The long-term effects of R&D subsidies on firm performance: evidence from a regression discontinuity design

ABSTRACT

Innovation is the most important factor for economic growth, but the government’s ability to affect innovation outcome is contested. This paper presents quasi-experimental evidence on the long-term effects of R&D subsidies for small businesses. We use firm level data on Swedish applicants of the Eurostars R&D program during the period 2008–2019. We find that R&D subsidy has a positive and significant effect on turnover, employment and the number of scientific and technology workers. We also find positive effects on export but with some time lags. The effect is stronger on firms that are expected to be financially constrained, such as small and younger firms. We find that the subsidy effect on turnover, employment and export last for more than 7 years after the end of the project, which is consistent with explanations based on long-term channels such as improved competitiveness through the market introduction of innovative products.

KEYWORDS:

• Innovation policy
• research and development subsidies
• impact evaluation
• regression discontinuity design
• small businesses

JEL:

• C20
• D04
• H71
• L52
• O31
• O38

Acknowledgements

We thank Marieke Bos for her careful review and valuable comments. We thank Camilla Andersson and seminar participants at the Swedish Agency for Growth Analysis and Swedish conference in economics for their feedbacks and comments. We would also like to thank the Eureka Network and Vinnova for providing data.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 The number of patent applications/approval has also been used as a proxi for innovation capacity of firms. Unfortunately, we do not have access to the patent data, and thus not included in this study.

2 For a more extensive review of the literature, including studies that us non-experimental methods, see Becker (2015), Bronzini and Iachini (2014), Bloom, Van Reenen, and Williams (2019) and Ziesemer (2021).

3 This study is not part of our literature review because of its methodology creates an uncertainty of the validity of the results. In the study they emphasize the fact that the control group of rejected applicants were rejected at the final stage and therefore similar to the awardees. However, there is a risk for a selection bias effect in this research design. The design of choice has most certainly decreased the risk of unobserved variable bias.

4 Once the funding decision is made, projects can ask for extension on their deadlines. If there are good reasons this is often approved. It is often extended by a few months according to Vinnova. There is no follow-up funding for Eurostars projects.

5 A natural alternative is to normalize the scores at the cut-off point, i.e., the minimum score values for the subsidized, so that the scores for every call will have a value of zero exactly at the cut-off point. However, given the small number of observations per call, requiring all calls to have a value of zero at the cut-off point will distort the score distribution by increasing the number of observations just at the cut-off point. Similar issue arises if we normalize at the maximum score point among the non-subsidized firms. This motivates our choice to use the midpoint between the maximum score for the non-treated and the minimum for the treated within calls (See Fort et al. 2022 for more discussion on the problem and solutions).

6 As will be shown later, excluding the above controls does not affect the main result, except the precision of our estimates.

7 In a few cases, firms win their first award in their second or further attempts. In this case, we allow such firms to be part of the control group from the year of the first application up to the year before their first grant award. Starting from the first grant year, they become part of the treatment group.

8 We will show that the main result is robust when excluding firms with multiple subsidy awards.

9 In addition, although larger firms can participate as partners, the main focus of the program is on small and medium sized firms.

10 The summary statistics for the years after the subsidy decision is presented in Online Appendix Table A1.

11 McCrary (2008) provides a formal test of manipulation using the density function of the running variable (score). However, such tests are not suitable when the running variable is discrete and the sample size is small.

12 In terms of percentage, the 0.7 log point increase is equivalent to a 101% increase in in turnover, which is obtained as follows: $(\exp (0.7)-1)\ast 100=101\text{\%}$.

13 In robustness section, we show results from estimation of a model similar to a difference in difference, where we estimate equation 2 by adding observations from three years before a subsidy. As shown in Figure 10, we find no significant effect on the year before a subsidy, and the effect only emerges after a subsidy decision.

14 We get a more precise estimate when excluding observation for the first year of subsidy.

15 That is obtained by taking the ratio of the estimated coefficients and the median employment (S&T workers). The median employment (S&T workers) for small firms is 3 (2) workers and for large firms it is 20 (8).