The importance of specialized intermediaries for gaining access to R&D tax credits: who uses intermediaries and what are the performance effects?

The importance of specialized intermediaries for

gaining access to R&D tax credits: who uses

intermediaries and what are the performance effects?

Master thesis

University of Groningen

Faculty of Economics and Business

By Koen Stamou

S2209383

Strategic Innovation Management

10/02/2020

Word count: 9856

Supervisor: dr. P. (Pere) Arque-Castells

Co-assessor: prof. dr. J. (Jordi) Surroca

2

Abstract

In the Netherlands, 85% of all R&D tax credit applications are filed by intermediaries. Claiming tax credits is costly and complex. Intermediaries play a crucial role in allowing firms access to government support programs. Yet, the literature often assumes that support programs are universally accessible. In this paper, I study the importance of intermediaries in allowing firms access to R&D tax credit programmes. For this research, application data has been collected at the largest Dutch application intermediary from the period between 2010 and 2018. In a first stage, I study the characteristics of firms that hire intermediaries for claiming R&D tax credits. This is important, because this allows us to understand what type of firm has difficulties in leveraging government incentives. By means of a probit regression I find that firm size and technological complexity negatively and significantly relate to the probability of intermediary selection. During the second stage of analysis, propensity matching is applied to identify twin firms from the control sample, and create a highly similar sub-sample to observe treatment effects from the intermediary studied in this paper. After the matching, difference-in-differences regressions are run, which provide no evidence for significant treatment effects during the period of analysis. However, the treated firms enjoy higher profits and produce more patents, which may have accrued before the period of analysis. This research adds fundamental insights on determinant firm characteristics for the probability of intermediary selection to the literature, and provides grounds for future research on the topic of the performance effects of using intermediaries to claim R&D tax credits.

Key words: R&D tax credit, R&D subsidy, KIBS, intermediary, firm-level, performance, policy

3

1. Introduction

Applying for R&D tax credit is a costly and complex process. Government support programs are at the core of policies OECD counties offer to simulate private R&D investment. Extant research argues that private R&D investment is of substantial importance for an economy its competitiveness (Hall & Lerner, 2010). This is mainly due to the fact that private innovative effort generate positive external effects in terms of knowledge spill-overs (Nelson, 1959). A share of value created as a result of private R&D efforts spill over, which disincentivizes firms to engage in costly R&D activities. For this reason governments from OECD countries interfere in the knowledge market with a wide range of policy instruments to compensate for the partial inappropriability of R&D investments. Tax credits, aimed at reducing the marginal costs of R&D activities, are arguably the most important instruments governments use. (Arrow, 1962). Governments claim that these tax incentive programs are universal, and that any firm satisfying the predetermined qualification criteria should be able to apply (for instance; Rijksdienst voor Ondernemers, 2019b). Yet, a recently published policy evaluation research conducted by De Boer et al. (2019) pointed out that in the Netherlands 85% of all tax credit applications for R&D credits (referred to as the WBSO programme) are submitted through specialized application intermediaries. This means, that although the Dutch government claims the WBSO programme to be universally applicable, firms seem to have a hard time applying for tax credit by themselves.

The purpose of this paper is to study the role that intermediaries play at allowing firms benefit from R&D tax credits. This study will look at the characteristics of firms that use intermediaries and if using intermediaries has a positive effect on performance. The literature primarily discusses the effectiveness of R&D tax credits in terms of complementarities: do R&D incentives stimulate or substitute private R&D investment (for instance; Aerts & Schmidt, 2008; Choi & Lee, 2017; Clausen, 2009). Why firms hire intermediaries to claim R&D tax credits them has not been studied in detail, but is potentially important because shadow costs exist in applying for public support. Firms might not be able to access public funding without their existence. To fill this gap, this study will scrutinize data from the largest specialized application intermediary operating in the Netherlands (due to confidentiality issues, I cannot mention the name). By matching application data provided by this specialized application intermediary to official tax records, this study will investigate which firms claim tax credits through the tax credit application intermediary discussed in this paper. Following this, using econometric difference-in-differences models, treatment effects are tracked, and performance implications are identified.

The reason why this question is of importance, is that if policy makers are to improve the universality of the WBSO instrument, it has to be uncovered which companies choose not to apply by themselves. Aerts & Schmidt, (2008); Busom, (2000), Takalo & Toivanen (2018) and Corsuelo & Ros, (2007) already found that firms find application procedures time consuming and costly. However, to

4

verify that this is the reason for the widespread use of specialized application intermediaries, additional research is needed. Studying which firms seem to have a hard time applying for tax credit by themselves, and the performance implications hiring the intermediary studied in this paper has, adds relevant insights to the literature on why firms might select intermediaries. These insights potentially allow policymakers to make the application process more universally accessible. Furthermore, studying which types of firms choose to select intermediaries is of particular value to the company who provided the data, since it helps them better understand their customer orientation and captures the treatment effects their services imply in terms of firm performance. Lastly, managers of firms that potentially qualify for R&D tax credit support benefit from the insights posed by this paper, as besides elaborating on which firms generally hire intermediaries to apply for R&D tax credit, implications for profits and revenues of doing so are identified.

This research uses a dataset consisting of 11.366 observations, containing data on sales, profit margins, patent output, number of employees, firm age and industrial complexity, over a period of ten years. Within this dataset, 918 observations are clients of the intermediary discussed in this paper, and the rest are controls. The controls can be described as firms that either did not apply for R&D tax credits, did apply for R&D tax credits by themselves, or firms that applied through another application intermediary than the one studied in this research. Ideally, I would have used non-intermediary firms as control sample, which would allow more accurate analyses, but given the fact that data in this field is rare and hard to come by, this was not possible. The data has been obtained from one of the largest specialized application intermediaries in the Netherlands, and has been enriched with official tax records obtained from Bureau van Dijk’s Orbis data service. The intermediary that offered the data preferred to remain anonymous in this paper. Hence, I will refer to it as “the intermediary studied in this paper”.

The analyses of this research are conducted in two consecutive stages. A first studies which type of firms hire intermediaries, and the second studies whether or not selecting an intermediary has positive performance effects. The results of the first the stage of analysis indicate that both firm size and industrial complexity negatively and significantly relate to the probability of claiming tax credits through an intermediary. The results of the second stage indicate that no significant treatment effect is identified during the period of analysis in firm performances. However, overall, treated firms have significantly higher profit margins and more patents.

One of the main insights of this paper is that predominantly smaller and technologically less complex firms hire intermediaries. This would support the notion that applying for R&D tax credit involves overcoming high fixed costs. High fixed costs would be relatively difficult to overcome for smaller firms. Technologically less complex firms apply less often, for less generous support than their high technologically complex counterparts (Herrera & Nieto, 2008). Hence it makes more sense for

5

technologically complex firms to internalize the application process for R&D tax credits, and duplicate the fixed application costs for themselves as compared to their technologically less complex counterparts. Specialized R&D tax credits application intermediaries allow firms to convert the high fixed costs of understanding the R&D tax credit application process in detail, into lower variable costs. Based on the results of this study, predominantly smaller and technologically less complex firms seem to do so. Furthermore, the fact that no significant performance based treatment effects have been uncovered in the second stage of analysis, but treated firms have significantly higher profit margins and patents, may mean that a part of the true treatment effect has been incurred before the period of analysis.

This paper is structured as follows. The next section presents what studies have been done so far on the topic of tax credit application intermediaries and government support in private R&D investment. Section 3 builds the theoretical framework of this paper. Section 4 elaborates on which data and variables I use to test the hypotheses, and section 5 will discuss the empirical strategies I employ to do so. In section 6 the results are given, which I discuss in section 7. Section 8 will conclude this paper. Following this, figures and tables are listed.

2. Literature review

2.1 R&D subsidy and tax credits

Throughout the years, many theorists have identified market failures in the market for knowledge development, justifying government intervention. For instance; Nelson (1959) already argued that the lack of appropriability of value captured by the public, but created by private R&D investment (due to knowledge spill overs), creates disincentives for firms to commit to investing in R&D. In a way, R&D acts as a collective good generating positive external effects that cannot be fully internalized (Arrow, 1962). Another market failure that adds to the aforementioned disincentive, is financial constraints that are present in the market for R&D activities. Hall & Lerner (2010) argue that there is a significant funding gap that cannot be solemnly resolved by venture capitalists, which, in the absence of government support seem to be the only party willing to commit resources to young, small firms intending to engage in R&D activities.

To correct for these market failures governments in OECD countries offer a wide range of policies. The most prominent ones include R&D subsidies (direct policies) and tax credits (indirect policies). On average, nearly a third of all innovating firms have done so with help of government funds. Government funds for R&D activities are primarily made available to innovating firms, either through direct subsidies or tax credits, in order to decrease the marginal cost of R&D (Hall & Van Reenen, 2000). In doing so, policymakers aim to correct for the disincentivizing market failures and stimulate private R&D efforts to a level that is socially desirable. Or in other words; conducive towards sustainable growth and competitiveness (Czarnitzki & Hussinger, 2004). In practice, this implies that

6

policymakers aim to partially fund projects that would not have taken place without subsidy, technologically improve technological laggards and declining regions, and support “national champions” (Clausen, 2009). A recent study by Lucking et al. (2018) demonstrates the significance. The ideal mix of direct versus non-direct measures largely depends on country specific features, and subsequently varies a lot between countries. However, a clear pattern is observable as argued by the OECD (2014); generally speaking, indirect fiscal policies seem to be generally more compliant with international trade regulations and more easily applicable as they are less discriminatory. This, according to the OECD (2014) to a large extent explains why countries seem to rely more and more on indirect fiscal measures in their innovation supporting policy mix.

With respect to the extent to which public R&D funding efforts succeed in stimulating private R&D expenditures, the literature remains ambiguous. In an ideal world, government support would exclusively flow to projects that suffer from market failures. When no market failures are present, publically funded intervention programs are likely to substitute private R&D expenditures. This phenomenon is referred to as the “crowding-out” effect. Although full crowding out effects are generally rejected (Aerts & Schmidt, 2008), several researches underline their significance (Dimos & Pugh, 2016). David et al (2000) reviewed existing literature, and confirm the literal ambiguity regarding the significance of crowding out effects. According to David et al. (2000) and Aerts & Schmidt (2008), the vast heterogeneity in findings regarding the significance of crowding-out effects can be explained as a result of widely varying estimators used in measuring effects, and the broadness of countries scrutinized, which all have their own specific policy mix. Although evidence with respect to the existence of crowding-out effects remains multi-sided, more recent studies do seem to support the effectiveness of R&D subsidy schemes and tax credits in terms of increasing R&D activity. For instance; Bloom et al. (2019) reviewed recent literature on effective policy mechanisms, and found that tax credits are generally found to be most effective. They argue that, based on recent findings in the literature, if the tax price of R&D falls with 10%, R&D activities in the long run increase with at least 10%. Another case for promoting public subsidies and tax credits was made by Lucking et al. (2018), who recently found that knowledge spill-over effects are present and significant: the socially marginal return of R&D is estimated to be 60%, whereas private marginal returns are estimated to be 15%. They argue, that since publicly beneficial knowledge spill-overs are substantial, subsidies and tax credits are essential parts of policy design.

2.2 Innovation policy in the Netherlands: WBSO

In this section, I review the institutional setting and public policy design in the Netherlands. The Dutch government aims to stimulate private R&D activities in several ways. The policy mix the Dutch government consists of direct subsidies in the form of grants to businesses, and indirect subsidies

7

in the form of tax advantages. Within the policy mix, the indirect, tax-based policy; “WBSO”, which is short for “Wet Bevordering Speur en Onderzoek” or translated: Stimulating Research and Development Act (Rijksdienst voor Ondernemers, 2017), is by far the most substantial. In 2019 the RVO (Rijksdienst voor Ondernemend Nederland, part of the Dutch chamber of commerce) budgeted the WBSO policy to account for 1.2 billion euros, which will increase to 1.285 billion euros in 2020 (Rijksdienst voor Ondernemers, 2019a).

The WBSO functions as a tax allowance on wage tax for entrepreneurs (only for their R&D activities), R&D personnel, and researchers who conduct technological-scientific research, and subsidizes activities that involve the development of new products, processes, services and software (Rijksdienst voor Ondernemers, 2019a). The WBSO is the most significant policy instrument the Dutch government employs in order to simulate R&D activities carried out by the private sector. Appelt et al, (2016) argues that in Holland roughly 87% of the policy mix consist of tax allowances (indirect subsidy instrument) which is the highest percentage globally.

2.3 Intermediaries

Although tax credit incentives are treated as universal support programs, in practice they are not. The reason for this is that high shadow application costs exist. This is because companies have to develop the competencies to successfully apply and leverage support programs, creating substantial fixed costs. Takalo & Toivanen (2018) found that in Finland tax credits did substantially increase R&D investment and subsequent knowledge spill-overs, increasing welfare. But after the aforementioned application costs are taken into account, virtually all positive welfare effects had been rendered obsolete. Earlier on, Takalo et al. (2013) have already proven that significant self-rejection, which implies that an eligible firm chooses not to apply for tax credits, exists as a result of implied shadow application costs. Another study in Spain by Corchuelo & Ros (2009) indicates that while 46% of their respondents conduct R&D, only 27.9% applied for R&D support. This low application rate can well be explained by the high implicit fixed costs incurred on applicants.

The high shadow costs and complexity in applying for government support generate demand for specialized intermediaries, and in the Netherlands, this demand is substantial: in 2017, 85% of all applicants have used an intermediary to claim R&D tax credits (De Boer et al., 2019). These intermediaries have a competitive advantage because they incur the high fixed costs of understanding the application procedure in great detail. Then they provide this knowledge to all applicants who only have to pay for a fraction of the fixed costs instead of having to duplicate them for themselves. Intermediaries allow firms to share the application costs. De Boer et al (2019) found the following to be the main motivations for companies to hire specialized application intermediaries. The highest ranking motivations for Dutch firms to choose for an application intermediary are time constraints,

8

insufficient experience in applying, unawareness of existing subsidy opportunities and that the costs of using an intermediary is comparable to the expenses if firms were to apply themselves. Even though the cost argument is 4th_{, not selecting an intermediary would force firms to overcome the first three} barriers, accruing the costs of doing that on top of the cost of the application process itself (which is already considered to be more or less equal to the fees of intermediary services by roughly 35% of the respondents).

As already argued in the relevance part of the introduction, very few studies have been conducted on the subject of these specialized intermediaries and their activities. Yet, these could be classified as “Knowledge Intensive Business Services”; or KIBS abbreviated, under which R&D services, in the broadest sense of the word, can be categorized (Cho et al.,2016). Cho et al. (2016) argue that R&D Services can be divided into two categories; contract R&D, who can be hired to conduct the full R&D processes for clients; and R&D support services. R&D support services consist of a broad range of areas, including patent management, technology and market research, R&D manpower supply and training, R&D consulting and R&D product design. For a definition, I rely on Miles et al. (1995), who have defined KIBS-firms as: “private companies or organizations relying heavily on professional knowledge i.e. knowledge or expertise related to a specific (technical) discipline or functional domain, and supplying intermediate products and services that are knowledge-based”

Although intermediaries that exclusively focus arranging funding through government incentives have been poorly addressed in existing literature, they could be categorized as R&D Consultants, since their core business is consulting clients on how to optimally arrange funding for R&D projects, and consult clients on how to proceed in the application procedures for government support projects.

3. Theory and Hypotheses

3.1 Demand for intermediaries

So why do firms hire specialized tax credit application intermediaries to arrange their public funding? This type of information is extremely hard to come by, since virtually no research on the subject of application intermediaries has been published. Most literature focusses on the bigger picture into the effectiveness of policies in terms of firm innovative performances (e.g. Charnitzki et al., 2011), complementarities between public and private R&D spending (e.g. Gonzales et al., 2008; Appelt et al., 2016), but very little on policy performances on the micro level (Czarnitzki et al, 2011; Appelt, et al., 2016; Gonzalez & Pazo, 2008).

9

Very relevant and rare information on why companies seem to fail to apply for government support is given by Tighe (2019), the CEO of a London based, specialized tax consultancy firm. He argues that companies generally fail to apply for innovation incentives because they do not realize their work is applicable for incentives, businesses do not know how to make the tax claim, are completely unaware of the existence of incentives, and are afraid that the process of applying is too resource consuming to make it worthwhile (Tighe, 2019). Especially the last explanation of why companies seem to reject themselves seems to be substantial, as argued by Takalo & Toivanen (2018) and Takalo et al. (2013). Developing competences to manage application from external government funding involves large fixed costs, costs that especially younger and smaller firms generally consider hard to overcome. This is in line with findings from De Boer et al., (2019) who found that applying for less than 500 applicable tax credit wage hours on average costs 1000 euros when hiring an intermediary, whilst it would take them 125 hours to apply by themselves. With slightly larger applications, between 500 and 3000 applicable hours, intermediaries on average charge 6000 euros, while applying by themselves would cost 750 hours.

Generally speaking, younger and smaller firms have a higher failure rate, which makes external funding from private parties hard to come by (Hall B., 2008; Lee, et al., 2012). But firms do need funding, especially for resource-intensive development processes. According to Takalo & Toivanen (2018), applying for government programmes requires extensive knowledge, and the development or acquisition of this knowledge incurs large fixed costs. This is in line with the findings of Smallbone et al., (1993), who argue that the use of consultants is relevant in cases where the costs of development of certain competencies is disproportionate compared to the strategic benefits in-house production of the focal developments offer. Smaller and younger firms will generally want to focus on their core activities. Adding to this, Muller & Zenker (2001) find that the absence of (in) tangible resources often motivates firms to turn to KIBS usage- which also applies to specialized application intermediaries. All in all; developing the competences required for a successful application process involve high fixed costs as argued in the literature review. Since resources are generally relatively poorly endowed with young and small firms, hiring intermediaries seems to be a logical way to convert high fixed costs into lower variable costs.

Based on this line of reasoning, I propose the following hypothesis:

H1a: Firm size and age negatively relates to the likelihood of specialized application intermediary usage

After having elaborated on how we expect firm size and age to relate to the likelihood of intermediary selection, I distinguish one more aspect that is believed to have potential in explaining the aforementioned: Technological intensity. In the innovation literature, technological opportunity is often defined as: “the advancement in scientific and technological understanding about one’s industry”

10

(Klevorick, et al, 1995). From this definition it already becomes clear that technological opportunity is shared within an industry, but can vary greatly across industries (Cohen, 1995). These differences explain why R&D intensity, and its nature deviate between industries. Hence, one would expect that R&D based subsidy application behaviour, which is highly dependent on the nature and intensity of the focal R&D activities, would deviate as well between industries. Generally speaking, firms in highly technologically complex industries apply for more subsidy (Silva, et al., 2017). This was argued earlier already by Busom (2000) and Herrerra & Nieto (2008), who found that firms operating in technologically complex industries and clusters generally receive more R&D support.

Now, it seems pretty straight forward that when firms in technologically complex industries apply for more subsidy, the demand for intermediaries arranging these subsidies increases as well. But there is more to this. Already a significant time ago, Teece (1988) argued that, especially in environments of increasing technological complexity, and increasing multi-disciplinarily of research activities, the most established, self-containing firms cannot thrive by themselves. This statement has been further researched by for instance Veugelers (1997), who found that firms in highly technologically advanced and complex environments have higher performance if they do not solemnly focus on internal R&D, but incorporate inputs from external R&D activities as well. Adding to this, The technological complexity and fast pace at which technological developments occur in high-tech industries, entail that the traditional internally focussed learning paradigm, in a sense, moved towards a paradigm where the learning takes place in a broad network of external partners. Basically, the locus of innovation moved from within the confines of the firm, to the status quo where innovation has become increasingly dependent on external input (Powell, et al., 1996)

This is where intermediaries demonstrate potential to be of significant value to innovating firms. As already argued by Muller & Zenker (2001), KIBS-firms, including intermediaries, can provide the focal firm with knowledge that is required to retain competitiveness in technologically complex industries. This entails that the application intermediary scrutinized in this paper, can provide clients with valuable information on what constitutes project success, and eliminate some of the high amount of risk involved in R&D activities. Since firms in technologically advanced industries are associated with comparatively higher R&D expenditures, I would expect that the aforementioned risk elimination would be most beneficial in technologically advanced industries.

Hence, I hypothesize the following:

H1b: Technological complexity is positively related to the likelihood of specialized application intermediary selection

11

3.2 Impact of intermediary usage on performance

The main use of application intermediaries, is to provide the focal company of opportunities and knowledge outside the boundaries of the focal firm (Ebersberger & Herstad, 2011). This clearly applies to applying for government R&D Support since applying for this involves expert knowledge on the institutional environment and how to operate in it (i.e. knowledge on how to apply, and what to apply for). In this sense, consultants (working for specialized application intermediaries) will inform companies on opportunities in terms of better financial performances. They have extensive knowledge on how to apply, what to apply for, and how much subsidy or tax credits a firm could and should apply for: they understand the application procedure in great detail. Generally, intermediaries have higher success rates in application outcomes (De Boer, et al., 2019). The consultants working at the intermediary studied in this paper also advise clients about feasibility of their projects, and have extensive knowledge of the focal industry. Besides solemnly filing client’s WBSO applications, they also advise companies on best practices and how the focal firm can best achieve its endeavours.

Besides exclusively providing firms with advice on how to leverage financial R&D incentives, consultants provide firms with valuable knowledge on what practices constitute to project success, and fill structural holes between different knowledge networks (Bianchi et al., 2016). In this sense, Bianchi et al. (2016) argue that consultants employed by the intermediary, besides providing companies with key knowledge on how to resolve the issue they are primarily hired for, they are a key source in linking the focal firm to potential partners that offer complementary, synergetic resources- tangible or intangible.

Given the fact that external consultants advise companies on how to realize their innovative endeavours, connect the focal firm to potential partners, and that firms involving intermediaries are more likely to get support, and generally get more generous grants, I hypothesize the following;

H2: Specialized application intermediary usage is positively related to firm performance. The conceptual model depicting the research approach is listed under figure 1: conceptual model in the figures and tables section at the end of this paper.

12

4. Data

This section describes the data that has been collected to test the hypotheses and which variables are used to test the hypotheses. An overview of all descriptive statistics is listed under table 1 in the figures and tables section at the end of this paper.

4.1 Data collection

4.1.1 Treatment sample

For this research, a database from the largest specialized application intermediary in the Netherlands is used. This intermediary handles about 10% of all applications in the Netherlands, and has offices equally distributed throughout the Netherlands. I cannot disclose more about the intermediary who offered the data due to confidentiality issues. The database has been exported by software the consultants use to fill in and submit the applications. The original database consisted of 2975 identical firms, and roughly 22000 applications in the period between 2012 and 2018. This dataset contains details on how many R&D employees a firm employs, how many projects the focal client is conducting, how much tax credit is applied for, how much is granted, and firm identifiers. Using the identifiers, I enriched this database with data on profits, total assets, patents, trademarks, number of employees, date of incorporation, and industry, using Bureau van Dijk its Orbis database. The Orbis database collects its data directly from the Dutch chamber of commerce, and thus consists of official tax records. The data retrieved from Orbis corresponds to the period 2010 to 2018. From the initial database, 2104 firms could be matched to Orbis. From this sample, 14000 applications were filed to the RVO (Dutch Entrepreneurial Agency), with an average max of 3657 hours per application. In total, the firms in the dataset applied for 50.7 million applicable hours of R&D tax credit in the period 2012 to 2018. In order to test the hypotheses, the aforementioned data extracted from Orbis is required.. For this reason, all firms that have been matched to Orbis, but do not include data on revenue, assets or employees have been removed from the sample. The final sample of intermediary using firms, consists of 918 unique applications, filed for 227 unique client firms. The average hours of R&D work in these applications is 19268 hours. This is considerably higher than before the cleaning. The main reason for this is that smaller firms, who are likely to apply for lower amounts, are not obliged to publish tax records, explaining the absence of their data in Orbis. For this reason, predominantly smaller firms are expected to have been dropped from the sample. Although most observations still range between 0 and 10000 hours of R&D work in the application, some very high values push the average upwards: the final treatment sample accounts for 17.7 million R&D work hours.

4.1.2 Control sample

In order to capture the treatment effect of the specialized application intermediary studied in this research, a control sample is required. To ensure data validity, the control sample has been extracted

13

from Orbis as well. This ensures that all financial indicators are based on officially published tax records. In this process, I sampled all Dutch firms with complete data over the last five years on assets, sales, profit margins, patents, number of employees and prime industries. This resulted in a control sample of 1530 firms, with 10448 unique observations. Since data completeness was used as a selection criterion, no further cleaning is required for the control sample. One drawback of this completeness criterion is that also for the control sample, only larger firms are obliged to publish financial data. Hence, also for the control sample, the firms considered in this study are relatively large.

The full sample used in the econometric analysis consists of 1757 firms, 227 are treated, and 1530 are not. The full dataset contains 11.366 observations in the period 2010-2018.

4.2 Measurement

4.2.1 Independent variables

Firm size

To test the first part of hypothesis 1a, a firm size variable has been computed, based on the number of full time employees the focal company employs. On average, firms in the complete sample have 1971 employees. Intermediary using firms (the treatment sample) have an average of 444 employees, compared to 2105 employees in the control sample.

Firm age

To test the second part of hypothesis 1a, an age variable has been computed. From Orbis, the date of incorporation has been extracted for all firms studied in this research. Based on this, the age has been computed in Stata SE. On average, firms in the sample are aged 15.9 years. The youngest firms in the total sample have been founded in the period of analysis, whilst the oldest is 181 years old.

Technological complexity

To test hypothesis 1b, the company’s NACE codes are converted in a binary technological complexity dummy. The NACE code is a categorization imposed by the European Union. The code consists of four digits, identifying the core activities the focal firm conducts. Based on this four-digit code, a binary variable indicating whether or not the main sector the focal firm operates in, classifies as technologically complex will be computed. This has been done according to the European Union’s classification of high-tech industries1 _{(Eurostat, 2020). Within the sample, 6926 observations took place in high-tech}

1 _{Full classification document available at:}

14

industries, and 4440 observations did not. From the intermediary firms 53.5% operated in high-tech industries, whilst their non-intermediary counterparts operated in high-tech industries for 61.63%.

4.2.2 Dependent variables

Intermediary selection

In order to test the first hypotheses, a treatment dummy has been computed. It will have value 1 if the focal firm has used the specialized application intermediary, and value 0 if the focal firm either; did not apply for tax credit, did apply but not through the intermediary studied in this paper, or did apply through another intermediary. Ideally I would use firms that apply for tax credit without help of an intermediary as control pins, but due to data availability constraints, this data was not collectible. Hence, the analysis will focus on the treatment effect of the specific intermediary studied in this paper, and will use a collection of firms that did not use the specific intermediary studied in this paper as control sample.

Firm performance

In order to measure innovation performance, patent output between the treatment group and controls will be compared. One significant drawback of patent output for the analysis is that Orbis only contains data on the aggregate amount of patents filed by the focal firm. There is no time dimension in Orbis for patent output, which makes observing true treatment effects of the intermediary in terms of patent output impossible. Therefore, although patent output differences between the treated firms and controls will be discussed, this study will focus on capturing treatment effects in terms of operating revenue and profit margin.

5. Empirical strategy

5.1 Stage one: probability of using intermediary (H1)

As can be witnessed in figure 1: conceptual model (in the figures and tables section at the end of this paper) the analysis will be done in two consecutive stages. In a first stage, I study the probability that a firm uses intermediary services as a function of firm characteristics. I run a probit regression that helps characterise the typology of firms that use intermediaries. In this probit regression, the binary intermediary selection dummy (intermediary selection=1) is used as dependent variable, and the technological intensity dummy, age variable and the amount of employees (as a size indicator) will be used as independent variables, allowing me to test hypotheses 1 a and b.

Equation 1 (probit regression):

15

After testing H1, propensity matching will be applied. The reason for this, is that I assume there is a substantial sampling bias resulting from the data availability. Or better, the lack of data availability. The fact that smaller firms are not obliged to publish their revenues and profit numbers implies that both the treatment and control sample are expected to be biased upwards in terms of size, and all size-related variables. Without matching, a part of the observed treatment effect could very well be attributed to firm features predicting treatment, rather than the treatment itself. Propensity matching aims to reduce treatment assignment bias by creating a subsample of controls that match the treatment sample. This new subsample consists of firms that have a high probability of being part of the treatment sample based on observed predictors. In this case, size will be the most prominent predictor, since this is where the most substantial bias is expected resulting from the fact that smaller firms do not have to publish their tax records. In a way, propensity matching mimics randomization in cases where randomization is not possible (as is the case in this research because of the data availability). Propensity matching matches the treatment firms to a firm that has not applied for tax credit, has applied on itself, or has applied through another intermediary, but has a high probability of being treated by the intermediary discussed in this paper. This allows me to control for a large share of the aforementioned sampling bias based on observables. However, there could still be bias resulting from unobservables. Based on these propensity scores, treatment firms are paired to their closest neighbour from the control sample for the next stage of analysis

5.2 Stage two: effect of intermediaries on performance

In the next stage of analysis, a difference-in-differences (DiD) analysis will be conducted. In the dataset used for this study, we have collected data ranging back from 2018 until 2010. Additionally, we have data on when the focal firm applies for tax credit through the intermediary studied in this paper. Based on this, a timing dummy has been computed, indicating when the focal firm starts receiving treatment from the intermediary. Hence, we have pre- and post-treatment data in terms of sales and profit margins from our treatment sample, and the same data for the non-treated controls. As a result of the propensity matching, the treatment firms have been assigned to a highly similar non-treatment counterpart. The DiD analysis predicts what the outcomes of the focal firm (based on its matched partner) would have been without treatment (in this case in terms of sales and profit margin), and tracks the difference between this value, and the observed value in the observations after treatment. The difference captured between the expected and observed outcomes is the treatment effect attributable to intermediary selection. The difference-in differences model can be summarized by the following equation:

Equation 2 (difference-in-differences regression):

16

Because DiD analysis uses similar controls operating in the same setting as the focal firm, unobserved effects like economic downturn or seasonal effects are implicitly controlled for, making it very well suited to scrutinizing effects of policy intervention.

6. Results

6.1 Stage 1

As argued in the previous section, during the first stage, a probit regression is run, with the intermediary dummy as dependent variable, and the technological complexity dummy, the age variable and the number of employees (as size indicator) as independent variables. The results are listed in table 2 in the figures and tables section, under model 1.

As can be seen in table 2, size is negatively relates to the probability of selecting an intermediary at a significance level of 1% (p=0.00, z=-3.58). This implies that the firms who have selected the intermediary discussed in this paper are significantly smaller in size than firms in the control sample (that is firms who did apply through another intermediary, applied for tax credit by themselves or did not apply). The results provide evidence that as hypothesised under H1a, the intermediary firms are smaller

Contradicting the expectations of H1a, firms using intermediaries firms seem to be older, although this result is not statistically significantly (p=0.8, z=0.25). Based on this, no evidence has been found that intermediary using firms differ in age from the controls. Hence, also taking account the aforementioned, H1a is partially accepted: intermediary using firms are smaller than the controls, but do not significantly differ in terms of age.

Regarding H1b, the results observed also contradict the predictions established in the conceptual part of this paper. Intermediary using firms seem to be of a technologically less complex nature than the controls at a significance level of 5% (p=0.049, z=-1.49). This witnessed effect is the opposite of what I expect under H1b, hence H1b is rejected. In the discussion section I will reflect on this outcome and elaborate on what this result might imply.

6.2 Stage 2

As discussed in the empirical strategy section of this paper, stage two of the analysis will consist of difference-in differences regressions to observe the treatment effects of intermediary selection in terms of revenue and profit margin. To correct for potential sampling bias resulting from the data availability issues discussed in the data section, propensity matching is applied. Using the psmatch2 analysis in Stata a probit regression was run to test the first hypotheses. Based on this probit model, the

17

propensity scores from the covariates (in this case size, age and technological intensity) with respect to the probability of intermediary selection are kept, and used to identify twin firms from the control sample. These sets of twin firms are used to compare forecasted and observed outcomes in terms of revenue and profit margin to capture treatment effects, by running difference-in-differences regressions. The results from the matched sample difference-in-differences regressions are listed table 2 under model 2 and 3. I run the same difference-in-differences regressions with the full sample to illustrate the effect of the propensity matching on the outcomes of the analyses. These results are listed in table 2 under model 4 and 5.

The first difference-in-differences model is run with the natural logarithm of operating revenue as dependent variable. Here we take the natural logarithm of the operating revenue to normalize the distribution. The results indicate that treatment by the intermediary actually reduces revenue. As can be seen in table 2 in the appendix under the second column, treated firms perform less as expected, (β=-.232, p=0.08) in terms of revenue, but only marginally significantly so (at a significance level of 10%)

The second difference-in-differences regression is run with the natural logarithm of the profit margins as dependent variable. Interestingly, in terms of profit margin, the intermediary firms perform better than the controls in general at a significance threshold of 5% (β=.315) based on the mean values. However, no significant treatment effect has been observed for the profit margins (β=-.1). Although firms selecting the intermediary seem to be more efficient, this is not attributable to treatment within the time frame of the analysis (that is between 2010 and 2018). Hence, hypothesis 2 is rejected

7. Discussion

As argued in the introductory section of this paper, the main aim of this study is to provide insights into which firms hire specialized application intermediaries to claim R&D tax credits, and what the effect of doing so is on firm performances. The results indicate that relatively smaller firms, operating in less technologically complex industries hire intermediaries. No significant treatment implications resulting from selecting the intermediary in terms of firm performances have been identified.

With respect to the first hypothesis, I am able to confirm that firms that select the intermediary discussed in this paper are significantly smaller in size. This finding is in line with the findings by De Boer et al. (2019) and Takalo & Toivanen, (2018) who underline the high fixed costs of applying for R&D tax credits. For smaller firms, the fixed costs of applying are much more difficult to recuperate than is the case for larger firms. Larger firms are likely to have larger development projects, potentially generating more revenue. Especially for smaller firms, converting the high fixed costs into lower variable costs by using an intermediary would seem attractive.

18

Another interesting finding is that under hypothesis H1b, intermediary firms actually proved to be of a less technologically complex nature than their non-intermediary counterparts. A reason for this could be that firms operating in technologically advanced industries generally have more formalized R&D processes (f.i. Jansen et al., 2006; Bianchi et al., 2016). For firms depending on R&D (as is likely in technological complex industries), it makes more sense to develop the infrastructure needed to internalize application related activities. This is because firms operating in technologically more complex environments apply for more generous R&D tax credit, and apply more often (De Boer, et al., 2019; Herrera & Nieto, 2008; Busom, 2000), allowing those firms to incur the fixed costs of applying over more, and more generous applications. Based on the findings under hypothesis H1b, it would seem that technologically complex firms are better off incurring the high fixed costs than paying a lower variable rate every time the firm applies for tax credits.

With respect to the results to performance implications identified in the second stage of analysis, no significant treatment effects are identified in terms of performance. As can be seen in table 2, model 2 and 3 in the figure and tables section. Revenue seems to slightly deteriorate as a result of treatment (but only marginally significantly) as pointed out by the difference-in-differences regression. This is only the case for the matched sample. After running the difference-in-differences regression with the full control sample, the negative treatment effects turn insignificant (see model 4 in table 2 in the tables and figures section).

Although the revenue of intermediary using firms seems to have deteriorated following treatment, the intermediary firms seem to be more efficient. This can be seen in the 3rd _{and 5}th _{model in} table 2. Here it reads that treated firms have significantly higher profit margins than the controls do (this counts for both the matched and unmatched control samples). No significant treatment effects have been identified by the difference-in-differences regressions, but there are some grounds to suspect that part of the treatment effect is embedded in the overall difference (the treated row in table 2). One drawback of using difference-in-difference regressions is that it requires pre- and post-treatment data to identify treatments effects. Within my sample, 227 intermediary using firms are used in the analysis, but only 119 observations contained pre- and post-treatment data and were used to measure treatment effects. The remainder have been a client throughout the entire period of analysis, were not used to identify treatment effects, but are used to establish the differences in sample means. Since these firms continuously received treatment, the treatment effects might be accrued in the profit margin means, rather than identified as stand-alone treatment effects. Within the sample, the profit margins of intermediary firms are significantly higher than those of the controls at a significance level 5% in the matched setting, and at a level of 1% in the unmatched setting (models 2 and 5 in table 2)

In line with this reasoning is that within the matched setting, patent output is significantly and positively associated with the probability of intermediary selection. This was tested by means of a probit

19

regression with the intermediary dummy as dependent variable and patent output as independent variable, in the matched sample setting. The results are listed under model 6 of table 2. The fact that intermediary firms produce significantly more patents, have lower revenues, but enjoy higher profit margins than the controls might imply that intermediary firms have realised process innovation. However, given the fact that there is no time dimension available in patent output and that data on revenues and profit margins dating back before the period of analysis is scarcely available, I cannot verify this to be true.

8. Conclusion

8.1 Theoretical implications

This study provides insights into which firms hire intermediaries to apply for R&D tax credits, and if this has a positive effect on firm performance. The results indicate that predominantly smaller, and technologically less complex firms decide to apply for tax credit through an intermediary. No significant treatment effects have been identified in terms of firm performance. Although specialized application intermediaries have been studied in other institutional settings (e.g. Takalo et al., 2013; Muller & Zenker, 2003), there is virtually no literature focussing on the role of these in the Netherlands (the only study I could find was conducted by De Boer et al. 2019). Yet 85% of all tax credit (WBSO) applications are filed through an intermediary (De Boer et al., 2019). Although the findings regarding the determinants of selecting an intermediary are much in line with the findings of Takalo et al., (2013) in Finland, and De Boer et al., (2019) in the Netherlands, the second focus of this paper on the effect of intermediary selection on firm performance is entirely novel, and therefore of substantial theoretical value.

8.2 Managerial implications

Managers could derive value from this research as it provides insights into current practices in application behaviour of Dutch firms. It provides key insights in the question of which firms potentially benefit from hiring intermediaries, and what type of firm is likely to be better off internalizing application related activities. Although the results posit that selecting the intermediary studied in this paper might not affect firm performances directly, they do give managers some idea about the high fixed costs they will have to overcome if they are to apply on themselves.

For policymakers, the findings entail that the design of the current WBSO programme might not be as universally accessible as they claim it to be. Although the Dutch entrepreneurial counsel argues that firms are very well able to apply themselves (Rijksdienst voor Ondernemers, 2019b), 85% of all applications go through an intermediary (De Boer et al., 2019). Based on the findings of this study, large firms and firms of high technological complexity seem to opt out of selecting the intermediary studied in this paper, which could be explained by the high fixed costs firms have to incur in order to

20

develop the competencies to apply for tax credit. If policymakers in the Netherlands aim to design a truly universal programme, improvement is needed to make applying for tax credit less costly and complex.

8.3 Limitations and future research

This study is subject to several limitations. The most significant limitations are caused by the availability of data. I was unable to obtain official data from the Dutch entrepreneurial counsel (RVO), which meant that for the control sample, I have collected data on all firms that have fully published records in the period between 2010 and 2018. Ideally, the control sample exclusively consisted of non-intermediary applicants, but given the lack of data availability, I had to suffice for non-non-intermediary applicants, non-applicants and applicants that applied through a different intermediary than the one studied in this paper. Therefore, this study is confined to capturing the treatment effect and selection determinants that apply to the intermediary discussed in this paper, and might not necessarily reflect the treatment effect of intermediaries in general when compared to non-intermediary firms. Second, as only the largest firms in the Netherlands have to publish their records, the analyses is based on the largest firms in the Netherlands, and the largest clientele of the intermediary discussed in this paper. I applied propensity matching and difference-in-differences techniques to correct for potential bias and control for firm-fixed effects, but this is based on observables. This means that some of the captured effects might still be subject to bias based on unobservable firm-fixed effects that given the data availability, could not be controlled for.

Future research could benefit from studying the implications of intermediary selection in more detail, based on a more comprehensive dataset. Knowing which firms are intermediary using, and which are not allows future research to study the implications of applying through intermediaries with more accuracy and in greater detail. Another relevant topic identified by the outcome of this research is what exactly causes the high fixed costs of applying. This would particularly be interesting for policymakers, since it would give them some grounds for improvement if policymakers are to make applying for tax credits more universally accessible. Lastly, although this paper is unable to identify treatment effects within the timeframe of the analysis, there are grounds to believe that a part of the treatment effect took place before the period of analysis, potentially explaining the statistically significant higher profit margins of intermediary using firms. From the data I was able to collect, this cannot be verified. Perhaps future research, based on a more comprehensive dataset, could look into this and provide more conclusive evidence for the performance effects of hiring specialized R&D tax credit intermediaries.

21

References

Aerts, C., & Schmidt, T. (2008). Two for the price of one?: Additionality effects of R&D subsidies: A comparison between Flanders and Germany. Research Policy, 37(5), 806-822.

doi:https://doi.org/10.1016/j.respol.2008.01.011

Appelt, S., Matej, B., Criscuolo, C., & Galindo-Rueda, F. (2016, 9 19). R&D Tax Incentives: Evidence on design, incidence and impacts. OECD Science, Technology and Industry Papers, 32, 0-32. Arrow, J. (1962). Economic welfare and the allocation of resources for invention. In R. R. Nelson, The

rate and direction of incentive activity: economic and social factors (pp. 609-625). N.J.: Princeton.

Bianchi, M., Croce, A., Dell'Era, C., Di Benedetto, C. A., & Frattini, F. (2016). Organizing for R&D outsourcing. Journal of Product Innovation Management, 33, 492-510.

Bloom, N., Van Reenen, J., & Williams, H. (2019). A Toolkit of Policies to Promote. Journal of Economic Perspectives, 33(3), 163–184.

Busom, I. (2000). An empirical evaluation of the effects of R&D subsidies. Economics of innovation and new technology, 9(2), 111-148.

Cho, C., Park, S. Y., Son, J. K., & Kee, S. (2016). R&D support services for small and medium-sized enterprises: The different perspectives of clients and service providers, and the roles of intermediaries. Science and Public Policy, 43(6), 859-871.

Choi, J., & Lee, J. (2017). Repairing the R&D market failure: Public R&D subsidy and the composition of private R&D. Research Policy, 46(8), 1465-1478.

Clausen, T. H. (2009). Do subsidies have positive impacts on R&D and innovation activities at the firm level? Structural change and economic dynamics, 20(4), 239-253.

Cohen, W. M. (1995). Empirical studies of innovative activity.

Corchuelo, M. B., & Ros, E. M. (2007). Incentivos fiscales a la inversión en I+D y su aplicación en Espana. Madrid: Instituto de Estudios Fiscales.

Czarnitzki, D., & Hussinger, K. (2004, August). The link between R&D subsidies, R&D spending and technological performance. ZEW - Centre for European Economic Research , 1-26.

Czarnitzki, D., Hanel, P., & Miguel Rosa, J. (2011). Evaluating the impact of R&D tax credits on innovation: A microeconometric study on Canadian Firms. Research Policy, 40, 217-229. David, P. A., Hall, B. H., & Toole, A. A. (2000). Is Public R&D a Complement or Substitute for Private

R&D? A Review of the Econometric Evidence. Research Policy, 29(4-5), 497-529. De Belastingdienst. (2020, 01 1). NACE codes. Retrieved from Belastingdienst:

https://download.belastingdienst.nl/belastingdienst/docs/bedrijfsactiviteiten_nace_codes_ ob4111b4pl.pdf

De Boer, P. J., Faber, D., Gielen, M., Van Dorsser, S., Den Hartog, P., Janssen, M., . . . Verspagen, B. (2019). Evaluatie WBSO 2011-2017. Den Haag: Ministerie van EZK.

Den Hartog, P. (2000). Knowledge-intensive business services as co-producers of innovation. International Journal of Innovation Managemen, 4, 491-528.

22

Dimos, C., & Pugh, G. (2016). The Effectiveness of R&D Subsidies: a Meta-Regression Analysis of the Evaluation Literature. Research Policy, 45, 797-815.

Ebersberger, B., & Herstad, S. (2011). Product innovation and the complementarities of external interfaces. European Management Review, 8(3), 117-135.

Eurostat. (2020, 1 1). Annex 3 – High-tech aggregation by NACE Rev.2 . Retrieved from Eurostat indicators on High-tech industry and Knowledge – intensive services :

https://ec.europa.eu/eurostat/cache/metadata/Annexes/htec_esms_an3.pdf

Gonzalez, G., & Pazo, C. (2008). Do public subsidies stimulate private R&D spending? Research Policy, 37, 371-389.

Griliches, Z. (1986). Productivity, R&D, and Basic Research at the Firm Level in the 1970's. American Economic Review, 76(1), 141-155.

Hall, B. (2008). The financing of innovation. In: Shane S (ed) Blackwell handbook of technology and innovation management. Oxford: Blackwell Publishers Ltd.

Hall, B. H., & Lerner, J. (2010). Chapter 14 - The Financing of R&D and Innovation. Handbook of the Economics of Innovation, 1, 609-639.

Hall, B., & Van Reenen, J. (2000). How effective are fiscal incentives for R&D? A review of the evidence. Research Policy, 29(4), 449-469.

Herrera, L., & Nieto, M. (2008). The national innovation policy effect according to firm location. Technovation, 28(8), 540-550.

Jansen, J., Van den Bosch, F., & Volberda, H. (2005). Exploratory innovation, exploitative innovation, and performance: Effects of organizational antecedents and environmental moderators. Management Science,, 52, 1661-1674.

Klevorick, A. K., Levin, R. C., Nelson, R. R., & Winter, S. G. (1995). On the sources and significance of interindustry differences in technological opportunities. Research policy, 24(2), 185-205. Laursen, K., & Salter, A. (2006). Open for innovation: the role of openness in explaining innovation

performance among UK manufacturing firms. Strategic Management Journal, 27(2), 131-150.

Lee, S., Hwang, T., & Choi, D. (2012). Open innovation in the public sector of leading countries. . Managment Decision, 147–162.

Mansury, M. A., & Love, J. H. (2008). Innovation, productivity and growth in US Business services: a firm level analysis. Technovation, 28(1-2), 52-62.

Muller, E., & Zenker, A. (2001). Business services as actors of knowledge transformation: The role of KIBS in regional and national innovation systems’. Research Policy, 30, 1501-16.

Nelson, R. (1959). The Simple Economics of Basic Scientific Research. Journal of Political Economy, 67(3), 297-306.

Powell, W., Koput, K., & Smith-Doerr, L. (1996). Interorganizational collaboration and the locus of innovation: networks of learning in biotechnology. Administrative Science Quarterly, 41, 116-145.

23

Rijksdienst voor Ondernemers. (2017). WBSO handleiding 2018. Zwolle: Rijksoverheid. Rijksdienst voor Ondernemers. (2019a). WBSO in 2020: flexibeler met stabiele voorwaarden.

Retrieved from Rijksdienst voor ondernemend Nederland:

https://www.rvo.nl/actueel/nieuws/wbso-2020-flexibeler-met-stabiele-voorwaarden Rijksdienst voor Ondernemers. (2019b). Wel of geen intermediar? Retrieved from RVO..nl:

https://www.rvo.nl/subsidie-en-financieringswijzer/wbso/aanvraagproces/wel-geen-intermediair

Shearmur, R., Doloreux, D., & Laperriere, A. (2015). Is the degree of internationalization associated with the use of knowledge intensive services or with innovation? International Business Review, 24, 457–65.

Silva, A. M., Silva, S. T., & Carneiro, A. (2017). Determinants of grant decisions in R&D subsidy programmes: Evidence from firms and S&T organisations in Portugal. Science and Public Policy, 44(5), 683-697.

Smallbone, D., North, D., & Leigh, R. (1993). The use of external assistance by mature SMEs in the UK: some policy implications. Entrepreneurship & Regional Development, 5, 279-295. Takalo, T., & Toivanen, O. (2018). Economics of Finnish Innovation Policy. VATT Institute for

Economic Research.

Takalo, T., Tanya, T., & Toivanen, O. (2013). Estimating the benefits of targeted R&D subsidies. The Review of Economics and Statistics, 95(1), 255-272.

Teece, D. J. (1988). Technological change and the nature of the firm. In G. Dosi, C. Freeman, R. Nelson, G. Silverberg, & L. Soete, Technical Change and Economic Theory. London: Pinter Publisher.

Tether, B., & Tajar, A. (2008). Beyond industry–university links: Sourcing knowledge for innovation from consultants, private research organisations and the public science-base,. Research Policy, 6(7), 1079-1095.

Tighe, M. (2019, 5 7). Why small businesses are missing out on millions in R&D tax credit relief. Retrieved from Small Business.co.uk: https://smallbusiness.co.uk/rd-tax-credit-relief-2547490/

Veugelers, R. (1997). Internal R&D expenditures and external technology sourcing. Research Policy, 5, 303-315.

24

Figures and Tables

25

Table 1: Descriptive Statistics

Variable N Mean Std.Dev. Min Max

All firms

Revenue 11366 635000 4110000 1.897 1.12e+08

Total assets 10240 674000 4820000 0 1.15e+08

Employees 11366 1971.718 19813.09 1 710000

Profit margin 11282 5.047 13.662 -99.266 100

Patents 11366 3.321 68.558 0 2527

Trademarks 11366 1.297 6.267 0 140

Age 11366 15.854 17.465 0 184

Intermediary using firms

Revenue 918 354000 1680000 18 2.43e+07 Total Assets 777 219000 620000 30.962 5490000 Employees 918 443.644 948.309 1 10480 Profit margin 885 4.397 14.238 -91.155 75.008 Patents 918 19.454 203.975 0 2527 Trademarks 918 2.806 12.368 0 140 Tax credit 918 19268.45 48663.81 0 428000 Age 918 16.91 17.966 0 120 Controls Revenue 10448 659000 4250000 1.897 1.12e+08

Total Assets 9463 711000 5010000 0 1.15e+08

Employees 10448 2105.981 20657.96 1 710000 Profit margin 10397 5.103 13.611 -99.266 100 Patents 10448 1.903 37.898 0 2527 Trademarks Age 10448 10449 1.164 15.761 5.392 17.419 0 0 140 184

26

Matched Controls All controls

*** P<0.01, ** P<0.05, *P<0.1

Table 2: Results

Model:

Dependent variable

Intermediary

dummy (ln)Revenue (ln)Profit (ln) Revenue (ln) Profit

intermediary dummy Employees -.000*** (.001) Age .001 (.002) Industrial complexity -.158** (.080) Treated -.318** .315** -.583*** .366*** (.133) (.146) (.114) (.107) Treated * post -.232* -.100 .044 -.135 (.123) (.159) (.125) (.120) Patents .505*** (.091) Observations 1,757 2,227 1,764 11,366 9,259 324