Research, Patents and the Financing of Ideas: Why is the EU Growth Potential so Low? Laura Bottazzi * Marco Da Rin ** Francesco Giavazzi * preliminary

Size: px
Start display at page:

Download "Research, Patents and the Financing of Ideas: Why is the EU Growth Potential so Low? Laura Bottazzi * Marco Da Rin ** Francesco Giavazzi * preliminary"

Transcription

1 Research, Patents and the Fnancng of Ideas: Why s the EU Growth Potental so Low? Laura Bottazz * Marco Da Rn ** Francesco Gavazz * prelmnary December 2001 Paper presented for the frst meetng of the Group of Economc Analyss of the European Commsson s Presdent. We thank Stefano Scarpetta for provdng us wth valuable data and Paolo Spada for excellent research assstance. (*) Unverstà Boccon, IGIER, and CEPR. (**) Unverstà d Torno and IGIER.

2 Executve Summary Knowledge matters for productvty, and productvty n R&D matters for the accumulaton of knowledge Lower output per worker n Europe, relatve to the US, does not depend on the avalablty of physcal captal. Both human captal and total factor productvty (TFP) are lower n Europe wth an mportant dfference: there s catchng up n human captal, whle n total factor productvty Europe s fallng behnd the US. The contrbuton of TFP to per-capta ncome n the EU, relatve to the US falls by 7 per cent between the 1980 s and the 1990 s. Thus the (small) ncrease n EU catch-up (relatve per-capta GDP ncreased from.79 to.82 n the perod) has been hndered by the slowdown n TFP. TFP depends on accumulated knowledge, as measured by the stock of valuable patents. In the producton of valuable patents Europe lags behnd the US not because expendture per worker employed n the R&D sector s lower (t s n fact twce as hgh as n the US), but rather because () the stock of accumulated knowledge s lower and () productvty n the R&D sector s lower. European productvty n the R&D sector s not only below the US level: t s also fallng behnd the US. Research and start-ups Qualty (measured by scentfc ctatons) s comparable to the US, but t does not translate nto a correspondng share of valuable patents, and n partcular t does not translate nto patents n the hgh tech sectors. The number and the degree of nnovaton of new start-ups are lmted. Important factors seem to be the exstence of regulatory mpedments and a form of organzaton where start-ups are more rarely ndependent and more often happen wthn the boundares of a larger establshed company ( ntra preneurshp vs entre preneurshp.) European venture captal Snce 1995 to 2000 the flow of venture captal nvestments n Europe has ncreased by a factor of 6, but the gap wth the US has become larger there nvestment has ncreased, over the same perod, by a factor of 24. The nature of venture captal n Europe s dfferent from the US: (1) the ndustry s at an nfant stage and thus lacks experence (for nstance the ablty to dentfy and nurture promsng deas); (2) contrary to the US, t s domnated by banks. The outcome s that quoted ventured-backed companes do not grow faster than non venture-backed ones. Ths, however, s not enough to conclude that natonal nnovatve capacty s hampered by a (venture captal) fnancng constrant: t could well be that the flow of deas that look for fnancng s scarce. (Tentatve) polcy mplcatons There seems to be somethng wrong n the productvty of European research: expendture s hgh but the producton of knowledge (at least n those areas that are more drectly valuable for mprovng the growth potental) t s a thrd than n the US. Two factors seem to be crtcal n an effort to ncrease the EU growth potental: the organzaton of research and the organzaton and fnancng of start-ups although ths paper s admttedly unable to wegh ther relatve mportance. 2

3 1. Introducton One of the man tenets of growth theory s that economes may experence perods of fast growth for two, qute dfferent, reasons. One s convergence towards hgher levels of per capta ncome through the accumulaton of physcal and human captal; the other s acceleraton n technologcal progress (see e.g. Barro and Sala--Martn, 1995). Convergence s typcally assocated wth 'catch-up' economes, whereas advanced economes typcally grow at sustaned rates only by explotng technologcal breakthroughs. Over the last century, for example, the US has gone through three perods of accelerated growth (Jovanovc and Rousseau, 2001). Growth durng the reconstructon whch followed World War II was manly convergence. The other two perods of fast growth occurred n correspondence of major technologcal nnovatons: electrcty at the turn of the XIXth century (the 'second Industral Revoluton') and nformaton and communcaton technology (ICT) almost one hundred years later. The extent to whch the recent 'jump' n technology has actually contrbuted to growth n productvty and to per capta ncome levels s stll to be agreed upon (Gordon, 2000, Jorgenson and Stroh, 2000), but a consensus s now emergng whch vews the 'IT Revoluton' as a major dscontnuty, at least for the US economy (Olner and Schel, 2000). The acceleraton of potental output growth n the Unted States n the 1990s, especally n the second half of the decade, has attracted much attenton because t entals pushng farther the producton possblty fronter (Scarpetta et al., 2000). Moreover, the U.S. experence contrasts wth the pace of potental output growth n Europe. In what follows we provde an nterpretaton for the reasons underlyng Europe s dsmal performance. (The same lne of reasonng has been followed n the 2001 European Compettveness Report, publshed by the Commsson n November 2001). We start from the economc concept of new deas, whch we measure as patents. These, n turn, are accumulated nto a stock measure knowledge. Whle knowledge s potentally useful for ncreasng the economc effcency of producton processes, t actually does so only through nnovaton, that s through the creaton of new frms. We therefore explore the lnks between knowledge, nnovaton, and growth. We start by decomposng per-capta GDP n the major ndustral countres and we thus retreve a measure of total factor productvty (the Solow resdual). We then show that knowledge (whch we measure wth the stock of ctatons of a country s patents that are regstered n the US) s an mportant determnant of the Solow resdual, and therefore of the dfferences n growth rates not accounted for by changes n the accumulaton of physcal and human captal. Whle ths s a well-known result n the lterature, we venture one step further, and look at how countres dffer n ther ablty to create new knowledge. We then ask why countres dffer n ther ablty to turn knowledge nto commercal products through the creaton of new, successful frms. We (tentatvely) dentfy the organzaton and the fnancngofeuropeanresearchastolkelycanddates. 3

4 2. Why s the EU growth potental so low? Much of the recent emprcal growth lterature has been devoted to explan the enormous dsparty of per capta ncome that s observed around the world. These dspartes--substantal and persstent over tme--run aganst many predctons of classcal growth theory. Notwthstandng the amount of research undertaken, a full explanaton of ths phenomenon has so far eluded economsts. The growth patterns of OECD countres over the last quarter century are helpful n llustratng ths pont. A major research project under way at the OECD has been lookng nto the determnants of economc growth over the last quarter century and of ts dspartes across member countres (see OECD, 2001a and Scarpetta et al. 2000). Ths secton brefly summarzes the results from the OECD project and of other recent emprcal studes, and puts them n perspectve. The ncreasng dspartes n trend (.e. cyclcally adjusted) growth rates across countres have been extensvely documented (see OECD 2001a). At a general level a consensus s emergng that the convergence story holds up well for countres whch started relatvely poor, and whch, over the last two decades, have caught up wth technologcally more advanced economes. However, convergence does not explan the sustaned growth rates of the Unted States, whose ncome levels, but also productvty levels, where hgh to begn wth. A dsparty has also emerged between the US and the European economes, snce the former has kept pushng ts growth potental farther, whereas the latter has fallen back and experenced dsmal growth rates. Changes n labor utlzaton help explan why trend growth fell back n some countres. For nstance, Italy, France and Belgum have ncreased ther labor productvty but ther employment rates have remaned low and workng hours have become shorter (OECD, 2000a,b). Dfferences n the utlzaton of captal have also played a role, snce the US has been the only major OECD economy to ncrease ts busness nvestment rate n the 1990s, wth a sharp upturn n the last fve years of the decade (OECD, 2000b). Also the change n the qualty of captal, n the form of technologcal progress emboded n ICT captal goods, has contrbuted to recent growth patterns. The rate of nvestment n hardware, for example, has nearly doubled between the 1980s and the 1990s (Coleccha, 2001), and a smlar pattern holds for software (OECD, 2001b). Changes n the qualty and quantty of labor and captal employed cannot, however, fully explan the pattern of varaton n growth rates. Estmates of total factor productvty (TFP) growth rates whch measure the porton of growth not accounted for by changes n captal and labor vared substantally across countres over 1980 to 1999 (Scarpetta et al., 2000). The US and Canada experenced ncreases of TFP growth over tme, rrespectve of how ths s measured, whereas France and Italy, for example, experence an unequvocal deteroraton. Other countres went through dfferent patterns, whch often change dependng on the way TFP s measured. In all cases, though, the contrbuton of TFP to growth has been far from neglgble. Attenton has therefore focussed on three factors whch are thought lkely to have nfluenced TFP n the recent past: ICT, technologcal nnovaton, and organzatonal changes. ICT has been sngled out as the most lkely canddate, snce technologcal progress n ths feld has been extremely fast and snce ICT nvestment has greatly accelerated n the 1990s. ICT goods contrbute to growth by ncreasng the qualty of the captal stock (Whelan, 2000). They also create benefts whch exceed ther purchase cost ('spllovers') and benefts whch extend beyond the frms whch use them ('externaltes'), as documented wth frm-level data by Brynjolfsson and Kemerer (1996) and Brynjolfsson and Htt (2000) and wth macro data by Olner and Schel (2000) among others. 4

5 Technologcal nnovaton, whch transforms scentfc and technologcal deas nto new products and processes, s very hard to measure. Common measures nclude nputs, lke R&D expendture or personnel, and (ntermedate) output, lke patents or ctatons. Whle nnovaton ncreases the qualty of physcal captal, recent research has started to look at the lnk between nnovaton and TFP. The European Commsson (2001) documents that the correlaton between nnovaton and productvty growth s statstcally weak f we consder drect measures of R&D (expendture, R&D personnel share of the labor force, patents, and publcatons per nhabtant), but becomes sgnfcant f we look at 'structural' measures lke PCs or Internet connectons per nhabtant, the share of frms engaged n cooperatve agreements, and the share of frms wth a contnuous engagement n research. Fnally, organzatonal changes, whle even more dffcult to measure, may also contrbute to TFP snce they change the way factors are combned. Entrepreneurshp s often consdered an mportant drver of organzatonal change, and has started attractng attenton from researchers (OECD 1998). 3. The producton of knowledge A smple growth accountng framework s useful to dentfy the reasons behnd dsparty of per capta ncome that s observed around the world. Wth growth-accountng, whch orgnates from the semnal work of Solow (1957), we can decompose the drvers of growth nto several factors: the amount and qualty of labor, the amount and qualty of captal, and the effcency wth whch these factors are combned ('Total Factor Productvty,' or TFP). TFP comprses technologcal and organzatonal progress as well as all other unmeasured or unaccounted factors. By usng growth accountng, we therefore decompose dfferences n output per worker nto dfferences n nputs and n productvty. The breakdown suggested by the aggregate producton functon s just the frst step n ths drecton. Assume that output Y n country s produced accordng to: α 1 α = ( ) (1) Y K BH Where K s the stock of physcal captal, H s the amount of human captal and B s a laboraugmentng measure of productvty. Ths equaton allows us to decompose dfferences n output per worker across countres nto dfferences n the captal-output rato, dfferences n educatonal attanment, and dfferences n productvty. We can then compute the level of productvty drectly by rewrtng the producton functon n terms of output per workers as n Hall and Jones (1999) (HJ): y æ K ö α 1 α = ç hb Y è ø (2) where h s human captal per worker. Early emprcal work by Barro (1991) and Mankw, Romer and Wel (1992), among others, demonstrates that accountng for levels of educaton goes a long way towards explanng the observed cross sectonal varaton n per capta ncome. Table 1 decomposes output per worker n each of the ffteen countres of our sample nto the three multplcatve terms of equaton (2). All terms are expressed as ratos to US values for ease of comparson. We use as a proxy for the human captal stock the average years of educaton n workng-age populaton, as n Donenech-de la Fuente (2000) and Scarpetta and Bassann (2001). They have constructed these tme seres to be consstent over tme, whch represent an mportant mprovement over the Barro-Lee estmaton of human captal. Data 5

6 show an upward trend n human captal for all OECD countres n our sample and a steady convergence to the US level. We have expressed the captal stock as a rato to the GDP to avod possble msnterpretaton: If a country experences an exogenous productvty shock that does not affect the nvestment rate of the economy, the ncrease n output per worker mght be erroneously attrbuted to the rse n the captal over labor rato that results. Table 1 shows that the dfference n output per workers among countres are due to dfferences n human captal per worker and to dfferences n productvty. The captal ntensty of most of the economes we consder s nstead hgher than that of the US. However dscrepances n physcal captal explan only a modest amount of the dfferences n output per worker snce t s only the square root of that dfferences that matters for the rato of output per worker 1. At the bottom of Table 1 we report the average and the standard devaton of the contrbuton of nputs and productvty to dfferences n output per worker, through tme. Accordng to ether statstcs average human captal has ncrease but the average level of productvty has not. For nne out of the ffteen countres productvty declned after 1990 vs-a-vs US. Our results dffer (slghtly) from the one obtaned by Hall and Jones (1999). The man dfference rests on the data we have used and n partcular on the seres on average human and physcal captal 2.However HJ decompose output per worker only for the year 1988 whle the tme dmenson of ths decomposton appears partcularly nterestng. From Table 1 we nfer that t s hard to escape the concluson that dfferences n productvty play an mportant role n creatng a dfferences n output per worker across countres. But why does productvty dffer so much across countres? One central dea of our report s that one of the fundamental determnants of a country long-run performance s ts ablty to create new productve deas. 1 Dfferences n nvestment rates get rased to the power of α/1-α. Snceα s 1/3, α/1-α s equal to ½. 2 Scarpetta et al. (2000) document the great extent to whch the mpact of labor on growth per capta depends on the adjustements for labor force partcpaton rates, employment rates, hours worked, etc. Even more dffcult s the measurement of the captal stock, whch depends on assumpton on ts composton, on deprecaton rates, on the amount of servces the stock s contrbutng to the economy. 6

7 Table 1 : Productvty Calculaton, ratos to US values Y/L=B(K/Y) α/1 α (H/L)) Y/L (K/Y) α/1 α h B DENMARK FINLAND FRANCE GERMANY ITALY NETHERLANDS NORWAY SPAIN SWEDEN UK European Average Standard dev USA AUSTRALIA CANADA JAPAN Source: OECD Statstcs and authors calculatons.

8 3.1 Knowledge and Total Factor Productvty Knowledge s a key factor, n the long run, to determne productvty growth. Therefore we can thnk that a country s Total Factor Productvty ( B ) s a functon, among other thngs, also of nnovaton. Consderng the log-lnear form of the TFP functon we can wrte: log ( TFP ( t) ) = a+ β log ( A(t) )+ θ log ( X (t) ) Where X (t) are country characterstcs whch could affect productvty, whle A (t) s the stock of knowledge accumulated by each country. If nnovaton s mportant for productvty growth we should observe that levels and changes n A (t) are correlated wth levels and changes n TFP. In order to control for country- and tme-specfc factors we estmate the partal elastcty of TFP (t) wth respect to A (t) controllng for country and tme effects as follows: log( TFP ( t ) = b log (A t)) + µ + θ (t) + u (t) The estmate of such elastcty b s reported n Table 2, where the stock of knowledge A (t) has been measured as the smple patent count 3, and where we have calculated TFP as n the prevous paragraph,.e. usng the U.S. shares for labor and captal ncome. 4 We employ three year averages of labor adjusted for the accumulaton of human captal (we obtan very smlar results even usng unadjusted labor stock). As we are usng only the most developed countres n the world ths assumpton does not seem out of place. We use the unverse of granted patents, released by the Unted States Patent offce between 1963 and These data are from the REI Data-set, verson 2000, compled by Jaffe and Trajtemberg. Patents are assgned to countres accordng to the resdence of ther frst nventor, and to a year accordng to applcaton dates (not the year n whch the patent was granted) as we are nterested n capturng the producton of an dea when t frst come around. The total number of these patents s about three mllons, and beng the Unted States the largest and most compettve economy, the nventons wth the potentally most proftable content are lkely to be patented wth the US Patent Offce. 3 To correct for the dfferent mportance of deas, we also defne the stock of knowledge correctng the count of patents by the yearly average ctatons that they receved n the frst three years. The stock of knowledge at the begnnng of the perod for whch we have data s - Pat (0) Pat (0) smply computed as A (t(0)) = å (1-δ ) = (1- g ) (g + δ ) 0 whch uses the growth rate of patentng n the country n the frst fve years and the deprecaton rate, assumng that patentng has been growng n the past at that average rate. Ths ntal stock s at best a rough estmate snce our data allow to start the computaton n To calculate TFP t s necessary to have sutable measures of partal output elastctes. Snce the latter are not observable the lterature has assumed them to be equal to ncome shares. Ths mples acceptng that product and mport market are perfectly compettve and that elastctes are constant along the whole perod and equal to the observed average. However drect estmaton rases a number of econometrc ssues that put nto queston the robustness of the results.

9 Table 2: Knowledge and TFP Dependent varable TFP = log(y) log(k) log(h) Log(A(t)) Country dummes Tme dummes R 2 Observatons 0.08 (0.018) Yes Yes As we nclude country and tme dummes, we are suffcently confdent that major cross-country dfferences n nsttutonal settngs and poltcal nsttutons s captured by these varables, such as common trends towards larger nternatonal ntegraton. The elastcty, dentfed, s nevertheless sgnfcant and equals As we fear that the yearly data mght dentfy the elastcty on short-term fluctuatons, we estmate the parameter wth the observatons averaged over three year ntervals. We take our results as evdence of the mportance of local nnovaton n leadng to a technologcal advantage,.e. that the level of knowledge related to nnovatve actvty mprove the productvty n the country. Total factor Productvty s, among other thngs, also a functon of local nnovaton. What determnes then each country ablty to create new deas and mprove the stock of knowledge? 3.2 The producton functon of deas We thnk that the proper way to answer ths queston must take two ssues nto account. The frst ssue means capturng nnovaton as the contrbuton to the stock of knowledge, whch requres to have a measure of ths stock and of how t evolves. The second ssue nvolves consderng, an aggregate 'producton functon' of new deas. Therefore we consder knowledge just as a partcular form of captal, whch s accumulated due to nnovaton, deprecates wth obsolescence and needs some nputs to be created. Our fndngs regard the creaton of knowledge n our sample of countres, ts dynamc behavor and the returns to factors n ts creaton. The lnk between the creaton of knowledge and productvty growth wthn countres wll be a second step n our nqury. Dffcultes arse from the lack of consensus on how to measure knowledge, from the less than precse data on R&D nputs, and from absence of any theoretcal predcton on what aggregate returns to scale should be n ths case. 5 In the extant lterature the only specfc contrbutons to the estmaton of a producton functon of deas across countres or regons are those of Porter and Stern (2000), Bottazz and Per (2001) and Porter, Stern and Furman (2001). We follow Bottazz and Per, and estmate the returns to factors n generatng new deas n the way n whch returns to factors n the aggregate producton functon are usually estmated. We therefore consder a producton functon of new deas where these are generated on the bass of R&D resources, human 'brans' and the stock exstng of Ideas. In order to focus on potental local spllovers whch are those whch would generate dvergence of country nnovaton rates, we consder as nput of the functon the knowledge generated wthn a country. The total stock of 'useful' deas n a country ncreases each perod due to the generaton of new deas but t also decreases due to the 'obsolescence' of exstng deas. Except for a few basc nventons, whch 5 The dea of constant returns due to replcaton, whch we use for the producton functons of good does not apply n ths case. 9

10 are probably ncorporated n later mprovements, the stock of knowledge useful for the producton of new deas depends on the amount of research. At a very general level we mght represent the change n 'non-obsolete' knowledge A(t) for a country as: A(t) = A(Brans, R & D spendng, A(t)) - Obsolescence) where the producton of new deas makes use of nputs such as the 'brans' (educated workers n R&D), the amount of resources spent n R&D, and the exstng (net) stock of knowledge 6. Obsolescence s the process by whch some deas get useless, ether because a more refned verson of the dea has been dscovered or because new versons of an dea have encompassed older ones. In order to construct the stock of avalable knowledge we consder the cumulatve amount of new patents, and assocate to each nnovaton the year the frst nventor appled for the patent. Obsolescence, on the other hand, s somewhat harder to capture. Part of the lterature focuses on 'creatve destructon,' whch holds that some deas become obsolete mmedately after the arrval of better ones (Aghon and Howtt (1992), whle some studes vew new deas as a complement to exstng ones (Stokey (1991), or even as unaffectng the value of exstng deas (Romer (1990). We approxmate obsolescence wth the same process used for the deprecaton of physcal or human captal. We choose a deprecaton rate whch s n the range of those estmated usng data on patent-ctatons (Caballero and Jaffe (1993) by choosng a deprecaton rate δ= We construct the stock of knowledge usng as a measure of new deas a pure patent count so that we consder the ncrease n useful knowledge as the count of patents n one year net of the deprecaton. To correct for the uneven mportance of dfferent deas, we correct the raw count of patents by the average yearly number of ctatons a patent receves n ts frst three years. We construct the stock of knowledge by usng the followng recursve equaton: Stylzed facts A (t + 1) = Pat (t + 1) + (1-δ )A (t) Before estmatng the deas producton functon we consder some stylzed evdence of the dynamc behavor of the stock of knowledge A (t) across countres and over tme. The evoluton of knowledge generated n each country should lead countres whch beneft from a larger pool of deas 'n the ar' to produce more new, for gven resources, and to exhbt hgher nnovaton rates. In Fgure 1 we plot the tme behavor of knowledge stock per employee n the largest ffteen patent producng countres, whch account for 96% of the world patentng. 6 The data on (tme-equvalent) research employees and on spendng n R&D (n constant dollars), are from the ANBERD and OECD databases. The data on total employment are from OECD. 7 We have also performed some robustness check by allowng the coeffcent of obsolence equal to 0.15 and 0.2. Our results does not change sgnfcantly. 10

11 Fgure 1: Knowledge per worker n 15 OECD countres, AUSTRALIA CANADA 0.8 DENMARK FINLAND FRANCE 0.6 GERMANY ITALY JAPAN 0.4 NETHERLANDS NORWAY SPAIN SWEDEN 0.2 UK USA Source: OECD Author's calculaton Source: OECD Statstcs and authors' calculatons

12 Consderng the dsperson across levels of knowledge per worker, the standard devaton of the dstrbuton has decreased from 1.08 n 1973 to 0.86 n It s apparent that countres such as Span and Fnland, startng wth low level of knowledge per capta have accumulated much faster than countres such as Germany and the US, whch were startng from a hgh level of knowledge. These facts casts already some doubts on the possblty of ncreasng returns to A(t). Ths suggests that the level of knowledge producton mght depend on the levels of R&D resources, so that there s a postve relatonshp between ther respectve growth rates, just as our model suggests. The dea that 'we buld on what we know' or, sayng t n a dfferent way, that knowledge s a publc good, has been powerful enough to brng people search at both the mcro and the macro level for the exstence of spllovers. Ths ssue has been extensvely analyzed n Bottazz and Per (2000) and t s not the object of our report. We are here nterested n estmatng returns to knowledge by frst lookng for drect evdence on the deas generatng process, and then by lookng at how productve deas translate on productvty growth through the creaton of new busnesses. Therefore there are two dfferent processes, one of generatng new deas and another of translatng these deas nto productvty gans. Ideas are patented as the nventor expects potental profts from them, but how these patents translate nto productvty ncreases s a second and separate step of the process. We turn to the frst ssue Estmates We focus on the producton functon of new deas (nnovaton), representng t as a Cobb- Douglas functon: λ α φ A(t) = S b(t) k(t) A(t) - δ A(t) The three factors dentfed as the nputs n the producton of new des are b(t), the total number of sklled workers employed n R&D (ths varable are the ''brans'') k(t), the amount of spendng per worker or average level of physcal-human captal endowment of a worker n R&D and A(t) the total stock of exstng useful deas whch could be used to mprove on exstng knowledge. S s a shft parameter that measures the natonal productve capacty n creatng new deas. The producton functon of new deas can then be wrtten as: A(t) α + δ = S b(t) A whch s the equaton we estmate ( n logs). We proceed by frst testng for statonarty of our data snce the asymptotc dstrbutons of estmators n panel regresson are affected by the presence of unt roots and contegraton. 8 Specfcally we test for non-statonarty aganst the alternatve that varables are trend statonary, where we allow for dfferent ntercept and tme dummes for each country. We use the unt root test proposed by Im, Pesaran and Shn (1995) whch allow each panel member to have a dfferent autoregressve parameter and dfferent short run dynamcs under the alternatve hypothess of trend statonarty. Before carryng out the tests, the data are purged of any common effect across countres. In no case can we reject the null hypothess that every countres has a unt root for the seres of n log levels. k(t) β A(t) φ 1 8 An mportant concern n testng for unt root s that, n small samples, the tests lack power. We address ths problem by takng the general form of the producton functon to be applcable to all members of the panel, but we do not mpose that the same parametrzaton apples unformly to all countres. Ths poolng condton allows us to substantally ncrease the power over tradtonal unt root and contegraton tests.

13 We estmated our producton functon of new deas usng panel contegraton. Contegraton between the creaton of new deas and worker n R&D, expendture per worker n R&D and the accumulated stock of knowledge means that there exst an error correcton mechansm, wth one of the mentoned varables adjustng to keep the long run equlbrum ntact. One way to thnk at ths relaton would be for some exogenous force to drve the ncrease n a natonal nnovatve capacty and to R&D expendture or workers to respond through a demand mechansm. Table 3: The producton functon of deas Dependent varable Log((dA(t)/A)+ δ) Log(A(t-1)) (0.069) Log(k(t-1)) 0.19 (0.073) Log(b(t-1)) 0.28 (0.093) Country dummes Yes Tme dummes Yes Prob(F-statstc) Observatons 300 Table 3 reports our results. We then decompose the creaton of new deas per worker n R&D n each country nto R&D captal ntensty, knowledge per worker and dfferences n productvty n the creaton of new deas (S). We use our estmates to obtan the value of the elastctes α and β: 9 β 1 α At () æ kt () ö æ At () ö = S ç ç bt () è At () ø è bt () ø Table 4 reports our decomposton and, to make comparsons easer, all terms are expressed as a rato to US values. Accordng to ths table deas per workers n Japan s about 76 percent of that n the Unted States. Japan has a hgher captal ntensty than the Unted States but only 80 per cent of knowledge per worker. Dfferences n nputs partally explan the lower Japanese creaton of new deas per worker, so Japan productvty s lower than the US productvty. Other OECD economes have much lower productvty levels than US. The average value of new deas to worker n Europe s close to 0.4 that must be compared wth the value of 1 for US. Interestngly n the producton of new valuable patents Europe lags behnd the US not because expendture per worker employed n the R&D sector s lower (t s n fact twce as hgh as n the US), but rather because the stock of accumulated knowledge s lower and productvty n the R&D sector s lower. European productvty n the R&D sector s not only below the US level: t s also fallng behnd the US. 9 In order to obtan the above decomposton we have mpose the condton that φ=1-α-β. 13

14 da/b=s(a/b) 1 α (k/b) β Table 4 : Patent productvty calculaton, ratos to US values da/b (A/b) 1 α (k/b) β S DENMARK FINLAND FRANCE GERMANY ITALY NETHERLANDS NORWAY SPAIN SWEDEN UK European avg Standard dev USA AUSTRALIA CANADA JAPAN USA Source: OECD Statstcs and authors' calculatons

15 4. Translatng knowledge nto growth Our econometrc exercse shows that knowledge s an mportant determnant of the Solow resdual but also that countres possess dfferental abltes n the producton of new deas. The next step conssts of gettng a good grasp of the process whch leads from the accumulaton of knowledge to the creaton of new busnesses and, through ths, to economc growth. What we need to understand s: () what determnes the dfferences n S n Table 4.e. what determnes the productvty of R&D; () what determnes the transformaton of new deas nto growth.e. natonal nnovatve capabltes. Unfortunately, economcs stll falls short of provdng clear gudance on ths process. On both counts, one would need mcro-level data to shed lght on these processes, whch requre unfoldng the process whch leads to the resduals n Tables 1 and 4. Rgorous economcs research has only just started tacklng these ssues. Most studes lnk nnovaton and growth at a macro level, where research hts the boundares of dmnshng returns. We therefore proceed to organze the (scant) avalable evdence n order to provde some gudance towards more nformed polcy makng. Economcs suggests at least two factors that may be responsble for the ablty to translate deas nto growth: the organzaton of research and ts fnancng. 4.1 The organzaton of research One ntrgung aspect of the nadequacy of European natonal nnovatve capabltes s the 'European Paradox,' frst dentfed by the EU Commsson's 1993 Green Paper on Innovaton. The Paradox conssts of the nablty of European frms to translate scentfc competence nto commercally successful nnovatons. The scentfc competence of Europe s reflected, for example, n the number of scentfc publcatons for mllon populaton, whch n 1998 was 609 for the EU, versus 739 for the US and 480 for Japan (European Commsson, 2000). The good scentfc performance of Europe n terms of pure scence s also reflected n the annual growth rate of the ts share of both publcatons and ctatons n the frst half of the 1990s: the former was 2.8% for the EU, -0.3 for the US, and 2.8% for Japan, whle the latter was 2.8% for the EU, -0.6 for the US, and 0.8% for Japan (European Commsson, 2000). However, sheer scentfc competence does not automatcally translate nto valuable technologcal deas. In 1998 the number of patents per mllon populaton whch were granted to an nnovaton by all the three patent offces (US, EU, and Japan, the 'trad' patents) was 88 for Japan, 49 for the US, and only 32 for the EU. Ths was mrrored by the exports of hgh-tech products per mllon populaton: 667 mllon euro for Japan, 571 for the US, and 301 for the EU (European Commsson, 2000). Another tellng fact emerges from the pattern of patents n hgh-tech felds. The share of patents granted to US and Japanese frms n hgh-tech felds s hgher than ther share of all patents, as emerges from Table 5 below.

16 Table 5: Shares of hgh-tech patents EU-15 US Japan EPO ( ) USPTO ( ) all Hghtech all hgh-tech 45% 34% 15% 11% 27% 40% 55% 56% 18% 26% 20% 27% Source: European Commsson (2000) The 'technologcal balance of payments,' defned as the balance of payments n hghtech goods, s another ndcator of the unsatsfactory performance of the EU. It s n fact nterestng to partton the growth rate of hgh-tech exports n the two perods and (see Table 6): whle the US nearly doubled ts exports growth rate between the two perods, the EU bascally stagnated. Table 6: Hgh-tech balance of payments, annual growth rates EU-15 US Japan % 12.0% 8.4% 15.5% 11.6% -0.5% Source: European Commsson (2000) What does ths apparent nablty to translate deas nto commercally successful nnovatons depend upon? Here the organzatonal dmenson of nnovaton may play a role. Europe s characterzed by a relatvely large degree of ntrapreneurshp,.e. creaton of new busnesses wthn establshed frms rather as ndependent start-ups. Ths contrasts wth the US pattern, where the entrepreneural channel has become partcularly mportant wth the acceleraton of nnovatve actvtes and the shortenng of product and process lfe-cycles (Baldwn, 1995). A lnk between entrepreneurshp and nnovaton has been documented by Audretsch (1995), among others. For the perod he fnds that the share of start-ups (new frms) s hgher n scence-based, fast-growng ndustres (such as chemcals, drugs, communcatons equpment, electronc components, computers, and scentfc nstruments) than n the more tradtonal and stable ndustres. Hs econometrc analyss also shows that start-ups are more frequent n ndustres wth hgh small-frm nnovaton rates and low total nnovaton rates. In other words, entrepreneurshp s hgher n ndustres where knowledge s less routnzed and technologcal opportuntes hgher such as scence based ndustres. 16

17 Acs and Audretsch (1990) show that nnovaton by small frms s more common n scence-based ndustres, where small frms contrbute a substantal share of total nnovaton. These ndustres are also characterzed by hgher attrton rates: there are more opportuntes for entry by new start-ups, but also lower chances of survval. In fact, there s consderable debate on the net contrbuton of entrepreneural companes to job creaton and economc growth, and a consensus has not yet been reached (see Audretsch and Thurk,1999). European polcy makers have extensvely dealt wth the obstacles to entrepreneural actvty (see the recent European Commsson's 'Innovaton n a Knowledge-based Economy' Communcaton). Here much has already been sad (OECD,1998) and we shall only touch upon the key ssues: competton: t has been shown that the nataltly rate of start-ups depends, among other thngs, on the degree of competton of an ndustry (Acs and Audretch, 1988). Therefore competton polcy does (ndrectly) contrbute to nnovaton; ntellectual property rghts (IPR): we know that nnovaton depends on frms' ablty to reap the benefts of costly research and of the development of new products and processes. Therefore the desgn of property rghts s an mportant determnant of the accumulaton of entrepreneural captal (see for example Eaton (1997)). The recent ntroducton of the European Patent has been an mportant step towards the harmonzaton of IPR legslaton---a sngle IPR system favours the commercal explotaton of deas; taxaton: there are several dmensons along whch taxaton affects nnovaton. One s the degree to whch taxaton affects the choce between dfferent forms of busness, such as ncorporaton vs partnershps. Another s burden of hgher labor taxes on labor and non-labor ntangble nvestments. Fnally, one should also consder the mpact of taxaton on the supply of both rsk captal and entrepreneurshp, especally through the captal gans tax; regulatons and legal barrers: cumbersome admnstratve practces make t costly to set-up a frm. Several OECD studes document a large varaton of such costs across countres (see OECD, 1998). Bankruptcy procedures are also mportant as they determne the cost of defaultng, and therefore of extng a busness. Bankruptcy rules, together wth the effcency of legal enforcement, also affect the cost of credt, and therefore the cost of nnovaton. Gromb and Scharfsten (2001), for example, show that the prevalence of ntrapreneurshp n Europe may depend also on the type of bankruptcy rules. 4.2 The fnancng of research Once deas are ready to be translated nto entrepreneural frms, these stll need fnance to grow and unfold ther growth potental. Avalablty of fnancal captal s crucal at ths stage. There s n fact a growng percepton that more than labor or product market rgdtes t may be the lack of adequate fnancng, both rsk captal and debt, that prevents Europe from producng a strng of successful entrepreneural stores such as those whch have been seen n the US. 17

18 The mportance of fnancal development for economc growth has been documented for several countres by Rajan and Zngales (1998). The lack of development of European captal markets s well known, but also the fragmentaton of credt markets has been shown to constran frms lendng (Detragache, Garella, and Guso, 1998). Two facets of the European stuaton seem partcularly troublng. One s the sheer lack of fnancal resources avalable for start-ups, especally for nnovatve, rsker frms. For smaller and younger companes, whch rely on external fnance at the early stages of ther lfe, ths consttutes a serous problem: Guso (1998) shows that European hghtech companes suffer from substantal credt ratonng. The other reason for concern s that the scarcty of rsk captal, of venture captal n partcular, may harm the nnovatve ablty of start-ups. Venture captal has been shown to beneft start-ups beyond the supply of fnance. In other words, there s a soft sde to venture captal that adds value to the hard fnancal sde: venture captalsts are often a coach for entrepreneural start-ups (Hellmann, 2000). In fact, Hellmann and Pur (2000, 2002) forcefully argue that venture-backed companes n Slcon Valley are faster n developng ther products and brngng them to market, and need less tme to hre professonal management. Smlarly, Kortum and Lerner (2000) argue that venturebacked companes n Massachussets develop a larger number of patents and more relevant patents than other start-ups. There s thus evdence that, at least n the US, venture captal adds to natonal nnovatve capabltes by affectng both the effcency of the knowledge producton functon (more and better patents per gven nputs) and the overall TFP. Several offcal polcy documents have addressed the obstacles to the flow of fnance towards entrepreneural frms n Europe. Most mportantly, the EU Commsson s 1998 Communcaton Rsk captal: A Key to Job Creaton n the European Unon, formulated a polcy agenda whch dentfes sx man obstacles for the development of Europe s supply of rsk captal to entrepreneural companes: fragmentaton of captal markets, regulatory barrers (such as lmts to nvest n rsk captal by nsttutonal nvestors), taxaton of captal gans, paucty of technologcal start-ups, lack of human captal among both venture captalsts and entrepreneurs, and cultural barrers. Smlar prescrptons have been put forth by other documents, up to the recent Lamfalussy Report., whch proposes a new procedure for the elaboraton of the Prospectus Drectve. The supply of fnancal captal towards entrepreneural companes s rghtly n the focus of the Commsson. When one looks at the statstcs for European venture captal and takes nto account venture captal proper only.e. dsregardng funds whch go nto management buy-outs and other prvate equty deals for mature companes a dscomfortng pcture emerges. Fgures 2 and 3 compare the development of funds rased and nvested nto venture captal n the EU and n the US over the 1990s. 18

19 Fgure 2: VC funds rased n the EU and US Source: authors' calculatons on EVCA and NVCA data. Data n mllons of current US dollars. Whle n the US the amount of funds rased ncreased an mpressve eghtyfold, n Europe t ncreased only twelvefold. In both economes growth was more pronounced n the latter part of the decade. Fgure 3: Venture captal nvestment n the EU and US Europe USA 19

20 Table 3 reveals that the pcture s even gloomer f we look at nvestment. Throughout the decade venture nvestments n the EU have grown a mere quarter of US nvestments. Therefore, the gap between Europe and the US has wdened, contrary to a popular belef. Table 7: Venture-backed companes n the EU and US US EU ,088 n.a ,294 n.a ,150 n.a ,191 n.a ,325 n.a ,002 n.a ,697 3, ,149 5, ,969 7, ,412 9,574 Source: EVCA and NVCA. Table 7 shows that the number of venture-backed companes s hgher n Europe than n the US. Ths means that the average fnancng nvested to start-ups s much lower n Europe than n the US. These patterns probably reflect the structural dfferences that exst between venture captal on the two sdes of the Atlantc (Lander 2001a). For nstance, the US venture captal ndustry has developed over half a century and has experenced sustaned growth for more than thrty years, whereas venture captal has been around n Europe for only two decades. Also, most US venture frms are small ndependent partnershps whch fnance themselves mostly from nsttutonal nvestors. In Europe, by contrast, nearly half of the funds come from banks or establshed companes. One suspects that such nsttutonal dfferences may translate n dfferent behavor as we fnd n the US experence (see Hellmann, 2002 and Hellmann, Lndsey and Pur, 1999). Ths s as all aggregate fgures can tell us. An nqury nto how venture captal works n practce, and nto ts ablty to nurture compettve frms, needs to resort to frm-level data. By lookng at mcro data one can study whether fnancal captal, n the form of venture nvestments, makes a dfference n the extent to whch nnovaton contrbutes to growth. So far ths has been possble only n the US, where commercal databases and ad hoc surveys have provded researchers wth valuable data. For example, a recent study of DRI-WEFA for the US Natonal Venture Captal Assocaton clams that 5.9% of the new jobs created n 2000 and 13.1% of the GDP are accounted for by venturebacked companes. In recent work, Bottazz and Da Rn (2001, 2002) provde a frst look at how European venture captal contrbutes to job creaton and growth so far terra ncognta. They look at companes whch lsted on Europe s new stock markets n Among 20

21 these, venture-backed companes do not seem to create more jobs or to grow faster than other companes. Whle subject to some methodologcal caveats, ths results show that somethng may be mssng n the way entrepreneural and fnancal captal nteract n Europe to create wealth and employment. At least three possble nterpretatons can be thought of. One s that European venture captal s stll an nfant ndustry whch needs to mature. A second nterpretaton s that the nature of European venture frms, mostly bank- or company-backed, makes them dfferent from US ones. Fnally, t could be that the problem rests not so much wth fnancal as entrepreneural captal: for nstance, Lander (2001b) offers, as mentoned above, and ntrgung nterpretaton of the dfference between the EU and the US n terms of the dfferent type of entrepreneurshp nduced by bankruptcy rules and socal norms: Rules whch penalze bankruptcy (lke n Europe) dscourage entrepreneurs wth rsker projects. We therefore conclude that the nterplay between the supply and the demand of dfferent types of fnancal captal by dfferent types of entrepreneurs should be better understood, and s lkely to be an mportant determnant of natonal nnovatve capabltes. 5. Conclusons Ths paper has documented the fall n the contrbuton of TFP to per-capta ncome n the EU, relatve to the US: a fall of 7 per cent between the 1980 s and the 1990 s. Thus the (small) ncrease n EU catch-up (relatve per-capta GDP ncreased from.79 to.82 n the perod) has been hndered by the slowdown n TFP: both physcal and human captal ntenstes have ncreased. Accumulated knowledge, that we measure wth the stock of valuable patents, s an mportant determnant of TFP. In the producton of these patents the EU s characterzed by low (and decreasng) relatve productvty. Ths does not depend on R&D expendture per R&D worker, whch exceeds by a factor of three the US level. What appears to explan the lower level of productvty s a relatvely lower accumulated stock of knowledge per R&D worker: a thrd than n the US, and a low and decreasng level of relatve total productvty ( natonal nnovatve capacty ). In the attempt to understand these developments the paper moves n two drectons: the organzaton of nnovaton and ts fnancng. On organzaton, the EU has sgnfcant scentfc competences (at least as compared to the US). These however translate nto a smaller and less nnovatve flow of new startups. Important factors seem to be the exstence of regulatory mpedments and a form of organzaton where start-ups are more rarely ndependent and more often happen wthn the boundares of a larger establshed company ( ntra preneurshp vs entre preneurshp.) On fnancng, the nature of venture captal n Europe s dfferent from the US. The ndustry s at an nfant stage and thus lacks experence, for nstance the ablty to dentfy and nurture promsng deas. Contrary to the US, t s also domnated by banks. The outcome s that quoted ventured-backed companes do not grow faster than non venture-backed ones. Flows are also lmted: snce 1995 to 2000 the flow of venture 21

22 captal nvestments n Europe has ncreased by a factor of 6, but the gap wth the US has become larger there nvestment has ncreased, over the same perod, by a factor of 24. Ths, however, s not enough to conclude that natonal nnovatve capacty s hampered by a (venture captal) fnancng constrant: t could well be that the flow of deas that look for fnancng s scarce. 22