Journal of Banking & Finance 30 (2006) 1309–1332 www.elsevier.com/locate/jbf Spanish Treasury bond market liquidity and volatility pre- and post-European Monetary Union q Antonio Dı́az a, John J. Merrick Jr. a b,* , Eliseo Navarro a Universidad de Castilla-La Mancha, Facultad de C. Económicas y Empresariales, 02071 – Albacete, Spain b School of Business, College of William and Mary, P.O. Box 8795, Williamsburg, VA 23187-8795, United States Received 19 April 2004; accepted 12 April 2005 Available online 27 June 2005 Abstract Spain enacted a number of important debt management initiatives in 1997 to prepare its Treasury bond market for European Monetary Union. We interpret the impacts of these changes through shifts in a bond liquidity ‘‘life cycle’’ function. Furthermore, we highlight the importance of expected average future liquidity in explaining Spanish bond liquidity premiums. We also uncover pricing biases that support the Spanish TreasuryÕs tactical decision to target high-coupon, premium bonds in its pre-EMU debt exchanges. Finally, we show that EMU has been associated with both a decrease in bond yield volatility and an increase in pricing efficiency.  2005 Elsevier B.V. All rights reserved. q This project developed while Merrick was at Baruch College. The authors acknowledge useful comments from participants in BaruchÕs Brown Bag Seminar series. Furthermore, Dı́az and Navarro acknowledge the financial support provided by Junta de Comunidades de Castilla-La Mancha grant PAC2002/01 and by Ministerio de Ciencia y Tecnologı́a grant BEC 2001-1599. * Corresponding author. Address: School of Business, College of William and Mary, P.O. Box 8795, Williamsburg, VA 23187-8795, United States. Tel.: +1 757 221 2891; fax: +1 757 221 2937. E-mail addresses: antonio.diaz@uclm.es (A. Dı́az), john.merrick@business.wm.edu (J.J. Merrick Jr.), eliseo.navarro@uclm.es (E. Navarro). 0378-4266/$ - see front matter  2005 Elsevier B.V. All rights reserved. doi:10.1016/j.jbankfin.2005.05.009 1310 A. Dı́az et al. / Journal of Banking & Finance 30 (2006) 1309–1332 JEL classification: G12; G15 Keywords: Bonds; Liquidity; Spanish public debt; European Monetary Union 1. Introduction This paper examines liquidity and volatility in the Spanish Treasury bond market within the context of debt policy shifts engineered by the Spanish government in preparation for entrance into European Monetary Union.1 The TreasuryÕs mid1997 debt management innovations were designed to make Spanish debt more attractive to the new class of Pan-European government bond investors created under European Monetary Union. These measures included (1) increases in the size of new issues, (2) increases in the time between bond issuance tranches, (3) development of a strips market and (4) institution of a new aggressive exchange policy to replace certain seasoned issues. A key purpose of this paper is to investigate the impact of SpainÕs debt management initiatives on both trading activity and valuation in its debt market. As it happens, SpainÕs concerns over properly preparing its markets for dramatic shifts in the relevant investor class under EMU turned out to be quite prescient. The share of Spanish government debt held by non-resident investors climbed from 25% in 1996 to 47% by February 2003.2 Analysis of SpainÕs actions and experiences during these special circumstances provides a number of specific insights on market structure and policy impacts of interest to both policymakers and academic researchers. To facilitate these insights, we estimate a model that relates individual Treasury issue market share of overall trading volume to a bondÕs age (the time since its initial auction) in a fashion best described as a liquidity life cycle. We then test for shifts in this liquidity life cycle function as a result of the TreasuryÕs debt policy innovations. We also estimate the structure of liquidity premiums in the different maturity sectors within the Spanish bond market and quantify the impacts of SpainÕs EMU-related debt management policy shifts on Spanish Treasury bond valuation. We conclude by examining the impacts of European Monetary Union on volatility and pricing efficiency in the Spanish Treasury market. 1 In general, prudent debt management by any sovereign requires attention to market structure and trading costs. Indeed, one of the three debt management goals espoused by the US Treasury is to ‘‘promote efficient markets’’ (see the US Treasury website). Likewise, the joint International Monetary Fund-World Bank guidelines for developing country debt management list an entire menu of regulatory and market infrastructure conditions designed to enhance debt market efficiency (see Box 5, ‘‘Relevant Conditions for Developing an Efficient Government Securities Market’’, in International Monetary Fund/ World Bank (2001)). 2 Source: Spanish Treasury (Tesoro Público). In contrast, during the transition to EMU, trading volume in the MEFFÕs (Mercado Español de Futuros Financieros) Spanish 10-year government bond future contract withered away. A. Dı́az et al. / Journal of Banking & Finance 30 (2006) 1309–1332 1311 1.1. Bond market liquidity proxies and model specification Liquidity is the somewhat amorphous financial market concept that embodies the ease with which a security can be traded within a short period of time without causing significant impacts on prices. Liquidity is valuable because of the associated savings of both trading costs and trading time. Theoretically, investors should require lower returns on assets with relatively high degrees of liquidity. The difference between the required return on liquid versus less liquid assets is called a liquidity premium. Issuers whose securities trade in liquid secondary markets should benefit through lower costs of capital. This effect should hold for both debt and equity securities and for both private and sovereign issuers.3 Operationally, analysis of potential liquidity effects in the cash bond markets involves choosing both a specific observable proxy for liquidity and a security valuation model. In this paper, we feature issue-specific trading volume market share and ‘‘auction status’’ proxies for liquidity. We test for the importance of liquidity effects by using these liquidity proxies to explain the valuation residuals from a standard term structure model. Our auction status approach attempts to follow the lead of the empirical literature for the US Treasury market, where the most recently auctioned or ‘‘on-the-run’’ issue in each maturity sector is distinguished from all other ‘‘offthe-run’’ issues. However, due to the special issuance system employed by the Spanish Treasury, we propose the need for three different status stages: ‘‘pre-benchmark’’, ‘‘benchmark’’ and ‘‘seasoned’’.4 Furthermore, following Goldreich et al. (2005), our empirical specifications stress the importance of distinguishing between current and expected future liquidity. As it happens, this distinction is critically important for understanding liquidity in the Spanish market. In particular, the pre-benchmark Spanish Treasury bond has a low share of overall trading volume at issue, but carries an expectation of a sharply increasing future market share. In contrast, the current benchmark bond has a high current share of market trading volume, but carries an expectation of a decreasing future share of market trading volume. Our main results regarding the impacts of SpainÕs debt policy changes concern both the liquidity life cycle of the typical bond and the value of liquidity. First, we show that a specific continuous, highly non-linear function of bond age explains the typical bondÕs changing market share of trading volume quite well. Moreover, we confirm that important structural changes took place in the Spanish market during the approach to monetary union. In particular, shifts in the liquidity life cycle modelÕs key parameters after 1997 show that Spanish debt market trading activity became more concentrated in benchmark bonds and reveal that the benchmark status period lengthened. 3 Sarig and Warga (1989), Amihud and Mendelson (1991), Warga (1992), Kamara (1994), Carayannopoulos (1996), Duffee (1998), Elton and Green (1998), Fleming (2001), Strebulaev (2001), Krishnamurthy (2002) and Goldreich et al. (2005) analyze liquidity in the US government debt markets. The liquidity of Spanish government debt has been studied by Alonso et al. (2004), who apply the Elton and Green (1998) methodology, and by Dı́az and Navarro (2002). 4 Alonso et al. (2004) propose similar stages for Spanish bonds. 1312 A. Dı́az et al. / Journal of Banking & Finance 30 (2006) 1309–1332 We also present a number of interesting results concerning liquidity value in the Spanish bond market. We show that our explicit life cycle function adds significant explanatory power to the literatureÕs standard bond auction status dummy variable approach. In particular, we use our estimated market share life cycle functions to project each issueÕs future liquidity. This allows us to test for an empirical relation between bond values and liquidity, while specifically distinguishing between the impacts of current versus expected future liquidity. Our results reveal that expected future liquidity is much more important than current liquidity for explaining relative Spanish Treasury bond values. In addition, we examine the valuation impacts of bond-specific characteristics such as the coupon rate and price premiums and discounts versus par. Our results for the 1993–1997 sample period detect statistically significant valuation biases confirming that Spanish investors favored discount bonds over premium bonds. These results lend support to the Spanish TreasuryÕs tactical decision to target high-coupon, premium bonds in its debt exchanges. Interestingly, we find that the impact of such bond-specific characteristics on value in the Spanish market decreased after European Monetary Union. Finally, we examine the impacts of European Monetary Union on the volatility of yields in the Spanish Treasury market. As anticipated by its early proponents, European Monetary Union led to dramatic falls in both yield levels and yield volatility for ‘‘Club Med’’ members such as Spain and Italy. For these countries, European Monetary Union membership decreased the relevant currency translation risks as well as the perceived bond default probabilities. Formal tests here based on the first-differences of yields strongly reject the null hypothesis of equal yield variances in our preand post-EMU periods. Such an impact on volatility is generally acknowledged (see Codogno et al., 2003). It is less widely recognized that European Monetary Union has also led to more efficient relative pricing in the Spanish Treasury bond market. Our results reveal that the residual variance of our liquidity and bond characteristicaugmented yield regressions fell sharply between our pre-EMU and post-EMU samples. Such improved pricing efficiency may be attributed to an important byproduct of European Monetary Union: the creation of a much larger universe of euro-based fixed income investors willing to focus attention on trading opportunities in any member market without the hindrance of currency risk. 2. Institutional features of the Spanish Treasury debt markets With total domestic Treasury debt of approximately €315 billion as of year-end 2002, Spain is the fourth largest euro zone sovereign debt market, trailing only Italy (€1061 billion), Germany (€745 billion) and France (€732 billion) in total par amount outstanding.5 Spanish Treasury debt consists of both bills and coupon- 5 Source: Security statistics, Bank for International Settlements, September 2003. A. Dı́az et al. / Journal of Banking & Finance 30 (2006) 1309–1332 1313 bearing notes and bonds. Letras del Tesoro (Treasury bills) are issued at discount with 6-, 12- and 18-month maturities. Bonos and Obligaciones del Estado (Treasury bonds and notes) bear annual coupon payments and have been issued for 3-, 5-, 10-, 15- and 30-year maturities. Letras del Tesoro and, since 1999, Bonos and Obligaciones have been traded free of withholding tax for non-residents and institutional investors. All Spanish Treasury debt is issued via competitive auction. However, the Spanish Treasury traditionally has built up the total par amount of each new security by keeping the same issue open over several (at least three) consecutive auctions. The securities issued through each tranche were fully fungible since they shared the same nominal coupon, interest payment and redemption dates, and security code. When the total nominal amount issued reached the appropriate target size, the corresponding security code was closed, and any further issuance took place using a new security. The secondary market for Spanish Treasury debt is known as Mercado de Deuda Pública Anotada or MDPA.6 Current practices in the Spanish market are the result of changes made by Spain as it prepared for entry into European Monetary Union. Under EMU, the Spanish Treasury recognized that it would have to compete directly with other euro zone sovereign debt issuers. Thus, beginning in mid-1997, the Spanish Treasury ‘‘prioritized the achievement of a more liquid and efficient public debt market’’.7 In practical terms, this meant undertaking a set of initiatives aimed to attract investor savings within the new single capital market. Among these initiatives, we highlight the following measures designed to increase the depth and liquidity of the market: 1. Reform of the Treasury market makers regime adapting it to the EMU rules and modifying the rights and obligations of public debt market makers and recognized dealers; 2. Enlargement of Spanish public debt trading platforms through ‘‘the advance in the blind segment of the Spanish market centred on the roll-out of a fully electronic trading system supporting automatic posting of public debt prices’’; 3. A change of the tax regime of the public debt; 4. An increase in the size of bond issues to total par amounts in the €11 billion to €12 billion range; 5. Organization of a Treasury strips market; 6. An increase in the range of issued Treasury maturities to include a 30-year bond; 6 The MDPA conducts trading through three systems. The first two are reserved for market members, while the third is for transactions between market members and their clients. The first member system is a ‘‘blind market’’ electronic trading system conducted without knowledge of the counterpartyÕs identity, while the second system channels all the remaining transactions between market members. The structure of the Spanish market is quite similar to the US Treasury market (see Fleming and Remolona, 1999 for details about the US Treasury market). 7 See ‘‘Memoria 2000’’ of Tesoro Público (http://www.mineco.es/tesoro/htm/deuda/Memorias/ indice_i.htm). 1314 A. Dı́az et al. / Journal of Banking & Finance 30 (2006) 1309–1332 7. A new government debt exchange policy designed to replace certain seasoned, low liquidity, high-coupon issues with new strippable, close-to-market coupon rate bonds. The debt exchanges provided a crucial mechanism through which to build par amount size in new issues with current market coupon levels that would be priced near 100% of par. This shift was designed to increase market liquidity and depth through two channels. First, the exchange policy ensured an adequate tradable supply of bonds priced near par (at the expense of premium bonds that some classes of investors avoid). Second, the debt exchanges produced the larger outstanding amounts of strippable bonds that were critical in supporting bond dealer stripping and reconstitution operations in the new strips market.8 3. Liquidity in the Spanish public debt market In this section, we first describe our database and discuss alternative proxies for liquidity in the Spanish Treasury bond market. We then analyze the impact of the new EMU-related debt management policy shifts on bond trading activity and bond liquidity. 3.1. The data The original database consists of 65,135 observations derived from actual transactions in all Spanish Treasury bills and bonds traded in MDPA (obtained from annual files made available by the Banco de España) over the period from January 1993 to December 2002. For each issue, the Banco de España database reports daily information on the number of transactions and both the nominal and effective trading volumes. The database also reports the maximum price, the minimum price and the average price for each issue computed from all MDPA transactions over each day in the sample. We match this information with each issueÕs coupon rate, maturity date, issue date and remaining coupon payment dates. We also track the par amount outstanding of each issue at the end of each month. Table 1 gives a brief overview of the average trading volume and par amounts outstanding where bonds are grouped by original issuance date term-to-maturity. The 10-year sector is the most actively traded maturity sector and accounts for about 41% of overall Treasury market trading. In this paper, we focus on trading activity in the three most active sectors: 10-, 5- and 3-years. 8 Government debt exchanges were conducted via competitive auction, in which the Treasury reserved the right to decide the cut-off price. (Also, in 2001 and 2002, exchange transactions were substituted by direct repurchases of the targeted high-coupon issues using the TreasuryÕs cash surpluses.) 1315 A. Dı́az et al. / Journal of Banking & Finance 30 (2006) 1309–1332 Table 1 Spanish Treasury market database summary for the 1993–1997 and 1998–2002 sample periods Bills 3-year bonds 5-year bonds 1993–1997 Daily data Average volume per traded issuea Average volume per sector # Traded issues per day # Observations 26.84 184.99 6.9 8539 94.26 440.49 4.7 5809 77.22 504.29 6.5 8117 Monthly data % Days traded per issue Average amount outstanding per issue Average market share per issue Average market share per sector Average # outstanding issues 22.9% 74.6% 78.5% 1548 4265 4490 0.2% 3.7% 3.2% 10.1% 23.5% 27.1% 38.12 6.3 8.3 Global information Total # issues in subsample 1998–2002 Daily information Average volume per traded issuea Average volume per sector # Traded issues per day # Observations Monthly information % Days traded per issue Average Average Average Average amount outstanding per issue market share per issue market share per sector # outstanding issues Global information Total # issues in subsample 255 18.33 97.53 5.3 6604 14.5% 16 136.23 485.81 3.6 4511 71.3% 12 9 99.50 708.73 7.1 8837 77.7% 13 15-year bonds 30-year bonds 24.94 34.05 1.4 1301 – – – – 95.2% 87.0% 5022 5695 5.1% 1.0% 38.0% 1.3% 7.5 1.5 10 126.38 136.76 641.86 542.14 5.1 8.1 6430 10,256 1194 6986 7409 0.1% 4.0% 3.9% 4.6% 20.1% 25.4% 43.2 5.0 6.6 203 10-year bonds 63.7% 3 37.63 93.64 2.5 3113 61.6% – – – – – – 61.71 84.47 1.4 1677 95.1% 6737 6262 6714 3.4% 1.0% 2.2% 43.2% 3.8% 3.0% 12.7 4.0 1.4 15 5 2 Average volume and amounts outstanding expressed in million € of par value. Issues are sorted by maturity sector of original issue. a Average calculated after excluding sample points for issues with zero volume on the given day. 3.2. Empirical liquidity proxies in the previous literature The literature recognizes a wide range of market condition variables and securityspecific characteristics related to bond liquidity. In the US Treasury market, a common liquidity proxy is a bondÕs bid–ask spread.9 Elton and Green (1998) suggest that the best proxy for liquidity is trading volume, though Fleming (2001) finds improved 9 See, for example, Shen and Starr (1998), Chakravarty and Sarkar (1999), Hong and Warga (2000), Chen and Wei (2001), Fleming, 2001), Gwilym et al. (2002) and Goldreich et al. (2005). 1316 A. Dı́az et al. / Journal of Banking & Finance 30 (2006) 1309–1332 performance using the number of trades instead.10 The literature also promotes bond age, auction status and issue size as relevant explanatory variables.11 For example, Fisher (1959) uses the amount of bonds outstanding on the basis of the potential correlation between the existing stock of a particular bond and the flow of trade in the bond. Sarig and Warga (1989) and Warga (1992) suggest that younger bonds are usually traded more frequently. Warga (1992) uses an auction status dummy variable that indicates whether or not an issue is ‘‘on-the-run’’ (i.e., the most recently issued security of a particular maturity). Amihud and Mendelson (1991) observe that bonds approaching maturity are significantly less liquid since they are ‘‘locked away’’ in investorsÕ portfolios. Importantly, Goldreich et al. (2005) emphasize expected liquidity over the full life of the issue – not just the current level of any liquidity measure – as the most relevant theoretical constructs for valuing bond liquidity. Here, we analyze the evolution of Spanish Treasury bond liquidity with respect to auction status and bond age. We argue that the two-stage (on-the-run/off-the-run) division traditionally used for US Treasury debt is not the most suitable choice for Spanish Treasury assets over our sample period since the Spanish Treasury built up its issues through a series of issuance tranches. Thus, the most recently issued security (the on-the-run) might have been only one-fourth or one-third of the size of the first off-the-run issue. This important relative issue size difference suggests that the on-the-run issue need not have the highest current liquidity.12 Moreover, detailed analysis of the data motivates a more sophisticated approach to modeling the evolution of a typical bondÕs trading activity over its life cycle. As it happens, this approach also allows us to build appropriate measures of expected average future liquidity to maturity for any issue on any trading date. Thus, this approach is particularly convenient for distinguishing between the values of current and future liquidity for Spanish Treasury bonds in the spirit of Goldreich et al. (2005). We use individual issue market share of total trading activity as the measurable proxy for relative liquidity.13 Many previous studies of bond market liquidity analyze raw trading volume (see, for example, Elton and Green, 1998; Fleming, 2001). We prefer the market share measure to raw volume measure since Spanish Treasury bond trading volumes trended higher over the 1993 to 2002 sample period. Scaling these individual issue volumes by total market volume both detrends these data and controls for week-to-week volume fluctuations that are unrelated to relative liquidity. Let the market share measure MSit for security i during week t be calculated as the ratio of the par value traded in bond i to the total par value traded 10 Shulman et al. (1993) uses trading frequency and Houweling et al. (2002) use the number and dispersion of quotes per day. 11 Other variables that have been used include the volatility of interest rates (Kamara, 1994) and the percentage growth of mutual funds (Fridson and Jónsson, 1995). 12 As it happens, in mid-2002 (near the end of our sample) Spain ultimately became confident enough to approach the market with large enough initial tranches such that new issues immediately become benchmarks. 13 We choose among volume-based measures since the Banco de España database does not include bid– ask quotes. A. Dı́az et al. / Journal of Banking & Finance 30 (2006) 1309–1332 1317 by all outstanding issues. The MSit variable allows us to compare the degree of liquidity among issues and to monitor the evolution of the liquidity of a given issue throughout its life. We divide the life of a Spanish Treasury bond issue into three status stages. We term the first stage to be the ‘‘pre-benchmark’’ period – beginning at the issue of a new bondÕs initial tranche (the bond is also by definition the ‘‘on-the-run’’ issue at this time). This pre-benchmark period covers the time during which the issueÕs market share increases with age, but still lies below the market share of the former onthe-run bond. The second stage is the ‘‘benchmark’’ period. This stage corresponds to the period during which the issue has the highest market share among all outstanding issues of the same original maturity. The last stage is the ‘‘seasoned’’ (post-benchmark) period. The seasoned stage corresponds to the period beginning the week that the particular bondÕs market share is eclipsed (by a newer and now sufficiently liquid issue) and ends at the bondÕs maturity. 3.3. The impact of EMU preparations on liquidity SpainÕs debt management policy changes generated important shifts in issuance tranches, outstanding issue sizes and the evolution of bond auction status stages. Panel A of Table 2 reports two sets of summary statistics on issue par amounts for both individual tranches and total issue sizes for each of the three main bond maturity sectors: 10-, 5- and 3-years. The 1993–1997 and 1998–2002 sample splits were chosen to reflect the two different Spanish issuance policy regimes. The latter subsample begins after the 1997 shift toward larger issue sizes. Note that tranche sizes were reasonably similar across the two regimes. However, the number of tranches increased so that the par amounts outstanding after the last tranche essentially doubled across board. The shifts in issuance policy also had an important effect on bond status. Panel B of Table 2 presents summary statistics on the evolution of the lengths of the crucial pre-benchmark and benchmark stages in a bondÕs life cycle. The average length of the pre-benchmark stage is similar across issuance regimes, but the length of the benchmark stage is considerably larger during the post-1997 period. 3.4. Modeling the market share function So far, we have classified bond liquidity using three discrete status categories. However, individual bond market shares of trading may be more generally modeled as smooth, non-linear functions of bond age. Here, we posit a parsimonious function to describe the behavior of individual bond market share (MSit) as a function of bond age (Ageit):14 14 Eq. (1) is inspired by forms arising from actuarial research on human mortality (see Heligman and Pollard, 1980). That literature uses this functionÕs ‘‘hump’’ to capture the impact of traffic accidents on mortality rates of 15–25-year-olds within a general mortality-versus-age relationship. 1318 Table 2 Effects of Spanish debt management changes on issuance and bond benchmark status Years to maturity in each tranche Minimum Maximum Amount outstanding after the first tranche Amount outstanding after the last tranche Panel A: Evolution of the issuance tranches and amounts outstanding (million € par value) by sector 1993–1997 3-year bond 2.21 3.54 3.09 900 5-year bond 4.21 5.54 5.00 934 10-year bond 9.42 10.54 10.01 890 15-year bond 12.71 15.63 14.20 245 30-year bond – – – – 1075 1285 1198 676 – – 1998–2002 3-year bond 5-year bond 10-year bond 15-year bond 30-year bond 1350 1686 2371 1691 2128 10,769 11,152 13,737 10,599 7532 2.22 4.29 8.96 12.64 28.24 Number of new issues 3.98 5.98 11.06 15.65 31.54 3.14 5.12 10.21 14.65 30.06 824 881 1031 575 620 Weeks between adjacent auctions Weeks that a bond keeps pre-benchmark status Minimum Weeks that a bond keeps benchmark status Average Minimum Maximum Average Minimum Maximum Average Panel B: Evolution of bond auction status by sector 1993–1997 10-year bond 8 13 43 5-year bond 8 6 82 3-year bond 9 6 69 33 30 26 3 5 1 28 81 67 16 27 17 12 2 4 49 87 92 33 41 37 1998–2002 10-year bond 5-year bond 3-year bond 50 54 38 3 0 0 34 21 27 16 16 17 31 32 32 58 75 68 40 53 53 5 5 4 38 31 4 Maximum 5385 6076 5830 4171 71 74 56 A. Dı́az et al. / Journal of Banking & Finance 30 (2006) 1309–1332 Average amount per auction (million €) Average A. Dı́az et al. / Journal of Banking & Finance 30 (2006) 1309–1332 2 Ageit MSit ¼ b1 exp½b2 ðAgeit  b3 Þ  þ b4  b5 þ uit . 1319 ð1Þ The first term allows for a hump in the liquidity profile during the first few years of bond life. The second term is a decreasing exponential function that describes the declining trading activity of the bond as it approaches maturity. The third term is a random error. The parameters in Eq. (1), with expected signs given in parentheses, can be interpreted as follows: 1. b1 measures the degree of concentration of trading activity in the benchmark bond, i.e., the size of the hump (b1 > 0); 2. b2 is inversely related to the length of the period over which a bond keeps the benchmark status, i.e., the width of the hump (b2 > 0); 3. b3 is the bond age at which the functionÕs first term (the liquidity hump) has the highest amplitude (b3 > 0); 4. b4 is the initial value for the exponential function component (b4 > 0); 5. b5 relates to the speed at which a seasoned bondÕs trading activity changes with time (1 P b5 > 0). For positive values of all parameters, the MS(Æ) function is also positive. We apply equation (1) to weekly data on individual bond issue shares of trading volume for all original-issue 10-, 5- and 3-year bonds in our database.15 Table 3 presents the parameter estimates for Eq. (1) for the two subsamples suggested by the changes in Spanish Treasury issuance policy. All reported individual t-statistics embody the Newey–West correction. Consider first the estimates for 10-year bonds over the 1993–1997 period presented in Panel A. The regressionÕs adjusted R-square of 71.2% reveals that, through the functional form given by Eq. (1), bond age does a very good job of explaining market share in the 10-year sector. As expected, all five coefficient estimates are positive in sign and significantly different from zero (all individual coefficient t-statistics have p-values of 0.00). The most interesting individual ^ ¼ 0.67 and b ^ ¼ 18.47. The b ^ estimate reveals that the peak in a estimates are b 3 1 3 typical 10-year bondÕs market share occurs two-thirds of a year after its first issue ^ implies that the shift in the 10-year bondÕs market share date. The estimate for b 1 versus its baseline value at this peak point is about 18.5%.16 The 10-year sectorÕs results for the 1998–2002 period are qualitatively similar. However, the key coefficients exhibit very interesting quantitative shifts. The new estimate for b1 implies that the peak shift in the 10-year bondÕs market share is now 23.5% (about 5% higher versus the earlier period) indicating that trading 15 While we prefer the market share measure, the results from using raw volume as the dependent variable were qualitatively similar. However, see Footnote 16 below. 16 Some experimentation showed that the reported parameter estimates are robust to the choice of specific starting values in the non-linear least squares estimation procedure. Ironically, the estimates generated when using raw volume as the dependent variable are highly sensitive to the initial set of parameters. 1320 A. Dı́az et al. / Journal of Banking & Finance 30 (2006) 1309–1332 Table 3 Estimates of the market share life cycle function Sample period: 1993–1997 Sample period: 1998–2002 Coefficients t-Statistics Coefficients t-Statistics Panel A: 10-year bonds b1 b2 b3 b4 b5 Adjusted R2 (%) # Observations 18.47 4.16 0.67 2.16 0.89 71.2 1686 (34.11) (12.00) (37.96) (6.80) (22.95) 23.50 2.43 0.83 4.85 0.69 76.3 3193 (24.88) (8.75) (35.68) (4.03) (16.65) Panel B: 5-year bonds b1 b2 b3 b4 b5 Adjusted R2 (%) # Observations 5.95 0.56 0.64 1.76 0.86 36.2 2020 (3.64) (2.52) (3.75) (0.81) (3.14) 10.43 3.00 0.89 4.69 0.75 63.6 1700 (15.74) (5.85) (32.27) (5.46) (20.54) Panel C: 3-year bonds b1 b2 b3 b4 b5 Adjusted R2 (%) # Observations 7.87 0.67 0.53 0.16 1.15 38.9 1502 (5.53) (2.78) (6.39) (0.10) (0.31) 11.11 1.83 0.88 1.06 1.10 60.5 1102 (18.75) (6.84) (32.65) (1.43) (4.35) Non-linear least squares regressions of the market share (MSit) of the bond i during week t on Ageit of the bond for two sample periods: 1993–1997 and 1998–2002. MSit is the trading volume of bond i during week t divided by total trading volume of all Treasuries during week t. The regression equation is Ageit MSit ¼ b1 exp½b2 ðAgeit  b3 Þ2  þ b4  b5 þ uit . Newey–West adjusted t-statistics are in parenthesis. activity became more concentrated in the benchmark bond in this later period. Moreover, the market share peak in the later period comes later (0.83 years versus ^ is shorter in the second period, suggesting that the 0.67 years). Furthermore, b 2 ^ is lower in the second benchmark bond tends to keep this status longer. Finally, b 5 period, indicating that the liquidity of seasoned bonds decays even faster than in the earlier period. Panels B and C of Table 3 present corresponding results for the original-issue 5-year and original-issue 3-year bond sectors, respectively. The adjusted R-square for each of the regressions reveals that bond age also provides important explanatory power for market share in these maturity sectors. The explanatory power is much larger in the later 1998–2002 sample. Moreover, as in the 10-year sector, the estimates for b1 imply that trading activity for both the 5- and 3-year sectors became A. Dı́az et al. / Journal of Banking & Finance 30 (2006) 1309–1332 1321 more concentrated in the benchmark bond in the later period. Moreover, in the later period, the market share peaks appear later in bond life as well. However, the estimates for b1 show that the magnitudes of the benchmark effects for 5- and 3-year bonds in both periods are less than one-half of the corresponding values for 10-year bonds.17 The point estimates presented in Table 3 suggest increases in both the height and width of the liquidity functionÕs ‘‘hump’’ in each sector in the later period. To examine whether the issuance policy shifts in mid-1997 had any effect on the structure of Spanish Treasury bond liquidity, we apply the Chow test for structural change to the coefficients in Eq. (1) after December 1997. The p-values for the Chow test statistics are 0.00 in all three sectors. Schmidt and Sickles (1977) argue that the Chow statistic may overstate the true test size in the presence of heteroscedastic residuals. Thus, we also review the results for the individual equation coefficient estimates b1 through b5 and their reported Newey–West corrected estimated coefficient standard errors. For each sector and subsample, consider the two standard error bounds around the estimated equation coefficients. For each sector there is at least one b estimate for which the two standard error bounds do not overlap across subsamples.18 Taken together, these results are consistent with the hypothesis that the mid-1997 changes in issuance policy induced significant shifts in the liquidity structure of the Spanish Treasury market. 4. Liquidity premiums and the impact of EMU In the search for liquidity premiums in bond pricing, care must be taken to control for other determinants of bond value. Previous researchers have used a variety of methods to isolate the value impacts of liquidity on bond pricing. For US bond markets, Amihud and Mendelson (1991), Kamara (1994) and Strebulaev (2001) have used the yield spread between a bill and a bond with similar term to maturity as the appropriate variable to be explained. Warga (1992) uses the mean return difference between a portfolio of seasoned bonds and a portfolio of the most recently issued securities with similar duration. Goldreich et al. (2005) study on-the-run versus off-the-run yield spreads adjusted for coupon and yield curve effects. Dı́az and Skinner (2001) use the differences between the yield-to-maturity of a bond and its theoretical yield as given by an explicit term structure model. Fleming (2001) also examines a yield spread calculated as the difference between the observed yield of 17 The point estimate of b5 for the 3-year note sector is greater than 1.0 in each subsample. Given the other parameter estimates, this value implies that a 3-year bondÕs market share would be greater than 100% if bond age were extrapolated over horizons greater than 4 years. Of course, original issue 3-year bonds would be ‘‘dead’’ by this time. Moreover, in either subsample, the hypothesis that b5 = 1.0 cannot be rejected at standard significance levels and, if we impose the constraint that b5 = 1.0, the re-estimated values of b1 through b4 are very close to the original values. 18 Specifically, the two standard error bounds do not overlap in four of five parameter cases for the 10year; one of five for the 5-year; and two of five for the 3-year. 1322 A. Dı́az et al. / Journal of Banking & Finance 30 (2006) 1309–1332 the on-the-run security and that predicted by a term structure model estimated with off-the-run bond prices. He finds that this yield differential is consistently correlated with a number of other liquidity proxies widely used in the literature. We begin our study of the yield impact of liquidity by estimating a daily term structure of interest rates using actual mean daily MDPA Treasury transactions prices. We include all the spot transactions that took place with Treasury bills and bonds during the day for all issues with a daily trading volume of at least than €3 million (500 million pesetas) and terms to maturity between 15 days and 15 years. We also include the one-week general collateral repo market interest rate to provide a liquid point at the very front of yield curve. We employ Nelson and SiegelÕs (1987) exponential model to fit the daily term structures. These daily term structure estimates do not incorporate any specific liquidity effects. Thus, the theoretical values for all bonds generated from these estimations are those produced by discounting coupon and principal payments according to fitted term structures that reflect an average liquidity level. The differences between actual bond yields and the theoretical ones can be understood as a liquidity effect plus an error term due to other factors. 4.1. Liquidity value versus current and expected future market share and auction status As recently emphasized by Goldreich et al. (2005), the price of a security depends on the flow of liquidity services generated over its entire life. Lifetime liquidity involves not only a securityÕs current of liquidity, but also the expected future path of liquidity. We continue to use a bondÕs share of trading volume as the relevant index of liquidity. Conveniently, the liquidity function of the previous section relates a bondÕs market share of trading volume to bond age, a deterministic variable. Thus, at any point in time, this function can be used to project the future path of liquidity of any individual bond. In turn, this approach permits identification of both current and expected future lifetime liquidity, and subsequently allows analysis of their separate impacts on bond valuation. Specifically, we define Et[MSi,t+j] as the week t conditional expectation of the market share of bond i during some future week t + j. Using Eq. (1), Et[MSi,t+j] can be expressed as ^ expbb ^ ðAge ^ 2 ^ ^Agei;tþj . Et ½MSi;tþj  ¼ b 1 2 i;tþj  b3 Þ c þ b4  b5 ð2Þ Furthermore, we define MSi;t;tþmit as the average lifetime expected market share for bond i from week t+1 through maturity week t + mit mit   1 X MSi;t;tþmit ¼ Et MSi;tþj ; ð3Þ mit j¼1 where mit is the number of weeks remaining until maturity for bond i as of the current week t. We invoke rational expectations on the part of investors and interpret equation (3) to incorporate the expected future market share function (2). Furthermore, based 1323 A. Dı́az et al. / Journal of Banking & Finance 30 (2006) 1309–1332 Market share of trading volume (%) 30 Current MS (1998–2002) 25 20 Current MS (1993–1997) 15 Avg. MS to maturity (1993–1997) 10 Avg. MS to maturity (1998–2002) 5 0 0.0 0.6 1.2 1.8 2.5 3.1 3.7 4.3 4.9 5.5 6.2 6.8 Bond Age (years) 7.4 8.0 8.6 9.2 9.8 Fig. 1. Average expected market share to maturity versus current market share for 10-years bonds based upon parameter estimates for the market share equations reported in Table 3. on the evidence of a regime shift in 1997, we match the appropriate set of parameters for the market share function from each subsample presented in Table 3 to generate the corresponding MSi;t;tþmit for that same subsample for use in our valuation equations below. In the spirit of a rational expectations model, this choice presumes that investors understood the nature and consequences of the shifts in SpainÕs debt management policies at the time they were publicly announced in 1997. Fig. 1 provides some insight into the MSi;t;tþmit variable for a newly issued 10-year-to-maturity bond. Fig. 1 plots two fitted series from the market share regressions for Eq. (1) generated using the parameter estimates for both the 1993–1997 and 1998–2002 sample periods. These series correspond to Et[MSi,t+j] of Eq. (2) for each subsample. Fig. 1 also plots the expected market share to maturity variable (i.e., MSi;t;tþmit ) generated for each subsample by Eq. (3) using the corresponding Et[MSi,t+j] profile. For each subsample, note the clearly defined differences between a 10-year bondÕs current market share and its average expected future market share to maturity, especially over the first year after initial issuance. Table 4 presents estimates of the following regressions for weekly data in both of our subsamples for 10-year sector issues with at least one year still remaining until maturity: Y dit ¼ /0 þ c1 OTRDoit þ /1 MSoit þ /2 MSi;t;tþmit þ mit ; Y dit ¼ /0 þ c2 PreBDoit þ c3 BDoit þ /1 MSoit Y dit ¼ /0 þ /1 MSoit þ /2 MSi;t;tþmit þ mit ; þ /2 MSi;t;tþmit þ mit ; ð4Þ ð5Þ ð6Þ where Y dit is the weekly average of daily differences between the actual and theoretical yields to maturity of bond i during week t; MSi;t;tþmit is defined as above; MSoit is an 1324 A. Dı́az et al. / Journal of Banking & Finance 30 (2006) 1309–1332 Table 4 Liquidity value impacts of current 10-year bond market share (MSoit ) and expected average future market share (MSi;t;tþmit ) with and without on-the-run auction status (OTRDoit ), pre-benchmark status (PreBDoit ) and benchmark status (BDoit ) dummy variables Sample period: 1993–1997 Sample period: 1998–2002 2 Coefficients t-Statistics Adjusted R (%) Coefficients t-Statistics Adjusted R2 (%) Constant OTRDit MSit MSi;t;tþmit 3.08 6.77 0.00 1.86 (15.90) (12.33) (0.10) (12.24) 20.43 3.12 0.18 0.06 2.30 (30.29) (0.50) (7.22) (25.87) 24.94 Constant PreBDit BDit MSit MSi;t;tþmit 3.08 8.39 5.81 0.00 1.86 (16.45) (9.66) (7.60) (0.09) (12.22) 18.76 3.12 1.79 0.87 0.06 2.30 (30.36) (2.50) (1.69) (6.81) (26.21) 25.42 Constant MSit MSi;t;tþmit 3.08 0.00 1.86 (13.89) (0.09) (11.01) 9.71 3.12 0.06 2.30 (30.28) (7.17) (25.80) 24.96 OLS regressions of the week t average difference between the actual and Nelson–Siegel theoretical yields to maturity for bond i (Y dit ) on the various defined variables for two sample periods: 1993–1997 and 1998– 2002. The regression equations are Y dit ¼ /0 þ c1 OTRDoit þ /1 MSoit þ /2 MSi;t;tþmit þ mit ; ð4Þ Y dit ¼ /0 þ c2 PreBDoit þ c3 BDoit þ /1 MSoit þ /2 MSi;t;tþmit þ mit ; ð5Þ Y dit ¼ /0 þ /1 MSoit þ /2 MSi;t;tþmit þ mit . ð6Þ Newey–West adjusted t-statistics are in parentheses and o superscript indicates an orthogonalized variable. orthogonalized version of MSit;19 OTRDoit is an orthogonalized version of a dummy variable set equal to 1.0 if bond i is the current on-the-run issue and zero otherwise; PreBDoit , is an orthogonalized version of dummy variable set equal to 1.0 if bond i is currently in its pre-benchmark stage and zero otherwise; and BDoit is an orthogonalized version of dummy variable set equal to 1.0 if bond i is currently in its benchmark stage and zero otherwise.20 These regressions explicitly distinguish between the contributions of the current versus expected future market share variables for relative bond value and permit tests of the marginal contributions of these market share variables to both 2-stage and 3-stage auction status dummy variables. Orthogonalized variables are used to control for the correlation among the candidate explanatory Thus, MSoit in (4) through (6) is the residual eit from the regression MSit = a0 + a1 + MSi;t;tþmit + eit. The OTRD0it variable is the residual eOTRDit from the regression OTRD i t = b 1 + b2MSit + b3MSi;t;tþmit þ eOTRDit ; and the other two orthogonalized dummy variables are defined in analogous fashion. 19 20 A. Dı́az et al. / Journal of Banking & Finance 30 (2006) 1309–1332 1325 variables. We expect the signs of /1 and /2 to be negative: the larger a bondÕs market share, the higher its liquidity, and so the lower its yield. Table 4 reports estimates of regressions (4) through (6) for original issue 10-year bonds. Here, incorporating the full explicit life cycle function adds significant explanatory power to the bond status dummies for most cases. Our results also reveal that expected future liquidity is much more important – in both magnitude and statistical significance – than current liquidity for explaining Spanish Treasury bond values. These patterns are most easily seen in regression (6), where only the market share variables appear. The explanatory power of Eq. (6) is quite high even though it excludes both forms of status dummy variables. However, note that in regressions (4) and (5), the new market share variables do not completely eliminate the contribution of the status dummy variables to overall explanatory power. Nevertheless, the contributions of the dummy variables are inconsistent across subsamples in sign and/or significance. In contrast, the key MSi;t;tþmit variableÕs coefficient estimates are always properly signed and highly significant.21 4.2. Impacts of other bond-specific characteristics As discussed in Section 2, a key component of the Spanish TreasuryÕs preparations for European Monetary Union was a series of exchange auctions designed to help quickly build large-sized benchmark issues. These exchange auctions replaced seasoned premium, high-coupon issues with new market-coupon bonds. Targeting premium, high-coupon issues as candidates for these exchanges would have been a sensible choice if such bonds traded cheaply in the market because of tax or other reasons. Here we investigate whether discounts and premiums from par had discernable impacts on Spanish Treasury bond pricing. We use Dit = Max(0, 100  Vit) and Pit = Max(0, Vit  100) to measure bond discounts and premiums, respectively, where Vit is the ‘‘clean’’ price of the ith bond. We also use C oit , an orthogonalized measure of the coupon rate (CRit) of the ith bond, to pick up any coupon-related 21 For original issue 5-year bonds (results not shown here for space reasons), the explanatory power of Eq. (6) is reasonably high even though it excludes both forms of status dummy variables. Again, expected future market share has a significantly more powerful impact on liquidity value than does current market share. In fact, in the first subsample, the current market share variable is wrongly signed and not significantly different from zero. Again, the results for regressions (4) and (5) suggest that the new market share variables do not completely eliminate the contribution of the status dummy variables to overall explanatory power. For original issue 3-year bonds (not shown here), term structure deviations are harder to explain using our set of liquidity proxies. At least for the 1993–1997, the impact of MSi;t;tþmit is properly signed and highly significant. (However, the coefficient for current market share is perversely signed and significant.) The results for the 1998–2002 subsample are disappointing since the overall explanatory power is low and the coefficient on MSi;t;tþmit in all three specifications is wrongly signed, though not statistically different from zero. The puzzles for the 3-year sector are not confined to the market share variables. The traditional status dummy approach also leads to wrongly signed and insignificant coefficients in the 1998–2002 subsample. 1326 A. Dı́az et al. / Journal of Banking & Finance 30 (2006) 1309–1332 Table 5 Value impacts of bond characteristics on 10-year bonds: price discount versus par (Dit); price premium versus par (Pit); coupon rate (C oit ); and strippable issue dummy variable (S oit ) Sample period: 1993–1997 Sample period: 1998–2002 Coefficients t-Statistics Adjusted R2 (%) Coefficients t-Statistics Adjusted R2 (%) Constant OTRDit MSit MSi;t;tþmit Couponi ðC oit Þ Discountit (Dit) Premiumit (Pit) Strippableit ðS oit Þ 3.28 6.30 0.01 0.99 1.04 0.29 0.05 (10.77) (11.97) (0.38) (4.68) (9.80) (12.37) (3.90) 34.24 1.71 0.36 0.02 1.52 0.18 0.27 0.08 0.04 (12.79) (0.85) (2.45) (19.02) (2.47) (11.61) (10.66) (0.19) 31.02 Constant PreBDit BDit MSit MSi;t;tþmit Couponi ðC oit Þ Discountit (Dit) Premiumit (Pit) Strippableit ðS oit Þ 3.47 6.30 3.47 0.00 1.14 0.88 0.29 0.05 (10.57) (6.52) (4.84) (0.13) (5.35) (9.20) (9.09) (3.85) 29.59 1.68 2.17 0.82 0.02 1.50 0.22 0.26 0.08 0.01 (12.72) (3.17) (1.69) (2.41) (19.18) (3.04) (11.43) (10.59) (0.02) 31.26 Constant MSit MSi;t;tþmit Couponi ðC oit Þ Discountit (Dit) Premiumit (Pit) Strippableit ðS oit Þ 3.89 0.00 1.18 0.84 0.37 0.07 (11.79) (0.24) (5.63) (8.79) (12.36) (5.72) 25.68 1.73 0.02 1.53 0.17 0.27 0.08 0.02 (13.69) (2.50) (19.77) (2.51) (12.88) (10.70) (0.10) 31.03 OLS regressions of the week t average difference between the actual and Nelson–Siegel theoretical yields to maturity for bond i (Y dit ) on the various defined variables for two sample periods: 1993–1997 and 1998– 2002. The regression equations are Y dit ¼ /0 þ c1 OTRDoit þ /1 MSoit þ /2 MSi;t;tþmit þ w1 C oit þ w2 Dit þ w3 P it þ w4 S oit þ mit ; Y dit ¼ /0 þ c2 PreBDoit þ c3 BDoit þ /1 MSoit þ /2 MSi;t;tþmit þ w1 C oit þ w2 Dit þ w3 P it þ w4 S oit þ mit ; Y dit ¼ /0 þ /1 MSoit þ /2 MSi;t;tþmit þ w1 C oit þ w2 Dit þ w3 P it þ w4 S oit þ mit . Newey–West adjusted t-statistics are in parentheses. The o superscript indicates an orthogonalized variable. (See notes to Table 4 for additional variable definitions.) effects not captured by Dit and Pit.22 Finally, the introduction of the Spanish strips market in January 1998 may have caused strippable bonds to trade at higher values than non-strippable issues. To investigate whether strip-related effects on value exist, we define the dummy variable Sit equal to +1 if the ith bond is strip market eligible 22 Thus, C oit is the residual eCit from the regression CRit ¼ c0 þ c1 Dit þ c2 P it þ eC it . A. Dı́az et al. / Journal of Banking & Finance 30 (2006) 1309–1332 1327 and zero otherwise. We include S oit , an orthogonalized measure of the strip eligibility dummy variable for the ith bond, in the 1998–2002 sample period regressions.23 Table 5 reports estimates of the following regressions for 10-year sector bonds in each of our subsamples: Y dit ¼ /0 þ c1 OTRDoit þ /1 MSoit þ /2 MSi;t;tþmit þ w1 C oit þ w2 Dit þ w3 P it þ w4 S oit þ mit ; ð7Þ Y dit ¼ /0 þ c2 PreBDoit þ c3 BDoit þ /1 MSoit þ /2 MSi;t;tþmit þ w1 C oit þ w2 Dit þ w3 P it þ w4 S oit þ mit ; Y dit ¼ /0 þ /1 MSoit þ /2 MSi;t;tþmit þ w1 C oit þ w2 Dit þ w3 P it þ w4 S oit þ mit . ð8Þ ð9Þ Generally, the results point to statistically significant differences in the valuation of discount versus premium bonds. Note especially the results for the 1993–1997 sample period that reflect the recent experience observed by the Spanish Treasury at the time of its exchange auction decision-making. The Dit variable is highly significant and indicates that investors favored bonds priced below par. Surprisingly, the coefficient for Pit is negative and significantly different from zero; but it is small in magnitude (about one-sixth the size of the coefficient on Dit). The coefficient on C oit is positive and highly significant, indicating an additional pricing bias against high-coupon bonds. These results confirm that discount bonds were favored over high-coupon, premium bonds.24 Thus, the estimates lend support to the Spanish TreasuryÕs decision to target high-coupon, premium bonds in its debt exchanges. Moreover, some changes in the magnitude, sign and significance of the Dit, Pit and C oit variables from the 1993–1997 sample to the 1998–2002 sample are quite interesting. For the 10-year bonds, both the magnitude and significance of the C oit fall precipitously, and the impact of Pit, while turning positive, remains small in magnitude.25 These patterns may indicate that market impacts of such bond-specific characteristics on value in the Spanish market have decreased since European Monetary Union ushered in a broader class of euro-based fixed income investors.26 23 Thus, S oit is the residual eS it from the regression S it ¼ d 0 þ d 1 Dit þ d 2 P it þ eS it . The results for the 5-year bond sector for this same 1993–1997 sample period reveal investor biases favoring discount bonds over high-coupon, premium bonds. Here, the estimated coefficients on the Dit and C oit variables are each statistically significant. The Pit variable is statistically insignificant. In the 1993–1997 sample, the results for the 3-year bond sector show that investors favored discount bonds (i.e., a statistically significant negative impact for Dit on bond yields) versus high-coupon, premium bonds (both C oit and Pit have statistically significant positive slope coefficients). 25 For the 5-year bonds, the coefficients on the Dit, and C oit variables switch signs and lose some significance. For 3-year bonds, the Dit variableÕs effect switches sign. Finally, for 3-year bonds, the estimated coefficient on S oit is both negatively signed and statistically significant, indicating a value premium for the strip feature. Evidence on the impact of the strip feature in the other sectors is mixed. The S oit variable is marginally significant for 5-year bonds and statistically insignificant for 10-year bonds. 26 Recall from Section 2 that bond trades for non-residents and institutional investors have been settled free of withholding tax for since 1999. 24 1328 A. Dı́az et al. / Journal of Banking & Finance 30 (2006) 1309–1332 4.3. The impact of EMU on Spanish bond market volatility Table 6 presents an expanded analysis of European Monetary UnionÕs effects on Spanish bond market pricing along two distinct dimensions regarding market volatility. We first quantify the impact of European Monetary Union on Spanish bond market yield volatility. Clearly, one major projected benefit for Spain under EMU was to be a dampening of market yield volatility in light of the elimination of currency crisis risk from bond pricing. Our sample of daily fitted term structures provides a particularly clean way in which to measure the volatility impact of EMU at different points along the zero coupon yield curve. Specifically, we create a daily time series of fitted zero coupon bond yields for annual maturities between 2 and 10 years. We then compute subsample means and standard deviations of these yield series and test for equality of yield variances across the two subsamples. While European Monetary Union actually officially began on January 1, 1999, market participants are widely viewed to have priced this merger as a fait accomplie before this date. We have kept the same two sample splits used before, interpreting that the market had fully priced in SpainÕs entry into EMU 1 year ahead of schedule.27 Panel A of Table 6 presents our yield volatility results based upon daily data for the 1993–1997 and 1998–2002 sample periods. The columns report estimated sample means and standard deviations of both yield levels (expressed in percentage points) and first-differences in yields. A profound downward shift in average yield levels and standard deviations across the zero coupon yield curve can be observed in the latter periods. Formal tests based upon the first-differenced data confirm that the hypothesis of equal variances in the 1993–1997 and 1998–2002 periods is easily rejected for all terms to maturity.28 For example, the standard deviation of yield first-differences for 5-year zero coupon bonds nearly halved. Clearly, as was anticipated by the planÕs early proponents, European Monetary Union has led to a dramatic fall in Spanish bond market yield volatility. Our second volatility investigation examines whether European Monetary Union has led to more efficient relative pricing. We gauge relative pricing efficiency through the residual variance of the Spanish bond yield regression equation (8). In particular, we examine the residuals from the estimated Y dit regressions (8) of Table 5 and test for equality of residual variances across the two subsamples. The null hypothesis is that the residual variances from the Y dit regressions (8) in the 1993–1997 and 1998–2002 subsamples are equal. We interpret this null hypothesis to mean that European Mon- 27 On March 25, 1998, the European Commission had recommended that 11 countries – Austria, Belgium, Finland, France, Germany, Ireland, Italy, Luxembourg, the Netherlands, Portugal and Spain – met the necessary conditions to adopt the single currency. On May 2, 1998, EU finance ministers officially announced the bilateral parities between the currencies of the euro zone. Of course, reasonable arguments exist for dating the sample split even before the start of 1998. The convergence of forward deposit interest rates between Spain and Germany was nearly complete by mid-1997. 28 Identical inferences come from other versions of tests for equality of variances in two samples (e.g., the standard F-test, BartlettÕs test, Siegel–TukeyÕs test and LeveneÕs test). Table 6 Impacts of European Monetary Union on Spanish Treasury market yield volatility and bond relative pricing efficiency Tests of equality of variances in the two sample periods for first-differences of daily fitted zero coupon yields Term Panel A 2 3 4 5 6 7 8 9 10 First sample period: 1993–1997 Second sample period: 1998–2002 First sample period: 1993–1997 Second sample period: 1998–2002 Mean Standard deviation Mean Standard deviation Mean · 102 Standard deviation Mean · 102 Standard deviation 7.72 7.90 8.09 8.26 8.39 8.50 8.59 8.66 8.72 2.68 2.69 2.67 2.63 2.59 2.54 2.49 2.45 2.40 4.03 4.24 4.43 4.60 4.75 4.88 4.99 5.09 5.17 0.68 0.64 0.60 0.57 0.53 0.50 0.48 0.45 0.44 0.74 0.70 0.67 0.64 0.61 0.58 0.56 0.55 0.53 0.0859 0.0819 0.0807 0.0816 0.0811 0.0787 0.0760 0.0755 0.0797 0.14 0.13 0.13 0.13 0.13 0.13 0.14 0.14 0.14 0.0500 0.0424 0.0437 0.0444 0.0440 0.0430 0.0421 0.0422 0.0437 Test for equality of variances Brown–Forsythe prob > F 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 Sample meansa and standard deviations for residualsfrom Y dlt regression estimates of Eq. (8) of Table 5 Tests of equality of variances in the two sample periods for residuals from Y dit regression estimates of Eq. (8) of Table 5 Original issue maturity sector (years) Brown–Forsythe prob > F Panel B 10 5 3 a First sample period: 1993–1997 Second sample period: 1998–2002 Mean · 102 Standard deviation Mean · 102 Standard deviation 0.000 0.000 0.000 0.035 0.060 0.072 0.000 0.000 0.000 0.031 0.034 0.034 A. Dı́az et al. / Journal of Banking & Finance 30 (2006) 1309–1332 Sample means and standard deviations for levels of fitted zero coupon yields (in %) 0.000 0.000 0.000 Regression residual sample means equal 0 by construction. 1329 1330 A. Dı́az et al. / Journal of Banking & Finance 30 (2006) 1309–1332 etary Union had no detectable impact on the dispersion of term structure arbitrage trading opportunities in the Spanish market. For each of the three original issue maturity sectors, Panel B of Table 6 presents tests of equality of residual variances from the Y dit regressions (8) in the 1993–1997 and 1998–2002 subsamples. The valuation equation residuals have lower estimated standard deviations in the latter period. The test statistics have p-values that indicate strong rejections of the null hypothesis of equal residual variances. We interpret these results as evidence that the new class of euro-based investors created under European Monetary Union has significantly increased pricing efficiency in the Spanish bond market.29 5. Summary and conclusions This paper has examined liquidity and volatility in the Spanish Treasury bond market in the context of debt policy shifts engineered by the Spanish government in preparation for entrance into European Monetary Union. Empirically detectable impacts of SpainÕs mid-1997 debt management initiatives exist for both trading activity and debt market valuation. We interpret these impacts through shifts in the coefficients of a liquidity life cycle model relating individual Treasury bond market share to a bondÕs age (the time since its initial auction). Test for shifts in this market share function as a result of the TreasuryÕs debt policy innovations clearly reject the hypothesis of no structural change in the post-initiatives sample period. We also estimate the structure of liquidity premiums in the different maturity sectors within the Spanish bond market. We investigate liquidity effects within a framework that values the lifetime flow of liquidity services and distinguishes between a securityÕs current liquidity and its average expected future liquidity. Our empirical results for 10- and 5-year Spanish Treasury bond sectors reveal statistically significant valuation impacts of expected future liquidity on current market value. Our expected future liquidity measure adds significant explanatory power to the traditional auction status dummy variable approach to assessing bond liquidity value. Our results for data from 1993 to 1997 detect statistically significant pricing biases confirming that discount bonds were favored over high coupon, premium bonds. These results lend support to the Spanish TreasuryÕs tactical decision to target high-coupon, premium bonds in its pre-EMU debt exchanges. Finally, we examine the impacts of European Monetary Union on volatility of yields in the Spanish Treasury market. First, we use our basic fitted term structures to show that the standard deviation of zero coupon bond yields declined dramatically after the market began pricing European Monetary Union as certain to occur. Formal tests based on the first-differences of yields strongly reject the null hypothesis of no change in variance from the pre-EMU period. Such an impact on volatility is 29 Changes in the Spanish tax regime beginning in 1999 (e.g., newly-issued Treasuries began trading free of withholding tax for domestic institutions) may also have helped increase pricing efficiency. A. Dı́az et al. / Journal of Banking & Finance 30 (2006) 1309–1332 1331 generally acknowledged (and had been forecasted by EMUÕs early supporters). It is less widely recognized that European Monetary Union has also led to more efficient relative pricing in the Spanish Treasury bond market. Our results reveal that the residual variance of our liquidity and bond characteristic-augmented yield regressions fell sharply between our pre-EMU and post-EMU samples. Such improved pricing efficiency may be attributed to an important byproduct of European Monetary Union: the creation of a much larger universe of euro-based fixed income investors willing to focus attention on trading opportunities in Spain. References Alonso, F., Blanco, R., del Rı́o, A., Sanchis, A., 2004. Estimating liquidity premia in the Spanish government securities market. European Journal of Finance 10 (6), 453–474. Amihud, Y., Mendelson, H., 1991. Liquidity, maturity, and the yields on US Treasury securities. Journal of Finance 46 (4), 1411–1425. Carayannopoulos, P., 1996. A seasoning process in the US Treasury bond market: The curious case of newly issued ten-year notes. Financial Analysts Journal 52 (1), 48–55. Chakravarty, S., Sarkar, A., 1999. Liquidity in US fixed income markets: A comparison of the bid–ask spread in corporate, government and municipal bond markets. Working paper, Federal Reserve Bank of New York. Chen, L., Wei, J., 2001. An indirect estimate of transaction costs for corporate bonds. Working Paper, University of Toronto. Codogno, L., Favero, C., Missale, A., 2003. Yield spreads on EMU government bonds. Economic Policy 18, 503–532. Dı́az, A., Navarro, E., 2002. La prima de liquidez en la Deuda del Estado española. Revista de Economı́a Aplicada (10), 23–58. Dı́az, A., Skinner, F.S., 2001. Estimating corporate yield curves. The Journal of Fixed Income 11 (2), 95– 103. Duffee, G.R., 1998. The relation between Treasury yields and corporate bond yield spreads. Journal of Finance 53 (6), 2225–2241. Elton, E.J., Green, T.C., 1998. Tax and liquidity effects in pricing government bonds. Journal of Finance 53 (5), 1533–1562. Fisher, L., 1959. Determinants of risk premiums of corporate bonds. Journal of Political Economy 67 (3), 217–237. Fleming, M.J., 2001. Measuring Treasury market liquidity. Working Paper, Federal Reserve Bank of New York, June. Fleming, M.J., Remolona, E.M., 1999. Price formation and liquidity in the US Treasury market: The response to public information. Journal of Finance 54 (5), 1901–1915. Fridson, M.S., Jónsson, J.G., 1995. Spread versus Treasuries and the riskiness of high-yield bonds. Journal of Fixed Income 5 (2), 79–88. Goldreich, D., Hanke, B., Nath, P., 2005. The price of future liquidity: Time-varying liquidity in the US Treasury market. Review of Finance 9 (1), 1–32. Gwilym, O., Trevino, L., Thomas, S., 2002. Bid–ask spreads and the liquidity of international bonds. Journal of Fixed Income, 82–91. Heligman, L., Pollard, J.H., 1980. The age pattern of mortality. Journal of the Institute of Actuaries (107), 49–80. Hong, G., Warga, A., 2000. An empirical study of bond market transactions. Financial Analyst Journal 56 (2), 32–46. Houweling, P., Mentink, A., Vorst, T., 2002. Is liquidity reflected in bond yields? Evidence from the Euro corporate bond market. Working Paper, Erasmus University, Rotterdam, April. 1332 A. Dı́az et al. / Journal of Banking & Finance 30 (2006) 1309–1332 International Monetary Fund and World Bank, 2001. Guidelines for Public Debt Management. Kamara, A., 1994. Liquidity, taxes, and short-term Treasury yields. Journal of Financial and Quantitative Analysis 29 (3), 403–417. Krishnamurthy, A., 2002. The bond/old-bond spread. Journal of Financial Economics 66, 463–506. Nelson, C., Siegel, A., 1987. Parsimonious modeling of yield curves. Journal of Business 60, 473–489. Sarig, O., Warga, A., 1989. Bond price data and bond market liquidity. Journal of Financial and Quantitative Analysis 24 (3), 367–378. Schmidt, P., Sickles, R., 1977. Some further evidence on the use of the Chow test under heteroskedasticity. Econometrica 45, 1293–1298. Shen, P., Starr, R.M., 1998. Liquidity of the Treasury bill market and the term structure of interest rates. Journal of Economics and Business 50 (5), 401–417. Shulman, J., Bayless, M., Price, K., 1993. Marketability and default influences on the yield premia of speculative-grade debt. Financial Management (Autumn), 132–141. Strebulaev, I.A., 2001. Many faces of liquidity and asset pricing: Evidence from the US Treasury securities market. Working Paper, London Business School, February. Warga, A., 1992. Bond returns, liquidity, and missing data. Journal of Financial and Quantitative Analysis 27 (4), 605–617.