Pecuniary Knowledge Externalities across European Countries—Are there Leading Sectors?

Abstract

This paper investigates empirically the occurrence of pecuniary knowledge externalities at the sector level across European economies. The main results suggest that, although some sectors can be considered as playing a particularly important role as a source of pecuniary knowledge externalities in the majority of examined countries, there exist significant national differences in the occurrence of these effects. Moreover, such external effects influence the dynamics of total factor productivity in downstream sectors and appear as a relevant source of growth in modern economies. As such, the concept of pecuniary knowledge externalities, as opposed to pure knowledge externalities postulated in the new growth theory, provides a new understanding of the growth process.

Keywords:

• Pecuniary knowledge externalities
• knowledge production function
• intermediate goods transactions

Notes

¹ See Kline and Rosenberg (1986).

² On the effects exercised by information technology on the US economy, see Jorgenson and Stiroh (2000), Oliner and Sichel (2000) and Jorgenson (2001). For the evidence (rather scarce) outside the USA, see Daveri and Silva (2004) who show that the reduction in prices of electronic components contributed greatly to the efficiency gains of ICT producers in Finland.

³ In some recent contributions authors discussed the relevance of costs necessary to imitate (Mansfield et al., 1981) to absorb knowledge (Cohen and Levinthal, 1990; Griffith et al., 2003), as well as of transaction, communication and interaction costs (Nelson, 1993). The presence of these costs, in addition to other costs deriving from the acquisition, when knowledge doesn't spill over freely between producers, is an important precondition for PKE that, nevertheless, occur only when the overall cost of external knowledge is significantly lower than the cost of its early generation.

⁴ For a more complete treatment of the concept of knowledge production function, see Patrucco (2009).

⁵ For a summary discussion of the subject, see Lipsey and Carlaw (2004). A number of studies, among which Jorgenson and Griliches (1967), Nelson (1981) and Hulten (2000) extensively discussed theoretical difficulties regarding especially methods and assumptions (to be) made on TFP.

⁶ According to Parisi et al. (2006), a possible way to overcome the assumption of perfect competition would imply the calculation of TFP with cost, instead of revenues, shares of factors. In the case of capital, this would require the computation of the shadow value of capital—a rather cumbersome operation, given the availability of data and market imperfections. Nevertheless, the authors go ahead with the operation, but obtain rather insignificant changes in the results.

⁷ For a more complete discussion on this issue, see Antonelli and Scellato (2007).

⁸ Among other contributions in this field, see Duguet (2006) and Parisi et al. (2006). In particular, Duguet examines the relation between the residual and other measures of innovation. Supported by a series of robustness checks, he finds, among others, that only radical, as opposed to incremental innovation significantly contributes to TFP growth. Parisi et al. (2006) analyze the effect of innovations on growth and report an evidence of a significant effect on productivity driven by the introduction of process innovation. The results hold independently on the fact that alternatively an augmented production function or standard measure of TFP growth, based on a Thörnquist index, is used.

⁹ It is not unreasonable to consider time lags of downstream TFP growth rates with respect to expenditure coefficients, as the positive effects associated with the internal exploitation of externally generated knowledge may occur with some delay. However, as the number of observations in each of 13 estimations is limited, the substitution of contemporary data with lagged variables would reduce considerably the dimension of panels. On the other hand, an extended analysis in the real business cycle literature concentrates on a contemporaneous time series analysis in order to find reasonable determinants of common economic fluctuations, as well as to answer a related question, whether independent sectoral variations in productivity may result in an aggregated fluctuation of GDP (Long and Plosser, 1983). While the analysis at a relatively high level of aggregation (in a six-sector model), even if not particularly strong, offers a positive evidence for this statement, there is no confirmation within a lower level of aggregation. Analogously, given a high degree of disaggregation in the present analysis, it is likely to exclude that the empirical results are driven by common co-movements between sectors. Such an interpretation notwithstanding, as robustness check, for each country an equation with lagged values of TFP growth rates has been estimated. Although to a lesser extent than in the original estimation, the results appeared to be still significant.

¹⁰ For some countries the time spread has been chosen differently: for Norway it was possible to construct a longer panel going from 1992 to 2005. For the Czech Republic, due to the missing data for the volume value added, for the employment and for the capital stock, the panel starts in 1996, but ends in 2006. For Finland, France, Germany and Sweden the time spread goes till 2005, while for Austria and Belgium two years, namely, 1996 and 1998, are missing due to unavailable Use tables for these years. Also for Spain and the UK some observations are missing, particularly 2002–2004 in the case of Spain and 2004 for the UK.

¹¹ In order to diminish concerns about risks of endogeneity, expenditure coefficients have been calculated separately for every year, for each single sector in all analyzed countries.

¹² Due to a low correlation coefficient in all estimations, the fixed effect method prevailed over the random effect method.

¹³ This empirical evidence on the chemical sector confirms the previous findings by Drejer (2000) and Düring and Schnabl (2000).

¹⁴ See Jorgenson and Stiroh (2000) and Oliner and Sichel (2000).

¹⁵ This additional estimation offers one more contribution to the robustness analysis. In particular, the main estimating equation is based on the contemporaneous effects, and as such it runs the risk of being driven by common economic movements considered in the real business cycle literature. However, the results from the additional estimation here performed, based on the same technological components as the main regression, seem to contradict such a possibility.