Soft Information in Loan Agreements

Abstract

In this study, we seek to understand whether soft information conveyed by contracting language found in private loan agreements is informative regarding borrower risk. We proxy for credit-risk-relevant soft information using Loughran and McDonald’s uncertainty measure. We first examine initial contract terms and find that, incremental to traditional summary measures of credit risk, increased contractual uncertainty is associated with higher initial loan spreads and a greater likelihood of using dynamic and performance-pricing covenants. We then turn to examine realized credit risk over the life of the loan and find that increased uncertainty is associated with a higher likelihood of future loan downgrades and loan amendments. We corroborate our results on the risk relevance of soft information by showing that the bid-ask spreads of loans trading on the secondary loan market are increasing in uncertainty. Overall, the evidence we provide is consistent with embedded linguistic cues in loan agreements publicly revealing the credit risk assessments of privately informed lenders.

Keywords

private information loan contracts textual analysis credit risk contract design

Introduction

Lenders and borrowers agree on contractual mechanisms during a loan’s negotiation process to help alleviate information asymmetry and address bondholder–stockholder conflicts. During these negotiations, borrowers typically share private, nonpublic information with lenders; this information is then factored into debt contract design. Contracting parties may further agree on including monitoring mechanisms that allow lenders to assess changes in borrowers’ credit circumstances over the life of the loan. In this article, we conjecture that the contracting language agreed upon between lenders and borrowers reflects the degree of information asymmetry between lending parties. In so doing, we seek to understand whether soft information conveyed by the contracting language found in private loan agreements is informative regarding borrower risk.

To proxy for credit risk relevant soft information in loan agreements that captures lenders’ concerns arising from information asymmetry, we use the intensity of words in the agreement that denote uncertainty following Loughran and McDonald (2011).¹ Consistent with our conjecture, we find that a higher frequency of uncertainty-related words is positively associated with the interest charged on the loan, measured as the spread over LIBOR at loan inception. Importantly, uncertainty provides explanatory power incremental to traditional summary measures of credit risk (e.g., size, leverage, credit rating, or the number of covenants).

Given that financial covenants are the standard mechanisms used in debt contracts to address agency conflicts between shareholders and debt holders, we next examine the relation between uncertainty and the use of covenants that are more likely to benefit lenders in adverse credit states. We find that uncertainty is positively associated with the use of dynamic covenants, which provide lenders with more control rights (Li, Vasvari, & Wittenberg-Moerman, 2016), and performance-pricing covenants, which potentially provide lenders with greater compensation, in the event of deteriorating firm credit performance. The evidence provided by our tests on initial contract terms supports the view that information pertaining to the risk of the borrower is embedded within the language of private debt contracts.

Moving beyond initial contract terms, we then examine the degree to which uncertainty is associated with realized risk throughout the life of the loan contract. The ability to screen ex-ante on the likelihood of adverse outcomes could aid a variety of external firm stakeholders. For example, such information would enable traders in the secondary loan market, such as managers of collateralized loan obligations (CLOs), to better manage the risk profile of their loan portfolios (Loumioti & Vasvari, 2015) or assist credit rating agency analysts with their soft rating adjustments when forming their overall rating assessments (Kraft, 2015).

First, we find that uncertain language in the initial contract is positively associated with future loan downgrades and contract amendments. Second, following Wittenberg-Moerman (2008), we examine loans trading in the secondary market and find that uncertain language is positively associated with secondary loan market bid-ask spreads. The evidence pertaining to ex-post outcomes reinforces our conclusion that soft information in private loan contracts is associated with the riskiness of the borrower. In sum, our results corroborate the notion that measureable linguistic features of contracts reflect latent risk regarding the creditworthiness of the borrower at both loan initiation and throughout the life of the loan.

Several recent studies examine aspects of debt contracts that are related to our study. Plumlee, Xie, Meng, and Yu (2015) examine the impact of privately communicated information about pending patents on loan pricing. In particular, they show that loan prices are lower for firms that have a patent pending. This is consistent with private information about the pending patent being communicated to lenders and, as a result, leading to lower interest rates charged on the loan. Carrizosa and Ryan (2015) primarily examine the determinants of private information in loan contracts. Consistent with the results of our study, they show that when borrowers are of higher risk or when there is high information asymmetry between borrowers and lenders, there is an increased flow of private information between the parties. Last, using a proprietary database, Minnis and Sutherland (2016) examine banks’ requests of private information from small commercial firms to study the role of financial statements as a monitoring mechanism. They find that when there is no mandate to prepare financial statements, banks’ requests for financial statements have a nonlinear relation with borrowers’ risk. That is, banks are more likely to request information from borrowers with medium risk level, and less likely to request information from low- and high-risk borrowers.

In supplemental analysis, we complement the work in these papers by examining the degree of monitoring that lenders and borrowers agree to put in place. Monitoring devices usually include the exchange of information during the life of the contract. For example, the borrower might agree to furnish the lenders with monthly financial statements or provide lenders with ongoing written communications with the auditor. To empirically capture this aspect of monitoring, we develop a list of terms that are indicative of information sharing between the lending parties and that plausibly serve to facilitate continual monitoring.² We then construct a secondary measure of soft information as the number of words in the loan agreements that contain these terms. Similar to our primary analyses, we find that greater contractual uncertainty is associated with a greater level of contractually specified borrower monitoring.

Our study makes several contributions to the literature on credit risk, debt contract design, and the cost of debt capital. First, we show that soft information in private loan contracts is useful in assessing the riskiness of the borrower, incremental to traditional summary measures of credit risk. Our evidence suggests that linguistic uncertainty captures borrower riskiness at the time of contract inception. Thus, our study informs the literature on credit risk by offering a novel source of information that can improve our ability to explain credit risk. Second, in demonstrating that soft information is useful in the context of debt contracting, we complement the work of Mayew, Sethuraman, and Venkatachalam (2015), who show that linguistic features of the MD&A section of annual reports are informative in assessing the viability of the firm as a going concern. Third, we are the first to provide evidence on the consequences of soft information embedded in loan agreements. In particular, we demonstrate that linguistic features of loan agreements have implications for realized credit risk over the life of the loan as well as for secondary loan market trading. By enhancing our understanding of credit risk beyond measures traditionally used to assess credit risk (e.g., credit ratings, distance to default, etc.), our study informs academics, investors, analysts, and regulators regarding risks ostensibly known to the borrowers and lenders, but not explicitly disclosed in the firm’s routine mandatory filings.

The remainder of the article is organized as follows. We provide a literature review and discuss our empirical predictions in the next section. We describe the data we employ in the section titled “Data.” In the section “Research Design,” we outline our research design, before providing empirical results in the section “Empirical Results.” Last, additional analyses are provided in “Supplemental Analysis” section. The section “Summary and Conclusion” concludes the article.

Literature Review and Empirical Predictions

Literature Review

Economic theory predicts that lenders will demand a higher price for loans made to high-risk borrowers. The riskiness of the borrower and its impact on price can take many forms. Borrower risk could be regarding the overall ability of the borrower to make interest payments and repay the principal. It could also be related to the inherent uncertainty regarding future states of the world that the borrower might face. In addition, borrower risk could increase as a result of greater information asymmetry between the lender and the borrower that reduces the lender’s ability to evaluate the borrower (e.g., Wittenberg-Moerman, 2008). While some forms of risk will result in explicit agreements between the borrower and the lender, others can be inferred from the actions of informed lenders, such as demanding detailed information from a borrower.

Several institutional features have evolved to overcome lenders’ difficulty in evaluating borrowers. For example, it has been argued that accounting conservatism shapes financial statements to be more informative to lenders (e.g., Ahmed, Billings, Morton, & Stanford-Harris, 2002; Watts, 2003; Zhang, 2008). Further, contractual provisions have emerged that are agreed upon between the borrowers and the lenders to help address lenders’ difficulty in evaluating borrowers, examples include debt covenants (Smith & Warner, 1979) and performance-pricing provisions (Asquith, Beatty, & Weber, 2005).

In addition, during the private loan negotiation process, there is an exchange of private information between borrowers and lenders. First, after a review of publicly available information, lenders demand private information that enables them to better evaluate borrowers’ credit risk. Private information, while reducing information asymmetry, might reveal to lenders a clearer picture of the nature of the borrowers’ future business risks and uncertainties. Second, as private information is revealed to lenders, and depending on lenders’ assessment of the borrower, the parties could agree on monitoring mechanisms that might involve future private information exchanges during the life of the loan (Carrizosa & Ryan, 2015; Minnis & Sutherland, 2016). For example, lenders might demand that borrowers furnish detailed reports about accounts receivable at prespecified future intervals. Hence, it is likely that the price of debt and/or contractual provisions will reflect either the information that is revealed to the lenders at the time of negotiations (Plumlee et al., 2015) or the existence of arrangements for future private information exchanges.

Empirical Predictions

The extent of the provision for future information exchanges likely varies with the degree of information asymmetry between the borrower and lender. As a result of the private information exchanged during a loan’s negotiation, the contracting language agreed upon between lending parties may publicly reveal latent information about the borrower. More specifically, if lenders are uncertain about the creditworthiness of the borrowers, the contracting language may contain more words that reflect uncertainty. Thus, we construct an empirical proxy based on such keywords in the contract following Loughran and McDonald (2011).

We expect our proxy for uncertainty to be related to several contractual features at loan inception as well as to credit risk events after loan inception. First, it is likely that contracts with higher degree of information asymmetry, as proxied for by contractual uncertainty, will command a higher price in the form of the spread above LIBOR that lenders charge on the loan.

Second, agency theory suggests that managers, acting in stockholders’ interest, have incentives to take actions that benefit stockholders to the detriment of bondholders (e.g., paying excessive dividends or issuing dilutive debt). As a result, debt contracts are designed to address the stockholder–bondholder conflict and protect lenders’ wealth. Financial covenants are one mechanism to address agency conflicts. For example, Li et al. (2016) show that dynamic covenants, whose thresholds evolve over the life of the loan, are meant to alleviate information asymmetries by committing the borrower to tighter thresholds with the passage of time. Similarly, Asquith et al. (2005) suggest that performance-pricing features in debt covenants play a similar role with respect to information asymmetry. Thus, we expect the use of both dynamic covenants and performance pricing to increase as the uncertainty about the borrower is higher.

Third, we conjecture that greater uncertainty regarding the borrowers’ creditworthiness at the time of contract initiation might indicate future negative events with respect to the loan, such as future loan downgrades and amendments. Specifically, because our uncertainty measure arises from the contract’s language, it potentially reflects information that is known by the contracting parties at contract inception. Therefore, it might reveal contracting parties’ assessment of borrowers’ creditworthiness. Prior research has documented that private loans are frequently amended (Roberts & Sufi, 2009). The renegotiation that precedes the amendment is likely a consequence of lenders’ uncertainty regarding the ongoing creditworthiness of the borrower. This assessment may be in part related to the borrower’s initial risk, either directly observed or latently perceived, at contract inception.

Finally, information asymmetry problems regarding borrowers could manifest when the loans are subsequently traded in the secondary market. We would expect that information frictions will be incorporated into pricing by increasing the bid-ask spreads on the traded loans. Thus, we also examine whether bid-ask spreads are increasing in the uncertainty reflected in the language of the loan contract at inception.

Data

We collect a comprehensive set of 11,173 private credit agreements directly from SEC filings between 2000 and 2010. Then, we match our private credit agreements to DealScan data, yielding 6,817 loan contracts that have the necessary private loan variables. We measure the extent of soft information in these loan contracts using Perl. We describe our detailed data collection and variable measurement procedures below.

Collection of Private Loan Agreements

The SEC requires public companies to include copies of all material contracts in their filings. Credit agreements typically appear as exhibits in 10-K, 10-Q, or 8-K filings. We use the search algorithm in Nini, Smith, and Sufi (2009; henceforth, NSS) with some modifications, following Beatty, Cheng, and Zach (2015), which we discuss below. As a result of the modification, our sample has a more comprehensive set of debt contracts than NSS.

First, we search all exhibits of 10-K, 10-Q, and 8-K filings in the SEC Edgar repository for the following 12 terms in capital letters: “credit agreement,” “loan agreement,” “credit facility,” “loan and security agreement,” “loan & security agreement,” “revolving credit,” “financing and security agreement,” “financing & security agreement,” “credit and guarantee agreement,” “credit & guarantee agreement,” “credit and security agreement,” or “credit & security agreement.” The last two terms are modifications to the terms used in NSS because these two terms result in legitimate credit agreements.

We then use a Perl program to ensure that the identification of the contracts as credit agreements is valid; if not, we delete them from the sample. To ensure the identification, we follow NSS’s requirement that the documents contain the term “table of contents” within 60 lines after the initial search terms. Unlike NSS, we allow “table of contents” to either be capitalized or not, because we found that in many credit agreements the term “table of contents” is not capitalized.

To ensure our data are of high quality, we perform a manual examination of all credit agreements with missing agreement dates or with agreement dates and filing dates that are more than 90 days apart. In these instances, we edit the agreement dates after manually identifying the date in the actual contract. In addition to revising the dates, the process also eliminates credit agreements that were filed twice.

Panel A of Table 1 details the sample of 11,173 contracts we collected from SEC filings between the years 2000 and 2010. Panel A reports the frequency of contracts by year as well as by the source filing in which the contract appeared. The distribution of contracts by year is fairly stable at around 10% until 2008. The number of debt contracts decreased substantially after 2008, possibly because of the financial crisis and the drying of credit markets. Further, the number of contracts in 2010 is low because the sample period for the study ends in June 2010. As for the source filings, about 48% of the contracts appear as attachments to 8-K filings, 32% are attached to 10-Q filings, and 20% are exhibits in 10-K filings.

Table 1.

Sample of Private Credit Agreements Collected From SEC Filings.

Panel A: Credit agreements by year and type of filings
Year	10-K	10-Q	8-K	Total (%)
2000	285	399	156	840 (7.5%)
2001	390	484	219	1,093 (9.8%)
2002	348	512	287	1,147 (10.3%)
2003	334	468	309	1,111 (9.9%)
2004	244	548	573	1,365 (12.2%)
2005	180	281	919	1,380 (12.4%)
2006	165	227	824	1,216 (10.9%)
2007	152	242	780	1,174 (10.5%)
2008	93	213	593	899 (8.0%)
2009	56	65	216	337 (3.0%)
2010	27	118	466	611 (5.5%)
Total (%)	2,274 (20.4%)	3,557 (31.8%)	5,342 (47.8%)	11,173 (100%)
Panel B: Sample selection
Total number of loan contracts collected from SEC filings			11,173 contracts
Total number of loan contracts that are covered by DealScan			6,817 contracts
Total number of loan facilities (contracts) with available variables for determinant Model (1)			5,629 facilities (4,206 contracts)
Total number of loan facilities (contracts) with available variables for Models (2) to (4)			4,915 facilities (3,749 contracts)
Total number of loan facilities (contracts) with available variables for Model (5)			835 facilities (572 contracts)

Note. This table presents the distribution of credit agreements by year and type of filings (Panel A) and sample selection (Panel B).

Matching Credit Agreements to DealScan

We restrict our sample to loan contracts that have other required variables from DealScan. We match our credit agreements with DealScan observations by GVKEY and contract date using the DealScan-Compustat linking table made available by Michael Roberts.³ We manually examine observations when multiple contracts were matched to a single or multiple DealScan deals on the same date. This step ensures that the matches between debt contracts and deals in DealScan are accurate. Out of the 11,173 credit agreements we identified, we are able to match 6,817 with the DealScan. For our main analyses, our sample is restricted to observations with required variables available. Panel B of Table 1 provides details of our sample construction for each analysis.

Measuring Soft Information in Credit Agreements

A growing body of work in accounting and finance uses techniques from computational linguistics to extract qualitative information from corporate disclosures. The applications vary widely and include general studies of annual reports (Lehavy, Li, & Merkley, 2011; Li, 2008; Miller, 2010), the MD&A (Brown & Tucker, 2011; Li, 2010; Mayew et al., 2015; Muslu, Radhakrishnan, Subramanyam, & Lim, 2014) risk disclosure sections of the annual reports (Beatty, Cheng, & Zhang, 2016; Campbell, Chen, Dhaliwal, Lu, & Steele, 2014; Kravet & Muslu, 2013), corporate press releases (Bozanic, Roulstone, & Van Buskirk, 2016; Bozanic & Thevenot, 2015; Davis, Piger, & Sedor, 2012; Davis & Tama-Sweet, 2012; Henry & Leone, 2016), and conference call transcripts (Davis, Ge, Matsumoto, & Zhang, 2015; Larcker & Zakolyukina, 2012). Collectively, the evidence in the literature suggests that important information can be gleaned from the qualitative aspects of financial disclosures. We extend this line of research to the private loan setting by extracting soft information from loan agreements that could be indicative of borrowers’ riskiness and creditworthiness.

As our proxy for soft information, we measure the extent of uncertain language reflected in loan contracts using the number of uncertain words (Uncertainty) based on the list of terms developed by Loughran and McDonald (2011). This variable has been used extensively in accounting and finance research in a variety of settings (e.g., Law & Mills, 2015; Purda & Skillicorn, 2015). To provide some descriptive support for the ability of our measure to capture credit risk, in untabulated analysis, we find that the number of uncertain words increases monotonically as long-term issuer credit ratings deteriorate. In addition to this measure, in supplemental analysis, we examine the extent of lender monitoring in loan contracts where we count the number of words in the contract that are related to lender monitoring (Monitoring). A full list of the monitoring keywords appears in Appendix B. We compiled this list after reading a random sample of loan contracts and identifying passages in which there is an explicit discussion of lender monitoring. Two such examples appear in Appendix C. In the first example, there is an agreement to provide the lender with an analysis of the borrower’s accounts and other detailed information. The second example shows an agreement to provide lenders with, among other things, the business plan and budget of the borrower and its subsidiaries.

Measuring Abnormal Uncertainty and Abnormal Monitoring

Firm and loan characteristics (e.g., firm size and loan amount) could partially explain the number of uncertainty or monitoring keywords that appear in loan contracts. As discussed previously, we are interested in capturing the incremental soft information that is not explained by these traditional firm and loan characteristics. Therefore, we estimate the following determinants Model (1) using OLS where we regress Uncertainty (Monitoring) on both firm and loan characteristics.⁴ Then, we use the unexplained component (i.e., the residuals), Abn_Uncertainty, (Abn_Monitoring) as our variables of interest for all main analyses.

\begin{array}{l} U n c e r t a i n t y (M o n i t o r i n g) = ρ_{0} + ρ_{1} S i z e + ρ_{2} B T M + ρ_{3} L e v e r a g e + ρ_{4} R O A + ρ_{5} A g e + ρ_{6} F i r m_ \\ R a t i n g + ρ_{7} N u m_A n a + ρ_{8} F S D_S c o r e + ρ_{9} L o a n_A m o u n t + ρ_{10} M a t u r i t y + ρ_{11} S e c u r e d + \\ ρ_{12} R e v o l v e r + α_{13} N u m_L e a d e r + α_{14} N u m_C o v + ρ_{15} L e n g t h + ε \end{array}

In Model (1), firm characteristics include Size, BTM, Leverage, ROA, Age, and Firm_Rating.⁵ Firm_Rating and ROA measure borrowers’ creditworthiness while Size, BTM, and Age capture the life cycle of the borrowers and, in part, borrowers’ public information environments. To some extent, a firm’s public information environment could potentially be viewed as a substitute for soft information in loan contracts. Therefore, we control for the number of analysts following (Num_Ana) and financial reporting quality (FSD_Score), where the latter is based on the level of financial statement error following Amiram, Bozanic, and Rouen (2015). We also include loan characteristics such as Loan_Amount and Maturity, as well as indicator variables for collaterals (Secure) and revolvers (Revolver). Furthermore, we control for the number of lenders (Num_Lender) and the number of financial covenants (Num_Cov). Finally, we control for the total number of sentences, that is, the length of the contract (Length). Please refer to Appendix A for detailed variable definitions. In Model (1) and all of our remaining models, we employ time, industry, and lead lender fixed effects with standard errors clustered by firm.

Research Design

Initial Loan Spread

In our first test, we examine the relation between initial loan spread (Spread) and our primary soft information measure: Abn_Uncertainty. We obtain Spread, the interest rate over LIBOR the borrower is charged on the loan, from DealScan. We estimate the following OLS Model (2):

\begin{array}{l} S p r e a d = α_{0} + α_{1} A b n_U n c e r t a i n t y + α_{2} S i z e + α_{3} B T M + α_{4} L e v e r a g e + α_{5} R O A \\ + α_{6} A g e + α_{7} T a n g + α_{8} F i r m_R a t i n g + α_{9} C F_V o l a t i l i t y + α_{10} B S M_P r o b \\ + α_{11} N u m_A n a + α_{12} F S D_S c o r e + α_{13} L o a n_A m o u n t + α_{14} M a t u r i t y \\ + α_{15} S e c u r e d + α_{16} R e v o l v e r + α_{17} N u m_L e a d e r + α_{18} N u m_C o v \\ + α_{19} P P_I n d + ε \end{array}

In Model 2, we include additional firm and loan control variables that could influence loan spread. We include tangibility (Tang) following Loumioti (2015). We add cash flow volatility (CF_Volatility) and firm’s distance-to-default probability, as implemented in Hillegeist, Keating, Cram, and Lundstedt (2004) using the Black-Scholes-Merton model (BSM_Prob), as additional controls for firm risk. We further include an indicator variable of whether there is a performance-pricing provision in the contract (PP_Ind). Other variables are the same as previously defined.

Dynamic Covenant and Performance Pricing

As discussed in “Empirical Predictions” section, we expect a higher degree of uncertainty to be associated with a higher likelihood of using dynamic covenants and performance-pricing provisions in loan agreements. For this test, we estimate the following probit Model (3):

\begin{array}{l} D y n a m i c_C o v (P P_I n d) = β_{0} + β_{1} A b n_U n c e r t a i n t y + β_{2} S i z e + β_{3} B T M \\ + β_{4} L e v e r a g e + β_{5} R O A + β_{6} A g e + β_{7} T a n g + β_{8} F i r m_R a t i n g + β_{9} C F_V o l a t i l i t y \\ + β_{10} B S M_P r o b + β_{11} N u m_A n a + β_{12} F S D_S c o r e + β_{13} L o a n_A m o u n t \\ + β_{14} M a t u r i t y + β_{15} S e c u r e d + β_{16} R e v o l v e r + β_{17} N u m_L e a d e r \\ + β_{18} N u m_C o v + β_{19} S p r e a d + ε \end{array}

where Dynamic_Cov (PP_Ind) is an indicator variable of having dynamic covenants (performance-pricing provisions) in the loan contract. We also include loan spread (Spread) in Model (3) as an additional control variable while all other variables are the same as previously defined.

Future Loan Downgrades and Amendments

To examine whether soft information is associated with realized credit risk, we collect information on loan downgrades and loan amendments from Moody’s Default and Recovery Database (DRD) and DealScan’s amendments database, respectively. To the extent that soft information reflects contracting parties’ assessment of borrowers’ creditworthiness, we expect the degree of uncertainty in debt contracts to be associated with future credit risk realizations, such as downgrades and contract amendments. Following Nikolaev (2015), we focus on contract amendments that require majority consent (i.e., 51% of votes).⁶ While amendments occur after the contract is put in place, it is likely that the probability of an amendment occurring is affected by the degree of uncertainty at loan initiation and by other contractual features. For example, the number of covenants is determined at the outset, but has implications for the probability of covenant violations and contract amendments. We estimate the following probit Model (4):

\begin{array}{l} D o w n g r a d e (A m e n d) = γ_{0} + γ_{1} A b n_U n c e r t a i n t y + γ_{2} S i z e + γ_{3} B T M + γ_{4} L e v e r a g e \\ + γ_{5} R O A + γ_{6} A g e + γ_{7} T a n g + γ_{8} F i r m_R a t i n g + γ_{9} C F_V o l a t i l i t y + γ_{10} B S M_P r o b \\ + γ_{11} N u m_A n a + γ_{12} F S D_S c o r e + γ_{13} L o a n_A m o u n t + γ_{14} M a t u r i t y + γ_{15} S e c u r e d \\ + γ_{16} R e v o l v e r + γ_{17} N u m_L e a d e r + γ_{18} N u m_C o v + γ_{19} P P_I n d + γ_{20} S p r e a d + ε \end{array}

where Downgrade (Amend) is an indicator variable taking the value of one if the loan was downgraded (amended) after issuance. All other variables are the same as previously defined.

Secondary Loan Market

Finally, we examine whether soft information in debt contracts has implications for loans subsequently traded on the secondary loan market. Following Wittenberg-Moerman (2008), we estimate Model (5) using OLS to examine the relation between Abn_Uncertainty and bid-ask spreads (BA_Spread) in the secondary loan market.

\begin{array}{l} B A_S p r e a d = θ_{0} + θ_{1} A b n_U n c e r t a i n t y + θ_{2} S i z e + θ_{3} B T M + θ_{4} L e v e r a g e + θ_{5} R O A \\ + θ_{6} A g e + θ_{7} N u m_A n a + θ_{8} R a t i n g_I n d + θ_{9} R e v o l v e r + θ_{10} R e p u t a t i o n \\ + θ_{11} D i s t r e s s + θ_{12} I n s t i t u t i o n + θ_{13} P r i m e_P u r p o s e + θ_{14} N u m_M M a k e r \\ + θ_{15} T i m e_M a t u + ε \end{array}

where Rating_Ind is an indicator variable set equal to one if the firm has a long-term issuer rating. We follow Wittenberg-Moerman (2008) and control for lead lender reputation (Reputation), loan distress (Distress), institutional investors (Institution), the primary purpose of the loan (Prime_Purpose), the number of market makers (Num_MMaker), and the time to maturity of the loan traded (Time_Matu). Other firm and loan variables are the same as previously defined.

Empirical Results

Descriptive Statistics

To measure Abn_Uncertainty (Abn_Monitoring), we regress the word count of Uncertainty (Monitoring) on firm and loan characteristics using determinants Model (1). Table 2 reports the results of this analysis. Most of the variables have predicted signs that are consistent with our expectations. For example, we find that ROA, firm age, firm rating, and analyst following are negatively associated with the number of uncertainty words used in loan agreements, while firm size and leverage are positively associated with it. This is consistent with the idea that higher profitability, higher credit quality, and better information environments are associated with lower degree of uncertainty. As for loan characteristics, we find that uncertainty words are more frequent for loans that are larger, have longer maturity, have collateral, and are revolving. Overall, our determinants model has a good explanatory power with an adjusted R² of 58.7%.

Table 2.

Determinants of Soft Information.

Variables	Predicted signs	Uncertainty
Size	±	2.896**
		(.011)
BTM	±	−0.105
		(.617)
Leverage	+	15.323***
		(.002)
ROA	−	−8.871**
		(.042)
Age	−	−0.131*
		(.094)
Firm_Rating	−	−0.434***
		(.004)
Num_Ana	−	−0.664***
		(.003)
FSD_Score	−	−1.032
		(.259)
Loan_Amount	±	9.041***
		(<.0001)
Maturity	+	0.420***
		(<.0001)
Secured	+	28.040***
		(<.0001)
Revolver	+	3.142*
		(.085)
Num_Lender	±	0.061
		(.745)
Num_Cov	±	0.798
		(.458)
Length	+	0.114***
		(<.0001)
Intercept		−34.788
		(.425)
Year FE		Yes
Industry FE		Yes
Lead Lender FE		Yes
N		5,629
Adjusted R²		.587

Note. This table presents estimation results of Model (1). The dependent variable is Uncertainty, which is the number of words appearing in the loan agreement that denote uncertainty (Loughran & McDonald, 2011). The residuals from this model represent abnormal uncertainty, Abn_Uncertainty. Other variable definitions are in Appendix A. All models include year, industry, and lead lender fixed effects and standard errors are clustered by firm.

p values are reported in parentheses below the coefficients. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively.

Table 3, Panel A provides descriptive statistics for the variables used in Model (1) and our subsequent analyses. On average, a loan agreement contains 271 uncertainty words as defined by Loughran and McDonald (2011), and 104 monitoring words from our dictionary in Appendix B. To help shed light on the measure’s construct validity, we note that Uncertainty is lower in the pre-crisis period (2000-2007) as compared with the post-crisis period (2008-2010). In addition, when we partition our sample into three groups based on borrowers’ credit ratings (conditional on firms having long-term issuer ratings), we find that Uncertainty increases monotonically as credit ratings deteriorate. An average contract is about 1,378 sentences long. Regarding the dependent variables of interest, the initial loan spread has a mean (median) of 203 (175) basis points. About 43% of loan agreements use dynamic covenant (Dynamic_Cov) while 72% of agreements have performance-pricing provisions (PP_Ind). On average, only about 1.3% of loans are downgraded after issuance, while roughly 39% are subsequently amended. The loan and firm control variables are largely consistent with prior literature. For example, the average loan maturity is about 46 months, 63% of loans are secured, and each contract contains about three financial covenants (Num_Cov).

Table 3.

Univariate Statistics.

Panel A: Descriptive Statistics.
Variable	n	25th percentile	50th percentile	75th percentile	Mean	SD
Uncertainty	5,629	197.000	260.000	323.000	271.158	120.306
Monitoring	5,629	69.000	95.000	129.000	103.723	52.115
Abn_Uncertainty	5,629	−42.794	−3.371	35.869	0.000	77.281
Abn_Monitoring	5,629	−22.027	−3.181	18.069	0.000	38.482
Length	5,629	988.000	1,279.000	1,601.000	1,378.090	684.487
Spread	5,440	87.500	175.000	275.000	203.308	148.037
Dynamic_Cov	5,629	0.000	0.000	1.000	0.427	0.495
PP_Ind	5,629	0.000	1.000	1.000	0.717	0.450
Downgrade	5,629	0.000	0.000	0.000	0.013	0.113
Amend	5,629	0.000	0.000	1.000	0.394	0.489
BA_Spread	835	0.431	0.538	1.000	0.878	0.996
Size	5,629	5.266	6.638	7.836	6.477	2.075
BTM	5,629	0.259	0.450	0.717	0.073	5.188
Leverage	5,629	0.154	0.290	0.436	0.315	0.236
ROA	5,629	−0.002	0.037	0.073	0.013	0.187
Age	5,629	6.000	13.000	31.000	17.147	12.801
Tang	5,621	0.128	0.261	0.485	0.322	0.238
Firm_Rating	5,629	1.000	9.000	16.000	8.958	8.914
CF_Volatility	5,627	11.131	31.959	102.680	123.466	321.395
BSM_Prob	5,082	0.000	0.017	0.920	3.105	9.782
Num_Ana	5,629	1.000	5.000	10.000	6.427	6.390
FSD_Score	5,629	2.238	2.767	3.332	2.843	0.852
Loan_Amount	5,629	17.728	18.826	19.807	18.676	1.586
Maturity	5,629	35.000	48.000	60.000	46.063	22.123
Secured	5,629	0.000	1.000	1.000	0.627	0.484
Revolver	5,629	0.000	1.000	1.000	0.576	0.494
Num_Lender	5,629	3.000	7.000	12.000	8.895	8.222
Num_Cov	5,629	2.000	3.000	3.000	2.650	1.156

Note. This table presents the descriptive statistics for the full sample used in Model (1) of Table 2. Please see Appendix A for variable definitions.

Table 3, Panel B presents the Pearson correlations for the variables used in our sample. We find some univariate evidence consistent with our expectations. That is, Abn_Uncertainty is positively associated with loan spread (Spread), the use of dynamic covenants (Dynamic_Cov) and performance-pricing provisions (PP_Ind), and the likelihood of future amendments (Amend). In addition, there is a strong positive association (0.38) between Abn_Uncertainty and Abn_Monitoring, which is consistent with the notion that higher degree of uncertainty leads to greater lender monitoring.

Panel B: Pairwise Correlation.
Variables	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)	(11)	(12)	(13)	(14)	(15)	(16)	(17)	(18)	(19)	(20)	(21)	(22)	(23)	(24)	(25)	(26)
Abn_Uncertainty (1)	1.00
Abn_Monitoring (2)	0.38	1.00
Length (3)	0.00	0.00	1.00
Spread (4)	0.04	0.11	0.16	1.00
Dynamic_Cov (5)	0.04	0.00	0.23	0.33	1.00
PP_Ind (6)	0.04	−0.06	0.05	−0.41	−0.05	1.00
Downgrade (7)	0.02	0.01	0.04	0.00	0.04	0.00	1.00
Amend (8)	0.02	0.01	0.13	0.10	0.14	0.00	0.04	1.00
BA_Spread (9)	0.03	0.05	−0.07	0.07	−0.12	0.00	0.16	0.02	1.00
Size (10)	0.00	0.00	−0.12	−0.55	−0.24	0.30	0.01	−0.13	−0.01	1.00
BTM (11)	0.00	0.00	−0.05	−0.10	−0.05	0.06	−0.10	−0.03	−0.01	0.22	1.00
Leverage (12)	0.00	0.00	0.11	0.24	0.22	−0.14	0.04	0.03	0.02	−0.16	−0.11	1.00
ROA (13)	0.00	0.00	−0.02	−0.28	−0.09	0.19	0.00	−0.03	−0.11	0.25	0.22	−0.24	1.00
Age (14)	0.00	0.00	−0.07	−0.24	−0.14	0.12	0.03	−0.05	0.01	0.29	0.01	−0.02	0.07	1.00
Tang (15)	−0.03	−0.02	−0.01	−0.01	−0.06	0.00	0.02	−0.01	0.05	0.04	−0.03	0.26	−0.02	0.06	1.00
Firm_Rating (16)	0.00	0.00	−0.10	−0.39	−0.17	0.19	0.06	−0.12	0.02	0.61	0.07	0.11	0.10	0.41	0.12	1.00
CF_Volatility (17)	−0.03	0.01	−0.11	−0.16	−0.14	0.07	0.02	−0.06	0.01	0.44	0.00	0.01	0.02	0.24	0.05	0.34	1.00
BSM_Prob (18)	0.01	0.04	0.04	0.37	0.06	−0.23	0.02	0.03	0.11	−0.34	0.06	0.19	−0.29	−0.12	0.00	−0.14	−0.06	1.00
Num_Ana (19)	0.00	0.00	−0.14	−0.39	−0.24	0.17	0.05	−0.10	0.04	0.71	0.07	−0.11	0.12	0.18	0.08	0.51	0.42	−0.16	1.00
FSD_Score (20)	0.00	0.00	−0.03	0.07	−0.04	−0.06	−0.02	0.00	0.07	−0.15	0.03	−0.04	−0.02	−0.15	0.08	−0.16	−0.10	0.01	−0.10	1.00
Loan_Amount (21)	0.00	0.00	−0.01	−0.44	−0.13	0.29	0.06	−0.04	−0.02	0.74	−0.02	0.04	0.15	0.28	0.14	0.58	0.39	−0.24	0.57	−0.16	1.00
Maturity (22)	0.00	0.00	0.21	0.02	0.22	0.09	0.05	0.12	−0.18	0.02	−0.03	0.06	0.08	−0.10	−0.02	−0.09	−0.09	−0.10	−0.06	−0.05	0.13	1.00
Secured (23)	0.00	0.00	0.27	0.54	0.34	−0.23	0.02	0.17	−0.04	−0.51	−0.06	0.16	−0.16	−0.31	−0.04	−0.45	−0.22	0.19	−0.43	0.07	−0.40	0.19	1.00
Revolver (24)	0.00	0.00	0.02	−0.23	−0.14	0.28	−0.01	0.04	0.24	0.00	0.02	−0.13	0.05	0.01	−0.01	−0.05	−0.06	−0.06	−0.02	0.01	0.07	0.13	−0.04	1.00
Num_Lender (25)	0.00	0.00	0.01	−0.33	−0.05	0.20	0.04	−0.03	0.06	0.52	−0.02	0.07	0.05	0.20	0.08	0.48	0.30	−0.16	0.46	−0.14	0.61	0.11	−0.29	0.02	1.00
Num_Cov (26)	0.00	0.00	0.22	0.34	0.44	−0.05	0.02	0.17	−0.10	−0.39	−0.03	0.12	−0.06	−0.21	−0.08	−0.35	−0.25	0.09	−0.35	0.03	−0.32	0.12	0.37	−0.04	−0.19	1.00

Note. This table presents the Pearson correlation for the full sample used in Model (1) of Table 2. Bold and italic indicate significance at 5% level. Please see Appendix A for variable definition.

Initial Loan Pricing

Table 4 reports estimation results of Model (2). Our independent variable of interest is Abn_Uncertainty, that is, the residuals from determinants Model (1) whose estimation results we report in Table 2. We find strong evidence that Abn_Uncertainty is associated with initial loan pricing, incremental to traditional proxies for credit risk. The coefficient on Abn_Uncertainty is 0.095 and is statistically significant at the 1% level. In terms of economic magnitude, holding all else constant, for a one standard deviation increase in Abn_Uncertainty, initial loan spreads increase by 7.3 basis points. In addition, we find that the control variables relating to the loan characteristics are mostly significant in directions that are consistent with prior literature. For example, we find that loans with longer maturities, collateral, and greater number of covenants are associated with higher spreads. In contrast, loans that have performance-pricing provisions as well as revolver loans are associated with lower spreads. Further, we find that the traditional proxies for credit risk, which we use as control variables, also load in the predicted directions. Specifically, Leverage, cash flow volatility (CF_Volatility), and the probability of default (BSM_Prob) are positively associated with Spread, while borrower size, ROA, and credit rating are negatively associated with spreads.

Table 4.

Soft Information and Initial Loan Spread.

Variables	Predicted signs	Spread
Abn_Uncertainty	+	0.095***
		(<.0001)
Size	−	−12.670***
		(<.0001)
BTM	±	3.164
		(.364)
Leverage	+	52.235***
		(<.0001)
ROA	−	−81.882***
		(.002)
Age	−	−0.149
		(.146)
Tang	−	0.602
		(.481)
Firm_Rating	−	−0.764***
		(.003)
CF_Volatility	+	0.034***
		(<.0001)
BSM_Prob	+	2.090***
		(<.0001)
Num_Ana	−	−0.380
		(.129)
FSD_Score	−	1.337
		(.269)
Loan_Amount	−	−1.720
		(.189)
Maturity	+	0.322***
		(.001)
Secured	+	74.653***
		(<.0001)
Revolver	−	−41.783***
		(<.0001)
Num_Lender	−	−0.954***
		(.001)
Num_Cov	±	12.405***
		(<.0001)
PP_Ind	−	−54.279***
		(<.0001)
Intercept		433.544***
		(<.0001)
Year FE		Yes
Industry FE		Yes
Lead Lender FE		Yes
N		4,915
Adjusted R²		.555

Note. This table presents the estimation results of testing the relation between Abn_Uncertainty and initial loan spread (Model 2). The dependent variable is Spread which is defined as the interest rate the borrower pays over LIBOR. Independent variable of interest is Abn_Uncertainty, that is, the residuals from Model (1) when Uncertainty is the dependent variable of interest. Uncertainty is the number of words in the loan contract which denote uncertainty (Loughran & McDonald, 2011). Other variable definitions are in Appendix A. All models include year, industry, and lead lender fixed effects and standard errors are clustered by firm.

p values are reported in parentheses below coefficients. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively.

Overall, the results in Table 4 document that greater Abn_Uncertainty is associated with higher spreads charged for private debt. These findings suggest that soft information as reflected in linguistic features of loan agreements captures lenders’ assessments of borrowers’ creditworthiness at contract inception. As such, these results support our predictions insofar as soft information appears to reflect important credit risk elements known to privately informed lenders and revealed publicly by linguistic characteristics of the loan agreement.

Dynamic Covenants and Performance-Pricing Provisions

Table 5 provides our second set of regression results investigating the relation between soft information as measured by Abn_Uncertainty and the use of dynamic covenants or performance-pricing provisions in the contract.

Table 5.

Soft Information and Covenants.

Panel A: Dynamic covenants
Variables	Predicted signs	Dynamic_Cov
Abn_Uncertainty	+	0.002***
		(.010)
Size	−	−0.122**
		(.036)
BTM	±	−0.053
		(.527)
Leverage	+	1.572***
		(<.0001)
ROA	−	−0.936***
		(.006)
Age	−	0.003
		(.272)
Tang	−	−0.962***
		(.002)
Firm_Rating	−	0.009
		(.171)
CF_Volatility	+	−0.000
		(.123)
BSM_Prob	±	−0.029***
		(<.0001)
Num_Ana	−	−0.036***
		(.005)
FSD_Score	−	−0.150***
		(.007)
Loan_Amount	±	0.160***
		(.007)
Maturity	+	0.018***
		(<.0001)
Secured	+	0.494***
		(.000)
Revolver	−	−0.558***
		(<.0001)
Num_Lender	±	0.030***
		(.002)
Num_Cov	+	0.740***
		(<.0001)
Spread	+	0.004***
		(<.0001)
Intercept		−3.215**
		(.049)
Year FE		Yes
Industry FE		Yes
Lead Lender FE		Yes
N		4,915
Pseudo R²		.454
ROC Curve		0.848
Panel B: Performance-pricing provisions
Variables	Predicted signs	PP_Ind
Abn_Uncertainty	+	0.002***
		(.003)
Size	±	0.000
		(.995)
BTM	±	−0.026
		(.633)
Leverage	±	−0.359
		(.176)
ROA	±	0.551*
		(.064)
Age	±	−0.003
		(.522)
Tang	±	0.445
		(.180)
Firm_Rating	−	0.012
		(.115)
CF_Volatility	+	0.000
		(.171)
BSM_Prob	±	−0.012**
		(.048)
Num_Ana	−	−0.020*
		(.073)
FSD_Score	−	−0.068
		(.108)
Loan_Amount	+	0.156***
		(.002)
Maturity	+	0.007***
		(.004)
Secured	±	−0.295**
		(.037)
Revolver	+	1.104***
		(<.0001)
Num_Lender	±	0.008
		(.393)
Num_Cov	+	0.282***
		(<.0001)
Spread	−	−0.005***
		(<.0001)
Intercept		10.694***
		(<.0001)
Year FE		Yes
Industry FE		Yes
Lead Lender FE		Yes
N		4,915
Pseudo R²		.311
ROC Curve		0.799

Note. This table presents the estimation results of testing the relation between Abn_Uncertainty, dynamic covenants, and performance-pricing provisions (Model 3). The dependent variables are indicator variables of Dynamic_Cov (Panel A) and PP_ind (Panel B). Independent variable of interest is Abn_Uncertainty, that is, the residuals from Model (1) when Uncertainty is the dependent variable of interest. Uncertainty is the number of words in the loan contract which denote uncertainty (Loughran & McDonald, 2011). Other variable definitions are in Appendix A. All models include year, industry, and lead lender fixed effects and standard errors are clustered by firm.

p values are reported in parentheses below coefficients. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively.

We find strong evidence that Abn_Uncertainty is positively associated with the likelihood of using both dynamic covenants (Panel A) and performance pricing (Panel B) in debt contracts. The coefficients on Abn_Uncertainty in both panels equal .002 and are statistically significant at the 1% level. In terms of economic magnitude, holding all else constant, for a one standard deviation increase in Abn_Uncertainty, the likelihood of using dynamic covenants (performance-pricing provisions) increases by 5.0% (5.2%) from its mean. Control variables are in directions that are consistent with our expectations. For example, we find that the use of dynamic covenants is negatively (positively) associated with firm size, ROA, tangibility, probability of default, analyst following, and financial reporting quality (leverage, loan amount, maturity, number of covenants, and loan spread). Finally, we note that the pseudo R² is 45.4% (31.1%) and that the ROC curve is 84.8% (79.9%) in Panel A (Panel B), suggesting that our model has high explanatory power.

Overall, the results in Table 5 demonstrate that a greater degree of lender uncertainty is associated with a higher likelihood of using both dynamic covenants and performance-pricing provisions in debt contracts. These results corroborate the initial loan spread results and thus add credence to the idea that soft information in loan contracts captures important credit risk elements known to privately informed lenders at contract inception.

Future Loan Downgrades and Amendments

In Table 6 we examine whether soft information in debt contracts is predictive of future loan rating downgrades and loan amendments. We find evidence that Abn_Uncertainty is associated with higher likelihood of future loan rating downgrades (Panel A) and amendments (Panel B), incremental to traditional proxies for credit risk. The coefficient on Abn_Uncertainty in Panel A equals .003 and is statistically significant at the 5% level while the coefficient on Abn_Uncertainty in Panel B equals .001 and is statistically significant at the 5% level. The effects are also economically significant. For example, holding all else constant, for a one standard deviation increase in Abn_Uncertainty, the likelihood of a future amendment increases by 3.7%. Again, fit statistics suggest that our model has high explanatory power. For example, the loan downgrades analysis (Panel A) has a pseudo R² of 24.2% and a ROC curve of 88.9%, respectively.

Table 6.

Soft Information and Realized Credit Risk.

Panel A: Loan downgrades
Variables	Predicted signs	Downgrade
Abn_Uncertainty	+	0.003**
		(.036)
Size	−	−0.162
		(.174)
BTM	±	−0.018
		(.808)
Leverage	+	1.628***
		(.005)
ROA	−	0.280
		(.337)
Age	±	0.020
		(.124)
Tang	±	−0.431
		(.624)
Firm_Rating	±	0.087***
		(<.0001)
CF_Volatility	+	−0.000
		(.255)
BSM_Prob	±	0.009
		(.275)
Num_Ana	±	0.068**
		(.028)
FSD_Score	±	−0.137
		(.465)
Loan_Amount	+	0.266**
		(.041)
Maturity	+	0.007
		(.133)
Secured	+	0.893***
		(.006)
Revolver	±	0.097
		(.762)
Num_Lender	±	−0.009
		(.596)
Num_Cov	+	0.219*
		(.091)
PP_Ind	±	−0.400
		(.226)
Spread	±	−0.001
		(.685)
Intercept		−28.663***
		(<.0001)
Year FE		Yes
Industry FE		Yes
Lead Lender FE		Yes
N		4,915
Pseudo R²		.242
ROC Curve		0.889
Panel B: Loan amendments
Variables	Predicted signs	Amend
Abn_Uncertainty	+	0.001**
		(.037)
Size	−	−0.129***
		(.001)
BTM	±	−0.004
		(.923)
Leverage	+	−0.251
		(.173)
ROA	−	−0.425**
		(.026)
Age	±	0.002
		(.540)
Tang	±	0.066
		(.742)
Firm_Rating	−	−0.013***
		(.008)
CF_Volatility	+	0.000
		(.123)
BSM_Prob	+	−0.002
		(.559)
Num_Ana	−	0.005
		(.552)
FSD_Score	−	−0.060*
		(.065)
Loan_Amount	+	0.142***
		(.000)
Maturity	+	0.010***
		(<.0001)
Secured	+	0.397***
		(<.0001)
Revolver	+	0.105*
		(.067)
Num_Lender	±	0.005
		(.322)
Num_Cov	+	0.187***
		(<.0001)
PP_Ind	−	0.072
		(.384)
Spread	+	0.001**
		(.052)
Intercept		−15.511***
		(<.0001)
Year FE		Yes
Industry FE		Yes
Lead Lender FE		Yes
N		4,915
Pseudo R²		.125
ROC Curve		0.674

Note. This table presents the estimation results of testing the relation between Abn_Uncertainty, loan downgrades and amendments (Model 4). The dependent variables are indicator variables of Downgrades (Panel A) and Amend (Panel B). Independent variable of interest is Abn_Uncertainty, that is, the residuals from Model (1) when Uncertainty is the dependent variable of interest. Uncertainty is the number of words in the loan contract which denote uncertainty (Loughran & McDonald, 2011). Other variable definitions are in Appendix A. All models include year, industry, and lead lender fixed effects and standard errors are clustered by firm.

p values are reported in parentheses below coefficients. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively.

These findings suggest that firms whose debt contracts contain more uncertain language are more likely to be downgraded or amended in the future. Thus, the evidence again supports our conjecture that the language in debt contracts is reflective of information about future realizations of credit risk, incremental to traditional measures of risk used in prior literature. As such, the linguistics-based contract characteristics that reflect credit risk relevant information known to privately informed lenders at contract inception appear to also be informative regarding the creditworthiness of the borrower throughout the life of the loan.

Secondary Loan Market Trading

In Table 7 we evaluate the relation between Abn_Uncertainty and bid-ask spreads of the loans that are subsequently traded in the secondary loan market and find evidence that Abn_Uncertainty is associated with higher bid-ask spread in the secondary loan market. The coefficient on Abn_Uncertainty equals .001 and is statistically significant at the 5% level. In terms of economic magnitude, holding all else constant, for an increase of one standard deviation in Abn_Uncertainty, the bid-ask spread increases by 8.8% from its mean. In addition, control variables are significant in the directions consistent with prior literature. For example, similar to Wittenberg-Moerman (2008), we find that bid-ask spread is negatively associated with ROA, firm age, the availability of credit rating, the number of market makers, and the time to maturity, while it is positively associated with revolver and distress indicators.

Table 7.

Soft Information and Secondary Loan Market Trading.

Variables	Predicted signs	BA_Spread
Abn_Uncertainty	+	0.001**
		(.025)
Size	−	0.005
		(.437)
BTM	±	−0.003
		(.270)
Leverage	+	0.027
		(.425)
ROA	−	−0.476**
		(.038)
Age	−	−0.005
		(.128)
Num_Ana	−	−0.008
		(.255)
Rating_Ind	−	−0.137
		(.109)
Revolver	+	0.332***
		(.0001)
Reputation	−	0.015
		(.415)
Distress	+	1.810***
		(<.0001)
Institution	+	−0.043
		(.209)
Prime_Purpose	+	−0.000
		(.498)
Num_Mmaker	−	−0.021
		(.110)
Time_Matu	−	−0.006***
		(<.0001)
Intercept		1.143***
		(.001)
Year FE		Yes
Industry FE		Yes
Lead Lender FE		Yes
N		835
Adjusted R²		.461

Note. This table presents the estimation results of testing the relation between Abn_Uncertainty and bid-ask spread in the secondary loan market (Model 5). The dependent variable is BA_Spread which is the average quarterly bid-ask spread of the traded loan. Independent variable of interest is Abn_Uncertainty, that is, the residuals from Model (1) when Uncertainty is the dependent variable of interest. Uncertainty is the number of words in the loan contract which denote uncertainty (Loughran & McDonald, 2011). Other variable definitions are in Appendix A. All models include year, industry, and lead lender fixed effects and standard errors are clustered by firm.

p values are reported in parentheses below coefficients. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively.

These results suggest that greater degree of lenders’ uncertainty at loan inception reflects similar information to that which traders subsequently use in pricing loans in the secondary market. As such, the secondary loan market trading results add to the body of evidence provided in further corroborating the notion that linguistic-based contract characteristics are informative at loan inception and throughout the life of the loan.⁷

Supplemental Analysis

Lender Monitoring

In the previous section, we provide consistent and strong evidence that soft information in loan agreements is associated with features of debt contract design, such as initial loan pricing, and conveys borrowers’ credit risk assessments, as reflected by its association with future consequences of borrowers’ riskiness. In this section, we examine whether greater uncertainty at loan initiation is also associated with stronger monitoring efforts as revealed in loan agreements. We conjecture that lenders will engage in more monitoring activities when faced with greater uncertainty at loan inception. For example, lenders could demand more forward-looking information such as budgets and forecasts. In addition, lenders can also request the borrowers to provide more timely information such as monthly financial statements or written communications with the auditor before the audit report is publicly available. As we expect that greater lender monitoring will also be reflected in contractual language, we measure lenders’ monitoring efforts by using the residuals (Abn_Monitoring) from Model (1) when examining the number of Monitoring keywords (Appendix B) as the dependent variable. Then, we regress Abn_Monitoring on Abn_Uncertainty and the same set of firm and loan characteristics.

We report the results in Table 8. We find that greater uncertainty at loan initiation is associated with greater lender monitoring efforts as suggested by the positive and significant coefficient on Abn_Uncertainty (p value = .02). This test further corroborates our main findings and provides additional evidence that soft information in loan agreements captures information about lenders’ credit risk assessments that may be incrementally informative to other external stakeholders of the firm.

Table 8.

Lender Monitoring.

Variables	Predicted signs	Abn_Monitoring
Abn_Uncertainty	+	0.165***
		(<.0001)
Size	±	1.669*
		(.053)
BTM	±	−0.478
		(.621)
Leverage	+	−1.371
		(.399)
ROA	−	3.042
		(.276)
Age	−	−0.010
		(.498)
Tang	−	−5.085
		(.179)
Firm_Rating	−	0.040
		(.429)
CF_Volatility	+	−0.000
		(.437)
BSM_Prob	+	0.069
		(.207)
Num_Ana	−	0.070
		(.315)
FSD_Score	−	0.718
		(.195)
Loan_Amount	±	−0.846
		(.280)
Maturity	+	0.014
		(.397)
Secured	±	−5.238***
		(.001)
Revolver	+	4.113***
		(.001)
Num_Lender	±	0.089
		(.425)
Num_Cov	±	−0.040
		(.976)
PP_Ind	−	−5.620***
		(.002)
Spread	+	0.046***
		(<.0001)
Intercept		−7.507
		(.709)
Year FE		Yes
Industry FE		Yes
Lead Lender FE		Yes
N		4,915
Adjusted R²		.145

Note. This table presents the estimation results of testing the relation between Abn_Uncertainty and lender monitoring (Abn_Monitoring). The dependent variable is Abn_Monitoring which is residuals from Model (1) when Monitoring is the dependent variable of interest. Monitoring is the total word count of monitoring keywords as shown in Appendix B. Independent variable of interest is Abn_Uncertainty, that is, the residuals from Model (1) when Uncertainty is the dependent variable of interest. Uncertainty is the number of words in the loan contract which denote uncertainty (Loughran & McDonald, 2011). Other variable definitions are in Appendix A. All models include year, industry, and lead lender fixed effects and standard errors are clustered by firm.

p values are reported in parentheses below coefficients. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively.

Robustness to an Alternative Loughran and McDonald (2011) Word List

Loughran and McDonald’s (2011) uncertainty word list contains the word “may.” In our private loan agreement setting, the word “may” is pervasive. While the sample mean for Uncertainty is 271, when we recompute Uncertainty but omit the word “may” from the word list, the mean for the revised variable, Uncertainty_Alt, drops considerably to 109, or 40% of the original sample mean. To ensure that our results are robust to this alternate specification, we compute Abn_Uncertainty_Alt and rerun the Spread analysis using the alternate measure. We find in Table 9 that our results remain unchanged. In addition, in untabulated analysis, we confirm that our inferences remain unchanged when we use Abn_Uncertainty_Alt instead of Uncertainty_Alt in the remaining tables (Tables 5 -8).

Table 9.

Alternative Measure of Uncertainty.

Variables	Predicted signs	Spread
Abn_Uncertainty_Alt	+	0.188***
		(.0001)
Size	−	−12.622***
		(<.0001)
BTM	±	3.333
		(.338)
Leverage	+	52.290***
		(<.0001)
ROA	−	−81.926***
		(.002)
Age	−	−0.151
		(.193)
Tang	−	0.074
		(.498)
Firm_Rating	−	−0.765***
		(.002)
CF_Volatility	+	0.034***
		(<.0001)
BSM_Prob	+	2.083***
		(<.0001)
Num_Ana	−	−0.382
		(.128)
FSD_Score	−	1.503
		(.244)
Loan_Amount	−	−1.790
		(.179)
Maturity	+	0.324***
		(.0004)
Secured	+	74.609***
		(<.0001)
Revolver	−	−41.749***
		(<.0001)
Num_Lender	−	−0.958***
		(.0002)
Num_Cov	±	12.460***
		(<.0001)
PP_Ind	−	−54.404***
		(<.0001)
Intercept		433.258***
		(<.0001)
Year FE		Yes
Industry FE		Yes
Lead Lender FE		Yes
N		4,915
Adjusted R²		.554

Note. This table presents the estimation results of testing the relation between Abn_Uncertainty_Alt and initial loan spread (Model 2). The dependent variable is Spread which is defined as the interest rate the borrower pays over LIBOR. Independent variable of interest is Abn_Uncertainty_Alt, that is, the residuals from Model (1) when Uncertainty_Alt is the dependent variable of interest. Uncertainty_Alt is the number of words in the loan contract which denote uncertainty (Loughran & McDonald, 2011) less the word count of “may.” Other variable definitions are in Appendix A. All models include year, industry, and lead lender fixed effects and standard errors are clustered by firm.

p values are reported in parentheses below coefficients. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively

Summary and Conclusion

Empirical research on credit risk traditionally captures firm and loan risk characteristics with proxies such as size, leverage, credit rating, or the number of covenants. Research to date has not examined the linguistic properties of private loan contracts and how such properties may publicly inform interested parties about the riskiness of those contracts. Given that a private lender has access to information beyond that residing in the public domain or in credit analysts’ models, there may be features of the contract that are based on that private information and that reveal the riskiness of the borrower, incremental to traditional summary measures of credit risk.

Borrowing from the field of computational linguistics, we extract soft information from private loan agreements to investigate its implications for loan contract design and borrower credit risk. The language embedded in loan contracts likely reflects lenders’ information asymmetry concerns arising from lenders’ private assessments of borrowers’ creditworthiness. To capture this soft information, we follow Loughran and McDonald (2011) and focus on the intensity of uncertain words found in private loan agreements. Hence, our empirical approach allows us to assess whether observable contractual features that arise from private negotiations between borrowers and lenders publicly reveal credit risk relevant information that may be of use to other external stakeholders of the firm.

We first examine how uncertainty relates to initial contract terms and find evidence that the spread at loan inception is positively associated with uncertainty. This relation persists even after controlling for several potential confounding factors, including traditional credit risk proxies, such as the distance to default and borrower credit rating, and employing a two-stage model to mitigate endogeneity concerns. We also find that uncertainty is positively associated with the use of dynamic and performance-pricing covenants. These results support the view that soft information appears to capture important credit risk elements known to privately informed lenders at contract inception.

We then examine how uncertainty relates to realized credit risk. We find that uncertainty is positively associated with future loan downgrades and future loan amendments, as well as with the bid-ask spreads of loans trading on the secondary market. These results further support the view that linguistic-based contract characteristics that reflect credit risk relevant information known to privately informed lenders at contract inception appear to also be informative throughout the life of the loan.

As we provide evidence that the linguistic features of loan contracts publicly convey soft information about the riskiness of the borrower, our article has the potential to add to the literature whose goal is to predict future (negative) credit-related events. Specifically, incorporating information that is embedded in the contract by negotiating parties holds promise for improving the current set of credit risk models (e.g., Altman, 1968; Bharath & Shumway, 2008; Hillegeist et al., 2004; Ohlson, 1980; Shumway, 2001). Hence, academics and investors can use the study’s findings to augment loan pricing models and models that attempt to predict default probabilities of firms. Indeed, practitioners are apparently already doing so (see, for example, Thomson Reuter’s StarMine Text Mining Credit Risk Model). Further, such information on latent credit risk may assist secondary loan market traders in their loan portfolio risk assessments (Loumioti & Vasvari, 2015) as well as credit rating agency analysts with their soft rating adjustments as they form their overall credit rating assessments (Kraft, 2015). These avenues are left for future research.

Footnotes

Appendix A

Appendix B

Appendix C

Acknowledgements

We thank the editor (Thomas Omer), two anonymous referees, Yakov Amihud, Dan Amiram, Joy Begley, Richard Carrizosa (JAAF discussant), Mei Cheng, Daniel Cohen, Trevor Harris, Maria Loumioti, Mike Minnis, Monica Neamtiu, Haresh Sapra, Oktay Urcan, Chris Williams, Helen Zhang, and participants at Hong Kong Polytechnic University and SUNY–Buffalo accounting workshops, the 2013 Tel Aviv University Accounting Conference, 2013 HKUST Accounting Symposium, and 2014 FARS Midyear Meeting for helpful comments and suggestions. Bozanic and Zach are research fellows at the National Center for the Middle Market (NCMM) and acknowledge the generous financial support provided by the Fisher College of Business and NCMM for the data used in this study.

Declaration of Conflicting Interests

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

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The author(s) received financial support from the National Center for the Middle Market (NCMM).

Notes

References

Ahmed

A. S.

Billings

B. K.

Morton

R. M.

Stanford-Harris

(2002). The role of accounting conservatism in mitigating bondholder-shareholder conflicts over dividend policy and in reducing debt costs. The Accounting Review, 77, 867-890.

Altman

E. I.

(1968). Financial ratios, discriminant analysis and the prediction of corporate bankruptcy. Journal of Finance, 23, 589-609.

Amiram

Bozanic

Rouen

(2015). Financial statement errors: Evidence from the distributional properties of financial statement numbers. Review of Accounting Studies, 20, 1540-1593.

Asquith

Beatty

Weber

(2005). Performance pricing in bank debt contracts. Journal of Accounting & Economics, 40, 101-128.

Beatty

Cheng

Zach

(2015). Non-recurring items in debt contracts (Working paper).

Beatty

Cheng

Zhang

(2016). Sometimes less is more: Evidence from financial constraints risk factor disclosures (Working paper).

Beatty

Ramesh

Weber

(2002). The importance of accounting changes in debt contracts: The cost of flexibility in covenant calculations. Journal of Accounting & Economics, 33, 205-227.

Bharath

S. T.

Shumway

(2008). Forecasting default with the Merton distance to default model. Review of Financial Studies, 21, 1339-1369.

Bozanic

Roulstone

Van Buskirk

(2016). Management earnings forecasts and other forward-looking statements (Working paper). The Ohio State University, Columbus.

10.

Bozanic

Thevenot

(2015). Qualitative disclosure and changes in sell-side financial analysts’ information environment. Contemporary Accounting Review, 32, 1595-1616.

11.

Brown

S. V.

Tucker

J. W.

(2011). Large-sample evidence on firms’ year-over-year MD&A modifications. Journal of Accounting Research, 49, 309-346.

12.

Campbell

Chen

Dhaliwal

Steele

(2014). The information content of mandatory risk factor disclosure in corporate filings. Review of Accounting Studies, 19, 396-455.

13.

Carrizosa

Ryan

(2015). Borrower private information covenants and loan contract monitoring and renegotiation (Working paper).

14.

Davis

A. K.

Matsumoto

Zhang

J. L.

(2015). The effect of managerial style on the tone of earnings conference calls. Review of Accounting Studies, 20, 639-673.

15.

Davis

A. K.

Piger

J. M.

Sedor

L. M.

(2012). Beyond the numbers: Measuring the information content of earnings press release language. Contemporary Accounting Research, 29, 845-868.

16.

Davis

A. K.

Tama-Sweet

(2012). Managers’ use of language across alternative disclosure outlets: Earnings press releases versus MD&A. Contemporary Accounting Research, 29, 804-837.

17.

Henry

Leone

(2016). Measuring qualitative information in capital markets research: Comparison of alternative methodologies to measure disclosure tone. The Accounting Review, 91, 153-178.

18.

Hillegeist

S. A.

Keating

E. K.

Cram

D. P.

Lundstedt

K. G.

(2004). Assessing the probability of bankruptcy. Review of Accounting Studies, 9, 5-34.

19.

Kraft

(2015). Rating agency adjustments to GAAP financial statements and their effect on ratings and credit spreads. The Accounting Review, 90, 641-674.

20.

Kravet

Muslu

(2013). Informativeness of risk disclosures in corporate annual reports. Review of Accounting Studies, 18, 1088-1122.

21.

Larcker

D. F.

Zakolyukina

(2012). Detecting deceptive discussions in conference calls. Journal of Accounting Research, 50, 495-540.

22.

Law

Mills

(2015). Taxes and financial constraints: Evidence from linguistic cues. Journal of Accounting Research, 53, 777-819.

23.

Lehavy

Merkley

(2011). The effect of annual report readability on analyst following and the properties of their earnings forecasts. The Accounting Review, 86, 1087-1115.

24.

(2008). Annual report readability, current earnings, and earnings persistence. Journal of Accounting & Economics, 45, 221-247.

25.

(2010). The information content of forward-looking statements in corporate filings—A Naïve Bayesian machine learning approach. Journal of Accounting Research, 48, 1049-1102.

26.

Vasvari

Wittenberg-Moerman

(2016). Dynamic threshold values in earnings-based covenants. Journal of Accounting & Economics, 61, 605-629.

27.

Loughran

T. I. M.

McDonald

(2011). When is a liability not a liability? Textual analysis, dictionaries, and 10-Ks. Journal of Finance, 66, 35-65.

28.

Loumioti

(2015). The use of intangible assets as loan collateral (Working paper). University of Southern California, Los Angeles.

29.

Loumioti

Vasvari

(2015). Managerial discretion in unregulated lending institutions: Evidence from the securitized loan market (Working paper).

30.

Mayew

W. J.

Sethuraman

Venkatachalam

(2015). MD&A disclosure and the firm’s ability to continue as a going concern. The Accounting Review, 90, 1621-1651.

31.

Miller

B. P.

(2010). The effects of reporting complexity on small and large investor trading. The Accounting Review, 85, 2107-2143.

32.

Minnis

Sutherland

(2016). Financial statements as monitoring mechanism: Evidence from small U.S. commercial loans. Journal of Accounting Research, Forthcoming.

33.

Muslu

Radhakrishnan

Subramanyam

K. R.

Lim

(2014). Forward-looking MD&A disclosures and the information environment. Management Science, 61, 931-948.

34.

Nikolaev

(2015). Scope for renegotiation and debt contract design (Working paper). University of Chicago, IL.

35.

Nini

Smith

D. C.

Sufi

(2009). Creditor control rights and firm investment policy. Journal of Financial Economics, 92, 400-420.

36.

Ohlson

J. A.

(1980). Financial ratios and the probabilistic prediction of bankruptcy. Journal of Accounting Research, 18, 109-131.

37.

Plumlee

Xie

Meng

(2015). Bank loan spread and private information: Pending approval patents. Review of Accounting Studies, 20, 593-638.

38.

Purda

Skillicorn

(2015). Accounting variables, deception, and a bag of words: Assessing the tools of fraud detection. Contemporary Accounting Research, 32, 1193-1223.

39.

Roberts

M. R.

Sufi

(2009). Renegotiation of financial contracts: Evidence from private credit agreements. Journal of Financial Economics, 93, 159-184.

40.

Shumway

(2001). Forecasting bankruptcy more accurately: A simple hazard model. Journal of Business, 74, 101-124.

41.

Smith

Warner

(1979). On financial contracting: An analysis of bond covenants. Journal of Financial Economics, 7, 117-161.

42.

Watts

(2003). Conservatism in accounting Part I: Explanations and implications. Accounting Horizons, 17, 207-221.

43.

Wittenberg-Moerman

(2008). The role of information asymmetry and financial reporting quality in debt trading: Evidence from the secondary loan market. Journal of Accounting & Economics, 46, 240-260.

44.

Zhang

(2008). The contracting benefits of accounting conservatism to lenders and borrowers. Journal of Accounting & Economics, 45, 27-54.