THE ADDED VALUE OF SRI SCREENING IN FIXED INCOME MARKETS?

(1)

THE ADDED VALUE OF SRI SCREENING IN

FIXED INCOME MARKETS?

An attribution approach

Bart Nijenkamp*

February 2012

Abstract

This thesis evaluates the performance of SRI fixed income portfolios compared to a conventional benchmark. I make a distinction between an equal weights portfolio and a progressive weights portfolio where bonds of very social responsible companies are assigned more weight. In relative terms, I find evidence of both SRI portfolios significantly underperforming the benchmark. However, this makes sense since SRI portfolios are less risky generating ceteris paribus less return. In addition, I find that absolute performance of the SRI portfolios is higher compared to benchmark since SRI bonds profited from the flight-to-quality momentum which was present in the dataset. Yet, robustness checks for three subsamples indicate that relative as well as absolute performance depends significantly on market circumstances.

JEL-classification: E43, G12, M14

Key words: social responsibility, social responsible investing (SRI), screening, SRI bond portfolios, conventional benchmark, performance evaluation, performance attribution

Supervisor: Dr. A. Plantinga 2nd Supervisor: Dr. H. Gonenc

*B.A. Nijenkamp, Student number: 1689290, Msc. Business Administration Finance, Department of Finance, University of Groningen. Corresponding address: Stokvisweg 15, 8111 RS, Heeten, The Netherlands.

Corresponding e-mail address: bartnijenkamp @ gmail.com. Acknowledgements:

(2)

B.A. Nijenkamp 1 University of Groningen

1. Introduction

Today’s institutional investors, especially in Europe, are developing and implementing a large number of social responsible investing policies in order to mark their investments as social responsible investments (SRIs). According to the 2010 European SRI study (Eurosif, 2010) the European ‘social responsible investing’ (SRI) market grew from € 2.7 trillion in 2007 to € 5 trillion in 2009. Different scholars, for instance Renneboog, Ter Horst and Zhang (2008), conclude that the growth is caused by two main factors, namely regulatory issues and “ethical consumerism”. Regarding the regulatory issues, public firms and pension funds are forced to publish social, ethical and environmental performance concerning their activities. Ethical consumerism refers to the behavior of people to buy products with an additional financial or non-financial premium to satisfy their personal values and idealistic standards. In other words, social responsible investing comes with moral benefits (non-financial!) for which some private investors and institutions are prepared to pay an additional premium which implies that the amount of premium indicates the extent of a company’s social responsible behavior. For investors this is likely to be a financial premium. Yet, this does not always suggest that investors are acting unconcerned regarding the premium costs in case of more or less the same investment opportunities.

(4)

B.A. Nijenkamp 3 University of Groningen with strong corporate management, reputational benefits and a forward-looking business style. All of these characteristics could be sources of superior firm performance.

Due to the existence of these two contrary views, a whole new field of empirical research was developed during the mid-nineties examining the financial performance of social responsible equity mutual funds relative to their conventional counterparts. A significant part of the research papers written suggest no significant out- or underperformance of SRI funds relative to their conventional counterparts implying social responsible equities do not generate superior or inferior returns (e.g. Hamilton, Jo and Statman (1993), Luther and Matatko (1994), Schöder (2004), Kreander et al. (2005) and Bauer, Koedijk and Otten(2005)).

Yet, during the last decades many scholars did not examine the financial performance social responsible fixed income securities. From one perspective this sounds very legitimate since investing in stocks comes with voting rights. Activist shareholders focusing on Environmental, Social and Governance (ESG) issues can influence a company’s management to strengthen ESG policy whereas bond holders cannot do that. On the other hand, this is rather odd since fixed income markets are significantly larger than stock markets and institutional investors experience increasing pressure on their policy regarding responsible investment decisions. Moreover, the basic elements of the previously mentioned views might also affect returns of bond portfolios. D’Antonio, Johnsen and Hutton (1997), Goldreyer, Parvez and Diltz (1999) and Derwall and Koedijk (2009) try to fill the gap. They all use different methodology to test the relationship between SRI screening and financial performance of bond portfolios. Unfortunately, they all find different outcomes.

The purpose of this thesis is to contribute to an expansion of research regarding social responsible bonds since the fixed income market accounts for an immense part of volume traded in the financial markets. SRI screening is fairly absent in fixed income markets relative to the stock market and this thesis intends to provide new insights regarding screened fixed income securities. The main objective of this research is to determine to what extent social responsible bonds out- or underperform their conventional counterparts. However, unlike for instance Derwall & Koedijk (2009), this thesis does not examine returns of existing fixed income mutual funds to determine out- or underperformance of social responsible fixed income securities. Instead, I create passively managed bond portfolios using a ‘best-in-class’ investment universe of companies issuing investment grade bonds denominated in Euros.

(5)

B.A. Nijenkamp 4 University of Groningen investment universe at that specific moment. The weights distribution regarding the portfolios is based on time-to-maturity, sector and the degree of a company’s social responsibility. Concerning the latter, weights are assigned both equally as well as progressively meaning that bonds of very social responsible companies have a larger weight. I examine the dataset as a whole, but I also perform robustness checks for three subsamples.

(6)

2. Literature review

This section highlights the theories challenging the relationship between SRI and the financial outcome of this investment strategy. More specific, a fixed income investment strategy. Subsection 2.1 points out the two common views that exist regarding the relationship between social responsible behavior and evaluates how these might affect financial performance and risk of firms, and indirectly the performance of asset portfolios. The existence of these views was a trigger for scholars to examine these relationship more in depth. Up until now, they primarily focused on social responsible equity mutual funds. Research results regarding these particular funds will be presented in subsection 2.2. Compared to equity mutual funds, financial performance of social responsible fixed income mutual funds was almost ignored by scholars while the fixed income market is much larger than the stock market. Only a few scientific papers were written regarding SRI fixed income. The papers’ results are presented in subsection 2.3. Ultimately, the hypotheses are presented in subsection 2.4.

2.1 Social responsibility and firm performance: Two views

The European think-tank for developing sustainability in European financial markets, Eurosif (2010), defines SRI as: “… a generic term covering any type of investment process that combines investors’ financial objectives with their concerns about ESG issues.” An SRI screening process is usually part of the investment process Eurosif (2010) mentions. Such a screening process is not uniform since many definitions of ‘social responsible investments’ or ‘sustainable and responsible investments’ exist. Basically, all screening approaches applied by investors are based on ESG factors, but they differ in how broad or narrow they define these factors. Therefore, Eurosif (2010) defines two types of SRI segments; a broad SRI segment and a core SRI segment. The broad SRI segment accounts for the largest part of growth of the SRI investments which suggests that a majority of SRI fund managers only applies a simple SRI screening approach (e.g. gambling and weapon-producing firms etc. are excluded from investment universes). Core SRI is defined more narrow. In addition to the simple SRI screening approach, an investor applies a more norms- and values/ethical-based or thematic screening implying a best-in-class SRI approach.

(7)

B.A. Nijenkamp 6 University of Groningen relationship between constraints due to societal norms and financial performance. In his model, employers not hiring particular people due to societal norms suffered financial damage relative to employers which did not impose such constraints. Scholars (i.e. Hong and Kacperczyk (2009)) draw a parallel between imposing such societal constraints and SRI constraints. This line of reasoning is supported by views related to the modern portfolio theory. This theory, based on the article of Markowitz (1952), serves as a starting point for many investors how they should manage their portfolio of assets. As we all know this theory states that more diversification within a portfolio will decrease aggregate risk given a certain expected return allowing investors to create an optimal portfolio which has the highest expected risk/return ratio. Rudd (1981) argues that exclusion of assets due to SRI constraints will limit diversification opportunities creating a suboptimal portfolio leading to a lower expected risk versus reward optimization for the SRI portfolio relative to a non-constrained conventional portfolio. This view therefore suggests that SRI portfolios might underperform their conventional counterparts.

More recent, scholars developed an alternative view while not withdrawing the modern portfolio theory. Although hard to localize and isolate, Kurtz (1997) argues that the existence of ‘informational effects’ cause investing in social responsible companies to be beneficial. He signals there could be a strong relationship between a social responsible policy and skillful management, current and future financial prosperity, and innovative capacity affecting a firm’s financial performance significantly. The benefits that come with these informational effects might offset the benefits of diversification from a financial return perspective since a firm’s valuation largely depends on its financial performance. Moreover, Derwall & Koedijk (2009) acknowledge the existence of these informational effects by stating that policies regarding corporate social responsibility are associated with strong corporate management, reputational benefits and a forward-looking business style. All of these characteristics could be sources of superior firm performance.

Besides the fact that financial return could offset diversification effects, Boutin-Dufresne and Savaria (2004) argue that a firm’s risk profile could cause a portfolio’s risk to go down. They state that social responsible companies are much more capable of avoiding for instance strikes, lawsuits or product and company boycotts thereby reducing the total risk of the firm relative to less social responsible firms.

(8)

B.A. Nijenkamp 7 University of Groningen outcomes for different types of asset classes. The sections below will point out scholars’ findings regarding the financial performance of social responsible investments.

2.2 Performance of ‘social responsible’ equities

Two decades ago scholars started an intensive quest trying to identify financial performance differences between SRI equity portfolios and conventional indices and funds. The empirical research shows broad consensus on this topic. A vast majority states that SRI equity portfolios do not out- or underperform their conventional counterparts. Hamilton, Jo and Statman (1993) find that SRI funds in the US with a longer history do not perform differently relative to non-SRI funds. In addition, they find that SRI funds with a shorter history underperformed. In addition, Luther and Matatko (1994) compare UK ethical funds with an all-share index and a small-cap index. They find no significant evidence that the ethical funds out- or underperform. For SRI funds consisting of mainly equities in Germany, Switzerland and the US Schröder (2004) does not find evidence for different performance numbers relative to a benchmark portfolio. A European study by Kreander et al. (2005) shows also no significant difference between SRI and non-SRI funds. Finally, Bauer, Koedijk and Otten (2005) find some evidence of SRI funds’ underperformance relative to their conventional counterparts in US. However, UK SRI funds significantly outperform their counterparts, while German SRI funds show no different returns relative to non-SRI funds (see Renneboog et al (2008) for a more expanded historical overview of empirical research).

(9)

B.A. Nijenkamp 8 University of Groningen 2.3 Performance of ‘social responsible’ fixed income

In addition to the main views that exist with respect to social responsible investing, some scholars have provided additional insights how aspects of social responsible behavior may have a specific influence on bond returns. More specific, do these aspects significantly influence the credit risk of a firm? Bauer and Hann (2010) find that a firm’s environmental issues can be associated with higher cost of debt issuing. Furthermore, they find that proactive practices regarding the environment relates to lower credit spreads. One might hypothesize that a parallel can be drawn between environmental issues and other CSR (corporate social responsibility) issues implying that CSR in general leads to a lower cost of debt. In addition, Goss and Roberts (2011) find a strong relationship between a firm’s poor CSR performance and higher spreads charged by banks providing a loan to a firm. They do not find any reverse relation. Moreover, Menz (2010) finds no substantial evidence that credit risk premia of social responsible firms differ from premia of non-social responsible companies.

Contrasting equity investing, not many scholars provide insight regarding the financial performance of SRI screened bonds relative to their conventional counterparts. D’Antonio, Johnsen and Hutton (1997) were the first to avoid the main stream SRI equity research and focus solely on returns of social responsible fixed income. They create a social responsible bond index based on firms appearing in the Domini 400 Index and compare the generated returns with a non-screened, conventional index. This conventional index has approximately the same duration. D’Antonio, Johnsen and Hutton (1997) implement a performance attribution model correcting for changes in the US risk-free term structure. They find that the social responsible index outperforms the conventional index. They signal that this outperformance is probably caused due to the fact that both indices have different levels of credit risk. The tested social responsible index had an overweight in BBB bonds while it had an underweight in AAA bonds relative to the non-screened index. Although D’Antonio, Johnsen and Hutton (1997) state that this difference might explain outperformance of the SRI index, they do not test empirical implications. Nevertheless, they conclude that SRI bond investors do not incur a financial performance penalty when one invests in bonds in accordance with an SRI screening method.

(10)

B.A. Nijenkamp 9 University of Groningen fund performance when a fund’s allocation shifts to more high yield debt if allowed by the fund’s investment mandate. Their factors are evaluated in excess of a 1-month treasury risk-free rate. Derwall and Koedijk (2009) do not find any evidence of out- or underperformance of SRI funds compared to non-SRI funds.

2.4 Hypotheses

Looking at the literature, one can conclude that the number of theories examining the particular influence of SRI screening on bond returns are rather limited and inconclusive. In addition, not many scholars have provided empirical evidence testing the influence of SRI screening on bond returns. The goal of this research is to provide additional insight and evidence regarding the influence of SRI screening on bond returns. Therefore, I compare the performance of a portfolio of SRI bonds with the performance of a conventional benchmark;

Hypothesis 1: H0: SRI screened bond portfolios do not significantly out- or underperform a conventional benchmark.

H1: SRI screened bond portfolios do significantly out- or underperform a conventional benchmark.

In addition to the finding of Barnett and Salomon (2006) stating that there is a non-linear relationship between the number of SRI screens and financial performance of a fund, this thesis furthermore assesses whether an SRI portfolio performs better against a conventional benchmark, when one assigns more weight to bonds of firms that show better social responsible behavior.

Hypothesis 2: H0: SRI screened bond portfolios with an additional weight for firms which show better social responsible behavior do not significantly out- or underperform a conventional benchmark.

(11)

3. Methodology

Although the number of papers written is limited, the different scholars mentioned in subsection 2.3 use three different models to evaluate performance of fixed income portfolios: a capital asset pricing model using a single index (Goldreyer, Parvez and Diltz (1999)), a multi-factor capital asset pricing model (Derwall and Koedijk (2009)) and a bond performance attribution model (D’Antonio, Johnsen and Hutton (1997)). Yet, one could question whether scholars should use a capital asset pricing model using a 1-month treasury as risk-free rate. In practice, bonds are mostly valuated on a risk-free benchmark instead of one rate for all bonds since the price of a bond is affected by time to maturity. A bond’s value is determined by the rate on the risk-free benchmark that matches the bond’s duration (i.e. duration-matched risk-free rate). For bonds denominated in US Dollars investors typically use the term structure of bonds issued by the US government as risk-free benchmark, while the yield curve based on German government bonds (i.e. German Bunds) serves as a risk-free benchmark for bonds denominated in Euros. Campisi (2009) states that the yield of a bond is the duration-matched risk-free rate plus the credit spread. Since yield is the inverse of a bond’s price, changes in the duration-matched free rate together with changes in credit spread determine ceteris paribus price fluctuations of bonds. A risk-free rate based on a 1-month treasury does not necessarily capture rate changes for other maturities on the risk-free term structure. This may significantly bias the coefficients related to the market index. Furthermore, the systemic risk captured by the coefficient of the market index does not necessarily say something about the direction of the changes in credit risk. Changes in credit risk directly affect a bond’s return. All of these inefficiencies can be captured using an fixed income performance attribution model comparable to the model D’Antonio, Johnsen and Hutton (1997) use since an attribution model for fixed income can make a distinction between risk related to differences in maturities and risk related to changes in credit spread. Campisi (2009) proposes the following model:

P D DMT D S



  



  



(1)

(12)

SRI C SRI SRI

t t i it j jt

R R    DMT  S (2) where the out- or underperformance of the SRI portfolio (

R

tSRI) relative to the conventional benchmark (

C t

R

) is captured by α. The coefficients belonging to the change in duration-matched risk-free rate (



DMT

itSRI) and the change in credit spread (



S

SRIjt ) signal whether these characteristics of the SRI portfolio significantly differ from the conventional benchmark. Hypothetically, it is assumed that both coefficients are zero. A coefficient higher than zero for



DMT

itSRI signals that interest rate changes in the risk-free yield curve have less impact on return i.e. the duration of the SRI portfolio is lower. A lower than zero coefficient signals a higher duration. Regarding

SRI jt

S



, interpretation of the coefficient depends on the sign of the statistic’s mean (i.e. a positive or negative sign). A significant positive coefficient indicates smaller credit spread increase relative to the conventional benchmark in case of a positive mean for the statistic or a smaller credit spread decrease if the mean is negative. If the coefficient is negative, a positive mean suggests higher credit spread increase whereas a negative mean for the statistic signals more credit spread decrease. In model 2 α relates to out- or underperformance due to a difference regarding carry return. Carry return consists of three income factors: income generated from cash flows, the pull-to-par effect, and changing rates at the time when cash flows are discounted (Daul, Sharp and Sørensen (2010)).

In addition, the relationship between price change and duration is not perfectly linear and therefore, the contribution of convexity to a price change might be substantial (Daul, Sharp and Sørensen (2010)). Thus, in addition to the model of Campisi (2009), Daul, Sharp and Sørensen (2010) propose to add a measure of convexity to a fixed income attribution model for straight bonds:

 

2

1

P D DMT D S C Y

2



  



  







(3) where C is convexity and ∆Y is the change in yield. Integrating this formula with formula 2 results in:

SRI C SRI SRI SRI 2

t t i it j jt k kt

(13)

B.A. Nijenkamp 12 University of Groningen where



Y

ktSRIis change in yield for the SRI portfolio. This coefficient is assumed to be equal to zero.

Implementing this derived formula implies that both the SRI portfolio and the conventional benchmark have the same credit risk. However, this is not necessarily the case as we saw studying the article of D’Antonio, Johnsen and Hutton (1997). Unfortunately, these scholars do not test the implications of different credit risks. Since



DMT

itSRIadjusts for different returns on different maturities, a discrepancy between the SRI portfolios yield and the conventional benchmark’s yield signals dissimilar levels of credit risk. Adding a control variable to the latter equation tests whether credit risks are the same and how a different credit risk influences α;

SRI C SRI SRI SRI 2 SRI C

t t i it j jt k kt l lt lt

R



R

 

  

DMT



 

S





0.5( Y )







(Y



Y )

(5)

where

Y

_ltSRIis the yield of the SRI portfolio and

Y

_ltc the yield of the conventional benchmark. The expected outcome for coefficient



_lis positive since a higher credit risk indicates ceteris paribus a higher return. Furthermore, I expect a non-significant coefficient since credit risks are assumed to be equal. Obviously, a statistically significant coefficient suggests non-similar levels of credit risk. Therefore, a significant positive coefficient for



_lsuggests that credit risks clearly differ from each other and that bonds with a higher level of credit risk deliver an additional carry return. However, if the coefficient were to be significantly negative, investing in bonds with a higher credit risk results in lower carry returns relative to bonds with a lower credit risk, indicating a flight-to-quality. This last market circumstance is typical for a period in which the economy is approaching recession (Næs, Skjeltorp and Ødegaard (2011)).

Both the SRI portfolio as well as the benchmark contain a large number of bonds. Consequently, both bond duration distributions may be quite different from each other. To correct for too much duration mismatch between the SRI portfolio and the benchmark, the SRI portfolio will be subdivided into five different maturity tranches consistent with the tranches that are subject to the benchmark. This leads to the following regression model:

5 5 5 5

SRI C SRI SRI SRI 2 SRI C

t t i it j jt k kt l lt lt

i 1 j 1 k 1 l 1

R R   dmt  s  0.5( y ) (y y )

   

  















 (6)

(14)

(15)

4. Data

After having described the methodology to determine out- or underperformance of an SRI portfolio, this sec-tion points out how bond data is collected, structured and modified to serve as input for the models presented. I collect data over the period of 01/07/2005 until 12/31/2010 leading to 3101 weekly observations. The first subsection sheds light on how bonds of social responsible companies are retrieved. In addition, subsection 4.2 reveals which filter criteria apply to the bonds. The next subsection considers benchmark data whereas subsection 4.4 discusses data collection, portfolio construction and gives insight in the statistics of the dataset.

4.1 SRI data

As already mentioned, the definition of SRI is not uniform and screening approaches differ from each other. For this research I use SRI registers provided by FORUM ETHIBEL/Vigeo for the period of measurement. Applying these registers as screening tools for portfolios can be classified as a best-in-class approach. Besides the fact that companies with controversial sources of income are filtered out, the registers contain companies that perform average or above when it comes down to a company’s behavior regarding ESG issues compared to other sector members. Moreover, within the register a distinction is made between average, above-average and best-in-class performers. Best-in-class performers are assigned to class A whereas the average performers are identified with class C. The SRI portfolios I construct use every first register issue of the congruent year, meaning that I will use a total of six FORUM ETHIBEL/Vigeo registers. Although these registers originally apply to stocks, they are also suitable for filtering bonds. Using the stocks’ ISIN numbers, I retrieve companies’ bond tickers from a Bloomberg terminal using companies’ stock tickers.

4.2 Constructing the bond universe

Not all companies issue public debt which means that not all companies have bond tickers. These companies are obviously erased from the registers. The remaining bond tickers serve as input for the bond search tool in Bloomberg. Every year a new set of bond tickers is constructed consistent with the six SRI registers. The search tool is subject to a number of filters. First of all, the SRI portfolios are reweighted every year. Therefore, the bonds should be issued before the congruent year of measurement and should not expire before the end of that year. Second, the bonds should be denominated in Euros. Furthermore, structured notes and non-straight

(16)

B.A. Nijenkamp 15 University of Groningen bonds are erased from the bond issues and only bonds with a fixed or zero coupon are selected. In addition, only investment grade bonds are selected. This means that at the time of rebalancing the portfolio at least one of the three large American rating agencies (S&P, Moody’s and Fitch) should rate the bond at BBB or higher. Moreover, bonds that do not have price quotes in Bloomberg are filtered out, since the absence of a quote signals to much illiquidity. Eventually, all bonds are assigned to their relevant maturity tranche. Table 1 shows the number of bonds per year per tranche.

Table 1 - Sample outline SRI bonds

This table shows the number of bonds per year and per maturity tranche.

Year 1-3y 3-5y 5-7y 7-10y 10y+

2005 76 67 46 24 12 2006 75 78 41 30 19 2007 125 108 56 47 36 2008 115 75 42 43 28 2009 147 145 89 72 46 2010 193 206 105 91 56 4.3 Benchmark data

The Barclays Capital Euro Aggregate Corporate indices serve as leading benchmarks. The indices have different maturity scopes consistent with the tranches mentioned in table 1. Using Barclays POINT I collect weekly returns based on reinvested coupons, durations and yields for the five indices which are equally weighted in an aggregate portfolio. Since we do not know yet whether multicollinearity is a problem, I show both individual yields as well as an aggregate yield. The aggregate yield is calculated as follows:

5 C lt l lt c l 1 lt 5 C l lt l 1 D w y Y w y   



(7)

where

Y

ltc is the aggregate yield, Dlt represents duration, wl the weight and y is the yield of every individual Clt

(17)

B.A. Nijenkamp 16 University of Groningen 4.4 Data collection and portfolio construction

For each of the SRI bonds I collect sector information, weekly mid clean price, accrued interests, coupons, mid yield-to-maturity, mid duration and time-to-maturity, all of which are obtained from Bloomberg. The returns of the individual bonds are determined as follows:

t ti t 1 V r ln V        (8)

where rti is return and V is the clean price plus accrued interest plus coupon which is reinvested every week assuming that the coupon is issued at the end of the week. In addition, the change in risk-free yield is retrieved by matching the bond’s time-to-maturity with time values on the interpolated implied German Bund yield curve. For maturities longer than twenty-five years it is assumed that the risk-free yield of the 25-Year Bund holds. Furthermore, the credit spread is collected by subtracting the risk-free yield from the yield-to-maturity. The spread change is obtained subtracting the spread value at t-1 from the value at t.

Eventually, for each maturity tranche two portfolios are created. One portfolio with equal weights regarding social responsible behavior and a portfolio with progressive weights meaning that class C bonds are assigned with a weight of 1 whereas class B and class A get weights of respectively 2 and 3. Regarding the first type portfolio, the weights are determined in accordance with the sector weight distribution of the indices. This distribution is presented in appendix table 1 (A1). For some sectors, especially for the small sectors ‘Diversified’ and ‘Technology’, there are not always bonds available matching the specific sector. To solve this issue, the

Table 2 - Statistics benchmark index

This table shows the statistics regarding the benchmark indices over a period from 01/07/2005 to 12/31/2010.AVariable

C t

R represents the aggregate return of the benchmark, c lt

Y the aggregate yield and c lt

y the yield belonging to a single

benchmark index. Observations are expressed as a number whereas all other statistics are expressed as a percentage. A

Maturity Tranche Variable Mean Median St.Dev Min. Max Obs.

(18)

B.A. Nijenkamp 17 University of Groningen weights of the empty sectors are distributed over the other sectors keeping the relative balance equal. Weights of the issues in the equal weighted portfolio is determined as follows:

i fsi i n i fs i 1 bw 1 w n bw              



 (9)

where wi is the weight of the bond issue belonging to a specific sector,

bw

_fsiis the benchmark sector weight congruent with issue’s sector and only applies if issues are present in that sector. The number of issues per sector in the portfolio is represented by ni. For the progressive weights portfolio the following formula applies:

i fsi j j n 3 fs j j i 1 j 1 bw ₁ w c bw n c             _    



 (10)

where wj is the weight of the issue belonging to the specific sector and specific SRI class, cj the weight assigned to the corresponding class and nj is the number of issues per sector per class in the portfolio. Using the weights we can construct 10 different sub portfolios. The return of such a portfolio is determined as follows:

n SRI t i t i 1 r w r  



 (11)

where

r

tSRIis the return on the portfolio. The aggregate return of the total portfolio ( SRI t

R

) is similarly constructed using equal weights. The sub portfolio’s risk-free rate change (



dmt

itSRI), credit spread change (

SRI jt

s



), yield change (



y

SRIkt ) and yield ( SRI lt

y

) are constructed using formula 12:

n it i it SRI i 1 xt n i it i 1 d w x x w x   



(12)

(19)

Table 3 - Statistics SRI portfolios per maturity tranche

This table presents the statistics regarding the SRI portfolios over a period from 01/07/2005 to 12/31/2010. Panel A

shows the statistics regarding the equally weighted portfolios whereas Panel B presents the statistics of the progressively

weighted portfolios. VariableRtSRIrepresents the aggregate return of all maturity tranches and variable

SRI t

r represents

the return of a single SRI portfolio. Furthermore, this table shows the statistics regarding the portfolios' risk-free rate

change (dmtitSRI), credit spread change (

SRI jt s

 ), yield change (ySRIkt ) and yield (

SRI lt

y ). Observations are expressed as a

number whereas all other statistics are expressed as a percentage.

Panel A - Equally weighted SRI portfolios

(20)

Table 3 (continued)

Panel B - Progressively weighted SRI portfolios

(21)

B.A. Nijenkamp 20 University of Groningen However, as stated in the methodology section, it is likely that there is multicollinearity among the same independent variables across the maturity tranches. From A2 one can conclude this is the case for both SRI portfolios. This means that the different maturity tranches have to be merged in order to prevent standard errors of independent variables from becoming biased. I use a weighted average for merging returns and a duration weighted average (see formula 12) to generate merged independent variables. Table 4 shows the merged statistics.

Table 4 - Merged statistics SRI portfolios

This table shows the merged statistics regarding two types of SRI portfolios: an equal weightsj

jASRI

portfolio (panel A) and a progressive weights SRI portfolio (panel B). The measurement period

ranges from 01/07/2005 to 12/31/2010. Statistics are provided for the portfolios' return ( SRI

t

R ), risk-free rate change (DMTitSRI), credit spread change (

SRI jt S

 ), yield change (YktSRI) and yield

( SRI

lt

Y ). Observations are expressed as a number whereas all other statistics are expressed as a

percentage.

Panel A - Equally weighted SRI portfolio

Variable Mean Median St.Dev Min. Max Obs.

SRI t

R

0.0895 0.1063 0.4679 -1.2179 1.4199 310 SRI it

DMT



-0.0047 -0.0037 0.0961 -0.3004 0.2702 310 SRI jt

S



0.0033 0.0013 0.0523 -0.2404 0.2697 310 SRI kt

Y



-0.0014 -0.0052 0.0932 -0.2884 0.2843 310 SRI lt

Y

4.5559 4.4347 0.7513 3.2620 6.1855 310

Panel B - Progressively weighted SRI portfolio

Variable Mean Median St.Dev Min. Max Obs.

(22)

B.A. Nijenkamp 21 University of Groningen -5% 0% 5% 10% 15% 20% 25% 30% 35% SRI EW SRI PW C

5. Results

This section presents the results regarding the performance of SRI portfolios relative to a conventional benchmark. First, let us observe absolute returns of both SRI portfolios and the conventional benchmark. Figure 1 shows the cumulative returns for the entire measurement period, where SRI EW represents the equal weights portfolio, SRI PW the progressive weights portfolio and C the conventional benchmark.

From figure 1 one can see that both SRI portfolios are not behaving very differently from each other with respect to absolute return. Furthermore, figure 1 signals that both SRI portfolios have outperformed the benchmark in absolute terms. On average, the weekly outperformance of the equally weighted portfolio is 0.0233 percent whereas the portfolio with progressive weights outperformed the benchmark by 0.0241 percent. Subsection 5.1 presents my main findings where I adjust these alphas for differences in duration, credit spread change, credit risk and convexity. In addition, in subsection 5.2 I perform robustness checks for three different subsample periods.

Figure 1 - Cumulative returns SRI portfolios and benchmark (in %) (01/07/2005 - 12/31/2010)

(23)

B.A. Nijenkamp 22 University of Groningen 5.1 Main findings

I examine six different regression models consistent with the derived formulas in the methodology section to evaluate the SRI portfolios’ out- or underperformance relative to the conventional benchmark. The initial model as presented by Campisi (2009) is based on a linear regression model. I evaluate whether an ordinary least squares (OLS) estimation model is appropriate given the dataset. Since Brooks (2008) states that linear structural models are unable to explain some important characteristics of financial data because of the existence of leptokurtosis, volatility clustering and leverage effects, I also test whether there are generalized autoregressive conditionally heteroscedastic (GARCH) effects involved. Table A3 presents present data regarding residual normality, autocorrelation, heteroscedasticity and goodness of fit using OLS estimation. I find strong statistical evidence for the presence of GARCH effects. As a consequence, I adopt a GARCH estimation model with Bollerslev-Wooldridge (1992) robust standard errors and covariance since residuals are not conditionally normally distributed (see table A4). Furthermore, Lehman Brothers went bankrupt in the week of Friday 19 September 2008. The sudden panic in the bond markets during this event had a severe impact on the data resulting in a unique residual outlier for that specific week. Therefore, I introduce a dummy variable, LEHMANDUM, and append it to the regression models in order to explain the outlier in the week in which Lehman Brothers went bankrupt. LEHMANDUM equals one for this specific week and zero for all other weeks. Table 5 shows the results using a GARCH estimation approach.

Examining the estimates of all six regressions, I find that coefficients for all explanatory variables are robust. The results indicate significant lower levels of duration for both SRI portfolios relative to the conventional benchmark. For the equally weighted portfolio these lower levels range from 0.34 to 0.49 whereas the SRI portfolio with progressive weights has a lower duration ranging from 0.34 to 0.46. Moreover, my findings signal a smaller credit spread increase for both SRI portfolios relative to the conventional benchmark. This indicates that the SRI portfolios generate a higher return on credit spread change than the conventional benchmark. With respect to convexity, I find no statistical evidence for the SRI portfolios having a different level of convexity compared to the benchmark. Results concerning control variable

Y

ltSRI–

(24)

Table 5 – Regression results GARCH estimation analysis

The sample period isj

jA01/07/2005 - 12/31/2010. Variable α represents the out- or underperformance of the SRI

portfolio when corrected for a selected number of explanatory variables; the coefficient for SRI

it

DMT

 indicates the

level of duration deviation with respect to the benchmark, SSRIjt corrects for a difference in credit spread change, the

coefficient for SRI 2

kt

0.5 ( Y )  signals a difference regarding convexity, and SRI lt

Y – c lt

Y indicates how differences in credit

risk affect out- or underperformance. LEHMANDUM corrects for the outlier in the week of 19 September 2008.j

jAPanel

A displays estimation results concerning the equal weights portfolio and panelj

jAB shows the results regarding the

progressive weights portfolio.j

jAMarquardt's iterative algorithm is used. Standard errors and covariance are

Bollerslev-Wooldridge (1992) robust.j

jA*** 1% significance; ** 5% significance; and * 10% significance.

j j

A A

Variable (1) (2) (3) (4) (5) (6) α 0.0013 -0.0031 0.0016 -0.0034 -0.0167*** -0.0148** (0.2270) (-0.6512) (0.2952) (-0.6428) (-2.6030) (-2.3381) SRI it

DMT



0.3378*** 0.4503*** 0.3391*** 0.4493*** 0.4749*** 0.4894*** (4.3184) (6.7672) (4.1249) (6.4385) (7.4027) (7.2510) SRI jt

S



1.2315*** 1.2323*** 1.1770*** 1.1691*** (4.6235) (4.5796) (4.5166) (4.4323) SRI 2 kt

0.5 ( Y )





-0.1461 0.0981 -1.1116 (-0.0951) (0.0757) (-0.8448) SRI lt

Y

–

Y

_ltc -0.1189*** -0.1286*** (-3.5342) (-3.5266) LEHMANDUM 3.2206*** 2.8975*** 2.5289*** 2.2895*** 2.7182*** 2.2951*** (8.1861) (8.4529) (6.3150) (6.4868) (10.6334) (8.6412)

(25)

B.A. Nijenkamp 24 University of Groningen Since estimates are robust across the different regressions, I adopt regression 6 as leading model to evaluate performance because it takes into account all of the discussed explanatory variables. The portfolios’ alphas belonging to regression model 6 indicate solid statistical evidence of both SRI portfolios underperforming the benchmark. The equal weights portfolios 0.0148 percent on a weekly basis whereas the progressive weights portfolio is beaten by 0.0137 percent. However, it is interesting to put this underperformance into perspective and see to what extent the different variables contributed. Therefore, I evaluate the underperformance of the SRI portfolios in more detail using a performance attribution table (see table 6). In this table I multiply the regression’s coefficients with the variables’ means to determine the impact of several variables on performance.

Table 6 – Performance attribution table

This table shows the contribution of the variables to out- or underperformance regarding sample periodj

j

A01/07/2005 - 12/31/2010. Panel

A shows the figures that apply to the equally weighted portfolio and panel Bj

jAshows the figures that apply to the progressively weighted

portfolio. The contribution figures indicate to what extent adjustments regarding duration (DMTitSRI), changes in credit spread (

SRI jt S

 ), convexity (0.5 ( Y ) ktSRI 2), the discrepancy in credit risk (

SRI lt

Y – Y ) and an outlier (LEHMANDUM) affected return. Variable α indicates ltc

the out- or underperformance in relative terms whereas SRI C

t t

R R is the out- or underperformance in absolute terms.j

jAContribution is expressed as a percentage.

Panel A - Equally weighted SRI portfolio Panel B - Progressively weighted SRI portfolio

Variable Mean Coefficient Contribution

SRI it

DMT



-0.0047 0.4894 -0.0023



DMT

_itSRI -0.0047 0.4623 -0.0022 SRI jt

S



0.0033 1.1691 0.0039



S

_jtSRI 0.0032 0.9715 0.0031 SRI 2 kt

0.5 ( Y )





0.0043 -1.1116 -0.0048

0.5 ( Y )





_ktSRI 2 0.0044 -1.4010 -0.0062 SRI lt

Y

–

Y

_ltc -0.3503 -0.1286 0.0450

Y

_ltSRI–

Y

_ltc -0.3263 -0.1489 0.0486 LEHMANDUM 0.0032 2.2951 0.0074 LEHMANDUM 0.0032 2.3932 0.0077 SUBTOTAL 0.0492 SUBTOTAL 0.0510 α -0.0148 α -0.0137 RESIDUAL -0.0111 RESIDUAL -0.0132 SRI C t t R R 0.0233

R_tSRIRC_t 0.0241

(26)

B.A. Nijenkamp 25 University of Groningen highlights the main difference between the absolute outperformance and the relative underperformance of the SRI portfolios. We can interpret the alphas as the expected additional carry return generated by the conventional benchmark if market circumstances would have been ‘normal’, i.e. if bonds with a higher credit risk would have generated ceteris paribus a higher return.

5.2 Robustness checks

In addition to the results regarding the complete dataset, I perform robustness checks for different subsamples. I argue that last decade’s financial crisis is a valid reason to split up the initial dataset. First, I assume that Lehman Brothers’ bankruptcy could mark a structural break in the dataset since this bankruptcy event immediately resulted in a period of unprecedented market uncertainty, which resulted in very illiquid financial markets and in very wide bid-ask spreads. In addition, credit spreads on corporate bonds immediately skyrocketed after Lehman Brothers went bankrupt. I check whether the week in which Lehman Brothers went bankrupt can be considered as a structural break using Chow’s (1960) breakpoint test. To perform this test, I use simple OLS estimation adopting α,



DMT

itSRI and

SRI jt

S



as variables. Ultimately, I indeed find evidence of a structural break for the variables in that specific week. Furthermore, I argue that markets gradually recovered from panic during the following year resulting in another structural breakpoint. Consequently, I perform a Quandt (1960) - Andrews (1993) unknown breakpoint test, using the same estimation technique and variables as for Chow’s breakpoint test. Results reveal evidence regarding an additional breakpoint in the week of the March 20th, 2009. Outcomes concerning the breakpoint tests can be found in table A5. Therefore, three subsamples are created in total. I choose to leave out the data gathered for the week of September 19th, 2008 to avoid using the dummy variable in subsamples 1 and 2. Statistics concerning the created subsamples can be found in table A6 and table A7 for the SRI portfolios and table A8 for the benchmark.

(27)

B.A. Nijenkamp 26 University of Groningen A11). Therefore, I estimate subsample 2 using OLS. OLS is also use for subsample 3. However, I find traces of heteroscedasticity using White’s (1980) specification (see table A12). Therefore, I use White’s (1980) specification to make the standard errors heteroscedasticity robust. Table 7 shows the estimation results regarding the three subsamples.

Table 7 – Regression results subsamples

Variable α represents the out-or underperformance of the SRIj

j

Aportfolios when corrected for a selected number of

explanatory variables; the coefficient forDMTitSRIindicates the level of duration deviation with respect to the

benchmark, SSRIjt corrects for a difference in credit spread change, the coefficient for

SRI 2 kt

0.5 ( Y )  signals a difference regarding convexity, and YltSRI–

c lt

Y indicates how differences in credit risk affect out- or underperformance. Panel A

displays estimation results concerning the equal weights portfolio and panelj

jAB shows the results regarding the

progressive weights portfolio.j

jARegarding subsample 1 (01/07/2005 – 09/12/2008), I use GARCH estimation.

Marquardt's iterative algorithm is used and standard errors and covariance are Bollerslev-Wooldridge (1992)j

jArobust.

With respect to subsample 2 (09/26/2008 – 03/13/2009), I use OLS estimation.j

jAFor subsample 3 (03/20/2009 –

12/31/2010)j

jAI use OLS estimation with White’s (1980) heteroscedasticity robust standard errors. *** 1% significance;

** 5% significance; and * 10% significance.j

jA A Panel A - Equally weighted SRI portfolio

Variable Subsample 1 Subsample 2 Subsample 3

α -0.0152** 0.4735 0.0662* (-1.9821) (0.8449) (1.6645) SRI it

DMT



0.5290*** 0.3260 -0.2907 (6.8982) (0.7122) (-1.2589) SRI jt

S



0.2420 1.2971* 0.2894 (0.5159) (1.9185) (0.6324) SRI 2 kt

0.5 ( Y )





-2.6440 6.7840 -0.3302 (-1.3808) (1.2675) (-0.0650) SRI lt

Y

–

Y

_ltc -0.2226*** 0.2008 0.3550*** (-3.8235) (0.4617) (3.2078)

Variable Subsample 1 Subsample 2 Subsample 3

(28)

B.A. Nijenkamp 27 University of Groningen Results indicate that the explanatory variables are not robust over time. Whereas the main results indicate a significant lower duration, only results for subsample 1 indicate a lower significant duration. For subsample 3 regression estimates for the progressively weighted SRI portfolio even show weak evidence of a higher duration. Regarding credit spread change, only results for subsample 2 reveal some weak evidence of the equally weighted portfolio experiencing a significant smaller credit spread increase compared to the benchmark. Furthermore, subsample 1 and 3 indicate that credit risks are not the same. However, for subsample 1 I find strong evidence to assume a flight-to-quality whereas for subsample 3 I find strong evidence to assume that riskier bonds deliver higher returns.

With respect to subsample 1, I find solid evidence of the equally weighed portfolio underperforming the benchmark by 0.0152 percent and weak evidence of the progressively weighted portfolio underperforming the benchmark by 0.0127 percent per week. Looking at table 8, consistent with the main sample, we see that this relative underperformance comes with an absolute outperformance mainly due to a flight-to-quality which was beneficial for the SRI portfolios. The contribution of the flight-to-quality entirely offsets the additional performance which the riskier benchmark would have gained if market circumstances would have been ‘normal’. The SRI portfolios’ absolute outperformance was respectively 0.0199 and 0.193 percent. Regarding subsample 2, I find that the alphas signal an outperformance for both SRI portfolios. However, these are not statistically significant. With respect to absolute performance, the equal weights portfolio beats the benchmark by 0.3176 percent whereas the portfolio with progressive weights outperformed the conventional benchmark by 0.3144 percent. Moreover, looking at the results for subsample 3, the estimates for the alphas indicate weak evidence to assume that both SRI portfolios outperform the benchmark by respectively 0.0662 and 0.0655 percent. In absolute terms, however, the SRI portfolios underperformed the benchmark by 0.0765 and 0.0724 percent. The significant positive coefficient for the variable

Y

ltSRI–

c lt

(29)

Table 8 – Performance attribution table subsamples

This table shows the contribution of the variables to out- or underperformance regarding three subsamples:j

jAsubsample 1 (01/07/2005 – 09/12/2008),

subsample 2 (09/26/2008 – 03/13/2009)j

jAand subsample 3 (03/20/2009 – 12/31/2010). Panel A shows the figures that apply to the equally weighted

portfolio and panel Bj

jAshows the figures that apply to the progressively weighted portfolio. The contribution figures indicate to what extent adjustments

regarding duration (DMTitSRI), changes in credit spread (

SRI jt S

 ), convexity (0.5 ( Y ) ktSRI 2) and the discrepancy in credit risk ( SRI lt

Y – Y ) affected return. ltc

Variable α indicates the out- or underperformance in relative terms whereas SRI C

t t

R R is the out- or underperformance in absolute terms. Contribution is

expressed as a percentage.j

jA

Subsample 1 Subsample 2 Subsample 3

Mean Coefficient Contribution

SRI it

DMT



0.0029 0.5290 0.0015 -0.0585 0.3260 -0.0191 -0.0071 -0.2907 0.0021 SRI jt

S



0.0055 0.2420 0.0013 0.0361 1.2971 0.0468 -0.0120 0.2894 -0.0035 SRI 2 kt

0.5 ( Y )





0.0031 -2.6440 -0.0082 0.0131 6.7840 0.0890 0.0043 -0.3302 -0.0014 SRI lt

Y

–

Y

_ltc -0.1991 -0.2226 0.0443 -1.3574 0.2008 -0.2726 -0.3942 0.3550 -0.1399 SUBTOTAL 0.0390 -0.1559 -0.1428 α -0.0152 0.4735 0.0662 RESIDUAL -0.0039 0.0000 0.0000 SRI C t t R R 0.0199

0.3176

-0.0765 Panel B - Progressively weighted SRI portfolio

Subsample 1 Subsample 2 Subsample 3

Mean Coefficient Contribution

(30)

6. Conclusion

Incorporation of social responsibility in portfolios is not a new phenomenon. Since the mid-nineties of last century scholars tried to evaluate performance of social responsible equity mutual funds. Yet, many did not examine the influence of social responsible screening on performance of bonds. I intended to add new insights with respect to the existing literature and determine whether the SRI portfolios out- or underperformed the conventional benchmark from 2005 until 2010 using a fixed income performance attribution model. Variables incorporated in this model adjust for different durations, different credit spread changes, other convexities and different levels of credit risk. First, I subdivide the dataset into five different maturity tranches matching the benchmark’s sector weights. Next, I identify two samples of bonds, i.e. a set of bonds with equal weights and a set of bonds with progressive weights. In total ten portfolios are created. However, due to multicollinearity problems, I merge the variables’ values of all five different maturity tranches for each type of portfolio weighting (i.e. equal and progressive weights). All maturity tranches are assigned equal weights. As a consequence, I examine the performance of one equally weighted SRI portfolio and one SRI portfolio with progressive weights relative to a conventional benchmark. I estimate the coefficients for the complete dataset and, in addition, I perform robustness checks regarding three subsamples. Regarding the complete dataset, I add a dummy variable to the regression equations for the bankruptcy of Lehman Brothers.

(31)

B.A. Nijenkamp 30 University of Groningen Results for subsample 2 indicate neither statistical proof of out- or underperformance, nor significant differences regarding credit risk. Estimates for subsample 3 even indicate weak evidence of the SRI portfolios outperforming the significantly riskier benchmark.

However, regarding the complete sample, we see that the SRI portfolios have outperformed the benchmark in absolute terms since there is significant evidence for a flight-to-quality in the period of measurement which was an advantage for the less risky SRI portfolios. Since the last decade’s financial crisis affects the data sample, one might argue that in times of growing economic uncertainty SRI bonds perform better in absolute terms since they gain from the flight-to-quality momentum. Findings regarding subsample 1 support this view. Subsample 1 marks the period prior to the recession which came along with the unprecedented uncertainty on the financial markets. Results for this subsample indicate a flight-to-quality which was advantageous for the SRI portfolios. The findings with respect to subsample 3 show the reverse implying that bonds with higher credit risks gained momentum and therefore, the benchmark gained extra return on top of the additional carry return it receives ceteris paribus due to a higher credit risk. This suggests that riskier non-screened bonds perform better when markets recover from panic in the financial markets.

These findings give a clear insight in how SRI bonds perform relative to a benchmark within different periods of time and what causes SRI bonds to out- or underperform in both relative as well as absolute terms. However, SRI screening approaches are not uniform and may result in different outcomes. Therefore, additional SRI research with respect to the fixed income markets is essential to create a more integrated view.

Limitations

The research I conducted is subject to some limitations. First, this thesis does not take into account expense ratios for managing a portfolio. Some argue that implementing social screens leads to an increase in expense ratios. Derwall and Koedijk (2009) state that the costs associated with screening do not lead to higher expense ratios. Since many different screening approaches can be adopted, the expense ratio may vary over the different SRI fixed income portfolios. Further research should be conducted.

(32)

References

Andrews, Donald W. K., 1993, Tests for parameter instability and structural change with unknown change point, Econometrica 61(4), 821-856.

Barnett, Michael L., and Robert M. Salomon, 2006, Beyong dichotomy: The curvilinear relationship between social responsibility and financial performance, Strategic Management Journal 27, 1101-1122.

Bauer, Rob, Kees Koedijk and Rogér Otten, 2005, International evidence on ethical mutual fund performance and investment style, Journal of Banking and Finance 29, 1751–1767.

Bauer, Rob, and Daniel Hann, 2010, Corporate environmental management and credit risk, Working Paper, Maastricht University.

Becker, Gary S., 1957, The economics of discrimination (University of Chicago Press, Chicago).

Bollerslev, Tim, and Jeffrey M. Wooldridge, 1992, Quasi-maximum likelihood estimation and inference in dynamic models with time-varying covariances, Journal of Econometrics 73, 151-184.

Boutin-Dufresne, François, and Patrick Savaria, 2004, Corporate social responsibility and financial risk, Journal of Investing 13(1), 57-66.

Brooks, Chris, 2008, Introductory econometrics for finance (University Press, Cambridge).

Campisi, Stephen, 2009, Fixed-income attribution made easy, Available at: http://www.cfawebcasts.org/ modules/Catalog/CourseDetails.aspx?ProductGroupID=5598 (accessed: November 18th, 2011).

Chow, Gregory C., 1960, Tests of equality between sets of coefficients in two linear regressions, Econometrica 28(3), 591-605.

D’Antonio, Louis, Tommi Johnsen and R. Bruce Hutton, 1997, Expanding socially screened portfolios: An attribution analysis of bond performance, Journal of Investing 6(4), 79–86.

Daul, Stephen, Nicholas Sharp and Lars Qvigstad Sørensen, 2010, Fixed income performance attribution, Working Paper, Riskmetrics Group.

(33)

B.A. Nijenkamp 32 University of Groningen Eurosif, 2010, European SRI study 2010 (revised edition), Available at: www.eurosif.org/research/eurosif-sri-study/2010 (accessed: November 21st, 2011).

Farrar, David E., and Robert R. Glauber, 1967, Multicollinearity in regression analysis: The problem revisited, The Review of Economics and Statistics 49(1), 92-107.

Goldreyer, Elizabeth F., Parvez Ahmed and J. David Diltz, 1999, The performance of socially responsible mutual funds: Incorporating sociopolitical information, Managerial Finance 25(1), 23–36.

Goss, Allen, and Gordon S. Roberts, 2011, The impact of corporate social responsibility on the cost of bank loans, Journal of Banking & Finance 35(7), 1794-1810.

Hamilton, Sally, Hoje Jo and Meir Statman, 1993, Doing well while doing good? The investment performance of socially responsible mutual funds, Financial Analysts Journal November/December, 62–66.

Hong, Harrison, and Marcin Kacperczyk, 2009, The price of sin: The effects of social norms on markets, Journal of Financial Economics 93, 15-36.

Luther, R.G., and J. Matatko, 1994, The performance of ethical unit trusts: Choosing an appropriate benchmark, British Accounting Review 26, 77-89.

Kreander, N., G. Gray, D.M. Power and C.D. Sinclair, 2005, Evaluating the performance of ethical and non-ethical funds: A matched pair analysis, Journal of Business, Finance and Accounting 32(7)&(8), 1465–1493.

Kurtz, Lloyd, 1997, No effect, or no net effect? Studies on socially responsible investing, Journal of Investing 6(4), 37-49.

Markowitz, Harry, 1952, Portfolio selection, Journal of Finance 7(1), 77-91.

Menz, Klaus-Michael, Corporate social responsibility: Is it rewarded by the corporate bond market? A critical note, Journal of Business Ethics 96, 117-134.

Næs, Randi, Johannes A. Skjeltorp and Bernt Arne Ødegaard, Stock market liquidity and business cycle, Journal of Finance 66(1), 139-176.

(34)

B.A. Nijenkamp 33 University of Groningen Renneboog, Luc, Jenke ter Horst and Chendi Zhang, 2008, Socially responsible investments: Institutional aspects, performance, and investor behavior, Journal of Banking and Finance 32, 1723–1742.

Rudd, Andrew, 1981, Social responsibility and portfolio performance, California Management Review 23(4), 55-61.

Schöder, Michael, 2004, The performance of socially responsible investments: Investment funds and indices, Financial Markets and Portfolio Management 18(2), 122-14.

(35)

Appendices

Appendix A: Sector Weights Benchmark

Table A1 – Sector weights benchmark

Sector weights of the Barclays Capital Euro Aggregate Corporate indices per year per maturity tranche. Weights are expressed as a percentage. Year Maturity Tranche Financial Communi-cations Consumer, Cyclical Consumer, Non-Cyclical Basic

Materials Industrial Utilities Energy Diversified Technology

(36)

B.A. Nijenkamp 35 University of Groningen Appendix B: Multicollinearity

Table A2 - Correlation matrix independent variables

The sample period is 01/07/2005 - 12/31/2010. From the shaded area one can see that correlations among the same type of explanatory variables are very large signaling multicollinearity.

Panel A - equally weighted distribution

Variable SRI it

dmt



SRI jt

s



0.5 ( y )





ktSRI 2 SRI lt

y

–

y

_ltc Variable Maturity

Tranche 1-3y 3-5y 5-7y 7-10y 10y+ 1-3y 3-5y 5-7y 7-10y 10y+ 1-3y 3-5y 5-7y 7-10y 10y+ 1-3y 3-5y 5-7y 7-10y 10y+

(37)

Table A2 (continued)

Panel B - progressively weighted distribution

Variable SRI it

dmt



SRI jt

s



0.5 ( y )





ktSRI 2 SRI lt

y

–

y

ltc Variable Maturity

Tranche 1-3y 3-5y 5-7y 7-10y 10y+ 1-3y 3-5y 5-7y 7-10y 10y+ 1-3y 3-5y 5-7y 7-10y 10y+ 1-3y 3-5y 5-7y 7-10y 10y+

THE ADDED VALUE OF SRI SCREENING IN FIXED INCOME MARKETS?