State Aid Causing Distinct Incentives? Quasi-Experimental Measurement of the Effects of Grants on Private Enterprises in North Macedonia

Abstract

The objective of this study is to measure the effects of state aid distributed to private enterprises in North Macedonia on the enterprises’ efficiency. We examine the governmental Plan for Economic Growth (PEG) pursued through the Financial Support of Investment Law (FSIL) and the Fund for Innovation and Technological Development (FITD). We rely on a rigorous impact evaluation method, whereby comparison groups are sourced from the pool of rejected applicants for the two programs. We pursue conditional matching on firms’ observables and then apply the difference-in-differences method to isolate the effect of the subsidy. FSIL State Aid showed largely ineffective, with producing hardly any difference in sales, investment, wages, or profits among recipients, except employment. These results demonstrate the absence of the incentive effect as it relates to FSIL state aid. On the other hand, the FITD State Aid showed to be considerably effective, as the recipients were found to have increased their sales revenue and their investment in technology and profits, corroborating the presence of the incentive effect as it relates to FITD state aid. Notwithstanding the legitimacy of the policy objective to equalize domestic and foreign firms in terms of their access to state aid, these findings document partial failure of the PEG to do so due to its structural problems.

JEL classification: H25, C90

Keywords

state aid incentive private enterprises North Macedonia

Introduction

In 2017, a new administration came to power in North Macedonia after a protracted political crisis (2015–2017). At the center of the crisis was a large-scale wire-tapping scandal revealing high-ranking officials’ potential corruptive and criminal activities. The political crisis took place at the end of the decade-long rule of the main right-wing political party in the country, whose economic policies were pro-business and included attraction of foreign direct investment, a low-regulation environment, and a low tax wedge. It was only in 2007 when North Macedonia developed the “Invest in Macedonia” policy to attract foreign direct investments. This policy featured direct tailor-made subsidies to foreign companies to invest in the newly-created industrial zones in the country. The country saw increased FDI inflow following the enacting of this policy, although the picture of these effects was somewhat blurred by effects of the Global Financial and the European Sovereign Crises. These increases brought about significant internal debate on state involvement in attracting foreign business. The debate centered on two main issues: the amount spent in subsidies and tax reliefs, and the government’s perceived neglect of the domestic firms, which were already falling behind due to low productivity and fragile international competitiveness.

The productivity gap between domestic and foreign-owned companies in North Macedonia is large. Preliminary observations based on the World Bank’s Enterprise Survey 2019 suggest that the productivity of domestic enterprises ranged between 15% and 20% of that of enterprises under full foreign ownership. Hence, the country in essence operates a dualistic economy: productive and competitive foreign-owned firms in the industrial zones which predominantly export; and fragile and domestically selling domestic firms located outside industrial zones. The “Invest in Macedonia” policy fully neglecting domestic companies also ignored the potential benefits of involving domestic companies in the value chains of the foreign ones. This neglect is borne out by the numbers: In 2017, only 2% of the purchases of the foreign firms were sourced from the domestic firms (Trajkovska and Petreski 2018), perpetuating the dual-economy problem.

The new government administration that came to power in 2017 introduced a pro-transparency position towards state aid granted to companies. It revealed publicly the amounts of the subsidies granted to FDIs which entered the industrial zones in the last decade and undertook steps to level the playing field for both foreign and domestic companies. The Plan for Economic Growth (PEG) was adopted in 2017 as a large state aid program to provide domestic companies fairly equal treatment as the foreign ones entering industrial zones. PEG awards state aid to domestic companies through three pillars: supporting investment and jobs (pillar 1), new market expansion (pillar 2) and innovation support (pillar 3). The first two pillars have been operationalized through the Financial Support of Investment Law (FSIL), adopted in 2018. The third pillar has been operationalized via the Fund for Innovational and Technological Development (FITD), with an assessed (programmed) value of about 100 million euro, or about 1% of GDP.

FSIL awards state aid to private companies in the form of partial reimbursement of an investment made in the preceding year. It is a more traditional state aid scheme where the basic “incentive” assumption is violated, that is, that the program generated incentive for the company to invest. Even if a company pursues an investment today because there is a subsidy to be obtained next year, the amount per company is linearly reduced if the total requested budget per year is higher than what the government planned to spend. Hence, any incentive to pursue an investment due to the subsidy is minimized. FITD, by contrast, awards a grant to companies which are to make an investment, as a share of the programmed costs, after the business plan is positively evaluated in a two-step procedure. It is unlikely that the applicants would undertake the investment (at least not in the same proposed form and size) if the grant is not awarded. Hence, for FITD, the incentive effect is present ex-ante, although it does not guarantee the program’s success.

The objective of this study is to evaluate the effects of the two state aid programs stemming from the Plan for Economic Growth, by referring to the notional (non-) existence of the incentive effect. To obtain this objective, we rely on a rigorous impact evaluation method, and more specifically on the difference-in-differences technique, to isolate the effect of the state aid on a range of efficiency indicators. Namely, we observe the enterprises that obtained state aid versus a comparison group of similar enterprises who commenced an investment or committed to invest but who did not get state aid. We source the latter from the pool of rejected applicants for the analyzed state aid programs. We analyze the two programs—for simplicity called FSIL State Aid and FITD State Aid—programs executed during or by the end of 2018, so that we are able to compare 2017 or 2018 (pre-support) and 2019 (post-support).

The study is the first of its kind in North Macedonia. As such, it is innovative in several ways and at multiple levels. At the academic level, the study enriches the literature on the subject with an evaluation of state aid programs in a transition country, whereby the low efficiency of the spending of public resources is frequently a key barrier to achieving higher and sustained growth. On a practical level, the study is the first in the country to rigorously evaluate any state aid program, let alone a study doing an evaluation to disentangle the actual incentive effects of policies. At the policy level, the study provides evidence that, despite the government’s goodwill desire to provide a level splaying-field for domestic companies and foreign ones, it partially failed in its attempt due to poor program design. Hence, the study has the potential to change the discourse surrounding the evaluation of state aid programs by policymakers in the future. At the level of public consideration, the study likewise has the potential to change the discourse of the public debate from subjective assessments and findings concentrated on mere amounts of money granted to data-driven and evidence-based evaluations that have state aid efficiency implications.

The study is structured as follows. state aid programs under evaluation describes the Plan for Economic Growth in order to display the features of the two state aid programs subject to evaluation. impact evaluation design provides a detailed overview of the applied impact evaluation technique and the underlying data. Results and discussion presents the results and offers a discussion. Conclusion and policy inference concludes and offers policy inferences.

State Aid Programs Under Evaluation

State aid must induce its beneficiary to change its behavior, that is, undertake an activity (an investment, an R&D project, etc.) that it would not have without the aid or would have done in a less desirable manner (Buelens et al. 2007; DG Competition 2013; Martin and Strasse 2005). When state aid causes this change in behavior, it is known as the incentive effect. In addition, the behavior has to lead to the achievement of the objective of the aid and not merely induce beneficiaries to undertake riskier projects that are less beneficial to society (Nicolaides 2009). Moreover, “it can readily be assumed that the aid is distortive in the sense that it provides the beneficiaries in question with windfall gains.” (DG Competition 2013, p. 9). The simplest way of checking the presence of an incentive effect ex-ante is by using the standard level of assessment—if the beneficiary applies for aid after starting the investment, there is no incentive effect (Nicolaides 2009). The absence of the incentive effect can be considered a failure to achieve the objective of aid, unless there are significant and positive indirect effects.

The Financial Support of Investments Law¹ (FSIL) of North Macedonia was adopted in May 2018 with the goal of promoting investments and exports and the creation of well-paid jobs in the private sector.² It consists of two sets of measures that are included under the first and the second pillar of the government’s Plan for Economic Growth 2018–2021 (PEG)

Measures Supporting Investments

1. Support for new jobs;

2. Support for establishing and promoting cooperation with suppliers from North Macedonia;

3. Support for establishing organizational forms for technological development and research;

4. Support for investment projects of significant economic interest;

5. Support for increasing capital investments and revenues; and

6. Support for purchasing assets of undertakings in difficult circumstances

Measures Supporting Exports

7. Support for increasing competitiveness on the market; and

8. Support for entering new markets and for sales growth.

Measures cover new productive investment in tangible and/or intangible assets related to establishing new business entity, extension of existing production capacity, introducing new products, substantive alteration of the entire production process and acquisition of fixed assets from a company in liquidation. Companies can apply for multiple measures at once and combine measures from the two sets. This is why the study analyzes the law as a single state aid program, which we denote “FSIL State Aid.” FSIL provides support only to successful companies which have already carried out an initial productive investment. Other eligibility criteria include: revenue growth in the last year compared to the 3-year average; and non-negative jobs growth over the same horizon.³ Companies with certain characteristics are excluded, such as: public companies, companies that use agriculture subsidies, companies that perform licensed activities, concession beneficiaries, producers of military equipment and of excise goods. The maximum amount to be awarded for a single recipient under FSIL ranges from 17% to 50% of the realized investment cost for investment ranging between EUR100 million and EUR50 million, respectively. In 2018, the average awarded amount was EUR40,000, while the maximum about EUR600,000.

The FITD promotes innovations and technological and human capital development in micro, small, and medium-sized enterprises (MSME), with the aim of increasing their productivity, as well as the creation of highly skilled jobs and increased overall competitiveness of the targeted enterprises. This fund’s activities partially form the third pillar of the PEG. Its activities include the following:

1. Co-financed grants for fast-growing SMEs, so-called “Gazelles”;

2. Co-financed grants for micro enterprises;

3. Co-financed grants for improvement of innovativeness;

4. Co-financed grants for professional development and practice for newly employed young people; and

5. Creating an environment and legal basis for the development of venture capital.

Measures from the PEG’s third pillar only apply to MSMEs, with the exception of the third measure (improvement of innovativeness) which allows large companies to apply as well. Hence, we are herewith only concerned with the third pillar of PEG, disregarding the utilization of these instruments for other programs of FITD. The process of application and granting is generally the same for all measures. First, FITD announces a public call containing information, such as the conditions and budget guidelines for granting. After enterprises submit their project proposals, they are graded on a scale from 0 to a 100 according to multiple criteria and those that score 51 or above move on to the next phase, wherein Committee members vote on which projects are to be funded. The following four criteria are key in the process of evaluation: technological progress and degree of digitalization; project quality; project team capacity; and impact. In 2018, the amounts awarded averaged EUR 130,000, with a maximum of EUR 325,000, and ranging between 16% and 70% of the total value of the project.

FSIL and FITD State Aid are essentially not mutually exclusive, but the same company cannot apply for both with the same investment project. In 2018, both programs exhausted the programmed budget.

Figure 1 presents the average difference between the two state aid schemes in employment, sales, and investment before and after the receipt of the state aid. Companies receiving the FSIL State Aid created more jobs on average (1.6 per company) compared to those receiving FITD state aid (1.4). In addition, the jobs created by FSIL state aid cost the government almost three times less to create (28 vs 70 thousand EUR per job created). Likewise, FSIL-supported companies exhibited larger absolute amounts of sales and investment, whereby the cost to the government per additional euro of sales and investment was 0.21 EUR for FSIL and 0.09 EUR for FITD. On the other hand, these unitary costs were considerably higher in FITD-supported firms: 1.02 EUR and 1.08 EUR, respectively. Therefore, at first sight, FSIL State Aid may be associated with larger multiplier effects than FTID state aid.

Figure 1.

Average difference in key outcomes and related cost effectiveness of state aid programs. Source: Author’s calculations.

In conclusion, the FSIL State Aid is a more cost effective subsidy program than FTID state aid. However, this observation may be premature and may push readers to an incorrect conclusion if the cost effectiveness is observed without also considering the control group—what has been happening in the comparator non–aid-receiving companies over the same period of time, for both types of aid. We now turn to an evaluation of such causal effects.

Impact Evaluation Design

Logic of Intervention and Outcome Indicators

Since our objective is to evaluate private enterprise state aid programs, we provide the following logic model (Figure 2). First, the program is announced and all enterprises who are eligible are invited to apply (block A). For FSIL State Aid, to be eligible means that companies conducted a productive investment in the previous year, experienced sales growth, and did not reduce their employees’ number by more than 5% annually. For FITD State Aid, to be eligible means that companies commit to making an investment (according to their project proposal they submit for funding) should they receive the state aid. However, measuring outcomes solely based on block A in Figure 2 may mean that what we observe is not an outcome of the program (state aid) but rather results which the enterprise would have achieved even without the program.

Figure 2.

Intervention logic. Source: Drafter by the author.

Second, the grantor may be applying some form of selection, or so-called “rationing” (block B in Figure 2). Usually, rationing is important for evaluation, since the pull of the rejected applications leads to a more credible inference about the impact of state aid than generic non-applicants, because, for example, a researcher is then able to observe whether a company will undertake a desired activity even in the event that they are rejected. (Bondonio and Martini 2012). FSIL State Aid does not apply rationing in this sense because all eligible applicants are awarded state aid. However, applicants are rejected if they fail to satisfy some of the additional criteria, including criteria related to maintaining the size of one’s staff, not having obtained any agricultural subsidies, not working in a licensed domain, and others. In case the demand for subsidies exceeds the budget, then a linear reduction of the budget is applied to all applicants who meet these minimum criteria, and all are granted some amount of money. The FITD State Aid, on the other hand, includes evaluation by a panel of domestic professionals on a 0–100 scale in the first step (where a threshold of 51 is applied to move on). In the second step, foreign experts review the evaluations of the first step and vote with the power to alter the decision of the first step on objective grounds.

Finally, a certain number of companies receives the state aid (block C in Figure 2). The question is whether the state aid could modify the behavior of the aided enterprises, or whether it was simply the enterprises who had already invested (or who may have been on the verge of making the investment) ran across the program and decided to apply. Our null hypothesis is that state aid produces plausible outcomes for the subsidized enterprise (block D on Figure 2) on several fronts, which correlate to what the FSIL State Aid and FITD State Aid programs are designed to encourage:

1. Increased revenues (sales revenues),

2. Increased investment in machinery and equipment (tangible assets),

3. Increased investment in technology (intangible assets),

4. Higher wages (payroll costs),

5. Higher employment (number of employees),

6. Increased gross profits,

7. Increased net profits,

8. Increased labor productivity (sales revenue over number of employees, which may signify that the workers in a given company became more productive due to the investment, which would be a reflection of the increase in overall efficiency. This thus correlates with an increase in the company’s total factor productivity; or that the company advance technologically so that more sales is generated with the same number of workers).

The literature is abundant of measuring these types of outcomes. Cassidy and Strobl (2004) and Criscuolo et al. (2019) find that investment subsidies for the manufacturing sector are effective in creating employment. Criscuolo et al. (2019) prove that such an increase in employment is related to a decrease in local unemployment generally, rather than it being a result of them ‘stealing employees’ from other firms or nearby areas. Einiö and Overman (2020) found that a place-based policy in the UK caused a transfer of employees from one deprived area to another, although the program was targeting local non-tradable companies.

Even if production and employment increase, productivity may not, as was the case in Bondonio and Martini (2012) and Criscuolo et al. (2019), Bondoni and Martini (2012) evaluate two programs: a national investment grant scheme in Italy known as “Law 488” and a set of support programs for SMEs in Piemonte. They find a lack of effect on productivity from “Law 488,” while the SME-Piemonte support programs show a modest increase in productivity for loans and interest rate subsidies, but not for grants. Either way, neither Bondonio and Martini (2012) nor Criscuolo et al. (2019) find a decrease in productivity. Bergström (2000) finds that productivity growth increases in the first year, but then takes a negative turn. Kuhn (2010) and Van Cayseele et al. (2014), when conducting similar studies, found that productivity increased. Srhoj et al. (2019) analyzed the effect of a SME grant scheme on productivity and other firm performance indicators in Croatia and found positive effects, which were particularly stronger for small firms.

The effects of these types of schemes on innovation have been mixed. González et al. (2005), Czarnitzki et al. (2007), and Görg and Strobl (2007) all find positive effects in terms of increasing R&D or of preventing the cessation of such activities. However, in Görg and Strobl (2007), these effects only apply to domestic companies that receive small grants. González et al. (2005) add that most subsidies go to companies that would have undertaken an R&D project anyway, and this lack of the incentive effect is also found by De Blasio et al. (2015). A similar conclusion is drawn by Wallsten (2000).

FSIL, however, is different from policies usually evaluated by the literature, in the sense that there is neither an evaluation process nor a competitive auction – like in Bondonio and Martini (2012)—nor it is a place-based policy—as in Criscuolo et al. (2019). It conceptually—not necessarily technically—resembles a nation-wide tax credit on investment, like the one early evaluated by Auerbach and Summers (1979) in the US. They conclude that the policy has been ineffective: it crowded-out non-favored investment that offset the increase in the capital stock built because of the tax credit. Later, empirical investigations offered no unique answer: Goolsbee (1998) documented increases in prices of investment foods rather than increases in real investment, while House and Shapiro (2008) showed that the tax credit had visible impact on capital expenditure.

Due to the risk stipulated under block C on Figure 2, we have a reason for concern—when the FSIL awards state aid for an investment already commenced, it may generate a deadweight loss for the society. This loss will likely appear to generate extra cash for the company, known as the windfall effect. Because of these considerations, we also use the following indicators as outcomes:

9. Working capital (current) ratio (current assets over current liabilities),

10. Cash ratio (cash over current liabilities),

These indicators are used in addition to the profit-related indicators already captured within the first group.

All variables are used in their differences between the 2 years observed for each program.

State aid may also produce area- and economy-wide effects (so-called indirect effects) as stipulated in blocks E and F of Figure 2, but they are beyond the scope of this study.

The Study’s Counterfactual, Assumptions, and Estimation Method

To identify the causal effect of the state aid programs, we need to compare the observed changes in the supported companies with the changes that “most plausibly” would have occurred over the same period of time, for the same firm, had it not received the state aid. However, it is not possible to observe the same company over the same period of time in both the “receiving” and “non-receiving” condition. This hypothetical situation is called the “counterfactual” and is not observable. The counterfactual change must be inferred by reconstructing the treatment group as closely and specifically as possible to form a comparison group of other enterprises who, despite not receiving any subsidy, are similar enough to represent what would have happened to subsidized firms had they not received the state aid.

To form this comparison group, in the FSIL case, we are able to utilize the rejected applicants who did not satisfy some of the criteria for the program and were rejected. There is another small group of applicants who were first accepted but then the contract for state aid was not executed (i.e., no disbursement of funds followed). There is no information about why this happened—if these companies did not satisfy some specific criteria or they had been offered another more favorable funding opportunity and hence withdrew. In the latter case, our evaluation would wrongly attribute the effect of another funding opportunity on changes in our indicators, which we clearly want to circumvent. To avoid this, we focus on the first group: companies who invested but failed to satisfy some of the criteria and were hence rejected. This may not be the ideal group though, particularly because they may differ from the selected on unobservables. Later, we argue that they may be similar on main unobservable “desire to invest,” as both pursued investment in the previous year. Yet, this may be too strong an assumption. It should also be noted that one of this group of companies is not used because it went bankrupt during the course of the study.

In the FITD case, we have all the applicants at our disposal, whose applications were ranked on a scale 0–100, whereby 51 is used as a threshold for moving to the second phase of evaluation. However, in the second phase, a company could be rejected even if it scored above 51. Moreover, to avoid the potential problem that companies who scored very high and those who scored very low are very different on unobservables, we retain for analysis all the companies who scored between 20 and 80: funded companies, who scored between 51 and 80, and rejected companies, who scored between 20 and 50. In the group of rejected applicants, we include those who scored 51–80 but who were ultimately rejected in the second round. Moreover, we cannot treat the 29 companies that were established in 2017 or later and these are dropped. Note that six companies are not used either because they appeared twice (once awarded, once rejected, for different projects) or because they went bankrupt.

Table 1 presents the construction of the treatment and comparison groups, as well the associated information.

Table 1.

Treatment and comparison groups for the two analyzed State Aid Programs.

	Financial Support of Investment Law State Aid	FITD State Aid
Type of method	Conditional difference-in-differences: Comparison group selected through propensity score matching (exact and near neighbor matching)
Treatment group	72	76
Treatment group	Supported companies (all)	Supported companies who scored 51–80 in the first evaluation stage (except 16 who were established in or past 2017)
Control group	60	66
Control group	Rejected applicants who failed on some of the additional criteria (but excludes applicants who were awarded but where subsequent disbursement did not occur)	Rejected applicants who scored 20–50 in the first evaluation stage and those who scored 51–80 but were rejected in the second evaluation stage (except 14 who were established in or past 2017)
Intuitive description of the method	Supported firms are matched with non-supported firms in identical sector (three-digit NACE Rev.2), size class (categorical ordering: Micro, small, medium, large), location (distance from the capital), age (years in operation). An estimate of the impact is obtained by difference in differences.
Differences between treated and non-treated firms controlled for by this method	Sector-specific economic trends; size effects (large and small firms may face different types of market failures); Geographic areas (proxy for possible socio–economic -institutional-transportation and labor cost-differences that may affect outcomes); Aging effects (accumulation of know-how, past retained earnings etc.) Unobserved characteristics that may lead to the decision to apply for the subsidy; Remaining unobserved differences between treated and non-treated firms, as long as they do affect the outcomes in a constant-over-time manner

Source: Drafted by the author.

The key assumption of our design is that in the absence of treatment, the average change in the outcome of interest for the treated group would have equaled the average change in the outcome for the untreated group (Blundell and Costa Dias 2000). This so-called parallelism assumption is counterfactual because the trend of the outcome in the treated group exists and is only observable in the presence of treatment. The parallelism assumption is thus ultimately unprovable (Bärnighausen et al. 2017). However, our comparison groups have the advantage of coming quite close to the parallelism assumption, in that before the intervention these firms likewise commenced or committed to an investment, which makes it easier to disentangle the effect of the investment from the effect of the state aid. This makes program participants as systematically similar to the rejected applicants as possible, hence reducing the selection bias.

Yet, our treated and comparison groups may still be different. One approach to completely equalize them is through propensity score matching (PSM). With PSM, firms from the comparison group are matched with the firms from the treated group on a propensity score. The propensity score is estimated using a probit regression model, in which treatment status is regressed on pre-intervention characteristics, which are observable: e_i = Pr(Z_i = 1|X_i). The estimated propensity score is the predicted probability of treatment derived from the fitted regression model (Austin 2011). PSM entails forming matched sets of treated and untreated subjects who share a similar value of the propensity score (Rosenbaum and Rubin 1985). PSM allows one to estimate the ATT (Imbens 2004).⁴

For this study, we utilize the following observable characteristics used in the probit regression: the firm’s age, location as represented through the distance from the capital, the firm’s size, and industry (three-digit NACE Rev.2 code). We assume groups are equal on unobservables, most notably on their desire to invest. If this is not the case, then results may be biased upwards and this is important to be borne in mind when interpreting the results. Since comparison applicants are fewer than the treated applicants, the option to use them more than once under PSM is applied.⁵ Hence, the one-to-one matching with replacement is conducted in two variants: first, a firm’s nearest neighbor, and second, within a caliper of 0.5 standard deviations of the firm’s estimated propensity scores (Rosenbaum and Rubin 1985). Nearest neighbor matching selects for matching the untreated subject whose propensity score is closest to that of the treated subject. Nearest neighbor matching within a caliper distance imposes the further restriction that the absolute difference in the propensity scores of matched subjects must be below some pre-specified threshold—the caliper distance. Austin (2011) discusses the methods for selecting untreated subjects in more great detail.

We should note that the parallelism assumption will be violated if individuals select into treatment and control groups based on factors that affect the change in the outcome of interest between the pretreatment and the post-treatment period (Bärnighausen et al. 2017). For instance, if eligibility for treatment were based on the values of the outcome in the pretreatment period, people might alter their behavior before a treatment starts to be eligible for evaluation. However, this is unlikely for both of the programs we evaluate: in the case of FSIL State Aid, 2018 was the first year when the program was deployed, for investment pursued in 2017. Hence, companies in 2017 or before did not know that such a state aid program was forthcoming, to make an adjustment of the outcome(s) in this period. For the FITD State Aid, selection into treatment and control is not based on factors that could affect the outcome(s) of interest, which are inherent to the firms themselves, but rather is based on a two-stage evaluation of the proposed projects for funding.

After matching, we estimate the difference of the differences between the treated and the comparison groups and between the 2 years, through the use of the psmatch2 command in Stata. The method is called difference-in-differences or DID, while the resultant estimate is called the average treatment effect of the treated or ATT. The method is common in quasi-experimental designs, that is, when randomization is not possible. The approach removes biases in post-treatment comparisons between the treatment and comparison group that could be the result of permanent differences between those groups, as well as removing biases from comparisons over time in the treatment group that could be the result of trends due to other causes of the outcome.

Data

As the FSIL State Aid was disbursed on December 26 to 28, 2018, we consider 2 years: 2018 and 2019. Using 2018 as the year pre-state aid is suitable because it is the year after the investment commenced in 2017, as well the year before the aid was awarded. By so doing, we avoid the problem of identifying the effects of the investment rather than of the aid, since all applicants (both approved and rejected) would have seen the effects of their 2017 investment in 2018.

As the FITD State Aid started being distributed in the second half of 2018 (in quarterly installments), we consider 2 years: 2017 and 2019. 2017 is used as the pre-state aid year, just to assure that any effects of the distributed grants in mid-2018 started demonstrating their effects by the end of the year. Using 2019 as the post-state aid year for both programs is a natural choice, as presently this is the last available year to evaluate.

Data were sourced from various institutions. Lists of the FSIL applicants were obtained from the Agency for Foreign Investment and Export Promotion, under the right to obtain information from government institutions for the public interest. A list of the FITD applicants was obtained from the Fund for Innovation and Technological Development under a Non-Disclosure Agreement. The full names and registration numbers were used to seek financial data from the Central Registry of North Macedonia, namely the Balance Sheet, the Income Statement, and the Statement of Changes in Equity for each company.

The appendix provides further details on the variables’ definitions and descriptive statistics.

Results and Discussion

Causal Effects of Financial Support of Investment Law State Aid

Table 2 presents the results of the tests measuring the significance of the differences in observables between the treatment and comparison groups. We work with four variables: the age of the firm and its distance from the capital, both of which are continuous; size of the firm, which is an ordinal variable; and finally the sector of operation, which is a categorical variable. Toward the bottom of the table, the Hotelling test is performed to assess the equality of the vector of means of all variables. Two of the four individual tests as well as the Hotelling test reject the null hypothesis that the means between the treatment and control group are equal, providing sufficient grounds to treat the two samples as different on observables.

Table 2.

Test of the randomization across treatment and comparison groups.

Ho: Means are Equal for the Treatment and Control Groups
Two-sample t test with unequal variances (p-value)
Age (years)	.0318
Distance from skopje (km)	.6170
Two-sample Wilcoxon rank-sum (Mann-Whitney) test (exact p-value)
Size (1 micro, 2 small, 3 medium-sized, 4 large)	.4704
Kruskal Wallis equality-of-populations rank test (p-value, with ties)
Sector (3-digit NACE Rev.2 classification	.0019
Hotelling T-squared test (p-value)
Vector of means of all variables	.0002

Source: Author’s calculations.

With these variables, we run a probit model, whereby the treated are used as the dependent variable. The coefficients are then used to predict a score which represents the probability that according to some demographic characteristics a company is selected to be awarded a FSIL State Aid. Figure 3 presents the distribution of these scores and depicts a clear difference between the treatment and the comparison group. Specifically, the comparison group is distributed along entire propensity score in a two-hump shape, while the treated group has a clear peak at a high propensity score with skewness to the left.

Figure 3.

Distribution of the propensity score—Financial support of investment law state aid. Source: Author’s calculations.

The propensity score is used for matching: when applied, both groups become highly comparable, as observed in Figure 4. In our case, the propensity scores almost overlap, and particularly when comparators are chosen based on the caliper procedure. At the bottom of the figure, the balancing test likewise fails to reject the null that the means between the treated and the control group after matching are identical.

Figure 4.

Assessing matching quality—Financial support of investment law state aid. Source: Author’s calculations.

In this figure, we present the main results. They are also presented in Table 3, but only the effect of the state aid on the range of outcomes is given for clarity. The first row presents the effect estimated on an unmatched sample for comparison purposes (otherwise invalid for the reasons explained above). The next rows focus on the average treatment effects on the treated (ATTs), based on the two estimators (nearest neighbor/caliper) and with the optional usage of the common support option, which all together leads to four variants.

Table 3.

Effects of the Financial Support of Investment Law State Aid—average treatment effect of the treated.

Difference in differences (decimal points)
		Revenue	Tangible assets	Intangible assets	Payroll cost	Employment	Labor productivity	Gross profit	Net profit	Current ratio	Cash ratio	Number of observations control/treatment
		(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)	(11)
Unmatched		−0.59*	−0.13*	0.15	−0.46	0.03	−0.77	−0.79	−1.40*	−0.12	−12.85*	60/72
ATT	Nearest neighbor	0.09	−0.00	0.12	0.09	0.13***	−0.04	−0.50	−1.29	0.06	−4.95	60/72
Caliper		0.05	0.01	−0.04	0.02	0.09**	−0.03	−0.95	−2.01	0.08	−6.72	60/54
ATT (common support)	Nearest neighbor	0.09	−0.00	0.12	0.09	0.13***	−0.04	−0.50	−1.29	0.06	−4.95	60/72
Caliper		0.05	0.01	−0.04	0.09	0.09**	−0.03	−0.95	−2.01	0.08	−6.72	60/54

aSource: Author’s calculations. *, ** and *** denote statistical significance at the 10%, 5% and 1%-age level, respectively. ATT = average treatment effect of the treated.

It is the general impression that companies which obtained the FSIL State Aid did not experience results different from the comparison group that could be ascribed to the receipt of a subsidy. All estimated effects except one are statistically insignificant at common levels, providing grounds to claim that the FSIL State Aid has been partially ineffective. The result validates the absence of the incentive effects for this program, agreeing with our pre-analysis predictions.

The only consistently positive result is that the FSIL State Aid likely improved job generation capacity in the recipient companies, partially supporting the notion that it supported operations and spurred an influx of working capital. These companies achieved better results in employment growth than their matching counterparts in the comparison group, though not when it came to wages. Specifically, employment in this group between 2018 and 2019 grew by an average of 6%, while in their matching non-recipient counterparts employment declined by 3% to 7%. A criterion for making an applicant eligible for the program is to have the number of employees in the year of application at least at the level of the previous year’s number. Since investment is typically related to employment growth, and the scheme excludes firms implementing labor saving investments, this result is hardly surprising.

These results may have proven to be insufficient, yet they demonstrate the policy’s benefits for at least three reasons. First, higher employment has been among the objectives of the Financial Support of Investment Law, and this finding validates that such a result has been attained. However, the absence of a differential in wages for the newly employed persons demonstrates that the objective of creating well-paid jobs due to the subsidy was not attained. Second, we expressed a concern that since FSIL State Aid is distributed only after an investment is made, and it may result in a significant deadweight loss for the society, as well as granting the recipient cash in a situation when the funding for the investment is likely to have been secured either way, and possibly by the applicant himself. These cash overflow may pump up a company’s profits, which would simply mean that the society pays for narrow private gains. However, the results do not lend support to this claim because there is no apparent difference in the cash on hand and profit growth with recipients more than of what has been observed in non-recipients.

In conclusion, when considered in its entirety, the FSIL State Aid resulted in improving economic efficiency only partially, yet it did not result in windfall gains. In the jargon of DG Competition (2013), the FSIL State Aid did not make recipients change their behavior, that is, undertake an activity that they would not have done without the aid or would do in a less desirable manner, and represents a clear example of the absence of an incentive effect.

Causal Effects of Fund for Innovation and Technological Development State Aid

Table 4 presents the results of the tests measuring the significance of the differences in observables between the treatment and comparison groups in the case of FITD State Aid. In addition to the four demographic variables used in the FSIL case, we consider the data points obtained for the project in the application phase. All tests reject the null hypothesis that the means between the treatment and comparison groups are equal, suggesting that the two samples are different on observables and cannot be compared directly.

Table 4.

Test of the randomization across treatment and comparison groups.

Ho: Means are equal for the treatment and control groups
Two-sample t test with unequal variances (p-value)
Age (years)	.0485
Distance from Skopje (km)	.0628
Grading of application (20/80 points)	.0001
Two-sample Wilcoxon rank-sum (Mann-Whitney) test (exact p-value)
Size (1 micro, 2 small, 3 medium-sized, 4 large)	.0113
Kruskal Wallis equality-of-populations rank test (p-value, with ties)
Sector (3-digit NACE Rev.2 classification	.0082
Hotelling T-squared test (p-value)
Vector of means of all variables	.0069

Source: Author’s calculations.

Figure 5 presents the distribution of the predicted score from the probit model, and depicts a clear difference between the treatment and the comparison groups. Namely, the comparison group is slightly tilted towards lower scores when compared to the treatment group, whereas the treatment group is apparently two-humped.

Figure 5.

Distribution of the propensity score—Fund for innovation and technological development state aid. Source: Author’s calculations.

When the matching algorithms are applied, both groups become highly comparable, as observed in Figure 6. At the bottom of the figure, the balancing test likewise fails to reject the null that the means between the treated and the control group after matching are identical.

Figure 6.

Assessing matching quality—Fund for innovation and technological development state aid. Source: Author’s calculations.

We present the main results in Table 5. We may already note that most of the differences in differences are significant, though only at the 10% level. Given the small sample and its large heterogeneity (even when treated versus comparators are observed), we do not consider this as significant grounds for concern.

Table 5.

Effects of the fund for innovation and technological development State Aid—average treatment effect of the treated.

	Difference in differences (decimal points)
		Revenue	Tangible assets	Intangible assets	Payroll cost	Employment	Labor productivity	Gross profit	Net profit	Current ratio	Cash ratio	Number of observations control/treatment
		(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)	(11)
Unmatched		1.49	0.41	0.63*	0.18	0.11	−0.06	−10.68	−6.61	−1.83	4.04	66/76
ATT	Nearest neighbor	1.74*	0.35	0.66*	0.10	0.10	0.22	12.21*	16.1**	2.11	7.86*	66/76
Caliper		2.85*	0.51	0.64*	0.19	0.20	0.43*	19.94	18.86*	−0.02	9.26*	66/63
ATT (common support)	Nearest neighbor	1.77*	0.35	0.67*	0.10	0.10	0.21	12.40*	16.4**	2.19	8.15**	66/75
Caliper		2.85*	0.51	0.64*	0.19	0.20	0.43*	19.94	18.86*	−0.02	9.26*	66/45

Source: Author’s calculations. *, ** and *** denote statistical significance at the 10%, 5% and 1%-age level, respectively.

Companies that obtained the FITD State Aid consistently achieved higher sales than their matched counterparts, and this effect is only due to the state aid they received. On average, between 2017 and 2019, their sales revenues grew between three and four times (depending on the matching procedure used), while the sales revenue of the matching comparators grew only between 40% and 50% (hence, 4–6 times less). Likewise, recipients have had consistently higher investment in intangible assets, which may be largely labeled as an investment in technology. Namely, their investment in intangible assets increased by about 50% between 2017 and 2019, while that of comparators declined by about 13%. Therefore, the FITD State Aid gave a clear advantage of the recipients to commence, sustain, or expand their investment in technology, which was the idea behind a grant program of this type. The finding is further important given that non-recipients declined their investment in intangible assets over the same period, which could either be an indication of shortage of funds to support such an investment or an abandonment of an investment project of technological level or character, which was presented in front of FITD and then rejected, indicating that had they been selected, the program may have, in fact, incentivized them to do something they otherwise would not have for whatever reason.

On the other hand, recipients of the FITD State Aid do not provide higher wages than their non-recipient comparators, nor did they invest in tangible assets (machinery and equipment) in a different manner than companies which ultimately did not receive the FITD State Aid. The latter may be an indirect indication that, on average, rejected applicants continued their investment despite not receiving the grant. Moreover, there is no evidence of higher employment growth among recipients of FITD aid. However, there is limited evidence that the labor productivity of the recipients increased due the FITD State Aid: the productivity of the treated companies increased between 50% and 80% between 2017 and 2019, while that of the matched comparators increased between 29% and 35%.

FITD State Aid increased the growth of profits and cash among recipients. The increase of the net profit is particularly astonishing: between 2017 and 2019 it grew between 17 and 19 times, while in non-recipient matching counterparts it only grew between 3% and 30%. In part, this profit surge may be driven by increased sales revenue and, to an extent, by labor productivity, and is thus affected by the lack of a differential increase in jobs and wages as a result of the FITD subsidy (increase of the capital share and no change in the labor share). Additionally, however, over the same period, FITD State Aid contributed to an astonishing increase in the cash ratio: for recipients it increased over 10 times, while for matched non-recipients it increased about 2.5 times. Therefore, both groups secured a source of cash, but the cashing of the latter has been apparently more cautious, probably because its source was of a more commercial (and hence more expensive) nature. However, when both large surges in the growth of profits and of cash in FITD State Aid recipients are observed in conjunction, it leaves space to claim that the subsidy, while spurring sales, productivity and investment in technology, also floods recipients with cash, part of which potentially supports higher profits. This conclusion should be given particular attention, including from the viewpoint of the cost effectiveness as presented in Figure 1: a euro of investment and of sales is paid for by more than a euro of FITD State Aid. Nevertheless, with this analysis we cannot observe if all or a part of this cash was obtained at year-end and was committed to pursuing further investment in the next year.

In conclusion, considered in its entirety, FITD State Aid resulted in improving economic efficiency. This corroborates the presence of the incentive effect. However, FITD State Aid was likewise accompanied by cash overflow, potentially suggesting that it may be over-cashing recipients, hence pumping up their profits, at government expense.

Conclusion and Policy Inference

The objective of this study is to measure the effects of state aid distributed to private enterprises in North Macedonia on the enterprises’ efficiency. We examine the governmental program for subsidizing companies under the Plan for Economic Growth, which is composed of state aid disbursed under the Financial Support of Investment Law (FSIL) of 2018 (which make up the first and second pillar of the Plan) and the one disbursed through the Fund for Innovation and Technological Development of North Macedonia (which makes up the third pillar of the Plan). Hence, we analyze two programs—labeled FSIL State Aid and FITD State Aid – implemented during or at the end of 2018, so that we are able to compare the years 2017 or 2018 (pre-support) and 2019 (post-support). We rely on a rigorous impact evaluation method. First, we create comparison groups from the pool of rejected applicants for the two programs because they best mimic the treatment group on unobservables: in both cases they had the same investment motive as they have also completed an investment (the FSIL case) or intended to make an investment (the FITD case). Moreover, in the FITD case, we use selection points to arrive at a more homogenous group of comparators. We pursue conditional matching on firms’ age, industry at three-digit NACE Rev.2, location (distance from the capital) and size, by using the procedure of nearest neighbor or a caliper (searching for matches within a range of standard deviations). After the matching, we pursue a difference-in-differences calculation to isolate the effect of the state aid on a range of efficiency indicators: growth in sales, investment, profit, employment, wages, productivity, and cash.

The FSIL State Aid showed partially ineffective. Our findings suggest that the state aid did not result in differences in sales, investment, wages or profits among recipients, compared to what has been observed among comparator non-recipients. These results demonstrate the absence of the incentive effect. However, FSIL State Aid did contribute to generating more jobs, as recipients steadily achieved a 9–13 percentage points higher employment growth than their matching counterparts over the same period. Yet, the absence of wage differentials in the jobs created due to the subsidy refutes the idea that the government accomplished the full objective of the program, which was to generate well-paid jobs. The cost for a gross job generated was nearly 28 thousand EUR. This suggests that, largely FSIL State Aid has been consumed in the generation of new jobs, rather than to make extra cash or profits, which is yet a positive sign given the partial ineffectiveness of the subsidy. No cash shower effects were discovered due to FSIL State Aid.

The FITD State aid showed considerably effective. Our findings suggest that due to the subsidy, recipients were able to increase their sales revenue and labor productivity. These results validate the presence of the incentive effect. The cost of an additional euro of sales generated, in gross terms, was slightly over a euro. However, no more jobs or higher wages were created in recipients than in what has been observed among matching comparator non-recipients. Likewise, the FITD aid has been found responsible for an increase in investment in intangible technology, which was particularly important finding given: i) the definitional role of FITD aid to spur technological growth and innovation; and ii) the declining investment in intangible assets among non-recipients over the same period. Overall, these positive developments induced by the FITD State Aid brought about a large increase in profits, far exceeding the profit growth in matching non-recipients. However, at the same time, FITD recipients were found to be experiencing cash overflow, which may suggest that excess profits were not exclusively driven by the increasing revenue, investment, and productivity, but also by the extra generated cash in the company due to the grant. As a cautionary point, though, we cannot exclude that cash accumulation from state aid disbursements represents a commitment for pursuing investment later because this investment would have occurred the subsequent year, which we did not observe in our data.

Important policy implications stem out of the findings of the effects of the two types of state aid. The finding that the FSIL State aid is partially ineffective is an important policy concern. It may signify that the subsidy needs to be modified to produce results, preserving the positive effect on employment generation. However, the lack of increase in sales, investment, wages, and profit effects suggests the need for an overhaul of the Financial Support of Investment Law. One possible method of reform would be to modify the disbursement of funds from ex-post to ex-ante, subject to project evaluation. On the other hand, given the effectiveness of FITD State Aid for sales, investment, and productivity, it may be beneficial for it continue to a large extent in its present form. One key area of possible intervention could be the need to strengthen rules related to assessment of the proposed costs in the project application budget, to avert any possibility of artificial inflation of such costs.

The study suffered some limitations. The key one is that it identifies the short-run effects of the state aid, since we only observed the year after the state aid was disbursed. It is hence wise that the longer-run effects are examined after the passage of sufficient time, yet without neglecting the early signs of failures documented herein. Other issues requiring deeper consideration include: i) companies who were rejected in 1 year, but who obtained funding in a subsequent year, which may be particularly important in the FSIL case whereby the application is determined by the investment already made; ii) the issue of related entities, particularly if both applied for the same or different state aid type and then obtained varied outcomes on their applications, may be examined from the viewpoint of the effect of the state aid obtained by one entity on the overall group of affiliated entities; and iii) the importance of the size of the subsidy should gain more prominence in further research, particularly given we suspected limited windfall effects in the FITD case.

Footnotes

Acknowledgments

I am thankful to Finance Think – Economic Research & Policy Institute for the opportunity to conduct this evaluation. The evaluation was conducted within a larger project Finance Think conducted under the auspices of the Balkan Trust for Democracy and the Norwegian Embassy in Belgrade. I particularly thank Desanka Dimitrova for helping me out with summarizing the literature and data management. I thank Blagica Petreski and Bojan Srbinoski for providing me with constructive critique on earlier drafts of this study. Any remaining errors are solely the author’s.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Marjan Petreski

Notes

Appendix

References

Austin

P. C.

(2011). An Introduction to Propensity Score Methods for Reducing the Effects of Confounding in Observational Studies. Multivariate Behavioral Research 46 (3): 399-424.

Auerbach

A. J.

Summers

L. H.

(1979) The Investment Tax Credit: An Evaluation. NBER Working Paper No. 404.

Bärnighausen

Oldenburg

Tugwell

Bommer

Ebert

Cara

Barreto

, et al. (2017). Quasi-Experimental Study Designs Series - Paper 7: Assessing the Assumptions. Journal of Clinical Epidemiology 89: 53-66.

Bergström

(2000). Capital Subsidies and The Performance of Firms. Small Business Economics 14: 183-193.

Blundell

Costa Dias

(2000). Evaluation Methods for Non-Experimental Data. Fiscal Studies 21 (4): 427-468.

Bondonio

Martini

(2012). Counterfactual Impact Evaluation of Cohesion Policy: Impact and Cost-Effectiveness of Investment Subsidies in Italy. Final Report to DG Regional and Urban Policy (EC).

Buelens

Garnier

Johnson

Meiklejohn

(2007). The Economic Analysis of State Aid: Some Open Questions. Brussels, Belgium: European Economy Economic Papers. 286/2007, 286, European Commission: Directorate General Economic and Financial Affairs (DG ECFIN).

Cassidy

Strobl

(2004). Subsidizing Industry: An Empirical Analysis of Irish Manufacturing. Journal of Industry, Competition and Trade 4 (2): 95-107.

Criscuolo

Martin

Overman

H. G.

Van Reenen

(2019). Some Causal Effects of an Industrial Policy. American Economic Review 109 (1): 48-85.

10.

Czarnitzki

Ebersberger

Fier

(2007). The Relationship Between R&D Collaboration, Subsidies and R&D Performance: Empirical Evidence from Finland and Germany. Journal of Applied Econometrics 22 (7): 1347-1366.

11.

De Blasio

Fantino

Pellegrini

(2015). Evaluating the Impact of Innovation Incentives: Evidence from an Unexpected Shortage of Funds. Industrial and Corporate Change 24 (6): 1285-1314.

12.

DG Competition (2013). Evaluation in the Field of State Aid - Concepts and Recommendations [Online]. Accessed January 15, 2021. https://ec.europa.eu/competition/consultations/2013_state_aid_modernisation/draft_guidance_paper_en.pdf ).

13.

Einiö

Overman

H. G.

(2020). The Effects of Supporting Local Business: Evidence from the UK. Regional Science and Urban Economics 83: 103500.

14.

Imbens

G. W.

(2004). Nonparametric Estimation of Average Treatment Effects Under Exogeneity: A Review. Review of Economics and Statistics 86: 4-29.

15.

Goolsbee

(1998). Investment Tax Incentives, Prices, and the Supply of Capital Goods. The Quarterly Journal of Economics, 113 (1): 121-148.

16.

González

Jaumandreu

Pazó

(2005). Barriers to Innovation and Subsidy Effectiveness. RAND Journal of Economics 36 (4): 930-950.

17.

Görg

Strobl

(2007). The Effect of R&D Subsidies on Private R&D. Economica 74 (294): 215-234.

18.

House

C. L.

Shapiro

M. D.

(2008). Temporary Investment Tax Incentives: Theory with Evidence from Bonus Depreciation. American Economic Review 98 (3): 737-768.

19.

Kuhn

(2010). An Analysis of Firm Growth Effects of the Danish Innovation Consortium SchemeInnovation: Analysis and Evaluation. Copenhagen, Denmark. Ministry of Higher Education and Science. Technical Report.

20.

Martin

Strasse

(2005). Community Policy of State Aid is it a Competition Policy? Concurrences, 3, 52-59.

21.

Nicolaides

(2009). Incentive Effect of State Aid: Its Meaning, Measurement, Pitfalls and Application. World Competition 32 (4): 579-591.

22.

Rosenbaum

P.R

(2002). Observational Studies (2nd ed.). New York, NY: Springer-Verlag.

23.

Rosenbaum

P. R.

Rubin

D. B.

(1983). The Central Role of the Propensity Score in Observational Studies for Causal Effects. Biometrika 70: 41-55.

24.

Rosenbaum

P. R.

Rubin

D. B.

(1985). Constructing a Control Group Using Multivariate Matched Sampling Methods that Incorporate the Propensity Score. The American Statistician 39: 33-38.

25.

Srhoj

Lapinski

Walde

J. F.

(2019). Size Matters? Impact Evaluation of Business Development Grants on SME Performance. Working Papers in Economics and Statistics, No. 2019-14.

26.

Trajkovska

Petreski

(2018) Bridging Gaps and Seizing Potential: Domestic SMES in the Supply Chain of Multinational Companies in the Country. Finance Think Policy Studies 2018-11/19. Skopje, North Macedonia: Finance Think - Economic Research and Policy Institute.

27.

Van Cayseele

Konings

Sergant

(2014). The Effects of State Aid on Total Factor Productivity growth. NBB Working.Paper No. 264.

28.

Wallsten

S. J.

(2000). The Effects of Government-Industry R&D Programs on Private R&D: The Case of the Small Business Innovation Research Program. The RAND Journal of Economics 31 (1): 82-100.