4.1 Data
Firm level data on public grants and subsidies in Sweden is collected and stored in the MISS database by the Swedish Agency for Growth Policy Analysis. The MISS database comprises information about the grant distributor and receiver, the size of the grant, and when the firm receives the payments. We link these data with yearly register data provided by Statistics Sweden (SCB) containing information on firms’ input and output, covering all firms in the economy. Hence, information about both the treated and the non-treated firms is collected from these sources.
In addition to firm level data, individual level data on workers’ education, wages, gender, and age is aggregated to the firm level and linked to firm data. Firm level and aggregated individual level data contains information on production, sales, employment, value added, investments, physical capital, profits, industry affiliation, educational attainment of the labor force, geographic location, etc., spanning the period 1997–2011. All datasets are linked using unique individual firm-year ID codes.
Out of the two analyzed programs “Win Now” and “Research & Growth,” Win Now is the smaller program directed at start-ups (firms younger than 1 year at the time of application), and its funds can be granted to SMEs that have developed a new product, method, or service that has not yet reached the market. The aim is to give start-ups a chance to survive in the market by providing financial aid during the commercialization process, which is intended to attract external capital and make the business successful in the future. Hence, future growth is one of the main purposes of the grant. However, the timeframe for achieving growth is not specified. Half of the granted amount should be allocated to business development, while the other half can be used for R&D activities. A total of 1309 firms applied for Win Now, and approximately 10% received support. Win Now has been granted 125 times during the period under study (2002–2010), and the average grant was 164,847 SEK ($18,458). A firm is only granted the subsidy once, and the maximum amount awarded is 300,000 SEK ($33,592). Win Now can also be seen as a springboard to its sister program Research & Growth, targeting slightly older firms.
5
The subsidy program Research & Growth targets small- and medium-sized innovative firms supporting developing projects, but support may also be awarded to pilot studies. Approximately 20% of the applicants were granted support, and the recipients consist mostly of SMEs that are already on the market. The purpose of Research & Growth is to support and promote innovation-driven growth within the subsidized firms. In all, the program provided 546 grants during the studied period (2005–2010), with an average grant of 543,321 SEK ($60,836). The project time is normally between 6 and 18 months.
Earlier empirical studies of the two programs Win Now and Research & Growth have produced mixed results. Some studies have found that they yield growth in employment and sales, while others have found negative outcome in, for example, employment, and productivity (Söderblom et al.
2015; Daunfeldt et al.
2016; Vinnova
2014). The different outcomes in these studies may be due to differences in empirical strategy, discussed in Daunfeldt et al. (
2016).
The innovation grants analyzed here are distributed to firms in cities as well as rural areas. In Table
1, all Swedish regions and municipalities are divided into three different groups:
Big Cities,
Support Areas A&B, and
Other Areas. This division is only used to provide an overview of the regional division of the grant sums. When carrying out the empirical analysis, the regions considered are the 60 Swedish functional labor market regions. The idea with the functional labor market regions is that most of the home-to-work commuting takes place within the region rather than across the borders. In addition, the labor market region should not be larger than it should be possible to, on a daily basis, commute between any two points within the region. For descriptive purposes, we choose to characterize Sweden in three dimensions, from dense city regions to rural areas. The first category,
Big Cities, contains the three largest cities in Sweden: Stockholm, Gothenburg, and Malmo. The second category,
Support Areas A&B, includes particularly vulnerable regions in Sweden. Vulnerable areas are those that have the right to apply for regional support.
6 The third and final group,
Other Areas, comprises the remaining in-between regions in Sweden.
Table 1Total and average grant sums, by group of regions
Big cities | 218 | 215,000,000 | 986,830 | 36.22 |
Support areas A&B | 45 | 22,800,000 | 506,648 | 11.75 |
Other areas | 358 | 288,000,000 | 803,490 | 23.89 |
Total: Sweden | 621 | 525,800,000 | 846,340 | 26.38 |
As seen in Table
1, both the size of the average grant and funding per employee is largest in the
Big Cities. There is a tendency to give more grants to firms in large cities compared to other firms, which is consistent with the internationally observed pattern regarding the distribution of R&D grants.
As noted above, the supply of skilled workers is a key component of innovation and growth. To measure the regional supply of high-skilled labor empirically, we construct a regional index. This index-variable is constructed as a revealed comparative advantage index (RCA-index), where a value above one indicates that a region is above the country average and vice versa. To simplify the interpretation of our regression variables, this index will be centered around zero in our analysis, and thus, a value above (below) zero will indicate that a region is above (below) the country average.
7 The measure of the relative supply of high-skilled labor is as follows:
$$ RCA- Skill=\left(\frac{L_{\mathrm{r}}^{\mathrm{H}}}{L_{\mathrm{r}}}\right)/\left(\frac{L_{\mathrm{Swe}}^{\mathrm{H}}}{L_{\mathrm{Swe}}}\right) $$
(1)
where the first term describes the share of high-skilled labor in the region, and the second term describes the share of high-skilled labor in the country. The use of RCA-indices is common within the international trade literature, starting with Balassa (
1965). In our setting, we might note that the RCA-index captures the relative concentration of skilled labor rather than factors behind this clustering. That is, from the governmental project coordinator perspective, when allocating grants, the spatial distribution of firms and human capital is taken as given.
8
Table
2 reveals a clear picture of the distribution of skilled labor across regions in Sweden: skilled labor is concentrated in the large cities, whereas the opposite is true for the (rural) areas granted regional support measures. The RCA-index (based on 60 labor market regions) and firm level variables, sources, mean values, and standard deviations are described in Table
3.
Table 2Average RCA-index, by region
Big cities | 0.22 |
Support areas A&B | − 0.31 |
Other Areas | − 0.04 |
Total: Sweden | 0.00 |
Table 3Variable description
ln(L) | Log. of number of employees (firm) | IFDB | 1.11 | 1.08 |
ln(Lp) | Log. of inflation adjusted value added per emp. (firm) | IFDB | 5.83 | 0.83 |
ln(sales) | Log. of sales (firm) | IFDB | 7.90 | 1.60 |
Wage premium | Mean of wage premium for skilled labor, divided by sni5 codes (firm) | LISA | 1.93 | 1.68 |
ln(K) | Log. of physical capital (firm) | IFDB | 4.95 | 1.90 |
RCA skill FA | (Skill FA/Emp FA)/(Skill Swe/Emp Swe) (regional) | LISA | 0 | 0.23 |
Post-treatment | 1 = period after support, 0 = otherwise (firm) | MISS, IFDB | 0.0004 | 0.02 |
Treatment | Annually awarded grant/sales (firm) | MISS, IFDB | 0.115 | 0.19 |
R&D/Ind. | Share of researchers by industry (industry) | LISA | 0.117 | 1.36 |
Profit quota | Operating profit/ production value (firm) | IFDB | − 0.52 | 69.3 |
Share of higher educ. | Number of higher educ./total (firm) | RAMS | 0.26 | 0.36 |
R&D int. SSY | Share of researchers by industry/ total number of emp. (Industry) | LISA | 0.01 | 0.09 |
ln(value added) | ln(L) in period (t − 1) (firm) | IFDB | 6.97 | 1.47 |
ln(W) | Log. of inflation adjusted value added (firm) | LISA | 5.15 | 0.79 |
4.2 Matching
As noted above, the R&D grants have both a specific purpose and are targeting a specific population of firms; hence, grants are not randomly distributed across firms. This, in turn, leads to the question of how to create a control group of similar firms. To handle this selection problem, we use coarsened exact matching (CEM) to create a control group of non-treated firms that, in all relevant aspects, are as similar as possible to the firms receiving grants. As shown by Coberly et al. (
2011), CEM matching usually outperforms both propensity score matching (PSM) and Mhalanobis distance matching (MDM) in terms of reducing the imbalance between the treatment and control group. The combination of its intuitive approach, good statistical properties, and easy-to-use has made CEM matching increasing popular in applied empirical work. For recent applications of this matching method, see Croce et al. (
2013), Cumming et al. (
2017), and Grilli and Murtinu (
2014).
9
The matching is based on variables that are relevant for both program participation and program outcomes, and the key idea is that all matching variables should be as similar as possible between the control and treatment groups. Unlike PSM, CEM does not estimate the probability of being treated, but instead it coarsens variables into strata and puts different weights on the control firms depending on how close they are to the treated firms (Iacus et al.
2011,
2012). Detailed descriptions of CEM can be found in Blackwell et al. (
2009) and Iacus et al. (
2011,
2012). The matching performed in this paper is a so-called one-to-one matching, which yields one control firm for each treated firm. Consequently, we do not need to take matching weights into consideration to adjust for differences in the number of observations between the treated and control groups. For each of the treated firms, we match on firm properties 1 year before the treatment, (
t − 1), with
t being the year a firm receives a grant (Caliendo and Kopeinig
2008). Firms receiving multiple R&D subsidies have been removed from the sample.
Results from the matching are presented in Table
4. As noted by Iacus et al. (
2011,
2012), the value of the imbalance test is subordinate to the change in imbalance as given by matching. As shown in Table
4, matching reduces the imbalance for all variables, suggesting that matching leads to a control group that is more similar to the treatment group than to the collection of all non-treated firms.
Table 4CEM matching results, matching imbalance, and number of matched pairs
ln(K) | | 0.04 (0.29) | |
Profit quota | 0.02 (0.11) | 0.03 (0.11) | 0.03 (0.11) |
ln(value added) | 0.04 (0.45) | | |
ln(W) | 0.10 (0.38) | | |
R&D int. SSY | 0.02 (0.22) | 0.01 (0.26) | 0.01 (0.26) |
ln(L) | | 0.03 (0.35) | 0.01 (0.35) |
Share of higher educ. | 0.02 (0.48) | 0.02 (0.49) | 0.01 (0.49) |
ln(capital intensity) | | | 0.07 (0.24) |
Overall (L1) | 0.56 | 0.38 | 0.46 |
Number of matched pairs |
Total Sweden | 481 | 464 | 468 |
Big cities | 182 | 168 | 166 |
Support areas A&B | 28 | 38 | 38 |
Other areas | 271 | 258 | 264 |
1–10 employees | 293 | 278 | 280 |
11–50 employees | 154 | 153 | 154 |
50+ employees | 34 | 33 | 34 |
Table
5 displays the matching results when the region of the treated firm is added as an (exact) matching variable, forcing the control firm to be in the same region as the treated firm. This creates a perfect balance between the treatment group and the control group on the regional variable. Having the “twin” firm located in the same region as the treated firm removes the possibility that subsequent changes in the development of the treated and control firms is due to location. We may also note that matching results from this matching strategy correspond strongly to the matching results presented in Table
4, where no geographical concern was included in the matching. However, forcing the control firm to be located in the same region as the treated firm reduces the number of matched pairs, which is seen when comparing the lower panels of Tables
4 and
5.
Table 5CEM matching results, including region as a matching variable
Region | 0.00 (0.19) | 0.00 (0.17) | 0.00 (0.17) |
ln(K) | | 0.08 (0.29) | |
Profit quota | 0.02 (0.11) | 0.04 (0.11) | 0.04 (0.11) |
ln(value added) | 0.06 (0.45) | | |
ln(W) | 0.10 (0.38) | | |
R&D int. SSY | 0.04 (0.22) | 0.04 (0.26) | 0.04 (0.26) |
ln(L) | | 0.07 (0.35) | 0.08 (0.35) |
Share of higher educ. | 0.06 (0.48) | 0.15 (0.49) | 0.17 (0.49) |
ln(capital intensity) | | | 0.14 (0.24) |
Overall (L1) | 0.71 (1.00) | 0.66 (0.99) | 0.71 (0.99) |
Number of matched pairs |
Total Sweden | 396 | 275 | 282 |
Big cities | 164 | 115 | 119 |
Support areas A&B | 18 | 15 | 16 |
Other areas | 214 | 145 | 147 |
1–10 employees | 241 | 178 | 183 |
11–50 employees | 127 | 80 | 82 |
50+ employees | 28 | 17 | 17 |
Finally, Table
6 provides the mean values of our outcome variables—employment, sales, and labor productivity—divided into six categories. The outcome variables are reported for all firms, treated firms, treated firms before treatment, treated firms after treatment, firms in the control group (original match), and firms in the control group when we include region as a matching variable. Here, we note that subsidized firms have slightly different mean outcome values before and after treatment and that there are some initial differences in the mean outcomes comparing the subsidized and control firms.
Table 6Mean values for dependent variables
ln(L) | 1.17 (1.09) | 2.21 (1.26) | 2.19 (1.28) | 2.30 (1.20) | 1.95 (1.34) | 1.84 (1.37) |
ln(sales) | 8.01 (1.55) | 8.91 (2.01) | 8.92 (2.01) | 8.89 (2.03) | 9.03 (1.81) | 8.70 (1.71) |
ln(Lp) | 5.89 (0.77) | 6.05 (0.73) | 6.04 (0.71) | 6.06 (0.81) | 6.12 (0.69) | 6.21 (0.80) |