 1: Total Trade 6

9 We shall present our results n sx sets of dagrams that nclude the plots of βˆ, ρˆ and the absolute value of ther rato. These sets wll nclude the plots for Total Trade, RMIG, LIG, CIG, EIRG and DIRG, and wll contan three dagrams n each set; namely, , for LFI and for RSCA. The plots are based on the three tables gven n the Appendx. The results for total trade are gven n Fgure 1. For the perod, the obtaned for LFI declnes to about 0.70 by 1980, then shows a sharp decrease to about 0.17 by 1987 and contnues declnng to 0.07 by 005. The declne n the varance rato s qute βˆ smlar but ρˆ s above βˆ at all ponts. Thus, there appears to be both β - and σ - despecalzaton of total exports wth respect to When we look at the results for the perod we fnd a smlar ncrease n β and σ -despecalzaton wth respect to 1990 but at a mlder pace, whch s not surprsng gven the nearness of to t n ths case. t1 Ths pcture does not change when we look at the RSCA results for the same perod. Thus, we may conclude that, both n terms of the specalzaton of exports wth respect to Turkey s total trade wth the world, gven by the LFI results and n terms of exports to the 15 EU countres, gven the RSCA results, there s a declne both n the pattern of specalzaton ndcated by the values of βˆ, whch les between 0 and 1 throughout both perods, and n the degree of specalzaton gven by the standard devaton ratos. When we turn to the results for the technologcal groupngs of the three-dgt sectors, we fnd, for the Raw Materal Intensve Goods sector n Fgure, that durng the perod, βˆ, based on the LFI ndex, declnes untl 1973 and then ncreases to a value close to unty by 1979, after whch there s a sharp decrease untl 1981 to a value around 0.30 after whch a more stable pattern s observed untl the end of the perod. The pattern for the standard devaton rato s qute smlar except that βˆ exceeds ρˆ durng the perod. Thus, β and σ -despecalzaton are agan the domnant patterns but, for a sx-year sub perod, even though the pattern of specalzaton decreases, ts degree ncreases. For the perod we observe a mlder but less steady ncrease n β and σ - despecalzaton. Decrease n the pattern and the degree of specalzaton s also observed for the RSCA results. In the case of Labour Intensve Goods n Fgure 3, for the perod based on the LFI results, βˆ shows a steady declne untl 1986, at whch pont t turns negatve, 7

10 ndcatng a reversal n the specalzaton pattern of exports. Ths reversal becomes a steady pattern after The declne n the standard devaton rato also contnues untl 1986 after whch we observe a steady path of σ -despecalzaton. For the shorter perod, , the specalzaton pattern that had been reversed n 1990 wth respect to 1969, s strengthened Fgure : Raw-Materal Intensve Goods 8

11 untl 1996, snce βˆ takes on values greater than unty, but a mld level of β -despecalzaton s observed for the rest of the perod. Interestngly, we fnd that, snce β > σ throughout the perod, there s σ -specalzaton even when there s β -despecalzaton. When we use the Fgure 3: Labour Intensve Goods 9

12 RSCA results, the pcture we obtan s not the same for the perod. There s a very mld ncrease n both β and σ -despecalzaton, the latter despecalzaton showng a Fgure 4: Captal Intensve Goods 10

13 decrease after Of course, whether ths reflects a pcture of the specalzaton pattern wth respect to 1969 that may have been reversed n 1990 cannot be determned because of lack of data. Fgure 5: Easy-to-Imtate Research Intesve Goods 11

14 For Captal Intensve Goods n Fgure 4, based on the LFI results and the perod, βˆ shows a decrease that becomes much more pronounced and steady after 1981, and reaches a value qute close to zero n 1995, agan takng on such low values from 001 Fgure 6: Dffcult-to-Imtate Research Intensve Goods 1

15 onwards. Snce ρˆ s never below βˆ, there sσ -despecalzaton throughout the perod. For , there s a steady declne n βˆ after 1993 that stablzes after 001. These pctures are not much dfferent from what we observe for the same perod n terms of the 1969-based results; the magntudes of βˆ and the standard error ratos are lower for the larger perod. The RSCA-based results are smlar but the declne n specalzaton s mlder. Turnng to Easy-to-Imtate Research Intensve Goods n Fgure 5, for the perod, the pattern of βˆ s a bt erratc, even turnng negatve for the subperod, and then contnung on takng values between 0.5 and 0. Thus, β -despecalzaton wth short perods of specalzaton reversal, s the pattern. For the standard devaton ratos, ther pattern becomes erratc after 1980, exceedng unty n 1986 and agan n 1987, remanng at such a level untl 1998, after whch an erratc pattern of σ -specalzaton s observed and appears to be stablzed at a very hgh level ( 3.5 ) for For the perod, we observe β -despecalzaton untl 1997 after whch we observe β -specalzaton that levels off at a pont above 1.5 for the perod. The pattern for the standard devaton ratos s very smlar; σ -specalzaton follows β -specalzaton. The RSCA results, on the other hand, show no evdence of ether type of specalzaton, wth βˆ rapdly declnng to a level below 0.40 n 1996, then jumpng to 0.50 n 1997 after whch t enters a steady declne that takes t to as low a level as 0.0. σ -despecalzaton, however, declnes, reachng a value of, approxmately, unty by 005. In other words, n the EIRG category, the specalzaton pattern of exports, n terms of Turkey s trade flows wth the world, s qute dfferent from the pattern one observes when measured n comparson wth the export performance of the EU15 as a group. Fnally, for Dffcult-to-Imtate Research Intensve Goods, durng the perod, βˆ takes on values close to zero untl 1979, after whch t starts takng on values grater than unty but shows an erratc pattern untl 1987, settlng nto a perod of reversal n specalzaton from that pont onwards. σ -despecalzaton s also at a very low level untl 1979 after whch we observe an erratc pattern of ncreasng hgh levels of σ -specalzaton, partcularly n reference to the sub perod of specalzaton reversal. β -specalzaton s observed for most of the shorter perod (the exceptons beng ) whle σ -specalzaton exsts for the whole perod. Agan, ths s not what s observed from the 13

16 RSCA results for the same perod; β -despecalzaton together wth σ -specalzaton s the pcture for the perod from 1993 onwards. The concluson reached for EIRG holds also for DIRG. 6. Conclusons 1. For total trade, both the LFI and the RSCA results show a declne n both the pattern and the degree of specalzaton. Ths holds for both Turkey s trade flows wth the world and n terms of the poston of ts exports vs-à-vs the 15 EU countres.. For raw-materal ntensve goods, β -despecalzaton takes place durng both and and, whle σ -despecalzaton s domnant, we observe σ -specalzaton durng Ths concluson holds for both the LFI and RSCA results. 3. For labour-ntensve goods, durng the perod and accordng to the LFI results, we fnd β -despecalzaton untl 1986 after whch the specalzaton pattern s strengthened. 4. For captal-ntensve goods all results ndcate both β and σ -despecalzaton. 5. In both categores of research-ntensve goods the LFI results ndcate σ - specalzaton after For DIRG, however, one notes a reversal n specalzaton patterns after 1990, whch appears to be strengthened after that year. The RSCA results are qute dfferent n both cases. References Balassa, B. (1965): Trade Lberalzaton and Revealed Comparatve Advantage, The Manchester School, 3, Casell, P. and A. Zaghn (005): Internatonal Specalzaton Models n Latn Amerca: The Case of Argentna, Tem d dscussone, No.558, Banca D Itala. Dalum, B., K. Laursen and G. Vllumsen (1998): Structural Change n OECD Export Specalzaton Patterns: De-specalzaton and Stckness, Internatonal Revew of Appled Economcs, 1(3), Erlat, G. and H. Erlat (003): Measurng Intra-Industry and Margnal Intra-Industry Trade: The Case for Turkey, Emergng Markets Fnance and Trade, 39(6),

17 Erlat, G. and H. Erlat (005): The Comparatve Advantage of Turksh Exports vs-à-vs the European Unon and OECD Countres, Paper presented at the 5 th Annual Conference of the MEEA, Phladelpha, USA. Erlat, G. and H. Erlat (006): The Pattern of Turksh Foregn Trade, Paper presented at the 6 th Annual Conference of the MEEA, Boston, USA. Flztekn, A. (006): Türkye de Açıklanmış Karşılaştırmalı Üstünlüklern Evrm (The Evoluton of Revealed Comparatve Advantage n Turkey), Uluslararası Ekonom ve Dış Tcaret Poltkası (Internatonal Economes and Foregn Trade Polcy), 1(1), Gagnon, J.E. and A.K. Rose (1995): Dynamc Persstence of Industry Trade Balances: How Pervasve s the Product Cycle?, Oxford Economc Papers, 47, 9-48 Hart, P.E. and S.J. Pras (1956): The Analyss of Busness Concentraton: A Statstcal Approach, Journal of the Royal Statstcal Socety, Seres A, 119, Laursen, K. (000): Do Export and Technologcal Specalzaton Patterns Co-evolve n Terms of Convergence or Dvergence? Evdence from 19 OECD Countres, , Journal of Evolutonary Economcs, 10, Lafay, G. (199): The Measurement of Revealed Comparatve Advantage n M.G. Dagenas and P.A. Muet (eds.): Internatonal Trade Modelng. London: Chapman and Hall.. 15

18 Appendx Raw Materal Intensve Goods SITC 0 Food and Lve Anmals SITC Crude Materal, Inedble, Except Fuels (excludng 6) SITC 3 Mneral Fuels, Lubrcants and Related Materals (excludng 35) SITC 4 Anmal and Vegetable Ols, Fats and Waxes SITC 56 Fertlzers (Other Than Those of Group 7) Labour-Intensve Goods SITC 6 Textle Fbres (Other Than Wool Tops and Other Combed Wool) and Ther Wastes (Not Manufactured Into Yarn or Fabrc) SITC 6 Manufactured Goods Classfed Chefly by Materal (excludng 6, 67, 68) SITC 8 Mscellaneous Manufactured Artcles (excludng 88, 87) Captal-Intensve Goods SITC 1 Beverages and Tobacco SITC 35 Electrc Current SITC 53 Dyeng, Tannng and Colourng Materals SITC 55 Essental Ols and Resnods and Perfume Materals; Tolet, Polshng and Cleansng Preparatons SITC 6 Rubber Manufactures, n.e.s. SITC 67 Iron and Steel SITC 68 Non-Ferrous Metals SITC 78 Road Vehcles (Includng Ar-Cushon Vehcles) Easy-to-Imtate Research-Intensve Goods SITC 51 Organc Chemcals SITC 5 Inorganc Chemcals SITC 54 Medcnal and Pharmaceutcal Products SITC 58 Plastcs n Non-Prmary Forms SITC 59 Chemcal Materals and Products, n.e.s. SITC 75 Offce Machnes and Automatc Data-Processng Machnes SITC 76 Telecommuncatons and Sound-Recordng and Reproducng Apparatus and Equpment Dffcult-to-Imtate Research-Intensve Goods 16

19 SITC 57 Plastcs n Prmary Forms SITC 7 Machnery and Transport Equpment (excludng 75, 76, 78) SITC 87 Professonal, Scentfc and Controllng Instruments and Apparatus, n.e.s. SITC 88 Photographc Apparatus, Equpment and Supples and Optcal Goods, n.e.s.; Watches and Clocks 17

20 Table 1: Regresson Results Based on LFI, Total Trade RMIG LIG CIG EIRG DIRG βˆ ρˆ ˆ β / ˆ ρ βˆ ρˆ ˆ β / ˆ ρ βˆ ρˆ ˆ β / ˆ ρ βˆ ρˆ ˆ β / ˆ ρ βˆ ρˆ ˆ β / ˆ ρ βˆ ρˆ ˆ β / ˆ ρ

21 Table 1: Regresson Results Based on LFI, (cont.) Total Trade RMIG LIG CIG EIRG DIRG βˆ ρˆ ˆ β / ˆ ρ βˆ ρˆ ˆ β / ˆ ρ βˆ ρˆ ˆ β / ˆ ρ βˆ ρˆ ˆ β / ˆ ρ βˆ ρˆ ˆ β / ˆ ρ βˆ ρˆ ˆ β / ˆ ρ

22 Table : Regresson Results Based on LFI, Total Trade RMIG LIG CIG EIRG DIRG βˆ ρˆ ˆ β / ˆ ρ βˆ ρˆ ˆ β / ˆ ρ βˆ ρˆ ˆ β / ˆ ρ βˆ ρˆ ˆ β / ˆ ρ βˆ ρˆ ˆ β / ˆ ρ βˆ ρˆ ˆ β / ˆ ρ

23 Table 3: Regresson Results Based on RSCA, Total Trade RMIG LIG CIG EIRG DIRG βˆ ρˆ ˆ β / ˆ ρ βˆ ρˆ ˆ β / ˆ ρ βˆ ρˆ ˆ β / ˆ ρ βˆ ρˆ ˆ β / ˆ ρ βˆ ρˆ ˆ β / ˆ ρ βˆ ρˆ ˆ β / ˆ ρ