Regional Specialization: Measurement & Application

Transcription

1 Schuan Unversty From the SelectedWorks of Zheng Lu (Chnese: 路征 ) September 5, 2012 Regonal Specalzaton: Measurement & Applcaton Zheng Lu Avalable at:

2 REGIONAL SPECIALIZATION: MEASUREMENT & APPLICATION Dr. ZHENG LU School of Economcs, Schuan Unversty

3 1.Introducton From 1990s, research on ndustral concentraton and regonal specalzaton developed very fast comng wth the development of new economc geography theory, agglomeraton economcs and spatal economcs. [See, Km (1995); Ellson and Glaeser (1997); Holmes and Stevens (2004); Amt (1998); Brülhart (2001); etc.] Many economsts have begun to study the problem of ndustral geographc dstrbuton and regonal specalzaton n Chna. Young (2000) and Poncet (2003) revealed that Chna s regonal specalzaton declned n 1980s and 1990s because of Chna s descendng domestc ntegraton n these perods. Whle other studes based dfferent measure methods argued that regonal specalzaton n Chna ncreased after the late 1980s (e.g. Brun and Renard, 2002; Naughton, 2003; Ba et al., 2004; Lang and Xu, 2005).

4 2. Revew: Varous Methods of Measurng Specalzaton Defntons: M N total No. of ndustres E r NE TE r TNE total No. of regons employment of ndustry n regon r M r 1 N 1 N E E r r M 1 r1 sr Er NE s r TE TNE Sr Er TEr S r NE TNE E r total employment of ndustry n the entre country total employment of all ndustres n regon r total employment of all ndustres n entre country employment share of ndustry n regon r n total employment of ndustry (ndustral regonal sze) employment share of regon r n aggregate employment (regonal sze) employment share of ndustry n regon r n all ndustres of regon r (regonal ndustral structure) employment share of ndustry n aggregate employment(natonal ndustral structure)

5 Hoover Localzaton Coeffcent (Hoover, 1936) Step 1. For any regon, calculate the employment share of ndustry n regon r n all ndustres employment of regon r, means, S E TE r r r and, the employment share of ndustry n all ndustres of the entre country, means, S NE TNE Step 2. Sort these two ratos n descendng order and calculate ther cumulatve value. Step 3. Draw these two cumulatve ratos on the vertcal axs and abscssa axs separately, thus we obtan a localzaton curve of ndustry Hoover localzaton coeffcent was defned as the rato of area between localzaton curve and 45-degree lne dvded by area of the trangle.

6 If nterregonal dstrbuton of an ndustry s totally unform, the coeffcent equals to 0, and f an ndustry concentrated entrely n one regon, the coeffcent s 1. Hoover Coeff.: =A/(A+B) Locatonal Gn Coeff.: =A Krugman (1991) used a smlar ndex whch s called Locatonal Gn Coeffcents to measure ndustry localzaton, the dstncton s Krugman s ndex just equals to the area between localzaton curve and 45-degree lne, thus ts value nterval s [0, 0.5].

7 Locaton Quotent (Haggett, 1965) SLQ r Er / TEr S NE / TNE S r An mportant defect of Smple LQ s that t does not consder the effect of regonal sze whch, as we knew, s an essental part of a coeffcent for analyzng the nterregonal trade relaton(round, 1978). In terms of varous defntons, who s regonal sze? s r TE TNE r

8 Spatal Gn Coeffcent (Ellson & Glaeser, 1994, 1997) Recall that If G=0, t mples the ndustry dstrbutes unformly across regons, a greater value ndcates a hgher spatal or geographcal concentraton. Snce t emphaszes departures of a gven ndustry s employment, spatal Gn coeffcent s of economc nterest, and moreover, t s easer to work wth than Gn coeffcent (Ellson and Glaeser, 1994). Then make a smple transformaton, M G s s s E NE 2 r1 r r r r sr r G r Note that Sr Er TEr N 1 TE TNE S S 2 r S NE TNE

9 Industral Specalzaton Index(Krugman, 1991) Industral specalzaton ndex measure specalzaton through measurng ndustral structure dfference of nter-regon. If ndustry structure of two regons are the same, the ndex equals to 0; If ndustry structure of two regons are complete rrelevant, t s equal to 2. N S S S r1 r2 1 r1 r2 Krugman s ndex of four major economc regons n Chna 2007 Eastern Chna Central Chna Western Chna Northeastern Chna Eastern Chna (+) (+) (+) Central Chna (+) Western Chna (+) 1993 Eastern Chna Central Chna Western Chna Northeastern Chna Eastern Chna Central Chna Western Chna Note: (+) means the ndex ncreased n 2007 comparatve wth that n Source: Calculated by Author based on 26 two-dgt ndustres.

10 Herfndahl Index Herfndahl ndex for ndustral localzaton s defned by H M r 1 s 2 For a gven ndustry, f every regon get the same share, Herfndahl equals to 1/M; If the gven ndustry only concentrates n one regon, Herfndahl equals 1. Smlarly, as regards a gven regon, Hr N 1 r S 2 Value nterval of Herfndahl ndex for regonal specalzaton s [1/N, 1]. If all ndustres n regon r have equal shares, ndex s equal to 1/N, and f regon r specalzed n only one ndustry, the ndex equals 1. r

11 E-G Coeffcent (Ellson & Glaeser, 1994, 1997) All coeffcents mentoned prevously have a common defect of mssng the mpact of frm locaton behavor, and as everyone knows, plant s dstrbuton s a very mportant factor of ndustral localzaton as well as regonal specalzaton. Fortunately, Herfndahl method s always used to measure ndustral concentraton usng frm s market share, thus f enterprse-level data s avalable, t enables us to construct a specalzaton ndex ncludng plant s geographc dstrbuton. Assume that there are m frms n ndustry and market share of frm k s z k. Then Herfndahl ndex s defne by H z 2 * k k, k 1, 2,

12 Fnally, E-G Coeffcent s defned as M 2 * G 1 sr H r1 M 2 * 1 s 1 r H r1 whch measures ndustry agglomeraton effect both ncludng effect of frm sze and regon sze. In terms of the constructng method of E-G coeffcent, Lu and Tao (2005) proposed a smlar ndex for measurng regonal specalzaton, G S H r N 2 * 1 S 1Hr 1 N 2 * r 1 r 1 Where And, E rk s employment or output of frm k n regon r. H E E * r k rk k rk

13 FLQ (Flegg & Webber, 1997) As dscussed prevously, a defect of LQ s that t does not consder the effect of regonal sze. In order to overcome ths problem, Flegg and Webber (1997) put forward a formula whch nvolves the effect of regon sze, where thus FLQ r SLQ * log 2 1TEr TNE Sr FLQ r log2 1 sr S r * δ s senstvty, δ s smaller, regonal scalar λ becomes more concave, and concde whle δ=0. Fnally, FLQ nvolves not only the relatve sze of supplyng and purchasng sector but also the regonal sze.

14 Selected Emprcal Applcaton on Chna s Economy Based on these methods, there are many emprcal studes for measurng Chna s regonal specalzaton. Although some studes found that Chna s regonal specalzaton declned n 1980s and 1990s because of Chna s descendng domestc ntegraton n these perods (Young, 2000; Poncet, 2003), most of studes argued that regonal specalzaton n Chna ncreased after the late 1980s. Brun and Renard (2002) and Naughton (2003) proved an accentuaton of regonal specalzaton between the late 1980s and early 1990s. Ba et al. (2004) found that regonal specalzaton n chna from 1985 to 1997 reversed at 1988 usng Hoover coeffcent, t regstered a sgnfcant ncrease after Usng Krugman s specalzaton ndex, Lang and Xu (2005) found except for Hanan and Hube, all other provnces experenced an ncrease n specalzaton from 1988 to Studes of Lu and Tao (2005) whch made use of and frm s data from 1998 to 2003, and studes of Fan (2007) whch used the ndex constructed by hmself from 1985 to 2004 also obtaned same conclusons. r

15 Short Comments We can fnd that almost all measurng methods dscussed above have a common character, ther man components are varous shares, and bascally, they are mproved based on LQ. Issue.1 LQ gnored the effect of regon sze. [Most of coeffcents ncludng FLQ have solved ths one.] Issue.2 As for some methods(e.g. E-G coeffcent), they are dffcult to use owng to the complcated computatonal process or the requrement of complete and hgh qualty data.[but both LQ and FLQ are easy to calculate] Issue.3 Some methods mentoned above(ncludng FLQ) gnored the mpact of ndustral regonal scale, namely,. s r Issue.4 Both LQ and FLQ cannot drectly descrbe the trend of ndustral geographc concentraton or regonal specalzaton n tme dmenson, they can only measure the concentraton of an ndustry n a gven regon, n other words, the result s a three dmenson dataset composed of tme, regon and ndustry. [How to solve Issue.3 and Issue.4 f we just have smple data?]

16 3. Coeffcents: A Method Improved from FLQ (Lu, Flegg & Deng,2011) How to overcome Issue.3? Lu, Flegg and Deng(2011) argued that although relatve regonal sze s a good adjustment factor of locaton quotent, t s stll too rough to measure ndustral geographc dstrbuton snce adjustment by regonal sze may lead to overestmaton or underestmaton of ndustral concentraton and regonal specalzaton. And actually, n terms of Gn coeffcent for concentraton and Hoover coeffcent for concentraton, we also can realze the mportance of regonal ndustral scale when economsts construct a coeffcent for measurng ndustral geographc concentraton. Thus, the most mportant factor should be regonal ndustral scale rather than regonal sze. Therefore, Lu, Flegg and Deng(2011) used ndustral regonal scale as an adjustment factor, t can be rewrtten as followng equaton. Sr FLQ r log 2(1 sr ) S

17 How to overcome Issue.4? Locaton quotents can be used to measure absolute ndustral geographc concentraton. For example, f LQ>1, t generally ndcates a hgh concentraton, and a hgher LQ mples a hgher concentraton. Lu, Flegg and Deng(2011) argued that, mathematcally, the specalzaton of an ndustry or a regon s not only related to the value of LQ but also related to ther varance (or dscrete degree): If the varaton of ther value for a certan ndustry s very bg, t ndcates that ths ndustry manly localzed n few regons, and ts spatal dstrbuton s more uneven; Whereas f the varaton s very small, t ndcates ts spatal dstrbuton s more unform; For a gven regon, f the varaton of ther value s very bg, t ndcates that ths regon manly specalzed n few ndustres. Consequently, the dsperson of locaton quotents value can reflect and measure the specalzaton of a regon and the localzaton of an ndustry.

18 Coeffcent of Varaton(CV) For non-negatve seres X x1, x2, x3,, xn the coeffcent of varaton s defned by where SD CV SD X x x x s standard devaton of X X s the mean If all x are equal, CV takes the lowest value 0; If all x but one are equal to zero, CV reaches the upper bound n 1 (Martn and Gray, 1971). (Note that t doesn t make sense f all are equal to 0)

19 Fnally, Coeffcent for ndustral localzaton s defned as M M 2 M FLQr FLQr FLQr M r1 M r1 M r1 Coeffcent for regonal specalzaton s defned as N N 2 N r FLQr FLQr FLQr N 1 N 1 N 1 where Sr FLQ r log 2(1 sr ) S

20 0 0, M 1 If ndustry dstrbutes unform n all regons and all regons have an equal regonal sze. M r 0 1 If ndustry dstrbutes only n one regon. r 0, N 1 If regon r has all ndustres and every ndustry has an equal ndustral regonal sze. r N 1 If regon r only specalzes n one ndustry. For proof of coeffcent nterval, see Lu, Flegg and Deng(2011).

21 4. Measurement of Regonal Specalzaton n Chna Usng coeffcents, measure result s showed n the fgure below: 1.65 Mean of lᵢ Mean of lᵣ Source: Lu, Flegg and Deng(2011)

22 r Regonal Dfference (, 2007) Source: Lu, Flegg and Deng(2011)

23 Industral Dfferences (, 2007) 2 [1, 2) <1 [1]Tobacco Manufacture ** [2]Chemcal Fber Manufacture [3]Petroleum and Natural Gas Extracton** [4]Coal Mnng and Washng Notes: [ ] s the ndustral rankng;** ndcates the ndex n 2007 decreased comparson wth n 1993,.e. the regonal specalzaton of ths ndustry reduced. [5]Manufacture of Communcaton Equpment, Computers and Other Electronc Equpment [6]Mnng and Processng of Non-Ferrous Metal Ores [7]Manufacture of Textle Apparel, Footware and Caps [8] Textle Manufacture [9]Mnng and Processng of Ferrous Metal Ores** [10]Manufacture of Measurng Instruments and Machnery or Cultural Actvty and Offce Work [11] Manufacture of Transport Equpment [12] Electrcal Machnery and Equpment Manufacture [13]Processng of Petroleum, Cokng, Processng of Nuclear Fuel [14] Manufacture of Metal Products [15]Smeltng and Pressng of Non-Ferrous Metals ** [16]Manufacture of General Purpose Machnery [17] Paper and Paper Products Manufacture [18]Smeltng and Pressng of Ferrous Metals [19]Food Manufacture ** [20]Mnng and Processng of Nonmetal Ores [21]Food Processng from Agrcultural Products [22]Beverage Manufacture [23]Manufacture of Specal Purpose Machnery [24]Manufacture of Non- Metallc Mneral Products [25] Medcne Manufacture [26]Raw Chemcal Materals and Chemcal Products Manufacture Source: Lu, Flegg and Deng(2011)

24 5. Remarks Every measure method has ts advantage. You can choose a sutable coeffcent accordng to your orgnal data. For example, f you have frm dataset, you can choose E-G coeffcent. Whereas f you just have some smple data, you can choose Herfndahl, Spatal Gn, coeffcent or some others. As regards Chna s regonal specalzaton, conclusons show that Chna s regonal specalzaton ncreased from the late 1980s, n partcular nland areas. Western Chna exhbts much hgher regonal specalzaton than do eastern coastal areas. The drect and man reason could be the uneven ndustral geographc dstrbuton: Most ndustres agglomerate n eastern developed areas, thus these regons do not specalze n few ndustres, but nland areas are stll at a low level stage of ndustralzaton and only focus on some resource-ntensve and labor-ntensve ndustres.

25 Core References Lu, Z., Flegg, T. and Deng, X. (2011). Regonal Specalzaton: A Measure Method and the Trends n Chna. MPRA Paper, No.33867, Unversty Lbrary of Munch, Germany. [all lteratures mentoned above are ncluded n ths paper, Artcle avalable: Lu, Z. and Deng, X. (2012). Comparatve Advantage, Scale Economy and Regonal Specalzaton: An Emprcal Analyss on Chna s Industres. Journal of Management Scence & Statstcal Decson, Vol.9, No.3, forthcomng. [Ths paper used Spatal Gn coeffcent to measure ndustral localzaton, orgnal artcle avalable: Flegg, T. and Webber, C. D. (1997). On the Approprate Use of Locaton Quotents n Generatng Regonal Input-Output Tables: Reply. Regonal Studes, 31, Ellson, G. and Glaeser, E. L. (1994). Geographc Concentraton n U.S. Manufacturng Industres: A Dartboard Approach. NBER Workng Paper, No [avalable: ; Fnal verson was publshed n Journal of Poltcal Economy, 1997, vol.105, pp ]