Master Thesis
MSc International Financial Management and MSc Finance
Country Risk Premium and Valuations in Emerging Markets
A Case Study of the Energy Industry in Russia
Abstract
2
Introduction
The size of the equity market risk premium (MRP) when estimating a firm’s cost of equity (COE) is an on-going debate in corporate finance (Damodaran, 2003; Estrada, 2007; Soenen and Johnson, 2008). Since a firm’s COE is one of the main inputs for the calculation of the weighted average cost of capital (WACC), the method applied by a company to measure it’s COE can have a substantial effect on valuation decisions (Ogier, Rugman and Spicer, 2004). Investment projects in emerging markets (EMs) further complicate the issue, as these investments involve risks not accounted for in the basic WACC formula, which has been developed based on data from developed markets (James and Koller, 2000; Cruces, Buscaglia and Alonso, 2012). As financial theory tells us, higher undiversifiable risk requires a higher return as compensation (Damodaran, 1999). Hence, the circumstances in EMs require the incorporation of various dimensions of risk in the cost of capital, making the estimation of the COE in EMs, and hence the execution of valuations in EMs, a complex task (Balas, 2014; Estrada, 2007). In spite of these higher risks and complexities, the growth of the economies and financial markets of emerging countries has attracted the attention of investors searching for higher returns and portfolio diversification. This increasing attention has intensified the debate regarding the country risk premium (CRP); should companies demand a different MRP when investing in EMs, and if so, how should various dimensions of country risk be incorporated in the COE (Balas, 2014; Damodaran, 2004, 2015)? Despite its shortcomings, the Capital Asset Pricing Model (CAPM) and the global CAPM have become a widely accepted approach among practitioners when estimating the required return on equity of investments in developed economies (Estrada, 2000; Harvey, 2005; Pereiro, 2002). With regard to evaluation of investment opportunities in EMs markets however, such a widely accepted model does not exist (Estrada, 2007). Although is seems that the majority of academics and practitioners agree that if country risk exists, it must be rewarded with a CRP over an equivalent investment in a developed economy (Naumoski, 2012), the question of whether or not this additional country risk does exist remains unanswered. And if the CRP exists, how to account for this additional risk? Although academics and practitioners generally seem to agree on the basics of risk in financial theory, they disagree when it comes to the approach taken to measure and incorporate the country risk into a framework (James and Koller, 2000; Keck, Levengood and Longfield, 1998) due to their emphasize on different facets (Naumoski, 2012).3 The lack of consensus regarding issues such as the degree of integration between financial markets, and the extent to which investors are diversified further complicate the debate (Damodaran, 2003, 2015). Moreover, the lack of a standard definition of EMS and risk in EMs complicates the measurement of risk, and hence, its conversion into an expected return on investment (Naumoski, 2012; Soenen and Johnson, 2008; Zenner and Akaydin, 2002). Due to these unresolved disputes no standard exists for the estimation of the COE in EMs, resulting in an abundance of models. Eight of the most applied models will be discussed, of which four are selected to be tested.
Since no general consensus exists on whether or not the CRP exists, and hence, on which model(s) to use in which situation, it is difficult for companies to determine how to evaluate projects in EMs (Soenen and Johnson, 2008; Zenner and Akaydin, 2002). Because the choice of model can have a significant impact on the valuation and investment decisions of a corporation (Ogier et al., 2004), it is important to have good reasons to use a certain model over its alternatives (Estrada, 2000). The aim of this thesis is therefore, first, to investigate if the CRP exists, and secondly which model provides the best estimates. The first question is examined through the execution of correlation and regression analyses based on panel data of 25 EMs. To answer the second question the focus will lie on Russia specifically, by applying the four selected models to the case study of Nord Stream (NS); an international consortium set up by five large players in the European energy industry, for the construction of two natural gas pipelines from Russia to Germany.
As the purpose is to provide NS with guidance on the practical application of the CRP, the focus in this thesis lies on the 8 most widely discussed CAPM-based models whereas the theoretical models are disregarded. An often-mentioned shortcoming of most models is the fact that they do not differentiate between industries, while the influence of country risk might differ per industry (Sabal, 2008; Cruces, Buscaglia and Alonso, 2012). As the focus of this case study lies on NS, this paper specifically focusses on the energy industry, and hence, overcomes this criticism. Clearly, the outcome of this analysis will be specific to NS and Russia.
4 results of the different tests and the fourth section discusses the implications the implications of these findings. In the fifth section a conclusion is provided. The seventh section discusses the limitations of this thesis and proposes directions for future research. The last section entails the references, appendices and an index of the abbreviations.
Literature Review
Although there are many different risk and return models in finance, often with differing assumptions, they all share a common set of ideas about risk (Abuaf, 2015; Damodaran, 1999). Firstly, all models agree with financial theory, which explains that when taking higher risk, investors expect a higher return (Donadelli and Prosperi, 2011). Next, risk is determined in terms of the variance of actual returns around the expected return (Damodaran, 1999). Moreover, the perspective of a well-diversified marginal investor is taken, meaning that risk exists out of firm specific and thus diversifiable risk, and out of market specific risk, which is non-diversifiable (Soenen and Johnson, 2008). Investor should be compensated with extra return only for the non-diversifiable risk. Hence, a required return consists of a risk free rate and a risk premium for the extra [non-diversifiable] risk taken (Estrada, 2000).
5 An implicit assumption of the original CAPM is the full integration of market. This means that an asset with a particular amount of risk will get the same expected return anywhere in the world, regardless of the country it is traded in (Estrada, 2000; Mongrut et al., 2010). In order to determine whether or not country risk is diversifiable, one should first decide whether or not the marginal investor in a country [the investor most likely to be trading on the equity market] is globally diversified (Damodaran, 2015; Ogier et al., 2004). If this is the case, the potential for global diversification exists, otherwise chances of diversifying country risk decrease considerably (Cruces et al., 2012; Naumoski, 2012). To estimate this, Damodaran (2015) follows Stulz (1999) by analysing if a market is open or segmented. In a segmented market, investors either will not or cannot invest abroad, resulting in a risk premium that is different for each country (Damodaran, 2003; Naumoski, 2012; Sabal, 2004). In open markets, on the other hand, investors can and will invest across countries, realizing global diversification and resulting in one universal global risk premium (Damodaran, 2003; Naumoski, 2012). Over the last couple of decades, barriers to trading across markets have dropped, providing investors with more opportunities to globally diversify (Salomons and Grootveld, 2003; Yang, Tapon and Sun, 2006). Nevertheless, not everyone agrees that this means that all investors are globally diversified. Although research has indicated that developed markets are becoming more and more integrated, the assumption of full integration of emerging country capital markets and the world market seems at odds with the evidence (Cruces et al., 2012; Fuenzalida and Mongrut, 2010; Yang et al., 2006). Additionally, although markets are becoming more open, research has indicated that investors have a home bias in their portfolios, leading markets to remain at least partially segmented (Fuenzalida and Mongrut, 2010; Koedijk, Kool, Schotman and Van Dijk, 2002). The general conclusion is that EMs are less integrated than developed markets (Ogier et al., 2004).
6 uncorrelated for country specific risk to be diversifiable (Bekeart, Harvey and Lundblad, 2001; Kruschwitz, Loffler and Mandl, 2013).
In summary, EMs and the world market appear to be partially integrated and correlated. Hence, the question arises whether or not this is enough for country risk to be non-diversifiable. Clearly, no consensus has been reached on this issue so far.
Practitioners vs. Academics: Discount Rate or Cash Flow Adjustment?
Although academics and practitioners depart from models with similar risk assumptions (Abuaf, 2015; Meldrum, 2000), they tend to emphasize different aspects, which has led to disagreements about how to account for country risk when conducting valuations in EMs (Keck et al., 1998; Meldrum, 2000). The main difference lies in the fact that practitioners use the CAPM and adjust discount rates for country risk, whereas academics do this in the cash flows (Abuaf, 2015; Koller et al., 2010).
Discount Rate Adjustment
7
Table 1
A difficulty that all measures face is the lack of data in EMs (Fuenzalida and Mongrut, 2010; Ogier et al., 2003). The data that is available for EMs is often only available for very short time samples (Salomons and Grootveld, 2003). Because of the high volatility characterizing financial markets in emerging economies these outcomes are deemed unreliable (Keck et al., 1998; Soenen and Johnson, 2008), especially because the shorter the time series, the greater the standard error (Damodaran, 20151).
Yet another criticism is that the addition of the CRP, measured with either one of the above proxies, implies that all projects, companies and industries are equally exposed to the country risk in a country (Abuaf, 2015; Harvey, 2005). As country risk exists out of factors such as politics, finance and economy, it is likely that country risk affects some business sectors than others (Koller et al., 2010; Sabal, 2008). It might even be that some sectors benefit where others are damaged (Cruces et al., 2012). 1 Damodaran estimated the global equity premium based on 112 years of data for 20 countries and found a standard error of 1.5%, which is still a rather wide range. Proxy Description Credit default swap [spread] (CDS)
The cost of insuring against a firm’s default, reflecting the probability of default (Giglio, 2011).
Relative equity market volatility (RV)
Scales the standard deviation of a local (emerging) stock market against the standard deviation of the stock market of a risk free country, like the US, multiplied with the equity risk premium used for US stocks (Damodaran, 2015).
Bond default spread (BS) The spread between the yield of an emerging country’s local US dollar denominated government bond and the yield on a risk free US dollar denominated government bond of similar maturity (Damodaran, 2003, 2015; Koller et al., 2010; Naumoski, 2012).
BS + Relative standard deviation (RVLBS)
8 Due to the lack of a standard definition of country risk, it is not clear how country risk should be accounted for (Kruschwitz et al., 2013; Meldrum, 2000). This often leads to arbitrary ad-hoc adjustments based on heuristics, intuition and perception (Harvey, 2005; Lessard, 1996; Pereiro, 2006). Theses adjustments are often in violation with the assumptions and theoretical foundation of the CAPM (Abuaf, 2015; Sabal, 2008). Another important point is that, when using the CAPM, the inclusion of a CRP into the discount rate presumes that the country risk is fully systematic, and thus undiversifiable (Cruces et al., 2012; Sabal, 2008).
[Weighted] Cash Flow Adjustment
Because adjusting the discount rate with a CRP lacks both theoretical and empirical evidence, “accounting for [country] risks in the cash flows through probability weighted scenarios provides both a more solid analytical foundation and a more robust understanding of how value might (or might not) be created” (James and Koller, 2000, p.81). Incorporating the country risk in the estimation of the cash flow scenarios allows managers to analyze specific risk and their impact on value providing more insights, whereas the various dimensions incorporated in the adjusted discount rate does not provide this information (Fuenzalida and Mongrut, 2010; Zenner and Akaydin, 2002). Furthermore, as mentioned, many country risks are idiosyncratic, meaning they do not equally affect all industries. While an adjusted discount rate does not account for idiosyncratic risk, the scenario adjusted cash flows does allow one to take these differences into account, avoiding over- or understatement of risk (Koller et al., 2010). Finally, as mentioned, the adjustment of the discount rate assumes that country risk is entirely systematic and cannot be diversified. Proponents of the adjusted cash flow approach (Koller et al., 2010) claim that most country risks, like devaluation, war and expropriation, are largely diversifiable and therefore better accounted for in the cash flow. The effective risk adjustment of cash flows therefore eliminates the need for adjusting the discount rate (Damodaran, 2015).
9 2003; Koller et al., 2010; Zenner and Akaydin, 2002). Another argument in favor of the adjusted discount rate is that in order to estimate truly unconditional cash flows it is necessary to consider all possible scenarios (Garcia-Sanchez, Preve and Sarria-Allende, 2010). As it is very difficult to picture the exact effects a country risk will have on a company’s cash flows, this implies that, just as the adjusted discount rate, adjusting cash flows involves ad-hoc and arbitrary decisions (Abuaf, 2011; Damodaran, 2003, 2015).
Global CAPM: Beta or Discount Rate Adjustment?
A third discussion evolves around the Global CAPM, which is a version of the traditional CAPM adapted to a global market. Supporters of this model believe that all companies should face the same global equity risk premium, regardless of where its operations take place, and that instead the differences in country risk is captured by the beta (Damodaran, 2015; Godfrey and Espinosa, 1996). Because the beta shows the responsiveness of a company to swings in the market, this would mean that a more risky stock from an EM should have a higher beta than a stock from a developed market. Although in first instance this seems logical, several arguments against this assumption have been raised.
First of all, the market beta is one, meaning that the average beta of each market has to be one. Hence, Damodaran (2015, p.46) claims “it would be mathematically impossible for betas to capture country risk”. A solution might then be to estimate a country’s beta against a global equity index (e.g. the MSCI), as theoretically speaking, there is a possibility this would capture country risk (Damodaran, 1999, 2015). Nevertheless, so far, research has not found evidence that they do (Pereiro, 2006). In contrast, research did find that companies in the same sector from all over the world tend to be scattered around a median with little statistical dispersion. (Abuaf, 2011). Moreover, since global equity indices are market weighted and companies from developed markets largely cover global market capitalization, this means that the betas from developed markets weigh more heavily, and hence the betas of developed markets will be higher. Therefore, the use of adjusted betas assuming one global equity premium leads to a lower country risk for EMs than developed markets (Damodaran, 2015; Erb et al.,1996).
11 A possible explanation for the different outcomes found is that tests have been subject to the lack of historical data in EMs, which limits the time sample of the tests. Furthermore, the earlier mentioned effects of downside risk might further blur the results (Estrada, 2000; Salomons and Grootveld, 2003). Hence, although the equity market might factor in a sizable CRP, it does not do so consistently (Harvey, 2004, 2005; James and Koller, 2000).
What is country risk and what is an emerging market?
As mentioned, no standard definition of country risk exists. For this reason, it has been difficult to find a measure of country risk. Clear from the earlier discussed literature is that it is generally assumed that country risk is only an issue for EMs, implying that developed markets are riskless. Empirical evidence shows this might be true. Whereas the relationship between several risk measures and the implied cost of capital in EMs was found highly statistically significant, no such a statistical significance was found for developed markets, indicating that these countries might not be risky (Harvey, 2004). Besides indistinctness about a definition for country risk, this raises the question; what is an EM and what is a developed market? A developed economy generally refers to a country with a high level of economic growth and security, being measured by GDP, standards of living, infrastructure, level of industrialization, but also by non-economic factors such as the Human Development Index (HDI) (International Monetary Fund, 2015; United Nations Development Program, 2014). Within the literature of CRP, EMs are not clearly defined. A characteristic often mentioned, however, is that EMs are perceived as more risky. The Financial Times (2015) describes an EM as a developing country that is growing and urbanising fast, “in which investment would be expected to achieve higher returns but be accompanied by greater risk”. Hence, it can be stated that EMs are defined by their characteristics of high returns in volatile and risky economies (Cruces et al., 2012; Lessard, 1996). But then what about for example Greece? As a high-income country, Greece is generally considered a developed country (Worldbank, 2015a), yet in its current volatile economy there are not many investors willing to invest in Greece (CNN, 2015; New York Times, 2015). Is this then not a CRP?
12 still differ. For example, Donadelli and Prosperio (2011) show that the mean (M) and the STD of monthly excess returns of equally weighted portfolios differ per continent2. The GDP growth rates in
13 Based on the above discussion, the questions ‘what is country risk?’, ‘does country risk exist?’ and ‘what is an EM?’ remain. Although there might be exceptions like Greece, for the purpose of this paper I do follow the mainstream theory by assuming that developed countries are risk free and do not need to be accounted for in a specific way. Moreover, based on the literature (e.g. Erb et al., 1995; Godfrey and Espinosa, 1996; Meldrum, 2000; Naumoski, 2912; Ogier et al., 2004) I assume that country risk includes political, economical, financial, social and institutional risks. Finally, for the case study described later, I apply Damodaran’s (2003, p.63) statement that “a company’s exposure to country risk should not be determined by where it is incorporated and traded” but by “a company’s operations” to determine NS’s country risk exposure.
The Models
Despite its shortcomings, the Capital Asset Pricing Model (CAPM) or the international CAPM has become the widely accepted approach among practitioners for the estimation of the required return on equity of investments in developed economies (Harvey, 2005; Pereiro, 2002). With regard to evaluation of investment opportunities in EMs however, there is no such widely accepted model (Estrada, 2007). Although from a theoretical point of view the scenario-adjusted cash flows are the best alternative, the majority of models accounting for country risk are variations of the CAPM (Abuaf, 2011; Zenner and Akaydin, 2002). There are a number of reasons for the popularity of the CAPM-based models;
14 measures is that, as rating agencies do not disclose their exact methodologies, there is no transparency, which reduces reliability (Naumoski, 2012; Revoltella et al., 2010), especially because credit ratings are subjective to perceptions of risk (Pereiro, 2001). Besides, credit ratings are not linear so they do not provide any information regarding the comparability of countries’ country risk (Naumsoki, 2012). Finally, since rating agencies are looking for stability they do not reflect immediate changes as they lag behind the market (Naumoski, 2012; Revoltella et al., 2010). As a result, it can be stated that sovereign credit ratings are based on historic data, making it a backward-looking measure, whereas the CRP is a forward-looking concept (Giglio, 2011; Naumoski, 2012; Soenen and Johnson, 2008).
Market-based measures are constantly updated, reflecting the current point of view of investors (Damodaran, 2015; Naumoski, 2012). The main disadvantage of market-based measures, however, is therefore that they show the same characteristics as the markets; volatility and at some times irrational behaviour (Damodaran, 2015; Revoltella et al., 2010). This is especially true for CDS spreads; although on the long run they move along with bond spreads, on the short run they move ahead of the bond market (Varga, 2009; Zhu, 2006). Nevertheless, market-based measures try to circumvent the limitations of sovereign credit ratings by aggregating individual default probabilities of financial institutions, making it a forward-looking measures, reflecting risk-neutral rather than objective default probabilities (Giglio, 2011; Sabal, 2008). Therefore, “the most widely used measures of systemic risk […] are market-based”(Giglio, 2011, p. 1), and for the same reason, this thesis focuses on market-based measures.
The 8 models that will be handled next have been chosen because they are the most widely discussed CAPM-based models in the literature (e.g. Balas, 2014; Estrada, 2007; Harvey, 2005; Pereiro, 2006; Ogier et al., 2004; Von Jenner, 2008). All the models account for currency risk by selecting a hard based currency like the US dollar (USD), and use bonds denominated in this currency (Estrada, 2007). Additionally, whenever the term CRP is used, this means one of the proxies for country risk premium mentioned in Table 1 can be used. In some cases the developer of the model specifically uses a certain proxy; this will be indicated when this is the case.
Home CAPM
15 The main advantage of the HCAPM is that it is easily applied. Nevertheless, one should be careful since the model is only applicable to an international investment of an investor that is concentrated in its home market, and perhaps to a very limited extent internationally diversified (Ogier, et al., 2004). Additionally, as mentioned, although research does not agree on the exact degree of integration, an overall conclusion is that there are no markets that are completely segmented; hence the applicability of this model decreases. Furthermore, this model does not consider the fact that the degree to which a company is affected by country risk might differ per type of industry (Von Jenner, 2008). Local CAPM Just as the HCAPM, the Local CAPM (LCAPM), assumes that markets are segmented (Fuenzalida and Mongrut, 2010; Von Jenner, 2008). Instead of benchmarks from the investor’s home country, however, the LCAPM bases its variables on the target country’s market information (Ogier et al., 2004; Pereiro, 2006). Hence, the expected return on equity becomes; ! !" = !"ℎ + !ℎ ∗ ! !"ℎ − !"ℎ + !"# Where; E(Rx) = Expected return on equity investment in target country RFh = Home market risk free rate
!ℎ = Equity beta of target company against the home country market E(RMh) = Expected market return on home market
CRP = Country risk of target country
! !" = !"# + !" ∗ [!(!"# − !"#]
Where; E(Rx) = Expected return on equity investment in target country RFx = Target market risk free rate
!! = Equity beta of target company against the target country market
16 When comparing this formula with the formula of the HCAPM one can see that the LCAPM formula does not have a CRP factor. Since the RF of the target country is based on the yield of the target country’s government bond, which implicitly includes a premium for the country risk of the target country, the CRP is already included in the LCAPM (Ogier et al., 2004). Although this seems as a large advantage as compared to the HCAPM, this model assumes that markets are efficient, which is not the case in EMs (Pereiro, 2001). Additionally, the variables of the LCAPM are difficult to estimate due to the lack of sufficient and reliable information in EMs (Funzalida and Mongrut, 2010; Sabal, 2004). Furthermore, since the LCAPM assumes segmented markets, the model is difficult to apply to the ‘real world’, as there are no completely segmented markets (Balas, 2014; Pereiro, 2006). Finally, an interesting comment by Ogier et al. (2004) is that the assumption of segmented markets implicates that country risk is diversifiable, whereas the automatic increase of the RF rate as a result of the implicit country risk in the government bond yield indicates the opposite.
Global CAPM
As opposed to the previous two models, the Global CAPM (GCAPM) assumes that the increased globalization has led to full integration. Hence, the different inputs of the model are based on the assumption of global supply and demand for all forms of capital (Harvey, 2005; Ogier et al., 2004). Consequently, the COE becomes; Although the incremental integration of EMs has led them to be more sensitive to global factors, as discussed, full integration does not exist. Additionally, full integration assumes the global market is perfect (Balas, 2014; Pereiro, 2001, 2006). In the light of conspicuous market imperfections, such as a deviation from the Purchasing Power Parity (PPP), these assumptions, and thus the model, do not hold (Fuenzalida and Mongrut, 2010; Harvey, 2005). Nevertheless, Stulz (1999) argues that the GCAPM can be applied anyway as most prices are determined globally instead of locally, which favors the GCAPM over the HCAPM and LCAPM.
! !" = !"# + ! ! ∗ [!( !"#) − !"#] + !"# Where; E(Rx) = Expected return on equity investment in target country RFg = Global market risk free rate
!" = Equity beta of target company against the global market
E(RMg) = Expected market return on global market
17
Hybrid CAPMs
The above three models are based entirely on home, local or global variables and assume that markets are either entirely segmented or entirely correlated. As discussed, empirical evidence seems to indicate that the situation in the ‘real world’ lies somewhere in between. The following models, also known as Hybrid CAPMs, assume markets are not completely integrated or segmented and combine global, home and local variables.
Lessard Model
The Lessard model (1996) suggests that the adjustment of country risk can be made in the stock beta rather than the by adding a CRP (Fuenzalida and Mongrut, 2010). For his model, Lessard makes a clear distinction between “symmetric” and “asymmetric” risks. Symmetric risks are market risks with similar up- and downside potential, such as fluctuations in GDP and interest rates. These risks are accounted for in the beta. The asymmetric risks on the other hand, are risk of which the potential downside impacts are greater than the potential upside effects, and are accounted for in the cash flows. Hence, the beta of the Lessard model is determined by multiplying the beta of the project with the country beta; Furthermore, Lessard uses the US market as the home benchmark. Consequently, the expected return on equity is; !" = !" ∗ !" Where; βa = Adjusted beta
βp = Project beta: correlation between the volatility of the project in the target market and the benchmark project in the home (US) market
βc = Country beta: correlation between the target market and the home (US) market
! !" = !"ℎ + !" ∗ [! !"ℎ − !"ℎ]
Where; E(Rx) = Expected return on equity investment in target country RFh = Home (US) market risk free rate
18 Although Lessard’s model enjoys some popularity, it has several shortcomings. First of all, Lessard assumes that it is possible to state a linear relationship between the stock return of an EM and the US market (Fuenzalida and Mongrut, 2010). Additionally, the model assumes that country risk can be accounted for in the stock beta, implying the assumption that country risk cannot be diversified (Damodaran, 2015; Fuenzalida and Mongrut, 2010). Both assumptions lack theoretical foundation. Moreover, although Lessard claims that asymmetric risk should be accounted for in the cash flows, he does not take into account that asymmetric risk might have an impact on symmetric risk. In other words, the model does not consider double counting (Pereiro, 2006). Finally, since this model partly relies on historical local data it will likely encounter the problem of limited data and its estimates will be backward looking. Godfrey and Espinosa model The Godfrey and Espinosa (1996) model (GEM), also known as the Bank of America model, identifies sovereign risk, business risk and currency risk as the three risks that impact investments in EMs. Just as Lessard, the GEM uses the US market as the home benchmark. Moreover, the sovereign risk can be accounted for by adjusting the RF with the BS between the target country and the home country; With regard to the business risk, Godfrey and Espinosa find that, compared to individual developed equity markets, correlation between individual emerging equity markets and a world equity portfolio are very low. Additionally, the volatility of EMs is relatively high. This results in EM betas that are lower than the betas of developed markets. To account for this Godfrey and Espinosa state; “In calculating the equity/business premium to be included in the discount rate, use an ‘adjusted’ beta that is equal to the ratio of an individual country’s equity volatility to that of the U.S. [home] market” (1996, p.88). It is likely that the volatility measure and the sovereign spread are both affected by fundamental economic and political developments. To avoid this, Godfrey and Espinosa adjust for double counting based on a study by Erb et al., (1996) stating that 40% of variation in equity volatility can be explained by variation in credit quality, meaning 60% is explained by volatility of the stock market, hence;
!"# = !"ℎ + !"# Where; RFa = Adjusted risk free rate
RFh = Home (US) market risk free rate
19 And the expected return on equity becomes; The largest disadvantage of this model is, as the authors themselves also indicate, that the model is only a rough estimation of the country risk, not making a distinction between different projects or industries (Estrada, 2007). Another criticism is that the use of the factor 0.6 is ad-hoc and indicates the assumption that it is acceptable to use a constant correction factor (Fuenzalida and Mongrut, 2010; Pereiro, 2001, 2006). Besides, this correction factor relies on historical information and might thus not be presentable for the current market situation (Sabal, 2004). Furthermore, a country risk premium is added to the RF without making any assumptions (Fuenzalida and Mongrut, 2010; Sabal, 2004). Goldman-Sachs Model The Goldman-Sachs model (GSM), developed by Mariscal and Hargis (1999), aims to overcome the problems of highly volatile data in EMs that are only available for a short time period. Hence, they developed a model that bases the determinants of discount rates on local and global variables. Just as in the GEM, the GSM model uses RV. Whereas Godfrey and Espinosa (1996) only calculate the volatility of the emerging stock market over the volatility of the home stock market, Mariscal and Hargis multiply this volatility ratio with the US market risk premium. Furthermore, instead of the 0.6 to adjust for double counting, Mariscal and Hargis subtract the correlation between USD denominated local stock and local sovereign bond returns. The final component of the adjusted beta is the beta of the target local company computed against the local stock market (βll). Hence;
!" = 0.6 ∗ !"!ℎ Where; βa = Adjusted beta
σx = Stock market volatility of target market σh = Stock market volatility of home (US) market
! !" = !"# + !" ∗ [(! !"ℎ − !"ℎ]
Where; E(Rx) = Expected return on equity investment in target country
E(RMh) = Expected market return on home (US) market
20 Furthermore, instead of only adding a CRP, the GSM adds an extra spread accounting for the company characteristics (Cc), which “can be thought of as a weighted average of the sovereign risk in the markets where the company derives its revenues” (Mariscal and Hargis, 1999, p.6). The COE therefore becomes;
For the calculation of Cc, Mariscal and Hargis use three global and five domestic indicators. The three global factors are global risk aversion, monetary policy tightness and commodity prices. Balance sheet, wealth, income statement, stability of cash flow and debt service history are the five domestic indicators; 𝐶𝑐 = 𝛼 + 𝛽1𝐵𝑆 + 𝛽2𝑊𝑒𝑎𝑙𝑡ℎ + 𝛽3𝐼𝑆 + 𝛽4𝑆𝑡𝑎𝑏𝑖𝑙𝑖𝑡𝑦𝐶𝐹 + 𝛽5𝐻𝑖𝑠𝑡𝑜𝑟𝑦𝐷𝑆 + 𝛽6𝐺𝑙𝑜𝑏𝑎𝑙𝑅𝐴 + 𝛽7𝑀𝑜𝑛𝑒𝑡𝑎𝑟𝑦𝑃𝑇 + 𝛽8𝐶𝑜𝑚𝑚𝑜𝑑𝑖𝑡𝑦𝑃 !"# = !"" ∗ !"#!" ∗ [! !"#$ − !"!"] ∗ (1 − !"#, !")
Where;βag = Adjusted Goldman-Sachs beta
βll = Equity beta of target company against the local market σx = Stock market volatility of target market σus = Stock market volatility of US market E(RMus) = Expected market return on US market RFus = US risk free rate Csl,bl = Correlation of target country’s local stock market and local bond market ! !" = !"#$ + !"# + !" + !"" ∗ !"#!" ∗ [! !"#$ − !"#$] ∗ (1 − !"#, !")
Where; E(Rx) = Expected return on target market
21 A positive features of this model is that it improves the GEM. By incorporating a double counting adjustment that is project based, Mariscal and Hargis respond to the criticism that the GEM assumes a constant correction factor and that the GEM does not account for difference between companies and industries (Estrada, 2007; Pereiro, 2006). A critical note from Harvey (2005) is that there is no theoretical foundation for this model. Nevertheless, he does not elaborate on this. The only criticism besides that is that not all countries have government bonds denominated in USD and that bond spreads only account for the probability of debt default (Fuenzalida and Mongrut, 2010; Harvey, 2005). These limitations are however a problem for most of the models.
Damodaran Model
22 Lambda is, just like the beta, scaled around one. The idea is that lambda measures the specific exposure of a company or project to a country’s risk. Damodaran describes three approaches on how to calculate lambda, the accounting earnings-, the revenue- and the market prices approach. Damodaran rules out the approach based on accounting earnings due to the shortcomings of accounting measures as a proxy for market value. The market prices approach is estimated as the slope between the stock returns and the bond returns of an EM. Limitations of this approach are large standard errors, and the requirement of liquid and widely traded bonds in a stable currency. Hence, the revenue approach is left;
Finally, in his model Damodaran does not only use CDSs, BSs or RVs as proxies for country risk. Instead, he prefers to use the BS and multiply it with the relative volatility of the local equity market as compared to the local bond market (RVL);
Where; E(Rx) = Expected market return on target market
RFg = Global (US) market risk free rate βg = Equity beta of target company against the global (US) market E(RMg) = Expected market return on global (US) market λ = Company specific exposure to country risk CRP = Country risk of target country λ = %!"#$, ! %!"#$, !"
23 Although the DM is theoretically sound given its assumptions, it does not cope with all the necessary issues (Fuenzalida and Mongrut, 2010). Who says, for example, that a firm’s exposure is only determined by its exports? Furthermore, the model requires input data that is in many cases private and requires the target country to have debt issued in USDs (or Euros) (Mongrut et al., 2010). Finally, the model requires that there should not be many episodes of financial crises as this would result in a very high RVL, and thus in very high COE. To overcome this problem, the DM uses what Walker (2003) calls the “Damodaran’s conjecture” which assumes a RVL equal to 1.50 (Damodaran, 2015). This means that, for this adjustment to be valid, the sensitivity of the stock return with respect of the bond yield spread should not differ statistically from 1.50 (Fuenzalida and Mongrut, 2010). Downside Risk Model Finally, the Downside Risk Model (DRM) by Estrada (2000, 2006) responds to the fact that “a number of recent empirical studies have shown that semi-deviation and downside risk measures can explain the cross-section of returns on US stocks and EMs” (2006, p.117), which is at odds with modern portfolio theory (Fuenzalida and Mongrut, 2010; Ogier et al., 2004). Modern portfolio uses the standard deviation to estimate total risk, and the beta to measure systematic risk. Nevertheless, Estrada claims that both measures can be questioned on theoretical, practical and empirical basis. The main argument for both proxies is that, although evidence indicates that risk is more costly on the downside, the standard deviation and beta give the same weight to upward- and downward swings (Estrada, 2000, 2006; Salomon and Grootveld, 2003). Estrada provides a downside version of the standard deviation (STDb), measuring the volatility below benchmark return B;
24 Either one of the measures can than be used to substitute the beta in the basic CAPM formula, giving;
With both measures Estrada shows that the downside risk has a larger effect than upside risk. The resulting ERPs differ in magnitude (STDb: 20-21% and Bb: 15-16%), however. Besides the comment that when using STDb the COE depends on its downside volatility relative to that of the market, whereas for the Bb this depends on its downside potential relative to that of the market, Estrada gives no indication with regard to which measures would be the best to use. Estrada (2000, p.1) states six reasons to support the DRM; “it is theoretically sound, it is very simple to implement, it can be applied both at the market level and at the company level, it is not based on subjective measures of risk, it can be tailored to any desired benchmark return, and it captures the downside risk investors want to avoid”. It should be stated, however, that the DRM is theoretically sound only given its assumptions. The fact that the model only uses one feature of returns in EMs points towards an incomplete approximation (Fuenzalida and Mongrut, 2010). Additionally, Estrada’s model is based on historical data. In spite of this, research has found that the DRM best fits the COE in Baltic countries (Balas, 2014). Furthermore, the findings of Estrada (2000) are consistent with the partial integration of EMs. Where; βb = Downside beta based on benchmark B N = Number of observations Rn = Individual stock returns target company for n number of observations RMn = Benchmark market stock returns for n number of observations BM = The benchmark return (M, 0.05 and 0) ! !" = !" + !" ∗ ! !" − !" and ! !" = !" + !" ∗ ! !" − !" where; E(Rx) = Expected market return on target market RF = Risk free rate
βb = Downside beta based on benchmark B σb = Semi-deviation based on benchmark B
25
Selection
An aspect of considerable importance for the selection of the models to be tested is that, as discussed, EMs are partially segmented. Specifically to the case study, limited research has been conducted regarding Russia’s state of segmentation. As the marginal investor in Russia can globally diversify (Ernst and Young, 2012), the correlation of the Russian stock market with the world stock market is what needs to be calculated in order to determine Russia’s state of segmentation.
26
Case Study
The majority of this research is built on the case study of Nord Stream AG (NS). NS entails two 1,224 kilometres long natural gas pipelines from Vyborg, Russia to Lubmin, Germany, which combined can transport 55 billion cubic meters of gas a year. To get to Lubmin, the pipelines pass through the exclusive economic zones of Russia, Finland, Sweden, Denmark and Germany, just as through the territorial waters of Russia, Denmark and Germany. The gas transported through the pipelines is monitored 365 days a year by the Control Centre, which, together with NS’s headquarter, is based in Zug, Switzerland.
NS was established in 2005 as an international consortium by five shareholders, for the planning, construction and subsequent operation of the two pipelines. The five shareholders of this Joint Venture (JV) are Gazprom (51%), E.ON SE (15.5%), Wintershall Holding GmbH (15,5%), N.V. Gasunie (9%) and Engie, formerly known as GDF SUEZ (9%). The construction of the first pipeline began in April 2010 and was completed in June 2011. The construction of the second pipeline commenced in May 2010, and was completed in April 2012. The total pipeline system required an investment of 7.4 billion Euros, financed for 30 percent through equity contributions by the shareholders, proportionate to their stakes in the JV. The remaining 70 percent was externally financed by banks and export credit agencies. As operator, NS does not own, buy or sell gas, but solely offers gas transportation capacities via its pipelines. The sole right to use this capacity has been granted to Gazprom Export, a 100% daughter operation of Gazprom, through a long-term contract. The five JV parties are therefore assured of a on forehand agreed return till 2032. Whereas capital intensive upfront investments secured by long-term contracts with (an)other “launching” customer(s) are rather unusual for most industries, it is common business for the oil, gas and electrics industry. Overall, it can thus be said that NS is a representative case for the energy industry.
The goal of NS is a direct and secure connection between the vast gas reserves in Russia and the energy markets in the European Union. The project has therefore been designated by the European Parliament and Council as being of ‘European interest’. Based on the European importance of NS, as well as the long-term contracts, it can thus be stated that the five shareholders are exposed to rather low levels of uncertainty.
27 implies that there is no possibility to diversify either physically or contractually. This is where country risk comes into play. Since the sole right to use the capacity of NS has been sold to Gazprom Export, a Russian based company, all revenues generated come from Russia. Although the construction and operational costs are less related to the country risk of Russia, the pipeline has already been constructed, and hence the revenue stream will be the most important determinant of the cash flows. Additionally, because the Russian government holds 50 percent of the shares of Gazprom, exposure to country risk is possibly even higher. Following Damodaran’s (2003) earlier mentioned statement that a company’s exposure should be determined by a company’s operations, it can be stated that the expected cash flows of NS are largely exposed to the country risk of Russia. The other countries involved are therefore of less interest. Furthermore, since the other countries involved can all be considered developed, their country risk is assumed zero and does therefore not need to be taken into account. From this point of view NS will be used to evaluate the different models that incorporate the CRP of Russia in the COE. Some additional notes important for later in this thesis are in place. First of all, although the capital structure of NS starts with a 70/30-debt/equity ratio, this structure changes. This is because NS is not really a corporation, but shows more similarities with a project; initial investments were required to set up NS, and afterwards the returns earned from the long-term contract with Gazprom Export are shared among the shareholders of NS, without requiring additional funding. With other words, debt slowly decreases till NS is 100 percent equity financed, constantly changing the D/E ratio. Secondly, taxes need to be paid to all countries NS passes through. Based on the tax rates NS needs to pay in each country, and the kilometres pipeline in each country, a blended effective tax rate has been calculated. Finally, the conservative assumption is made that the terminal value of NS after 2032 is small. This is because the long-term contract with Gazprom Export is designed in such a way that each shareholder earns back its investment during the contractual period. Even if Gazprom Export, or another shipper, would enter into a new contract with NS, it is not likely that the shipper in question would agree to pay the same return, as is the case in the current contract. Because the shareholders no longer need to earn back an investment, it is therefore expected that only maintenance and operational costs will be compensated.
28
Data & Methods
This thesis investigates two separate, but linked issues. The first one concerns the question whether or not a country risk premium exists, and if so, which country risk measure proxies this risk best? The second part of this thesis focuses on which model best predicts country risk. Before discussing which methodologies are used and why, the raw data and the necessary modifications are reviewed, so one knows exactly which inputs are used. The raw data for the two issues will be discussed together. The modifications and methods will be explained for each issue separately.Raw data
For the developed-, world- and the emerging equity markets, the Morgan Stanley Capital International (MSCI) Total Return Indices (gross dividend) are used (Table 4). All returns are denominated in USDs. As there is no generally accepted definition of what entails an EM, the selection of EMs is based on the countries that are included in the MSCI Emerging Market. Additionally, based on the Dow Jones Emerging Market Index the selection of EMs is supplemented with Argentina and Morocco to create a larger sample, which is desirable due to the short historical availability of EM data. The focus of this study lies on the behaviour of EMs themselves, and not necessarily in comparison with developed markets. Nonetheless, the World-, and United States MSCI are included, since these indices are generally considered the best estimate of risk free markets, and can therefore be used as benchmarks.
Following previous research (e.g. Erb et al, 1995; Donadelli and Prosperi, 2011), the 10 year USD denominated generic government bond yields and the 5 year USD denominated CDS are used. For Russia there is no available data of a 10year USD denominated government bond, therefore both the 7 and 15 year Russian USD denominated government bonds are included. All the above data is monthly and required from Bloomberg.
Testing the four models requires additional data. Because the NS project commenced in 2010, the time frame for the case study is 2010-2032. Nevertheless, to see if the estimates of the different models are consistent over time, additional data is collected for the period from 2001-2014, as longer time period allow for better detection of possible trends.
29 bonds. Since NS is a long-term investment, the appropriate measures used are bonds based (Donadelli and Prosperi, 2011).
Table 4
30 Specific to the models, for the GSM, the GDP per capita, GDP growth, International Reserves Assets, Total Gross External Debt, exports, and inflation of Russia are obtained. Also the US BB corporate bond spread, the Goldman Sachs Commodity Index, and the US Treasury Bill yield are acquired for the GSM. All the above is annually and obtained from Bloomberg. Finally, an economic report by the Rabobank (2013) on the Russian Crisis in 1998 is used for the data on Russia’s default history.
31 Following Damodaran (2003), different relative volatility (RV) rates are calculated. First, the annualized STD of each EM is divided by the annualized STD of the US, which results in the relative volatility using the US as a benchmark (RVU). The local relative volatility (RVL) is calculated by dividing the STD of a particular EM stock market by the STD of the same EM’s bond market. The RVL can than be multiplied with the BS to arrive at Damodaran’s ‘Default Spread + Relative Standard Deviation’ (RVLBS). As mentioned in the literature part, as a shortcut, Damodaran uses the ratio 1.5 instead of actually calculating the RVL. Comparing the RVLs calculated for the different EMs indicates that they often differ from 1.5 and per country. Furthermore, is his 2015 paper Damodaran states that the stock/bond market volatility was approximately 1.88, but that he continues to use the historical value of 1.5. For the above reasons, the shortcut is not deemed exact enough and the actual RVLs estimated will be used.
The CDS, BS and RVLBS values are all included with their current value, as well with a 1, 2, 3 and 6 month lag. The reason for this is that previous research has used different lags (e.g. Erb et al., 1995; Diamonte et al., 1996; Donadelli and Prosperi, 2011), but there is no clarity on which lag is the best. Hence, including different lags provides the possibility to compare the different outcomes, and perhaps even establish which shows superior predictive power.
Next, the downside beta (Bb) and the downside SD or semi-deviation (STDb) are estimated as explained in the Appendix of Estrada’s article (2006), using the formulas discussed in the “Models” section. As Estrada suggests, the mean (M), 0.05 and 0 are used as benchmarks, resulting in Bb-M, Bb-0.05, Bb-0, and STDb-M, STDb-0.05 and STDb-0. Since in this part of the study the focus lies on the behavior of countries and not on the behavior of companies, Bb and STDb are estimated for each country against the MSCI World.
Finally, due to the historically short sample periods available, the different country datasets are combined into a panel dataset so, scientifically and statistically speaking, better estimates can be obtained with the correlation and regression analyses. Because the data for the different countries is not altogether available from the same date, different panel datasets are created, each covering different time samples. As mentioned, all datasets run until 30 September 2015. The first and the Where;σ = Standard deviation country returns
N = Number of observations
xi = Individual country returns for ith number of observation
32 second panel datasets start in 1995 and 2006, and focus on the Bb, STDb, STD and RVL. The third and fourth dataset shift the focus towards the CDS and start in 2003 and 2009. The fifth until eighth panel look at the bond spread, of which the fifth and sixth start in 2011, and the seventh and eighth in 2014. The difference between the fifth and the sixth, and the seventh and the eight panels is that the later ones of each pair exclude Russia from the panel. The reason for this is that there are no 10 year Russian USD denominated government bonds available, leading to the inclusion of both the 7 and the 15 year version. The 2011 dataset includes only the 15 year bond, while the 2014 dataset includes both the 7 and the 15 year bond. The in- and exclusion of the bonds provides the
opportunity to see if they make a difference, possibly indicating the influence of the annuity of the bond. See Appendix 3 for the exact composition. Which model estimates country risk best? The different variables discussed above will be used again for the Russian-, US- and World Market. Besides that, the inputs for the standard COE and WACC formula, as well as the specific inputs per model are required. Input COE and WACC As mentioned, the RFs and RMs provided by Credit Suisse are long-term averages. For both the world and the US measures this means the average from 1900 onwards, and for the Russian measures the average from 1995 onwards. Clearly, the Russian measures are likely to be less stable due to the shorter time period, which should be kept in mind. Since the focus of this thesis lies on the effects of country risk, the RF and MR will be kept constant to assure that the differences per year are caused by differences in the betas and country risk measures. The data from Credit Suisse Global Investment Returns Yearbook 2015 are used so the average of Russia is based on the longest time period available. The cost of debt is kept constant over time for the same reason.
33 for Europe Fluxys Belgium, National Grid, CEZ AS, Red Electrica Corp., SNAM and Enagas SA are selected.
The next step, is the estimation of NS’s beta based on the returns of NS’s Russian and European peer group. To control for currency risk, the Russian betas are calculated using both Ruble and USD returns, and the European betas are calculated with Euro and USD returns. Using a rolling window, the betas are calculated by using the standard formula;
34 As the difference between the MSCI US and the S&P500, and the MSCI RUS and the MICEX are very small, I continue by using the betas against the US and Russian MSCI. This also assures consistency. To continue, the peer companies’ betas must be unlevered. Generally speaking, corporations assume that the debt is risk free, meaning the debt beta is zero and does not need to be taken into account when unlevering (Koller, et al., 2010). In the case of NS, the credit spread on the loan is small, therefore common practice is followed and the debt beta is assumed zero. A second assumption concerns the influences of taxes. According to Koller et al., (2010), taxes can be considered equal throughout time when the capital structure of a company stays approximately the same, meaning that the influence of taxes could also be excluded when unlevering betas. As discussed, this is not the case for NS. As a robustness check, the betas are unlevered both, with and without using taxes, using the formulas; Including taxes in the calculation results in slightly higher betas than when excluding taxes (Appendix 9A-C). Because the beta unlevered with the effect of tax is the better approach from a theoretical point of view, the focus will lie on betas unlevered with tax. In line with theory and practice, the median of the peer companies’ unlevered betas is taken, as this reduces in the influence of outliers. To stay consistent, relevering of the betas is also done using the effective tax rate calculated for NS. As a robustness check, the same is done while using the average unlevered beta. Although small, differences exist, indicating that the chosen approach does indeed protect against the influence of outliers. Furthermore, comparing the levered betas (Appendix 9A-C) shows that, although the median betas of the US and World market lie close to each other, the World market beta is somewhat larger. Additionally, the median Russian levered betas are by far the lowest. !" = 1 + 1 − !" ! ! !" & !" = [1 + ! !]!" Where;βe = Levered equity beta
βu = Unlevered equity beta
35 Model specific inputs Following the GSM, Russia’s Net External Debt is calculated by subtracting its International Reserves Assets from its Total Gross External Debt. The ‘Monetary Policy Tightness’ of the model is calculated by subtracting the US T-Bill yield from the US T-Bond Yield. Furthermore, the economic report by the Rabobank on the Russian Crisis in 1998 states that in august 1998 the Russian government announced a default on short term Treasury Bills and longer-dated Rouble denominated bonds. Therefore, a default dummy of 1 is applied. The BS for both the 7 and the 15 year Russian USD denominated government bonds compared to the US government bond has been estimated previously, just as the volatility of the Russian stock market relative to the volatility of the US stock market. Calculating the correlation of the Russian stock market with Russian bond market provides the last input for the GSM. Due to too much missing data for the 7year USD denominated Russian government bond, only the correlation 15year bond will be used.
For the DM it is necessary to calculate the export revenues of NS as percentage of its total revenues. Because the capacity of the pipelines is all contracted to Gazprom Export, one could say that all revenues are acquired in Russia. Claiming all revenues are made within Russia would thus mean ‘full’ exposure to Russia’s country risk. However, this does not seem entirely correct since Gazprom Exports uses the pipelines to export part of its gas. Therefore, the Gazprom Export’s gas export sales divided by its total gas revenues are used. This percentage is then divided by Russia’s annual exports as percentage of GDP per capita to arrive at the lambda.
With regard to the DRM, the required return index for NS is created based on the fixed dividend pay-out in March each year. Besides the Bbs and STDbs estimated earlier for the Russian market (DRMRUS) relative to the World and US market, Bb and STDb are also estimated using NS’s return index (DRMNS), using the World-, US-, and Russian markets as benchmarks.
Missing data
36 kept in the sample unless this leads to an entire year missing. If this is the case, the country in question will not be included, or included from a later date onwards.
37 of the standard errors is low and that using random effects is not appropriate. Finally, a redundancy fixed effects test confirmed that time fixed effects is the right approach for this data.
The majority of the data is not normally distributed. To secure consistency, all regressions are estimated first with the original measure, and afterwards with their LNs. Furthermore, as a robustness check, all the tests are performed using pooled panel regression as well. As an additional robustness check, both versions are re-estimated for samples where the upper and lower 5% is excluded to control for the influence of outliers. Unfortunately, Eviews does not offer a heteroscedasticity test for panel data. This should therefore be taken into account when interpreting the results.
Case Study: which model predicts the country risk the best?
Now all the data is ready, the four models can be applied to the case of NS. The first step is to estimate the COEs and the CRPs by following the different formulas. Because US data is often referred to as a proxy for the global market, the GCAPM, GSM and DM are estimated using both US and global data. By doing this it is possible to see how large the differences are, and what impact this assumption has. Finally, as the GSM and DM model focus on the BS and the RVLBS respectively, these measures will be used. Because the GCAPM does not specify the correct measure for the CRP, the CDS, BS, RVU and RVLBS will be applied. No time lags will be used as these are not indicated by the models. Specific to the DRM; because Estrada (2006) does not specify whether the Bb or the STDb is a better downside risk measure, both are estimated. Furthermore, Estrada does not position himself either about whether one should use the global, US or local market at a benchmark3. Thus, the DRMNS risk measures relative to all three markets are used. Because the return index of NS largely depends on dividend incomes, the index is not very volatile, which might not be a good approximation of the country’s risk. Therefore, the DRMRUS risk measures relative to the world and US market are also used. This means that the DRM is applied to case study in multiple ways; both the Bb and the STDb measures are estimated for the DRMNS and the DRMRUS sample, using different stock markets as benchmark.
Next to comparing the COEs and CRPS of the different models with each other and over time, a subsequent way to measure the outcomes is to compare the WACCs. The WACCs are calculated using the formula;
38 Normally, the D/V and E/V ratios used are target ratios. As discussed, in the case of NS, debt slowly decreases, implying a target ratio of 100% equity. It is exactly because of this that the WACCs are calculated per year. For the same reason, the actual ratios that are known due to the long-term contracts are the ones used.
As mentioned, the debt/equity ratio of NS fluctuates. Therefore, the WACC needs to be calculated per year until 2032 (Koller et al., 2010; Ogier et al., 2004). To achieve this, the assumption is made that after 2014 all parameters, except the debt/equity ratio, stay constant. The reasoning is that, for measures such as the CDS and the bond yield, the current situation is a better prediction of the future than an average. When, for example, the CDS has grown over the years, it does not seem correct to use an average to forecast future values, as this might underestimate the country risk when there is no reason to assume that the country risk will decrease in the future. When the WACCs are estimated, the EV can be estimated using Discounted Cash Flow Analysis (DCF). The theoretical approach to arrive at the correct EV is backward iteration (Koller et al., 2010; Ogier et al., 2004), starting the DCF in 2032 and working back to the year of valuation. Five different starting dates for valuation are used; 2010, 2011, 2012, 2013 and 2014. The resulting EV cannot be compared over the years, as they depend on the forecasted cash flows. What can be compared, are the WACCs. In order to do so, however, one WACC per model, per starting date is required. This can be done by estimating the implied WACC (e.g. Harvey, 2004). Using the ‘what if analysis, goal seek’ function in excel is used to estimate the WACC that arrives at an EV very close to the EV found when using backward iteration is estimated.
The final step is to determine which model provides the best outcome. In order to do this, the ‘correct’ WACC is required. Clearly, this actual WACC is not known. However, assuming that the market is right, the ‘correct’ EV can be approximated by applying the EBITDA multiples of peer
39 companies to NS’s EBITDA. Next, the same ‘what if analysis, goal seek’ function can be applied to arrive at a market implied WACC, which approximates the correct WACC of NS.
In this case, the choice of peer companies is less straightforward, because the median multiples of the two peer groups differ significantly, meaning that the Russian multiples are a lot lower. As multiples also provide an indication of the riskiness of a company, a possible explanation for these differences is that the EV of the Russian companies does include at least a part of the country’s risk. On the other hand, the European peers are better comparable. Including both the European and the Russian peer multiples allows for comparison.
Results
Does the CRP exist? CorrelationFirst of all, the line charts of the different MSCI’s can be found in Appendix 2A through to 2J. Appendix 3 presents the correlation of the individual EMs with the World market. A correlation analysis provides outcomes on a scale of -1(-100%) to 1(100%). Consequently, Appendix 4 shows that the positive correlation of EMs and the world market has increased over the years. In the 2005-2015 sample, only Colombia, Morocco and Qatar show a correlation below 50%. To continue, the robustness check of the risk measures shows that, although with different degrees, all correlations are positive. Bb-0 has the lowest correlations (>25%) with the other risk measures, whereas the correlations between the BSs and CDSs are the highest (aprox. 80%). Moving on, the correlation analyses of the risk measures with the local market return can be found in Appendix 4. Bb-M shows no clear sign. Whereas 1995, 2014 and 2014-Russia show a positive sign, the remaining years show negative signs. For Bb-0.05 and Bb-0 the correlation is negative; Bb-0.05 ranges from -9.2% till -16.5%, and Bb-0 is somewhat smaller ranging from -1.7% till -7.1%. The 2014 sample of Bb-0 is an exception, showing a positive correlation, although with very small values. The signs for the STDb show the same pattern; STDb-M shows mixed signals, whereas STDb-0,05 and STDb-0 show a clear negative correlation. The RVU also shows mixed signals, ranging from -7.7% till 4.6%.
CDS and the BS without a lag have a negative correlation with total returns, ranging from -8.9% till -40 19.4% and -9.1% till -20.2% respectively. With the exception of the 2014 sample excluding Russia, the relation of total return with the 1-, 2-, 3- and 6 month lag of both the CDS and the BS is positive, ranging from 2.0% till 13.7%. Although the correlations show a clear positive relationship, it is not clear how many months one should lag the CDS or BS to find the best estimator of positive returns on an EM’s stock market. The 2003 sample shows an increasing positive correlation from the CDS 1 month lag all the way until the 6month lag. In the 2009 sample, the order is exactly the opposite. In the 2011 sample including Russia, both the CDS and the BS show the largest correlation for the 1 month lag, declining till the 3 month lag, and than increasing again for the 6 month lag. The 2011 sample excluding Russia shows the highest correlation for the CDS and BS with a 6 month lag, whereas the 2014 example including Russia, the positive relation of the CDS and BS is the largest for the 1 month lag. The 2014 sample that excludes Russia contrast all the other samples, showing a negative relationship for all lags.
Regression
In Appendix 6A to H the results for the fixed time effects panel regression can be found. The results when using the original values and LNs of the risk measures are presented, as well as the results when in- and excluding the upper and lower 5%. In the majority of cases, taking the LN results in normal distribution or improvement of the data. When the LN does not improve the distribution of a measure, the original value provides better estimates, both speaking in terms of statistical significance and size of the error terms. The results of the better-distributed measures will therefore be discussed. Additionally, since it is generally assumed that longer time periods provide better estimates, the focus per risk proxy will lie on longest data set available. Only when the shorter time period data result in significantly different outcomes, this will be discussed. Finally, the pooled panel regression results are not reported due to limited space, nevertheless, about 85% of pooled regression results showed the same signs.
Before continuing, an informative note for the interpretation of the outcomes is necessary (Brooks, 2008); the dependent variable (total return) is always used as a LN. Therefore, when using independent variables in their original value, the model becomes log-linear4. This means that 1 unit
increase in independent variable x results in a 100 * β2% increase in dependent variable Yt. When using the independent variable as a LN, the model becomes a double log5, meaning that a 1%
increase in the independent variable x results in an β2% increase of dependent variable Yt. This thus
4 ln(Yt) = α1 + β2x
