A Schumpeter-inspired approach to the construction of R&D capital stocks

Abstract

A new method for constructing R&D capital stocks is proposed and tested. Following Schumpeter, the development of R&D capital stocks is modelled as a process of creative destruction. Newly generated knowledge is assumed not only to add to the existing R&D capital stocks but also, by displacing old knowledge, to destroy part of that capital. This is in stark contrast to the perpetual inventory method, which postulates a constant rate of depreciation. We compare both methods by estimating the impact of R&D and spillovers on output of 9 industries in 12 OECD countries, and find that the new approach leads to more sensible and robust results.

Acknowledgements

We thank Holger Görg, Paul Gregory, Bronwyn H. Hall, Almas Heshmati, Pierre Mohnen, Wolfram Schrettl, Philipp J. H. Schröder, Sabine Stephan, the participants of the NBER productivity lunch seminar and an anonymous referee for helpful comments and suggestions. The usual disclaimer applies.

Notes

¹ Only a part of the differences can be explained by different data sources and aggregation levels used.

² Only a few studies have recently addressed the problem of determining the depreciation rate (Nadiri and Prucha, 1996), the gestation lag (Esposti and Pierani, 2003) and the impact of the assumed depreciation rate on the estimation results (Hall and Mairesse, 1995).

³ Based on the work of Terleckyj (1974, 1980) a small number of studies use R&D expenditures or R&D intensities as a proxy for the R&D capital stock.

⁴ Please note that there are other external factors which might lead to a depreciation of the R&D capital stock and therefore our approximation of the R&D capital stock might be too high. Examples of such external factors are the oil price shock in the 70s as pointed out by Sterlacchini (1989), the transformation in Eastern Europe, or changes in the legislation as pointed out by Bitzer (2005).

⁵ The depreciation rate of the R&D capital stock can be obtained by setting the investments in R&D to zero. Thus, . The right-hand term is obviously not constant.

⁶ Further estimations with depreciation rates of 5, 15 and 20% have been carried out as well. The results are not significantly different from those reported later in this article.

⁷ In Bitzer (2005) the sensitivity of results with respect to the specification of different time lags is tested. The results turn out to be quite robust with respect to the variation of time lags.

⁸ The sensibility of the SIM referring to the substitution rate was tested in Bitzer (2005). Estimations with substitution rates of 0.95, 0.90 or 0.80 did not produce significantly different results.

⁹ As the arguments of Zellner et al . (1966) hold for an aggregated CD-production function, (6) can be estimated by OLS producing consistent estimates of the parameters.

¹⁰ Estimations showed that the R&D capital stocks are only significant if they enter the estimation with a lag of at least 1 year. This is in line with the findings of Coe and Helpman (1995) and van Pottelsberghe de la Potterie and Lichtenberg (2001).

¹¹ Note that since and are constructed as linear combinations from , this also automatically leads to a rejection of the unit roots hypotheses for and .

¹² The estimation results of the fixed group effects are available upon request.