Chapter 4. An Overview of the Dynamic GTAP Data Base: The Data Base Construction and Aggregation Programs

Transcription

1 Chapter 4 An Overvew of the Dynamc GTAP Data Base: The Data Base Constructon and Aggregaton Programs Robert A. McDougall, Terre Walmsley, Alla Golub, Elena Ianchovchna and Ken Itakura 1. Introducton The GDyn Data Base, used wth the Dynamc GTAP model, s based on the GTAP Data Base (Dmaranan, 2005), whch descrbes the global economy n a gven year (2001 for the GTAP 6 Data Base). The GTAP Data Base s augmented wth addtonal data requred for the Dynamc verson of the GTAP model. Ths chapter covers the preparaton of these addtonal elements of the GDyn Data Base, 1 and the manpulaton and aggregaton of the GDyn Data Base wth varous utlty programs. The focus s on the addtonal data requrements, wth mnmal dscusson of the dynamc parameters. Readers nterested n the underlyng GTAP Data Base are referred to Dmaranan (2006). The GTAP data base (Dmaranan, 2006) s changed n the followng three ways to obtan a database for GDyn: a) x new arrays, lsted n table 1, are added to the standard GTAP data fle. The values of regonal savngs also change from those n the standard GTAP Data Base. The dmenson of the array however does not change. These changes are dscussed n secton 2 below. b) A new parameters fle contans the seven new parameters used n the dynamc theory (Table 2). The methodology for estmatng econometrcally, and n some 1 We have occason at varous ponts to refer to specfc versons of the GDyn data base. We use a verson numberng system of the form <GTAP release>:<gdyn verson>. The element <GDyn verson> s of the form <GDyn major verson>.<gtapdyn mnor verson>. We declare a new major verson to mark a major dfference n data content; a new mnor verson, to mark a change n format or a mnor change n content. For example, verson 6:1.0 of the GDyn data base s based on release 6 of the GTAP data base. Verson 6:2.0 s based on the same GTAP release, but ncorporates more recent GDyn specfc data. At the tme of wrtng, the current verson s 6:2.0.

2 cases calbratng these parameters, was examned n detal n chapter 3. In ths chapter we outlne the procedure for assgnng values to these parameters n the standard GDyn Data Base. These changes are dscussed n secton 3 below. c) One parameter array from the standard GTAP parameters fle (Dmaranan, 2005) RORDELTA s not used n the new nvestment theory and can be deleted from the fle. Table 1: New Arrays n the andard Data fle Header name Coeffcent name Dmensons Unts Descrpton KHAT KHAT REG year -1 normal captal stock growth rate RRGT RORGTARG REG year -1 target gross rate of return YQHT YQHTRUT REG $U Mllons YQTF YQTFIRM REG $U Mllons YQHF YQHFIRM REG $U Mllons ncome of regon from global trust ncome of global trust from frms n regon ncome of regon from local frms

3 Table 2 Contents of the Dynamc Parameters Fle Coeffcent Name Dmensons Descrpton INC REG 2 ntal ncome LAMBKHAT REG coeffcent of adjustment n estmated normal growth rate LAMBRORGE REG coeffcent of adjustment n expected rate of return LAMBRORG REG coeffcent of adjustment n rate of return RORGFLEX REG elastcty of rate of return to captal wth respect to captal stock RIGWQH REG rgdty of allocaton of wealth by regonal household RIGWQ_F REG rgdty of source of fundng of enterprses 2. The GDyn Data Fle We construct the new GDyn data as follows. Frst, we procure statstcs for foregn ncome recepts and payments; from these and standard GTAP data we obtan the arrays: YQHT, the ncome from equty owned by each regonal household n the global trust; YQTF, the ncome from equty owned by the global trust n each regons frms; and YQHF, local household ncome from ownershp of equty n local frms. We then recognze that foregn ncome and recepts affects regonal ncome. nce regonal ncome must equal regonal expendture, regonal expendture must also change when foregn ncome s ncorporated. Accordngly, we calculate new values for ntal regonal ncome, and regonal expendtures are adjusted through a change n regonal savngs. The adjusted data on savng s stored n the standard data fle under header AVE. econd, the normal rate of growth n the captal stock (KHAT) s estmated. The normal rate of growth n the captal stock s perceved by nvestors and assumed to converge to 1) the actual normal rate of growth n the captal stock, and 2) the actual captal stock growth rate n the long run equlbrum. To determne ths perceved normal 2 REG denotes number of regons n the GDyn model and GTAP data base.

4 growth rate of captal, tme-seres estmates for nvestment are collected and used to construct tme seres data for captal stock. The normal growth rate of captal (KHAT) s then set equal to the resultng geometrc average growth rate of captal. Thrd, the target gross rate of return, RORGTARG, s set unformly across regons and equal to the world average actual gross rate of return. In prevous versons of the GDyn Data Base (6:1.0 and lower) the expected gross rate of return (RORGEXP) was also ncluded n the GDyn Data Base. In the release 6:2.0 of the GDyn Data Base, RORGEXP s calculated wthn the GDyn model code. Ths change was made to allow users to specfy dfferent values of LAMBRORG, upon whch RORGEXP depends. RORGEXP s set ndvdually n each regon so as to reconcle the nvestment level mpled by the nvestment theory wth the level recorded n the data base. Each of these stages s dscussed n detal below. Income and savng For the GDyn data base verson 6:2.0 1, the data reference year s The standard country GDP data set s based on the macro country data set used n buldng the standard GTAP data base verson 6, supplemented by estmates for 2001 from the CIA World Factbook. For foregn ncome recepts and payments, we use the World Development Indcators (WDI) data set from the World Bank. For the GDyn data base verson 6:2.0, we use the 2002 edton of the WDI. We extract two data seres: Foregn ncome recepts (BoP, current U$) Foregn ncome payments (BoP, current U$) We obtan foregn recepts data for 154 countres, and foregn payments data for 155. The foregn ncome payments and recepts shares n total ncome are also used to determne the share of foregn and domestc ownershp of regonal captal stocks. An alternatve would be to use wealth/ownershp data to mply the shares of ncome. 4 Unfortunately, the set of countres for whch wealth/ownershp data are avalable s small and data for the correct base year unavalable. For example the Kraay et al. (2000) 3 Verson 5:1.0 s based on data from Ths approach was used n earler versons of the GDyn Data Base.

5 database contans 1997 data for only 38 countres, sgnfcantly less than the 154 countres for whch data on ncome are avalable. For ths reason we choose to use ncome data and mply from ths data wealth. The foregn ownershp shares of regonal captal obtaned from the foregn ncome payments and recepts WDI data for 1997 and 2001 are compared to the shares from Kraay et al. (2000) n fgure 1. The data show that the share of foregn ownershp s hgher when wealth (Kraay et al., 2000) s used to calculate the shares, as opposed to ncome from the WDI, and that the 2001 shares of foregn ownershp are hgher than the 1997 shares, reflectng the fact that the share of foregn ownershp has grown over ths perod. We should draw the attenton to the comfortng fact that the shares obtaned from Kraay et al. data and the ones from WDI data follow the same cross-country pattern. New foregn ncome and savng data are constructed n fve steps and dscussed n detal below: fllng n mssng values n the foregn ncome recepts data, fllng n mssng values n the foregn ncome payments data, balance foregn ncome recepts and payments, aggregate foregn ncome data from countres to GTAP regons, and calculate new values for ncome earned by domestc captal, total regonal ncome, and savng, n each regon. The average rato of foregn ncome recepts to GDP (FYRFFACT) s calculated, for countres for whch we have foregn ncome data. Foregn ncome recepts data are extended to cover all standard countres, assumng that for the mssng countres, the rato of foregn ncome recepts to GDP s equal to the average rato for countres for whch data are avalable, that s, calculatng the mssng value of foregn ncome recepts for country c as FYRFFACT*GDP(c). We take the GDP values from the standard country GDP data set. We use a smlar calculaton to extend the foregn ncome payments data to cover all standard countres.

6 Fgure 1: Comparson of hare of Foregn Ownershp based on Income and Wealth Australa New Zealand Chna Japan Korea Phllppnes ngapore Thaland Inda r Lanka Canada UA Mexco Venezeula Argentna Brazl Uruguay Austra Fnland France Germany WDI 1997 Kraay et al (2000) 1997 WDI 2001 UK Ireland Italy Netherlands ource: GDyn Data Base versons 5:1.0 and 6:2.0 and Kraay et al (2000). Portugal pan Turkey outh Afrca The foregn ncome data are unbalanced. That s, the world total for foregn ncome recepts s not equal to the world total for foregn ncome payments. A new common target total s calculated as the geometrc mean of the two new totals 5. The data s then balanced by rescalng both recepts and payments to match the common target total, adjustng the values for all countres. The balanced foregn ncome data are then aggregated across countres to the GTAP regons. Ths gves us the arrays YQHT and YQTF. Income from captal n each regon (VOA) s then obtaned from the standard GTAP standard data fle. ubtractng foregn ncome payments (YQTF) from ths we then obtan estmates for local ncome from domestc captal n each regon (YQHF). In so dong, we assume mplctly that all foregn ncome payments are ncome from captal. 5 Total of all factor ncome payment/recepts for all standrad countres (ncludng countres for whch data was avalable and countres for whch data was flled).

7 Takng account of the new foregn ncome payments and recepts data, we calculate new values for total ncome, INC, and for savng n each regon, AVE: Where: AVE(r) INCOME(r) - PRIVEXP(r) - GOVEXP(r) INCOME( r) VOA(, r) + YQHHLD( r) + TAXREV ( r) ENDWNA r REG r REG Normal Rates of Growth n Captal tocks For the normal rate of growth n the captal stock, KHAT, we use tme seres nvestment data (referred to below as the macro tmes seres data 6 ); as well as nvestment and captal stocks data for 2001 from the GTAP Data Base (referred to below as the sngle-year macro data set 7 ). For verson 6:2.0 of the GDyn data base, the macro tme seres data are avalable for 117 countres for the perod from the WDI, whle the 2001 nvestment data from the standard GTAP data base are avalable for 226 countres 8 (we refer to ths lst of 226 countres as the GTAP standard lst of countres). For each regon we defne the normal rate of growth n the captal stock as the rate at whch the captal stock n that regon can grow, wthout any ncrease or decrease n the rate of return. We store a regon-specfc parameter KHAT representng the normal rate of growth as estmated by nvestors n the data fle. We proceed n fve steps dscussed n detal below: 6 The macro tme seres data set contans tme seres nvestment, GDP and other macro varables. In ths chapter however we are only nterested n the nvestment data. 7 The sngle-year macro data set also contans other macro data, ncludng GDP, nvestment, captal stocks, consumpton, government spendng etc. 8 The sngle-year macro data for 2001 uses ts own country classfcaton, whch s ncomplete relatve to the lst of 226 countres upon whch the GTAP Data Base s based. We fll n the gaps usng GDP data from the World Bank whch are consstent wth the GTAP data. For nvestment, we calculate a scalng factor equal to the rato of two sums over the countres covered n the sngle-year macro data set. The numerator s the sum of nvestment values from the macro data set; the denomnator s the sum of GDP values from the standard country GDP data set. We apply ths scalng factor to GDP values from the standard country GDP data set to fll n the nvestment values mssng from the sngle-year macro data set. We use smlar methods to fll n the mssng countres n the macro data for captal stocks, GDP, and the rest. These procedures are undertaken as part of the standard GTAP data base constructon procedures (Dmaranan, 2005).

8 fll n mssng values and dsaggregate country groups n the macro tme seres data, from the captal stock data n the sngle-year macro data set, and the nvestment data n the macro tme seres data set, construct tme seres data for captal stocks, aggregate the captal stock tme seres data from countres to GTAP regons, and calculate the geometrc average rates of growth n captal stocks, and set KHAT equal to those average growth rates. As mentoned above the macro tme seres data set s avalable for 117 countres/regons. The dataset s therefore ncomplete and somewhat aggregated relatve to the standard GTAP country classfcaton of 226 countres (notably, t treats the ovet Unon as a sngle country). We use data from the completed sngle-year macro data set to fll n mssng values and dsaggregate aggregate values. We fll n the mssng values n the nvestment tme seres by calculatng for each year n the tme seres data a country-generc nvestment scalng factor equal to the rato of two values for total nvestment n countres covered by the tme seres data, one value taken from the tme seres data set (denomnator), the other from the sngle-year macro data set (denomnator). We then apply these scalng factors to nvestment values for the requred countres taken from the sngle-year macro data set. To dsaggregate tme seres nvestment estmates for country aggregates, we use shares calculated from the nvestment data n the sngle-year data set. We fll n the mssng countres n the GDP tme seres by calculatng for each year n the tme seres data a country-generc GDP scalng factor, usng the same approach as that taken for the nvestment scalng factor. To dsaggregate tme seres nvestment estmates for country aggregates, we use shares calculated from the GDP data n the sngle-year data set to apporton the aggregated nvestment tme seres data across the countres n the aggregated regon. We then use the sngle-year macro data set and the dsaggregated (226 country level) macro tme seres data set to construct tme seres data for captal stocks n each country. nce the two nput data sets may use dfferent unts, we frst put them on a

9 common bass. To do ths we rescale the nvestment tme seres usng a country-specfc scalng factor, the rato of the GDP value from the sngle-year macro data set to the GDP value for the same reference year from the macro tme seres data set. Ths ensures that the 2001 tme seres nvestment data s equal to the nvestment n the 2001 sngle-year macro dataset and accounts for any dfferences n unts. We then construct the captal stock tme seres data usng the perpetual nventory method, takng for captal stock base values, the captal stock data from the sngle-year macro data set, for gross nvestment, the rescaled nvestment tme seres, and for the deprecaton rate, a separately suppled country-generc value. We aggregate the captal stock tme seres data from countres to GTAP regons. For each regon, we set KHAT equal to the average actual rate of growth between the frst and last year covered by the captal stock tme seres. The normal growth rates (KHAT) are wthn the range -1.9% to 14.5%, except Mozambque whch has a KHAT of 22%. 9 Target Rate of Return Fnally we nclude n the GDyn data fle a regon-dmenson array RRGT. Ths s the array contanng data on the target gross rate of return used by nvestors n allocatng captal between regons (RORGTARG). For each regon, we set the target rate of return equal to the world average rate of return, whch depends on the global earnngs to captal (EKTOT) relatve to the global value of captal (VKTOT). where: EKTOT RORGTARG VKTOT Captal r REG r REG VOA(j,r) VK(r) VOA(,r) s the value of output at agents prces of captal n regon r; VK(r) s the value of captal stocks n regon r. 9 Ths s most lkely due to the very low value of captal stocks n Mozambque n 2001 relatve to the hgh value of nvestment n 2001.

10 Thus, n the data base, though not necessarly n the model theory, target rates of return are unform across regons. 3. Parameters The new behavoral parameters must then be ncluded n a new dynamc parameters fle. For all the arrays n the parameters fle, except for ntal ncome INC, the values set are based on the results of the emprcal analyss undertaken n chapter 3. For a more detaled explanaton of the parameters used n the GDyn model the reader s referred to chapter 3. Here we summarze the GDyn parameters and provde a bref account of ther chosen default values n the GDyn Data Base. INC The frst new parameter, lsted n Table 1, s INC. Ths parameter s the ntal ncome level n U$ mllons n the ntal GDyn data base. It s used for calculatng welfare measures n mult-perod experments and remans unchanged over the smulaton and tme. Lagged Adjustment Parameters The lagged adjustment parameters nclude: LAMBRORG, the coeffcent of adjustment n actual rate of return; LAMRORGE, the coeffcent of adjustment n expected rate of return; and LAMBKHAT, coeffcent of adjustment n perceved normal growth rate of captal stock. In the standard GDyn 6:2.0 Data Base, LAMBRORG and LAMBRORG are set equal to 0.4, and LAMBKHAT s set equal to 0.2 for all regons. Usng the calbraton procedures, outlned n chapter 3, we fnd that these values lead to emprcally consstent convergence of the rates of return n a typcal GDyn aggregaton (approx. 10 regons), and reduce the occurrence of negatve nvestment. Those readers wshng to calbrate these parameters to ther own aggregaton or alter the values to obtan alternatve rates of convergence are referred to Chapter 3 for further detals on the calbraton procedure and effects assocated wth alterng these parameters. Flexblty of Gross Rate of Return RORGFLEX s the elastcty of rate of return to captal wth respect to captal stock. Ths parameter s set equal to 1 for all regons, whch was shown n chapter 3 to be close to the true value of RORGFLEX gven the settngs for LAMBRORG stated above.

11 The true values of RORGFLEX can be obtaned va calbraton followng the technques outlned n chapter 3. It s hghly recommended that users do not alter RORGFLEX arbtrarly, as large dfferences between the chosen and true values can adversely affect convergence propertes of the model (see chapter 3). 10 Rgdty Parameters There are two rgdty parameters n the GDyn model: RIGWQH, whch s the rgdty of allocaton of wealth by regonal household; RIGWQ_F, the rgdty of source of fundng of enterprses. We use the emprcal estmates of rgdty parameters (as specfed n Table 7 n Chapter 3) to obtan rgdty parameters for the 226 standard GTAP countres. The followng procedure s used: Rgdty estmates for 57 countres and 5 regons, ncludng Benelux, were provded n chapter 3. After applyng the rgdty parameters of Benelux to Belgum and Luxembourg, data for 59 countres are avalable. Of these 59 countres, 42 countres are contaned n the GTAP 6 Data Base. For these 42 countres n the GTAP 6 Data Base the country specfc rgdty estmates come from Table 7 n Chapter 3. The rgdty parameters for the 4 regons 11 from Table 7 n chapter 3 were then assgned to the 16 GTAP regonal groupngs used n the producton of the GTAP 6 Data Base accordng to Table 3 below 12. The remanng standard countres (226 less 42) were then assgned the rgdty parameters of the regon (one of 16) to whch they belonged It should also be noted that calbrated RORGFLEX s dependent on LAMBRORG and hence changes n LAMBRORG affect the true value of RORGFLEX. If users choose calbrated RORGFLEX, as opposte to RORGFLEX1, and wsh to alter LAMBRORG, they are advsed to recalbrate RORGFLEX to obtan new true value. 11 All Industralzed countres, All Latn Amerca, All East Asa Pacfc and All. 12 Note that the rgdty estmates for the remanng 17 (59 less 42) countres not n the GTAP 6 Data Base were not used due to aggregaton problems. The alternatve would have been to allocate estmates to all 226 countres and then aggregate the parameters of the countres to obtan regonal estmates. Ths method was not used because of the lack of country estmates to aggregate for those regons and the dffcultes n obtanng data for weghtng such an aggregaton. 13 nce all the regonal rgdty parameters were smlar, ths allocaton method dd not yeld any real dfferences between regons and hence where countres wthn the regon were not ncluded n the econometrcs the parameters for ALL were used.

12 Table 3: Allocaton of Rgdty Parameters across 16 GTAP Regonal Groupngs Based on the followng countres/regons from RIGWQH RIGWQ_F Table 7 n chapter 3: Oceana All East Asa ALL EAP outh East Asa ALL EAP outh Asa All North Amerca All outh Amerca All LAC Central Amerca CRI, LV, GTM, HND, NIC Carbbean All Europe All Eastern Europe All Former ovet Unon All Mddle East All North Afrca All outhern Afrca All Central and Eastern Afrca All Western Afrca All The rgdty parameters for the 226 countres were then aggregated up to the 87 regons n the GTAP 6 Data Base, usng smple averages. nce all countres not covered n the GTAP Data Base as separate regons, but presented as regons compostes, are assgned regonal averages, the weght used to aggregate are rrelevant. 3. Creatng Aggregatons of the GDyn Data Base and Parameters The aggregaton program developed for the GDyn model allows the user to aggregate up the data base and parameters from the GTAP regonal and sectoral level to

13 an aggregaton whch can be used n ther analyss/ applcaton. In developng the aggregaton procedure, ssues arse concernng the choce of aggregaton formulae. It s not obvous that all parts of the aggregate data base should be constructed by aggregatng the orgnal data base. Alternatve methods such as calbraton suggest themselves for some parameters. Furthermore, for some parameters, t s not obvous what aggregaton formulae should be used. In general, we calculate aggregate parameters as weghted averages of dsaggregate parameters; but the choce of weghts s not always obvous. In earler versons, the parameters KHAT and RORGFLEX were set by an teratve procedure nvolvng calbraton smulatons wth the GDyn model. Ths was done after aggregaton. Ths however cuts across our goal of provdng a quck and easy aggregaton procedure. In the latest verson of the GDyn Data Base, we determne the aggregate KHAT and RORGFLEX parameters not by post-aggregaton calbraton but by applyng smple aggregaton formulas. Wth ths approach, KHAT and RORGFLEX are unlkely to match exactly wth the underlyng or true" parameters. Ths rases no partcular problems for KHAT. The model theory does not requre that t should agree wth the true normal rate of growth n the captal stock, snce KHAT s a self-correctng parameter that vares through tme to elmnate dvergences between the perceved and true rate of normal growth. RORGFLEX, on the other hand, s not self-correctng and problems assocated wth the aggregaton of RORGFLEX and the other parameters may arse as dscussed n chapter 3. KHAT The parameter KHAT s updatable and therefore held n the data fle. In aggregatng t, we use captal stock weghts calculated from the array VKB n the data fle. AKHAT(n) r n regon n VKB(r).KHAT(r) AVKB(n) r REG, n AREG Where: AKHAT(n) s the aggregated KHAT for regon n; VKB(r) s the value of captal stocks for country r (n aggregated regon n); AVKB(n) s the aggregated value of captal stocks for the regon n; and

14 KHAT(r) s the normal growth rate of captal for country r (n aggregated regon n). RORGFLEX The parameter RORGFLEX represents the elastcty of the expected rate of return wth respect to the captal stock. The aggregate elastcty, φ log R E log K Where: R E represents the expected (gross) rate of return and K the captal stock. Lettng the subscrpt run over the correspondng dsaggregate regons, we have R E E K E K R E, K K where E represents earnngs; so d logr E d logr + E, E, E, d logk - d logk (1) Where: E, represents the share of regon n aggregate captal earnngs. Then, assumng unform relatve change n the captal stock, assumng that s, that for all dsaggregate regons, d logk d logk, we have: d logr o φ E E, logr d logr E, E, E, E, logk E, logr logk o for RORGFLEX, earnngs weghts are approprate. More precsely, snce the rate of return n queston s the expected not the actual rate of return, we should use expected not actual earnngs weghts. Note that the assumpton of unform captal stock growth s not nnocuous: we should have obtaned a dfferent aggregaton formula f we had assumed, for example, unform relatve change n the rate of return. LAMBRORGE The parameter LAMBRORGE s used to defne the error-correcton effect for the expected rate of return: E, φ

15 logre t EΛ R E logk t Then: Where: E s a measure of error. Recallng equaton 1, and puttng logk, we have agan: t d logr E E, d logr E, Λ RE 1 E 1 E E, E, logr E Λ t RE, E, of adjustment Then, assumng unform errors E (or no correlaton between errors E and speed Λ Λ RE, Λ R E E, ), we have: RE, o agan, expected earnngs weghts are approprate. Gven the parameters for all regons are set equal to 0.2 the aggregaton procedure s not relevant under the standard settngs; however, f changes are made to specfc countres/regons ths would become an ssue. Ths s also true for the aggregaton of LAMBRORG and LAMBKHAT. LAMBRORG The parameter LAMBRORG represents the dervatve of the requred rate of growth n the rate of return wth respect to a measure of dvergence between the expected and the target rate. At any tme t, nvestors hold a percepton or expectaton not only of the current rate of return, but also of the rate of return at future tmes. They requre that ths expected rate of return should grow through tme at a rate that depends on the dvergence between the expected and the target rate. If the expected rate exceeds the target rate, the expected rate should grow at a postve rate; f the target rate exceeds the expected rate, the expected rate should declne. The dervatve of the rate of growth n the expected rate of return wth respect to the measure D of dvergence s LAMBRORG or Λ RR.

16 Now: Λ R R D d logre d t so, agan recallng equaton 1, and assumng logk logk t t, we have: Λ R R D E, d logr dt E, have: Then, gnorng any effect of change n dvergence on the earnngs shares, we Λ R Λ R E, R R, o once agan, we use expected earnngs weghts. As n the case of LAMBRORGE, LAMBRORG parameters for all regons are set equal to 0.2. LAMBKHAT The parameter LAMBKHAT or Λ represents the dervatve of the expected Kˆ normal rate of growth n the captal stock wth respect to a measure of error n expectatons: Λ Kˆ d log Kˆ E dt d log Kˆ K, E dt K, Λ K, o for LAMBKHAT we use captal stock weghts. Agan the parameters for all regons are set equal to 0.2. RIGWQH The parameter RIGWQH descrbes the rgdty of the allocaton of the wealth of the regonal household between local and domestc assets. To aggregate t we use the wealth of the regonal household (coeffcent WQHHLD) for weghts. RIGWQ_F

17 The parameter RIGWQ_F descrbes the rgdty of the sourcng of funds of enterprses between locally and foregn sources. To aggregate t we use the value of the domestc captal stock (coeffcent WQ_FIRM) for weghts 14. References Dmaranan, Betna V., Edtor (2005). Global Trade, Assstance, and Producton: The GTAP 6 Data Base, Center for Global Trade Analyss, Purdue Unversty. 14 If we derved the weghts to be used wth RIGWQH and RIGWQ_F wth the same level of care as the earler parameters, we would obtan some strange formulae, n whch the weghts sum to less than one, except n the specal case that the composton of wealth (for RIGWQH) or sourcng of funds (for RIGWQ_F) s unform across dsaggregate regons wthn dsaggregate regons. Ths s smlar to the stuaton we meet wth domestc / mport or factor substtuton elastctes (detals avalable on request). In those stuatons n standard GTAP, we gnore the complextes, and use plan weghts summng to one; we do the same here. Then the choce of weghtng coeffcent can be justfed by ths argument: For RIGWQH, as the share of some dsaggregate regon n the aggregate regon's wealth approaches one, the aggregate regon's rgdty parameter should approach that dsaggregate regon's parameter; smlarly for RIGWQ_F and domestc captal.