New research in endogenous growth theory and economic geography has highlighted the importance of technological spillovers for understanding growth and economic performance. This research involves both theoretical and empirical work, drawing upon diverse branches of economics, including trade theory and industrial organization. A joint CEPR workshop with the Université de Lausanne held on 27/28 January, brought together leading theoretical and empirical economists in the field to consider such issues as the conduct of anti-trust policy vis-à-vis cooperation in R&D, the impact of R&D spillovers on productivity, and the spatial distribution of spillovers. The conference was organized by Damien Neven (Université de Lausanne and CEPR) and David Audretsch (WZB and CEPR), and funding was provided by the European Commission's SPES programme. A volume of papers presented is now available.*

In the first paper, `Scale, Scope and Spillovers: Determinants of Research Productivity in Pharmaceuticals', written with Iain Cockburn, Rebecca Henderson (MIT) examined the determinants of research productivity in the pharmaceutical industry using detailed internal firm data. The authors find that larger research efforts are more productive, not only because of conventional economies of scale, but also due to important economies of scope. The latter are realized by sustaining a diverse portfolio of research projects and by capturing internal and external spillovers of knowledge. This pattern has become more pronounced as the industry has moved from `random' to `rational' techniques of drug discovery. The authors also find large and persistent heterogeneities among firms, both in research productivity and in the structure of their research portfolios. Thomas von Ungern (Université de Lausanne) questioned the generally held view that the pharmaceutical industry is particularly innovative. He suggested that more research should be done to assess the social rate of return on R&D investment in the industry, arguing that econometric results are indicative of decreasing returns rather than economies of scale. He also suggested that a structural model should be applied in view of the long lag between research and innovative output.

In `R&D Productivity: A Survey of the Econometric Literature', written with Pierre Mohnen, Jacques Mairesse (ENSAE-CREST) presented the various approaches adopted by econometricians for assessing the impact of R&D investment on the productivity of firms and industries. It emerges that, even within a given framework, few of the studies are fully comparable. The survey underlines the usefulness of the rarely used dual approach based on cost functions: this should complement and improve the results of the primary approach based on production functions. The survey also confirms that R&D helps to improve the productivity of firms, and that there is evidence of positive externalities across firms and industries. Overall, it appears that the private and social rates of returns on research capital are greater than those on physical capital. Since the knowledge capital stock is much smaller than the physical capital stock, its contribution to overall productivity is still relatively modest, however. Yannis Katsoulacos (Athens University of Economics and Business and CEPR) found the survey a good introduction to the subject, but regretted that no clear link between theory and empirical work was established. It was also suggested that the distinctive nature of technological spillovers should be clarified: input versus output spillovers, and endogenous versus exogenous spillovers.

In `The Demands for Factors of Production and the Pursuit of Industrial R&D', James Adams (US Bureau of the Census and University of Florida) examined the impact of industrial R&D on the composition of inputs, using induced innovation theory. His model consists of R&D firms enjoying some market power with costs described by translog cost functions. The paper uses this to estimate factor bias induced by R&D, that is, the impact of R&D spending on the cost share of four inputs: white and blue collar labour, structures and equipment. The main findings are that total R&D is biased towards white collar labour and equipment. When R&D spending is treated as a dependent variable, equipment prices discourage R&D, the price of white collar labour has no effect, while blue collar and capital structure prices encourage R&D. These findings are broadly consistent with the idea that R&D is factor saving in structures and blue collar labour. However, applied product R&D reveals a more complex pattern: in this case equipment and structure prices no longer matter.

Iain Cockburn (University of British Columbia) wondered whether the constant R&D/sales ratio was truly a random walk or simply reflected arbitrary budgeting decisions by firms. He also pointed out that data gathering and aggregation by the Census Bureau could pose serious problems in the estimation procedure. This is illustrated by the variable used to measure investment in equipment, for which the estimated elasticity of substitution has the wrong sign in 30% of the cases. Cockburn also noted that the theoretical model consisted of single product firms, while the panel included multi-product firms that engaged in product and process innovation.

In `Science-Based Diversity and the Geography of Innovation', Maryann Feldman (Carnegie Mellon University) and David Audretsch examined the geography of innovation at the city level. Their paper suggests that industries that rely on similar academic departments tend to cluster together spatially. In particular, the degree to which an industry is specialized within a city, the extent to which the industrial base of the city is diversified, and the degree of local competition for local factors of production all shape the innovative output of a particular industry located in a specific city. Based on a large database of innovation citations for the US for specific cities and disaggregated manufacturing industries, the authors find considerable evidence that innovative output tends to be greater in cities &nbspwith &nbspdiversified complementary &nbspproduction. Alfred Kleinknecht (Vrije Universiteit Amsterdam) indicated that similar results emerge from detailed survey data from the Netherlands. Jacques Mairesse suggested that the presence of relevant academic departments within a city might be acting as a proxy for infrastructure.

In `R&D Collaboration and Specialization in the European Community', Frank Lichtenberg (Columbia Business School) examined the geographical distribution of R&D collaboration in the EUREKA programme. The data appear to be consistent with the hypothesis of spatial concentration of R&D collaboration. Cluster analysis is also consistent with existing estimates of international R&D spillovers. The paper also examines and contrasts the role of organizations (universities, and small and large firms) in the innovation process. The main findings are that university research is more basic than company research, and that in most cases, large firms' research is more basic than that of small firms. The evidence is also &nbspconsistent &nbspwith &nbspthe hypothesis that the optimal firm size necessary for commercialization exceeds that necessary for applied R&D.

Maryann Feldman indicated that government sponsored programmes such as EUREKA alter the incentives to collaborate. For example, there is a premium on having a poor member of the EU in proposed ventures. In this context, she wondered whether innovation diffusion, rather than collaboration, was the objective of such programmes. Damien Neven asked whether it would be possible, on the basis of this sample, to identify the matching functions of public authorities in Member States and the Commission. David Audretsch suggested focusing on countries which possess a comparative advantage in knowledge generation.

In `Information Revelation, R&D Cooperation and Technology Policy', written with Yannis Katsoulacos, David Ulph (University College London and CEPR) examined the process of R&D competition and cooperation in the presence of spillovers. Unlike most of the literature, the authors treat these spillovers as endogenous and under firms' control. They show that as soon as spillovers are endogenous, it is essential to make a number of distinctions: between information sharing, research coordination and cooperation; between substitute and complementary research paths; and between firms in the same industry or in different sectors. The authors find that allowing cooperation is sufficient to induce full information sharing/research coordination, so that the justification for public subsidies lies in encouraging firms to increase R&D when it is insufficient. Their analysis suggests that R&D subsidies are desirable to increase the level of research, but are not necessary to correct information problems.

Raymond De Bondt (Katholieke Universiteit Leuven) stressed the importance of this research programme by pointing out that in practice, firms do control information flows. He regretted that information exchange and spillovers were only modelled in the profit function, however. He also suggested that the stability of the cooperative solution be examined more thoroughly. On the basis of the paper's result, he drew some policy implications regarding the EU's technology policy. He wondered whether subsidizing R&D was not diverting entrepreneurial activity from wealth creation into rent seeking of EU grants. Iain Cockburn stressed that appropriating spillovers was costly, so that subsidizing R&D could contribute to reducing the cost of benefiting from spillovers.

In `Public Policy Towards R&D in Oligopolistic Industries', written with Dermot Leahy, Peter Neary (University College Dublin and CEPR) examined the free market and socially optimal outcomes in a dynamic oligopoly model that allows for non-linear demands and spillovers. The authors find that first-best optimal subsidies to R&D are higher when firms play strategically against each other and lower when they cooperate on R&D, at least when spillovers are high. Optimal subsidies are lower when firms play strategically against the government, however. The authors also find that second-best optimal subsidies are higher than first-best ones. Furthermore, simulations of their model suggest that the welfare cost of lax anti-trust policy is high in all cases.

Damien Neven showed that the arguments developed in the paper depend on whether cost-reducing R&D and output were strategic complements or substitutes in this two-stage game. On the one hand, cooperation solves the externality problem associated with spillovers, that there is insufficient R&D because of limited appropriability. On the other hand, cooperation reduces the strategic value of R&D-induced cost reductions at the output competition stage (the business stealing effect), which in turn will reduce R&D. He argued that these findings have important policy implications: cooperation would reduce rather than increase R&D spending, because profit shifting would be less of a motivation. In this context, R&D ventures should be viewed as a form of collusion rather than cooperation.

In `R&D Spillovers and Productivity in German Manufacturing Firms', Dietmar Harhoff (Universität Mannheim and ZEW) estimated the effect of a firm's own R&D and of R&D spillovers on the labour productivity of German manufacturing firms. Using a panel of 443 firms observed every two years from 1979–89, the author finds that R&D spillovers do not have a homogeneous effect on all firms in the sample. Roughly half of them appear to experience &nbsppositive productivity effects from R&D spillovers, and these tend to be located in high technology industries. Some of the results suggest a more intricate pattern, however, consistent with the `absorptive capacity' hypothesis: that the effect of spillovers on productivity may be contingent on the firm's own R&D capital stock. Bronwyn Hall (University of California, Berkeley) noted the quality of this empirical study but found some results difficult to interpret. For example, industry and pool dummies are substitutes in the factor demand equation. She suggested adapting the framework of analysis to non-atomistic agents by introducing strategic interactions between firms. James Adams argued that sources of learning are very diversified and that the pooling technique used in the paper might therefore misrepresent this phenomenon.

In `The Demand and Supply of Knowledge: Innovations, Patents and Cash Flow in a Panel of British Companies', written with Paul Geroski and Chris Walters, John Van Reenen (University College London) examined the interrelationship between the patenting activities, innovative capacity and financial resources of a panel of British firms. The authors estimate a dynamically recursive model, using observable measures of innovations, patents and cash flow. They find that lagged patents are significant predictors of current innovations, but lagged innovations do not affect the conditional expectations of current patents. They also report that innovations are influenced primarily by advances in the science base as &nbspmeasured by R&D intensity and spillovers. Innovations depend on the flow of patents, but are more sensitive to temporary shocks such as changes in demand. It also emerges that innovations have a greater impact on cash flow than patents. In addition, both innovations and patents show a strong pattern of history dependence. Simulations of the model suggest that policies geared towards promoting technological advance have only a limited impact.

Marco Vivarelli (Università Cattolica del Sacro Cuore) pointed out that some innovations are not patented, suggesting that the sequence of events might be different from that modelled by the authors. He noted that the explanatory power of some regressions was very low. Alternative estimation techniques, such as full information models or seemingly unrelated regressions, could shed additional light on these issues. In his view, the impact of public intervention (technology policy) is underestimated in the simulations. Rebecca Henderson argued that instead of using a count number of patents, the authors should use citations instead.