TITLE. Performance aspects of Greek bond mutual funds

Transcription

1 TITLE Perormance aspects o Greek bond mutual unds Dritsakis Nikolaos, University o Macedonia Grose Christos, University o Macedonia Kalyvas Lampros, Bank o Greece and University o Macedonia Dr. Dritsakis Nikolaos 1 Associate Proessor o Econometrics University o Macedonia Department o Applied Inormatics P.O. Box 1591, 54006, Thessaloniki, Greece Phone: drits@uom.gr Mr. Grose Christos Research Associate University o Macedonia Department o Applied Inormatics P.O. Box 1591, 54006, Thessaloniki, Greece Phone: grose@uom.gr Mr. Kalyvas Lampros Bank Examiner Bank o Greece Department or the Supervision o Credit and Financial Institutions 3, Amerikis Street, 10250, Athens, Greece Phone: lkalyvas@bankogreece.gr Research Associate University o Macedonia Department o Applied Inormatics P.O. Box 1591, 54006, Thessaloniki, Greece 1 Corresponding author

2 Perormance aspects o Greek bond mutual unds Abstract This paper examines the perormance characteristics o Greek bond unds and the impact o und lows on portolio returns. The evidence shows that on average bond unds do not oer risk-adjusted proits exceeding the returns o the benchmark index, consistent with the US and international evidence. Returns beore ees are slightly superior to the returns o the benchmark index, but when ees are considered they lag considerably. The security selection and market timing skills o und managers are also tested using both an unconditional and a conditional model to test or the impact o public inormation variables. We also ind that und lows impact negatively on market timing. JEL Classiication: G15, G23 Keywords: perormance evaluation, bond unds, conditional model, und low 2

3 1. Introduction It is indubitable that the dramatic advance o portolio evaluation techniques, which involves as its primary evolution the replacement o gross returns calculations by detailed explorations o risk and return and the sources o each, has compelled investment organizations to incorporate evaluation as an integral part in any decisionmaking process. Perormance evaluation literature concerning active unds has been widely addressed worldwide and the majority o studies conirm the inability o active unds to outperorm market benchmarks. Among the major studies exploring the perormance evaluation literature we ind articles by Cumby and Glen (1990), Malkiel (1995), Gruber (1996), Blake and Timmerman (1998), Sawicki and Ong (2000). The Greek perormance evaluation literature is relatively small and is basically limited to studies on equity mutual unds (Milonas, 1999; Artikis, 2002). The same applies or the international perormance evaluation literature since most studies analyze either the equity unds market or diversiied unds that invest both on equities and non-equities securities. It is thereore surprising that bond unds perormance has received almost no coverage especially given the act that they comprise 1/3 o the unds invested on mutual unds in the Greek market. The total value o unds invested in this category o investments based on the latest data oered by the Association o Greek Institutional Investors is in excess o 6 billion euros and represents 5% o Greek GDP. This study attempts to ill the gap in the evaluation literature and provides useul insights both or investments proessionals in Greece and abroad since the Greek market has recently been classiied amongst the developed markets. This evolvement 3

4 has led to the signiicant increase o unds invested in Greece and Greek 10 year bond returns have experienced a signiicant decrease in yield summing up to just 30 basis points above the corresponding German bonds that are used as benchmark. Most studies evaluating the perormance o bond mutual unds examine the U.S. market and their common conclusion is that bond unds do not outperorm passive benchmarks (Blake et al., 1993; Elton et al., 1995). Detzler (1999) investigates the returns characteristics o global bond portolios using various benchmarks and inds that no abnormal returns are observed. Other studies take up the issue o the dierence between high-grade and low-grade unds (Cornell and Green, 1991) but conclude that no signiicant perormance dierences are ound. In contrast Blume and Keim (1987) and Blume et al. (1991) ind that indeed lower grade bond portolios earn signiicantly higher returns than higher grade bond portolios. In order to evaluate the returns structure o bond mutual unds in Greece both an unconditional and a conditional method is used. Most studies on perormance measurement use the unconditional approach even though as Ferson and Schadt (1996) point out this may lead to biases due to common time variation in managed und risks and risk premia. Few studies worldwide have utilized the conditional method and none have applied it to bond unds. In this respect this study oers a unique approach examining the Greek market with both methodologies, oering thus more robust results. The conditional model in addition to the benchmark index incorporates public inormation variables namely treasury note yield, term structure o interest rates, the Athens Stock Exchange (ASE) returns as a measure o economic activity and a dummy variable or the month o January. 4

5 In this study we also estimate the impact o und lows on the perormance o Greek bond unds. The limited studies concerning the impact o und lows conclude that exogenous und low shocks have a negative impact on perormance (Edelen, 1999). Edelen goes urther stating that in perormance measurement o open-end unds liquidity driven trading adversely aects market timing and security selection. The negative market timing elements according to Edelen are attributed to liquidity motivated trading where und managers are eager to invest the inlows o unds. Such phenomena are more evident at periods o extreme market volatility. Other studies concerning low include Warther (1995), Ferson and Schadt (1996) and Edelen and Warner (1998). In this study we attempt to interpret the impact o lows on perormance both using traditional perormance methods and based on publicly available inormation. The rest o this paper is organized as ollows. Section 2 explains the methodology used in measuring investment perormance or Greek bond unds. In Section 3 the data used in the analysis are outlined, while Section 4 analyses the empirical results. Comments on the implications o the indings are given in the Conclusion. 2 Methodology 2.1 Perormance Measurement Unconditional Measures The traditional literature on perormance evaluation has been based on the work by Jensen (1968) where the abnormal excess return on unds is estimated by alpha, which is calculated by regressing the returns o the und in excess o the risk-ree rate on the 5

6 returns o the market portolio in excess o the risk-ree rate. Jensen s model is outlined as ollows: R = α + β R + ε (2.1) m where R is the return o the active und in excess o the risk-ree rate α is the excess return coeicient o the active und during the examination period β measures the sensitivity o the und returns relative to the benchmark portolio used R m is the return o the market portolio in excess o the risk-ree rate ε is the residual term Treynor and Mazuy (1968) extended Jensen s work by proposing the use o a quadratic term measuring the market timing element o und managers behaviour. They suggested that when markets are expected to grow und managers are likely to invest more heavily on the market portolio o risky assets. On the contrary they are expected to hold a smaller proportion o the market portolio assets when they expect markets to all. Thereore Treynor and Mazuy try to capture both market timing and security selection elements in und managers behaviour. Where the coeicient γ is signiicantly positive market timing elements are present. Their model is the ollowing: R = α + β R + γ R + ε m 2 m (2.2) 6

7 where γ measures the market timing elements 2.2 Perormance Measurement Conditional Measures Initially Ferson and Schadt (1996) proposed the use o public inormation variables to extend the unconditional perormance evaluation methods to conditional perormance evaluation methods that cater or the changes in risk over time. Indeed in other studies as well it is ound that when unconditional methods are used or perormance evaluation negative alphas are ound more oten whereas when a conditional method is used results are dierent (Becker, Ferson, Mayers & Schill, 1999). Using the conditional approach the predictive ability o publicly available inormation is utilised and potential biases incorporated in unconditional methods are bypassed. Fund managers are usually able to identiy such market anomalies as noted rom public inormation variables and when these are exploited they may be granted or superior results. In order to account or the publicly known variables included in the conditional model to the traditional perormance evaluation model a vector o lagged public inormation variables is added. The calculation o the model involves the multiplication o the market return variable with each o the lagged public inormation variables irst. R = α + β R + δ ( R P 1 ) + ε (2.3) m m t where 7

8 α measures the risk-adjusted perormance δ measures the coeicients o the lagged public inormation variables P t 1 is the vector containing the public inormation variables lagged or one period There have been a number o empirical studies concerning the public inormation variables that should be used in conditional models. These studies have primarily relied on the analysis o equity unds but their indings could be extended to bond unds as well. Chen, Roll and Ross (1986) and Fama and French (1993) initially investigated actors that inluence stock returns. Elton et al. (1995) emphasised that actually the same actors that are used to explain stock returns should be used on nonequity unds analysis as well. Ferson and Schadt (1996) propose the use o the ollowing lagged public inormation variables or the measurement o the conditional alpha or mutual unds treasury note yield, term structure o interest rates, dividend yield, a corporate quality yield spread and a dummy variable or the month o January. According to Sawicki and Ong (2000) Australian equity unds returns are analysed but the corporate yield spread is not included as an explanatory actor o und returns and as a broader measure o economic activity they use both the dividend yield and the stock exchange General Index instead in separate regressions. In our study only the Athens Stock Exchange returns are used as an explanatory variable, since they are considered as a good proxy or industrial production and corporate proitability and data on the market s dividend yield were not readily available. Alternative studies regarding the Greek market have utilised alternative indices, instead o the Athens Stock Exchange General Index, arguing that the ASEGI cannot approximate the market portolio (Artikis, 2003). For the purpose o 8

9 introducing an economic conditions variable-a proxy or industrial production, corporate proitability and general economic growth though, the ASEGI is indubitably the best possible measure as reerred in other studies as well (Fountas and Segredakis, 2002). Furthermore, unreported analysis that we perormed, using as explanatory variables the lagged market returns and the dependent variable lagged returns (Jegadeesh and Titman, 2001), provide not signiicant results and are not included in the analysis. Thereore, in (3) we use as public inormation variables the term structure o interest rates, the Treasury note yield, the Stock Exchange returns and a dummy variable or the month o January. In act Sawicki and Ong (2000) ind that stock exchange returns have stronger explanatory power relative to the dividend yield. The January anomaly is a pervasive, well-documented anomaly in the inancial markets. The anomaly has been acknowledged in stock returns, corporate bond returns and yields, as well as municipal bond returns (Kihn, 1996). A number o studies ind supporting evidence o the existence o such an anomaly in the corporate bond market (Chang and Pinegar, 1986; Chang and Huang, 1990; Fama and French, 1993; Maxwell, 1998), underlying its importance in perormance measurement. Consequently the inclusion o a dummy variable or January was considered necessary to test or the existence o an analogous anomaly in the Greek bond market. An additional conditional perormance model is estimated to test or market timing adding the squared market returns relative to model (3). R = + β Rm + δ RmP ) t 1 α ( + γ R + ε (4) 2 m 9

10 We run the above regressions both on net returns (net o expenses) and gross returns basis. This means that bond mutual unds returns are also calculated including ees that are charged both in entering a bond und and when the investor liquidates his investment. When expenses are calculated returns are signiicantly altered and we test on separate regressions the nature o the inluence both or the unconditional and conditional models reported above. 2.3 Fund Flows and Perormance An area in the empirical inance literature that has attracted increased coverage in recent years is the eect o und lows on perormance measurement. It is important to analyze this inluence so as to be able to predict the nature o changes occurring on unds perormance as a result o inlows and outlows, which are very common especially during periods o bull and bear markets. The inclusion o und low as an explanatory variable in measuring perormance gives better insight to the liquidity service provided to stakeholders o mutual unds. Edelen (1999) inds a negative relationship between und-lows and perormance driven by the act that und managers engage in liquidity-motivated trading in periods o increased inlows thus ignoring the market timing element. For this reason market timing coeicients are in most cases negative and insigniicant. Warther (1995) and Edelen and Warner (2001) ind a positive correlation between market returns and und lows strengthening the negative market timing eect mentioned previously. These results render clear the inability o either an unconditional or conditional approach that ignores lows to identiy the market timing ability o a und. 10

11 An additional question posed by many researchers is why there is a negative covariance between und betas and market returns (Ferson and Schadt, 1996). This means that und managers reduce their market betas when they expect market returns to go up and vice versa. The most commonly accepted explanation is that when market returns are expected to rise there is an inlow o unds and managers usually do not invest the incoming unds on time so that portolio betas all. The size o the all in betas is analogous to the size o inlows (Sawicki and Ong, 2000). An alternative explanation cites the possibility o this negative covariance being attributed to diering betas o the underlying assets comprising the und portolio (Ferson and Schadt, 1996). In this study we incorporate und lows as an explanatory variable both or conditional and unconditional models testing or their inluence on returns. We include net und lows meaning that we calculate the dierence between actual inlows and outlows thus being able to make inerences o und low activity on the managed und s returns. During the period under examination bond unds experienced signiicant dierences in the unds under managed thus making our conclusions more robust. In order to estimate net und lows we calculate the dierences in total unds under managed within time intervals t and t-1 adding the appreciation/depreciation o the und during this period under examination. The sum derives rom the ollowing: N = F F (1 + R )) (2.5) t t ( t 1 t where N t represents the net und lows at period t 11

12 F t is the total unds under managed at period t F t 1 is the total unds under managed in the previous period R t is the returns achieved rom t-1 to t We include the und low measurement term scaled by the weekly und size ( S ) both in the unconditional and conditional models including the market timing term to control or the inluence o und low on this term. Regressions are modiied accordingly as ollows: R = α + β R + γ R + λ ( S ) R + ε (2.6) m 2 m 2 m R 2 = + β Rm + δ RmPt 1 ) + γ Rm + λ ( S ) α ( R + ε (2.7) 2 m 3. Data The empirical analysis uses weekly data rom 27 bond mutual unds that operated in the Greek market or the seven-year period In addition, data or another 15 unds that did not operate throughout the aorementioned period were used to test or the known survivorship bias eect in our data. The large data set spans a period during which interest rates experienced both an upward trend, during the period, and a downward trend ater In this way we believe that the data set is representative o the changes that occurred in the interest rates market during the last decade. Initially we tested or the our-year period and empirical indings showed that results were only slightly worse regarding und managers ability to beat 12

13 passive benchmarks. The data were obtained rom the Greek Institutional Investors Association. The market value o assets invested on bond mutual unds on December 2003 was 6 billion euros representing 5% o Greek GDP and 30% o the total o unds invested in mutual unds in Greece. According to und managers, the most widely used benchmark index to measure bond unds returns is the BONDEX comprising o a portolio o government and corporate bonds. Fund managers enorce active management strategies to outperorm the BONDEX, attempting to improve their perormance concerning security selection and market timing. For the inclusion o the public inormation variables, used in the conditional model, we used the ASE General Index returns, ATHIBOR and EURIBOR interest rates and 10-year Greek Government Bond yields spanning the period rom 01/01/1997 to 26/12/2003. The aorementioned data provided by the Bloomberg Proessional Database. The 3-month EURIBOR rate (the interbank borrowing rate in the Euro market) was used as a proxy or the interest rate or the period 5/1/01-26/12/03, since this is likely to approximate the rate aced by wholesale traders in the speciic market. In order to orm the most accurate representation o the Greek money market, we used the 3- month ATHIBOR or the period beore the adaptation o Euro (1/1/97-29/12/00). The bond yield sampling procedure, employed in this study, selects the nearest to 10 year maturity Greek Government bond and thereater the yield to maturity or the speciic bond. As ar as the ASE General Stock index is concerned, we used the logarithmic returns: 13

14 pt r = ln t (3.1) pt 1 where r t is the weekly return p t is the end o the week price p t 1 is the lagged end o the week price because it has been shown that they ollow a smoother approximation o the normal distribution (Dacorogna et al, 2001). All data were selected on a Friday closing prices basis. In the cases where those prices were absent, we substituted them with the previous day s closing prices. However, some errors are inevitable because o the unavailability o simultaneous quotes or stocks, bonds and interest rates markets, since the closing time o three markets does not coincide. Moreover, the phenomenon o the asynchronous trading o the stocks included in the ASE General Index is relected on it. Hence, an entire sequence o 209 weekly values was constructed in this manner. 4. Empirical Results The summary o results or the sample o bond unds in Greece is presented in Table 1. The 7-year examination period starts rom January 1997 to December 2003 and results are reported both beore ees and costs are charged by the unds and ater these ees are incorporated in the prices. The 27 unds included in the sample represent the total o bond unds operating in the Greek market throughout the examination period. 14

15 Thereore results might suer rom survivorship bias since unds that were merged or terminated during the 7-year period under examination are not included. Normally, this renders the results more avourable than would be the case had terminated unds been included. This issue is revisited throughout the paper. Both the unconditional and conditional models are analysed with and without the und low inluence. The Table includes unds displaying positive and signiicant, positive but insigniicant as well as negative signiicant and insigniicant coeicients. Signiicance levels are reported at the 95% conidence interval. Results appear to be signiicantly dierent when results are reported beore ees and when ees are incorporated in the prices. The alpha coeicient appears to be positive and signiicant or all but two o the unds when ees are not included but this result is dramatically reversed when ees are included with 17 unds exhibiting signiicantly negative returns. This is expected since ees appear to reduce signiicantly any returns achieved rom active trading. Thereore, when ees are accounted or the majority o unds do not exhibit superior to the benchmark index risk-adjusted perormance. The same conclusion is reached both using the unconditional and the conditional perormance measure. The average ees charged by und managers or individual investors amount to approximately 2 percent including the ees charged when entering the und and those imposed when liquidating your share in the und. It should be noted however that 30 percent o the unds in the sample still exhibit positive and signiicant returns even ater ees are accounted or indicating that during the examination period some unds did indeed beat passive benchmarks. INSERT TABLE 1 15

16 As ar as the market timing and security selection elements are concerned results veriy that when ees are included und managers on average underperorm as a result o security selection. The market timing element however shows that both beore and ater ees 30 percent o the unds exhibit positive and signiicant market timing, which is probably the driving orce behind the 30 percent o the unds having above normal risk-adjusted returns. When und low is considered, perormance estimates do not dier notably. Around hal o the unds appear to have negative λ coeicients indicating the negative impact o und lows on perormance. On the other hand the other hal o the unds exhibit positive coeicients both or the unconditional and conditional models, irrespective o whether costs are included or not. However, the act that over 70 percent o the λ coeicient results are not signiicant, either positive or negative, indicates that und low activity has a relatively limited role in und perormance or the majority o the sample. However, there still exists a small percentage o unds where perormance estimates improve when low is included. In addition, market timing results are almost identical irrespective o the und low coeicient. The overall results signiy that the majority o active Greek bond unds do not outperorm passive benchmarks consistent with other indings or the American bond market (Blake et al., 1993). However a signiicant part o the unds included in the sample exhibit positive perormance even when ees are considered both using the unconditional and conditional model. 16

17 The results rom Table 2 indicate that on average bond unds have slightly positive and signiicant alphas or the unconditional model. Alphas are almost identical when the conditional model is used but results are not signiicantly dierent rom zero on average. Ater ees are considered, unds exhibit 1,5 percent underperormance and results are signiicant both or the unconditional and conditional model. An interesting point is that unds alphas appear to improve when a conditional model is used. However, this shit o alphas to the right is not large enough to alter the main conclusion derived rom this Table that bond unds, on average, ail to surpass passive benchmarks. Furthermore, this shit is reported in other studies in the bond evaluation literature (Ferson and Schadt, 1996; Beckers et al., 1999) but in our study this shit appears to be even smaller. Bond unds returns appear to be particularly low especially during the irst quartile o the data set which coincides with the period when the Greek stock market experienced extraordinary growth, while during the same period bond returns were comparatively low, which might explain this notable dierence. INSERT TABLE 2 Unreported diagnostic test results or the regression estimates o Table 2 gave statistically signiicant results regarding serial correlation above the ten percent signiicance level or the majority o unds. Results or unctional orm and heteroskedasticity are signiicant in approximately 80% o the unds analysed, underlying the robustness o our indings. However, results are not important when normality is considered indicating that the residuals are not normally distributed, a 17

18 inding however very common in the evaluation literature. Only 10 percent o the examined unds exhibit signiicant normality results when the unconditional model is used. Results are almost identical when the conditional model is used. Finally, in analysing the stationarity o all the time series comprising our data set we ound that almost all were stationary. In Table 3 the cross-sectional averages o the coeicients used as conditional variables are presented coupled with the number o unds exhibiting signiicant results in its case. The conditional variables used are consistent with Sawicki and Ong (2000), with the exception o the dividend yield that is omitted and the ASEGI returns being used instead as a measure o economic conditions. Regression results show that the only signiicant variable explaining bond und returns is the ASEGI returns as a measure o the economic conditions prevailing in the Greek economy during the examination period. However, it should be stated that t-stat results prove that even the economic conditions variable is only marginally signiicant. The same result applies or the sample including ees, as the economic conditions variable is important at the ive percent signiicance level, with t-ratio results being even slightly higher, still though marginally signiicant. Thereore, the economic conditions variable appears to be a signiicant explanatory variable or Greek bond und returns. The economic conditions variable is an important positive determinant o bond und returns or over hal o the sample. This result comes in accordance with the indings o Sawicki and Ong (2000), who document that 48 o equity and balanced unds appear to have positive inluence rom the Australian Stock Exchange dividend yield. 18

19 Coeicients or the term structure o interest rates are positive or both samples, with the results provided or the sample including ees being more robust but still not signiicant. The Treasury note yield coeicient, however, is marginally negative or the sample without ees and positive or the sample including ees. On the contrary, a negative January coeicient is observed or both samples, results proving again not signiicant. The act that empirical tests do not support the existence o a January eect in the results could be due to the act that Greek bond mutual unds invest primarily on government bonds rather corporate bonds, where this anomaly is welldocumented. INSERT TABLE 3 Results presented so ar cover solely unds that operated in the Greek market throughout the sample period. However, in the perormance evaluation literature it is stated that the inclusion o non-surviving unds in the sample examining the return characteristics o unds, results in average alphas being reduced. This means that on average the exclusion o non-surviving unds rom the sample data causes und returns to be overstated. In our data sample, 15 additional unds did not operate throughout the seven-year examination period. Interestingly, when these unds are included in our model results are inluenced with average alphas being slightly higher or the total sample. Our indings are signiicant at the 1 percent signiicance level. We reach the same conclusion when the conditional model results are used. In act, or the conditional model average alphas are even higher. This result contrasting the indings o Elton et al. (1996) could be attributed to the act that the majority o these unds did not cease to exist due to poor perormance and inability to attract new 19

20 investors. The mere reason was that during the period mergers occurred in the Greek banking sector, where Mutual Fund Management companies, being the subsidiaries o big banking groups, were subsequently merged or economies o scale reasons. Thereore, it is reasonable to expect that the inclusion o these unds did not inluence results negatively since in most cases they were perorming well compared to the rest o the market. INSERT TABLE 4 The results are analytically presented in Table 4 where we notice that non-surviving unds are better perormers both at the unconditional and conditional level. On the total sample level however results are not signiicantly altered and the depicted proit opportunities are marginal in any case. In Table 5 we see how perormance is attributed between the market timing and security selection elements o the behaviour o und managers. When considering the sample without ees on average bond unds deliver positive returns that can be attributed to security selection. We also note a positive market timing element depicted rom the signiicantly positive γ coeicient in our sample, even though results are marginally statistically signiicant. Contrary to previous indings we ind thereore that the security selection process does not lessen managers ability to choose the right timing or their investments and achieve risk-adjusted excess returns, but as we will later see this result is not supported by all the evidence. For the conditional model, the α coeicient is positive and the γ coeicient negative but results are not signiicant or both the unconditional and the conditional model 20

21 indicating that risk-adjusted excess returns are not achievable through either improved security selection estimates or correct market timing ability on the part o und managers. On the whole, perormance estimates or the conditional model are better as indicated by the Pearson correlation coeicient. This result highlights a known inding in the perormance evaluation literature (see Ferson and Schadt, 1996) that documents a negative covariance between und betas and market returns, which is particularly puzzling since it underlines a tendency or und managers to reduce (increase) their portolios average betas when market returns are expected to grow (diminish). INSERT TABLE 5 When ees are considered both the unconditional and the conditional model results show that returns are signiicantly negative, indicating negative security selection. Market timing estimates, on the other hand, are positive but insigniicant. The Pearson correlation coeicient is ound negative and signiicant or the sample including ees as well. This negative relationship between market timing and security selection, ound in all regression analysis results in Table 5 is also documented in other papers in the perormance evaluation literature (Henriksson (1984), Coggin et al. (1993)) and underlines the act that the ability to choose securities correctly does not ensure a positive market timing ability and vice versa. In an attempt to postulate the reasons that cause und betas and market returns to be negatively correlated we examine the impact o und lows on perormance as indicated in Table 6. Findings show that or the unconditional model the und low 21

22 eect is negative and signiicant, whereas the coeicient measuring the market timing element is positive and also signiicant. However, the conditional model including low gives us negative sign or the und low coeicient and positive market timing coeicient but results are not signiicant thus not enabling us to draw any deinite conclusions. INSERT TABLE 6 These results are in line with Edelen (1999) that underlines that the und low coeicient should be negative and the corresponding market timing estimate positive i the liquidity component eect is considered o minimal importance in the market timing process. Thereore, we can iner that a negative relationship between und betas and market returns stems rom the eagerness o und managers to invest inlows irrespective o the correct market timing estimates that would have been conducted had there not been these new money lows. Edelen (1999) indeed documents that decreased market timing estimates are observed when mutual unds have new money lows and vice versa. 5. Conclusion This paper has examined the ability o active bond mutual unds in Greece to achieve proits above the benchmark index returns. The evidence showed that on average bond unds do not oer risk-adjusted proits exceeding the returns o the benchmark index. This was proved using both an unconditional and a conditional model to test or the return characteristics o bond unds. Even though conditional models provide 22

23 marginally improved results the general conclusion that unds do not beat passive benchmarks remains valid. Furthermore, when ees are included in the prices the marginal proits derived beore their inclusion vanish completely and unds appear to underperorm on average. The obtained results ater ees are included are almost identical, irrespective o the model used. Our results proved not to suer rom survivorship bias since the inclusion o the unds that did not operate throughout the sample period did not aect average alphas negatively probably because these unds seized to exist or reasons beyond bad perormance and small clientele, such as mergers. Results on the impact o security selection and market timing showed that the security selection skills do not necessarily mean that this und should lack market timing ability. Moreover we attempt to derive the reasons that cause und betas and market returns to be negatively correlated examining the impact o und lows on perormance, the results showing that this is attributed to the eagerness o und managers to invest new lows in periods o high inlows. Reerences Artikis, P., 2002, Evaluation o Equity Mutual Funds Operating in the Greek Financial Market, Managerial Finance, vol. 28, no. 2, Artikis, P., 2003, Measuring Risk in the Greek Bond Market An Alternative Approach, Managerial Finance, vol. 29, no.9, Becker, C., W. Ferson, D. Myers and M. Schill, 1999, Conditional Market Timing with Benchmark Investors, Journal o Financial Economics, vol. 52, no. 1,

24 Blake, C., E. Elton and M. Gruber, 1993, The Perormance o Bond Mutual Funds, Journal o Business, vol. 66, no. 3, Blake, D. and A. Timmerman, 1998, Mutual Fund Perormance: Evidence rom the UK, European Finance Review, vol. 2, Blume, M. and D. Keim, 1987, Lower Grade Bonds: Their Risks and Returns, Financial Analysts Journal, vol. 43, no. 4, Blume, M., D. Keim and S. Patel, 1991, Returns and Volatility o Low-Grade Bonds , Journal o Finance, vol. 46, no. 1, Chang, E. and R. Huang, 1990, Time-Varying Return and Risk in the Corporate Bond Market, Journal o Financial and Quantitative Analysis, vol. 25, no. 3, Chen, N., R. Roll and S. Ross, 1986, Economic Forces and the Stock Market, Journal o Business, vol. 59, no. 3, Coggin, T., F. Fabozzi and S. Rahman, 1993, The Investment Perormance o U.S. Equity Pension Fund Managers: An Empirical Investigation, Journal o Finance, vol. 48, no. 3, Cornell, B. and K. Green, 1991, The Investment Perormance o Low-Grade Bond Funds, Journal o Finance, vol. 46, no. 1, Cumby, R. and J. Glen, 1990, Evaluating the Perormance o International Mutual Funds, Journal o Finance, vol. 45, no. 2, Dacorogna M., R. Gencay, U. Muller, R. Olsen and O. Pictet, 2001, An Introduction to High Frequency Finance (Academic Press, San Diego). Detzler, M., 1999, The Perormance o Global Bond Mutual Funds, Journal o Banking and Finance, vol. 23,

25 Edelen, R. and J. Warner, 2001, Aggregate Price Eects o Institutional Trading: A Study o Mutual Fund Flow and Market Returns, Journal o Financial Economics, vol. 59, no. 2, Edelen, R., 1999, Investor Flows and the Assessed Perormance o Open-End Mutual Funds, Journal o Financial Economics, vol. 53, no. 3, Elton, E., M. Gruber, S. Das and M. Hlavka, 1993, Eiciency with Costly Inormation: A Reinterpretation o Evidence rom Managed Portolios, Review o Financial Studies, vol. 6, no. 1, Elton, E., M. Gruber and C. Blake, 1995, Fundamental Economic Variables, Expected Returns, and Bond Fund Perormance, Journal o Finance, vol. 50, no. 4, Elton, E., M. Gruber and C. Blake, 1996, Survivorship Bias and Mutual Fund Perormance, Review o Financial Studies, vol. 9, no. 4, Fama, E. and K. French, 1993, Common Risk Factors in the Returns on Stocks and Bonds, Journal o Financial Economics, vol. 33, Ferson, W. and R. Schadt, 1996, Measuring Fund Strategy and Perormance in Changing Economic Conditions, Journal o Finance, vol. 51, no. 2, Fountas, S. and K. Segredakis, 2002, Emerging Stock Markets Return Seasonalities: The January Eect and the Tax-loss Selling Hypothesis, Applied Financial Economics, vol. 12, Gruber, M., 1996, Another Puzzle: The Growth in Actively Managed Mutual Funds, Journal o Finance, vol. 55, no. 3, Henriksson, R., 1984, Market Timing and Investment Perormance: An Empirical Investigation, Journal o Business, vol. 57, no. 1, Jegadeesh, N., and S. Titman, 2001, Proitability o Momentum StrategiesQ\: An Evaluation o Alternative Explanations, vol. 56, no.2,

26 Jensen, M., 1968, The Perormance o Mutual Funds in the Period , Journal o Finance, vol. 23, Kihn, J., 1996, The Financial Perormance o Low-Grade Municipal Bond Funds, Financial Management, Summer, Malkiel, B.G., 1995, Returns rom Investing in Equity Mutual Funds 1971 to 1991, Journal o Finance, vol. 50, no. 2, Maxwell, W., 1998, The January Eect in the Corporate Bond Market: A Systematic Examination, Financial Management, vol. 27, no. 2, Milonas, N., 1999, Greek Mutual Funds, Theory and Practice (Greek Banks Association edition). Sawicki, J. and F. Ong, 2000, Evaluating Managed Fund Perormance Using Conditional Measures: Australian evidence, Paciic-Basin Finance Journal, vol. 8, no. 3 4, Treynor, J. and K. Mazuy, 1966, Can Mutual Funds Outguess the Market? Harvard Business Review, vol. 44, Warther, V., 1995, Aggregate Mutual Fund Flows and Security Returns, Journal o Financial Economics, vol. 39, no. 2 3,

27 Table 1. Evaluation o Greek bond unds in the 7-Year Period to December 2003 This Table shows the summary results o the sample o bond unds operating in the Greek market throughout the 7-year examination period that exhibit perormance estimates signiicant at the 5 percent signiicance level. Thereore in each column we observe which unds out o the 27 included in the sample have negative but insigniicant, positive but insigniicant, negative and signiicant and positive and signiicant results. In the irst section o the Table, results are reported beore ees whereas in the second section o the Table ees are included in the estimated regressions. For equations 2.2 and 2.4 the alpha (α) and gamma (γ) coeicients reported measure the returns o the bond unds and the market timing ability respectively or the unconditional and the conditional model. The results including the und low inluence on perormance estimates, indicated by lambda (λ), are reported or equations 2.6 and 2.7. Alpha (α) captures und managers security selection ability, gamma (γ) denotes the market timing ability o managers, while lambda (λ) reers to the und low variable eect. Unconditional ignoring low (Eq. 2.2) Unconditional (including low) (Eq. 2.6) Conditional (ignoring low) (Eq. 2.4) Conditional (including low) (Eq. 2.7) Without ees α γ α γ λ α γ α γ λ Negative & Insigniicant Positive & Insigniicant Negative & Signiicant Positive & Signiicant Funds in sample With ees Negative & Insigniicant Positive & Insigniicant Negative & Signiicant Positive & Signiicant Funds in sample ** Signiicant at

28 Table 2. Overall risk-adjusted perormance o Greek bond unds This Table provides the average returns or the sample data, alphas being reported in percentage terms. Returns are shown both using the unconditional and the conditional model beore and ater ees. The examination period is spanned into three quartiles to estimate returns during dierent periods o the data set. The coeicient beta (β) measures the systematic risk o the unds while R 2 results report the adjusted R 2 results. Model Mean α t-stat SD α Min α Q1 α Q2 α Q3 α Max α Mean β Mean R 2 Beore ees Unconditional *** Conditional Ater ees Unconditional *** Conditional *** *** Signiicant at

29 Table 3. Cross-sectional averages o the conditional variable coeicients In this Table the public inormation variables coeicients are reported and more particularly the results. Economic represents the ASE General Index returns, Term is the term structure o interest rates, Treasury Note is the Treasury note yield while January represents the dummy variable coeicient. Beore ees Ater ees Coeicient t-stat Coeicient t-stat Economic ** Economic ** Term Term Treasury Note Treasury Note January January ** Signiicant at

30 Table 4. Analysis o the perormance o surviving and non-surviving bond unds This Table presents the results both or unds that survived throughout the examination and or those that did not survive. The reported alphas are given in percentage terms coupled with t- statistic and standard deviation results. Category No. unds Mean α t-stat SD α Unconditional model Non-surviving *** Surviving *** Dierence Total *** Conditional model Non-surviving *** Surviving *** Dierence Total *** *** Signiicant at

31 Table 5. Security Selection and Market Timing perormance o Bond unds The ollowing Table reports the descriptive statistics results without the impact o lows (equations 2.2 and 2.4) and with the impact o lows (equations 6 and 7). The alpha (α) coeicient results represents the returns derived through the use o security selection only while the gamma (γ) coeicient measures returns deriving rom market timing. The Pearson correlation coeicient is shown by ρ. Mean t-stat SD Min Q1 Q2 Q3 Max Unconditional model beore ees (ignoring low) α *** γ * ρ(α,γ) *** Conditional model beore ees (ignoring low) α γ ρ(α,γ) *** Unconditional model ater ees (ignoring low) α *** γ ρ(α,γ) *** Conditional model ater ees (ignoring low) α * γ ρ(α,γ) *** *** Signiicant at 0.01 * Signiicant at

32 Table 6. Security selection, market timing and und low This Table provides the results or the alpha (α), gamma (γ) and lambda (λ) coeicients or equations 2.6 and 2.7. Alpha denotes returns in percentage terms, gamma is the market timing element and lambda measures the und low eect Coeicient Mean t-stat Unconditional beore ees α *** γ * λ *** Conditional beore ees α *** γ λ *** Signiicant at 0.01 * Signiicant at