The influence of R&D investments on firms’ financing Evidence from the US and Germany

(1)

The influence of R&D investments on firms’ financing

Evidence from the US and Germany

University of Groningen

Faculty of Economics and Business

Department of Economics, Econometrics and Finance

MSc Business Administration Finance

11 August 2012

Name:

A.J. Nederlof

Student number:

1474928

First supervisor:

dr. L. Dam

(2)

ABSTRACT

I investigate how firms’ R&D investments influence their financing structure using a comprehensive panel data sample of 910 German and US firms in the period 1971 2010. To this end I use different methods, including standard least squares regressions and instrumental variable regressions. I present evidence that investments in R&D determine the level of firms’ equity, transactional debt and relational debt. The effect of R&D on financing by means of convertible securities remains inconclusive. Focusing on several time frames and using limited dependent variable methods does not affect these results. Furthermore, with respect to the impact of R&D on equity financing, I find a significant difference between German and US firms. On the other hand, the analyses provide inconclusive evidence with respect to differences between German and US firms regarding the effect of R&D on financing through transactional debt, relational debt and convertible securities.

(3)

1. Introduction

Firms that align their governance structure with their research and development (“R&D”) strategy perform better than those that are misaligned (David et. al., 2008). Therefore, firms tend to align their governance structure to R&D investments. A firm’s governance structure is determined by (inter alia) the use of different financial instruments, and thus by its financial structure. By implementing and extending current literature, this study examines the influence of R&D on firms’ financing and provides an insight in the influence of investments in R&D on the use of four different financing instruments.

Further, access to different forms of external financing should be considered as a source of competitive advantage, particularly when firms heavily depend on external financing (Barney, 1991; Folta and Janney, 2004). Information and agency costs raise the costs of external capital and/or may even limit access to external funds (Myers and Majluf, 1984; Jensen and Meckling, 1976). Because R&D investments extensively place a burden on firms’ financing, are subject to substantial information asymmetries and often require more capital than is internally generated by operating cash flows, these firms rely on external capital to overcome the gap between required capital and available capital (Thornhill and Wang, 2010). Consequently, the availability and cost of external capital is relevant in financing R&D investments.

In this thesis I examine the effect of investments in R&D on firms’ choice of financing instruments using a panel data set consisting of 910 German and US firms from 1971–2010. The financing instruments are decomposed into equity, transactional debt, relational debt and convertible securities (four financials instruments). The focus of this thesis is on the influence of internally developed R&D and does not include, for example, the effects of R&D knowledge spill-over between universities and firms. To deal with potential endogeneity issues, I use both standard least squares regressions and instrumental variable methods to extract the impact of R&D on the aforementioned financing instruments. These models control for firm specific characteristics, macro influences and the influence of time. Finally, this thesis extends current empirical literature by including firms from different industries, and by also covering the influence of R&D on short term debt.

(5)

investments in R&D generate intangible assets which poorly suffice as collateral. TCE assumes that firm-specific (intangible) assets have lower resale value than general assets (Williamson, 1988). This results in lower resale value of R&D knowledge in case of bankruptcy, and causes debt providers to be more reluctant to invest in R&D intensive firms (Kochhar, 1996; Long and Malitz, 1985). Vincente-Lorente (2001) confirm that specific and opaque resources created by internally generated R&D, are inversely related to financial leverage.

Recently, the focus of research has shifted from a general distinction of debt and equity (Balakrishnan and Fox, 1993; Long and Malitz, 1985; Vincente-Lorente, 2001), to different forms of debt and equity financing. Building on the distinctions made by Boot (2000), David et al. (2008) argue that debt should be treated heterogeneous, with distinctions between relational and transactional debt. They examine the differences between governance structures of relational debt and transactional debt, especially focusing on the relationship between lenders and borrowers. They prove that failing to account for debt heterogeneity can have substantial impact on the returns of investments in R&D and thus provide evidence that, additional to capital providers´ investment decisions, the board’s financing choices to align the debt structure with the R&D investments are important.

Thornhill and Wang (2010) further develop the idea of debt heterogeneity and take a more comprehensive view of debt and equity financing, categorizing financing instruments into four governance groups based on the degree of intervention barriers and appropriation discrepancy. However, several caveats in Thornhill and Wang’s study make further research necessary. First, Thornhill and Wang (2010) only focus on the petroleum industry. The essence of my thesis takes into consideration all R&D investing firms, irrespective of the industry the firms is active in.1 Secondly, I also consider the influence of R&D on short term debt financing. Firms heavily rely on short term bank financing, like overdrafts. Not recognizing the short term debt would probably bias the results of total financing. Third, I make a comparison between the US and Germany. The US is commonly known as a market based system, in which firms rely on financial markets, like financing by means of transactional debt and equity. Germany on the other hand should be qualified as a bank based system, in which firms rely more on relational debt (Allen and Gale, 2001). They argue that market based systems suffer less from information asymmetries between investors and management. Since the different governance structures according to TCE are based on the resale value of assets and information asymmetry with respect of R&D projects, information asymmetry on markets may also explain differences in the effects between US and Germany. Since German investors suffer from higher information asymmetries I expect to find stronger effects for the respective financing instruments for German firms than for US firms.

(6)

Consistent with expectations, I find evidence that R&D expenses are able to explain the capital structure of firms with respect to some of the financing instruments. I find that R&D has a positive effect on equity and has a negative effect on transactional debt and relational debt financing. This applies to both Germany and the US. In the main regression results, I do not find significant effects of R&D on convertible securities. However, I do find that R&D has a significant positive effect on financing through convertible securities for US firms, which was contrary to an expected negative sign. This unexpected effect may be caused by the influence of the convertibility of debt, which enables investors to profit from successful R&D investments and rising stock prices and therefore reduces the appropriation discrepancy. The robustness tests confirm to a large extent that governance structure of firms can be explained by R&D investments, although the effects are not always confirmed for all subtests.

Furthermore, I find evidence that there are indeed differences between German and US firms with respect to equity financing. However for the other financing instruments, the tests cannot reject the hypothesis that the effect is the same across both countries.

(7)

2. Background literature

In this chapter I present an overview of the theory underlying the hypotheses to be presented in chapter 3. First, paragraph 2.1 discusses TCE and the governance of debt and equity. Afterwards, I point out the importance of R&D investments and deliberate about the use of financing instruments in relation to the protection of R&D knowledge against exchange hazards. Further, I discuss (empirical) literature regarding R&D in connection with firms’ capital structure in paragraph 2.2. Paragraph 2.3 reviews the differences between four financing instruments and describes these in a framework. Finally, paragraph 2.4 focuses on the differences between leverage in bank based versus market based financial systems, which form a basis for the identification of hypotheses with respect to differences between Germany and the US.

2.1. Governance and transaction cost economics

The modern theory of capital structure has been developed by Modigliani and Miller (1958), in a theory stating that, in the absence of taxes, bankruptcy costs, agency costs, and asymmetric information, and in an efficient market "the average cost of capital to any firm is completely

independent of its capital structure and is equal to the capitalization rate of a pure equity stream of its class”. However, the assumptions underlying this capital structure irrelevance theorem are unlikely to

exist in reality. Apart from taxes and transaction costs, informational asymmetries and incentive problems will likely create distortions in capital markets. As a result, information asymmetry and agency costs raise the cost of external financing (debt and equity), and have a significant impact on the functioning of capital markets and the manner in which firms are financed. This paragraph explains the different governance of debt and equity according to TCE and furthermore considers the differences between relational and transactional debt.

2.1.1. Transaction cost economics and agency theory

In explaining firms´ capital structure, the agency theory is widely used. Since the theoretical underpinnings of this thesis are based on TCE, I will delilberate on this theory and shortly explain the commonalities and differences between the agency theory as theorized by Jensen and Meckling (1976) (“AT”), and TCE as described by Williamson (1988).

(8)

The most important differences between AT and TCE are the following. In the first place, TCE regards the transaction as the basic unit of analysis (Williamson, 1988), while for AT the basic unit of analysis is the individual agent (Jensen, 1983). TCE aligns transactions (which differ in their attributes) with governance structures (the costs and competencies) in a discriminating way, of which asset-specificity is the most important condition. The joining of incomplete contracting with asset specificity is distinctively associated with TCE and not with AT. Secondly, the distinctive TCE orientation intends to reduce the maladaptation costs, costs which are incurred when transactions drift out of alignment, through judicious choice of governance structure (market, hierarchy or hybrid), rather than realigning incentives and pricing the associated costs out. Thirdly, AT does give little concerns for dispute resolution and focuses on ex ante contract approaches. TCE on the other hand more focuses on dispute avoidance and the machinery for processing disputes and therefore TCE has a focus for harmonizing ex post contractual relations.

In this thesis, I focus on the influence of R&D investments on firms´ financing in connection with their governance structure. R&D investments lead by definition to a high amount of non-deployable assets within a firm (Williamson, 1988). Since the influence of the degree of asset-specificity within the firm is one of the most important differences between the AT and the TCE, I build on the assumptions and theoretical foundations of TCE.

2.1.2. Transaction cost economics

According to TCE, debt and equity should be treated as different governance structures rather than as financing instruments (Williamson, 1988). Williamson therefore examines financial leverage from a TCE perspective. Simplifying the reality, he assumes that there are only two forms of finance that projects or firms should be financed with, entirely with debt or entirely with equity. In order to examine the differences for firms that are financed differently, this theory further assumes that debt and equity have different governance characteristics. Debt is a governance structure including the following conditions: (1) stipulated interest payments at regular time intervals, (2) the business should continuously meet certain liquidity needs, (3) sinking funds will be set up and the principal will be paid at the expiration date, (4) in case of default, the debt-holder will exercise pre-emptive claims against (assets of) the firm. In case of failure to pay scheduled payments, the firm defaults and should liquidate. On the other hand, firms have the option to use equity, which from a governance structure point of view should be characterized as follows: (1) it bears the residual value of the firm in earnings and asset liquidation, (2) it contracts for the duration of the entire life of the firm, and (3) the board of directors is awarded to equity.2 Therefore, Williamson (1988) concludes that debt is a simple and

2_{The shareholders elect the board, have the power to replace the management, decide on management compensation, have access to internal}

(9)

market-like governance structure, while equity is a more administrative and complex structure with higher setup costs.

According to Williamson, asset-specificity influences the use of debt and equity. TCE states that asset specificity reduces the redeployment value of assets in the event of default and will therefore adjust the terms of firms’ debt financing adversely (Williamson, 1988). In order to reduce the negative effects of investments in specialized assets, firms will forego some of the specialized investments in favor of greater redeployability and favorable debt terms. Therefore, the cost of capital is reduced by using equity due to the monitoring function and its claim on the residual firm value.

Williamson furthermore states that although TCE assumes that the costs of both debt and equity rise with asset specificity, debt financing rises more rapidly. In a rule-governed regime with debt, firms will be forced into bankruptcy, while an equity regime causes more value-enhancing decisions, for example decisions based on added control systems (see also the differences between transactional and relational debt under 2.1.3). Consequently, Williamson argues that firms with a significant amount of redeployable assets will be financed with debt, while equity is preferred as assets become less redeployable.3 Therefore, debt holders select firms with lower asset specificity or require higher interest rates (Titman and Wessels, 1988; Balakrishnan and Fox, 1993; Vicente-Lorente, 2001), and a negative relationship exists between financing through debt and R&D.

2.1.3. Differences between transactional debt and relational debt

Although the treatment of all debt as transactional is not an unreasonable generalization for US firms, this theory lacks generalizability to contexts in which firms are commonly financed with relational debt. This also means that not all sorts of debt face the same (negative) relationship for R&D as described under paragraph 2.1.2. The contexts to which this applies are for example small US firms and firms in many other developed nations in the world, including but not limited to Germany (Allen and Gale, 2001). The distinction between different kinds of debt, including diverging governance implications, yields a more applicable theoretical framework. In this paragraph I will discuss the differences between transactional and relational debt in order to explain why transactional debt is similar to the debt Williamson described, while relational debt is more characterized as hierarchical governance.

Transactional debt lending is viewed as arms-length finance focusing on single transactions, instead of considering long-term information intensive relationships like relational debt providers (Boot, 2000; Boot and Thakor, 2000). Consequently, transactional debt lenders are subject to more extensive information asymmetries than relational debt lenders. In contrast to relational debt lenders,

3_{In contrast to the AT (Jensen and Meckling, 1976), TCE assumes that debt financing is the natural financing instrument while equity is a}

(10)

transactional debt lenders are not specialized in screening and monitoring firms. Furthermore, they are generally more dispersed than relational debt lenders, which may enlarge the free-rider problem and reduce the monitoring function of the transactional debt lenders. Therefore, the main distinguishing characteristics of transactional and relational debt are dispute resolution, adaptation and monitoring (David et al., 2008).

Firstly, dispute resolution refers to the methods parties resolve disputes. Transactional debt lenders rely strictly on contractual terms and court intervention and only consider the direct returns of debt. Relational debt lenders try to resolve disputes with clients bilaterally, preferably without court intervention. Relational debt lenders take into consideration the ongoing relationship with the borrower, and consider the indefinite duration of the future revenues and thus are motivated to help the struggling clients with renegotiations and relaxing loan terms. Additionally, the relational debt lenders are more prepared to provide the borrowers with additional funds (Boot, 2000). From a practical point of view, debt renegotiation requires agreement with a few other relational debt lenders, which is more feasible than reaching agreement with many unknown transactional debt lenders. Besides, forbearance may enhance the bank’s reputation to attract new clients, which might induce relational debt lenders to be less reluctant to provide struggling companies with additional capital (Chemannur and Fulghieri, 1994). Transactional debt lenders lack the motivation to be forbearing, due to the single tie with the company, and they lack the ability to be forbearing, because the shares are diffusely held. Empirical evidence supports the intuition that relational debt lenders exercise forbearance for distressed firms and make strategic investments to successfully support the restructuring of the company and avoidance of bankruptcy (Hoshi, Kashyap and Scharfstein, 1990; Gilson, John and Lang, 1990).

(11)

Thirdly, relational debt and transactional debt distinguish on the monitoring function for the borrower. Transactional debt providers are diffuse and the lenders lack the scale of economies that justify demanding monitoring. Consequently, the monitoring function of transactional debt providers is relatively small. Transactional debt lenders particularly rely on monitoring through objective criteria which show conformity with the debt terms as defined by Williamson.4 Relational debt providers on the other hand have an incentive to monitor the company elaborately, because the exercise of forbearance and administrative control requires access to detailed information. For example, multiple interactions and seats on client’s board of directors may give relational debt lenders access to proprietary information, which relational debt lenders lack.

As relational debt and transactional debt have different characteristics, I also expect different effects of R&D on the use of these respective financing instruments, which will be discussed under paragraph 2.3 and chapter 3. So far, the differences mainly relate to an investor’s point of view. In the next chapter, I also consider on the differences of different financing from a firm’s perspective.

2.1.4. R&D and appropriability of debt

Firms invest in R&D because they aim to build capabilities that enhance competitive advantages, which may lead to superior performance compared to (in)direct competitors (David et. al., 2008). Building competitive advantages is widely recognized as one of the main characteristics of R&D development. By using these competitive advantages, firms tend to increase their ability to outperform competitors in the market to increase firm’s profits, also in the long run.

Another characteristic of R&D investments is that the nature of R&D investments is risky (Miller and Bromiley, 1990), and subject to serious exchange hazards. In order to mitigate exchange hazards, R&D investments require strong governance safeguards (Hill and Snell, 1988). In accordance with David et. al. (2008), I make a distinction between three different types of hazards of R&D, which influence a firm’s choice for a particular type of governance (different sorts of debt or equity), namely: (i) asset specificity, which is the extent to which investments lose value when redeployed to alternate uses, (ii) uncertainty about the states of nature and the behavior of other parties, and (iii) the appropriability of the returns arising from a transaction.

Firstly, investments in R&D create knowledge-based intangible assets that have the greatest value when used in connection with the other firm’s complementary assets (Helfat, 1994). Consequently, R&D investments do not serve as good collateral for lenders, as non-redeployable assets lose considerable value in case of bankruptcy (Long and Malitz, 1985). The obligation to meet regular payments following from debt financing can deteriorate financial flexibility, and the chance of liquidity problems of the firm might induce managers to cut down ongoing R&D programs (O’Brien,

(12)

2003), amongst other things the possibility of disrupting the continuity of vital R&D investments (Dierickx and Cool, 1989).

Secondly, the measurement of success of R&D investments is difficult, because of the considerable time lag between the date of investment and the date of payoff and because of the considerable number of external factors and events which may appear during the investment period and can affect the pay back of the principal amount (Laverty,1996; Hill and Snell, 1988). Additionally, the criteria to measure success involve highly qualitative judgments, including the probability of success, harmony with existing technologies, the match with other current activities, the strategic importance of a program to a firm (Osawa and Murakami, 2002), and indirect or spillover benefits (Oral, Kettani, and Lang, 1991). Potential adverse selection and moral hazard problems (the borrowers who will likely suffer credit risks are most likely to seek financing, respectively borrowers may try to shift risks to lenders after the financing has been obtained) result in the request of more detailed qualitative and subjective information from lenders.

Finally, the leakage of information of R&D programs can lead to imitation of products by competitors (Teece, 1986), and therefore the returns to R&D investments are subject to weak appropriability. Legal safeguards such as patents are often ineffective, as competitors are able to avoid the contents of patents (Levin, Klevorick, Nelson and Winter, 1987). Providing information to lenders that the firm is making appropriate investments in R&D requires public disclosures of detailed data on the confidential R&D programs, and would consequently weaken the appropriability regime for the investments and deteriorate managers’ motivation to make R&D investments (Bhattacharya and Chiesa, 1995).

(13)

2.2. Empirical findings

In line with the governance of debt and equity described under paragraph 2.1, the difference between transactional debt and relational debt and the safeguards required by R&D investments mentioned in paragraph 2.1.4, evidence has been provided that R&D investing firms prefer equity financing above debt financing (Balakrishnan and Fox, 1993; Vincente-Lorente, 2001; David et. al., 2008; Thornhill and Wang, 2010).

Using a panel data sample of 295 single business firms operating in 30 different SIC industries, Balakrishnan and Fox (1993) find a negative effect of R&D intensity on leverage. They conclude that firms that tend to invest heavily in R&D, and thus potentially create more intangible and firm specific assets, will find it more difficult to fund R&D investments with debt. Vicente-Lorente (2001) investigate the influence of internally developed R&D on firm’s financial leverage using a panel data consisting of 260 observations obtained from 52 non-financial firms listed on the Spanish stock exchange based on TCE. Similar to Balakrishnan and Fox (1993), they conclude that internally developed R&D has a significant negative effect on financial leverage.

More recently, David et al. (2008) investigated the influence of R&D on the governance structure of firms. In contrast to earlier studies, their research took into account the different governance structure of relational and transactional debt. The treatment of divergent governance implications of relational and transactional debt yields a broadlier applicable theoretical framework from which they conclude that relational debt shows similar governance effects as equity and transactional debt is more equivalent to the governance of debt described in current governance studies about capital structure (Williamson, 1988; Rajan and Zingales, 1995).

(14)

2.1.3, the rights attached to certain kinds of debt vary. With respect to relational debt, Boot (2000) points at the ongoing interaction between banks and firms. He emphasizes that in relationship banking borrower specific information is available to the bank and that the bank has the ability to monitor the use of funds, which may result in successful intervention in the conduct of management. Thirdly, TCE involves financing of investments. It however fails to recognize the specific character of R&D investments which returns also influence the firm’s success in the long run. Building firm-specific competitive advantages may lead to superior performance (Ravenscraft and Scherer, 1982). TCE does not acknowledge this significant role of R&D investments.

In consideration of the aforementioned shortcomings, Thornhill and Wang (2010) identify four different financing groups, based on the degree of intervention barriers to intervene in the conduct of management and appropriation discrepancy with respect to the firm’s assets. The first indicates the extent to which providers of capital are able to intervene in the conduct of management and to what extent they are able to monitor management. The latter expresses the extent to which capital providers are residual claimants in the firm’s assets. The two-by-two categorization results in the quadrant as presented under in Figure 1. In the next paragraph, I will discuss the characteristics of four different financing instruments from the quadrant and the associated financial instruments.

In te rv e n ti o n b a rr ie rs IV - Convertible securities

- high intervention barriers - Low appropriation discrepancy

II - Transactional debt

- High intervention barriers - High appropriation discrepancy

I - Common Equity

- Low intervention barriers - Low appropriation discrepancy

III - Relational debt

- Low intervention barriers - High appropriation discrepancy Appropriation discrepancy Low High L o w H ig h

Figure 1: based on the model of Thornhill and Wang (2010), this two-by-two categorization of financing instruments is based on appropriation discrepancy and intervention barriers. Appropriation discrepancy indicates the extent to which providers of capital are able to intervene in the conduct of management and to what extent they are able to monitor management. Intervention barriers express the extent to which capital providers are residual claimants to the firm’s assets.

2.3. Four financing instruments

(15)

commercial loans are popular financing instruments. In (relational) bank based systems like Germany, bank loans are popular in particular (Allen and Gale, 2000). This thesis divides financing instruments into groups based on appropriation discrepancy and intervention barriers. This two-by-two categorization of financial instruments of large listed firms in the US and Germany, results in four categories. In this paragraph I categorize financial instruments with similar characteristics in the same quadrant, which is based on the model set up by Thornhill and Wang (2010).

2.3.1. Quadrant I - Equity

Especially equity, in the form of shares in the company, fits well into Quadrant I. The shareholders have access to the firm’s proprietary information and can relatively simply intervene in the conduct of management. For example, shareholders have the right to elect the management board pro-rata to their votes in the general meeting of shareholders, the right to replace the management board or individual managers, the right to decide on management compensations, the right to access to internal performance of the company on a timely basis (mostly during the annual meeting of shareholders), the right to authorize audits for special investigation purposes, the approval right for important investments and operating proposals before implementation and in other respects bears a decision-review and monitoring relationship to the company’s management (Fama and Jensen, 1983; Williamson, 1988). Therefore, equity has low intervention barriers to the conduct of management. Regarding appropriation discrepancy, holders of equity are entitled to the residual value of the firm. Since R&D investments help build firm specific resources and capabilities, these R&D investments might also improve the firm-specific resources and capabilities of the firm (see paragraph 2.1.4) resulting in more firm value. Thus, equity holders have low appropriation discrepancy to the residual value of the firm. Based on these characteristics, equity is categorized with low appropriation discrepancy and low intervention barriers and fits perfectly in Quadrant I.

2.3.2. Quadrant II - Transactional debt

(16)

Corporate bonds, including notes, subordinates and debentures, classify as transactional debt, since these instruments are issued in specific monetary denominations on capital markets and sold to individuals and monetary institutions. The maturity of the securities determines the fixed contract duration (David et al, 2008). The payoff structure is simple, as bond holders are entitled to the principal amount and a periodical interest component. Bond holders are thus entitled to direct returns available from holding the securities. Furthermore, bond holders do not have the opportunity or the incentive to monitor the borrower intensively, since they have only simple ties with the borrower. Therefore, holders of capital market debt fit perfectly in Quadrant II with high appropriation discrepancy and high intervention barriers.

2.3.3. Quadrant III - Relational debt

The definition of relational debt is based on the definition of relationship banking defined by Boot (2000). Relationship banking includes normal lending services to companies but also other lending services, such as financial lease services. Relationship banking should be defined as the provision of a financial party that (i) invests in obtaining customer-specific information often via proprietary information, (ii) that the financial party evaluates the investments with the borrower over time and across products and (iii) that the financial party gathers information beyond readily available public information. Especially the first and last characteristics are important for the model used. Relationship lenders create an opportunity to benefit from the informational advantage built up over time. Next to banks, also other financial parties, like finance companies and financial lease providers, can be classified as relational debt providers, which is in line with Carey et al. (1998).

(17)

2.3.4. Quadrant IV – Convertible securities

The fourth quadrant is characterized by low appropriation discrepancy and high intervention barriers. In line with Thornhill and Wang (2010), I classify preferred stock and convertible debt as convertible securities. Generally, holders of preferred stock do not have the same rights as common stock holders to intervene in the conduct of management. Although holders of preferred stock have access to proprietary information regarding the company, they have limited voting rights and power to intervene in the conduct of management. On the other hand, they are entitled to the residual claims of the company and have a preferred right to the dividends of the company. Preferred stock holders thus have high intervention barriers and low appropriation discrepancy.

Similar to holders of transactional debt, convertible debt holders have limited access to the firm’s proprietary information. They lack the option to intervene in the conduct of management because the convertible debt contracts are established with similar characteristics as transactional debt. Thus, the intervention barriers for convertible debt holders are high compared to relational debt holders. On the other hand, convertible debt has a preliminary dividend right and possesses the option to exchange the security for the firm’s common stock. If the company performs well, the convertible debt holders can convert their securities in common stock of the company and will be entitled to the firm’s residual value. In other words, the convertible debt holders are entitled to the upside potential of R&D investments. The option to convert into common stock eliminates the appropriation discrepancy which is attached to transactional debt. Therefore, convertible debt should also be placed in Quadrant IV. In connection with the framework, I explain the differences between market and bank based systems the next paragraph. Additionally, I consider the different characteristics of financing instruments in connection with the different financial systems in chapter 3.

2.4. Market and bank based systems

The latter paragraph chapter discussed the framework presented by Thornhill and Wang (2010). In their sample Thornhill and Wang (2010) only consider US firms. David et al (2008) recognize that in many other developed nations in the world statements about treating debt may be less generalizable. For example, there is a wide variation between the structures of financial systems around the world. The two polar extremes are the German model, in which intermediaries like banks play a major role, and the US model, in which the financial markets dominate (Allen and Gale, 1995).

(18)

literature in their paper. In this paragraph I point out several significant differences between the German and US financial based systems.

Market based systems are characterized by the importance of capital markets and low concentrated banking industry. In bank-based systems on the other hand, banks play a major role in the financing of firms (Allen and Gale, 1995), compared to capital markets. Although a traditional comparison can be made between US and EU countries (including Germany), Europe has shown a clear expansion of the arm’s length financing over the last three decades (Rajan and Zingales, 2003). Despite this change they conclude that that relationship-based financing still remains predominant in Europe.

Allen and Gale (2001) set out important differences between the market-based and bank-based financial systems. First, they show that internal financing is by far the most important source of funds in all countries. This is in line with the pecking order theory developed by Jensen and Meckling (1976). In order to understand the importance of banks in both countries Allen and Gale show, by examining the savings and the holdings of financial assets per country, that the importance of banks in Germany is significantly bigger than in the US, and that on the other hand the importance of the capital markets in the US is of more importance than in Germany.

Secondly, they state that the extensive use of the main bank system in Germany means that long-lived relationships between large firms and banks are commonplace. US banks have less long-term relationships with firms. This causes that US banks are less involved in the conduct of management than German banks are, which is in line with Boot (2000), who also emphasizes the relative importance of German banks compared to US banks.

(19)

banks in bank-based systems have more proprietary information than banks in market based systems, which better enables these banks to intervene in the conduct of management (Boot, 2000).

(20)

3. Hypotheses

In this section, I identify several hypotheses to answer the main questions. These hypotheses form the framework for further research in the following chapters. Since I research different financial systems, I expect different influences of R&D on the use of the distinguished financing instruments in US firms compared to German firms. In the paragraphs below I will formulate in the aggregate twelve hypotheses, for each country per governance instrument one hypothesis, and one hypothesis per financing instrument to compare the effects across countries.

3.1. Equity

Common equity has low appropriation discrepancy and low intervention barriers. Further, it has been argued that investments in R&D lead to more specific assets. TCE suggest that asset-specificity of R&D investments hinder firm’s access to debt financing and will therefore raise firm’s ratio of equity to total financing (Williamson, 1988). Secondly, because equity does not force firms to pay regular interests, firms shall prefer financing through equity due to reduced financial obligations relating to interest payments. This may help firms to buffer failures in R&D projects. Finally, equity is an adequate safeguard against the hazards posed to R&D. Therefore, equity is a good financing instrument for R&D intensive firms. Consecutively, I formulate the following hypothesis with respect to common equity:

Hypothesis 1: US firms face a positive relationship between R&D and financing through common equity.

Furthermore, information asymmetry between the management board and investors on German capital markets is higher than in on US capital markets due to extensive filing regulation in the US. Due to this higher information asymmetry investors have more difficulty to judge the resale value of intangible assets existing pursuant to R&D. Therefore, I expect that German firms shall to a greater extent be financed with equity than US firms (see under 3.2), which has the following consequences for financing through equity for German firms:

Hypothesis 2a: German firms face a positive relationship between R&D and financing through common equity.

Hypothesis 2b: German firms face a stronger positive relationship between R&D and financing through common equity than US firms.

3.2. Transactional debt

(21)

firms to limit (other) financial obligations to mitigate risk of financial distress. Furthermore, since transactional debt holders are not entitled to firm´s residual value, they do not have an incentive to accept additional risk and lower resale value of assets associated with R&D. Furthermore, they lack the ability to intervene in the conduct of management. Finally, transactional debt does not provide an adequate safeguard against the hazards posed to R&D, because of which management may choose for relational debt instead of transactional debt. Therefore, I formulate the following hypothesis with respect to transactional debt:

Hypothesis 3: US firms face a negative relationship between R&D and financing through transactional debt.

Information asymmetry between management and investors on German capital markets is higher than on US capital markets, due to extensive filing regulation in the US. Therefore, German investors should be more reluctant to finance firms with transactional debt than US investors, which has the following consequences for financing through equity for German firms:

Hypothesis 4a: German firms face a negative relationship between R&D and financing through transactional debt.

Hypothesis 4b: German firms face a stronger negative relationship between R&D and financing through transactional debt than US firms.

3.3. Relational debt

Relational debt has high appropriation discrepancy and low intervention barriers. As R&D investments increase the ability to outperform competitors in the market, relational debt lenders stimulate firms to some extent to innovate since the longstanding relationship depends on the success of the firm. Furthermore, relational debt provides adequate safeguards against the hazards posed to R&D (David et. al., 2008), because of which relational debt better suffices the governance requirements of R&D investing firms than transactional debt. Further, firms investing in R&D need to maintain financing flexibility, which can be realized by relational debt (David et. al., 2008). On the other hand relational debt lenders have access to the firm’s proprietary information to some extent (Boot, 200; David. et.al, 2008), but are still outsiders of the firm and because they can only to some extent intervene in the conduct of management (Stiglitz and Weiss, 1981; Williamson 1988). Besides, the assets generated by R&D investments still remain intangible, which means that the TCE assumption holds again suggesting that relational debt lenders may tend to avoid much R&D investments. This leads to the following hypothesis with respect to relational debt:

(22)

Although I hypothesize a negative relationship between R&D and financing by means of relational debt, it has been argued that relational debt better suffices the requirements of firms and investors than transactional debt with respect to firms that invest in R&D. Therefore, I expect a less negative effect of R&D for relational debt than for transactional debt.

The proprietary information obtained by US banks is lower compared to German banks, since German banks in general have stronger and more long-standing relationships with firms than US banks (Boot, 2000). This has the following consequences for financing through relational debt for German firms:

Hypothesis 6a: German firms face a negative relationship between R&D and financing through transactional debt.

Hypothesis 6b: German firms face a less negative relationship between R&D and financing through relational debt than US firms.

3.4. Convertible securities

Convertible securities are qualified as instruments with low appropriation discrepancy and high intervention barriers. In line with the arguments used for transactional debt, holders of convertible securities are expected to be reluctant to invest in firms that invest heavily in R&D as they cannot impede risky decisions of the management board and they have a senior claim compared to equity on the firm’s assets in case of bankruptcy. Unlike transactional debt lenders, holders of convertible debt do have the option to exchange their debt into shares, which eliminates the appropriation discrepancy. Because of this, investors profit from the potentially improved residual value of the firm due to R&D investments. This effect may mitigate a part of the negative expected sign and therefore I formulate the following hypothesis with respect to transactional debt:

Hypothesis 7: US firms face a negative relationship between R&D and financing through convertible securities.

Since information asymmetry on German capital markets is higher than on the US markets, investors should be more reluctant to finance firms with convertible securities, which has the following consequences for financing through convertible securities for German firms:

Hypothesis 8a: German firms face a negative relationship between R&D and financing through convertible securities.

Hypothesis 8b: German firms face a stronger negative relationship between R&D and financing through convertible securities than US firms.

(23)

4. Methodology

Using a comprehensive panel data set consisting of 910 firms (136 German firms and 774 US firms), I investigate the influence of R&D investments on firms’ financing structure. This sections starts with explaining the structure and advantages of a panel data set. The next paragraph describes the panel data models in general. The last paragraph describes the models that I use to test the hypotheses.

4.1. Panel analysis

In line with research by Thornhill and Wang (2010), this study uses a panel data set. Panel data has the differentiating aspect of containing a cross-sectional dimension as well as a time-series dimension. The panel contains time series and cross-sectional dimensions of US and German firms from 1971 through 2010. For this study the use of panel data has several advantages over pooling the data together (Baltagi, 2008). First, pooling data from the panel data set disregards the heterogeneity between firms and over time which eventually can result in biased results. Second, panel data are more informative, have more variability within the sample, results in less collinearity among the variables and increases the degrees of freedom resulting in more efficiency of the regression results. Because the cross-section dimensions add significant variability, collinearity plays a minor role in panel datasets. Especially, this characteristic is important for this study because it is highly probable that firms have constant R&D expenses over years resulting in more or less constant financing constructions. The panel data helps reduce the variability problem. Third, panel data is better able to identify and measure effects than pure single cross-section and time-series data. Panel data can discriminate between the effect of time and entities on the variables. This is especially relevant for the US sample since this sample contains several long cross-sections. One of the disadvantages of panel data is that macro panels, which are panels with long time series (20-30 years), do not recalibrate for cross-company dependence and may lead to misleading results.

For the US, the length of the panel data varies from 1 observation till 40 observations. The German firms in the sample better fit the description of a micro panel (maximum of 12, minimum of 1 and average of 6.5), which contains a large number of companies ( ) over a short period of time ( ). For the US, the panel also fits the description of a micro panel. Although this sample has more length (maximum of 40, mean of 12.5 and minimum of 1), the length of the samples is not enough to cause dependence problems (Baltagi, 2008). As most of the cross-sections have a length of around 10 years the dependence problems shall most likely not error the estimated coefficients.

4.2. Panel techniques

(24)

Furthermore, I recognize that there might be an endogeneity problem between firms’ financing and R&D investments. Application of OLS will lead to biased coefficient estimates when endogenous variables are included. Therefore, I conduct two stage least squares tests with instrumental variables (“IV-2SLS”) to reduce the endogeneity problem, as a robustness check. Finally, I use a non-linear panel Tobit model. OLS is a linear model which assumes a linear relationship between all relevant variables. Although the distribution of equity is almost normal, this is far from normal for transactional debt, relational debt and convertible securities. I observe a large amount of zero values in the sample which may cause parameter estimates in linear models to be biased and inconsistent. A non-linear model like a truncated Tobit regression model takes these characteristics into account. In the next sections, I discuss the use of OLS, IV-2SLS and non-linear Tobit methods in more detail. In this thesis, I do not make use of dynamic model of generalized method of moments. In the empirical literature, the nature of a firm’s capital structure has been tested using dynamic econometric models because it has been suggested that firms periodically readjust their capital ratio towards a target ratio that reflects the costs and benefits of the financing instruments for a particular company (Gonzalez and Gonzalez, 2011). Another reason to use dynamic models is that the explanatory variables with dynamic models do not necessarily have to be strictly exogenous, but may be correlated with past and current realizations of the explanatory variable (Roodman, 2006). Despite these advantages of dynamic models, the object of this thesis and the data sample construction does not need the use of a dynamic model. In the first place, my object is to measure the impact of R&D on firms’ financing and the object is not to focus on the deviation from a target ratio. Secondly, a dynamic panel data model requires a small T and large N data set. Due to many large ’s (especially for the US data set) the applicability of a dynamic model is not more efficient than the use of a standard fixed effects model (Roodman, 2009). Furthermore, my dataset contains several gaps between firm years. The use of first-difference estimators transforms and materially weakens the results of dynamic models because it magnifies gaps in unbalanced panels. The use of second orthogonal deviations might overcome this problem, but would also reduce the sample dataset (Arellano and Boover, 1995).

(25)

4.3. Methods

After describing the testable relationships, this paragraph explains the OLS, IV-2SLS and panel Tobit methods used to test the hypotheses.

4.3.1. Testable relationships

To study the impact of R&D investments on financing structure of a firm, I describe the tested models below. In each model, one of the four financing instruments is the dependent variable. These four financing instruments are equity, transactional debt, relational debt and convertible securities. The independent variable consists of R&D expenses. Furthermore, control variables have been added. In the literature, several methods have been specified to examine the effects of my control variables on firms´ financing. Additionally to OLS, I use IV-2SLS and panel tobit analyses as robustness checks. The variable abbreviations are described in paragraph Table A in the Appendix (p. 58). The four models are as follows:

(1)

(2)

(3)

(4)

Furthermore, I include year dummies to correct for potential systematic effects through the years.

4.3.2. Random and fixed effects model

In panel data estimators, there are generally two approaches which are useful for analyzing financial data, namely random and fixed effects. Fixed and random effects allow for heterogeneity between firms and can be applied to a large number of cross-sectional observations with short time periods. In a fixed effects model, the intercepts control for cross-sectional specific characteristics (Carter-Hill, 2008). The fixed effects model decomposes the disturbance term, , into an individual specific

effect, and a remainder error term, , which varies over time and entities and captures the

relationship that is left unexplained about the dependent variable. This formula adjusted for fixed effects is stated:

(26)

sample. The intercept thus arises from a common intercept, α, plus a random variable εi. This random variable varies cross-sectionally, but is assumed to be constant over time. The heterogeneity does not come from an added dummy variable as in the fixed effects model, but comes from a random walk of the error term.5 The general formula for a random effects model is stated

The underlying conceptualization of fixed and random effects is that the estimators give consistent results if there is no correlation between and the explanatory variables. In principle the random effects model should produce more efficient estimation results for two reasons. The transformations which are used in a random effects model will not remove the time-invariant explanatory variables, and induces impact on . Secondly, compared to a fixed effects model, a random effects model uses fewer parameter estimates. This saves degrees of freedom and should result in more efficient estimations. Brooks (2008) states that in in principle a random effects model is more appropriate when the entities in the sample are selected randomly from the population, and that the fixed effect model suits better when the entities constitute the entire population.

Eventually, in both models the estimators should converge to the true parameter values of in large samples. However, if any correlation exists, the random effects model gives inconsistent estimators, while the fixed effects model still produces consistent estimators. (Carter-Hill, 2011). By the means of a Hausman test, I formally check for the applicability of the random or fixed effects model. The test takes into account the correlation between the error component and the regressors in the random effects model. Therefore the test compares the coefficient estimates of the random effects model with the fixed effects model. The null hypothesis of no correlation should be rejected if indicates a value lower than 5%, resulting in inefficient estimators from the random effects model. Accordingly, if the Hausman test is rejected, I use a fixed effects model. The Hausman statistics are presented together with the test results of the relevant model. Finally, I included year dummies in all models to take into account possible autocorrelation between the years.

4.3.3. Instrumental variable – 2SLS

Theoretical evidence supports a possible endogenous relationship between R&D expenditures and equity financing (Seifert and Gonenc, 2012). This means that the total of resources spent on R&D is determined by firms’ financing, but R&D also influences firms’ financing. The effect of financing on R&D may occur for example if a firm is financed heavily with debt which does not allow for additional R&D investments to be made due to the risky nature of R&D investments or because potential lenders approval to invest is required. Another example, Jensen and Showalter (2004) show

5_{Assumptions of the new cross-sectional error term are: zero mean, independent of the individual observation error term, has a constant}

(27)

a model in which the amount of leverage affects total R&D expenditures. They argue and empirically show that in a patent race, debt reduces the amount of R&D expenditures. In the presence of endogeneity, results from the OLS and panel Tobit (to be discussed hereinafter) models might yield biased and inconsistent results (Brooks, 2008). In order to deal with two way causality, the standard ordinary least square models are modestly changed.

Instrumental variables can be used to address problems with omitted variables which are correlated with but are unobserved (not included in the regression), and to handle simultaneous causality bias problems. To serve as a valid instrument, the instrument must be exogenous (( ( ) ) and must be correlated with the endogenous explanatory variable ( ( ) For a strong instrument, there should be strong theoretical underpinning that the instrument is uncorrelated with the error term of the regression. Testing the validity of instruments is hard. Testing this is not possible with one instrumental variable since there are no unbiased estimators for . We can however test several other conditions of the instrument. In the first place, we can test the covariance between the instrument and the explanatory variable using the following condition: ( ) . Secondly, in the first stage regression, the instrument should have a significant predicted estimator.

Because I use more than 1 instrument in this research, I use a Sargan statistic for over identifying the restrictions. This test examines the validity of the instruments by excluding instruments. The joint null hypothesis of the test is that the instruments are uncorrelated with the error term and that the excluded instruments are correctly excluded from the estimated equation. The Sargan test obtains the residuals from the IV-2SLS and regresses residuals on all exogenous variables. The test statistic is calculated as , whereby q indicates the number of instrumental variables minus the number of endogenous explanatory variables. A rejection of the test casts doubt on the validity of the instruments. Hansen extended the Sargan test and made it consistent for heteroskedasticity. Because my tests are robust for heteroskedastistic standard errors, I use Hansen’s J statistic to test the validity of the instrumental variables. I add the results to the relevant tables in the results chapter.

With 2SLS – IV, all exogenous regressors are regressed on using OLS and the predicted values of (indicated as ) are computed. In the second stage, is regressed on all exogenous variables and the fitted values of the endogenous variable ( ̂). The first stage regression results are available upon request.

(28)

problem. In order to test for the presence of endogeneity, I perform a Durbin-Wu-Hausman test. Using OLS, the fitted values of the reduced form equations are added and tested as additional explanatory variables in the IV-2SLS models. Additionally, the Durbin-Wu-Hausman test examines whether the coefficient of the fitted values are significantly different from zero and are robust to violations of conditional homoscedasticity. If the null hypothesis is rejected, the use of IV-2SLS is justified. Because the four models have different dependent variables, the presence of endogeneity depends on the model specification.

4.3.4. Limited dependent variable

I also test the models with limited dependent variable methods, namely a panel Tobit model. Tobin (1958) introduced this method to solve problems with truncated and censored data. A Tobit model can be used when limited dependent variables are used of which the values are roughly continuous over strictly positive values but is zero for a nontrivial fraction of the population (Wooldridge, 2002). In my dataset, the value of the dependent variable should be between 0 and 1 since the number consists of a ratio value, the amount of financing of a certain kind compared to the total financing of the firm. Besides, for at least three of the four models, the dependent variable contains a lot of zero values. The zero values represent the absence of a certain kind of financing method within a company. A panel Tobit model is appropriate therefore and in my thesis functions as a robustness test. The latent underlying regression for the relevant financing method is:

With the observed dependent variable such that:

if , and if

represents the ratio of that relevant method of financing compared with the total financing of the

firm i, at year t. differs across cross/sectional units to capture firms specific effects, similar to the ordinary least squares system.

(29)

4.3.5. Other robustness tests

Further, I perform robustness checks under standard least squares regressions for two different time frames and for firms having a total assets value of more than 250 million dollar in a particular year. The first time frame is selected from 1971–1998, which means that this time frame only includes US firms, since I only have data from German firms in the period 1999–2010. The second time frame consists of all firms in the period from 1999–2010 and therefore includes all German and US firms for which I have data available in that time frame. Finally, the last robustness check consists of all firms having more than 250 million dollar of total assets. Since larger companies usually attract more different sorts of financing, this test might give better insight in financing of larger firms.

4.4. Variables

4.4.1. Dependent variables

To measure the impact of R&D investments on each of the four financing instruments in the four models, this study divides financing into four different groups. These groups are selected in accordance with the model set up in the literature section and divide the instruments based on intervention barrier and the appropriation discrepancy characteristics. The dependent and independent variables, which are used in the various models, are described based on CompuStat descriptions of the items.

(30)

Model 3 uses the ratio relational debt over total financing (relational debt / total financing) as a dependent variable. This variable includes bank loans and capitalized lease obligations. For US firms, this is a summation of CompuStat items other debt (dlto) and capitalized lease obligations (dlco). These items in total include all firms’ relationship lending. For Germany, the sample includes all debt obligations towards banks, long-term obligations from capitalized leases and debt contracts from private sponsors. These items are hand-collected from annual reports.

Model 4 uses the ratio of convertible securities over total financing (convertible debt/ total financing), because of similar characteristics of preferred stock (pstk) and convertible debt (dcvt). These items are summed and used for the calculation of total value of convertible securities. This is in line with Thornhill and Wang (2010).

4.4.2. Independent variable

All models comprise one independent variable. R&D investments are measured by R&D expenses over total sales (xrd/sales). This measurement of R&D encompasses especially fundamental research for new products. For example, it excludes R&D related to improve the quality of existing products. Because fundamental research will in essence bear the more risk than developmental R&D, this is a good measure for R&D investments.

The CompuStat item xrd represents all company’s costs incurred during the year that relate to the development of new products or services. The item excludes software expenses and amortization due to software costs. Furthermore, the item excludes customer government-sponsored R&D of which reimbursable indirect costs, extractive industry activities, such as prospecting, acquisition of mineral rights, drilling and mining. It also excludes engineering expenses which are routine, ongoing efforts to define, enrich, or improve the qualities of existing products and marketing related research expenses. Due to possible endogeneity between financing and R&D, I use instrumental variables for R&D. I control for endogeneity by using two lags of R&D. The use of lags of a certain variable is a frequently used manner to avoid endogeneity problems. Although serial correlation can be a concern when using lags of an endogenous variable as an instrument, the overidentification tests help confirm that the lags of R&D were indeed exogenous. In addition, I notice that R&D remains stable over years, which is also confirmed by high correlations with lags of R&D (L1.R&D 0.89 and L2.R&D 0.84)

4.4.3. Control Variables

(31)

Internal funding. Additionally, I control for macro-economic developments using Gdp growth, which reflects the annual percentage growth of the gross-domestic product.6

Only firms with total assets above 100 million dollars are included in the sample, which indicates that only big firms with certain market power are included. The inclusion of Size in the control variables is mainly based on the findings of Titman and Wessels, 1988; Rajan and Zingales, 1995; Frank and Goyal, 2003) who argue that large firms are relatively more diversified and less prone to bankruptcy. This suggests that large firms will be higher leveraged. Further, the costs of issuing debt and equity securities are related to firm size because larger firms have easier access to the capital markets, and borrow could therefore at more favorable interest rates (Ferri and Jones, 1979). This variable is the natural logarithm of total assets in millions of dollars (at).

Growth rate positively influences investment flexibility of firms (Titman and Wessels, 1988; Balakrishnan and Fox, 1993). Equity controlled firms have a tendency to invest sub-optimally to expropriate wealth of firm’s bondholders. Firms active in growing industries are likely to have higher agency costs due to higher flexibility in investment choices. Therefore, firms with higher growth rates are likely to have more equity than debt financing. Thus, growth rates are negatively related to leverage. Myers noted that the problem would be mitigated by the use of short term debt. Besides Jensen and Meckling (1976), Warner (1977) and Green (1984) argued that agency costs are negatively related to convertible debt usage, which suggests a positive relationship between convertible securities and growth rate. Growth rate is measured by (Salest–Salest-1)/(Salest-1), where Salest and Salest-1 represent the firm’s sales in year t and t-1.

To account for the impact of debt financing, Non-debt tax shield is included. Debt can be used to avoid paying tax due to interest deductibility (Titman and Wessels, 1988). De Angelo and Masulis (1980) have argued that tax deductions for depreciation and investment tax credits are substitutes for the tax benefits of debt financing. Consequently, debt is negatively related to the non-debt tax shield relative to the expected cash flows. The variable is defined by the ratio of depreciation and amortization (dp) divided by total assets (at).

Collateral influences the access to debt capital. Titman and Wessels (1988) argue a positive relationship between collateral and leverage. Myers and Majluf (1984) put forward that there may be unobserved associated costs related with issuing securities when there is an agency problem between management and investors. Assets with known values avoid agency problems and associated costs. Eventually, firms with assets ready for collateral are expected to take advantage of the possibilities to

6_{In contrast to Thornhill and Wang (2010), I do not include an ownership variable in the model because I was not able to obtain ownership}

The influence of R&D investments on firms’ financing Evidence from the US and Germany