The macroeconomic indicators and economic performance of frontier markets can differ completely from country to country. In Table 11.1, several frontier markets are compared in terms of their GDP per capita in nominal 2014 dollars along with GDP per capita purchasing power parity (PPP)-adjusted values. These two measures help us better understand the living standards in selected countries. There are huge differences in per capita GDP levels among frontier markets. In addition, the PPP-adjusted figures are considerably higher than the nominal ones: sometimes more than twofold. As Table 11.1 indicates, there are huge gaps in GDP levels among countries—for instance, the GDP per capita levels of Bangladesh, Ghana, Zimbabwe, and Pakistan are considerably lower than those of Bahrain, Kuwait, and Saudi Arabia.

Frontier markets are spread across five regions: Africa, Asia, Eastern Europe, Latin America, and the Middle East. African and Middle Eastern countries have high potential due to abundant resources, including oil and gas resources, and a propitious investment climate. Frontier markets in the European region neighbor developed countries, and as a consequence can attract capital, thanks to being part of the Euro zone. Asian frontier markets also have a good potential to perform successfully, as was observed in the cases of China and India. However, there are almost 20 countries with a population of more than 280 million that are not represented in frontier market indices.

Political and economic policies play a key role in defining a country as a frontier market. Time after time, markets with relatively small market caps have been motivated to develop their policy regimes to attract foreign investment (Ghose, 2004). During the last decade, many frontier markets have started to demonstrate rapid growth by carrying out some effective legal and institutional reforms, particularly by reviewing tax policies and reducing restrictions on capital inflow. Today 17 frontier economies, including Ghana (7.6%), Qatar (6.5%), and Kazakhstan (6.0%) have the highest average annual GDP growth among 20 fast-growing markets (World Bank, 2015). This trend was confirmed by recent GDP forecasts by the International Monetary Fund (IMF).

Portfolio investors have always kept developing markets under close attention. However, usually many of them do not consider frontier markets as potential investment markets because of high risks. Speidell (2011) draws attention to the fact that these markets, aside from certain challenges, have favorable advantages for investment diversification. Speidell (2011) argues that the risk of investing in frontier markets is not much more than that in emerging countries. He suggests carefully observing these markets, relying on the Heritage Foundation’s Index of Economic Freedom. Table 11.2 illustrates the economic freedom index figures of some of the emerging and frontier markets with the most significant changes over the last 12 years. Surprisingly, closer investigation of Table 11.2 indicates that frontier markets have relatively higher positive changes in the overall index (10%), while in emerging markets the increase is 6%. Several frontier markets—such as Croatia, Kuwait, and Estonia—managed to increase their positions in the world ranking by demonstrating a remarkable change of more than 50%. However, in spite of high ratings in 2003, Ukraine, Nigeria, and Bahrain had negative changes. Meanwhile, emerging economies such as Brazil (−11%), China (−18%), Egypt (−23%), and India (−4%) experienced deceleration in economic freedom index scores.

Rational investors search for alternative investment markets, and frontier markets may offer fresh investment opportunities, especially after the recent GFC (Shukla, 2013). Recently, some researchers have focused on frontier markets with the motivation of studying possible investment opportunities. For example, Speidell and Krohne (2007), also known as pioneers in investigating frontier markets, analyzed the correlation between frontier markets and the world, and proposed that frontier markets might suggest possible diversification benefits, since these markets have limited interrelations with the world. Later, this proposition was supported and deeply discussed in empirical research by Berger et al. (2011). Moreover, Chen et al. (2014) attempted to examine the determinants of potential integration between frontier markets and developed markets while considering the financial crisis effect in the integration process. This study suggests that some frontier market officials should develop their existing stock markets. However, Marshall et al. (2015) stated that frontier markets have relatively high transaction costs and offered negligible diversification benefits during the financial crisis. Furthermore, they emphasized that the liquidity issue in frontier markets is an important consideration. This view is also supported by Samarakoon (2011), whose research was based on analyzing the interdependence of US shocks and frontier markets. He claimed that the recent financial crisis had more infectious effects on frontier markets relative to emerging markets. On the other hand, de Groot et al. (2012) documented that stock returns in frontier markets cannot be interpreted by virtue of the effect of global shocks.

Currently the total population of frontier markets has reached 1.2 billion people, constituting almost 22% of the world’s population. The percentage share of the young population in frontier markets is significantly higher than that in developing markets, which is an important fact to be noticed (World Bank, 2014). Possible arguments in favor or against frontier markets are summarized in Table 11.3.

These economies are classified based on different criteria. Investment capability is the essential criteria determining a country’s classification, while other macroeconomic indicators (eg, aggregate and per capita GDP) play more of a secondary role. Market cap–weighted frontier market benchmarks from these index providers mainly consist of stocks listed in Africa, the Middle East, the CIS (the Commonwealth of Independent States, or former Soviet republics), and less-developed Asian countries, such as Pakistan, Vietnam, and Bangladesh. While frontier markets in African and Asian countries tend to have very low per capita GDP figures, the oil-producing states in the Middle East (also known as the Gulf Cooperation Council, or GCC, countries) are relatively wealthy. The reason these resource-rich nations are considered frontier markets is because they have restrictive foreign ownership limits on their exchange-listed companies.

Several factors could undermine the diversification benefits of frontier markets. Primarily, increased integration of frontier markets with the global financial system and the world economy could induce diminished diversification. However, Speidell and Krohne (2007) find no evidence of correlation between frontier markets and developed markets equities. Jayasuriya and Shambora (2009) investigated diversification advantages across markets and document that investors could benefit by diversifying their portfolios with frontier markets. A recent study by Cheng et al. (2009) observed that there is minor integration within emerging and frontier markets using CAPM (capital asset pricing model) variations. In addition to that, Berger et al. (2011) saw no indication that the frontier markets are becoming more integrated over time, even after allowing for structural breaks, and advocate that investors could improve their positions with the diversification benefits of these markets.

Another factor is the increased transaction costs resulting from the lower liquidity and depth of frontier markets, that can neutralize diversification benefits. Speidell (2011) observed that bid-ask spreads in frontier markets could be as enormously high as 10–12%, based on a case study of Kenya and Ukraine. It has also been documented that transaction costs in frontier markets are 3 times those in the United States. However, investors can improve their portfolios and benefit by investing in frontier markets, despite high transaction costs. For instance, Marshall et al. (2013) detected that combining frontier markets equities with international portfolios can generate higher Sharpe ratios. Consequently, frontier markets turn out to be better suited to a diversification strategy than emerging markets, since the diversification benefits of the latter could disappear when transaction costs are accounted for (de Roon et al., 2001).

There are other costs beyond transaction costs associated with frontier markets investments (Schoenholz, 2010; Speidell, 2011). Regulatory and trading costs emerge from ownership constraints, local administrative obligations, custody requirements, and the need to employ domestic brokers. In particular, emerging and frontier markets are very vulnerable to substantial equity price risk and exchange rate risk exposure. In addition, an operating stock index derivatives market may be absent in most emerging and frontier economies, making it more difficult to manage foreign equity price risk. Moreover, investors also have to cope with adjustment systems that are often subject to errors, vulnerable corporate governance standards, and political distress. In spite of all of these barriers, investment opportunities in frontier markets remain appealing. Growth performance data for various economies classified by type of market are shown in Fig. 11.1.^c The median annual growth rates smoothed over 3 years for 52 economies in the 2002–07 period in Fig. 11.1 were consistently positive, indicating that these countries have experienced no growth deceleration. Furthermore, since 2007 all market economies except for frontier market economies have endured a significant decline in income per capita growth, especially in 2009 and 2010. Thus most countries suffered from substantial slowdowns, but the growth deceleration in the United States and developed EMEA (Europe, the Middle East, and Africa) markets was much more severe. By contrast, median annual growth rates for frontier markets experienced no growth deceleration in the pre- and postcrisis periods. Frontier markets are more attractive investment markets since frontier markets outperform other markets in terms of income per capita growth.

Emerging economies have been displaying rapid real and monetary growth and continue to be appealing to investors in developed countries who are looking for opportunities to improve portfolio performance (Wang and Low, 2003). The developing world was growing more rapidly than developed economies even in the period preceding the financial crisis of 2008–09 and the economic deterioration that followed it. For instance, transitional and emerging economies largely contributed to the world’s economic growth in 2009 (UN, 2011). Table 11.5 displays the magnitudes of economic growth data displayed in Fig. 11.1. Frontier markets on average are growing at least 2 times higher than developed EMEA markets, indicating better investment opportunities.

Value-at-risk (VaR) estimates are provided to support further investigation of investment opportunities in frontier markets. Building conditional volatility under the framework of a GARCH model is the approach adopted in the RiskMetrics methodology. For most of the financial time series, volatility tends to exhibit clustering behavior, with prolonged periods of high and then low volatility. This type of behavior in financial time series was initially observed and suggested by Engle (1982) through the use of an autoregressive conditional heteroskedastic (ARCH) process. The ARCH process was modified by Bollerslev (1986) so that the conditional variance is not only explained by movements in the past errors but also by lagged conditional variances. GARCH (generalized autoregressive conditional heteroskedastic) modeling has since been widely used in empirical applications for the second moment in financial time series. Following the RiskMetrics methodology, the conditional variance is measured by a GARCH(1,1) model with restrictions on parameters; that is, by forcing the constant to be zero and with the parameters summing to unity. The methodology is formally known as integrated GARCH (IGARCH) (Pownall and Koedijk, 1999):

Instead of estimating unconditional volatility utilizing an equally weighted moving average, the RiskMetrics methodology uses exponential weights, putting more weight on the most recent observations. The deceleration rate of the exponential weights depends on the decay factor, thus indicating the persistence with which a shock will decay over time. Academic literature suggests setting the decay factor to 0.94 and 0.97 when using, respectively, daily and monthly data (Christoffersen, 1998). This is the clear advantage of the RiskMetrics method, since only one parameter is to be estimated that facilitates the estimation process and provides for greater robustness against estimation error. Thus, the VaR measure for different markets using the RiskMetrics approach is estimated (Fig. 11.2). The estimate of 1% VaR using RiskMetrics was considerably lower for frontier markets in contrast to that for other markets, implying a relatively smaller risk level in the postcrisis period since 2008. However, these markets are vulnerable to financial turmoil and distress, as a tremendous increase in the 1% VaR estimate was observed during the crisis. Nevertheless, the potential investment benefits of frontier markets are appealing to investors.

Examining return correlation across different markets can support the argument that frontier markets are attractive to potential investors. To that end, return covariances across MSCI indices for different markets and the world are estimated and presented in Fig. 11.3. It shows the time series of covariance between the monthly Emerging Markets–World MSCI (EM/World), frontier markets–World MSCI (FM/World), US–World MSCI (FM/World) and Europe–World MSCI (Europe/World) for the period Jan. 3, 2003–Jun. 27, 2006, using the DCC model based on EWMA correlations with a smoothing constant of 0.94 designed by RiskMetrics (1996). Even taking into account the high covariance of frontier markets and world market returns during the crisis, the covariance for frontier markets was significantly lower than that for emerging markets. In addition, the return covariance of frontier markets is the lowest for the rest of the periods: 2002–08 and 2011–14.

In this section we make use of analytical tools to explore the spillover effect of changes in the US stock exchange over frontier market countries. Academics and practitioners have to study such spillover effects carefully, since a strong shock striking a financial market of especially a high-output country is transmitted to others at different levels and lags. The aforementioned spillover effect can be studied via the coordinated behavior of markets with changing lags.

There are various techniques used to analyze spillover effects, and we have opted to employ the GARCH framework first, as the majority of researchers have done. There are a voluminous number of applications of GARCH models to research spillover effects. Hamao et al. (1990), for instance, document the spillover running from the US to Japanese markets in one of the earliest studies. A study by da Veiga and McAleer (2005) tests for the existence of volatility spillover among the S&P 500, FTSE 100, and Nikkei 225 indices using daily data for the period over 12/10/1992 to 7/7/2003 with a VARMA–GARCH specification and reports the volatility spillover from the FTSE 100 to both the S&P 500 and the Nikkei 225 and from the S&P 500 to the FTSE 100. Similarly, Abou-Zaid (2011) analyzed the dissemination of stock index volatility from the United States and United Kingdom to selected MENA (Middle Eastern and North African) emerging markets. In another article Bhar and Nikolova (2007) use the GARCH-in-mean specification to question whether world index returns have an impact on the variance of returns in Brazil, Russia, and India. In another analysis of such an effect, Theodossiou and Lee (1993) report evidence of volatility spillovers from the S&P 500 to all other indices. Three families of the GARCH framework are used to investigate the spillover effects: GARCH-in-mean, VECH, and BEKK parameterizations. In the VECH parameterization, assuming that ɛ_it is the error term of country i at day t:

In this formulation, H_t is the conditional variance matrix. Basically there are two main approaches to model the spillover effect on $H_t^{1/2}$. In the first one—the VECH model of Bollerslev et al. (1988)—the columns of the argument belonging to the VECH function are stacked to the column vector.

In the BEKK parameterization of the multivariate GARCH(1,1), the spillover effect is modeled as:

Note that there are numerous GARCH specifications possible, ranging from the EGARCH of Nelson (1991) to the VARMA–GARCH model of Ling and McAleer (2003). In this article we adopt the GARCH-in-mean model to determine the spillover effect on the return series. In this specification R_i,t is the stock market index return of country i for day t, computed as for each of the frontier market countries; that is, i = Argentina, Bahrain,..., Vietnam, where P_i,t is the closing price of the index for country i on day t:

We have multiplied the $\ln (R_{i,t}/R_{i,t-1})$ in order to amplify the figures. Indeed, the first step is to test for the existence of the ARCH effect to see whether the series under focus has changing volatility or not. We have executed this test, and reported that the null hypothesis of “no ARCH effect” is rejected by all but one series highly significant; that is, with a p-value of about 0.0000. The only exception was Croatia, with a p-value of more than 10%. We plot the series for the USA, Argentina, Kenya, and Romania without loss of generality (Fig. 11.4). The graphs very explicitly show that there is changing variance over the whole range of 1215 observations. We present the results of ARCH effect and stationarity tests in Table 11.6 for each frontier market country. In ARCH effect testing the test statistic, nR², follows the χ² distribution. We report the test statistic as well as the p-value of the test in Table 11.6 directly from the EViews output.

Similarly, the stationarity of series belonging to each country is tested in order to apply Granger causality and to get rid of spurious regression. In testing for stationarity, we check for the existence of a unit root claimed by the null hypothesis; that is, the series is nonstationary. The augmented Dickey–Fuller test statistic and its p-value are listed in the same table. Note that none of the series belonging to the frontier market countries or the USA has a unit root; that is, they are all stationary. We test for the ARCH effect and stationarity of the US series. The null is rejected in both tests with p-values less than 0.000.

The stylized GARCH specification has two equations, one for the mean and one for the variance. In this study we adopt:

where ${\varepsilon }_{i,t}\sim N(0,H_{i,t})$.

In this formulation, ɛ_i,t is the error term and H_i,t is the GARCH term. H_i,t is included in the mean equation to investigate the mean spillover effect of US stock exchange index returns on frontier market countries. And similarly, ${\varepsilon }_{\text{US},t}^2$ is included in the variance equation to explore the volatility spillover effect. We have estimated the GARCH-in-mean equations using EViews 8 and have listed the outputs of all estimation results in Table 11.7. Note that θ is the coefficient representing the mean spillover of the US stock market on the frontier market country’s stock exchange. Numbers right below the coefficient estimates are the probability values belonging to the significance tests of coefficients; that is, very low p-values, especially those less than 5%, prove that the coefficient is significant and the mean spillover effect is documented. The mean spillover is proven for Argentina, Bulgaria, Croatia, Estonia, Kazakhstan, Kuwait, Lithuania, Morocco, Romania, Serbia, Slovenia, Tunisia, and Ukraine highly significant; that is, p-values are less than 1%. We reject the null hypothesis of θ = 0 for Kenya and Nigeria, but at less than a 10% significance level. The mean spillover effect is not documented for Bahrain, Bangladesh, Lebanon, Pakistan, Sri Lanka, and Vietnam. There are two countries positioned at the borderline: Jordan (p-value = 14.4%), and Oman (p-value = 17.4%).

Regarding the volatility spillover, GARCH-in-mean estimation results reveal that the volatility spillover is documented for Bahrain, Bulgaria, Estonia, Kazakhstan, Kenya, Lebanon, Lithuania, Mauritius, Morocco, Nigeria, Oman, Pakistan, Romania, Serbia, and Slovenia with the exact level of significance less than or equal to 1%. In addition, the same is true for Croatia, Jordan, and Kuwait, with p-values of less than 5%, for Bangladesh with 6.5%, for Argentina with 12%, for Sri Lanka with 9.3%, and for Vietnam with 8.7%. The two exceptions are Tunisia with 70.9% and Ukraine with 14.8%. All in all, the volatility spillover is proven for 21 out of 24 frontier market countries with a less than 10% p-value.

We proceed to the Granger causality test between returns for the USA and for frontier market countries. Theoretically, in order to test for Granger causality between the USA and frontier market i;

The null hypothesis of “R_i does not Granger cause R_USA” in the first equation asks for the testing of β₁ = β₂ = … = β_m = 0. Similarly, the second equation tests for “R_USA does not Granger cause R_i.” We again use EViews 8 with lag length of 3 days (Table 11.8).

Argentina, Bahrain, Bangladesh, and Tunisia have no Granger causality with the USA. Almost all of the remaining countries, including Bulgaria, Croatia, Estonia, Jordan, Kazakhstan, Kenya, Kuwait, Lebanon, Mauritius, Morocco, Oman, Pakistan, Romania, Slovenia, Sri Lanka, Ukraine, and Vietnam have unidirectional causality with the USA; that is, there is causality from the USA to these countries. Finally, Lithuania, Nigeria, and Serbia have bidirectional causality with the USA. Indeed, these findings are analogous to the previous ones in the majority of cases.

Our analyses show that there is significant mean spillover effect for Argentina, Bulgaria, Croatia, Estonia, Kazakhstan, Kuwait, Lithuania, Morocco, Romania, Serbia, Slovenia, Tunisia, and Ukraine. Yet there are two countries, Jordan and Oman, positioned at the borderline, and the mean spillover effect is not documented for Bahrain, Bangladesh, Lebanon, Pakistan, Sri Lanka, and Vietnam.

Concerning the volatility spillover, GARCH-in-mean estimation reveals that significant volatility spillover is detected for Bahrain, Bulgaria, Estonia, Kazakhstan, Kenya, Lebanon, Lithuania, Mauritius, Morocco, Nigeria, Oman, Pakistan, Romania, Serbia, and Slovenia. The same is documented for Croatia, Jordan, and Kuwait, with lesser degree of significance. The Granger causality test is performed to identify the interaction between the United States and frontier markets. Argentina, Bahrain, Bangladesh, and Tunisia have no Granger causality with the USA. Almost all remaining countries—including Bulgaria, Croatia, Estonia, Jordan, Kazakhstan, Kenya, Kuwait, Lebanon, Mauritius, Morocco, Oman, Pakistan, Romania, Slovenia, Sri Lanka, Ukraine, and Vietnam—have unidirectional causality with the USA; that is, there is causality from the USA to these countries. Finally, Lithuania, Nigeria, and Serbia have bidirectional causality with the USA. Indeed, these findings are consistent with those of previous studies in most cases.

The linkages between frontier and developed markets are expected to strengthen in the future. The IMF World Economic Outlook Database (2010) recognizes that immature markets often provide upside potentials. Investing in frontier markets, that tend to be smaller, illiquid, and more restrictive than emerging markets, indeed induces high risks beyond those confronted in emerging markets. In this regard, the study gives insights into the investment risk possibilities involved in investing in frontier markets.