Beyond segmentation: The case of China s repo markets

Transcription

1 Journal of Banking & Finance 31 (2007) Beyond segmentation: The case of China s repo markets Longzhen Fan a, Chu Zhang b, * a School of Management, Fudan University, Shanghai, China b The Hong Kong University of Science and Technology (HKUST), Department of Finance, Clear Water Bay, Kowloon, Hong Kong Received 10 February 2006; accepted 23 June 2006 Available online 18 October 2006 Abstract This paper explores the reasons behind the discrepancy between interest rates in China s two repurchase agreement (repo) markets, the interbank repo market and the exchange-traded repo market. The repo rates in the exchange market are at times, significantly higher than those in the interbank market, especially in the first three years of the sample period. While market segmentation clearly hinders arbitrage, the causes of the repo rate discrepancy are related to the alternative investment opportunities available to market participants and to the volatility differences in the repo rates. Ó 2006 Elsevier B.V. All rights reserved. JEL classification: G12 Keywords: Repurchase agreements (repos); Market segmentation; Alternative investment opportunities; Interest rate volatility 1. Introduction In this paper, we study the pattern of interest rates in China s two repurchase agreement (repo) markets: the exchange-traded repo market and the interbank repo market. * Corresponding author. Tel.: ; fax: address: czhang@ust.hk (C. Zhang) /$ - see front matter Ó 2006 Elsevier B.V. All rights reserved. doi: /j.jbankfin

2 940 L. Fan, C. Zhang / Journal of Banking & Finance 31 (2007) Although they have a short history, the repo markets in China have grown rapidly. One interesting observation is that the repo rates on virtually identical transactions in the two markets are often very different in the sample period of While partial market segmentation is a usual culprit for such differences in repo rates in China, market segmentation itself does not create systematic differences in prices of securities with the same future payoffs. It merely prevents arbitrage from taking place. The source of the price differences is often traced to factors such as tax differences, default probability differences, alternative investment opportunities, liquidity differences, and so on. Determining what causes the differences in rates in China s two repo markets is important in order to eliminate this type of market imperfection and to guide the future development of the financial markets in China. Moreover, understanding the causes and effects of a phenomenon of this sort has implications for other emerging and transitional markets as well. We identify two sources for the discrepancy in China s two repo markets. The first one is alternative investment opportunities. The repo markets in China are mainly markets for participants to borrow or lend money, rather than to borrow or lend securities. In the exchange repo market, the reasons for borrowing money are mostly related to stock trading. During periods when hot new stocks are issued, the exchange repo rates are extremely high. The period of frequent hot issues and the period of higher exchange repo rates coincide. Variables that measure new issue subscriptions explain much of the rate differences across the two repo markets. The reasons for participants to trade in the interbank market, are diverse and mundane. The interbank repo rates initially follow the general trends in the macro economy without much fluctuation. As the interbank market develops and more instruments are introduced, the interbank repo rates also fluctuate more with the market conditions, making the two repo markets more integrated. The second source for the repo rate discrepancy across the two markets is the volatility difference between the two repo markets. The exchange repo rates were a lot more volatile than those in the interbank market in period. As a result, the exchange repo rates carried a risk premium over the interbank repo rates to compensate the lenders in the exchange repo market for bearing the volatility risk. Anticipated premiums in future short-term rates also added to the current long-term rates through expectations for the exchange-traded repo market. After 2002, the volatility of the exchange repo rates abated and fell to the level comparable to that of the interbank repo rates. The systematic differences between the two markets were also reduced. In combination, the variables indicative of alternative investment opportunities and conditional volatility explain more than 40% of the variations in the rate discrepancies across the two repo markets. Although repos are important in financial markets, the literature on repo markets is relatively small. Duffie (1996) analyzes the specialness of repo contracts driven by security borrowing and gives a brief account of the US repo market. Jordan and Jordan (1997) provide further supportive evidence of the relationship between the repo specialness and the price richness of collateral in the US, as theorized in Duffie (1996). Buraschi and Menini (2002) examine the specialness due to liquidity risk in the German repo market. Fleming and Garbade (2003) explain a new breed of repos, the so-called general collateral finance repos, in reducing transaction costs. Longstaff (2000) finds that the US repo rates conform to the expectations hypothesis, contradicting evidence from the US Treasury market. Fan and Zhang (2006) confirm that term premiums in repo rates are indeed small using China s interbank repo data. These studies, however, do not touch on the issue of market segmentation. On the other hand, the literature on market segmentation focuses

3 L. Fan, C. Zhang / Journal of Banking & Finance 31 (2007) mainly on stock markets, especially in emerging markets, where there are typically two classes of shares available, but ownership restrictions are imposed either on domestic or foreign investors, or on both. These restrictions create market segmentation and sustain price differences. Other related studies include Kidwell and Koch (1983), who investigate the segmentation of regionally issued US municipal bonds from the US Treasury market, and Simon (1991), who studies the segmentation between the US Treasury bill market and the cash management bill market. Our study is unique in the sense that the phenomenon we analyze is market segmentation in two virtually identical fixed-income markets. This paper is organized as follows. Section 2 provides institutional details on China s government bond markets and the two repo markets that use government bonds as collateral. Section 3 analyzes the repo rates in the two markets and their discrepancies. It shows that the discrepancies exhibit strong persistence initially in the sample period and decline over time. Section 4 explores the relationship of the repo rate discrepancies with alternative investment opportunities and repo rate volatilities. The last section concludes the paper. 2. China s government bond and repo markets China s government bonds consist of Treasuries, policy financing bonds (PFB), and central bank notes (CBNs). Treasuries are issued by the Ministry of Finance to the public for general purposes and are mostly medium- to long-term bonds. PFBs are bonds issued by the so-called policy banks as government agencies for development projects in special economic sectors such as agriculture, import export, and in special regions, especially rural and underdeveloped areas. CBNs are short-term notes issued by the central bank for implementation of monetary policies, available since All of these bonds are backed by the central government and have the highest quality in China. While in the early days, some Treasuries were bearer bonds, the bonds that have been issued more recently are mostly electronically registered bonds. There were two secondary bond markets in China before 1997 and there are three now since Before 1997, the Shanghai Stock Exchange and the Shenzhen Stock Exchange were the marketplaces where both government bonds and corporate bonds were traded. An over-the-counter (OTC) bearer bonds market also existed in which retail Treasury bonds could be traded between banks and investors, who were either firms or individuals. All individuals, companies and institutions participated in the exchange market through brokers. Commercial banks were heavy traders in the bond market and, sometimes, indirectly in the stock market despite regulations against such activities. An enormous amount of bank-owned cash flowed in and out of the exchange bond market and allegedly caused huge volatility in bond prices without fundamental reasons. At the end of 1996, commercial banks were ordered by the government to leave the exchange market. In the middle of 1997, an interbank bond market was established in which commercial banks could trade Treasuries and PFBs. Before the summer of 1999, commercial banks were the only participants in the interbank market. In August 1999, the central bank allowed 32 non-bank financial institutions such as insurance companies, mutual funds and securities firms to have access to the interbank market and the number of members in this market increased to 308. In 2002, some big non-financial firms were also given access to the interbank market and they accounted for about 5% of 850 members of the interbank market. By the end of 2005, the total number of the members in the interbank market reached 1750.

4 942 L. Fan, C. Zhang / Journal of Banking & Finance 31 (2007) Of the three bond markets, the exchange market and the interbank market are dominant. All the Treasuries are earmarked for either exchange trading, interbank wholesale trading, or bank OTC bearer bond trading when they are initially issued. The market for bank OTC bearer bond trading is relatively small and is not discussed below. The Treasuries traded in the exchange market are registered with and are cleared through the China Securities Depository and Clearing Corporation Ltd (SD&C), while all the Treasuries traded in the interbank market are registered with and are cleared through the China Government Securities Depository Trust and Clearing Corporation Ltd (CDC). PFBs and CBNs are traded only in the interbank market and are cleared through CDC. In terms of the market share, the interbank bond market is larger than the exchange bond market. While some Treasuries are restricted to being traded only on the interbank market and some are restricted to being traded only on the exchanges, there are some Treasuries, less than 20% of the total, that are listed on both markets. If a Treasury bond in the custody of the clearinghouse of one market is to be traded on the other market, a custody transfer is necessary before trading takes place, which typically takes two days. Although the secondary bond market was instituted in 1991 when the two national exchanges were established, the market value of outstanding Treasuries and trading volumes were small before There were only four Treasury bonds traded in the exchanges at the beginning of Panel A of Table 1 lists the face amount of the three types of government bonds outstanding since In the backdrop of the two markets for Treasuries, there are two corresponding repo markets. Repo transactions are a recent phenomenon in China, mostly used by institutions for short-term financing and by the central bank as a tool in open market operations. While some institutions also use repos to increase leverage, such activities are not commonplace in China s repo markets. Trading of exchange traded repos started at the end of 1993 with Treasuries as collateral. Trading is anonymous with a guarantee from the exchange. Trading of Interbank repos began in 1997 with Treasuries, PFBs and later CBNs as collateral. Unlike the classical repos in the US, the repos traded in China before 2003 were all cash driven, meaning that the motive for trading was to borrow or to lend money, with the traded bonds used as collateral only. The ownership of the collateral did not transfer to the buyer. This particular feature hindered arbitrage to a large extent if the repo rates differed across the two markets. An arbitrage profit could be made only by Table 1 Outstanding government bonds and trading volume of repos Category (A) Year-end outstanding face amount of government bonds (billion Yuan) Treasuries PFBs CBNs (B) Annual trading volume of repos (billion Yuan) ER BR Panel A of this table lists the year-end outstanding amount of government bonds in each category: Treasuries, PFBs and CBNs traded in exchanges and the interbank market. Panel B presents the trading volume of Shanghai stock exchange traded repos (ER) and interbank repos (BR). The sample period is 1999 through The Chinese Reminbi is pegged at 8.27 Yuan per US dollar during the sample period.

5 L. Fan, C. Zhang / Journal of Banking & Finance 31 (2007) those who have access to both bond markets and own bonds listed on both markets. The required custody transfers further deterred such arbitrage activities. In 2004, US-style repos were introduced into China s market for which the ownership of the collateral was transferred from the sellers to the buyers. The trading volume of US-style repos has been low and the impact of this type of repo on repo rates is yet to be seen. There are various repo maturities prevailing in both markets. One-week repos have been the most popular so far. The exchange market has repos with maturities of less than one week. The interbank market also trades three-week and two-month repos. Overnight repos were approved for the interbank market in 2003 and their trading volume has increased quickly. Panel B of Table 1 lists the total trading volume of all repos in the two markets. 1 ER stands for exchange-traded repos and BR stands for interbank repos. As the table indicates, the trading volume of the exchange-traded repo market was less than half of that of the interbank repo market in Due to the problem of non-performing loans in China s banking industry, interbank repo transactions have surpassed interbank borrowing and lending without collateral as the most important form of interbank borrowing and lending transactions. The background of our analysis can be summarized as follows. The post-1997 bond and repo markets began as two polarized markets. In the exchange market, all entities were allowed to trade except for commercial banks. In the interbank market, commercial banks dominated, while individuals and most non-financial firms were not allowed. Over time, however, this situation gradually changed. Some non-bank financial institutions began to gain access to both markets since the second half of Repo rates and cross-market discrepancies In this paper, we analyze the weekly observations of repo rates of four maturities that are available in both markets in the sample period of January 2000 through December The maturities are one week, two weeks, one month, and three months. The market before 2000 was too small to warrant detailed study. We first adjust the quoted exchange and interbank repo rates for their day-count differences and then convert all the rates to continuously compounded rates. 2 For exchange traded repos, the daily opening rate and closing rate are averaged first and the averaged daily rates during a week are then averaged again using trading volumes as weights. For interbank repos, since the daily average rate is available, these daily averages are averaged over a week using daily trading volumes as weights. Fig. 1 plots the weekly observations of one-week repo rates in both the exchange market and the interbank market. The rates for other maturities follow similar patterns, so they are not plotted in the figure to avoid clustering. 1 Exchange-traded repos in the table include only those from the Shanghai Stock Exchange. The total trading volume in the Shenzhen Stock Exchange was about 6% of that of the Shanghai Stock Exchange in 1999, 10% in 2000, 1% in 2001, and less than 0.1% afterwards. Thereafter, the exchange-traded repos include only those from the Shanghai Stock Exchange. 2 The exchange repo rates are quoted on the actual/360 basis while the interbank repo rates are quoted on the actual/365 basis. In addition, exchange repo rates are quoted as street yield, unadjusted for holidays, while interbank repo rates are quoted as true yield, adjusted for holidays. The adjustment is important because, in the sample period, China had long national holidays that lasted for more than a week.

6 944 L. Fan, C. Zhang / Journal of Banking & Finance 31 (2007) ER007 BR Fig. 1. Exchange and interbank one-week repo rates. This figure shows the weekly observations of the one-week repo rates in the exchange-traded market, ER007, and the interbank market, BR007, from January 2000 to December We see an obvious discrepancy between the exchange-traded repo rate and the interbank repo rate in Fig. 1. In the first half of the sample period, the interbank repo rate is quite smooth with little variation, while the exchange-traded repo rate is much more volatile than its interbank counterpart. In the second half of the sample period, however, the interbank repo rate becomes more volatile and is more in line with that of the exchange market. We examine the statistical properties of the repo rates in this section and leave the discussion about the causes of the cross-market differences in the next section. Table 2 reports the descriptive statistics of the one-week and three-month repo rates of the two markets and their differences. The two-week and one-month repo rates normally stand between the one-week and three-month ones with similar properties and are omitted from the table. Panel A gives the means, standard deviations, and percentiles of the repo rates, ERs and BRs, and the cross-market differences, denoted as DRs, for the entire sample period. Panels B and C give the same descriptive statistics for two subperiods. Over the entire sample period, the average repo rates are low, between 2% and 3%. Across all repo terms, the standard deviations of the exchange repo rates are much greater than those of the interbank repo rates and the magnitude of the difference declines in maturity. The exchange repo rates exhibit positive skewness, while the interbank repo rates have a mixed pattern. The average cross-market rate differences range from 30 to 44 basis points. These differences are quite large, considering that the average rates themselves are roughly between 2% and 3%. The standard deviations of the rate differences are large, greater than those of the interbank rates.

7 L. Fan, C. Zhang / Journal of Banking & Finance 31 (2007) Table 2 Descriptive statistics for the repo rates and their differences Variable Mean St. dev Min q 1 Median q 3 Max (A) ER BR DR ER BR DR (B) ER BR DR ER BR DR (C) ER BR DR ER BR DR Panel A of this table presents the means, standard deviations, minimums, lower quartiles (q 1 ), medians, upper quartiles (q 3 ), and maximums of the weekly observations of 7-day and 91-day repo rates in the exchange market (ER), the interbank market (BR) and their differences (DR) for the entire sample period, January 2000 through December Panel B presents the same statistics for January 2000 through December Panel C presents the same statistics for January 2003 through December The subperiod results in Panels B and C provide more details than the overall pattern shown in Fig. 1. In the first subperiod, January 2000 through December 2002, the exchange repo rates are much higher than the bank repo rates on average and the difference increases with the term. The patterns of the standard deviations look much like those in Panel A for the entire sample period. In the second subperiod, January 2003 through December 2005, the average differences are small. The standard deviations of the interbank repo rates are slightly higher than those of the exchange repo rates. An interesting observation is that, while the cross-market differences are much smaller on average in the second subperiod, their sample standard deviations are not small at all. In fact, the standard deviations are of similar magnitude as those of the repo rates themselves. There appear to be some differences across the two subperiods. Another way of characterizing the repo rate differences is through their persistence over time. The statistical persistence in the cross-market repo rate differences may provide clues about the nature of the driving forces behind these differences. The persistence pattern shows whether the repo rate differences are caused by pure random shocks in the demand and supply of funds across the two markets, or by more systematic factors that have longlasting effects. Panel A of Table 3 gives the sample autocorrelations of the repo rate differences up to the sixth order for the entire sample period. As the table shows, all the rate differences

8 946 L. Fan, C. Zhang / Journal of Banking & Finance 31 (2007) exhibit strong persistence. For the entire period, the autocorrelation for the one-week rate begins at about 0.7 for the one-week lag and decays gradually to 0.3 for the six-weeks lag. The sample autocorrelations for repo rates with other terms follow a similar pattern. As the term increases, the sample autocorrelations appear greater and decay more slowly over the time lag. Panels B and C of Table 3 report the sample autocorrelations for the two subperiods. In the first subperiod, the autocorrelations begin at about the same level as those in Panel A for the entire sample period, but they decay faster. The second subperiod has smaller autocorrelations than the first subperiod has. One of the reasons for the subperiod sample autocorrelations to be smaller than those for the entire period is the potential non-stationarity of the rate differences over the entire sample period. From Fig. 1 and Table 2, we see that, on average, the differences are positive in the first subperiod and near zero in the second subperiod. This potential non-stationarity contributes to the apparently strong persistence. Since the non-stationarity is more evident in the entire sample period than in each subperiod, it affects the sample autocorrelations more for the entire period. To explore the idea of persistence in the repo rate differences further, we estimate a vector auto-regressive (VAR) model for the four repo rate differences in the form presented in Table 3. The goodness-of-fit measures, R 2, are reported in the last column of Table 3 for each of the four individual equations. For the entire sample, the R 2 ranges from 0.6 to 0.8. For the first subperiod the R 2 is slightly smaller for the one-month rate and the threemonth rate, while for the second subperiod the R 2 s are even smaller. Like the sample autocorrelations, the R 2 s indicate that, in the first subperiod, there is strong persistence in the cross-market repo rate differences, while the persistence becomes weaker in the second Table 3 Persistence of cross-market repo rate differences q 1 q 2 q 3 q 4 q 5 q 6 Adj-R 2 (A) DR DR DR DR (B) DR DR DR DR (C) DR DR DR DR This table presents the first six sample autocorrelations of cross-market repo rate differences, x t =(x 1t, x 2t, x 3t, x 4t ) 0 = (DR007 t, DR014 t, DR028 t, DR091 t ) 0. It also reports the goodness-of-fit, adj-r 2, for each equation in the VAR model, x t ¼ a 0 þ X4 l¼1 A l x t l þ e t :

9 L. Fan, C. Zhang / Journal of Banking & Finance 31 (2007) subperiod. For the entire sample, some of the apparently strong persistence is due to the potential non-stationarity The causes of cross-market repo rate discrepancies There are potentially various causes of the repo rate differences in China s two markets. Some of them, however, can be ruled out immediately. First of all, since repo interests are not subject to tax, the tax rate difference is not an issue. Default premiums are unlikely the reason because the exchange traded repo rates tend to be higher than the interbank repo rates while the transactions in the exchange are guaranteed by the exchange. Over time, interbank repo trading volume has increased much faster than that of the exchange traded repos, while the differences between exchange repo rates and interbank repo rates have declined to zero on average. This further rules out liquidity in the sense of trading volumes as the major cause for the repo rate differences. The notion that alternative investment opportunities cause the prices of identical securities to differ has always been an interesting one in the literature on market segmentation. In the case of China s repo market, an important alternative investment opportunity is in the stock market. The sample period of January 2000 through December 2005 witnessed the rise and fall of the emerging Chinese stock market. The 1990s and the early 2000s were the boom period for the stock market in China. The Shanghai Stock Exchange A-Share Index peaked in June 2001 at While the average price earning ratio was still high by conventional standards, the stock value languished for the next four years in the sample period. By June 2005, the index slid to , dropping to about half of its value from its peak. While stock market performance in general had some impact on the repo rates, a particularly relevant sector was the market for initial public offerings (IPOs). The initial returns on IPOs were extremely high during the first half of the sample period. Before the summer of 2002, IPOs were issued to all investors and were allocated using lotteries when oversubscribed. In the hot market of the 1990s and early 2000s, oversubscription was commonplace. In order to improve the odds of obtaining shares, investors would apply for many more shares than they actually wanted. Between January 2000 and June 2002, the fifth percentile of the oversubscription rate was 36.5 times and the ninety-fifth percentile was times. Investors were required to submit funds at the time of an IPO application and would get refunds for the unsuccessful portion of their applications. Sources of funding for speculating in the IPO market were limited, however. Bank loans for individuals and small private enterprises were hard to obtain without collateral. Approval for bank loan applications were given to those with legitimate purposes and the use of funds after approval would be closely monitored. Under the circumstance, a natural source of cash for IPO subscriptions was the exchange-traded repo market, accessible by most investors who owned Treasuries, except for banks. The one-week repo was extremely useful because it took a few days for the subscription and allocation to go through for a typical IPO. The high borrowing costs in the exchange repo market did not seem to matter because they were outweighed by the expected IPO returns. 3 Another way to present the persistence of the variable under study is through the impulse response functions from the VAR model, which summarizes the effect of unreported parameters. The details of impulse functions that reveal similar persistence properties are available upon request.

10 948 L. Fan, C. Zhang / Journal of Banking & Finance 31 (2007) ER S Fig. 2. Stock market new issue subscriptions and the one-week exchange repo rate. This figure shows the stock market new issue subscriptions (in 100 billion Yuan), 10S t, and the one-week exchange repo rate, ER007 t, from January 2000 to December In the summer of 2002, however, China s Security Regulatory Committee changed the rules governing IPO issuance. Instead of being issued to all investors, at least 50% of IPO shares were required to be issued only to existing shareholders (of other companies) according to the amount of their holdings and funds were not required to be submitted until the allocation was confirmed. For the remaining 50% or less issued to the public, the old rules still applied. Most issuers chose to allocate 100% to existing shareholders. The change in the rules substantially reduced IPO oversubscription and, in turn, the reliance on the repo market for funding IPO purchases. We consider the public IPO subscription, denoted as S t and measured in trillion Yuan, that characterizes these new issue stock market activities. Fig. 2 plots S t along with the one-week exchange repo rate during the sample period. As the figure shows, the total subscription amount is huge in the first half of the sample period, and there were only a few IPOs in the second half that were issued to the public. The two series correspond well at times in the first half of the sample period, indicating that IPO subscription is indeed a reason for the large fluctuation in the exchange repo rates, although it obviously does not explain all of the fluctuations. 4 4 Another variable that helps explain repo rate differences is the expected IPO returns. In an earlier version of the paper, we used a simple ad hoc model that measures expected IPO returns in a week as a weighted average of realized IPO returns in the past weeks. The expected IPO returns measured this way indeed play a significant role in explaining the repo rate differences when used alone, but lose much of their significance when combined with the actual IPO subscriptions.

11 L. Fan, C. Zhang / Journal of Banking & Finance 31 (2007) While stock market activities are an important reason for exchange-traded repo rates to fluctuate, they are not as important for interbank repo rates. In understanding the driving force of interbank repo rates, we consider the interest rates on the central bank reserves and the CBN rates, as they represent alternatives to trading repos for interbank traders. Unlike many other countries, China s central bank pays all the commercial banks interest on their reserves (required and surplus) and charges a much higher interest rate for any shortage of the required amount. This reserve interest rate, denoted as R t, is set by the central bank and changes occasionally. Fig. 3 plots the rate for the sample period. Since the transfer of funds between banks and the central bank is within the system and is extremely easy, the reserve surplus obviously provides an alternative for trading repos and the reserve interest rate serves as a lower bound for the repo rates. In April 2003, the central bank started to issue CBNs. These are short-term instruments issued on the weekly basis to institutional investors in the interbank market. Since their inception, CBNs have become an important alternative to interbank repos. The market perceives CBNs as even less credit-risky than repos because they are directly issued by the central bank. The CBN rates, determined by auction on every Wednesday, fluctuate with the interbank repo rates because they are subject to similar demand and supply shocks. The three-month CBN has the shortest maturity. Its rate, denoted as N t, is also plotted in Fig. 3, along with the threemonth interbank repo rate. We can spot a close, though not perfect, relationship between the CBN rate and the repo rate BR091 R N Fig. 3. Central bank rates and the three-month interbank repo rate. This figure shows central bank reserve interest rate, R t, the three-month CBN rate, N t, and the three-month interbank repo rate, BR091 t, from January 2000 to December 2005.

12 950 L. Fan, C. Zhang / Journal of Banking & Finance 31 (2007) To combine the effects of the central bank reserve interest rate, R t, and the CBN rate, N t, we define a new variable, C t = max(r t,n t ). This new variable will be loosely termed as the central bank interest rate and used as an indicator of alternative investments for interbank repos. Unlike the relationship between actual IPO subscriptions and the exchange repo rates that has an obvious causal connection, the relationship between the central bank rate and interbank repo rates is simply a correlation. The advent of CBN trading also coincided with the beginning of the interbank repo market as a trading system open for a larger number of participants. As many new members of the interbank market had access to other markets, the interbank market became more integrated with other markets. From Fig. 1, we see that in the first half of the sample period, the exchange repo rate is more volatile than the interbank repo rate and the cross-market difference is mostly positive, while in the second half of the sample period, the two repo rates are about equally volatile and the cross-market difference is close to zero. This provides a clue that the cross-market repo rate volatility difference might be a reason for the cross-market repo rate differences themselves. The idea that interest rates are related to the volatility of the short-term interest rate has a sound theoretical foundation. Longstaff and Schwartz (1992), for example, use the conditional volatility of the instantaneous rate as a factor that, together with the short-term rate itself, determines the entire term structure of interest rates. Another reason the conditional variance of the one-week rate effects the repo rates of longer terms is that repo rates of longer terms contain expectations of future short-term rates, which will be affected by the future conditional variances of the short-term rate, which in turn are partially determined by the current conditional variance of the shortterm rate. Empirical evidence also supports the model s implications. In particular, Fan and Zhang (2006) find that the short-term rate volatility is a determining factor of the term premiums in China s interbank repo market. For whatever reason, if the short-term rate volatilities across China s two repo markets are different, then the difference is likely to cause the rates of all maturities to differ across the two markets. We construct an AR(1)-EGARCH(1,1)-in-mean model to estimate the conditional variance, h e t, for the one-week exchange repo rate, and the conditional variance, hb t, for the one-week interbank repo rate. Different specifications of the conditional variance may produce different estimates of model parameters, but the differences in resulting conditional variances have little impact on the analysis below. Therefore, the details of the model are not reported here. Fig. 4 plots the conditional standard deviations, rather than the variances, for better graphic presentation. p ffiffiffiffi The conditional standard deviation of the one-week exchange repo rate, h e t, is high with greater fluctuation in the first half of the sample period and becomes lower in the second half with reduced fluctuation. On qtheffiffiffifficontrary, the conditional standard deviation of the one-week interbank repo rate, h b t, is very low until August 2003, at which point there is an obvious jump to a level comparable to the conditional standard deviation of the exchange repo rate. It remains low after May The patterns revealed in Fig. 4 conform well to the patterns seen in Fig. 1. It should be noted that, although h e t and h b t are the conditional variances of the oneweek rates, they can be viewed as multipliers of the conditional variances of the repo rates of other terms as well because of the strong correlations among repo rates of all the terms. Regression results (not reported here) confirm that the conditional variance

13 L. Fan, C. Zhang / Journal of Banking & Finance 31 (2007) (h e ) 1/2 (h b ) 1/ Fig. 4. Repo rate volatilities. This figure shows the estimated conditional standard deviation of one-week repo p ffiffiffiffi qffiffiffiffi rates for the exchange repo market, h e t, and the interbank repo market, h b t, from January 2000 to December of the one-week exchange repo rate, h e t, contributes to the determination of the exchange repo rates of all the terms considered here. Similarly, the conditional variance of the one-week interbank repo rate, h b t, contributes to the determination of the interbank repo rates of all the terms. The coefficients for the same term but across the two markets are different. As a result, the cross-market repo rate differences are related to both conditional variances. It is more interesting, however, to determine if their difference contributes to the differences in the repo rates. We therefore define H t ¼ ln h e t ln h b t. We combine all the variables that may have explanatory power for the cross-market repo rate differences in the following regression: DR t ¼ w 0 þ w 1 S t þ w 2 C t þ w 3 H t þ e t ; ð1þ where DR t is the cross-market repo rate difference of a particular term. Since the explanatory variables are correlated with each other, the joint effects may interact. It is worth noting that the model is by no means the best model conditioned on all currently available information. It is meant to be explanatory, so that variables such as the lagged repo rates are omitted. Variables that explain both exchange repo rates and interbank repo rates, but not their differences, are not included either. Table 4 reports the regression results. The result in Panel A for the entire sample period shows that the coefficients of S t are positive and very significant for all the maturities. The coefficients of C t are negative

14 952 L. Fan, C. Zhang / Journal of Banking & Finance 31 (2007) Table 4 The effects of alternative opportunities and conditional volatility differences w 0 w 1 w 2 w 3 Adj-R 2 (A) DR (6.75) (5.49) ( 6.11) (1.23) DR (5.87) (5.75) ( 5.56) (2.16) DR (4.71) (5.86) ( 5.42) (4.13) DR (1.66) (6.18) ( 2.28) (7.30) (B) DR (0.20) (5.05) (0.84) DR (0.84) (5.08) (1.01) DR (1.09) (4.69) (1.49) DR (3.47) (4.01) (2.31) (C) DR (8.21) (1.21) ( 6.82) ( 0.14) DR (6.43) (1.07) ( 5.82) (0.13) DR (5.49) (2.00) ( 5.98) (2.06) DR (2.06) (0.91) ( 3.09) (2.48) This table presents parameter estimates of the model DR t ¼ w 0 þ w 1 S t þ w 2 C t þ w 3 H t þ e t ; where DR t is the cross-market repo rate difference of a particular term, S t is the new stock issue subscription (in trillion Yuan), C t is the central bank interest rate, and H t ¼ ln h e t ln h b t is the difference between the log of conditional variances of the exchange one-week repo rate and of the interbank one-week repo rate. The numbers in the parentheses are Newey West t-ratios with three lags. and significant. 5 When used alone, H t is very strong in explaining the rate differences (not reported in the table). When all three explanatory variables are included in the regression, the role of the volatility difference, H t, is reduced and becomes insignificant in the oneweek rate difference, but it remains significant for other maturities. The goodness-of-fit measure, R 2, stays above 40%. The subperiod results reported in Panels B and C appear to be a little weaker than those for the entire sample. For the first subperiod, January 2000 through December 2002, since C t rarely varies, it is highly collinear with the constant term and is left out of the equation. 5 These two explanatory variables are also used to explain exchange and interbank repo rates separately. The results are not reported but are summarized here. Basically, IPO subscriptions, S t, explain exchange repo rates very well, but explain very little about interbank repo rates. On the contrary, the central bank rate, C t, explains the interbank repo rates very well, but not much about the exchange repo rates.

15 The IPO subscription, S t, is still useful in explaining the cross-market repo rate differences. H t is marginally significant when used alone, but not so when combined with S t. Recall that, in the first half of the sample period, h b t does not have much variation. The role of H t thus mainly comes from h e t, which is correlated with S t. For the second subperiod, January 2003 through December 2005, the IPO subscription is not significant while the central bank rate is. H t is useful only in explaining the one-month and three-month rate differences. Overall, the goodness-of-fit measures for the subperiods are much lower than those for the entire sample. The reduction is most substantial for the three-month repo rate difference. The fact that the subperiod results are weaker than those for the entire period suggests that part of the strong relationship between the rate differences and the explanatory variables comes from the cross-period differences. The week-to-week variation in the explanatory variables is useful in explaining the cross-market rate differences, but it is not as strong as what the results for the entire sample period indicate. 5. Conclusions In this paper, we exploit a rare opportunity to examine China s exchange-traded and interbank-traded repo markets that have identical products but different interest rates. For the sample period of January 2000 through December 2005, the apparent difference across the two markets is not arbitraged away due to market segmentation. When the sample period is divided into two halves, we see striking differences between the two. In the first half, the differences between the exchange repo rates and the interbank repo rates are positive on average and the standard deviations of the differences are large. On the other hand, the cross-market repo rate differences in the second half of the sample period are near zero on average, with standard deviations only slightly smaller than those in the first half. In addition, the cross-market repo rate differences exhibit much stronger persistence in the first half of the sample period than they do in the second half. This is indicative that there are systematic forces in the first subperiod that drive apart the repo rates of the two markets. Over time, we observe a trend that the repo rates differences across the two markets become smaller on average, but remain volatile from time to time. Two types of variables are helpful in explaining the cross-market repo rate differences. One type is the measures of alternative investment opportunities. The exchange repo market was heavily used by investors in the IPO market in the first subperiod due to institutional arrangements. New issue subscriptions and central bank interest rates explain much of the rate differences across the two repo markets. The other type of variables consists of the volatility difference in the short-term repo rates across the two markets. The volatility is important because it is a component of the risk premium in the interest rates. In combination, these variables explain more than 40% of the variations in the rate discrepancies across the two repo markets. Subperiod results are slightly weaker, however, indicating that the high goodness-of-fit measure for the entire sample period is partly due to the cross-period differences in these variables and partly due to the week-to-week variation within each subperiod. Acknowledgments L. Fan, C. Zhang / Journal of Banking & Finance 31 (2007) We thank Warren Bailey, Douglas Foster, Haizhou Huang, Jun Pan, the participants of the 2005 China International Conference in Finance, the seminar participants at HKUST,

16 954 L. Fan, C. Zhang / Journal of Banking & Finance 31 (2007) and two anonymous referees for their helpful comments and suggestions. Longzhen Fan acknowledges financial support from the National Science Foundation of China Grant NSFC Chu Zhang acknowledges financial support from the Direct Allocation Grant DAG03/04.BM36 of HKUST. All remaining errors are ours. References Buraschi, A., Menini, D., Liquidity risk and specialness. Journal of Financial Economics 64, Duffie, D., Special repo rates. Journal of Finance 51, Fan, L., Zhang, C., The Chinese interbank repo market: An analysis of term premiums. Journal of Futures Markets 26, Fleming, M., Garbade, K., The Repurchase agreement refined: GCF repo. Federal Reserve Bank of New York Current Issues in Economics and Finance 9, 1 7. Jordan, B.D., Jordan, S.D., Special repo rates: An empirical analysis. Journal of Finance 52, Kidwell, D.S., Koch, T.W., Market segmentation and the term structure of municipal yields. Journal of Money, Credit and Banking 15, Longstaff, F., The term structure of very short rates: New evidence for the expectations hypothesis. Journal of Financial Economics 58, Longstaff, F., Schwartz, E., Interest rate volatility and the term structure: A two-factor general equilibrium model. Journal of Finance 47, Simon, D.P., Segmentation in the Treasury bill market: Evidence from the cash management bills. Journal of Financial and Quantitative Analysis 26,