Empirical examination of the Greek non-government bond market Nicholas T. Kokores Panteion University of Athens, Greece ABSTRACT The introduction of the euro has forced the development of new products and greater investor demand for instruments with higher return/risk profiles and greater mathematical and computational complication such as credit derivatives, asset backed securities and inflation linked bonds. In Greece the bank debt securities have been the dominant force in the non-government bond market, due to the reshaping of regulatory framework under the ongoing Basel II process and the consolidation of European capital markets after the initiation of monetary union. This paper offers an empirical examination of the development of this market using macroeconomic factors such as interest rates and GDP. Furthermore, this research builds upon the existing studies by looking into the underlying factors that explain the price differential in the non-government bond market and which have implications for the mathematical modeling of individual bonds. I) Introduction After the introduction of euro many categories of investors that prefer assets denominated in the local currency where able to increase the diversification of their investment within the euro area. This case study focuses on the Greek market but for comparative purposes we include findings from the American and the European bond markets too. The government bond market still retains the leading position in the Eurobond market mainly due to the creditworthiness of the borrowers, the high liquidity of the issues and the well-developed derivatives market. -1- The secondary market activity mainly takes place in the “over the counter” (OTC) market. Non-government issuing activity has increased substantially since the introduction of the euro. Specifically, according to ECB data from end-2000 to end-2003, the outstanding amount of non-government debt securities issued by euro area residents rose from 3,052 billion to 4,558 billion euros, which is an increase of almost 50%.1 One explanation is that, the low interest rate environment present in the bond and money markets drove investors towards higher-yielding investments like the corporate bonds. However, the lack of a common hedging instrument remains a serious obstacle for the corporate bond market. Corporate and financial bonds are often issued in international financial centers, such as Luxemburg, which has the predominant position among the banking locations. There is no argument about the domination of bank debt securities in the non-government bond market for all euro area countries (ECB 2004a). Furthermore, among Greek bank bonds, the percentage of small issues – below 500 million euros – was about 90% while in the rest of the euro-area was well below 40% in 2004 (ECB 2004b). In addition to the above, financial firms, such as banks, tend to have leverage ratios above 90% where on the other hand, non-financial institutions use much lower debt ratios. Another major difference between bank bond and government bond sector is the coupon structure. According to ECB data 65% of euro-area government bonds still have fixed rate coupons, while in financial bond market this percentage is substantially lower (ECB 2004a). Therefore, the patterns of coupon structures in government and financial bond market do not allow for a direct calculation of yield differentials. In order to surpass this obstacle a transformation of the standard yield calculation is needed that is presented latter on. 1 For a detailed analysis of the euro-area bond market see (ECB 2004a). -2- II) Theoretical Background and Methodology The theoretical literature on corporate bond pricing (credit risk valuation) can be segmented into two categories: 1) structural (contingent claim) credit risk models and 2) reduced form models. The classical “structural” method of valuation for default risky bonds is based on the works of Merton (1974), Longstaff and Schwartz (1995) and Collin-Dufresne and Goldstein (2001) while the more recent “reduced form” model is used by Duffie and Singleton (1998) and Lando (1998). One of the pivotal differentiations between these models is that the first one attempts to model the asset value and capital structure of the firm, whereas the latter uses proxy variables for the capital structure in order to model yield spreads on the basis of differing default probabilities. This paper is based mainly on the second category but we also use findings from the first category. In this paper debt securities (bonds) refer to securities other than shares excluding financial derivatives, which should cover issues by resident entities (originator) irrespective of the currency and market of issuance. Moreover, we do not use bond indices for our analysis since they tend to present a filtered universe of the outstanding debt markets due to the criteria of bond selection that are enforced. Our sample consists of weekly data covering prices of Greek bank bonds for the period from September 2000 up to October 2004. The bond prices are collected from Bloomberg Database that produces composite prices for each bond. Bonds classified as non rated are excluded from the analysis. Furthermore we eliminate such bonds as securitized bonds and quasi-government bonds. Yield differentials (spreads) are mainly created by the government bond yields themselves, as the yield spread of euro area government bonds over German bonds – also called “Bunds” – which are the “benchmark” for the euro-area bond market. The major obstacle that we face when trying to calculate yield spreads for floating rate bonds is that we cannot readily predict the cash flows because the coupon payments reset periodically, based on a reference rate such as the threemonth “Euribor” rate (Euro Interbank Offered Rate) posted daily by the ECB. The way to overcome this difficulty is to compute the discount margin, which is the -3- margin relative to the base index rate such that the present value of cash flows equals the price plus accrued interest of the bond. A second way is to calculate the yield to maturity spread of a bond as the difference between the yield to maturity of the floating rate bond and the yield to maturity of the index rate: YTMSpread= YTMFloat – YTMIndex. In this study we use the second calculation method. The variables that we are going to examine as the driving forces of this yield spread are presented below. III) Description of proxy variables and testable hypotheses 1) Default risk parameter For modeling default risk, one proxy variable is implemented in the current study. We are using the ratings announced by Standard & Poor’s and Moody’s which capture the effect of both probability of default and the recovery rate. Recent research incorporates ratings as a parameter of default risk. Specifically, Elton et al. (2001) use Moody’s Standard & Poor’s (S&P’s) bond ratings in order to depict a proxy of company specific financial ratios. However, in this study we are going to use a combined rating scale. The first step is to transform S&P’s and Moody’s ratings into two distinct numerical scales (for S&P’s from AAA=1 down to D=21 and for Moody’s from Aaa=1 down to D=21). We then combine the two scales into one composite measure by averaging the two numerical ratings for each bond. For instance an entity rated Aa1=2 and AA=3 would be rated 2.5. Therefore, we assume that S&P’s and Moody’s ratings indicate the same level of risk. Furthermore we expect a positive correlation between this variable and the yield spread meaning that a higher number on the derived rating scale (poorer actual rating) will result to a higher yield spread for the bond. 2) Liquidity parameter According to many academics, an investor should expect some compensation for the liquidity offered in the corporate bond markets versus the government bonds market. Specifically, Collin-Dufresne et al. (2001) find evidence that liquidity significantly influences credit spreads changes. Liquidity is also -4- important because some institutional investors might not be able to participate in a corporate bond issue if it is of less than a specific benchmark size, commonly set to 500 million euros. The ease with which a security can be traded in the market (liquidity) is best depicted by the difference between the bid price (the price at which the holder can sell securities) and the offer price (the price at which the purchaser can buy securities) also known as the bid-offer spread. However, for the “over the counter” market this spread is difficult to be calculated with accuracy since there is a lack of data from the bond dealers. Therefore we use an indirect measure of liquidity, namely, nominal amount outstanding for each bond issue. We expect a negative relationship where the higher outstanding amount should be a driver towards lower yield spreads. 3) Maturity parameter Another reason why bonds might be perceived by investors as carrying different risk is the “age” of the bond meaning the term to maturity. However, there is no definitive evidence from the finance literature that bonds tend to differentiate in relation to their maturity. Nevertheless, the maturity structures of each issuer are closely linked to the debt requirements that are faced by these issuers. Furthermore, a common restriction imposed in most of the empirical analysis on bond markets is to exclude bonds with “optionality” (call or put options).2 However, in this analysis we include callable bonds by shortening their maturity down to the call date since in all the Greek outstanding issues of our sample the call feature shows a 100% certainty of execution. We expect a positive relationship, where the increase of the bond’s maturity results to higher yield spreads. 2 The alternative might be to construct a model that explicitly prices the option effect but most of the empirical analysts prefer to examine the determination of credit risk separately from the option pricing. -5- 4) Interest rates parameter There is an argument that lower interest rates are usually associated with a weakening economy and therefore higher credit spreads. Specifically, according to evidence provided by Duffee (1998) for fixed rate corporate bonds, a negative relationship exists between changes in credit spreads and interest rates. We use two proxies for the interest rate variable, one for the short term rates and another one for the slope of the curve (term structure of interest rates). As a proxy for short term interest rates we use the three-month “Euribor” rate to capture trends in the closely related to bonds, money market and also, mainly because around 90% of the Greek financial bonds are floating rate notes that have coupon payments as a spread based on such a rate. As a proxy for the slope of the interest rate curve we use the spread between the ten year and the two year yields of the euro-area government bond curve. 5) Parameter for macroeconomic factors We are going to use two proxies for the macroeconomic environment. The first one is the monthly rate of change for the annualized Gross Domestic Product (GDP) inputted in order to capture the effect of output growth, and the second one is the monthly rate of change for the annualized Consumer Price Index (CPI) to capture the effect of inflation. We expect the first one to be negatively and the second one to be positively correlated with the changes of the yield spread. IV) Results In Table 1 below we present the findings of our model estimation in relation to each proxy. The partial regression coefficient estimates β1, β2,... ,βκ=7, are the parameters which quantify the individual effect of each of these explanatory variables on Y (x1, x2, x3, …, xn=6, explanatory variables). The dependent variable “Y” is calculated as the natural logarithm of the difference between yield to maturity of the floating rate bond and the yield to maturity of the base index (YTMSpread= YTMFloat – YTMIndex) in order to depict rates of change for the credit spread. -6- Table 1: Regression results for the filtered sample Expected sign Coefficient t-value -8.5392 -365.04 Variable Intercept Rating Outstanding Amount Maturity / Call Date ST Interest Rates Curve Slope + +0.1489 175.13 - -4.7312 -41.10 + +0.4225 67.59 - +0.1327 -5.7625 85.78 -12.39 GDP_an - -7.35 E-6 -2.35 CPI_an + +6.90 E-5 4.77 P-value Data source 0,0091 Published announcements 0,0087 Author’s calculations 0,0127 Companies’ Publications Companies’ Publications 0,0191 Author’s calculations 0,0375 Bloomberg, ECB 0,0288 Bloomberg, ECB National Statistical Service 0,1132 -Greece/ Bloomberg National Statistical Service 0,1283 -Greece/ Bloomberg As it is depicted above, all coefficients are significant except for the GDP and CPI indicators that are not significant even at the 5% significance level. An explanation might be that even though the bonds of our sample all have Greek originators, the macroeconomic factors that affect them might be closely related to the overall euro-area rather than just the Greek one. In addition to the above, all have the predicted sign except for short term interest rates which resulted to the opposite sign from the one expected, signaling that changes in the three-month money market rate cause spreads to move in the same direction. A qualitative explanation of this might be that, low short term interest rate levels induce investors to shift their preferences from low to high-risk assets. Such a trend results in a decrease of the yield spreads differentials. Overall the model has a R2 equal to 0.7823 and an adjusted R2 of 0.7582, while the model’s Durbin-Watson statistic is 2.06. The interest rate premium therefore compensates the investors for the higher default risk of a financial institution with a poorer credit rating. Furthermore the decomposition of credit spreads shows that the risk premium increases substantially for lower-rated debt. -7- V) Conclusions This study mathematically tests a model that examines the relationship between Greek financial bonds yield spreads and its dependence to specific underlining factors: default, liquidity, maturity, interest rates and macroeconomic indicators. The results add to the previous work that suggests the significance of these factors. Furthermore, the empirical findings emphasize the deepening of the Greek non-government bond market. Thus we can conclude that, although the European and the Greek non-government bond markets differ in term of number of bonds, the empirical results for both markets are quite similar. References Collin-Dufresne, P., Golstein, R., and Martin S., (2001). The determinants of credit spread changes, Journal of Finance, 56, 2177-2208. Duffie, D., & Singleton, K. J. (1998). Econometric modeling of term structures of defaultable bonds, Working Paper, Graduate School of Business, Stanford University. ECB, (2004a). The euro bond market study, December 2004. ECB, (2004b). Developments in euro-denominated bond issuance in 2004, Quarterly Bulletin, No 79, October – December 2004. Elton, E., Gruber, M., Agrawal, D., Mann, C., (2001) Explaining the rate spread of corporate bonds, Journal of Finance, pp.247-277. Lando, D. (1998). On Cox processes and credit risky securities. Review of Derivatives Research, 2 (213), 99-120. Longstaff, F. and Schartz, E., (1995). Valuing risky debt: A new approach, Journal of Finance, 53, 1213-1242. Merton, R. (1974). On the pricing of corporate debt: the risk structure of interest rates. Journal of Finance, 11, 288-300. -8-