Firm innovation and covenant tightness

Abstract

This study explores the association between firm innovation and loan covenant strictness. We find that lenders construct stricter contracts for firms filing more patents, consistent with lenders imposing more oversight on firms when they enter the commercialization stage after having demonstrated their inventiveness. Our results hold under propensity score matching and entropy balancing, and when exploiting the American Inventors Protection Act as a shock affecting unrelated banks’ access to patent filing information. The relationship we document is stronger when the lender has more expertise and for firms with higher default risk. We demonstrate that borrowers’ patent filings are associated with more future R&D and capital investment and with a higher likelihood of their acquiring firms in the industry of their patent filings. Our results are consistent with the theoretical prediction that lenders interpret patent filings as indicative of high inventive potential that requires stricter discipline and oversight by lenders in order to be converted into actual business success, and with them designing debt contract terms accordingly.

1 Introduction

A large economics literature going back to at least the 1940s identifies technological innovation as a primary source of productivity gains and economic growth (Schumpeter 1942; Solow 1957). Much of the economic growth in the United States and Europe can be attributed to significant innovations by established corporations (e.g., Baumol 2002; Hyytinen and Maliranta 2013). Given the vital economic role of innovative firms, it is critical to understand their access to bank debt financing and the extent to which their innovation can be impeded by capital market frictions. This issue is salient, given that banks are playing an increasingly important role in both financing the firms engaging in innovation (Houston and James 1996; Johnson 1997; Kerr and Nanda 2015) and shaping corporate innovation policies (e.g., Mazzucato and Semieniuk 2018; Deleidi et al. 2020; Geddes and Schmidt 2020).

Prior studies investigating how debtholders respond to firm R&D activities report mixed results. However, recent studies focusing on firms’ patents document a negative association between patents and loan prices (e.g., Plumlee et al. 2015). This finding is consistent with patents being viewed as a sign of technological success, which increases the likelihood of future profitability, and also with patents being valuable as collateral. However, these results are incomplete because patents at best signify interim successes of the innovation process. To turn these interim successes into viable commercial products, additional and often substantial effort and resources are typically required. If banks were to favorably price patents without constraints, we would observe cycles of debt financing leading to patents—and further debt financing followed by yet more patents. Given that we do not empirically observe such cycles and, in fact, observe that firms with more patents typically have lower leverage than their industry peers with fewer patents, it is likely that researchers may be overlooking important constraints embedded into lending contracts. Our study attempts to fill this gap.

Patents can serve as a reliable sign of successful idea generation, which is best thought of as the first stage of a multi-stage process of innovation, and therefore they represent only one of the necessary inputs. The ultimate success of the process requires that the subsequent stage—implementation—be carried out in a focused and disciplined manner. This is so because effective implementation critically depends on high-quality managerial decisions that include coordinating the efforts of all parties and making appropriate future capital investments. Because successful idea generation at the first stage enlarges the managerial action space—and, in particular, the range of inappropriate actions that could be taken by the insiders (e.g., Gabler and Poschke 2013; Castro et al. 2015; Caggese 2019)—it increases the importance of stricter disciplining devices at the implementation stage. Accordingly, banks should design debt contracts that facilitate idea generation yet provide stronger incentives to carry out their implementation effectively and efficiently.

The literature has long argued that covenants can play an important role in mitigating information asymmetry and agency problems in debt contracting (Jensen and Meckling 1976; Smith and Warner 1979; Smith 1993). Empirical studies have identified debt covenant strictness (or tightness, as it is often referred to) as a unique characteristic of debt contracts, distinct from other contractual provisions such as the price of debt (e.g., Murfin 2012; Prilmeier 2017), and have suggested that the primary purpose of covenant strictness is to discipline borrowing firms’ behavior (e.g., Chava and Roberts 2008). The theoretical arguments explaining the unique role of debt covenant strictness start with the observation that because the contracting parties face uncertainty, which is always present when firms are engaged in genuine innovation, they cannot specify all possible future states of the world in advance. Thus, the large incomplete contracting literature emphasizes the reality that formal contracts—debt contracts in our setting—are always incomplete.

A recent extension of this literature highlights that because the success of innovation requires that contracting parties make optimal decisions contingent upon the information that emerges at interim stages (e.g., Gibbons and Henderson 2012), they can improve the efficiency of their relationship if they supplement the formal contract with an informal one supported by the value of the future relationship, usually termed a relational contract. This arises because relational contracts can be based on information that is available to contracting parties even if it is not verifiable in the court of law (Fuchs 2007; Levin 2003). Strict debt covenants, which constitute a part of the formal contract, compel the parties to renegotiate their relational—i.e., informal—contract based on relevant new information. When each of the parties performs in a mutually agreeable manner, covenant violations are waived and the relationship continues (Watson et al. 2020; Kostadinov 2021). In other words, covenants play a unique role in debt contracting because they increase the self-enforcing range of the relational contract and thereby improve the efficiency of a contractual relationship. This off-equilibrium role of covenant strictness stands in contrast with other contractual provisions, such as the price of debt, that are binding in equilibrium. Building upon the theoretical results outlined above, we argue that stricter covenants serve a distinct function of focusing firm managers’ attention on carrying out the implementation stage of the innovation process in an efficient manner. In other words, tighter covenants allow lenders to provide a valuable nontrivial input into the innovation process even though they may lack the requisite technical expertise to monitor the implementation process directly.¹

In this study, we explore the association between borrowers’ patent filings and loan contract design—specifically the loan covenant strictness. We find that banks issue contracts with higher covenant strictness for firms filing more patents. The effect we document is economically important: a one-standard-deviation increase in Patent Filings (the natural log of a borrower’s patent filings plus one) is associated with a 0.014 increase in the covenant strictness, which translates into approximately a 4.70% increase compared with the average covenant strictness.

We conduct several robustness tests with different alternative empirical specifications and identification strategies to mitigate potential endogeneity concerns. First, in order to mitigate the concern of endogenous matching, we perform propensity score matching and entropy balancing by controlling for the known difference between firms with higher and lower innovation outputs. Second, we use the American Inventors Protection Act (AIPA) as a shock to the availability of patent filing information to unrelated banks, and we test the difference in reactions between related and unrelated banks to firms’ patent filings before and after the AIPA. These results reinforce our findings in the baseline model and provide further support for a causal relation between patents and covenant strictness. These various tests impose an empirical structure that reduces the likelihood that our results are purely driven by selection bias, client characteristics, or omitted variables.

We also conduct several cross-sectional tests to see whether the effect varies across different banks or firms. We predict that our observed effect will be stronger when banks have more relevant expertise or when borrowers’ default risk is higher and, therefore, the disciplining effect of stricter covenants is more valuable. Consistent with these predictions, we find that our results are stronger when there is an “industry-expert” or “tech-expert” bank in the loan syndicate or when at least one bank in the syndicate previously lent to the borrower. We also find that our results are stronger when borrowers have higher default risk.

We next consider the simultaneous determination and substitution effect between covenant strictness and loan spread. First, we explicitly examine spread-covenant combinations for firms in different stages of innovation. We find that firms are more likely to have [high spread, low covenant] in the innovation stage (first stage) and [low spread, high covenant] in the commercialization stage (second stage), supporting the argument that lenders give more leeway to firms in the first stage of innovation while imposing tighter monitoring to help facilitate implementation of technological success in the second stage of innovation. Second, we consider the substitution effect between spread and covenant strictness. Consistent with the intuition in the prior literature (e.g., Gigler et al. 2009), we find that a substitution effect between covenant strictness and loan spread does exist in our sample, but our main findings are robust to controlling for it. Third, we consider the simultaneous determination of spreads and covenant strictness by incorporating seemingly unrelated regression (SUR), two-stage least squares (2SLS), and three-stage least squares (3SLS) analyses. Our main results hold. Overall, these findings are consistent with our argument that covenant strictness and loan spreads serve distinct purposes and that banks intentionally choose different combinations of these loan terms based on the innovation stage of a borrower.

We next explore potential channels. We find that patent filings are associated with higher future R&D, capital investment, and M&A activities in the industry of the filed patents. These results show that, as argued above, demonstrable inventiveness is not the end of firm innovation, because the ultimate success critically depends on the commercialization stage, which is subject to managerial discretion and thus brings about incremental uncertainty for the lenders. These results are consistent with the theoretical predictions, derived under a wide variety of modeling specifications (e.g., Rantakari 2012, 2016; Hörner and Samuelson 2013; Gomes et al. 2016; Guo 2016; Halac et al. 2016; Henry and Ottaviani 2019; Zambrano 2019), that creative idea generation increases the need for effective disciplining mechanisms to translate the ideas into business success and that banks are aware of this need and consider it when designing debt contracts.

Our study makes several contributions to the literature. First, we contribute to the body of research related to the determinants of debt contract strictness (e.g., Murfin 2012; Demerjian and Owens 2016; Carrizosa and Ryan 2017) and its relation to other characteristics of debt contracts such as pricing, maturity, and the frequency of contract renegotiation (e.g., Smith and Warner 1979; Gârleanu and Zwiebel 2009; Roberts 2015). In particular, our study emphasizes an important conceptual distinction between covenant strictness—best understood as an off-equilibrium threat intended to be negotiated away if both parties are acting in accordance with their relational (i.e., informal) contract—and loan pricing, which characterizes the contracting parties’ actions on the equilibrium path according to their formal contract.

Further, while prior studies focus on creditors’ capital positions (e.g., Murfin 2012), our study provides a new perspective on the nature of lender screening that is more in line with the emerging literature on the role of financial institutions in shaping corporate innovation (e.g., Mazzucato and Semieniuk 2018; Geddes and Schmidt 2020). Our results provide evidence consistent with lenders viewing patent applications positively, as a sign of technological success, but also as but an interim input into a longer and often continuous innovation process. Thus, lenders increase covenant strictness to protect themselves from the downside risks inherent in the commercialization stage. Our study highlights the importance of considering both stages of the innovation process—idea generation and implementation—when studying the design of financing arrangements, because each of the two stages is subject to its own underlying logic and thus requires its own governance mechanisms.

Last, in recent years, a new literature has emerged investigating the link between debt financing and innovation (see, e.g., Acharya and Subramanian 2009; Amable et al. 2010; Loumioti 2012; Grilli et al. 2018).² Mann (2018) provides evidence that innovative firms obtain loans more frequently and that patents are often used as collateral. Similarly, Hochberg et al. (2018) report that patents are used as collateral for debt, and Robb and Robinson (2014) document that bank financing is a significant source of startup capital.³ Whether and how debtholders price firms’ inventions and internally generated intangible assets is a timely, relevant issue that calls for empirical investigation, especially because a firm’s long-term success and survival largely depend on its innovation competitiveness in the knowledge-based economy.

However, no previous studies have examined the impact of patents on non-price terms of loan contracts. This study provides a new angle to understand how banks value firm patenting success, and we contribute to the understanding of the link between debt financing and innovation. We shed light on this issue by showing that banks respond to patenting success with more intensive and focused disciplining mechanisms in the form of stricter debt covenants.⁴ Our study extends prior work in this area (Plumlee et al. 2015) by providing a fuller picture of how banks set contract terms for innovative firms, and our findings partially explain why innovative firms generally prefer lower leverage (Titman and Wessels 1988).

2 Hypothesis development

2.1 Financing the innovation process

Innovation has long been documented as the driving force for economic development and firm growth, and it can increase gross output and revolutionize economic structure (Schumpeter 1942; Solow 1957). Commentators have also suggested that innovation produces new business opportunities and leads to productivity growth (Abramovitz 1956; Kirzner 1997). A large body of research suggests that better access to financing can stimulate firm innovation and further promote economic growth (Benfratello et al. 2008; Brown et al. 2009; Amore et al. 2013; Chava et al. 2013b; Grilli et al. 2018). However, researchers in economics and finance have long argued that innovative activities are difficult to finance in a freely competitive market (see, e.g., Hall (2005) for a review of this literature) because innovation creates wide information gaps between insiders and capital markets owing to its idiosyncratic (Holmström 1989), intangible (Eberhart et al. 2008), and confidential (Bhattacharya and Ritter 1983; Carpenter and Petersen 2002) nature.

There is also an ongoing debate over the merits and demerits of various financing channels in supporting firm innovation. The proponents of equity financing argue that deep uncertainty characterizing genuine innovation and the lack of collateral makes equity financing a better choice for innovative firms (Brown et al. 2009). Moreover, international evidence suggests that innovation-intensive industries come up with higher innovation levels in countries with more developed equity markets (Hsu et al. 2014). However, the proponents of debt financing counter that financial intermediaries can better motivate and monitor firms through contract design and thus further promote firm innovation (De la Fuente and Marin 1996). Additionally, banks can ensure that corporate innovation projects adhere to borrowers’ core business activities and increase firm value by impeding empire building (Gu et al. 2017).

A related line of research maintains that, as creditors, banks lack motivation to support aggressive innovation associated with high returns and high risk because they are only entitled to a fixed payment and therefore focus their attention on downside risks. Bergemann and Hege (2005) show that firms with risky investments prefer arm’s-length financing. Additionally, Encaoua et al. (2000) and Atanassov et al. (2007) argue that the banking system is generally unsuitable to support innovation, owing to a shortage of adequate instruments by which to evaluate the quality of innovative projects. Prior studies on debtholders’ reactions to firms’ R&D activities find mixed results (Shi 2003; Eberhart et al. 2008; Liano 2013). Some studies suggest that the positive effects of innovation on firms’ earnings dominate the negative effects, such as a lower liquidation value for intellectual capital (Liu and Wong 2011). On the other hand, Hsu et al. (2015) find that bondholders can recognize the value of innovation and ask for lower interest spreads for innovation-competitive firms.

Patent records are often used to measure firms’ innovation performance. Because patents represent exclusive rights to use certain knowledge in a technologically competitive economy and reflect firms’ intangible intellectual assets and market prospects, patent records provide useful information about the output of corporate innovation. An additional advantage is that patent information is not directly subject to accounting manipulation for short-term financial reporting purposes. Further, patents are valuable and can potentially be sold off in the event of default. Recent studies focusing on firms’ patents generally find that debtholders favorably price patents. In particular, Francis et al. (2012a) demonstrate that borrowers with higher innovation capability enjoy lower bank loan spreads and fewer collateral requirements, and Hsu et al. (2015) find that the quantity, impact, originality, and generality of a firm’s patent portfolio are negatively related to bond premiums. Additionally, Plumlee et al. (2015) find that banks charge borrowers with forthcoming patents a lower spread, and Chava et al. (2017) show that an exogenous enhancement in the value of borrowers’ patents, either through greater patent protection or creditor rights over collateral, results in cheaper loans.

However, the results published to date raise additional questions because the ultimate outcome of the innovation process is contingent upon its successful implementation, which, in general, requires both appropriate managerial effort and subsequent financing. If banks were to price patents as if successful implementation was irrelevant, we would observe a spiraling pattern where more patents lead to more bank financing, engendering even more patents and more bank financing. Given that this is not the case in reality, the implication is that research to date may have overlooked some important elements within debt contracts.

2.2 From technological success to business success

We argue that being successful at generating creative ideas does not, in and of itself, guarantee their successful implementation and future business success. The reason is that each of the activities—idea generation and idea implementation—is subject to its own underlying logic and prone to its own set of pitfalls, and accordingly requires its own governance mechanisms (e.g., Ding and Eliashberg 2002; Loch and Kavadias 2008; Baer 2012). The fundamental differences between the two activities have even prompted some commentators to suggest that they should be intentionally separated in order to avoid the “contamination” of the idea-generation phase by the practical considerations typical of the implementation phase (e.g., Puranam et al. 2006; Andriopoulos and Lewis 2009; Fang et al. 2010). The distinctive characteristics of the appropriate governance mechanisms are best seen in terms of binding versus slack constraints, as we argue below.

The first activity—generating candidate solutions—requires that participants come up with a multitude of ideas and discuss their advantages and disadvantages in a frank manner over the course of no-holds-barred argumentation. Binding resource constraints limit both the range of proposed ideas and the intensity of discussions and thus have an overall detrimental effect on the number and quality of candidate solutions proposed (Schneider and Veugelers 2010; Chava et al. 2017; García-Quevedo et al. 2018; Grilli et al. 2018). Thus, the generation of candidate solutions works best in an unstructured environment with as few binding constraints as possible.

In contrast, the second stage of the innovation process—implementation—requires a disciplined and calculated approach and thus works best in structured environments, with the assignment of clear goals and responsibilities that serve as incentive constraints that are binding at the optimum. This is so for the following reasons.

• 1. Because creative ideas increase information asymmetry between investors and firm managers, they exacerbate the severity of agency problems that can arise at the implementation stage. Such problems include suboptimal capital investment decisions (e.g., Gale and Hellwig 1985; Aghion and Bolton 1992) and misappropriation of the returns to innovative activities by the employees directly involved in them (e.g., Coff 1999; Chadwick 2017).

• 2. Managing innovative activities requires substantial cognitive efforts: the managers should be able to focus their attention on unfamiliar technological developments (Eggers and Kaplan 2009; Csaszar and Levinthal 2016); construct mental models required to navigate the emerging technological landscapes and communicate them to the employees (Gavetti et al. 2005; Thrane et al. 2010); and then implement and continually maintain in working condition organizational policies and routines that enable the integration of knowledge generated in different locations (Van den Bosch et al. 1999; Lenox and King 2004; Gardner et al. 2012; Aggarwal et al. 2017).

• 3. Decision-makers—especially those directly involved in conducting research—often find it difficult to terminate unsuccessful projects to which they become emotionally attached, when such termination is warranted (Adner and Levinthal 2004; Guler 2007).

Therefore, to translate the ideas generated in the first stage into commercial success, the firm has to find adequate means of countering the problems just described.

2.3 Lender options: Price and non-price loan terms

Owing to the sheer number of actions that should be taken and the level of technical expertise required to assess the extent to which they are appropriate, direct, step-by-step monitoring of the implementation stage by creditors is impracticable even if they have access to confidential information. Nonetheless, banks have access to private information in a form that they can interpret and use in contracting, both formal and informal (e.g., Hadlock and James 2002; Bharath et al. 2008; Beatty et al. 2010). Specifically, banks can use both the price and non-price terms of bank loans to deal with borrower risk.

The logic underpinning our empirical predictions follows from two streams of work, one primarily empirical and one theoretical. The first stream argues that banks use debt covenants as a relatively inexpensive and more practicable form of monitoring that does not require technical expertise (e.g., Dichev and Skinner 2002; Frankel et al. 2008; Chava et al. 2010; Demerjian 2017). If a borrower violates a covenant, it finds itself in “technical default” and the loan contract is usually renegotiated but can, in rare cases, be terminated (e.g., Chen and Wei 1993; Beneish and Press 1995; Chava and Roberts 2008). Covenant slack, which is defined as the difference between the required threshold value and the actual value of the covenant measure, is considered an ex post proxy for borrower riskiness or the degree of agency conflicts. Demiroglu and James (2010) argue that loans to observationally riskier firms may also have stricter covenants because stricter covenants provide lenders the option to reassess the loan and take action for even modest deteriorations in performance. Even though, as noted above, a bank would find it difficult to directly monitor the effectiveness of managerial actions at the implementation stage of the innovation process, it can use covenant violations to reassess the effectiveness of the firm’s governance mechanisms. Consistent with the prediction that covenant strictness serves as a form of monitoring, the literature documents that firm-level governance mechanisms (Ge et al. 2012), board monitoring (Fields et al. 2012; Francis et al. 2012b), government monitoring (Black et al. 2004), and analyst coverage (Francis et al. 2021) are negatively associated with covenant strictness.

Another stream of research, briefly outlined in the introduction, views debt covenants through the lens of incomplete contracting theory. For example, Gârleanu and Zwiebel (2009) analyze a model where the contract allocates the right to make an ex post investment decision to one of the parties and interpret the situation where such a right is allocated to the creditor as consistent with tight debt covenants that are waived. Empirical studies confirm the relevance of the incomplete contracting framework to the study of innovation (e.g., Frésard et al. 2020). Recently, theoretical studies extend the incomplete contracting framework by focusing on the relational—i.e., informal—contracts that complement formal (incomplete) contracts. These studies show that tight covenants can serve as an effective means to enlarge the self-enforcing range of the relational contract (e.g., Watson et al. 2020; Kostadinov 2021). This latter theoretical approach replaces an earlier stylized representation where all decision rights are assigned to only one contracting party with a more nuanced setting where the relational agreement requires ongoing inputs from both contracting parties.⁵ Thus, the lending agreement continues on the equilibrium path, while tight covenants are used as an off-equilibrium threat (that is, they are never binding in equilibrium). This representation is also consistent with the empirical studies documenting the importance of relational contracts for successful innovation (e.g., Gibbons and Henderson 2012; Lee 2018).

It follows from both streams of research discussed above that loan interest spreads (i.e., price) and stricter covenants are conceptually distinct and, thus, can be used to address the two different stages of the innovation process independently. Whereas the former is designed to facilitate the generation of new ideas, the latter is intended to improve the efficiency of their implementation. Viewed through this lens, lower loan prices and stricter covenants are in fact complementary tools that are designed to exert a joint positive effect on the innovation process. Indeed, the more innovative are the ideas generated at the first stage, the more difficult they are to evaluate and implement (Danneels and Kleinschmidtb 2001; Criscuolo et al. 2017) and, thus, the more important it is to have effective disciplining mechanisms in place at the second stage of the innovation process. It follows that when the quality of the firm’s creative team increases for exogenous reasons, the equilibrium response is a further lowering of the price of the loan accompanied by the tightening of the covenants.

Formally, we state our hypothesis as follows:

• H1: Patent filings are associated with tighter debt covenants.

3 Research design and sample selection

3.1 Research design

In order to test H1, we estimate the following model:

$${Covenant\ Strictness}_i={\beta}_0+{\beta}_1{Patent\ Filings}_i+{\beta}_2{Controls}_i+{\epsilon}_i$$

(1)

where Covenant Strictness is provided by Demerjian and Owens (2016), who measure covenant strictness as the simulated probability of violating at least one covenant during the quarter after loan initiation. Specifically, the financial measures underlying each covenant are simulated and then the probability of covenant violation is estimated by the frequency of violation instances generated by those forecasts.⁶Patent Filings is the natural log of one plus the number of patents filed in the year before the facility starts. We include current loan contract terms and borrower characteristics in the year prior to the facility issuance as control variables in our analyses. We also include a variety of fixed effects specifications, including year, loan type, loan purpose, and industry.

3.2 Data sources and sample selection

We obtain patent data from Stoffman et al. (2022).⁷ This patent data includes patents where the authors have matched the assignees to firms in CRSP over the period from 1926 to 2017. Thus, we limit our sample to those observations with non-missing Compustat data and non-duplicate PERMCOs, matched through the linking table provided by WRDS. The firm-level data comes from Compustat, and the loan-level data comes from Dealscan. By using the Compustat-Dealscan link provided by Chava and Roberts (2008), we are able to match firm-level variables to facility borrowers. We use facilities as the unit of analysis because we examine covenants as one of the tools that lenders use from a basket of loan terms, with loan spread being the primary tool lenders use to compensate for default risk. As we emphasize above, covenants and interest spreads serve different purposes, so, to be consistent with our arguments throughout the paper, we conduct our empirical tests at the loan facility level.⁸ We limit our loan facility sample to loans issued to US-domiciled non-financial firms (SIC code outside of 6000 to 6999) over the period from 1995 to 2016. We drop those facilities where we cannot identify a lead arranger (bank). In addition, we focus our analysis on the loan facilities of borrowers with at least one patent filed within 10 years before a loan issuance. Therefore, if a firm does not file any patents in the previous 10 years, then it does not enter our sample and is not included as a control firm. We include this restriction because we are interested in comparing borrowers in the commercialization stage to those in the innovation stage, conditional on borrowers being involved in the innovation process. We are less interested in comparing innovative firms to non-innovative firms.⁹ After imposing the requirement of non-missing data related to the calculation of Strictness and control variables, we have 7200 observations in our final sample.

3.3 Descriptive statistics

Table 1 reports the descriptive statistics for our sample. The sample average of covenant strictness (probability of covenant violation) is 0.298. Averaged loan spread is 146.38 basis points over LIBOR, loan size is $147.59 M, loan maturity is 40.85 months, the number of participants is 9.836, and covenant index is 2.418. Of the loans in our sample, 40.6% are from unrelated banks, and 67.3% include a performance pricing provision. All variables are defined in detail in Appendix Table 11.

Table 1 Summary descriptive statistics

Table 2 reports the Pearson correlation coefficients for our sample variables. Many of the control variables are significantly correlated. In particular, Covenant Strictness is positively correlated with Spread, Unrelated, Covindex, Lev, CFO volatility, and Rating, and negatively correlated with Amount, Maturity, Nbanks, Provision, R&D, Size, ROA, MTB, Current, and Z-score.

Table 2 Pearson correlation matrix

4 Empirical results

4.1 Patent filings and debt covenant strictness

Table 3 presents the results of estimating Eq. (1), with Column 1 dropping loan spread as a control variable and Column 3 adding lender fixed effects. Consistent with H1, the coefficients are positive and significant across all specifications, suggesting a positive association between Patent Filings and covenant strictness. In particular, this association holds when controlling for loan spread, and our results are not driven by lender invariant preferences. In Column 2, our main specification, which includes all control variables and year (facility year), loan purpose, loan type, and industry (Fama-French 48 industry) fixed effects, the coefficient on Patent Filings is 0.009 and statistically significant. In terms of economic significance, a one-standard-deviation increase in Patent Filings is associated with a 0.014 increase in the covenant strictness, which translates into approximately a 4.70% increase compared with the average covenant strictness.

Table 3 Patent filings and debt covenant strictness

The coefficients on many of the control variables have the expected signs and are statistically significant. Covenant Tightness is negatively associated with number of banks, ROA, and current ratio and positively associated with unrelated banks, covenant index, leverage ratio, CFO volatility, and a borrower’s credit rating.

Taken together, the results in Table 3 provide evidence consistent with our prediction in H1 that patent filings are associated with stricter debt covenants.

4.2 Endogeneity concerns

Despite the robustness of our findings in Table 3, our main results may still be subject to specification and endogeneity concerns. Specifically, one potential concern is that our results are driven by differences in the characteristics of firms with high patent filings and low patent filings. That is, firms with many patents might possess characteristics that are different from those of the firms with fewer patents, and this may affect the covenant strictness in their loans. We adopt propensity score matching and entropy balancing in order to mitigate this concern. Furthermore, we also use the AIPA as a shock to the availability of patent filing information for unrelated banks to further mitigate endogeneity concerns.

We use propensity score matching (PSM) to construct a matched sample using a nearest-neighbor propensity score match, with the scores given by a logit model in which the dependent variable High Patent is an indicator variable that takes a value of one (zero) if Patent Filings is (not) in the highest tercile of all facilities in a given year. In the PSM model, we include all firm-level control variables from the baseline model of Eq. (1) as well as industry and year fixed effects, and we use the predicted probabilities of the model as the propensity score. Additionally, we require no replacement for the control sample selection and the maximum distance of matched propensity score to be 0.01.

Table 4 presents the results of the propensity score matching analysis. Specifically, Panel A reports parameters estimated from the logit model used in estimating the propensity scores for high versus low patent filings, and Panel B presents the efficiency of the propensity score matching process. While the differences between the two groups are significant for many variables before matching (Column 3), they become insignificant after the matching process in the multivariate comparisons (Column 6), suggesting that our matching process is efficient. Panel C of Table 4 presents the regression results of Eq. (1) using the matched sample. We use the indicator variable High Patent in Column 1 and Patent Filings in Column 2. The variables of interest are significantly positive in both specifications, leading to similar inferences as in our main test (Table 3). Together these PSM results further suggest that our results are unlikely to be driven by the different characteristics of firms with high versus low patents.

Table 4 Propensity score matching for high patent and low patent firms

We next apply entropy balancing (EB) for High Patent and Low Patent firms. Entropy balancing helps to obtain a high degree of covariate balance (e.g., imposing constraints in adjusting the first, second, and even higher moments of the covariate distributions) on firm characteristics and, thus, generates well-balanced samples where High Patent and Low Patent firms are not significantly different on these characteristics.

Table 5 presents the results of the EB method. Panel A illustrates the efficiency of entropy balancing, showing that the first, second, and third moments between the High Patent and the Low Patent firms do not greatly vary after balancing. Panel B presents the regression results of estimating Eq. (1) using the EB sample. Consistent with our main findings in Table 3, the coefficients are significantly positive, as predicted, for both the indicator variable High Patent and Patent Filings.

Table 5 Entropy balancing for high patent firms and low patent firms

Finally, we use the AIPA as a shock to the availability of patent filing information only for unrelated banks. The AIPA was enacted on November 29, 2000, and represents an expansion of firms’ patent disclosure requirements by increasing both the timeliness and scope of patent disclosures (Hegde and Luo 2018). Before the AIPA, only US-granted patents were disclosed at the time of grant. Since the enactment of the AIPA, firms must typically disclose US patent applications 18 months after filing, regardless of whether the applications are eventually granted or not. Thus, patent filings were private information before the AIPA but became public information after the AIPA. Specifically, unrelated banks did not have access to private patent filing information before the AIPA, but they do have access to that information after the AIPA. However, related banks were always able to obtain borrowers’ patent filing information. Therefore, if banks do react to patent filing information, then we should observe that unrelated banks only react after the AIPA and that related banks react during the whole sample period.

To test how related and unrelated banks react differently to patent filings before and after the AIPA, we use a [−5, 5] window around AIPA. To eliminate ambiguity, we drop year 2001 and define [1996, 2000] as the pre-period and [2002, 2006] as the post-period.¹⁰ We also divide our facility sample into related bank facilities (at least one bank in the loan syndicate lent to the firm in the prior three years) and unrelated bank facilities (no bank lent to the borrower in the prior three years).

We present the results of this analysis in Table 6. Column 1 reports the analysis for related banks. The coefficient on Patent Filings is significantly positive, and the coefficient on Patent Filings*Post AIPA is not significant. This finding is consistent with our argument that related banks considered borrowers’ patent filings before the AIPA, imposing stricter covenants on borrowers with more patent filings, and this reaction did not change after the AIPA because their information availability did not change. Column 2 reports this test for the sample of loans issued by unrelated banks. In contrast to the results in Column 1, the coefficient on Patent Filings is not significant, but the coefficient on Patent Filings*Post is significantly positive. These results are consistent with unrelated banks not having patent filing information before the AIPA and therefore not reacting to firms’ patent filings, but then learning this information after the AIPA and imposing stricter covenants based on firms’ patent filings accordingly. Taken together, these results are consistent with our predictions and reinforce the findings from our baseline model, further strengthening our claims of a causal relation between patent filings and covenant strictness.

Table 6 Exogenous shock: American inventors protection act

Together, the variety of tests we conduct in Tables 4, 5 and 6 serve to mitigate potential endogeneity concerns in our setting. The consistency of our inferences across all our tests increases our confidence in asserting that patent filings lead banks to include stricter debt covenants in loan contracts.

4.3 Patent filings and debt covenant strictness: Cross-sectional tests on lender and borrower characteristics

In order to better understand the relationship between patent filings and covenant tightness, we next explore the effect of a variety of bank and firm characteristics on the relation between patent filings and covenant strictness. In motivating H1, we argue that lenders use stricter covenants in order to discipline and monitor firms with more patent filings, consistent with higher monitoring needs during the commercialization process. Prior studies show that banks are different from other types of investors and that lenders accumulate knowledge and information in the process of lending to borrowers. Banks that are experienced in lending to high-patent borrowers have greater knowledge and expertise related to these types of firms. For example, they may more accurately estimate the value of borrowers’ innovation, and they may possess expertise in how best to exercise control rights (ex., when to cut R&D) in the event of a covenant violation (e.g., Chava et al. 2013a). It follows that lenders with the requisite experience and expertise may better know how to value patent filings and evaluate the related risks. Furthermore, Glode et al. (2012) suggest that financial expertise will lead to higher efficiency in acquiring information. We argue that banks with more expertise and experience are better equipped to understand their monitoring needs during the commercialization stage, so they impose stricter covenants on borrowers with more patent filings.

To investigate the effects of bank experience and expertise, we use three measures: 1) whether there is at least one “industry-expert” lead bank in a syndicate, where “industry-expert” lead banks are banks whose total lending (amount) to a borrower’s industry is in the highest tercile of all lead banks’ lending to that industry in the prior year, 2) whether there is at least one “tech-expert” lead bank in a syndicate, where “tech-expert” lead banks are those whose total number of patents filed by clients (borrowers) in a borrower’s industry is in the highest tercile of all lead banks’ lending to that industry in the prior year, and 3) whether there is at least one lead bank in a syndicate that lent to the firm in the prior three years.

We interact these bank characteristics with Patent Filings, rerun our tests, and report the results in Panel A of Table 7. In all three cases, we observe that the covenant strictness effect is stronger when lenders have more experience and expertise: the interaction term between Patent Filings and the indicator variable of interest is significantly positive in all cases. These results are consistent with banks intentionally viewing covenant strictness as a cost-effective tool for managing the risks associated with commercialization after achieving technological success.

Table 7 Cross-sectional tests – lender expertise and borrower default risk

Furthermore, we argue that lenders will view patent filings with even more caution when a borrower has a higher default risk, strengthening the relation we previously documented. We use three measures to capture default risk: 1) high ROA volatility, 2) high current ratio volatility, and 3) high interest coverage ratio volatility. All the three measures are calculated as the standard deviation of the underlying volatility in the prior three quarters, and we define facilities with high default risk as those whose volatility ratios are in the highest tercile of all facilities in an industry in a given year. We interact these borrower characteristics with Patent Filings and rerun our analysis, reporting the results in Panel B of Table 7. In all cases, the effect of patent filings on covenant strictness is stronger for borrowers with higher default risk.

4.4 Robustness checks

In order to validate that our findings are not driven by specific research design choices, we also provide several robustness tests. First, we expand our sample to (1) all firms with at least one patent filing during our sample period and then to (2) all firms over our sample period, with no restrictions based on patent filing history. The results in Panel A of Table 8 show that our findings are robust to these different sample choices. Second, we rerun our tests at the package level and provide the results in Table 8 Panel B. We estimate our main regression in Column 1, the PSM analysis in Column 2, and the EB analysis in Column 3. All our results hold at the package level. Third, we employ different measures of covenant strictness and report the results in Panel C of Table 8. We first use another widely used covenant strictness measure following Murfin (2012), who uses covenant slack and the covariance of covenants to calculate the probability of covenant violation in Column 1. We then use the covenant tightness measure following Demiroglu and James (2010), who measure covenant tightness using the two most commonly employed debt covenants—debt-to-EBITDA ratio and current ratio—in Column 2.¹¹ Using both alternative measures, we find a positive relationship between patent filings and covenant strictness or tightness. Last, in untabulated tests, we confirm that our results are also robust to different standard error clustering methods (e.g., firm, bank, industry, or year).

Table 8 Robustness tests

4.5 Separating the effect of interest spread from loan covenants

Consistent with Plumlee et al. (2015), who find that banks provide lower interest spreads to firms with patents pending approval, we also find, in our sample, that banks provide lower spreads to borrowers with higher patent filings (untabulated). However, this may raise the concern that patent filings impact loan spreads and covenant strictness simultaneously. In order to consider these two effects together, we first use a multinomial logistic regression method to test how banks choose different combinations of covenant strictness and loan spread in the first stage (innovation) and second stage (commercialization). Specifically, we define firms as being non-innovative if they have zero R&D expense and no patent filings in the prior three years; first stage–focused (Innovation) if they have positive R&D expense but no patent filings in the prior three years; and second stage–focused (Commercialization) if they have positive R&D expense and at least one patent filing in the prior three years.

We then combine the choice of loan spread and covenant strictness into one multinomial choice variable. Specifically, we set annual median-split dichotomous indicators for [high spread, low spread] and [high strictness, low strictness] and classify each loan facility into one of four possible groups: [high spread, high strictness], [high spread, low strictness], [low spread, high strictness], and [low spread, low strictness]. We then run a multinomial logit regression with [high spread, high strictness] and [low spread, low strictness] as the base group. We first compare innovative firms (first stage) with non-innovative firms as the benchmark, then compare commercialization firms (second stage) with innovative firms (first stage) as the benchmark.

Panel A of Table 9 presents the multinomial logistic regression results. Innovation is an indicator variable equal to one if a firm is first stage–focused, and zero if a firm is non-innovative. Commercialization is an indicator variable equal to one for second stage–focused firms, and zero for first stage–focused firms. In Columns 1 and 2 we find that, compared to non-innovative firms, first stage–focused firms are more likely to have the loan contract combination [high spread, low strictness] and less likely to have the combination [low spread, high strictness], with the difference being statistically significant. In Columns 3 and 4, we find that, compared to first stage–focused firms, second stage–focused firms are more likely to have the loan contract combination [low spread, high strictness] and less likely to have the combination [high spread, low strictness], also with the difference being statistically significant. These findings support our argument that lenders give more leeway to firms in the first stage of innovation with less binding covenant constraints and may use higher loan spreads to compensate for innovation risk. However, for borrowers in the second stage, lenders are more likely to impose tighter monitoring to help facilitate implementation of technological success and, in turn, to require a lower interest spread.

Table 9 Interest spread and covenant strictness for innovative firms

In our second test to address simultaneity, we acknowledge that covenant strictness and loan spreads may be not only simultaneously determined but also interdependently determined. Agency theory suggests that there is likely to be a tradeoff between the covenant restrictions imposed by a loan contract and the interest rate (Jensen and Meckling 1976; Myers 1977; Smith and Warner 1979). Notably, Matvos (2013) finds that loan covenants are priced and that borrowers are more likely to include a covenant when doing so allows them to obtain a larger reduction in the interest rate. Therefore, the primary concern is whether and to what extent there is a substitution effect between covenant strictness and spread. Relatedly, do covenant strictness and interest spread perfectly substitute, or do they serve different purposes?

In order to take the substitution effect of covenant strictness and loan spread into consideration, we follow a structural estimation methodology following Lee (1978) and Prilmeier (2017). Specifically, we first create a covenant strictness binary choice model:

$${I}_i=\lambda \left({y}_{lowstrict,i}-{y}_{highstrict,i}\right)+{Z}_i^{\prime}\xi -{v}_i$$

(2)

where I_i equals one if covenant strictness is higher than the annual median, and zero otherwise, and (y_{lowstrict, i} − y_{highstrict, i}) is the increase in the log of the all-in-drawn yield spread by including low strictness covenants instead of high strictness covenants, or, in other words, the covenant strictness price. The coefficient λ captures the extent to which borrowers are more likely to have stricter covenants when the resulting spread increase from low strictness covenants is large. The vector Zi contains other control variables related to covenant strictness.

To estimate the covenant strictness price (y_{lowstrict, i} − y_{highstrict, i}), we create two loan pricing equations. The first is for loans with low strictness (below annual median covenant strictness), and the second is for loans with high strictness (above annual median covenant strictness), with both as a function of loan pricing determinants Xi:

$${y}_{lowstrict,i}={X}_i^{\prime }{\pi}_1+{u}_{1i}$$

(3)

$${y}_{highstrict,i}={X}_i^{\prime }{\pi}_2+{u}_{2i}$$

(4)

Following Lee (1978) and Prilmeier (2017), (y_{lowstrict, i} − y_{highstrict, i}) can be estimated in two steps. In the first step, Eqs. (3) and (4) are substituted into Eq. (2), and the resulting reduced-form probit model is estimated. The estimated linear predictor $\hat{\uppsi}$ is used to compute the Inverse Mills’ ratio, defined as $\phi \left(\hat{\uppsi}\right)/\left(1-\Phi \left(\hat{\uppsi}\right)\right)$ for loans with low covenant strictness and $-\phi \left(\hat{\uppsi}\right)/\left(\Phi \left(\hat{\uppsi}\right)\right)$ for loans with high covenant strictness.¹²

In the second step we estimate the two loan equations, correcting for selection bias by inserting the appropriate inverse Mills’ ratio into each equation. Using the estimated coefficients and excluding the effect of the inverse Mills’ ratio, we can calculate the predicted interest spreads for the entire sample under both high and low covenant strictness scenarios. We then winsorize predicted (y_{lowstrict, i} − y_{highstrict, i}) at the 1% and 99% levels, and we include it in the covenant strictness and interest spread models to control for the substitution effect of interest spread and covenant strictness.

Following Prilmeier (2017), we include additional control variables in the covenant strictness and interest spread specifications.¹³ In the interest spread regression we add the prevailing spread, defined as the natural log of average all-in-drawn spread of all loans issued in the syndicated loan market over the six-month period prior to a loan issuance (Avg Spread). The average market interest rate primarily reflects changes in institutional investors’ demand for syndicated loans (Ivashina and Sun 2011). At the same time, it seems reasonable to argue that average past interest spreads should not affect the covenant strictness choice for any given loan directly, other than through their effect on loan spread (Prilmeier 2017). Similarly, we add the average covenant strictness of all loans issued by the bank in the syndicated loan market over the six-month period prior to a loan issuance (Avg Strictness). We do this following prior studies which find that different lenders have their own preferences related to covenant strictness (Bushman et al. 2021; Christensen et al. 2021; Ma et al. 2022).

For brevity, we only present the results of the final covenant strictness and interest spread regressions in Panel B of Table 9. We find that the coefficient of (y_{lowstrict, i} − y_{highstrict, i}) is significantly positive on covenant strictness in Column 1 and significantly negative on interest spread in Column 2. These results indicate that the higher the cost of low covenant strictness, the more likely it is that a borrower chooses stricter covenants to lower interest spread. Importantly, even after controlling for the substitution effect between spread and covenant strictness in the regression, we still find that more patent filings lead to stricter covenants and lower spreads. These results confirm that a substitution effect does exist in our sample, but that interest spread and covenant strictness are not perfect substitutes. Together, these results are consistent with lenders intentionally imposing stricter covenants and lower spreads on firms in the commercialization stage.

In addition, we use several other methods to further mitigate concerns related to the simultaneous determination of covenant strictness and loan spreads. First, we estimate a system of two equations where covenant strictness and interest spreads are simultaneously determined using seemingly unrelated regression (SUR) analysis following Chan et al. (2013). This analysis allows the error terms of the two equations to be correlated. We provide the results of this test in Panel C of Table 9. Our inferences do not change.

Second, we use two-stage least squares (2SLS) to estimate the joint determination of covenant strictness and loan spread following Chan et al. (2013) and Hollander and Verriest (2016). Unlike SUR, the estimation of 2SLS requires instrumental variables for the two dependent variables. Following Chan et al. (2013) and Hollander and Verriest (2016), we use the natural log of average all-in-drawn spread of all loans issued in the syndicated loan market over the six-month period prior to a loan’s issuance (Avg Spread) as the instrument for loan spread, and Nbanks as the instrument for covenant strictness. Saavedra (2018) argues that the larger a loan syndicate, the higher the cost of renegotiation after a covenant violation, so lenders will impose looser covenants when the number of banks in a syndicate is large. We also drop Nbanks from the original control variables from the covenant strictness and interest spread regressions. For the sake of brevity, we only report the second stage results in Panel D of Table 9. Our inferences do not change.

Finally, we use three-stage least squares (3SLS) following Costello and Wittenberg-Moerman (2011) to take both the endogenous determination of covenant strictness and the correlation between the error terms of the two functions into consideration. The results are shown in Panel E of Table 9. Again, our inferences do not change.

4.6 Future R&D, capital expenditures, and M&a

We next further explore the effect that patent filings have on future R&D, capital expenditures, and M&A. In Table 10, we present the results of the association between patent filings and future R&D, capital expenditures, and the number of M&A deals. We find that patent filings are associated with an increase in future R&D (Columns 1 to 3) and capital expenditures (Columns 4 to 6). We further test whether the number of patent filings in an industry increases firms’ likelihood of acquiring target firms from that industry over the next one-, two-, or three-year period. We present the results in Columns 7 to 9, and we show that patent filings increase this type of acquisition over at least the next three years.

Table 10 Patent Filings and Debt Covenant Tightness: Future R&D, Capital Expenditures, and M&A

These findings provide at least a partial explanation for lenders’ desire to increase the strictness of debt covenants included in debt contracts. While patents, on average, signal a technological success, they may also lead to more R&D, capital investments, and M&A to finish the commercialization of the related technology. When lenders include stricter debt covenants in debt contracts, their ability to gain control rights in the event that financial performance deteriorates increases. As discussed above, lenders have a large menu of options from which to select when designing debt contracts. The fact that we observe stricter debt covenants in the contracts of borrowers with high firm patent filings is consistent with lenders viewing this dimension of the debt contract as an efficient way to manage portfolio risk.

5 Conclusions

This study explores the association between firm innovation and loan covenant tightness. We document that lenders include stricter covenants in the loan contracts of borrowers filing more patents, consistent with lenders imposing more oversight on firms demonstrating greater innovation. Our study provides a new perspective on the nature of lender screening and debt contract design, in line with the emerging literature on the role of financial institutions in directing corporate innovation (e.g., Mazzucato and Semieniuk 2018; Geddes and Schmidt 2020).

Our results provide evidence consistent with lenders positively viewing patent filings as a sign of technological success related to innovation. However, lenders increase covenant strictness to protect themselves from the downside risks inherent in the implementation or commercialization stage. Our study highlights the importance of considering both stages of the innovation process—idea generation and implementation—when studying the design of financing arrangements, because each of the two stages is subject to its own underlying logic and thus requires its own governance mechanisms, which are to be designed jointly.

Our findings are also consistent with recent developments in the incomplete contracting literature related to relational contracts. For example, Gibbons and Henderson (2012) argue that because the success of innovation requires that contracting parties make optimal decisions contingent upon the information that emerges at interim stages, the parties can improve efficiency by supplementing formal contracts with informal ones, based on the value of the future relationship. Strict debt covenants—a component of the formal contract—compel the parties to renegotiate their relational—i.e., informal—contract based on relevant new information (Watson et al. 2020; Kostadinov 2021). In other words, covenants play a unique role in debt contracting because they increase the self-enforcing range of the relational contract and thereby improve the efficiency of a contractual relationship. Our main tests are motivated by these theoretical findings, and our empirical results are consistent with them.

We perform a variety of additional tests to confirm the robustness of our main results. For example, we confirm that our results hold under PSM and entropy balancing and when using the American Investors Protection Act in November 2000 as a shock to the availability of patent filing information for unrelated banks. In cross-sectional tests, we find that the relation between innovation and covenant tightness is stronger when the lender has more expertise and experience and for borrowers with higher default risk.

We also conduct several tests related to the simultaneous determination of covenant strictness and interest spreads. Our results are robust to estimating our effects using SUR, 2SLS, and 3SLS. Overall, we find that covenant strictness and interest spreads do act as substitutes, but not perfect substitutes. Also, we document that borrowers in the early “innovation” stage are more likely to have higher interest spreads and lower covenant strictness, while borrowers in a later “commercialization” stage are more likely to have higher covenant strictness but lower interest spreads. These results are consistent with covenant strictness and interest spread serving different purposes and with lenders intentionally choosing certain combinations of loan terms based on a borrower’s innovation phase. These results extend the intuition from recent studies that document a lower spread for firms with pending patent approvals (e.g., Plumlee et al. 2015).

Consistent with the observed preferences of lenders, we document that a higher number of patent filings is associated with greater future R&D, capital investment, and the likelihood of M&A activity in the industry of the filed patents. Taken together, our results are consistent with lenders interpreting patents as indicative of the firm’s technological potential and, being aware of the need for stricter discipline at the subsequent implementation stage, designing the debt contract terms accordingly. Overall, our results are consistent with lenders tightening the debt covenants of highly inventive firms in order to better focus the firms’ managers’ attention upon the risks associated with turning innovative ideas into commercial successes.

Appendix

Table 11 Variable definitions