WPS4283 Policy Research Working Paper 4283 Innovation Shortfalls William Maloney Andrés Rodríguez-Clare The World Bank Latin America and the Caribbean Region Office of the Chief Economist July 2007 Policy Research Working Paper 0-3762 Abstract There is a common perception that low productivity or that should be expected given the country's specialization low growth is due to what can be called an "innovation and accumulation patterns? This is the question the shortfall," usually identified as a low rate of investment authors tackle in this paper. First, they show a simple in research and development (R&D) when compared way to estimate the R&D gap that can be explained by with some high innovation countries. The usual reaction a country's specialization pattern, illustrating it for the to this perceived problem is to call for increases in R&D case of Chile. For this country they find that although investment rates, usually specifying a target that can its specialization in natural-resource-intensive sectors be as high as 3 percent of GDP. The problem with this explains part of its R&D gap, a significant shortfall analysis is that it fails to see that a low R&D investment remains. Second, the authors show how a calibrated rate may be appropriate given the economy's pattern of model can be used to determine the R&D gap that specialization, or may be just one manifestation of more should be expected given a country's investment in general problems that impede accumulation of all kinds physical and human capital. If the actual R&D gap of capital. How can we know when a country suffers is above this expected gap, then one can say that the from an innovation shortfall above and beyond the ones country suffers from a true innovation shortfall. This paper--a product of the Office of the Chief Economist, Latin America and the Caribbean Region--is part of a larger effort in the region to Measure innovation effort. Copies of the paper are available free from the World Bank, 1818 H Street NW, Washington, DC 20433. Please contact Veronica Cornago, room I8-168, telephone 202-458-4039, fax 202-522-7528, email address vcornago@worldbank.org. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The authors may be contacted at wmaloney@worldbank.org. July 2007. (40 pages) The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Innovation Shortfalls* William Maloney, World Bank and Andrés Rodríguez-Clare, IADB *We would like to thank Gerardo Esquivel, Rodrigo Fuentes, and Reinhilde Veugelers as well as participants to the "Conference on R&D and Innovation in the Development Process" (Barcelona, June 2005) for useful comments. We also thank Lucas Siga for expert research assistance. 1. Introduction There is a common perception that low productivity or low growth in both developed and underdeveloped countries is due to a lack of innovation. Exhibit number one in favor of this argument is a relatively low R&D investment rate as a share of GDP, a common proxy for innovative activity. Some recent publications, for example, have noted with concern that Latin American countries invest an average of roughly .4% of GDP on R&D, whereas most OECD countries' R&D investment rates hover around 2% of GDP.1 The reaction to this perceived problem is to recommend increasing R&D spending up to some target rate based on the R&D investment rates of "high innovation" countries. Thus, in pursuit of "turning the EU into the most competitive knowledge-based economy in the world" the March 2002 meeting of the European Council in Barcelona announced a goal of increasing the average RDI from 1.9% to 3% by 2010 so as to close the gap with the US (2.7%) and Japan (3.0%).2 For another example we can cite the recent speech by President Ricardo Lagos of Chile, who stated that his country should reach an R&D investment rate of 1.5% by the year 2010.3 The problem with these benchmarking exercises and targets is the failure to recognize that R&D investment is just one more activity whose level is determined by the economy's pattern of specialization as well as by overall economic incentives and distortions. Should we expect the natural-resource abundant economies of South America to invest as much in R&D as the manufacturing oriented countries of East Asia? Should we think of "innovation policies" for countries whose low R&D investment rates may just be one more manifestation of general problems that impede accumulation of all kinds of capital? How can we know when a country suffers from an innovation shortfall above and beyond the ones that should be expected given the country's specialization and accumulation patterns? These are the questions we tackle in this paper. Because of 1See OECD 2004, World Bank Institute 2005, De Ferranti et. al. 2003 2See OECD (2004) 1 Chile's commitment to the innovation agenda, and its previous far reaching micro- reforms, we have chosen this country to serve as illustration of the analysis that we propose. Thus, the paper can be seen as both as a general exploration of ways to identify innovation shortfalls, and as an analysis of whether Chile suffers from one. We begin in the next section by exploring the relationship between R&D investment rates and specialization patterns. Based on data at the OECD level we show that R&D investment rates vary tremendously across sectors, going from 0.2% of value added in "publishing, printing, and reproduction of recorded media" and "construction" to 25% in "pharmaceuticals" and 27% in "office, accounting and computing machinery." Clearly, if for exogenous reasons a country is specialized in sectors characterized by low innovation, then one should not be surprised or concerned to find that it spends relatively little on R&D. Ideally, one would directly compare R&D investment rates in the same sector across countries, but unfortunately such data is not available for LDCs. Thus, we perform an indirect analysis, asking how much would R&D investment rates fall in OECD countries if they had the economic structure of a resource rich Latin America country, using Chile as an example. The results suggest that compositional matters are relevant, but do not explain a large part of the country's innovation shortfall. Section 3 turns to the question of whether low R&D investment rates should be seen as "innovation shortfalls" caused by environments that adversely affect innovation, or as part of a broader problem of low accumulation in all types of capital. Put differently, the question is whether a country's low R&D investment results from problems common to accumulation overall, or whether in fact the activity of innovation itself is somehow especially impeded. Clearly, the appropriate policy depends critically on the answer to this question. Consider, again, the case of Chile, which had an R&D investment rate of 0.4% of GDP in 1995, a rate that is well below the comparable number for the U.S. (2.5% in 1995). It is 3Ricardo Lagos, Discourse on the State of the Nation, May 21st, 2005, (see 2 tempting to conclude that Chile has an innovation shortfall until one notices that Chile also has low investment rates in human and physical capital, as reflected in a composite capital-output ratio of 50% of the U.S. level.4 Perhaps whatever leads Chile to have low investment rates in human and physical capital may also explain its low R&D investment rate. Figure 1 extends this analysis to 48 countries for which we have the required data for 1995. It shows that there is a positive relationship between R&D investment as a share of GDP and the composite capital-output ratio. Again, the conclusion is that ­ at least for some countries ­ low R&D investment rates could be just part of a more general accumulation problem. The challenge that emerges, then, is to find a way to identify true innovation shortfalls and separate them from cases of low overall accumulation (or "accumulation problems").5 For this purpose we propose the use of a model developed and calibrated by Klenow and Rodríguez-Clare (2004) that captures the interactions among accumulation of different types of capital, including "knowledge capital," and allows for both barriers to general accumulation and barriers that are specific to the accumulation of knowledge. We explore the model's implications for Latin American countries and find some countries that seem to suffer from true innovation shortfalls. It is important to emphasize, however, that the results from this analysis should be seen as extremely preliminary, as the exercise suffers both from large measurement error intrinsic to the use of international datasets such as those of Barro and Lee (2000) and the Penn World Tables 6.1, as well as model misspecification for some countries. To explore these issues, we consider again the case of Chile and show that some particular adjustments to the model are necessary for a more appropriate analysis. After such adjustments, the conclusion that emerges is that http://www.gobiernodechile.cl/21mayo2004/indice_discursos.asp). 4The composite capita-output ratio includes both human and physical capital, and is given by k h(K /Y) /( 1- ) , where h is human capital per worker (measured as h = es , where is the Mincer coefficient and s is the average years of schooling of the adult population), K is the stock of capital, Y is total output, and is the share of capital in output (see below for a formal derivation). 5In part, this paper is inspired by the recent contribution by Hausmann, Rodrik and Velasco (2005), titled "Growth Diagnostics," in which they argue that different countries may be affected by different binding constraints to growth, and that the goal of development economics should be to identify those constraints. A significant difference in our approach compared to theirs is that whereas they argue that one should look at shadow prices, we instead look at quantities but use theory to infer the corresponding shadow prices. 3 Chile does indeed suffer from a true innovation shortfall: its R&D investment rate is significantly below what would be expected given its stocks of human and physical capital. Still, our goal here is not so much to reach a solid conclusion for any particular country, as to illustrate how this kind of analysis might be undertaken. We believe that even for the case of Chile, a more careful analysis is called for. Finally, Section 4 briefly discusses possible explanations of true innovation shortfalls that have been presented in the literature, again focusing on the case of Chile. We end by calling for more research both to improve the methodology to identify innovation shortfalls, and for finding the most important determinants of such shortfalls in specific countries. 2. R&D and the Pattern of Specialization As mentioned in the introduction, there is enormous variation in R&D investment rates across sectors. This is a reflection of the fact that, just as with physical and human capital, some sectors are intensive in knowledge capital relative to others. This has important implications for comparisons of R&D investment rates across countries. In a multi-sector economy with international trade, one may have different countries specializing in sectors with differing possibilities for technological change, so that one would observe significant gaps in R&D investment rates that are consistent with factor price equalization and hence similar wage levels (see Grossman and Helpman, 1991). Although income levels would be higher in the economy with higher R&D rates (since it would enjoy a higher stock of knowledge capital per worker), under reasonable conditions it would not be appropriate for countries to encourage the growth of high- R&D sectors. We discuss this point by briefly reviewing the relevant theory in the next subsection. The following subsection looks at the R&D data at the sector level to gauge the extent to which differences in specialization patterns may explain a significant part of the R&D insufficiency one observes in LDCs such as Chile. 4 R&D rates in a multi-sector economy: implications from trade theory Although our interest is R&D and knowledge capital, we present this discussion in the more familiar setting of investment and physical capital. The framework we use for this discussion is the Hecksher-Ohlin (HO) model with capital accumulation. We later add Ricardian productivity differences, differences in income taxes, and externalities. Consider a model of a small open economy, two factors of production (capital and labor), and two goods that differ in their capital intensity. Let the two goods be 1 and 2, with 1 being labor intensive and serving as the numeraire. Also, assume that capital is produced with a technology that is identical as the one to produce good 1, and assume that there is no depreciation, so that the rental rate of capital is equal to the net rate of return to capital (see Findlay, 1995). Technology is identical in the small economy and in the Rest of the World (RW). International prices are determined in the RW. From the Stolper-Samuelson Theorem we know that under certain regularity conditions there is a unique and positive relationship between the rate of return to capital and the relative price of 2. If the instantaneous intertemporal discount rate is , then we know that the long-run equilibrium in the RW is such that the rate of return to capital is , which pins down the relative price of 2. Coming now to our small open economy, Figure 2 illustrates the equilibrium analysis. The horizontal axis measures k, the capital-labor ratio, and the vertical axis measures r, the rate of return to capital. Curve ri(k) denotes the equilibrium rate of return to capital given an economy-wide capital-labor endowment of k if there is complete specialization in sector i. Decreasing marginal returns to capital imply that the curves ri(k) are downward sloping. Let ki be defined implicitly by ri(k) = . Standard HO analysis tells us that if k was exogenous, then the allocation of resources between sectors 1 and 2 is determined by k: if k is equal to k1 or lower there would be complete specialization in 5 good 1, whereas if k is equal to k2 or higher there would be complete specialization in good 2. There is factor price equalization if k is in between these two extremes, whereas the rate of return is higher (lower) than the world's rate of return if k is lower than k1 (higher than k2 ). For any level of k between the two extremes, the allocation of capital between sectors 1 and 2 is just such that the equilibrium rate of return is equal to . Hence, the equilibrium rate of return as a function of the capital-labor ratio endowment is given by the fat line in the figure. When there is endogenous capital accumulation in the small open economy, we see that if the rate of discount is the same as in the RW, then there is indeterminacy in the steady- state capital-labor ratio, and the economy is indifferent among all these points. On the other hand, if the economy has a discount rate higher than , meaning it is more impatient than the RW, then it will have a capital-labor ratio lower than k1, it will be specialized in the labor intensive good, and it will have a higher equilibrium rate of return to capital. The same would occur if the small economy has an income tax that is higher than in the RW. Thus, the first result we derive here is that comparative advantage in a long run HO model is determined from the policies and preferences that affect capital accumulation. Translated to R&D terminology, countries with more favorable policies towards innovation would be specialized in R&D intensive sectors, have higher R&D investment rates in each sector and for the whole economy and attain higher income levels. But consider now what happens if there are Ricardian productivity differences. This is relevant because, as has been pointed out in the literature, when the capital stock is endogenous, then the long-run Production Possibilities Frontier becomes flat. Hence, just as in the Ricardian model, sector-specific productivity differences would completely determine comparative advantage and the pattern of specialization. To see this, imagine that our small economy has a Ricardian productivity that is lower in sector 2 relative to the RW. The analysis for the small economy is exactly as above, but "as if" the international price of good 2 was lower. A lower international price for good 2 implies that the curve r2(k) is shifted downward relative to the original curve, with points k1 and 6 k2 moved to the right, so that the new equilibrium-returns curve as a function of k has the same shape as above, but moved South East. This implies that if the small economy has the same discount rate as the RW, it will specialize in good 1 and use the same capital- labor ratio in the production of this good as the RW. The relevance of this finding for our goal of understanding the role of specialization patterns in explaining R&D gaps across countries is the following: if a country's lower R&D investment rate is entirely caused by its specialization in low-R&D-intensive sectors, then we should find that the whole R&D gap can be explained by the country's sectoral structure, with the same R&D investment rates across each sector. In other words, if we think of an economy's R&D investment rate as the weighted average of the sectoral R&D rates with weights given by the shares of each sector in total output, then a Ricardian story would lead to differences in R&D rates entirely driven by differences in the weights but not in the sectoral R&D rates. Below we will check whether this is indeed the case. Summing up, an economy can specialize in good 1 either because it has a higher discount rate, a higher income tax, or a Ricardian comparative advantage in good 1. In the first two cases one would find that both sector specific R&D gaps and sectoral composition differences explain differences in R&D investment rates across countries, whereas in the later case (i.e., Ricardian comparative advantage) sectoral composition would be solely responsible for international differences in R&D spending. Note also that the only case where it could make sense to do something about the fact that the economy is specialized in a low R&D sector is if this is caused by a higher income tax, otherwise, no intervention is justified. Externalities Now imagine that there are sector-specific (Marshallian) externalities. As shown in Rodríguez-Clare (2005), the problem that may arise in this case is that the economy may have a comparative advantage in good 2 and experience a coordination failure that keeps 7 it specialized in sector 1. This is of course the classic analysis of sector-specific externalities and trade, where an economy may be in a bad equilibrium, specialized in a sector where it doesn't have a comparative advantage. If this were the case, then a policy inducing specialization in sector 2 would lead to higher investment rates, a higher steady state capital-labor ratio, and a higher TFP arising from specialization in the sector with comparative advantage. What does this tell us for the case of an LDC? Translating again to R&D terminology, if the LDC has a Ricardian comparative advantage in R&D intensive sectors, but there are sector specific and local spillovers, then it could make sense to think of a policy to induce a reallocation of resources towards the more R&D intensive sectors. This would lead to an increasing R&D investment rate. But does it make sense to think that Chile, for example, has a Ricardian comparative advantage in more R&D sectors? Probably not! There is a case that can be made for a policy to induce specialization in high R&D sectors. The previous argument applies to the case where externalities or R&D spillovers are entirely within industry. A different result emerges if R&D generates positive economy-wide (i.e., inter-industry) spillovers. In that case, it is easy to show that an economy could be justified in sacrificing efficiency through specialization in sectors where it doesn't have a comparative advantage, to attain higher R&D investment rates and enjoy the associated spillovers. In fact, some of the discussion that took place in the U.S. when it was feared that it was losing its edge in semiconductors can be interpreted in this way, with commentators like Laura Tyson arguing that semiconductors generate strong inter-industry externalities, and that therefore it is important to have a domestic semiconductor industry even if this runs against comparative advantage (Borrus, Tyson and Zysman, 1986). For this to be a valid argument, however, it would be necessary that knowledge spillovers associated with R&D be stronger across domestic firms than across firms in different countries. Indeed, if spillovers are international, then it would clearly not make sense for 8 a country to intervene, for any market failures would be international in scope, and hence national economy policy is clearly not the correct type of intervention. Although there is some controversy on this matter, with ­ for instance ­ Irwin and Klenow (1994) finding that learning-by-doing spillovers in the semiconductor industry are as strong internationally as domestically, our reading of the more general evidence leads us to think that domestic spillovers are stronger, since knowledge spillovers are clearly attenuated by distance (Audretsch and Feldman, 2003). Ultimately, then, this is an empirical matter. If R&D spillovers go beyond sectors but stay mostly within borders, one cannot easily discard policies to push resources towards high R&D sectors. Of course, favoring high-R&D sectors may not be the most standard way to encourage R&D; a more conventional approach would be simply to subsidize R&D. But if for practical reasons the latter approach is not advisable, then perhaps a sectoral approach is relevant. As with any policy option, however, there are significant costs and risks that would have to be carefully considered. Before going any further with policy discussions, however, it is necessary first to explore whether in fact there are significant systematic differences in R&D intensities across sectors, and whether this can explain a significant part of LDCs' (and Chile's, in particular) shortfalls in R&D. A look at sector level R&D data We first examine how R&D investment varies across sectors for the OECD since LDC data do not yet permit this kind of exercise. Table 1 reveals several important stylized facts. First, there is a wide range of average R&D investment rates by sector from around .1% in Services, apparel or publishing to almost 30% in pharmaceuticals; office, accounting and computing equipment; and air and spacecraft. Second, there is tremendous variation of investment rates within sectors. In manufacturing in the 9 aggregate, for example, Spain holds up the bottom with 2% while Sweden tops the list at close to 12%. Looking within one sector, pharmaceuticals, Spain invests under 10% of value added while Sweden invest more than 40%. Third, overall, individual sectoral investment rates have tended to rise suggesting an increasing intensity in the use of knowledge in the production of these products. Fourth, sectoral composition matters appears to matter. The declining aggregate R&D investment rates across this period, partially a phenomenon of this particular sample cut, is driven by the fact that OECD countries have moved heavily into services ­ over 4 percentage points of total non- agricultural value added in many OECD countries.6 Hence, as a first pass, increases in aggregate R&D investment rates occur both from increasing R&D investment in existing sectors and shifting into more intensive sectors. A more careful decomposition of differences in aggregate R&D Investment (RDI) rates within the OECD suggests a combination of both elements with wide variations across countries (see figures 3 and 4). The RDIs in the US and France are higher than the mean largely due to higher investment rates within the mean set of sectors, while Finland and Korea's high RDI, and Canada's, Australia, Netherlands and Norway's lower RDIs are due largely to compositional effects--electronics in the former, perhaps natural resources in the latter. Among the newly emerging eastern European countries, countries, Poland, Spain, Czech Republic, the deficit is due almost entirely to low investment rates within sectors. Within the manufacturing sector (not shown), the story is again mixed. Again, the deficits of the younger countries are due largely to low within-sector investment rates. Among the wealthier countries there is a mix with Germany's superiority and to a lesser extent Japan's due largely to within sector rates while others, again, Finland, Belgium Canada and the US, being driven more prominently by sectoral composition. To get a feel for what is happening in Chile, for which we lack comparable rates of R&D investment at the sector level, we take an indirect route. The first column of Table 2 applies Chile's industrial structure to the sectoral investment rates in each country in the 6See Maloney, 2005. This trend represents a continuation of that identified by Bernard and Jones (1996) 10 OECD. Column 3 relates this simulated value to the actual. It is first clear that structure is not everything. Norway's predicted level is within 10% of its actual suggesting that the fact that its RDI is double that of Chile is significantly due to low investment rates within existing sectors. However, it is also clear that structure matters. On average, simulated OECD aggregate investment rates are just under 60% of those observed and Finland and Germany, are roughly 30% of their actual. Table 3 asks which sectors are most responsible for these very large disparities by applying the average OECD sectoral investment rates to the difference between Chile's and the aggregate OECD sectoral participation rates. Virtually the entire difference can be accounted for by Chile's very low participation in the electronics and transport sectors, both of which show very high average investment rates. The fact that Chile has not added Nokia to its forestry industry the way Finland did explains the vast difference in the two countries simulated rates. In summary, Chile's low R&D comes in part from its specialization in sectors with low R&D intensity, but this is not the whole explanation, as there is also a significant gap that comes from lower R&D investment rates at the sector level. In the next section we explore whether such lower R&D investment rates can be seen as consequences of a specific innovation shortfall or of a broader problem of low accumulation of all kinds of capital. 3. A model of knowledge capital accumulation There is a long literature that tries to understand the relative contribution of capital accumulation and productivity growth to economic growth. More recently, research has focused on what is sometimes called "development accounting," the goal of which is to understand the determinants of income differences across countries at a particular point in from 1970-1990. 11 time. In particular, the exercise explores whether a country's low income level is due to low investment in physical capital, human capital, or to a low TFP level. One problem with development accounting is that it is almost never acknowledged that TFP, just as the stock of physical and human capital, is the result of investments in some kind of capital, perhaps "organizational" capital or technology. In other words, TFP is also the result of accumulation of some sort. To tackle this issue and undertake a more meaningful development accounting exercise, Klenow and Rodríguez-Clare (2004) formulate a model in which TFP is the result of accumulation decisions. The authors used the model to explore the relevance and magnitude of international spillovers, and also to understand whether policies that affect appropriability in general, together with exogenous differences in the relative price of investment goods and investment levels in human capital, can explain the international variance of income levels, or whether one also had to postulate significant differences across countries in their treatment of innovation and technology adoption. The conclusion was that this latter element was important: to explain differences in labor productivity across countries, one has to assume that there are significant cross-country differences in policies or institutions that affect the cost of technology adoption. In this section we turn our attention to a slightly different matter. We are interested in applying the framework of Klenow and Rodríguez-Clare to understand the different reasons behind an LDC's low income level. Perhaps there are some countries where low income is due to low appropriability, others where low income is due to low human capital, others where it is mainly due to a high relative price of investment, and yet others where low income is due to a high implicit cost of technology adoption. In a sense, we are interested in exploring this framework to conduct a sort of "R&D diagnostics," so that one can see whether a country suffers from low R&D beyond what would be expected given its low investment in other types of capital. We take the case of Chile for an illustration of this methodology, and to discuss its advantages and disadvantages, as well as the way in which it is sensitive to different assumptions. 12 We first explain briefly the main workings of the model. As customary, we postulate a Cobb-Douglas production function of the form Y = K (AhL)1 , where Y is total output, - K is the physical capital stock, A is a technology index, h is average human capital per worker, and L is the total labor force. We follow the Mincer specification, so that h = es , where s is years of schooling, assumed constant and exogenous. Output can be used for consumption (C), investment (I), or research (R), Y = C + pI + R , where p is the relative price of investment and is assumed constant through time. Physical capital is accumulated according to: K& = I - K . The only thing left to specify is the way that A evolves. A complete description is beyond the scope of this paper, and the reader is referred to Klenow and Rodríguez-Clare (2004). Here we just provide a brief sketch. First, there is a world technology frontier, denoted by A*, that increases thanks to the R&D performed in all countries. The rate of growth of A* is denoted by gA. Second, each country's A relative to the world level ­ which we denote by a = A/ A* ­ is determined by the country's efforts in technology adoption, which we equalize to a broad concept of R&D.7 Thus, R&D in our model has two functions: it contributes to increasing the world's technology level frontier and it allows the country to come closer to the world's frontier (i.e., decrease a ). Given that R&D is more effective in increasing the country's A when the country has a lower relative A level (i.e., there are benefits of backwardness), then low R&D does not translate into lower growth, but rather into a lower steady state relative A, with all countries in steady state growing at a common rate. Moreover, there is also a "free flow" of ideas from the rest of the world to any particular country, and this happens at a rate denoted by . It is also assumed that the basic productivity in R&D is the same across countries, although the actual labor productivity 7The reader may be concerned here that this formulation implies that all TFP differences result from differences in technology adoption. Below we explore this issue quantitatively. 13 in R&D may differ due to differences in the amounts of physical and human capital. We denote this basic productivity in R&D by . Thus, A& = (R / L + A)(1- A/ A*) In steady state we have: (1) a =1- gA sRk + where sR is R&D as a share of GDP (i.e., sR R /Y ) and k h(K /Y) /( 1- ) is the "composite" capital-output ratio (incorporating both physical and human capital). As usual, y Y / L = Ak , so that labor productivity is the product of the technology index and the capital-output ratio. This expression takes into account that ­ just as in the neoclassical model ­ an increase in A leads to an increase in the rate of return to capital, so that to bring the economy back to steady state an increase in the capital-labor ratio is called for. The full effect of an increase in A, once one takes into account the indirect effect through the induced capital accumulation, is a proportional increase in labor productivity (see Klenow and Rodríguez-Clare, 1997). But here there is an additional interaction between A and k, since a positively affected by k . The reason for this is that R&D uses the same technology as production of output, which relies on human and physical capital, hence a high level of k makes R&D more effective in accumulating A. Thus, this model incorporates both the effect of technology on capital accumulation, and the reverse effect from capital accumulation to increased technology adoption. Third, a country's R&D investment is the sum of R&D performed by firms, who undertake R&D together with accumulation of physical capital to maximize the present value of their future stream of profits, which are equal to total income net of wages paid and net of taxes. Apart from general income taxes, there are also policies and institutions that affect the cost of R&D, which we capture by the parameter , so that the unit cost of R&D in terms of units of output is 1+ . Apart from this implicit R&D tax, we allow for an R&D externality, so that a firm's A increases not only thanks to its own R&D but also 14 thanks to R&D performed by other firms in the economy. We use a parameter between zero and one to capture this externality, with = 0 implying no externalities and = 1 implying full externalities, in the sense that A is determined completely by average R&D efforts among all the firms in the economy. The firm's decision about how much to invest is determined by a dynamic optimization problem, which yields two first order conditions: one for investment in physical capital, and one for R&D. The first order condition for investment in physical capital yields the following steady state restriction: (2) p(K /Y) = r + 1- where is the tax on profits, and r is the equilibrium steady state real interest rate, which is assumed equal across countries. Assuming a common interest rate across countries, and using data for each country for p and k, equation (2) yields an implicit for each country. Note that and r are "interchangeable" ­ that is, the model cannot differentiate between low accumulation due to high taxes or low finance, since both work through the same channels. As mentioned, however, we assume that r is the same across countries, so that all international differences in the "nominal" capital-output ratio are explained by differences in tax rates. The second first order condition determines R&D, and hence relative A in steady state. This condition is: (3) (1-)k(1- a) - ga/(1- a) +(1- a) = r where = (1- )(1- ) /(1+) is a composite distortion term that captures the effect of taxes and externalities. To see this better, the difference between the social and the private rate of return to R&D can be shown to be equal to: 15 ~ r - r = (1- )(1-)k(1- a) + gL where gL is the rate of growth of the labor force. The wedge between the social and private rates of return to R&D is thus composed of two components: the first component is generated by taxes and the domestic R&D externality, as captured by the term , whereas the second term is associated with the rate of growth of firms, which in the model is equal to the rate of growth of the labor force, and arises because of an assumption in the model that new firms are born with a productivity equal to the average productivity of existing firms. Equation (3) determines the relative A level of a country given its measured levels of k, and the two tax parameters and . Calibration For the calibration, we follow Klenow and Rodríguez-Clare in having =1/3, = 0.085, = 0.08, r = 0.086, g = = 0.015, = 0.38, and = 0.55. The interested reader can consult that paper to understand the details of this calibration. Here we just provide a brief explanation. The values used for the parameters , , and are standard in the literature. The interest rate is obtained by noting that with a tax rate of 25% in the U.S. (i.e., = 0.25) and given data for the capital-output ratio and the relative price of investment in the U.S., then equation (2) implies r = 0.086. The steady state growth rate of A*, g, is obtained from the average growth of TFP in the OECD for the period 1960-2000. We assume that = g to generate reasonable steady state properties. Finally, parameters and are calibrated to U.S. data. In particular, these parameters are set so as to have that the social rate of return to R&D in the U.S. be three times the net private rate of return given an R&D subsidy of 20% (i.e., = -0.2 ), and given an R&D investment rate in the U.S. of 2.5%.8 8Assuming that the U.S. has a 20% subsidy on R&D may be questioned for two reasons. First, because although this is the statutory rate (see Hall and Van Reenen, 2000), the effective rate is much lower. Second, because since we are considering a broad concept of R&D, then the actual rate would be even lower. It turns out, however, that this is not too relevant for our main conclusions. We recalibrated the model with a U.S. R&D tax of 0%, and the results do not change in any significant way. 16 Initial Results Table 4 presents the results of this exercise for several Latin American countries plus the U.S. We emphasize that these results are only suggestive, since they are affected by the measurement error intrinsic to international databases such as those of Barro and Lee (2000) and the Penn World Table 6.1. Although useful to generate broad international stylized facts, such databases are too noisy to be reliable in undertaking a country- specific analysis. Moreover, although the calibrated model is a good approximation for broad international patterns, it may be way off for particular countries. A serious analysis about a particular country necessarily entails obtaining better data and adjusting the model for country idiosyncrasies. We illustrate the relevance of this for the case of Chile below. Columns 1-3 of Table 4 come from Barro-Lee data on human capital and the Penn World Tables, using =1/3, = 0.085, and a procedure to construct capital stocks as described in Klenow and Rodríguez-Clare (2004). Column 4 calculates the income tax implied by equation (2) above assuming that all countries have the same interest rate as in the U.S., calibrated above as r = 8.6%. The country with the lowest implied income tax is Mexico, which has a physical capital-output ratio equal to that of the U.S. in spite of having a relative price of capital that is twice as high. The only way for this to be an equilibrium is to have an income tax much smaller than that of the U.S. Column 5 presents the composite capital-output ratio k h(K /Y) /( 1- )as a ratio of the U.S. level. Column 6 uses equation (3) to calculate the value of a assuming that all countries had the same implicit tax on R&D as the U.S. (i.e., = -0.2 ), and presents it also as a ratio of the U.S. level. Column 7 shows the associated R&D investment rate, using equation (1). Column 8 shows the product of relative k and relative a, which yields labor productivity relative to the U.S. Thus, for example, if Chile had = -0.2 , given its levels of human capital, the relative price of investment, and the (real) physical capital- 17 output ratio K/Y, then its labor productivity would be 38% of that of the U.S. Column 9 presents the social rate of return to R&D given = -0.2 . The exercise continues in columns 10-13 of Table 4. Columns 10 and 11 show labor productivity and technology level A calculated directly from the data expressed as ratios of corresponding U.S. levels, respectively. (The level of A is obtained from y and k by applying y = Ak). Column 12 calculates the R&D investment rate implied by a country's "measured" a using equation (1). Finally, column 13 shows the R&D tax necessary for the model to be consistent with this R&D investment rate. Comparison of columns 6- 8 with columns 10-12 reveals the impact of innovation policies and regulations, and column 13 summarizes this comparison in a single index. Finally, column 14 presents the implied social rate of return to R&D. As way of illustration, consider the case of Peru. According to the model, with = -0.2 Peru's labor productivity would be 61% of the U.S. level, rather than the 18% recorded in the data; the reason for this is that given its (implied) low income tax ( = 9%), a 20% R&D subsidy (i.e., = -0.2 ) would lead Peru to an R&D investment rate of 3%, implying a steady state technology index equal to 93% of the U.S. level. In contrast, Peru's actual R&D rate is only 0.4%, implying a level of A of only 28% of the U.S. level, and hence a labor productivity of only 18% of the U.S. level. For this to be an equilibrium phenomenon, the model requires an R&D tax of 154%, which implies a social rate of return to R&D of 51%. Thus, Peru appears to suffer from a true innovation problem, that is, a case of policies and institutions that negatively affect broad R&D. Something quite different happens in Chile. In this case the labor productivity that would obtain with a 20% R&D subsidy would be 38%, which is very similar to what is recorded in the data. In both the hypothetical and "real" cases, the implied R&D investment rate is close to 2%. In line with this, the model's implied R&D tax for Chile is ­24%. Thus, according to this exercise, Chile's problem is entirely driven by its low h and its high 18 implicit income tax . In other words, it is an accumulation rather than an innovation problem. On the other extreme, we find El Salvador: given its low levels of h and K/Y (and hence a very low k equal to only 0.3 relative to the U.S.), one would expect El Salvador to have a low relative A level (38% of the U.S. level), yet one observes a high relative A of 72%, implying a high R&D investment rate of 3.3%. Hence, it must have policies and institutions that favor R&D: the model implies that El Salvador enjoys an R&D subsidy of 53%, significantly higher than the one in the U.S. To summarize the previous results, the exercise suggests (again, remember these results are only suggestive; more elaborate country-specific analysis is required to explore individual countries) very high R&D taxes in Ecuador, Mexico, Panama and Peru, and medium R&D taxes in Argentina, Bolivia, Brazil, and Venezuela. Chile, Colombia, El Salvador and Uruguay appear to have favorable R&D institutions and regulations. The first group of countries has an innovation problem, whereas the problem in the later group is one of accumulation. The first group of countries would benefit enormously from adopting policies and regulations more favorable to innovation. For example, Panama's R&D investment rate would increase from 0.6% to 2.9% of GDP if it could go from = 1.17 to = -0.2 , leading to an increase in its labor productivity relative to the U.S. from 27% to 63%. Of course, this is not to say that this is a simple matter of innovation or tax policy; as we discuss below, the institutions and regulations that determine the effective R&D implicit tax are much more complex. For the group of countries with favorable innovation institutions and regulations, there is little to gain from additional efforts in this dimension.9 9Of course, this doesn't mean that these countries should not continue to improve their innovation policies. Even with a 20% R&D subsidy, the social rate of return is relatively high. For example, the implied rate of return to R&D in Chile is 26%, just as in the U.S., and considerably higher than the private rate of return. Clearly, it makes sense to provide even stronger support to R&D. The point is that this is no longer the source of divergence from U.S. productivity levels. 19 The next subsection explains columns 15-17 of Table 4. The last column of the table shows the measured R&D investment rate. All the implied R&D investment rates of column 12 are higher than the measured ones in column 18. This reveals that measured R&D is significantly lower than the model's implied R&D including technology adoption efforts. This should not be surprising: measured R&D only considers a small portion of overall innovative and technology adoption efforts, since the formal definition of R&D excludes investments that one would normally want to include as technology adoption.10 Indeed, one advantage of the approach taken here is that the "R&D" measure we back out is really a more general measure of innovative effort that is mapped to the TFP measures plugged into the model. In this way, we avoid some issues complicating innovation diagnostics mentioned earlier. First, as mentioned earlier, we include technology adoption efforts that are likely to be left out of the formal measurement of R&D. Second, we implicitly take into account international differences in effectiveness with which R&D is turned into useful knowledge, resulting ­ among other factors ­ from differences in the fraction of R&D that is financed by governments across countries. The role of distortions The model we have used so far assumes that all TFP differences across countries result from differences in R&D or technology adoption. Thus, it leaves no room for distortions acting through other channels, such as trade barriers that decrease efficiency directly or regulations that leads firms to adopt suboptimal combinations of inputs. We believe that an interesting area for future research is precisely to explore ways to identify the relevance of barriers to technology adoption and direct distortions for international TFP differences. For now, we undertake a simple exercise: we want to know the distortions 10One complication that arises here is that although we have a broad definition of R&D, we nevertheless use the measured 2.5% R&D investment rate in the U.S., as well as the official R&D subsidy in the calibration. There are two reasons why we believe this not to be a serious problem. First, because the bias in the measurement of R&D must be much stronger in LDCs than in the U.S., since the main problem arises from the lack of measurement of technology adoption as opposed to innovative efforts. Second, because the results do not change in a significant way if we instead calibrated the model to a broader concept of R&D in the U.S. (see footnote 8). 20 that would be necessary to account for observed productivity levels if countries had the same R&D policy and institutions as the U.S. (i.e., = -0.2 ). We model distortions as a factor z that directly reduces output: Y = K (zAhL)1 . - Everything else is as in the model presented above. The analysis of steady state equilibrium is exactly as above as if human capital per worker was zh, instead of h. This implies that now k zh(K /Y)/(1 -) . For any particular country we can then ask: what is the value of the distortions variable z such that the data and the model are consistent if we also impose = -0.2 . The result is presented in column 15, whereas columns 16 and 17 present the implied R&D investment rate and the associated social rate of return to R&D. Consider Peru again. Instead of being a case of failed development due to perverse innovation policies and institutions, it is now seen as an economy plagued by distortions that by themselves explain a labor productivity level of 30% of the U.S. level.11 More generally, the countries that in the previous exercise (column 13) were classified as having the highest levels of , are now portrayed as having the lowest levels of z (i.e., highest distortions). The problem with this analysis is that it is hard to know what specific distortions, and through what channels, would generate such enormous static productivity losses. Moreover, it is hard to compare this to the result that ­ in the opposite extreme, without distortions ­ a "barriers to technology adoption" explanation of Peru's low productivity would entail = 1.55 , which implies firms face a cost of R&D in terms of output that is approximately three times higher than in the U.S. This is because although the distortions analysis only tells us the overall productivity loss resulting from unknown distortions, the barriers to technology adoption analysis tells us the specific "wedge" needed to create the technological backwardness consistent with that productivity loss. 11This effect of distortions takes into account its total effect, both the direct effect through a lower TFP, and the indirect effect through a lower capital stock given a constant capital-output ratio (and constant rate of return to capital). 21 In summary, although the model without distortions used for the analysis of the previous subsection suggests that some Latin American countries suffer true innovation problems, an alternative explanation is that rather than lack of innovation, these economies suffer from severe distortions that directly lower TFP. More research is necessary to understand how to disentangle static distortions from barriers to technology adoption. For now, we proceed (mostly) under the assumption that distortions play no role in explaining low productivity levels. Potential pitfalls of using international databases and common parameters: the case of Chile As mentioned above, one limitation of the analysis we have conducted so far is that it relies on international databases and assumed common parameters. Although this is fine for the purpose of establishing stylized facts, it is not satisfactory when analyzing a particular country. We now consider two specific issues. First, some countries may exhibit a high measured TFP as result of their large endowment of natural resources. Ideally, one would correct for this to make the results comparable across countries. Second, although the assumption of a constant Mincer coefficient may be a good approximation when studying broad regularities in the data, this is no longer the case when one is interested in a particular country. In that case, it is much better to use the particular Mincer coefficient for the country in question. In this section, we explore these two issues for the case of Chile, assuming first that there are no distortions (i.e., z = 1). First consider the impact of natural resources. In the case of Chile, it is clear that a significant part of its GDP is not so much the result of using human and physical capital according to the production function above but rather the result of "using" its large endowment of mineral resources. According to the Central Bank, mining contributed 6.7% of GDP in 1999, whereas according to the 1998 Household Survey employment in this sector accounted for 1.6% of total employment. Assuming that the physical and 22 human capital stocks per worker were the same in mining as in the rest of the economy, then this implies that pure natural resources in mining account for approximately 5% of GDP. Table 5 shows the results of this adjustment. The first row replicates the exercise above, while the second row shows the adjusted results. We see that the implied capital-output ratio increases, implying a drop in the implicit income tax from 37% to 34%. Also, there is a small increase in k, and a small decrease in the relative technology index. For this new relative technology index to be consistent with the model above, it is necessary to have a smaller R&D subsidy, calculated now to be 9%. Thus, Chile's "problem" remains one of accumulation and not one of innovation. The third row of Table 5 turns to the second adjustment mentioned above. We consider Chile's estimated Mincer coefficient rather than the common coefficient imposed for the exercise in Table 1. The estimated TFP (and hence the estimated technology index A) is quite sensitive to the Mincer coefficient. For example, according to Arellano and Brunner (1999) the Mincer coefficient in Chile is close to 0.12. If we use this coefficient, then h increases from 1.85 to 2.39, which by itself would imply a decline in A of 23%. Together with the mining adjustment above, a falls to 51% of the U.S. level. The R&D investment rate and R&D implicit tax that go with this (according to the model) are 0.8% and 50%.12 Chile now appears to have an innovation problem. Is the upward adjustment to the Mincer parameter driving these results reasonable? Theory offers little advice: on the one hand, educational quality is likely to be lower in Chile than in the OECD; on the other hand, education stock is lower and hence, ceteris paribus, the return should be higher. What we can say is that the finding has empirical precedent. The adjusted rate is the same as the one Bils and Klenow (2000) and borrowed from Psacharopoulos (1994) and substantially below Lam and Schoeni's (1993) estimate for Brazil. 23 What would be the required distortions to explain Chile's low TFP after the previous adjustments? Row 4 of Table 5 presents an exercise similar to the one performed in the previous section, to determine the distortions that would be necessary to explain Chile's lower TFP level given an R&D subsidy of 20%. The result is that distortions would have to be such as to reduce Chile's labor productivity by 27%. Although we do not have anything rigorous to say about whether this number is reasonable or not, our feeling is that it would be hard to argue that Chile is so much more inefficient than the United States as to generate such a large direct fall in TFP. Still, this clearly remains an open question for research. In summary, adjusting for the impact of natural-resource abundance and a higher than average return on schooling, the analysis for Chile changes radically: these adjustments lead to a lower TFP, a lower implied R&D investment rate, and a higher "innovation tax." More broadly, the analysis suggests that Chile's low labor productivity is the result of (1) a high income tax that leads to a lower capital-output ratio that by itself would lead to a labor productivity level 16% lower than the U.S. level, (2) a lower average mean years of schooling of the adult population that by itself would lead to 11% lower labor productivity than in the U.S., and a combination of (3) distortions that would cause a decline in labor productivity of 27%, and (4) unfavorable policies and institutions for innovation that would lower R&D from 1.9% to 0.8% and labor productivity by 27%. 4. Explaining Innovation Problems The previous sections have explored the idea that some countries suffer from policies and institutions that adversely affect innovation and technology adoption, resulting in lower productivity relative to high-innovation countries. We applied the developed framework to Chile and found that, after some adjustments ­ and assuming that distortions are not 12Interestingly, the 0.8% implied R&D investment rate is now close to the measured rate for Chile, which 24 unusually large ­, this country appears to suffer from this problem. What might be the market, government or other failures that make it somehow more difficult in Chile to accumulate the factors associated with a higher TFP relative to its accumulation of human and physical capital? Four broad categories come to mind: labor market rigidities, lack of human resources, lack of credit, and absence of policies to internalize externalities. We discuss these in turn. Labor market rigidities. The recent theoretical literature on explaining international TFP differences has pointed to "barriers to technology adoption," by which it is usually meant labor market rigidities that prevent firms from adopting new technologies that would negatively affect particular groups of workers (see, for example, Parente and Prescott 1994). Indeed, in a recent survey in Chile, firms cited resistance to change and costs of reducing employment as barriers to adopting technologies (Benevente 2004). This is also consistent with studies showing that Chile's costs of severance are substantially above the OECD (Heckman and Pages-Serra 2004), although they are substantially below much of the rest of Latin America. Recent empirical work on the impact of the rigidities is somewhat mixed. Caballero, Cowan, Engel and Micco (2004) finds that job security hampers the creative destruction process and that moving from the 20th to 80 percentile in job security cuts one percent from annual productivity growth. For Chile, they calculate that raising flexibility to US levels would lead to an initial gain between 2 and 4% and permanent gains in the structural rate of growth of .3%. On the other hand, working at the firm level in Argentina, Galiani (2005) somewhat surprisingly finds only a fragile relationship between the degree of rigidity in union contracts and innovative behavior by firms. Nor can Europe's labor legislation be termed flexible. A fair reading of the limited evidence probably suggests continuing agnosticism on the true magnitude of these effects. averaged 0.6% for 1990-2000 (see Lederman and Saenz, 2003). 25 Credit markets. According to the recent survey mentioned earlier, firms in Chile do not undertake more innovation because of the associated high technical risk and long gestation periods. This could be seen as broadly mapping into the market failures standard in the literature: individual firms cannot handle the lumpiness, risk and long gestation periods of innovation projects. This points clearly to credit market failures. Recent micro-estimates for Chile by Benevente, de Gregorio and Nuñez (2005) suggest that own rates of return to R&D are high. In particular, they estimate rates of around 30% to R&D, whereas the (gross) returns to physical capital are 16%. The higher private rates of return to R&D may be due to its higher risk, but it may also be associated with the fact that it is harder to finance, both because of higher risk together with absence of venture capital, and because of the fact that R&D leads to the accumulation of assets that are harder to use as collateral. Innovation surveys again suggest that the vast majority of financing of innovative activities is internal suggesting potentially an inability to share risk. That said, there is no consensus on why specialized institutions, such as venture capital, have not taken hold in Chile. Some VC firms have folded allegedly for lack of "deal flow" suggesting that there is inadequate financing at the early stages of idea development that would generate demand downstream. However, recent entrants into the market suggest that deal flow is adequate, but that the design of previous VC operations failed to pay sufficient attention to the provision of complementary management and mentoring services that are generally part of VC packages.13 Further, legislation has tied the development of specific institutions to the intermediation of pension funds assets and hence burdened them with inappropriate regulation on risk taking (Arrau 2002). This said, it is worth highlighting that VC is virtually absent in Spain and Italy so it is difficult to assert that this missing market is an insuperable barrier to gains in TFP. 13Discussions with Eduardo Bitran and Patricio Arrau respectively 26 Lack of human resources: Ideally, one could simply look at wages and their distribution according to levels of schooling and professions to see whether there is a scarcity of human resources crucial to innovation and technology adoption. The problem is that this presumes that demand and supply are independent, whereas in this case it is likely that ­ at least to some extent ­ supply creates its own demand, and demand depends on the supply of human resources. Very innovative firms are often spinoffs of university research. Managers in firms with a taste for innovation are likely to have an academic background. In Finland, the most important dimension of U-private sector linkages are reported to be masters students doing their theses in the firms. It may be that such exchanges help define the frontier of the field and possible areas for innovation investment. We can easily imagine a country where entrepreneurs have little idea of where the frontier is and thus available investment opportunities and as a result, have no demand for the products of the science establishment. In this case, no excess demand for the products of a scientific establishment will appear. Multiple equilibrium models consistent with this type of idea have been elaborated by Howitt and Mayer (2005). Policies and institutions. A critical problem in the area of innovation is the existence of externalities. This implies that policies and institutions that internalize such externalities are crucial. Perhaps LDCs, and Chile in particular, suffer from policies and institutions that do not perform this function. Although Fundación Chile and the national Development Corporatin ( Corporación de Fomento, CORFO) are recognized for work in this area, the overall effort may be insufficient. An important area is that of collaboration and linkages between universities and the private sector. University private ­ sector collaboration is a common way of shifting the long term risk of basic science or difficult to appropriate investments from the individual firm. Clearly, universities are also the source of qualified personnel, the lack of which is also cited as a barrier to innovation. In an inversion of what is found in the OECD, in Chile most research is done by, and most researchers are found in, universities rather than the private sector and there is evidence that firms have difficulty accessing either. In 27 theory, skills shortages would be revealed by a high wage premium for scientists and engineers although analysis to date has only identified the general rising premium to tertiary education found globally. However, surveys of private firms suggest, for example, that Chile ranks globally very low on collaboration of the private and university sectors.14 Though most major universities have offices to promote linkages, only the Universidad de Concepción has any incentives for faculty to collaborate with the private sector in promotion criteria.15 5. Conclusion Countries do have innovation shortfalls, but we have argued that their diagnosis requires more than simple unconditional comparisons of R&D or other related indicators. Standard issues of comparative advantage influence the optimal level of knowledge accumulation and generation. Further, even if we establish conditionally low levels of innovation, it is not immediately clear whether the problem pertains particularly to this factor, or whether there are barriers to accumulation more generally that need to be addressed. We offer approaches to both issues and illustrate their application for Chile, a country which is currently thinking seriously about improving its innovation policy. The results that emerge suggest that this country does indeed suffer from a true innovation shortfall, and we offer some tentative ideas on what may be causing it. Clearly, we don't have the last word on the issue and we encourage further refinements both in technique and data. 14World Economic Forum (various). Only 12% of Chilean firms have signed agreements with universities compared to for instance, 40% in Finland (de Ferranti et al 2003). 15Mullin (2005) 28 References Arellano, M. S. and M. Braun (1999), "Rentabilidad de la Educación Formal en Chile" Cuadernos de Economia 36(107):685-724. Arrau, P. (2002), "Agenda Pro-Crecimiento, Propuestas de Reformas al Mercado de Capitales II. Mimeo, Santiago, Chile. Audretsch, D and M. Feldman, (2003), "Knowledge Spillovers and the Geography of Innovation," forthcoming in the Handbook of Urban and Regional Economics, volume 4, North Holland. Barro, R. and J. W. Lee, (2000), "International Data on Educational Attainment: Updates and Implications," National Bureau of Economic Research Working Paper 7911, Cambridge, Massachusetts. Benavente, J. M., (2004), "Technology Innovations in Chile :Evidence from a National Survey," mimeo, Universidad de Chile. Santiago. Benevente, J. M., J. de Gregorio and M. Nuñez (2005), "Rates of Return for Industrial R&D in Chile" mimeo, Universidad de Chile, Santiago. Bernard, A.B. and C. I. Jones (1996), "Apples and Oranges: Productivity convergence and measurement across Industries and Countries." American Economic Review 86:1216-1237. Bils, M. and P. J. Klenow (2000), "Does Schooling Cause Growth?" The American Economic Review 90(5):1160-1183. Borrus, M, L D. Tyson and J. Zysman (1986), "Creating Advantage: How Government Policies Shape International Trade in the Semiconductor Industry" in P. Krugman ed. Strategic trade policy and the new international economics (1986): 91-113, Cambridge, Mass., and London: MIT Press. Caballero, R. J., E. M. R. A. Engel, and A. Micco (2004), Microeconomic Flexibility in Latin America, Economic Growth Center Yale University Discussion paper No 884. Comin, D. (2004), "R&D: A Small Contribution to Productivity Growth" NBER Working Paper 10625. De Ferranti, D., G. Perry, I Gill, J.L Guasch, W.F. Maloney, C. Sanchez P-Paramo, N Schady (2003), "Closing the Gap in Education and Technology, World Bank, Washington. Findlay, R. (1995), Factor Proportions, Trade, and Growth, MIT Press, Cambridge. 29 Galiani, S. (2005), "Unions and R&D, New Evidence Using Panel Data from Argentina." Mimeo, Udesa and World Bank. Grossman, G. M. and E. Helpman (1991), Innovation and Growth in the Global Economy, MIT Press, Cambridge. Hall, B.H. and J. Van Reenen (2000), "How Effective Are Fiscal Incentives for R&D? A Review of the Evidence," Research Policy 29: 449-469. Hausman,R, D. Rodrik and A. Velasco (2004), "Growth Diagnostics" mimeo, Harvard University. Howitt, P and D. Mayer-Foulkes (2005), R&D, Implementation, and Stagnation: A Schumpeterian Theory of Convergence Clubs, Journal of Money, Credit, and Banking 37 (1): 147-77. Heckman, J. J and C. Pages (2004), "Introduction" in J.J. Heckman and C. Pages eds. Law and Employment: Lessons from Latin America and the Caribbean, NBER and U Chicago Press, Chicago. Irwin, D and P. J. Klenow (1994), "Learning-by-Doing Spillovers in the Semiconductor Industry, Journal of Political Economy 102 (6): 1200-1227. Klenow, P. and A. Rodríguez-Clare, 1997, "Has the Neoclassical Revolution Gone Too Far?," NBER Macroeconomics Annual 1997, Brookings Institution. ______________________________, 2005, "Externalities and Growth," NBER Working Paper No. 11009 (forthcoming in Handbook of Economic Growth, edited by P. Aghion and S. Durlauf, North Holland). Lam, D. and R. F. Schoeni (1993), Effects of Family Background on Earning and Returns to Schooling: Evidence from Brazil" Journal of Political Economy 101(4) 710-740. Lederman, D. and L. Saenz (2002), "A Database of Innovation Indicators for the World," Office of the Chief Economist, Latin America and the Caribbean Region, The World Bank. Mullin, J. (2005), "University Policies in Latin America for the Promotion of R&D Links with Firms: A Five Country Analysis." Mimeo, World Bank. OECD (2004), Science, Technology and Industry Outlook, Paris. Parente, S.L. and E. Prescott (1994), "Barriers to Technology Adoption and Development" Journal of Political Economy 102:298-321. 30 Psacharopoulos, G. (1994), "Returns to Investment in Education: A Global Update," World Development, 22(9): 1325-43. Rodriguez-Clare, A., (2005),"Clusters and Comparative Advantage: Implications for Industrial Policy," manuscript, IADB. World Bank Institute (2005), Knowledge Assessment Methodology http://info.worldbank.org/etools/kam2005/. World Economic Forum (various) The Global Competitiveness Report 2001-2002. Harvard University, Center for International Development. Geneva: World Economic Forum. 31 Table 1: R&D investment rates by sector and standard deviation Sector Median R&D investment rate Standard Deviation 1985 2000 1985 2000 Manufacturing, Utilities, Construction and Services 0.014 0.013 0.005 0.007 TOTAL MANUFACTURING 0.060 0.067 0.025 0.035 Food products, beverages and tobacco 0.012 0.013 0.007 0.007 Food products and beverages 0.005 0.012 0.002 0.008 Tobacco products 0.004 0.004 0.004 0.000 Textiles, textile products, leather and footwear 0.007 0.011 0.005 0.010 Textiles 0.009 0.016 0.005 0.012 Wearing apparel, dressing and dyeing of fur 0.001 0.010 0.000 0.022 Leather, leather products and footwear 0.003 0.004 0.003 0.022 Wood, paper, printing, publishing 0.006 0.007 0.005 0.006 Wood and products of wood and cork 0.003 0.005 0.006 0.011 Pulp, paper, paper products, printing and publishing 0.006 0.007 0.006 0.007 Paper and paper products 0.007 0.008 0.013 0.012 Publishing, printing and reproduction of recorded media 0.001 0.002 0.001 0.005 Chemical, rubber, plastics and fuel products 0.101 0.103 0.033 0.055 Coke, refined petroleum products and nuclear fuel 0.057 0.027 0.042 0.023 Chemicals and chemical products 0.130 0.147 0.045 0.074 Chemicals excluding phamaceuticals 0.098 0.067 0.038 0.039 Pharmaceuticals 0.248 0.251 0.103 0.158 Rubber and plastics products 0.023 0.028 0.044 0.060 Other nonmetallic mineral products 0.016 0.015 0.015 0.014 Basic metals and fabricated metal products 0.015 0.016 0.009 0.010 Basic metals 0.026 0.029 0.012 0.016 Iron and steel 0.024 0.022 0.012 0.020 Nonferrous metals 0.047 0.028 0.040 0.023 Fabricated metal products, except machinery and equipment 0.011 0.009 0.006 0.011 Machinery and equipment, instruments and transport equipment 0.119 0.145 0.048 0.062 Machinery and equipment, n.e.c. 0.043 0.067 0.030 0.028 Electrical and optical equipment 0.178 0.242 0.066 0.153 Office, accounting and computing machinery 0.243 0.274 0.089 0.783 Electrical machinery and apparatus, nec 0.091 0.080 0.239 0.053 Radio, television and communication equipment 0.231 0.186 0.094 0.438 Medical, precision and optical instruments, watches and clocks 0.119 0.154 0.083 0.101 Transport vehicles 0.103 0.085 0.094 0.067 Motor vehicles, trailers and semitrailers 0.104 0.101 0.068 0.070 Other transport equipment 0.160 0.124 0.177 0.072 Building and repairing of ships and boats 0.022 0.025 0.012 0.027 Aircraft and spacecraft 0.289 0.212 0.287 0.079 Railroad equipment and transport equipment n.e.c. 0.042 0.094 0.049 0.076 Furniture; manufacturing n.e.c. 0.005 0.025 0.000 0.007 ELECTRICITY, GAS AND WATER SUPPLY 0.006 0.006 0.007 0.005 CONSTRUCTION 0.001 0.002 0.002 0.002 TOTAL SERVICES 0.002 0.003 0.001 0.002 Simple Correlation 0.9539 Spearman Correlation test. Prob > |t| = 0.0000 Source: OECD Structural Analysis Data Base 32 Table 2: European R&D investment rates with Chile's economic structure (1995-1990 average) Estimated RDI Estimated/ Country using Chilean Observed Observed shares Australia 0.007 0.008 0.886 Belgium 0.007 0.014 0.471 Canada 0.007 0.011 0.645 Czech Republic 0.005 0.008 0.550 Germany 0.004 0.017 0.259 Denmark 0.011 0.015 0.750 Spain 0.002 0.005 0.509 Finland 0.008 0.021 0.365 France 0.007 0.015 0.433 United Kingdom 0.010 0.014 0.724 Italy 0.005 0.006 0.846 Japan 0.010 0.020 0.531 Korea 0.006 0.019 0.329 Netherlands 0.006 0.012 0.507 Norway 0.011 0.012 0.929 Poland 0.002 0.003 0.486 Sweden 0.014 0.030 0.475 United States 0.011 0.019 0.567 Note: Applies Chile's sectoral shares in value added to OECD Country's R&D investment rates (RDI). Source: UNCTAD, Central Bank of Chile 33 1 foera cene 06. 20 0 Sh ferfid -0 01.0- 0.00 00.0 00.0 03.0- 00.0 01.0- 01.0 06.0- 01.0 10.0 00.0 000. 100. 61.0 17.0 520. 100. 000. -0. 1.00- 500. 00.1 )-a 9 6 (b*c 02. -0 00.0- 100.0- 0000. 0200.- 5 01.0- 0000. 3 7 9 6 6 00.0- 00.0 02.0- 00.0 00.0 0000. 2 00.0- 030.0 7700. 143.0 181.0 0300. 0100. 110 -0. 40.00- 2200. 8140. -sera se Shna arhS )a 1 6 's b-( 03 100 5 6 1 0 2 2 520. Me ilehC -0. 000.- 000.0 -0. 030.0- 000.- 000.0 000.- 00.0 010.- 00.0 300.0 000.0 000.- 2000. 150.0 220.0 2100. 0200. 0100. -0 70.00- 6600. 0300. OECD vs. niIDRna DCEO )c( 9 0 00.0 01.0 31.00 400.0 0600. 8 2 4 02.0 00.0 00.0 0910. 9 02.0 3400. 2200. 5 5 02.0 01.0 110.0 2 05.0 351.0 990.0 4 6 01.0 01.0 020.0 60.00 3 00.0 Me Chile se edifissacleerh in nt arhS ane DCEOni )b( 2800. 0200. 700.0 5000. 2 00.0 080.0 120.0 0600. 0 02.0 060.0 8 00.0 700.0 030.0 1000. 5100. 180.0 420.0 8100. 0800. 0100. 3 06.0 820.0 0270. 0301. M sewelton:c.e..n investme nisera eilhC )a( 590.0 080.0 700.0 060.0 5 000. 310.0 210.0 120.0 9 010. 710.0 7 5 000. 000. 300.0 120.0 130.0 300.0 200.0 060.0 060.0 000.0 411.0 530.0 366.0 000.1 R&D Sh aggregate in tn fferenceid pmeiuqed e.lihCrofnoitagergesedfolevelelb ssiop for ruffogni eli st st st s responsible gearevebd dydnagnisse raewtoofdnastc leufraelcund anyrenih c.e..n duo ucdorpre anstcudo pr stc .c.e.n,tn ChfoknaBlartn mi st dr,le papd gnihsilb muelotr duo ucdorplaren mactpecxe,stcudo slat prlat quipmeed s,utarappadnayerni tn .c.e.ngnirutca esthgihehtnodseabsinoicte Ce, pmei uf AD Sectors rot anst par anr pud duo me 3: sec Table ucdorpdooF ucdo proccaboT s tilex apgniare Te W prrehtael,rehtaeL krocdnadoowfostc prd chaml man;e g pape,pluP angnitnirP andoo W ucdorplacimehcdnaslacimehC pedenifer, ke prscitsapldnareb cairt Equtrops uritn s selsrotceS itie CTNU:ec Co ubR leestdnanoIr usorrefnoN icllatem-nonrehtO medetacirbaF anyrenihca M eclE anrT urF inlcyceR noitcurtsnoC tilU secivreS latoT s:etoN uroS ) 18 (D Rs)(z 4%0. 4%0. 9%0. 6%0. 3%0. 1%0. 3%0. 4%0. 1%0. 3%0. 3%0. 5%0. 5%2. of 2 .s 17 RRS)(z % % % % % % % % % % % % % rate 19 21 16 27 21 18 14 18 17 26 30 17 26 velel cialos P TF 16 Rs 4%2. 1%1. 9%2. 9%1. 9%1. 6%1. 0%3. 6%1. 3%1. 9%1. 6%1. 6%2. 5%2. ande Imp. 15 z 50. 70. 60. 11. 80. 40. 40. 40. 30. 71. 11. 50. 01. associatedeth comnid )2( and ure 14 R % % % % % % % % % % % % % 37 30 27 26 25 41 38 47 51 16 28 32 26 tax SR )2( R&D 13 % % % 4% % 5% 8% 5% % Phi 61 18 41 -2 -4% 3% 5% 0% 91 12 11 15 -5 -2 53 -2 easmreihtdeliy ) atht 12 (2 exercise sR 2%1. 7%0. 7%1. 0%2. 5%1. 7%0. 1%1. 6%0. 4%0. 3%3. 7%1. 3%1. 5%2. liedpmieht 11 )D( % % % % % % % % % % % % 0% rate,t A 63 30 64 72 53 39 57 39 28 72 65 61 10 enm onsitrostdi ve ) accounting 10 (Dy % % % % % % % % % % % % 0% 44 12 33 39 22 23 37 27 18 24 35 36 invest hatub 10 U.S.eth th )1( R&D 9 grow RRS % % % % % % % % % % % % % asyids S. 21 23 17 27 22 20 16 21 20 24 29 19 26 U.eht sub lied new ) asyd 8 A (1y % % % % % % % % % % % % 0% 66 22 47 38 26 51 67 63 61 13 33 55 10 R&De impeth bsi 4: ) n su 7 (1 sn sR 0%3. 9%1. 5%3. 8%1. 1%2. 9%2. 0%4. 9%2. 0%3. 2%1. 5%1. 2%3. 5%2. obtai R&D latio Table 6 )1( to % % % % % % 1% % % % % % 0% calcu A 94 57 90 69 63 87 10 92 93 38 61 91 samethevah 10 odelm sameeht tries the ve data. ownd 5 k 70. 40. 50. 50. 40. 60. 70. 70. 70. 30. 50. 60. 01. coun has the an all and on 04) 4 % % % % % % % % % % tau data 11 28 -7% 9% 37 23 12 -1 14 10 36 44 4% 25 that countrie 3 K/Y 51. 80. 71. 11. 90. 61. 71. 51. 61. 70. 11. 51. 71. ngimussa the all directly (20ear ing sed us where ba uez-Clg 2 p 21. 81. 31. 11. 61. 11. 41. 21. 11. 02. 01. 31. 90. 1 h 12. 61. 51. 91. 51. 71. 81. 02. 91. 51. 91. 81. 72. Rodrídna or calculationsera D. alculationsc calculationsera y an lculationsacera are iab r a y Klenow nti ia untr liv ilz Co Arge Bo Bra Chile mlooC do o ma adv ua uela R&otn esehT Ecua Mexic Pan Peru SallE ugrU A esehT These These Venez US (1) (2) retur (z) (D) Source: Table 5: Exploring limitations of international databases and common parameters, the case of Chile 1 2 3 3 4 5 6 7 8 9 K/Y z k Rel. k Data rel. Y/L Data rel.a ImpliedsR SRR Chile (1) 1.15 37% 1 1.98 0.55 39% 72% 2% -24% 26% Chile (2) 1.21 34% 1 2.04 0.56 37% 66% 1.7% -9% 28% Chile (3) 1.21 34% 1 2.62 0.72 37% 51% 0.8% 50% 44% Chile (4) 1.21 34% 0.73 1.93 0.53 37% 70% 1.9% -20% 25% Source: Klenow and Rodríguez-Clare (2004) and authors calculations 3 Figure 1: R&D investment rates versus capital-output ratios 5 4 3 2 R&D/GDP 1 0 0 1 2 3 4 Composite capital-output ratio Figure 2 r r2(k) r1(k) k k1 k* k2 4 Figure 3: Differences in aggregate R&D investment rates (RDI) from OECD mean 0.02 0.01 0 stralia lgiumCanadaepublGe ic any ark ain nce rm nm Sp Ja Au Be FinlandFraKingdomItaly panKorearlandsNorwayPoland denStates the Sweted -0.01 ech R De ited Ne Uni Cz Un -0.02 Figure 4: Contribution of economic structure vs sectoral RDI to deviations from OECD mean aggregate RDI 2 1.5 1 0.5 0 rmanenmark y ain e -0.5 Italy panKo Ja rway Australia lgiumanadepublGe a ic Be C PolandwedenStates R D Sp nlandFrancngdom Fi Ki Netherl rea andsNo S -1 Czech ited United Un -1.5 -2 Economic Structure RDI 5 Policy Research Working Paper Series Title Author Date Contact for paper WPS4263HIV/AIDSandSocialCapitalina AntonioC.David June2007 A.David Cross-SectionofCountries 82842 WPS4264FinancingofthePrivateSectorin ConstantinosStephanou June2007 S.Coca Mexico,2000­05:Evolution, EmanuelSalinasMuñoz 37474 Composition,andDeterminants WPS4265TheStructureofImportTariffsinthe OleksandrShepotylo June2007 P.Flewitt RussianFederation:2001­05 32724 WPS4266TheEconomicCommunityofWest SimpliceG.Zouhon-Bi June2007 S.Zouhon-Bi AfricanStates:FiscalRevenue LyngeNielsen 82929 ImplicationsoftheProspective EconomicPartnershipAgreement withtheEuropeanUnion WPS4267FinancialIntermediationinthe HeikoHesse June2007 G.Johnson Pre-ConsolicatedBankingSectorin 34436 Nigeria WPS4268PowertothePeople:Evidencefrom MartinaBjörkman June2007 I.Hafiz aRandomizedFieldExperimentofa JakobSvensson 37851 Community-BasedMonitoringProject inUganda WPS4269ShadowSovereignRatingsfor DilipRatha June2007 N.Aliyeva UnratedDevelopingCountries PrabalDe 80524 SanketMohapatra WPS4270Jump-StartingSelf-Employment? RitaAlmeida June2007 A.Bonfield EvidenceamongWelfareParticipants EmanuelaGalasso 31248 inArgentina WPS4271Construction,Corruption,and CharlesKenny June2007 C.Kenny DevelopingCountries 33540 WPS4272Migration,Remittances,Poverty, DavidMcKenzie July2007 M.Sasin andHumanCapital:Conceptualand MarcinJ.Sasin 36877 EmpiricalChallenges WPS4273RulesofOriginandtheWebofEast MiriamManchin July2007 L.Yeargin AsianFreeTradeAgreements AnnetteO.Pelkmans-Balaoing 81553 WPS4274AreLaborRegulationsDriving MohammadAmin July2007 S.Narsiah ComputerUsageinIndia'sRetail 88768 Stores? WPS4275CanForeignLobbyingEnhance KishoreGawande July2007 V.Cornago Development?TheCaseofTourism WilliamMaloney 84039 intheCaribbean GabrielV.MontesRojas WPS4276HumanCapital,TradeLiberalization, TomKrebs July2007 V.Cornago andIncomeRisk PravinKrishna 84039 WilliamMaloney WPS4277ClimateChangeAdaptationinAfrica: SungnoNiggolSeo July2007 P.Kokila AMicroeconomicAnalysisof RobertMendelsohn 33716 LivestockChoice WPS4278EndogenousIrrigation:TheImpactof PradeepKurukulasuriya July2007 P.Kokila ClimateChangeonFarmersin RobertMendelsohn 33716 Africa WPS4279TheImpactofClimateChangeon SungnoNiggolSeo July2007 P.Kokila LivestockManagementinAfrica: RobertMendelsohn 33716 AStructuralRicardianAnalysis Policy Research Working Paper Series Title Author Date Contact for paper WPS4280 GovernanceMattersVI:Aggregate DanielKaufmann July2007 R.Bonfield andIndividualGovernance: AartKraay 31248 Indicators,1996-2006 MassimoMastruzzi WPS4281 CreditGrowthInEmergingEurope: SophieSirtaine July2007 I.Hafiz ACauseForStabilityConcerns? IliasSkamnelos 37851 WPS4282 InnovationShortfalls WilliamMaloney July2007 V.Cornago AndrésRodríguez-Clare 84039