Accumulation of R&D Capital and Dynamic Firm Performance: A Not-so-Fixed Effect Model

Klette, Johansen
1998 Annales d Économie et de Statistique  
Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen
more » ... von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Abstract: Considering the observed patterns of R&D investment, we argue that a model which allows for a positive feedback from already acquired knowledge to the productiveness of current research, fits the empirical evidence better than the standard model that treats knowledge accumulation symmetrically to the accumulation of physical capital. We present an econometric framework consistent with a positive feedback in the accumulation of R&D capital. The empirical model is econometrically simple and less data-demanding than the standard framework. Our estimates show a significant positive effect of R&D on performance and a positive feedback effect from the stock of knowledge capital. We calculate the depreciation rate and the rate of return to knowledge capital for our alternative framework, and compare our estimated rate of return to results obtained within the standard framework. Over the last 10-15 years, we have seen an outburst of econometric research on R&D investment and productivity; see Griliches (1995) for a recent survey of the many insights that have emerged from this line of research. Much of this research follows the framework outlined in Griliches (1979). In this paper we argue that this econometric framework should be modified and extended in various ways. In particular, considering the empirical evidence on the patterns of R&D investment, we argue that a model which allows for a positive feedback from already acquired knowledge to the productiveness of current research fits the empirical evidence better than the standard model that treats knowledge accumulation symmetrically to the accumulation of physical capital. Positive feedbacks in knowledge accumulation have recently been considered in the literature on macroeconomic growth by Romer (1990), Milgrom, Qian and Roberts (1991), and Jones (1995). Their argument is that this feedback mechanism can explain the persistent differences in productivity between countries or industries, and why some industries or countries suddenly gain momentum and go through phases of high growth. Our analysis is concerned with a related phenomenon at the micro level; how can we rationalize that some firms are persistently, often for a long period, more productive than other firms, as shown e.g. by Bailey, Hulten and Campbell (1992). Similarly, why do some firms persistently carry out considerable R&D, while other firms in the same industry never report any R&D investments? Empirically, it is widely observed that there are large differences in R&D effort across firms within narrowly defined industries, and that these differences in R&D effort are persistent over time. Nelson (1988) has pointed out that this co-existence of innovators and imitatorsas he calls themis a puzzle for the standard framework for productivity analysis at the micro level. We argue that positive feedbacks in knowledge accumulation can be one explanation for the persistency of performance differences at the micro level, parallel to the cited arguments presented in the macro growth literature. The co-existence of innovators and imitators can within our framework be considered a consequence of the stochastic nature of knowledge production in combination with a positive feedback from past R&D success to the productiveness of current R&D. We present a simple, alternative specification for the accumulation of R&D capital that differs from the standard specification in the R&D productivity literature. After an analysis of R&D investment for both the standard and our alternative specification, we show that our alternative specification better fits the empirical patterns with persistent differences in R&D activity between firms in the same industry. The second main part of this paper presents an empirical analysis of R&D, productivity and performance that uses our alternative specification 3 for R&D investment and knowledge accumulation. In this empirical analysis we also alter and augment the standard framework as presented in Griliches (1979Griliches ( , 1995, in other ways, by explicitly incorporating the demand side and both process and product innovations. We have estimated this empirical model on a new data set that links R&D investment at the line-ofbusiness level (within each firm) to plant level data on productivity. The results show that R&D investment is a significant determinant of dynamic performance and that the appropriable part of R&D capital depreciates quite rapidly. The analysis presented here is in several ways an extension of the analysis presented in Klette (1996): First, the present paper presents a formal analysis of optimal R&D investment when the accumulation process allows for the feedback mechanism in our alternative specification. Second, we present an empirical analysis of R&D investment to illustrate the empirical importance of our respecification. Third, the formal analysis of optimal R&D investment leads us to a formula for calculating the private rate of return to R&D investment. Finally, the empirical analysis in section 4 is carried out on a new data set that links R&D data at the line of business level to plant level data for the period 1980-92 (while Klette, 1996, used only a single cross section of R&D data for 1989). This new, larger data set allows us to explore a number of specification issues and formal econometric tests that were not possible with the limited data set used in Klette (1996). The rest of our paper is organized as follows: In section 2, we examine patterns of R&D investments. After discussing R&D investment in the standard model of knowledge accumulation, we present a dynamic programming analysis of optimal R&D investment for our alternative specification of the accumulation process. This analysis is then confronted with empirical patterns of R&D investment in the second half of section 2. Having concluded that our alternative specification of knowledge accumulation fits the empirical data better, we proceed to the empirical analysis of R&D, productivity and performance in sections 3 and 4, based on our alternative specification of knowledge accumulation. For comparison, we start in section 3 with a standard analysis of R&D and productivity, following Griliches (1979, 1995) and Hall and Mairesse (1995). Section 4 contains the main part of our analysis of R&D and performance, where we spell out the empirical framework and present the econometric results. We add some final comments in section 5. 2 Investment in R&D capital and performance 2.1 Persistent cross sectional differences in R&D investment: Theory The standard framework treats the accumulation of knowledge capital in the same way as that of physical capital, using the "perpetual inventory" process as a common framework. Formally, 4 Kit+1 = Kit ( 1 -+ Rit (1) where Kit and Rit represent knowledge capital and R&D investment for firm i in year t. We will argue that the standard framework contradicts the widely observed pattern that the same firms tend to persistently carry out above (or below) average amounts of R&D, say, relative to their sales. This persistence in the differences in R&D intensities between firms within the same industry is hard to rationalize on the basis of a knowledge accumulation process as specified in equation (1 ). To clarify our point, assume a Cobb-Douglas production function, Qt (1)itX2 1 XK , where Qt is output, (kit a productivity term, and Xit inputs. A firm's rate of return to knowledge capital can then be calculated as aK t This expression implies that if we consider two firms which differ only in their knowledge capital stocks at the beginning of a period, the firm with the lowest R&D capital stock should have the highest return on an increase in its capital stock. Since equation (1) implies that a unit of R&D investment generates a unit of R&D capital, one should expect highest investment by the firm with the smallest R&D capital stock. Note that the argument above is valid even if firms differ in terms of productivity, (Dit-A second weakness of the model is its treatment of other factors that could account for persistent differences in the level of the R&D activity. Such factors are often captured by socalled fixed effects in empirical research on firm level data. While the presence of fixed effects can make the model consistent with the observed cross sectional differences in R&D activity, they are not very satisfactory. First, econometric studies of R&D and productivity based on models with fixed effects often give weak, if significant results, and the estimates are often not robust; see the survey by Mairesse and Sassenou (1991) . Second, while our model suggests a mechanism generating persistent differences in R&D investment, models with fixed effects only account for such differences without offering any explanation how such differences have been generated. An alternative specification of knowledge accumulation A possible explanation for the observation that a high return on knowledge capital does not lead to R&D investments is that the relationship between R&D investment and knowledge capital is more complex than in equation (1). There is an alternative to the perpetual inventory model of capital accumulation that suggests that old capital and new investment are complementary inputs in the production of new capital. This view seems particularly relevant for the accumulation of knowledge capital, as noticed by Griliches
doi:10.2307/20076123 fatcat:k4pxxn75e5gjpfjztn6lqx5xzm