OECD DEVELOPMENT CENTRE

Transcription

1 OECD DEVELOPMENT CENTRE Workin Paer No. 288 INNOVATION, roductivity and economic develoment in latin america and the caribbean by Christian Daude Research area: InnovaLatino February 2010

2 Innovation, Productivity and Economic Develoment in Latin America and the Caribbean DEVELOPMENT CENTRE WORKING PAPERS This series of workin aers is intended to disseminate the Develoment Centre s research findins raidly amon secialists in the field concerned. These aers are enerally available in the oriinal Enlish or French, with a summary in the other lanuae. Comments on this aer would be welcome and should be sent to the OECD Develoment Centre, 2 rue André Pascal, PARIS CEDEX 16, France; or to dev.contact@oecd.or. Documents may be downloaded from: htt:// or obtained via (dev.contact@oecd.or). THE OPINIONS EXPRESSED AND ARGUMENTS EMPLOYED IN THIS DOCUMENT ARE THE SOLE RESPONSIBILITY OF THE AUTHOR AND DO NOT NECESSARILY REFLECT THOSE OF THE OECD OR OF THE GOVERNMENTS OF ITS MEMBER COUNTRIES OECD (2010) Alications for ermission to reroduce or translate all or art of this document should be sent to rihts@oecd.or CENTRE DE DÉVELOPPEMENT DOCUMENTS DE TRAVAIL Cette série de documents de travail a our but de diffuser raidement aurès des sécialistes dans les domaines concernés les résultats des travaux de recherche du Centre de déveloement. Ces documents ne sont disonibles que dans leur lanue oriinale, anlais ou français ; un résumé du document est rédié dans l autre lanue. Tout commentaire relatif à ce document eut être adressé au Centre de déveloement de l OCDE, 2 rue André Pascal, PARIS CEDEX 16, France; ou à dev.contact@oecd.or. Les documents euvent être télécharés à artir de: htt:// ou obtenus via le mél (dev.contact@oecd.or). LES IDÉES EXPRIMÉES ET LES ARGUMENTS AVANCÉS DANS CE DOCUMENT SONT CEUX DE L AUTEUR ET NE REFLÈTENT PAS NÉCESSAIREMENT CEUX DE L OCDE OU DES GOUVERNEMENTS DE SES PAYS MEMBRES OCDE (2010) Les demandes d'autorisation de reroduction ou de traduction de tout ou artie de ce document devront être envoyées à rihts@oecd.or 2 OECD 2010

3 OECD Develoment Centre Workin Paer No. 288 TABLE OF CONTENTS ACKNOWLEDGEMENTS... 4 PREFACE... 5 RÉSUMÉ... 7 ABSTRACT... 8 I. INTRODUCTION... 9 II. DEVELOPMENT ACCOUNTING III. WHAT IS BEHIND THE TFP GAP? IV. OTHER DEVELOPMENT DIMENSIONS: LIFE EXPECTANCY CONVERGENCE V. TFP GAPS AND INNOVATION SHORTFALLS VI. CONCLUSIONS APPENDIX REFERENCES OTHER TITLES IN THE SERIES/ AUTRES TITRES DANS LA SÉRIE OECD

4 Innovation, Productivity and Economic Develoment in Latin America and the Caribbean ACKNOWLEDGEMENTS I am rateful to Laura Alfaro, Mauricio Cárdenas, Francesca Castellani, Eduardo Fajnzylber, and William Maloney, as well as articiants of October 2009 INNOVALatino Exerts Meetin in Buenos Aires, for insihtful comments and suestions. As usual, all remainin errors are my sole resonsibility. 4 OECD 2010

5 OECD Develoment Centre Workin Paer No. 288 PREFACE Just rior to the current lobal financial crisis,latin America enjoyed several banner years of economic rowth. Even so, the reion s lon-term rowth erformance, when comared to that of other develoin reions or to OECD countries, remains too low to trier a substaintial shift towards hiher levels of develoment. Whatis holdin the reion back? In macroeconomic terms, there are two ossible exlanations. First, it could be that Latin American economies are exeriencin slow rowth of oulation and slow rates of investment these are the factors of roduction that allow outut to row. Alternatively, it could be that economies in the reion do a comaratively oor job of combinin those factors of roduction. Which is it: slow factor accumulation or low roductivity? This workin aer by Christian Daude of the OECD Develoment Centre identifies low roductivity as the culrit reventin Latin American economies from converin to standards of livin observed in other emerin economies and in the OECD. Thus, this aer s conclusion reminds us of the fundamental imortance of understandin better the barriers to innovation and technoloy adotion on the one hand, and the identification of olicies that increase economic efficiency and sur economic rowth, on the other. These are the twin oals of the Innovalatino initiative, a joint venture between the OECD Develoment Centre and the INSEAD Business School. The first reort of the Innovalatino roject is set to be resented at the Euroean Union/Latin American and Caribbean Summit in Sain in May Daude s aer is a backround study for that reort.. In this aer, Daude exlores why roductivity rows more slowly in Laitn America than elsewhere. He arues that economic and olitical institutions, as well as differences in human caital and access to finance, exlain the reion s a with resect to OECD countries reardin the diffusion of new technoloies. In an interestin extension, the aer looks beyond roductivity rowth to consider the determinants of life exectancy; aruably, this is a more immediately relevant measure of well-bein than roductivity or GDP er caita, and one in which the diffusion of new technoloies and ractices is also crucial. Daude shows that Latin American countries have been much more successful in romotin loner lives than in raisin roductivity. Even controllin for initial levels of life exectancy, the level of develoment, the availability of human resources in the health sector, and the decline in infant mortality all linked to rather low-tech innovations on averae, the reion shows a hiher seed of converence to the frontier than the rest of the world. This raises the curious question of why technoloical chane in the health sector seems to sread more easily than knowlede in other arts of the economy. Althouh more detailed research is needed, Daude suests that the comaratively better erformance in imrovin health may owe to fewer olitical and economic OECD

6 Innovation, Productivity and Economic Develoment in Latin America and the Caribbean barriers to knowlede diffusion at the national and international levels in this sector than in other sectors of the economy. Our hoe is that the results of the Innovalatino roject, and of this workin aer in articular, will romote fruitful olicy dialoue that will ultimately lead to an inflexion in Latin America s roductivity rowth, with benefits at all levels of the social yramid. Javier Santiso Director and Chief Develoment Economist OECD Develoment Centre February OECD 2010

7 OECD Develoment Centre Workin Paer No. 288 RÉSUMÉ En Amérique latine, le PIB ar habitant n a eu de cesse deuis lusieurs décennies de reculer ar raort à celui des ays à hauts revenus et d autres ays de références. Les mauvaises erformances de la réion en terme de croissance sont l une des rinciales raisons our lesquelles la réduction de la auvreté, et de façon énérale le niveau de vie, sont bien lus faibles que ceux observés dans les ays. Dans cet article, nous exlorons certaines des raisons otentielles de cette mauvaise erformance à l aide de techniques comtables de déveloement. Les résultats tendent à montrer que la rinciale cause de l absence de converence de la réion est la roductivité totale des facteurs. Afin de rechercher ourquoi ces ays n ont as comblé leur retard de roductivité, nous analysons les déterminants des technoloies de diffusion, et en articulier internet et les technoloies de téléhonie mobile. Les résultats emiriques montrent que les institutions, la caacité d absortion (caital humain) et les contraintes financières sont les rinciales variables exlicatives de l écart qui existe entre les ays de l OCDE et ceux de l Amérique latine concernant la diffusion de ces technoloies. Nous exlorons éalement la erformance de la réion en matière de santé, mesurée ar l évolution de l esérance de vie, et le rôle sécifique joué ar l innovation et l adotion technoloique. Finalement, un exercice de calibrae d un modèle de croissance endoène nous ermet d évaluer jusqu à quel oint la différence de revenu ar tête au sein de la réion est due à des roblèmes d allocation des facteurs ou à des distorsions qui diminuent les incitations à innover. Les résultats varient fortement d un ays à l autre au sein de la réion. Si our certains ays nous mettons en évidence un «manque d innovation», our d autres, la faible accumulation de facteurs demeure le rincial roblème. Latine Classification JEL: O10; O30; O47 Mots clé: roductivité totale des facteurs; innovation; croissance économique ; Amérique OECD

8 Innovation, Productivity and Economic Develoment in Latin America and the Caribbean ABSTRACT GDP er caita in Latin America has been fallin behind hih-income countries and other benchmarks for decades and the reion s mediocre rowth erformance is one of the main reasons why overty reduction, and livin standards more enerally, in the reion is well below that observed in eer countries. In this aer, we exlore some of the otential roots of this oor erformance by usin develoment accountin techniques. The results oint towards total factor roductivity as the main culrit for the reion s lack of converence. In order to investiate what causes the lack of roductivity catch-u, we analyse the determinants of technoloy diffusion, in articular of internet and mobile hone technoloies. The emirical results show that institutions, absortion caacity (human caital), and financial constraints are the main exlanatory variables of the diffusion as in these technoloies between the OECD and Latin America. We also exlore the erformance of the reion in terms of health outcomes, reflected in the evolution of life exectancy, and the secific role layed by technoloical innovation and adotion. Finally, a calibration exercise of an endoenous rowth model allows us to assess the extent to which the reion s er caita income a is due to roblems in factor accumulation or distortions that reduce the incentives to innovate; the results oint to very different situations across countries in the reion. While for some countries we find evidence of innovation shortfalls, other countries roblems concentrate around low factor accumulation. JEL Classification: O10; O30; O47 Keywords: Total factor roductivity; innovation; economic rowth; Latin America 8 OECD 2010

9 OECD Develoment Centre Workin Paer No. 288 I. INTRODUCTION Economic rowth in Latin America and the Caribbean (LAC) has been low and volatile for several decades. As shown in Fiure 1 and 2, between 1960 and 2008, GDP er caita in the averae country in the reion multilied by a factor of just 2.1, which imlies an averae annual rowth rate of GDP er caita of around 1.5 er cent (in PPP terms). This contrasts sharly with the rowth exerience of East Asian countries. While in 1960 LAC was on averae 60 er cent richer than the averae East Asian country, in 2008 GDP er caita in East Asia was almost 3 times that in the LAC reion! While art of this incredible diverence is due to the stron erformance of East Asia, when comared to Hih-Income OECD countries 1 or countries that had levels of GDP er caita very similar to those in the reion in 1960, 2 the main facts still hold: on averae, there is a steady decline in Latin America s GDP er caita relative to any of these rous. 3 Fiure 1. Evolution of GDP er caita in PPP terms across reions (index 1960 = 100) 1000 LAC East Asia Peers 1960 Hih Income OECD Source: Own calculations based on Penn World Tables version 6.2 and World Develoment Indicators This rou includes Australia, Austria, Belium, Canada, Denmark, Finland, France, Germany, Greece, Hunary, Ireland, Italy, Jaan, Korea, Netherlands, New Zealand, Norway, Portual, Sain, Sweden, United Kindom and United States. This rou includes Aleria, Fiji, Greece, Hon Kon, Hunary, Iran, Jaan, Jordan, Portual and Sinaore. In the aer we use simle eometric means for rou averaes to reresent the averae country in each reion. OECD

10 Innovation, Productivity and Economic Develoment in Latin America and the Caribbean Furthermore, as Fiure 3 shows, even countries that are reional success stories (e.. Chile and Dominican Reublic since the late 1980 s and early 1990 s, resectively) exhibit only a modest erformance, recoverin only art of the lost round in the earlier decades, when comared to other exeriences around the world. Fiure 2. Ratio of LAC GDP er caita versus benchmarks East Asia Peers 1960 Hih Income OECD Source: Own calculations based on Penn World Tables version 6.2 and World Develoment Indicators. Fiure 3. Ratio of GDP er caita (PPP) versus eer countries Arentina Brazil Chile Panama Dominican Reublic Venezuela Source: Own calculations based on Penn World Tables version 6.2 and World Develoment Indicators. 10 OECD 2010

11 OECD Develoment Centre Workin Paer No. 288 Desite the fact that the eriod resented the hihest er caita GDP rowth rate in LAC in the last 50 years, the fiures above show that this ositive news has so far had a modest imact on closin the develoment a with resect to other reions. Growth durin this eriod in LAC has clearly been hiher than in the eer countries (Fiures 2 and 3), but comared to East Asia or even the Hih-Income OECD (Fiure 2), the erformance has been ood, but far from bein quantitatively imortant so far. The resent aer exlores the some of the otential exlanations behind this disaointin erformance of the reion. In section II, we show some develoment accountin exercises that oint towards total factor roductivity (TFP), the efficiency with which factors of roduction are combined in the economy, as the main reason for the reion s lack of converence in GDP er caita to the frontier. Section III exlores some of the exlanations behind the reion s ersistent TFP a. Furthermore, we analyse the factors that exlain the a in the adotion and diffusion of new technoloies in the reion with resect to OECD countries. Of course, GDP er caita is not the only measure of economic roress; accordinly in section IV, we discuss the converence of life exectancy in the reion comared to the rest of the world. In articular, we relate the ood erformance in the reion on this extensive marin to the discussion on TFP and converence of the revious sections. Section V analyses the existence of innovation as in the reion based on the calibration of an endoenous rowth model. In section VI, we conclude and discuss some olicy imlications of our main results. OECD

12 Innovation, Productivity and Economic Develoment in Latin America and the Caribbean II. DEVELOPMENT ACCOUNTING 4 What lies behind the disaointin erformance of LAC discussed in the revious section? As we will show, accordin to standard develoment accountin exercises, mediocre TFP is the main culrit in most countries of the reion. Growth and develoment accountin techniques traditionally rely on a decomosition on outut based on a areate roduction f(.) with constant returns to scale that mas accumulated roduction factors (hysical and human caital, denoted by K and H resectively) into final outut (Y). In articular, we assume a Cobb-Doulas technoloy that takes the followin form: 1 1 Y f ( K, H) AK H AK ( hl). (1) We follow the literature since Hall and Jones (1999) by assumin that the stock of human caital can be measured by the roduct of the quality of the labour force (h) and the labour force (L). 5 The arameter A is TFP, which reresents the efficiency with which factors of roduction are combined in the economy. As it is standard in the literature (see Klenow and Rodriuez-Clare, 2005), we assume that the caital share (α = 1/3) is the same across countries and constant over time. This is a standard assumtion in the literature which is mainly based on the evidence for the United States. While Gollin (2002) shows that there is a sinificant variation across countries in this arameter, this variation does not follow any articular attern. In articular, once entrereneurshi and informality are taken into account, it is not related to the level of develoment (GDP er caita). Usin series of hysical and human caital, 6 we comute our measure of TFP by: A Y (hl) 1 K. (2) Unfortunately, internationally comarable data on hysical caital formation and human caital are available until 2005, such that the last rowth sur cannot be analysed entirely here. However, as discussed in the revious section, GDP er caita relative to all benchmarks has not chaned too much in the eriod, such that the rather structural aroach taken here focusin on trend TFP assures that the results remain valid The analysis in this section draws heavily on revious work by Daude and Fernández-Arias (2010). Of course, this assumes erfect substitutability between skilled and unskilled labour. See Caselli (2005) for the imlications of usin a lower elasticity of substitution between different levels of skills. See the aendix for details reardin the data. 12 OECD 2010

13 OECD Develoment Centre Workin Paer No. 288 The evolution of TFP comuted usin equation (2) for the averae LAC country is shown in Fiure 4. As the fiure shows, TFP in the averae country of the reion rew until the late 1970s and has declined since then, reboundin somewhat only in recent years Fiure 4. Evolution of TFP for the averae LAC country (1960 = 1) Source: Daude and Fernández-Arias (2010). It is imortant to oint out that the evolution hihlihts an imortant asect of TFP: in the standard rowth literature it is assumed to measure technoloical roress, absolute declines are difficult to interret in this way. Thus, a more eneral interretation of TFP is required. In articular, we think that the correct interretation is that measured TFP reflects also the deree of efficiency with which markets and institutions work to combine and allocate roductive factor in the economy. Clearly, under this broader interretation, efficiency can decline in absolute terms for a roloned eriod of time, as we observe for the case of LAC. As Fiure 5 shows, there is almost an even slit between countries in the reion that in 2005 exhibit hiher levels of TFP than in 1960 (8 cases) and those who have lower levels of efficiency today than in the 1960s (10 cases). Furthermore, the reional leader in terms of TFP rowth erformance (Chile) increased it level of TFP be around 55 er cent, which is aroximately equivalent to a 1 er cent rowth rate er annum durin the eriod, almost twice the rate of TFP rowth for the United States. 8 While Chile is the only country in the reion where TFP rew faster than in the United States, TFP rowth in the US is close to the median in our samle, which imlies that in more than 30 countries TFP rowth was actually hiher, such that 7 Part of this end-of-samle rebound is driven by the nature of the Hodrick-Prescott filter used to analyse these data. Therefore, this increase in the recent eriod should not be extraolated, iven that the estimates would robably be revised sinificantly once more data would become available. 8 Nevertheless, the United States continues to be the country with the hihest TFP level in OECD

14 Innovation, Productivity and Economic Develoment in Latin America and the Caribbean converence to the frontier is not a rare event. This is in line with the imortance of roductivity converence as a factor of converence in GDP er caita emhasised by Bernard and Jones (1996). Fiure 5. TFP ratios (2005 versus 1960) Chile United States Panama Brazil Dominican Reublic Ecuador Uruuay Bolivia Colombia Peru Jamaica Averae LAC Mexico Arentina Costa Rica Parauay El Salvador Venezuela Honduras Nicaraua Source: Own calculations based on Daude and Fernández-Arias (2010). 85 Fiure 6. Averae LAC TFP relative to United States (er cent) Source: Daude and Fernández-Arias (2010). In Fiure 6 and 7, we show TFP levels for the averae country in LAC over time and for each country as of 2005 with resect to the United States (which is usually considered the frontier), resectively. As it can be seen, TFP has been fallin behind since the early 1980s from 14 OECD 2010

15 OECD Develoment Centre Workin Paer No. 288 an efficiency of almost 80 er cent to only 52 er cent in Thus, a iven amount of roduction factors roduces only half the outut in the tyical LAC country than it would be feasible roducin in the United States. There is however a relatively lare disersion across countries in the reion in terms of TFP. For examle, in Chile or Costa Rica TFP is around ¾ of US TFP, while at the other end of the sectrum in Honduras or Peru it is only around er cent. These as are hue, iven that they imly that Honduras would trile it income er caita, if it could close the efficiency a with resect to the United States. 9 Fiure 7. TFP relative to the United States in 2005 (er cent) Chile Costa Rica Dominican Re. Arentina Uruuay El Salvador Mexico Brazil Colombia Panama Venezuela Averae LAC Parauay Nicaraua Jamaica Ecuador Bolivia Peru Honduras Source: Daude and Fernández-Arias (2010). In eneral, from a develoment ersective we are interested in decomosin and comarin GDP er caita (y). For this urose equation (1) can be written in er caita terms: Y K 1 L 1 y A h Ak h f, (3) N L N where the last term (f) is the share of the labour force in the total oulation. 10 Physical caital investment clearly deends on the level of TFP. To take this indirect effect of TFP into account, followin Klenow and Rodriuez-Clare (1997) the roduction function (3) can be rewritten in intensive form as: y A hf, (4) 9 10 It is imortant to notice that the disersion in TFP as is smaller than GDP er caita as in the reion (the coefficient of variation for TFP as is half that of GDP er caita a). This ratio in turn results from the share of workin ae oulation (a demorahic factor) and the rate of its articiation in the labour force, an economic decision which for simlicity we consider exoenous. OECD

16 Arentina Bolivia Brazil Chile Colombia Costa Rica Dominican Reublic Ecuador El Salvador Honduras Jamaica Mexico Nicaraua Panama Parauay Peru Uruuay Venezuela Per cent Innovation, Productivity and Economic Develoment in Latin America and the Caribbean where κ is the caital-outut ratio (K/Y) which does not deend on TFP in the neoclassical rowth model and does therefore reflect distortions secific to hysical caital accumulation. 11 With resect to the frontier or any other benchmark (denoted by * ), we have: y * y A h f * *, (5) * * A h f In Fiure 8, we resent the lo-decomosition based on equation (5) by country with resect to the United States. As the rah shows, TFP exlains most of the a in income er caita with resect to the United States, but other factors are still imortant. On averae, TFP account for 56 er cent of the a, hysical caital distortion for around 11, human caital for 24 er cent, and demorahic and labour force articiation factors for the remainin 8 er cent of the income er caita a. 100 Fiure 8. Decomosition of GDP er caita as versus United States (2005) TFP Imediment to investment (к) Human caital (h) Labor force Intensity (f) Source: Daude and Fernández-Arias (2010). There are also interestin differences across countries in the reion. On the one hand, in countries like Peru, Ecuador, Panama, Honduras and Venezuela, TFP is a major exlanation of the income er caita a, with a contribution between around 60 and 70 er cent. On the other hand, in economies like Costa Rica, Dominican Reublic and El Salvador, the main exlanation 11 This follows from simle rofit maximisation in the neoclassical rowth model. Observe that for now we assume human caital to be exoenous. Thus, human caital just like TFP has now two effects on outut: a direct effect and an indirect effect throuh hysical caital investment which is increasin in h. 16 OECD 2010

17 OECD Develoment Centre Workin Paer No. 288 come from factor distortions. Chile is also an interestin case. While TFP exlains 39 er cent of the income er caita a with resect to the United States, human caital exlains nearly a third of it a hih contribution shared also with Brazil, Costa Rica and the Dominican Reublic. Furthermore, labour market articiation (rather than demorahic) factors lay an imortant role in exlainin the income er caita a in Chile (almost 24 er cent of the a). This is driven by the fact that female labour articiation is one of the lowest in the reion (as of 2005 only 38.4 er cent of women of 15 years and older work in Chile, comared with a 52 er cent averae in the reion, and 52.6 er cent in hih income countries, accordin to the World Develoment Indicators of the World Bank). While this could artially reflect references over leisure and work decisions as well as culture asects, it seems reasonable to exect that a lare art of this lower articiation of women in the economy is caused by distortions whose elimination would be welfare imrovin. Furthermore, as Fiure 9 shows, over time the imortance of TFP in exlainin the averae income a of the reion with resect to the United States has increased from around 37 er cent in 1960 to 55 er cent nowadays, with hysical caital distortions accountin for a rather constant 10 er cent of the a, and human caital reducin its imortance from 39 er cent to 24 er cent. 12 Fiure 9. Decomosition of GDP er caita a (averae LAC versus United States) 100 TFP Imediment to hysical investment (к) Human Caital (h) Labor force Intensity (f) Source: Daude and Fernández-Arias (2010). Summin u, standard develoment accountin techniques show that TFP is the main reason why the reion is not catchin u. If countries in the reion could eliminate the restrictions behind these low levels of TFP, welfare measured by income er caita would 12 Of course, as mentioned above, the otential as in the quality of education are catures by TFP rather than human caital (see Fiure 10 below). OECD

18 Innovation, Productivity and Economic Develoment in Latin America and the Caribbean almost trile (increasin by a factor of 2.7) for the averae LAC country. For examle, oeratin at TFP levels similar to the United States would increase Honduras income er caita (around USD in PPP terms for 2005) to a similar level of Chile currently (almost USD in PPP terms)! Even for a country such as Chile the ains are substantial: oin from USD er caita to almost USD Certainly, these sinificant fiures raise immediately a series of questions. First, if ains from TFP imrovements are so lare, why have overnments in the reion not made the necessary chanes and reforms to increase efficiency, as well as absorb and create more knowlede? Even if TFP catures broader institutional issues, many of the ideas and arranements that seem to work in terms of roductive efficiency could be relatively well-known and do not have to be invented from scratch, requirin therefore only limited investments of resources. Alternatively, the second rou of questions refer to whether technoloical knowlede is really available at low or no cost of adotion? Third, related the revious question, is slow catch u with the frontier due to lack of absortion caacities or insufficient investment in ideas and R&D? We discuss some of these issues and suortin evidence in the next section. 18 OECD 2010

19 OECD Develoment Centre Workin Paer No. 288 III. WHAT IS BEHIND THE TFP GAP? There are different ways to aroach analytically the questions outlined above, which otentially lead to very different olicy imlications. Followin the work by Parente and Prescott (1994 and 1999) and Prescott (1998), a ossible interretation is that knowlede is available to develoin countries, but not fully incororated because of vested interests that reduce cometition and the economic incentive to oerate at the frontier (see also Krussell and Rios- Rull, 1996). Thus, from this ersective what is needed is a clear understandin of the olitical and economic forces that hold back the institutional framework and market forces to adot the best roduction ractices. An alication of this aroach for the case of Latin America is Cole et al. (2005) who arue that trade rotectionism, monooly ower and lack of cometition are the main drivers of the ersistently low levels of TFP in the reion. Related to this literature that emhasised areate inefficiencies, there is a literature that has focused on the interaction between distortions and the existence of heteroeneous firms, which lead to misallocations of factors (see Restuccia and Roerson, 2008). This literature tends to focus on barriers to entry for new and more roductive firms; barriers to exit that maintain inefficient firms in the market. For examle, Hsieh and Klenow (forthcomin) find that TFP in manufacturin could be between 30 to 60 er cent hiher in China and India, if factors were reallocated amon firms to reroduce the US roductivity distribution. A recent reort by the Inter-American Develoment Bank shows that these tyes of misallocations are also sinificant in the manufacturin sector in Latin America (see Inter-American Develoment Bank, 2010). A quite different emhasis comes from the literature labelled as aroriate technoloies. 13 This literature focuses on the issue that technoloical roress and innovation mainly take lace in develoed countries where factor endowments are different. 14 Furthermore, technoloical roress is often non-neutral with resect to factor endowments, 15 and there miht be little substitutability between factors either caital-labour or skilled-unskilled labour as emhasised also by Caselli and Coleman (2006). Clearly, the concetual debate reardin these issues is not yet settled in the literature. Caselli and Coleman (2006) and Jermanowksi (2007) show that relaxin the assumtion of an areate Cobb-Doulas roduction function (makin factors substitution harder), as well as allowin technical chane to be biased towards the relative factor endowments of the develoed world, reduce the contribution of TFP to income The term was coined by an early contribution by Atkinson and Stilitz (1969). See Basu and Weil (1998) for caital-labour mismatches and Acemolu and Zilibotti (2001) for skilledunskilled labour. Skill-biased technoloical roress lays and imortant role in Acemolu and Zilibotti (2001). OECD

20 Innovation, Productivity and Economic Develoment in Latin America and the Caribbean as. However, the quantitative imlications of these chanes seem to be rather limited, such that TFP often continues to be the most relevant factor in exlainin cross-country income disersion. For the case of Latin America, Daude and Fernández-Arias (2010) resent nonarametric estimations of the relative efficiency, usin data enveloe techniques, that ive icture very similar to standard techniques in terms of the evolution over time and the size of the contribution of TFP to the develoment a with resect to the frontier. III.1. Correlation analysis between TFP and otential drivers Next, we exlore the relationshi between TFP rowth and levels with a series of variables that have been associated in the literature as otential drivers of roductivity. In articular, the estimations in Table 1 focus on TFP rowth durin the latest decade for which data are available ( ). In all reressions, we include the initial level of TFP to allow for converence. Althouh the oint estimates show a neative coefficient, consistent with the idea that countries further away from the frontier rew faster durin this decade, it is not sinificant at conventional levels of confidence. Columns 1 to 7 introduce the exlanatory variables one-byone, in order to exlore the simle correlation with TFP rowth, while in column 8 we introduce all of them at once. Table 1. TFP rowth and exlanatory variables ( ) Variable (1) (2) (3) (4) (5) (6) (7) (8) (9) Initial level of lo TFP Technoloical creation index (0.95) (0.19) (0.30) (0.74) (1.35) (0.43) (1.05) (1.26) (0.94) (2.37)* (0.29) (0.81) Trade/GDP (%) (0.55) (0.82) (0.59) FDI/GDP (%) Private Sector Domestic Credit/GDP (2.12)* (1.00) (0.88) (1.66) (0.61) (0.42) Reulatory Quality (2.31)* (0.55) (1.18) Inflation Averae years of schoolin (2.67)** (2.11)* (0.07) (1.88)+ (0.25) (0.50) LAC Dummy (3.58)** Observations R-squared Notes: + sinificant at 10%; * sinificant at 5%; ** sinificant at 1%. Absolute value of robust t statistics in arentheses. All reressions include a constant not reorted here. The deendent variable is TFP rowth (in los) between 1995 and All exlanatory variables refer to initial 1995 values. For Trade/GDP, FDI/GDP, credit/gdp and inflation, averae values between 1990 and 1994 from World Develoment Indicators are used. Reulatory Quality is the 1996 value of the index reorted by Kaufmann et al. (2009). The technoloical creation index comes from Archibui and Coco (2004), while averae years of schoolin refer to the averae for the oulation older than 15 years from the Barro and Lee (2000) dataset. 20 OECD 2010

21 OECD Develoment Centre Workin Paer No. 288 Column 1 shows that an index of technoloical creation index by Archibui and Coco (2004), which basically reflects atents ranted er million oulation in the United States and scientific ublication er million oulation (i.e. knowlede outcome variables). Interestinly, TFP rowth is ositively and sinificantly correlated with this index. The imlied manitude is economically sinificant too; a one-standard-deviation imrovement in the index (e.. increasin from the level of Chile or Arentina to that of France) would imly an increase by 0.3 ercentae oints in the annual TFP rowth rate. Next, we consider two variables often related to international diffusion of knowlede: lobal trade and FDI links (for a survey see Keller, 2004). While we do not find a sinificant correlation with trade oenness, FDI shows a sinificant and ositive correlation with TFP rowth. The quantitative effect is similar to that of technoloical activity: a one-standarddeviation increase boosts annual TFP by 0.2 ercentae oints. While for rivate credit, we do not find a sinificant effect, we do so in the case of the reulatory quality, macro instability and uncertainty (roxied by the inflation rate) and human caital to analysis the caacity to absorb technoloies. Aain, in terms of economic manitudes an imrovement in any of these variables yields a similar effect on TFP rowth of around an additional 0.3 ercentae oints er annum. Table 2. TFP levels and exlanatory variables (2005) Variable (1) (2) (3) (4) (5) (6) (7) (8) (9) UNCTAD Technoloy Index (7.59)** (1.91)+ (1.66)+ Trade/GDP (%) (1.20) (0.53) (0.50) FDI/GDP (%) Private Sector Domestic Credit GDP Reulatory Quality (6.56)** (1.18) (0.59) (0.58) Inflation (2.06)* Averae year of schoolin (4.85)** (0.74) (0.72) 0.11 (8.99)** (1.99) (0.37) (1.85)+ LAC Dummy (0.14) Observations (0.96) R-squared (0.27) (0.98) Notes: + sinificant at 10%; * sinificant at 5%; ** sinificant at 1%; Absolute value of robust t statistics in arentheses. All reressions include a constant not reorted here. The deendent variable is TFP rowth (in los) between 1995 and All exlanatory variables refer to initial 1995 values. For Trade/GDP, FDI/GDP, credit/gdp and inflation, averae values between 1990 and 1994 from World Develoment Indicators are used. Reulatory Quality is the 1996 value of the index reorted by Kaufmann et al. (2009). The technoloical creation index comes from Archibui and Coco (2004), while averae years of schoolin refer to the averae for the oulation older than 15 years from the Barro and Lee (2000) dataset. OECD

22 Innovation, Productivity and Economic Develoment in Latin America and the Caribbean When all factors are considered jointly, all of them, excet for inflation, lose their sinificance, robably due to multicollinearity roblems. Of course, this simle exercise is too limited to identify causal links, but nevertheless it is interestin to oint out that all eiht variables (includin the initial TFP level) exlain only 15 er cent of the total variation in TFP rowth rates across countries. 16 Thus, the correlates suested by economic theory exlain only a little art of the disersion in rowth rates. In articular, as column 9 of Table 1 shows, the low rowth of TFP durin this eriod in Latin America and the Caribbean is not exlained by these correlates. When we include a dummy for the reion, it is neative and hihly sinificant, showin that the reion s undererformance is not catured by the variables considered. In Table 2, we follow a similar aroach, but considerin TFP levels instead of rowth rates. As the results show, there is a ositive and sinificant correlation between TFP levels and technoloical activity, the availability of domestic credit, reulatory quality, macroeconomic stability and human caital. When we include all of them toether, only the technoloy index and institutions remain sinificant at a 10 er cent level. It is interestin to oint out that in contrast to TFP rowth rates the correlates seem to exlain a lare fraction of the cross-country variation in TFP levels. R&D related activities, credit, institutions and human caital exlain over 30 er cent of the variation considered one-by-one. The correlates exlain jointly over 50 er cent. 17 This evidence coincides also with Klenow and Rodríuez-Clare (2005) who find that R&D sendin as a share of GDP correlates ositively with income er caita and TFP levels, but there is no sinificant correlation with TFP and income rowth. 18 They arue that the fact that income and TFP levels rather than rowth rates are more correlated to investment and R&D exenditure, combined with the observed lare co-movement in rowth rates, is evidence in favour of endoenous rowth models with international knowlede sillovers. Furthermore, it is interestin to oint out that the dummy for LAC is not sinificant when included with all other exlanatory variables. Based on column 8, the differences in R&D activity would account for around 38 er cent of the difference in TFP between the averae OECD country and the averae Latin American country in our samle, while differences in the institutional quality would account for around 26 er cent. As mentioned above, our measure of TFP would also contain differences in the quality of education across countries. While international comarable data on the quality of human caital is difficult to have, we can use PISA scores as a roxy of it. 19 In articular, in Fiure 10 we lot This contrasts with the much hiher R-squares commonly reorted in cross-country rowth reressions (see e.. chater 12, Barro and Sala-i-Martin, 2004). Of course, many of these variables could clearly be endoenous. For examle, in most models of endoenous human caital, investment in human caital is an increasin function of the level of roductivity. Similar aruments can be raised for R&D, or the demand for credit. They also resent similar evidence reardin hysical investment. Investment ratios are stroner correlated with income er caita (across countries and time) than with the corresondin rowth rates. The OECD Proramme for International Student Assessment (PISA) is an internationally standardized assessment that was jointly develoed by articiatin countries and administered to 15-year-olds in 22 OECD 2010

23 OECD Develoment Centre Workin Paer No. 288 the averae score in the test for 2006 aainst the TFP levels for There is a ositive correlation (with a coefficient of 0.65), such that as ointed out above, differences in the quality of human caital miht well be included in our measure of TFP. This seems to be in line with Hanushek and Woessmann (2009) who arue that the low quality of education is one of the drivers of low GDP er caita rowth in Latin America. However, PISA scores are not sinificant when included in multile reressions like those shown in Table 2. Furthermore, from quantitative viewoint, Caselli (2005) shows that the differences in the quality of education have to be unreasonable lare in order to exlain a sinificant fraction of the observed TFP as. Fiure 10. TFP levels and PISA scores Ireland Tunisia Chile Israel Hon Kon Italy France United States Belium Austria United Kindom Norway Sain Denmark Netherlands Australia Sweden Germany Finland Portual Canada Jaan Greece Hunary New Zealand Arentina Turkey Uruuay Mexico Colombia Brazil Korea, Reublic of Jordan Indonesia Thailand Averae PISA score 2006 Source: Own calculations based on Daude and Fernández-Arias (2010) and OECD PISA 2006 reort. III.2. Evidence from the diffusion of new technoloies In this subsection, we analyse the diffusion of new technoloies, to exlore further the issue of whether countries in the reion have a roblem in terms of absortion caacity or whether framework conditions reduce areate efficiency throuh less diffusion. In articular, we consider the use of the internet and mobile hone technoloies. Fiure 11 shows the evolution of internet users er 100 eole and mobile hone subscritions er 100 eole for Hih Income OECD (HI-OECD) countries and Latin America and the Caribbean (simle schools. PISA assesses how far students near the end of comulsory education have acquired some of the knowlede and skills that are essential for full articiation in society. In all cycles, the domains of readin, mathematical and scientific literacy are covered not merely in terms of mastery of the school curriculum, but in terms of imortant knowlede and skills needed in adult life. (for more details see htt:// OECD

24 Innovation, Productivity and Economic Develoment in Latin America and the Caribbean averaes). As shown in the left-hand anel, there is a considerable a in the diffusion of internet usae between both reions. As of 2007, the averae LAC country is at a state similar to the averae HI-OECD almost ten years ao, with a user rate of almost 30 er cent versus almost 40 er cent in HI-OECD. Furthermore, the sloes do not indicate any increased seed in the adotion of the internet in the last coule of years. This contrasts with the situation of mobile hones, where the reion seems to be catchin u raidly with HI-OECD countries. Of course, this difference miht be due to a lare variety of factors. For examle, the cost involved in the access to the internet miht be larer than buyin a mobile hone. Furthermore, the required skills to oerate a ersonal comuter miht be hiher than the skills needed to master cellhones technoloies. It is imortant to oint out that in rincile the diffusion and adotion of these new communication technoloies could have limited imact on economic rowth and develoment if they are just standard consumtion oods that do not increase the efficiency of labour or the flow of information. However, there is some recent evidence that broadband internet as well as mobile hones have a sinificantly ositive imact on economic rowth. Qian et al. (2009) find that a 10 er cent increase in broadband enetration increases rowth by around 1.4 ercentae oints, while for mobile hones the imact is somewhat smaller: 0.6 ercentae oints of additional GDP rowth iven an increase of 10 er cent in the enetration rate for develoin countries, accordin to Waverman et al. (2005) Fiure 11. Comarative diffusion of new technoloies across reions Internet users er 100 eole Mobile hone subscritions er 100 eole Hih Income OECD LAC Hih Income OECD LAC Source: World Develoment Indicators database. Next, we analyse the determinants of the diffusion of these relatively new technoloies, considerin an emirical model similar to Chon and Micco (2003) that includes the variables used in our analysis of TFP determinants and some additional variables of interest that have been indicated in the economic literature to have an imact on technoloical diffusion. Table 3 resents the estimates for the case of internet usae. The results show that human caital lays a sinificant role under all secifications considered. For examle, the estimates in column 1 imly that a one-standard-deviation increase in the averae years of schoolin in the 24 OECD 2010

25 OECD Develoment Centre Workin Paer No. 288 oulation (a chane that would brin Parauay to Arentina s level) is associated with an 8.4 ercentae oint increase in the ercentae of internet users in the oulation. Another relevant variable is the availability of fundin. Domestic credit to the rivate sector as a share of GDP is sinificant in all secifications, and imlies that a one-standard-deviation increase in financial develoment leads to a 5-6 ercentae oint increase in internet usae. Meanwhile, trade and FDI exosure, as well as market size (GDP) and macro stability do not have a sinificant imact on the diffusion of the internet across countries. Overall, the results in eneral are in line with a recent study by Chinn and Fairlie (2010) on this issue, in terms of the factors that matter to exlain the internet diffusion a. Table 3. Determinants of internet users er 100 inhabitants (2006) Variables (1) (2) (3) (4) (5) (6) (7) (8) GDP (lo) (0.29) (0.60) (0.02) (1.01) (0.27) (0.70) (0.08) (1.04) Averae years of schoolin Domestic credit to rivate sector/gdp (%) (4.22)** (2.09)* (1.71)+ (4.04)** (3.93)** (2.11)* (1.70)+ (3.52)** (2.31)* (2.27)* (1.94) (2.08)* (2.26)* (2.15)* (1.89)+ FDI/GDP (%) (2.07)* (1.23) (1.26) (0.85) (0.96) (1.36) (1.48) (0.96) (0.98) Trade/GDP (%) (1.18) (1.63) (1.32) (0.91) (1.23) (1.74)+ (1.38) (0.91) Inflation (0.42) (0.34) (0.23) (0.44) (0.49) (0.53) (0.31) (0.37) Reulatory Quality (3.84)** (3.40)** (2.96)** (3.39)** (3.66)** (3.17)** (2.90)** (3.35)** GDP er caita (lo) Procedures to start a business (2.31)* (2.26)* (2.87)** (2.28)* (2.75)** (1.92) (1.92)+ LAC dummy (1.79)+ (0.40) (0.88) (0.31) (0.19) Observations R-squared Notes: + sinificant at 10%; * sinificant at 5%; ** sinificant at 1%; Absolute value of robust t statistics in arentheses. All reressions include a constant not reorted here. Notes: All exlanatory variables refer to averaes between 2001 and 2005, with the excetion of Averae years of schoolin which refers to 1999 data. Procedures to start a business come from the World Bank s Doin Business database ( See Table 1 or 2 for the remainin variables. Institutional factors seem to be the most imortant to understand the different atterns in internet usae. The quality of reulations has a statistically and also economically sinificant imact, imlyin an increase by around 10 ercentae oints if reulations are imroved by one standard deviation (e.. imrovin from Peru s level to that of Sain). Furthermore, the costs of startin a business roxied by the number of rocedures to start a business have also a sinificant effect, althouh the quantitative imlications are smaller, with a one-standarddeviation reduction imlyin an increasin in the internet usae rate of around 3 ercentae OECD

26 Innovation, Productivity and Economic Develoment in Latin America and the Caribbean oints. 20 Finally, when includin a dummy for LAC, we do not find it to be sinificant. This imlies that the model which has a very ood overall fit does a ood job in exlainin the averae a between the reion and the rest of the world. This result contrasts with Chon and Micco s findin of a ositive and sinificant residual for LAC countries for the late 1990s. In the left-hand-side anel of Fiure 12, we show a decomosition of the a in internet enetration rates between OECD and LAC countries based on equation (1) in Table 3. As shown, the quality of reulations exlains almost 50 er cent of the a, while human caital and access to finance exlain each around a quarter of the a. Fiure 12. Decomosition of as between OECD and LAC countries Internet user a Finance 25% Human Caital 28% Others 0% Reulatory Quality 47% Human Caital 28% Mobile hone subscrition a Finance -1% Others 9% Reulatory Quality 62% Source: Own calculations based WDI, Doin Business and Barro-Lee datasets. In Table 4, we resent the equivalent estimates for mobile hones, which show some interestin differences comared to the case of internet users. First, althouh human caital is sinificant in some secifications, once controllin for the level of develoment (GDP er caita) it loses its sinificance. Second, financial develoment does not have any sinificant effect on mobile hones. Third, there is some evidence that lobal linkaes (trade) has a ositive effect on mobile hone enetration, althouh the effect is rather small (an increase of trade to GDP by 10 ercentae oints increases the enetration rate of mobile hones by 0.7 ercentae oints). Finally, reulation and institutional asects seem to be one of the main drivers of the cross country differences in mobile hone subscritions. In articular, as shown in the riht-hand-side anel of Fiure 12, the quality of reulations exlains 62 er cent of the remainin a in mobile hone subscritions between the OECD and LAC countries, while human caital differences exlain around 30 er cent. As already mentioned, in contrast to the case of the internet, access to finance does not lay a sinificant role. 20 While it could be arued that multicollinearity is a roblem here, iven that this variable miht be correlated with reulatory quality and GDP er caita, even excludin these two variables, raises the effect just to 4.5 ercentae oints. Actually, the correlation with GDP er caita and Reulatory Quality is just and -0.52, resectively. 26 OECD 2010