The Volcker Rule and Market-Making in Times of Stress

Finance and Economics Discussion Series Divisions of Research & Statistics and Monetary Affairs Federal Reserve Board, Washington, D.C. The Volcker Rule and Market-Making in Times of Stress Jack Bao, Maureen O’Hara, and Alex Zhou 2016-102 Please cite this paper as: Bao,Jack,MaureenO’Hara,andAlexZhou(2016). “TheVolckerRuleandMarket-Making inTimesofStress,”FinanceandEconomicsDiscussionSeries2016-102. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/FEDS.2016.102. NOTE: Staff working papers in the Finance and Economics Discussion Series (FEDS) are preliminary materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the research staff or the Board of Governors. References in publications to the Finance and Economics Discussion Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers.

The Volcker Rule and Market-Making in Times of Stress1 Jack Bao Federal Reserve Board Maureen O’Hara Johnson Graduate School of Management Cornell University Xing (Alex) Zhou Federal Reserve Board September 2016 Abstract Focusing on downgrades as stress events that drive the selling of corporate bonds, we document that the illiquidity of stressed bonds has increased after the Volcker Rule. Dealers regulated by the Rule have decreased their market-making activities while non-Volcker-affected dealers have stepped in to provide some additional liquidity. Furthermore, even Volcker-affected dealers that are not constrained by Basel III and CCAR regulations change their behavior, inconsistent with the effects being driven by these other regulations. Since Volcker-affected dealers have been the main liquidity providers, the net effect is that bonds are less liquid during times of stress due to the Volcker Rule. JEL classification: G14, G21, G23, G24, G28 Keywords: Volcker Rule, Corporate Bond Illiquidity, Regulation, Capital Commitment, Dealer Inventory, Market-Making, Financial Crisis 1 We thank Manuel Adelino, Sergey Chernenko, Dan Covitz, Alie Diagne, Darrell Duffie, Josh Gallin, Adam Kolasinski, Jongsub Lee, Nellie Liang, Kleopatra Nikolaou, and seminar participants at the Federal Reserve Board for helpful comments and discussions. The views expressed herein are those of the authors and not necessarily of the Federal Reserve Board of Governors or its staff. All errors are our own. 1

1. Introduction Among the many regulatory changes following the financial crisis, few are more controversial than the Volcker Rule. Enacted as part of the Dodd-Frank Act, the Volcker Rule was intended to limit bank risk-taking by restricting or prohibiting certain speculative activities. Critics (for example, Duffie [2012]) contended that an unintended consequence of the Rule could be diminished bond market liquidity, resulting from a reduction in banks’ market making activities. Advocates of the Rule disagreed, arguing that non-Volcker affected dealers could compensate for any market making reductions, leaving liquidity essentially unchanged. Recent empirical studies of post-crisis market behavior (e.g., Trebbi and Xiao (2015), Bessembinder, Jacobsen, Maxwell, and Venkataraman (2016), and Dick-Nielsen and Rossi (2016)), however, find conflicting evidence of the effect of regulations on bond market liquidity. In this paper, we focus specifically on the implementation of the Volcker Rule and its impact on bond market liquidity, particularly in times of market stress. We argue that fully understanding the impact of the Volcker Rule on market liquidity requires understanding how liquidity behaves in the face of severe conditions, or exactly when liquidity is needed most. As shown by recent research, liquidity deterioration was particularly pronounced during the height of the Financial Crisis.2 Practitioners and policymakers alike have noted that illiquidity in times of market stress may be the more relevant metric for gauging market stability and performance.3 The main motivation and first major contribution of our paper is to study whether illiquidity is relatively worse in periods of stress after the Volcker Rule was 2 See Bao, Pan, and Wang (2011), Dick-Nielsen, Feldhutter, and Lando (2012), and Friewald, Jankowitsch, and Subrahmanyam (2012). 3 See recent comments by Deutsche Banc Research (2016) and testimony by Powell (2016) that even if liquidity is high in normal conditions, it may become more troublesome in periods of stress. 2

implemented. Motivated by Ellul, Jotikasthira, and Lundblad (2011) who find evidence of forced selling of downgraded bonds induced by regulatory constraints imposed on insurance companies, we use downgrades of corporate bonds to junk as stress events where liquidity is demanded by clients. Focusing on regulation-induced sales has the added advantage of plausibly preventing investors from optimally timing their trades, thereby providing a more reliable estimate of the liquidity conditions that investors face. Our focus is on a difference-in-differences test comparing the illiquidity of downgraded corporate bonds to a baseline control group both before-and-after the Volcker Rule was implemented. In particular, the first difference is the difference in price impact between a set of bonds recently downgraded to speculative-grade from investment-grade and a set of BB bonds used to control for the general level of illiquidity.4 The second difference is between the post- Volcker difference and the pre-Volcker difference. Our results show that bond liquidity deterioration around rating downgrades has worsened following the implementation of the Volcker Rule. We find such adverse effects whether we benchmark to the pre-crisis period or to the period just before the Volcker Rule was enacted, and we find that the relative deterioration in liquidity around these stress events is as high during the post-Volcker period as during the Financial Crisis. Given how badly liquidity deteriorated during the financial crisis, this finding suggests that the Volcker Rule may have serious consequences for corporate bond market functioning in stress times. The second motivation and contribution of our study is to understand how the Volcker Rule induced changes in dealer behavior, and particularly to identify any differential effects on Volckeraffected vs. non-Volcker-affected dealers. Because the Volcker Rule applied only to banks with 4 Results are similar if we instead use bid-ask spreads. 3

access to government backstops (such as deposit insurance or Federal Reserve borrowing), other dealers without such access can continue to trade and could, in principle, step in to provide additional liquidity in cases where the lines between permissible market-making and prohibited proprietary trading are blurred. Using a unique data set with dealer identities, we present evidence that non-Volckeraffected dealers have been providing more liquidity during post-Volcker stress times. In the post- Volcker period, the relative share of dealer-customer trades taken by non-Volcker dealers has increased. Dealers affected by the Volcker Rule see a statistically significant increase in agency trades, or trades in which the dealer has pre-arranged an offsetting trade so as not to have inventory risk. For non-Volcker dealers, we see no such effects on agency trades in the post-Volcker period. We also find that Volcker-affected dealers significantly reduce their capital commitment, while non-Volcker dealers commit more capital in market-making. Combined with our results on the increased illiquidity during the post-Volcker period, these results suggest that while non-Volcker dealers have stepped in (as proponents of the Volcker Rule suggested would happen), opponents of the Volcker Rule were correct in arguing that the change would not be immediate.5 At least during stress times, this new participation is not yet enough to offset the decreased liquidity in bond market trading. Finally, a third goal of our research is to disentangle the effects of the Volcker Rule from those of other important regulations on dealer bond market behavior. We do so by focusing particularly on the implementation period as compared to the period just before implementation and also by splitting dealers by their exposure to Basel III. Though most banks’ capital ratios are significantly above Basel III minimums, increased Basel III capital requirements along with 5 See the Federal Register (2014) publication on the Volcker Rule for details of comment letters. Liquidity deterioration was particularly severe during the height of the Financial Crisis. 4

Comprehensive Capital Analysis and Review (CCAR) requirements may potentially mean that some banks will reduce their market-making activities because of CCAR constraints. These constraints arise from the fact that dealers are required to meet minimum capital requirements in stress scenarios. Thus, to ensure that our results are not driven by banks constrained by the start of Basel III implementation (along with existing CCAR requirements), we split Volcker-affected dealers into those that are CCAR-constrained and those that are not. We find that capital commitment has decreased significantly for dealers that have neither failed CCAR tests nor been given a conditional pass. Thus, our results are unlikely to be driven by banks adjusting their business to remedy failed CCAR tests. Our paper is most closely related to three recent studies on regulation and liquidity, all of which focus on the general regulatory environment following the global financial crisis. Studying general trends in corporate bond market liquidity, Trebbi and Xiao (2015) argue that liquidity has not deteriorated following post-crisis regulations. Bessembinder, et al. (2016) provide a similar finding, but also add an examination of dealer behavior. They find that while there is little evidence of increases in transactions costs, there is evidence that dealers behave differently as new regulations have been implemented. Dick-Nielsen and Rossi (2016) study liquidity provision around index exclusion events, finding that liquidity has deteriorated post-crisis. All three papers provide evidence of how liquidity and market-making has changed in post-crisis years following the passage of reform rules, but in contrast to these studies, our focus is on isolating the specific effects arising from implementation of the Volcker Rule. Our main results relate to comparing the post-Volcker implementation period to the period just before Volcker implementation, whereas both Bessembinder et al. (2016) and Dick-Nielsen and Rossi (2016) focus on the years prior to 5

Volcker Rule implementation.6 Trebbi and Xiao’s (2015) sample extends to the end of 2014, but they also do not focus on the Volcker Rule implementation period. Furthermore, our use of the regulatory version of TRACE, with dealer identities, allows us not only to split dealers by those that are directly affected by the Volcker Rule and those that are not, but also to identify which dealers were potentially affected by other regulations such as Basel III and CCAR. The evidence in our study suggests that there are significant costs to the proprietary trading ban in the Volcker Rule, but it is important to note that we do not do any welfare analysis to assess whether the Volcker Rule is overall net positive or net negative for financial markets and the economy.7 One obvious potential benefit of the Volcker Rule is the ban of risky trades by institutions that could eventually seek government support if their risky trades led to significant losses. Such analysis requires modeling the trade-off between the social cost to the loss of liquidity in corporate bond markets and the societal benefit of safer banks and is beyond the scope of our study. The rest of the paper is organized as follows. In Section 2, we discuss the Volcker Rule and its potential impact on market-making in the corporate bond market. In Section 3, we describe our data sources and variable construction. In Section 4, we examine changes in liquidity around times of stress. In Section 5, we examine how the behavior of Volcker-affected and non-Volckeraffected dealers changes with the implementation of the Volcker Rule. We also discuss Basel III and CCAR regulations. Section 6 concludes. 6 In fact, both papers discuss their results as being related to an anticipation of new regulations. Our results, in contrast, look at the implementation of the Volcker Rule. 7 There are, of course, costs to not having regulation. For example, Chernobai, Ozdagli, and Wang (2016) show that operational risk events increased during the gradual deregulation of bank holding companies from 1996 to 1999. We are, however, unaware of any studies quantitatively measuring the costs of allowing banks to participate in proprietary trading. 6

2. Potential Impact of the Volcker Rule on the Corporate Bond Market As part of the Dodd-Frank Act, passed July 21, 2010, section 13 (the “Volcker Rule”) was added to the Bank Holding Company Act of 1956. Section 13 generally prohibits banking entities from engaging in proprietary trading or having ownership or relationships with hedge funds and private equity funds. Implementation of section 13, however, was not immediate and followed a laborious process. On January 18, 2011, the Financial Stability Oversight Council (2011) released a study of its recommendations for implementing section 13. The Treasury, Board of Governors, FDIC, and SEC worked with the CFTC in formulating a proposal before releasing a version for comments in the Federal Register (2011) in November 2011. In December 2013, final regulations were issued, and final regulations with details of market participants’ comments were released in the Federal Register (2014) on January 31, 2014. On April 1, 2014, the Volcker Rule became effective with banks of at least $50 billion in trading assets required to report some quantitative metrics starting July 2014. By July 21, 2015, large banks were required to be fully compliant with the Volcker Rule. During the conformance period, banks were required to make good faith efforts to conform to the new rules.8 Hence, we expect to already see some effects of the Volcker Rule starting in April 2014. Other research (e.g., Bessembinder et al. (2016) and Dick-Nielsen and Rossi (2016) has argued that anticipation of new regulation implementation could lead to earlier changes in dealer behavior. Though we expect the impact to be the greatest once the implementation period requires dealers to begin reporting metrics on market-making activity, our tests do not preclude the possibility of some changes in dealer behavior prior to rule implementation. In particular, our tests 8 See Federal Reserve Board (2016). 7

are designed to gauge the additional impact of Volcker Rule implementation, mainly benchmarking to the period just before implementation. The intent of the Volcker Rule is to prohibit banking entities with access to the discount window at the Federal Reserve or to FDIC insurance from engaging in risky proprietary trading. It is important to keep in mind that not all financial firms are covered. For example, an Oliver Wyman and SIFMA (2011) study lists Cantor Fitzgerald & Co., Daiwa Capital Markets, Jefferies & Co., and Nomura as explicitly not covered. It is also the case that not all trading activities are precluded. Recognizing that some activities are necessary for the market to function normally, the Volcker Rule includes an explicit set of permitted activities. The most relevant one for this paper, and arguably the most controversial, is a provision that permits market-making. Essentially, affected dealers can trade securities in a way to facilitate client-driven transactions, but cannot transact in a way intended to make profits based on the price appreciation of securities. A major difficulty in implementing the market-making exception is distinguishing allowed market-making from prohibited proprietary trading. The Financial Stability Oversight Council (2011) proposed a number of principles to distinguish between the two. Among these are that market-making should have rapid inventory turnover with the vast majority of profits from bidask spreads rather than profits from inventory appreciation. Proprietary trading is likely to have more modest turnover with significant profits from inventory appreciation. The FSOC also proposed a number of metrics including measures of inventory aging, customer-initiated trade ratios, and revenue from customer-initiated flows. The final law requires establishment of an internal compliance program and the reporting of seven sets of metrics: (1) Risk and Position Limits and Usage, (2) Risk Factor Sensitivities, (3) Value-at-Risk and Stress VaR, (4) 8

Comprehensive Profit and Loss, (5) Inventory Turnover, (6) Inventory Aging, and (7) Customer Facing Trade Ratio. Critics of the Volcker Rule noted many gray areas in the rule and further argued that ambiguity in how the rule would be enforced was likely to be detrimental to market liquidity. Furthermore, though the intent of market-making and proprietary trading may be different, observationally, they are difficult to distinguish. In fact, some argued that proprietary trading could be deemed “risky market-making.” Duffie (2012) writes, “… an attempt to separate ‘legitimate and acceptable’ market-making from ‘speculative and risky’ market-making is not productive, in my opinion.” Duffie and other commenters suggested that the Volcker Rule could be particularly problematic in illiquid markets such as corporate bond markets. Duffie notes that whereas the average half-life of order imbalance in equities is three days, for investment grade corporate bonds it is roughly two weeks. Thus, metrics based on measures such as inventory aging and inventory turnover could be particularly problematic for market-making in corporate bonds. Furthermore, dealers who fear violating the Volcker Rule could be unable to properly manage inventory. One of the guidelines for the Volcker Rule is meeting “near-term customer demand.” But absent perfect predictions about future customer demand, market makers may be hesitant to acquire bonds in advance of a predicted spike in customer demand. The final rule also presents complications for fulfilling customer demand because of the required internal compliance metrics. The Federal Register (2014) notes that trades exceeding internal limits “should not be permitted simply because it responds to customer demand.” Instead, a banking entity is required to have escalation procedures that include “demonstrable analysis and approval.” Such regulations mean that market makers will find it particularly difficult to respond to large sells in the market. 9

One initial proposal that was dropped in the final rule was a requirement for detailed revenue attribution. This included identifying revenue attributable to the bid-ask spread as opposed to price appreciation. While the final rule no longer has such a requirement, it does have a profit and loss attribution requirement that focuses on revenue generation patterns. Abnormal patterns could raise a red flag and lead to further review. Given the illiquid nature and infrequent trading patterns in corporate bonds, this could potentially cause issues for market makers, particularly when a significant subset of its bonds has a severe order imbalance. In summary, Volcker Rule requirements have the potential to impact the behavior of dealers covered by the rule and lead to less liquid markets. Ambiguity as to what is legal marketmaking and what is prohibited proprietary trading may exacerbate the problem by pushing dealers toward more conservative trading strategies. New rules favoring customer-facing trades may discourage dealers from using the interdealer market, while inventory-based metrics may lead dealers to reduce their inventory exposure. Perhaps most pertinent to our study, the requirement that dealers set internal limits may result in dealers being unable to respond to increased customer demands during times of stress. With all of these theoretical reasons why the Volcker Rule may damage corporate bond liquidity, particularly in times of stress, we turn to assessing whether the empirical evidence is consistent with these concerns. 3. Data and Sample Description To examine how the Volcker Rule has affected corporate bond liquidity in stress periods, we focus on bond trading around times when a bond was downgraded from investment-grade to speculative-grade. Insurance companies, the dominant investors in the corporate bond market, face regulatory restrictions when investing in the corporate bond markets. The National Association of Insurance Commissioners (NAIC) classifies corporate bonds into six risk categories (NAIC1 to 10

NAIC6) based on their credit ratings, and requires insurance companies to maintain a higher level of capital when investing in bonds in a higher risk category.9 In addition, insurance companies are usually required to invest no more than 20% of their assets in bonds below NAIC2, i.e., speculative-grade bonds. Ellul, Jotikasthira, and Lundblad (2011) find that rating downgrades to speculative-grade can trigger fire sales in the bond market since greater capital requirements and other regulatory constraints prompt widespread divestment by insurance companies. Such regulation-induced fire sales generate high demand for liquidity, and can cause substantial price pressure in the absence of adequate liquidity provision. We obtain the rating history file from Mergent’s Fixed Income Securities Database (FISD) for the period from January 2006 to March 2016. This data file provides the announcement date of rating actions by the three largest rating agencies: Standard & Poor’s (S&P), Moody’s, and Fitch. We focus on fixed coupon corporate bonds with semi-annual coupon payments, $1000 par amount, and fixed maturity. These bonds are issued in U.S. Dollars by U.S. firms in the following three broad FISD industry groups: Industrial, Finance, and Utility. We exclude from our sample the following bonds: convertible or putable bonds, private placements, asset-backed issues, and issues which are part of a unit deal. Since rating agencies differ with respect to the timing of rating actions, we follow Ellul, Jotikasthira, and Lundblad (2011) and define the rating change event as the date of downgrade from investment-grade to speculative-grade announced by the first acting rating agency. We then extract data from FINRA’s TRACE database on corporate bond transactions during the one month following each rating downgrade. These data provide detailed information 9 Bonds rated AAA, AA, A are in NAIC risk category 1 (NAIC1). NAIC2-NAIC5 correspond to BBB, BB, B and CCC rated bonds respectively. Bonds rated CC or lower belong to NAIC 6. The capital charge for NAIC1 to NAIC6 is 0.4%, 1.3%, 4.6%, 10%, 23% and 30%, respectively. 11

on all secondary market transactions in the downgraded bond, including bond CUSIP, trade execution date and time, trade price and quantity, a buy or sell indicator, an indicator for agency or principal trade, and an indicator for inter-dealer trade. In addition, the data also contain information on dealers for each trade and, in the case of inter-dealer trades, both sides of the trade. Our version of TRACE is the regulatory version of TRACE, which has dealer identities. The standard version of TRACE, while including flags for dealer-customer and interdealer trades, does not identify the dealer(s) involved in a trade. Knowing dealer identities allows us to separately analyze liquidity provision by Volcker-affected and non-Volcker-affected dealers. Lastly, for each of the rating downgrades in our sample, we obtain characteristics information, including total par amount outstanding, issuance date and maturity date, from Mergent FISD. To examine bond liquidity during stress times, we estimate the average price impact during the one-month post-downgrade period in the spirit of Amihud (2002): (cid:1842)(cid:1870)(cid:1861)(cid:1855)(cid:1857)(cid:1835)(cid:1865)(cid:1868)(cid:1853)(cid:1855)(cid:1872) (cid:3404) (cid:2869) ∑ (cid:3015)(cid:3284) (cid:3435)(cid:3017)(cid:3284),(cid:3295)(cid:2879)(cid:3017)(cid:3284),(cid:3295)(cid:3127)(cid:3117)(cid:3439) , (cid:3036) (cid:4666)(cid:3015)(cid:3284)(cid:2879)(cid:2869)(cid:4667) (cid:3047)(cid:2880)(cid:2870) (cid:3018)(cid:3284),(cid:3295) where (cid:1842) and (cid:1843) represent the price (per $1000 of par value) and par amount (in thousands) of (cid:3036),(cid:3047) (cid:3036),(cid:3047) the t-th trade in bond i, and (cid:1840) represents the total number of trades during the one month following (cid:3036) the downgrade of bond i.10 In calculating the price impact measure, we exclude the following transactions: when-issued, cancelled, subsequently corrected, reversed trades, and exclude interdealer trades. Following Bessembinder, Kahle, Maxwell, and Xu (2009) and Ronen and Zhou (2013), we remove trades with $100,000 or less in par amount to avoid the substantial noise that these small trades introduce into prices. 10 Because we have transaction-level data with trade direction, we modify our calculation of the Amihud (2002) measure to use transaction-level data (as in Dick-Nielsen, Feldhutter, and Lando (2012)) and also use signed trades rather than using absolute changes in prices. As in Amihud (2002), which is based on the theoretical model of Kyle (1985), we aim to capture liquidity by using the response of price to order flow. 12

Table 1 presents our final sample of rating downgrades after matching FISD’s rating history file with FINRA’s TRACE data. A total of 687 bonds by 218 firms were downgraded from investment grade to speculative-grade during the period from January 2006 to March 2016. Moody’s acted first in 375 bonds, followed by S&P, which downgraded 247 bonds, and then Fitch who acted first for the remaining 89 bonds. Out of the 687 bonds, 356 were downgraded by one notch, and 157 were downgraded by two notches. The remaining 174 bonds were downgraded by three or more notches. We divide our sample period into five sub-periods: Pre-crisis Period (January 1, 2006 – June 30, 2007), Crisis Period (July 1, 2007 – April 30, 2009), Post-crisis Period (May 1, 2009 – July 20, 2010), Post-Dodd Frank Period (July 21, 2010 – March 31, 2014), and Post-Volcker Period (April 1, 2014 – March 31, 2016). We focus on comparing bond liquidity during the Post- Volcker Period with that during the other four sub-periods prior to the effective date of the Volcker Rule.11 The designations of the four pre-Volcker sub-periods are generally consistent with existing studies (e.g., Dick-Nielsen, Feldhutter, and Lando (2012), and Bessembinder et. al. (2016)).12 As pointed out by Trebbi and Xiao (2015), using exact dates of regulatory policies to study the impact of regulation on market liquidity is potentially complicated by anticipatory or delayed responses by market participants. For example, bank dealers might have become more conservative in market-making in anticipation of the rule prohibiting proprietary trading. In addition, regulators gave market participants over one year to fully comply with the Volcker Rule. Thus, using the effective date of the Volcker rule allows us to capture only partial effects of the Volcker Rule on bond liquidity and biases against finding results. The complete effects (including 11 Since our focus is on examining bond liquidity during the one-month following each downgrade, we exclude those downgrade events that happened during the last month in each of the five sub-periods. 12 The regulatory period that other papers study largely coincides with the period we classify as the Post-Dodd Frank Period. 13

both implementation and anticipation-related actions) could be larger than our empirical methodology captures. Table 1 shows how the distribution of sample of rating downgrades across the five subperiods. A total of 182 bonds were downgraded during the post-Volcker Period. The number of downgraded bonds increases from 114 for the Pre-Crisis Period to 210 for the Crisis Period, and then declines to 68 and 113 for the Post-Crisis Period and Post-Dodd Frank Period, respectively. 4. Liquidity around Stress Events Studying the effect of Volcker Rule on corporate bond liquidity during stress times is challenging since liquidity of the bond market might have changed over time for reasons unrelated to the post-crisis regulations. To account for the potential influence of such time trends, we use a difference-in-differences methodology by first comparing the price impact in the BB bonds newly downgraded from BBB with that in the existing BB bonds, and then examine how their differences have change from the Pre-Volcker periods to the Post-Volcker Period. Specifically, for each downgrade event, we calculate the average PriceImpact in bonds which were rated BB by the acting rating agency during the same one-month period, labeled as PriceImpactControl. i PriceImpactDiff is the first difference and is defined as the difference in PriceImpact between the i downgraded bond i and other BB bonds during the same one-month period (cid:1842)(cid:1870)(cid:1861)(cid:1855)(cid:1857)(cid:1835)(cid:1865)(cid:1868)(cid:1853)(cid:1855)(cid:1872)(cid:1830)(cid:1861)(cid:1858)(cid:1858) (cid:3404) (cid:1842)(cid:1870)(cid:1861)(cid:1855)(cid:1857)(cid:1835)(cid:1865)(cid:1868)(cid:1853)(cid:1855)(cid:1872) (cid:3398)(cid:1842)(cid:1870)(cid:1861)(cid:1855)(cid:1857)(cid:1835)(cid:1865)(cid:1868)(cid:1853)(cid:1855)(cid:1872)(cid:1829)(cid:1867)(cid:1866)(cid:1872)(cid:1870)(cid:1867)(cid:1864) . (cid:3036) (cid:3036) (cid:3036) We then compare PriceImpactDiff over different periods. 4.1. Univariate Analysis of Price Impact Measures Table 2 shows that the average (cid:1842)(cid:1870)(cid:1861)(cid:1855)(cid:1857)(cid:1835)(cid:1865)(cid:1868)(cid:1853)(cid:1855)(cid:1872)(cid:1830)(cid:1861)(cid:1858)(cid:1858) is 0.016 during the Post-Volcker (cid:3036) Period. This is substantially higher than the mere 0.003 during the Pre-Crisis period. It is also higher than the 0.007 and 0.011 for the Post-Dodd Frank Period and the Post-Crisis Period, and 14

only slightly smaller than the 0.018 for the Crisis Period. To benchmark these numbers, consider two trades at $1000 and $1016 (per $1000 in face value), respectively. Suppose that the second trade is for $1,000,000 in face value. This gives a price impact measure of (1016 – 1000)/1000 = 0.016. The changes in (cid:1842)(cid:1870)(cid:1861)(cid:1855)(cid:1857)(cid:1835)(cid:1865)(cid:1868)(cid:1853)(cid:1855)(cid:1872)(cid:1830)(cid:1861)(cid:1858)(cid:1858) across sub-periods mainly reflect changes in (cid:3036) (cid:1842)(cid:1870)(cid:1861)(cid:1855)(cid:1857)(cid:1835)(cid:1865)(cid:1868)(cid:1853)(cid:1855)(cid:1872) of the downgraded bonds, rather than those of the BB bonds in the control sample. (cid:3036) For our sample of downgraded bonds, (cid:1842)(cid:1870)(cid:1861)(cid:1855)(cid:1857)(cid:1835)(cid:1865)(cid:1868)(cid:1853)(cid:1855)(cid:1872) was 0.007 during the Pre-Crisis Period. It (cid:3036) jumped to 0.03 during the Crisis-Period, but has since declined to 0.021 in the Post-Crisis Period, and further to 0.015 in the Post-Dodd Frank Period. However, following the implementation of the Volcker Rule, the trend of declining price impact reversed: (cid:1842)(cid:1870)(cid:1861)(cid:1855)(cid:1857)(cid:1835)(cid:1865)(cid:1868)(cid:1853)(cid:1855)(cid:1872) increased to 0.024, (cid:3036) higher than any of the Pre-Volcker sub-periods except for the Crisis Period. This finding is intriguing given that (cid:1842)(cid:1870)(cid:1861)(cid:1855)(cid:1857)(cid:1835)(cid:1865)(cid:1868)(cid:1853)(cid:1855)(cid:1872)(cid:1829)(cid:1867)(cid:1866)(cid:1872)(cid:1870)(cid:1867)(cid:1864) did not change from the Post-Dodd Frank Period to (cid:3036) the Post-Volcker Period. In fact, the changes in (cid:1842)(cid:1870)(cid:1861)(cid:1855)(cid:1857)(cid:1835)(cid:1865)(cid:1868)(cid:1853)(cid:1855)(cid:1872)(cid:1829)(cid:1867)(cid:1866)(cid:1872)(cid:1870)(cid:1867)(cid:1864) over time for the control (cid:3036) sample of BB bonds not in stress are consistent with that documented in Bessembinder et al. (2016). In sum, bond liquidity around stress events have deteriorated since the Volcker Rule took effect. 4.2. Regression Analyses To check the statistical significance of the changes in (cid:1842)(cid:1870)(cid:1861)(cid:1855)(cid:1857)(cid:1835)(cid:1865)(cid:1868)(cid:1853)(cid:1855)(cid:1872)(cid:1830)(cid:1861)(cid:1858)(cid:1858) from the pre- (cid:3036) Volcker sub-periods to Post-Volcker Period, and also to control for the influences of other factors on bond liquidity during stress times, we conduct regression analyses in this section to further study the Volcker Rule effect on corporate bond liquidity. We create four dummy variables for the four sub-periods after the Pre-Crisis Period: Crisis, Post-Crisis, Post-Dodd Frank, and Post-Volcker. Crisis takes the value of one if a rating 15

downgrade occurred during the Crisis Period, and it takes the value of zero otherwise. The other three sub-period dummies are created in a similar way. We then regress (cid:1842)(cid:1870)(cid:1861)(cid:1855)(cid:1857)(cid:1835)(cid:1865)(cid:1868)(cid:1853)(cid:1855)(cid:1872)(cid:1830)(cid:1861)(cid:1858)(cid:1858) on the (cid:3036) four sub-period dummies, and a host of control variables. First, although all our sample bonds were downgraded from investment grade to speculative-grade, they differ from each other in terms of both pre-downgrade rating and the number of notches downgraded. Since such differences can affect bond trading following the downgrade announcement, and hence the PriceImpactDiff measure, we include as control i variables Previous Rating and ΔRating, which refer to the rating of the bond prior to the downgrade and the number of notches by which it was downgraded, respectively. Second, we control for bond characteristics, including (the log of) number of years since issuance (Log Age), number of years until maturity (Log Time to Maturity), and total par amount outstanding (Log Amount Outstanding). Lastly, we include into the regressions several variables that capture general market conditions during the same one-month period following each downgrade. These variable include aggregate market index returns, such as the return to the S&P 500 Index (SP500 Index Return), the return to the Barclays Capital U.S. Investment-Grade Corporate Bond Index (IV Bond Index Return) and the Barclays Capital U.S. High-Yield Corporate Bond Index (HY Bond Index Return). We also include changes in market volatilities, such as the change in CBOE stock market volatility index (ΔVIX), the change in the volatility of the Barclays Capital U.S. Investment-Grade Corporate Bond Index (ΔIV Bond Volatility) and the Barclays Capital U.S. High-Yield Corporate Bond Index (ΔHY Bond Volatility), and the change in 3 month LIBOR rate (Δ3M LIBOR). Changes in market volatilities and interest rates are calculated by comparing the one-month following a downgrade 16

to the one-month prior to the downgrade.13 Since bonds issued by the same firm are usually downgraded at the same time, we cluster the standard errors at the firm level. Column I of Table 4 presents the result from this regression analysis. We find that bond characteristics affect the PriceImpactDiff measure, with older bonds and bonds with longer timei to-maturity experiencing lower liquidity following their downgrade, while larger issues enjoy higher liquidity. Previous Rating and Rating Change do not have a significant impact on the PriceImpactDiff, and neither do the macro-economic variables. i More importantly, the coefficient for all four sub-period dummies are positive and significant at the 5% level, suggesting that bond liquidity around stress events has significantly deteriorated since the beginning of financial crisis. Consistent with the summary information presented in Table 2, the magnitude of the coefficient for sub-period dummies first declines monotonically from Crisis to Post-Dodd Frank, but then increases from Post-Dodd Frank to Post- Volcker. Tests on the differences in the coefficients on sub-period dummies show that the coefficient for Post-Volcker is significantly higher than that for Post-Dodd Frank, and it is not statistically significantly different from that for Crisis and Post-Crisis. These results suggests that bond liquidity around stress events has worsened following the implementation of the Volcker Rule, and it has deteriorated to a level similar to that during the financial crisis. To confirm that the increase in PriceImpactDiff for Post-Volcker is mainly driven by i higher price impact for the downgrade bonds, rather than lower price impact for BB bonds, we run the regression by using either PriceImpact or PriceImpactControl as the dependent variables. i i These results are presented in Columns II and III, respectively. For the sample of downgraded bonds, Post-Volcker has a significantly higher coefficient than Post-Dodd Frank. The coefficient 13 For ease of reference, we also provide a definition of all of these independent variables in Table 3. 17

for Post-Volcker is not statistically different from Crisis and Post-Crisis. These results mirror those from using PriceImpactDiff as the dependent variable. Meanwhile, for the sample of BB bonds i not experiencing any rating changes, there is no significant difference in the coefficients for Post- Volcker and Post-Dodd Frank, both statistically and economically. Taken together, these results are consistent with Volcker Rule degrading liquidity in the bond market around times of stress. 4.3. Robustness Checks We conduct three tests to examine the robustness of our results on post-Volcker bond liquidity changes. First, we investigate the sensitivity of our results to an alternative measure of liquidity, Realized Spread. Second, we study whether using the compliance date instead of the effective date of the Volcker Rule affects our results. And finally, we use a matched sample approach to conduct the difference-in-differences test. 4.3.1 Alternative Liquidity Measure Measuring liquidity in financial markets is challenging. The fact that most bonds do not trade often makes it even hard to measure liquidity in the bond market as almost all the existing bond liquidity measures rely on transaction data.14 The reliability of these liquidity measures varies with the amount of trades used in estimation. In this section, we estimate a measure of Realized Spread which has relatively low requirements on trade frequency. Specifically, for each downgraded bond, we first calculate the daily Realized Spread by taking the difference between volume weighted average customer buy prices (Ask) and volume weighted average customer sell prices (Bid) during the one-month following the downgrade. To avoid the noise embedded in small trades, we exclude trades with $100,000 or less in par amount. We then average the daily spread across days within the one-month period to get an event level estimate: RealizedSpread. i 14 One notable exception is Mahanti et al. (2008) who propose a latent liquidity measure for corporate bond by using the holding-weighted average turnover rate of bond portfolio of each fund that holds the bond. 18

For each downgrade event, we also calculate the average RealizedSpread in bonds which were rated BB by the acting rating agency during the same one-month period. We then subtract the average BB bond RealizedSpread from that of the downgraded bond to get a SpreadDiff measure. i We regress SpreadDiff on the four sub-period dummies and all the control variables and i the results are presented in Column I of Table 5. The coefficients for sub-period dummies declines from 0.166 for Crisis to 0.066 for Post-Crisis, and further to 0.051 for Post-Dodd Frank. However, the downward trend of RealizedSpread reverses following the implementation of the Volcker Rule. The coefficient of Post-Volcker is 0.09, which is higher than that for the Post-Dodd Frank at the 10% level. Therefore, liquidity as captured by RealizedSpread also seems to have deteriorated post-Volcker. 4.3.2 Alternative Definition of Post-Volcker Period The final Volcker Rule became effective April 1, 2014, but the compliance date for banks to fully conform their proprietary trading activities to the Volcker Rule was July 21, 2015. To examine how any lagged reaction of market participants to regulation during the gap between the effective date and compliance date affects our results, we use the compliance date of the Volcker Rule to redefine Post-Volcker period. Specifically, Post-Dodd Frank period now is from July 21, 2010 to July 20, 2015 and Post-Volcker period is from July 21, 2015 to March 31, 2016. The other sub-periods are defined as earlier. Column II of Table 5 again provides evidence of deteriorating liquidity following Volcker Rule. The coefficient of Post-Volcker is 0.026, more than double that of Post-Dodd Frank (0.011) and the different is statistically significant at the 10% level. Also similar to the results from using the Final Rule Effective date to define Post-Volcker, the coefficient of Post-Volcker is not significantly different from that of Crisis and Post-Crisis. 19

4.3.3 Alternative Approach for the Difference-in-Differences Test In examining how liquidity in downgraded bonds has changed over time, we compare each downgraded BB bond with a sample of BB bonds not experiencing any recent rating changes. Although both downgraded bonds and bonds in the control group have the same rating, they can differ in other key attributes, which could affect their liquidity. To account for this possibility, we use a matched sample approach by comparing each downgraded BB bond with a sub-sample of the BB bonds that are similar to the downgraded BB bond in terms of time-to-maturity, total par amount outstanding, and age. Specifically, we first segment BB bonds in the control group into three time to maturity categories: short-term (maturing within one year), medium-term (with time to maturity greater than one year by no more than seven years), and long-term (maturing over seven years). Within each maturity category, we further segment bonds into three size categories: small issue, medium issue, and large issue, using $0.5 Billion and $1.5 Billion in total par amount outstanding as the cutoffs. Finally, we divide bonds within each size category into new issues and seasoned issues, depending on whether its time since issuance is greater than one year. Therefore, we form a total of eighteen bond groups in the control sample based on time to maturity, amount outstanding, and age. We then calculate PriceImpactDiff for each downgraded bond by taking the difference between the PriceImpact of the downgraded bond and the average PriceImpact of BB bonds from the matching group during the same one-month period. Column III of Table 5 shows that using the matched sample approach has little impact on our results. We continue to observe that following Volcker Rule, the marginal deterioration in bond liquidity during stress times is as severe as during the financial crisis period. 5. Dealer Behavior Around Stress Events 20

In this section, we study how the behavior of dealers has changed around Volcker Rule implementation and, importantly, compare the behavior of Volcker-affected dealers and non- Volcker-affected dealers. In Subsection 5.1, we discuss how we identify whether a dealer is Volcker-affected and in Subsections 5.2 and 5.3, we document the change in behavior across the two groups over time. Finally, we discuss other regulations in Subsection 5.4. 5.1 Identifying Volcker-Affected Dealers A key issue is identifying which broker-dealers are subject to the Volcker Rule. This is a non-trivial task as full lists of Volcker-affected institutions are not published. In a study of the Volcker Rule, Oliver Wyman and SIFMA (2011) provide a list of 21 liquidity providers and whether they categorize as affected by the Volcker Rule.15 Of these 21 banks, they identify four (Cantor Fitzgerald & Co., Daiwa Capital Markets, Jefferies & Co., and Nomura) that are not affected by the Volcker Rule. Among those affected are major bank holding companies such as Goldman Sachs and Morgan Stanley. However, this list is far from complete as TRACE data identifies hundreds of dealers transacting in the bond market. To determine whether other broker-dealers are covered by the Volcker Rule, we follow the principle that the Volcker Rule was designed to prevent institutions with access to government backstops from participating in proprietary trading. The two most prominent backstops mentioned in the Federal Register (2014) discussion of the Volcker Rule are FDIC insurance and access to the Fed’s discount window. We start with the broker-dealers on the Oliver Wyman and SIFMA (2011) list and add to it the top 300 broker-dealers in terms of trading volume; together, these broker-dealers account for 97% of total bond market trade volume. We then search both the FDIC’s database of FDIC-insured banking institutions and the National Information Center’s 15 We reproduce this list in Table 6. 21

institution database to see which of the 300 broker-dealers were subject to the Volcker Rule. The former is relatively straightforward. If a broker-dealer, or more likely an affiliate (i.e., a commercial bank with the same parent holding company) is listed as having FDIC insurance, we code it as Volcker-affected. The latter is more complicated as the NIC database contains “banks and other institutions for which the Federal Reserve has a supervisory, regulatory, or research interest…” Thus, not all institutions in the database are necessarily Volcker-affected. We look for institutions coded as National Banks, State Member Banks, Bank Holding Companies, and Financial Holding Companies and treat these as Volcker-affected. Among the main types of institutions in the NIC database that we do not treat as Volcker-affected are Securities Broker/Dealers and Domestic Entity Other. As a third source, we search the Federal Reserve Board’s Resolution Plans website (https://www.federalreserve.gov/bankinforeg/resolutionplans.htm) to identify large bank holding companies under Fed supervision that must submit a living will.16 Combining results from our manual search with the list in Oliver Wyman and SIFMA (2011) results in approximately 45% of the top dealers being determined to be Volcker-affected. 5.2 Dealer Trading Activities We start by documenting basic dealer trading patterns around downgrade-to-speculative (“stress”) events in Table 7. In the month following a downgrade, the average turnover of downgraded bonds is close to 40% in the Post-Volcker period, higher than any of the other four periods in our sample. It is also much higher than the 9% monthly turnover of BB bonds during the Post-Volcker period, consistent with significant selling by insurance companies in the period just after downgrades to speculative-grade. However, the proportion of total trading volume that 16 Living wills are mandated by Dodd-Frank to prevent taxpayer bailouts in the future. Thus, the fact that regulators require a living will suggests that these are institutions with government backstops. 22

is dealer-customer (as opposed to interdealer), 62%, is roughly in line with the other periods of our sample. Of perhaps more interest, we compare Volcker-affected and non-Volcker-affected dealers in the other panels of Table 7. Our focus is on the proportion of dealer-customer trading handled by Volcker vs. non-Volcker dealers and also the dealers’ use of agency trading. The underlying evidence in Ellul, Jotikasthira, and Lundblad (2011) is that around stress events, some institutional investors (e.g., insurance companies) sell bonds due to regulatory constraints. Dealers then intermediate these trades, and potentially hold inventory in bonds when selling demand exceeds buying demand. Both the proportion of customer-dealer trades and the percentage of agency trades address how dealers react to customer demands. Volcker-affected dealers tend to be larger than non-Volcker dealers and handled 93% of dealer-customer volume around stress events in the pre-crisis period. Over our sample period, we see a gradual decline in the share of dealer-customer volume handled by Volcker-affected dealers. By the Post-Volcker period, non-Volcker dealers were handling almost one quarter of the dealercustomer volume. Though the increasing volume handled by non-Volcker dealers is consistent with Volcker-affected dealers scaling back their market-making due to the Volcker Rule, we cannot rule out the explanation that there has been a gradual time series change in the dealer business that has led the smaller, non-Volcker dealers to take a greater share of dealer-customer volume. Next, we turn to how agency trading has changed over time for Volcker and non-Volcker dealers. Agency trading occurs when a dealer has lined-up a counterparty to immediately offset a trade with a customer. For example, if an insurance company decides to sell a downgraded bond, a dealer in an agency trade would line-up another customer (or dealer) to purchase the bond. In 23

such a case, dealers do not commit capital or take on any inventory risk. A principal trade, in contrast, involves dealers taking on one side of a trade without pre-existing knowledge that they will be able to unwind the trade quickly. We follow Harris (2015) and Bessembinder et al. (2016) and define a trade as an agency trade if it is offset by another trade in the opposite direction within one minute. Our empirical results indicate that Volcker-affected dealers have increased the proportion of their total volume that is done on an agency basis. Pre-crisis, only 12% of the volume traded by Volcker-affected dealers was in agency trades. This number jumped to a little over 15% with the onset of the Financial Crisis and stayed fairly flat until jumping again to almost 23% with enforcement of the Volcker Rule. The sudden jump in the proportion of volume done as agency trades is suggestive of a causal effect of the Volcker Rule on Volcker-affected dealers’ willingness to hold bonds on their balance sheet without pre-arranging an offsetting trade. Non-Volcker dealers, in contrast, have seen a decline in the proportion of trades that they do on an agency basis. During the Pre-Crisis Period, almost half of the trades done by non-Volcker dealers around stress events were done as agency trades. By the Post-Volcker Period, this percentage had dropped to 29%. To more formally study the changes in agency trades across time for Volcker and non- Volcker-affected dealers, we run a regression of proportion of trades that are agency trades on period dummies and controls. Our base regression is, (cid:1842)(cid:1870)(cid:1867)(cid:1868)(cid:1867)(cid:1870)(cid:1872)(cid:1861)(cid:1867)(cid:1866) (cid:1867)(cid:1858) (cid:1853)(cid:1859)(cid:1857)(cid:1866)(cid:1855)(cid:1877) (cid:1874)(cid:1867)(cid:1864)(cid:1873)(cid:1865)(cid:1857) (cid:3036) (cid:3404) (cid:2010) (cid:3397)(cid:2010) (cid:1829)(cid:1870)(cid:1861)(cid:1871)(cid:1861)(cid:1871)(cid:3397) (cid:2010) (cid:1842)(cid:1867)(cid:1871)(cid:1872)(cid:3398)(cid:1829)(cid:1870)(cid:1861)(cid:1871)(cid:1861)(cid:1871)(cid:3397)(cid:2010) (cid:1842)(cid:1867)(cid:1871)(cid:1872)(cid:3398)(cid:1830)(cid:1867)(cid:1856)(cid:1856) (cid:1832)(cid:1870)(cid:1853)(cid:1866)(cid:1863) (cid:2868) (cid:2869) (cid:2870) (cid:2871) (cid:3397) (cid:2010) (cid:1842)(cid:1867)(cid:1871)(cid:1872)(cid:3398)(cid:1848)(cid:1867)(cid:1864)(cid:1855)(cid:1863)(cid:1857)(cid:1870)(cid:3397)(cid:2011)(cid:1850)(cid:3397)(cid:2035), (cid:2872) 24

where the unit of observation is a stress event, the dependent variable is the proportion of volume done by either Volcker or non-Volcker dealers done on an agency basis, and the omitted period dummy is the Pre-Crisis Period. X represents a vector of control variables that are the same as defined in Table 3 and used in Table 4. Our variable of interest is β , which directly measures the 4 difference in the proportion of volume done on an agency basis between the Post-Volcker Period and the Pre-Crisis Period. Also of interest is the difference between β and the coefficients on the 4 other sub-period dummies. The regression results are presented in Table 8. In the first column, the dependent variable is the proportion of agency trading done by Volcker Rule affected dealers after stress events. The coefficient on the Post-Volcker dummy is 0.133, indicating a 13 percentage point increase in volume done on an agency basis relative to the pre-crisis period. This change is slightly larger than the 11 percentage point increase without controls in Table 7. Importantly, we also see that the coefficient on the Post-Volcker dummy is also significantly larger than for the other periods in our sample. The Post-Dodd Frank period has a statistically significant coefficient of 0.046, indicating a 4.6 percentage point increase in agency trades compared to the Pre-Crisis Period, but also much smaller than the Post-Volcker Period. The nine percentage point increase in agency trading from the Post-Dodd Frank Period to the Post-Volcker Period is both statistically and economically significant. In the second column of Table 8, we re-run our agency trade regression, but instead consider the proportion of trades done on an agency basis by non-Volcker-affected dealers. While the coefficients on all of the sub-period dummies are negative and the amount of agency trading done by non-Volcker dealers is smaller (in magnitude) during the Post-Volcker Period as compared to any other period, we do not find any statistical significance. In particular, unlike 25

Volcker-affected dealers, we do not find a sharp jump in the proportion of agency trading for non- Volcker dealers upon implementation of the Volcker Rule. If anything, we find the opposite, at least in terms of point estimates. Our results are consistent with the Volcker Rule inducing Volcker-affected dealers to shift from principal to agency trading as a way to avoid inventory imbalance. 5.3 Dealer Capital Commitment A more direct measure of dealers’ willingness to hold inventory imbalances is the timeweighted capital commitment. In the one month following a stress event, we calculate for each dealer the absolute deviation from starting inventory. The intuition is that if a dealer starts with a particular desired inventory level, the first purchase moves the dealer above this desired inventory level, but a following sell will again move the dealer back towards the desired inventory level. The actual desired inventory level is unobservable, so our implicit assumption is that the starting level of inventory is optimal. To calculate how far a dealer is from the starting inventory level, we simply take the accumulated buys and subtract the accumulated sells from the starting point. To calculate the time-weighted capital commitment, we then average the absolute distance from the starting inventory, weighting by the amount of time the inventory level is held. While our measure is similar to the dealer capital commitment measure in Bessembinder et al. (2016), it is important to note that we measure capital commitment over the course of a month while they construct a daily measure. Their measure implicitly assumes that the starting point at each day is the optimal inventory, whereas our monthly measure allows for inventory to continue to move away from optimal inventory over the course of a few days. In particular, if a dealer has purchased a large volume of a bond in a day and has not sold this volume to another customer or 26

dealer, the dealer still has significant capital commitment the next day. Once we calculate bonddealer level capital commitment, we sum across dealers for a given stress event. In Table 9, we report regressions of time-weighted dealer capital commitment on subperiod dummies and controls separately for Volcker-affected and non-Volcker-affected dealers, similar to our proportion of agency volume regressions. The units for time-weighted dealer capital commitment are the number of bonds, with each bond being $1000 in face value. In the first column, we find that dealer capital commitment by Volcker-affected dealers has declined in all periods relative to the pre-crisis period. Dealer capital commitment is roughly $10 million in face value lower on average for a downgraded bond during the Crisis, Post-Crisis, and Post-Dodd Frank periods as compared to the Pre-Crisis period.17 For the Post-Volcker Period, this decline is $20 million in face value relative to the Pre-Crisis Period. The lower capital commitment for the Post- Volcker Period is also statistically larger for the Volcker Period than it is for the Crisis, Post-Crisis, and Post-Dodd Frank periods. Thus, while there is a large and sudden drop in capital commitment from the Pre-Crisis to the Crisis Period, there is also a large and sudden drop from the Post-Dodd Frank Period to the Post-Volcker Period, suggesting that there was a significant shift in Volckeraffected dealers around the implementation of the Volcker Rule. Column II of Table 9 considers whether capital commitment has changed for non-Volcker dealers. Our results indicate that capital commitment has actually increased for non-Volcker dealers during the Post-Volcker period, in contrast to Volcker-affected dealers. 5.4 Capital Commitment and Basel III 17 As a benchmark, the average capital commitment of Volcker-affected dealers in BB-rated bonds during non-stress periods is $6 million. For non-Volcker dealers, it is $1.6 million. 27

A potential concern in trying to isolate a Volcker Rule implementation effect is that, in the post-crisis period, a number of reforms were passed to regulate the finance industry. In particular, the Comprehensive Capital Analysis and Review (CCAR) process began in 2011, requiring bank holding companies (BHCs) to submit capital plans to the Federal Reserve. The capital plan requires that the BHC is able to maintain minimum capital requirements even under stress scenarios, providing a stiff test of a BHC’s regulatory capital. The punishment for not passing a CCAR test is that the BHC is not allowed to make capital distributions unless the Federal Reserve indicates in writing that it allows the distribution.18 Each year, the Federal Reserve publishes a list of BHCs that have either failed their CCAR tests or received only a conditional pass. In January 2014, the start of Basel III implementation went into effect, adding additional capital requirements above what was required in Basel II. In conjunction with CCAR regulations, this potentially made banks more capital constrained and may have caused BHCs to change their market-making businesses.19 To test the hypothesis that it was the combination of Basel III and CCAR that is driving our results on dealers, we split dealers into those that were CCARconstrained and those that were not. We classify any bank that failed a CCAR test or was given a conditional pass in 2014 or 2015 as CCAR-constrained.20 If BHCs change their market-making behavior in response to changing capital requirements, then we would expect CCAR-constrained banks to lower their capital commitment more than BHCs that were able to pass their CCAR tests. In Table 10, we find that both dealers that passed the CCAR tests and dealers that failed or conditionally passed the CCAR test had lower capital commitment in the Volcker Rule 18 Historically, the Federal Reserve has continued to allow failed BHCs to continue capital distributions at the same rate as in the past. Effectively, the main constraint is that these BHCs cannot increase their capital distributions. 19 We thank Darrell Duffie for suggesting the CCAR linkage to us. 20 BHCs receiving a conditional pass are required to remediate deficiencies and resubmit a new capital plan later in the year. Thus, such BHCs would have similar incentives to BHCs that fail CCAR tests. 28

implementation period as compared to the Post-Dodd Frank Period. However, the decline in capital commitment is higher for the dealers that passed CCAR tests (roughly $4.5 million) than for dealers that failed or conditionally passed CCAR tests (roughly $2.6 million), a result at variance with the prediction above.21 Hoarding capital to pass the CCAR test is thus not supported as an explanation for decreased dealer capital commitment in bond trading. 6. Conclusion In this paper, we study the impact of Volcker Rule implementation on corporate bond illiquidity and dealer behavior. Our main finding is that the Volcker Rule has a deleterious effect on corporate bond liquidity and dealers subject to the Rule become less willing to provide liquidity during stress times. While dealers not affected by the Volcker Rule have stepped in to provide liquidity, we find that the net effect is a less liquid corporate bond market. We also rule out that the effects are due to the implementation of Basel III in conjunction with CCAR requirements. Our study focuses on events where investment-grade bonds are downgraded to speculativegrade to capture plausible events of forced selling. Using these stress events, we find that downgraded bonds exhibit a larger price impact of trading than a control group of BB bonds. More importantly, the relative level of the excess price impact is larger after the Volcker Rule is implemented than the period just before the Volcker Rule is implemented. Indeed, we find the disturbing result that illiquidity in stress periods is now approaching levels see during the financial crisis. 21 It is possible that the BHCs that passed their CCAR tests chose to change their capital commitment in anticipation of Basel III, prior to the actual implementation, whereas the BHCs that failed did not. Nevertheless, this also predicts that if BHCs commit less capital to try to pass CCAR tests, we should still see stronger declines in capital commitment during the Volcker implementation period for those BHCs that failed CCAR tests, relative to those BHCs that passed CCAR tests. 29

Examining individual dealer behavior allowed us to rule out the possibility that our results are driven simply by time series changes in dealer behavior. We find that following Volcker Rule implementation Volcker-affected dealers are less involved in dealer-customer trades, use a greater proportion of agency trades, and are less willing to commit capital. Non-Volcker dealers pick-up a greater proportion of dealer-customer trades and do not have statistically significant changes in their use of agency trades or willingness to commit capital. Splitting Volcker-affected dealers into those who have failed CCAR tests in 2014 and 2015 and those who have not, we find that capital commitment among downgraded bonds has decreased more for dealers that passed CCAR tests, a result inconsistent with a Basel III explanation for decreased bond market liquidity. Overall, our results show that the Volcker Rule has had a real effect on dealer behavior, with significant effects only on those dealers affected by the Volcker Rule and not all bond dealers. 30

Table 1: Sample Description This table provides a description of the corporate bonds downgraded from investment grade to speculative-grade by one of three major credit rating agencies (S&P, Moody’s, and Fitch) over the period from January 1, 2006 to March 31, 2016. Data on historical rating changes by the three major rating agencies are obtained from Mergent’s Fixed Income Securities Database (FISD). We use the date of announcement by the rating agency who acted first to define the downgrade event. We divide the full sample period into five subperiods: Pre-crisis Period (January 1, 2006 – June 30, 2007), Crisis Period (July 1, 2007 – April 30, 2009), Post-crisis Period (May 1, 2009 – July 20, 2010), Post-Dodd Frank Period (July 21, 2010 – March 31, 2014), and Post-Volcker Period (April 1, 2014 – March 31, 2016). For the full sample period and each of the sub-periods, we present the number of bonds downgraded and number of firms whose bonds were downgraded. We report the number of bonds in which S&P, Moody’s, or Fitch was the first to take action. Note that multiple rating agencies can downgrade a bond on the same day. We also report the number of bonds that were downgrade by one notch, by two notches, and by three or more notches (which are in the “other” column) respectively. # of # of by by by by One by Two others Bonds Firms S&P Moody's Fitch Notch Notches full sample period January 1, 2006 – March 31, 2016 687 218 247 375 89 356 157 174 by sub-periods Pre-crisis Period (January 1, 2006 – June 30, 2007) 114 45 36 50 37 54 17 43 Crisis Period (July 1, 2007 – April 30, 2009) 210 57 105 100 8 97 80 33 Post-crisis Period (May 1, 2009 – July 20, 2010) 68 16 3 61 4 46 3 19 Post-Dodd Frank Period (July 21, 2010 – March 31, 2014) 113 45 51 41 33 64 28 21 Post-Volcker Period (April 1, 2014 – March 31, 2016) 182 55 52 123 7 95 29 58 31

Table 2: Univariate Analysis of Corporate Bond Liquidity Following Downgrades This table analyzes the liquidity of a bond during the one-month following its downgrade from investment-grade to speculative-grade over the period from January 1, 2006 to March 31, 2016. We measure bond liquidity by using a price impact measure, PriceImpact = (P – P )/Q, where (cid:1842) and (cid:1843) refers to the price (per $1000 of par value) and par amount (in thousands) of the trade at time t respectively. t t-1 t (cid:3047) (cid:3047) Retail-sized trades, i.e., those with par amount less than $100,000 are excluded from calculation in order to avoid the noise they tend to carry as suggested by Bessembinder et al. (2009). We first calculate the PriceImpact measure for each trade, and then average it across the trades within the one-month following each downgrade to get an event level estimate. For each downgrade event, we also calculate the average (cid:1842)(cid:1870)(cid:1861)(cid:1855)(cid:1857)(cid:1835)(cid:1865)(cid:1868)(cid:1853)(cid:1855)(cid:1872) in bonds which were rated BB by the acting rating agency during the same one-month period, and is labelled as PriceImpactControl. PriceImpactDiff is the difference in PriceImpact between the downgraded bond and other BB bonds. We divide the full sample period into five sub-periods: Pre-crisis Period (January 1, 2006 – June 30, 2007), Crisis Period (July 1, 2007 – April 30, 2009), Post-crisis Period (May 1, 2009 – July 20, 2010), Post-Dodd Frank Period (July 21, 2010 – March 31, 2014), and Post-Volcker Period (April 1, 2014 – March 31, 2016). Since our focus is on examining bond liquidity during the one-month following each downgrade, we exclude those downgrade events that happened during the last month in each of the sub-periods. We test whether (cid:1842)(cid:1870)(cid:1861)(cid:1855)(cid:1857)(cid:1835)(cid:1865)(cid:1868)(cid:1853)(cid:1855)(cid:1872)(cid:1830)(cid:1861)(cid:1858)(cid:1858) for each sub-period is statistically different from zero and report the p-value. We also conduct tests on the differences in the (cid:1842)(cid:1870)(cid:1861)(cid:1855)(cid:1857)(cid:1835)(cid:1865)(cid:1868)(cid:1853)(cid:1855)(cid:1872) liquidity measures between two sub-periods and report the p-value. Since bonds issued by the same firm are usually downgraded at the same time, we cluster the standard errors at the firm level. PriceImpact PriceImpact PriceImpact Number Number p-value Control Diff of Bonds of Firms Pre-crisis Period 0.007 0.004 0.003 0.062 114 45 Crisis Period 0.030 0.012 0.018 0.000 210 57 Post-crisis Period 0.021 0.011 0.011 0.129 68 16 Post-Dodd Frank Period 0.015 0.008 0.007 0.000 113 45 Post-Volcker Period 0.024 0.008 0.016 0.007 182 55 32

Table 3: Independent Variable Definitions This table provides detailed definitions of independent variables used in the tables below. Dependent variables are defined in the respective tables that they are used in. Variable Definition Crisis Dummy variable equal to 1 for July 1, 2007 to April 30, 2009 and 0 otherwise. Post-crisis Dummy variable equal to 1 for May 1, 2009 to July 20, 2010 and 0 otherwise. Post-Dodd Frank Dummy variable equal to 1 for July 21, 2010 to March 31, 2014 and 0 otherwise. Post-Volcker Dummy variable equal to 1 for April 1, 2014 to March 31, 2016 and 0 otherwise. Previous Rating The rating of a downgraded bond before the downgrade from investment-grade to speculative-grade. A numeric value is assigned to each notch of credit rating, with 1, 2, 3, 4 … denoting AAA, AA+, AAA, AA- … respectively. ΔRating The number of notches that a bond was downgraded. Age The log of the number of years since issuance for a bond. Time-to-Maturity The log of the number of years to maturity for a bond. Amount Outstanding The log of the total amount outstanding in $thousands. S&P 500 Index Return The return of the S&P 500 over the one-month post-downgrade period. It is expressed in decimal form. IV Bond Index Return The return to the Barclays Capital U.S. Investment-Grade Corporate Bond Index over the one-month post-downgrade period. It is expressed in decimal form. HY Bond Index Return The return to the Barclays Capital U.S. High-Yield Corporate Bond Index over the one-month post-downgrade period. It is expressed in decimal form. ΔVIX The change in CBOE stock market volatility index from the onemonth pre-downgrade period to the one-month post-downgrade period. ΔIV Bond Volatility The change in the standard deviation of the Barclays Capital U.S. Investment-Grade Corporate Bond Index Return from the onemonth pre-downgrade period to the one-month post-downgrade period. It is expressed in decimal form. ΔHY Bond Volatility The change in the standard deviation of the Barclays Capital U.S. High-Yield Corporate Bond Index Return from the one-month predowngrade period to the one-month post-downgrade period. It is expressed in decimal form. Δ3M LIBOR Change The change in the 3 month LIBOR rate (in percentage) from the one-month pre-downgrade period to the one-month postdowngrade period. 33

Table 4: Corporate Bond Liquidity Following Downgrades This table analyzes how corporate bond liquidity evolves during the period from January 1, 2006 to March 31, 2016, especially following the effective date of Volcker rule. The dependent variables for Columns I – III are PriceImpactDiff, PriceImpact, and PriceImpactControl, respectively. PriceImpact is price impact of trading in a downgraded bond in the month after the downgrade. PriceImpactControl is the average price impact for BB-rated corporate bonds in same horizon as PriceImpact. PriceImpactDiff is the difference between PriceImpact and PriceImpactControl. The primary independent variables of interest are dummy variables for Crisis, Post-crisis, Post- Dodd Frank, and Post-Volcker. Detailed definitions of all of the independent variables are provided in Table 3. Since bonds issued by the same firm are usually downgraded at the same time, we cluster the standard errors at the firm level. III. I. PriceImpactDiff II. PriceImpact PriceImpactControl Estimate p-value Estimate p-value Estimate p-value Intercept -0.024 0.339 -0.001 0.491 0.022 0.000 Crisis 0.017 0.003 0.022 0.000 0.006 0.000 Post-crisis 0.016 0.024 0.024 0.002 0.007 0.000 Post-Dodd Frank 0.010 0.010 0.013 0.001 0.004 0.000 Post-Volcker 0.021 0.002 0.025 0.000 0.004 0.000 Previous Rating -0.002 0.361 -0.004 0.252 -0.002 0.000 ΔRating 0.001 0.348 0.000 0.405 0.000 0.176 Log Age 0.005 0.015 0.005 0.014 0.000 0.213 Log Time to Maturity 0.008 0.010 0.008 0.014 0.000 0.142 Log Amount Outstanding -0.016 0.000 -0.016 0.000 -0.001 0.018 SP500 Index Return 0.006 0.477 -0.002 0.493 -0.008 0.227 IV Bond Index Return -0.021 0.460 -0.031 0.443 -0.010 0.347 HY Bond Index Return -0.152 0.299 -0.182 0.267 -0.030 0.015 ΔVIX -0.001 0.242 -0.001 0.201 0.000 0.142 ΔIV Bond Volatility 0.313 0.457 1.163 0.345 0.850 0.009 ΔHY Bond Volatility 0.146 0.476 -0.500 0.418 -0.645 0.004 Δ3M LIBOR -0.015 0.251 -0.012 0.300 0.003 0.024 Number of Observations 687 687 687 R2 0.079 0.079 0.079 Post-Volcker vs Crisis 0.296 0.384 0.062 Post-Volcker vs Post-crisis 0.315 0.454 0.000 Post-Volcker vs Post-Dodd Frank 0.037 0.040 0.477 34

Table 5: Robustness Checks on Liquidity Following Downgrades This table presents results from robustness checks of the analyses on bond liquidity changes following the implementation of the Volcker Rule. In Column I, we use an alternative measure, Spread, to capture bond liquidity. For each downgraded bond, we first calculate daily Spread by taking the difference between volume weighted average customer buy prices (Ask) and volume weighted average customer sell prices (Bid) during the one-month following the downgrade. We then average the daily spread across days within the month to get an event level estimate. For each downgrade event, we also calculate the average Spread in bonds which were rated BB by the acting rating agency during the same one-month period. We then subtract the average BB bond Spread from that of the downgraded bond to get a SpreadDiff measure. This is the dependent variable for the regression in Column I. In Column II, we used the compliance date for banks to conform their proprietary trading activities and investments in and relationships with non-legacy covered funds under the Volcker Rule, which is July 21, 2015, to define Post-Volcker period. Specifically, Post- Dodd Frank period is from July 21, 2010 to July 20, 2015 and Post-Volcker period is from July 21, 2015 to March 31, 2016. The other sub-periods during our sample are defined as earlier. In Column III, we compare each downgraded bond to the average of other BB rated bonds with similar time to maturity, amount outstanding, and age when calculating the PriceImpactDiff measure. We first segment bonds into three time to maturity categories: short-term (maturing within one year), medium-term (with time to maturity greater than one year by no more than seven years), and long-term (maturating over seven years). Within each maturity category, we further segment bonds into three size categories: small issue, medium issue, and large issue, using $0.5 Billion and $1.5 Billion in total par amount outstanding as the cutoffs. Finally, we divide bonds within each size category into new issues and seasoned issues, depending on whether its time since issuance is greater than one year. Therefore, we form a total of eighteen bond groups based on time to maturity, amount outstanding, and age. We then calculate (cid:1842)(cid:1870)(cid:1861)(cid:1855)(cid:1857)(cid:1835)(cid:1865)(cid:1868)(cid:1853)(cid:1855)(cid:1872)(cid:1830)(cid:1861)(cid:1858)(cid:1858) for each downgraded bond by taking the difference between the PriceImpact of the downgraded bond and the average (cid:1842)(cid:1870)(cid:1861)(cid:1855)(cid:1857)(cid:1835)(cid:1865)(cid:1868)(cid:1853)(cid:1855)(cid:1872) of BB bonds from the matching group during the same one-month period. Downgrade events that happened during the last month in each of the newly defined subperiods are excluded since the one-month following those downgrades overlapped with the next sub-period. The dependent variable in Column II is PriceImpactDiff as in Table 4. The primary independent variables of interest are dummy variables for Crisis, Post-crisis, Post-Dodd Frank, and Post-Volcker. Detailed definitions of all of the independent variables are provided in Table 3 Since bonds issued by the same firm are usually downgraded at the same time, we cluster the standard errors at the firm level. 35

I. Measuring II. Use Compliance III. Use Matched Liquidity Date to Define Bonds to Calculate by Spread Post-Volcker PriceImpactDiff Estimate p-value Estimate p-value Estimate p-value Intercept -0.229 0.165 -0.033 0.285 -0.035 0.266 Crisis 0.166 0.002 0.016 0.003 0.013 0.013 Post-crisis 0.066 0.152 0.017 0.020 0.013 0.061 Post-Dodd Frank 0.051 0.156 0.011 0.007 0.008 0.028 Post-Volcker 0.090 0.056 0.026 0.007 0.019 0.004 Previous Rating 0.008 0.349 -0.001 0.418 -0.001 0.450 ΔRating -0.022 0.004 0.001 0.353 0.000 0.462 Log Age -0.009 0.292 0.005 0.013 0.005 0.023 Log Time to Maturity 0.076 0.002 0.008 0.008 0.008 0.006 Log Amount Outstanding -0.004 0.433 -0.017 0.000 -0.015 0.000 SP500 Index Return -0.328 0.186 0.016 0.436 0.013 0.443 IV Bond Index Return -0.440 0.375 -0.071 0.368 -0.006 0.489 HY Bond Index Return 0.932 0.152 -0.176 0.269 -0.158 0.286 ΔVIX 0.007 0.145 -0.001 0.198 -0.001 0.221 ΔIV Bond Volatility 10.592 0.221 -0.009 0.499 -0.096 0.487 ΔHY Bond Volatility -21.171 0.025 0.336 0.445 0.705 0.382 Δ3M LIBOR 0.105 0.101 -0.018 0.223 -0.020 0.184 Number of Observations 647 687 687 R2 0.058 0.079 0.074 Post-Volcker vs Crisis 0.050 0.196 0.212 Post-Volcker vs Post-crisis 0.283 0.217 0.255 Post-Volcker vs Post-Dodd Frank 0.075 0.070 0.034 36

Table 6: Major Liquidity Providers in the Corporate Bond Market and Volcker Rule This table provides a list of 21 major securities dealers and whether they are subject to the Volcker Rule. Dealers affected by the Volcker Rule are prohibited from participating in proprietary trading, but have a market-making exception. Non-affected dealers are not subject to bans on proprietary trading or market-making. Source: "The Volcker Rule restrictions on proprietary trading: Implications for the US corporate bond market" presentation by Oliver Wyman and SIFMA. Dealers Affected by Volcker Rule Dealers Not Affected by Volcker Rule Bank of Nova Scotia Cantor Fitzgerald & Co. Barclays Capital Daiwa Capital Markets Americas BMO Capital Markets Jefferies & Company BNP Paribas Securities Nomura Securities International Citigroup Global Capital Markets Credit Suisse Securities (USA) Deutsche Bank Securities Goldman, Sachs & Co. HSBC Securities (USA) J.P. Morgan Securities Merrill Lynch, Pierce, Fenner & Smith Mizuho Securities USA Morgan Stanley & Co. RBC Capital Markets RBS Securities SG Americas Securities UBS Securities 37

Table 7: Trading Activity by Dealers Affected by Volcker Rule and those not affected by Volcker Rule This table presents summary information on trading activities by dealers affected by Volcker rule and those not affected by Volcker rule during each of the five sub-periods between January 1, 2006 to March 31, 2016: Pre-crisis Period (January1, 2006 – June 30, 2007), Crisis Period (July 1, 2007 – April 30, 2009), Post-crisis Period (May 1, 2009 – July 20, 2010), Post-Dodd Frank Period (July 21, 2010 – March 31, 2014), and Post-Volcker Period (April 1, 2014 – March 31, 2016). All Trade refers to the aggregate trade volume by all dealers, including both inter-dealer trade and dealer-customer trade, during the one-month following each rating downgrade. To control for the effect of issue size on trade volume, we first divide the aggregate one-month trade volume for each downgrade event by the total par amount outstanding of the downgraded bond, and then average it across bonds within each sub-period. We report the percent of Dealer-Customer trade volume out of the total trade volume for all dealers (D-C Trade), as wells as for dealers affected by Volcker rule and those not affected by Volcker rule separately. In addition, for the two groups of dealers, we also report their respective share of the total dealer-customer trade volume (Share of Total D-C Trade), and the percentage of their dealer-customer trade that is effectively agent trade (Dealer Agency Trade). We classify a trade as being effectively agent if it offset by another trade that occurred within one minute with the same trade size by the same dealer but with opposite trade direction. This one-minute algorithm is similar to that used in Harris (2015) and Bessembinder et al. (2016). Dealers Not Affected by Volcker Full Sample Dealers Affected by Volcker Rule Dealer Dealer All D-C D-C Share of Total Agency D-C Share of Total Agency Trade Trade Trade D-C Trade Trade Trade D-C Trade Trade (%) (%) (%) (%) (%) (%) (%) Pre-crisis Period 0.300 65.601 77.428 93.371 12.104 22.119 6.629 46.404 Crisis Period 0.277 70.037 83.295 89.117 15.413 37.803 10.883 45.326 Post-crisis Period 0.219 62.324 75.958 84.569 15.543 29.192 15.431 35.779 Post-Dodd Frank Period 0.306 53.913 67.224 79.728 15.965 33.063 20.272 33.118 Post-Volcker Period 0.383 62.032 75.608 76.297 22.709 48.722 23.703 29.403 38

Table 8: Volcker Rule and Agency Trades This table analyzes how dealers’ willingness to arrange trades on a principal basis change following Volcker Rule. We first estimate for each dealer the proportion of dealer-customer trade volume completed on effectively agent basis (expressed in decimals) during the one-month following each downgrade. For each downgrade, we divide the active dealers into two groups: those affected by Volcker rule and those that were not. We then average the percent of agency trade across dealers within each dealer group, and use them as the dependent variables in Columns I and II. The primary independent variables of interest are dummy variables for Crisis, Post-crisis, Post-Dodd Frank, and Post-Volcker. Detailed definitions of all of the independent variables are provided in Table 3 Since bonds issued by the same firm are usually downgraded at the same time, we cluster the standard errors at the firm level. I. Dealers Affected II. Dealers Not Affected by Volcker Rule by Volcker Rule Estimate p-value Estimate p-value Intercept 0.219 0.163 -0.041 0.454 Crisis Period 0.030 0.176 -0.026 0.361 Post-crisis Period 0.037 0.199 -0.062 0.175 Post-Dodd Frank Period 0.046 0.049 -0.048 0.232 Post-Volcker Period 0.133 0.000 -0.077 0.114 Previous Rating -0.016 0.223 0.035 0.147 ΔRating -0.011 0.086 0.017 0.087 Log Age 0.016 0.083 0.026 0.078 Log Time to Maturity 0.006 0.384 -0.008 0.391 Log Amount Outstanding -0.035 0.006 -0.020 0.209 SP500 Index Return -0.164 0.280 -0.627 0.110 IV Bond Index Return -0.520 0.292 1.421 0.226 HY Bond Index Return 0.688 0.117 -0.821 0.232 ΔVIX -0.002 0.232 -0.008 0.114 ΔIV Bond Volatility 4.730 0.386 -2.517 0.451 ΔHY Bond Volatility 20.346 0.004 3.674 0.409 Δ3M LIBOR -0.082 0.105 0.016 0.431 Number of Observations 687 687 R2 0.073 0.031 Post-Volcker vs Crisis 0.003 0.212 Post-Volcker vs Post-crisis 0.019 0.375 Post-Volcker vs Post-Dodd Frank 0.003 0.266 39

Table 9: Dealer Capital Commitment around Volcker Rule Implementation This table analyzes how dealers’ willingness to commit their own capital to bond trading changes following Volcker rule. During the one-month following a bond’s downgrade, we first calculate for each dealer, the absolute value of a dealer’s accumulated principal buy volume and accumulated principal sell volume at the time of each of the dealer’s trades in the downgraded bond (in thousands of dollars of face value). We then average the absolute difference between accumulated buys and accumulated sells across trades within the one-month for each dealer, weighting each observation by the time for which the capital is committed. Trades that were not offset prior to day end hence received larger weight in the capital commitment calculation. For each downgrade, we divide the active dealers into two groups: those affected by Volcker rule and those that were not. We then aggregate each dealer’s capital commitment measure within each dealer group, and use them as the dependent variables in Columns I and II. The primary independent variables of interest are dummy variables for Crisis, Post-crisis, Post-Dodd Frank, and Post-Volcker. Detailed definitions of all of the independent variables are provided in Table 3. Since bonds issued by the same firm are usually downgraded at the same time, we cluster the standard errors at the firm level. 40

I. Dealers Affected II. Dealers Not Affected by Volcker Rule by Volcker Rule Estimate p-value Estimate p-value Intercept 45675.140 0.030 6371.760 0.003 Crisis Period -9025.310 0.003 -617.660 0.012 Post-crisis Period -11231.500 0.000 609.720 0.023 Post-Dodd Frank Period -12891.800 0.000 1879.880 0.000 Post-Volcker Period -20127.890 0.000 2131.810 0.000 Previous Rating -894.800 0.337 -391.460 0.029 ΔRating 1371.910 0.025 -117.810 0.099 Log Age -5690.310 0.000 -321.400 0.034 Log Time to Maturity 4250.750 0.002 50.050 0.387 Log Amount Outstanding 16734.870 0.000 966.710 0.000 SP500 Index Return 55463.810 0.004 3324.580 0.072 IV Bond Index Return 38394.350 0.301 -13601.550 0.162 HY Bond Index Return -149809.210 0.000 -8738.820 0.085 ΔVIX -589.090 0.033 -42.560 0.110 ΔIV Bond Volatility -528626.130 0.269 -94993.340 0.214 ΔHY Bond Volatility 633910.390 0.166 33231.020 0.348 Δ3M LIBOR 2868.600 0.232 -721.820 0.161 Number of Observations 687 687 R2 0.472 0.031 Post-Volcker vs Crisis 0.000 0.000 Post-Volcker vs Post-crisis 0.000 0.068 Post-Volcker vs Post-Dodd Frank 0.000 0.086 41

Table 10: Capital Commitment by Volcker Affected Dealers: The Effect of CCAR Testing This table analyzes how CCAR regulations affect capital commitment among Volcker affect dealers. During the one-month following a bond’s downgrade, we first calculate for each dealer, the absolute value of a dealer’s accumulated principal buy volume and accumulated principal sell volume at the time of each of the dealer’s trades in the downgraded bond (in thousands of dollars of face value). We then average the absolute difference between accumulated buys and accumulated sells across trades within the one-month for each dealer, weighting each observation by the time for which the capital is committed. Trades that were not offset prior to day end hence received larger weight in the capital commitment calculation. For each downgrade, we divide the Volcker affected dealers into two groups: those who passed the CCAR testing in both 2014 and 2015, and those either failed or conditionally passed the CCAR test in at least one year. We then aggregate each dealer’s capital commitment measure within each dealer group, and use them as the dependent variables in Columns I and II. The primary independent variables of interest are dummy variables for Crisis, Post-crisis, Post-Dodd Frank, and Post-Volcker. Detailed definitions of all of the independent variables are provided in Table 3. Since bonds issued by the same firm are usually downgraded at the same time, we cluster the standard errors at the firm level. 42

I. Volcker Affected II. Volcker Affected Dealers who Dealers who Passed the Failed/Conditionally CCAR Test Passed the CCAR Test Estimate p-value Estimate p-value Intercept 21561.340 0.065 23872.770 0.027 Crisis Period -4255.380 0.022 -5114.490 0.001 Post-crisis Period -7449.130 0.000 -4186.190 0.008 Post-Dodd Frank Period -6207.900 0.002 -7171.690 0.000 Post-Volcker Period -10732.840 0.000 -9816.140 0.000 Previous Rating 368.070 0.382 -1240.680 0.138 ΔRating 808.620 0.035 588.190 0.039 Log Age -3098.010 0.001 -2670.770 0.000 Log Time to Maturity 2146.430 0.014 2300.440 0.000 Log Amount Outstanding 12345.580 0.000 4428.690 0.000 SP500 Index Return 45166.660 0.000 10757.080 0.153 IV Bond Index Return 11309.940 0.415 33251.730 0.185 HY Bond Index Return -104937.140 0.000 -46102.670 0.019 ΔVIX -481.870 0.004 -101.680 0.277 ΔIV Bond Volatility 153290.210 0.393 -636766.770 0.071 ΔHY Bond Volatility 279028.120 0.239 377606.390 0.118 Δ3M LIBOR 3094.440 0.107 -336.950 0.433 Number of Observations 683 683 R2 0.420 0.369 Post-Volcker vs Crisis 0.002 0.000 Post-Volcker vs Post-crisis 0.020 0.000 Post-Volcker vs Post- Dodd Frank 0.003 0.007 43

References Amihud, Y. (2002). Illiquidity and stock returns: cross-section and time-series effects. Journal of Financial Markets 5, 31-56. Bao, J., J. Pan, and J. Wang (2011). The Illiquidity of Corporate Bonds. Journal of Finance 66, 911 – 946. Bessembinder, H., S. Jacobsen, W. Maxwell, and K. Venkataraman (2016). Capital Commitment and Illiquidity in Corporate Bonds. Working Paper, Arizona State University and Southern Methodist University. Bessembinder, H., K. M. Kahle, W. F. Maxwell, and D. Xu (2009). Measuring Abnormal Bond Performance. Review of Financial Studies 22, 4219 – 4258. Chernobai, A., A. Ozdagli, and J. Wang (2016). Business Complexity and Risk Management: Evidence from Operational Risk Events in U.S. Bank Holding Companies. Working Paper, Syracuse and Federal Reserve Bank of Boston. Deutsche Banc Research (2016). Surviving less liquid markets. March 17, 2016. Dick-Nielsen, J., P. Feldhutter, and D. Lando (2012). Corporate bond liquidity before and after the onset of the subprime crisis. Journal of Financial Economics 103, 471 – 492. Dick-Nielsen J. and M. Rossi (2016). The Cost of Immediacy for Corporate Bonds. Working Paper, Copenhagen Business School and Texas A&M University. Duffie, D. (2012). Market-making Under the Proposed Volcker Rule. Working Paper, Stanford University. Edwards, A. K., L. E. Harris, and M. S. Piwowar (2007). Corporate Bond Market Transaction Costs and Transparency. Journal of Finance 62, 1421 – 1451. Ellul, A., C. Jotikasthira, and C. T. Lundblad (2011). Regulatory pressure and fire sales in the corporate bond market. Journal of Financial Economics 101, 596 – 620. Federal Register (2011). Prohibitions and Restrictions on Proprietary Trading and Certain Interests in, and Relationships With, Hedge Funds and Private Equity Funds. Vol. 76, No. 215, November 7, 2011. Federal Register (2014). Prohibitions and Restrictions on Proprietary Trading and Certain Interests in, and Relationships With, Hedge Funds and Private Equity Funds. Vol. 79, No. 21, January 31, 2014. 44

Federal Reserve Board (2016). Volcker Rule: Frequently Asked Questions, http://www.federalreserve.gov/bankinforeg/volcker-rule/faq.htm, March 4, 2016. Financial Stability Oversight Council (2011). Study & Recommendations on Prohibitions on Proprietary Trading & Certain Relationships with Hedge Funds & Private Equity Funds, January 2011. Friewald, N., R. Jankowitsch, and M. Subrahmanyam (2012). Illiquidity or credit deterioration: A study of liquidity in the U.S. corporate bond market during financial crises. Journal of Financial Economics 105, 18 – 36. Harris, L. (2015). Transactions Costs, Trade Throughs, and Riskless Principal Trading in Corporate Bond Markets. Working Paper, USC. Kyle, A.S. (1985). Continuous Auctions and Insider Trading. Econometrica 53, 1315-1335. Mahanti, S., A. Nahikkar, M. Subrahmanyam, G. Chacko, and G. Mallik (2008). Latent liquidity: A new measure of liquidity, with an application to corporate bonds. Journal of Financial Economics 88, 272 – 298. Oliver Wyman and SIFMA (2011). The Volcker Rule restrictions on proprietary trading: Implications for the US corporate bond market, December 2011. Powell, J. H. (2016). Trends in Fixed-Income Markets. Testimony before the Subcommittee on Securities, Insurance, and Investment, and Subcommittee on Economic Policy, Committee on Banking, Housing and Urban Affairs, U.S. Senate, Washington, D.C., April 14, 2016. Ronen, T., and X. Zhou (2013). Trade and information in the corporate bond market. Journal of Financial Markets 16, 61-103. Trebbi, F. and K. Xiao (2015). Regulation and Market Liquidity. Working Paper, University of British Columbia. 45