REVEALED FACTOR INTENSITY INDICES AT THE PRODUCT LEVEL

UNITED NATIONS CONFERENCE ON TRADE AND DEVELOPMENT POLICY ISSUES IN INTERNATIONAL TRADE AND COMMODITIES STUDY SERIES No. 44 REVEALED FACTOR INTENSITY INDICES AT THE PRODUCT LEVEL by Mho Shrotor UNCTAD Bolormaa Tumurchudur UNCTAD Olver Cadot Unversty of Lausanne, CEPR, CEPREMAP and CERDI UNITED NATIONS New York and Geneva, 2010

NOTE The purpose of ths seres of studes s to analyse polcy ssues and to stmulate dscussons n the area of nternatonal trade and development. The seres ncludes studes by UNCTAD staff, as well as by dstngushed researchers from academa. In keepng wth the obectve of the seres, authors are encouraged to express ther own vews, whch do not necessarly reflect the vews of the UNCTAD secretarat or ts member States. The desgnatons employed and the presentaton of the materal do not mply the expresson of any opnon whatsoever on the part of the Unted Natons Secretarat concernng the legal status of any country, terrtory, cty or area, or of ts authortes, or concernng the delmtaton of ts fronters or boundares. Materal n ths publcaton may be freely quoted or reprnted, but acknowledgement s requested, together wth a reference to the document number. It would be apprecated f a copy of the publcaton contanng the quotaton or reprnt were sent to the UNCTAD secretarat at the followng address: Chef Trade Analyss Branch Dvson on Internatonal Trade n Goods and Servces, and Commodtes Unted Natons Conference on Trade and Development Palas des Natons CH-1211 Geneva Swtzerland Seres edtor: Khallur Rahman Chef, Trade Analyss Branch UNCTAD/ITCD/TAB/46 UNITED NATIONS PUBLICATION ISSN 1607-8291 Copyrght Unted Natons 2010 All rghts reserved

Abstract Ths paper descrbes a data-ntensve methodology to generate ndces that ndcate revealed factor ntensty of traded goods, at the most dsaggregated level of product classfcaton (SITC 5-dgt or HS 6-dgt). We construct the ndces by calculatng, for each good, a weghted average of the factor abundance of the countres that export ths good, where the weghts are varants of Balassa s Revealed Comparatve Advantage ndex. In dong so, we take advantage of recent mprovements n the avalablty of data on aggregate natonal endowments of prmary factors (captal, educaton and labour force) usng, nter ala, Summer and Heston s PWT (verson 6.2), Barro-Lee s latest database, the World Bank and COMTRADE databases. The resultng ndces are avalable on the UNCTAD webste. Key words: Internatonal trade, factor endowments, factor ntenstes, product classfcatons JEL Classfcaton: F1

Acknowledgements The authors are grateful to Alessandro Ncta, Marco Fugazza, Sudp Ranan Basu and Joerg Mayer for ther valuable suggestons and comments. Any mstakes or errors reman the authors own. v

Contents 1. Introducton...1 2. Natonal factor endowments: physcal captal, human captal and natural resource endowment...3 2.1 Measurng natonal factor endowments...3 2.1.1 Captal stock...3 a. Constructon of captal stock...3 b. Descrpton of captal stock and ts relablty...4 2.1.2 Human captal stock...7 a. Measures for human captal stock...7 b. Descrpton of human captal stock...9 2.1.3 Natural resource endowment...12 2.2 Cluster analyss of natonal factor endowments...13 3. Estmatng the revealed factor ntensty ndces...18 3.1 Methodology...18 3.1.1 Caveats...19 a. Lmtatons of Balassa s ndex...19 b. Dealng wth agrcultural dstortons...20 3.1.2 Data coverage...21 3.2 Results revealed factor ntensty ndces...23 4. Concluson...29 Appendx A Appendx tables and fgures...31 Appendx B Dataset of ndex of revealed factor ntensty...40 References...45 v

Lst of fgures Fgure 1. Real GDP per worker vs. real captal stock per worker (n logs)... 6 Fgure 2. Real GDP per worker vs. human captal stock (n logs)... 11 Fgure 3. Captal stock per worker vs. human captal stock (n logs)... 11 Fgure 4. Dendogram of Ward s cluster analyss: countres... 14 Fgure 5. Dendogram of Ward s cluster analyss: ndustral sectors... 25 Lst of tables Table 1. Real captal stock and real captal stock per worker estmates... 5 Table 2. Summary statstcs (n 2000 Unted States dollars)... 5 Table 3. Growth rate of real captal stock per worker, by ncome group... 5 Table 4. Growth rate of real captal stock per worker, by regon... 6 Table 5. Real GDP per worker vs. real captal stock per worker, by ncome group... 6 Table 6. Correlaton of alternatve captal seres wth our new real captal stock per worker seres... 7 Table 7. Summary statstcs, average years of schoolng... 9 Table 8. Average years of schoolng by regons... 10 Table 9. Annual average growth rate of average years of schoolng, by regons... 10 Table 10a Land database coverage... 12 Table 10b Natural captal, year 2000... 13 Table 11. Summary statstcs of factor endowments for the perod 1971 2003... 13 Table 12. Calnsk and Harabasz and Duda and Hart stoppng rules result... 14 Table 13. Summary of clusters... 16 Table 14. Effect of Vollrath s correcton on RCI ndex... 20 Table 15. Balanced data coverage... 21 Table 16. Wde (unbalanced) coverage for each endowment... 22 Table 17. Wde (unbalanced) coverage... 22 Table 18. Summary statstcs of revealed factor ntensty ndces, year 2000... 23 Table 19. Smple averages of factor ntensty ndces, by SITC 1 dgt ndustres... 23 Table 20. Rankng of ndustres n terms of factor ntensty ndces... 24 Table 21. Stoppng rule result... 25 Table 22. Summary statstcs of the clusters... 26 Table 23a. Percentages of SITC 1 dgt ndustres, by clusters... 26 Table 23b. Percentages of clusters, by SITC 1 dgt ndustres... 26 Table 24a. Percentages of SITC 2 dgt ndustres, by clusters... 27 Table 24b. Percentages of clusters, by SITC 2 dgt ndustres... 28 v

Appendx tables: A1. Countres ncluded n the sample...31 A2. Calnsk and Harabasz and Duda and Hart stoppng rules result (Arable land per captal excluded from the varable lsts)...33 A3. Summary of clusters (arable land per captal excluded from the varable lsts)...34 A4. Summary statstcs of revealed factor ntensty ndces, year 2000...36 A5. Calnsk and Harabasz and Duda and Hart stoppng rules result (HS classfcaton)...36 A6. Summary statstcs of the clusters (HS classfcaton)...37 A7. Percentages of HS sectons ndustres, by clusters (HS classfcaton)...37 A8. Percentages of HS chapters ndustres, by clusters (HS classfcaton)...38 B1a. Revealed factor ntensty ndces SITC...40 B1b. Revealed factor ntensty ndces database HS...41 B2a. Revealed factor ntensty ndces database (1994, 2000) SITC...41 B2b. Revealed factor ntensty ndces database (1994, 2000) HS...42 B3a. Endowment database, by country...42 B3b. Endowment database, by country (1994 and 2000) (wth data on natural captal and ts components from the World Bank)...43 Fgure A1. Dendogram of Ward s cluster (natural resource excluded from the varables)...33 Fgure A2. Dendogram of Ward s cluster analyss (HS classfcaton)...36 v

1. Introducton The process of export dversfcaton has long been a maor research ssue n nternatonal economcs. In recent years, we have seen a renewed nterest n the nature and the process of export dversfcaton. For nstance, Klnger and Lederman (2005) nvestgate the role of nnovaton n export dversfcaton. They fnd that off-the-fronter nnovaton (e.g. the ntroducton of new export products) s more common among low-ncome countres than among hgh-ncome countres. Consstent wth the U-shape pattern of the relatonshp between export dversfcaton and natonal ncome, as descrbed n Imbs and Waczarg (2003), overall export dversfcaton ncreases at low levels of development but declnes wth development after a mddle-ncome pont. Hausmann and Klnger (2006) suggest that changes over tme n the revealed comparatve advantage of countres are assocated wth the pattern of relatedness (measured n terms of productve factors) determned across products. As countres change ther export mx, there s a strong tendency to move towards related goods rather than to goods that are less related. They also suggest that a partcular product s proxmty to exstng areas of comparatve advantage s one of the most sgnfcant determnants of whether a country wll develop an advantage n that product n the future. These approaches provde fresh thnkng nto old ssues and that s very welcome. However, the debate remans largely dsconnected from tradtonal theory of factor-content of trade consderatons, namely the Hecksher-Ohln model of comparatve advantage based on relatve factor endowments. 1 A potental danger of such approaches s as usual wth nductve reasonng that one may end up nferrng supposedly general laws from statstcal relatonshps that may or may not hold out of sample, and, ultmately, advsng polcy on the bass of these emprcal fndngs. Ths study ams to provde a tool to fll the gap between tradtonal, theory-based approaches and newer eclectc ones by developng a tme-seres database of the ndces of revealed factor ntensty (RFI) of export products, usng a wealth of raw data accumulated n the last two decades. The ndces are constructed as follows. Frst, we collected (and updated whenever necessary) raw data on natonal factor endowments of physcal captal, human captal and natural resource endowment for countres of whch data was avalable. Combnng these data gves a panel database of factor endowments at the country level, spannng close to 100 countres over three decades (our balanced panel covers 92 countres from 1971 to 2003). Second, we calculated a revealed factor ntensty for each product at a dsaggregated level of product classfcaton (we used both Unted Natons Standard Internatonal Trade Classfcaton (SITC) 5-dgt and Harmonzed System (HS) 6-dgt), usng the factor endowments of countres exportng that product (from step 1). The dea s that a product exported predomnantly by countres that are rchly endowed wth human captal s revealed to be ntensve n human captal. To wegh natonal factor endowments n the averagng, we adopted a methodology frst proposed by Hausmann, Hwang and Rodrk (2007), usng as weghts a slghtly modfed verson of revealed comparatve advantage (RCA). Usng RCAs, nstead of smple export weghts, flters out scale effects. 1 See Feenstra (2004), chapter 2 for a survey of that older lterature. 1

Beng a weghted average of factor endowments, our measure s senstve to the country coverage of the endowments database. However, there s a trade-off between the one wth a large sample sze and the one whch s smaller n sze but wthout any mssng values. Therefore, we propose two versons of our revealed factor ntenstes: (a) a wde one, based on the wdest annual country coverage; and (b) a consstent one, based on a balanced panel of data. We were also careful to weed out, as much as we could, the effect of subsdes and other trade dstortons. Because these dstortons are prevalent n agrculture, we used the World Bank s new Agrcultural Dstortons database (Anderson et al., 2008) and elmnated observatons where RCAs were obvously drven by polcy. Wthout ths correcton, we would have hgh revealed human captal ntenstes for agrcultural goods whose exports are subsdzed by rch countres. The resultng RFI ndces are presented and analyzed n varous ways n the paper. We beleve the value added of the RFI ndces s that t wll make possble to control for Heckscher Ohln effects n analyss of trade dversfcaton n a way that was not possble before. The outlne of the paper s as follows. Chapter 2 provdes a detaled descrpton of the constructon of natonal factor endowments. Chapter 3 provdes a descrpton of the constructon of the ndex of revealed factor ntensty, and dscusses caveats. It also uses cluster analyss to explore broad groupngs of products on the bass of ther revealed factor ntenstes. The explanaton of the database of the ndces of revealed factor ntensty s attached to the paper. The database of the ndces s accessble and can be downloaded from UNCTAD webste (http://r0.unctad.org/dtc/tab/ndex.shtm). 2

2. Natonal factor endowments: physcal captal, human captal and natural resource endowment 2.1 Measurng natonal factor endowments 2.1.1 Captal stock a. Constructon of captal stock Ths secton descrbes the dervaton of our database of aggregate (natonal) captal stock estmates. In general, two methods are avalable: (a) drect measurement through surveys and (b) perpetual nventory method (PIM). Because drect measures are not everywhere avalable, we use the PIM. The PIM reconstructs captal stock estmates from nvestment flows by addng up, recursvely, current nvestment to the prevous perod s captal stock, approprately deprecated. The method rases (nter ala) two problems. One s the ntal estmate of the captal stock, the other s the choce of the deprecaton rate. We have followed the approach of Easterly and Levne (2001, henceforth EL) 2 and replcated ther captal stock estmates usng the updated verson 6.2 of the Penn World Table (PWT) whch provdes aggregate nvestment fgures 3 for 159 countres. Let K t and I t be respectvely the real captal stock and nvestment flow of country n perod t. The captal-accumulaton equaton s K t+ = ( 1 δ ) K, t + I 1 t, where δ s the rate of deprecaton. Followng EL, we assume that country s at ts steady-state dk captal-output rato, whch mples that t / K t = dyt / Yt dk. Snce t= I t δk t, then dk t/ K t = I t / K t δ * * * * *. At the steady-state growth rate, be g = dy Y = I K δ, we can wrte g * * * * δ * * * K κ I Y ι = = δ (1) Y 2 They used the PWT 5.6 captal stock data, based on dsaggregated nvestment and deprecaton statstcs for 64 countres. They also constructed captal stock fgures for more countres usng aggregate nvestment fgures. 3 It would have been desrable to use dsaggregated nvestment seres (especally our nterest s havng the seres for non-resdental nvestment), but the PWT only provdes wth the aggregate nvestment seres. Though the verson 5.6 provdes captal stock for non-resdental, t covers much less countres and perods. 3

where * ι s the nvestment rate and * κ s the captal-output rato. The latter can thus be wrtten as κ = * ι + δ * * g (2) * Followng EL, we construct g the steady-state growth rate as a weghted average of the country s average growth rate durng the frst 10 years for whch the PWT have output and nvestment data and the world growth rate. That s, * ( 1 λ) g g g W = λ + (3) where bars represent values averaged over the sample s frst 10 years. The world growth rate s computed as 0.0423. Followng Easterly et al. (1993), we set λ at 0.25. We compute * ι smlarly as the average nvestment rate durng the frst 10 years for whch there s data. Fnally, we get an estmate of the ntal captal stock K = κ Y% (4) * 0 0 where Y % 0 s the average real output value between 1950 and 1952 rather than smply the frst observaton n the sample perod n order to reduce the nfluence of busness-cycles. For countres where output and nvestment data do not start untl 1960, everythng s moved down one decade. As for δ, we agan follow EL n assumng a deprecaton rate of 7 per cent. b. Descrpton of captal stock and ts relablty Table 1a shows the coverage of our estmates of the real captal stock. We cover 159 countres, 154 of whch have more than 30 years of tme seres. In order to construct a seres for the real captal stock per worker (K/L), we used an ndrect approach usng real gross domestc product (GDP) per worker (Y/L), GDP per capta (Y/P) and populaton (P) from the PWT to nfer the numbers of workers (L), all from the PWT. Because the PWT has mssng data for GDP per worker, the coverage was further reduced to 152 countres, 140 of whch have 30 years of tme seres or more (see table 1b), and 136 of whch have data over the common sample perod 1971 2003. 4

Tme perod Table 1. Real captal stock and real captal stock per worker estmates a. b. Number of Start of Countres tme seres Tme perod Number of Countres Start of tme seres 14 2 1990 1 2 1996 17 1 1988 14 2 1990 23 1 1981 16 1 1988 24 1 1980 21 1 1977 30 1 1971 24 3 1971 31 1 1973 24 3 1977 33 38 1971 30 2 1971 34 15 1971 31 1 1973 43 19 1961, 62 33 38 1971 44 11 1961, 60 42 1 1962 45 2 1960 43 28 1961 48 4 1956 44 3 1960 49 2 1955 47 1 1952 50 2 1954, 55 48 4 1956 51 1 1954 49 3 1955 52 3 1952, 53 50 2 1954 53 20 1951, 52 51 1 1953 54 35 1951 52 5 1952 159 53 51 1951 152 Table 2 shows summary statstcs for the 136 countres for whch we have data for the common tme perod 1971 2003. Table 2. Summary statstcs (n 2000 Unted States dollars) Varable Obs Mean Std. Dev. Mn Max Real GDP per capta 136 8 061 8 309 521 42 419 Real Captal Stock per worker 136 31 307 37 116 376 160 177 Table 3 shows the growth of our estmates of the real captal stock per worker by ncome group over 1971 2003. Low-ncome countres have had the slowest growth, wth negatve growth n the 1980s and the 1990s. Table 3. Growth rate of real captal stock per worker, by ncome group (annual average percentage change at 2000 US dollar) Growth rate of Real Captal stock per worker (per cent) Income group 1971-1981 1982-1992 1993-2003 1971-2003 Hgh ncome: OECD 2.8 1.6 2.2 2.2 Hgh ncome: nonoecd 2.0 1.0 2.0 1.6 Upper mddle ncome 2.4-0.4 1.7 1.2 Lower mddle ncome 3.8 0.4 0.2 1.4 Low ncome 1.9-0.6-0.3 0.3 Table 4 decomposes ths rate of growth n terms of ncome groups and regonal breakdown. The general trend s a steady declne n rates of captal accumulaton per worker, wth some recovery n the decade between 1993 and 2003. It s notable that the growth rate of captal stock vares consderably across developng regons, as well as across perods wthn a developng regon. 5

Table 4. Growth rate of real captal stock per worker, by regon Growth rate of Real Captal stock per worker (per cent) 1971-1981 1982-1992 1993-2003 1971-2003 Hgh ncome countres 2.6 1.4 2.1 2.0 Low and mddle ncome East Asa & Pacfc 4.3 3.0 2.8 3.4 East Asa & Pacfc (wthout Chna) 2.6 1.4 2.1 3.2 Europe & Central Asa 6.3 1.5 0.7 2.7 Latn Amerca & Carbbean 2.0-1.1 1.2 0.6 Mddle East & North Afrca 6.0 0.1-2.2 1.2 South Asa 3.9 3.6 3.2 3.6 Sub-Saharan Afrca 1.8-0.2 0.8 0.7 In order to check the plausblty of our estmates, we plot the sample-perod average, per country, of real GDP per worker aganst the real captal stock per worker, both n logs (fgure 1). The real GDP per-worker s a proxy to aggregate labour productvty of the country. Thus, they should be correlated f our estmates are reasonable. Fgure 1. Real GDP per worker vs. real captal stock per worker (n logs) Output per worker, 2000 US dollars 7 8 9 10 11 BRN QAT SAU LUX ARE USA OMN BHR FRA AUT BELNOR NLD CHE PRI ISRITA CAN MAC BRBGAB MLT ANT IRL HKG GBRNZL BHSISL SWE DNK AUS DEU ESP JPN FIN SGP CYP ARG PRT TTOGRC MUS CRI ZAF URY MEX CHLVEN HUN SWZ DZA KOR IRN JOR BLZ PAN MYS SUR BRA NIC COL TUN CUB GTM SLV DOM PRYNAMFJI PER ECU POL ROM TURBWA IRQ EGY MAR JAM PNG CPV PHL GNQ BOL ZWE THA CMR SYR CIV MDV HND IDN LKA PAK SLB GIN COG IND COM SEN SLE LSOMNG CHN NGA LBRSDNMRT LAO BENGHAKEN SOM PRK NPL TGO ZMB RWA MOZ MDGNER ZAR CAF UGA GMB TCD MLI BDI BFA MWI GNB ETH TZA KHM BTN 6 8 10 12 Real Captal stock per worker, 2000 US dollars The scatter plot shows that they are ndeed hghly correlated. 4 Table 5 presents the data aggregated by ncome group, whch agan shows a plausble degree of correlaton. Table 5. Real GDP per worker vs. real captal stock per worker, by ncome group (n 2000 Unted States dollars) Income Group Real GDP per worker Real Captal Stock per worker Hgh ncome: OECD 39 934 92 454 Hgh ncome: nonoecd 42 842 53 368 Low ncome 2 715 2 470 Lower mddle ncome 9 121 11 738 Upper mddle ncome 16 958 23 470 4 We have done the same fgures for three dfferent perods (1971 1981 1982 1992 and 1993 2003) to see whether the correlaton was mantaned over the perods. The postve correlaton was observed. 6

As a further check, table 6 shows the correlaton between our seres and alternatve estmates. It can be seen that the degrees of correlaton are all above 90 per cent. Table 6. Correlaton of alternatve captal seres wth our new real captal stock per worker seres Our estmates of Real Captal Stock per worker Our estmates of Real Captal Stock per worker a 1 Replcaton of Klenow-Rodrguez Clare Nehru-Dhareshwar Larson et al. 0.9854 (3 856) 0.9307 (2 382) 0.7878 (1 411) Replcaton of Klenow- Rodrguez Clare 1 0.9153 (1 896) 0.9381 (1 323) Nehru- Dhareshwar 1 0.9580 (1 138) Larson et al. 1 Note: a As we descrbed n the text, ths s our update of Easterly and Levne (2001). Larson et al. (2000) covers 62 ndustral and developng countres for the years 1967 92. Nehru Dhareshwar (1993) covers 92 ndustral and developng countres from 1960 1990. In brackets are the number of observatons. 2.1.2 Human captal stock a. Measures for human captal stock There are varous types of proxes that have been used for measurng human captal. These nclude lteracy rates, school enrolment ratos, educatonal attanment and average years of schoolng. Among those, the last one average years of schoolng s the most popular, partly because of the avalablty of large datasets n terms of country coverage and the length of perod for whch data s avalable. There are several data sets on educatonal attanment. The avalable datasets can be dvded nto two groups dependng on whether they make use of (a) census/survey data, whch are the only drect numbers avalable together wth school enrolment rato; or (b) only the school enrolment rato. The frst group (Kyracou 1991 5 ; and Barro and Lee, 1993, 2001) reles on census numbers whenever those are avalable, and flls n mssng values usng a regresson of average years of schoolng on lagged enrolment rates. However, ths procedure s vald only when the relatonshp between these two varables s stable over tme and across countres, whch s not often the case. As an alternatve, Barro and Lee use an accuracy test based on a sample of 30 countres wth relatvely 5 Kyracou (1991) estmated the average years of schoolng of the labor force for a sample of 111 countres for the perod of 1965-1985 at fve-year ntervals. He uses UNESCO census data and Psacharopoulos and Arragada (1986) attanment fgures to estmate average schoolng years on school enrollment ratos. Psacharopoulos and Arragada (1986) reports data on educatonal composton of the labor force n 99 countres and provdes estmates of average years of schoolng. The man drawback s that they provde only one tme-seres observaton n most countres. 7

complete census numbers n order to fll n mssng values. 6 As such, Barro and Lee s data may be more robust than Kyracou s, although ths s largely a matter of udgement. The second group (Lau, Jamson and Louat, 1991; Lau, Bhalla and Louat 1991; and Nehru, Swanson and Dubey, 1995) uses only school enrolment ratos to construct human captal stock seres. 78 Ther PIM s a sophstcated verson of Barro and Lee, but they gnored census data on educatonal attanment. Based on Krueger and Lndahl s (2001) estmates of the relablty of the Barro and Lee and Kyracou datasets, we chose to use Barro and Lee s data, although there are arguments n favour of both. The latest verson of the dataset, descrbed n Barro and Lee (2001), ncorporates varous mprovements n the procedure used to fll n mssng values. 9 De la Fuente and Doménech (2001) and Cohen and Soto (2000) provde useful ndcatons on how to clean up the avalable 10 11 census/survey data. Barro and Lee estmated two sets of educatonal attanment rates at fve-year ntervals from 1960 for dfferent levels of educaton for overall populatons aged over 15 and over 25 6 Barro and Lee use a PIM that starts wth the survey numbers as benchmark stocks, and then use the school enrolment ratos to estmate the changes from the benchmarks. Ths method s vulnerable to naccuraces n the underlyng data on gross enrolment ratos. They assess ts accuracy for the 30 countres for whch they have complete census estmates for 1960, 1970 and 1980 as follows. Frst, they use the benchmark values for 1960 (1970) and PIM n the forward drecton to estmate attanment n 1970 (1980), yeldng forward-flow estmates. Second, they start wth benchmark values n 1970 (1980) and use PIM backward to estmate attanment n 1960 (1970), yeldng backward flow estmates. Then they compare the accuracy of these two estmates wth forecasts from smple lnear trends: extrapolatons from 1960 and 1970 to an estmate for 1980 and from the values for 1970 and 1980 to an estmate for 1960. They also estmated lnear nterpolatons from the values for 1960 and 1980 to estmates for 1970 and ran several regressons of the observed values of varous levels of educatonal attanment n 1960, 1970 and 1980 for the 30 countres on the estmates generated from forward- and backward-flow and lnear extrapolaton and nterpolaton methods. They found that lnear extrapolatons for 1960 and1980 were nsgnfcant n all cases, and so was the backward-flow estmate for 1970. By contrast, the forward-flow estmate was sgnfcant n all cases for 1980, and the forward-flow and lnear nterpolaton for 1970 were ontly sgnfcant n all cases. For more detals see Barro and Lee (1993). 7 Nehru, Swanson and Dubey (1993) ntroduced several mprovement n Lau, Jamson and Louat s procedure. Frst, they collect more data on school enrolment pror to 1960 and therefore they do not have to rely on the backward extrapolaton. Next, they dd some adustment for grade repetton and drop-outs. 8 Lau, Jamson and Louat (1991) and Lau, Bhalla and Louat (1991) use a PIM and annual enrolment data to construct educatonal attanment seres. Ther PIM uses age-specfc survval rates constructed for representatve countres n each regon. 9 The Barro and Lee (2001) dataset mproves on ther earler estmates n a number of respects. Frst, fll-n procedure for mssng values now uses gross enrolment ratos, adusted for repeaters. Second, n the constructon of average years of schoolng, they now take account of changes of school duraton over tme wthn countres. 10 De la Fuente and Doménech (2001) construct educatonal attanment seres for the adult populaton of a sample of 21 OECD countres coverng the perod 1960 1995. Ther approach has been to collect all the nformaton that could be found on educatonal attanment n each country, both from nternatonal publcatons and from natonal sources and use t to reconstruct a plausble attanment profle for each country. 11 Cohen and Soto (2000) construct a dataset for a sample of 95 countres coverng the perod 1960 2000 at 10-year ntervals. The key methodology s to mnmze the extrapolaton and keep them as close as possble to those drectly avalable from natonal census. They collect census/survey data from UNESCO, the OECD s n-house educatonal database and webstes of natonal statstcal agences. Ther estmates refer to the 15 64 age group. 8

respectvely. 12 For each level, the attanment rate s defned as the percentage of the relevant subpopulaton (over 15 or over 25) havng been enrolled up to a specfc level of educaton but no further (.e. those who dd not pursue any further educaton than the gven level). 13 Barro and Lee estmate the average years of schoolng usng attanment data as follows: _ (1/2) ( 1) ( 1 2) ( DUR DUR 1 DUR 2 (1/ 2) DUR ) h ( DUR DUR 1 DUR 2 DUR ) h Av yrs = DUR h + h + DUR + DUR h + DUR + DUR + DUR h + p p cp p s s p s s cs + + + + + + + + p s s h h p s s h ch where h represents the percentage of populaton wth dfferent degree of educatonal attanment descrbed by subscrpts. That s, for each h, the th level of educaton s the hghest attaned: =p for ncomplete prmary educaton, cp for completed prmary educaton, s for the frst cycle of secondary educaton, cs for the second cycle of secondary, h for ncomplete hgher educaton and ch for completed hgher educaton. DUR s the duraton n years of the th level of schoolng : =p for prmary, s1 for the frst cycle of secondary, s2 for the second cycle of secondary and h for hgher. 14 Because the Barro and Lee dataset only gves values for each fve years, we used a technque of nterpolaton/extrapolaton to obtan yearly fgures from 1960 to 2004 for 105 countres. 15 Note that the data s not adusted for educaton qualty. Educaton qualty vares across countres, and avalable data s too fragmentary to be exploted systematcally. 16 b. Descrpton of human captal stock Table 7 shows the summary statstcs of the estmated human captal stock seres. It can be seen that there s a large varaton across countres n the average number of years of schoolng. Table 7. Summary statstcs, average years of schoolng Varable Obs Mean Std. Dev. Mn Max Average Years of School 4 710 4.65 2.94 0.04 12.30 Country averages over sample perod Varable Obs Mean Std. Dev. Mn Max Average Years of School 105 4.64 2.72 0.44 11.02 12 Barro and Lee (2001) provde data for the populaton aged 25 and over and for the populaton aged 15 and over. The earler verson of Barro and Lee provded the data only for the populaton aged 25 and above n order to obtan the wdest possble coverage. However, focusng only on the populaton aged 25 and over was gnorng the fastest growng segment of the labour force n the developng countres. Therefore, the latest verson of Barro and Lee also provdes the data for the populaton aged 15 and over whch corresponds better to the labour force for many developng countres. 13 The raw data on educatonal attanment come from ssues of UNESCO Statstcal Yearbook, whch reports census and survey data by age and sex. 14 See Barro and Lee (1993, 2001) for detals. 15 For Benn and Egypt, we extrapolated only untl 1965 and 1970 respectvely, snce the extrapolatons backward were resultng n a negatve numbers. Congo, Gamba and Chna were extrapolated backward from 1975 untl 1960 and Rwanda from 1970 untl 1960. 16 Studes by Nehru, Swanson and Dubey (1995) and Cohen and Soto (2001) show that there s a hgh degree of correlaton between Barro and Lee estmates and other estmates of educatonal stocks. 9

Table 8 shows average years of schoolng broken down both by decade (1971 1981, 1982 1992, and 1993 2003) and by ncome group/regon. The data has 34 hgh-ncome countres (24 Organzaton for Economc Cooperaton and Development (OECD) and 10 non-oecd), 17 upper mddle-ncome countres, 28 lower mddle-ncome countres, and 24 low-ncome countres. The low and mddle-ncome group s further broken down nto sx regons: East Asa and Pacfc (8 countres), Europe and Central Asa (3 countres), Latn Amerca and Carbbean (21 countres), Mddle East and North Afrca (6 countres), South Asa (6 countres), and sub-saharan Afrca (25 countres). The average years of schoolng rses n all regons. Among the low- and mddle-ncome countres, sub-saharan Afrca, South Asa and Mddle East and North Afrca have the lowest averages but show the hghest growth rates (see table 9). Table 8. Average years of schoolng by regons Regon 1971 1981 1982 1992 1993 2003 1971 2003 Hgh ncome countres 6.53 8.08 9.05 7.56 Low and mddle ncome countres East Asa & Pacfc 3.01 4.49 5.54 4.02 Europe & Central Asa 5.13 6.71 7.63 6.17 Latn Amerca & Carbbean 3.45 4.80 5.64 4.35 Mddle East & North Afrca 1.39 3.30 5.04 2.79 South Asa 1.59 2.44 3.08 2.18 Sub-Saharan Afrca 1.44 2.48 3.30 2.17 Table 9. Annual average growth rate of average years of schoolng, by regons Regon 1971 1981 1982 1992 1993 2003 1971 2003 Hgh ncome countres 1.3 1.1 0.8 1.1 Low and mddle ncome countres East Asa & Pacfc 1.8 2.4 1.3 1.9 Europe & Central Asa 1.3 1.5 0.8 1.2 Latn Amerca & Carbbean 1.7 1.7 1.1 1.5 Mddle East & North Afrca 5.1 5.0 2.8 4.5 South Asa 2.6 2.0 1.9 2.3 Sub-Saharan Afrca 2.8 3.2 1.7 2.6 As a relablty check, fgure 2 shows average output per worker at the country level (averaged over the sample perod) aganst average years of schoolng. The correlaton s, as expected, qute hgh. 10

Fgure 2. Real GDP per worker vs. human captal stock (n logs) Average years of schoolng -1 0 1 2 3 NZL USA CAN SWENOR POL DNKCHE AUS KOR HUN BRBFIN JPN PHL FJI ARG CYP IRL GBR DEU ISR GRC ISL AUT NLD FRA PANURY CHL TTO MLT LKA ECU PER MYS ZAF ESP SGP ITA CHN BOL THA PRY CRI JOR MEXVEN MUS JAM ZMBLSO SYR DOMCOL SWZ BHR PRT IND COGIDNZWE TUR BWA SLV NICBRA GHA HND EGY KEN IRQ IRN TUN DZA MWI GTM UGAZAR CMR TGO SEN PAK LBR RWA PNG CAF BENSLE SDN GMB NPL MOZ MLI NER 7 8 9 10 11 Real GDP per worker, n 2000 US dollar As a further check, fgure 3 shows years of schoolng aganst the real stock of captal per worker, both n logs. It can be seen that the relatonshp s postve, reflectng correlaton wth a thrd varable (ncome levels), but also concave: there s more deepenng of physcal captal than human captal n the north-east of the scatter plot. Fgure 3. Captal stock per worker vs. human captal stock (n logs) Average years of schoolng -1 0 1 2 3 UGA RWA MOZ NZLUSA CAN SWE POL DNK AUS NOR CHE BRB HUN KOR CYP IRL GBRISRDEU JPN FIN PHL FJI ARG GRC ISL AUT NLD FRA PAN URYCHL MLT TTO ZAF PER LKA PRY CRI MYS ECU ESPITA SGP CHN BOL THA JOR MUS MEXVEN SYR LSO DOMSWZ COLJAM BHR ZMB IND IDNSLVTUR ZWEBWA GHA EGY HND COG PRT NIC BRA KEN IRQ IRN TUN DZA MWI GTM SEN ZAR CMR TGO PAK LBR PNG CAF SLE BEN SDN GMB NPL NER MLI 6 8 10 12 Real Captal per worker, n 2000 US dollar 11

2.1.3 Natural resource endowment To measure the natural resource endowment n a country, we use the data on arable land taken from the World Bank s World Development Indcators (WDI). The seres we used arable land hectares per person s presented n 1,000 ha per person, and covers 203 countres over the perod of 1961 2005. 17 Out of those countres, 164 have 45 years or more of data (see table 10a). Table 10a. Land database coverage Tme perod Number of countres Start of tme seres 2 1 2004 3 9 2003 6 2 2000 11 2 1995 13 3 1993 14 19 1992 16 1 1990 26 1 1980 42 1 1961 45 164 1961 203 One ustfcaton for usng arable land s that t does not stay the same over tme for each country as t reflects land development or desertfcaton. However, the avalablty of arable land tself s not a perfect measure of natural resource endowments of a country. Therefore we also look nto a database on natural resource captal from the World Bank s volumes Expandng the Measure of Wealth (1997) and Where s the Wealth of Natons? (2006). They offer, among others, a database on natural captal for over 100 countres. Though the database covers only two years (1994 and 2000), t provdes us wth the most complete measure of natural resource endowments to date and t could be used as a good ndcator. Natural resource captal n the database conssts of non-renewable resources (subsol assets, ncludng ol, natural gas, coal, and mneral resources), cropland, pastureland, forested areas (ncludng areas used for tmber extracton and non-tmber forest products), and protected areas. 18 Natural captal values are gven per capta and are based upon country-level data on physcal stocks, and estmates of natural resource rents are based on world prces and local costs. Table 10b presents the total values of natural captal and ts components by ncome groups. Whle the value of natural captal per capta s substantally hgher n hgh-ncome countres than low ncome ones, the percentage of cropland and pastureland n total natural captal s sgnfcantly hgher n low ncome countres. 17 Arable land (hectares per person) ncludes land defned by the Food and Agrculture Organzaton of the Unted Natons (FAO) as land under temporary crops (double-cropped areas are counted once), temporary meadows for mowng or for pasture, land under market or ktchen gardens, and land temporarly fallow. Land abandoned as a result of shftng cultvaton s excluded. See WDI explanatons. 18 For detals, see World Bank (2006). 12

Table 10b. Natural captal, year 2000 (Unted States dollars per capta) Income group Subsol assets Tmber resources Nontmber forest resources Protected areas Cropland Pastureland Total Natural Captal % of Cropland and Pastureland n total natural captal Low-ncome countres 325 109 48 111 1 143 189 1 925 69% Mddle-ncome countres 1 089 169 120 129 1 583 407 3 496 57% Hgh-ncome countres (OECD) 3 825 747 183 1 215 2 008 1 552 9 531 37% World 1 302 252 104 322 536 4 011 51% Sources: The World Bank (2006), Table 1.2. Notes: Ol states are excluded. Both of the data, arable land and natural captal (and ts components), are gven per capta. In order to make them consstent wth our measures for physcal captal measured per worker, we have merged the data wth PWT to nfer the numbers of workers and converted them nto values of relatve endowments per worker. Gven the data avalablty, we use only arable land as a measure of natural resource endowment for the panel data for the common perod, that s from 1971 (1988 n HS) to 2003. But we also calculated RFI ndces separately for the years 1994 and 2000, usng the natural resource data. 2.2 Cluster analyss of natonal factor endowments We now examne whether our data are reasonable and realstc estmates of natonal factor endowment by usng cluster analyss. In order to avod scale effects (the fact that the range of a varable affects ts nfluence n the clusters defnton), we have standardzed all endowment varables to a mean of zero and a standard devaton of one. Ths prevents the captal stock, whch has a much wder range than the other two varables, from domnatng the clusterng procedure (see table 11). Table 11. Summary statstcs of factor endowments for the perod 1971 2003 Varable Obs Mean Std. Dev. Mn Max Captal Stock per worker 95 34 306 37 768 426 147 540 Human Captal Stock 95 5.04 2.81 0.51 11.58 Arable Land per worker 95 0.74 0.80 0.00 5.79 Two of the general types of clusterng methods are herarchcal and partton. Herarchcal clusterng methods create herarchcally related sets of clusters. Partton clusterng methods separate the observatons nto mutually exclusve groups. We appled here two alternatve algorthms to explore the endowment data s structure. Frst, we used a dstance-based agglomeratve clusterng algorthm known as Ward s method. The resultng dendrogram s shown n fgure 4. 13

Fgure 4. Dendrogram of Ward s cluster analyss: countres L2squared dssmlarty measure 0 50 100 150 200 250 G1 n=3 Dendrogram for L2wlnk cluster analyss G2 n=17 G3 n=20 G4 n=39 G5 n=16 Long vertcal lnes at the top of the dendrogram ndcate group of countres that are strongly dssmlar, and shorter lnes ndcate those that are less dssmlar. Fgure 4 suggests fve broad country groupngs. More formally, we appled two stoppng rules whose results are shown n table 12: Calnsk and Harabasz pseudo-f ndex, and Duda and Hart Je(2)/Je(1) ndex. The best stoppng level s gven by the maxmum value of the pseudo-f ndex or, alternatvely, by the mnmum value of the Je(2)/Je(1) ndex. Both support a fve-group structure. 19 Table 12. Calnsk and Harabasz and Duda and Hart stoppng rules result Calnsk & Harabasz Duda & Hart Numbers of Clusters Pseudo-F Numbers of Clusters Je(2)/Je(1) Pseudo T-squared 2 68.8 2 0.59 25.9 3 67.5 3 0.45 43.1 4 73.6 4 0.62 32.2 5 74.3 5 0.19 4.2 6 72.1 6 0.49 38.2 7 71.3 7 0.39 22.0 8 73.9 8 0.57 11.5 9 78.1 9 0.68 8.6 10 77.7 10 0.58 7.2 11 76.6 11 0.00. 12 75.8 12 0.63 14.1 13 76.2 13 0.58 5.0 14 76.1 14 0.63 7.6 15 76.6 15 0.14 18.0 19 The groupng nto clusters s even clearer when we exclude arable land use per capta from the endowment varables. See (appendx tables A2 and A3 and fgure A1). 14

Based on the workng hypothess that fve clusters s a natural partton of the data, we have used the k-means partton method (where k s specfed by the user, whch n our case we take as 5) to form the clusters teratvely. We base the partton on the Eucldean dstance metrc (also known as the Mnkowsk dstance metrc wth argument 2). The general form for the dstance metrc between observaton and centrod usng p varables s gven by 1/ n n p d = Xm Xm (5) m= 1 Table 13 descrbes the fve clusters and shows that the partton s farly natural. Cluster 1 s made of low-ncome countres and lower mddle-ncome countres (except Bahran and Portugal, whch are hgh-ncome countres but wth a very low endowment of land and a relatvely low captal and human captal endowment). It s characterzed by low captal and human captal endowments and wth the lowest endowment of arable land (all n relatve to labour). Cluster 2 s made of lower mddle-ncome countres wth a few low-ncome countres. The dfference of ths cluster from the above one s that t has the second hghest endowment of arable land. 20 Cluster 3 essentally conssts of upper mddle-ncome countres. Clusters 4 and 5 consst of OECD countres. The only dfference between these clusters s that two countres n cluster 4 (Canada and Australa) own a large land endowment n addton to large physcal and human captal endowments. 20 Turkey, whch s an upper mddle-ncome country, s ncluded n cluster 2, because (a) t has the lowest physcal- and human-captal endowment among the upper mddle ncome countres; and (b) ts arable land endowment s one of the hghest n the mddle-ncome countres. 15

Clusters Countres n Clusters Table 13. Summary of clusters Number of countres Captal Stock Human Captal Stock Arable land World Bank Income Group (1) (2) (3) (4) (5) (6) (7) Benn Low ncome Bahran Hgh ncome: nonoecd Bolva Lower mddle ncome Brazl Upper mddle ncome Botswana Upper mddle ncome Chna Lower mddle ncome Congo, Rep. Lower mddle ncome Colomba Lower mddle ncome Costa Rca Upper mddle ncome Domncan Republc Lower mddle ncome Egypt, Arab Rep. Lower mddle ncome Ghana Low ncome Gamba, The Low ncome Guatemala Lower mddle ncome Honduras Lower mddle ncome Indonesa Lower mddle ncome Inda Lower mddle ncome Jamaca Upper mddle ncome Jordan Lower mddle ncome Kenya Low ncome 1 Lbera 41 9 366 3.08 0.47 Low ncome Sr Lanka Lower mddle ncome Lesotho Lower mddle ncome Mal Low ncome Mozambque Low ncome Maurtus Upper mddle ncome Malaw Low ncome Ncaragua Lower mddle ncome Nepal Low ncome Pakstan Low ncome Papua New Gunea Low ncome Portugal Hgh ncome: OECD Rwanda Low ncome Senegal Low ncome Serra Leone Low ncome El Salvador Lower mddle ncome Swazland Lower mddle ncome Thaland Lower mddle ncome Uganda Low ncome Congo, Dem. Rep. Low ncome Zmbabwe Low ncome Central Afrcan Republc Low ncome Cameroon Lower mddle ncome Algera Lower mddle ncome Iran, Islamc Rep. Lower mddle ncome Iraq Lower mddle ncome Nger Low ncome 2 Paraguay 13 11 942 2.68 1.50 Lower mddle ncome Sudan Lower mddle ncome Syran Arab Republc Lower mddle ncome Togo Low ncome Tunsa Lower mddle ncome Turkey Upper mddle ncome Zamba Low ncome 16

Clusters Countres n Clusters Number of countres Captal Stock Human Captal Stock Arable land World Bank Income Group (1) (2) (3) (4) (5) (6) (7) Argentna Upper mddle ncome Barbados Hgh ncome: nonoecd Chle Upper mddle ncome Cyprus Hgh ncome: nonoecd Ecuador Lower mddle ncome Span Hgh ncome: OECD F Upper mddle ncome Greece Hgh ncome: OECD Hungary Hgh ncome: OECD Ireland Hgh ncome: OECD 3 Korea, Rep. Hgh ncome: OECD 22 36 781 6.76 0.65 Mexco Upper mddle ncome Malta Hgh ncome: nonoecd Malaysa Upper mddle ncome Panama Upper mddle ncome Peru Lower mddle ncome Phlppnes Lower mddle ncome Poland Upper mddle ncome Trndad and Tobago Hgh ncome: nonoecd Uruguay Upper mddle ncome Venezuela, RB Upper mddle ncome South Afrca Upper mddle ncome 4 Australa Hgh ncome: OECD 2 90 764 10.32 4.59 Canada Hgh ncome: OECD Austra Hgh ncome: OECD Swtzerland Hgh ncome: OECD Germany Hgh ncome: OECD Denmark Hgh ncome: OECD Fnland Hgh ncome: OECD France Hgh ncome: OECD Unted Kngdom Hgh ncome: OECD Iceland Hgh ncome: OECD 5 Israel 17 101 711 8.74 0.51 Hgh ncome: nonoecd Italy Hgh ncome: OECD Japan Hgh ncome: OECD Netherlands Hgh ncome: OECD Norway Hgh ncome: OECD New Zealand Hgh ncome: OECD Sngapore Hgh ncome: nonoecd Sweden Hgh ncome: OECD Unted States Hgh ncome: OECD 17

3. Estmatng the revealed factor ntensty ndces We now proceed to use our endowment data to buld our revealed factor ntensty (RFI) ndces of export products, usng a methodology nspred by Hausmann, Hwang and Rodrk s ndex of revealed technology content (PRODY). 3.1 Methodology Our Revealed Factor Intensty (RFI) ndces for each traded good s calculated as a weghted average of the factor abundance of the countres exportng that good, wth a varant of Balassa s Revealed Comparatve Advantage (RCA) ndces as weghts. The ratonale for usng a varant of RCA ndces as opposed to straght export shares ( X X ) s to ensure that country sze does not dstort the rankng of goods. For example, both Chna and Togo produce and export the 5 dgt SITC product category 65394, Fabrcs,woven,of vegetable textle. In year 2000, the export value of Chna for ths product was US$ 96 mllon, whereas Togo s export value was only US$ 0.1 mllon. However, ths product consttuted only 0.02% of total Chnese exports, compared to 0.05% for Togo. Therefore the ndex allows us to weght Togo s factor abundance more heavly than the Chnese factor abundance (37% for Togo, 17% for Chna) n calculatng the revealed factor ntensty level of the product, even though Chna s exports are bgger than Togo s. Thus, the revealed captal ntensty ndex of good s calculated as where K s country s captal stock, K k = ω L (6) L s ts labor force, and the weghts are gven by X X ω =. (7) ( X X ) That s, ω s a varant of Balassa s RCA for country n good. Balassa s ndex s where X s country s exports of good, world exports of good, and X = X whch s ( X X ) of Balassa s ndex, whch s X X RCA = (8) X X X = X s country s aggregate exports, X s s world aggregate exports. The denomnator ofω,,.e. the sum of product s shares across countres, s not dentcal wth that X X,.e. the share of product n world trade. In so dong, we use a trck frst used by Hausmann, Hwang and Rodrk (2007) whch ensures that the weghts add up to one, as 18

X X 1 = = ( X X ) ( X X ) ( X X ) = 1 ω. (9) Ths elmnates a problem of a large values of RCA ndces arsng from the values that are very close to zero n the denomnator (a product s share n world trade) at the dsaggregated level (lke Hausmann et al.). Smlarly, the revealed human captal ntensty ndex s gven by h = ω h (10) where h s the average years of schoolng acheved by the average person. The revealed land ntensty ndex, fnally, s calculated usng arable land per person, where l s the arable land (n hectares) per person. l = ω l (11) Two ssues are worth mentonng. Balassa ndces have been crtczed because (a) countres and commodtes are double-counted; and (b) they are based on gross exports, whereas (as the argument goes) t should be based on net exports nstead. Second, our ndex s potentally dstorted by export subsdes, and agrcultural exports are a partcularly severe problem. We deal wth both n turn. 3.1.1 Caveats a. Lmtatons of Balassa s ndex Vollrath (1987, 1989) suggested slghtly amended versons of the ndex. One elmnates the double countng: RCA Vollrath1 = X X X X (12) where X and X stands for country s exports net of ts exports of good (that s, X = X X ) and X stand respectvely for exports of good by all countres except and world exports net of country s. He also suggested the followng verson of the ndex, encompassng both mport and export dmensons of comparatve advantage: RCA Vollrath2 = X X X X M M M M (13) We tred both specfcatons and decded to reect them. The frst one makes lttle dfference and s not worth the complcaton. The second, by contrast, ntroduces consderatons whch make t unsutable for our purposes. To see ths, consder a world of three countres: France, Germany and Ghana, wth ntra-ndustry trade n telecom equpment between Germany and France, 19

and no trade n that product between ether of them and Ghana. Germany s a slght net exporter and France s a slght net mporter. In calculatng the revealed captal ntensty of telecom equpment, France and Germany wll cancel out each other, ther aggregate weght beng zero. The revealed captal ntensty wll then be ndetermnate. Ths s not a far-fetched example. Table 14 shows how usng Vollrath s second correcton for the RCA yelds a lower revealed captal ntensty for SITC 5-dgt 86198 (nstruments for physcal or chemcal analyss, traded by 96 countres) than for SITC 4-dgt 4217 (rape, colza and mustard ols). Table 14. Effect of Vollrath s correcton on RCI ndex Country SITC 4-5 dgt Export Import (xª/x)/ (xª/x) (mª/m)/ (mª/m) (xª/x)/ (xª/x)- (mª/m)/ (mª/m) Captal Stock per worker RCI Korea, Rep. of 86 198 11 275 239 806 0.17% 2.22% -2.05% 84 821-1'739 New Zealand 86 198 1 332 15 152 0.35% 1.61% -1.26% 88 927-1'123 Norway 86 198 7 351 34 717 0.38% 1.50% -1.12% 152 748-1'713 Canada 86 198 189 653 349 152 2.01% 2.18% -0.17% 110 351-188 Instruments for physcal or chemcal analyss 86 198 8 168 913 7 656 878 100.00% 100.00% 42 041 Rape, colza and mustard ols 4 217 51 161 b. Dealng wth agrcultural dstortons The last example rases an addtonal ssue. Many agrcultural commodtes end up wth hgh revealed captal and human captal ntenstes because they are exported by rch countres who subsdze them (export subsdes have been ltgated out for most manufactured products so we gnore them). Such outcome does not arse from comparatve advantage, but rather a result of drect polcy nterventon. We attempt to correct for dstortons n agrcultural prces usng a new database on agrcultural dstortons publshed by the World Bank n October 2008. 21 The database provdes, among others, a nomnal rate of assstance (NRA) for a number of agrcultural products for developed and developng countres over the perod 1955 2005. The agrcultural product coverage ncludes 70 per cent of agrcultural and food value added excludng hghly processed food, beverages and tobacco, and agrcultural crops, of those countres ncluded n the sample. 22 21 See Anderson, et al. (2008) for detals of the database. We would lke thank Kym Anderson and Ernesto Valenzuela for ther knd e-mals wth very useful clarfcatons to our questons on the database. 22 A smlar database has been provded systematcally for the last two decades by the OECD secretarat, whch provdes Producer Support Estmates (PSEs) and Consumer Support Estmates (CSEs). However, these estmates are gven only for a few key products, and for a much smaller number of countres (only for hgh ncome countres and fve non-european Unon developng countres) for the years from 1986 2005. 20

The NRA s measured as the unt value of producton at the dstorted prce less ts value at the undstorted free market prce expressed as a fracton of the undstorted prce. 23 NRA t k t E* P(1 + tk ) E* P t = = τ k (14) E* P where E s the domestc currency prce of foregn exchange and P s the foregn currency prce of good k n the nternatonal market and τ s the ad-valorem equvalent of the array of tarffs and kt domestc tax and subsdes affectng good k n country n year t, whch can be postve or negatve. It s typcally postve n hgh-ncome countres subsdzng and protectng agrculture, and negatve n low-ncome countres taxng thers (Anderson, 2008). Our correcton conssted of weedng out observatons (country product pars) characterzed by nonzero NRAs n order to keep only undstorted RCAs. 3.1.2. Data coverage In constructng the database, we faced a trade-off between wdth and consstency n country-endowment data. On the one hand, we are nterested n havng ndces for as many countres as possble, to gve a wdth to the database. On the other hand, to track the evoluton of RFI ndces for each good over several years, we need to have a complete (.e. balanced) panel of data on endowments of a gven set of countres for the same length of years, to ensure that the ndces are constructed n comparable ways. However, f there s systematc bas n the selecton of countres n the panel (say, f lowncome countres are underrepresented n the data), RFI ndces wll be based aganst factors of whch low-ncome countres are poorly endowed. Ths may not necessarly alter the rankng of goods by RFI, but wll affect the relatve ntenstes. In order to mnmze ths bas, the wdecoverage (unbalanced) panel ncludes, each year, all the countres for whch data are avalable n that year. Table 15 gves the largest number of countres wth a common set of tme perods for all trade and endowment data (the balanced panel). The resultng 92 countres are tracked over a sample perod of 33 years (1971 2003). Table 15. Balanced data coverage Number of years Tme perod Number of Countres Captal Stock 33 1971 2003 136 Human Captal Stock 45 1960 2004 105 a Land 45 1961 2005 165 b Ths study coverage 33 1971 2003 92 c Notes: a Egypt has 35 years of tme seres over the perod 1970 2004. b Benn has 40 years of tme seres over the perod 1965 2004. c WITS does not provde trade data for all these 33 years for 3 countres such as Lesotho, Swazland and Botswana. 23 OECD s PSEs are calculated as a fracton of the dstorted value; that s, PSE = t m ( 1+ tm ) and for a postve t m t s smaller than NRA and s necessarly less than 100 per cent, whch s not the case for the NRA. See Anderson, Kurzwel, Martn, Sandr and Valenzuela (2008) and OECD (2007). 21

Table 16 shows the wdest range of countres for each factor endowment (the unbalanced panel). For example we have the captal stock data startng from 1951, but the number of countres whch have the data vares from year to year. The number of countres havng all three endowments data vares between 76 and 99, dependng on the years (bottom lne of the table 16). Table 16. Wde (unbalanced) coverage for each endowment Range of years Range of number of countres Captal Stock 1951 2003 51 141 Human Captal Stock 1960 2004 103 105 Arable Land 1961 2005 165 203 All three endowments 1961 2003 76 99 year. Table 17 presents the number of countres covered n the unbalanced data sets for each Year Table 17. Wde (unbalanced) coverage Number of countres Year Number of countres 1961 76 1983 99 1962 77 1984 99 1963 77 1985 99 1964 77 1986 99 1965 78 1987 99 1966 78 1988 99 1967 78 1989 99 1968 78 1990 99 1969 78 1991 99 1970 79 1992 98 1971 98 1993 98 1972 98 1994 98 1973 99 1995 98 1974 99 1996 98 1975 99 1997 98 1976 99 1998 98 1977 99 1999 98 1978 99 2000 99 1979 99 2001 98 1980 99 2002 98 1981 99 2003 98 1982 99 Fnally, as regards product classfcaton, we calculated the ndces usng two dfferent classfcaton schemes: Revson 1 of the Unted Natons Standard Internatonal Trade Classfcaton (SITC 5-dgt) and the Harmonzed System (HS88/92 6-dgt). Each has ts own advantages and dsadvantages. SITC provdes longer years of trade statstcs (snce 1962) wth fewer revsons than the HS, thus has the advantage of gvng maxmum comparablty over the sample perod. HS gves us a more dsaggregated product classfcaton, at the 6-dgt level, than SITC. Whereas there are only over 1,000 products at the 4-5 dgt products of the SITC classfcaton, there are over 5,000 products at the HS 6-dgt level. 22

Our SITC database covers 1971 2003, and our HS6 covers 1988 2003, wth few countres untl 1992. 3.2 Results revealed factor ntensty ndces We now llustrate the results of our RFI ndces for the year 2000. Table 18 shows the summary statstcs of the RFI ndces for each good n the SITC classfcaton (correspondng tables of results usng the HS classfcaton are gven n appendx A). Table 18. Summary statstcs of revealed factor ntensty ndces, year 2000 (SITC classfcaton) Varable Obs Mean Std. Dev. Mn Max rhc 1166 7.07 1.62 1.52 11.21 rc 1166 60 257 30 726 2 608 149 916 rnr_land 1166 0.61 0.35 0.07 4.18 rnr_nc 1166 14 768 8 998 2 028 73 993 rnr_sa 1166 4 826 5 428 31 61 315 rnr_pc 1166 6 909 4 011 1 087 43 272 rhc rc rnr_land rnr_nc rnr_sa rnr_pc Revealed human captal ntensty Revealed (physcal) captal ntensty Revealed natural resource ntensty land Revealed natural resource ntensty natural captal Revealed natural resource ntensty sub-ol assets Revealed natural resource ntensty pastured and crop land level. Table 19 shows smple averages of RFI ndces for 10 ndustres at the SITC-1 aggregaton Table 19. Smple averages of factor ntensty ndces, by SITC 1 dgt ndustres stc1 SITC 1 dgt descrpton RHCI RCI RNRI_land RNRI_nc RNRI_sa RNRI_pc 0 Food and lve anmals 6.27 39 067 0.79 15 428 4 422 8 314 1 Beverages and tobacco 6.95 52 538 0.61 15 070 4 614 7 704 2 Crude materals, nedble 6.37 42 159 0.74 16 382 5 256 7 640 3 Mneral fuels, lubrcants 6.94 47 869 0.69 20 925 12 070 6 187 4 Anmal and vegetable ols and fats 5.67 34 756 0.74 12 748 3 795 6 709 5 Chemcals 7.66 72 169 0.59 16 641 6 119 7 354 6 Manufact goods classfed chefly 7.06 62 059 0.55 13 667 4 342 6 380 7 Machnery and transport equpment 8.23 87 231 0.56 15 474 4 705 6 998 8 Mscellaneous manufactured artcles 7.04 60 941 0.46 11 818 3 607 5 814 9 Commod. & transacts. not class. acc 7.60 75 250 0.79 18 288 7 253 6 115 The revealed captal ntensty (RCI) ndces and the revealed human captal ntensty (RHCI) Indces appear hghly correlated. 23

Factor ntensty rankngs are reported n table 20 (a-f) for all 10 ndustres and three factors (also for factors that are calculated usng the World Bank data on natural captal). Resultng rankngs are plausble. For nstance, machnery and transport equpment or chemcals are revealed as ntensve n captal and human captal. By contrast, food and lve anmals, anmal and vegetable ols and fats, or crude materals have the lowest RFI ndces for captal and human captal, but rank near the top n terms of land ntensty. Table 20. Rankng of ndustres n terms of revealed factor ntensty ndces a. RHCI b. RCI Rank SITC 1 dgt descrpton RHCI Rank SITC 1 dgt descrpton RCI 1 Machnery and transport equpment 8.23 1 Machnery and transport equpment 87 231 2 Chemcals 7.66 2 Commod. & transacts. not class. acc 75 250 3 Commod. & transacts. not class. acc 7.60 3 Chemcals 72 169 4 Manufact goods classfed chefly 7.06 4 Manufact goods classfed chefly 62 059 5 Mscellaneous manufactured artcles 7.04 5 Mscellaneous manufactured artcles 60 941 6 Beverages and tobacco 6.95 6 Beverages and tobacco 52 538 7 Mneral fuels, lubrcants 6.94 7 Mneral fuels, lubrcants 47 869 8 Crude materals, nedble 6.37 8 Crude materals, nedble 42 159 9 Food and lve anmals 6.27 9 Food and lve anmals 39 067 10 Anmal and vegetable ols and fats 5.67 10 Anmal and vegetable ols and fats 34 756 c. RNRI (Arable Land) d. RNRI (Total Natural Captal) Rank SITC 1 dgt descrpton RNRI_ land Rank SITC 1 dgt descrpton RNRI_nc 1 Commod. & transacts. not class. acc 0.79 1 Mneral fuels, lubrcants 20 925 2 Food and lve anmals 0.79 2 Commod. & transacts. not class. acc 18 288 3 Anmal and vegetable ols and fats 0.74 3 Chemcals 16 641 4 Crude materals, nedble 0.74 4 Crude materals, nedble 16 382 5 Mneral fuels, lubrcants 0.69 5 Machnery and transport equpment 15 474 6 Beverages and tobacco 0.61 6 Food and lve anmals 15 428 7 Chemcals 0.59 7 Beverages and tobacco 15 070 8 Machnery and transport equpment 0.56 8 Manufact goods classfed chefly 13 667 9 Manufact goods classfed chefly 0.55 9 Anmal and vegetable ols and fats 12 748 10 Mscellaneous manufactured artcles 0.46 10 Mscellaneous manufactured artcles 11 818 e. TNRI (Subsol Assets) f. RNRI (Pastureland and Cropland) Rank SITC 1 dgt descrpton RNRI_sa Rank SITC 1 dgt descrpton RNRI_pc 1 Mneral fuels, lubrcants 12 070 1 Food and lve anmals 8 314 2 Commod. & transacts. not class. acc 7 253 2 Beverages and tobacco 7 704 3 Chemcals 6 119 3 Crude materals, nedble 7 640 4 Crude materals, nedble 5 256 4 Chemcals 7 354 5 Machnery and transport equpment 4 705 5 Machnery and transport equpment 6 998 6 Beverages and tobacco 4 614 6 Anmal and vegetable ols and fats 6 709 7 Food and lve anmals 4 422 7 Manufact goods classfed chefly 6 380 8 Manufact goods classfed chefly 4 342 8 Mneral fuels, lubrcants 6 187 9 Anmal and vegetable ols and fats 3 795 9 Commod. & transacts. not class. acc 6 115 10 Mscellaneous manufactured artcles 3 607 10 Mscellaneous manufactured artcles 5 814 24

We now turn to cluster analyss to explore whether ndustres can be clustered nto naturally homogenous groups n terms of the RCI and RHCI ndces,.e. factor ntensty, usng the same algorthms as n the prevous secton. Fgure 5 shows that Ward s dendrogram gves sx welldentfed clusters of products at the fnest dsaggregated level of the SITC Rev 1 (4-5 dgts). Fgure 5. Dendrogram of Ward s cluster analyss: ndustral sectors L2squared dssmlarty measure 0 1000 2000 3000 Dendrogram for L2wlnk cluster analyss G1 G2 G3 G4 G5 G6 G7 G8 G9 G10 G11 G12 G13 G14 G15 G16 n=76n=59n=67n=119n=37n=66n=98n=72n=86n=52n=39n=67n=94n=51n=90n=93 The number of clusters s valdated by the stoppng-rule results shown n table 21. The sxcluster soluton s the most favourable under Calnsk and Harabasz Pseudo-F ndces, and to a lesser extent under Duda and Hart ndces. Table 21. Stoppng rule result Calnsk & Harabasz Duda & Hart Numbers of Clusters Pseudo-F Numbers of Clusters Je(2)/Je(1) Pseudo T-squared 2 1 712 2 0.39 1 144 3 2 308 3 0.45 630 4 2 137 4 0.49 456 5 2 195 5 0.35 388 6 2 374 6 0.69 142 7 2 186 7 0.71 83 8 2 053 8 0.40 244 9 1 964 9 0.54 138 10 1 919 10 0.65 105 11 1 880 11 0.62 96 12 1 864 12 0.70 99 13 1 840 13 0.60 101 14 1 812 14 0.52 127 15 1 799 15 0.58 89 Table 22 shows summary statstcs for the clusters ust dentfed. They are ordered from the least ntensve n captal and human captal (cluster 1), to the most ntensve n both captal and human captal (cluster 6). 25

Table 22. Summary statstcs of the clusters Cluster 1 Cluster 2 Cluster 3 Cluster 4 Cluster 5 Cluster 6 Revealed Captal Intensty Index 10 783 23 386 39 601 58 076 81 636 102 341 Revealed Human Captal Intensty Index 3.29 5.20 6.37 7.38 8.16 9.05 Number of Goods 72 167 224 273 237 193 Tables 23 and 24 show the ndustry composton of the clusters at the SITC-1 and SITC-2 levels respectvely. Two ndustres account for 50 per cent or more for all clusters except cluster 3 (49 per cent), wth the hghest proporton accounted for by the top 2 ndustres n clusters 4, 5 and 6 (54 per cent n each). Table 23 a. Percentages of SITC 1 dgt ndustres, by clusters SITC sectors at 1 dgt Cluster 1 Cluster 2 Cluster 3 Cluster 4 Cluster 5 Cluster 6 0 Food and lve anmals 26 17 18 10 3 4 1 Beverages and tobacco 1 1 0 2 1 0 2 Crude materals, nedble 31 26 17 12 5 7 3 Mneral fuels, lubrcants 0 1 5 3 0 1 4 Anmal and vegetable ols and fats 8 4 3 1 1 0 5 Chemcals 4 11 8 13 25 18 6 Manufact goods classfed chefly 17 25 31 36 29 26 7 Machnery and transport equpment 0 1 4 7 22 28 8 Mscellaneous manufactured artcles 10 15 14 18 13 14 9 Commod. & transacts. not class. acc 3 0 0 0 1 2 100 100 100 100 100 100 b. Percentages of clusters, by SITC 1 dgt ndustres SITC sectors at 1 dgt Cluster 1 Cluster 2 Cluster 3 Cluster 4 Cluster 5 Cluster 6 0 Food and lve anmals 15 22 31 20 6 6 100 1 Beverages and tobacco 8 8 8 50 25 0 100 2 Crude materals, nedble 14 27 23 20 7 9 100 3 Mneral fuels, lubrcants 0 5 55 32 0 9 100 4 Anmal and vegetable ols and fats 25 29 25 13 8 0 100 5 Chemcals 2 11 10 21 35 21 100 6 Manufact goods classfed chefly 4 12 21 28 20 15 100 7 Machnery and transport equpment 0 1 7 14 39 39 100 8 Mscellaneous manufactured artcles 4 15 18 29 18 16 100 9 Commod. & transacts. not class. acc 25 0 0 0 38 38 100 Examnaton of tables 23 and 24 (partcularly the latter) shows that cluster composton s far from perfectly overlappng wth ndustry composton. Factor ntenstes vary substantally not ust between, but also wthn ndustres, and ths pattern remans at all levels of dsaggregaton. For nstance, an ndustry sector, SITC 65 (textle yarns, fabrcs, made-up artcles) covers a wde varety of goods whose factor contents vary from the least human/physcal captal ntensve to the most captal ntensve (table 24b). Ths suggests that analyses of the factor content of trade, whether motvated by the emprcal valdaton of trade models or by polcy advce, should best be carred out at hgh degrees of dsaggregaton. 26

SITC 1 dgt descrpton SITC 2 dgt Table 24 a. Percentages of SITC 2 dgt ndustres, by clusters SITC 2 dgt descrpton Cluster 1 Cluster 2 Cluster 3 Cluster 4 Cluster 5 Cluster 6 0 Lve anmals 1.4 0.0 1.3 0.0 0.0 0.5 1 Meat and meat preparatons 1.4 0.6 2.2 1.5 0.4 0.5 2 Dary products and eggs 0.0 0.6 0.9 1.1 0.0 0.0 3 Fsh and fsh preparatons 1.4 0.6 0.4 0.7 0.0 0.0 0 Food and lve anmals 4 Cereals and cereal preparatons 1.4 3.0 1.8 1.5 1.3 1.0 5 Frut and vegetables 9.7 4.2 7.1 2.2 0.4 1.0 6 Sugar, sugar preparatons and honey 1.4 0.6 0.9 0.4 0.0 0.0 7 Coffee, tea, cocoa, spces 8.3 4.2 0.9 0.0 0.4 0.0 8 Feed.-stuff for anmals 1.4 1.2 0.4 1.5 0.4 0.5 9 Mscellaneous food preparatons 0.0 1.8 1.8 0.7 0.0 0.0 1 Beverages and 11 Beverages 0.0 0.0 0.4 1.5 1.3 0.0 tobacco 12 Tobacco and tobacco manufactures 1.4 0.6 0.0 0.7 0.0 0.0 21 Hdes, skns and furskns, raw 2.8 1.2 0.4 0.7 0.0 0.5 22 Ol-seeds, ol nuts and ol kernels 5.6 2.4 0.0 0.4 0.0 0.0 23 Crude rubber 1.4 0.0 0.0 0.4 0.8 0.0 24 Wood, lumber and cork 4.2 1.2 1.3 0.4 0.4 1.0 2 Crude materals, 25 Pulp and paper nedble, except f 0.0 0.6 0.9 1.5 0.0 1.0 26 Textle fbres, not manufactured 5.6 6.0 4.0 1.8 0.8 1.0 27 Crude fertlzers and crude mnerals 2.8 5.4 4.0 2.9 2.1 1.0 28 Metallferous ores and metal scrap 5.6 4.8 3.1 1.5 0.4 1.6 3 Mneral fuels, lubrcants and relat 4 Anmal and vegetable ols and fats 5 Chemcals 6 Manufact goods classfed chefly b 7 Machnery and transport equpment 8 Mscellaneous manufactured artcles 9 Commod. & transacts. not class. acc 29 Crude anmal and vegetable materals 2.8 4.2 2.7 2.2 0.0 1.0 32 Coal, coke and brquettes 0.0 0.0 0.9 0.7 0.0 0.5 33 Petroleum and petroleum products 0.0 0.6 3.1 1.8 0.0 0.5 34 Gas, natural and manufactured 0.0 0.0 0.9 0.0 0.0 0.0 35 Electrc energy 0.0 0.0 0.4 0.0 0.0 0.0 41 Anmal ols and fats 0.0 0.6 0.4 1.1 0.4 0.0 42 Fxed vegetable ols and fats 5.6 2.4 1.3 0.0 0.4 0.0 43 Anmal and vegetable ols and fats, 2.8 1.2 0.9 0.0 0.0 0.0 51 Organc chemcals 2.8 6.6 3.1 9.2 13.1 10.4 52 Inorganc chemcals 0.0 0.0 0.9 0.0 0.0 0.0 53 Dyeng, tannng and colourng materals 0.0 1.2 0.4 0.7 2.1 1.0 54 Medcnal and pharmaceutcal products 0.0 0.6 0.0 0.0 3.0 1.0 Perfume materals, tolet & cleansng 55 preparatons 1.4 0.6 0.9 0.7 0.0 0.0 56 Fertlzers, manufactured 0.0 1.8 0.9 0.0 0.0 0.0 57 Explosves and pyrotechnc products 0.0 0.0 0.4 0.4 0.4 0.5 58 Plastc materals, etc. 0.0 0.0 0.0 0.4 2.1 0.5 59 Chemcal materals and products 0.0 0.6 0.9 1.5 4.2 4.7 61 Leather and leather manufactures, nes 2.8 1.8 1.3 1.5 0.4 0.0 62 Rubber manufactures, nes 0.0 0.0 1.3 1.5 1.3 0.0 63 Wood and cork manufactures 1.4 1.2 2.7 1.5 0.8 0.5 64 Paper, paperboard and manufactures 0.0 0.0 0.4 1.8 3.0 3.1 65 Textle yarn, fabrcs, made-up artcles 9.7 12.6 7.6 8.1 4.6 1.6 66 Non-metallc mneral manufactures, nes 0.0 4.8 3.6 4.8 6.3 4.7 67 Iron and steel 0.0 1.2 3.6 5.5 4.2 5.7 68 Non-ferrous metals 2.8 1.2 5.4 3.7 3.0 6.7 69 Manufactures of metal, nes 0.0 2.4 5.4 7.3 5.5 4.1 71 Machnery, other than electrc 0.0 0.0 0.4 2.9 13.1 16.6 72 Electrcal machnery 0.0 0.0 1.8 2.9 5.1 5.7 73 Transport equpment 0.0 0.6 2.2 1.1 4.2 5.7 Santary, plumbng, heatng and lghtng 81 fxtures 0.0 0.0 0.9 0.7 0.8 0.0 82 Furnture 0.0 0.6 0.0 0.7 0.4 0.0 83 Travel goods and handbags 0.0 0.0 0.4 0.0 0.0 0.0 84 Clothng 4.2 7.8 1.3 1.8 0.0 0.0 85 Footwear 0.0 0.6 1.3 0.4 0.0 0.0 Scentf & control nstrum, photographc 86 apparatus 0.0 0.0 0.4 4.0 5.9 8.8 89 Mscellaneous manufactured artcles 5.6 6.0 9.4 9.9 5.5 5.2 94 Anmals, nes, ncl. zoo anmals 1.4 0.0 0.0 0.0 0.0 0.0 95 Frearms of war and ammunton 1.4 0.0 0.0 0.0 1.3 1.0 96 Con, other than gold con 0.0 0.0 0.0 0.0 0.0 0.5 100.0 100.0 100.0 100.0 100.0 100.0 27

SITC 1 dgt descrpton 0 Food and lve anmals SITC 2 dgt b. Percentages of clusters, by SITC 2 dgt ndustres SITC 2 dgt descrpton Cluster 1 Cluster 2 Cluster 3 Cluster 4 Cluster 5 Cluster 6 0 Lve anmals 20.0 0.0 60.0 0.0 0.0 20.0 100.0 1 Meat and meat preparatons 7.7 7.7 38.5 30.8 7.7 7.7 100.0 2 Dary products and eggs 0.0 16.7 33.3 50.0 0.0 0.0 100.0 3 Fsh and fsh preparatons 20.0 20.0 20.0 40.0 0.0 0.0 100.0 4 Cereals and cereal preparatons 5.3 26.3 21.1 21.1 15.8 10.5 100.0 5 Frut and vegetables 17.9 17.9 41.0 15.4 2.6 5.1 100.0 6 Sugar, sugar preparatons and honey 20.0 20.0 40.0 20.0 0.0 0.0 100.0 7 Coffee, tea, cocoa, spces 37.5 43.8 12.5 0.0 6.3 0.0 100.0 8 Feed.-stuff for anmals 10.0 20.0 10.0 40.0 10.0 10.0 100.0 9 Mscellaneous food preparatons 0.0 33.3 44.4 22.2 0.0 0.0 100.0 1 Beverages and 11 Beverages 0.0 0.0 12.5 50.0 37.5 0.0 100.0 tobacco 12 Tobacco and tobacco manufactures 25.0 25.0 0.0 50.0 0.0 0.0 100.0 21 Hdes, skns and furskns, raw 25.0 25.0 12.5 25.0 0.0 12.5 100.0 22 Ol-seeds, ol nuts and ol kernels 44.4 44.4 0.0 11.1 0.0 0.0 100.0 23 Crude rubber 25.0 0.0 0.0 25.0 50.0 0.0 100.0 24 Wood, lumber and cork 25.0 16.7 25.0 8.3 8.3 16.7 100.0 2 Crude materals, 25 Pulp and paper 0.0 11.1 22.2 44.4 0.0 22.2 100.0 nedble, except f 26 Textle fbres, not manufactured 12.5 31.3 28.1 15.6 6.3 6.3 100.0 27 Crude fertlzers and crude mnerals 5.7 25.7 25.7 22.9 14.3 5.7 100.0 28 Metallferous ores and metal scrap 14.8 29.6 25.9 14.8 3.7 11.1 100.0 3 Mneral fuels, lubrcants and relat 4 Anmal and vegetable ols and fats 5 Chemcals 6 Manufact goods classfed chefly b 7 Machnery and transport equpment 8 Mscellaneous manufactured artcles 29 Crude anmal and vegetable materals 8.7 30.4 26.1 26.1 0.0 8.7 100.0 32 Coal, coke and brquettes 0.0 0.0 40.0 40.0 0.0 20.0 100.0 33 Petroleum and petroleum products 0.0 7.1 50.0 35.7 0.0 7.1 100.0 34 Gas, natural and manufactured 0.0 0.0 100.0 0.0 0.0 0.0 100.0 35 Electrc energy 0.0 0.0 100.0 0.0 0.0 0.0 100.0 41 Anmal ols and fats 0.0 16.7 16.7 50.0 16.7 0.0 100.0 42 Fxed vegetable ols and fats 33.3 33.3 25.0 0.0 8.3 0.0 100.0 43 Anmal and vegetable ols and fats, 33.3 33.3 33.3 0.0 0.0 0.0 100.0 51 Organc chemcals 2.1 11.5 7.3 26.0 32.3 20.8 100.0 52 Inorganc chemcals 0.0 0.0 100.0 0.0 0.0 0.0 100.0 53 Dyeng, tannng and colourng materals 0.0 16.7 8.3 16.7 41.7 16.7 100.0 54 Medcnal and pharmaceutcal products 0.0 10.0 0.0 0.0 70.0 20.0 100.0 Perfume materals, tolet & cleansng 55 preparatons 16.7 16.7 33.3 33.3 0.0 0.0 100.0 56 Fertlzers, manufactured 0.0 60.0 40.0 0.0 0.0 0.0 100.0 57 Explosves and pyrotechnc products 0.0 0.0 25.0 25.0 25.0 25.0 100.0 58 Plastc materals, etc. 0.0 0.0 0.0 14.3 71.4 14.3 100.0 59 Chemcal materals and products 0.0 3.8 7.7 15.4 38.5 34.6 100.0 61 Leather and leather manufactures, nes 15.4 23.1 23.1 30.8 7.7 0.0 100.0 62 Rubber manufactures, nes 0.0 0.0 30.0 40.0 30.0 0.0 100.0 63 Wood and cork manufactures 6.3 12.5 37.5 25.0 12.5 6.3 100.0 64 Paper, paperboard and manufactures 0.0 0.0 5.3 26.3 36.8 31.6 100.0 65 Textle yarn, fabrcs, made-up artcles 8.6 25.9 21.0 27.2 13.6 3.7 100.0 66 Non-metallc mneral manufactures, nes 0.0 15.1 15.1 24.5 28.3 17.0 100.0 67 Iron and steel 0.0 4.3 17.4 32.6 21.7 23.9 100.0 68 Non-ferrous metals 4.3 4.3 26.1 21.7 15.2 28.3 100.0 69 Manufactures of metal, nes 0.0 7.0 21.1 35.1 22.8 14.0 100.0 71 Machnery, other than electrc 0.0 0.0 1.4 11.1 43.1 44.4 100.0 72 Electrcal machnery 0.0 0.0 11.4 22.9 34.3 31.4 100.0 73 Transport equpment 0.0 3.3 16.7 10.0 33.3 36.7 100.0 Santary, plumbng, heatng and lghtng 81 fxtures 0.0 0.0 33.3 33.3 33.3 0.0 100.0 82 Furnture 0.0 25.0 0.0 50.0 25.0 0.0 100.0 83 Travel goods and handbags 0.0 0.0 100.0 0.0 0.0 0.0 100.0 84 Clothng 12.5 54.2 12.5 20.8 0.0 0.0 100.0 85 Footwear 0.0 20.0 60.0 20.0 0.0 0.0 100.0 Scentf & control nstrum, photographc 86 apparatus 0.0 0.0 2.3 25.6 32.6 39.5 100.0 89 Mscellaneous manufactured artcles 4.7 11.8 24.7 31.8 15.3 11.8 100.0 94 Anmals, nes, ncl. zoo anmals 100.0 0.0 0.0 0.0 0.0 0.0 100.0 95 Frearms of war and ammunton 16.7 0.0 0.0 0.0 50.0 33.3 100.0 9 Commod. & transacts. not class. acc 96 Con, other than gold con 0.0 0.0 0.0 0.0 0.0 100.0 100.0 28

4. Concluson We constructed a database of revealed factor ntensty (RFI) ndces for each export good at a very detaled dsaggregaton level, usng data from up to 99 countres for the perod between 1961 and 2003. To calculate the ndces on factor endowments, we used (and n some cases updated or made estmates) data from dfferent sources: (a) Barro and Lee s dataset on educatonal achevements; (b) Easterly and Levne s estmates on natonal captal stock; and (c) the World Bank s World Development Indcators. Frst, we constructed two country-endowment datasets: a wde one wth the maxmum number of countres n each year, and a consstent one wth 92 countres wth full data over 33 years (1971 2003). Second, usng these data, we followed Hausmann, Hwang and Rodrk s (2007) methodology to construct the RFI ndces for all export goods at the fnest dsaggregaton level avalable n harmonzed trade data: SITC-5 and HS-6. For each good and relatve factor (captal/labour, human captal and land/labour), the RFI ndces s calculated as a weghted average of the relatve factor abundances of the countres exportng that good, usng slghtly modfed versons of Balassa s RCA ndces as weghts. Our RFI ndces allow us to systematcally classfy products accordng to ther factor ntenstes, at the most dsaggregated level of product classfcaton. Ths s an advantage over other ad hoc attempts, as the degree of factor ntensty can wdely vary wthn an ndustry, e.g. as classfed at the HS 2-dgt level. Thus, we beleve the RFI ndces generate a more economcally meanngful categorzaton of products that can be used for polcy advce as well as postve trade analyss. The RFI ndces could be used for many purposes. As mentoned n the ntroducton, t wll enable us to revst the ssue of export dversfcaton wth a more standard, theory-based approach,.e. takng nto account the effect of changes n relatve factor endowment, than the recent eclectc approaches wth nductve reasonng. For nstance, one could explore to whch extent export dversfcaton proceeds from changes n comparatve advantage. Ths could have nterestng mplcatons to polcymakers and export-promoton agences when they need to dentfy or prortze sectors for export dversfcaton. A recent study by Cadot, Carrère and Strauss-Kahn (verson March 2009) has used our RFI ndces to verfy a conecture that dversfcaton n mddle to hgh ncome countres may smply reflect a slow adustment to changes n ts comparatve advantage. The paper confrms a robust hump-shaped relatonshp between export dversfcaton and the level of ncome,.e. export dversfcaton contnues up untl a certan level of ncome, but then stops and moves to export specalzaton as ncome ncrease. Then, usng the RFI ndces, they were able to suggest the reason behnd the hump shape was because countres fal to close a tal of export lnes that no longer belong to ther comparatve advantage. Ther export bundles are therefore artfcally nflated. That s, the slow adustment of producton/export lnes may explan the hump-shaped relatonshp between dversfcaton and development. In addton, ssues that can be explored usng the RFI ndces would nclude: (a) how does the captal content of exports evolve wth ncome levels? (b) are there systematc devatons lnked e.g. to governance falures (an ant-captal bas)? and (c) does the factor content of trade vary wth ts destnaton (e.g. Southern countres could export more captal-ntensve goods to other Southern countres than to Northern ones)? 29

Appendx A. Appendx tables and fgures Table A1. Countres ncluded n the sample World Bank Country Code Country Name 1 ARG Argentna 2 AUS Australa 3 AUT Austra 4 BEN Benn 5 BHR Bahran 6 BOL Bolva 7 BRA Brazl 8 BRB Barbados 9 CAF Central Afrcan Republc 10 CAN Canada 11 CHE Swtzerland 12 CHL Chle 13 CHN Chna 14 CMR Cameroon 15 COG Congo 16 COL Colomba 17 CRI Costa Rca 18 CYP Cyprus 19 DEU Germany 20 DNK Denmark 21 DOM Domncan Republc 22 DZA Algera 23 ECU Ecuador 24 EGY Egypt 25 ESP Span 26 FIN Fnland 27 FJI F 28 FRA France 29 GBR Unted Kngdom 30 GHA Ghana 31 GMB Gamba, The 32 GRC Greece 33 GTM Guatemala 34 HND Honduras 35 HUN Hungary 36 IDN Indonesa 37 IND Inda 38 IRL Ireland 39 IRN Iran, Islamc Republc of 40 IRQ Iraq 41 ISL Iceland 42 ISR Israel 43 ITA Italy 44 JAM Jamaca 45 JOR Jordan 46 JPN Japan 47 KEN Kenya 31

World Bank Country Code Country Name 48 KOR Korea, Republc of 49 LBR Lbera 50 LKA Sr Lanka 51 MEX Mexco 52 MLI Mal 53 MLT Malta 54 MOZ Mozambque 55 MUS Maurtus 56 MWI Malaw 57 MYS Malaysa 58 NER Nger 59 NIC Ncaragua 60 NLD Netherlands 61 NOR Norway 62 NPL Nepal 63 NZL New Zealand 64 PAK Pakstan 65 PAN Panama 66 PER Peru 67 PHL Phlppnes 68 PNG Papua New Gunea 69 POL Poland 70 PRT Portugal 71 PRY Paraguay 72 RWA Rwanda 73 SDN Sudan 74 SEN Senegal 75 SGP Sngapore 76 SLE Serra Leone 77 SLV El Salvador 78 SWE Sweden 79 SYR Syran Arab Republc 80 TGO Togo 81 THA Thaland 82 TTO Trndad and Tobago 83 TUN Tunsa 84 TUR Turkey 85 UGA Uganda 86 URY Uruguay 87 USA Unted States 88 VEN Venezuela, Bolvaran Rep. of 89 ZAF South Afrca 90 ZAR Congo, Dem. Rep. of 91 ZMB Zamba 92 ZWE Zmbabwe 32

Fgure A1. Dendrogram of Ward s cluster (natural resource excluded from the varables) L2squared dssmlarty measure 0 50 100 150 200 250 Dendrogram for L2wlnk cluster analyss 19364987113556789245968922869263385303921121804365903620274729424477126347810743264750705460561588937683665831483415915357178145468418735172386407261422945282961871637795253538 Table A2. Calnsk and Harabasz and Duda and Hart stoppng rules result (arable land per capta excluded from the varable lsts) Calnsk & Harabasz Duda & Hart Numbers of Clusters Pseudo-F Numbers of Clusters Je(2)/Je(1) Pseudo T-squared 2 166.93 2 0.34 70.42 3 201.84 3 0.39 83.75 4 207.08 4 0.52 17.82 5 188.53 5 0.57 12.11 6 188.06 6 0.47 10.33 7 182.31 7 0.42 19.40 8 182.86 8 0.64 17.14 9 191.07 9 0.50 14.88 10 200.20 10 0.35 12.96 11 209.29 11 0.28 54.34 12 223.14 12 0.36 17.87 13 227.71 13 0.31 6.53 14 229.34 14 0.33 8.09 15 233.77 15 0.38 8.18 33

Table A3. Summary of clusters (arable land per capta excluded from the varable lsts) Countres n Cluster Number of countres Captal Index Human Captal Index World Bank Income Group World Bank Regonal Group Benn Low ncome Sub-Saharan Afrca Central Afrcan Republc Low ncome Sub-Saharan Afrca Cameroon Lower mddle ncome Sub-Saharan Afrca Gamba, The Low ncome Sub-Saharan Afrca Guatemala Lower mddle ncome Latn Amerca & Carbbean Iraq Lower mddle ncome Mddle East & North Afrca Kenya Low ncome Sub-Saharan Afrca Lbera Low ncome Sub-Saharan Afrca Mal Low ncome Sub-Saharan Afrca Mozambque Low ncome Sub-Saharan Afrca Cluster 1 Malaw 22 2 632 1.65 Low ncome Sub-Saharan Afrca Nger Low ncome Sub-Saharan Afrca Nepal Low ncome South Asa Pakstan Low ncome South Asa Papua New Gunea Low ncome East Asa & Pacfc Rwanda Low ncome Sub-Saharan Afrca Sudan Lower mddle ncome Sub-Saharan Afrca Senegal Low ncome Sub-Saharan Afrca Serra Leone Low ncome Sub-Saharan Afrca Togo Low ncome Sub-Saharan Afrca Uganda Low ncome Sub-Saharan Afrca Dem. Rep. of the Congo Low ncome Sub-Saharan Afrca Australa Hgh ncome: OECD.. Austra Hgh ncome: OECD.. Canada Hgh ncome: OECD.. Swtzerland Hgh ncome: OECD.. Germany Hgh ncome: OECD.. Denmark Hgh ncome: OECD.. Fnland Hgh ncome: OECD.. France Hgh ncome: OECD.. Unted Kngdom Hgh ncome: OECD.. Cluster 2 Ireland 20 98 827 8.86 Hgh ncome: OECD.. Iceland Hgh ncome: OECD.. Israel Hgh ncome: non-oecd.. Italy Hgh ncome: OECD.. Japan Hgh ncome: OECD.. Netherlands Hgh ncome: OECD.. Norway Hgh ncome: OECD.. New Zealand Hgh ncome: OECD.. Sngapore Hgh ncome: non-oecd.. Sweden Hgh ncome: OECD.. Unted States Hgh ncome: OECD.. Bahran Hgh ncome: non-oecd.. Bolva Lower mddle ncome Latn Amerca & Carbbean Brazl Upper mddle ncome Latn Amerca & Carbbean Botswana Upper mddle ncome Sub-Saharan Afrca Chna Lower mddle ncome East Asa & Pacfc Congo, Rep. Lower mddle ncome Sub-Saharan Afrca Colomba Lower mddle ncome Latn Amerca & Carbbean Costa Rca Upper mddle ncome Latn Amerca & Carbbean Domncan Republc Lower mddle ncome Latn Amerca & Carbbean Algera Lower mddle ncome Mddle East & North Afrca Egypt, Arab Rep. Lower mddle ncome Mddle East & North Afrca Ghana Low ncome Sub-Saharan Afrca Honduras Lower mddle ncome Latn Amerca & Carbbean 34

Countres n Cluster Number of countres Captal Index Human Captal Index World Bank Income Group World Bank Regonal Group Indonesa Lower mddle ncome East Asa & Pacfc Inda Lower mddle ncome South Asa Iran, Islamc Rep. of Lower mddle ncome Mddle East & North Afrca Cluster 3 Jamaca 32 15 043 3.90 Upper mddle ncome Latn Amerca & Carbbean Jordan Lower mddle ncome Mddle East & North Afrca Sr Lanka Lower mddle ncome South Asa Lesotho Lower mddle ncome Sub-Saharan Afrca Maurtus Upper mddle ncome Sub-Saharan Afrca Ncaragua Lower mddle ncome Latn Amerca & Carbbean Portugal Hgh ncome: OECD.. Paraguay Lower mddle ncome Latn Amerca & Carbbean El Salvador Lower mddle ncome Latn Amerca & Carbbean Swazland Lower mddle ncome Sub-Saharan Afrca Syran Arab Republc Lower mddle ncome Mddle East & North Afrca Thaland Lower mddle ncome East Asa & Pacfc Tunsa Lower mddle ncome Mddle East & North Afrca Turkey Upper mddle ncome Europe and Central Asa Zamba Low ncome Sub-Saharan Afrca Zmbabwe Low ncome Sub-Saharan Afrca Argentna Upper mddle ncome Latn Amerca & Carbbean Barbados Hgh ncome: non-oecd.. Chle Upper mddle ncome Latn Amerca & Carbbean Cyprus Hgh ncome: non-oecd.. Ecuador Lower mddle ncome Latn Amerca & Carbbean Span Hgh ncome: OECD.. F Upper mddle ncome East Asa & Pacfc Greece Hgh ncome: OECD.. Hungary Hgh ncome: OECD.. Korea, Rep. Hgh ncome: OECD.. Cluster 4 Mexco 21 35 393 6.70 Upper mddle ncome Latn Amerca & Carbbean Malta Hgh ncome: non-oecd.. Malaysa Upper mddle ncome East Asa & Pacfc Panama Upper mddle ncome Latn Amerca & Carbbean Peru Lower mddle ncome Latn Amerca & Carbbean Phlppnes Lower mddle ncome East Asa & Pacfc Poland Upper mddle ncome Europe and Central Asa Trndad and Tobago Hgh ncome: non-oecd.. Uruguay Upper mddle ncome Latn Amerca & Carbbean Venezuela (Bolvaran Rep. of) Upper mddle ncome Latn Amerca & Carbbean South Afrca Upper mddle ncome Sub-Saharan Afrca 35

Table A4. Summary statstcs of revealed factor ntensty ndces, year 2000 (HS classfcaton) Varable Obs Mean Std. Dev. Mn Max rhc 5009 7.32 1.68 0.79 11.57 rc 5009 66 948 33 406 1 407 165 297 rnr_land 5009 0.57 0.32 0.10 4.84 rnr_nc 5009 14 380 8 823 1 859 89 591 rnr_sa 5009 4 679 5 858 7 74 197 rnr_pc 5009 6 744 3 549 854 46 780 Fgure A2. Dendrogram of Ward s cluster analyss (HS classfcaton) L2squared dssmlarty measure 0 5000 10000 15000 Dendrogram for L2wlnk cluster analyss G1 n=218 n=139 G2 n=705 G3 n=312 G4 n=659 G5 n=454 G6 n=354 G7 n=627 G8 n=701 G9 n=151 G10 n=506 G11 n=183 G12 Table A5. Calnsk and Harabasz and Duda and Hart stoppng rules result (HS classfcaton) Calnsk & Harabasz Duda & Hart Numbers of Clusters Pseudo-F Numbers of Clusters Je(2)/Je(1) Pseudo T-squared 2 8 881 2 0 3 903 3 9 068 3 0 2 821 4 9 395 4 1 1 500 5 8 807 5 0 1 068 6 8 658 6 1 709 7 8 493 7 1 673 8 8 118 8 1 555 9 7 698 9 1 606 10 7 443 10 1 442 11 7 328 11 1 353 12 7 336 12 1 489 13 7 191 13 1 630 14 7 085 14 1 540 15 7 044 15 1 296 36

Table A6. Summary statstcs of the clusters (HS classfcaton) Cluster 1 Cluster 2 Cluster 3 Cluster 4 Revealed Captal Intensty Index 18 619 42 873 72 971 105 368 Revealed Human Captal Intensty Index 4.45 6.51 7.91 9.04 Number of Goods 769 1 333 1 559 1 348 Table A7. Percentages of HS sectons ndustres, by clusters (HS classfcaton) HS sectons Secton Descrptons Cluster 1 Cluster 2 Cluster 3 Cluster 4 1 Lve anmals 4.7 4.7 3.8 2.7 2 Veg. 13.5 6.1 3.6 2.1 3 Fats and Ols 2.6 1.5 0.6 0.2 4 Bev. & Tobac. 4.6 5.6 3.6 1.1 5 Mneral 4.4 4.9 2.2 1.3 6 Chemcal 5.9 9.6 17.5 22.9 7 Plastcs 0.7 2.9 5.4 4.5 8 Leather 3.1 2.3 0.8 0.5 9 Wood 2.9 2.1 1.2 0.6 10 Paper 0.7 2.5 3.7 4.0 11 Textle 42.0 22.7 9.7 2.4 12 Footwear 1.6 2.9 0.3 0.0 13 Stone & Glass 1.0 3.9 2.6 2.8 14 Precuos Stones 0.9 1.5 0.8 0.8 15 Base Metal 6.1 11.4 13.3 13.4 16 Machnery 2.6 6.6 18.8 26.8 17 Trans.Eq 0.8 1.7 3.5 3.6 18 Optcal 0.4 3.0 4.7 8.4 19 Arms 0.1 0.1 0.5 0.5 20 Msc. 1.3 4.1 3.3 1.2 21 Works of Arts 0.3 0.2 0.2 0.0 100.0 100.0 100.0 100.0 37

Table A8. Percentages of HS chapters ndustres, by clusters (HS classfcaton) Sectons Chapters Chapter Descrptons Cluster 1 Cluster 2 Cluster 3 Cluster 4 1 Lve anmals 2 Veg. 1 Lve anmals 0 0 0 0 2 Meat and edble meat offal 1 1 1 1 3 Fsh and crustaceans, molluscs 3 2 1 2 4 Dary produce; brds eggs; natural honey 0 1 1 0 5 Products of anmal orgn 0 1 0 0 6 Lve trees and other plants 1 0 0 0 7 Edble vegetables 2 2 1 0 8 Edble fruts and nuts 2 1 1 0 9 Coffee, tea, maté and spces 3 1 0 0 10 Cereals 1 0 0 0 11 Products of mllng ndustry 1 1 0 1 12 Ol seeds and oleagnous fruts 2 1 1 1 13 Lac; gums, resns 1 0 0 0 14 Veg. plantng materals 1 0 0 0 3 Fats and Ols 15 Anmal or vegetable fats and ols 3 2 1 0 4 Bev. & Tobac. 5 Mneral 6 Chemcal 7 Plastcs 8 Leather 9 Wood 10 Paper 11 Textle 16 Preparatons of meat, of fsh 0 1 0 0 17 Sugars and sugar confectonery 0 1 0 0 18 Cocoa and cocoa preparatons 1 0 0 0 19 Prep. of cereals, flour 0 0 1 0 20 Prep. of vegetables, fruts, nuts 1 2 1 0 21 Msc. edble preparatons 0 0 1 0 22 Beverages, sprts and vnegar 0 1 1 0 23 Waste from food ndustres 1 1 0 0 24 Tobacco 1 0 0 0 25 Salt, sulfur, earths and stone 3 2 1 1 26 Ores, slag and ash 1 1 1 0 27 Mneral fuels, mneral ols 1 2 0 0 28 Inorganc chemcals 2 3 4 4 29 Organc chemcals 1 3 6 12 30 Pharmaceutcal products 0 0 1 1 31 Fertlzers 1 1 0 0 32 Tannng and dyeng extracts 0 1 1 1 33 Essental ols and resnods 1 1 1 0 34 Soap, organc surface-actve agents 1 0 1 0 35 Albumnodal substances 0 0 0 0 36 Explosves; pyrotechnc products 0 0 0 0 37 Photographc and cnematographc goods 0 0 1 2 38 Mscellaneous chemcal products 0 1 1 2 39 Plastcs and artcles thereof 0 1 4 3 40 Rubber and artcles thereof 1 2 1 1 41 Raw hdes and skns 2 1 0 0 42 Leader of leather 1 1 0 0 43 Furskns and artfcal fur 0 0 1 1 44 Wood and artcles of wood; wood charcoal 2 1 1 1 45 Cork and artcles of cork 0 1 0 0 46 Manufactures of straw, of esparto 1 0 0 0 47 Pulp of wood 0 1 0 0 48 Paper and paperboard 0 2 3 3 49 Prnted books, newspapers, pctures 0 0 1 0 50 Slk 0 1 0 0 51 Wool, fne or coarse anmal har 0 1 1 0 52 Cotton 10 3 1 0 53 Other vegetable textle fbers 2 1 0 0 54 Man-made flaments 1 2 1 0 38

Sectons Chapters Chapter Descrptons Cluster 1 Cluster 2 Cluster 3 Cluster 4 55 Man-made staple fbers 4 3 2 1 56 Waddng, felt and nonwovens 1 1 1 0 57 Carpets 1 1 0 0 58 Specal woven fabrcs 1 1 1 0 59 Impregnated, coated, covered textle fabrcs 0 0 1 1 60 Kntted and crocheted fabrcs 0 1 1 0 61 Apparel and clothng 7 4 0 0 62 Apparel and clothng, not kntted and crocheted 9 4 0 0 63 Other textles 5 1 0 0 64 Footwear 1 2 0 0 12 Footwear 65 Headgear 0 0 0 0 66 Umbrellas 0 1 0 0 67 Prepared feathers 1 0 0 0 68 Stone, plaster, cement 1 2 1 1 13 Stone & Glass 69 Ceramc 0 1 0 1 70 Glass 0 1 2 1 14 Precous Stones 71 Precous stones 1 2 1 1 15 Base Metal 72 Iron and steel 2 4 4 5 73 Artcles of ron or steel 1 2 3 2 74 Copper 1 1 2 1 75 Nckel 0 0 0 1 76 Alumnum 0 1 1 0 78 Lead 0 0 0 0 79 Znc 0 0 0 0 80 Tn 0 0 0 0 81 Other base metals 0 0 0 2 82 Tools, mplements, cutlery 1 2 1 1 83 Msc. artcles of base metal 1 0 1 0 16 Machnery 84 Nuclear reactors 2 3 10 21 85 Electrcal MAshnery 0 3 9 5 86 Ralway and Tramway 0 0 1 1 17 Trans.Eq 87 Vehcles other than ralway and tramway 0 1 3 1 88 Arcraft, spacecraft 0 0 0 1 89 Shps and boats 0 0 0 1 90 Optcal 0 1 3 6 18 Optcal 91 Clocks and watches 0 1 1 2 92 Muscal nstruments 0 1 1 0 19 Arms 93 Arms 0 0 1 1 94 Furnture 1 1 1 0 20 Msc. 95 Toys and games 0 2 1 1 96 Msc. manu.artcles 0 2 1 0 21 Works of Arts 97 Works of art 0 0 0 0 100 100 100 100 39

Appendx B. Dataset of ndex of revealed factor ntensty The dataset that contans the revealed factor ntensty (RFI) ndces s avalable on the UNCTAD webste (http://r0.unctad.org/dtc/tab/ndex.shtm). The dataset s avalable both n Stata and Excel. For each format, the dataset conssts of three folders, ttled (a) SITC; () HS; and (c) RFII_1994and2000. Folders SITC and HS The SITC folder contans the RFI ndces for each good calculated at the fnest dsaggregated level of the SITC. Though the fnest level of the SITC s a 5-dgt level, not every 4- dgt level s dvded nto 5 dgts. Therefore our data conssts of a mx of four and fve dgts. The ndces are gven n a separate fle for each year (called year_stc_ndces.dta f n Stata, and year_stc_ndces.xml). The year coverage s from 1971 to 2003. The HS folder contans the RFI ndces for products classfed at the HS 6-dgt level. The year coverage s from 1988 to 2003. For those ndces calculated from a wde or unbalanced dataset, the flename s unb_year_stc_ndces.dta. Table B.1 (a and b) provde the descrpton of all the varables ncluded n these SITC and HS folders. Table B1 a. Revealed factor ntensty ndces a. SITC (fle name = year _stc_ndces.dta or year _stc_ndces.xml) Varable name Varable descrpton 1 product SITC code at ether 4 or 5 dgt level 2 productname Correspondng product descrpton 3 dgt Number of dgt (ether 4 of 5) 4 stc45 SITC code at ether 4 or 5 dgt level, n strng form 5 stc1 Correspondng SITC 1 dgt code 6 stc1_desc SITC 1 dgt product descrpton 7 stc2 Correspondng SITC 2 dgt code 8 stc2_desc SITC 2 dgt product descrpton 9 stc3 Correspondng SITC 3 dgt code 10 stc3_desc SITC 3 dgt product descrpton 11 Export World Export of the product 12 Import World Import of the product 13 rnc Revealed Human Captal Intesty Index 14 rc Revealed Psyccal Captal Intesty Index 15 rnr_land Revealed Natural Resource Intensty Index 16 percentage Percentage of excluded exports, due to the lack of data, n total exports 40

Table B1 b. Revealed factor ntensty ndces database b. HS (fle name = hs_ year _ndces.dta or hs_ year _ndces.xml) Varable name Varable descrpton 1 year year 2 product HS code at sx dgt level 3 h0productname Correspondng product descrpton 4 sect Correspondng HS secton code 5 sect_desc HS secton descrpton 6 hs2 Correspondng HS 2 dgt code 7 Export World Export of the product 8 Import World Import of the product 9 rhc Revealed Human Captal Intesty Index 10 rc Revealed Psyccal Captal Intesty Index 11 rnr_land Revealed Natural Resource Intensty Index 12 percentage Percentage of excluded exports, due to the lack of data, n total exports Folder-RFI ndces for the years 1994 and 2000 The RFI ndces calculated usng addtonal data on natural resources from the World Bank are gven n a folder called RFII_1994and2000. The World Bank data s avalable only for the years 1994 and 2000. The lst of the varables and ther descrptons are gven n Table B.2 (a-b). a. SITC Table B2. Revealed factor ntensty ndces database (1994, 2000) Varable name Varable descrpton 1 product SITC code at ether 4 or 5 dgt level 2 productname Correspondng product descrpton 3 dgt Number of dgt (ether 4 of 5) 4 stc45 SITC code at ether 4 or 5 dgt level, n strng form 5 stc1 Correspondng SITC 1 dgt code 6 stc1_desc SITC 1 dgt product descrpton 7 stc2 Correspondng SITC 2 dgt code 8 stc2_desc SITC 2 dgt product descrpton 9 stc3 Correspondng SITC 3 dgt code 10 stc3_desc SITC 3 dgt product descrpton 11 Export World Export of the product 12 Import World Import of the product 13 rnc Revealed Human Captal Intesty Index 14 rc Revealed Psyccal Captal Intesty Index 15 rnr_land Revealed Natural Resource Intensty Index (Arable Land) 16 rnr_nc Revealed Natural Resource Intensty Index (Total Natural Captal) 17 rnr_sa Revealed Natural Resource Intensty Index (Subsol Assets) 18 rnr_pc Revealed Natural Resource Intensty Index (Pastureland and Cropland) (fle name = 1994 _stc_ndces.dta or 1994_stc_ndces.xml) (fle name = 2000 _stc_ndces.dta or 2000_stc_ndces.xml) 41

b. HS (fle name = hs_1994_ndces.dta or hs_1994_ndces.xml) Varable name Varable descrpton 1 year year 2 product HS code at sx dgt level 3 h0productname Correspondng product descrpton 4 sect Correspondng HS secton code 5 sect_desc HS secton descrpton 6 hs2 Correspondng HS 2 dgt code 7 Export World Export of the product 8 Import World Import of the product 9 rnc Revealed Human Captal Intesty Index 10 rc Revealed Psyccal Captal Intesty Index 11 rnr_land Revealed Natural Resource Intensty Index (Arable Land) 12 rnr_nc Revealed Natural Resource Intensty Index (Total Natural Captal) 13 rnr_sa Revealed Natural Resource Intensty Index (Subsol Assets) 14 rnr_pc Revealed Natural Resource Intensty Index (Pastureland and Cropland) (fle name = hs_1994_ndces.dta or hs_1994_ndces.xml) (fle name = hs_2000_ndces.dta or hs_2000_ndces.xml) Fles country endowments In addton, we have attached our newly constructed database on countres endowments (called endowments_all.dta and endowments1994_2000.dta ; the same fle name for Excel). Table B.3 (a-b) presents the lst of varables and the descrptons. Table B3 a. Endowment database, by country Varable name Varable descrpton 1 socode PWT 6.2: Country Code 2 countryname Country name 3 year Year 4 phys_cap_pw Physcal Captal Stock per Worker 5 hum_cap Average Years of Schoolng 6 land_pw Arable Land hectares per worker 7 workers Number of Workers 8 phys_cap Physcal Captal Stock 9 land Arable Land hectares 10 regon World Bank Regon Classfcaton 11 ncome World Bank Income Classfcaton 12 group World Bank Income Classfcaton 42

b. Endowment database, by country (1994 and 2000) (wth data on natural captal and ts components from the World Bank) Varable name Varable descrpton 1 socode PWT 6.2: Country Code 2 countryname Country name 3 year Year 4 workers Number of workers 5 phys_cap_pw Physcal Captal Stock per Worker 6 hum_cap Average Years of Schoolng 7 land_pw Arable Land hectares per worker 8 nc Natural Captal, $ per worker 9 sa Subsol Assets, $ per worker 10 tr Tmber Resources, $ per worker 11 ntr Non Tmber Resources, $ per worker 12 pa Protected Areas, $ per worker 13 p Pastureland, $ per worker 14 c Cropland, $ per worker 15 pc Pastureland and Cropland, $ per worker 43

References Anderson K (2008). Dstorted agrcultural ncentves and economc development: Asa s experence. CEPR DP 6914. Anderson K, Kurzwel M, Martn W, Sandr D and Valenzuela E (2008). Measurng dstortons to agrcultural ncentves, revsted. World Trade Revew. 7(4): 675 704. Balassa B (1965). Trade lberalzaton and revealed comparatve advantage. The Manchester School of Economc and Socal Studes. 33: 99 123. Barro RJ and Lee JW (1993). Internatonal comparsons of educatonal attanment. Journal of Monetary Economcs. 32: 5 23. Barro RJ and Lee JW (2000). Internatonal data on educatonal attanment: updates and mplcatons. CID Workng Paper. No. 42. Harvard Unversty. Barro RJ and Lee JW (2001). Internatonal data on educatonal attanment: updates and mplcatons. Oxford Economc Papers. 53(3): 541 563. Carrère C, Strauss-Kahn V and Cadot O (2007). Export dversfcaton: what s behnd the hump? Mmeo, Unversty of Lausanne. Cohen D and Soto M (2001). Growth and human captal: good data, good results. OECD Development Centre Workng Papers. 179, OECD Development Centre. De La Fuente A and Doménech R (2001). Educatonal attanment n the OECD, 1960-1995. CEPR Dscusson Paper, 3416. Easterly W, Kremer M, Prtchett L and Summers LH (1993). Good polcy or good luck: country growth performance and temporary shocks. Journal of Monetary Economcs, 32(3), 459-483. Easterly W and Levne R (2001). It s not factor accumulaton: stylzed facts and growth models. The World Bank Economc Revew. 15(2): 177 219. Feenstra R (2004). Advanced Internatonal Trade. MIT Press. Hausmann R, Hwang J and Rodrk D (2007). What you export matters. Journal of Economc Growth. 12: 1 25. Hausmann R and Klnger B (2006). Structural transformaton and patterns of comparatve advantage n the product space. Mmeo. Imbs J and Waczarg R (2003). Stages of Dversfcaton. Amercan Economc Revew. 93(1): 63-86. Nehru V and Dhareshwar A (1993). A new database on physcal captal stock: sources, methodology and results. Revsta de Analss Economco. 8(1): 37 59. Nehru V, Swanson E and Dubey A (1993). A new database on human captal stock: sources, methodology, and results. Polcy Research Workng Paper 1124. World Bank. Washngton, D.C. 45

Klenow P and Rodrguez-Clare A (1997). The neoclasscal revval n growth economcs: has t gone too far? In Natonal Bureau of Economc Research Macroeconomcs Annual 1997, edted by Ben S.Bernanke and Julo Rotemberg: 73 103, Cambrdge, MA: MIT Press. Klnger B and Lederman D (2005). Dversfcaton, nnovaton and mtaton off the global technologcal fronter. Mmeo. Krueger A and Lndahl M (2001). Educaton for growth: why and for whom? Journal of Economc Lterature. Amercan Economc Assocaton. 39(4): 1101 1136. Kyracou GA (1991). Level and growth effects of human captal: a cross-country study of the convergence hypothess. Unpublshed. New York Unversty. Larson DF, Butzer R, Mundlak Y and Crego A (2000). A cross-country database for sector nvestment and captal. World Bank Economc Revew. 14: 371 391. Lau L, Jamson D and Louat F (1991). Educaton and productvty n developng countres: an aggregate producton functon approach. Workng Paper Seres, 612, World Bank. Lau L, Bhalla S and Louat F (1991). Human and physcal captal stock n developng countres: constructon of data and trends. Draft mmeo. World Development Report, World Bank. OECD (2007). Producer and consumer support estmates, OECD database 1986 2006. Avalable at www.oecd.org/document/59/0,3343,en _2649_33727_39551355_1_1_1_1,00.html. OECD (2008). Agrcultural polces n OECD countres: at a glance. Psacharopoulos G and Aragada AM (1986). The educatonal composton of the labor force: an nternatonal comparson. Internatonal Labor Revew. 125: 561 574. Vollrath TL (1987). Revealed compettveness for wheat. Economc Research Servce Staff Report, No AGES861030, U.S. Dept. of Agrculture, Economc Research Servce. Washngton, D.C. Vollrath TL (1989). Compettveness and protecton n world agrculture. Agrcultural Informaton Bulletn, 567, Unted States Dept. of Agrculture, Economc Research Servce. Washngton, D.C. World Bank (1997). Expandng the measure of wealth: ndcators of envronmentally sustanable development. World Bank (2006). Where s the wealth of natons? Measurng captal for the twenty-frst century. 46

UNCTAD Study Seres on POLICY ISSUES IN INTERNATIONAL TRADE AND COMMODITIES No. 1 No. 2 No. 3 No. 4 No. 5 Erch Supper, Is there effectvely a level playng feld for developng country exports?, 2001, 138 p. Sales No. E.00.II.D.22. Arvnd Panagarya, E-commerce, WTO and developng countres, 2000, 24 p. Sales No. E.00.II.D.23. Joseph Francos, Assessng the results of general equlbrum studes of multlateral trade negotatons, 2000, 26 p. Sales No. E.00.II.D.24. John Whalley, What can the developng countres nfer from the Uruguay Round models for future negotatons?, 2000, 29 p. Sales No. E.00.II.D.25. Susan Teltscher, Tarffs, taxes and electronc commerce: Revenue mplcatons for developng countres, 2000, 57 p. Sales No. E.00.II.D.36. No. 6 Bt Bora, Peter J. Lloyd, Mar Pangestu, Industral polcy and the WTO, 2000, 47 p. Sales No. E.00.II.D.26. No. 7 No. 8 No. 9 No. 10 No. 11 No. 12 Emlo J. Medna-Smth, Is the export-led growth hypothess vald for developng countres? A case study of Costa Rca, 2001, 49 p. Sales No. E.01.II.D.8. Chrstopher Fndlay, Servce sector reform and development strateges: Issues and research prortes, 2001, 24 p. Sales No. E.01.II.D.7. Inge Nora Neufeld, Ant-dumpng and countervalng procedures Use or abuse? Implcatons for developng countres, 2001, 33 p. Sales No. E.01.II.D.6. Robert Scollay, Regonal trade agreements and developng countres: The case of the Pacfc Islands proposed free trade agreement, 2001, 45 p. Sales No. E.01.II.D.16. Robert Scollay and John Glbert, An ntegrated approach to agrcultural trade and development ssues: Explorng the welfare and dstrbuton ssues, 2001, 43 p. Sales No. E.01.II.D.15. Marc Bacchetta and Bt Bora, Post-Uruguay round market access barrers for ndustral products, 2001, 50 p. Sales No. E.01.II.D.23. No. 13 Bt Bora and Inge Nora Neufeld, Tarffs and the East Asan fnancal crss, 2001, 30 p. Sales No. E.01.II.D.27. No. 14 No. 15 Bt Bora, Lucan Cernat, Alessandro Turrn, Duty and quota-free access for LDCs: Further evdence from CGE modellng, 2002, 130 p. Sales No. E.01.II.D.22. Bt Bora, John Glbert, Robert Scollay, Assessng regonal tradng arrangements n the Asa-Pacfc, 2001, 29 p. Sales No. E.01.II.D.21. 47

No. 16 Lucan Cernat, Assessng regonal trade arrangements: Are South-South RTAs more trade dvertng?, 2001, 24 p. Sales No. E.01.II.D.32. No. 17 Bt Bora, Trade related nvestment measures and the WTO: 1995-2001, 2002. No. 18 Bt Bora, Ak Kuwahara, Sam Lard, Quantfcaton of non-tarff measures, 2002, 42 p. Sales No. E.02.II.D.8. No. 19 No. 20 No. 21 No. 22 No. 23 No. 24 No. 25 No. 26 No. 27 No. 28 No. 29 No. 30 No. 31 No. 32 No. 33 Greg McGure, Trade n servces Market access opportuntes and the benefts of lberalzaton for developng economes, 2002, 45 p. Sales No. E.02.II.D.9. Alessandro Turrn, Internatonal trade and labour market performance: Maor fndngs and open questons, 2002, 30 p. Sales No. E.02.II.D.10. Lucan Cernat, Assessng south-south regonal ntegraton: Same ssues, many metrcs, 2003, 32 p. Sales No. E.02.II.D.11. Kym Anderson, Agrculture, trade reform and poverty reducton: Implcatons for Sub-Saharan Afrca, 2004, 30 p. Sales No. E.04.II.D.5. Ralf Peters and Davd Vanzett, Shftng sands: Searchng for a compromse n the WTO negotatons on agrculture, 2004, 46 p. Sales No. E.04.II.D.4. Ralf Peters and Davd Vanzett, User manual and handbook on Agrcultural Trade Polcy Smulaton Model (ATPSM), 2004, 45 p. Sales No. E.04.II.D.3. Khall Rahman, Crawlng out of snake pt: Specal and dfferental treatment and post-cancun mperatves, 2004. Marco Fugazza, Export performance and ts determnants: Supply and demand constrants, 2004, 57 p. Sales No. E.04.II.D.20. Lus Abugattas, Swmmng n the spaghett bowl: Challenges for developng countres under the New Regonalsm, 2004, 30 p. Sales No. E.04.II.D.38. Davd Vanzett, Greg McGure and Prabowo, Trade polcy at the crossroads The Indonesan story, 2005, 40 p. Sales No. E.04.II.D.40. Smonetta Zarrll, Internatonal trade n GMOs and GM products: Natonal and multlateral legal frameworks, 2005, 57 p. Sales No. E.04.II.D.41. Sam Lard, Davd Vanzett and Santago Fernández de Córdoba, Smoke and mrrors: Makng sense of the WTO ndustral tarff negotatons, 2006, Sales No. E.05.II.D.16. Davd Vanzett, Santago Fernandez de Córdoba and Veronca Chau, Banana splt: How EU polces dvde global producers, 2005, 27 p. Sales No. E.05.II.D.17. Ralf Peters, Roadblock to reform: The persstence of agrcultural export subsdes, 2006, 43 p. Sales No. E.05.II.D.18. Marco Fugazza and Davd Vanzett, A South-South survval strategy: The potental for trade among developng countres, 2006, 25 p. 48

No. 34 No. 35 No. 36 No. 37 No. 38 No. 39 Andrew Cornford, The global mplementaton of Basel II: Prospects and outstandng problems, 2006, 30 p. Lakshm Pur, IBSA: An emergng trnty n the new geography of nternatonal trade, 2007, 50 p. Crag VanGrasstek, The challenges of trade polcymakng: Analyss, communcaton and representaton, 2008, 45 p. Sudp Ranan Basu, A new way to lnk development to nsttutons, polces and geography, 2008, 50 p. Marco Fugazza and Jean-Chrstophe Maur, Non-tarff barrers n computable general equlbrum modellng, 2008, 25 p. Alberto Portugal-Perez, The costs of rules of orgn n apparel: Afrcan preferental exports to the Unted States and the European Unon, 2008, 35 p. No. 40 Baley Klnger, Is South-South trade a testng ground for structural transformaton?, 2009, 30 p. No. 41 No. 42 No. 43 No. 44 Sudp Ranan Basu, Vctor Ognvtsev and Mho Shrotor, Buldng trade-relatng nsttutons and WTO accesson, 2009, 50 p. Sudp Ranan Basu and Monca Das, Insttuton and development revsted: A nonparametrc approach, 2010, 26 p. Marco Fugazza and Norbert Fess, Trade lberalzaton and nformalty: New stylzed facts, 2010, 45 p. Mho Shrotor, Bolormaa Tumurchudur and Olver Cadot, Revealed factor ntensty ndces at the product level, 2010, 55 p. 49

Unted Natons Publcatons All orders from North Amerca, Latn Amerca, the Carbbean and Asa and the Pacfc should be sent to: Unted Natons Publcatons Room DC2-853, 2 UN Plaza New York, NY 10017, USA Telephone: (212) 963-8302, Toll Free 1-800-253-9646 (North Amerca only) Fax: (212) 963-3489 E-mal: publcatons@un.org Customers n Europe, Afrca and the Mddle East should send ther orders to: Secton des Ventes et Commercalsaton Bureau E-4, CH-1211 Geneva 10, Swtzerland Telephone: 41 (22) 917-2613/2614 Fax: 41 (22) 917-0027 E-mal: unpubl@unog.ch For further nformaton, please vst: http://www.unctad.org/tab 50

QUESTIONNAIRE UNCTAD Study seres on POLICY ISSUES IN INTERNATIONAL TRADE AND COMMODITIES (Study seres no. 44: Revealed factor ntensty ndces at the product level) Readershp Survey Snce 1999, the Trade Analyss Branch of the Dvson on Internatonal Trade n Goods and Servces, and Commodtes of UNCTAD has been carryng out polcy-orented analytcal work amed at mprovng the understandng of current and emergng ssues n nternatonal trade and development. In order to mprove the qualty of the work of the Branch, t would be useful to receve the vews of readers on ths and other smlar publcatons. It would therefore be greatly apprecated f you could complete the followng questonnare and return to: Trade Analyss Branch, DITC Rm. E-8076 Unted Natons Conference on Trade and Development Palas des Natons CH-1211 Geneva 10, Swtzerland 1. Name and address of respondent (optonal): 2. Whch of the followng descrbes your area of work? Government Prvate enterprse nsttuton Internatonal organzaton Not-for-proft organzaton Publc enterprse Academc or research Meda Other (specfy) 3. In whch country do you work? 4. Dd you fnd ths publcaton Very useful Of some use Lttle use to your work? 5. What s your assessment of the contents of ths publcaton? Excellent Good Adequate Poor 6. Other comments: