Productivity and R&D: an econometric evidence from Spanish firm-level data

Abstract

This article analyses the relationship between productivity growth and R&D investments of Spanish manufacturing firms during the 1990s. The theoretical model is a version of the Cobb–Douglas production function in its growth rate form. The purpose is to estimate the rate of return to R&D expenditures. The econometric specification is a distributed lag model. The estimation applies the generalized method of moments method. The main empirical finding is that a positive and significant role is played by R&D expenditures on productivity growth. The rate of return to R&D expenditures is 26.598%.

Acknowledgements

Authors thank the Fundación SEPI from Spain for the access to used empirical data. The authors are also grateful for the comments received from the participants at the 6th Annual EUNIP Conference (Turku, Finland, December 2002); V Jornadas de Economía Laboral (Tarragona, Spain, July 2003); 7th ISINI International Congress (Lille, France, August 2003); and XIX Jornadas de Economía Industrial (Castellón, Spain, September 2003). Nonetheless, results and conclusions of this article are exclusive responsibility of the authors.

Notes

¹ Data extracted from Eurostat.

² In this way, see the technological capital model from Griliches (1979), which, in addition to the usual productive factors, provides in the production function another differentiated productive factor that could be called research capital, technological capital or R&D capital.

³ In this approach, researchers usually make different estimates by assuming several values for this depreciation rate.

⁴ See Clark and Griliches (1984) and Griliches and Lichtenberg (1984).

⁵ See Hall and Mairesse (1995), for example. They use a depreciation rate of 15%.

⁶ See, for example, Bessen (2001).

⁷ See Mairesse and Sassenou (1991).

⁸ See in the Appendix the industry classification of the two-digit level used by the ESEE.

⁹ Measuring the output through the value added is frequent in the economic literature. See, for example, Odagiri and Iwata (1986), Hall and Mairesse (1995), Mairesse and Hall (1996) and Rouvinen (2002).

¹⁰ However, the value of the rate of return is somewhat lower than that obtained by the corrected values. Nonetheless, the difference in the estimates for adjusted and nonadjusted data is practically cancelled if the corrections are only made for the labour variable, but not for the rest of variables with a similar problem. For more information, see Hall and Mairesse (1995) or Smith et al. (2004).

¹¹ The use of the book value of the equipment in real terms as a proxy of the physical capital stock is frequent in the economic literature. See, for example, Clark and Griliches (1984), Hall and Mairesse (1995), Mairesse and Hall (1996), Beneito (2001), Wakelin (2001) and Parisi et al . (2002).

¹² The firms’ total R&D expenditures refer to those made by both private funds and public sector subsidies. They are deflated by the aggregate Industrial Price Index provided by the INE.

¹³ See Clark and Griliches (1984).

¹⁴ It is defined in the ESEE as average percentage of utilisation–during the year–of the firm's standard capacity.

¹⁵ Econometric arguments also support the introduction of lags to the technological effort variable. On the one hand, due to the correlation that might exist between the actual investment in R&D and the value added of the period. On the other hand, the presence of the value added in both sides of the equation might generate biases in the coefficient of the technological effort variable. See Mairesse and Hall (1996).

¹⁶ See, for example, Griliches and Lichtenberg (1984) and Rouvinen (2002).

¹⁷ See, for example, Mairesse and Hall (1996) and Wooldridge (2002, especially Chs. 8, 11 and 14).

¹⁸ Significant problems relative to outliers were not found.

¹⁹ A balanced panel would contain an econometrically insufficient number of firms.

²⁰ This is similar to the published results of Clark and Griliches (1984) and Hall and Mairesse (1995), for example. The empirical results with other lags are available from the authors by request.

²¹ Several instruments were analysed and those with the best econometric results were chosen. They are similar to those applied in other studies, which also use three or four lags for some explanatory variables of the model. For instance, see Mairesse and Hall (1996).

²² The inclusion of the capacity utilization variable might be capturing the different characteristics of each industry and, as a consequence, reducing the significance of dummy variables.

²³ This result was also presented by Smith et al . (2004) for Denmark and Wakelin (2001) for the United Kingdom. As it is argued by the latter, one might attribute that result to the exclusion of raw materials and intermediate products from the production function.

²⁴ See Mairesse and Sassenou (1991) concerning articles about rate of return to R&D and Atella and Quintieri (2001) in relation to articles about elasticity of total factor productivity with respect to R&D capital stock.

²⁵ See Mairesse and Sassenou (1991) and Hall and Mairesse (1995).

²⁶ See Nadiri (1993).