A. Performance-Sensitive Debt

Performance-sensitive debt (PSD) is a type of debt in which interest payments depend on a measure of a borrower’s financial health (e.g., credit ratings, debt-to-cash flow ratio). The idea behind this type of debt obligation is to delay the costly renegotiation that results from borrowers violating covenants. In general, if a borrower’s financial health deteriorates, the interest rate associated with the PSD obligation increases, thus compensating the debt holder for the additional risk.

PSD contracts became common in the early 1990s. Now, the total size of the PSD market is over $2 trillion. In this article, we focus on the $900 billion market in which interest payments depend exclusively on credit ratings.

In PSD contracts, interest rates are contractually linked to measures of a borrower’s financial health through a pricing grid. Figure 1 shows the pricing grid from The Walt Disney Co.’s syndicated 5.25-year revolving credit facility issued on Feb. 23, 2005. The loan amount was $2.25 billion. The facility’s pricing grid indicates that The Walt Disney Co. can be subject to pay five different interest rates, depending on its long-term senior debt rating. For example, if the firm’s credit rating by S&P is AA- or better, then The Walt Disney Co. is subject to an interest rate of 11.5 bp over the London Interbank Offered Rate (LIBOR). If their credit rating is between A and A+, the interest rate increases to 13 bp over LIBOR. The highest interest rate that The Walt Disney Co. may be required to pay under this facility is 30.0 bp over LIBOR, which is an increase triggered by a credit rating deterioration to BBB- or worse.

FIGURE 1Example Pricing Grid

Figure 1 shows the pricing grid from The Walt Disney Co.’s syndicated 5.25-year revolving credit facility issued on Feb. 23, 2005. Graph A shows that there is no change in the loan’s interest rate if the firm’s credit rating changes from A+ to A. Graph B shows that the loan’s interest rate changes from LIBOR +13 bps to LIBOR +14 bps if the firm’s credit rating changes from A to A−. The 1 bps increase translates into additional borrowing costs of as high of $225,000 per year.

FIGURE 1 Long description

The figure contains two tables and a footer note.

Graph A. One-Notch Credit Rating Downgrade From A plus.

The table has six columns: Firm's name (Disney), Quarter (2006 Q 1), Current rating (A plus), Minimum rating, Maximum rating, and Spread over L I B O R (b p s).

* Row 1: Minimum AA minus, Spread 11.5.

* Row 2 (highlighted in blue): Minimum A, Maximum A plus, Spread 13.

* Row 3: Minimum A minus, Maximum A minus, Spread 14.

* Row 4: Minimum BBB minus, Maximum BBB plus, Spread 17.5.

* Row 5: Maximum BBB minus, Spread 30.

A blue U-shaped arrow to the right of rows 2 and 3 points back to row 2 with the text: Downgrade does not change the loan's interest rate.

Graph B. One-Notch Credit Rating Downgrade From A.

The table structure is identical but the Current rating is A.

* Row 1: Minimum AA minus, Spread 11.5.

* Row 2 (highlighted in blue): Minimum A, Maximum A plus, Spread 13.

* Row 3: Minimum A minus, Maximum A minus, Spread 14.

* Row 4: Minimum BBB minus, Maximum BBB plus, Spread 17.5.

* Row 5: Maximum BBB minus, Spread 30.

A red curved arrow to the right points from row 2 down to row 3 with the text: Downgrade does change the loan's interest rate.

Footer note: Loan amount = $2.25 billion, 1 b p s = up to $225,000 in annual savings.

B. Credit Rating Agencies and Conflicts of Interest

The main role of CRAs is to perform a continual assessment of firms’ creditworthiness. CRAs communicate these assessments to financial markets in the form of credit ratings. Credit ratings are important for the rated firm because they affect the firm’s ability to access capital through the value placed on them by investors, either for informational or regulatory reasons. Importantly, it is not the investor who pays for the issuance of the credit rating but rather the rated firm itself. This issuer–pays business model of credit ratings has been subject to extensive criticism because it can generate a conflict of interest that can lead to CRAs catering to the firms that pay them.

In our setting of credit rating-based PSD loans, this conflict arises when a credit rating downgrade by a CRA causes a change in the pricing schedule that increases the interest rate paid by the borrower. For example, in the pricing grid from Figure 1, there is no change in Disney’s borrowing costs (i.e., the interest rate would remain 13 bp over LIBOR) if the firm is downgraded from a credit rating of A+ to a credit rating of A. However, if the firm is downgraded to A−, its borrowing costs would increase to 14 bp over LIBOR, implying additional annual interest expenses of up to approximately $225,000 if the entire credit facility were drawn. This number is about half the median annual cost of a downgrade in our sample (i.e., $520,000). We hypothesize that, all else being equal, a CRA will be more reluctant to downgrade a firm if the downgrade causes higher costs for the borrower. Moreover, the misalignment of incentives should be even more severe when a CRA’s rating is the one that leads directly to an increase in borrowing costs.

Credit ratings are supposed to be an objective and high-quality assessment of borrower creditworthiness. CRAs claim to meet the highest standards of integrity, independence, objectivity, and transparency (U.S. Securities and Exchange Commission (2002)). CRAs are aware that credit ratings can negatively affect borrowers’ funding costs (Moody’s Investors Service (2002)). However, CRAs’ guidelines are explicit in that credit ratings are issued independently of the effect that they may have on the borrower (U.S. Securities and Exchange Commission (2002), S&P Global Ratings (2018)). Similarly, CRAs acknowledge the conflicts of interest that arise from the issuer–pays business model (Department of Justice (2015), (2017)). However, fear of diminished reputation and regulatory penalties is often thought to mitigate these conflicts. Section C of the Supplementary Material includes excerpts from CRA documents, regulator documents, congressional hearings, DOJ press releases, and settlement statements of facts discussing these issues.

CRAs catered to underwriting banks by inflating the credit ratings associated with RMBSs and CDOs, making them important contributors to the financial crisis (Griffin et al. (Reference Griffin, Nickerson and Tang2013)). Recently, both S&P and Moody’s agreed to settlements with the DOJ in which they i) explicitly acknowledged failing to adhere to their own standards when rating securitized products and ii) renewed their commitment to credit rating quality (see Section C of the Supplementary Material).

The catering behavior of CRAs during the run-up to the financial crisis has partly been attributed to the complexity of securitized products (Griffin (Reference Griffin2019)). In this article, we study whether the conflicts of interest affecting CRAs prevail in other less complex markets, such as the PSD market. Importantly, pricing data are more reliable and available for PSD loans than for RMBSs and CDOs, which makes the PSD setting ideal for studying whether market participants recognize and internalize the conflicts of interest, both at origination and in the secondary market. Finally, we also study whether the DOJ settlements were effective in holding CRAs to their renewed commitment to credit rating quality, a question that cannot be answered by studying the RMBS and CDO markets because those markets disappeared before the settlements.

A. Primary Data

We obtain loan-level data on performance pricing (i.e., pricing grids) from LPC DealScan. Recall that we focus on loans with performance provisions that depend exclusively on senior credit ratings. To identify where borrowers stand in the pricing grid, we assign to each loan the corresponding borrower’s long-term senior unsecured credit rating from Bloomberg.Footnote ⁹ We consider the credit ratings issued by S&P and Moody’s. Following Tchistyi et al. (Reference Tchistyi, Yermack and Yun2011), for cases in which borrowers have received credit ratings from both CRAs, we consider the better of the two ratings as relevant for pricing purposes.

We then use the pricing grids to assign each loan in each quarter the current interest spread and the interest spread that would prevail after a credit rating downgrade. Some contracts give borrowers a choice between different types of reference rates for determining interest spreads (e.g., LIBOR, the bank’s prime rate). Since interest spreads over LIBOR are by far the most common, we use this market rate as our baseline measure for interest spreads.Footnote ¹⁰ We use $ cost\hskip0.35em of downgrade $ to refer to the difference between the spread at the prevailing credit rating and the spread at the credit rating one notch below it.

B. Secondary Data and Final Sample

Initially, we obtain pricing grid data for 31,005 facilities originated between 2000 and 2016. We link these facilities to Compustat using the latest DealScan-Compustat linking table (Chava and Roberts (Reference Chava and Roberts2008)), which leaves 27,141 loans. We then require all relevant loan and firm information to be nonmissing, which leaves 3,700 loans. Of these loans, 2,075 have pricing grids that depend exclusively on credit ratings. Finally, we require borrowers to have credit ratings issued by S&P or Moody’s.Footnote ¹¹ The final sample consists of 1,988 loans spanning 22,760 loan-quarter observations.Footnote ¹² The sample consists of 510 distinct borrowers, of which 415 obtain at least two loans, with an average of 4.1 loans per borrower. Among these, 387 firms have overlapping loans, meaning that they hold more than one loan outstanding at the same time during the sample period.

Panels A and B of Table 1 describe the final sample. The mean increase in borrowing costs following a one-notch downgrade is 13.7 bp, with an interdecile range of 25 bp. Moreover, the cost of downgrade is nonzero 80% of the time. Downgrades are rare: The likelihood of a borrower being downgraded in a given quarter is, on average, only 2.6%. The median credit rating throughout the sample is BBB. The average loan size is $914.0 million, 80% of loans are revolvers, and conditional on being downgraded, the average borrower is downgraded within 4.8 quarters of issuing a loan. The Appendix describes these and the remaining variables used in the article and specifies the numerical equivalence assigned to each credit rating (e.g., AAA/Aaa = 1, AA+/Aa1 = 2, AA/Aa = 3).

TABLE 1Data Summary

TABLE 1 Long description

The table is divided into three sections.

Panel A. Borrower-Quarter Characteristics (N = 22,760 for most rows):

* Cost of downgrade (bps): Mean 13.74, Std. Dev. 12.07, p10 0.00, p50 12.50, p90 25.00.

* Leverage: Mean 0.31, Std. Dev. 0.17, p10 0.12, p50 0.30, p90 0.51.

* Total assets ($ billion): Mean 23.42, Std. Dev. 56.15, p10 2.08, p50 8.81, p90 47.02.

* Intangibles over assets: Mean 0.21, Std. Dev. 0.20, p10 0.00, p50 0.14, p90 0.55.

* Profitability (R O A): Mean 0.01, Std. Dev. 0.02, p10 -0.00, p50 0.01, p90 0.03.

* 1(downgrade) (percent): Mean 2.57, Std. Dev. 15.81, p10 0.00, p50 0.00, p90 0.00.

* Issuer credit rating (numeric): Mean 9.16, Std. Dev. 2.63, p10 6.00, p50 9.00, p90 12.00.

* R & D ($ million, N = 10,401): Mean 283.81, Std. Dev. 620.20, p10 0.00, p50 65.92, p90 718.10.

Panel B. Loan Characteristics (N = 1,988 for most rows):

* No. of financial covenants: Mean 1.46, Std. Dev. 0.83, p10 0.00, p50 1.00, p90 2.00.

* Loan amount ($ million): Mean 914.04, Std. Dev. 910.40, p10 150.00, p50 550.00, p90 2200.00.

* 1(secured): Mean 0.16, Std. Dev. 0.37, p10 0.00, p50 0.00, p90 1.00.

* Average cost of downgrade: Mean 14.82, Std. Dev. 8.50, p10 6.83, p50 12.50, p90 25.00.

* 1(revolver): Mean 0.80, Std. Dev. 0.40, p10 0, p50 1, p90 1.

* Time to downgrade (quarters, N = 455): Mean 4.77, Std. Dev. 5.14, p10 0.00, p50 3.00, p90 12.00.

Panel C. Variation Within Groups (N = 22,760):

* 1(downgrade) (percent): Overall Std. Dev. 15.81, Within Credit Rating 15.59, Within Firm 15.03, Within Year 15.76.

* Cost of downgrade (bps): Overall Std. Dev. 12.07, Within Credit Rating 10.94, Within Firm 8.84, Within Year 11.90.

* Time to downgrade (quarters): Overall Std. Dev. 4.39, Within Credit Rating 4.30, Within Firm 3.56, Within Year 4.17.

Finally, Figure 2 shows the distribution of credit ratings at origination. The most common rating is BBB, followed by AA. Overall, 93% of observations are rated at or above investment grade.

FIGURE 2Distribution of Credit Ratings at Origination

Figure 2 shows the distribution of credit ratings at the time of loan origination for our sample of credit rating-based PSD loans. The credit rating scale is simplified by combining the credit ratings within each letter credit rating category. For example, we combine the initial credit ratings of A+, A, and A− into one group, A.

FIGURE 2 Long description

The Y-axis is labeled Ratings at origination and lists categories from top to bottom: A A, A, B B B, B B, B, C C C, C C, and D. The X-axis is labeled Fraction and ranges from 0 to 0.6 in increments of 0.2.

* A A: A very short bar near 0.01.

* A: A bar extending to approximately 0.25.

* B B B: The longest bar, extending to approximately 0.56.

* B B: A bar extending to approximately 0.13.

* B: A short bar extending to approximately 0.03.

* C C C: A negligible bar near 0.005.

* C C: A negligible bar near 0.

* D: A negligible bar near 0.

The distribution is heavily concentrated in the investment-grade B B B and A categories, with a significant drop-off for ratings below B B.

C. Empirical Framework

Pricing grids vary within loan contracts (and therefore the change in borrowing costs associated with a credit rating downgrade also varies). Thus, to study whether the conflicts of interest of CRAs affect the credit ratings they issue, we estimate specifications of the form

(1) $$ 1{(downgrade)}_{i,t}={\beta}_1 cost\hskip0.35em of\hskip0.35em {downgrade}_{i,l,t}+{X}_{i,l,t}^{\prime}\Gamma +{\unicode{x025B}}_{i,l,t}, $$

where $ 1{(downgrade)}_{i,t} $ is an indicator that takes the value of 1 if borrower $ i $ is downgraded in quarter $ t $ , and 0 otherwise. The independent variable of interest is $ cost\hskip0.35em of\hskip0.1em {downgrade}_{i,l,t} $ , a measure of the increase in the loan spread that would result from a one-increment credit rating downgrade for loan $ l $ in quarter $ t $ . Thus, this variable captures the time-varying degree of the CRAs’ conflicts of interest for each loan. The higher the increase in a firm’s borrowing costs following a one-notch downgrade, the stronger the incentive for the CRA to avoid the downgrade. The variable $ {X}_{i,l,t} $ is a vector of time-varying loan- and firm-level characteristics and fixed effects. In secondary tests, we also estimate specifications in which the dependent variable is the time interval in which the firm is downgraded after loan origination.

Identifying conflicts of interest in PSD is challenging for multiple reasons. First, firms could self-select into PSD either because they have inside information (Manso et al. (Reference Manso, Strulovici and Tchistyi2010), Begley (Reference Begley2012)) or because they are overly optimistic (Adam et al. (Reference Adam, Burg, Scheinert and Streitz2020)). Even within PSD contracts, a naïve comparison of the probability of a downgrade between firms that self-selected into ratings-based PSD and those that did not is likely to be biased, with the former set of firms being downgraded less frequently. Consistent with this concern, Table 2 shows that firms with credit rating-based PSD loans differ significantly from those with accounting ratio-based PSD loans across a range of observable characteristics. Firms with credit rating-based PSD loans tend to be larger, less leveraged, and more likely to rely on larger, unsecured loans compared to firms whose PSD terms are based on accounting ratios.

TABLE 2Accounting Ratio-Based PSD Loans Versus Credit Rating-Based PSD Loans

TABLE 2 Long description

The table consists of four columns: Metric, Accounting Ratio-Based Mean, Credit Rating-Based Mean, and Difference.

* Leverage: 0.41 vs 0.31, difference 0.10.

* Total assets in dollars billions: 4.92 vs 23.42, difference minus 18.50.

* Intangibles over assets: 0.29 vs 0.21, difference 0.09.

* Profitability R O A: 0.005 vs 0.009, difference minus 0.004.

* R and D in dollars millions: 61.53 vs 283.80, difference minus 220.96.

* Number of financial covenants: 2.12 vs 1.52, difference 1.24.

* Loan amount in dollars millions: 350.9 vs 894.5, difference minus 543.64.

* 1 secured: 0.88 vs 0.18, difference 0.70.

* Cost of one grid in rating or ratio, b p: 18.36 vs 13.74, difference 4.63.

* Number of observations: 21,789 for Accounting Ratio-Based, 22,760 for Credit Rating-Based, totaling 44,549.

Statistical significance is indicated by asterisks on the difference values.

Second, there is another type of selection often referred to as rating shopping. A firm may approach a number of CRAs and then choose to do business with the agency that offers the most positive credit rating.Footnote ¹³ As a result, credit ratings are overly optimistic due to the unintentional errors inherent in the credit rating process rather than CRAs yielding to conflicts of interest.

Our empirical setting allows us to mitigate both of these selection concerns. Since both selection concerns are borrower-specific, we include borrower (i.e., firm) fixed effects in all specifications, effectively exploiting within-borrower variation in the impact of a credit rating downgrade on loan spreads. However, other variables likely correlated with downgrade likelihood (e.g., borrower quality, managerial optimism) are time-varying, so they are not accounted for by the borrower fixed effects. Thus, we include time-varying variables of financial health in $ {X}_{i,l,t} $ , such as the firm’s current credit rating, size (measured by the log of total assets), profitability, asset tangibility, and leverage.

In addition, because the probability of a credit rating downgrade varies with the business cycle (Bar-Isaac and Shapiro (Reference Bar-Isaac and Shapiro2013)), we include year–quarter fixed effects in $ {X}_{i,l,t} $ .Footnote ¹⁴ We also control for loan-level characteristics that are jointly determined at origination, including loan amount, number of financial covenants, whether the loan is secured, loan type (e.g., revolver or term loan), and deal purpose. Finally, in Section IV.C, we estimate a more stringent specification that includes loan fixed effects.Footnote ¹⁵

D. Sources of Variation

The most stringent specification in our initial analysis includes fixed effects for the current credit rating, borrower, and year–quarter. Thus, we exploit variation in credit rating downgrade probabilities and in borrowing costs from three separate sources. That is, we exploit variation in these variables i) across loans and across time within credit ratings, ii) across loans and across time within borrower, and iii) across loans within time. In this section, we show that, even with these tight sets of fixed effects, there is enough variation in the data to identify the effects of CRAs’ conflicts of interest in PSD.

Panel C of Table 1 shows within-group standard deviations of the dependent and independent variables that are most relevant to our regressions. The within-credit rating standard deviation of $ 1(downgrade) $ is 15.6 pp, which is very similar to its overall standard deviation of 15.8 pp. Similarly, the within-credit rating standard deviation of $ cost\hskip0.35em of downgrade $ is 10.9 bp (compared to 12.1 bp overall). Figure 3 shows the distribution of the cost of a one-notch downgrade by credit rating category at origination. Borrowers rated above A+ or below BBB− exhibit a high incidence of zero downgrade costs. In contrast, firms rated between A+ and BBB− face downgrade costs more frequently and show greater variation in these costs.

FIGURE 3Distribution of Cost of Downgrade by Rating

Figure 3 shows the distribution of the cost of a one-notch credit rating downgrade across the different credit rating groups (at origination). The bottom row tabulates the fraction of observations within each credit rating group that have a cost of downgrade of zero.

FIGURE 3 Long description

The Y axis is labeled Cost of downgrade (in b p) with a scale from 0 to 100. The X axis contains four credit rating groups. From left to right, the data shows:

* Greater than A plus: The box is very low, near 0. The percentage with 0 cost is 70 percent. There are two outliers near 12 and 25 b p.

* Bracket A plus, A minus bracket: The box ranges from approximately 5 to 10 b p with a median around 6. The percentage with 0 cost is 13 percent. Several outliers extend up to 45 b p.

* Bracket B B B plus, B B B minus bracket: The box ranges from approximately 10 to 22 b p with a median around 15. The percentage with 0 cost is 6 percent. Numerous outliers are clustered between 40 and 100 b p.

* Less than B B B minus: The box is larger, ranging from 0 to 25 b p. The percentage with 0 cost is 62 percent. Outliers are present at 75 and 100 b p.

Panel C of Table 1 also shows that the time required for the firm to be downgraded follows the same pattern as downgrades and the cost of a downgrade and that there is significant variation in the previous three variables within firm and within year. Overall, there is more than sufficient variation in the data to expect that our specifications will have power.

A. Probability of Downgrades

Table 3 presents the results from estimating different variants of equation (1). Recall that the dependent variable is $ 1(downgrade) $ , which indicates whether a borrower is downgraded by a CRA in a given quarter. The independent variable of interest is $ cost\hskip0.35em of downgrade $ , a variable that measures a firm’s increase in borrowing costs following a one-notch downgrade. For ease of exposition, $ 1(downgrade) $ is multiplied by 100, and $ cost\hskip0.35em of downgrade $ is standardized so that the coefficients represent the effect (in percentage points) on the probability of a downgrade associated with changing the cost variable by one standard deviation. Standard errors are clustered at the borrower (i.e., the firm) level.

TABLE 3Probability of Downgrade and Cost of Downgrade

TABLE 3 Long description

The table presents regression results for the dependent variable 1 open parenthesis Downgrade close parenthesis across four columns labeled 1 through 4.

* Row 1: Cost of downgrade. Values are minus 0.755 in model 1, minus 0.701 in model 2, minus 0.617 in model 3, and minus 0.674 in model 4. All values are significant at the 1 percent level as indicated by three asterisks.

* Row 2: Standard errors for cost of downgrade. Values are 0.237, 0.213, 0.210, and 0.209 respectively.

* Row 3: Firm F E. Yes for all models.

* Row 4: Year dash quarter F E. Yes for all models.

* Row 5: Rating F E. No for model 1; Yes for models 2, 3, and 4.

* Row 6: Loan controls. No for models 1 and 3; Yes for models 2 and 4.

* Row 7: Firm controls. No for models 1 and 2; Yes for models 3 and 4.

* Row 8: Number of observations. 22,760 for all models.

* Row 9: R-squared. 0.09 for model 1, 0.11 for model 2, 0.12 for model 3, and 0.12 for model 4.

* Row 10: Mean of dependent variable. 2.57 for all models.

Column 1 of Table 3 presents the results from our most basic specification, which includes only firm and year–quarter fixed effects. The point estimate on $ cost\hskip0.35em of downgrade $ is $ - $ 0.76, indicating that a one standard deviation increase in borrowing costs in the eventuality of a one-notch downgrade decreases the quarterly downgrade likelihood by almost 0.8 pp. This effect is statistically significant at the 1% level and is economically important: It represents a 30% decrease relative to the sample mean quarterly downgrade likelihood of 2.57%.

The previous specification presents initial evidence that CRAs are more reluctant to downgrade the credit ratings of borrowers who are the most likely to be negatively affected. However, one remaining concern is that the various features of loan contracts are determined at origination, and these features could correlate with the likelihood of future downgrades. To address this concern, column 2 adds controls for loan-level characteristics such as loan amount, loan type (revolver or term loan), the number of financial covenants, and deal purpose.Footnote ¹⁶ Moreover, the specification also includes issuer credit rating fixed effects that capture the average probability of a downgrade at each credit rating level. The coefficient of $ cost\hskip0.35em of downgrade $ changes slightly to $ - $ 0.70 pp.

Another remaining concern is that borrower quality is time variant and therefore not accounted for by the firm fixed effects. Thus, to mitigate this concern, columns 3 and 4 include firm-level control variables that capture a borrower’s time-varying ability to repay. Specifically, the specifications include firm size (i.e., log of assets), leverage, profitability (i.e., return on assets), and asset tangibility (i.e., intangibles divided by assets). In column 3, we replace the loan-level controls with the firm-level controls. The point estimate on $ cost\hskip0.35em of downgrade $ is $ - $ 0.62 pp. Finally, column 4 presents our most complete specification, which includes the full set of fixed effects and control variables. The coefficient of $ cost\hskip0.35em of downgrade $ is $ - $ 0.67 pp, statistically significant at the 1% level. Overall, the results in Table 3 show that CRAs are significantly less likely to downgrade their clients when the downgrade is more costly to the client.

In the Supplementary Material, we present a series of robustness checks for the previous results. Table IA.2 in the Supplementary Material shows that results hold in a subsample of loans that excludes loans rated above A $ + $ and below BBB $ - $ , that is, those ratings where zero downgrade costs are more prevalent. Table IA.3 in the Supplementary Material shows that the results hold for the subsample of revolver loans, with slightly weaker significance ( $ p $ -value of 5.1%) in the sample of term loans.Footnote ¹⁷ Tables IA.4 and IA.5 in the Supplementary Material show that the main results are robust to alternative constructs of the variable for the costs of a downgrade. Specifically, these alternate constructs are i) constructing the variable based on two-notch downgrades (instead of one notch) and ii) constructing the variable as a dollar cost divided by assets. In Table IA.6 in the Supplementary Material, we analyze the firms rated exclusively by S&P and Moody’s separately, and we find similar regression coefficients as Table 3 in both samples. In Table IA.7 in the Supplementary Material, we conduct a placebo test using a sample of accounting ratio-based PSD loans and find that the steepness of the pricing grid is not associated with a higher probability of a credit rating downgrade for these loans, where conflicts of interest for CRAs are absent.

Finally, in Table IA.8 in the Supplementary Material, we examine the effect of cost of downgrade on a related outcome variable, namely, the number of quarters between origination and the eventual downgrade. Consistent with the previous results, the table’s most stringent specification (in column 4) shows that a one standard deviation increase in the borrowing costs following a one-notch downgrade is associated with 155 additional days (1.72  $ \times $  90) before the borrower is downgraded by a CRA.

B. Profitability of Borrowers and Decisive Credit Ratings

Next, we investigate whether CRAs are more likely to yield to conflicts of interest when rating clients that are potentially more profitable for the agency. We consider two measures of client profitability: i) the natural logarithm of the total volume of loans issued by the firm in the past 4 years and ii) an indicator variable for whether the firm is a high-yield issuer (i.e., an issuer rated BB+ or lower).Footnote ¹⁸ We supplement our main specification with these measures as well as with the interaction of these variables and $ cost\hskip0.35em of downgrade $ .

Table 4 presents the estimation results. Specifically, the negative (statistically significant) coefficients on the interaction terms in columns 1 and 2 indicate that CRAs are more likely to yield to conflicts of interest when dealing with more profitable clients. This result is confirmed by the negative coefficient on the triple interaction in column 3 of Table 4.

TABLE 4Probability of Downgrade and Cost of Downgrade: Profitability of Clients

TABLE 4 Long description

The table presents coefficients and standard errors for three regression models.

Row 1: Cost of downgrade. Model 1: minus 0.256 (0.243). Model 2: minus 0.509 with two asterisks (0.210). Model 3: minus 0.183 (0.247).

Row 2: Cost of downgrade times loan volume. Model 1: minus 0.029 with two asterisks (0.012). Model 2: empty. Model 3: minus 0.021 with one asterisk (0.012).

Row 3: Cost of downgrade times 1(high-yield issuer). Model 1: empty. Model 2: minus 2.459 with three asterisks (0.874). Model 3: 0.159 (1.225).

Row 4: Cost of downgrade times loan volume times 1(high-yield issuer). Model 1: empty. Model 2: empty. Model 3: minus 0.423 with three asterisks (0.090).

Row 5: Loan volume. Model 1: minus 0.012 (0.019). Model 2: empty. Model 3: minus 0.017 (0.018).

Fixed Effects and Controls: All three models include Yes for Firm F E, Year-quarter F E, Rating F E, Loan controls, and Firm controls.

Summary Statistics:

Number of observations: 22,760 for all models.

R-squared: 0.13 for all models.

Mean of dependent variable: 2.57 for all models.

Finally, we exploit variation in the CRAs issuing the decisive credit rating, which occurs when i) a borrower is rated by only one CRA or ii) two CRAs issue different ratings, with the higher rating being decisive (following Tchistyi et al. (Reference Tchistyi, Yermack and Yun2011)). If conflicts of interest affect downgrade decisions, the decrease in downgrade likelihood should concentrate when the CRA’s rating determines borrowing costs.

To test this, we introduce an indicator variable $ 1\left( decisive rating\right) $ , which is 1 when the CRA is decisive. We then interact this variable with $ cost\hskip0.35em of downgrade $ . The results, shown in Table IA.9 in the Supplementary Material, indicate that CRAs are more hesitant to downgrade when their rating is decisive. For example, when S&P is decisive, the interaction term is statistically significant at $ - $ 1.2 pp, suggesting that S&P is particularly hesitant to downgrade in these cases. A similar pattern is observed for Moody’s.

In sum, these results suggest that CRAs are more reluctant to downgrade when their rating directly affects borrowing costs.

C. Loan Fixed Effects Regressions

Our previous analysis relies on within-borrower (i.e., firm) variation as a primary source of variation for identification purposes. This reliance on borrower fixed effects is intuitive from an economic perspective and consistent with the pricing analysis that we conduct below, which does not allow accommodating a tighter set of fixed effects. However, in the context of our probability of downgrade regressions, the possibility remains that borrowers self-select into particular types of PSD contracts at different points in time. To address this concern, we estimate a more stringent regression specification that includes loan fixed effects (instead of firm fixed effects). This allows accounting for unobserved dimensions of loan contracts and selection into contracts at origination (i.e., inference stems from borrowers who experience a change in cost of downgrade after loan parameters are set).

The estimation results are presented in Table 5. Specifically, column 1 shows a statistically significant coefficient on $ cost\hskip0.35em of downgrade $ of an even greater magnitude than that in column 4 of Table 3. The coefficient of $ - $ 1.43 indicates that a one standard deviation increase in borrowing costs in the eventuality of a one-notch downgrade decreases the quarterly downgrade likelihood by slightly over 1.4 pp. Once again, results strongly support the idea that CRAs yield to the conflicts of interest that result from the issuer–pays business model of credit ratings.

TABLE 5Loan Fixed Effects Regressions

TABLE 5 Long description

The table presents results for the dependent variable 1(Downgrade) across four numbered columns. Columns are grouped by their source: Column 1 refers to Table 3, while Columns 2, 3, and 4 refer to Table 4.

Independent Variables:

* Cost of downgrade: -1.426 (standard error 0.642) in Column 1; -0.921 (0.702) in Column 2; -0.915 (0.652) in Column 3; -0.508 (0.717) in Column 4.

* Cost of downgrade times loan volume: -0.031 (0.022) in Column 2; -0.012 (0.023) in Column 4.

* Cost of downgrade times 1(high-yield issuer): -5.244 (1.709) in Column 3; -2.444 (1.705) in Column 4.

* Cost of downgrade times loan volume times 1(high-yield issuer): -0.462 (0.105) in Column 4.

* Loan volume: -0.033 (0.025) in Column 2; -0.039 (0.025) in Column 4.

Fixed Effects and Controls (All columns marked ‘Yes’):

* Loan F E

* Year-quarter F E

* Rating F E

* Firm controls

Summary Statistics:

* Number of observations: 22,650 for all columns.

* R-squared: 0.11 for Columns 1, 2, and 3; 0.12 for Column 4.

* Mean of dependent variable: 2.57 for all columns.

Columns 2–4 of Table 5 report regressions analogous to those in Table 4. All coefficients of interest except that when profitability of borrowers is measured using loan volume (column 2) remain strong and statistically significant.Footnote ¹⁹ Overall, the inferences from previous sections remain unchanged with the inclusion of loan fixed effects in our regressions.Footnote ²⁰

D. Potential Alternative Explanations

Our evidence indicates that CRAs are significantly less likely to downgrade credit ratings when the downgrade is more costly to the client. This result is consistent with the proposition that CRAs yield to the conflicts of interest that result from the issuer–pays business model of credit ratings. This interpretation is further supported by the fact that the previous result i) is driven by those instances in which the CRA is the decisive one for the determination of the borrowing costs, ii) is stronger when CRAs rate clients that are potentially more profitable for the agency, and iii) holds when exploiting variation in the cost of downgrade within loan contracts. In this section, we address a number of potential challenges to our interpretation of the results.

1. Selection into Steeper Pricing Grids

A first potential challenge to our interpretation of the results is that managers with positive inside information about future firm fundamentals may not just self-select into PSD contracts, but also into steeper pricing grids (Begley (Reference Begley2012)). While the loan fixed effects regressions account for the average steepness of the loan’s pricing grid, we further explore this concern. To do this, we construct a measure of firm creditworthiness based on firm fundamentals. First, we regress observed credit ratings on firm variables, including leverage, size, profitability, cash flow, and asset tangibility, as well as industry and year–quarter fixed effects. Then, we use this model to generate a predicted value, that is, our measure of firm creditworthiness. If managers have inside information about an imminent improvement in firm fundamentals after loan issuance, firm creditworthiness should improve after issuance. Thus, if selection into steeper pricing grids is affecting our previous results, the measure of creditworthiness should decrease more for firms with loans with high cost of downgrade.Footnote ²¹

Figure 4 plots the progression of the measure of creditworthiness over the 4-year period following loan issuance for firms with high (above median) and low (below median) cost of downgrade.Footnote ²² The figure shows overall a slight deterioration of creditworthiness during the lifetime of the loan that affects both groups of firms. In addition, there are no distinguishable differences in the progression of the measure between high- and low-cost downgrade firms. In fact, the average difference in the measure of creditworthiness across both groups of firms is only 0.18 notches.

FIGURE 4Borrower Creditworthiness Over Time

Figure 4 shows the measure of creditworthiness for firms with loans with high (i.e., above median) and low (i.e., below median) costs of downgrade at origination. To construct the measure of creditworthiness, observed (numeric) credit ratings are regressed on firm variables, including leverage, size, profitability, cash flow, and asset tangibility, as well as industry and year fixed effects. Numeric credit ratings reflect the number of notches away from AAA (rating = 1), with a rating of 9 corresponding to BBB and a rating of 8 corresponding to BBB+. The measure of creditworthiness comes from obtaining the fitted values from the regression model. The average cost of a downgrade for the high-cost group (circles) is 22.3 bp, whereas the average cost of a downgrade for the low-cost group (triangles) is 6.7 bp. The analysis is restricted to the first 4 years of each loan’s life. The shaded areas denote the 95% confidence interval for each estimation.

FIGURE 4 Long description

The x-axis is labeled Quarters since loan issuance, ranging from 0 to 16 in increments of 4. The y-axis is labeled Fundamentals-implied creditworthiness (A A A = 1), ranging from 8 to 10 in increments of 0.5.

Two data series are plotted with shaded 95% confidence intervals:

* High cost of downgrade: Represented by a red line with circular markers. It begins at a numeric rating of 9.0 at quarter 0, remains relatively flat until quarter 4, and then trends upward steadily to approximately 9.4 by quarter 16.

* Low cost of downgrade: Represented by a blue line with triangular markers. It starts lower at approximately 8.85 at quarter 0, dips slightly until quarter 3, then follows a similar upward trajectory as the high-cost group, ending at approximately 9.2 at quarter 16.

The high-cost group consistently maintains a higher numeric rating (indicating lower creditworthiness) than the low-cost group throughout the 16-quarter period. The shaded confidence intervals for both groups overlap significantly between quarters 4 and 16.

Next, we provide two additional tests to further mitigate selection concerns. First, we exploit the fact that managers are more likely to possess insider information that is relevant for the short run than for the long run (e.g., Larcker, Levy, Quinn, Tayan, and Taylor (Reference Larcker, Levy, Quinn, Tayan and Taylor2021)).Footnote ²³ Specifically, in Table 6, we repeat our main analysis while dropping observations 1, 2, 4, and 8 quarters after loan origination. If managers select into steeper pricing grids because of positive insider information, the relationship between cost of downgrade and probability of downgrade should concentrate in the quarters shortly after loan origination and therefore excluding the early quarters from the sample should weaken the effects shown in Table 3. Inconsistent with this concern, Table 6 shows that the relationship between cost of downgrade and probability of downgrade remains essentially the same irrespective of the time window that we investigate.

TABLE 6Probability of Downgrade and Cost of Downgrade: Different Time Windows

TABLE 6 Long description

A data table with four columns labeled 1 through 4, representing time windows excluding the 1st Quarter, 1st Halfyear, 1st Year, and 2nd Years respectively.

Row 1: Cost of downgrade coefficients.

- Column 1: minus 0.635 with three asterisks (standard error 0.204).

- Column 2: minus 0.586 with three asterisks (standard error 0.222).

- Column 3: minus 0.726 with three asterisks (standard error 0.251).

- Column 4: minus 0.590 with two asterisks (standard error 0.283).

Rows 2 through 6: Fixed Effects and Controls.

- Firm F E, Year dash quarter F E, Rating F E, Loan controls, and Firm controls are all marked as Yes for all four columns.

Row 7: Number of observations.

- Column 1: 20,657.

- Column 2: 18,759.

- Column 3: 15,275.

- Column 4: 9,856.

Row 8: R-squared values.

- Column 1: 0.13.

- Column 2: 0.12.

- Column 3: 0.12.

- Column 4: 0.11.

Row 9: Mean of dependent variable.

- Column 1: 2.24.

- Column 2: 2.04.

- Column 3: 2.00.

- Column 4: 1.79.

Second, we follow the corporate governance literature and exploit measures of CEO optimism based on options holdings. Specifically, we proxy for CEO optimism using i) the fraction of CEOs that hold 67% deep-in-the-money options (Campbell, Gallmeyer, Johnson, Rutherford, and Stanley (Reference Campbell, Gallmeyer, Johnson, Rutherford and Stanley2011)) and ii) the fraction of CEOs that hold in-the-money options with less than one year to expiration (Malmendier and Tate (Reference Malmendier and Tate2008)). We add interactions between these measures of CEO optimism and cost of downgrade to our main specification and report the estimation results in Table IA.11 in the Supplementary Material. If CEOs with positive information regarding their firm prospects select into steeper pricing grids, one would expect that the majority of the effect of cost of downgrade is driven by optimistic CEOs (i.e., the coefficients on the interaction terms should be negative and the coefficient on the standalone cost of downgrade should lose its predictive power). However, this is not what Table IA.11 in the Supplementary Material shows. Overall, the results in this section are inconsistent with the proposition that our previous results are driven by self-selection of executives with positive inside information into steeper pricing grids.

Finally, one implication of the previous results is that credit rating inflation should cause ratings to become less accurate over time. That is, a wedge should eventually occur between objective credit quality and nominal credit ratings, similar to the effect caused by credit rating inflation in the RMBS market (Griffin, Nickerson, and Tang (Reference Griffin, Nickerson and Tang2013)). We examine this possibility in Table IA.12 in the Supplementary Material. For each firm in our sample, we proxy for credit quality using Merton’s distance-to-default and then estimate regressions of distance to default on the cost of downgrade. A higher cost of downgrade is associated with a lower distance to default, though this relationship is statistically significant only in the specification with industry-year fixed effects and not in the baseline regression. This approach is appropriate, as default risk is typically highly correlated within industries at any given time, owing to common economic shocks and similar cost and revenue structures among peers. Within each credit rating notch, the pattern in the more stringent specification is generally consistent with the idea that conflicts of interest distort the incentives of CRAs to accurately rate firms.

2. Undisclosed Firm Information

A second potential challenge to our interpretation of the results is that the results could be driven by firms hiding negative financial information from the CRAs rather than by CRAs catering to their clients. CRAs issue credit ratings based on information provided by a borrower, and they do not verify this information (see Section C of the Supplementary Material). Firms whose borrowing costs would experience the highest interest rate increase in the eventuality of a credit rating downgrade have the strongest incentive to hide detrimental information, which could explain why CRAs are less likely to downgrade firms when the downgrade is potentially more costly.

To address this concern, we exploit an observable negative shock to the creditworthiness of a subset of the firms in our sample. Specifically, between the third quarter of 2014 and the end of 2015, commodities experienced a dramatic loss in value, with the Dow Jones Commodity Index plummeting by 50%. Arguably, this loss negatively affected firms that depended on commodity values to generate profits, such as firms from the oil, gas, and mining sectors. Consistent with this proposition, there was significant public concern about the prospects of these affected firms during this time.Footnote ²⁴ The idea of this test is to compare the CRAs’ behavior when they rate firms that were affected by the commodity shock (which the public saw as deteriorated) with the CRAs’ behavior when rating the remaining, unaffected firms. Since the shock was visible and highly public, the ability of borrowers in the affected sectors to hide adverse information was particularly low. If our results are driven by unobserved (i.e., hidden) firm information, we should see a weaker link between the cost of a downgrade and the likelihood of a downgrade among borrowers in the affected sectors during this downturn.Footnote ²⁵

In Table 7, we supplement our main specification with an indicator variable for firms in industry sectors that depend on commodity values ( $ 1(commodities) $ ), an indicator variable that captures the time when commodity values fell ( $ 1\left( commodities\hskip0.35em shock\right) $ ), and the interaction of these indicators with $ cost\hskip0.35em of\hskip0.35em downgrade $ .Footnote ²⁶ The coefficient on the standalone $ cost\hskip0.35em of downgrade $ remains statistically significant and similar in magnitude to the estimates in Table 3 in all specifications. Consistent with falling commodity values adversely affecting borrowers, the coefficient on $ 1(commodities)\times 1\left( commodities\hskip0.35em shock\right) $ is positive, ranging from 0.7 to 6.2 pp. However, the coefficient on the triple interaction ranges from $ - $ 5.7 to $ - $ 3.9 pp (always statistically significant at the 1% level), indicating that downgrades were concentrated among borrowers for whom the cost of a downgrade is lower. Since arguably all market participants were well aware of these commodity setbacks, these results are inconsistent with the idea that our main results are driven by firms hiding negative financial information. In contrast, these results are consistent with conflicts of interest.

TABLE 7Probability of Downgrade and Cost of Downgrade: Observable Shock to Borrower Quality

TABLE 7 Long description

The table presents regression results for the dependent variable 1(Downgrade) across four columns labeled 1 through 4.

Key variables and coefficients (standard errors in parentheses):

* Cost of downgrade: -0.629*** (0.238) in model 1; -0.640*** (0.217) in model 2; -0.540** (0.211) in model 3; -0.626*** (0.215) in model 4.

* 1(commodities) times 1(commodities shock): 6.279*** (1.186) in model 1; 3.994*** (1.023) in model 2; 1.121 (1.318) in model 3; 0.841 (1.293) in model 4.

* Cost of downgrade times 1(commodities) times 1(commodities shock): -5.774*** (1.442) in model 1; -4.476*** (1.375) in model 2; -4.048*** (1.492) in model 3; -4.024*** (1.498) in model 4.

* Cost of downgrade times 1(commodities shock): -0.453 (0.620) in model 1; -0.111 (0.576) in model 2; -0.044 (0.552) in model 3; 0.018 (0.553) in model 4.

Fixed Effects (F E) and Controls:

* Firm F E: Yes for all models.

* Year-quarter F E: Yes for all models.

* Rating F E: No for model 1; Yes for models 2, 3, and 4.

* Loan controls: No for models 1 and 3; Yes for models 2 and 4.

* Firm controls: No for models 1 and 2; Yes for models 3 and 4.

* Other interactions: Yes for all models.

Summary Statistics:

* Number of observations: 22,760 for all models.

* R-squared: 0.09 for model 1; 0.11 for model 2; 0.12 for model 3; 0.13 for model 4.

* Mean of dependent variable: 2.57 for all models.

We further examine the possibility that the hiding of negative financial information drives our results in the Supplementary Material. First, we conduct a test based on earnings surprises. Following a similar idea to that of our previous test, we use negative earnings surprises as an unanticipated shock to firm information and quality. Results are presented in Table IA.15 in the Supplementary Material. Similar to the results in Table 7, Table IA.15 in the Supplementary Material shows that in times of negative earnings surprises, the negative relationship between cost of downgrade and probability of downgrade is strongly amplified.

Second, we exploit cross-sectional differences in the ease with which firms can hide or manipulate financial information. Specifically, the idea of this test is to compare i) the CRAs’ behavior when rating firms that are relatively more opaque in nature (i.e., firms that have more room to manipulate their financials) to ii) the CRAs’ behavior when rating the remaining, less opaque firms.

We use a firm’s intangibles divided by total assets and a firm’s R&D expenses as proxies for a firm’s “opaqueness” (Cañibano, Garcia-Ayuso, and Sanchez (Reference Cañibano, Garcia-Ayuso and Sanchez2000), Wyatt (Reference Wyatt2005)). In columns 1 and 2 of Table IA.16 in the Supplementary Material, we supplement our main specification with indicator variables for above-median values of these proxy variables, as well as with the interaction between these indicators and $ cost\hskip0.35em of downgrade $ . The coefficients on both interaction terms are statistically insignificant, inconsistent with the proposition that our main results are driven by firms hiding negative financial information from the CRAs.

5. Internalization of Externalities

The final challenge to our interpretation of the results is that they could be driven by CRAs internalizing negative externalities to the firms they downgrade. Specifically, a credit rating downgrade can reduce a firm’s access to credit, which subsequently may undermine the firm’s financial health, thereby leading to a self-fulfilling prophecy of creditworthiness deterioration and downgrades (Manso et al. (Reference Manso, Strulovici and Tchistyi2010)). While CRA guidelines explicitly forbid considering the effects of rating actions when issuing or changing ratings (see Section C.4 of the Supplementary Material for excerpts from CRA guidelines), the possibility remains that CRAs differentially deviate more from their guidelines when rating firms with a higher cost of downgrade. While the negative externalities of downgrades should be most prevalent for firms that are close to losing their investment grade classification and thus the fact that our results hold for borrowers irrespective of their initial credit rating mitigates the aforementioned concern (see column 3 of Table IA.16 in the Supplementary Material), we implement two additional empirical tests.

First, we exploit the fact that the internalization of externalities concern should be absent for credit rating upgrades, since these only have positive externalities for borrowers. To test this idea, we first construct a sample of loans with upgrade grids, that is, loans where borrower credit rating at origination is toward the lower end of the grid.Footnote ²⁸ We define two variables: i) $ 1(upgrade) $ , which takes the value 1 if the borrower experiences a credit rating upgrade by a CRA in a given quarter, and ii) benefit of upgrade, which measures the lowering in interest rate spreads following a one-notch increase in a credit rating. We then regress $ 1(upgrade) $ on the benefit of an upgrade and report the estimation results in Table 8. Column 1 uses the full sample, while column 2 restricts the sample to loans with upgrade grids.

TABLE 8Probability of Upgrade and Benefit from Upgrade

TABLE 8 Long description

The table consists of two columns labeled 1 and 2 under the primary header 1 open parenthesis Upgrade close parenthesis.

Row 1: Benefit of upgrade times 1 open parenthesis upgrade grid close parenthesis. Column 1 is 0.763 with a standard error of 0.996. Column 2 is empty.

Row 2: Benefit of upgrade. Column 1 is minus 0.484 with three asterisks and a standard error of 0.165. Column 2 is 2.107 with one asterisk and a standard error of 1.234.

Row 3: 1 open parenthesis upgrade grid close parenthesis. Column 1 is 2.651 with three asterisks and a standard error of 0.988. Column 2 is empty.

Row 4: Firm F E. Both columns are Yes.

Row 5: Year minus quarter F E. Both columns are Yes.

Row 6: Rating F E. Both columns are Yes.

Row 7: Loan controls. Both columns are Yes.

Row 8: Firm controls. Both columns are Yes.

Row 9: Number of observations. Column 1 is 22,702. Column 2 is 1,889.

Row 10: R-squared. Column 1 is 0.05. Column 2 is 0.15.

Row 11: Mean of dependent variable. Column 1 is 1.70. Column 2 is 3.71.

We find that loans with upgrade grids are more likely to receive upgrades, though the effect is statistically insignificant in the full sample and significant at the 10% level in the subsample. When we include loan fixed effects (Table IA.19 in the Supplementary Material), the effect in the full sample becomes significant at the 5% level.Footnote ²⁹

Next, building on the previous test, we estimate an ordered logit regression, which allows us to incorporate both the cost of downgrades and the benefit of upgrades using a continuous variable. The dependent variable is the change in credit rating, defined such that positive values indicate downgrades and negative values indicate upgrades. Table 9 presents the results. Column 1 reports OLS estimates for completeness and as a benchmark, while column 2 reports the ordered logit results.Footnote ³⁰ Consistent with our earlier findings, the coefficients on the cost of downgrade are negative in both specifications.

TABLE 9Ordered Logit Regression

TABLE 9 Long description

The table consists of three columns. The first column lists variables and controls, while the second and third columns represent results for O L S 1 and Ordered Logit 2 respectively under the header Worsening in Rating.

* Cost of downgrade: O L S 1 shows minus 0.013 with two asterisks (standard error 0.006). Ordered Logit 2 shows minus 0.141 with two asterisks (standard error 0.066).

* Firm F E: Yes for O L S 1, No for Ordered Logit 2.

* Industry F E: No for O L S 1, Yes for Ordered Logit 2.

* Year-quarter F E: Yes for both models.

* Rating F E: Yes for both models.

* Loan controls: Yes for both models.

* Firm controls: Yes for both models.

* Number of observations: 22,760 for both models.

* R-squared: 0.11 for O L S 1, not applicable for Ordered Logit 2.

* Mean of dependent variable: 0.02 for both models.

Overall, these results are not only inconsistent with the idea that CRAs systematically internalize negative borrower externalities but also suggest that any conflicts of interest may be, to some extent, symmetric.

Second, if CRAs were to internalize the externalities of their downgrade decisions, CRAs should be especially hesitant to issue downgrades when uncertainty is high, for fear of causing unnecessary spillover damages to firms. Thus, if our results are driven by CRAs being more conservative when rating firms with a higher cost of downgrade, then these firms should experience fewer credit rating reversals after a downgrade. To test this conjecture, we introduce the indicator $ 1(reversal) $ , which takes the value 1 if a firm receives a credit rating that offsets a previous credit rating change during the lifetime of the same loan (i.e., there was a previous credit rating downgrade followed by a credit rating upgrade or vice versa), and 0 otherwise. We then estimate regressions of $ 1(reversal) $ on $ cost\hskip0.35em of downgrade $ and report the estimation results in Table IA.21 in the Supplementary Material. We find no relationship between cost of downgrade and the reversal of credit rating decisions.

E. Additional Proxy for Conflicts of Interest

In this section, we consider that, since loans are often rolled over at maturity, a credit rating downgrade shortly before the maturity date can be particularly costly for the borrower, as it negatively affects the terms of a new loan. Accordingly, one would expect loans closer to maturity to be less likely to be downgraded. This relationship also provides a useful benchmark for interpreting the cost-of-downgrade results.

In Table 10, we regress the downgrade indicator on a continuous time-to-maturity variable (column 1) and its interaction with the cost of downgrade (column 2). The results show that loans closer to maturity are less likely to be downgraded, consistent with the idea discussed above, and this effect remains robust after controlling for the cost of downgrade. In economic terms, each additional quarter of remaining maturity is associated with an increase in the probability of downgrade of 9.8–10.9 basis points.

TABLE 10Probability of Downgrade and Time-to-Maturity

TABLE 10 Long description

The table presents regression results for the dependent variable 1 open parenthesis Downgrade close parenthesis across two columns labeled 1 and 2.

Row 1: Time to maturity open parenthesis quarters close parenthesis shows a coefficient of 0.098 with three asterisks in column 1 and 0.109 with three asterisks in column 2. Standard errors are 0.024 for both.

Row 2: Cost of downgrade multiplied by time to maturity is only present in column 2 with a coefficient of minus 0.061 with two asterisks and a standard error of 0.027.

Row 3: Cost of downgrade is only present in column 2 with a coefficient of minus 0.418 with one asterisk and a standard error of 0.244.

The following rows indicate fixed effects and controls, all marked as Yes for both columns: Firm F E, Year minus quarter F E, Rating F E, Loan controls, and Firm controls.

The bottom section provides summary statistics:

Number of observations is 22,756 for both columns.

R squared is 0.12 for column 1 and 0.13 for column 2.

Mean of dependent variable is 2.57 for both columns.

The previous results provide an alternative perspective on the effect of the cost of downgrade: as shown in Section IV.A, a one standard deviation increase in borrowing costs in the event of a one-notch downgrade reduces the quarterly probability of downgrade by nearly 0.8 pp, equivalent to shortening loan maturity by approximately 7.4 quarters. Overall, these findings support the conclusion that CRAs face significant conflicts of interest, which meaningfully affect their rating behavior.

A. Loan Pricing and the Cost of Downgrades in the Primary Market

We investigate the relationship between loan spreads and pricing grid design at origination. We introduce the variable $ average\hskip0.35em cost\hskip0.35em of downgrade $ , computed as the average cost of downgrade across all rating levels in the initial grid. Thus, this variable is a measure of the ex ante (i.e., expected) conflicts of interest of CRAs captured by the average steepness of the pricing grid, as opposed to a measure of the ex post (i.e., realized) conflicts of interest of CRAs over the life of the loan.Footnote ³¹ Table 11 shows the results of regressions in which the dependent variable is the loan’s interest rate spread at origination and the independent variable of interest is $ average\hskip0.35em cost\hskip0.35em of downgrade $ . As in previous analyses, the variable is standardized, and standard errors are clustered by firm. We retain only one observation per loan (i.e., at origination). Importantly, while the four specifications in Table 11 vary in terms of control variables and fixed effects, they all include firm fixed effects. Thus, the estimations effectively compare loan contracts with different pricing grids by the same borrower.

TABLE 11Loan Pricing at Initiation and Conflicts of Interest

TABLE 11 Long description

The table presents regression results for Loan Spread across four columns labeled 1 through 4.

* Row 1: Avg. cost of downgrade. Column 1 is 4.683 with three asterisks. Column 2 is 3.823 with three asterisks. Column 3 is 4.106 with three asterisks. Column 4 is 3.685 with three asterisks.

* Row 2: Standard errors for Avg. cost of downgrade. Column 1 is 0.576. Column 2 is 0.438. Column 3 is 0.512. Column 4 is 0.425.

* Row 3: Firm F E. All columns are marked Yes.

* Row 4: Year-quarter F E. All columns are marked Yes.

* Row 5: Rating at origination F E. Column 1 is No. Columns 2, 3, and 4 are Yes.

* Row 6: Loan controls. Column 1 is No. Column 2 is Yes. Column 3 is No. Column 4 is Yes.

* Row 7: Firm controls. Column 1 is No. Column 2 is No. Column 3 is Yes. Column 4 is Yes.

* Row 8: No. of obs. Column 1 is 1,702. Column 2 is 1,697. Column 3 is 1,701. Column 4 is 1,697.

* Row 9: R-squared. Column 1 is 0.77. Column 2 is 0.82. Column 3 is 0.81. Column 4 is 0.83.

* Row 10: Mean of dependent variable. Column 1 is 115.94. Columns 2, 3, and 4 are 116.00.

The point estimates on $ average\hskip0.35em cost\hskip0.35em of downgrade $ are positive and statistically significant at the 1% level in the four specifications, ranging from 3.7 to 4.7 bp. Thus, the regression results are consistent with borrowers and lenders being aware of and pricing the conflicts of interest of CRAs.

The previous result can seem somewhat surprising at first, since it indicates that, all else being equal, a steeper pricing grid is detrimental to the borrower (i.e., borrowing costs are higher at origination). Since interest rates increase when the borrower is downgraded, PSD loans effectively embed an insurance mechanism for the lender against the deterioration of the borrower’s financial health. If the pricing grid is steeper, the insurance payment is higher. Intuitively, a loan with a larger premium should be cheaper for the borrower, not more expensive. However, the previous result can be understood in the context of our analysis. Our analysis compares only loans that have a pricing grid to begin with. Specifically, our regressions compare loan spreads across loans that have pricing grids of varying steepness, and the coefficient estimates reflect the decreasing relative efficiency of pricing grids. Compared to a very small increase in interest rates for downgrades, a large increase in interest rates after downgrades is less efficient, since, as the results in Section IV indicate, CRAs are more reluctant to downgrade their clients when the costs of a downgrade are higher.

B. Loan Pricing and the Cost of Downgrades in the Secondary Market

The trading of loans in the secondary market has increased substantially in recent years (Beyhaghi and Ehsani (Reference Beyhaghi and Ehsani2017)).Footnote ³² One concern with these trades is that originating banks may have an advantage over investors in the secondary market because they have superior information regarding loan quality (Dahiya, Puri, and Saunders (Reference Dahiya, Puri and Saunders2003)). While some investors are well-informed and even have access to insider information (Addoum and Murfin (Reference Addoum and Murfin2020)), many, such as mutual funds and loan funds, are nonbank investors (who tend to be less informed). Previously, we showed that the potential for credit inflation is, to some extent, priced at origination. We now investigate whether this pricing of the potential for credit inflation persists once loans are traded in the secondary market.

We obtain secondary market loan pricing data from Refinitiv’s Loan Pricing Corporation (LPC). The secondary market data consist of self-reported information from brokers that quote daily prices on loans. While a quote does not guarantee a trade, brokers tend to make markets for more liquid loans, and any loan in the LPC database is almost surely traded at some point.Footnote ³³ We match these data to our sample of PSD loans using a proprietary linking table provided by Refinitiv. The Refinitiv linking table matches 4.1% of the loans in our sample (i.e., 75 loans) to the secondary market data. While this matching rate may seem low, it is similar to the 4.9% matching rate that we find for all loans in DealScan and to the matching rate reported by other research using similar data.Footnote ³⁴ Section A of the Supplementary Material details the matching procedure. Although the sample is small, it supports useful—albeit mostly descriptive—analyses, which are still valuable given the limited evidence on whether secondary market participants price conflicts of interest.

We construct a quarterly panel with quoted prices for the traded loans in our sample and compare the prices of PSD loans with high (i.e., above-median) downgrade cost at origination to the prices of PSD loans with low (i.e., below-median) downgrade cost at origination. Specifically, we estimate a regression where the dependent variable is loan price (measured as the average of the mid-prices quoted each quarter) and the independent variables of interest are the interactions of an indicator for above-median cost of downgrade and indicators for each quarter since loan issuance. The regression includes the same set of control variables and fixed effects as in column 4 of Table 3 (i.e., our most stringent specification) plus loan fixed effects, and we consider the first 4 years of each loan’s lifetime.Footnote ³⁵

Figure 5 shows the coefficients for each quarterly interaction along with their corresponding 95% confidence interval based on standard errors clustered by loan. We use the first quarter’s estimate as our coefficient of reference. The price difference in this quarter is 1.3 pp, meaning the zero line in this graph represents low cost of downgrade loans priced 1.3% higher than high cost of downgrade loans. Thus, since each interaction coefficient captures the average difference in price between PSD loans with high cost of downgrade and PSD loans with low cost of downgrade, a positive coefficient in a given quarter means that the differences in prices between the two groups of loans increased in that quarter relative to the difference that prevailed in the first quarter. Note that an increase in price is equivalent to a decrease in yield. Thus, a positive coefficient implies a decrease in the initial yield premium associated with loans that have a high cost of downgrade.

FIGURE 5Dynamics of Secondary Market Prices

Figure 5 shows the dynamics of the pricing of loans with high cost of downgrade in the secondary market. Loan price (measured as the average of the mid-prices quoted each quarter) is regressed on the interactions of an indicator for above-median cost of downgrade and indicators for each quarter since loan issuance. The regression includes the same set of control variables and fixed effects as in column 4 of Table 3 plus loan fixed effects, and the first 4 years of each loan’s lifetime are considered. The coefficients associated with the interactions are denoted by solid circles, and the vertical bars denote the corresponding 95% confidence interval (based on standard errors clustered by loan). The dashed vertical line denotes the average time from origination to first sale for loans with above-median cost of downgrade.

FIGURE 5 Long description

A line graph where the Y axis is labeled Effect of high cost of downgrade (% of initial price) with a scale from minus 2 to 8. The X axis is labeled Quarters since issued with a scale from 2 to 16. A horizontal red baseline is set at 0. A vertical dashed line is positioned between quarters 3 and 4.

Data points are represented by solid blue circles with vertical 95% confidence interval bars.

* At quarter 2, the point is slightly below 0.

* At quarter 3, the point is just above 0.

* From quarter 4 to quarter 10, there is a steady linear increase from approximately 0.5 to 1.5.

* At quarter 11, there is a noticeable jump to approximately 2.

* From quarters 12 to 16, the points continue a general upward trend, reaching nearly 4 by quarter 16.

* The vertical confidence interval bars widen significantly as time progresses, particularly from quarter 11 onward, indicating increased variance in the later quarters.

The figure shows that there is no change in price differences between high and low cost of downgrade loans in the first three quarters after loan origination. Coincidently, the average time from origination to first recorded sale for loans with above-median cost of downgrade is 100 days (denoted by the dashed vertical line in the figure). From the fourth quarter onward, coefficient estimates turn positive, consistent with a decrease in the premium associated with high cost of downgrade loans as they are sold to secondary market participants. On average, these loans become about 2% more expensive than low cost of downgrade loans 3 years after the first quarter of trading. Overall, the point estimates in Figure 5 are consistent with secondary market participants overpaying for above-median cost of downgrade loans (and therefore pricing less of the conflicts of interest of CRAs) relative to originating banks.

C. Probability of Trade and Time to First Trade

To complement the previous analysis, we also study whether there is a relationship between a loan’s cost of downgrade, its probability of being traded, and the speed at which it is first traded. We continue to focus on the 75 loans in our sample for which we have secondary market pricing data. Since this sample is relatively small and the nature of this analysis is cross-sectional, we lack the additional observations provided by the quarterly-panel structure from the previous section. Consequently, we display the results graphically.

We begin by investigating whether banks are more likely to sell loans with higher CRA conflicts. To facilitate the comparison between loans that are traded and loans that are not traded, we match the 75 loans with similar nontraded loans using a nearest neighbor matching technique based on all continuous control variables.Footnote ³⁶ We then split these 150 observations into three groups based on their cost of downgrade and plot the probability of a loan being sold by the originating bank for each group in Figure 6.

FIGURE 6Probability of Loan Sale

Figure 6 shows the probability of loan trade in the secondary market by terciles of average cost of downgrade. The sample consists of 75 traded loans and 75 nontraded loans. The nontraded loans are selected so that they resemble the traded loans using a nearest neighbor matching framework based on firm characteristics (current credit rating, size, profitability, asset tangibility, and leverage) and loan characteristics (amount, number of financial covenants, loan type, and whether the loan is secured or not).

The figure shows a nonlinear relationship between our measure of CRA conflicts and the likelihood of a loan being traded. Loans in the highest group of cost of downgrade have a 58% probability of being traded in the secondary market, almost 10% higher than that of the other two groups with lower downgrade costs.

Next, we investigate whether banks, conditionally on selling a loan, are selling loans with higher CRA conflicts faster. The underlying assumption for this analysis is that originating banks sell parts of the loan on the same day a dealer quotes a price for the loan for the first time. Thus, we compare the time elapsed between loan origination and when the loan gets quoted for the first time across groups of loans based on costs of downgrade. As before, we split the sample into three groups based on cost of downgrade and plot the average time-to-first-trade for each group in Figure 7.

FIGURE 7Time to First Sale

Figure 7 shows the average time-to-first-trade (days) in the secondary market by terciles of average cost of downgrade. The sample consists of 75 traded loans.

Once again, the figure shows a nonlinear relationship between our measure of CRA conflicts and the outcome variable. Loans in the lowest group of cost of downgrade stay on banks’ balance sheets for about 470 days before getting quoted, whereas loans in the other two groups experience their first quote after about 200 days, on average. This is a 270-day difference, equivalent to a 57% relative change. Overall, the results in this section are consistent with banks being more likely to sell loans with a higher potential for credit rating inflation. The results are also consistent with these loans being traded faster. Note that we only observe quotes for the most liquid loans that have brokers providing daily prices. Since 40% (i.e., about 8 times as many loans as in our sample) are traded (Beyhaghi and Ehsani (Reference Beyhaghi and Ehsani2017)), the actual number of loans and investors affected by CRA conflicts is arguably quite large.