POLICY INTERVENTIONS, LIQUIDITY, AND CLIENTELE EFFECTS IN THE CHINESE CORPORATE CREDIT BOND MARKET*

JINGYUAN MO

MARTI G. SUBRAHMANYAM

We study the impact of policy interventions on credit bond liquidity and

pricing in the interbank and exchange markets in China during the period of 2006-

2017. First, we introduce a novel measure of liquidity and find that the levels of

liquidity, measured by this proxy, vary across different categories of credit bonds.

Second, we show that government policies exerted a profound influence on

liquidity during the various phases of international and domestic liberalization.

Liquidity effects responded strongly to the relaxation of quota limits and to the

simplification of approval procedures on foreign investment into the interbank

market, while, surprisingly, not to the two crackdowns of illegal trading activities.

The break dates in liquidity, endogenously determined from a structural break test,

also correspond to the dates of several major policy announcements. We find that,

in general, liquidity effects became more pronounced as more foreign investment

took place into the interbank market and during more stressful market conditions.

Third, we find that, on average, credit bonds in the exchange market have higher

yield spreads than those in the interbank market, which we attribute to cross-market

clientele effects. Finally, within each market, enterprise bond investors, on average,

earn higher yield spreads than do corporate bond investors, which we attribute to

within-market clientele effects. Keywords: bond markets; yield spread; liquidity; policy intervention; clientele effects; China. JEL classification: G01, G12, G15, G18

[Preliminary! Please do not circulate!]

1. INTRODUCTION

Chinas financial sector is dominated by the state-controlled banking industry and the government-backed securities markets, and is, therefore, fundamentally different from the free economy structure of the financial sector in the U.S., Western Europe, and other large market economies. In spite of the large size of its economy and its clout in global trade, aggregate marketable financial assets account for only 15% of total wealth holdings in China in 2018, while the corresponding figure for the U.S. is approximately 65%.4 At the end of 2017, Chinese marketable financial assets had a total market capitalization of 20.8 trillion U.S. dollars, with the proportion of the three main constituent financial assets being 57%, 37% and 6%, for bonds, stocks and mutual funds, respectively.5 The fact that the bond market is larger than the stock market in China should not be surprising, as the fixed income market is largely dominated by debt securities issued by the central government and various local government entities, including state-owned enterprises (SOEs).6 While the equity market in China has approximately doubled in market capitalization over the past decade, the Chinese fixed income market has grown almost six-fold during the same period. Market analysts forecast that the size of Chinas fixed income market could potentially become quadruple that of its equity market by 2025.7 These developments have made the Chinese bond markets the dominant segment of the domestic financial market, even while remaining much more under-researched than the Chinese stock market. Our contribution here is an attempt to address this lacuna.

All corporate credit bonds in China are traded in the secondary market in two major venues: the interbank market and the exchange market. Due to the severe segmentation of the two markets in China, the extent to which each market is open to foreign investment differs and varies over time. Between the two trading venues, the exchange market is more liberalized, even though it accounts for only a small percentage of both market capitalization and trading volume of Chinese bonds. On the other hand, the more mainstream interbank market has experienced a liberalization process at a glacial pace. Despite the announcement of a series of liberalization policies since 2002, such as the simplification of approval procedures and the gradual augmentation of investment quota limits, the percentage of foreign investors holdings in the interbank bond market was merely 1.1% in June 2017. A year after the initiation of the Bond Connect trading platform in July 2017, this figure increased to 1.9%, although remaining quite small. The fact that the exchange market was opened to foreign investment in the early 2000s, while the interbank market was not, provides a perfect natural experimental setting to investigate whether the liberalization process in the interbank market, though slow, has significantly affected the levels of liquidity, as well as the pricing of liquidity and clientele effects into bond yield spreads, in the Chinese corporate credit bond market.

7

Over the past two decades, Chinese financial markets have received increasing attention from investors globally, as China attempts to integrate its financial system with the rest of the world. As the domestic Chinese economy grew rapidly in recent years, even while it is also gradually opened up to foreign investors, the ability of the government to control and to stabilize the financial markets using targeted measures is bound to diminish. It is, thus, of great importance to study how government policies affected the credit bond market in the past, from which valuable lessons can be learned to forecast the impact of future policy announcements on the Chinese credit bond market, on the broader Chinese bond market, or even on the Chinese economy as a whole.

In this paper, we utilize a unique dataset, covering a 12-year period between January 2006 and December 2017, obtained from the China Foreign Exchange Trade System (CFETS), an arm of the Peoples Bank of China (PBOC), and employ a wide range of liquidity measures to examine the impact of liquidity effects on the pricing of credit bonds over different phases of liberalization and economic conditions. Financial market shocks, culminating in the global financial crisis, as well as the Eurozone sovereign debt crisis, have also triggered dramatic turbulence in financial markets around the world. Along with other major economies, China adopted a stimulus program during the 2008 global financial crisis to revive its economy, with a particular focus on the fixed income market, which was hitherto often disproportionally affected during economic downturns. It is, thus, essential to quantify the magnitude of changes in liquidity effects during recessions and crisis periods. In addition, bond markets in China are well-known to be largely segmented across its two actively trading venues. Due to the structural differences along various dimensions between the two markets, such as the eligibility of foreign investors to participate, trading mechanisms, trading rules, etc., the same type of credit bond may exhibit significant variations in prices across the two markets, even after controlling for the independent variables that are well known to affect yields to maturity. Therefore, it is also of interest to examine whether such a gap in yield spreads, which we attribute to cross-market clientele effects, exists; and, if yes, whether this yield spread differential changes across different phases of liberalization, and under different economic conditions. In addition to investigating the cross-market variation of clientele effects, it is also worthwhile to examine how the yield spread differential between different types of bonds within the same market, which we attribute to within-market clientele effects, shifts across different phases of liberalization and economic conditions.

Our findings are novel for a variety of reasons. First, we find that the levels of liquidity, measured by a wide range of liquidity proxies, of the different categories of bonds are not equal across the two trading venues. Second, we show that government policies exerted a profound influence on the levels of liquidity during the various phases of international liberalization in the interbank market and domestic liberalization in the exchange market. In particular, we find that liquidity improved dramatically in the interbank market, as foreign investments grew in the past decade, while we fail to observe a corresponding improvement of liquidity in the exchange market that is attributable to foreign participants, even though it was opened to foreign investors earlier than the interbank market, and had a longer period of foreign investment participation.

The remainder of this paper is organized as follows. Section 2 presents a review of the evolution of the Chinese corporate credit bond market. Section 3 discusses the academic literature on topics related to liquidity of corporate bond markets, including the sparse evidence on the Chinese markets. Section 4 motivates and states the testable hypotheses in this paper. Section 5 explains, in detail, the composition of our dataset and reports the descriptive statistics of the variables. Section 6 introduces our novel liquidity measure, and outlines the research methodology that we employ in our analysis. Section 7 presents our discussion of the empirical results on the levels of liquidity, and the liquidity effects in bond pricing. Section 8 presents a parallel discussion of the empirical results relating to cross-market and within-market clientele effects. Section 9 concludes the paper.

2. CHINESE CORPORATE CREDIT BOND MARKETS

In this section, we provide a brief description of the two active trading venues for bonds in China, the interbank market and the exchange market, as well as the four types of corporate credit bonds traded across the two markets.8 We also discuss the three different phases of international liberalization between 2006 and 2017 and the two crackdowns of agent-holding transactions in the interbank market in 2013 and 2016 and the domestic liberalization in the exchange market in 2015, before we present the liquidity proxies that we use to measure liquidity in the largely illiquid Chinese corporate credit bond market.9

2.A The Interbank Market and The Exchange Market

The two trading venues for Chinese corporate credit bonds differ from each other in various aspects, of which the following six are the major ones: regulatory structure, market participants, trading instruments, trading mechanism, trading rules, and collateral policy.

First, the interbank market is regulated by the Peoples Bank of China (PBOC) and uses the China Foreign Exchange Trade System (CFETS) as its trading platform, while the exchange market is regulated by the China Securities Regulatory Commission (CSRC) and uses the Shanghai and Shenzhen Stock Exchanges as trading platforms. Second, market participants in the interbank market are mostly institutional investors, such as banks, securities companies, insurance firms, mutual funds and large financial institutions, while investors in the exchange market are mostly household investors. Institutional investors can invest in both markets, while household investors can only invest in the exchange market. As a result, the interbank market has a much higher level of demand for liquidity than the exchange market, leading to a lower liquidity premium. Controlling for all other variables,

8 9

we expect, therefore, to observe a potentially lower volume-weighted yield spread in the interbank market than in the exchange market. Third, the two markets also differ in terms of the particular instruments traded. Both enterprise bonds and corporate bonds are traded in the exchange market, while only enterprise bonds are traded in the interbank market.10 Fourth, investors in the exchange market trade through the clearinghouse of the exchange, and hence have little counter-party risk. The interbank market is a telephone market, in an over-the-counter setting, in which traders call several counterparties when they plan to sell or buy bonds, as the large trading volume induces them to search for the best price. Fifth, transactions in the interbank market are carried out based on inquiries, which involve market makers who provide bilateral quotes. In the exchange market, however, bond transactions are matched through a centralized computer system, similar to stocks; transactions are settled based on clean prices. Lastly, the interbank market has strict restrictions on instruments that can be pledged as collateral. During normal times, both government bonds and corporate credit bonds with an AAA rating can be pledged as collateral, while during crises, only interest rate securities are accepted as collateral.11 In the exchange market, however, most high-quality credit bonds can be pledged as collateral at a certain discount rate (haircut) at all times.12 Of course, the haircut would vary over time and across bonds.

2.B A Taxonomy of Corporate Credit Bonds in China

In this paper, we study all four types of corporate credit bonds in China: enterprise bonds and medium-term notes in the interbank market and enterprise bonds and corporate bonds in the exchange market. It should be noted that both corporate bonds and enterprise bonds would be categorized as corporate bonds in the U.S. and the European bond markets. In China, they have different names mainly because of their respective issuers and regulatory bodies.

Although both types of bonds are issued by entities with corporate credit, corporate bonds are mostly issued by limited liability, joint-stock companies (listed and unlisted), while enterprise bonds are mostly issued by state-owned enterprises and state-holding companies, both at the national and provincial level. The two types of bonds also differ according to their financing purposes: enterprise bonds are issued to finance projects approved by the central government or local governments, such as infrastructure projects and government projects, while the issuance of corporate bonds enjoys more flexibility, in terms of debt payment, long-term investment, and projects that benefit the future growth of the corporations. In addition, enterprise bonds are traded in both the interbank market and the exchange market, while corporate bonds are traded only on the exchange. As a result, issuers can choose to issue enterprise bonds in both markets, and approximately 90% of enterprise bonds thus became dual-listed in both markets since 2005, the year that the exchange market started to admit enterprise bond issuance. However, enterprise bonds issued by the same entity cannot be

traded across the two markets; they are assigned different trading codes across the two markets and can be regarded as two distinct bonds, even if their terms are identical. This gives us a perfect setup to identify the pricing discrepancy across the two segmented venues. The counterpart to exchange market corporate bonds in the interbank market, debt-financing instruments of non-financial corporates (DFINFCs), are further classified into two types of instruments: commercial paper and medium-term notes. Commercial paper is issued by non-financial firms and functions as a short-term financing tool, similar to commercial paper in the U.S., with a maturity of less than one year. Due to the short-term feature of the maturity of commercial paper, we only consider medium-term notes as the interbank corporate counterpart in this paper.

[Insert Figure 1 Here]

2.C Liberalization of the Chinese Bond Markets

Over the past decade, China had gradually opened its bond markets to international institutional investors. However, the past decade also witnessed a series of major and minor announcements regarding how regulators planned to reform the bond markets, including granting access to more qualified foreign investors, simplifying approval procedures, and increasing foreign investment quota limits. It is, thus, important to identify which announcements had the greatest impact on the changes in the fundamental characteristics of the bond market. Figure 1 presents the timeline of the major policy changes that will be discussed and studied in detail in this paper.13 In particular, we define the following three subperiods of liberalization stages in the interbank market:

Phase 1 limited foreign institutional investors participation (prior to Mar. 2012) Phase 2 foreign central banks participation (Mar. 2012 ~ Jul. 2015) Phase 3 further expansion to most foreign institutional investors (Jul. 2015 ~ Dec. 2017)

Despite the sequence of regulatory changes relating to foreign investors in Figure 1, only 1.1% of bonds in the interbank market were held by them by June 2017. In July 2017, however, Bond Connect, a trading platform that allows offshore investors and mainland China investors to invest in each others bond markets, was established in Hong Kong. The northbound link of this channel, allowing offshore investors to invest in the domestic China interbank bond market, was opened on July 3, 2017.14 One year later, we observe an increase of foreign investors holdings to 1.9%, a substantial increase of 70%, compared to the original percentage of holdings.

Despite being more open to foreign investments, the exchange market accounts for a very small percentage of the trading volume, and is usually ignored in academic research in China. However, a sequence of regulatory changes in the interbank bond market also pushed forward the development of the exchange market at the same time. For example, in the case of corporate bonds, in terms of both issue amount and trading volume, the interbank market

accounts for more than 90% of the entire bond market. However, The Corporate Bond Issuance and Trading Regulations, File No. 113, issued in January 2015, largely increased the issuance of corporate bonds in the exchange market. Moreover, not only did File No. 113 lower restrictions on issuers, offering methods and offering period of corporate bonds, it also simplified the issuance approval procedure. More importantly, the change of regulations allowed many unlisted companies to issue corporate bonds in the exchange market, which substantially lowered their financing cost. We take these structural changes into account by defining the following two subperiods in the exchange market, separated by May 2015.15 Given these developments, it is also necessary to incorporate the exchange market into our analysis. Despite the governments announcement that unlisted firms would be allowed to issue corporate bonds in January 2015, the first bond issue by unlisted firms did not occur until May 2015, which defines our break date.16

Prior to 2015 only listed companies could issue corporate bonds (before May 2015) After 2015 both listed and unlisted firms could issue corporate bonds (May 2015 ~ Dec. 2017)

2.D The Crackdown of Agent-holding Transactions

While the government is aiming to develop the Chinese corporate credit bond market, some financial institutions, inevitably, make use of legal loopholes inherent in the guidelines to meet their own objectives. One typical example is that banks and security firms all have the incentive to boost trading volume, one of the most important grading criteria in the institutions regulatory ratings, in the assessment of the qualification to underwrite sovereign bonds as primary dealers. This has led to the emergence of agent-holding activities since 2010.17 Agent-holding transactions, which consist of transferring ownership of a security to another party, similar to a repurchase agreement-like transaction, are strictly not illegal when used as a normal means of conducting bank business; however, they would be prohibited if investors use them to cover up bad bond investments or to satisfy private uses, not within the qualified categories of investment. The precise rules constitute a fine line between what are considered to be normal business activities, and what may be deemed to be illegal, in which case the authorities may come down heavily on the offenders.

The interbank bond market underwent two major crackdowns of such agent-holding practices in the past decade, in 2013 and 2016, respectively. The 2013 crackdown began in early 2013, which led to a drastic drop in trading volume of almost 90%, as compared to that prior to the announcement of the crackdown. The situation stabilized thereafter, before agent-holding transactions re-emerged in 2015, which led to a second crackdown in 2016, leading to, once again, a 70% decline in trading volume. It seems reasonable to conjecture that these crackdowns would exert a negative impact on incentives to trade and, hence, on market liquidity. We will, therefore, examine the impact of agent-holding crackdowns on the

levels of liquidity and the extent to which liquidity is priced in bond yield spreads for all four categories of credit bonds in Section 6.

2.E Liquidity in the Chinese Bond Markets

To measure liquidity in the extremely illiquid Chinese corporate credit bond market, we adopt a wide range of liquidity proxies proposed in the paper by Friewald, Jankowitsch and Subrahmanyam (2014). We include bond characteristics variables and trading activity variables that have been used as liquidity proxies by many researchers in the literature. The bond characteristic variables that we use include age, time to maturity, coupon rate, duration, convexity, and rating dummies. The trading activity variables that we use include the trading interval, the zero-return measure, the zero-trade measure, and daily trading volume. We also employ several liquidity measures proposed by previous researchers, including the Amihud ratio in Amihud (2002), the price dispersion measure in Jankowitsch, Nashikkar and Subrahmanyam (2011), the high-low spread estimator in Corwin and Schultz (2012), the interquartile range estimator in Han and Zhou (2007), and the CHL estimator in Abdi and Ronaldo (2017). Detailed definitions of these liquidity proxies are provided in Appendix C.

3. LITERATURE SURVEY There is a vast literature on liquidity effects in bond pricing, especially in the U.S. market and, to a lesser extent, the European market. Amihud and Mendelson (1986) are the first to study the impact of illiquidity (measured by the bid-ask spread) on asset-pricing by analyzing a model of investors with different trading horizons. Their study shows that higher returns are required on stocks with higher bid-ask spreads and, thus, higher liquidity (lower bid-ask spreads) corresponds to higher stock prices. Testing their earlier model, Amihud and Mendelson (1991) demonstrate empirically that a negative relationship exists between illiquidity and asset pricing, and argue for the economic importance of liquidity even in an efficient market. Amihud (2002) also suggests that illiquidity premium may partly explain the expected stock return, and that stock returns and contemporaneous unexpected illiquidity are negatively correlated. This work was expanded and elaborated upon by numerous researchers in a variety of stock markets around the world.18

Since the inception of the Trade Reporting and Compliance Engine (TRACE) database disseminated by the Financial Industry Regulatory Authority (FINRA) for the U.S. corporate bond market in 2002, researchers began to focus more on empirically analyzing liquidity effects in the corporate bond market. A growing number of studies have emerged as researchers are now able to analyze bond price information following post-trade transparency, due to the establishment of the TRACE database. Edwards, Harris and Piwowar (2007) are among the first to document that bid-ask spreads reduced significantly

after transaction prices became available. They further provide evidence that bonds that are more recently issued have higher credit ratings, less time-to-maturity and more liquidity, compared to bonds that are otherwise. Mahanti et al. (2008) propose a novel liquidity measure, known as latent liquidity, defined as the weighted average of bond turnovers of investors, and find that the new measure is suitable for bond markets with sparse transactions. Bao, Pan and Wang (2011) also propose a liquidity measure, the gamma estimator, and show that a large variation of yield spreads can be explained by movements in bond prices, a transitory component that the gamma estimator is able to extract.

In the over-the-counter (OTC) market environment, Jankowitsch, Nashikkar and Subrahmanyam (2011) model price dispersion effects in OTC markets, and demonstrate that transaction prices deviate from the market-wide valuation in the presence of inventory and search costs. They also document that their measure fits the data well in the U.S. corporate bond market. Duffie, Garleanu and Pedersen (2007) also show that illiquidity discounts are higher when it is difficult to find counterparties, sellers have less bargaining power, and there is a paucity of qualified owners. The unique dataset that we use in this paper is broadly similar to a restricted version of the TRACE dataset in the U.S. market, which we will discuss in detail below. In particular, our dataset includes a summary of the transaction data in both the exchange market and the interbank market, which enables us to compare liquidity effects in both markets.

Liquidity has always been of primary concern during periods of financial distress in markets, especially in the case of bond markets, which are not that liquid even in normal times. It is, therefore, of interest to study how liquidity effects change in times of financial crisis. Some of the first papers to investigate this after the 2008 financial crisis are by Bao, Pan and Wang (2011), Friewald, Jankowitsch and Subrahmanyam (2012), and Dick-Nielsen, Feldhutter and Lando (2012). These authors show that trading costs peak during financial crises and largely affect yield spreads, especially for bonds with low credit quality. Friewald, Jankowitsch and Subrahmanyam (2012) also find that liquidity effects account for a fairly large proportion of the total variation of corporate bond yield spread changes, and that the impact of liquidity measures on bond prices is much larger in times of financial distress. In this paper, we show that a similar argument holds in the Chinese corporate credit bond market: liquidity effects tend to be more pronounced during crisis periods. We also add depth to the broad statement relating to such effects by highlighting their differential effects in the two trading venues and in different segments of the market.

Our paper also examines the impact of government policies on the levels of liquidity in the Chinese corporate credit bond market, a topic that has been studied in the U.S. context. For example, Duffie (2012) shows that restrictions imposed by regulators through the Volcker rule on the U.S. corporate bond market significantly affected the magnitude of bid-ask spreads. Bao, OHara and Zhou (2016) find that stressed bonds become more illiquid as market liquidity provisions by dealers affected by the inception of the Volcker rule weakened. The difference between this literature and our paper is that, rather than examining this issue from a micro-economic perspective, we investigate the issue in the Chinese bond markets from a broader macro-economic perspective. We do so by studying how liquidity effects

change as the Chinese government gradually lifted certain restrictions and loosened the rules governing foreign institutional investor participation in the bond markets.

As investors became increasingly aware of the importance of liquidity in the U.S. and European bond markets, it captured the attention of not only academic researchers, but also industry practitioners and policy makers, including those involved with the Chinese bond markets, around the issue of liquidity. However, very little rigorous research has been conducted thus far in the Chinese bond markets; most of the existing studies remain qualitative. This is understandable given that the low data quality, and the low frequency of bond transactions in the secondary markets, inhibit the application of rigorous econometric methodologies. In addition, the extant literature on the Chinese bond markets does not discuss data selection or filtering procedures. This is an important omission given that data filtering is essential for cleaning the data to reaching robust conclusions, especially since data quality is a challenge with the Chinese bond market data. All of these concerns, however, will be addressed in our paper.

Despite this handicap due to the non-availability of reliable data, a few recent studies provide a good qualitative description of the Chinese bond markets. At a general level, Amstad and He (2018) provide a comprehensive introduction to Chinese bond markets, and Hu, Pan and Wang (2018) present an empirical overview of the Chinese capital market. A few other papers have also examined the pricing of corporate bonds with respect to liquidity risk factors. Some early studies focus on the stock market and bond markets liquidity spillover effects. Wang and Wen (2010) provide evidence of the liquidity linkage between stocks and government bonds in China. They show that macroeconomic shocks are transmitted from one market to another. Several other papers study more directly the liquidity effects on corporate bond pricing by employing liquidity risk factors. He and Shao (2012) investigate the effect of liquidity risk on Chinas corporate bond yield spread, covering the period from April 2007 to September 2009. This empirical study reveals a negative correlation between liquidity and corporate bond yield spreads in the exchange market, with the correlation significantly enhanced during the subprime crisis. However, the study only uses a variation of the Amihud measure, instead of a wide range of liquidity proxies. Another paper is by Wang and Wen (2016), which compares the effect of liquidity risks on the pricing of corporate bonds and enterprise bonds. This study does include three liquidity measures as explanatory variables, in addition to trading activity variables, such as trading volume and trading interval, which are well known to affect the bond yield spread, in the panel regression analysis, but does not find any significant statistical relationship between liquidity and bond yields.

Our study employs a wide range of liquidity measures that have been proven to be appropriate and useful for the U.S. and European bond markets. Friewald, Jankowitsch and Subrahmanyam (2012) are the first to apply a range of commonly utilized liquidity measures to study the yield spreads of corporate bonds. Schestag, Schuster and Uhrig-Homburg (2016) also find that most low-frequency liquidity proxies measure transaction costs equally well, and that high-frequency measures are also strongly correlated. We focus on the low-frequency measures in our research in the paper, due to the constraints of data availability and quality. Our liquidity measures include the Amihud measure in Amihud and Mendelson

(1986), the high-low spread estimator in Corwin and Schultz (2012), the price dispersion measure in Jankowitsch, Nashikkar and Subrahmanyam (2011), the CHL measure in Abdi and Ronaldo (2011), and the inter-quartile bid-ask spread estimator in Han and Zhou (2007).

It should be noted that our paper can be distinguished from the prior literature on Chinese corporate bond markets along several important dimensions. First, we employ a much larger transaction dataset for all four types of corporate credit bonds in the two actively traded venues in China than any previous paper on Chinese bond markets, which typically focus only on a certain sub-segment of these markets. Second, our paper is related to the literature on the liquidity effects on bond pricing. Most of the existent liquidity measures are tested in the context of well-developed corporate and sovereign bond markets, such as in the U.S. and Europe, while we test them in the corporate credit bond market in China, the second largest economy in the world. We also develop a novel liquidity measure, using a set of widely used liquidity proxies, which adapts better to the Chinese bond markets with extremely low trading frequency and high data noise. Third, our dataset spans a total of 12 years, covering several major policy announcements by the Chinese government, especially regarding foreign investors participation in the interbank market, and on domestic regulations in both the interbank and exchange markets. The development of the regulations allows us to identify the changes in the levels of liquidity and the evolution of liquidity effects and clientele effects, while the power of our set of independent variables helps us in explaining bond yield spreads over the different phases of liberalization and varying economic conditions.

4. HYPOTHESES

In this section, we present an overview of the research questions that we pose, and the hypotheses that we test, in our empirical research. In particular, we examine the validity of specific arguments regarding the levels of liquidity of different categories of credit bonds in both markets, and the impact of these liquidity effects, as well as clientele effects, on bond yield spreads.

Hypothesis 1: Liquidity effects are priced significantly in Chinese corporate credit bond yield spreads. Liberalization, including allowing foreign investors to participate, simplifying approval procedures, increasing limits on investment quota, and lowering requirements on bond issuers, improved the levels of liquidity and the extent to which liquidity effects are priced into yield spreads in the Chinese corporate credit bond market.

Corporate credit bonds in China are traded across two trading venues, the interbank market and the exchange market. The exchange market, which was opened much earlier, still accounts for less than 5% of the aggregate trading volume in the secondary markets, while the interbank market has been experiencing liberalization at a more gradual pace. Even after the interbank bond market was opened to foreign investors, its constraints, such as complex

approval procedures and limited investment quotas, have largely hindered the participation of foreign investors. Even though the extent of foreign investors participation may be small compared to the developed markets, it would still be interesting to investigate whether the slow pace of liberalization and the low percentage of foreign investment have significantly affected the levels of liquidity in the interbank market. We conjecture that, with liberalization, the removal of complex approval procedures and the lifting of limits on the foreign investment quota have caused an improvement in liquidity in the corporate credit bond market, despite the low aggregate level of foreign investment. In addition, it has been documented in the literature that liquidity is an important priced factor in the U.S. and European corporate bond markets. While some existing research papers published in Chinese journals have documented the influence of some widely-used liquidity measures on the pricing of corporate credit bonds in the Chinese bond markets, most of these analyses remain preliminary, and often only at the general descriptive level. Furthermore, no paper so far has studied the evolution of liquidity as a priced factor for explaining the bond yield spread over different phases of liberalization. We intend to test this issue rigorously by conjecturing that liquidity is priced significantly in the corporate credit bond markets. We also speculate that the degree to which liquidity is priced into bond yield spreads varies during different time periods and market conditions.

Hypothesis 2: Across the two markets, there exists a consistent yield spread differential for the same category of credit bond (enterprise or corporate), due to cross-market clientele effects. The improvement in liquidity over time in the interbank market, as a consequence of liberalization, leads to a decrease in this yield spread differential.

The vast difference in total trading volume across the two markets is due to the heterogeneity in market participants and the different trading behaviors between institutional investors and household investors. Large financial institutions, such as commercial banks, hedge funds, and insurance companies, mostly invest in the interbank market, even though they can also access the exchange market. On the other hand, household investors can only invest in the exchange market. Moreover, as mentioned in Chen et al. (2018), household investors mostly speculate in the exchange market, while sophisticated institutional investors in the interbank market trade whenever they have to. This leads to much more trading activities in the exchange market, while the trading volume remain negligible as compared to that in the interbank market. As a result, the interbank market is deeper but lacks immediacy, as compared to the exchange market. The heterogeneity in market participants and the different trading behaviors of market participants across the two markets motivate the first conjecture in Hypothesis 2. In addition, the liberalization of the Chinese corporate credit bond market occurred in stages. Since the exchange market was liberalized first, bonds traded there also experienced an improvement in liquidity earlier. We further conjecture that, prior to the opening of the interbank market to foreign investors, credit bonds in the interbank market had a positive yield spread differential over those in the exchange market, due to their limited exposure to the foreign investor clientele. Later on, as the interbank market was liberalized, we anticipate that the enhancement in liquidity would lead to a decline in the yield spread differential, a time-series effect that is captured by the

second conjecture in Hypothesis 2. This provides us with a perfect natural experimental setting to examine the evolution of this cross-market clientele effect. We expect the yield spread differential to shrink in magnitude, and could even reverse in sign as the interbank market becomes more open.

Hypothesis 3: Within each of the two venues, there exists a consistent yield spread differential across bond categories (enterprise and corporate), due to within-market clientele effects. This yield spread differential is largely susceptible to policy changes within each market, and varies under different economic conditions.

The fairly constant composition of institutional market participants in the interbank market (commercial banks, insurance companies, hedge funds, etc.) and the close-to-constant presence of household/individual investors in the exchange market motivate Hypothesis 3. Controlling for the variation in foreign investment exposure (i.e., the foreign institutional investor clientele), we investigate the variation in the yield spread differential between enterprise bonds and corporate bonds within each market. We investigate whether this variation is due to market participants strict preferences for one type of credit bond over another, and whether this yield spread gap persists or varies over time due to shifts in preferences, and hence net demand, during different phases of liberalization and a changing economic environment.

5. DATA DESCRIPTION In this section, we present the unique dataset that we have assembled for this paper. Our dataset covers virtually all end-of-day transactions from all types of bonds in both the interbank market and the exchange market, including government bonds, local-government bonds, enterprise bonds, corporate bonds, financial bonds, etc. We focus on all corporate credit bonds that are actively traded in the two venues: enterprise bonds, corporate bonds, and medium-term notes.19 We also use the yields to maturity of government bonds in both the exchange market and the interbank market to bootstrap the zero-coupon rates (de-couponed rates) in each market. Recall that the source of the dataset is CFETS, an arm of the PBOC, which functions in a manner similar to the FINRA-developed platform, TRACE, which has been instrumental in facilitating the mandatory reporting of secondary market transactions in the U.S.20

5.A Data Regulation and Composition

Our dataset contains two subsets and covers a period of 12 years, from January 2006 to December 2017. The beginning date of the transaction data for each type of bond may vary,

depending on the date that transactions associated with that type of bond were first required to be reported to CFETS. The first subset contains fundamental bond characteristic variables, which include trading date, issue date, time-to-maturity (remaining), duration, convexity, coupon rate, bond ratings, rating agencies, type of interest rate (floating or fixed), trading market, and special clauses. Many of these characteristics are frequently used as simple liquidity proxies in the literature. The second subset contains detailed information about all end-of-day bond transactions, including trading code, daily trading volume, highest and lowest traded prices (clean and dirty), opening and closing prices (clean and dirty), and yield to maturity based on the closing dirty price, all of which are widely utilized in the literature on bond market liquidity.

Due to the data regulations of the PBOC, CFETS is only permitted to provide end-of-day transaction data. In other words, whether a bond is traded once or more than once during a day, it appears only once as a bond-day observation in our data set. As a result, we only know what bonds are traded during a day, but cannot discern how many times each bond was traded on that day. This restriction certainly poses some serious challenges to our study, such as constructing certain liquidity measures, since detailed trading information during a day of each traded bond is not available. However, these are the best data available for the Chinese bond market, and we have to conduct our research in the presence of this constraint. We, therefore, employ a complex filtering procedure to ensure the soundness of our final dataset, which is described in Appendix A.

5.B Completing Missing Data Points of Yield to Maturity

One of the key variables employed in our study is the yield to maturity of each bond. However, in the early years of our dataset between 2007 and 2009, quite a few data points of yield to maturity are missing for all types of bonds. We explicitly compute each missing yield to maturity by following the instructions in the official document issued by the PBOC. As in other bond markets, on each trading day, the dirty price is related to the clean price through the following equation:

dirty price = clean price +

where is the accrued interest and is computed as (

365/ ); is the number of days

between the last coupon payment date and the trading date; is the frequency of coupon payments per year; and is the amount of the next coupon payment. The dirty price is related to the yield to maturity through the following equation:

=

(1 + / )1+

(365/)

+ /

(1 + /)+

(365/)

1

=0

where PV is the dirty price; y is the annualized yield to maturity; is the coupon payment per period ; f is the frequency of coupon payments per year; d is the number of days between the trading date and the next coupon payment date; N is the total number of coupon payments; and FV is the notional face value of the bond. The yield to maturity y does not

have a closed form solution, except if the trading date is within the final coupon payment period. In other words, if only one cash flow remains (i.e., = 1), the yield can be found explicitly as:

=365

(

+

1)

Moreover, for enterprise bonds and corporate bonds in the two markets, the volume-weighted average yield in each week is calculated as follows:

=

where is the volume-weighted average yield of bond k in week t; is the yield

corresponding to day k of bond k in week t; and is the volume of day k in week t.

5.C Rating Agencies and Bond Rating Quantification

Similar to bond ratings in the U.S., bonds with higher default risk are given lower ratings, in principle. Bonds in China are rated by six credit rating agencies: China Chengxin Credit Rating Group, Shenzhen United Ratings, Shanghai Brilliance Credit Rating, Dagong Global Credit Rating, Pengyuan Credit Rating, and China Orient Golden Credit Rating. Their credit ratings are similar to those of Standard and Poors (S&P) and range as follows: AAA, AAA-, AA+, AA, AA-, A+, A, A-, BBB+, BBB, BBB-, BB+, BB-, B+, B, B-, CCC, CC, and C.21 In the Chinese bond markets, bonds with ratings below BBB are considered to be speculative-grade bonds, while bonds that have ratings equal to or above BBB are considered as investment-grade bonds. To control for credit risk differentials, Friewald, Jankowitsch and Subrahmanyam (2012) quantify the ratings by creating a measure called Rating Number, being multiples of 1, i.e., AAA = 1, AAA- = 2, AA+ = 3, etc. This, however, is valid if and only if we assume that: (1) all rating agencies have the same rating criteria (i.e., if a bond is given a rating x by rating agency 1, it receives the same rating x from all other rating agencies); and (2) the credit differential between two adjacent ratings is approximately equal. To circumvent this problem, we create an indicator variable for each rating. For instance, an indicator measure for AAA is assigned 1 if the bond is rated AAA, and 0 otherwise.22 This approach certainly gives us the most freedom to map from ratings to explanatory variables, and eases the concern that assigning a specific number to each credit rating may be too restrictive.23

[Insert Figure 2 here]

22

Figure 2 displays graphically the distribution of ratings for the four types of credit bonds in the two markets, after the filtering procedure described in Appendix A. We observe that almost all traded bonds traded have a rating above or equal to AA-. The total number of bonds traded in the interbank market is approximately six times larger than that in the exchange market, which again shows that the former is the dominant market of the two. We also observe a vast decrease in the number of bonds traded in the exchange market, with corporate bonds declining about 20% and enterprise bonds declining nearly 65%, due to early redemption before maturity in the exchange market. These bond transaction records are discarded through the filtering procedure.

5.D Bootstrapping the Zero Rates

In all of the papers related to bond markets in China, researchers have adopted the convention of using the yields to maturity of the Chinese government bonds as a proxy for the risk-free rates of the relevant maturity, in line with industry practice. Since all government bonds in the Chinese bond market have non-zero coupon payments, government bond yields are all coupon-implied, i.e., yields to maturity with coupon payments considered. This, in particular, renders government bond yields as an inappropriate and incorrect proxy of the risk-free rate, due to the associated coupon effects. In this paper, we therefore de-coupon the original coupon-implied government yields to create the respective zero rates (i.e., bootstrapping the zero curves). The resultant zero rates can be more properly used as the correct proxy of the risk-free rates in China. In this paper, we consider both sets of yields the coupon-implied yields to maturity and the zero rates as proxies for the risk-free rates and investigate the differences between the resultant bond yield spreads, as a robustness check.24

To the best of our knowledge, we are the first to point out this issue in the literature related to Chinese financial markets, and are also the first to explicitly compute the zero rates in the interbank market and the exchange market. Not surprisingly, for a given trading day, the correlation between the coupon-implied yield curve and the bootstrapped zero curve is very close to one, as the latter is obtained through de-couponing the former. In the exchange market, after winsorizing the time series at the 95% confidence interval, the correlation lies between 0.905 and 1, with a mean of 0.996. In the interbank market, after winsorizing the correlation time series at the 95% confidence interval, the correlation lies between 0.916 and 1, with a mean of 0.999.

5.E The Nelson-Siegel Three-factor Estimation

The dependent variable in our regression analysis is the yield spread of a particular type of bond, which is defined as the difference between the yield to maturity of that bond and the zero rate corresponding to the same time to maturity on the same trading day. However, in most cases, the two sets of maturities do not exactly match. For instance, a bond

24

that is traded on one trading day may have a time to maturity of 3.22 years, while there is no government bond traded on the same day with exactly the same time to maturity. Without knowing the corresponding zero rate, it would be impossible to calculate the yield spread for that bond on that day. In order to better capture the non-linear relationship between bond yields and time to maturities and to obtain zero rates of all possible maturities on a given trading day, we fit our daily data with a Nelson-Siegel three-factor model, as in Nelson and Siegel (1987) as modified by Diebold and Li (2006). The three-factor estimation for any given maturity is given by:

() = 00 + 11 + 22

where

0 = 1, 1 =1 exp()

, 2 =

1 exp()

exp()

The three factors in the Nelson-Siegel model, the level factor 0, the slope factor 1 and the curvature factor 2, are interpreted as latent dynamic factors. The parameter is the time-dependent decay factor that determines the level of contribution of 0,, 1, and 2. It should be noted that while is often taken to be time-invariant in many previous studies, we allow this decaying factor to be time-varying on a daily basis, similar to what was originally implemented in Nelson and Siegel (1987).

For each type of bond with a total number of N trading days, we generate a N 1000 grid which contains a term structure of zero rates for each trading day, and for all maturities ranging from 0.01 year to 10 years, with an increment of 0.01 year. In this way, we can match a zero rate to the yield of every traded bond on each trading day until maturity.

6. METHODOLOGY

In this section, we present the various explanatory variables that we use in the regression analysis of liquidity. Of the various liquidity proxies discussed in the literature review section, a few that require tick-by-tick data cannot be employed in this paper due to the difficulty in obtaining sufficiently detailed, frequent data for the Chinese bond markets. However, using our dataset, we can still construct many of the widely-utilized liquidity proxies, generally used in the literature. Further, we also use these proxies to develop a novel liquidity measure, the first principal component of these proxies, which we further use to analyze the liquidity of different types of credit bonds in China. Along the lines of Friewald, Jankowitsch and Subrahmanyam (2012), the explanatory variables in this paper are classified into three groups: bond characteristic variables, trading activity variables, and liquidity metric variables. In addition, we discuss our panel regression model and the structural break test used to statistically identify the dates when structural changes are likely to have occurred. We then present the specifications of our panel data regressions to explore the time-series

properties of our dataset. In the next section, we first present the summary statistics of the yields for different maturities, and study, in general, the discrepancy between weekly average yields of bonds across the two markets, followed by empirical evidence from the panel regression models.

6.A Explanatory Variables

The explanatory variables in this paper, classified into three groups, bond characteristic variables, trading activity variables, and liquidity metric variables, are either obtained from the original data set or are constructed using the available information.25

Bond characteristic variables include coupon, duration, maturity, and age. Although these are individually crude indicators of the liquidity levels of bonds, they do provide some intuitive sense of the information conveyed by basic bond features about potential liquidity. In general, we expect bonds with longer maturities to be less liquid because they are often purchased by buy-and-hold investors with medium- to long-term objectives, bonds with higher coupons to be more liquid due to demand from income-seeking investors, and on-the-run (recently issued) bonds to be more liquid than off-the-run bonds. However, these measures are mostly used for cross-sectional comparisons, as many of them exhibit little variation over a short period of time. Trading activity variables include daily trading volume, trading interval, zero-return measure, and zero-trade measure. They provide information about liquidity based on bond transaction details, with higher trading activity generally indicating greater liquidity. In addition to daily trading volume, which is given in the raw data set, we create three additional variables: trading interval, zero-return, and zero-trade. For a particular bond, trading interval is defined as the number of days between two consecutive trades; if the bond is not traded on a given day, its trading interval on that day is assigned NA. The zero-return measure for a bond on a given day is an indicator variable that is assigned one if there is price variation on that day (the highest dirty price is different from the lowest dirty price), and is assigned zero if otherwise. The zero-trade measure for a bond in a week is defined as the number of days in the week that the bond is not traded divided by five, the number of trading days in most weeks. Liquidity measures used in this paper include the modified Amihud ratio, the interquartile range estimator, the price dispersion measure, the high-low spread estimator, and the CHL estimator. Detailed descriptions of these five proxies are given in Appendix C.

6.B Subperiods of Interest

We are interested in how the explanatory power of the independent variables and the statistical significance of the coefficients of these variables change under the different phases of liberalization. In addition to investigating this issue, we compare liquidity effects in the credit bond market during regular market conditions, recession periods and crisis periods, across both markets. We adopt the Organization of Economic Cooperation and Development (OECD) recession indicators for the Chinese economy to define the three recession periods

25

between 2006 and 2017.26 Regular market periods are defined as the complement of the joint recession periods. Between 2006 and 2017, the Chinese economy witnessed two major financial crises, the 2008 global financial crisis and the 2015 Chinese stock market crash, which we further classify as crisis periods. We anticipate liquidity effects to be more pronounced during crisis periods than during recessions, and to be more pronounced during recessions than during regular market conditions.

6.C A Novel Liquidity Measure

We develop a novel liquidity measure, which we term as Liquidity PC, defined as the first principal component of the five liquidity proxies discussed in Appendix C. The reason for us to adopt this measure in the Chinese corporate credit bond market is multi-fold. First, the Chinese bond markets are still in a nascent state, with trading rules and regulations that continue to be refined. This renders some liquidity measures that have been proven to have worked well in more liquid markets to not deliver the expected results in the Chinese bond markets, which is characterized by low trading frequency, despite the large size of the market. Second, the transaction data that we obtain from CFETS are not tick-by-tick, which means that some of the intra-day trading information is not captured by our dataset. This inevitably generates noise in our end-of-day transaction records, even after we aggregate them on a weekly basis. Third, no single liquidity measure is able to capture the overall level of liquidity in a particular bond market, while every single liquidity measure captures a certain dimension of liquidity. In other words, the information that one liquidity measure can reflect in the market is largely limited, especially when the market, such as the Chinese bond markets, is extremely illiquid. It is entirely possible that a subset of liquidity proxies works well to capture liquidity variation for one particular type of credit bond, while another subset of liquidity proxies works better to capture liquidity variation for another type of credit bond. Extracting the first principal component, thus, enables us to obtain as much information from the set of liquidity proxies as possible, while avoiding the inclusion of noisy information when all individual liquidity proxies are included in the model. For the same reasons, we construct another principal component measure for the set of trading activity variables used in this paper, which we term as Trading PC. Before demonstrating the validity of our new measure as a significant pricing factor in bond yield spreads in Section 7, we will examine, at a general level, the levels of liquidity in the Chinese corporate credit bond markets in Section 7.B.

6.D Panel Regression Model on the Levels of Liquidity

We first perform a simple panel regression to analyze how liquidity levels change with time in the Chinese bond market. The dependent variable is the first principal component of liquidity measures, and the independent variable is the number of the week during our sample period. We then investigate whether breaks in the level of liquidity exist in our

sample around important policy announcement dates by performing a structural change breaks test proposed by Andrews (1993) (the sup-F test), based on the following model:

(Liquidity PC) = 0 + 1 Week + (1)

Briefly, this test corresponds roughly to a Chow (1960) test. However, while in the Chow test, the date of structural change break is pre-specified, the sup-F test leaves the structural break date unknown, a priori, and endogenously detects the date that is most likely to constitute a structural change in liquidity. This date is identified as the date with the largest Chow test-statistic, and the presence of a structural break is tested by comparing that dates test statistic with a non-standard distribution. More specifically, this sup-F test performs a Chow test for the relation in the above regression equation on each week in the data sample. If the null hypothesis of no-structural-break can be rejected, the date with the largest Chow test statistic is selected as the structural break date. To implement this test, for each policy announcement we truncate the data into two subperiods: the first one covers a one-year period before the announcement date, and the second one covers a one-year period after the announcement date. Ideally, we expect to see a structural change in liquidity level around, slightly before, or after the announcement date. Figure 4 and Figure 5 display the test results.

6.E Panel Regression Model on Bond Yield Spreads

We next rely on a panel data regression approach to analyze bond yield spreads. Our panel consists of the pooled time-series of bond yield spreads as the dependent variable and bond characteristic variables, the first principal component of trading activity variables, the first principal component of liquidity measures, the VIX measure, and rating dummies as independent variables. To overcome the problem of heteroscedasticity and serial correlation, we apply a Newey-West correction to our model. We adopt six regression specifications, as shown in Table 3, which allow us to identify the incremental explanatory power of each additional variable. The full model is Model 6 and is performed in the following structure:

(Yield spread) = 0 + 1 Maturity + 2 Maturity2 + 3 Age + 4 Coupon

+ 5 (Trading PC) + 6 (Liquidity PC)

+ 7 VIX + ( )

+ (2)

Note that subscript k denotes the kth bond-week observation in our panel dataset, and superscript s varies from 1 to S, where S is the number of distinct credit ratings. The key variables here are Trading PC and Liquidity PC. A significant positive (negative) coefficient of Trading PC indicates that higher trading activity leads to a higher (lower) bond yield spread, while a significant positive (negative) coefficient of Liquidity PC indicates that more liquidity should be compensated by a lower (higher) yield spread. Our regression results are shown in Table 3, Panels A to E, where the coefficient estimates of Rating dummies are not reported.

We also conjecture that liquidity becomes a more significant pricing factor after foreign investment took place in the interbank market, and after unlisted firms were allowed

to join the exchange market. To test this, for credit bonds in the interbank market, we run the specifications in equation (2) for the three phases of liberalization, and for credit bonds in the exchange market, we run the specifications on the two subperiods separated by May 2015. The regression outputs are presented in Table 4. As a robustness check, we let the data endogenously identify whether structural break changes exist in the relationship described in equation (2) over the past decade by performing a structural change break test (the sup-F test) around important policy announcement dates. Figure 6 and Figure 7 display the structural breaks test results.

To compare changes in the pricing of liquidity during different economic conditions, for all four credit bonds in the interbank market and the exchange market, we run the specification in equation (2) on the three subperiods, which characterize regular market conditions, recessions and crisis periods, respectively. The results are reported in Table 5. We also examine the relative increase and absolute increase in the model explanatory power, measured by R-squared, by comparing the specification in equation (2) with the same specification, but with the exclusion of Trading PC and Liquidity PC of the four credit bonds during different market conditions. The results are reported in Table 10.

6.F The Cross-market Clientele Effects

We are also interested in whether there exists a consistent yield differential between enterprise bonds in the exchange market and those in the interbank market, and also between corporate bonds in the exchange market and medium-term notes in the interbank market, after controlling for all independent variables. We attribute such yield spread differentials, if they exist, to cross-market clientele effects, generated by the heterogeneity of market participants across the two markets and associated frictions. To do so, we merge the two datasets for enterprise bonds across the two markets and create an additional independent variable, IBorEX, a dummy variable that is equal to one if the observation comes from the interbank market, and is assigned zero if it comes from the exchange market. Again, we rely on the panel regression approach and include Duration as an additional independent variable to ensure the robustness of our results. To overcome the problem of heteroscedasticity and serial correlation, we apply the Newey-West correction. The full model specification is:

(Yield spread) = 0 + 1 IBorEX + 2 Maturity + 3 Maturity2 + 4 Age

+ 5 Coupon + 6 Duration

+ 7 (Trading PC) + 8 (Liquidity PC)

+ 9 VIX + ( )

+ (3)

The key variable of interest here is IBorEX. A significant and positive (negative) value of 1 indicates that the average yield spread in the interbank market is higher (lower) than that in the exchange market. Regression results of enterprise bonds across the two markets are displayed in Table 6, where the coefficient estimates of Rating dummies are suppressed. Regression results over different subperiods of interest are presented in Table 7. This enables

us to identify whether the average yield spread differs across different liberalization phases and different economic conditions.

6.G The Within-market Clientele Effects

We are also interested in whether there exists a consistent yield differential between enterprise bonds and corporate bonds in the exchange market and in the interbank market, respectively. We attribute this yield spread differential, if it exists, to within-market clientele effects, generated by the time-varying preference of participants across the two markets. To do so, we merge the enterprise bond dataset with the corporate bond dataset in each market to obtain two new datasets, one in each market. We also create an additional independent variable, ENorCO, a dummy variable that is equal to one if the observation is an enterprise bond transaction record, and is assigned zero otherwise. Again, we rely on the panel regression approach, and include Duration as an additional independent variable to ensure the robustness of our results. To overcome the problem of heteroscedasticity and serial correlation, we again apply the Newey-West correction to our model. The full model specification is:

(Yield spread) = 0 + 1 ENorCO + 2 Maturity + 3 Maturity2 + 4 Age

+ 5 Coupon + 6 Duration

+ 7 (Trading PC) + 8 (Liquidity PC)

+ 9 VIX + ( )

+ (4)

The key variable of interest here is ENorCO. A significant, positive (negative) value of 1 indicates that the average yield spread of enterprise bonds is higher (lower) than that of corporate bonds. Regression results of the two markets are displayed in Table 8, where the coefficient estimates of Rating dummies are suppressed. The regression results over different subperiods are presented in Table 9. This allows us to identify whether this yield spread differential varies over the liberalization phases and varying economic conditions.

7. EMPIRICAL RESULTS: LIQUIDITY EFFECTS This section outlines the examination of Hypothesis 1, and presents the empirical results of measuring the levels of liquidity and the pricing of liquidity effects in the Chinese corporate credit bond market. To test Hypothesis 1, we present the summary statistics of the actual yields at different maturities and study the discrepancy between weekly average yields of bonds across the two markets. We then present empirical evidence to explore the time-series properties of the levels of liquidity, and examine whether there exist structural changes in liquidity around major policy announcement dates. Finally, we present empirical evidence to study liquidity effects as a price factor. We determine whether there exist structural

changes in the relationship specified in equation (2) around major policy announcement dates in both markets.

7.A Validity of the Principal Component Approach

Before we proceed, we first take a look at Table 1, which presents the correlations between the variables of interest and the two principal components, one from trading activity and one from liquidity proxies. First, we observe consistent negative correlations between Trading PC and Liquidity PC, with the correlations in the exchange market being consistently lower than those in the interbank market. Second, we are also interested in how each principal component correlates with each of its constituents. To confirm that Trading PC can be utilized as a proxy for trading activity, we would expect it to have negative correlations with its constituents, Volume, Zero-return, Zero-trade, and Interval, which are known to be negatively correlated with trading activity. Trading PC has the anticipated signs of correlation with its constituents, except for its negative correlations with the Volume variable in the exchange market, although relatively small. This could be largely explained by the much more volatile, erratic, and noisier trading volumes in the exchange market, which is populated with small household/individual investors. We confirm that Liquidity PC can be used as a proxy for liquidity by checking that it has negative correlations with its constituents, Amihud, Price dispersion, IQR, and CHL, which are well-known illiquidity measures. Liquidity PC indeed does have consistent negative signs of correlation with its constituents for all four credit bonds. The above findings justify the use of Trading PC and Liquidity PC as appropriate measures of trading activity and liquidity, respectively.

[Insert Table 1 Here]

7.B Data-based Yields by Maturity Buckets

In the first step of our analysis, we compute the weekly volume-weighted yields to maturity for each type of bond in the two markets. We also compare the difference in the yields to maturity between enterprise bonds in the interbank market and the comparable bonds in the exchange market, by pooling the daily observations in the two markets into 13 fixed maturity buckets.27 Table 2 presents the cross-sectional descriptive statistics of the mean, standard deviation, minimum, maximum, median value, and 25th and 75th percentile for the data-based yields to maturity of each bucket. Since the traded bonds in our dataset contain less than 1% of bonds with maturities longer than 10 years, we omit them from the summary and present the descriptive statistics of bonds with maturities less than or equal to 10 years. For bonds with multiple trades in one week, we estimate the yields from individual daily transactions by computing a volume-weighted average for that week. For each bucket, we calculate the cross-sectional descriptive statistics in each week, and report the time series average of each statistic.

[Insert Table 2 Here]

In the interbank market, for medium-term notes and commercial paper, the average yields increase in maturity monotonically from 4.77% to 5.20%, and from 4.42% to 4.84%, respectively, although the increments are small. However, for enterprise bonds, yields peak at medium-term maturities (up to 6.16%) and then gradually decrease to 5.6%. In the exchange market, yields again peak at medium-term maturities for both enterprise bonds and corporate bonds. A cross-market comparison of enterprise bonds indicates a lower average yield for maturities less than five years, and a higher average yield for maturities of six years or more. A cross-market comparison of corporate bonds indicates that there exists a virtually uniform yield difference between the two markets, with corporate bonds in the exchange market having a consistently higher average yield than those of their counterparts, except for the last two maturity buckets.28

7.C Structural Breaks in the Levels of Liquidity

Figure 3 displays the levels of liquidity of the four types of credit bonds. For corporate bonds in the exchange market, we observe a strong increase in the overall level of liquidity since 2014, preceded by a gradual decrease in liquidity since 2009. The former improvement is related to the policy announcement that allowed unlisted companies to issue corporate bonds in the exchange market in January 2015 (green bar), which also led to a sharp increase in the number of bonds traded in 2015. There is no major policy announcement targeted at enterprise bonds in recent years and, indeed, we observe no perceptible change in the level of liquidity. We do, however, find a slight increase in the level of liquidity since 2016, which we ascribe to the liquidity spillover from the vast increase in the trading activity of corporate bonds within the exchange market. In the interbank market, we observe a steady increase in liquidity for both enterprise bonds and medium-term notes immediately after the 2008 financial crisis, due to the stimulus program from the Chinese government to boost the economy in its aftermath. For commercial paper, the level of liquidity fluctuates much more wildly, possibly due to their extremely short maturities and their function as a store of short-term liquidity. Following the introduction of 18 RMB-qualified foreign institutional investors (RQFIIs) with their investment quota increased from 30 to 80 billion U.S. dollars in March 2012 (red bar) and the complete removal of approval requirements and quota limits for large QFIIs in July 2015 (orange bar), we observe a slight increase in the liquidity of enterprise bonds and medium-term notes, and a significant uptick in liquidity in the commercial paper segment. Interestingly, the crackdown of spurious agent-holding activities in 2013 does not appear to have strongly affected liquidity, as seen visually from the graph, although we cannot draw a clear conclusion based on this figure alone, and should let the data speak for themselves through a formal structural break test.

[Insert Figure 3 Here]

28

To formally test Hypothesis 1, and verify that the structural changes in the relationship in equation (1) did indeed occur around the policy announcement dates, we perform a structural break test (the Sup-F test) as discussed in the previous section on the model specification to endogenously detect the break date from the dataset. Figure 4 displays the test result for the exchange market. For corporate bonds, we find that, from a statistical perspective, the test indicates a structural break in December 2015. As mentioned above, in early 2015, the government announced the policy that allowed unlisted firms to issue corporate bonds in the exchange market, which led to an increase of liquidity in the corporate bond exchange market. The Chinese stock market crash in June 2015 also played a vital role in causing investors to shift their investments to safer, low-yield assets, such as high-quality bonds. For enterprise bonds, we find that the test indicates a structural change in May 2016, which verifies the conjecture that liquidity spillover to enterprise bonds occurred due to increased corporate bond trading activity within the exchange market. Figure 5 displays the tests results for the interbank market. Note that we excluded commercial paper from this analysis, due to its short-term nature. For both enterprise bonds and medium-term notes, the break dates of the announcement in 2012 occurred around late April and early May, respectively. The break date of the announcements in July 2015 and February 2016 for enterprise bonds occurred in early March 2016, immediately after the second announcement date, while the break date for medium-term notes occurred in late September 2015, lying between the two announcement dates. For the early 2013 crackdown of agent-holding activities, the two break dates occurred in late May and late April that year, respectively.

All of the detected structural break dates are significant at the 1% level. These dates are identified purely based on the statistical evidence, as the dates for which the Chow test statistics are most significant coincide either exactly with, or are very close to, the actual policy announcement dates. Our evidence suggests, therefore, that policy announcements in the corporate credit bond market do significantly alter the levels of liquidity.

[Insert Figure 4 and 5 Here]

7.D Panel Regression Results of Liquidity Effects

Table 3, Panels A to D, presents the panel regression results. Each panel contains six different empirical specifications. The specification in Model 1 uses only the VIX measure as the independent variable. It is used as a base case to compare other specifications against and to explore the increase in explanatory power after including additional variables. Model 1 has a reasonable level of explanatory power in each of the five cases, probably due to the inclusion of rating dummies and the VIX measure. The coefficients of the Intercept term in Models 1 and 2 are exceedingly high, due to the lack of the coupon rate as an independent variable, one of the primary determinants of yield to maturity. The specifications in Model 2 and Model 3 include only bond characteristic variables. We observe an absolute improvement in R-squared from 5.2% to 16% in each of the four cases, showing the incremental explanatory power of the Coupon variable. In addition, all bond characteristic variables are statistically significant in explaining bond yield spreads in Model 3.

Model 4 includes the additional independent variable Trading PC. All coefficients of Trading PC are positive and statistically significant at the 1% level, indicating that higher trading activity leads to higher yield spreads. Moreover, the absolute improvement in the R-squared, as compared to Model 3, varies from 1.3% to 2.1% in the exchange market, and is close to zero in the interbank market. For enterprise bonds in the exchange market, a one standard deviation increase in trading activity corresponds to an increase of 20.7 bps in the bond yield spread, which is three times more than the increase of 5.8 bps in the case of its counterpart in the interbank market. For corporate bonds in the exchange market, a one standard deviation increase in trading activity leads to an additional 15.4 bps in the bond yield spread, which is five times more than the increment of 2.5 bps for medium-term notes. This shows that trading activities play a much more important role in explaining bond yield spreads in the exchange market than in the interbank market.

The positive coefficient of Trading PC, while consistent with the findings in studies in the U.S. corporate bond market (e.g., Friewald, Jankowitsch, and Subrahmanyam (2012)) and, more generally, in research on global bond markets, may appear counter-intuitive. One explanation for this result may be that bond markets, in general, are far less liquid than stock markets, indicating that trading activity variables may not always be reliable proxies for liquidity. For instance, in a market where a particular market does not trade often, an increase in bond yield of an illiquid bond may attract trading interest and a temporary uptick in trading volume and a decline in trading interval. Since bond markets in China are especially illiquid, trading activity variables may thus not always correlate well with actual levels of liquidity, rendering the sign of Trading PC to be inverted. Another explanation could be that bonds that are traded more frequently (the more liquid ones) do not necessarily have large trading volumes, due to potentially limited outstanding amounts. Since the data of outstanding amounts are not available in our dataset, the daily trading volume, unscaled for the outstanding amounts, a key constituent in Trading PC, may be a crude measure of aggregate volume traded per day. This may cause the Trading PC to be a noisy, or even inappropriate, measure to capture the actual levels of liquidity. A daily trading volume measure that is scaled by outstanding amounts could possibly lead to negative coefficients of Trading PC, as we would expect to observe, in equilibrium. Unfortunately, we are unable to confirm this due to a lack of data.

Compared to Model 3, Model 5 includes an additional independent variable, Liquidity PC. All of the coefficients of Liquidity PC are negative and statistically significant at the 1% level, indicating that higher liquidity in the market is compensated with lower yield spreads. For enterprise bonds in the exchange market, a one standard deviation increase in liquidity leads to a decline of 5.1 bps in the yield spread, which is approximately half of the corresponding 9.4 bps decline in the interbank market. For corporate bonds in the exchange market, a one unit increase in liquidity leads to a decline of 7.6 bps in the yield spread, which is slightly higher than the 4.9 bps decline for medium-term notes. Thus, in the interbank market, liquidity effects are priced more strongly into the yields of enterprise bonds, while in the exchange market, liquidity effects are priced more into the yields of corporate bonds.

Model 6, the full model, includes the results for both Trading PC and Liquidity PC. The coefficients of the two measures are statistically significant at the 1% level in all cases,

indicating that higher trading activity and lower liquidity lead to higher yield spreads. Moreover, as anticipated, the signs remain unchanged, and the magnitude of the coefficient estimates of the two principal components decrease, as compared to the ones in Model 4 and Model 5, where only of one of the two measures is included. The magnitude of the coefficient of Liquidity PC stays the same as compared to that in Model 5. In terms of R-squared, as we move from Model 3 to Model 6, for enterprise bonds and corporate bonds in the exchange market, we find a relative improvement of 4.6% and 2.5%, respectively. For enterprise bonds and medium-term notes in the interbank market, the relative improvements are 3.2% and 0.7%, respectively. This shows that trading activity and liquidity effects, combined, explain a larger variation in bond yield spreads in the exchange market than in the interbank market. In terms of the magnitude of the increase, a one standard deviation increase in each measure leads to an aggregate increase of 17.2 bps and 10.1 bps in the bond yield spreads of enterprise bonds and corporate bonds, respectively, in the exchange market. In the interbank market, however, a one standard deviation increase in each measure leads to an aggregate decrease of 4.2 bps and 2.1 bps in the yield spreads of enterprise bonds and medium-term notes, respectively. Overall, we find that liquidity is an important priced factor driving yield spread variation in the corporate credit bond market in China. Trading activity and liquidity may explain a fair proportion of the bond yield spreads in levels. The impact of trading activity dominates that of liquidity on bond yield spreads in the exchange market, while it is the other way around in the interbank market.

[Insert Table 3 Here]

7.E Liquidity Effects under Different Phases of Interbank Liberalization

We apply the panel regression model, which includes the credit rating dummies, to analyze the changes in the liquidity effects of bond yields during different phases of liberalization. We study how liquidity effects change as more foreign investment occurred in China during Phase 1, Phase 2, and Phase 3 of the liberalization process. Table 4 reports the panel regression results during the three phases of liberalization. In Panel A, we observe that both Trading PC and Liquidity PC are mostly statistically significant. Moreover, the coefficient estimates in Phase 2 and Phase 3 are generally more statistically significant (i.e., with larger t-statistics) than those in Phase 1, except for Liquidity PC of medium-term notes. The magnitude of the coefficient estimates gradually increases from Phase 1 to Phase 2, and from Phase 2 to Phase 3, except for enterprise bonds in Phase 3 and medium-term notes in Phase 2. These observations indicate that liquidity effects became more significantly priced in bond yield spreads as the interbank market became more liberalized. Comparing the magnitude of coefficients of Trading PC and Liquidity PC shows us that, despite being both statistically significant, liquidity is priced more strongly than trading activity in the interbank market, which is the same as what we find in the combined regression. Another interesting finding is that R-squared decreased from Phase 1 to Phase 2 in all cases. Our explanation for this is from the perspective of both market participants and trading activity: As the interbank market was firstly opened to large foreign institutional investors during the first transition, the explanatory power of existing variables to explain yield spreads in the interbank market

decreased as omitted-variable bias increased. However, the R-squared decreased from Phase 2 to Phase 3 for enterprise bonds and medium-term notes. This finding concurs precisely with the policy announcement at the second transition, which permitted most foreign investors to enter the interbank market, given that many of them held medium-term to long-term investment objectives. These new-entrant foreign investors concentrated their investments in enterprise bonds and medium-term notes, consonant with their investment objectives, and lowered the explanatory power of existing variables to explain medium- to long-term credit bond yield spreads in Phase 3.

In the case of the exchange market, for which results are presented in Panel B, both Trading PC and Liquidity PC are statistically significant at the 1% level in almost all cases, except for the Liquidity PC of enterprise bonds prior to 2015. Moreover, the coefficient estimates of both measures became more significant (larger t-stat) after 2015, and their magnitudes also increased. These observations indicate that trading activity and liquidity effects became more significantly priced in bond yield spreads, as unlisted firms were allowed to issue corporate bonds in the exchange market. They also verify our conjecture that liquidity effects became more significantly priced in enterprise bonds due to potential within-market liquidity spillover effects from the increase in trading activity of corporate bonds.

As a robustness check, to verify whether the two policy announcements, which separate the three phases did, in fact, cause structural changes in the relationship in equation (2), we perform the same structural break test to endogenously detect the break dates. Figure 6 displays the test results for the exchange market. For corporate bonds, we find that, from a statistical perspective, the test indicates a structural break in late May 2015. As discussed, the announcement date when unlisted firms were henceforth allowed to issue corporate bonds in the exchange market in early 2015, while the first corporate bond issued by an unlisted firm did not occur until May 22, 2015, exactly coincides with the break date detected from the test. For enterprise bonds, we find that the test also indicates a structural change in May 2015, which verifies the conjecture that a liquidity spillover occurred due to increased corporate bond trading activity.

Figure 7 displays the test results for the interbank market. We next perform the test only for enterprise bonds. The break date for the announcement in April 2012 occurs around early 2013, showing a lagged effect on the announcement. The break date of the agent-holding crackdown announcement in early 2013 of enterprise bonds occurred in January 2013, coincident with the policy announcement date. For the July 2015 announcement of removing the approval procedure and quota limits, the break date detected by the test again coincides with the actual announcement date. Again, all detected structural break dates are significant at the 1% level. Our evidence suggests that policy announcements in the corporate credit bond market do significantly alter the relationship between the bond yield spread and the set of independent variables illustrated in equation (2).

[Insert Figure 6 and 7 Here]

[Insert Table 4 Here]

7.F Liquidity Effects under Different Economic Environments

We next explore whether liquidity effects are stronger during periods of recession and financial distress, the last conjecture in Hypothesis 1. We perform this investigation by consid