Analyst coverage and syndicated lending

According to the latest annual Shared National Credit (SNC) review conducted by federal banking regulators—the Federal Reserve System (FRS), the Federal Deposit Insurance Corporation (FDIC), and the Office of the Comptroller of the Currency (OCC)—the total syndicated loan market was $4.8 trillion in terms of commitments outstanding as of the third quarter of 2019.

Yuan (2006) shows that increased analyst coverage is associated with decreased bank loan spreads. Like Coyne and Stice (2018), his findings are mainly based on OLS regressions without correcting for the endogenous determination of analyst coverage and loan spread. Although Yuan tries to address the endogeneity in supplementary tests via 2SLS regression, the variables excluded from his second stage regressions (i.e., instrumental variables) include stock price and market-to-book ratio, which are likely associated with loan spreads. Yuan (2006) does not distinguish between lead arrangers’ and participant lenders’ differential information access as we do.

To illustrate the pricing of a syndicated loan, consider a loan initially launched by the lead arranger with a target spread of LIBOR +150 to LIBOR +200. Assume that the lead does not know the demand with certainty, having only priors regarding the distribution of the demand based on comparable transactions and market conditions. Once the bidding opens, potential participants submit bids that are tiered by spread. For example, say bank A puts in $25 million at LIBOR +200 and bank B puts in $15 million at LIBOR +150. At the end of the process, the lead will total up the commitments and decide where to price the loan. If the loan is oversubscribed (i.e., total commitment amounts exceed the required amounts) at LIBOR +150, the lead will adjust the spread down. If the loan is undersubscribed even at LIBOR +200, then the spread will have to increase further to attract more funding. As we elaborate below, the number of analysts following the borrower can plausibly influence participants’ perception of the agency conflicts within the lending syndicate.

For participants to receive any material nonpublic information (MNPI), such as financial projections, from the lead arranger, they must stay on the “private side of the wall,” which prohibits them from trading in the borrower’s other securities. Some lenders opt to stay on the “public side of the wall” to reserve their right to trade the borrower’s other securities (Bushman et al. 2010). These lenders do not receive MNPI.

Reputation in the syndicated loan market may help mitigate lead arrangers’ perverse incentives because their bad actions could damage future syndication businesses. However, such a correction is not always timely, and not all lead arrangers’ bad actions are caught.

One oft-studied information source is the borrower’s credit ratings provided by rating agencies. However, not all borrowers are rated, and analyst research can be incrementally informative to lenders even for rated borrowers. Lui et al. (2012) show that changes in equity analysts’ risk rating lead credit rating downgrades but not vice versa, suggesting that analyst research can be a substantive source of new information that is incremental to what is conveyed by changes in credit ratings. Participant lenders also use borrower financial statements. However, such information is low frequency (quarterly) and largely backward-looking because financial accounting aims to report a firm’s past performance and not to forecast its future performance (Ball and Shivakumar 2008). ^.

For research on the informational content of analyst research, see, among many others, Gleason and Lee (2003), Bradshaw (2004), Frankel et al. (2006), Howe et al. (2009), and Lui et al. (2012).

Banking regulators can penalize banks for not imposing inspection or visitation provisions in credit agreements as failure to do so likely undermines the bank’s safety and soundness. Nearly all credit agreements contain such a provision as part of affirmative covenants. A sample visitation clause provided by the Loan Syndication and Trading Association takes the following form: “The Borrower will, and will cause each Subsidiary to, permit any representatives designated by the Administrative Agent or any Lender…to visit and inspect its properties, to examine and make extracts from its books and records, and to discuss its affairs, finances, and condition with its officers and independent accountants, all at such reasonable times and as often as reasonably requested.” (see Bellucci and McCluskey 2016, Section 7.5.2).

He and Tian (2013) use a similar 12-month gap when constructing annual firm-level variables.

We use the Roberts Dealscan-Compustat Linking Database from the Wharton Research Data Services (WRDS) platform (Chava and Roberts 2008).

Compared to matching without replacement, matching with replacement yields better covariate balancing, but it could lead to larger variances of the estimated treatment effect since the same control observation can be used multiple times. Clustering standard errors at the firm level adequately addresses the potentially large variances in the error term due to repeat observations in our sample. In an untabulated robustness test, we form a matched sample without replacement and redo the main analysis. Our results continue to hold, though the magnitude of the DiD estimators for both dependent variables drops in half compared to that in the main tests.

Treatment and control samples also have different numbers of loan facilities because firms can obtain multiple facilities in a loan package and we treat each loan facility as a separate observation.

The number of lead arrangers for loans in our sample varies widely, with an interquartile range of 2 to 9, a median of 4, and a mean of 6.7. In practice, there are sometimes more senior or less senior positions within the non-participant group that we classify as “lead” arrangers, but each is senior to (and likely has an informational advantage over) participant lenders (François and Missonier-Piera2007).

In an untabulated analysis, we find that treated firms’ analyst coverage decreases by about one analyst after brokerage mergers, compared to control firms, for at least three years after the mergers.

In untabulated analyses, we add year fixed effects to the main models. In the loan spread analysis, the coefficient on TREAT×POST remains significant in both economic (coefficient = 0.140) and statistical terms (p value = 0.065). In the participant syndicate share analysis, the coefficient on TREAT×POST is economically meaningful (coefficient = −0.061) but not statistically significant at conventional levels (p value = 0.175).

Gunn et al. (2017) find that firms switch to high-quality auditors after a reduction in their analyst coverage to mitigate increased information asymmetries. We investigate whether our results are driven by auditor switching. In our sample, only two treatment firms switched from a non-Big N auditor to a Big N auditor (which is a variable that Gunn et al. use to capture firms’ switching to a high-quality auditor). Our results do not change when removing observations associated with the two firms. Also following Gunn et al., we check whether the effect is driven by firms switching to a larger audit office. We lose about half of the observations because audit office data from Audit Analytics are only available from year 2000 on. There are only five treatment firms that experienced an audit upgrade, and our results are the same when excluding those observations. Therefore, we do not believe the increase in loan spread is driven by firms switching auditors after analyst departure.