Bidding against the odds? The impact evaluation of grants for young micro and small firms during the recession

Abstract

Impact evaluations of business development grants targeting young firms have been somewhat neglected in the literature. While most research studies focus on the impact of research and development grants, a larger percentage of young firms would benefit from grants that assist them in business development activities. In this paper, we examine the impact of small business development grants on young small firm survival, turnover growth, labor growth, and access to external finances. We study this topic in the context of a long recession in Croatia (2009 to 2014), which makes it possible to better observe the effect of the public instrument intervention. Results show positive effect on firm survival and on obtaining long-term bank loans and no significant effects on firm performance. The grant scheme was most successful for firms newest to the market.

Appendices

Appendix A

Table 9 Basic information on the grant schemes

Table 10 Distribution of government grants

Table 11 Result comparison between our main specification and other matching algorithms

Table 12 Estimation of ATT (single- and multiple-treated)

Appendix B: The derivation of Rosenbaum bounds

Let us assume that the probability of the treatment D for observation i is a function of observed vector of covariates x_i and unobserved variable u_i. More precisely, P(D_i = 1| x_i, u_i) = F(βx_i + γu_i), where F is the logistic function and γ is the effect of the unobserved variable on the probability of treatment. When γ = 0, this means that the study is free of hidden bias and the selection into treatment is determined solely by x_i. When γ ≠ 0, two observations, say i and j, which have the same covariates x_i = x_j, can have different probabilities of treatment if u_i ≠ u_j. Since F is logistic, the odds of treatment for the two observations are $\frac{P_i}{1-{P}_i}$ and $\frac{P_j}{1-{P}_j}$, and the odds ratio is given by $\frac{P_j\left(1-{P}_i\right)}{P_i\left(1-{P}_j\right)}=\frac{e^{\beta {x}_i+\gamma {u}_i}}{e^{\beta {x}_j+\gamma {u}_j}}={e}^{\gamma \left({u}_i-{u}_j\right)}$. If the unobserved variable does not exert any influence (i. e. if γ = 0), or if u_i = u_j, then ${e}^{\gamma \left({u}_i-{u}_j\right)}=1$. Rosenbaum (2002) showed that the following bounds can be put on the odds ratio$:\kern0.5em \frac{1}{\Gamma}=\frac{1}{e^{\gamma }}\le \frac{P_j\left(1-{P}_i\right)}{P_i\left(1-{P}_j\right)}\le {e}^{\gamma }=\Gamma$. Both observations have the same probability to be in treatment only if Γ = e^γ = 1. If for example Γ = e^γ = 2,that means that the probability that observation i receives treatment can be up to twice as big as the probability for observation j, regardless of the fact that probability should be the same for both units according to the observables, which is the result of hidden bias. This is how Rosenbaum bound Γ measures the extent of the hidden bias.