1
Valuation and Country Risk: Empirical Evidence regarding the
Country Risk Premium
Master Thesis
MSc Business Administration
Specialization: Finance
University of Groningen
Faculty of Economics and Business
Author: Huub Thoben Student Number: S1540122
2
Valuation and Country Risk: Empirical Evidence regarding the
Country Risk Premium
Huub Thoben
1ABSTRACT
This study examines the influence of country specific risk factors on stock returns, using the Country Risk Premium (CRP) as proposed by Damodaran, as measure for country risk. I focus on two parts. The first concerns the ability of four different CRP measures to embrace factors of country risk as proposed in earlier literature. It seems that CRP measures are unable to capture all country risk specific factors as specified by the International Country Risk Guide (ICRG) for 63 countries in the period 2005-2010. A second set of tests looks at the additional explanatory power of the CRP in a CAPM and Three Factor Model framework. It is concluded that a Local CAPM or Three Factor Model approach works better than a Global variant with addition of a CRP. Critics on the CRP on basis of theoretical objections from earlier studies are now expanded by critics on empirical grounds.
JEL classification: G12, G15
Key words: country risk, country risk premium, valuation, emerging markets, capital asset pricing model
1
3
1. Introduction
This paper covers the subject of country risk premiums. The country risk premium (CRP) is an additional risk premium that can be imposed when equities in developing or emerging markets are valued. The discount rate in a particular valuation, which is derived from frameworks like the Capital Asset Pricing Model (CAPM), is raised by an additional premium for country risk (Damodaran, 2011). Damodaran argues that the CRP should reflect the additional risk that is associated with investing in companies in these markets. Macroeconomic factors like economic turmoil and exchange rate risk or unstable political systems can cause investors to be careful to invest in developing or unstable markets like Asia, Eastern Europe, Africa or Latin America. Because it is assumed that more risk is involved with investing in these developing markets, it can be argued that investments in those countries should have a higher return (Damodaran, 2011; Harvey, 1994). Aswath Damodaran is one of the scientists that argue strongly in favor of imposing a CRP for the markets mentioned above. In his article (2003, later adjusted versions up to 2011) he argues that a CRP on equities is needed to account for the additional risk, if this country risk cannot be diversified away, when it cannot be adjusted for by adjusting cash flows and when it is not measured by the Global CAPM.
However, in their critical paper on the CRP, Kruschwitz et al. (2010) argue that imposing a CRP is theoretically nor empirically supported. Their main discussion focuses on the validity of financial models like the CAPM and the violations of these models that arise when a CRP is imposed. Next to this, Kruschwitz et al. (2010) criticize the lack of empirical evidence in the argumentation of Damodaran. However, they are not able to provide any empirical argumentation themselves; the critique mainly focuses on the theoretical problems with the approach of Damodaran. These problems are further discussed in Section 2.
In this paper I critically analyze the CRP concept, by evaluating the concept in the light of different theories that are common in the financial world, like the CAPM and IAPM, Arbitrage Pricing Theory (APT), the Three Factor Model by Fama and French (1992) and the extended Four Factor Model by Carhart (1997). Next to this, I test the relevance of the CRP concept in pricing equities.
The next two research question are the main focus of this study:
Do the different measures of the CRP embrace concepts that are defined as country risk variables by previous
research?
4
Do the measures of CRP improve the explanatory power of a Global and Local CAPM and Three Factor Model
framework?
There is a debate in existing literature whether financial markets have to be considered as (partially) integrated or segmented. Different authors like Harvey (1994), Bekaert and Harvey (1995) and Serra (2000) find that markets are at least not completely integrated. The different authors conclude that the amount of integration with world markets differs across countries and over time, where a Global approach to the CAPM implies totally integrated markets. Damodaran (2011) states, that a Local approach to the CAPM is not appropriate for emerging markets. This is because of the general lack of data for these markets. He prefers a Global approach to the CAPM, but with the addition of a CRP for emerging markets. This second question addresses the problem of which model to use and the importance and relevance of adding a CRP to a Global framework.
To answer the first research question, data for the independent variables (country risk variables) is collected from the PRS Group that provides risk profiles on countries using a set of 22 macroeconomic variables. These profiles are comprised into three main Risk Factors, namely the Economic, Financial and Political Risk Factor. The last factor is not available for free, so instead of using data from the PRS group, data from the Worldwide Governance Indicators (WGI) project of the World Bank is collected. The WGI provide an estimate on six variables for Political and Governance Risk. The relevant dependent variables, the CRP measures, are derived from papers of Damodaran (2011), Cosset and Roy (1991) and Domowitz et al. (1998). Then, panel least squares regressions are used to test if the CRP measures are capable of capturing country risk characteristics for the total sample of companies and two subsamples. Tests show that the measures Damodaran proposes do not have similar results. The Government Bond Spread seems to be best able to capture country risk variables, followed by the Credit Default Swap (CDS) Spread. When the regressions are run again for emerging and developed markets separately, it seems that the CRP measure works better for emerging markets than for developed markets, since the R-squared values are larger in the group of emerging markets. Next, the CRP measures are added to a Global CAPM and Three Factor framework. It seems that R-squared values rise when a CRP is added, however this increase is not significant. A relative large and significant underperformance relative to the market is seen within the emerging markets. A result opposing the theory of Damodaran (2011) is that the Local framework has much higher R-squared values than the Global framework. Thus, the Local CAPM and Three Factor Model seem to be a better fit to the data than a Global framework with the addition of a CRP, even when the emerging markets are considered separately.
5
methodology used to test the usage of the CRP, where Section IV gives an overview of the data that is used for the different tests. In Section V the results are presented and discussed. Section VI concludes.
2. Literature Background
This paper is about country risk. However, in the literature there is no one single definition of country risk available. Country risk is described generally by Damodaran (2011) and others as the additional risk that is accompanied with an equity investment in a particular developing (or emerging) country versus another developed country. Meldrum (2000) defines country risk as particular imbalances that increase the risk of a shortfall in the expected return of a cross-border investment. Authors like Bilson et al. (2001), Feder and Uy (1985) and Diamonte et al. (1996) refer to outside sources when defining country risk. Country risk is stated as the potential, extra risks of international business operations by the International Country Risk Guide of the PRS Group. Euromoney defines country risk as the investment risk that is related to the political and economic stability of a country. Different definitions of country risk are available, but these definitions are all embracing terms that state that additional or extra risk related to investments abroad and influences of these risks on returns are part of the definition of country risk. Because this study focuses on the methods of Damodaran (2011) to deal with country risk, the definition of Damodaran is leading in this paper.
6
He states that country risk can be split up into two components: Economic and Political Risk. Vij (2005) uses seven variables to replicate the country risk rating of Euromoney. Again, variables related to GNP, Debt service, exports and Political Risk are considered to be relevant. Vij concludes that the Political Risk rating is by far the most important factor that influences the country risk rating. Furthermore, GNP per capita is important in explaining risk ratings, but that is contrary to conclusions by Cosset and Roy (1991), who argue it is not the most important factor.
Country risk is a hard to define concept. It is not clear which particular factors are the most important in determining country risk; as stated earlier, there is no single definition available. If country risk is so ambiguously defined, should it be taken into account in valuations of equity? It is questioned by Kruschwitz et al. (2010) whether country risk should be taken into account and how this should be done. Imposing a CRP, as proposed by Damodaran (2011) could be a solution to this question. He states that three arguments are used against imposing a CRP. In his discussion he tries to refute these arguments and explain why the CRP is a correct measure for country risk. The following discussion deals with these three arguments and the considerations of Damodaran regarding the critique.
The first argument relates to the way in which country risk has to be dealt with in valuations. According to Koller et al. (2010), Damodaran (2011) and Pereiro (2002) there are in fact two ways to deal with country risk: • The first way is to account for the additional risk in the denominator of the valuation formula. This means
that the discount rate is adjusted with a CRP.
• The second method is to adjust the estimated cash flows from a firm with measures of country risk. In the literature there is a debate whether the first or second method should be preferred. James and Koller (2000), among others, argue that the second method of adjusting cash flows should be preferred. According to James and Koller, adjusting the discount rate with a CRP is a highly arbitrary procedure and adjusting cash flows is a more accurate way of dealing with country risk. According to Damodaran (2011), this argument is not valid. He argues that expected cash flows, when scenario analysis is used to allow for possible poor outcomes, are not a risk adjusted cash flow. These expected cash flows should be further adjusted for the risk aversion associated with the average investor, by calculating certainty equivalents of these cash flows. When calculating these certainty equivalents, the risk associated with a particular country has to be calculated in the form of a CRP. Therefore, according to Damodaran, also the procedure of adjusting cash flows needs an estimate of a CRP.
7
(market) risk. Consequently, if you want to include a CRP in the cost of capital, the country risk that is measured with this CRP should be non-diversifiable. It is debated in literature whether this is the case. According to Damodaran (2011), the ability to diversify away country risk depends on the amount of global diversification in the portfolio. Damodaran argues it can be questioned if the marginal investor is globally diversified, based on conclusions of Stulz (1999). Stulz finds that there is a home bias apparent among investors; the average investor prefers investing in his own country or region. Even if the marginal investor is globally diversified it can be questioned if country risk can be completely diversified away. Damodaran points here to the contagion effect, where sudden changes in one market affect the environment of another market. This statement is supported by findings of Kaminsky and Schmukler (2002) who state that contagion effects are particularly present within specific regions. James and Koller (2000) and Kruschwitz et al. (2010) question the statement of Damodaran by stating that many of the risks associated with countries are in fact diversifiable. Kruschwitz et al. do acknowledge that only non-diversifiable risk has to be taken into account in the cost of capital and are therefore against imposing corrections for country risk. As pointed out by Kruschwitz et al., a main lack of Damodaran’s line of argumentation is that he does not provide any evidence of the statement that country risk is not diversifiable. Damodaran simply argues that markets have to be ‘uncorrelated’ to reach diversification. However, modern portfolio theory by Markowitz (1952), states that diversification is possible even when securities are correlated (positively or negatively)
The third argument relates to other asset pricing models that are available from literature. Kruschwitz et al. (2010) argue that adding a CRP to the discount rate seriously violates classic asset pricing models. In the 20th century, academics developed different models to calculate expected returns of securities. During the 1960’s, Sharpe (1964) and Lintner (1965) developed the Capital Asset Pricing Model (CAPM). The CAPM is by academics and practitioners considered to be an equilibrium model. According to this theory the return of a security equals, ) ( , , , ,t ft j mt ft j R R R R = +β − (1)
In this formula
R
f,t equals the risk free rate.R
m,t is the market return, when calculated from the past8
The Arbitrage Pricing Theory is developed by Ross (1976). In contrast to the CAPM, the APT is a multi-factor model. The theory suggests that the expected return of a stock can be calculated as a linear function of factors that represent for instance macro-economic variables or market indices. The CAPM is essentially a one-factor variant of the APT. One could say that the inclusion of a CRP can be seen as a form of a multi-factor model. However, as Kruschwitz et al. (2010) and Ross (1976) state, it is not possible in the APT to give the factors of an APT formula a name or identity, because they argue that these identities are different across economies and over time. Kruschwitz et al. therefore argue that you cannot just interpret one or more of the factors in an APT regression as a country risk premium. For both the CAPM and the APT it is proven that they are linear combinations of variables. Damodaran (2011) assumes that the CRP also has a linear structure but he does not prove this in any way.
Fama and French (1992) introduce the Three Factor Model. They link the rate of return of different portfolios to three factors. The model is an extension of the CAPM, where the other two factors (related to Size and Book-To-Market Ratio) explain the variation of stock classes that are found to do better than the market on average. This model is extended by Carhart (1997) to a Four Factor Model that captures the momentum factor as an additional variable. The Three Factor Model shows a linear relationship between portfolio return and the three factors. However, in contrast to APT these factors are named and labeled. Kruschwitz et al. (2010) do not consider this model in their critique paper. Fama and French do show that it is possible to name different variables that influence stock returns.
The debate on whether to impose a CRP or not is ongoing (Damodaran, 2011; Kruschwitz et al., 2010) and there are arguments in advance of and against imposing a CRP. The main arguments by Damodaran are that adjusting cash flows do not take the risk aversion of the investor into account and that country risk is not diversifiable. Kruschwitz et al. (2010) argue that the argumentation of Damodaran is not scientifically correct and they explicitly provide theoretical evidence that supports their arguments.
The next question of how to evaluate country risk and how to calculate a CRP is not so straightforward either. It is clear from literature (Sharpe, 1964 and Lintner, 1965), that there is one single definition of the CAPM. Kruschwitz et al. (2010) do admit that it can be difficult to estimate the market portfolio, something that is also discussed by Roll (1978). However, they argue there is at least a clear definition of this concept. Regarding country risk and the CRP there is no clear definition of how to come up with an estimate.
Damodaran (2011) argues that there are several ways to add the CRP to a CAPM framework. The basic approach is to calculate the return of a particular security with the CAPM and then add a CRP to this return to adjust for the risk of a particular country where the security is traded, so
CRP R
R R
9
where Re,tis the return on equity at time t, Rf,US,t is the risk free rate (return on US treasury bills) at time t,
β
is a measure of risk exposure, RGlobalMarket,t is the return on an index that comprises all available assets in the world at time t. The CRP is the risk premium associated with country risk in a particular country. The model above has large similarities with the model of Adler and Dumas (1983) and Solnik (1973). The International Asset Pricing Model (IAPM) of Adler and Dumas (1983) argues that investors are generally able to invest in all available securities worldwide. However, investors from different regions or zones (regions or zones are defined as areas with similar purchasing power units or currencies) have heterogeneous believes about real returns because of purchasing power deviations. Therefore each region has a different short-term bond which is used in portfolio construction. If the CRP in (2) equals the difference between the US Treasury Bill and a regional short term bond, the model is quite similar to the IAPM approach of Adler and Dumas (1983).In literature, all kinds of varieties on (2) exist. In Ogier, Rugman and Spicer (2004), they are as follows:
CRP R
R R
Re,t = f,Global,t+β*( GlobalMarket,t− f,Global,t)+ (3)
This variant is labeled Global CAPM approach in this study. The difference between (2) and (3) lies in the assumption of the Risk Free Rate. In the case of (3), the Risk Free Rate for all investors is assumed to be equal. Since this model is generally implemented by taking differences in currency (i.e. inflation) and interest rates into account, the model is again very close to an IAPM approach.2
) (
* , , ,
, ,
,t f Foreignt ForeignMarkett f Foreignt
e R R R
R = +β − (4)
The model in (4) will be interpreted as the Local CAPM approach in this study. One of the main assumptions of Adler and Dumas (1983) is that there is at least partial world market integration. This justifies the choice of stating that securities’ returns,Re,t, should be compared with an Index that comprises all securities available in the world, RGlobalMarket,t, as is in (3). Equation (4) assumes that world markets are strictly separated. Security returns should be compared to a local market index, RForeignMarket,t, where the term local can be defined as a market with its own currency and its own inflation rates, a similar definition as is used by Adler and Dumas (1983). There is no CRP added in equation 4, because local data is used. The local Risk Free Rate (which is then the local government bond or a similar instrument) and the local market index return are
2 It has to be noted here that Damodaran (2011) argues that the US Treasury Bill can generally be seen as the
10
expected to contain the country risk elements. A problem with this model is that there is often not sufficient data available to come up with reliable estimates for RForeignMarket,t and Rf,Foreign,t (Damodaran, 2011).
CRP R
R R
Re,t = f,Home,t+β*( HomeMarket,t− f,Home,t)+ (5)
This is a model where the characteristics of the country of the particular investor are taken into account. The model is essentially the same as the Global CAPM from (3). However, the global measures for the risk free rate and the market premium are replaced by measures from the investor’s home country. Merely, this is done to deal with the home bias that is apparent under investors, according to Damodaran (2011) and Stulz (1999).
Next to the previous equations, Godfrey and Espinosa (1996) propose a modification of the traditional CAPM to cope with country risk. In their model, the relative volatility approach, they propose to adjust the CAPM in two ways: ad CreditSpre R R R Ri,t = f,US,t+βa,i*( GlobalMarket,t − f,US,t)+ (6) Firstly, the credit spread of a particular country, as measured by the spread over sovereign debt instruments, is added to the equation. This is to reflect the credit quality of a particular country. As a second step, Godfrey and Espinosa adjust the β. The β becomes equal to the volatility of the country under investigation divided by the volatility of the US equity market, multiplied by 0.6 to deal with potential ‘double counting’ of risks. This number gives the β for a particular country’s equity market. Next to the problem that the 0.6 multiplier is arbitrarily chosen, the major problem with this model is that it considers the return in a specific country instead of the return on one single security.
All the above models (except the Local CAPM model of (4) and the relative volatility approach by Godfrey and Espinosa as in (6)) include a CRP. Damodaran (2011), Cosset and Roy (1991) and Domowitz et al. (1998) suggest different methods to come up with an estimate of the CRP.
11
companies traded on the Athens Stock Exchange are as close to default as the Greek bonds are supposed to be.
• The Country Credit Rating (CCR) as given by a rating agency like Moody’s. Damodaran himself points out that a drawback of this method is that rating agencies often lag the market. As a result, numbers are compared that do not correspond to the same time period.
• The last method to come up with a CRP is to take the relative standard deviation of a particular developing country with respect to a developed market (Damodaran, 2011 takes the US as developed market). Next, the relative volatility number has to be multiplied with the Market Risk Premium of the developed market to get the Market Risk Premium of the developing country. The CRP is then the difference between the Market Risk Premium of the developing country minus the Market Risk Premium of the developed country (the US), so
US US countryX MRP CRP=( −1)× σ σ (7)
where σ equals the standard deviation. The main problem with this approach is that you need an estimate of the standard deviation of an emerging market. As mentioned before, Damodaran (2011) states that it is hard to obtain a reliable estimate of market returns (and therefore also standard deviations) in emerging markets, because there is less data available. This lack of data in emerging markets was precisely one of the reasons that the Local CAPM does not work and a CRP measure is needed.
Essentially, it can be seen that the relative volatility approach from Godfrey and Espinosa (1996) is a mix between the different measures provided above. They use both the relative volatility and the credit spread in their assessment of country risk.
The equations (2), (3) and (5) imply that every firm in a specific country has the same exposure to country risk. This can however be debated, something Damodaran points out himself. Damodaran (2011) indicates that there are three different ways of dealing with country risk exposure per firm. The first method adds the CRP to the CAPM equation in the way (2), (3) and (5) indicate: all firms have the same exposure to the country risk variable. The second method indicates that the CRP depends on the β of a firm. Thus, a firm with a higher exposure to market risk (a higher beta), has a higher exposure to country risk, so
) ) (( * , , , , , , R R R CRP
12
One of the major problems Kruschwitz et al. (2010) point out is that the assumptions of the CAPM are explicitly violated. Within this second method, Damodaran has a different market return for different countries, whereas the CAPM argues that there is one single market return available. The third method (Lambda Method) is to come up with a separate measure of country risk per firm by multiplying the CRP with a λ, which differs for every company depending on, for instance, the amount of sales or production in a specific country. So,
) ( * ) ( * , , , , , , R R R CRP
Ret = f Globalt+β GlobalMarkett− f Globalt +λ (9)3
According to Kruschwitz et al. (2010), the Lambda Method has a major problem in its definition for Lambda. As stated earlier, within the CAPM, there is a clear definition of the β. Damodaran does not give a clear definition of the λ; the investor is left in the dark of which measure to use for λ. Damodaran points out that the λ should be structured in the same way as the β, so with values around 1. Furthermore, he gives numerous options of what accounting measure should be used to define lambda, for instance turnover or production, but does not give a sound theoretical foundation for either one of these measures (Kruschwitz et al. 2010).
There is a vast amount of empirical evidence available on the influence of country specific variables on stock returns. Next to the earlier discussed empirical research on the relation between several political and economic factors and country risk ratings, research has been conducted to investigate the relation between country level characteristics and equity returns. Moreover, studies have published about the lack of integration of world markets, a crucial assumption in the work of Damodaran (2011). The much cited paper by Harvey (1994) explains that the higher returns in emerging markets are not explained by the asset pricing theory. A reason Harvey mentions is that the assumption of complete integration of world capital markets is violated. This is one of the crucial assumptions of the CAPM. Bekaert and Harvey (1995) elaborate further on this statement. They present a model where the excess return of a security in a particular country depends on the covariance with the world market and the covariance with the local market. This could be seen as a combination of (3) and (4), both a local and global index is used in their analysis. Within their dataset of twelve emerging markets, the assumption of total integration is rejected for all countries. In later work, Estrada (2002) investigates the betas of 27 different emerging and developing countries. Estrada uses a Global approach as in (3), while adding local risk variables. He finds that the average beta against the MSCI is 0.92, which would indicate that the average emerging and developing market has a
3 The global measures in (7) and (8) can be changed with ‘home’ measures, just as in (5). As stated earlier, using
13
beta lower than 1. This implies that some developed markets should have a higher beta than 1, which supports the findings of Damodaran (2011). This result is supportive to conclusions of Harvey (1994) that markets are not fully integrated. In later papers, Erb, Harvey and Viskanta (1995, 1996b) elaborate on the assumption of integrated world markets. They find that the Country Credit Risk rating is a measure that can predict future stock returns. However, they find a positive correlation between credit risk and beta. This indicates that countries with a high credit risk have a low beta, which they indicate as counterintuitive. Kaminsky and Schmukler (2002) have similar findings in their study and conclude that higher credit ratings yield lower returns. An additional result is that a change in outlook also implies changes in returns. Moreover, they find that rate changes show contagion effects. This means that the change of the credit rating in one country can have an impact on returns in other (often closely associated) countries. Previous research thus shows that markets are not fully integrated and that country specific factors can influence returns. The studies by Erb, Harvey and Viskanta (1995 and 1996b) and Kaminsky and Schmukler (2002) follow a different approach than presented by Damodaran (2011). In their analysis, the Global or Local CAPM is not considered; instead they only analyze country specific factors. A main assumption of Damodaran’s theory (2011) on country risk is the lack of integration of world markets. This assumption can be supported by previous research.
14
(2001) find ambiguous results for different markets. In their model, where the expected return of a country depends on local as well as global factors, they find that for different countries, different factors are relevant in explaining returns. Since the global factor is set equal to the Global Market Index, the model can be interpreted as a Global CAPM approach with the addition of local factors. The exchange rate is the only measure that is pretty consistent among all countries, but it is not so clear that for all countries the country specific factors are as important as stated earlier.
In research related to governance procedures and country differences, James and Koller (2000) mention that poor investor rights should also be related to country risk. Indeed, Klapper and Love (2004) find that investor rights are related to governance practices in particular countries. These governance practices are then significantly associated with firm performance and valuation in the market. In countries with weaker legal systems, the relationship between firm level governance and performance is even more pronounced. These findings are supported by conclusions of Hail and Leuz (2006), who also find that countries with stronger regulation and more extensive disclosure have lower cost of capital. The following Table 1 gives a general overview of the related literature and the main conclusions drawn from testing. Articles are sorted on the method used with respect to (3) and (4) in this paper.
Table 1: Overview Empirical Literature on Country Risk research
Authors Period Countries Model Methods Variables/Data Important results
Bekaert and Harvey (1995) Sep. 1979 – March 1995 47 equity markets focus on 12 emerging markets
The model implies that the expected excess return of a security in country i depends on the covariance with the world market (CAPM) and the (co)variance with local market. This depends on the size of φ which is a measure for the integration with world markets.
To infer φ from the data the Hamilton model (1990) is used. Moreover, φ is a time-varying measure, so integration can differ across time.
The market returns are gathered from the MSCI and IFC (World Bank).
The model is highly significant. For 6 countries the constant transition probabilities are rejected, mainly due to data from the local market. Bilson et al (2001) Feb. 1985 – Dec. 1997 20 emerging markets
The authors use the following model:
where the return is dependent on N global factors and K local factors.
Test for commonality of markets. Start with a PCA, where 11 variables are being reviewed. Global Market index with local factors added to the equation
N is supposed to be 1, the World Market Index. K equals 4, namely money supply, goods prices, real activity and exchange rates..
Ambiguous results for different markets for the model with 5 variables. When conducting the PCA with 4 components, the model appears stronger.
Serra (2000) Jan. 1990 – Dec. 1996
26 emerging markets
The return of a particular stock in a particular country and industry is evaluated using:
where I and C are dummy variables for specific countries and industries under consideration. α equals the global level return for a particular period (Global CAPM).
Simple regressions are performed, where countries and industries act as dummy variables.
Weekly data from Emerging Markets Data Base. SIC codes are used to discriminate between industries.
Emerging markets’ indices are driven by country factors and cross-market correlation does not seem to be affected by the industrial composition of indices. There are no regional effects.
Estrada (2000)
28 emerging markets
The downside risk approach is proposed:
In this model only downside risk is taken into account when measuring risk. B is equal to a benchmark return.
Simple linear regressions on all nine risk variables. Estimate three models with significant variables StDev and Downside Risk compared to beta.
Data from MSCI, monthly returns are used. Nine risk measures are used: Beta, StDev, Idiosyncratic risk, Size and Downside risk (5 different measures)
In the risk-return model, Beta has much lower returns then StDev and Downside risk approach. Results are consistent with assumption of partially integrated markets Diamonte et al. (1996) Jan. 1985 – June 1995 45 markets; 21 developed and 24 emerging
- From three portfolios per quarter based
on downgrade or upgrade in political risk.
Political Risk measure from ICRG.
Difference in return of up and down grade portfolio 11% in emerging and 2.5% in developed markets. Erb et al. (1995) March 1980 – Dec. 1993 40 national equity markets
- Four CCR portfolios are formed, ranging
from high to low rating. Portfolios are rebalanced monthly and a half year lag is used to ensure full information exposure.
Data from the MSCI and IFC. CCR is the Country Credit Rating gathered from Institutional Investor.
The CCR is able to distinguish between performances of portfolios. Next to this the CCR is correlated with volatility and is able to predict future volatility.
Erb et al. (1996a)
Jan. 1984 – July 1995
117 countries The model is as follows:
where the return of a change portfolio is regressed to different lagged variables that represent macro-economic risk indicators.
A cross sectional and time-series analysis is done through a fixed effects model.
Three measures of risk according to the ICRG, a composite index of the three and the CCR from S&P and Moody’s.
All the variables from IRCG are positively correlated with returns. Especially the financial and the composite index have a strong significant influence on returns. Erb et al. (1996b) Oct. 1979- Sep. 1995 135 developed and emerging
markets where the return of a country’s stock market is regressed against the country credit rating.
A log linear regression analysis Data from the MSCI and IFC.
CCR is the CCR from Institutional Investor.
The slope coefficient is significantly different from 0. Negative relation between the CCR and the return. Kaminsky and Schmukler (2002) Jan 1990 – June 2000 16 emerging markets (which suffered from crises in 1990s)
The reaction of country premiums and stock returns to changes in ratings, outlook and US interest rates is studied.
Next to this, two regressions are performed to test contagion effects between countries w.r.t. credit rating and US interest rates.
Pooled panel regressions are used to estimate the equations.
Credit Ratings from Moody’s, S&P and Fitch. 244 changes in ratings and outlooks are considered.
Rating is significant, higher rating yields lower returns. Higher US rates increase returns. There is evidence of contagion effects, especially within regions. Hail and
Leuz (2006)
Jan. 1992 – Dec. 2001
40 countries In the model the implied cost of capital is regressed against all sets of country and control variables.
OLS regressions are performed. All kinds of robustness checks are performed with respect to macroeconomic variables.
Stock returns are from I/B/E/S. Data on control variables from DataStream and data for regulation and legal institutions is from the LaPorta, Lopez and Schleifer (2006) database.
Countries with more extensive disclosure and stronger regulation have a lower cost of capital. The control variables are mostly significant, total model describes almost 60% of total cross-sectional differences. Klapper and Love (2004) 2000 - 2001 14 emerging markets and 374 companies
First a measure of firm level governance is created by. Nest they study the relationship between firm-level governance, the country level legal environment, and firm performance by
Qf = a + b1(Gov) + b2(Eff) + b3(Gov x Eff) + ε Where Qf equals Tobin’s Q.
OLS regressions, with interaction effects between firm governance and country legal systems.
Firm level governance data by CLSA , three measures of country level legal efficacy. Worldscope data for returns.
16
Fama (1991) explains that the selection of the initial factors that are associated with country risk is subject to criticism on the grounds of subjectivity and the arbitrary nature of the selection process. He states that this is an unavoidable problem associated with this area of research. The convenience of having one estimate (the CRP) for country risk is therefore very high. This paper will investigate if the CRP is a measure for country risk. Variables that are used by previously mentioned authors will be regressed on measures for the CRP that are proposed by Damodaran (2011). Next, the measures of CRP will be added to a Global and Local CAPM or Three Factor Model framework to test its relevance in pricing equities. This is an addition to earlier research done by, among others, Bilson et al. (2001) who regressed global as well as local factors on equity returns. This way, the discussion on the relevance of CRP can also focus on empirical evidence rather than only underlying theoretical implications as is done by Kruschwitz et al. (2010).
With respect to the two research questions from the Introduction section, general hypotheses can be constructed which will be worked out in detail in the following section. Recall that the first research question relates to the ability of the CRP measures of Damodaran to capture country risk variables from earlier studies. Based on the above analysis of the influence of country specific factors on stock returns and the statement by Damodaran (2011) that the CRP is a measure for country risk, the following general hypothesis is expected to hold,
General H1: Country specific factors, from earlier literature, that influence stock prices have significant
explanatory power over the CRP measures as proposed by Damodaran.
The second research question relates to the explanatory power of the CRP in a Global and Local CAPM framework. Earlier studies by among others Erb, Harvey and Viskanta (1995, 1996b) and Serra (2000) conclude that country specific variables have significant explanatory power in regression on stock returns in a Global CAPM setting as in (3). Next to this, Damodaran (2011) argues that a Local CAPM approach should not lead to better estimates, because a general lack of data. The general hypothesis regarding research question 2 is as follows.
General H2: The additional CRP to a Global CAPM or Three Factor Model leads to better estimation results
and a better fit to the data. Subsequently, the Local CAPM approach will not perform better than the Global
approach.
17
explosive.4 To profit from investments in these high growth markets, it is crucial to know associated risks. Damodaran advocates the CRP concept for dealing with this country risk concept. The relevance and use of a CRP on equities in emerging markets is doubted by different academics. The main problems lie in the theoretical evidence that is not provided, according to Kruschwitz et al. (2010). Next to this, Damodaran does not prove empirically that the CRP is justified. Therefore this paper contributes to existing literature by addressing this empirical lack of evidence on CRP and I try to fill the gap between Damodaran’s propositions and the critique by Kruschwitz et al. (2010).
3. Methodology
This section is split up in two parts. The first part elaborates on Damodaran’s proposed measures for the CRP and to which extent they embrace country risk definitions. The second part focuses on the explanatory power of CRP measures in extended versions of the CAPM and Three Factor Model for explaining differences in stock returns.
3.1 Country Risk Tests
As stated earlier, James and Koller (2000), Cruces (2002), Cosset and Roy (1991), Vij (2005), Feder and Uy (1985) and Damodaran (2011) all give a different interpretation of the country risk concept. However, it is argued that country risk should embrace concepts like high inflation, macroeconomic volatility, political instability etc. An important question is if the different measures Damodaran provides as estimate of the CRP, are actually measuring these concepts. An important next question is then which measure captures the concepts best.
3.1.1 Country Risk Model
Country risk is represented by a set of variables that is used by the PRS Group in its International Country Risk Guide (ICRG) and by the Worldwide Governance Indicator (WGI) project of the World Bank. Since 1980, the PRS Group produces risk ratings for over 140 countries that are important for international business. These risk ratings are centered on three different pillars, namely Economic, Financial and Political Risk and comprise information on 22 variables which are largely collating with the variables mentioned in earlier research. ICRG ratings are used by, among others, Erb et al (1996a) and Diamonte et al. (1996) in academic research regarding country risk. As Klapper and Love (2004) and Hail and Leuz (2006) argue, differences in governance procedures are also explaining differences in stock returns in countries. Considering the statement of James
18
and Koller (2000) that shareholder rights are also related to country risk, I will add a measure for governance to the equation. WGI has yearly data from 2002 onwards on 213 countries, and uses ICRG data as one of the inputs for their measures for Political Risk and Governance. The above discussion leads to the following model that is estimated,
ε
β
β
β
× + × + × ++
=c EconomicRisk FinancialRisk PoliticalRisk
CRP 1 2 3 (10)
where Economic Risk and Financial Risk are measured by respectively the Economic Risk Factor and Financial Risk Factor of the ICRG. The variable Political Risk will be equal to the variables used by the WGI. The Data section hereafter will provide more insight in the variables used and their components. From the discussion above that is related to General H1 from Section 2 and from statements of Damodaran (2011), it is expected that all the three Risk Factors have significant explanatory power. Therefore the first hypothesis is:
H1: The Economic, Financial and Political Risk Factors have significant impact on the measures of CRP.
As pointed out earlier, Damodaran (2011), Cosset and Roy (1991) and Domowitz et al. (1998) suggest different methods to come up with an estimate of the CRP. According to their statements, the CRP is tested with three different measures, namely the Credit Default Swap (CDS) spread, the spread on 10 year Government Bonds between the local Bond and a Bond from a developed market (Damodaran (2011) identifies the US as the developed market) and the Relative Volatility, which is measured as in (7).5 Since Damodaran (2011) finds all three measures a good approximation for a CRP, it is expected that there will be no differences between the results of regressions on different measures of CRP. This leads to the second hypothesis:
H2: There are no significant differences between the coefficients and R-squared values of three separate
regressions on the different CRP measures as proposed by Damodaran (2011).
To test which variable captures country risk concepts best, a panel data least squares regression is performed. Since the next section shows that (10) is estimated using data on 63 countries over six consecutive years, this is the appropriate statistic estimation method. In accordance with Brooks (2008), fixed and random effects’ testing is dealt with in an appropriate way, together with statistics for autocorrelation and heteroscedasticity. To test if R-squared values and coefficients differ between regressions (as in H2), T-tests statistics on differences in mean values are provided and evaluated.
5 The fourth measure, the Country Credit Risk Rating from Moody’s S&P or Fitch, is not available for free over
19
3.1.2 Robustness checks
In (10), the Political Risk Factor is represented by the first component from a Principal Component Analysis that is explained further in Section 4. The model as in (10) is tested on robustness regarding different outcomes of the PCA; an additional second component of Political Risk is added to the equation and tested for significance.
The influence of adding a CRP to classic finance models is particularly significant when countries have high spreads on CDS or Government Bonds, or high volatility. These countries often fall into the category of emerging markets. To check whether results differ in different types of countries, two robustness checks are performed on the results of the estimation of (10). The total sample is split up in two parts, namely developed and emerging markets. The first method of splitting the total sample is by taking the definition of the World Bank for emerging markets and split the sample in two parts accordingly.6 This leads to 31 countries in the developed market group and 32 countries are marked as being emerging. Damodaran (2011) compares prime and subprime countries in his analysis. The second method to split up the sample is by countries with a AA+ or AAA rating7 per 8/2011 (developed markets) and countries with a lower rating (emerging markets). This leads to a total of 45 countries being marked as emerging and 18 as developed. 3.2 CAPM and Three Factor Model Tests
The second part of this research investigates whether the CRP helps to explain security returns. The CAPM model and the extended Three Factor Model by Fama and French are two widely used models to explain security returns. Damodaran (2011) states that a Global CAPM framework as in (3), without a CRP, is not a valid tool in emerging markets. This is merely because of the low integration with world markets, a statement that is supported by the findings of Harvey (1994). Next to this, he argues that the Local CAPM variant does not hold due to the lack of data in emerging markets.
3.2.1 Global CAPM and Three Factor Model Tests
To test whether the statement on Global CAPM is valid for my sample, (3) will be tested on the available dataset. Fama and French (1992) argue that proxies for Size and book-to-market-value (BTMV) are significant in explaining returns on stocks. It is proven by them and others that Small Cap firms often outperform Large Cap firms. Firms with a high BTMV are determined to be value stocks, whereas firms with a low BTMV are often labeled growth stocks. Fama and French (1992) conclude that firms with a higher BTMV generally
6 See Table A1 in Appendix A for which countries are marked as emerging and developed by the World Bank. 7 Because the US has a AA+ rating, but is viewed as being mature/prime by Damodaran (2011), this group of
20
outperform firms that have a low BTMV. For this reason, the Global CAPM framework will be extended to incorporate influences of both variables of Fama and French (1992) in the models. The equation becomes:
ε
β
β
β
− + + + + = − f t j mt f t t t t j R c R R Size BTMV R , , ( , , ) 1* 2* (11)where Rj,t−Rf,tequals the excess return of a particular security on the Risk Free Rate. βj(Rm,t−Rf,t)equals the beta of a security multiplied with the market risk premium (MRP). Returns are based on US dollar prices of securities, as was the case in Damodaran’s (2011) paper. Size is measured as the natural logarithm (ln) of the Market Capitalization in millions of US dollars at date t. The ln of Market Capitalization is used because, ln-data tends to be more normally distributed than the raw data in, in this case, dollar values (Brooks, 2008). BTMV equals the book value of equity divided by the market value of equity at date t, based on US dollar values. In empirical studies regarding the Three Factor Model and in the Fama and French (1992) paper, portfolios are used to incorporate the effects of Size and BTMV. Because effects on a firm level are explicitly important in this study (the effects of incorporating a CRP on firm level are tested), it is desirable that results are not blurred by making portfolios. Literature on explaining stock returns often adds the Carhart (1997) momentum factor to the Three Factor Model of Fama and French. This momentum factor is often used to check mutual fund performance, since the momentum factor reflects returns on a portfolio long in last year’s winners and short in the last year’s losers. The momentum factor, which would equal relative performance of one period versus one or more periods before, is less relevant in explaining individual security returns with firm-based factors, so I will make use of the Fama and French Three Factor Model and I +will leave the momentum factor out of consideration. To incorporate the CRP in the equation, the left hand side of the equation is reduced by the CRP estimate. This way, the extra explanatory power of the CRP can be tested in a CAPM and Three Factor Model setting. Damodaran (2011) expects that the CRP will increase the power of the Global CAPM and Three Factor Model Framework. From Table 1 it is seen that research by Bilson et al. (1996), Bekaert and Harvey (1995) and Serra (2000) add local factors to a Global CAPM framework, where a world index is used as proxy of the market index. Equation (11) is comparable to these studies; a local variable (CRP) is added to a Global CAPM framework, which is contrary to the earlier studies, also extended to incorporate influences from Size and BTMV. In line with the General H2 from Section 2 and above discussion on Country Risk measures, the following hypothesis is expected to hold:
H3: Adding CRP measures to a Global CAPM and Three Factor Model environment will increase explanatory
power of these models (as measured with R-squared)
21
Brooks (2008), fixed and random effects’ testing is dealt with in an appropriate way, together with statistics for autocorrelation and heteroscedasticity. Independent samples T-tests are then used to compare results between different regression estimates which incorporate the different measures of CRP.
3.2.2. Local CAPM and Three Factor Model Tests
As mentioned before, Damodaran (2011) states that a Local CAPM approach as in (4), where returns are denominated in the currency of the firms origin and are regressed on a local Risk Free rate and a local Market Index, does not work for emerging markets. Due to lack of data, unreliable estimates show up for countries that are marked as being emerging. Damodaran therefore clearly favors the Global CAPM approach with the addition of a CRP as in (3) for explaining variation in returns. For the years under investigation, sufficient data is available to come up with a Local CAPM approach, which is later extended to incorporate the Size and BTMV effects from the Fama and French (1992) model. The previously stated equation (11),
ε
β
β
β
− + + + + = − f t j mt f t t t t j R c R R Size BTMV R , , ( , , ) 1* 2* (12)is used with different definitions of the variables used. In (12) Firm Returns (Rj,t) and Market Returns (Rm,t) are denominated in local currencies, as opposed to the dollar value in (11). Moreover, Rm,t equals the return on a local index. The Risk Free Rate (Rf,t) equals the local 90 days short term interest rate and Market Cap and BTMV statistics are stated in local currencies at date t. The Local approach is comparable with analyses done in papers of Erb, Harvey and Viskanta (1996a) and Klapper and Love (2004) where only local variables were used in explaining differences in returns across companies. Because of the argumentation of Damodaran (2011) and Harvey (1994) that a Local CAPM setting will not improve results and in the line of the General H2 from Section 2, it is expected that the following hypothesis will hold:
H4: The Local CAPM or Three Factor Model approach will not lead to higher R-squared values (explanatory
power) than a Global approach.
The statistic procedures that are used for estimating (12) are similar to the ones used for estimating (11) and stated in Section 3.2.1. R-squared values of regressions in (11) are than compared to the R-squared values in (10) to evaluate H4. This is done by calculating T-test statistics on differences for all available regressions. 3.2.3 Robustness checks Global and Local Tests
22
Robustness tests are performed on results from the Global model in (11) and the Local model in (12) for the two subsamples of developed and emerging markets and these robustness checks are compared to results of the models in (11) and (12) on their relevance.
According to Damodaran (2011) and Ogier, Rugman and Spicer (2004), the Local CAPM variant as in (12) does not need an additional CRP. Local data, by means of local Risk Free Rates and local Market Index Returns in local currencies, are expected to reflect all country risk that is apparent and known to the market. Therefore, adding CRP estimates (in the form of a 10 year Government Bond Spread or a CDS spread) and Country Risk variables (variables of the ICRG as used in the Country Risk Tests of (10)) should not have a significant impact on the Local CAPM or Three Factor Model approach. Several robustness checks will deal with this statement by adding CRP estimates and Country Risk variables to (12).
4 Data analysis
This section describes the data used to conduct the analyses from Section 3. As described before, this study focuses on two topics, namely the appropriateness of certain measures when describing country risk factors and the importance of the CRP in explaining returns in a Global and Local CAPM setting. Both topics have different data requirements and they are discussed separately.
4.1 Country Risk Test
There are over a hundred stock exchanges in the world. From these stock exchanges, a list of 76 exchanges was composed using Damodaran (2011) and other empirical literature regarding country risk as mentioned in Section 2.8 As mentioned before, Damodaran considers four different variables as a potential measure for the CRP: the CDS spread, the spread on 10 year Government Bonds, the Country Credit Rating (CCR) as produced by Moody’s, S&P or Fitch and relative volatility of a stock market versus a mature market (Damodaran uses the USA as mature market). Data on the CDS spread, 10 year Government Bonds and stock market indices are collected from Thomson Reuters DataStream. Some countries’ yields on 10 year Government Bonds are collected from the sites of National Banks. The initial sample of 76 exchanges is reduced to 63 for which at least two of the three CRP measures are available. For 13 countries no data on CDS spreads or Government Bonds were available, so these countries are deleted from the sample. Table A1 in Appendix A provides an overview of the exchanges used in this study and the country where the exchange is located. Table 2 below provides descriptive statistics of the three variables that measure CRP.9 Damodaran (2011) takes the US as
8 From http://www.wikinvest.com/wiki/List_of_Stock_Exchanges The 76 exchanges were also chosen, because
they were available in DataStream. The exchanges excluded are not to be found in the DataStream database.
9 Appendix B, Table B1 gives an overview of Dependent and Independent variables in (10) with an exact definition
23
the mature market in the relative volatility approach, but fails to provide a statement on which market index should be taken. Therefore, both the large indices, the S&P500 and the Dow Jones are taken as a robustness check.
Table 2: Descriptive statistics dependent variables Country Risk Test
Descriptive statistics for the independent variables. CDS spread and Gov Bond spread is measured as the spread in percentage points over the US rate and US 10y Treasury Bond respectively. The relative volatility is calculated as in (9) and is also in percentage points. Yearly data for 63 countries for the 2005-2010 periods.
Statistic CDS Spread Gov Bond Spread Relative Volatility Dow Jones Volatility S&P500 Relative
Mean 1,7509 1.8746 0,2289 0,6832 Median 0,9189 0.6000 0,4988 0,6383 Maximum 37,7571 20.2400 12,7725 17,4765 Minimum 0,0291 0.0000 -48,6316 -55,9747 Std. Dev. 3,3839 3.0010 4,5145 5,1219 Skewness 7,1252 2.9556 -3,8034 -3,6547 Kurtosis 66,0442 14.2117 39,9904 43,4380 Jarque-Bera 58661,16 1894.258 22402,5 26526,0 Probability 0.0000 0.0000 0,0000 0,0000 Observations 337 283 377 377
The highest spreads on both CDS and Government Bonds are from Venezuela and the lowest spreads can be contributed to countries from the Euro area, Japan and Switzerland. The Relative Volatility measure produces large outliers, since CRPs of over -50% exist. These numbers coincide with the financial crisis during 2008, were large negative returns are apparent. These large negative values are found in Russia and Iceland. It can be seen that all the variables are highly non-normally distributed. Therefore estimation results should be interpreted with care and with this anomaly in mind.
24
Table 3: Correlations between dependent variables Country Risk Model
Correlations between independent variables with pairwise deletion. Correlations and Number of Observations are given. * indicates significant at 5% **indicates significant at 1%
CDS Spread Gov Bond
Spread Relative Volatility Dow Jones Relative Volatility S&P500 CDS Spread 1,0000 378
Gov Bond Spread 0,6583 ** 1,0000
378 378 Relative Volatility Dow Jones -0,2226 * -0,0334 1,0000 378 378 378 Relative Volatility S&P500 -0,1996 * -0,0383 0,9808 ** 1,0000 378 378 378 378
The correlations in Table 3 indicate that not all measures are positively correlated with one another. The high positive correlation between the two Relative Volatility measures does make sense. The Government Bond Spread is only correlated with the CDS spread, not with the Relative Volatility. The lower, though significant negative correlations between the spread on Government Bonds and the Relative Volatility measure, indicate that these are not perfect substitutes either.
25
4.1.1 Economic and Financial Risk Factors
As mentioned before, the Economic and Financial Risk Factor each consist of five different variables.
Economic Risk Factor Financial Risk Factor
GDP per Capita Foreign Debt as % of GDP
Real GDP Growth Foreign Debt Service as % of Export
Annual Inflation Rate Current Account as % of Export
Budget Balance as % of GDP Net International Liquidity
Current Account as % of GDP Exchange Rate Stability
Points scored on each variable are added and a total score per country is reported. Table 4 reports the descriptive statistics of both Risk Factors in total. For descriptive statistics of the individual components and the correlations between all the independent variables, see Tables C1-C4 of Appendix C and Table D1 in Appendix D. Note that the relationship between Risk Score and risk association is negative. Higher scores on Risk Factors mean lower associated risk.
Table 4: Descriptive statistics Economic and Financial Risk Factor Descriptive statistics of the Economic and Financial Risk Factor of the ICRG. 63 countries from 2005-2010 are included. Note that the maximum score on each risk factor can be 50 points.
Statistic Economic Risk Factor Financial Risk Factor
Mean 37.51 38.40 Median 37.50 38.50 Maximum 49.50 48.50 Minimum 18.50 19.50 Std. Dev. 5.15 4.63 Skewness -0.39 -0.50 Kurtosis 3.37 3.56 Jarque-Bera 11.65 20.86 Probability 0.0031 0.0000 Observations 378 378
26
4.1.2 Political Risk Factor
The Political Risk Factor is constructed with data from the WGI project.10 The WGI project defines six variables on which each country is scored, namely
• Rule of Law (RL) • Regulatory Quality (RQ)
• Political Stability and Absence of Violence (PV) • Government Effectiveness (GE)
• Control of Corruption (CC) • Voice and Accountability (VA)
Since the WGI uses ICRG data as one source of input data, the Political Risk Scores from ICRG are available on the website of WGI. WGI comprises the 12 variables of the ICRG into the six variables mentioned above. Table 5 describes the variables of WGI. 11
Table 5: Descriptive statistics Political Risk Factor
Description of the six variables of the WGI project. 63 countries from 2005-2010 are included. Note that the maximum scores on the factors are 100.
Statistic GE PV RL RQ VA CC Mean 74.21 74.47 73.72 82.72 82.49 55.03 Median 75.00 75.09 83.33 90.91 87.50 50.00 Maximum 100.00 92.04 100.00 100.00 100.00 100.00 Minimum 25.00 40.53 16.66 13.64 16.66 16.66 Std. Dev. 22.06 9.36 19.95 18.47 19.40 20.39 Skewness -0.32 -0.68 -0.65 -1.46 -1.24 0.44 Kurtosis 2.16 3.71 2.63 5.15 3.93 2.01 Jarque-Bera 17.53 36.88 29.44 206.73 110.93 27.37 Probability 0.0002 0.0000 0.0000 0.0000 0.0000 0.0000 Observations 378 378 378 378 378 378
All variables are highly non-normally distributed, mainly because of the fact that there are a lot of countries that score high on the risk factors and because there are a few outliers on the downside. It is notable that the average score on CC is much lower than the average scores on the other variables.
27
One of the characteristics of the WGI data is that there is no single risk factor available. All the six variables are viewed independently, whereas the ICRG is able to merge all the variables into one measure. To be able to merge the six variables into one single indicator for Political Risk, a Principal Component Analysis is conducted12. The results of this analysis are presented in Table 6.
Table 6: Principal Components Analysis (PCA) for the six Political Risk measures
Number Value Difference Proportion Cumulative Value Cumulative Proportion
1 3.8565 3.0823 0.6428 3.856 0.6428 2 0.7742 0.1721 0.1290 4.630 0.7718 3 0.6021 0.2780 0.1003 5.232 0.8721 4 0.3241 0.0895 0.0540 5.556 0.9262 5 0.2346 0.0261 0.0391 5.791 0.9653 6 0.2085 --- 0.0347 6.000 1.0000
This PCA shows that comprising variables into components could be relevant here. From Malhotra (2009) we know that only the components with an eigenvalue greater than 1 are of interest. He also states that it is desirable that the components that are chosen explain at least 60% of total variance. With regard to both statements, this means that only the first component is chosen. Therefore, all the variables can be merged into one single indicator. However, adding the second component adds 13% to total explained variance (which is quite substantial), but this component has an eigenvalue of less than 1. From Appendix E, Table E2 it can be seen that the second component is very dependent on the value of only one variable, namely PV. Therefore, a PCA is conducted with five variables (excluding PV). Also, this results in the solution of one component and explains 72% of total variance. Because Malhotra (2009) states that the general rule is that eigenvalues have to be greater than 1 and total explained variance has to be over 60% and because WGI suggests that all variables are measures of political risk and governance procedures, one component is chosen from the PCA analysis with six variables13.
4.2 CAPM and Three Factor Model Tests
To see whether the CRP is significant in explaining security returns, a test of the Global and Local CAPM (and the extended Three Factor Model) is conducted. For the 63 countries that are mentioned in Section 3.1, there are 55 which have prices on a security level available for the years 2006-2010 (or part of this period) in Thomson Reuters DataStream. The countries, together with the appropriate index and the number of available companies, are summed in Appendix A, Table A1.
12 From Table E3 in Appendix E it can be seen that the correlations between the 6 variables, are between 0,32 and
0,77 and are all significant at the 1% level. Detailed PCA analysis in Appendix E, Tables E1 and E2
28
4.2.1 Global CAPM and Three Factor Model Tests
For the 2213 available firms, quarterly data is gathered for the period January 2006 - December 2010. Next to the stock prices, DataStream is used to gather information about Market Capitalization and BTMV for the incorporation of the Fama and French (1992) variables. US Dollar denominated values are used in the analysis. To deal with large outliers in Market Capitalization and BTMV, the top and bottom 0,5% of the values is set equal to the closest observation available thereafter. Table 7 gives descriptive statistics for the Global CAPM and Three Factor Model Tests variables. As can be seen from Table 7, it seems that all the variables are non-normally distributed. Since our sample is relatively large, small deviations from the normal distribution lead to rejection of the null hypothesis of a normal distribution. According to the central limit theorem, a large sample will be approximately normally distributed, so in this it can be assumed a normal distribution even with a rejection of the null hypothesis with the Jarque-Bera test (Newbold et al, 2003).14
Damodaran (2011) shows that the returns on stocks (S&P500, 11.31%) and the risk free rate (US Treasury Bill, 3.70%) are on average much higher for the period 1928-2010 than they are in my sample. This fact can be largely contributed to the huge impact of the recent financial crisis on stock markets. Next to this, Damodaran provides estimates of average returns on shorter time periods, i.e. the past 30, 20 and ten years. These estimates show that the average return on stocks is declining over time. The low minimum returns can be mainly contributed to firms that do no longer exist or to firms from less developed markets. So are the high returns, which are found in countries like Pakistan, Romania and Egypt. For market capitalization, it seems that the larger companies are mainly from developed markets, whereas smaller companies are from less developed markets. High BTMV firms can be found mainly in less developed countries like the Philippines and Thailand, whereas the low BTMV are spread across all different kind of countries; developed as well as emerging.
14 The non-normally distributed data that is used later in this paper will also be assumed normal by the explanation
