Preliminary, please do not quote Yield Spreads on Government Bonds Before and After EMU Lorenzo Codogno (Bank of America) Carlo Favero (IGIER, Bocconi University and CEPR) Alessandro Missale (Florence University and IGIER) January 21, 2002 Abstract Contrary to market perceptions, this paper provides evidence that yield differentials on euro-zone bonds are mostly due to risk-related domestic and international factors, as opposed to liquidity factors. Monthly series point to the existence of different, slowly evolving means, as well as to the importance of common international factors in the determination of fluctuations around the means. Slowly evolving means are related to the ratio of government securities to the total outstanding securities. At higher frequencies, yield differentials appear to be explained by common (credit-) risk related international factors, such as the US asset swap spreads and US and Euro-zone policy rates. Only the yield spread between on-the-run and off-the-run benchmark bonds (of the same issuer) appears to significantly affect yield spreads across member States; traditional liquidity indicators, such as bid-ask spreads, trading volumes and outstanding amounts have no effect. 1 1. INTRODUCTION Euro-zone government bond markets have rapidly changed as a result of global trends in the financial industry and the process of European Monetary integration. The introduction of the euro in January 1999 eliminated exchange risk between the currencies of participating member States, thereby creating the conditions for a substantially more integrated public debt market in the euro area. The government bond market may be considered as the "building block" that allows the nongovernment sector to grow. Therefore, even in a situation of shrinking government borrowing requirements that reduce supply of paper, the government bond market plays a key reference role. In terms of size and issuance volume, the euro-area government bond market is comparable to the US Treasuries market. However, it appears still segmented, with public debt management decentralized under the responsibilities of 12 separate agencies, and characterized by different issuing techniques (see Favero, Missale and Piga 1999). The persistence of yield differentials and their behavior over time may be perceived as the evidence of this segmentation. Understanding the factors that explain current yield differentials and their behavior over time has important implications for financial markets in general and may trigger policy responses with the aim to achieve higher financial markets integration. The aim of this paper is to study the determinants of observed yield spreads in the euro-area government bond markets. To this end we attempt to decompose yield spreads into the components that are expected to explain spread levels and movements, i.e. credit risk and liquidity (while in a future version of the paper we plan to also look at deliverability into futures contracts). In particular, we examine the relative role of international and liquidity factors that may be important in determining the yields on government bonds, by looking at the impact of US asset swap spreads and policy rate, European policy rate and liquidity factors such as trading volumes, outstanding amounts, spreads between on-the-run and off-the-run benchmarks, etc. Finally, we examine the effect of institutional changes regarding the functioning of primary and secondary markets. Yield spreads provides a simple measure of the market's assessment of the risk of default and the extent of financial integration between EMU member States. Therefore, a study of the determinants of yield spreads can shed light on a number of issues. A first issue is whether the risk of default has actually increased with the Monetary Union as countries have lost, with monetary independence, the option of printing money to pay for their debts. This can be evaluated by looking at the component of yield spreads, which does not depend on expected devaluation, before and after the EMU. Secondly, yield spreads capture the degree of financial integration. A main issue is whether a market for European public and private bonds as large and liquid as the US bond market will ever develop. In fact, to the extent that bonds issued by different member States are perceived as imperfect substitutes, the goal of creating such a market is frustrated. While persisting differences between the expected return on government bonds issued by different member States is a cause of concern independently of the source of segmentation, distinguishing between the credit-risk component and the liquidity premium is crucial for policymaking. Yield spreads reflecting persisting differences in credit ratings suggest that fiscal rectitude and coordination imposed by the Stability and Growth Pact is either ineffective or lack credibility. On the other hand, yield spreads reflecting differences in the liquidity of government bonds indicate that fiscal efforts may not be enough to deliver a "same-bond" market. More importantly, differences due to liquidity premia are likely to stay and prevent the creation of a European bond market unless member States agree on a more coordinated approach to the choice of issuing procedures, market-making requirements, futures markets, repo facilities etc. 2 To address these issues we rely on asset swap spreads (the difference between the yield on government bonds and the interest rates from the asset-swap curve) that allow to separate the liquidity and credit-risk components from expected exchange-rate depreciation so as to examine such components over a long horizon. Then, a decomposition between the liquidity premium and the credit-risk component is attempted by modeling their behavior in relation to a number of factors potentially affecting the creditworthiness of the government but not the liquidity of its debt. By focusing on potential determinants of liquidity and the credit-risk separately, we try to identify and measure the most important factors that have affected the yield spreads over the last decade (with monthly data) and, in particular, in the EMU period (with daily data). We find, contrary to market perceptions, that yield differentials on euro-zone bonds are mostly due to risk-related domestic and international factors, as opposed to liquidity factors. Monthly series point to the existence of different, slowly evolving means, as well as to the importance of common international factors in the determination of fluctuations around the means. Slowly evolving means are related to the ratio of government securities to the total outstanding securities. At higher frequencies, yield differentials appear to be explained by common credit-risk related international factors, such as the US asset swap spreads and US and Euro-zone policy rates. Only the yield spread between on-the-run and off-the-run benchmark bonds (of the same issuer) appears to significantly affect yield spreads across member States; traditional liquidity indicators, such as bidask spreads, trading volumes and outstanding amounts have no effect. 2. MEASUREMENT Before the introduction of the Euro, yield differentials within Europe had been determined by four main factors: expectations of exchange rate fluctuations, different tax treatment of bonds issued by different countries, credit risk and liquidity. Different tax treatments were eliminated or reduced to a negligible level during the course of the 90s. The introduction of the Euro in January 1999 eliminated the first factor (i.e. expectations on exchange rate fluctuations) creating the conditions for a substantially more integrated public debt market in the Euro area. Hence, yield differentials are now mainly determined by two remaining factors: credit risk and liquidity. Measuring the impact of credit risk and liquidity on Euro-area government bond yield differentials requires the comparative analysis of two types of interest rates. We consider the yield on government bonds and interest rate swaps of the same term to maturity. Using monthly data over the sample period January 1992 to August 2001, we report these variables for Austria, Belgium, Finland, France, Germany, the Netherlands, Ireland, Italy, Portugal and Spain respectively in Figures 1 and 2. The figures show that fixed interest rates on asset swaps have converged fully given that there is only one reference swap market since the launch of the Euro. Although smaller than in the past, spreads between yields on government bonds still persist. This supports the interpretation that differentials between fixed interest rates on swaps before Monetary Union are a direct measure of expectations on exchange rate devaluation. In principle, a “pure” measure of expected exchange rate changes could also be computed by comparing the yields on long-term bonds issued by the same supranational institution (such as the World Bank or the European Investment Bank) in different currencies. In fact, supranational issues are by definition free of any country-specific default risk. However, we prefer to concentrate on asset swaps, as they are homogenous and readily available for almost all EMU countries. Supranational issues in individual currencies are not comparable for all the different countries and are not issued on a continued basis, while yield differentials on government bonds are computed on the latest issues of the benchmarks (the so-called "on the run” issues). 3 Two approaches can be taken to measure the importance of liquidity and credit-risk factors in determining yield differentials. Considering Germany as the reference country, if we want to remove the component due to expected exchange-rate depreciation we can either focus on: RAS ti,T Rti,T RtGER RSWt i,T RSWt GER ,T ,T or, alternatively ASti,T Rti,T RSWt i,T The first approach looks at the relative asset swap spread, RAS, which is obtained by subtracting, from the difference in yields, R, between bonds of country i and Germany, the difference between fixed rates on asset swaps, RSW. This measures the combined importance of liquidity and default risk factors. It is worth noting that the second part of the formula directly measures the expected depreciation of the exchange rate. From 1999 onwards, as swaps differentials converged to zero, RAS converged to the differentials between bond yields. The second approach, represented by the asset swap spread, AS, provides a more detailed disaggregation by taking the swap curve as the reference point. This does not necessarily imply that the swap curve is a totally "neutral" reference point nor that it is the benchmark for government bonds. In fact, the asset swap spread also depends on the credit standing of the banking system and benchmark issues of major countries are in general more liquid and thus allow more accurate pricing. Nevertheless, interest rate swaps provide a well-arbitraged homogeneous curve and asset swap spreads a measure that is shielded from the effect of expected fluctuations in exchange rates. This allows us to use long-time series, which includes the pre-EMU period. Moreover, this approach clearly shows the contribution to yield spread developments due to each country as opposed to a combination of the two. The choice between the asset swap spread, AS, and the relative asset swap spread, RAS, can be seen as a choice on the best level of aggregation at which our empirical investigation on the determinants of yield differentials is to be conducted. In fact, given the identity: RAS ti,T ASti,T AStGER ,T ASt,T reflects two factors, i.e. liquidity of government bonds and credit risk relative to the banking sector in country i. RASt,T reflects four factors, i.e. liquidity and credit risk of government issues in country i relative to the country i banking sector, along with liquidity and credit risk of government issues in Germany relative to the German banking sector. It is instructive to compare the behaviour of these two measures over time and across different regimes. Figure 3 reports the time-series behaviour of the two series for the non-German countries in our sample, while Figure 4 visualises the inter-country correlation before and after the EMU regime. Figures 3 and 4 clearly show that the high positive correlation between the two measures in the preEMU period disappears in the EMU. In principle, this evidence could be explained by a higher volatility of the credit risk (and the liquidity premium) in the pre-EMU period. In fact, as shown in Figure 3 and, more clearly in Figures 5a, 5b, 5c and 5d , a stronger correlation between the domestic asset swap spread and the German spread is clearly the driving force behind the absence of correlation in the EMU period. Indeed, it appears that government bonds have been viewed as 4 closer substitutes, a sign of increased integration. In particular, as shown in Favero, Giavazzi, Spaventa (1997), since fluctuations in the expectations of exchange rate devaluation were also usually resulting in fluctuations of the credit-risk premia, it is conceivable that the EMU has reduced this source of segmentation between the credit-risk premium of domestic and German bonds. Naturally, such evidence requires us to choose between the two different measures. We favour asset swap spreads, AS, because identifying the significance of movements in a country’s asset swap spread separately is easier than identifying the significance of the combined effect encapsulated in the differential.1 For example, consider the Russian Crisis of Autumn 1998. The asset swap spread in Germany jumped to reach a peak of 70 basis points below Libor from an historical average of 25, as a consequence of spreading fears on the stability of the German banking system. The asset swap spread in Italy, Ireland, Portugal and Spain did not show any such deviation from their historical means in that period. Greece was not perceived as close to reaching EMU entry, and thus suffered dramatically, with the Greek asset swap spread recording the impact. This example makes also clear that the choice of the asset swap as the informational variable is to be taken cautiously in presence of fears of instability of the domestic banking sectors. To conclude, we favour the more disaggregated asset swap spread measure. This means that we model total yield differentials in the Euro area by investigating separately the behaviour of the two asset swap spreads which determine them. 3. UNDERSTANDING MOVEMENTS IN YIELD SPREADS Asset swap spreads tend to fluctuate around a long-run slow-moving component. We report the behaviour over time of asset swap spreads in the Euro area measured at monthly frequencies in Figures 5a, 5b, 5c and 5d. Asset swap spreads of EMU countries had been rather volatile before 1998, as can be seen in Figure 5a. We believe the erratic behaviour is related to: i) market segmentation; ii) tax distortions; iii) inefficient pricing and, more interestingly; iv) interaction between credit risk and exchange-rate risk. Some sovereign borrowers used to have captive domestic markets, and domestic banks/companies were induced to buy domestic bonds either by regulatory constraints or by their assets/liabilities matching preferences. This probably explains the behaviour of spreads in several countries, and among others in France, during the pre-1998 period. The first and second factors are key in explaining asset swap behaviour in Italy. The latter factor was at play in Portugal and Ireland, at least before 1998. Since 1998, the behaviour of asset swap spreads has been less erratic and more homogeneous across countries. At the beginning of 1998, EMU was almost completely discounted by financial markets, which probably explains the stronger correlation. The correlation can be seen in Figure 5b, where only Greek asset swap spreads behave as outliers. Figure 5e shows daily observations. In Figures 5c and 5d, we have grouped countries according to the level of their asset swap spread and plotted these variables for the two groups along with the German asset swap spread. All these figures strongly suggest the existence of common fluctuations around different, slowly evolving, means. 1 Moreover, we have worked so far on the assumption that benchmark bonds have the same duration. Different maturity/duration of benchmark issues causes differences in yield spreads that are not related to either credit or liquidity considerations, but simply to the shape of the yield curve. To avoid this distortion and any bond-specific factor, we took yields on constant maturity bonds (see Appendix 1). 5 To provide further insights to this observation we have considered a more international perspective by considering the US asset swap spreads. Figure 6a reports the level of asset swap spreads for the US and Germany, while Figure 6b also includes Italy. Asset swap spreads of German 10-year benchmarks have shown a high correlation with US spreads since the beginning of the 1990s, i.e. 0.71 between January 1990 and August 2001 on monthly data. This correlation has become stronger over time (0.84 between January 1998 and August 2001). The “sensitivity” of US asset swap spreads to this common international factor has been roughly twice as much as that of German asset swap spreads and the series appear to be increasingly correlated (Figure 6a). To conclude, the international evidence confirms the existence of different, slowly evolving means, as well as the importance of common international factors in the determination of fluctuations around the means. Slowly evolving means are likely to be related to economic fundamentals, such as debt and deficit to GDP ratios, and the ratio of the stock of outstanding government securities to the total outstanding securities. 4. IDENTIFYING CREDIT RISK Credit risk in the Euro area can be identified by looking at the role of slowly evolving variables and a common international factor; i.e. variables not related to liquidity. Our strategy for extracting information from asset swap spreads is based on separate modeling of evolving means and common model of fluctuations. We relate means to economic fundamentals and we claim that different means in the asset swap spread, AS, reflect different credit risk across countries. We circumstantiate this claim by considering as a monthly available measure of the economic fundamentals the ratio of the stock of outstanding government securities to the total outstanding securities. Figure 7 shows the high correlation of the 10-year US asset spread with the ratio of US outstanding government securities over total outstanding securities. Figure 8 reports the ratio of government securities to total securities for Germany, Italy, Europe and the US. The graphical evidence suggests not only a correspondence between the levels of such ratio and the levels of asset swaps but also the possibility of a common long-run trend. Interestingly, also deviations from the long-run trends seem to be correlated across countries. Such evidence suggests the possibility that, over the long-run, the equilibrium value of the asset-swap spread is determined by the credit risk and that the relative supply of government to total securities captures long-term fluctuations in this variable. Furthermore, the correlation of deviations from long-run equilibria suggests the possibility of an international determinant of short-run movements in asset swap spreads that are not related to structural factors. Such common fluctuations might instead be referred to phenomena often quoted by market participants such as “appetite for risk”, which have not so far received attention in the academic literature. Our interpretation of the graphical evidence suggests naturally a dynamic specification of the following type: AStUS a0 1 a1 a1 AS US a2 1 a1 RATIOtUS u1t t 1 AStGER b0 1 b1 b1 AS tGER b2 1 b1 RATIOtGER b3 1 b1 AStUS a2 RATIOtUS u 2t 1 ASti c0 1 c1 c1 AS ti1 c2 1 c1 RATIOti c3 1 c1 AStUS a2 1 a1 RATIOtUS c4 1 c1 AStGER b2 RATIOtGER AStUS a2 RATIOtUS u3t This Vector Error Correction model has the following features: 6 1. In the steady state the asset swap spreads are determined only by the domestic economic fundamentals as captured by the ratios variables. The steady state solution of the model is: AStUS a0 a2 RATIOtUS AStGER b0 b2 RATIOtGER ASti c0 c2 RATIOti 2. In the short-run international, German and local factors might have an impact on the determination of the fluctuations of the asset swap spread. We propose a recursive structure in which deviations from equilibrium in the US have a spill-over effect to Europe. We identify such an international factor by assuming that the US asset swap spread does not respond to fluctuations in European asset swap spreads. We then consider German data to extract a potentially common European factor, by assuming that Germany plays within Europe the role played by the US in the world and that the European factor is orthogonal in the long-run to the US factor. Lastly, we consider the other member States of the Union to evaluate the residual role of country specific factors in the determination of asset swaps. 3. This general specification will allow, in Section 5.3, an evaluation of the importance of liquidity factors by augmenting the system with liquidity-related variables such as bid-ask spreads, trading volumes, outstanding amounts, and yield spreads between on-the-run and off-the-run benchmark bonds. To illustrate the potential of the dynamic model let us start by concentrating on the US-German block only. Given the recursive structure of the system this is done at no cost in terms of consistency of estimators. The estimation of the model for US and Germany over the sample 1995:1-2001:8 delivers the following results: a0 = -2.35(0.78), a1 = 0.91(0.06), a2 = 7.00(3.38) b0 = 0.51(0.45), b1 = 0.76(0.06), b2 = 1.86(0.61), b3 = 0.70(0.17) We report in Figures 9-11 the actual asset swap spreads with their fitted long-run component along with the deviations from the long-run component. Estimation of the US-German block clearly shows that long-run trends in asset swap spreads are determined by the economic fundamentals as captured by our ratio variable. The short-run dynamics of the US spread is significantly determined by the disequilibrium with respect to the long-run trend. Interestingly, deviations from long-run trend in the US are an important determinant of the short-run dynamics in the German AS: such variable is significant in our specification along with disequilibium with respect to local fundamentals. As shown in Table 1, the extension of the model to other countries confirms the importance of local fundamentals in the determination of the long-run trends. However, it is also clear that international factors (such as deviation from long-run trends of US and German asset swaps) are much more important than local factors in the determination of short-run fluctuations of asset swap spreads in the euro-area. Interestingly, results are rather robust to the inclusion of the variable EXCH (the difference of fixed interest rates on swaps between each country and Germany) which capture expected exchange rate depreciation and allows to separate the regime of convergence to EMU from the properly defined EMU regime. 7 4.2 Long-run trends The interesting question on the long-run determinants is what is measured by the ratio of public bond outstanding to total bonds. Are we proxying the default risk related to the size of government debt or are we capturing relative supply effects between private and public bonds? We shall answer these questions by assessing how our results in the previous section are affected by considering different variables to capture the long-run trend. We shall consider variables more directly related to the relative supply effect, such as the ratio of government bonds outstanding to private bonds outstanding. We shall then assess the importance of variables more directly related to the risk premium on government bonds such as the government debt to GDP ratio and the deficit to GDP ratio. Credit ratings by major rating agencies may also provide insight. 4.3 Short-run fluctuations The strong international correlation among short-run fluctuations in differentials seems to point toward the importance of fluctuations over-time of a variable labeled “appetite for risk” by market participants. We plan to investigate the role of this variable by assessing the co-movement of our measured deviations from long-run trends with variables commonly thought as proxies for the risk appetite of investors. In particular we shall consider corporate spreads, the slope of the yield curve, the level of long-term interest rates and some naïve direct measure of the appetite for risk such as the rank correlation between returns on assets and their variance. 5. IDENTIFYING THE LIQUIDITY PREMIUM The econometric evidence, based on monthly data, leaves very little role for liquidity: spreads seem to be determined by economic fundamentals in the long-run and by common international factors in the short-run. The role of liquidity-related factors has, however, to be tested explicitly. This is impossible at monthly frequencies since data on volumes and bid-ask spreads are available only from 1999 onwards. Therefore, in this and the following sections, we concentrate on daily data from the EMU regime in order to test the relative importance of international and liquidity factors in the determination of spreads. Of course the role of macroeconomic fundamentals cannot be assessed at daily frequencies, but this should not be a source of econometric problems as the long-run trend are slowly evolving and they are not too different from constants over a two-year period. To identify the relative importance of international and liquidity related factors in explaining asset swap spreads we first focus on international factors and, then, in the next section, we test for the introduction of a number of liquidity-related factors. 5.1 International factors In our baseline specification, we consider US asset swap spreads, and US short and long-term interest rates as the variables depicting the international common trend. Equally, local factors are captured by Euro-zone short and long-term interest rates. The structure of our baseline model is as follows: 8 US US AS tUS c11 c12 AS tUS 1 c13 Rt 1 c14 rt 1 u1t US GER EU AS tGER c 21 c 22 AS tGER c 24 AS tUS 1 c 23 AS t 1 c 25 Rt 1 c 26 rt 1 u 2 t GER GER EU AS ti,T ci1 ci 2 AS ti1 ci 3 AS tUS ci 4 AS tUS ci 6 AS tGER 1 c i 5 AS t 1 c i 7 Rt 1 c i 8 rt 1 u it R i t RtGER AS ti AS tGER i 3( Italy ), 4( Spain), 5(Be lg ium), 6(Austria), 7(Finland), 8(The Netherlands), 9(France), 10(Portugal), 11(Ireland) We apply our baseline model to 10-year benchmarks and report the results in Table 2. The equations are estimated simultaneously by Seemingly Unrelated Regression Equations over the sample period June 1999 to August 20012. The results clearly show the importance of international factors in determining Euro-zone asset swap spreads3. This is strengthened by the pattern of the correlation of residuals. The co-variation between residuals in the Euro area raises some doubts on the local nature of possibly omitted variables. To evaluate the capability of the model to explain yield differentials and assess the relevance of different factors we use dynamic simulations under different scenarios. Dynamic simulation of asset swaps allows results to be presented in terms of yield differentials as the two variables are linked by an identity. The results of dynamic simulations are reported in Table 3. In the first simulation, we report spreads obtained by a dynamic simulation of the model assuming no modification in the exogenous variables. We take the evidence of zero mean in the difference between the averages of the actual and simulated series as a sign of the correct specification of our model. Actual and dynamically simulated series are reported in Figure 12. The reported results illustrate the capability of the model to track the evolution of all spreads. The pattern of correlation of the difference between actual and simulated series reflects that of the residuals of the econometric model. Such evidence strengthens our point on the importance of common factors in the determination of the fluctuations in yield differentials not explained by our model. Having assessed the model by simulation, we proceed to experiment with it and evaluate the impact on yield differentials of an increase in US government bond swap spreads (Scenario 1), a generalised tightening of monetary policies (Scenario 2), and an increase of international long-term interest rates (Scenario 3). The results of these simulations reveal that monetary policy affects yield differentials substantially, while the impact of a (one standard deviation) change in US government bond spreads is more limited. They also show that the levels of the yields on long-term government bonds have no significant effect. 5.2 Liquidity Liquidity is claimed to have become the most priced element in spread valuation since the launch of the Monetary Union, but it is not clear what is meant by it. The econometric evidence of our baseline model points strongly towards the importance of international factors in determining 2 The initial point of the sample is determined by the availability of measures for liquidity, which we are going to discuss in the next section. 3 Following our simple and intuitive definition of “international factors”, we have also tried to include an international measure of appetite for risk in our model, and found it not significant. We have considered the measure described in Kumar M.S. and Persaud A.(2001), based on the correlation between the rank of assets performance and the rank of their historic volatility (where a value of -1 implies perfect risk aversion and a value of +1 implies risk seeking behaviour by investors). The use of such measures is consistent with the evidence that models capable of generating slow countercyclical variation in the risk premia have been successful in explaining a wide variety of dynamic asset pricing phenomena (Campbell J.Y. and Cochrane,1999). 9 fluctuations in asset swap spreads in the Euro area. Can liquidity-related factors increase the explanatory power of international factors for yield differentials in the Euro area? To answer this question we must first find a measure of liquidity. While there is no generally accepted definition of liquidity, according to academic literature, there are four dimensions of liquidity4: i. immediacy, speed with which a trade of a given size at a given width is completed; ii. depth, maximum size of a trade for any given bid-ask spreads; iii. width, bid-ask spread, the cost of providing liquidity; iv. resiliency, i.e. how quickly price movements revert to “normal” levels after a large transaction and how quickly the imbalances in transaction flows dissipate. Market breadth is also referred to as the market ability to absorb large buy/sell orders without largescale price movements. All dimensions tend to interact, and there is no single measure of liquidity. It has been argued that, in the context of government bond securities, liquidity may be best thought of in terms of the cost of supplying immediacy. Liquidity is closely linked to secondary as well as primary market efficiency, i.e. having transparent issuing procedures with a high degree of predictability and a small number of large issues. Liquidity is influenced by the market structure as well as security-specific factors, i.e. the amount outstanding etc. These four dimensions of liquidity are difficult to record directly, and therefore other measures are usually considered to gauge liquidity conditions. These are: a) bid/ask spread; b) trading volume; c) turnover ratios (total trading volume divided by the stock of securities outstanding, i.e. the number of times the market "turns over" in the period); and d) trading intensity (number of transactions that take place over a set period). To study liquidity effects we identify a number of quantitative and qualitative proxies of liquidity or variables that may have affected liquidity premia according to the above classification and depending on the availability of data. The variables we consider comply with the requirements of our study, and are available across the Euro area (see Appendix 1 on the data set for more information). Such variables are: i) trading volume; ii) outstanding amount of the specific benchmark; iii) bid/ask spread, and; iv) yield spreads between off-the-run and on-the-run securities. We measure yield differentials between on-the-run and off-the-run securities by taking the difference between the benchmark on-the-run and the previous benchmark5. This measure has a potential problem in that the off-the-run security has a shorter maturity implying that term-structure effects might obscure the impact of liquidity on the differential. However, as we model spreads on German long-term rates, we consider all liquidity measures relative to the German ones. We consider yield differentials between on-the-run and off-the-run securities and differentials between bid-ask spreads for each country and those for Germany. Consistently, we take ratios of daily volumes and outstanding amounts of the benchmark in each country to those of the German benchmark. This transformation reduces the term-structure effect as the non-German tends co-move with that of the German term structure. 4 These definitions are taken from Gravelle (1999c). 5 An analogue measure for the US market has been proposed by Fleming (2001). 10 5.3 Testing for liquidity effects We combine the international (credit-risk related) factors with the liquidity factors in an econometric model of yield differentials in the Euro area (instead of asset swap spreads as in Section 4). We estimate by Seemingly Unrelated Regressions a nine-equation model on a sample of daily data covering the period between June 1999 and August 2001. We start from a baseline specification in which only international factors are considered, obtained by regressing yield differentials on a constant, on the lagged dependent variable, on contemporaneous and lagged US asset swap spreads and on the lagged level of Euro-zone policy rates. We then extend the model by adding our four measures of liquidity. The results from estimation are reported in Table 4. Table 4: Modelling 10-year yield spreads in the Euro-area NO liquidity OUTSTANDING VOLUME ONOFF BIDASK SPREAD Coef. S.E. Coef. S.E. Coef. S.E. Coef. S.E. Coef. S.E. ITA constant lagged d.v. ASSUSA ASSUSA-1 SHORT LIQUIDITY 0.0011 0.9324 -0.0257 0.0230 0.0048 0.0028 0.0087 0.0091 0.0091 0.0008 -0.0003 0.9308 -0.0255 0.0229 0.0050 0.0009 0.0040 0.0089 0.0091 0.0091 0.0009 0.0002 0.0004 0.9350 -0.0250 0.0210 0.0040 0.00003 0.0030 0.0114 0.0094 0.0095 0.0009 0.00003 -0.0048 0.9360 -0.0230 0.0165 0.0047 0.0400 0.0037 0.0086 0.0090 0.0091 0.0007 0.0145 0.0050 0.9317 -0.0260 0.0240 0.0042 0.0005 0.0107 0.0168 0.0116 0.0114 0.0022 0.0015 SPA constant lagged d.v. ASSUSA ASSUSA-1 SHORT LIQUIDITY 0.0152 0.8931 0.0013 0.0017 0.0043 0.0045 0.0151 0.0126 0.0127 0.0009 0.0167 0.8930 0.0011 0.0016 0.0042 -0.0020 0.0061 0.0152 0.0126 0.0126 0.0010 0.0052 0.0091 0.8870 -0.0100 0.0140 0.0061 0.0000 0.0046 0.0210 0.0130 0.0131 0.0015 0.0003 0.0060 0.8770 0.0010 -0.0040 0.0050 0.0740 0.0049 0.0162 0.0123 0.0123 0.0009 0.0160 0.0137 0.9320 -0.0110 0.0140 0.0022 -0.0030 0.0110 0.0182 0.0113 0.0112 0.0019 0.0050 BEL constant lagged d.v. ASSUSA ASSUSA-1 SHORT LIQUIDITY 0.0120 0.8687 -0.0448 0.0337 0.0047 0.0047 0.0160 0.0144 0.0145 0.0009 0.0076 0.8632 -0.0435 0.0343 0.0056 0.0047 0.0057 0.0164 0.0145 0.0146 0.0012 0.0033 0.0090 0.8630 -0.0530 0.0406 0.0052 -0.0007 0.0050 0.0219 0.0151 0.0152 0.0013 0.0003 0.0080 0.8650 -0.0370 0.0250 0.0055 0.0300 0.0057 0.0160 0.0143 0.0144 0.0010 0.0180 -0.0150 0.8410 -0.0470 0.0280 0.0110 -0.0015 0.0152 0.0329 0.0162 0.0163 0.0037 0.0085 AUS constant lagged d.v. ASSUSA ASSUSA-1 SHORT LIQUIDITY 0.0013 0.8561 -0.0099 -0.0066 0.0052 0.0038 0.0208 0.0128 0.0129 0.0010 0.0005 0.8560 -0.0096 -0.0065 0.0052 0.0026 0.0047 0.0208 0.0129 0.0129 0.0010 0.0090 0.0070 0.8430 -0.0136 -0.0020 0.0050 -0.0009 0.0049 0.0252 0.0159 0.0160 0.0013 0.0007 -0.0070 0.8560 -0.0100 -0.0080 0.0060 0.0730 0.0050 0.0205 0.0128 0.0128 0.0011 0.0210 0.0110 0.8390 -0.0057 -0.0095 0.0044 -0.0060 0.0193 0.0314 0.0206 0.0206 0.0036 0.0100 FIN constant lagged d.v. ASSUSA ASSUSA-1 SHORT LIQUIDITY 0.0350 0.8738 -0.0260 0.0327 -0.0002 0.0064 0.0155 0.0147 0.0147 0.0008 0.0351 0.8733 -0.0257 0.0327 -0.0002 0.0015 0.0078 0.0156 0.0148 0.0147 0.0009 0.0110 0.0130 0.8230 -0.0188 0.0250 0.0070 -0.0001 0.0110 0.0310 0.0178 0.0178 0.0021 0.0008 0.0350 0.8660 -0.0180 0.0241 0.0000 0.0044 0.0069 0.0159 0.0145 0.0146 0.0008 0.0190 0.0391 0.7910 -0.0200 0.0320 0.0037 -0.0030 0.0159 0.0308 0.0170 0.0168 0.0029 0.0070 NET constant lagged d.v. ASSUSA ASSUSA-1 SHORT LIQUIDITY 0.0171 0.8575 -0.0910 0.0871 -0.0001 0.0060 0.0181 0.0183 0.0184 0.0010 0.0164 0.8555 -0.0910 0.0870 -0.0001 0.0010 0.0066 0.0185 0.0182 0.0183 0.0009 0.0030 0.0286 0.7990 -0.0910 0.0860 -0.0012 0.0000 0.0069 0.0273 0.0188 0.0189 0.0010 0.0040 0.0020 0.8661 -0.0810 0.0720 0.0018 0.0480 0.0073 0.0172 0.0186 0.0186 0.0011 0.0163 0.0830 0.7620 -0.0720 0.0810 -0.0080 0.0009 0.0203 0.0355 0.0189 0.0188 0.0033 0.0060 FRA constant lagged d.v. ASSUSA ASSUSA-1 SHORT LIQUIDITY 0.0056 0.7535 0.0104 -0.0126 0.0073 0.0086 0.0259 0.0290 0.0291 0.0017 -0.0023 0.7510 0.0100 -0.0120 0.0080 0.0030 0.0169 0.0264 0.0290 0.0291 0.0026 0.0060 0.0140 0.7230 0.0580 -0.0142 0.0200 -4.4E-04 0.0119 0.0373 0.0454 0.0448 0.0053 0.00420 0.0100 0.7540 0.0190 -0.0152 0.0070 0.1000 0.0087 0.0257 0.0290 0.0291 0.0017 0.0370 0.0890 0.6700 -0.0300 0.0500 0.0410 0.0470 0.0138 0.0413 0.0507 0.0510 0.0055 0.0290 POR constant lagged d.v. ASSUSA ASSUSA-1 SHORT LIQUIDITY 0.0361 0.8314 -0.0028 -0.0030 0.0040 0.0100 0.0196 0.0300 0.0301 0.0016 0.0379 0.8303 -0.0027 -0.0030 0.0039 -0.0020 0.0144 0.0197 0.0300 0.0301 0.0018 0.0300 0.0610 0.7880 0.0086 -0.0126 0.0018 -0.00450 0.0135 0.0317 0.0407 0.0410 0.0035 0.0001 0.0165 0.8100 -0.0170 -0.0060 0.0054 0.1573 0.0094 0.0192 0.0278 0.0278 0.0015 0.0156 -0.0700 0.8600 -0.0700 0.0500 0.0240 -0.0020 0.0600 0.0560 0.0470 0.0477 0.0044 0.0100 IRL constant lagged d.v. ASSUSA ASSUSA-1 SHORT LIQUIDITY 0.0243 0.8314 -0.1005 0.0867 -0.0025 0.0066 0.0182 0.0205 0.0206 0.0011 0.0237 0.8312 -0.1014 0.0882 -0.0025 0.0020 0.0090 0.0186 0.0205 0.0206 0.0012 0.0120 0.0169 0.8299 -0.0860 0.0710 -0.0010 0.0200 0.0072 0.0173 0.0203 0.0205 0.0014 0.0100 VOLUMES are the ratios of trading volume of bond issued by country i to those issued by Germany recorded by MTS. OUTSTANDING are the ratios of outstanding amounts on specific benchmark bonds in country i to the outstanding amount of the German benchmark. ONOFF is the difference in yield spread between on-the-run and off-the-run bonds issued by country i and yield spreads between on-the-run and off-the-run bonds issued by Germany. BIDASK is the difference between the bid/ask spread in country i and that in Germany. The extension of the model to the inclusion of liquidity indicators shows that the significance of international factors is always robust of the inclusion of liquidity indicators and only the relative onthe-run/off-the-run yield differentials have a generalised significant effect in addition of the international factors. The impact is positive everywhere, showing that an increase in the differential between on-the-run and off-the-run securities in country i relative to that of Germany widens the yield differentials between country i 10-year bonds and German 10-year bonds. 11 The US asset swap spread is significant for Italy, Belgium, Austria, Finland, the Netherlands, Portugal and Ireland. This variable has a negative impact: an increase in 10-year US asset swap spreads widens yield differentials between non-German and German Euro countries. France and Spain are the only countries whose differentials on German bonds are insulated from this effect. Higher short-term interest rates in the Euro area have also a significant impact widening yield differentials, the only exception being Finland, the Netherlands and Ireland where, however, the short rate is not significant. The negative results on three out of four liquidity indicators may be due to the lack of comprehensive data for the whole market (including over-the-counter activity). The only data available on traded volumes and other liquidity indicators for a sufficiently long time period and on a daily basis are from EuroMTS.6 This subset of information is not an unbiased sample of underlying market activity. Moreover, measuring the liquidity premium is also difficult because liquidity and credit risk interact with each other. The lack of liquidity amplifies the effect of risk. This is because liquidity variables, such as the bid-ask spread, reflect the risk borne by market makers in managing unbalanced positions. As credit risk increases so does the risk they face. Indeed, there is substantial anecdotal evidence of bid-ask spreads opening during crisis periods with larger increases for assets and bonds of lower quality. Obviously this argument can also be applied to the only significant measure for liquidity. A (credit-related) flight-to-quality argument might also be used to interpret the significance of our off-the-run/on-the-run yield differentials. To shed further light on this issue we report in Table 5 the correlation matrix of the on-the-run/off-the-run yield differentials for each country, and the international factors which we relate to credit risk. Table 5: Correlation of the ONOFF liquidity measures with the international factors ITA ITA IRE FRA FIN BEL AUS NET POR SPA ASS10USA SHORT IRE 1.00 0.24 0.14 0.68 0.48 0.53 0.60 0.35 0.48 0.69 -0.33 FRA 0.24 1.00 0.21 0.01 0.48 0.30 0.59 0.14 0.43 0.29 -0.78 FIN 0.14 0.21 1.00 0.04 0.28 0.12 0.09 0.08 0.15 -0.01 -0.03 BEL 0.68 0.01 0.04 1.00 0.43 0.46 0.52 0.29 0.54 0.61 -0.16 AUS 0.48 0.48 0.28 0.43 1.00 0.40 0.55 0.20 0.37 0.18 -0.53 NET 0.53 0.30 0.12 0.46 0.40 1.00 0.50 0.23 0.46 0.36 -0.44 POR 0.60 0.59 0.09 0.52 0.55 0.50 1.00 0.27 0.54 0.51 -0.75 SPA 0.35 0.14 0.08 0.29 0.20 0.23 0.27 1.00 0.24 0.30 -0.14 0.48 0.43 0.15 0.54 0.37 0.46 0.54 0.24 1.00 0.55 -0.44 ASS10USA SHORT 0.69 -0.33 0.29 -0.78 -0.01 -0.03 0.61 -0.16 0.18 -0.53 0.36 -0.44 0.51 -0.75 0.30 -0.14 0.55 -0.44 1.00 -0.33 -0.33 1.00 The liquidity measure for each country is the difference in yield spread between on-the-run and off-the-run bonds issued by country i and yield spreads between on-the-run and off-the-run bonds issued by Germany, the international factors are the US asset swap spread and the 1-month policy rate in Europe (short). The reported correlation does not solve the issue. In fact, the correlation of liquidity indicators with international factors is only moderate, and features a considerable amount of cross-country variability. However, there is also a rather strong pattern of cross-country correlation for these 6 EuroMTS is the electronic trading platform that started operating in April 1999 on Euro-zone government bond markets and that was the only available source of information for homogeneous indicators of liquidity. The rest of the market has been, and still is, mainly over-the-counter and no reliable, consistent and sufficiently long series can be obtained on liquidity indicators. EuroMTS provided time series long enough to perform statistical analysis, i.e. from spring 1998 (some countries only) to present. Other electronic systems have not been taken into consideration as they started operating only very recently, and therefore time series long enough to perform statistical analysis are not available. EuroMTS data are available for the 10-year maturity, although some data are also available for 2/3-year, the 5-year, and the 30-year benchmarks in the main government bond markets. 12 measures, suggesting global factors, such as flight-to-quality phenomena, might be important in explaining its variability. 6. IDENTIFYING LIQUIDITY FROM CASE STUDIES While there is no generally accepted definition of liquidity, we can broadly define liquidity risk premia as the additional yield required by market makers for the risks and costs in managing their bond positions.7 The less liquid a bond, the higher the yield it must offer. Relevant factors that affect the return and risk of such activity include: i. the features of the market where bonds are traded, from their infrastructures, the admission and trading rules, the clearing and settlement of transactions, to the regulatory frameworks, etc.; ii. the trading volume, which is related to the outstanding amount of bonds with the same characteristics, among other things; iii. the efficiency of the repurchase agreement and futures markets available to manage the risks of dealers' activity; iv. the primary market to the extent that the issuing techniques affect dealers' incentives to trade the particular government’s securities. The first group of factors is an important area for policy reform, since the creation of a fully integrated European bonds market probably hinges on the definition of common regulatory frameworks, trading rules, clearing and settlement arrangements. We do not know how to test the impact of these factors, but the market perception is that such factors are not too important. With regard to the second source of liquidity differentials, in the previous section we have investigated whether differences both in trading volumes and in the outstanding amounts of benchmark bonds (relative to German benchmarks' volumes and amounts) affect yield spreads without finding any significant effect. In this section, we turn to the other sources of liquidity differentials. Besides trading volumes, the risks and costs faced by market makers in managing their bond positions depend on: i) the efficiency of the futures market and of the repurchase agreement market; ii) institutional aspects of primary and secondary markets, and repo facilities of the last-resort type that may affect dealers’ incentives to trade the securities of particular governments. 6.1 Futures markets A proper functioning of associated derivatives markets facilitates the active trading and management of interest rate risk. Where a well-developed futures market exists, market makers can manage their positions using futures, thereby enhancing their ability to carry out inventory-risk management in the cash market, which, in turn, promotes better liquidity. Trading activity in the futures markets may also increase activity in the cash markets due to arbitrage activity on the basis.8 Equally, a well-developed futures market depends on a deep underlying cash market. This mutually reinforcing process results in large liquid issues, which are deliverable into an actively traded futures contract, commanding a price premium. In the Euro area the Bund futures contract has become predominant. German government bonds, which have become the de-facto benchmark in 7 Market makers, i.e. those players making secondary markets on bonds, may not be the same institutions being part of the placement consortium (also called group of primary dealers or specialists in some countries), and may also be different from the more closer group that gets mandates for international issues or for Treasury’s related activities (M&A etc.). 8 There appears to be different positions on this issue. Some believe a liquid future market can withdraw liquidity from the underlying cash market, as speculative trades would find trading in the future market cheaper. 13 the 10-year sector, appear to command a sizeable premium versus other sovereign issues due to this "derivative factor". Although we planned to analyse the impact on the derivative market, so far we have not been able to collect sufficient and reliable data on overall trading volumes in the cash markets on a daily basis (including over-the-counter transactions) to compare with trading volumes in derivatives markets.9 6.2 Institutional innovations in primary and secondary markets Institutional aspects concerning primary markets may affect dealers' incentives to trade the securities of particular governments and thus impact on their liquidity. For example, the regularity and predictability of the issuance process is valued by market makers since it allows for better planning of their activities and for a more effective risk management. Sovereign borrowers who announce in advance the types of bonds (with indicative amounts) that they are going to issue may gain a liquidity premium for their bonds, as market makers prefer to trade such securities. It is indicative that with the advent of EMU and the implied increase of competition, issuance calendars have been introduced in all EMU member States (see Favero, Missale and Piga 1999). The liquidity that market makers are willing to provide by actively trading the bonds of specific issuers may also depend on institutional aspects not regarding the primary market. For instance, market makers may positively value the existence of a repurchase agreement facility in the form of a window at the Treasury, say, a repo facility of the last-resort type. It must be realised, however, that measuring the extent of reduction in cost and risk that is provided by specific features of calendars, issuing procedures and placement techniques is practically impossible; at best we can deal with qualitative differences among sovereign borrowers. Since institutional aspects of debt management rarely change over time and, when they do, such changes are part of broader reforms involving several features of debt management, as it was the case with the advent of EMU, it becomes difficult to identify their effects on liquidity. For instance, the introduction of dummies to account for differences in the placement techniques across member States would provide no useful information, since it would be equivalent to attributing any unexplained differences in yield spread to country-specific factors. Furthermore, even if we took for granted that differentials in liquidity premia depend on debt management, the exact source of such differentials would be difficult to identify since countries differ along several dimensions of debt management. However, such difficulties can however be overcome if any specific aspect of debt management changes during the sample period (and if such changes do not occur contemporaneously and in all countries). Innovations regarding the placement technique, the issuing method, the introduction of risk-management facilities, etc. all provide important events that possibly affect the liquidity of the securities issued by the reforming government. The impact of such events on liquidity can be studied with the introduction of dummy variables in the equation system for yield-spreads. 9 The only available source of information for homogeneous indicators of liquidity was EuroMTS, the electronic trading platform that started operating in April 1999 on Euro-zone government bond markets. The rest of the market has been, and still is, mainly over-the-counter and no reliable, consistent and sufficiently long series can be obtained on liquidity indicators. EuroMTS provided time series long enough to perform statistical analysis, i.e. from spring 1998 (some countries only) to present. Other electronic systems have not been taken into consideration as they started operating only very recently, and therefore time series long enough to perform statistical analysis are not available. 14 Events characterising debt management during the sample period from June 1999 to August 2001 have been identified through the answers provided by the debt managers of EMU member States to the questionnaire reported in Appendix 2. The questionnaires (when incomplete or lacking the indication of the precise day of the innovation) have been integrated with the information in the annual reports and other documents at the web sites of the debt offices. Unfortunately, a complete account of innovations is not possible since answers have not yet been received from Ireland and Spain. The questionnaire allows us identification of innovations in the following areas: 1) innovations regarding the primary market, in particular, changes in the issue calendar, in the issuing method or in its characteristics; 2) innovations regarding the secondary market, such as changes in admission or trading rules of wholesale markets and changes in the requirements (market making obligations) of dealers; 3) institutional reforms of the repurchase agreement market or the introduction of a repurchase agreement facility in the form of a window at the Treasury; 4) the delegation of debt management to a different institution; i.e. an independent agency; The questionnaire asks for the date when the change in debt management over the above areas became effective. This allows us to create step dummies; i.e. 0-1 "event variables", that take the value of 1 after an institutional change. Including these variables into the model of yield-spreads allows us to assess the importance of institutional aspects as determinants of yield-spreads across member States. 6.3 Innovations in primary markets Primary market policy varies widely across EMU member States. For instance, while most countries issue bonds by auctions, the tap issuing technique is still in use in the Netherlands and small issuers, like Austria, Belgium, Finland, Greece and Portugal use a bank syndicate for the launch of new bond lines. Most countries adopt multiple price auctions and hold a second round auction with non-competitive bidding which is restricted to primary dealers with higher status, but in Italy and Finland the uniform-price technique is used and no second round auction is held in Germany. Despite these differences very little has changed since the advent of EMU. We have investigated whether: 1) the issuance calendar has been modified; 2) the issuing technique has changed, for example, from auction to a bank syndication; 3) the auction method has changed from multiple to uniform pricing or in the method to determine the total amount allotted; 4) the auction framework has changed; i.e. with the introduction of an electronic auction system or with a shortening of the time period before the notification of the results; 5) second-round non-competitive auctions have been modified with an extension of the time available for placing bids and with an increase in the amount of bonds available in the secondround; 6) the requirements that primary dealers must meet to participate in the auctions; in particular, whether they must be based locally. 6.3.1 Event study: calendars and auction cancelled Issuance calendars, whose importance has been discussed above, have not been an area of reform. Few countries introduced or extended the calendar in preparation for EMU but since 1999 no 15 changes have been made. In particular, we asked: i) whether there has been a move from a quarterly (or semi-annual calendar) to a yearly calendar; ii) whether bond types and indicative issue amounts have been more closely specified and; iii) whether the time period (in cases where an exact date is not announced) has been more narrowly defined. By contrast, some countries have reported instances where an auction (previously scheduled in the calendar) has not taken place. These events are likely to take place because of lower deficits and decreasing debts with very long average maturities. The problem clearly shows up at low levels of debt as pointed out by the answer of Finnish debt managers that auctions were cancelled because of no need for financing, unfortunately without providing the dates of such episodes. But, surprisingly, in December 2001 a bond auction was not held in Italy for the same reason. The cancellation of a previously scheduled auction directly affects liquidity, if it regards a reopening of an existing issue, and, more generally, it may lead market makers to change their plans with possible effects on the incentive to trade the securities of the issuing country. Although the answer from Finland was imprecise and the Italian episode falls outside the sample period, the Austrian and Belgian authorities provided the following useful information for an eventstudy: Episodes of cancelled bond auctions: Country Date AUSTRIA 11-11-2000 BELGIUM 25-01-1999 (out of sample) 27-09-1999 20-12-1999 31-01-2000 29-01-2001 28-05-2001 26-11-2001 (out of sample) Reason Fulfillment of funding needs Syndication Syndication Fulfillment of funding needs Syndication Syndication Syndication Fulfillment of funding needs To account for such episodes we have included (in the equation system for the 10-maturity spreads) point dummies for the date when the cancelled auction was supposed to have taken place. We choose a point dummy taking the value of 1 on the days of the auctions and 0 otherwise (instead of a step dummy as for other event studies) because such episodes are expected to have transitory effects as opposed to institutional changes of a permanent nature. The specification used for the system of yield spreads is that based on the international factors and the on-off-differentials so as to control for the other factors affecting credit risk and liquidity. The point dummy for the cancellation episodes of Belgium is significant at the 10% level and has a negative coefficient delivering a long-run impact of minus 2 basis points. This unexpected result could be explained by the fact that we do not use, as would be more appropriate, the dates when the cancellations were announced. It is however worth observing that 4 out 5 episodes in our sample result in a substitution of a syndicated placement for the cancelled auction. Since the syndication technique eliminates the auction risk to which primary dealers (and thus most market makers) are exposed, it may lead to an increase in the demand for the bonds of the same line (if a reopening) as those which are going to be placed through syndication. 6.3.2 Event study: change in the auction method and targeted issue size The issuing techniques adopted by EMU member States have remained the same since 1999 when smaller issuers, namely, Austria, Belgium, Finland and Portugal moved from auctions to syndicated placements because of common currency denomination and visibility problems. In fact, the main 16 features of the issuing method have also remained the same; no country reports a change in the pricing system and/or in the determination of the total amount allotted in first-round competitive auctions. The only relevant innovations which have taken place regard the targeted issue size and the auction framework, namely, the introduction of an electronic auction system or with a shortening of the time period before the announcement of the results. The advent of EMU and the resulting stronger competition for raising funds have led countries to increase the size of their benchmark bonds in order to achieve a greater liquidity. Large borrowers now have benchmarks around euro 15 billion, while the target for smaller issuers is set at euro 5 billion, that is the requirement for admission at trading on the Euro-MTS. Accordingly, the issue sizes are large, around euro 1.5 billion for small borrowers and over euros 2 billion for large borrowers. However, many of these changes have taken place before 1999. Austria and Belgium are the only countries (among respondents) which report an increase in the target for the size of new issues, both since January 2000. Quoting the Belgian answer: "The number of auctions has been reduced. For the auctions up to January 2000, we used to have monthly auctions, with 3 lines issued at every auction. Since January 2000, we have cut down the number of auctions to 6 and we issued only 2 lines at each auction. The average size has doubled from what it used to be." This is an important innovation that provides a clear event study Episodes of increased issue size: AUSTRIA 1-01-2000 BELGIUM 1-01-2000 To account for such episodes we have included (in the equation system for the 10-maturity spreads) step dummies taking the value of 1 since 1/1/2000. As usual we augment the specification that is based on the international factors and the on-off-differentials. The step dummy for the change in the issue size targeted by Belgium has a negative coefficient, as expected, but is not significant. By contrast, in the case of Austria the coefficient is significant with a t-stat of 1.9, but its sign is negative, indicating, if anything, a deterioration of the liquidity of Austrian bonds relative to German Bunds. Though disappointing, this result confirms, once more, the difficulties in finding any effect of liquidity on yield spreads. The introduction of an electronic auction system or the shortening of the time period between the end of the bidding period and the notification of the results can provide further evidence on liquidity effects. Indeed, the costs of hedging the risk of unbalanced bond positions taken at times of bond auctions may depend on the length of time between the end of bidding and the announcement of the result. An electronic auction system has been used in Belgium since May 2000. In the same year in France the time needed for the announcement of the auction results has been reduced from 10 to 5 minutes, the shortest in the euro area. More recently, in Italy the time for auction results has been reduced from 20 to 15 minutes. A faster release of auction results may reduce the risk faced by primary dealers (and thus by most market makers) and may increase the incentive to trade the bonds of the most efficient country. This suggests that episodes of improvements in the infrastructure of auction systems could be used as event studies. 17 Episodes of reduction of time period before notification of auction results: Country Episode Date BELGIUM Introduction of electronic auction system 29-05-2000 FRANCE Reduction from 10 to 5 minutes 6-01-2000 ITALY Reduction from 20 to 15 minutes 14-06-2001 As usual, a step dummy for each country has been included in the equation system for yield spreads to account for such episodes. The dummy for the introduction of an electronic auction system in Belgium has a positive coefficient, though the impact on the yield spread is not significant. The reduction in the time needed for the notification of the results in Italy also appears to be associated with a higher yield spread. The impact is significant, though its magnitude is small with a long-run impact of 4 basis points. Only in the case of France does a faster release of auction results reduce the yield spread. The effect is significant at the 5% level and the long-run impact on the yield spread is minus 4 basis points. 6.3.3 Event study: change in second-round non-competitive auctions The advent of EMU has led sovereign borrowers to compete for the services of market makers since they are crucial for the bonds to be placed, quoted and traded. More concessions have been given to primary dealers with higher status (who are also market makers) because of their crucial role in promoting national bonds abroad and attracting the demand of European investors. A few countries, such as Italy for example, no longer require that primary dealers be locally based. With the start of EMU, the time available for non-competitive bidding in second-round auctions (to which only dealers with highest status can participate) has been extended de facto in countries like France and Spain. In the Netherlands, where bonds are issued by tap, a Primary Dealer System was created in January 1999 and dealers receive a fee depending on the volume of bonds taken. The system is expected to change in 2002 with the use of non-competitive bids. Since 1999 policies regarding the localization of primary dealers have not been changed (with the exception of Greece where local residence is no longer required since January 2001). On the contrary, the time available for non-competitive bidding in second-round auctions has been extended in Austria, Greece and Italy. This provides an interesting case study of whether concessions given to primary dealers may increase the incentive to trade and place the bonds of generous governments with a positive impact on their liquidity. Episodes of extended time for placing bids in second-round auctions: AUSTRIA 13-2-2001 GREECE (not usable) 1-01-2001 ITALY 30-3-2000 To study the effect of such episodes on the yield spreads of Austria and Italy, we have augmented with two step dummies the 10-year system based on the international factors and the on-offdifferentials. The extension of the bidding period does not appear to have any significant effect on the yield spread, both in Austria and in Italy. A longer period to place non-competitive bids appears to reduce the yield spread in Austria but the coefficient for the step dummy is not significant. The same holds true for Italy, where the yield spread is even wider after the institutional innovation. 18 6.4 Innovations in secondary markets and repo markets Institutional changes regarding the functioning of secondary markets are expected to have the strongest impact on liquidity. However, as we focus on bond yields recorded on the electronic MTS markets (Euro-MTS and local MTS), it is unlikely that institutional changes which do not affect the framework, the admission and trading rules of such markets, may have an impact on yield spreads. We introduced in the questionnaire, sent to the debt managers, a number of questions regarding the functioning of local secondary markets in order to identify events that could possibly affect liquidity. We asked debt managers whether: 1) a domestic electronic securities market exists and the date when it became operative; 2) admission or trading rules in domestic wholesale markets have changed; 3) requirements for becoming market-makers have been modified since 1999; 4) the number of market makers has changed; 5) a well established repo market for bonds exist and the date when it became operative; 6) a repo facility of the last resort type exist at the Treasury and the date of introduction. The study did not allow us to identify significant innovations in the structure of secondary markets except for the start of operations of domestic MTS markets. Indeed, episodes of changes in admission and trading rules in wholesale markets coincide with the beginning of the operations of domestic MTS markets. 6.4.1 Domestic MTS markets The increasing automatisation of trading is also an important factor of liquidity. Since 1999, following the example of Italy and the success of Euro-MTS, electronic trading has developed in Belgium, France, the Netherlands and Portugal. Since national MTS markets are electronically linked with the Euro-MTS, their start has increased overall MTS trading volumes on the country’s benchmarks. The MTS system offers a simple interface with other MTS markets (including EuroMTS), allowing for the use of a common platform for the trading of debt instruments of EMU member States, while simultaneously keeping local specificities such as local management and supervision, market makers, etc. As a consequence of the development of local MTS, the number of market-makers, trading the same bonds, has increased. This has probably led to higher trading volumes and better pricing and has possibly reduced the yield on bonds issued in the country of the national MTS. Beginning of the operations of domestic MTS markets MTS Amsterdam 9-1999 MTS Belgium 5-05-2000 MTS France 25-04-2000 MEDIP/MTS Portugal 24-07-2000 To evaluate the impact on yield spread of the introduction of national MTS electronic markets, we have introduced three dummies in the 10-year maturity system based on the international factors and the on-the-run/off-the-run differentials (in short ‘on-off-differentials’). Specifically, we have augmented the equations for the Belgian, French and Portuguese yield spreads with three different dummies, one for each equation, taking the value of 1 since the date of the beginning of the MTS for the corresponding country. Instead, we could not include a dummy for the start of MTS Amsterdam since it begun operating in September 1999, which also marks the start of the series for the Dutch spread. 19 The regression results show that the step dummies for the domestic Belgian and Portuguese MTS markets significantly affect the yield spreads of the corresponding countries while the start of the MTS France has had no impact on the French yield spread. This finding is not surprising, since most trade on French bonds occur on the Euro-MTS while the opposite is true in the case of Belgium and Portugal where the trading volume (of the corresponding bonds) on the domestic MTS is substantially higher than that on the Euro-MTS. The start of MTS Belgium appears to have added liquidity to the market, leading to a reduction in the yield spread. The dummy is indeed significant at the 10% level and has a negative coefficient delivering a long-run impact of 2 basis points. Also the start of MEDIP/MTS Portugal significantly affects the Portuguese yield spread, but contrary to expectations, the effect is a widening of the spread up to 6 basis points in the long run. This result is surprising since the start of MEDIP has been associated with a considerable increase in trading volume and thus of a greater liquidity. However, Portuguese data on yields are very volatile until July 2000, the start of the domestic market, and this may be an indication of the poor quality of recording. Unfortunately, this problem is likely to affect the estimates concerning the other episodes of institutional changes in Portugal. Focusing on the more reliable event-studies of Belgium and France we observe that when the introduction of the domestic market significantly add liquidity as in the case of MTS Belgium the yield spread is reduced. However, the effect of increased trading volumes do not appear substantial, only 2 basis points, a result which implicitly confirms our previous estimates of the effect of trading volumes on yield spreads. Focusing on the more reliable event-studies of Belgium and France, we observe that the introduction of the domestic market significantly adds liquidity as is the case of MTS Belgium where the yield spread is reduced. However, the effect of increased trading volumes do not appear substantial, only 2 basis points, a result which implicitly confirms our previous estimates of the effect of trading volumes on yield spreads. 6.4.2 Market making requirements Interestingly, with the notable exception of Greece, the requirements for becoming market makers does not appear to have changed since the advent of EMU. The number of bond lines to be quoted, the amounts to be traded or the minimum amounts of bidding in auction (if applicable) or qualitative requirements have remained the same. Some countries have changed the requirements in January 1999 but have since left such requirements unaltered. In Greece a number of quantitative and qualitative requirements have been introduced but data on Greek yields are available only since February 2001. This leaves us with no main event to study. Only in Finland has a narrower bid-ask spread been introduced for market makers; since January 2001 the bid-ask spreads have been reduced to 4, 6 and 8 cents depending on the maturity of the bond being traded. Episode of narrower bid-ask spread requested to market makers: FINLAND 1-01-2001 A step dummy has been included in the equation for the Finnish yield spread, taking the value of 1 since January 2001. The very low t-stat for the coefficient of the step dummy clearly shows that the stricter requirement for market makers has had no effect on the liquidity and the yield spread of Finnish bonds. 20 6.4.2 Event study: introduction of a repo window at the Treasury In all EMU member States, with the exception of Greece, well functioning repurchase agreement markets for bonds existed before 1999 and further facilities have been provided through access contracts to Euro-MTS repo facilities. However, the latter cannot be considered as independent events to be studied separately from the development of domestic MTS. Interesting case studies are instead provided by the recent introduction of repo facilities of the last resort type available to primary dealers at the Belgian and Portuguese Treasuries. These repo windows should further reduce the risk faced by market makers and allow them to take larger bond positions. This should increase the volume of trading with positive effects on liquidity. The event studies correspond to the following episodes. Introduction of a repo facility of last resort type: BELGIUM 1-04-2001 PORTUGAL 7-08-2000 A step dummy for each country has been included in the 10-year system to evaluate the effect of such episodes on the yield spreads. The introduction of the repo facility in Belgium appears to have had a positive effect on the liquidity of the Belgian benchmark. However, though the effect is significant at the 10% level, the magnitude is negligible with a long-run impact of 2 basis points. The introduction of the same facility by the Portuguese debt office is also significant (at the 5% level) but it displays the opposite effect on the yield spread which rises with a long-run impact of 8 basis points. In fact, the widening of the Portuguese yield spread after mid 2000 is surprising especially since it follows the beginning of the operations on the domestic MEDIP/MTS. A possible explanation of this finding is the low quality of data recording until mid 2000 (see above). 6.4.3 Event study: repurchase programs Decreasing debts and very long maturities have led to a strong contraction of financing needs in the euro area. It is becoming problematic to build large benchmark issues just at a time when the desired volume of bond issues has increased as a result of pan European markets. Consequently, it is becoming common practice to repurchase old illiquid bonds in exchange for new ones, so as to top up more recent bond lines. By reducing the number of funding instruments and concentrating the outstanding debt in few bond lines, such operations should increase the overall liquidity of the debt as well as that of the remaining benchmark bonds. Buy back operations have been carried out by countries like Belgium, Finland, Ireland, Italy and Spain much before the start of the EMU. The Belgian debt agency was probably the first to announce a proper exchange program with dates of operations and types of bonds to be repurchased set in advance according to a specified calendar. After the start of EMU France and Portugal also announced repurchase programs in July 2000 and in January 2001, respectively. The first reverse auction took place in France the 5th of September 2000, while the first repurchase operation in Portugal was the set up of a buyback window for Eurobonds on the 8th of February 2001. The first reverse auction for domestic bonds was instead held the 11th of April 2001. While this date is too close to the end of the sample period to conduct an event study (as our sample ends in June 2001), the start of the French program provides an ideal case to examine the impact of buyback operations on the overall liquidity of the debt. 21 Episodes of first reverse auctions: FRANCE PORTUGAL (not usable) 5-09-2000 11-04-2001 To study the how the launch of the French repurchase program affected the yield spread we have included a step dummy in the system based on the international factors and the on-off-differentials. The coefficient on the step dummy is negative and significant at the 5% level showing that the program has indeed been perceived as a reduction of the instrument fragmentation of the French debt with a positive effect on liquidity. Indeed the long-run impact on the yield spread of minus 5 basis point appears substantial. 6.4.4 Delegation of debt management to an independent agency In many countries debt management is delegated to an agency that can be internal or external to the Treasury. Although there is little evidence (unlike for monetary policy) on the effects of delegation in debt management, it is believe that independent agencies subtract debt management from government pressure, eliminating the temptation to use private information in order to reduce funding costs at the expense of investors and make debt management more transparent. Independent debt agencies, especially if external, operate under performance contracts (or are assigned specific targets) and are closely monitored and must be accountable to the Treasury. As a result of delegation, cost minimisation becomes the main goal of debt managers and this gives rise to a more active but also more transparent debt management. The announcement of delegating debt management to an independent agency may thus send a signal to the market of a new regime characterised by a more transparent and more efficient, cost effective debt management. Over our sample period there have been three episodes of delegation. In Greece, debt management has been delegated to an external agency since July 1999. In Germany, an external debt agency was founded in October 2000 and started operating in June 2001. In France debt management has been delegated to an internal agency since February 2001. The change in responsibility has been associated with the announcement of an interest-swap program that started in October 2001, and with an intensification of the repurchase program. Establishment of an independent agency: FRANCE 8-02-2001 GERMANY 1-10-2000 GREECE (not usable) 1-07-1999 Since the shift in regime is expected to affect the German yield and thus all countries spreads, we introduce a step dummy, which takes the value of 1 from the date in which the German debt agency was established (October 2000). Actually, the agency became operational only in June 2001 but this date could not be tested, since it is roughly the end of our sample period. The step dummy for the introduction of the French agency has only been included in the equation for the French yield spread. The step dummy for the German agency is significant at the 5% level for all countries, with the only exception of Belgium. The coefficient is always positive indicating a widening of the yield spread as a consequence of the delegation of German debt management to the external agency. Consistent with this evidence, the equation for France shows that the yield spread decreases following the 22 establishment of the internal debt agency, after controlling for the change in regime in Germany. Indeed, the dummy for the French agency is negative and significant at the 5% level indicating a long-run impact of about four basis points. This evidence seems to suggest that the delegation of debt management to an external agency has been perceived as a change in regime characterised by a more efficient and transparent debt policy. This is consistent with the belief by many observers that, despite its benchmark status, the German debt is not efficiently managed with too large a number of instruments and an uneven maturity structure. The move to an independent agency may have been viewed as the first step in closing the gap with the more modern management of other member States. This interpretation is however in contrast with the political environment in which the decision to delegate debt management has been taken by the German Finance Ministry. As held by many observers, the move appears to be motivated by the aim to subtract the management of the German debt from the Bundesbank that, though not formally, has long played the main role. To conclude, there is strong evidence of an increase in yield spreads relative to Germany since the last quarter of 2000 with the exception of the French and Belgian yields. In principle, this increase can be attributed to the announcement of the external debt agency in Germany, but the debate that surrounded its creation casts serious doubts about this interpretation. However, if the reason for the increase in yield spreads must be found elsewhere, what can safely be concluded is that liquidity effects (with the important exception considered below) are hardly part of the explanation. We have examined the most significant events that one would expect to have an impact on yield spreads and found little evidence of their importance except for innovations occurring in Belgium and France. The important exception is the impact on yield spreads of changes in trading volumes in futures markets that we have not been able to examine because of data availability. 7. CONCLUDING REMARKS In this paper we have shown that euro-zone yield on government bonds, more precisely, their credit-risk (and liquidity) component have been increasingly correlated across issuers, a sign of enhanced integration which is possibly explained by the elimination of the risk of exchange-rate fluctuations. Interestingly, the risk and liquidity component appears to have been greater after the EMU. Our study suggests, contrary to market perceptions, that yield differentials on euro-zone bonds are mostly due to credit risk as opposed to liquidity factors. The analysis of monthly series points to the existence of different, slowly evolving means, as well as to the importance of common international factors in the determination of fluctuations around the means. Slowly evolving means are related to economic fundamentals such as the ratio of government securities to the total outstanding securities and, most likely, to debt and deficit to GDP ratios. At higher frequencies yield differentials appear to be almost completely explained by common credit-risk related international factors such as the US and German asset swap spreads and US and EU policy rates. The main source of yield differentials appears to be the different sensitivity to common, credit risk-related, international factors. By contrast, only the yield spread between onthe-run and off-the-run benchmark bonds (of the same issuer) appears to significantly affect yield spreads across member States: traditional liquidity indicators, such as bid-ask spreads, trading volumes and outstanding amounts have no effect. 23 Attempts to introduce dummies, depicting structural changes in secondary and primary market structure, that may have increased the liquidity of the bonds of the reforming country, were also not particularly successful. Most of the times such changes are not significant and, when significant, their impact appears small. However, such evidence does not allow us to conclude that liquidity is irrelevant. “Structural” liquidity factors could indeed explain the different sensitivities to international factors. For instance, the hedging facilities provided by the Eurex futures market are expected to have an impact on German yields relative to other member States, even though we have not been able to introduce any variable capturing such effect. Indeed, the basis risk of managing a position is not present in the case of German bonds, while it is non-negligible for the other members. This may well have justified increasing liquidity premia over the past few years, and episodes of flight to quality may have heightened the importance of the basis risk in hedging non-German bond positions with Eurex contracts. Furthermore, the fact that bid-ask spreads and trading volumes do not significantly affect yield spreads may simply reflect a data availability problem. Indeed, the liquidity indicators that we have used may not be representative of the overall market situation; we had to rely on indicators taken from the MTS database instead of the more relevant but unavailable data on over-the-counter activity. On the other hand, evidence of little differences in liquidity premia could not be spurious; it may reflect an equilibrium phenomenon. It is worth noting that the need to compete for funds in European markets with instruments characterised by a common currency denomination has led EMU sovereign borrowers to grant concessions to primary dealers. The bargaining power of primary dealers has increased because of their crucial role in promoting national bonds abroad. Most countries no longer require primary dealers to be locally based. Furthermore, the time available for non-competitive bidding in second round auctions has been extended in a number of countries. Syndication and even mandates for privatisation of public assets may all be perceived as ways to ‘pay back’ some of the costs of doing business to institutions providing liquidity to the government bond market. Other elements may come as part of tradeoffs. Although such concessions may not show up in bond spreads, they may be considered as offsetting part of the overall costs of market making10 and therefore be part of a liquidity premium that does not show up in market data. 10 Market makers, i.e. those players making secondary markets on bonds, may not be the same institutions being part of the placement consortium (also called group of primary dealers or specialists in some countries), and may also be different from the more closer group that gets mandates for international issues or for Treasury’s related activities (M&A etc.). 24 REFERENCES Campbell J.Y. and Cochrane (1999) “By force of habit: a consumption-based explanation of aggregate stock market behaviour” Journal of Political Economy, 107, 205-251 Favero C.A, Giavazzi F. and L. Spaventa (1997) “High Yields: The Spread on German Interest Rates” The Economic Journal, 107, 443, 956-986. Favero C.A., Missale A. and G. Piga (1999) “EMU and Public Debt Management: One Money One Debt?” CEPR Policy Paper No.3, December Fleming M.J. (2001) “Measuring Treasury Market Liquidity”, paper available at www.newyorkfed.org/rmaghome/economist/fleming/fleming.html Fleming M.J. and E.M. Remolona (1999) “Price Formation and Liquidity in the U.S. Treasury Market: The Response to Public Information”, Journal of Finance, 54, 5, 1901-15 Gravelle T. (1999a) “Liquidity of the Government of Canada Securities Market: Stylized Facts and Some Market Microstructure Comparisons to the United States Treasury Market, Bank of Canada Working Paper No. 99-11. Gravelle T. (1999b) “The Market Microstructure of Dealership Equity and Government Securities Markets: How They Differ” in Market Liquidity: Research Findings and selected Policy Implications BIS-CGFS Study No. 11, May. Gravelle T. (1999c), “Markets for Government of Canada Securities in the 1990s: Liquidity and Cross-Country Comparisons,” Bank of Canada Review (Autumn) 9-18. Kumar M.S. and A. Persaud (2001) “Pure Contagion and Investors’ Shifting Risk Appetite: Analytical Issues and Empirical Evidence”, IMF Working Papers 25 Appendix 1: The data set The daily data set we use runs from June 1, 1999 to August 30, 2001. We consider observations in each main segment of the yield curve, i.e. 3, 5, and 10 years, for all EMU members except Luxembourg and Greece. As for Luxembourg, data simply do no exist, while for Greece reliable series are not available. Our sources of data are Datastream and Prisma. We use mainly Prisma for daily asset swap and yield data. As for monthly data we use Datastream/Thomson Financial. It provides the longest series among the data providers we took into consideration, although not many maturities are available. All working days in which prices are recorded for several markets, either in London or domestically, are included in our EMU-wide database. Week-ends and holidays shared across Europe are therefore excluded. When it is a market holiday in a specific country while not in the others and no prices are recorded in London, we consider the observation as missing value. Data from Datastream are not subject to any filtering procedure while row data form Prisma have been e filtered only to cut off clear outliers (3 standard deviations away from previous day’s observation). Missing observations and outliers have been dropped in the estimation. Swap and bond rates are mid-rates (average between bid and ask) recorded daily before market closing. Key issues for our data set are: Interbank credit standing affecting asset swap spreads Government bond asset swap spreads are relative measures that reflect government-specific factors as well as the credit standing of the interbank market or, more generally, that of the domestic private sector. While the credit standing of the interbank market is generally perceived to have remained stable during the 1990s in the Euro area, this element may not have been totally negligible in specific sub-periods and may have played a role in the determination of Euro-area asset swaps. This was probably the case during the Russian crisis in 1998. Moreover, before Monetary Union the credit standing of domestic banking systems was not perceived to be totally homogeneous among current EMU member states. This may have given rise to non-homogeneous changes in asset swap spreads across EMU countries and therefore may have affected differentials for reasons not linked to the variables considered in our study. We assume this ‘interbank’ factor to have remained largely negligible on swap differentials during the period before the launch of the single currency, while for the post-EMU period under examination, this problem has clearly disappeared. Different approaches to calculating asset swap spreads There are three main approaches that can be used to calculate asset swap spreads for benchmark maturities: (1) what we call the ‘quick and dirty’ approach, (2) the ‘constant maturity bond’ approach, and (3) the ‘benchmark’ approach. There are trade-offs when choosing one approach over the other. 1) Calculating ‘quick and dirty’ differentials between swap and bond rates can be considered sufficiently reliable for most purposes. However, it causes several problems when trying to estimate and compare asset swap spreads among countries and for high frequency data. First, the maturity of benchmark bonds can significantly differ from the standard maturity of swaps (this problem may be partly offset by interpolating swap rates). Second, comparing bonds with 26 different maturities and duration can produce differences in yield differentials, and therefore in asset swap spread differentials, that are linked to the shape of the yield curve rather than to credit, liquidity or other factors. 2) The second approach would call for an estimation of the term structure of interest rates from swap rates, the zero coupon curve from bond data, and then ‘constant maturity bonds’ for each of the main benchmark positions. This may give results that are too reliant on the model used. The major problem relates to the estimation of zero coupon curves in the bond market, as a certain degree of subjectivity must be introduced. For example, using models based on exponential splines necessitates judgements on inflection points that would be subjective and difficult to make retrospectively. Other methods that do not take into account the specifics of each government bond curve risk losing a lot of information. Finally, matching liquidity indicators specific to a single benchmark bond may not be appropriate. Despite these drawbacks, this approach gives the opportunity to analyse asset swap data that always refer to the same point along the term structure of interest rates. Prisma benchmark data that we choose for our study are based on this approach. Benchmark bond yields are calculated by using a bond model which uses least square fitting of a smooth curve through pre-defined maturities at 1.5, 4.5, 8.5 and 28.5 years. This gives constant maturity bond yields for each maturity. The values have been estimated to be almost indifferent to the model specification. We find it appropriate for the purpose of our study, while inevitable distortions are perceived as acceptable. 3) The ‘benchmark’ approach refers to the use of asset swap spreads for specific benchmark bonds. This has the advantage of relying less on any model of term structure or any approximation of benchmark yields (only modelling of the swap term structure is required), and allows matching of specific indicators of liquidity to the bond to which they refer. Given a pricing curve, asset swap spreads are defined as the constant spread we have to add on the curve in order to correctly re-price a bond. p ai cf i df ti e oast i i Where: p ai cfi df(ti) ti = bond spot price = bond accrued interest = i-th bond cash flow = swap curve discount factor at time t = time between the bond settlement date and the i-th bond cash flow payment date on an ACT/365 basis. 4) Since there is no closed form solution, the equation is solved numerically. The drawbacks of this approach are mainly related to the somewhat subjective definition of a benchmark bond and to the bond-specific nature of the measure which may not be representative of the market situation in that maturity range for the sovereign borrower. Data providers assign the benchmark status to bonds according to different criteria, and therefore bonds are benchmarks in different time intervals. Liquidity criteria used to decide whether the bond is ‘on-the-run’ or ‘off-the-run’ are inevitably subjective. In addition, the benchmark status also depends on the admission criteria of electronic markets. Moreover, the distortion due to the shape of the curve is not eliminated, although it only refers to the term structure of asset swap spreads. In our modelling exercise we also tried to use Prisma bond-specific asset swap data calculated according to this approach for the on-the-run/off-the-run liquidity indicator. However, there were missing data in the series that suggested use of Datastream/Thomson Financial yield series for this indicator. 27 3. Volumes data Bond markets have been mainly over-the-counter in the past, and therefore few statistics are available on a consistent basis and for a long period of time across EMU countries. In addition, the market share of electronic markets substantially differs among EMU countries, and therefore even the information contents of data are not the same. Settlement data on volumes are not useful for our purposes as volumes on repurchase agreements, basis trading, refinancing operations with the central bank, and primary market operations are all included, and it is almost impossible to net these volumes out to obtain pure cash transactions on the secondary market. Although other electronic initiatives were launched, up until very recently the EuroMTS 11 trading platform was the only one with a significant presence across several EMU countries and therefore our sole source of homogeneous data. The rest of the market was, and still is, mainly over-thecounter and no reliable, consistent, and sufficiently long series can be obtained on liquidity indicators. Nevertheless, EuroMTS only started operating on March 30, 1999, and at the beginning only the bonds of three countries were included. It was only by February 2000 that the sample expanded to cover nine EMU countries. Therefore, we have no choice but to further shorten our daily database in order to include liquidity indicators and narrow the maturities that have been considered. Euro/MTS data are available for the 10-year maturity, although some data are also available for 2/3-year, 5-year, and 30-year benchmarks in main government bond markets. A table at the end of the Appendix summarises volumes traded by EuroMTS in EMU government bonds and the date when electronic trading started. Volumes for benchmark issues record big jumps in the series even when we allow for a shift in the benchmark status only when there is a sustainable pick-up in trading volumes.12 EuroMTS has a rough idea of its market share in each country, although there is no official reliable source of information on total daily trading volumes (the sum of electronic and over-the-counter volumes). As can be seen in the table at the end of the Appendix, EuroMTS business is still concentrated on the Italian market. In our liquidity indicator on volumes we take relative volumes, and this is perceived to have offset the impact the larger volumes on the Italian market as the relative trend in volumes is what is captured in the model. 4. Bid-ask spreads EuroMTS was able to calculate and provide us with daily averages of market makers’ bid-ask spread of ‘proposals’ inputted into the system for each benchmark security. This has involved significant computational work, and is the best proxy of ‘market best’ bid-ask spreads. While this information is extremely useful and important for the purpose of our study, considerable doubts remain on the significance of these data in depicting liquidity conditions in the market. Market makers are ‘forced’ to quote securities with a tight bid/ask spread according to specific electronic market obligations, although these ranges are not really binding for benchmark bonds. 11 EuroMTS refers to the sum of the activity of MTS SpA, EuroMTS Ltd and other MTS companies in EMU countries. 12 Days in which volumes are higher than the previous benchmark and which mark the turn in relative volumes. 28 However, ‘market best’ bid/offer spreads are generally much narrower than quoting obligations, and therefore we perceive our indicator may not be biased for this reason. A more serious distortion relates to the averaging process of bid/ask spreads of proposals during the day. The introduction of unusually large bid/ask proposals into the system may affect the results, and therefore data were filtered to cut off clear mistakes or abnormal spreads. Still, large bid/ask spreads during periods when the market is not particularly active tend to have a disproportionate effect on the daily averages. To increase the significance of bid/ask spreads on proposals, we also tried to weigh them by means of the number of proposals inputted into the system during the day. Unfortunately, there are structural differences in domestic markets that allow for disproportionately strong activity (high number of daily proposals) when basis trading is particularly active, as it is the case in Germany. Therefore, this weighting did not work. The bid/ask spread on ‘market best’ would be our preferred measure, but this information is not yet available. More fundamentally, we believe there is a problem of significance of EuroMTS trading for each specific market. Trading executed on EuroMTS is perceived to be highly significant for capturing overall market conditions in Italy and Belgium for instance. While we perceive the activity of German government bonds is not fully representative of conditions in the market, It is believed that in each specific market price discovery differs. In the German market, price discovery is probably in the futures market. In Italy and Belgium it is in the electronic cash market. In other countries it is still in the over-the-counter inter-dealer brokerage market. Moreover, these conditions appear to change over time. 29