Market Discipline through Credit Ratings and Too-Big-to-Fail in Banking

Do credit ratings help enforce market discipline on banks? Analyzing a uniquely comprehensive dataset consisting of 1,081 rating change announcements for 154 international financial institutions between January 2004 and December 2015, we find that rating downgrades for internal reasons, such as adverse changes in the operating performance or capital structure of banks, are associated with a significant CDS spread widening. However, this widening only occurs for banks that are not perceived as to be Too-Big-to-Fail (TBTF). Our findings question the reliability of credit ratings as a tool to discipline TBTF banks and suggest that regulatory monitoring should remain the main mechanism for disciplining these banks.


Introduction
Credit ratings play a significant role for banks as they provide important information on the overall riskiness of a bank to investors and its counterparties. If the rating of a bank is low, its refinancing costs will be high, while it may simultaneously lose business to other banks due to its high counterparty risk. 1 Therefore, having a comparatively high and stable rating is of great importance for any financial institution.
However, banks (not unlike any other rated entities) are often only partially able to influence their own credit rating. A bank's rating can, for example, be tied to the sovereign rating of the country (in which the bank has its main operations) due to the sovereign rating ceiling. 2 Moreover, a deteriorating economic environment or changes to the methodology CRAs use to rate financial institutions are also beyond the control of banks and may lead to adverse rating changes (e.g., King, Ongena and Tarashev, Forthcoming). Nevertheless, even though banks operate in a highly cyclical environment, they still have the ability to influence certain factors that determine their rating.
In this study, we therefore test whether market discipline is relevant for Too-Big-to-Fail banks and how implicit TBTF guarantees may affect credit default swap (CDS) markets in particular. Particularly in the context of market discipline, credit ratings play a major role because they factor into the calculation of capital requirements for interbank claims and serve as a benchmark for the internal rating approach (see e.g., Basel Committee on Bank Supervision, 2006Supervision, , 2011. The immediate motivation for our study is that a bank that is considered systemically important, however, may not need to make any adjustments even if it was downgraded for internal reasons, because changes to its credit rating may not matter much as a result of an implicit assumption by market participants that the bank will be bailed out by the government if there is a threat of bankruptcy (e.g., Hett and Schmidt, 2017). 3 We use CRAs' credit rating announcements, i.e., rating downgrades or upgrades, as easily visible events to market participants, to investigate whether the creditworthiness of a bank is increasing or deteriorating. We analyze the reaction of the banks' CDS spreads, which are a direct measure of a bank's creditworthiness and default probability. The CDS market is viewed as one of the most prevalent channels through which new, credit relevant information can be transmitted and which contains unique firm information that is not captured by other markets (Lee, Naranjo, Velioglu, 2018). Indeed, given its informed trading (Acharya and Johnson, 2007), any deterioration or improvement in a bank's creditworthiness should therefore immediately be reflected in a widening or tightening of the CDS spreads. Consequently, examining the CDS market as one of the most prevalent, market-based measures of credit risk offers valuable insights into the way market participants evaluate rating changes in light of potential TBTF considerations.
Prior research on the effect of rating change announcements on firms' CDS spreads generally documents a significant widening of CDS spreads as a result of credit rating downgrades. 4 The majority of prior studies, however, does not focus on one specific industry or the financial sector in 3 TBTF was first acknowledged by the U.S. Comptroller of the Currency in 1984 following a U.S. Senate hearing when he defended the bailout of Continental Illinois by the Federal Deposit Insurance Corporation and the U.S. government. Since then, research on the existence, effects, and consequences of TBTF increased significantly (e.g., O'Hara and Shaw, 1990;Morgan and Stiroh, 2005;Brewer and Jagtiani, 2013). 4 For example, Hull, Predescu and White (2004), Norden and Weber (2004), Galil and Soffer (2011), Finnerty, Miller and Chen (2013), and Norden (2017. In contrast, for rating upgrades, the majority of prior studies fails to observe a significant CDS market reaction to rating upgrade announcements (Hull et al., 2004;Norden and Weber, 2004) and only more recent evidence points to a small, but significant, tightening of CDS spreads following a rating upgrade (Finnerty et al., 2013;Galil and Soffer, 2011). particular, thereby neglecting the possibility that, for example, rating announcements for closely regulated banks may lead to different CDS market reactions. TBTF considerations are strongest in the banking sector and the expectation of a government bailout in case of financial distress may shape investors' reactions to downgrade announcements in particular. Especially for TBTF banks, rating downgrades may not serve as an appropriate channel through which market discipline can be exerted on banks given the existence of implicit TBTF guarantees by governments. As the studies of Rime (2005) and Ueda and Weder di Mauro (2013) document, the credit ratings of large banks are distorted as a result of TBTF considerations. As a consequence, the CDS spreads of TBTF banks may not widen appropriately following a rating downgrade due to expectations of government bailouts. Therefore, studying the effect of rating changes on the CDS spreads of financial institutions will offer valuable insights into the efficacy of rating changes as an appropriate channel through which market discipline is enforced on financial institutions. A potential break down of one of the mechanisms of market discipline inevitably questions the viability of credit ratings as an appropriate tool to discipline banks for excessive risk taking and should consequently be a concern to regulators and policy makers alike.
We analyze the effect of TBTF considerations on credit rating changes for financial institutions in two steps. First, we examine how banks' CDS spreads react to rating change announcements, depending on the reason of the rating change. It may well be that the CDS spread reactions differ if the rating changes are due to external reasons outside the direct control of the bank, such as a deteriorating economic climate or changes to the CRA's rating methodology, or internal reasons within the purview of a bank, such as a poor operating performance. Second, we investigate whether TBTF benefits exist in debt capital markets by differentiating between CDS spread reactions of TBTF and non-TBTF banks to rating downgrades. Rating downgrades may have a limited Electronic copy available at: https://ssrn.com/abstract=2928113 effect on the perceived creditworthiness of TBTF banks, rendering downgrades an ineffective mechanism for the enforcement of market discipline.
We construct a uniquely comprehensive dataset consisting of 1,081 rating change announcements, 782 rating downgrades and 299 rating upgrades, for 154 international financial institutions between January 2004 and December 2015. The analysis of the CDS spread reaction to rating changes reveals that only rating downgrades due to internal reasons, such as adverse changes in the operating performance or capital structure of the bank, are associated with a large and significant widening of CDS spreads. For TBTF banks, however, rating downgrades attributed to internal reasons do not lead to any CDS spread changes, while a significant widening of CDS spreads is observed for non-TBTF institutions. The results also hold after controlling for other factors that may influence the CDS spread reaction to rating changes, such as a bank's initial rating or its leverage. The loading factors for the variables in the regression analysis suggest that, all else being equal, following a rating downgrade due to internal reasons, the CDS spread of a non-TBTF bank will widen approximately 67 basis points (bps) more than the CDS spread of a TBTF institution.
This, for example, implies that for a non-TBTF bank with a letter rating equivalent to an A, the CDS spread will widen by roughly 86 bps, compared to 19 bps for a TBTF bank in case of a one notch rating downgrade. The results are robust to different methodology specifications, such as alternative empirical approaches, model specifications and time-varying changes in the regulatory environment.
Our findings make multiple contributions to the existing literature. First, we contribute to the empirical literature on TBTF by analyzing whether the effects of credit rating downgrades differ depending on the importance of banks. By investigating the CDS market reaction for TBTF and non-TBTF banks to credit rating changes, we go beyond investigating the effect of bank size on rating Electronic copy available at: https://ssrn.com/abstract=2928113 levels (e.g., Völz and Wedow, 2011;Ueda and Weder di Mauro, 2013) and directly test whether size influences the market reaction itself. We document that TBTF banks, which are larger and more important, possess a comparative advantage in credit markets over their smaller peers. Third, and most importantly, by showing that TBTF banks' CDS spreads are not affected by rating downgrades, we present novel evidence that rating downgrades are not an appropriate mechanism through which market discipline can be enforced on banks. This breakdown in one of the mechanisms of market discipline questions the viability of credit ratings as a tool to curb banks' excessive risk taking and highlights the importance of regulatory discipline as the main mechanism for monitoring TBTF banks. Second, we add to prior research on the effects of rating announcements on the CDS spreads of firms (e.g., Hull. et al., 2004;Norden, 2017). By focusing exclusively on financial institutions, we are able to offer new insights into the importance of the reasons for rating announcements and how they affect the refinancing costs of firms.
The rest of this paper is structured as follows. Section 2 describes the data set and methodology and Section 3 presents the empirical results on the effect of rating changes on banks' CDS spreads, depending on the reason of the rating announcements. Section 4 provides an analysis of the differential impact of rating downgrades on the CDS spreads of TBTF and non-TBTF banks, while Section 5 provides several robustness tests, including a comparative analysis of the stock and bond market reactions. Section 6 concludes the paper.

Sample construction
Our analysis is based on a global sample of banks with a long-term issuer rating by S&P, Moody's and/or Fitch with CDS spread data available through Thomson Reuters Composite EOD. Thomson Reuters offers data on CDS spreads beginning in 2004 and our analysis covers the 12-year time Electronic copy available at: https://ssrn.com/abstract=2928113 We identified a total of 3,393 rating change and rating review announcements by S&P, Moody's and Fitch between the years 2004 and 2015. The total sample is split into 2,186 rating downgrade and rating review for downgrade announcements and 1,207 rating upgrade and rating review for upgrade announcements. Figure 1 shows the total number of rating downgrades and upgrades during our entire investigation period. On average, each bank in our sample receives approximately 7 downgrades and 4 upgrades over all three major CRAs, which equals one rating announcement every 7 months. The average time period between two announcements is 272 trading days with a median of 120 trading days. Most rating downgrades occur between the third quarter of 2008 and the second quarter of 2009, the height of the recent financial crisis. Prior to the crisis, upgrades occurred more frequently than downgrades and a spike in the number of upgrades can be observed during the second quarter of 2007, just before the start of the financial crisis. During the crisis, very few rating upgrades are observed. Following the crisis, still fewer rating upgrades than downgrades are observed. Moreover, the number of rating downgrades spiked again during the fourth quarter of 2011 and the second quarter of 2012, the height of the European sovereign debt crisis.
Particularly Greek and Spanish banks were frequently downgraded, their average rating decreased by 7 and 5 notches, respectively. Since the third quarter of 2012, the number of downgrades appears to have stabilized. Figure 1 also shows that the average bank rating increased between 2004 and the first quarter of 2008 from A-/A3 to A/A2. However, following the financial crisis, the average rating decreased by more than two notches and stood at BBB+/Baa1 at the end of 2015.
This is an indication that CRAs started to more carefully assess the credit risks of banks in the wake of the financial crisis and are now less willing to provide high credit ratings. This is in line with the findings of Dimitrov, Palia and Tang (2015) who show that after the Dodd-Frank act CRAs issue lower ratings. Moreover, the finding is also in line with the reputation model developed by Morris (2001), suggesting that CRAs in the post-Dodd-Frank period may be more protective of their reputation. It is, however, noteworthy that the average rating of a non-US and non-European financial institution increased by roughly 1 notch during our observation period, possibly reflecting the increased importance of banks headquartered outside of the U.S. and Europe.
[Please insert Figure 1 around here] Our starting sample contains all 3,393 rating review and rating change announcements by the three major CRAs for our sample of 212 financial institutions. As the focus of the present study is on the CDS market reaction to rating change announcements, we drop all rating reviews from our sample in a first step. Next, all events are dropped for which CDS data is not available or not in sufficient quality during the estimation or event period. In a further step, we eliminate all events for which the reason for the rating change could not be properly determined. Finally, we exclude all events for which accounting data for the subsequent regression analyses are missing. This leaves us with a final sample of 1,081 rating change announcements for 154 different financial institutions, of which 28 are considered TBTF, from 36 different countries for our analysis: 782 rating downgrades and 299 rating upgrades (Table A.1 in the Appendix offers a detailed overview of the number of events by country and year). 9 Table 1 provides an overview of the sample selection procedure. 9 TBTF appears to be universal phenomena and in this context national regulatory systems only play a subordinated role. The G-SIBs, for example, are compromised of the globally largest financial institutions, irrespective of their country of origin and regulation of these institutions goes beyond the national level. Furthermore, prior research also does not differentiate between different legal systems (e.g., Cubillas, Fernandez and Gonzalez, 2017;Ueda and Weder di Mauro, 2013), thereby acknowledging the universality of TBTF.
[Please insert Table 1 around here] In line with Goh and Ederington (1993), Bannier and Hirsch (2010), and Imbierowicz and Wahrenburg (2013), we explicitly focus on the reason behind a rating announcement. We allocate the rating change announcements into either internal or external reasons. Both reasons are divided into three subcategories. Internal reasons are rating announcements due to changes in capital structure, operating performance, or M&As, while external reasons are a result of circumstances that are outside the direct control or influence of the bank. They are further divided into rating announcements due to changes in the macroeconomic or regulatory environment, adjustments to the sovereign rating, and changes in the methodology the CRA uses to assess the banks' ratings. We use 76 keywords that are frequently mentioned as a reason and sort them in order of appearance in the press release of the CRA (see also Table A.2 in the Appendix for a full list of all keywords by category). In case more than one keyword appeared in the press release, we follow Imbierowicz and Wahrenburg (2013) and assign the event to the first keyword mentioned in the press release and allocate the keywords to one of the six categories. 10 Goh and Ederington (1993) divide rating announcements into two broad categories: deterioration or improvement in the firm's earnings and in actions or decisions that result in a change in the firm's leverage. Both categories are part of our internal reasons. We add rating change announcements because of merger activity to our internal reasons, as M&As are usually actively pursued by a firm and can affect the operating performance and capital structure of a company in multiple ways. In contrast, external reasons are a result of circumstance that are outside the direct control or influence of the bank. These include an improvement or deterioration in the economic climate or changes in the overall regulatory framework, adjustments to the CRAs' rating methodology, and rating changes due to adjustments in the sovereign rating of the country in which the bank has its main operations or headquarter. Table 2 provides an overview of the different reasons for the rating change announcement by year. 373 (35%) of the events in the sample are due to internal reasons, while 708 (65%) events are due to external reasons. This already shows that approximately two thirds of the rating change announcements are outside of the direct control of the bank. 2007. This is in line with the observation in Figure 1 that the average bank rating underwent a pronounced increase during those years. Rating upgrades are largely driven by reasons outside of the direct control of the bank, as upgrades due to adjustments in the CRAs' rating methodology and due to sovereign rating changes account for 153 of the 299 upgrades in our sample. Overall, the distribution of rating changes by year and reasons underlines the highly cyclical business of firms operating in the financial sector.
[Please insert Table 2 around here]

Empirical methodology
We use CDS spreads to determine the debt market reaction to rating announcements. CDS are a viable, market-based measure of the credit risk of firms and are extensively used by investors in order to hedge and invest in credit risk (Gilchrist and Zakrajsek, 2013). One of the main advantages of CDS is the ease with which to buy credit protection compared to shortening bonds. In addition, Electronic copy available at: https://ssrn.com/abstract=2928113 prior research documents that information in the CDS market tends to spill over to bond spreads (Blanco, Brennan and Marsh, 2005), that CDS generally reflect new information more quickly than bond markets (e.g., Longstaff, Mithal and Neis, 2005) and also contain unique information not readily present in other markets (Lee et al., 2018).
In line with the vast majority of existing literature, we use the five-year tenor senior CDS mid spread in U.S. dollars (e.g., Longstaff et al., 2005;Jorion and Zhang, 2007;Demirgüc-Kunt and Huizinga, 2013) as it is by far the most commonly traded and liquid tenor in the market.
We analyze CDS spread changes as a measure for the daily change in the credit risk of a bank as the goal of this study is to document the changes in banks' CDS spreads subsequent to rating announcements. Following, among others, Jorion and Zhang (2007) and Norden (2017), we use daily absolute spread changes , for bank i at time t: where , , , and , | are the adjusted, realized, and normal spread changes, respectively, for bank i's CDS on day t. The normal spread change is the expected spread change conditional on the information set and is estimated using a 120-trading day period prior to the first day of the event window: We calculate cumulative adjusted CDS spread changes (CASCs) by adding daily ASCs from day τ1 to τ2. CASCs are computed for each day of the event window [τ1; τ2] in [-10…,+10]. In order to test whether the mean CASCs differ significantly from 0, we use the standard cross-sectional parametric t-test as well as the nonparametric Wilcoxon signed-rank test. The difference between two samples are tested using the two sample t-test and the Mann-Whitney U test. [Please insert Table 3 around here]

The effect of credit rating changes on banks' CDS spreads
Electronic copy available at: https://ssrn.com/abstract=2928113 By further dividing the sample into downgrade announcements due to internal reasons (Table 3 Panel B) and external reasons (Table 3 Panel  into the banks' CDS spreads. 12 Moreover, the difference between rating downgrades due to internal and external reasons is highly significant for almost all event windows (Table 3 Panel D). Table 4 shows the CDS event study results for rating upgrade announcements. For the entire sample of rating upgrades, a small, but significant tightening of CDS spreads can be observed. However, compared to rating downgrades, the effect of rating upgrades is low and no significant CDS spread changes can be observed during the [-10;+10] day event window, suggesting that rating upgrades have little to no impact on banks' CDS spreads.
Splitting the sample of upgrade announcements into upgrades due to internal reasons (Table 4 Panel B) and external reasons (  Table 4 Panel D). 13 [Please insert Table 4 around here] Overall, credit rating downgrades of banks lead to a large, significant widening of CDS spreads, while rating upgrades are only associated with small and at best weakly significant tightening of CDS spreads. This finding, and the asymmetric nature of the CDS market reaction to rating change announcement, is in line with large parts of the literature (e.g., Hull et al., 2004;Finnerty et al., 2013). Figure 2 illustrates this result. The results, however, also document that the widening of CDS spreads following rating downgrades is entirely driven by downgrade announcements due to 13  internal reasons (see also Figure 2 Panel A). Rating upgrade announcements due to external reasons are associated with a small, but significant tightening of CDS spreads, while rating upgrades for internal reasons lead to a small widening of CDS spreads (see also Figure 2 Panel B). Yet, compared to the CDS market reaction to rating downgrades, the reaction to rating upgrades can considered to be negligible. 14 [Please insert Figure 2 around here]

The impact of rating downgrades on the CDS spreads of Too-Big-to-Fail versus regular banks
The previous section clearly demonstrated that only rating downgrade announcements attributed to internal reasons lead to clear CDS market reactions. The CDS market reaction to rating upgrades is negligible compared to downgrades. We therefore forgo a deeper analysis of the effect of rating upgrades and largely concentrate our further examinations on the impact of TBTF rating downgrade announcements. Focusing on internal reasons has the additional advantage that these downgrade announcements relate to a single bank only, which allows for clearer conclusions with regard to the effect of TBTF consideration on the CDS spread of financial institutions. Rating change announcements due to external reasons usually affect a large number of banks simultaneously, which in turn may blur results and therefore not allow for a clear isolation of the effect of TBTF.
In particular, we anticipate that rating downgrade announcements for banks that are considered TBTF will have little to no impact on the CDS spreads of these institutions, as market participants are likely to expect a government bailout in case the bank faces bankruptcy. For non-TBTF banks, on the other hand, we expect a significant widening of CDS spreads.
In this context, the definition of TBTF is important in order to properly assess the effect of TBTF considerations on rating changes. Prior research that focused exclusively on the U.S. frequently used the 11 largest banks by assets to define TBTF banks (e.g., O'Hara and Shaw, 1990;Morgan and Stiroh, 2005). For international samples, the more recent definition of Global Systemically Important Banks (G-SIBs) 15 is often used (e.g., Moenninghoff, Ongena and Wieandt, 2015). The size of a bank's total assets is clearly important to be perceived as being TBTF, and several studies that use samples consisting of international banks (e.g., Demirgüc-Kunt and Huizinga, 2013;Cubillas et al., 2017) focus on some measures based on total assets. As we likewise have an international sample of financial institutions from 36 different countries, we define TBTF by the total assets of a bank in its country of origin relative to other banks in that country. In particular, we treat a bank as being TBTF if it was among the three largest banks in its country of origin as measured by the total assets at the end of the year prior to the rating change. We believe it is reasonable to assume that a national government would not allow one of the three largest financial institution in the country to fail, as the repercussion for the financial system are potentially too severe, thereby justifying a government bailout. This definition of TBTF is also used by Cubillas et al. (2017). Table 5 presents the results of the CDS event study comparing rating downgrades due to internal reason for TBTF and non-TBTF banks. It can clearly be seen that the CDS market reaction to rating downgrades for TBTF banks is small and largely insignificant (see Table 5 Panel A). There appears to be a weakly significant widening in CDS spreads for TBTF banks during the [-30;-11] day event window, which may indicate some information leakage, but the spread changes are comparatively small. 16 In contrast, the CDS spreads of non-TBTF banks undergo a significant widening during almost all event windows (see Table 5 Panel B [Please insert Table 5 around here]

CDS event study results
The difference between the CDS market reaction to rating downgrade announcements between TBTF and non-TBTF banks is also very pronounced (see Table 5 Panel C). With the exception of the announcement day and the [-30;-11] and [-1;+1] day event windows, respectively, the mean and median CASC are significantly lower for TBTF banks than for non-TBTF banks. The difference in the mean CASC amounts to -76.30 bps during the [-10;+10] day event window, giving a strong indication that CDS market participants distinguish between banks that are likely to receive a bailout by their country's government in case they experience financial distress and those that 16 We also calculate the CASC for the [-30;-11] day event window in this section and several of the following sections in order to alleviate concerns with regard to potential leakage. For this event window the expected spread changes are computed using a 120-trading day period from t=-150 to t=-31.
are unlikely to receive such a treatment. The assumption of TBTF status by CDS market participants leads to a very subdued market reaction for TBTF banks, providing evidence that the market discipline mechanism, which downgrades by CRAs should provide, is actually not functioning properly or not present at all. 17 Figure 3 Panel A illustrates the CDS spread development for TBTF and non-TBTF banks during the [-30;+10] day event window for internal downgrade reasons. There is a very pronounced widening of CDS spreads for non-TBTF banks, starting approximately five days prior to the rating downgrade announcement and ending five days following the rating announcements. In stark contrast, the CDS spreads of the TBTF banks remain comparatively stable and show no distinct pattern overall. 18 Figure 3 Panel B shows the CDS spread development for external downgrade reasons divided into TBTF and non-TBTF banks. It can be seen that the tightening of CDS spreads is driven by non-TBTF banks following the downgrade announcements. This tightening may be due to market participants anticipating multi-notch downgrades, which subsequently did not occur.
[Please insert Figure 3 around here]

Regression results
The observed CDS spread changes following rating downgrade announcements may be driven by more factors than just TBTF considerations. In order to account for these factors, we run several regressions to determine the impact of different definitions of TBTF on our results and of other 17 Table A.5 in the Appendix shows the difference between TBTF and non-TBF banks for rating downgrades due to external reasons. For external reasons, the tightening of CDS spreads is largely driven by non-TBTF banks, while for TBTF banks largely insignificant and negligible reactions are observed. 18 The results also hold when splitting the sample of rating downgrades for internal reasons into different regions and countries (see Figure A.2 in the Appendix). Noting the comparatively small sample sizes, the difference in absolute spread changes between TBTF and non-TBTF banks varies between the different regions and countries and downgrades appear to matter more to TBTF banks in smaller countries (e.g. the Netherlands and Italy).
rating-, event-, and bank-specific variables, including the most prominent variables used in prior literature. The ordinary least squares (OLS) regression takes the following form: both also binary variables that take the value of 1 if the bank was the largest bank or among the five largest banks as measured by total assets in its country of origin at the end of the year prior to the rating change announcement, respectively, and take the value of 0 otherwise. In addition, we Z , is defined as a vector of bank-specific variables that includes OPAQUENESS, which is defined similarly as in Morgan (2002) and Bannier, Behr and Güttler (2010): it takes the value of 0 if the bank has a rating by S&P, Moody's and Fitch, 1 if the banks has a rating by two of the three CRAs, and 2 if the bank is rated by only one CRA 20 . As financial institutions are considered to be opaque compared to other firms (see e.g., Morgan, 2002;Hirtle, 2006), we introduce this variable to test whether differences in opaqueness matter. Further bank-specific variables are ROA, defined as the bank's return on assets (ROA) in the year prior to the rating announcement, LEVERAGE, defined as the ratio of total liabilities to total equity of the bank at the end of the year prior to event, LIQUIDITY, defined in analogy to Kapadia and Pu (2012) as the ratio of non-zero daily CDS spread changes to the total number of trading days during the 120-day estimation period, 21 and LIABILITIES/GDP, which is a binary variable that is equal to 1 if the ratio of the bank's total liabilities divided by the GDP of the bank's country of origin in the year prior to the event exceeds 20 Morgan (2002) and  use split ratings between the CRAs, while we differentiate between the number of available ratings as a proxy for opaqueness. However, we also ran the regression using split ratings and analyst coverage as measures of opaqueness. The results remain unchanged. 21 Further tests showed that there is no significant difference in the liquidity measure between estimation period and the event window or between TBTF and non-TBTF banks. 0.5, and 0 otherwise. This variable controls for the systemic importance of a bank and for potential to Too-Big-to-be-Rescued considerations (see e.g., Völz and Wedow, 2011;Cubillas et al., 2017).
The balance sheet data and a bank's ROA are taken from Bankscope and GDP data is obtained from the World Bank database. Additional bank-specific variables include GOVERNMENT, which is a binary variable that takes the value of 1 if the bank is a policy bank or the government holds significant stake in the bank (e.g., Royal Bank of Scotland following a capital injection by the British government in October 2008), and 0 otherwise, IB, which is equal to 1 if the bank's first two digits of the SIC code start with 62, and 0 otherwise, and EU, equal to 1 if the firm's headquarter is located in Europe, and 0 otherwise. The variable IB essentially captures investment banks and asset managers, while EU accounts for the fact that European governments provided blanket guarantees to depositors and creditors of ailing financial institutions during the crisis, potentially reinforcing existing TBTF considerations. The interaction term TBTF X INTERNAL is negative and highly significant in all model specifications, providing robust evidence that the CDS spreads of TBTF banks are considerably less affected by rating downgrades than those of non-TBTF banks. It is also noteworthy that the effect of TBTF is most pronounced when using the variable TBTF1, as the coefficient for this variable is larger compared to the coefficients for TBTF and TBTF5. In line with expectations, the more banks are included in the definition of TBTF, the lower the coefficient, with the coefficient for 22 The descriptive statistics for the variables used in the regression analyses are presented in Table A.6 in the Appendix.
We also ran the same regressions only using the sample of 242 observations for internal reasons. The results are presented in Table A.7 in the Appendix and further support the results for the overall sample. TBTF5 X INTERNAL generally being the lowest, showing that only the largest banks in a country benefit most from TBTF considerations.
The rating-specific variables also have a significant influence on the observed CDS spread changes. The coefficient for INTERNAL is highly significant and positive in all models, further confirming that rating changes due to internal reasons lead to a significant widening of CDS spreads for non-TBTF banks (this is also in line with the CDS development observed in Figure 3 Panel A). Furthermore, the coefficient for RATING is negative and significant in all model specifications, indicating that a higher rating is associated with lower CDS spread changes. In addition, the coefficient for the variable REVIEW is likewise negative in all model specifications, but only significant in the first three model specifications. These estimates show that a rating review prior to the rating downgrade announcement leads to a significantly lower CDS market reaction, suggesting that CDS market participants may have anticipated the rating change. The coefficient for [Please insert Table 6 around here] Among the event-specific variables, only the coefficient for CRISIS is significant in Models 4 through 6. The positive sign of the coefficient documents that CDS spreads of financial institutions widened significantly if the rating downgrade occurred during the recent financial crisis. 23 The coefficients for the other event-specific variables lack significance, indicating that the CRA that announced the downgrade does not influence the observed CDS spreads surrounding a rating change.
With respect to the bank-specific variables, the coefficient for the variable LIABILITIES/GDP is positive and significant in several of the regression models. This provides some evidence that Too- Overall, TBTF considerations appear to have a significant effect on CDS spread changes surrounding credit rating downgrades, even after controlling for other potential influence factors. This provides further evidence that market discipline mechanisms, such as credit rating downgrades, do not appear to work for financial institutions that are perceived to be TBTF and that markets assume 24 Even in 2017, almost a decade after the end of the financial crisis, particularly European governments still prefer to bail out financial institutions rather than letting them go bankrupt, even if they are not TBTF (e.g., Italy in 2017 (Financial Times, 2017)). 25 We also ran the regressions dropping all observations where the variable GOVERNMENT is equal to 1, as banks with partial government ownership may be regarded as TBTF simply due the government's involvement. The results are largely unchanged.
implicit government guarantees. Even if the creditworthiness of these institutions deteriorates and their credit rating is lowered, this does not lead to a corresponding widening in CDS spreads, giving TBTF banks a clear advantage over non-TBTF banks in refinancing and funding operations.
Therefore, the enforcement of regulatory discipline appears to be of particularly importance for TBTF banks. Nonetheless, some of the rating-specific variables, such as the number of notches a rating was downgraded, rating review announcements prior to the rating change, and the bank's prior rating, still possess a significant influence on CDS spreads. In contrast, event-specific and bank-specific variables have a limited impact at best on the observed CDS spread changes surrounding rating downgrades. 26

Stock and bond market reactions
We conduct an equity and bond event study to test whether these markets react in a similar fashion to the CDS market to bank rating changes. 27 The results of the stock and bond event study are shown in Table 7 and Table 8, respectively. The stock event study documents for the sample of 606 events that rating downgrades are generally perceived as negative events and that downgrades due to internal reasons lead to more pronounced negative reactions than those for external reasons.
This corroborates the results of the CDS event study. Further differentiating downgrades due to internal reasons into those for TBTF and non-TBTF banks shows that the reaction is similar, even 26 We also ran the regression removing the countries with the least observations from our sample in order to compare TBTF and non-TBTF firms in the same country as there may not be enough counterfactual non-TBTF firms with rating downgrade information in smaller countries. We drop the observations from Kazakhstan (3 observations), Denmark (2), Canada, Norway, Russia, Korea, Turkey (1 each) from our sample and repeated the OLS regression. We find that the results remain qualitatively the same. 27 Compared to the analysis using CDS spreads, the number of events is lower for the stock and bond event studies as not all sample banks are listed on a stock exchange and not all banks had bond data in sufficient quality available. For reasons of brevity, the stock and bond event study methodologies are described in more detail in Appendix B.
though the overall stock price development is less negative for TBTF banks than non-TBTF banks during the [−10;+10] day event window. The negative returns indicate that stockholders may benefit less from potential TBTF considerations, as debt and credit providers are usually the direct beneficiaries of a bailout. In addition, the results indicate that the stock market reaction is largely confined to the short term event windows as only these event windows show significant returns.
[Please insert Table 7 around here] The results of the bond event study for the sample of 524 events are presented in Table 8 and show that rating downgrades are generally perceived to be negative events, as bond prices experience a small but significant drop in value. This further supports the results of the CDS event study. Differentiating the sample of downgrades due to internal reasons again into TBTF and non-TBTF banks shows that there is no meaningful bond reaction for TBTF banks during the [−10;+10] day event window, while for non-TBTF banks a significant decline in bond prices can be observed.
The difference is significant for the [−5;+5] day event window. This result provides further confirmation of the reaction observed in the CDS market and offers additional support that particularly debt and credit markets take TBTF considerations into account.
[Please insert Table 8 around here] The results for the stock and bond event study are a graphically illustrated in Figure 4. The graphics show that stock as well as bond markets react to rating downgrades, particularly if they are due to internal reasons. Furthermore, the difference between TBTF and non-TBTF banks is persistent and in line with the observations of the CDS market. Therefore, TBTF considerations appear to play an important role in all three markets, whereby TBTF considerations are more prevalent in the CDS and bond market than the stock market.
[Please insert Figure 4 around here]

CDS market reaction to rating review announcements
Rating changes are usually carefully deliberated by CRAs and their ratings are supposed to look through the business cycle to ensure a certain degree of rating stability (Löffler, 2004). This, however, raises concerns with regard to the timeliness of credit rating changes. In this context, rating reviews allow CRAs to provide market participants with more timely information with respect to the creditworthiness of a firm, without resorting to direct rating changes (Boot, Milbourn and Schmeits, 2006). 28 Therefore, in order to alleviate concerns with respect to the timeliness of rating changes, we also look at the effect of rating review announcements on banks' CDS spread changes. 29 To obtain our sample for rating reviews, we start with the 1,029 rating review announcements that we collected from the websites of the three major CRAs and employ the same selection procedure as described in Section 2.1. We focus on rating reviews for downgrades, which leaves us with a final sample of 446 review announcements.
For the entire sample of rating reviews for downgrade, a significant widening of banks' CDS spreads can be observed (see Table A.9 in the Appendix). Splitting the sample into reviews for downgrade due to internal reasons) and external reasons leads to similar results as for the respec- 28 The three major CRAs, S&P, Moody's, and Fitch use different terminologies to describe the rating review process: S&P places a firm on "CreditWatch", while Moody's places a firm's rating on "Watchlist", and Fitch on "Rating Watch". 29 Prior research shows that CDS spreads widen following rating reviews for downgrade, analogous to rating downgrades, with most studies finding that review announcements have a similar effect to actual rating downgrades (see e.g., Hull et al., 2004;Norden and Weber, 2004;Galil and Soffer, 2011). For rating reviews for upgrade, the results are ambiguous, with Hull et al. (2004) and Norden and Weber (2004) failing to observe significant CDS market movements following rating reviews for upgrade, while Galil and Soffer (2011) document that rating reviews for upgrade lead to a significant tightening of CDS spreads.
tive rating downgrade announcements. Our prior result that CDS market participants react differently to rating announcements attributed to internal and external reasons is confirmed (see also Figure A.3 in the Appendix). The reaction to rating review for downgrade announcements, however, is overall less pronounced than for actual downgrade announcements.

Alternative methodologies and model specifications
We use multiple methodology modifications to further validate our results. As an alternative to the CDS event study methodology described in Section 2.2, we use benchmark adjusted CDS spread changes, an empirical set up that follows Hull et al. (2004), Jorion and Zhang (2007), and Norden (2017). This methodology adjusts the observed CDS spread changes by deducting the changes of a CDS spread index of the same rating class as the bank's initial rating using the equation: , , where is the adjusted CDS spread change of bank i on day t, , is the observed CDS spread of bank i on day t and It is the relevant CDS spread index for the rating class on day t. 30 This methodology is well suited for internal reasons, as these rating announcements focus on one bank only, thereby avoiding distortions in the benchmark index due to multiple banks being affected simultaneously, as often happens for external reasons. We therefore limit our examination to rating downgrades due to internal reasons. The results are line with those described in Section 30 The daily CDS spread index corresponds to the equally weighted cross-sectional mean of all CDS spreads for each of the six letter rating classes AAA/AA, A, BBB, BB, B, and CCC or below, and is composed of the CDS spreads of all banks with available data through Thomson Reuters EOD and a long-term issuer rating of either S&P, Moody's, or Fitch, excluding the event firm. AAA and AA rated banks are combined into one rating class, due to the small sample size of these firms. Our approach with respect to the benchmark also implies that the benchmark for each event is composed of all banks that have a rating by the CRA that announced the rating change (e.g., rating announcements by Moody's are compared with a benchmark consisting only of banks with a rating by Moody's).
4. 31 We again document that for TBTF banks rating downgrade announcements do not influence their CDS spreads, while the CDS spreads of non-TBTF banks widen significantly.
In addition, we test whether our results hold when using subordinated CDS spreads instead of senior ones. Even though, the five-year tenor is the most liquid trading segment in the CDS market, traders and bank supervisors may also focus on information contained in CDS spreads on subordinate bank debt given that they are more risk-sensitive. We find that our results are consistent when using the five-year subordinated CDS, as we only observe a significant market reaction for non-TBTF banks but not for TBTF ones. 32 With respect to our regression models, for the sub-sample of internal rating changes, we interact the TBTF variable with a set of other variables that are potential drivers of the observed CDS market reaction to test the robustness of the TBTF variable. This analysis aims to determine whether TBTF considerations are the "true" driver of the CDS market reaction to rating downgrades. To this end we interact TBTF with the standardized variables BANKSIZE, RATING and LIQUIDITY, as well as the variables, SOVEREIGN RISK and CDS TRADER. The variable BANKSIZE is defined as the banks total assets in billion U.S. dollars at the end of the year prior to the rating change and tests whether larger banks in general are more likely to be rescued. 33 The variable SOVEREIGN RISK is defined as GDP divided by total national debt of the bank's country of origin and captures the possibility that countries with high levels of national debt may not be in a financial position to bail out a struggling bank. The variable CDS TRADER is a binary variable 31 The results for market reaction using the benchmark model are shown in Table A .8. 32 Given the limited data availability for subordinated CDS spreads, our sample for internal rating changes is reduced from 242 to 95 observations. Table A.10 provides the results using the five-year subordinated CDS. 33 BANKSIZE can be used as a variable in our regression analysis as we define TBTF to be banks that are among the three largest banks in their country of origin in the year prior to the event. As a result, some large financial institutions are not considered TBTF because they are smaller than other banks in their country of origin, even though they may still be larger by assets than the three largest banks in another country. equal to 1 if the bank was among the active CDS traders based on quotes submitted by dealers in Bloomberg and accounts for the possibility that large banks which are involved in the CDS market may try to mitigate any adverse effects to their CDS spreads prior to the rating announcement. 34 The results show for all model specifications that the coefficient for TBTF is statistically significant, negative and large, indicating that TBTF considerations are the actual driver behind the observed CDS market reactions (see Table A.11 in the Appendix). This further supports the notion that TBTF is the main driver of the CDS market reaction to downgrade announcements and suggests that the reaction of CDS market participants to rating downgrade announcements is driven to a large extent by TBTF assumptions.

Changes in the regulatory environment
Following the recent financial crisis, bank regulation was tightened, especially for large, international banks with a global footprint. On 4 November 2011 the Financial Stability Board (FSB) released a list of financial institutions whose failure would lead to severe repercussion in the global financial system and disrupt the global economy, the so-called G-SIBs. The official acknowledgment of the existence of TBTF institutions may have alleviated concerns around the threat that a failure of these institutions potentially pose for the global economy and hence, market discipline may play a more prominent role again. 35 We divide our examination period into two sub-periods, one prior to the official G-SIB announcement on 4 November 2011, the pre-G-SIB period, and one following the official announcement, 34 We standardized each variable by deducting the mean value and dividing by the standard deviation of that variable. For binary variables, we only deducted the sample mean. 35 We also split the sample into three different time periods: the post-G-SIB period. We focus our investigation on the sample for internal reasons, which gives us a sample of 180 events in the pre-G-SIB period and 62 events in the post-G-SIB period. 36 It can be seen that CDS spreads widen significantly during both periods (see Table A.12 in the Appendix), even though the widening of CDS spreads is less pronounced during the post-G-SIB period.
For TBTF banks, no significant CDS market reaction can be observed to rating downgrade announcements during both time periods. In contrast, for non-TBTF banks, a significant widening of CDS spreads can be observed during both investigation periods. Figure A.5 graphically illustrates the CDS spread development before and after the official G-SIB designation. It is noteworthy that the CDS market reaction appears less pronounced during the period following the official G-SIB designation, but there is no significant difference in the CDS market reaction of non-TBTF banks between the pre-and post-G-SIB period. The same observation is also made for TBTF banks. This suggests that there was no significant change in the market participants' perception of the importance of TBTF banks vis-à-vis non-TBTF banks during both periods.

Conclusion
We analyze the effect of credit rating changes on the CDS spreads of banks to test whether rating changes are an appropriate channel through which market discipline can be enforced. To this end, we construct a comprehensive sample of 1,081 credit rating change announcements between 2004 and 2015 for an international sample of financial institutions. 36 Splitting the sample period results into the period prior to the leaked announcement of the G-SIB list on 30 November 2009 by the Financial Times (Financial Times, 2009) and the period following the leaked announcements leads to qualitatively similar results with 146 events for the period prior to the leaked list and 96 events following the leaked list. As a further robustness check, we control whether the Dodd-Frank act had a significant impact on our results. However, the results are similar to the ones reported.
Our analyses show that credit rating downgrades are overall associated with a large, significant widening of CDS spreads, while rating upgrades are associated with only a small tightening of CDS spreads. However, by further dividing our sample into the reasons for the rating change announcements, we find that the widening of CDS spreads following rating downgrade announcements is entirely driven by downgrades due to internal reasons, such as adverse changes in the operating performance or capital structure of the bank. Rating changes due to external reasons, such as changes in the economic environment, have almost no effect on banks' CDS spreads. In addition, even though a significant tightening of banks' CDS spreads can be observed surrounding rating upgrades, the effect is negligible compared to rating downgrades.
We further examine whether the rating downgrades due to internal reasons actually function as a proper channel for the enforcement of market discipline on financial institutions. Particularly for large banks, TBTF considerations may influence the reaction of CDS market participants to rating downgrade announcements, as government bailouts are anticipated if the financial position of these institutions deteriorates. In line with this reasoning, we find that CDS spreads of TBTF banks show no reaction to rating downgrade announcements. In contrast, non-TBTF banks experience a significant widening of CDS spreads surrounding rating downgrades. Even after controlling for other factors that potentially influence CDS spread changes, TBTF banks still experience significantly smaller spread changes than non-TBTF banks. Similar observations are also made for the stock and bond market. Therefore, TBTF banks have limited incentives to curb excessive risk taking as market participants do not appear to penalize these banks through a widening of CDS spreads, even if the bank's credit rating is lowered. Simultaneously, TBTF banks have significant refinancing and funding advantages over their non-TBTF peers.
These results provide evidence that for TBTF banks, rating downgrades are not an appropriate mechanism through which market discipline can be enforced on these institutions, given market participants' assumption of implicit government guarantees. This shows the importance of regulatory discipline as the main tool for controlling the risk taking of TBTF banks and raises the question whether markets can be expected to effectively discipline large, globally interconnected financial institutions.
Brewer, E., Jagtiani, J., 2013. How much did banks pay to become Too-Big-to-Fail and to become    +5] day event windows are shown to account for potential information leakage prior to the announcement and to account for adjustments to CDS spreads following the official announcement, respectively. The mean and median cumulative adjusted CDS spread changes (CASC) are shown in basis points (bps) and tested for significance using the parametric t-test and the nonparametric Wilcoxon signedrank test (Wilcoxon). The equality of means and medians of rating downgrade announcements due to internal and external reasons are tested for statistical significance using the two sample t-test and the Mann-Whitney U test (Panel D). * , ** , *** denote statistical significance at the 10%, 5%, and 1% level, respectively.  +5] day event windows are shown to account for potential information leakage prior to the announcement and to account for adjustments to CDS spreads following the official announcement, respectively. The mean and median cumulative adjusted CDS spread changes (CASC) are shown in basis points (bps) and tested for significance using the parametric t-test and the nonparametric Wilcoxon signedrank test (Wilcoxon). The equality of means and medians of rating upgrade announcements due to internal and external reasons are tested for statistical significance using the two sample t-test and the Mann-Whitney U test (Panel D). * , ** , *** denote statistical significance at the 10%, 5%, and 1% level, respectively.

CDS market reaction to rating downgrade announcements due to internal reasons for TBTF and non-TBTF banks
This table shows the results of the short-term CDS market reaction for the sample of 242 rating downgrade announcements attributed to internal reasons, divided into downgrade announcements for TBTF banks (Panel A) and non-TBTF banks (Panel B). A bank is defined as TBTF if it was among the three largest banks in its country of origin as measured by the total assets at the end of the year prior to the rating downgrade announcement. The short-term event windows [-10;+10], [-5;+5], and [-1;+1], as well as the announcement day [0;0] are shown to capture the market reaction surrounding the rating downgrade announcement. In addition, the [-30;-11], [-5;-1] and [+1;+5] day event windows are shown to account for potential information leakage prior to the announcement and to account for adjustments to CDS spreads following the official announcement, respectively. The mean and median cumulative adjusted CDS spread changes (CASC) are shown in basis points (bps) and tested for significance using the parametric t-test and the nonparametric Wilcoxon signed-rank test (Wilcoxon). The equality of means and medians of rating upgrade announcements due to internal and external reasons are tested for statistical significance using the two sample t-test and the Mann-Whitney U test (Panel C). * , ** , *** denote statistical significance at the 10%, 5%, and 1% level, respectively.  , and TBTF5 are binary variables equal to 1 if a bank was among the three largest, the largest, or among the five largest banks in its country of origin as measured by the total assets at the end of the year prior to the rating change announcement, respectively, and 0 otherwise. INTERNAL is a binary variable that takes the value of 1 if the rating downgrade is due to internal reasons and 0 if the rating downgrade is due to external reasons. RATING is defined as the bank's rating prior to the rating downgrade on a 23 step numerical scale (AAA/Aaa=23, AA+/Aa1=22,…, C and lower=1), NOTCHES is defined as the absolute difference between the old and new rating based on the 23 step numerical scale, REVIEW is equal to 1 if the rating change was preceded by a rating review and 0 otherwise and INVESTMENT TO NON-INVESTMENT DOWNGRADE is equal to 1 if the rating downgrade resulted in a non-investment grade status of the bank (e.g. BBB-/Baa3 and above to BB+/Ba1 and below), and 0 otherwise. S&P and FITCH are both equal to 1 if the downgrade announcement was made by S&P or Fitch, respectively, and 0 otherwise, and CRISIS takes the value 1 if the event occurred between January 2008 and December 2009 and 0 otherwise. OPAQUENESS takes the value of 0 if the bank has a rating by S&P, Moody's and Fitch, 1 if the banks has a rating by only two of the three CRAs, and 2 if the bank is rated by only one CRA, ROA is defined as the bank's ROA in the year prior to the rating announcement, LEVERAGE is the ratio of total liabilities to total equity of the bank at the end of the year prior to event, LIQUIDITY is the ratio of non-zero daily CDS spread changes to the total number of trading days during the 120-day estimation period, and LIABILITIES/GDP is a binary variable equal to 1 if the ratio of the bank's total liabilities divided by the GDP of the bank's country of origin in the year prior to the event exceeds 0.5 and 0 otherwise, GOVERNMENT is a binary variable that takes the value of 1 if the bank is a policy bank or the government holds significant stake in the bank and 0 otherwise, IB is equal to 1 if the bank's first two digits of the SIC code start with 62 (e.g., investment banks and asset managers) and 0 otherwise, and EU is equal to 1 if the firm's headquarter is located in Europe and 0 otherwise. The standard errors are corrected for heteroskedasticity and given in parentheses. * , ** , *** denote statistical significance at the 10%, 5%, and 1% level, respectively.

Figure 1: Total number of bank rating change announcements and average bank rating
This figure shows the total number of bank rating change announcements during the investigation period from 1st January 2004 to 31st December 2015 on a quarterly basis. The data is based on 2,364 rating announcements that were collected for the 212 banks in the sample during the investigation period that possessed a long-term issuer rating by S&P, Moody's and/or Fitch. The figure also shows the average rating of the 212 sample banks during the entire investigation period.

Figure 2: CDS spread development during the [-10;+10] day event window
This figure shows the mean cumulative adjusted CDS spread change (CASC) development in basis points (bps) for the entire sample of 782 rating downgrade announcements (Panel A) and 299 rating upgrade announcements (Panel B) during the [-10;+10] day event window surrounding the rating change announcement t=0. The rating downgrade and rating upgrade announcements are divided into announcements due to internal and external reasons.

Figure 3: CDS spread development during the [-30;+10] day event window for TBTF and non-TBTF banks
This figure shows the mean cumulative adjusted CDS spread change (CASC) development in basis points (bps) for the sample of 242 rating downgrades attributed to internal reasons and 540 rating downgrades attributed to external reasons during the [-30;+10] day event window surrounding the rating change announcement t=0, divided into the CASC development for TBTF and non-TBTF banks. A bank is defined as TBTF if it was among the three largest banks in its country of origin as measured by the total assets at the end of the year prior to the rating downgrade announcement.

Figure 4: Development of the equity returns and bond returns around rating downgrade announcements
This figure shows the average cumulative abnormal return (ACAR) and the average cumulative abnormal bond return (ACABR) development in percent for the sample of 156 events for the stock event study and 164 events for the bond event study for rating downgrades attributed to internal reasons (Panel A) and 450 events for the stock event study and 352 events for the bond event study for rating downgrades attributed to external reasons (Panel B) during the [-10;+10] day event window surrounding the rating change announcement t=0. Both samples are divided into the ACAR and ACABR development for TBTF and non-TBTF banks. A bank is defined as TBTF if it was among the three largest banks in its country of origin as measured by the total assets at the end of the year prior to the rating downgrade announcement.

Market discipline through credit ratings and Too-Big-to-Fail in banking
Electronic copy available at: https://ssrn.com/abstract=2928113  The internal reasons are further subdivided into capital structure, operating performance and M&A, while the external reasons are split into environment, sovereign, and methodology. The reason for a rating change by a CRA is identified using a keyword search in the corresponding press release. 76 keywords are used which are frequently mentioned as a reason and sorted in order of appearance in the press release. If more than one keyword appeared in a press release, the event is attributed to the first keyword, as it is assumed that the most important reason is mentioned first.
In case a press release did not explicitly include one of the keywords, the reason is manually matched to the closest category.

Panel A: Internal reasons
Capital structure Balance sheet structure, capital structure, credit profile, debt reduction, financial policy, financial position, financial profile, financial risk, financial risk profile, financial structure, funding profile, leverage, parent guarantee, risk profile Operating performance Business portfolio, business risk, business risk profile, capital efficiency, cash flow generation, competition, competitiveness, demand, diversification, earnings, efficiency, growth potential, integration, liquidity, market position, operating environment, operating performance, products, profit margin, profitability, restructuring, retained earnings, revenues, sales, trading loss

M&A reasons
Acquire, acquiring, acquisition, bid, deal, M&A, merged, merger Panel A: External reasons Environment Condition, Dodd-Frank, economic downturn, environment, fundamental trends, law, market condition, market development, reform, regulation, regulatory, systemic risk Sovereign Authorities, propensity country, financial support, follows the downgrade, follows the revision of, follows the upgrade, government, likelihood of a downgrade, provide support, receive support from, sovereign Methodology Current criteria, JDA, methodology, rating criteria, revised bank criteria

Table A.3: CDS market reaction to rating downgrade announcements by reason
This table shows the results of the short-term CDS market reaction for the 782 rating downgrade announcements by the reason for the downgrade announcements. The reasons are divided into internal reasons and external reasons. The internal reasons include rating changes as a result of changes in capital structure (Panel A), operating performance (Panel B), or M&A activity (Panel C). External reasons include rating changes due to changes in the macroeconomic or regulatory environment (Panel D), adjustments to the sovereign rating (Panel E), and changes in the methodology the CRA uses to assess the ratings of banks (Panel F). The short-term event windows [−10;+10], [−5;+5], and [−1;+1], as well as the announcement day [0;0] are shown to capture the market reaction surrounding the rating downgrade announcement. In addition, the [−5;−1] and [+1;+5] day event windows are shown to account for potential information leakage prior to the announcement and to account for adjustments to CDS spreads following the official announcement, respectively. The mean and median cumulative adjusted CDS spread changes (CASC) are shown in basis points (bps) and tested for significance using the parametric t-test and the nonparametric Wilcoxon signed-rank test (Wilcoxon). * , ** , *** denote statistical significance at the 10%, 5%, and 1% level, respectively.

: CDS market reaction to rating upgrade announcements by reason
This table shows the results of the short-term CDS market reaction for the 299 rating upgrade announcements by the reason for the upgrade announcements. The reasons are divided into internal reasons and external reasons. The internal reasons include rating changes as a result of changes in capital structure (Panel A), operating performance (Panel B), or M&A activity (Panel C). External reasons include rating changes due to changes in the macroeconomic or regulatory environment (Panel D), adjustments to the sovereign rating (Panel E), and changes in the methodology the CRA uses to assess the ratings of banks (Panel F). The short-term event windows [−10;+10], [−5;+5], and [−1;+1], as well as the announcement day [0;0] are shown to capture the market reaction surrounding the rating upgrade announcement. In addition, the [−5;−1] and [+1;+5] day event windows are shown to account for potential information leakage prior to the announcement and to account for adjustments to CDS spreads following the official announcement, respectively. The mean and median cumulative adjusted CDS spread changes (CASC) are shown in basis points (bps) and tested for significance using the parametric t-test and the nonparametric Wilcoxon signed-rank test (Wilcoxon). * , ** , *** denote statistical significance at the 10%, 5%, and 1% level, respectively.

: CDS market reaction to rating downgrade announcements due to external reasons for TBTF and non-TBTF banks
This table shows the results of the short-term CDS market reaction for the sample of 540 rating downgrade announcements attributed to external reasons, divided into external announcements for TBTF banks (Panel A) and non-TBTF banks (Panel B). A bank is defined as TBTF if it was among the three largest banks in its country of origin as measured by the total assets at the end of the year prior to the rating downgrade announcement. The short-term event windows [-10;+10], [-5;+5], and [-1;+1], as well as the announcement day [0;0] are shown to capture the market reaction surrounding the rating downgrade announcement. In addition, the [-30;+10], [-5;-1] and [+1;+5] day event windows are shown to account for potential information leakage prior to the announcement and to account for adjustments to CDS spreads following the official announcement, respectively. The mean and median cumulative adjusted CDS spread changes (CASC) are shown in basis points (bps) and tested for significance using the parametric t-test and the nonparametric Wilcoxon signed-rank test (Wilcoxon).The equality of means and medians of rating upgrade announcements due to internal and external reasons are tested for statistical significance using the two sample t-test and the Mann-Whitney U test (Panel C). * , ** , *** denote statistical significance at the 10%, 5%, and 1% level, respectively.   Table 6 for the sample of the 782 rating downgrades.
TBTF, TBTF1, and TBTF5 are binary variables defined as 1 if a bank was among the three largest, the largest, or among the five largest banks in its country of origin as measured by the total assets at the end of the year prior to the rating change announcement, respectively, and 0 otherwise.
INTERNAL is a binary variable that takes the value of 1 if the rating downgrade is due to internal reasons and 0 if the rating downgrade is due to external reasons. RATING is defined as the bank's rating prior to the rating downgrade on a 23 step num erical scale (AAA/Aaa=23, AA+/Aa1=22,…, C and lower=1), NOTCHES is defined as the absolute difference between the old and new rating based on the 23 step numerical scale, REVIEW is defined as 1 if the rating change was preceded by a rating review and 0 otherwise and INVESTMENT TO NON-INVESTMENT DOWNGRADE is equal to 1 if the rating downgrade resulted in a non-investment grade status of the bank (e.g. BBB-/Baa3 and above to BB+/Ba1 and below), and 0 otherwise. S&P and FITCH are both equal to 1 if the downgrade announcement was made by S&P or Fitch, respectively, and 0 otherwise, and CRISIS takes the value 1 if the event occurred between January 2008 and December 2009 and 0 otherwise. OPAQUENESS takes the value of 0 if the bank has a rating by S&P, Moody's and Fitch, 1 if the banks has a rating by only two of the three CRAs, and 2 if the bank is rated by only one CRA, ROA is defined as the bank's ROA in the year prior to the rating announcement, LEVERAGE is the ratio of total liabilities to total equity of the bank at the end of the year prior to event, LIQUIDITY is the ratio of non-zero daily CDS spread changes to the total number of trading days during the 120-day estimation period, and LIABILITIES/GDP is a binary variable equal to 1 if the ratio of the bank's total liabilities divided by the GDP of the bank's country of origin in the year prior to the event exceeds 0.5 and 0 otherwise, GOVERNMENT is a binary variable that takes the value of 1 if the bank is a policy bank or the government holds significant stake in the bank and 0 otherwise, IB is equal to 1 if the bank's first two digits of the SIC code start with 62 (e.g., investment banks and asset managers) and 0 otherwise, and EU is equal to 1 if the firm's headquarter is located in Europe and 0 otherwise.  TBTF, TBTF1, and TBTF5 are binary variables equal to 1 if a bank was among the three largest, the largest, or among the five largest banks in its country of origin as measured by the total assets at the end of the year prior to the rating change announcement, respectively, and 0 otherwise.
RATING is defined as the bank's rating prior to the rating downgrade on a 23 step numerical scale (AAA/Aaa=23, AA+/Aa1=22,…, C and lower=1), NOTCHES is defined as the absolute difference between the old and new rating based on the 23 step numerical scale, REVIEW is equal to 1 if the rating change was preceded by a rating review and 0 otherwise, and INVESTMENT TO NON-INVESTMENT DOWNGRADE is equal to 1 if the rating downgrade resulted in a non-investment grade status of the bank (e.g. BBB-/Baa3 and above to BB+/Ba1 and below), and 0 otherwise. M&A and CAPITAL STRUCTURE are both equal to 1 if the reason for the rating downgrade announcement relate to M&A activity or changes to the bank's capital structure, respectively, and 0 otherwise, S&P and FITCH are both equal to 1 if the downgrade announcement was made by S&P or Fitch, respectively, and 0 otherwise, and CRISIS takes the value 1 if the event occurred between January 2008 and December 2009 and 0 otherwise.
OPAQUENESS takes the value of 0 if the bank has a rating by S&P, Moody's and Fitch, 1 if the banks has a rating by only two of the three CRAs, and 2 if the bank is rated by only one CRA, ROA is defined as the bank's ROA in the year prior to the rating announcement, LEVERAGE is the ratio of total liabilities to total equity of the bank at the end of the year prior to event, LIQUIDITY is the ratio of non-zero daily CDS spread changes to the total number of trading days during the 120-day estimation period, and LIABILITIES/GDP is a binary variable equal to 1 if the ratio of the bank's total liabilities divided by the GDP of the bank's country of origin in the year prior to the event exceeds 0.5 and 0 otherwise, GOVERNMENT is a binary variable that takes the value of 1 if the bank is a policy bank or the government holds significant stake in the bank and 0 otherwise, IB is equal to 1 if the bank's first two digits of the SIC code start with 62 (e.g., investment banks and asset managers) and 0 otherwise, and EU is equal to 1 if the firm's headquarter is located in Europe and 0 otherwise. The standard errors are corrected for heteroskedasticity and given in parentheses. * , ** , *** denote statistical significance at the 10%, 5%, and 1% level, respectively.  +5] day event windows are shown to account for potential information leakage prior to the announcement and to account for adjustments to CDS spreads following the official announcement, respectively. The cumulative adjusted CDS spread changes are calculated by adding daily benchmark adjusted spread changes using ASCi,t = (CDSi,t -CDSi,t-1) -(It -It-1), where ASCi,t is the adjusted CDS spread change of bank i on day t, CDSi,t is the observed CDS spread of bank i on day t and It is the relevant CDS spread index for the rating class on day t. The daily CDS spread index corresponds to the equally weighted cross-sectional mean of all CDS spreads for each of the six letter rating classes AAA/AA, A, BBB, BB, B, and CCC and below, composed of the CDS spreads of all financial institutions with available data through Thomson Reuters EOD and a long-term issuer rating of either S&P, Moody's, or Fitch, excluding the event firm. The mean and median CASC are shown in basis points (bps) and tested for significance using the parametric t-test and the nonparametric Wilcoxon signed-rank test (Wilcoxon). The equality of means and medians of rating downgrade announcements for TBTF and non-TBTF banks is tested for statistical significance using the two sample t-test and the Mann-Whitney U test (Panel C). * , ** , *** denote statistical significance at the 10%, 5%, and 1% level, respectively.    ). TBTF is a binary variable equal to 1 if a bank was among the three largest banks in its country of origin as measured by the total assets at the end of the year prior to the rating change announcement and 0 otherwise. BANKSIZE is defined as the bank's total assets in billion U.S. dollars at the end of the year prior to the event, while RATING is defined as the bank's rating prior to the rating downgrade on a 23 step numerical scale (AAA/Aaa=23, AA+/Aa1=22,…, C and lower=1), LIQUIDITY is defined as the ratio of non-zero daily CDS spread changes to the total number of trading days during the 120-day estimation period, SOVEREIGN RISK is defined as GDP divided by total national debt of the bank's country of origin, and CDS TRADER is a binary variable equal to 1 if the bank was among the active CDS trader based on quotes submitted by dealers in Bloomberg subtracted by the mean number. All variables were standardized by deducting the mean value and dividing by the standard deviation of that variable. For binary variables, only the sample mean was deducted. The standard errors are corrected for heteroskedasticity and given in parentheses. * , ** , *** denote statistical significance at the 10%, 5%, and 1% level, respectively. 6.59 *** 5 . 4 4 *** 5 . 3 9 *** 5 . 0 2 *** 6 . 5 7 *** 2.54 ***

Table A.12: CDS market reaction to rating downgrade announcements for TBTF and non-TBTF banks for the time period prior to and following the official G-SIB designation
This table shows the results of the short-term CDS market reaction for the sample of 242 rating downgrade announcements attributed to internal reasons (Panel A), divided into downgrade announcements for TBTF banks (Panel B) and non-TBTF banks (Panel C). The sample is split into the time period prior to the official G-SIB announcement on 4 November 2011, the pre-G-SIB period, and the time period following the official announcement, the post-G-SIB period. A bank is defined as TBTF if it was among the three largest banks in its country of origin as measured by the total assets at the end of the year prior to the rating downgrade announcement. The short-term event windows [-10;+10], [-5;+5], and [-1;+1], as well as the announcement day [0;0] are shown to capture the market reaction surrounding the rating downgrade announcement. In addition, the [-30;-11], [-5;-1] and [+1;+5] day event windows are shown to account for potential information leakage prior to the announcement and to account for adjustments to CDS spreads following the official announcement, respectively. The mean and median cumulative adjusted spread changes CASC are shown in basis points (bps) and tested for significance using the parametric t-test and the nonparametric Wilcoxon signed-rank test (Wilcoxon). The equality of means and medians of rating downgrade announcements during the pre-G-SIB period and post-G-SIB period is tested for statistical significance using the two sample t-test and the Mann-Whitney U test (MWU test). * , ** , *** denote statistical significance at the 10%, 5%, and 1% level, respectively.

for TBTF and non-TBTF banks by region or country
This figure shows the mean cumulative adjusted CDS spread change (CASC) development in basis points (bps) for the sample of 242 rating downgrades attributed to internal reasons during the [-10;+10] day event window surrounding the rating change announcement t=0, split by region (Asia and Europa) and country (United States, United Kingdom, France, Germany, Italy, the Netherlands) and divided into the CASC development for TBTF and non-TBTF banks. A bank is defined as TBTF if it was among the three largest banks in its country of origin as measured by the total assets at the end of the year prior to the rating downgrade announcement. Electronic copy available at: https://ssrn.com/abstract=2928113 ;+10] day event window surrounding the rating change announcement t=0. The samples are further split into TBTF and non-TBTF banks. A bank is defined as TBTF if it was among the three largest banks in its country of origin as measured by the total assets at the end of the year prior to the rating downgrade announcement.

Asia
Panel A: Pre-G-SIB period Panel B: Post-G-SIB period (Bessembinder, Kahle, Maxwell, and Xu, 2009 where N is the number of events. To test whether ACABRs differ significantly from 0, we use the parametric t-test as well as the nonparametric Wilcoxon signed-rank test. The difference between two samples are tested using the two sample t-test and the Mann-Whitney U test. There may also be other benefits from being perceived as TBTF. The chairman of HSBC acknowledged prior to the official G-SIB designation that the banks that are officially considered to be of systemic importance would likely have a more concentrated flow of business to them (Financial Times, 2011b). The Bank of China even viewed its designation as a G-SIB as offering the bank an advantage over its domestic peers (Financial Times, 2011a). Moreover, besides obtaining more business, benefits of being considered TBTF may be particularly pronounced in debt and credit markets. Here, banks potentially benefit from their TBTF status and obtain more favorable refinancing rates. To this end, the then CEO of Deutsche Bank, Josef Ackermann, stated that he was glad that Deutsche Bank would likely be considered a G-SIB, as this would lead to benefits with regard to its refinancing and depository business (Handelsblatt, 2011). It may therefore well be that the perception of being a large, systemically important bank will lead to a comparative advantage for these institutions, particularly in debt markets with respect to refinancing and funding operations.
In case TBTF banks possess a funding advantage over their non-TBTF banks, this may indicate that debt markets do not impose an adequate level of market discipline on large banks. The early studies of Avery, Belton and Goldberg (1988) and Hannan and Hanweck (1988) point in this direction, as the rates paid on insured certificates of deposit had little relation to risk measures of banks and the yields of certificates of deposit issued by larger banks were generally lower than the yields of those issued by smaller banks during the 1980s. During the early 1990s, however, investors started to exert some market discipline as the yields on subordinated debentures of financial institutions increased with higher levels of risk (Flannery and Sorescu, 1996). Similarly, Morgan and Stiroh (2005) document that the bond ratings of the 11 U.S. financial institutions that were considered TBTF following the testimony of the U.S. Comptroller of the Currency in 1984 were able to issue new debt at significantly lower rates than other banks. There are also indications that the costs financial institutions associate with market discipline are lower than the costs they associate with regulatory discipline . 2 Rime (2005) finds that TBTF considerations significantly lift a bank's issuer rating, in some cases resulting in a bonus of several rating notches. This rating bonus significantly reduces a bank's refinancing and funding costs, effectively giving these banks a competitive advantage over their smaller peers. In a similar vein, Ueda and Weder di Mauro (2013) find that the funding advantage of TBTF banks was around 60 bps over their smaller peers at the end of 2007. By the end of 2009 and in the wake of the financial crisis, this advantage increased to 80 bps. This provides evidence that TBTF considerations undermine the effect of market discipline (Rime, 2005). Acharya, Anginer and Warburton (2016) document that bond credit spreads are not sensitive to risk for large financial institutions, while the spreads are sensitive to risk for smaller institutions and non-financial sector companies. This effectively shows that bondholders of large financial institutions anticipate government bailouts in case of a bank failure, while simultaneously allowing large banks to refinance and fund themselves at subsidized rates. As such, empirical research shows that TBTF considerations exist, not only in equity, but also in debt markets. Yet, there are also potential downsides for a bank to be considered TBTF, such as higher regulatory scrutiny and increased regulatory requirements

Swiss Finance Institute
Swiss Finance Institute (SFI) is the national center for fundamental research, doctoral training, knowledge exchange, and continuing education in the fields of banking and finance. SFI's mission is to grow knowledge capital for the Swiss financial marketplace. Created in 2006 as a public-private partnership, SFI is a common initiative of the Swiss finance industry, leading Swiss universities, and the Swiss Confederation.