Regulations and the “Too-Big-to-Fail” Problem: Evidence from the Dodd–Frank Act

Abstract

Before the enactment of the Dodd–Frank Act, firm size was taken into account by rating agencies in determining the credit ratings of banks. Therefore, the “too-big-to-fail” problem was, at least partially, reflected in big banks’ elevated ratings, which are more than justified by intrinsic creditworthiness. What is unclear is whether the bond market still gives an additional discount in yield to big banks over and above the lower yield spread that is already reflected in the elevated credit ratings due to their size. In this study, we examine this question and document a significant incremental yield discount for large banks even after controlling for credit ratings. Furthermore, we find that big banks with lower ratings pay lower borrowing costs than non-big banks with higher ratings. This additional discount, however, mostly disappeared after the Dodd–Frank Act.

1. Introduction

The U.S. Treasury Department defines the “too-big-to-fail” (TBTF) problem as “a situation in which regulators are unwilling to inflict losses on uninsured depositors or even non-deposit creditors in a troubled bank for fear of adverse macroeconomic consequences or the instability of the financial system as a whole” (United States Department of Treasury, 1991). In the U.S. the TBTF problem first occurred during the 1984 Continental Illinois National Bank crisis, when the bank’s total liabilities were USD 33 billion, while only USD 3 billion of deposits was insured by the Federal Deposit Insurance Corporation (FDIC). The fear that the failure of Continental Illinois may pose a systemic risk to the entire financial system prompted the Federal Reserve System, the FDIC, and the Comptroller of the Currency to work with twenty-four banks and to bail out Continental Illinois eventually.

As Berlin et al. (1991) pointed out, while there are many issues with the U.S. deposit insurance system, one of the most severe problems is that some banks are perceived as TBTF, and one of the consequences of the TBTF problem is that the market discipline on big banks is removed. Several studies have shown that the borrowing costs as measured by yield spreads on bonds issued by big banks tend to be lower than what can be justified by their risk (Anginer & Warburton, 2014; Balasubramnian & Cyree, 2011, 2014; Flannery & Sorescu, 1996; Santos, 2014). The lower cost paid by big banks, which is lower than what can be justified by their risk, is an indication of the removal of market discipline.

Furthermore, before the enactment of the Dodd–Frank Wall Street and Consumer Protection Act (henceforth the Dodd–Frank Act), credit rating agencies took firm size into account in determining the credit ratings of banks; thus, big banks tended to receive a credit rating higher than what could be justified by their creditworthiness1. Therefore, the U.S. government’s willingness to bail out large banks was, to some degree, reflected in the credit ratings received by large banks.

Since it is well known in the finance literature that firms with higher credit ratings tend to pay lower borrowing costs, large U.S. banks are therefore likely to enjoy lower interest costs through their elevated credit ratings, reflecting size-related advantages in perceived creditworthiness. What remains unclear is whether investors in the bond market apply an additional yield discount to debt issued by large banks beyond what is already reflected in credit ratings and other observable risk characteristics. This distinction between rating-based borrowing cost advantages and market-based yield discounts constitutes the central question of this paper.

The primary contribution of this paper is to disentangle rating-based borrowing cost advantages from market-based yield discounts associated with bank size and to examine how this incremental TBTF discount evolves in response to post-crisis regulatory reform. Using a comprehensive sample of U.S. bank bond issuances, we show that large banks benefit not only from higher credit ratings attributable to size, but also from an additional yield discount in bond markets that persists even after controlling for credit ratings and other risk characteristics. Moreover, we document that, prior to the Dodd–Frank Act, lower-rated large banks often borrowed at lower yields than even higher-rated smaller banks, highlighting a distortion in market pricing consistent with TBTF expectations. These results suggest that the TBTF problem is reflected not only in big banks’ credit ratings but also in bond market pricing over and above credit ratings. Finally, we find that this incremental yield advantage largely disappeared following the implementation of the Dodd–Frank Act, suggesting that post-crisis regulation has been effective in mitigating TBTF-related distortions in bond market pricing.

The rest of the paper is organized as follows: In Section 2, we review the literature; in Section 3, we outline the research methodology; in Section 4, we describe the data and the definitions of variables used in the study; in Section 5, we present the empirical results; and in Section 6, we conclude the paper.

2. Literature Review

2.1. Market Discipline and Bond Market Pricing

Flannery and Sorescu (1996) find that, after government intervention and the rescue of Continental Illinois in 1984, banks’ bond yield spreads were no longer sensitive to risks. The finding suggests that the bailout of Continental Illinois removed market discipline against banks’ risk-taking behavior and created the TBTF perception in the bond market. Balasubramnian and Cyree (2011) show that after Long Term Capital Management (LTCM) was bailed out by sixteen financial institutions coordinated by the Federal Reserve Bank of New York, the discount in yield spread paid by big banks doubled. The finding also suggests the market discipline was removed after the LTCM rescue. Baker and McArthur (2009) report that after the Congress passed the Troubled Asset Relief Program (TARP) to bail out big banks after the 2008 financial crisis, big banks’ costs of deposit are significantly lower than those of smaller banks.

Acharya et al. (2013) report that, while bond credit spreads are sensitive to risk for most financial institutions, credit spreads lack risk sensitivity for the largest institutions. The results indicate that the bondholders of major financial institutions have an expectation that the government will shield them from losses and, as a result, they do not accurately price the risk of bonds issued by the largest banks. Anginer and Warburton (2014) report that the risks are only reflected in the bond yield spreads for mid-size and small institutions, not the large ones. Their findings also suggest the removal of market discipline against large institutions and indicate the existence of a TBTF problem.

Penas and Unal (2004) find evidence of TBTF from banks’ mergers and acquisitions. They report that when a medium-sized bank acquires another bank and this pushes the size of the combined bank beyond the threshold of a TBTF bank, the reduced borrowing cost after the acquisition is highest—higher than the mega-banks that have already reached the TBTF status, and higher than small banks acquiring another bank—but the combined bank size still remains below the TBTF threshold. Brewer and Jagtiani (2013) find that banks are even willing to pay an additional premium for mergers so that they can reach the status of being commonly viewed as “too-big-to-fail”. More broadly, the TBTF literature is closely related to macro-financial research on financial intermediation, cyclical risk-taking, and regulatory regimes. Konstantakopoulou (2023) shows that regulatory conditions shape banks’ risk-taking over the business cycle, with weaker regimes amplifying systemic risk. Our findings complement this work by showing that the Dodd–Frank Act reduced TBTF-related pricing distortions in bond markets and strengthened market discipline.

2.2. Credit Rating Agency Behavior and Size-Related Rating Effects

A related strand of the literature examines how credit rating agencies incorporate firm size and implicit government support into bank credit ratings. Prior to the Dodd–Frank Act, rating agencies explicitly considered the likelihood of government support when assessing the creditworthiness of large banks, resulting in ratings that were higher than those implied by standalone fundamentals. As a result, large banks benefited from lower borrowing costs not only through bond market pricing but also through elevated credit ratings.

Santos (2014) reports that investors in the bond market perceive the largest banks to be too big to fail. There are, however, a few differences between our study and Santos’s paper (2014). The first difference is that the sample used in Santos’s study ends in 2009, and therefore Santos’s study does not address the effect of regulatory changes—such as the Dodd–Frank Act—on investors’ perceptions of the potential for government bailout of big banks, while in our paper we examine how the Dodd–Frank Act affected investors’ perception of the “too-big-to fail” problem. The second difference between our paper and Santos’s study (2014) is that we examine the seriousness of the TBTF problem by using a number of rating categories to measure the additional discount in yield spread received by big banks in the bond market. In our paper, we find that the magnitude of the extra discount in yield spread received by big banks from the bond market over and above the discount already reflected in elevated credit ratings is equivalent to two more additional rating notches. The third difference is that we use several more control variables than Santos (2014) to control for the factors that may have an impact on yield spread beyond credit ratings and bank size2. Regardless of the differences in data and methodology between our paper and Santos’s study (2014), both studies reach the same conclusion that there is strong evidence of the TBTF phenomenon prior to the implementation of the pre-Dodd–Frank Act.

2.3. Regulatory Interventions and Post-Crisis Reforms

Several studies examine how regulatory reforms have affected the TBTF phenomenon. Balasubramnian and Cyree (2014) examine the effect of the Dodd–Frank Act on the TBTF effect. There are two major differences between our paper and that of Balasubramnian and Cyree (2014). The first difference is that they do not use credit ratings in their analyses. As a result, the TBTF problem they measure is the total effect of big banks on yield spread, while in our study we include credit rating among the independent variables. Consequently, in our study we measure the extra (marginal) effect of bank size in addition to the size effect that has already been reflected in elevated credit ratings due to the size of big banks. As a result, the size effect in our study is smaller than the size effect in Balasubramnian and Cyree’s (2014). The second difference between our paper and Balasubramnian and Cyree’s (2014) is the same as the difference between ours and Santos’s study (2014) in that we examine the seriousness of the TBTF problem by using a number of rating categories to measure the additional discount in yield spread received by big banks in the bond market. Regardless of the difference in methodology between our paper and the study by Balasubramnian and Cyree (2014), both papers find the TBTF phenomenon before the enactment of the Dodd–Frank Act and the alleviation of the TBTF phenomenon after the Dodd–Frank Act.

More recently, Berndt et al. (2025) examine globally systemically important banks in the U.S. over the period of 2001–2022 and find that the TBTF premium diminished following the Global Financial Crisis. However, their sample ends in 2022 and therefore does not reflect the 2023–2024 regional banking turmoil; by extending the dataset through mid-2025, we capture the renewed safety premium and shifted market perceptions associated with the regional bank crisis in 2023–2024; thus, our study offers a more up-to-date assessment of the TBTF phenomenon. These studies suggest that post-crisis regulatory reforms not only affected bank-level borrowing costs but also played a broader role in reshaping financial intermediation and risk-taking dynamics, with implications for systemic stability over the business cycle.

3. Research Methodology

To test whether big banks receive an extra discount in yield from the bond market in addition to the discount already reflected in the elevated credit ratings assigned by rating agencies, we take several steps in our analysis. First, we start from a preliminary regression:

YieldSpread_it = intercept + β1Big_it + β2Issue_it + β3Market_t + YearDum + ε_it

(1a)

where the YieldSpread_it is the yield spread on bond i issued on day t and the yield spread is defined as the difference between the yield to maturity of bond i on day t minus the yield to maturity on a Treasury security with the same maturity day as bond i. The baseline model follows the standard reduced-form bond yield spread framework commonly used in the corporate and bank bond pricing literature, in which yield spreads are modeled as a function of issuer size, bond contract characteristics, and prevailing market conditions. This specification enables examination of whether large banks face lower bond yield spreads than smaller banks after controlling standard bond and issuer characteristics, providing an initial test of size-related borrowing cost advantages. Big_it, a dummy variable for bank size, is equal to 1 if a bond is issued by a big bank (as measured by asset size) and 0 otherwise. In testing Equation (1a), if a big bank benefits from its size in the bond market, we expect the coefficient associated with Big_it to be significantly negative. We use four size indicators (the largest 3, 5, 10, and 20 banks in the banking industry) to examine the TBTF problem.

Issue_it is a vector of bond issue-related control variables. Following Balasubramnian and Cyree (2014), we control for term to maturity, coupon rate, and issue size, which proxy for interest rate risk, cash-flow structure, and liquidity effects, respectively. Specifically, we expect yield to be positively correlated with term to maturity, coupon rate (Buse, 1970; Saunders, 2001), and issue size. Likewise, Markett is a vector of market-level control variables designed to capture time-varying macro-financial conditions and aggregate risk premia. These include average S&P 500 returns, Treasury yield, the slope of the yield curve (defined as 30-year Treasury yield minus 2-year Treasury yield), and implied market volatility in S&P 500 options, as suggested by Balasubramnian and Cyree (2014). In Appendix A, we describe the definition and expected sign of each of the above-mentioned variables and the reason for including them as control variables.

We expect β1, the coefficient associated with big banks in the preliminary analysis (Equation (1a)), to be significantly negative. That is, the yield spreads on bonds issued by big banks, after controlling for other factors that have an impact on yield spread, tend to be lower. However, to answer the question whether there is an additional yield discount on bonds issued by big banks beyond the existent discount which has already been reflected in elevated credit ratings, we need to include the variable credit ratings in the regression analyses. Including credit ratings allows us to net out the borrowing cost advantage already embedded in ratings so that the coefficient on bank size captures the incremental yield discount beyond ratings. Accordingly, the coefficient on the size indicator can be interpreted as the market-implied TBTF discount that is not explained by credit quality. Therefore, we further include the variable CreditRating in the following regression Equation (1b):

YieldSpread_it = intercept+ β1Big_it + β2Issue_it + β3Market_t + β4CreditRating_it + YearDum + ε_it

(1b)

By explicitly controlling for credit ratings, this model isolates whether large banks receive an additional yield discount in the bond market beyond the borrowing cost advantage already implied by elevated ratings. If there is an additional yield discount for big banks on top of credit ratings, we expect β₁, the coefficient associated with the size indicator, still to be significantly negative. We also examine whether the enactment of the Dodd–Frank Act reduces this additional yield discount. We split the data sample into pre-Dodd–Frank Act (pre-DFA) and post-Dodd–Frank Act (post-DFA) sub-periods. Comparing estimates before and after the enactment of the Dodd–Frank Act allows us to assess whether post-crisis regulatory reforms reduced or eliminated the incremental TBTF-related yield discount in bond markets. If the Dodd–Frank Act does alleviate the TBTF problem, we would expect the significance of size indicators to shrink or even disappear during the post-Dodd–Frank Act period.

We conduct a series of robustness tests, including alternative size definitions, subsample analyses, and additional control variables, to ensure that our main findings are not driven by model specification or omitted risk factors. As a robustness check, we conduct a two-stage regression as follows. In the first-stage regression (Equation (2a)), yield spread is estimated as a function of bond issue-related and market-related control variables, as well as credit rating, without size-related variables.

YieldSpread_it = intercept + β1Issue_it + β2Market_t + β3CreditRating_it + ε_it

(2a)

The ε_it represents the yield that cannot be explained by the independent variables in Equation (2a). In the second stage (Equation (2b)), we regress the residuals, ε_it, obtained from the first-stage against bank size. If the bank size dummy variable is still significantly negative in the second-stage regression, then we can conclude that the bond market gives an additional yield discount to big banks.

ε_it = intercept + αBig_it + YearDum + δ_it

(2b)

The two-stage approach has the advantage of allowing us to explicitly measure the extra yield discount from size effect beyond the existent discount from elevated credit ratings.

To address potential endogeneity in credit ratings and bank size, we further conduct an additional robustness test using a dynamic panel GMM framework. This approach incorporates a lagged dependent variable and relies on internal instruments constructed from lagged values of potentially endogenous regressors, allowing for dynamic adjustment in bond yield spreads under stricter identification assumptions. We estimate both Blundell–Bond system GMM and Arellano–Bond difference GMM specifications while maintaining the same set of bond-specific and market control variables as in the baseline model. The results from these dynamic GMM estimations are reported in Section 5.

4. Descriptions of Data and Variables

We collected newly issued bonds from the Thomson Financial SDC Platinum Global New Issues database over the period 1 January 1991 through 30 June 2025. Following Cantor and Packer (1996), new debt issues under USD 10 million and with less than one year of maturity are excluded, as are issues with significant equity features, equipment trusts, collateralized mortgage obligations, government-guaranteed issues, variable rate issues, ESOP, lease certificates, and bond issues denominated in non-U.S. dollars. Further, as in Morgan (2002), we restrict our sample to issues rated by both Moody’s and Standard & Poor’s (S&P), the two major rating agencies in the rating industry. In addition, we deleted a few issues where the rating difference between Moody’s and S&P’s is greater than three notches. After applying these criteria to the database, the final sample contains 2939 new bond issues from 405 U.S. banks between 1 January 1991 and 30 June 2025.

Bond issue-related variables used in the study and their expected effect on yield spread are explained in Appendix A of the paper. Those variables include terms to maturity (measured in months), yield spreads (in basis points), coupon rates, issue size (in millions of dollars), and Treasury yields. Also, as in Morgan (2002), letter ratings by the two agencies are mapped onto a single numeric scale, with better credit quality indicated by lower numbers, and the mapping is described in Appendix B of the paper3. The rating gap is defined as Moody’s numeric rating minus the S&P numeric rating. Given two bonds with the same S&P rating, when the Moody’s numeric value is higher (and thus there is lower credit quality according to the Moody’s rating), the higher the credit risk and the higher the yield spread expected.

Following Balasubramnian and Cyree (2014), we also include S&P 500 return (SPRET), Treasury slope (TSLOPE)—defined as the difference between the 30-year rate and the 2-year rate—and the implied volatility of S&P 500 index options (VIX) as control variables. The two variables S&P 500 return and Treasury slope (TSLOPE) are used as proxies for the state of the macroeconomy, and they are collected from the CRSP and U.S. Department of Treasury websites, respectively. VIX is used as a proxy for market volatility, and the data are collected from the Chicago Board of Options Exchange (CBOE) website.

In

Table 1. Summary statistics for the newly issued bond-specific characteristics, market variables, and other control variables. New bond issue data were obtained from the Thomson Financial SDC Platinum Global New Issues database over the period 1 January 1991 through 30 June 2025. As in Cantor and Packer (1996), bonds included in the sample are fixed-rate, U.S. dollar-denominated, plain vanilla bonds without any derivative features or credit enhancement features such as collateral or sinking funds, equipment trusts, government-guaranteed issues, variable rate issues, ESOP, or lease certificates. Further, as in Morgan (2002), debt issues included in the sample are restricted to issues rated both by Moody’s and Standard & Poor’s, the two major rating agencies in the rating industry. In addition, we delete a few observations where the rating difference between Moody’s and S&P’s is greater than three notches. Letter ratings by the two agencies are transformed into a numeric scale, and higher letter ratings correspond to lower numbers. The definitions of each variable and numeric ratings are described in Appendix A and Appendix B.

Panel A: All Sample of Banks 1 January 1991–30 June 2025
Variable	Mean	Min	25th Pctl	Median	75th Pctl	Max	Std Dev
Yield Spread in Basis Point	90.167	−205.000	43.000	78.000	118.000	773.000	77.349
Maturity in Years	6.856	0.997	2.792	5.027	10.025	49.271	5.777
Coupon Rate in Basis Point	524.867	10.000	370.000	555.000	675.000	1450.000	213.665
Issue Size in USD Millions	0.437	0.010	0.082	0.200	0.500	3.750	0.588
S&P Rating	5.770	1.000	4.000	6.000	7.000	17.000	2.178
Rating Gap	−0.268	−5.000	−1.000	0.000	0.000	5.000	1.136
Treasury Yield in Basis Point	438.475	−9.500	275.000	472.900	597.800	833.500	198.860
SPRETM	0.011	−0.146	−0.016	0.014	0.036	0.112	0.038
VIX	18.346	10.125	13.495	16.919	22.110	44.795	5.883
TSLOPE	1.519	−0.870	0.530	1.210	2.640	3.910	1.185
Panel B: Sample of Banks 1 January 1991–2 December 2009 (Pre-Dodd–Frank Act)
Variable	Mean	Min	25th Pctl	Median	75th Pctl	Max	Std Dev
Yield Spread in Basis Point	79.824	−205.000	33.000	69.000	105.000	773.000	76.460
Maturity in Years	6.748	0.997	1.962	5.027	10.027	49.271	5.882
Coupon Rate in Basis Point	592.074	22.000	500.000	612.500	700.000	1450.000	185.069
Issue Size in USD Millions	0.238	0.010	0.050	0.150	0.250	3.127	0.339
S&P Rating	5.540	1.000	4.000	5.000	7.000	16.000	2.174
Rating Gap	−0.388	−5.000	−1.000	0.000	0.000	5.000	1.107
Treasury Yield in Basis Point	521.613	16.000	439.450	553.350	626.850	833.500	148.396
SPRETM	0.010	−0.146	−0.017	0.012	0.036	0.112	0.039
VIX	18.988	10.818	13.606	17.744	23.202	44.795	6.129
TSLOPE	1.341	−0.660	0.490	0.880	2.360	3.660	1.140
Panel C: Sample of Banks 3 December 2009–30 June 2025 (Post-Dodd–Frank Act)
Variable	Mean	Min	25th Pctl	Median	75th Pctl	Max	Std Dev
Yield Spread in Basis Point	119.886	−26.000	75.000	103.000	150.000	693.000	72.062
Maturity in Years	7.191	1.011	4.556	5.034	10.016	30.085	5.429
Coupon Rate in Basis Point	316.976	10.000	210.000	311.250	416.200	1009.000	154.111
Issue Size in USD Millions	1.053	0.012	0.500	0.900	1.500	3.750	0.747
S&P Rating	6.680	1.000	6.000	7.000	8.000	17.000	1.947
Rating Gap	0.207	−4.000	−1.000	0.000	1.000	4.000	1.127
Treasury Yield in Basis Point	199.578	−9.500	112.750	180.900	272.000	548.000	113.842
SPRETM	0.014	−0.093	−0.010	0.018	0.036	0.108	0.034
VIX	16.358	10.125	13.467	14.837	18.059	36.530	4.504
TSLOPE	2.069	−0.870	1.520	2.160	2.970	3.910	1.155

we report summary statistics on bond-specific characteristics, market proxies, and other control variables for the entire sample period first, and we then divide the sample into two sub-periods. We choose the period before 2 December 2009 as the pre-Dodd–Frank Act sub-period because on 3 December 2009 Congressman Barney Frank formally introduced the bill “The Wall Street Reform and Consumer Protection Act of 2009” to the U.S. House of Representatives.

In Panel A of Table 1, we present the summary statistics for the full sample, which includes all banks from 1 January 1991 to 30 June 2025, while in Panels B and C we report the summary statistics for the pre- and post-Dodd–Frank Act sub-periods, respectively. During the post-Dodd–Frank Act sub-period, which is also known as the post-2008 financial crisis period, coupon rates on bank-issued bonds and Treasury yields are significantly lower than the pre-Dodd–Frank Act sub-period due to the Federal Reserve Bank’s easy monetary policy, while the yield spreads are significantly higher due to increased risk caused by the 2008 crisis. The mean yield spread increased from 79.82 to 119.89 basis points and the median yield spread increased from 69 to 103 basis points from the pre- to post-Dodd–Frank Act sub-periods, respectively.

Table 1 also indicates that banks are better-rated by S&P during the pre- rather than the post- Dodd–Frank Act sub-period by about one notch (a change from 5.54 to 6.68—recall that a lower numerical score implies better credit quality). The rating gap also changes from negative to positive, which suggests that Moody’s also became more conservative in assigning credit ratings on bank debt after the Dodd–Frank Act. Overall, the results in Table 1 suggest that banks’ bond issue characteristics substantially changed after the enactment of the Dodd–Frank Act. These results are consistent with the study of Balasubramnian and Cyree (2014), which also suggests that the Dodd–Frank Act has been effective in reducing the TBTF problem and the discounts in yield spreads.

5. Empirical Results

To examine whether there is a TBTF problem and whether there is a size effect on yield spread over and above what credit rating agencies have already taken into account in assigning ratings, we run several regressions on our data before the enactment of the Dodd–Frank Act. Our analysis focuses on whether bank size continues to explain borrowing costs after controlling credit quality and other bond-specific and market conditions.

Table 2. The effect of banks’ TBTF problem on yield spread during the pre-Dodd–Frank Act (pre-DFA) period. The pre-DFA period is defined as the period prior to 3 December 2009 when “The Wall Street Reform and Consumer Protection Act of 2009” was introduced to the U.S. House of Representatives by Congressman Barney Frank. The dependent variable is the yield spread on newly issued bonds. We present the results with and without the control of credit rating separately. We present the parameter estimates and the respective probabilities. We examine the TBTF problem for the largest 3, 5, 10, and 20 banks.

Sample: Banks 1 January 1991–2 December 2009 (Pre-Dodd–Frank Act)
Dependent Variable = Yield Spread in Basis Points
	Without Control of Credit Rating								With Control of Credit Rating
	(1)		(2)		(3)		(4)		(5)		(6)		(7)		(8)
	Estimate	Pr > |t|	Estimate	Pr > |t|	Estimate	Pr > |t|	Estimate	Pr > |t|	Estimate	Pr > |t|	Estimate	Pr > |t|	Estimate	Pr > |t|	Estimate	Pr > |t|
Intercept	−10.5110	0.0123	−10.8212	0.0098	−10.5499	0.0121	−10.7281	0.0107	−11.2481	0.0070	−11.5654	0.0054	−11.2815	0.0068	−11.4950	0.0058
Big 3 Banks	−2.6041	0.0047							−2.1130	0.0227
Big 5 Banks			−2.3070	0.0049							−1.6752	0.0417
Big 10 Banks					−2.1076	0.0070							−1.6723	0.0323
Big 20 Banks							−1.9204	0.0121							−1.4102	0.0659
Ln(MAT)	2.6408	<0.0001	2.6653	<0.0001	2.6407	<0.0001	2.6301	<0.0001	2.6460	<0.0001	2.6425	<0.0001	2.6423	<0.0001	2.6194	<0.0001
Coupon	0.9680	<0.0001	0.9678	<0.0001	0.9683	<0.0001	0.9684	<0.0001	0.9541	<0.0001	0.9540	<0.0001	0.9542	<0.0001	0.9543	<0.0001
Issue Size	2.8442	0.0021	2.7132	0.0031	2.6402	0.0039	2.6005	0.0044	3.1417	0.0006	3.0091	0.0010	2.9803	0.0011	2.9328	0.0013
S&P Rating									0.9186	<0.0001	0.9122	<0.0001	0.9221	<0.0001	0.9217	<0.0001
Rating Gap									0.2559	0.4015	0.3033	0.3168	0.2927	0.3345	0.3013	0.3208
Treasury Yield	−0.9593	<0.0001	−0.9587	<0.0001	−0.9592	<0.0001	−0.9592	<0.0001	−0.9478	<0.0001	−0.9474	<0.0001	−0.9476	<0.0001	−0.9476	<0.0001
SPRET	−9.7910	0.1678	−9.3818	0.1864	−9.5808	0.1772	−9.6759	0.1731	−8.2420	0.2420	−7.9414	0.2598	−8.0491	0.2533	−8.1390	0.2482
VIX	0.1395	0.0743	0.1359	0.0825	0.1369	0.0804	0.1362	0.0822	0.1772	0.0228	0.1752	0.0246	0.1752	0.0246	0.1756	0.0244
TSLOPE	0.5090	0.5561	0.6016	0.4843	0.6112	0.4775	0.6590	0.4432	0.1720	0.8413	0.2706	0.7517	0.2554	0.7653	0.3065	0.7198
Year Dummy	Yes		Yes		Yes		Yes		Yes		Yes		Yes		Yes
Adj. R2	0.9783		0.9783		0.9780		0.9782		0.9787		0.9787		0.9783		0.9787
N	1856		1856		1856		1856		1856		1856		1856		1856

presents baseline regressions in which the yield spread on bank-issued bonds is regressed on indicator variables for large banks, defined as the top three, five, ten, and twenty banks in the industry, along with a standard set of bond-specific and market-related controls. Bond-specific control variables include term to maturity (in months), coupon rate (in basis points), issue size (in millions of dollars), and yield (in basis points) on Treasury securities which have the same maturity date as the bond. The market-related control variables include the monthly returns on the S&P 500 index, implied volatility (VIX) on S&P 500 index options, and the slope of the Treasury yield curve measured by the difference between 30-year and 2-year Treasury yields.

Panel A of Table 2 reports specifications that exclude credit ratings. The estimated coefficients on the big-bank indicators are uniformly negative and statistically significant, indicating that bonds issued by larger banks carry lower yield spreads. Moreover, the magnitude of the estimated discount increases monotonically with bank size, consistent with a graduated size-related borrowing advantage. These estimates alone do not distinguish between rating-based borrowing advantages and market-based pricing effects, since larger banks also tend to receive higher credit ratings. To address this concern, Panel B augments the baseline specifications with two measures of credit quality: the issuer’s S&P credit rating and the numerical rating gap between Moody’s and S&P.

The results in Panel B show that credit ratings are significantly related to the yield spreads, with lower-rated bonds commanding higher yields. More importantly, after controlling credit ratings, the bank size variables remain statistically significant, indicating that the borrowing cost advantage of large banks cannot be fully explained by their superior credit quality. Because rating agencies explicitly incorporated firm size into bank credit ratings during this period, the residual size effects in Panel B capture an incremental yield discount beyond that already embedded in ratings. The monotonic pattern across size thresholds suggests that this discount increases with institutional scale, consistent with bond market perceptions of implicit support for the largest banks. In addition, all the significant control variables exhibit signs consistent with expectations.4

To further provide a robust test of the argument that the bond market gives an additional discount in yield to big banks, we conduct a two-stage regression analysis5. In the first stage, yield spreads are regressed on credit ratings (measured by S&P ratings and the Moody’s & S&P rating gap) along with the full set of bond-specific and market controls, but excluding bank size. The resulting residuals capture the component of yield spreads not explained by credit ratings. In the second stage, these residuals are regressed on bank size indicators to test whether large banks receive an incremental yield discount beyond that embedded in ratings.

In

Table 3. The effect of banks’ TBTF problem on yield spread during the pre-Dodd–Frank Act (pre-DFA) period. The pre-DFA period is defined as the period prior to 3 December 2009 when “The Wall Street Reform and Consumer Protection Act of 2009” was introduced to the U.S. House of Representatives by Congressman Barney Frank. The dependent variable is the yield spread on newly issued bonds. We present the results for the pre-DFA period from a two-stage analysis, with the first stage not controlling the bank size and the second stage examining the size effect. We present the parameter estimates and the respective probabilities. We examine the TBTF problem for the largest 3, 5, 10, and 20 banks.

Sample: Banks 1991–2 December 2009 (Pre-Dodd–Frank Act)
Stage One: Dependent Variable = Yield Spread in Basis Points
	Estimate				Pr > |t|
Intercept	−16.1720				<0.0001
Ln(MAT)	2.0537				<0.0001
Coupon	0.9635				<0.0001
Issue Size	3.4593				<0.0001
S&P Rating	0.7729				<0.0001
Rating Gap	0.4973				0.0881
Treasury Yield	−0.9546				<0.0001
SPRET	−8.8861				0.1760
VIX	0.1407				0.0027
TSLOPE	0.3322				0.2258
Adj. R2	0.9778
N	1856
Stage Two: Dependent Variable = Residuals in Stage One Regression
	(1)		(2)		(3)		(4)
	Estimate	Pr > |t|	Estimate	Pr > |t|	Estimate	Pr > |t|	Estimate	Pr > |t|
Intercept	5.6389	0.0018	5.5411	0.0021	5.7773	0.0015	5.6766	0.0018
Big 3 Banks	−1.8184	0.0319
Big 5 Banks			−1.4807	0.0507
Big 10 Banks					−1.4859	0.0408
Big 20 Banks							−1.2609	0.0761
Year Dummy	Yes		Yes		Yes		Yes
Adj. R2	0.0194		0.0190		0.0192		0.0186
N	1856		1856		1856		1856

we present the results for the two-stage regression analysis for the pre-Dodd–Frank Act period. In the first-stage regression, all the significant independent variables have the correct signs. In particular, the coefficients associated with S&P ratings are statistically significant. In the second stage, the residual yield spreads remain significantly related to bank size. The estimated coefficients are negative and economically meaningful, with a monotonic pattern across size thresholds, indicating that larger banks receive larger residual yield discounts even after purging rating-related effects. The results in Table 3 suggest that even the component of yield spreads unexplained by credit ratings remains systematically associated with bank size. Taken together with the results in Table 2, the two-stage analysis provides solid confirmation that bond investors extend an additional yield discount to large banks beyond the advantages already reflected in credit ratings.

After identifying the presence of the TBTF problem before the enactment of the Dodd–Frank Act, we further examine how severe the TBTF problem is. The severity of the problem can be measured by three levels of severity:

• The first level of severity is reached when a big bank has the same credit rating as a non-big bank and it pays the same borrowing cost as the non-big bank. Since the big bank already benefits from elevated ratings due to its size, even when it pays the same borrowing cost as a non-big bank, it still benefits from its size and the TBTF problem still exists.
• The second level of severity is reached when a big bank has the same credit rating as a non-big bank and it pays a lower cost than a non-big bank in the bond market. This is a more severe TBTF problem than the first level because the big bank not only benefits from an elevated credit rating due to its size, but it further benefits from the bond market by paying a lower borrowing cost even though it has the same credit rating as a non-big bank. In our study, we identified such a second-level TBTF problem in the bond market before the enactment of the Dodd–Frank Act.
• The third level of severity is reached when a big bank has a lower credit rating than a non-big bank while it still can borrow at a lower cost than the non-big bank in the bond market. This is the most severe form of TBTF problem because the big bank not only benefits from elevated credit rating due to its size, but it also further benefits from the bond market by paying a lower borrowing cost even though its credit rating is lower than the non-big bank. In our study, we identified such a third-level TBTF problem in the bond market before the enactment of the Dodd–Frank Act.

These severity levels show that TBTF distortions extend beyond rating-based advantages and can dominate bond market pricing even when big banks exhibit lower observable credit quality. The three-level TBTF severity framework offers a systematic and economically grounded approach to characterizing the extent of market-based distortions associated with bank size. For regulators, the progression across levels distinguishes between residual advantages already internalized by rating agencies and more severe breakdowns of market discipline in which bond investors extend preferential pricing beyond, or even contrary to, credit ratings. For investors, the framework clarifies whether observed yield differentials are driven by fundamental credit risk or by expectations of implicit public support. This three-level structure advances the TBTF literature by moving beyond binary classifications and enabling a nuanced evaluation of regulatory effectiveness across varying degrees of systemic distortion. The test results showing the second level and the third level of yield discount are presented in

Table 4. The effect of banks’ TBTF problem on yield spread during the pre-Dodd–Frank Act (pre-DFA) period. The dependent variable is the yield spread on newly issued bonds. We compare the coefficients of big banks with non-big banks with equal ratings, one-level higher ratings and two-level higher ratings. in Panels A, B, and C, respectively. Wald tests are used to examine the significance of the difference between the coefficients. We examine the TBTF problem for the largest 3, 5, 10, and 20 banks. ** and *** indicate significance level at 5% and 1% respectively.

Sample: Banks 1991–2 December 2009 (Pre-Dodd–Frank Act)
Dependent Variable = Yield Spread in Basis Points
Panel A	(1)		(2)		(3)		(4)
	Estimate	Pr > |t|	Estimate	Pr > |t|	Estimate	Pr > |t|	Estimate	Pr > |t|
Intercept	−9.1046	0.0328	−9.3147	0.0288	−9.0137	0.0348	−9.2261	0.0307
big3bank_A+_or_higher (Beta1)	−4.8839	<0.0001
big3bank_A_or_lower	0.0038	0.9980
Nonbig3bank_A+_or_higher (Beta3)	−1.3317	0.0387
big5bank_A+_or_higher			−4.0042	<0.0001
big5bank_A_or_lower			−0.6421	0.6402
Nonbig5bank_A+_or_higher			−1.3819	0.0358
big10bank_A+_or_higher					−4.1365	<0.0001
big10bank_A_or_lower					−0.4593	0.6968
Nonbig10bank_A+_or_higher					−1.3125	0.0481
big20bank_A+_or_higher							−3.7630	0.0001
big20bank_A_or_lower							−0.4217	0.7165
Nonbig20bank_A+_or_higher							−1.3269	0.0488
Ln(MAT)	2.6580	<0.0001	2.6328	<0.0001	2.6512	<0.0001	2.6182	<0.0001
Coupon	0.9649	<0.0001	0.9648	<0.0001	0.9651	<0.0001	0.9652	<0.0001
Rating Gap	−0.0470	0.8774	0.0255	0.9327	0.0191	0.9496	0.0220	0.9422
Issue Size	2.8490	0.0021	2.6943	0.0033	2.6543	0.0037	2.5983	0.0045
Treasury Yield	−0.9580	<0.0001	−0.9572	<0.0001	−0.9577	<0.0001	−0.9575	<0.0001
SPRET	−9.3123	0.1888	−9.2020	0.1945	−9.1746	0.1957	−9.2248	0.1936
VIX	0.1458	0.0620	0.1427	0.0685	0.1414	0.0708	0.1417	0.0708
TSLOPE	0.3829	0.6577	0.5444	0.5265	0.5079	0.5548	0.5791	0.5001
Year Dummy	Yes		Yes		Yes		Yes
Wald Test: Beta1 = Beta3	9.82	***	6.74	***	7.99	***	6.17	**
Adj. R2	0.9781		0.9784		0.9784		0.9780
N	1856		1856		1856		1856
Panel B	(1)		(2)		(3)		(4)
	Estimate	Pr > |t|	Estimate	Pr > |t|	Estimate	Pr > |t|	Estimate	Pr > |t|
Intercept	−9.4496	0.0263	−9.5805	0.0243	−9.3809	0.0279	−9.6043	0.0243
big3bank_A+_or_higher (Beta1)	−4.4800	<0.0001
big3bank_A_or_lower	0.2152	0.8883
Nonbig3bank_AA−_or_high(Beta3)	−1.2570	0.0710
big5bank_A+_or_higher			−3.6105	0.0002
big5bank_A_or_lower			−0.4262	0.7545
Nonbig5bank_AA−_or_higher			−1.3538	0.0600
big10bank_A+_or_higher					−3.7264	0.0001
big10bank_A_or_lower					−0.2080	0.8582
Nonbig10bank_AA−_or_higher					−1.1445	0.1135
big20bank_A+_or_higher							−3.3750	0.0003
big20bank_A_or_lower							−0.1790	0.8760
Nonbig20bank_AA−_or_higher							−1.1555	0.1133
Ln(MAT)	2.6628	<0.0001	2.6268	<0.0001	2.6579	<0.0001	2.6326	<0.0001
Coupon	0.9657	<0.0001	0.9656	<0.0001	0.9660	<0.0001	0.9661	<0.0001
Rating Gap	−0.0371	0.9043	0.0331	0.9135	0.0143	0.9626	0.0162	0.9577
Issue Size	2.8756	0.0019	2.7079	0.0032	2.6551	0.0037	2.5990	0.0045
Treasury Yield	−0.9584	<0.0001	−0.9576	<0.0001	−0.9583	<0.0001	−0.9581	<0.0001
SPRET	−9.1453	0.1971	−8.9069	0.2095	−8.8297	0.2135	−8.8528	0.2126
VIX	0.1432	0.0667	0.1399	0.0741	0.1392	0.0755	0.1393	0.0758
TSLOPE	0.3593	0.6777	0.5201	0.5453	0.4837	0.5740	0.5520	0.5207
Year Dummy	Yes		Yes		Yes		Yes
Wald Test: Beta1 = Beta3	7.04	***	4.15	**	5.53	**	4.22	**
Adj. R2	0.9781		0.9780		0.9780
N	1856		1856		1856		1856
Panel C	(1)		(2)		(3)		(4)
	Estimate	Pr > |t|	Estimate	Pr > |t|	Estimate	Pr > |t|	Estimate	Pr > |t|
Intercept	−7.2695	0.0888	−7.3522	0.0847	−7.3470	0.0857	−7.5976	0.0755
big3bank_A+_or_higher (Beta1)	−4.5040	<0.0001
big3bank_A_or_lower	0.0907	0.9525
Nonbig3bank_AA_or_higher (Beta3)	−3.6927	<0.0001
big5bank_A+_or_higher			−3.7047	0.0001
big5bank_A_or_lower			−0.5420	0.6891
Nonbig5bank_AA_or_higher			−3.9041	<0.0001
big10bank_A+_or_higher					−3.8522	<0.0001
big10bank_A_or_lower					−0.3132	0.7867
Nonbig10bank_AA_or_higher					−3.5209	0.0002
big20bank_A+_or_higher							−3.5002	0.0001
big20bank_A_or_lower							−0.2929	0.7973
Nonbig20bank_AA_or_higher							−3.4971	0.0003
Ln(MAT)	2.6157	<0.0001	2.6131	<0.0001	2.6407	<0.0001	2.6164	<0.0001
Coupon	0.9647	<0.0001	0.9645	<0.0001	0.9648	<0.0001	0.9650	<0.0001
Rating Gap	0.0931	0.7596	0.1967	0.5170	0.1685	0.5797	0.1668	0.5836
Issue Size	2.8620	0.0019	2.6787	0.0034	2.6632	0.0035	2.6025	0.0043
Treasury Yield	−0.9582	<0.0001	−0.9573	<0.0001	−0.9579	<0.0001	−0.9577	<0.0001
SPRET	−8.9259	0.2065	−8.7587	0.2154	−8.5706	0.2257	−8.5823	0.2254
VIX	0.1402	0.0714	0.1350	0.0834	0.1356	0.0821	0.1346	0.0849
TSLOPE	0.2286	0.7910	0.3944	0.6456	0.3749	0.6624	0.4550	0.5956
Year Dummy	Yes		Yes		Yes		Yes
Wald Test: Beta1 = Beta3	0.34		0.02		0.07		0.01
Adj. R2	0.9782		0.9782		0.9782		0.9781
N	1856		1856		1856		1856

.

Since most big banks are rated as A+ or higher in our sample, we compare the big banks in this rating group (i.e., A+ or higher) with non-big banks in the same rating group (i.e., A+ or higher). We find that the borrowing costs paid by big banks are lower than the borrowing costs paid by non-big banks in the same rating group.

Panel A partitions bonds into four groups based on bank size and credit rating using bonds issued by smaller banks with ratings of A or lower as the baseline category. This structure allows direct comparison of yield spreads between large and small banks within the same rating group. The estimated coefficients indicate that large banks with ratings of A+ or higher pay significantly lower yields than smaller banks with the same ratings. Wald tests confirm that these differences are statistically significant.

Panel B extends the analysis by comparing large banks rated A+ or higher with smaller banks rated AA− or higher. This comparison allows us to test whether large banks can borrow at lower yields than smaller banks with superior ratings. We compare β₁, the coefficient for bonds issued by big banks and rated A+ or higher, with β₃, the coefficient for bonds issued by non-big banks rated AA− or higher. The results in Panel B show that β₁ is again smaller than β₃, and the Wald tests examining the differences between β₁ and β₃ suggest that the differences are statistically different. Therefore, the results in Panel B suggest that the borrowing costs paid by big banks are not only lower than non-big banks in the same rating category (as shown in Panel A), but also lower than non-big banks with a higher rating category.

Panel C further examines how far this yield equivalence extends by comparing large banks rated A+ or higher with smaller banks rated AA or higher. The results in the panel indicate that the yield spread on bonds issued by big banks rated A+ or higher is still slightly lower than bonds issued by non-big banks rated AA or higher, but the difference is no longer statistically significant, as indicated by the Wald tests. The results in Panel C suggest that the borrowing costs paid by big banks rated A+ are similar to the costs paid by non-big banks rated AA, two notches higher. So the results in Table 4 suggest that big banks, regardless of their elevated credit ratings by rating agencies, can still borrow from the bond market at lower costs, not only compared to those non-big banks with the same credit rating but also those non-big banks with higher credit ratings. In other words, big banks’ lower borrowing costs are reflected in three levels: first, they receive higher credit ratings; secondly, their borrowing costs are lower than non-big banks in the same rating categories; thirdly, their borrowing costs are lower even than non-big banks with higher rating categories.

After showing evidence of the TBTF problem in the bond market prior to the Dodd–Frank Act, we then continue to examine the question of whether the TBTF problem is alleviated by the Act. We analyze the data for the post-Dodd–Frank period from 3 December 2009 until 8 March 2023, the onset of the regional bank crisis6, and present the results in

Table 5. The effect of banks’ TBTF problem on yield spread during the post-Dodd–Frank Act (post-DFA) and pre-regional bank crisis period. The post-DFA and pre-regional bank crisis period is defined as the interval beginning after 3 December 2009, when the Wall Street Reform and Consumer Protection Act of 2009 was introduced in the U.S. House of Representatives by Congressman Barney Frank, and ending prior to the outbreak of the regional banking crisis on 8 March 2023. The dependent variable is the yield spread on newly issued bonds. We present the results with and without the control of credit ratings separately. We present the parameter estimates and the respective probabilities. We examine the TBTF problem for the largest 3, 5, 10, and 20 banks.

Sample: Banks 3 December 2009–8 March 2023 (Post-Dodd–Frank Act, Pre-Regional Bank Crisis)
Dependent Variable = Yield Spread in Basis Points
	Without Control of Credit Rating								With Control of Credit Rating
	(1)		(2)		(3)		(4)		(5)		(6)		(7)		(8)
	Estimate	Pr > |t|	Estimate	Pr > |t|	Estimate	Pr > |t|	Estimate	Pr > |t|	Estimate	Pr > |t|	Estimate	Pr > |t|	Estimate	Pr > |t|	Estimate	Pr > |t|
Intercept	−3.5004	0.6481	−3.5004	0.6481	−3.8120	0.6191	−3.9158	0.6096	−2.2896	0.7993	−1.4476	0.8724	−1.5422	0.8641	−1.8327	0.8389
Big 3 Banks	−0.6727	0.3333							−1.3425	0.1028
Big 5 Banks			−0.6727	0.3333							−0.6552	0.4289
Big 10 Banks					−0.9937	0.1596							−0.6898	0.4202
Big 20 Banks							−1.1212	0.1263							−0.8770	0.3212
Ln(MAT)	0.7819	0.2526	0.7819	0.2526	0.8964	0.1944	0.8734	0.2036	1.1104	0.1937	1.1176	0.1938	1.1177	0.1937	1.1245	0.1903
Coupon	0.9820	<0.0001	0.9820	<0.0001	0.9795	<0.0001	0.9791	<0.0001	0.9638	<0.0001	0.9661	<0.0001	0.9654	<0.0001	0.9646	<0.0001
Issue Size	1.4767	0.0007	1.4767	0.0007	1.4733	0.0006	1.4530	0.0006	1.6486	0.0009	1.5601	0.0016	1.5460	0.0016	1.5379	0.0017
S&P Rating									0.7545	0.0175	0.6038	0.0578	0.6077	0.0553	0.6260	0.0455
Rating Gap									0.4125	0.3098	0.2504	0.5250	0.1903	0.6140	0.1854	0.6217
Treasury Yield	−97.4997	<0.0001	−97.4997	<0.0001	−97.2279	<0.0001	−97.1700	<0.0001	−95.4628	<0.0001	−95.7048	<0.0001	−95.6224	<0.0001	−95.5290	<0.0001
SPRET	9.3931	0.3420	9.3931	0.3420	9.9488	0.3145	10.0662	0.3088	8.6965	0.4843	7.8427	0.5290	7.9066	0.5258	8.1578	0.5128
VIX	−0.2524	0.0236	−0.2524	0.0236	−0.2456	0.0275	−0.2424	0.0297	−0.3003	0.0267	−0.3132	0.0210	−0.3102	0.0222	−0.3051	0.0246
TSLOPE	1.3321	0.2628	1.3321	0.2628	1.2623	0.2887	1.2865	0.2791	0.6267	0.6665	0.5376	0.7125	0.5372	0.7127	0.5699	0.6957
Year Dummy	Yes		Yes		Yes		Yes		Yes		Yes		Yes		Yes
R2	0.9900		0.9900		0.9900		0.9903		0.9888		0.9887		0.9887		0.9888
N	670		670		670		670		670		670		670		670

.

Panel A estimates yield spreads as a function of bank size and control variables without credit ratings, while Panel B adds credit ratings to isolate any incremental size-related yield effects. In both panels, the coefficients associated with big banks become statistically insignificant. These results indicate that, unlike the pre-DFA period, large banks no longer enjoy a statistically detectable yield advantage once standard risk controls are applied. Nevertheless, to ensure that this conclusion is not sensitive to model specification, we conduct the same two-stage regression framework as a robustness test. The results for the post-Dodd–Frank two-stage analyses are presented in

Table 6. The effect of banks’ TBTF problem on yield spread during the post-Dodd–Frank Act (post-DFA) and pre-regional bank crisis period. The post-DFA and pre-regional bank crisis period is defined as the interval beginning after 3 December 2009, when the Wall Street Reform and Consumer Protection Act of 2009 was introduced in the U.S. House of Representatives by Congressman Barney Frank, and ending prior to the outbreak of the regional banking crisis on 8 March 2023. The dependent variable is the yield spread on newly issued bonds. We present the results for the post-DFA period from a two-stage analysis, with the first stage not controlling the bank size and the second stage examining the size effect. We present the parameter estimates and the respective probabilities and examine the TBTF problem for the largest 3, 5, 10, and 20 banks.

Sample: Banks 3 December 2009–8 March 2023 (Post-Dodd–Frank Act, Pre-Regional Bank Crisis)
Stage One: Dependent Variable = Yield Spread in Basis Points
					Estimate		Pr > |t|
Intercept					−5.6633		0.2266
Ln(MAT)					0.7141		0.3751
Coupon					0.9710		<0.0001
Issue Size					1.5473		0.0011
S&P Rating					0.5662		0.0495
Rating Gap					0.0417		0.8995
Treasury Yield					−0.9595		<0.0001
SPRET					15.3389		0.191
VIX					−0.1127		0.2871
TSLOPE					1.0482		0.0352
Adj. R2					0.9887
N					670
Stage Two: Dependent Variable = Residuals in Stage One Regression
	(1)		(2)		(3)		(4)
	Estimate	Pr > |t|	Estimate	Pr > |t|	Estimate	Pr > |t|	Estimate	Pr > |t|
Intercept	−0.9283	0.8655	−0.9283	0.8657	−0.9283	0.8657	1.5435	0.4079
Big 3 Banks	−0.9643	0.1829
Big 5 Banks			−0.3934	0.5816
Big 10 Banks					−0.4604	0.5457
Big 20 Banks							−0.6272	0.4341
Year Dummy	Yes		Yes		Yes		Yes
Adj. R2	0.0202		0.0173		0.0174		0.0174
N	670		670		670		670

, and the results confirm our conjecture, as we expected. Taken together, the results in Table 5 and Table 6 indicate that the incremental TBTF-related yield discount observed prior to the Dodd–Frank Act largely disappears in the post-reform period, suggesting a meaningful restoration of market discipline7.

Soon after the Dodd–Frank Act was implemented, rating agencies adopted a major policy change: They abolished the practice of incorporating bank size into credit rating assignments. This shift in rating methodology creates an opportunity to assess whether the disappearance of the yield discount observed in Table 5 and Table 6 is attributable to the Dodd–Frank Act itself or to the rating agencies’ policy change.

Therefore, we further separate the post-Dodd–Frank Act period into two sub-periods: one period before and one period after rating agencies’ decision to abolish the policy, so that we can decide whether the disappearance of the discount in yield was caused by the Dodd–Frank Act or by rating agencies’ decision to abolish the practice of using banks’ size in determining credit ratings8. The results for the two separate sub-periods are presented in

Table 7. The effect of banks’ TBTF problem on yield spread between the post-Dodd–Frank Act and the pre-S&P rating change period. The post-DFA period is defined as the period after 3 December 2009, when “The Wall Street Reform and Consumer Protection Act of 2009” was introduced to the U.S. House of Representatives by Congressman Barney Frank. The pre-S&P rating change period is defined as the period beginning on 9 November 2011, when S&P published its new bank rating criterion of no longer taking the possibility of government bailout into account in determining banks’ ratings. The dependent variable is the yield spread on newly issued bonds. We present the results with and without the control of credit rating separately and examine the TBTF problem for the largest 3, 5, 10, and 20 banks.

Sample: Banks 3 December 2009–9 November 2011 (Post-Dodd–Frank Act and Pre-S&P Rating Change)
Dependent Variable = Yield Spread in Basis Points
	Without Control of Credit Rating								With Control of Credit Rating
	(1)		(2)		(3)		(4)		(5)		(6)		(7)		(8)
	Estimate	Pr > |t|	Estimate	Pr > |t|	Estimate	Pr > |t|	Estimate	Pr > |t|	Estimate	Pr > |t|	Estimate	Pr > |t|	Estimate	Pr > |t|	Estimate	Pr > |t|
Intercept	96.5452	0.2430	83.2042	0.3241	91.2608	0.2674	93.4090	0.2592	28.2999	0.6978	32.8428	0.6551	28.8727	0.6910	27.1396	0.7096
Big 3 Banks	−3.9778	0.4662							0.6918	0.8861
Big 5 Banks			−3.4564	0.5238							1.6042	0.7365
Big 10 Banks					−5.3608	0.3543							1.6364	0.7576
Big 20 Banks							−3.47388	0.5471							−1.7380	0.7568
Ln(MAT)	−4.5571	0.5026	−3.9017	0.5740	−2.6534	0.7099	−3.39719	0.6363	−3.6994	0.5394	−4.0798	0.5056	−4.2373	0.4994	−3.2830	0.5945
Coupon	0.9734	<0.0001	0.9717	<0.0001	0.9635	<0.0001	0.96861	<0.0001	0.8381	<0.0001	0.8377	<0.0001	0.8401	<0.0001	0.8302	<0.0001
Issue Size	5.1136	0.1711	4.7360	0.1897	4.4871	0.1973	4.2540	0.2220	2.9284	0.3913	2.8152	0.3861	2.9391	0.3545	3.3658	0.2948
S&P Rating									5.8333	0.0079	5.9262	0.0076	5.8824	0.0076	6.0279	0.0088
Rating Gap									10.8845	0.0006	11.0116	0.0005	11.1117	0.0006	10.4897	0.0009
Treasury Yield	−0.9456	<0.0001	−0.9445	<0.0001	−0.9367	<0.0001	−0.9424	<0.0001	−0.7937	<0.0001	−0.7931	<0.0001	−0.7954	<0.0001	−0.7854	<0.0001
SPRET	59.9923	0.3137	62.4618	0.2937	62.3210	0.2919	66.4622	0.2656	146.7714	0.0107	147.7301	0.0096	148.1429	0.0098	145.2145	0.0102
VIX	−1.8673	0.0695	−1.7452	0.0928	−1.8523	0.0704	−1.8509	0.0723	−0.5956	0.5224	−0.6271	0.4996	−0.5811	0.5322	−0.5924	0.5232
TSLOPE	−13.3243	0.4143	−10.7869	0.5146	−13.0294	0.4221	−13.0569	0.4243	−5.7924	0.6824	−6.6597	0.6410	−5.7496	0.6839	−5.8191	0.6802
Year Dummy	Yes		Yes		Yes		Yes		Yes		Yes		Yes		Yes
R2	0.9853		0.9853		0.9854		0.9852		0.9893		0.9893		0.9893		0.9893
N	50		50		50		50		50		50		50		50

and

Table 8. The effect of banks’ TBTF problem on yield spread during the post-S&P rating change period. The post-S&P rating change period is defined as the period after 9 November 2011, when S&P published its new bank rating criterion of no longer taking the possibility of government bailout into account in determining banks’ ratings. The dependent variable is the yield spread on newly issued bonds. We present the results with and without the control of credit rating separately and examine the TBTF problem for the largest 3, 5, 10, and 20 banks.

Sample: Banks 10 November 2011–30 June 2025 (Post-S&P Rating Change)
Dependent Variable = Yield Spread in Basis Points
	Without Control of Credit Rating								With Control of Credit Rating
	(1)		(2)		(3)		(4)		(5)		(6)		(7)		(8)
	Estimate	Pr > |t|	Estimate	Pr > |t|	Estimate	Pr > |t|	Estimate	Pr > |t|	Estimate	Pr > |t|	Estimate	Pr > |t|	Estimate	Pr > |t|	Estimate	Pr > |t|
Intercept	1.0945	0.8293	1.9177	0.7057	2.1467	0.6727	2.3878	0.6384	−0.2918	0.9542	0.9339	0.8543	1.1546	0.8208	1.4015	0.7831
Big 3 Banks	−0.5535	0.3163							−1.0361	0.0744
Big 5 Banks			−0.9686	0.0700							−0.9494	0.0749
Big 10 Banks					−1.0914	0.0463							−0.96565	0.0793
Big 20 Banks							−1.2804	0.0237							−1.21551	0.0315
Ln(MAT)	1.0469	0.0518	1.1855	0.0296	1.1723	0.0306	1.1623	0.0313	1.1688	0.0298	1.2533	0.0213	1.21909	0.0244	1.22284	0.0233
Coupon	0.9835	<0.0001	0.9816	<0.0001	0.9804	<0.0001	0.9797	<0.0001	0.9731	<0.0001	0.9736	<0.0001	0.9733	<0.0001	0.97212	<0.0001
Issue Size	1.1042	0.0014	1.1951	0.0006	1.1665	0.0007	1.1507	0.0007	1.1778	0.0007	1.1889	0.0006	1.14668	0.0008	1.1413	0.0007
Moody Rating									0.4782	0.0095	0.3679	0.0350	0.3364	0.0555	0.3518	0.0438
Treasury Yield	−97.9267	<0.0001	−97.7361	<0.0001	−97.5544	<0.0001	−97.4721	<0.0001	−96.7484	<0.0001	−96.8044	<0.0001	−96.7378	<0.0001	−96.5992	<0.0001
SPRET	3.6589	0.6453	3.7710	0.6346	4.1218	0.6034	4.0783	0.6068	3.4149	0.6660	3.5778	0.6511	3.8934	0.6228	3.8743	0.6240
VIX	0.0006	0.9949	−0.0018	0.9855	0.0006	0.9954	0.0035	0.9720	0.0201	0.8414	0.0136	0.8928	0.0147	0.8842	0.0181	0.8573
TSLOPE	1.2376	0.2052	1.1408	0.2429	1.1989	0.2186	1.2103	0.2137	1.3502	0.1653	1.2481	0.2007	1.3023	0.1812	1.3132	0.1000
Year Dummy	Yes		Yes		Yes		Yes		Yes		Yes		Yes		Yes
R2	0.9917		0.9917		0.9918		0.9920		0.9918		0.9917		0.9917		0.9918
N	609		609		609		609		609		609		609		609

, respectively.

The results in Table 7 show that the coefficients on large bank indicators are statistically insignificant across specifications, regardless of whether credit ratings are included as control variables. The results in Table 7 suggest that the disappearance of the discount in yield for big banks started before the rating agencies’ formal decision to remove bank size from their rating methodologies.

Table 8 covers the period after this policy change and extends through 30 June 2025. This period includes the 2023 regional banking crisis, which began with the failure of Silicon Valley Bank on 8 March 2023 and culminated in the resolution of the First Republic Bank on 1 May 2023. Those episodes triggered a brief but substantial flight-to-safety in the markets, driving investors to large and systemically important banks from regional financial institutions. Although short-lived, this regional banking crisis reactivated a modest “too-big-to-fail” phenomenon, as investors perceived large banks to be better insulated from deposit outflows and liquidity stress. Moreover, the perception is likely to persist beyond the crisis window. Therefore, the results in Table 8 display a re-emergence of yield discount for large banks even after rating agencies no longer used size as a determinant for credit ratings. Heightened liquidity stress among regional banks and the resulting flight-to-quality in bond markets appear to have rewarded large institutions with a safety premium. Altogether, the results in Table 7 and Table 8 suggest that the absence in yield discount is confined to the relatively calm post-Dodd–Frank period but the absence in yield discount was reversed when systemic stress intensifies.

As an additional robustness check, we estimate dynamic panel models using Blundell–Bond system GMM and Arellano–Bond difference GMM to address potential endogeneity of credit ratings and bank size, persistence in yield spreads, and reverse causality between bank size and borrowing costs.

Table 9. Dynamic panel GMM robustness tests. (A) System GMM specification diagnostics. This table reports specification diagnostics for the Blundell–Bond system GMM estimations of bond yield spreads. The dependent variable is the yield spread in basis points. All models include a lagged dependent variable and the same bond-specific and market control variables as in the baseline specifications. Credit ratings and TBTF size indicators are treated as endogenous and instrumented using their own lagged values. To ensure feasible identification and limit instrument proliferation, the sample is restricted to issuers with at least five usable time observations, and the instrument lag depth is capped (MAXBAND = 3). Reported diagnostics include tests for serial correlation in first-differenced errors and the Hansen (Sargan) test of overidentifying restrictions. The number of instruments is also reported. (B) Coefficient estimates. This table reports coefficient estimates from dynamic panel GMM regressions of bond yield spreads on lagged yield spreads, TBTF size indicators, bond-specific characteristics, and market-level controls. The models are estimated using Blundell–Bond system GMM and Arellano–Bond difference GMM, treating credit ratings and, in alternative specifications, TBTF indicators as endogenous and instrumenting them with their own lagged values. All specifications include year fixed effects.

(A)
Model	AR(1) z	AR(1) p	AR(2) z		AR(2) p	Hansen/Sargan p		# Instruments
Pre-DFA System GMM	−2.72	0.0065	—		—	0.3818		XX
Pre-DFA Diff GMM	−2.51	0.0121	—		—	0.3483		XX
Post-DFA System GMM	−1.80	0.0717	—		—	0.1878		XX
Post-DFA Diff GMM	−2.02	0.0429	—		—	0.3913		XX
(B)
	Panel A Subsample: Banks Pre-DFA				Panel B Subsample: Banks Post-DFA
Variable	Coefficient	Std. Error	t-stat	p-value	Coefficient	Std. Error	t-stat	p-value
Lagged Yield Spread	−0.0828	0.0207	−4.01	<0.001	0.0146	0.0397	0.37	0.7143
Big3Bank	7.6262	12.1321	0.63	0.5304	2.3475	8.9109	0.26	0.7924
Big5Bank	−25.4619	10.4397	−2.44	0.0157	−8.1321	5.568	−1.46	0.1451
Big10Bank	−25.4007	10.81	−2.35	0.0199	−21.8004	11.2627	−1.94	0.0537
Big20Bank	−30.4076	6.0478	−5.03	<0.001	−0.6592	6.7716	−0.1	0.9225
Controls	Yes				Yes
Year Dummy	Yes				Yes

A reports the standard GMM specification diagnostics, including tests of serial correlation and instrument validity, while Table 9B reports the coefficient estimates, with Panel A presenting results for the pre-DFA period and Panel B presenting results for the post-DFA period. The diagnostic statistics in Table 9A indicate no evidence of instrument invalidity, and the serial correlation tests behave as expected for dynamic panel specifications. The coefficient estimates in Table 9B Panel A shows that, in the pre-DFA period, broader TBTF indicators, such as Big5, Big10, and Big20, remain negative and statistically significant in the system GMM specifications, implying economically significant yield discounts even after accounting for dynamic adjustment and internal instrumentation. The Big3 indicator is not statistically significant in the dynamic framework, due to its limited within-issuer time variation and the reliance of GMM estimators on within-panel variation for identification. Table 9B Panel B shows that, in the post-DFA period, TBTF indicators are generally insignificant and economically small across GMM specifications, consistent with a substantial reduction in size-related yield discounts following post-crisis regulatory reform. In sum, the dynamic GMM evidence reinforces our main conclusion that TBTF-related yield discounts are present prior to the Dodd–Frank Act but largely disappear in the post-crisis regulatory environment.

6. Conclusions

Before the enactment of the Dodd–Frank Act, credit agencies used bank size as one factor in determining banks’ credit ratings. As studies have shown that higher credit ratings are associated with lower borrowing costs, big banks obviously can issue bonds at a lower yield due to the higher credit ratings that are higher than can be justified based only on their intrinsic creditworthiness. Therefore, the TBTF problem has already been reflected in big banks’ credit ratings. In this paper we further examine whether the bond market gives an additional yield discount to big banks over and above the lower yield that has already been reflected in the higher credit ratings. Using a two-stage regression analysis, we identified an additional yield discount for bonds issued by big banks before the Dodd–Frank Act.

To further understand the TBTF problem, we investigate the issue from a different angle. We compare the yield spread on bonds issued by big banks with the bonds issued by non-big banks. We find that big banks not only pay a lower yield than non-big banks with the same credit rating, but they also even pay a lower yield than non-big banks with higher credit ratings. In other words, we identified three levels of discount in yield received by big banks. Given that big banks already receive a higher credit rating due to their size and given that studies have documented that bonds with higher ratings pay a lower yield, even when the yields on bonds issued by big banks are the same as non-big banks, big banks receive a discount in yield due to their elevated ratings; this is the first level of discounts big banks receive. When the yield on bonds issued by big banks is lower than the yield on bonds issued by non-big banks with the same rating, this is the second level of discounts received by big banks. In our study, we not only find that the yields on big banks’ bonds are lower than the yields on non-big banks with the same ratings, we find that the yields on big banks’ bonds are even lower than the yields on non-big banks’ bonds with higher credit ratings; this is the third level of discounts received by big banks.

We also find that this additional yield received by big banks disappeared after the enactment of the Dodd–Frank Act. Also, soon after the enactment of the Dodd–Frank Act, rating agencies abolished their policy of using banks’ size as a determinant of banks’ credit ratings. That change in policy provides another opportunity to examine whether the disappearance of the big-bank yield discount was due to the Dodd–Frank Act or due to credit agencies’ changes in policy. We examined the issue and found the disappearance of the yield discount received by big banks started before, not after, credit rating agencies’ changes in policy. In summary, the evidence suggests that post-crisis regulation meaningfully restored market discipline by weakening expectations of government support in normal market conditions. Viewed through our three-level severity framework, Dodd–Frank effectively reduced the more severe forms of TBTF pricing, although episodes of systemic stress may still trigger temporary flight-to-safety behavior. More broadly, conceptualizing TBTF as a continuum rather than a binary outcome clarifies how moral hazard, regulatory credibility, and financial stability evolve across regulatory regimes and market environments.