Persistence Effects of Innovation Activities

Part of the ZEW Economic Studies book series (ZEW, volume 38)


Recent empirical evidence indicates that firm performance in terms of productivity is highly skewed and that this heterogeneity is persistent over time (for an overview, see Dosi et al., 1995; Bottazzi et al., 2001; Bartelsman and Doms, 2000). Since innovation is seen as a major determinant of firm’s growth, one hypothesis is that the permanent asymmetry in productivity is due to permanent differences in the innovation behaviour. However, little is known so far about the dynamics in firms’ innovation behaviour, and the evidence is mostly based on patents (see Geroski et al., 1997; Malerba and Orsenigo, 1999; Cefis, 2003a).


Service Sector State Dependence Innovation Activity Exit Rate Individual Heterogeneity 


Unable to display preview. Download preview PDF.

Unable to display preview. Download preview PDF.


  1. 120.
    Theories which focus on how firms accumulate technological capabilities may also be considered as “success breeds success” theories since technological capabilities might substantiate sustained competitive advantages (Teece and Pisano, 1994). However, learning can also occur as a result of unsuccessful innovations.Google Scholar
  2. 121.
    In contrast to most other kinds of sunk costs, firms can strategically decide upon the amount of R&D expenditure. Costs incurred in this manner are, therefore, referred to as endogenous sunk costs (Sutton, 1991).Google Scholar
  3. 127.
    R&D expenditure accounted for 50–55% of innovation expenditure in the period under consideration; see Gottschalk, Janz, Peters, Rammer, and Schmidt (2002).Google Scholar
  4. 129.
    Manez Castillejo et al. (2004) reported transition rates for only small and large firms. Using a weighted average, one would get an exit rate of about 17% and an entry probability of 8%.Google Scholar
  5. 130.
    This result coincides with the decline in the share of innovators at the aggregate level, see Fig. 2.1. A main cause for this somewhat astonishing development was a severe shortage of high-qualified personnel in 2000, hampering a large number of small and medium-sized firms in their innovative efforts (see Janz, Ebling, Gottschalk, Peters, and Schmidt, 2002).Google Scholar
  6. 133.
    But, Monte Carlo studies have shown that this estimator performs quite well for 8 or more time periods. The estimator is based on the idea of getting a reparametrisation in such a way that the incidental parameters are information orthogonal to the other parameters which reduces the order of the bias of the maximum likelihood estimator without increasing its asymptotic variance (see Cox and Reid, 1987).Google Scholar
  7. 134.
    The latter approach, i.e. using the joint density of (yi1,..., yiT) given (yi0, xi) is not the same as treating yi0 as non-random though it includes the model with non-random yi0 as special case (see Wooldridge, 2002: 494).Google Scholar
  8. 135.
    Instead of \( \bar x_i \) , the original estimator used xi = (xi1,..., xiT) in eq. (5.3) but time averages are allowed to reduce the number of explanatory variables (see Wooldridge, 2005).Google Scholar
  9. 141.
    See Veall and Zimmermann (1996) for a discussion of the interpretation and importance of different Pseudo R2 measures in binary choice models.Google Scholar
  10. 142.
    I used the procedure proposed by Wooldridge (2002), i.e. I added the lead of the corresponding variable and tested on the significance of the coefficientGoogle Scholar

Copyright information

© Physica-Verlag Heidelberg New York 2008

Personalised recommendations