Investment plan for Europe : prospects of project bonds as a new way of funding corporate investments : an investor's perspective

(1)

Amsterdam Business School

MSc Finance: Corporate Finance Track

Master’s Thesis

1

st

July 2018

Title of the Thesis:

Investment Plan for Europe: Prospects of Project Bonds as a New

Way of Funding Corporate Investments: an investor’s

perspective.

Student Name: Paulina Klaudia Willak Student Number: 11586389

(2)

Investment Plan for Europe: Prospects of Project Bonds as a New

Way of Funding Corporate Investments: an investor’s perspective.

Abstract

This paper examines the dynamics of project bond monthly promised excess returns, as measured by yield spreads, and the risks of holding such debt instrument in an institutional investor’s portfolio. I compare the yield spreads of project vs. matched corporate bonds, and I explore the potential drivers of the discovered price difference. By using a hand-collected subset on 37 publicly traded project bonds with matching corporate bonds issued in the period of 2015-2018, I then show that project bonds add diversification benefits to portfolios. The bonds are in fact a safer investment than corporate bonds given the same credit rating, issue size, and maturity. The source of this price differential lies in unobservable and unique to project bonds risk factors other than liquidity and credit risk. However, project bonds do not offer liquidity risk premium and are found not to have any market hedging abilities. Lastly, the obtained results from a mean-spanning test show that project bonds can improve the capital market line when added to a representative well-diversified long-term investor’s portfolio. Therefore, the overall findings of the study support the use of project bonds across the EU infrastructure plans as they can benefit both the institutional investors, as well as Europe’s economic and social prosperity.

Statement of Originality

This document is written by Paulina Klaudia Willak who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business of the University of Amsterdam is responsible solely for the supervision of completion of the work, not for the contents.

(4)

1. Introduction

The bond trading activity and the outstanding amount of bonds issued have significantly increased over time, demonstrating that nowadays fixed-income positions play a more significant role in investors’ portfolios than ever (Prager et al., 2018). Over the past several years, institutional investors, in particular, have made extensive use of such financial instruments in portfolio composition, increasing the proportion of debt securities to the overall financial assets from 28.07% in 2013 to almost 31% in 20161_{(OECD, 2017). Because of the}

growing importance and scope of bond capital markets, understanding the dynamics of emerging debt securities is crucial to evaluate the true risks and returns they bring to institutional investors’ funds (Bai, Bali and Wen, 2016). This study focuses on bonds of a specific and exotic nature issued since 2015 across the EU area, called project bonds. Project bonds are standardized securities, issued by Public-private-partnership (PPP) infrastructure companies within transport, energy and ICT sector. They are a growing source of long-term funding for infrastructure projects (OECD, 2015), promoted by European Commission (EC) and European Investment Bank (EIB) through the EU 2020 Project Bond Initiative (PBI). What makes the project bonds distinct from any other corporate bonds is that the former are standalone debt obligations whose risk is solely determined on the riskiness of the infrastructure plans alone, rather than specified by the overall firm risk (AFME, 2012).

According to EIB (2018), project bonds would perfectly pair with long-term horizon investors’ profile given the long maturities the bonds have and could add-on to the return and diversification of investment portfolios. Therefore, targeted investors are those with long-term liability structures and regulated rating requirements for their investments, i.e., institutional investors, such as insurance companies, mutual funds or pension funds.

Despite the claimed attractiveness of project bonds, private-sector investors have been reluctant to invest in this specific type of bond. As of January 2017, 444 project bond operations were approved with a total mobilized investment of €170 billion, meeting only 54% of the investment target levels of PBI (Ibid.). Institutional investors fail to fully take advantage of the project investment’s potential mainly because they are not familiar with investments in the infrastructure sector. The lack of publicly-available knowledge on how to specifically benchmark project bonds creates uncertainty. An additional difficulty is that infrastructure assets substantially vary in risk-return profiles (Thierie and De Moor, 2016). There is thus a

(5)

need for more thorough analysis of the risk–return characteristics of project bonds, with taking into account the extensive heterogeneity across bonds. Although a number of papers already exist to assess the attractiveness and constraints of the 2020 Initiative for investors (Vassallo et al., 2017 or Scanella, 2012), they are only based on a qualitative approach and fail to assess project bonds’ relative performance to other debt instruments. Also, no direct empirical evidence exists which would evaluate the channels through which project bonds can benefit institutional bodies. Through my study, I would like to fill in this need for comparable and quantitative analysis. My paper is the first one to analyse the performance of project bonds with regards to traditional corporate bonds. Consequently, it can provide a comprehensive guide for institutional investors whether or not to include this fixed-income instrument in their investment funds.

The main research question this paper tries to answer is to what extent is it beneficial for institutional bodies to hold these bonds in investment portfolios. More specifically, I examine the dynamics of project bond returns, as measured by yield spreads, and inherent risks like credit, liquidity and market risks. The analysis is conducted vis-à-vis traditional corporate bonds, matched in terms of credit rating, maturity, and issue size. This study investigates whether project bonds have a substantially different risk-return profile and whether they have the ability to provide significant liquidity premium or hedge the market risk more efficiently, as opposed to traditional bonds. By additionally conducting a mean-variance spanning test, I establish the diversification benefits of this debt security and its capability to expand the efficient frontier of a long-term investment portfolio.

For the purpose of my empirical analysis, I hand-collect a comprehensive monthly dataset on project bonds and for each matched one conventional corporate bond for the period 2015-2018. By formulating five hypotheses, I investigate the value-add properties of project bonds. For the first four hypotheses, a modified empirical framework of Chen, Liao and Tsai (2011) is employed. Firstly, I evaluate and compare the yield spread performance of project bonds to the matched conventional bonds. Secondly, using a cross-sectional model, I investigate whether credit and liquidity risk fully explain the volatility of yield spreads, or whether other unobservable factors affect the riskiness of project bonds. Thirdly, I examine whether project bonds’ liquidity risk offers a significant premium. Next, I develop a model of bond sensitivity to the economic cycle to see the degree of systematic risk exposure project bonds have on top of the traditional corporate bonds. Lastly, by applying the regression framework of mean-variance spanning developed by Huberman and Kandel (1987), I

(6)

statistically test the diversification benefits project bonds can add to institutional investors’ portfolios.

My major empirical findings are as follows. First, project bonds are in fact a safer investment than matched corporate bonds on the basis of credit rating, issue size, and maturity. The source of this ‘mispricing’ lies in factors that cannot be explained and captured by credit or liquidity risk. Second, project bonds do not provide a significantly better market hedge than conventional bonds and do not offer liquidity risk premium. Lastly, by using a mean-variance spanning test, I drive the conclusion that project bonds can increase the diversification benefits when added to a representative institutional investor’s portfolio, as they improve the capital allocation frontier.

The results are robust to alternative set of matched corporate bonds, partial-sample estimations and a different measure of liquidity. Consequently, the overall findings of the study support the use of project bonds across the EU as they can benefit both the institutional investors in terms of expanding the efficient frontier of their portfolios, and are also able to boost Europe’s economic and social prosperity coming from the increased financing of infrastructure plans.

The remainder of this paper continues as follows. Section 2 will describe in detail what project bonds are and what is the rationale in issuing them across the EU. It will also contain a discussion of previous studies around the topic of project bonds, yield spreads, and portfolio selection theory which will contribute to the development of the five hypotheses. Section 3 presents the various methodologies used to test the derived hypotheses. Section 4 specifies the data used and its summary statistics. Next, Section 5 offers the empirical results, while Section 6 provides robustness checks for obtained findings. Finally, Section 7 shows conclusions and potential limitations of the study.

2. Literature Review

For the purpose of my study, three different streams of academic literature are useful in order to build extensive and comprehensive background information for understanding my study topic. They also allow to directly derive the five hypotheses tested. The first line of literature review involves relevant theoretical details about project bonds and their surrounding research. Next, the literature on the dynamics of bond yield spreads is encompassed, followed by academic studies of portfolio selection theory.

(7)

2.1. Project Bonds and the 2020 Project Bond Initiative

The Institute for European Environmental Policy (IEPS) estimates that until 2020, at least €1.8 trillion euros needs to be raised in order to finance future investments within transport, energy and ICT sector (Sauter, Illés and Nunez, 2014). This infrastructure modernisation will ensure EU’s prosperity and stable competitive positioning on a global scale. Conventionally, the predominant instrument to fund infrastructure projects were loans provided by commercial banks (Khmel and Zhao, 2016). However, the global financial crisis of 2007-08 has resulted in rigorous regulations, constraints and lending requirements, as further implemented by Basel III reforms. Such reforms imposed minimum capital requirements, obliging banks to hold more capital against their assets. They also limit the size of the activities a bank can develop compared to its own capital by introducing specific liquidity coverage and leverage ratios (Council of the European Union, 2017)2_{. Due to these investment limitations,}

infrastructure projects can no longer be solely funded by traditional bank debt which in turn increases the costs of project funding, slowing down the efficient capital allocation process and potentially the entire EU growth. In other words, the resulting lending pressure and suboptimal level of infrastructure capital provision limit social development by preventing businesses and people from making the most of economic opportunities. Without making adequate investments to improve and maintain European facilities, the ability and capacity to produce goods declines, causing the long-term growth to stagnate (Jones, 2017). Therefore, there is an urge to explore new ways to promote private sector financing of infrastructure projects, without the compromise of increasing government funding, thus public indebtedness.

EU 2020 Project Bond Initiative (PBI) is a collaboration between the European Investment Bank (EIB) and European Commission (EC) aiming to revitalise financing of sizeable European infrastructure projects. Alternatively called as an ‘Investment Plan for Europe’ (IPE), it was commenced in November 2014 in order to mobilize public and private investments over the 2015-2017 horizon (EIB, 2012). One of the ways to achieve this goal is to expand the potential investor base, which can be done by reaching out to capital markets.

The fruits of this alternative source of financing are so-called ‘Project Bonds’. Project bonds are standardized securities, issued by Public-private-partnership (PPP) infrastructure companies within transport, energy and ICT sector. They are an emerging part of infrastructure finance and a growing source of long-term funding for infrastructure-related projects (OECD,

2_{Basel III framework introduced a minimum capital requirement of 8% on total risk-weighted assets}

and imposes a leverage ratio (calculated as Tier 1 capital divided by the total of on and off-balance assets less intangible assets) of 3% (see Mann, 2014)

(8)

2015). They are either traded on secondary or in OTC markets where the dealers directly issue bonds to interested investors. A typical project bond is a non-recourse debt instrument with long tenure and an investment grade rating, assessed by multiple credit rating agencies. Its creditworthiness depends on the cash flow performance of the vehicle, with more risky projects (rated as Baa, Ba or lower)3_{having credit enhancement mechanisms attached to them to reduce}

their overall credit risk.

Project bonds are a unique and novel mechanism for the Euro-area bond market. The particularity in their nature arises from the fact that the source of bond repayment are the revenues generated by the infrastructure project itself, which originate from either direct users of the infrastructure facilities (i.e., by using toll roads, or entering into energy contracts), or from contract payments coming from the government for which the project was developed. This is in contrast to “corporate finance,” in which the source of payment is dependent on the whole balance sheet of an operating company (Arca, 2013). Consequently, what makes the project bonds distinct from any other corporate bonds is that the former are standalone debt obligations whose risk is solely determined by the riskiness of the infrastructure plans alone, rather than the overall firm risk. In this sense, project bonds could happen to be riskier because the probability of loss to credit holders is higher when relying on the performance of one specific project versus on a diversified portfolio of projects (OECD, 2015). The risk is intensified because of the nature of the infrastructure projects themselves. Assets attached to transport, energy, and ICT sectors are usually volatile and may not have predictable cash flows as they have high construction risks and costs, especially in the pre-development and early-construction stages of the projects. Most of the institutional investors are not prepared to bear construction and start-up risks and thus require at least an investment-grade rating for the bonds. The second disadvantage relates directly to the issuer of bonds itself. The entire proceeds of a project bond are usually received as a one-off payment at the beginning of the project construction. When unexpected sunk costs arise during the progression of the project, the company needs to absorb the negative carry resulting from the difference between the offering yield of the long-term debt and the short-term ad-hoc investments (Buljevich and Park, 1999). Hence, the credit holders may not receive their coupon payments on time or in full. The additional source of this uncertainty is the fact that large infrastructure projects suffer from a substantial under-management of risk in all stages of the value chain and throughout the life

3_{This paper uses Moody’s credit rating scale (i.e., from highest to lowest grade): AAA, Aa1, Aa2, Aa3,}

(9)

cycle of a project (Beckers and Stegemann, 2013). This poor risk assessment and allocation leads to higher materialized costs and private-financing shortages during the construction phase, putting the initial providers of finance into jeopardy.

However, the aforementioned risk is mitigated thanks to EIB’s provision of subordinated debt instruments which enhance the solvency of senior debt issued by PPP firms. This project bond credit enhancement (PBCE) programme helps to achieve a robust level of creditworthiness attractive to institutional investors (Schoen, Dusina and Castillo-Bernaus, 2013). The mechanism of refining the credit-standing of projects relies on separating the debt of the project company into two tranches: senior and subordinated. The EIB would provide a subordinated tranche, which essentially improves the overall credit quality of the senior bonds, and therefore increases their credibility. The tranche is provided either in the form of a loan given to the infrastructure private company or by way of a conditional credit line which can be introduced if the cash flows generated by the project are insufficient to cover the construction costs or to guarantee the debt service repayment (Liberadzki, 2015). So, despite the bonds being issued by companies operating in reasonably risky sectors, they become safer as the default risk is reduced due to sovereign credit control.

Assuming that the government is effective in supporting any deteriorating project performance through its PBCE programme, including project bonds in investment portfolios can have a number of advantages. Unlike commercial bank debt that typically provides floating rate interest, project bonds, with only a few exceptions, are fixed rate instruments (Smith et al., 2015). Usually fixed rate bonds, in contrast to floating ones, hold more interest rate risk as their real returns are sensitive and tied to the performance of the overall market interest rates (SEC, 2013). However, these fixed coupon payments also mean that when included in the capital structure, they offer a predetermined and solidified constant return (Fabozzi and Mann, 2012) that in turn can help investors hedge the interest risk appropriately and stabilise the debt proceeds. In such sense, they eliminate the level of uncertainty as the value of future interest payments is known today. Project bonds may also benefit the sponsors by providing longer tenure than might be available in commercial banks. Institutional bodies tend to be yield-oriented and match maturities of the bonds against their future capital requirements. For this reason, they are not interested in being prepaid prior to maturity. Given that historically project bonds have been payable at final maturity only rather than having an amortised principal over time, they perfectly match the needs of such investors (Morrison, 2016). The fact of the yield-focused investment decision-making also implies that project bonds do not have to be liquid,

(10)

because in any case institutional sponsors lock up their predictable annuity by holding them until maturity.

Project bonds are not a new concept – they were already commercialised in 1996 and used for project finances in the regions such as the North and South Americas, along with Asia4_{. Nevertheless, project bonds in the EU area are only an emerging topic, where a more}

extensive empirical work needs to be done. Due to the novelty of this type of debt obligation in the EU, little academic research has been conducted around the area of project bonds and their potential value-add to institutional investors across Europe. Most of the existing literature relies on conceptual and theoretical discussions about the project bond’s performance.

For example, Scanella (2012) investigates the uniqueness of project bonds in the EU market, identifies the economic reasons for using project finance and evaluates the proposed financial support from the EIB and EC to boost the European project bond market. The author recognises that predictable cash flows and a strong credit rating attached to the project bonds are the most desirable stable characteristics crucially determining the success and the liquidity of the bonds. Vassallo et al. (2017) evaluate the Project Bond Initiative using the SWOT methodology. They analyse the information coming from both pilot case studies, the responses of institutional investors and other stakeholders to the bond programme. The researchers identify project bonds as a well-matched debt instrument for the investment needs of institutional bodies, with the PBI collaboration as suitable means to achieve adequate bond ratings. However, there are also some weaknesses which undermine the attractiveness of project bonds. For most of the project bonds, the information about risk allocation and construction risks is not transparent enough (Beckers and Stegemann, 2013). Investors do not have a satisfactory knowledge of the financed projects which in turn can reduce the demand for project bonds. Smith et al. (2015) examine the increasing use of project bonds in a selection of diverse regions and infrastructure markets. By looking at the input of prominent investors, bankers, and sponsors, the authors plot the global rise in non-bank debt in greenfield and brownfield projects. The authors conclude that project bonds operate very differently, depending on the geography, industry segments, and country-specific structural and legal features. For this reason, the picture for project bonds is not consistent across the globe meaning that empirical studies concerning the performance of project bonds in the US or Latin America are not representative of the European landscape.

(11)

The only paper conducting empirical research on EU project bonds is of Li, Abraham and Cai (2017). The authors build a structural model which estimates the probability of project bond default. Secondly, they price the credit default swap (CDS) using the risk-neutral valuation method and lastly, they evaluate the bond characteristics on the probability of default and CDS premium. Their study shows that if CDS is fairly priced, it is beneficial for government and companies to implement such bond financing. Its literature contribution is that it explores the role of government debt guarantee to enhance project bond ratings under PPPs.

Gathering all the pieces of academic research, the main difference between corporate and project finance lies within the way assets are being financed, and in the credit enhancement mechanism attached to the infrastructure fixed-income securities. In project finance, the assets are funded as stand-alone entities rather than as part of the consolidated corporate balance sheet. The project needs to generate a satisfactory level of funds to cover all operating costs and debt obligations, while still providing acceptable returns on funds dispensed by institutional investors (Scanella, 2012). Therefore, because of this lack of risk diversification, one can argue that this project-specific credit risk is the main reason as to why the returns, as measured by the yield spread, could be higher for project bonds. In this case, investors would need to be compensated by holding the risk of the entire project, without the possibility of keeping the risk of the whole infrastructure firm which is most likely to have more satisfactory overall risk management.

However, there is an opposing force which could actually make the project bonds less risky than corporate bonds. The source of this dynamism lies in the non-recourse nature of project bonds. This crucial distinction of the project bonds means that nobody else, but the debtholders, have access to the cash flows generated by the infrastructure project itself in the form of coupon payments and are provided to specific information regarding the project cash flow expectations. In contrast, with traditional corporate bonds, managers are the ones with a direct entitlement to use these cash flows according to their personal preferences, i.e., for things other than repaying the bonds obligations. This possibility of cash flow diversion and extraction by managers can make the traditional bonds riskier than their project bond equivalents. These opposing arguments create doubt about which of the effects, if any, dominate in the yield spreads determination.

Nevertheless, because project and corporate bonds are perfectly matched in my dataset concerning their credit risk (measured by the agency credit grade), issue size and maturity, it is predicted that no differences should be observed between the pricing of two bonds. The

(12)

opposing effects should in theory perfectly balance each other out and be irrelevant in this setting. Therefore, one should expect no significant differences within the bond returns, as expressed by yield spreads. This idea leads to my first hypothesis: project bonds have the same yield spread as the matched corporate bonds.

2.2. Yield Spread Dynamics

Yield spreads are defined as the difference between the yield to maturity of a given bond and a risk-free government bond (Kim and Stock, 2014). They are reflecting the additional benchmark needed to compensate for a non-risk-less investment. This measure thus gives a tangible scale of returns investors demand over and above the risk-free asset when holding risky bonds. Yield spread is not fixed over time and rather constantly changes, depending on a number of factors. These can include the supply and demand curves for particular bonds, credit risk and the overall state of the economy. In free markets, where international capital movements have no limitations, the yield spreads on project bonds and comparable corporate bonds issued at one point in time in the same country, denominated in the same currency and of the same risk should be identical (Hillier, 2013), as predicted by my first hypothesis. However, if market environment is imperfect, rational borrowers would seek to issue their bonds in the market with lower growth and investors would shift their funds to the market with higher returns potential. This shifting by issuers and investors would move the yield rates back to the equilibrium level in the long run. Such condition could not hold in the short-run due to liquidity, regulation, tax, firm-specific or bond-specific differences which affect particular fixed-income securities and their required additional compensation for exposure. For example, the higher the credit risk of a bond, the higher the yield spread demanded by investors in order to take on additional risk. Conversely, the higher the liquidity of a security (thus the easiness by which the bond is tradable), the lower the return needed by investors (IDFC, 2006).

The academic literature sheds light on several determinants of the spread of yields between different classes of bonds. The academic yield spread analysis involves, among others, comparing time to maturity, liquidity, creditworthiness, and correlation with macroeconomic conditions of different types of bonds (i.e. Sorensen (1979); Mithal, and Neis (2005); Chen, Lesmond and Wei (2007), or Weinstein (1983)).

Fisher (1959) was one of the first researchers to analyse, using the term structure and liquidity preferences theory, the risk premium of bonds and showed that bond yields are

(13)

positively correlated with time to maturity. As long-term fixed-rate bonds are likely to be subject to greater price fluctuation than short-term notes, they offer higher yields to compensate for the risk involved. These conclusions were later supported by, i.e., Sorensen (1979). However, this relationship is not always positive. Helwege and Turner (1999) find that yield spread decreases with maturity for speculative-grade bonds. In other words, they argue long-term speculative-grade bonds are more creditworthy than short-long-term speculative-grade bonds for the same credit rating. This is because junk bonds with longer maturity have much more time to recover from unfavourable and downside shocks than short-frame bonds. Therefore, their probability of becoming healthy and more reliable is higher.

Regarding the bond riskiness, both Longstaff, Mithal and Neis (2005) and Favero et al. (2010) find that the yield of the corporate spread is mainly delineated by default risk. Shah and Kebewar (2013) attempted to quantify the magnitude of the default risk, and by constructing an efficient meter evince it explains almost half of the change in the yield spreads (precisely 48%). The measure which is used widely used by researchers when estimating the bond default risk is the credit rating grade. Some papers try to establish their risk-capturing abilities. For example, Carey and Hrycay (2001) investigate the consistency between an agency rating and bank default model. By applying a scoring and mapping methodology, they show that the credit grading sufficiently picks up the bonds’ default risk. The study was later confirmed by Godlewski (2007) who corrected the model by dealing with mechanical and informational bias. This proves that agency rating grade is one of the most efficient proxies for expressing the bond credit risk. In the context of my study, the aforementioned academic results imply that project and corporate bonds exhibit no significant differences in the riskiness since they were perfectly matched regarding credit rating. Therefore, my second hypothesis is: the riskiness of project and corporate bonds is the same.

In terms of bond liquidity risk, Chen, Lesmond and Wei (2007) empirically prove that less liquid bonds earn higher yield spreads, and an enhancement in liquidity causes a significant reduction in bond returns. The results remain significant even after controlling for mutual bond-specific, firm-bond-specific, and macroeconomic variables, and are robust to issuers' fixed effect and different roots of endogeneity. The outcomes support the concern in the default risk literature that default risk determinants can wholly explain neither the level nor the dynamics of yield spreads. Chen, Liao and Tsai (2011) empirically prove that the momentous driver of the yield spread differential is the internal liquidity of bonds, after controlling for the remaining bond characteristics. Given the fairly low trading volume, lower issue size and relative unpopularity of project bonds (OECD, 2015), the bid-ask spread are expected to be larger than for

(14)

conventional bonds, which tend to trade at much higher frequencies. This, in turn, should translate in higher liquidity return compensation, as measured by yield spreads, for institutional investors. Putting all the shreds of evidence together, my third hypothesis is as follows: project bonds have lower liquidity than corporate bonds, as reflected by the negative impact on the yield spread. In general, illiquidity is perceived as a risk factor which needs to be rewarded because of the inability to buy and sell bonds at fair prices quickly. However, if you have a long horizon of investment, just like institutional investors do, buying less liquid assets make sense because this inability to promptly buy and sell bonds at a fair price is not desirable, yet rewarded with a liquidity premium (Ibbotson et al., 2015). Therefore, if the hypothesis turns out to be supported empirically, higher benefits can be obtained from a driver that in theory is relatively irrelevant for institutional investors.

Another risk dimension that influences bond returns is the exposure to the business cycle movements. The sensitivity of bonds to the business cycle is the degree to which fixed-income securities are exposed to systematic risk. An example of an early empirical analysis that examines the level of systematic risk in traditional bonds is one of Weinstein (1983). By conducting cross-sectional bond analysis, he explored the extent to which interest rate and default risk explain the variation in bonds. His results show that bonds have relatively low and slightly positive systematic risk, which mostly comes from the exposure to interest and default risk. Brooks, Ingram and Copeland (1983) explore this default risk dimension and find evidence of an inverse relationship between credit ratings and the beta coefficient (systematic risk) of CAPM model. They demonstrate bond ratings may be an adequate measure of systematic market risk, especially for bonds with low liquidity. Later on, Blume, Keim and Patel (1991) conducted a comparative analysis between stocks and different types of bonds and found that a diversified portfolio of low-grade bonds with long-term maturity is subject to lower sensitivity to unexpected changes in macroeconomic variables such as interest rate than the indices of long-term Treasury bonds, long-term high-yield corporate bonds, S&P 500 stocks, and small stocks. Longstaff and Shwartz (1995) expanded this relationship and showed credit spreads for firms with similar default risk could vary significantly if the assets of the firms have different correlations with changes in interest rates. Their statistical model explains why bonds with similar credit ratings, but different industries or sectors of operation have significantly varying credit spreads. Linking to these industry differences, more recent studies (Russ, Thambiah and Foscari (2010), and Oydele, Adair and McGreal (2014)) compute efficient portfolio frontiers using monthly return on infrastructure indices. They demonstrate

(15)

due to their lower correlation to the macroeconomic environment. Given that project bonds compromise both infrastructure asset and fixed-income characteristics, they would, in theory, be even less correlated with the business cycle than traditional corporate bonds. This claim about project bonds gives rise to the following fourth hypothesis: project bond yields are less affected by economic cycles than corporate bonds. This, successively, would mean the bonds can serve as a hedging strategy against market risk.

2.3. Portfolio Selection Theory

The study of modern portfolio selection theory was initiated and developed by Markovitz (1952) who documented that an asset’s risk-return profile should not be assessed as standalone security but instead compared to the already existing investment portfolio. The risk of a portfolio should be measured by comparing the covariance of its returns to the returns given by the market portfolio. Thus, the appropriate measure of a single security risk is the market risk it holds and its degree of covariance to the existing portfolio (Sharpe, 1963). Later on, Evans and Archer (1968) have attempted to quantify the rate at which the exposure to systematic risk can be reduced by adding an asset into a portfolio, and showed it takes the form of a hurriedly diminishing asymptotic function.

More recent studies have been involved in establishing further factors that institutional investors should consider when trying to diversify their portfolios. For example, some authors have evaluated the advantages of expanding the set of tradable assets from the domestic to the international equity markets. Santis and Gerard (1997) estimate that a portfolio which consists of securities from all around the world increase annual portfolio gains by 2,11%. They also prove that global markets have not been entirely affected by the increasing level of integration. Security returns are much less correlated across countries than within the country. This is so because economic, political and institutional country-specific factors affecting security returns tend to vary significantly, resulting in relatively low correlations among domestic and foreign financial products (Eun and Resnick, 1984). De Roon, Nijman and Werker (2001) find that U.S. investors can benefit from including emerging markets assets in their well-diversified international portfolio of developed market assets. However, such diversification benefits vanish when investors face short sales constraints and transaction costs. Even if international diversification is beneficial, French and Poterba (1991) point out that in most of the cases, investors are reluctant to hold international portfolios and tie up a substantial proportion of their wealth in domestic markets, thus are subject to home bias.

(16)

Diversification can also occur by adding different classes or sets of assets, which can potentially yield a portfolio mix coming from stocks, fixed income, money market and real estate. Denis, Denis and Yost (2002) document that firms are increasing the extent of their diversification, by both investing in multiple lines of business (industrial diversification) and across different national markets (global diversification). This search for new assets classes and countries may be due to increased co-movements of large-cap stocks which reduces the diversification benefits, as reported by Eun, Huang and Lai (2008). They additionally suggest that holding smaller-cap stocks is a good alternative for international portfolio diversification. Finally, de Roon, Nijman and Werker (2003) show positive and significantly improved performance of all-equity portfolios after including bonds whose risk-adjusted yield differential is positive.

However, the diversification benefits can turn out not to be substantial when holding assets subject to high correlations of country factors, as pointed out by Lessard (1976). By examining the covariance structures of equity returns across international markets, he demonstrates the diversification benefits occur when investors hold securities among countries with segmented markets, but the effect of reducing non-systematic risk vanishes once a portfolio holds assets subject to highly integrated markets. Therefore, since institutional investors’ portfolios already consist of securities issued by countries that are highly united such as the US and European countries, I would not expect the project bonds to cause a substantial expansion of the efficient frontier. It is because project bonds’ performance is subject to infrastructure plans within the EU which, due to Economic and Monetary Union (EMU), have considerable country synchronisation when it comes to equity and debt capital markets (Adjaoute and Danthine, 2004; Hardvouelis, Malliarpulos and Priestley, 2006). Hence, my fifth and last hypothesis is the following: project bonds give no additional diversification benefits to institutional investors, as their portfolios are sufficiently diversified.

Overall, my analysis aims to contribute to the existing literature in the following ways. To begin with, my study is the first one to fill the need for comparable and quantitative analysis of the risk-return profile of project bonds. It provides a comprehensive guide for institutional investors whether or not to include this fixed-income instrument in their investment funds. Apart from this paper, no empirical study exists which assesses the various risks (credit, liquidity and market risks) of project bonds and puts them into the context of traditional bonds that are already known to institutional investors. All my derived hypotheses, which ultimately depict the contribution and aim of this paper, are graphically presented in Figure 1.

(17)

Figure 1

Contribution and aim of this paper.

The figure graphically presents the contribution and aim of this paper. The empirical setting is divided upon five hypotheses: first one (H1) measures which effect, if any, dominates in project bond price differential against matched corporate bonds. Second (H2) and third (H3) investigate the project bonds’ riskiness and liquidity profiles, respectively. Fourth hypothesis (H4) tests the bond’s sensitivity to the macroeconomic conditions. The last one (H5) establishes project bonds’ diversification benefits. The (+) and (-) signs represent respectively the expected positive and negative effects the specific bond characteristics will have on the regression outcomes.

(18)

3. Methodology

3.1. Calculating the Yield Spread

The yield spread calculation is based on the method adopted by (Finnerty and Nunn, 1985). The yield spreads on project and corporate bonds with the same credit risk are hypothesised to be equal. This means that if the monthly 10-year German government bond yield (i.e., the risk-free rate Rf) was deducted from the corresponding Yield to Maturity of

Project Bonds (YTMPB) and Corporate Bonds (YTMCB), we would end up with the same yield

spread (YSpread) for both debt instruments:

YSpread = YTMPB - Rf = YTMCB – Rf

The bonds in question form a set of matched pairs of project and corporate bonds sharing the same characteristics. These comparable bonds are matched, listed by priority as follows: Moody’s credit rating, size, issue date and maturity date, as well as the currency. Such pairing should, in theory, translate into no difference in the yield spreads, as the following internal bond features should sufficiently capture all the bond risk factors.

3.2. Yield Spread of Project vs. Corporate Bonds

In order to explore the source of project bond yield spreads, an empirical approach inspired by Chen, Liao and Tsai (2011) is used. I regress the yield spreads, a measure of promised excess returns to investors calculated in Section 3.1. (YSpreadi,t), on issue-specific

measures. My model differs from the paper as it excludes most of the company-specific data. The reason for excluding the company-specific data is that project bonds have a non-recourse nature, so their performance is unaffected by the issuing firms themselves. The only macroeconomic variable left in the model is the European stock volatility index. The proxy used for liquidity is the difference between the bid and ask prices quoted on the market (BidAski,t), and the proxy for assessing the bonds’ credit risk is the Moody’s credit rating grade

assigned to each bond (Ratingi). The modification of the aforementioned paper’s approach also

lies in the inclusion of a project bond binary dummy variable (ProjectBondi) to see the

additional effect these debt instruments have on the yield spread, as compared to the matched corporate bonds. In other words, the dummy variable allows seeing whether project bonds have a significantly different yield spread relative to corporate bonds. Consequently, what the

(19)

regression ultimately wants to establish, is whether there is any return difference of project bonds vis-à-vis corporate bonds, and can be written as:

YSpreadi,t = 0 + 1(BidAski,t) + 2(Maturityi) + 3(Ratingi) + 4(Couponi) +

5(StockVolatilityt) + 6(ProjectBondi) + i,t

where BidAsk is the difference between the bid and ask closing prices of each time-bond observation, Maturity is the starting number of years until bond maturity, Rating is the Moody’s credit rating grades measured by a dummy that takes a value of 1 when the bond has an Aa2 rating, 2 if assigned an A2 grade etc. until the last value of 9 explaining the sample’s lowest B1 rating and 0 otherwise. The highest credit rating of the dataset (Aa2) is the default one;

Coupon is a fixed annual percentage rate payable on a bond, and lastly, StockVolatility is the

EURO STOXX 50 Volatility Index (VSTOXX). The period considered is from January 2015 to March 2018.

Following the academic literature on bond liquidity (i.e. Chen, Lesmond and Wei, 2007), I predict the control variables to behave as follows: the bid-ask spread should be positively related to the yield spread, as more illiquid assets need to promise higher returns. As the term structure and liquidity preference theory (i.e. Fisher, 1959) suggests, maturity should be positively correlated with the yield spread. Higher coupon rate should translate into higher yield spread, and higher credit rating grade must negatively impact the dependent variable due to the reduced credit risk of bonds. Finally, higher volatility index implies more market uncertainty which in turn should push up the yield spreads.

Based on the regression, the null hypothesis is that there is no difference in the yield spread between project and corporate bonds, and can be derived as:

H1: 6 = 0

I use a double-sided t-test because of the unknown direction of the yield spread effect. If the null hypothesis is rejected, I can conclude that project bonds, despite controlling for all the internal differences in characteristics, still produce a significantly different yield spread with respect to traditional corporate bonds. If the hypothesis is rejected, there is compelling evidence that the returns of project and corporate bonds vary significantly, even after

(20)

accounting and controlling for principal internal characteristics and movements of the overall stock volatility.

Such empirical analysis has two-fold advantages. First of all, it gives a full picture of the determinants of the project bond yield spreads. Secondly, and more importantly, it allows for a direct comparison of project bonds vis-à-vis corporate bonds. It provides the additional effect project bond could have on the yield spreads, a piece of information which is crucial in evaluating the true value-added characteristics of such bonds for institutional investors. Using only project bonds in the equation, as done by the past academic studies (i.e. Vassallo et al., 2017), does not satisfactory reflect what investors can sincerely get out of investing in project bonds. By directly comparing the performance of project bonds to traditional corporate bonds, investors can know whether project bonds, securities of a particular and exotic nature, can add value to their existing portfolios.

3.3. Riskiness of Project vs. Corporate Bonds

Establishing the pricing of the project bond riskiness with regards to traditional corporate firstly involves computing a p-value of the difference of the historical yield spread volatilities, between project and corporate bonds. The yield spread volatility is treated as a proxy for the overall riskiness of bonds. Such preliminary analysis will give indications whether there are some significant differences between the bonds’ risk profiles:

diff = MeanYSVol(PB) - MeanYSVol(CB)

where MeanYSVol(PB) is the average standard deviation in yield spread (YS) of each project bond, while MeanYTMVol(CB) measures the average standard deviation in YS for each corporate bond. The hypothesis is evaluated on the period 2015-2018.

The following equation is tested:

H: diff = 0

If the p-value of difference turns out to be under 0.05, it means the yield spread volatility of both bond types is significantly different.

Next, I identify what measures to adopt as credit and liquidity risk for the cross-sectional regression model of the historical volatility of the yield spread. One of the most

(21)

commonly used proxy to capture credit risk is the bond’s agency credit rating. The measure of liquidity risk, however, is not equally straightforward. It is vital to note that the definition of liquidity and the way it is measured are not the same as for liquidity risk (Li et al., 2009). Liquidity is the degree to which a bond can be sold or bought within a short time frame at a price close to its consensus value, and the central proxy to estimate it is the difference between the closing ask and bid prices of a security across time (Foucault, Pagano and Roell, 2013). Liquidity risk, on the other hand, is the “unpredictable variation in transaction costs arising from idiosyncratic and institution-specific shocks” (Acharya, 2006) and the threat of low liquidity at the time when investors want to trade their securities. One way to estimate the liquidity risk of bonds is first to compute the market liquidity factor which can be derived from taking the residuals of an autoregressive integrated moving average (ARIMA) model on the monthly difference of the illiquidity measure. This is in line with Amihud (2002); however, my study modifies the regression and applies average bid-ask spread of individual bonds as a liquidity proxy. Thus, the following model is performed:

BidAskt = 0 + 1(BidAskt-1) + 2(BidAskt-2) +

+ 3(BidAskt-1) + t + 1t-1

where BidAskt = BidAskt - BidAskt-1 is the monthly average difference in bid-ask spread of all

bonds between time t and the first lag; BidAskt-1 and BidAskt-2 is the second and third

bid-ask spread difference, respectively; BidAskt-1 is the first lag of the average bid-ask spread across

corporate and project bonds, while the t-1 denotes the moving average constituent which

addresses the serial correlation issue. The expression t is the main component of interest as it

will represent the market liquidity factor needed to estimate the bonds’ overall liquidity risk. The liquidity risk of the subsequent regression will be represented by both the market liquidity factor, as measured by the model residuals, and a bond-specific factor which is approximated by the average bond bid-ask spread.

Once the liquidity factor is extracted from the model’s residuals and assigned to each bond, I perform a cross-sectional linear regression of the historical volatility of the yield spread (YSVolatilityi) on recognised credit and liquidity risks. The project bond dummy variable

(ProjectBondi) is added to see whether risk factors incorporated in project bonds, other than

related to credit and liquidity, are still priced within the model. By entangling each individual measurable risk factors, the regression tries to provide a rough picture whether credit and

(22)

liquidity risks are able to explain the volatility of bond returns fully, or whether there is some additional risk component of the project bonds, which makes the relative pricing significantly different. Therefore, the third hypothesis is tested using the following cross-sectional regression:

YSVolatilityi = 0 + 1(MarketLiquidityFactori) + 2(BidAski) +

+ 3(Ratingi) + 4(StockVolatilityi) + 5(ProjectBondi) + i

where YSVolatilityi is the historical average standard deviation of the bonds’ yield spreads;

MarketLiquidityRiski, is the previously obtained market liquidity proxy; BidAski is the

bond-specific liquidity proxy. The liquidity measures taken together form a liquidity risk proxy.

Ratingi is the assigned Moody’s credit rating grade measuring credit risk, and StockVolatilityi

are the monthly closing prices of CBOE S&P 500 Volatility Index (VIX) which controls for the market return volatility.

The coefficient tested is:

H2: 5 = 0

If the t-test is rejected, it can be concluded that project bonds exhibit other risk profile characteristics than traditional bonds, which is the opposite of what is predicted by the hypothesis.

The last testing of this subsection involves looking at the average Sharpe ratio (S), a measure scrutinising the performance of a security after risk adjustment (Sharpe, 1966). It is calculated for project and corporate bonds. In the context of my study, the Sharpe ratio is computed as:

S = 𝐸(𝑌𝑇𝑀−𝑅𝑓) √𝑣𝑎𝑟(𝑌𝑇𝑀)

where YTM is the average volatility in yield to maturity across bonds, Rf is the risk-free rate

measured by the yield on 10-year German government bond and var(YTM) is the volatility of yield to maturity calculated using average standard deviation across time. Then, the p-value of difference is computed in the same way as with yield spread volatility:

(23)

diff = MeanS(PB) - MeanS(CB) H: diff=0

The test is performed to get an indication of the relative risk-return profiles of corporate and project bonds. However, it needs to be noted that yield to maturity might not reflect the return on a bond. It is rather a projected rate of return on the assumption that the fixed-income security will be held until its maturity date and not called in the meantime (Ross, 2017). In other words, the YTM can be viewed as an effective rate of return which takes into account the market value of a bond.

3.4. Liquidity Premium of Project vs. Corporate Bonds

My research continues with testing whether institutional investors can gain from holding project bonds by capturing the liquidity premium, assuming these bonds suffer from a limited trading activity.

First, the p-value of difference in the mean bid-ask spread need to be calculated to see whether liquidity varies significantly among the bonds:

diff = MeanBASpread(PB) - MeanBASpread (CB)

where MeanBASpread(PB) is the average bid-ask spread of each project bond, MeanBASpread(CB) measures the bid-ask spread for each corporate bond.

The liquidity difference is tested by:

H: diff = 0

Secondly, the same cross-sectional sample is used as in the previous subsection, but it is regressed on the average yield spread. I also remove the market liquidity factor and add an interaction term between project bond dummy and liquidity measure. The regression thus examines whether the liquidity coming from project bond is the driver of significantly different pricing of the bonds and is derived as:

YSi = 0 + 1(BidAski) + 2(Ratingi) + 3(ProjectBondi) +

(24)

The hypothesis is tested on the following coefficient:

H3: 4 = 0

The rejection of the hypothesis and a positive coefficient would suggest that liquidity of project bonds offers a premium, which could be captured by institutional investors without any compromise, given their long-term investment horizons.

3.5. Market Risk of Corporate vs. Project Bonds

To investigate whether there is any correlation amid yield spread and the business cycle, I use the same dataset as in Hypothesis 1. I regress the yield spreads (YSpreadi,t) on the measure

of the business cycle, and add an interaction term with project bonds to explore the explanatory power macroeconomic movements have on the project bond yield spread. The variables used as a proxy for describing the market movements are GDP growth and various European Stock Volatility Indices. Separate panel regressions are estimated for each of the proxies, and after controlling for the bond-specific characteristics, they are computed as follows:

YSpreadi,t = 0 + 1(BidAski,t) + 2(Maturityi) + 3(Ratingi) + 4(Couponi) +

+ 5(StockVolatilityt) + 6(BusinessCyclet) + 7(ProjectBondi) +

+ 8(ProjectBondi x BusinessCyclet) + i,t

where BidAsk is the difference between the bid and ask closing prices of each time-bond observation, Maturity is the starting number of years until bond maturity, Rating is the Moody’s credit rating grades measured by a dummy that takes a value of 1 when the bond has an Aa2 rating, 2 if assigned an A2 grade etc. until the last value of 9 explaining the sample’s lowest B1 rating and 0 otherwise. The highest credit rating of the dataset (Aa2) is the default one;

Coupon is a fixed annual percentage rate payable on a bond; StockVolatility is the Euro Stoxx

50 Volatility Index (VSTOXX); BusinessCycle is a measure of the overall macroeconomic movements, approximated by GDP growth per country of bond issue and the return of stock indices such as Euro Stoxx 50 (SX5E), Euronext 100 (N100) or MSCI Europe.

By testing 8, I can see whether the business cycle is statistically significant in

(25)

H4: 8 = 0

If the project bonds exhibit market hedging abilities as predicted by the hypothesis, I will reject the null hypothesis. This would mean that the business cycle, as expressed by various proxies, has some explanatory power in the yield spreads for project bonds, over and above the corporate bonds. The significantly greater co-movements with the business cycle indicate project bonds could have potential market hedging abilities. Hence, with the appropriate use of investment positions, they could be used as a worthy vehicle to reduce portfolios’ macroeconomic exposure from unfavourable shocks.

3.6. Mean-Variance Spanning Test

To test whether project bonds have an impact on investor diversification, I perform a mean-variance frontier analysis to evaluate the diversification properties of project bonds for international institutional investors, whose portfolios already consist of traditional bonds and stocks. The analysis is conducted using a mean-variance spanning regression, developed by Huberman and Kandel (1987). The principal notion of the test is that a set of “K” risky assets spans a larger set of “N + K” risky assets if the minimum-variance frontier of the “K” assets is similar to that of the “K + N” assets5_{(Zhou and Kan, 2012). The word ‘spanning’ in this sense}

means that the mean-variance frontier created by “K+N” assets overlaps the frontier created by the benchmark asset alone. This interrelating frontier suggests that the additional set of assets does not improve the tangency portfolio, meaning there are no diversification benefits when holding a more extensive pool of assets. In the context of my study, the performed regression statistically tests the impact the introduction of additional project bonds has on the efficient frontier of an investment opportunity set of K benchmark assets, which in this case is a portfolio made up of the global bond and stock indices: Barclays Capital Aggregate Bond Index and the MSCI World Index respectively. Such portfolio is assumed to represent a well-diversified portfolio held by an average institutional investor. The regression used to test the mean-variance spanning is the following:

Return(ProjectBondi,t) = 0 + 1Return(TraditionalBondsi,t) +

+ 2Return(Stocksi,t) + i,t

5_{In the Huberman and Kandel (1987) notation, “K” is defined as a benchmark asset, while “N” denotes}

(26)

where the Return(ProjectBond) is the dependent variable calculated as a total average monthly yield to maturities of project bonds across time (a synthetic project bond index is therefore computed), TraditionalBonds is the total monthly yield to redemption of conventional bonds included in the Barclays Capital Goods, while Stocks refer to the total monthly returns of stocks included in the MSCI Europe). In the conventional mean-spanning notation,

Return(ProjectBond), 0, and  are N x 1 vectors, Return(LocalBonds) is a K x 1 vector, and

1 along with a 2 is an N x K matrix. The vectors r, R, and  are random. The random vector

 is uncorrelated with the random vector R, and the expected value of each element of  is 0. The hypothesis tested is whether the minimum-variance frontier of a set of benchmark assets of a base portfolio of institutional investors (which is assumed to consist of the world bond and stock indices) is the same as the minimum-variance frontier of the benchmark assets plus a set of additional project bonds issued in the EU. In other words, the spanning test allows us to see whether the inclusion of any project bond improves the tangency of a conventional portfolio made up by an institutional body investing in the global range of securities. If it does, it would mean that such debt instruments enhance return and reduce the variance and risk, since the tangency portfolio steepens and maximises the Sharpe ratio.

In order to evaluate the extent of portfolio improvement and whether the frontier remains unchanged when including the additional asset of interest, the value of both the intersection and the coefficient of local bond returns is assessed. More precisely, the last hypothesis involves testing the following coefficients:

H4: 0=0, and 1+2=1

If I fail to reject the hypothesis, it means that the returns on project bonds can be perfectly mirrored by the returns of the traditional institutional investor’s portfolio (Petrella, 2005). In other words, the outcome of holding the set of additional project bond assets simply reproduces the frontier of the investor’s already existing portfolio. This means project bonds would not have a substantial diversification benefit in expanding the efficient portfolio, or, put differently, institutional investors already have a sufficiently diversified portfolio. Hence, project bonds would not add any benefits in terms of higher returns or lower risk as compared to the overall set of investments. Empirically speaking, the lower the p-value of the coefficients, the higher the diversification benefits (Belousova and Dorfleitner, 2012) as the

(27)

differences in returns will be more statistically significant. Consequently, if H4 is true, the regression should result in statistically significant values of zero for the intercept (0=0) and

one for the sum of the beta coefficients (1+2=1). In contrast, if 0=0 or 1+2=1, it will

indicate adding the new asset to the benchmark portfolio would not be as valuable for the institutional investors because the asset does not give any substantial diversification benefits, therefore spans the portfolios of long-term investors.

4. Data and Descriptive Statistics

The dataset consists of monthly observations of project, and matched corporate bonds along with macroeconomic variables such as 10-year government yields, GDP growth, stock volatility and return indices. It covers the time period from 01/01/2015, the date when the ‘Investment Plan for Europe’ Initiative was officially commenced (although a couple of project bonds were issued in 2014, or even in 2013), until the latest data available on DataStream, thus March 2018.

4.1. Project Bonds Data Collection

One of the most challenging parts of my study was obtaining the full information about project bonds. As mentioned earlier, project bond information is mostly confidential and private, with only a small percentage of bonds being actually traded on an exchange. Therefore, the details of project bonds are not sufficiently transparent to the broader public.

The data collection firstly involved hand-picking the names of the project bonds from various EIB and EC brochures, investment banks’ newsletters and other financial institutions research documents around the topic of project finance6_{. Unfortunately, these sources provide}

only basic information about the bonds (i.e., coupon rate and project location at most). After collecting 180 project bonds motivated by the 2020 Initiative, I try to identify these bonds on DataStream in order to obtain a complete time-series data of variables such as price and yield to maturity. I make sure the bonds acquired from DataStream are in fact project bonds by matching them in terms of the coupon rate and date of issue with the previous sources. Next, I

6_{The project bond names mainly come from Credit Agricole, ‘Project Bond Newsletters’, EY ‘Ad-hoc}

Audit of the pilot phase of the Europe 2020 Project Bond Initiative’, White Case ‘Project bonds: Their growing role in global infrastructure finance’, Afme ‘Guide to Infrastructure Financing’, S&P ‘Europe's Investment plan surges to €500 billion but is it working?’ and Gatzert N., Kosub T. (2016). The Impact of European Initiatives to Promote Infrastructure Investments from the Insurance Industry’s Perspective.

(28)

exclude all project bonds with unavailable time-series dimension and those which refer to infrastructure plans that cannot be attributed to one specific country (i.e. some projects such as highways or railways involve multiple EU countries). In addition, I dismiss any bond-time observations with missing ask and bid prices or credit rating information. Any outliers that may bias the coefficients in the models are also eliminated, such as negative coupon rates and issue sizes. Overall, out of all possible 444 project bonds, the panel data could be obtained only for 37 project bonds. The whole list and full description of project bonds used can be found in Appendix Table A.1.

I choose monthly, instead of daily bond data frequency. Although it is widely shown that using daily bond data significantly increases the power of the tests, relative to the monthly data (i.e. Bessembinder et al., (2008); Flannery and Protopapadakis (2002)), Shiller (1989) stresses out that less frequent sampling may be more beneficial in cases where the bond data length is only of few years, which is the case in the context of project bonds. In addition, given project bonds are fairly illiquid securities, it is recommended to use weekly or monthly data for a better overview, as the idea of their liquidity may be biased with a more frequent sampling. Bao, Pan and Wang (2011) also collect monthly bond data to assess the illiquidity of junk corporate bonds and the frequency choice is motivated by the same arguments.

Out of all collected project bonds, the majority of them (nineteen) involve projects within the transport sector, such as the construction of highways, railways, and ports. Fourteen bonds are issued to finance energy projects in the EU area like renewable energy power plants or oil and gas extractions. Only three project bonds encompass ICT projects of improving telecommunication systems. This small proportion makes sense given that projects in ICT sectors are considered to be most risky regarding stable cash flow generation; therefore, bond issuing is not considered to be an optimal financing solution. The last one involves ‘other’ infrastructure project of a housing regeneration in the United Kingdom. In terms of currency issuing, 70% of bonds are denominated in Euro, followed by British pounds (10%) and US dollars (5%). More than half of the bonds are traded on either Deutsche Boerse AG, London or Luxembourg Exchange.

Only 32.5% of project bonds have an investment-grade credit rating (i.e. A3 or above). The lowest recorded credit rating is Baa3 and is shared among five project bonds. By looking at the spread of the credit ratings, it seems that, despite the sovereign credit enhancement mechanisms available, most of the bonds fail to have a credit rating desired by institutional investors. This could be the reason as to why institutional bodies are actually reluctant to invest-

(29)

the bonds seem to do not deliver the promised robust creditworthiness as required by these lending bodies.

4.2. Matched Corporate Bonds

In order to determine whether project and corporate bonds are priced differently, the bonds should perfectly emulate regarding their internal, as well as external characteristics. For this reason, to build a comparable subsample of conventional bonds, for each project bond, I search in DataStream for one corporate bond with the most corresponding bond attributes. The matching priority is as follows: first, the bonds need to be issued and mature in similar periods, with a maximum allowed discrepancy of two years. Secondly, the bonds need to have the same credit rating7_{to reflect the similarities in the creditworthiness of both types of bonds. Thirdly,}

they should have similar issue size, which in this case stands for a proxy of liquidity and their fairly equal level of attractiveness to the overall debt market. Ideally, they are denominated in the same currency and traded on the same exchange. However, these conditions do not necessarily need to be met. Nonetheless, the more accurate the matching of corporate bonds, the more meaningful comparisons can be drawn from the empirical analysis. The final set of the matched corporate bonds is listed in Appendix Table A.2.

4.3. Macroeconomic Data and Market Indices

The data about the GDP growth was obtained mainly through DataStream platform, but for smaller or less developed economies, where the data was limited, it was retrieved from other sources like Eurostat, OECD, Federal Reserve and IMF Data. Some of the figures for individual countries could only be obtained on a quarterly or annual basis – if this was the case, then the monthly observations were computed as an average of the less frequent data points8_.

Data about stock volatility (EURO STOXX 50 Volatility Index) is acquired through STOXX.com, while the returns on stock indices such as Euro Stoxx 50, Euronext 100 or MSCI Europe were attained from DataStream. Both the closing prices of CBOE S&P 500 Volatility Index (VIX), and the Fama-French three factors (SMB, HML and MKTRF) are downloaded from Wharton Research Data Services (WDRS). The Barclays Capital Aggregate Bond Index monthly returns were also obtained from DataStream, while the German 10-year monthly

7_{If Moody’s credit rating is not available for a bond, I translate other agencies’ ratings into Moody’s}

equivalent grade (i.e., S&P’s AAA grading is converted into Moody’s Aaa; AA+ into Aa1, etc.).

8_{For example, if the unemployment rate for Bangladesh was of 4.2% in Q12015, then the average}

Investment plan for Europe : prospects of project bonds as a new way of funding corporate investments : an investor's perspective

Amsterdam Business School

MSc Finance: Corporate Finance Track

Master’s Thesis

1

July 2018

Title of the Thesis:

Investment Plan for Europe: Prospects of Project Bonds as a New

Way of Funding Corporate Investments: an investor’s

perspective.

Table of Contents

Investment Plan for Europe: Prospects of Project Bonds as a New

Way of Funding Corporate Investments: an investor’s perspective.

Abstract

1. Introduction

2. Literature Review

3. Methodology

4. Data and Descriptive Statistics