Threat of termination and firm innovation

Abstract

This study investigates how the ex-ante threat of termination affects firm performance in innovation measured by number of patents and citations. Empirical results show that the threat of termination is negatively associated with both measures of firm innovation. This relation however is sensitive to industry structure. The negative effect of the threat of termination on innovation is statistically significant only for high-tech firms. For low-tech firms there is no statistically significant relation between the threat of termination and firm innovation. One plausible explanation is that high-tech firms are inherently risky and have higher rates of project failure. Adding the risk of higher threat of termination makes the manager more risk averse and forces her to avoid investing in value increasing innovations. Managers in low-tech firms don’t face such pressures. The policy implication is that high-tech firms should lower threat of termination and increase tolerance for project failure to encourage innovation.

Appendix

1.1 Truncation bias in patent counts

The truncation bias in patent counts exists because there is an average lag of 2 years between the time an application is filed and the time the patent is granted. This is not a problem in the beginning of our sample (1990) as the average patent filed by the year 1990 was granted by the year 1992. However, as we approach towards the end of the sample the data does not report the patents that will eventually be granted in the coming few years. We therefore observe large number of missing patent data in the last years of the sample. Hall et al. (2002, 2005) show that correcting the truncation bias in patent counts is simple and straight forward. We follow Hall et al. (2005) to correct the truncation in patent counts. The method involves in calculating weight factors based on the application-grant distribution of patents in the sample and then multiplying number of patent counts by the respective weight factors. The patent counts are corrected for the years 2001–2006. Based on Hall et al. (2005) the following formula is used to correct for the truncation in patent counts.

$$\begin{aligned} \text {Patent}_{\mathrm{t}}=\frac{patent_t}{\mathop \sum \nolimits _{k=0}^{2005-t} weight_k } \quad 2006 \le \mathrm{t} \ge 2001 \end{aligned}$$

(3)

1.2 Truncation bias in citations

The truncation bias in citations exists because citations to a patent keep on coming long after the patent was granted but we only observe citations in the data that are available by the last year of the data. Hall et al. (2002, 2005) use the “quasi-structural” approach to correct the truncation bias in citations. They first estimate the life time citation-lag distribution over the entire data period (1976–2006). Given this distribution they estimate the total citations of any patent for which only a portion of its citations life is observed by dividing the observed citations by the fraction of the population distribution for the time interval during which the citations are observed. A detailed description of the method can be found in Hall et al. (2002). We don’t need to correct for the truncation bias in citations in our sample as the updated NBER data now provides truncation corrected data on citations using Hall et al. (2002).