Page 25 - IDEA Study 8 2017 Direct subsidies and R&D output in firms

context. In other words, after the matching the treated and untreated firms’ difference in propensity to apply for Czech IP protection holds, albeit with a smaller magnitude as could be expected, but their difference in propensity to apply for international IP protection fizzles out. Table 4. Propensity score matching estimates by the office of IP protection (NN 3 estimator, conventional) IP protection Czech IP protection International IP protection Program Year Treated me IMPULS (t) 0.185 (t+1) 0.168 (t+2) 0.213 TIP (t) 0.228 (t+1) 0.287 (t+2)† 0.302 ALFA (t) 0.285 (t+1)† 0.296 (t+2)† 0.263 IMPULS (t) 0.023 (t+1) 0.025 (t+2) 0.028 TIP (t) 0.042 (t+1) 0.042 (t+2)† 0.042 ALFA (t) 0.042 (t+1)† 0.042 (t+2)† 0.031 Untreated Difference N 20,104 20,104 20,104 20,783 20,783 20,783 16,027 16,027 16,027 20,104 20,104 20,104 20,783 20,783 20,783 16,027 16,027 16,027 0.152 0.033 0.139 0.028 0.154 0.058** 0.221 0.007 0.196 0.091*** 0.191 0.111*** 0.239 0.046 0.205 0.091*** 0.159 0.103*** 0.033 -0.010 0.025 0.000 0.029 -0.002 0.039 0.003 0.025 0.018* 0.023 0.019* 0.048 -0.007 0.033 0.009 0.016 0.015 (0.025) (0.025) (0.028) (0.027) (0.024) (0.027) (0.028) (0.030) (0.028) (0.011) (0.011) (0.012) (0.018) (0.011) (0.012) (0.017) (0.013) (0.010) Note: The effect of the subsidy programme is the difference between the estimated probabilities of applying for IP protection for treated and untreated firms. Positive values imply “crowding-in” effects: R&D subsidies stimulate IP applications, which would not be produced, if the subsidy was not provided. The effects are reported for the first (t), second (t+1) and third (t+2) year after the start of funding from the respective programme. Abadie Imbens robust standard errors in parentheses; * p<0.05, ** p<0.01, *** p<0.001. † IP application data are not complete due to the publishing delay. In the next step, we control for unobserved time-invariant fixed effects with the help of a difference-in-differences estimation. Using this method, we are able to account for differences that are constant, such as the firms’ birth characteristics, or tend to change slowly over time, such as their innovative behaviour, but which are not represented by the variables in hand. More specifically, we combine the baseline 23