Chapter 8
Are European Frontier Markets Efficient?
D. Bond
K. Dyson Ulster Business School, Ulster University, Londonderry, United Kingdom
Abstract
With the collapse and dismemberment of the USSR in the early 1990s, and the move to free market economies in many Eastern European countries, a number of new frontier stock markets emerged. In the early years of their existence, there was great interest in whether these new frontier stock markets were informationally efficient and therefore potential investment opportunities. Unfortunately, some of the early econometric tests used to assess the level of informational efficiency had little statistical power to reject the null hypothesis. Therefore, this chapter reinvestigates these stock markets, using longer time series and more powerful econometric techniques that have been developed since those earlier studies. Specifically, the possibility of long memory and bubble behavior (irrational exuberance) is investigated.
Keywords
Eastern Europe
informational efficiency
random walk
fractional integration
bubbles
1. Introduction
The concept of frontier markets is a nebulous term and can carry many meanings. This chapter uses (2013, pp. 3) definition of frontier markets, namely that they are a subset of emerging markets, with lower market capitalization and liquidity or more investment restrictions than more established emerging markets, or both, depending on the country under consideration. Within Europe, the countries that broadly fit these definitions are predominantly the Eastern European, former Soviet bloc, excommunist countries.
As such, these Eastern European frontier markets, by their nature, are smaller in size and lower in liquidity than other comparable markets in the rest of Europe (and indeed the world) and are therefore not highly correlated with other markets, or indeed each other, thus increasing their attraction to investors searching for higher potential yield while lowering the overall volatility of their investment portfolios. Crucially, all such frontier markets exhibit high growth potential and consequently higher expected returns. Partially, these qualities arise from the fact that frontier markets remain relatively local in character, often shaped primarily by internal economic and political dynamics, as these countries have embraced capitalism after the fall of the Soviet bloc in the 1990s; (Speidell and Krohne, 2007; Shadbolt, 2013, eg, for a full exposition of the case for—and against—investing in such frontier markets.)
As more and more international investors have moved into these Eastern European frontier markets, a natural question is: are the financial markets in these countries informationally efficient, or are there inefficiencies from which investors might profit? A corollary question is: have bubbles developed in these markets as they rush to attract investment and “hot money floods in,” without investors performing the necessary and appropriate due diligence? This study therefore examines these two questions for Eastern European frontier markets in the period between the late 1990s and mid-2015.
The study is structured as follows: Section 2 examines the theory of informational market efficiency; Section 3 examines the literature in the area of frontier markets and informational efficiency; Section 4 discusses the methodology; while Section 5 examines the data used in the study; the results are discussed in Section 6; and finally, the conclusions and recommendations for further study are summarized in Section 7.
2. The Theory of Informational Market Efficiency
While, on the face of it, the concept of informational market efficiency appears relatively simple, the realities are somewhat different, a point made clear by Timmerman and Granger (2004).
Research into the informational efficiency of markets originated with the paper of Samuelson (1965) and the study of Bachelier into commodity markets in the early 1900s (Cootner, 1964). A problem with all of this literature is that the definition of what comprises an efficient market has changed subtly through time (Fama, 1965a,b; for two early definitions). These differences in definition simply compound the problem of investigating whether markets are informational efficient.
To provide a foundation for both his, and future, research into the concept of information efficiency, Fama (1970) refined and defined what is meant by “information.” He did so by classifying the information on the basis of three “levels” of availability—namely historical, present, and insider information—in his efficient market hypothesis (EMH). These three levels of information comprise information sets and are self-encompassing, with the higher levels incorporating the lower sets. Therefore, evidence in support of a higher level of informational efficiency provides evidence that the lower level(s) hold(s) also. Thus, by definition, evidence against a lower level of efficiency also topples the level(s) above.
In a weak-form efficient market, the information set comprises prior (historical) share prices or returns. Investors cannot earn excessive returns from detecting patterns in share prices or returns, and consequently from developing trading rules that attempt to exploit the perceived patterns, as no such patterns should exist; thus technical analysis should not yield consistent excess profits.
In a semistrong-form efficient market, the information set comprises all publicly available information, including historical information, thus encompassing weak-form efficiency. Investors cannot earn excess returns from trading in such publicly available information, as it is incorporated instantaneously into the share price once made public, leaving no time to profitably trade from such information; thus fundamental analysis should not yield consistent excess profits.
In a strong-form efficient market, the information set comprises information privy only to the company insiders, as well as all historical and publicly available information. Investors cannot earn excessive returns from trading in such information, as it will also, as with the other information sets, be incorporated instantaneously into the share price.
The empirical study of share prices, their returns and volatility, and consequential levels of informational efficiency mirrors the development of statistical analysis of time series. The traditional view of share prices has been that they follow some variant of a martingale/random walk process (Granger, 1992); for inference purposes, see Campbell et al. (1997) for variants on the random walk process. As more powerful tests have rejected the simple random walk models in empirical studies, more complex econometric models have allowed for the incorporation of some sort of “memory” in the share price formation process. Such models contradict the weak-form variant of the EMH (Gil-Alana, 2006; Floros et al., 2007).
The mounting contradictory evidence, and doubts as to the efficacy of simple martingale/random walk models, have combined and resulted in a reappraisal of what can be defined as an informationally efficient market (Jensen, 1978; Malkiel, 1992; Timmerman and Granger, 2004). As such, these works shift the definition of an informationally efficient market away from an economic framework to a set-theoretic one. This is best summarized by Timmerman and Granger (2004), who, building on the earlier research, provide the following definition:
A market is efficient with respect to the information set, Ωt, search technologies, St, and forecasting models, Mt, if it is impossible to make economic profits by trading on the basis of signals produced form a forecasting model in Mt defined over predictor variables in the information set Ωt and selected using a search technology in St.
This definition requires that prices fully reflect all information to the point that the marginal returns from acquiring the information equal the marginal costs. All of these definitions do not permit profitable arbitrage: a requirement under the earlier definitions of informational efficiency. Fama (1991, 1998) argues that this is a more acceptable approach than that espoused by the likes of Grossman (1976), Grossman and Stiglitz (1980), and Keiber (2007): namely, that truly informationally efficient markets are impossible. Their counterargument is a deceptively simple one: namely, that if arbitrage opportunities do not exist, then there is no profit to be gained from gathering information and trading on it (because if share prices fully reflect all information, then information and trading costs must equal zero). If this is the case, then there will be little motivation to trade and, in the extreme, markets will collapse. The reason why this does not occur, they argue, is that the degree of informational inefficiency determines the level of effort that investors are willing to expend to garner, and then trade, on the information. The profits earned by these investors accrue, according to Black (1986), because of “noise traders” where “noise traders” are those individuals who trade on what they consider to be relevant market information but which, in reality, is simply “noise” in the market and thus irrelevant to the price formation process.
The doubts that have been raised as to whether financial markets are informationally efficient have led to a number of theoretical alternatives to the EMH being proposed. Among these are the theory of fair markets (TFM) and the adaptive markets hypothesis (AMH).
The TFM (Frankfurter, 2006) proposes that informational markets should be considered fair, rather than efficient. The AMH proposes a reconciliation of the EMH with the behavioral alternatives, as proposed by the likes of Shefrin (2000), Statman et al. (2008), and others, by applying evolutionary theory to financial markets.
The groundwork in AMH begins with Farmer and Lo (1999), with formalization of the theory by Lo (2004), who couples bounded rationality and “satisfying” theory (Schwartz, 2004) with evolutionary dynamics to create a more realistic model of the price formation process. Market prices under the AMH reflect as much information as is dictated by the combination of environmental conditions and the number, and nature, of market participants. Thus, profitable opportunities in such markets will be a time-varying function of environmental market conditions. The AMH will therefore permit behavioral biases to exist, with the following implications for market participants: (1) The risk–reward relationship will be inherently unstable through time; (2) arbitrage opportunities will evolve and disappear through time; (3) the profitability of investment strategies will vary through time; (4) markets will continuously innovate; and (5) the prime objective of all firms—in Darwinian terms—is survival.
In summary, the AMH, by permitting arbitrage, provides justification for the activities of traders who, by their activities, ensure that markets remain relatively efficient through time. Thus, the Grossman–Stiglitz–Keiber explanation for the persistent existence of traders is compatible, and consistent, with AMH. Further, by permitting informational efficiency in markets to be a time-varying function, a natural explanation is provided for those studies that find conflicting evidence regarding market efficiency over varying time periods; Timmerman and Granger (2004) and Floros et al. (2007) for a concise summary of such studies.
The AMH is therefore a theory which would seem to be more consistent with the actual evidence of how financial markets operate than is the EMH, which has been amended over the past 40 or so years by subtly redefining what is meant by the term “efficiency” in the light of new, and contradictory, empirical evidence.
3. The Empirical Literature on Frontier Markets and Informational Efficiency
The striking feature of the empirical literature on informational efficiency and frontier markets is the paucity of material available. Very little research—whether on a European or other basis—has been carried out into frontier markets and specifically the efficiency of how these markets process information; an example of such an early study, albeit of an emerging market, namely Russia, can be found in Kratz (1999).
3.1. Informational Efficiency and Frontier Markets
Three studies have examined the informational efficiency of stock markets in the Indian subcontinent and surrounding areas. The first study, by Mobarek and Keasey (2000), examines the stock market in Bangladesh, while the second and third studies (Akbar and Baig, 2010; Rehman and Khidmat, 2013) examine the stock market in Pakistan. All three studies, to varying degrees, have issues with respect to data availability, relatively small sample size, thin trading, and methodologies employed, and therefore must be treated with a degree of caution.
Mobarek and Keasey (2000) examine the Bangladesh stock exchange for weak-form market efficiency from 1988 to 1997. The study uses both indices and the returns on 30 actively traded companies, testing using both parametric [autocorrelation, autoregression, and autoregressive integrated moving average (ARIMA)] and nonparametric (runs test) techniques. Irrespective of test and sample used, the authors find evidence that the Bangladesh stock market, for the period examined, is not weak-form efficient. Akbar and Baig (2010), for the neighboring Pakistan stock market from 2004 to 2007, again find evidence supportive of informational inefficiencies at the semistrong-form level. However, evidence contrary to this is provided by Rehman and Khidmat (2013), albeit for a larger sample period: 2001–11.
A number of studies have examined whether African markets are weak-form efficient, the majority of which have focused on the South African market. The evidence, based on weekly data, supports the argument for this market being weak-form efficient in the 1980s and 1990s (Dickinson and Muragu, 1994; Smith et al., 2002; Jefferis and Smith, 2005; Magnusson and Wydick, 2002). Interestingly, one study, that of Appiah-Kusi and Menyah (2003), finds that the South African market appears not to be weak-form efficient from 1990 to 1995. This study also provides results that would appear to indicate that the markets of Botswana, Ghana, and the Ivory Coast all seem to exhibit evidence that weak-form efficiency did not hold during the early and mid-1990s. The findings for Ghana and Botswana are consistent with those of Magnusson and Wydick (2002).
Evidence supportive of weak-form efficiency in other African markets has been provided for Kenya, Zimbabwe, Egypt, Morocco, and Mauritius during the 1990s (Appiah-Kusi and Menyah, 2003), and for Kenya during the 1980s (Kiweu, 1991; Dickinson and Muragu, 1994). However, evidence contrary to weak-form efficiency in African markets has been provided for Egypt, Morocco, and Mauritius during periods in the 1990s (Smith et al., 2002; Bundoo, 2000; Asal, 2000).
A criticism of the African studies discussed is that there are questions as to the quality of available data, the issue of thin trading, and the use of weekly return data in many of the studies. These issues are partially addressed by Miambo and Biekpe (2007), who examine weak-form efficiency in 11 African stock markets (Egypt, Kenya, Zimbabwe, Morocco, Mauritius, Tunisia, Ghana, Namibia, Botswana, BVRM [Bourse Régionale des Valeurs Mobilières regional stock exchange], and Ivory Coast) using daily data from 1997 to 2002, with sample size varying between 9 and 54 stocks. The issue of thin trading is addressed by calculating the returns on a trade-to-trade basis, and adjusting for the variability in trade interval lengths. This adjustment is achieved by weighting the trade-to-trade returns by the number of days between trades. Miambo and Biekpe (2007) observe thin trading in all markets, particularly in Namibia and Botswana, with all but a few of the markets exhibiting evidence contrary to weak-form market efficiency (the exceptions are Kenya and Zimbabwe). However, methodologically, the econometric procedures used to examine whether the markets exhibit evidence of weak-form efficiency are the usual serial correlation and runs tests used in the prior studies.
Jarrett (2010) examines whether four small Pacific-Basin markets (Singapore, Malaysia, South Korea, and Indonesia) exhibit evidence of weak-form efficiency. The data for the study covers the period 1985–2000, with sample size ranging from 390 stocks (Indonesia) to 900 (Malaysia). The study uses a simple autoregressive conditional heteroscedasticity (ARCH) model to examine the returns for each market, and the author concludes that the markets examined exhibit predictable properties, contrary to the weak-form definition of efficiency.
De Groot et al. (2012) investigate the returns of more than 1400 stocks in 24 frontier markets (including Eastern Europe) for the period 1997–2008, examining the impact of value, momentum, and size-based investment strategies in each market.
The value strategy involves grouping the stocks in each market into portfolios based on three characteristics: namely, historical ratios: (1) book-to-market ratio, (2) earnings-to-price ratio, and (3) dividend-to-price ratio. For the value strategy to be successful, stocks with high ratios should, on average, have higher returns than stocks with low ratios.
The momentum strategy involves grouping the stocks into portfolios based on past returns. Stocks with higher past returns are expected to have higher future returns.
The size effect involves grouping the stocks into portfolios based on market capitalization of equity. Stocks with relatively low market capitalization should, based on the strategy, have higher returns than stocks with relatively large market capitalizations.
Each of the strategies outlined is inconsistent with both weak-form and semistrong-form market efficiency, as defined earlier.
De Groot et al. (2012) find that portfolios constructed on either a value or a momentum basis, even after adjusting for transaction costs, generate statistically significant excess returns of between 5% and 15%, depending on the frontier market portfolio examined. Evidence is also found for a significant size effect. The study further investigates whether the excess returns can be explained by risk factors, but finds no supportive evidence for such an effect. Such evidence is therefore contrary to information efficiency holding in the frontier markets investigated.
Okicic (2014) examines the stock returns, and associated volatility, for a number of indices in Eastern European frontier and emerging markets (Bosnia and Herzegovina, Bulgaria, Croatia, Czech Republic, Hungary, Macedonia, Montenegro, Poland, Romania, Serbia, Slovakia, and Slovenia) from 2005 to 2013, employing a simple ARIMA and ARCH methodology. The results of the ARCH analysis provide evidence that there is a “leverage effect,” namely that negative shocks in the market increase volatility proportionality more than positive shocks do—a result inconsistent with semistrong-form efficiency. Further, based purely on the ARIMA analysis and associated Ljung-Box statistics, the author rejects the null hypothesis that there is no autocorrelation in the returns, and concludes that the markets exhibit signs of weak-form inefficiency.
To summarize, a number of frontier market studies have found evidence supportive of informational inefficiencies in frontier markets, depending on the time period examined, both in Eastern Europe and elsewhere. However, all of these studies, to varying degrees, have issues with respect to data sources, samples, thin trading, and/or methodologies employed. With respect to data, many of the studies which examine the African markets, for instance, do not use standardized validated data sets, but instead rely on material supplied directly by the relevant stock market, without external validation. However, most of these studies do recognize the issue. With respect to methodology, a number—but not all—of the studies that test for weak-form efficiency use statistical approaches with relatively low power; and, as far as the authors of this present study are aware, there are no studies that examine the relative time series available for evidence of long memory. This study, therefore, attempts to address these issues by using more recently developed econometric techniques and a standardized and externally validated database.
4. Methodology
The concept of market efficiency is initially explored using the standard tests for basic martingale behavior of the various stock market indices in each of the frontier economies. The tests aim to establish the value of d, the size of the lag operator Ld [eg, L1(Xt) = Xt–1] needed to transform the index into a stationary ergodic series. The EMH implies that d should be 1 for the level series of the indices and zero for their returns. If d is 1, the level series follow a random walk and the movement of their returns is completely random. If this is the case, then no advantage can be gained by studying the past movements of the index.
First, the hypothesis that the level indices follow a random walk is explored by using the ADF–GLS test of Elliott et al. (1996). This is a more powerful test than the traditional augmented Dickey–Fuller (ADF) test. The null hypothesis is H0: d = 1 against the alternative H1: d = 0. The test allows for two possibilities: nonzero mean and trend stationarity. In both cases, for the market to be efficient, the null hypothesis should not be rejected. The probability distribution of the test statistic without a trend follows that derived in MacKinnon (1996) and can easily be found. For the trend-stationary version of the test, the probability distribution of the statistic is more complex, and ready figures are not available except for the standard 10, 5, and 1% values. As the power of unit root tests is generally low, the alternative of testing the null hypothesis that H0: d = 0 against the alternative that H1: d = 1 is conducted using the traditional Kwiatkowski–Phillips–Schmidt–Shin (KPSS) test (Kwiatkowski et al., 1992). For the markets to be efficient, the null hypothesis should not be acceptable for the level series and should not be rejected for the returns.
Given the knife-edge nature of the traditional I(0)/I(1) testing processes, this study then considers whether the series might be fractionally integrated. That is, 
and can take noninteger values; thus:
and can take noninteger values; thus:
When 
, the series is mean reverting and is said to have long memory. That is, the current value of the series depends on previous values of the series going back many time periods. Thus, within a market, knowledge of previous values can give an advantage and the market would not be efficient. When , it has been shown that the test of whether a series has long memory is either a simple z-score test or t-test of the hypothesis H0: d = 0 against the alternative, H1: 0 < d < 0.5. It is the first if the local Whittle estimator (Robinson, 1995) of d is used in the construction of the test statistic, and the second if the GPH estimator (Geweke and Porter-Hudak, 1983) is used. If the markets are efficient, then the value of d should be zero.
, the series is mean reverting and is said to have long memory. That is, the current value of the series depends on previous values of the series going back many time periods. Thus, within a market, knowledge of previous values can give an advantage and the market would not be efficient. When , it has been shown that the test of whether a series has long memory is either a simple z-score test or t-test of the hypothesis H0: d = 0 against the alternative, H1: 0 < d < 0.5. It is the first if the local Whittle estimator (Robinson, 1995) of d is used in the construction of the test statistic, and the second if the GPH estimator (Geweke and Porter-Hudak, 1983) is used. If the markets are efficient, then the value of d should be zero.It is well known that tests concerning the size of d can give misleading results if the series exhibit either structural breaks or bubble behavior (Smith, 2005; Qu, 2011). In particular, the tests might suggest that the series is I(1) or I; 0 < d < 1 when actually the series is I(0) with either a structural break or bubble behavior for the period. Identifying what is happening offers many challenges methodologically. For example, if there is bubble behavior, d is likely to be greater than 1 and deriving a test statistic becomes problematic.
Recent approaches to the identification of bubbles have included techniques based on fractional integration (Cunado et al., 2005; Frammel and Kruse, 2011; for details) and on sequential unit roots testing (Phillips et al., 2011). In the first approach, the return series is tested for nonstable fractional integration (0.5 < d). This is an implication of bubble behavior in the levels series. In the second approach, the levels series is tested for a mildly explosive root using a right-tailed unit root test.
In Phillips and Yu (2011) the issue of dating the time line of financial bubbles is explored. They modify the technique proposed by Phillips et al. (2011) and provide a methodology for identifying bubble behavior with consistent dating of bubbles’ origination and collapse. The paper also provides a methodology for testing for bubble migration, which could be used to examine possible contagion of individual stock prices. The methodology is based on the analysis of the mildly explosive stochastic process developed in Phillips and Magadalinos (2007a,b). The methodology is to recursively estimate:
To test for an explosive root, the critical values of the standard Dickey–Fuller test are obtained for the right-tailed alternative hypothesis H1: δ > 1 rather than the normal left-tailed test H1: δ < 1. The regression in the first recursion uses τ0 = ⌊nr0⌋ observations for some fraction r0 of the total sample, where ⌊…⌋ denotes the integer part of the argument. Subsequent regressions build on this original data using successive observations giving a sample of size τ = ⌊nr⌋ r0 ≤ r ≤ 1. The standard Dickey–Fuller t-test can be written as:
where 
is the least squares estimate of δ based on the first τ observations, 
is the corresponding estimates of σ, and 
.
is the least squares estimate of δ based on the first τ observations, is the corresponding estimates of σ, and .To test for the existence of a bubble, Phillips and Yu (2011) suggest the 
test that compares the supremum statistics 
with the right-tail critical values obtained from the limit distribution 
where W is a standard Wiener process, and 
is a demeaned Wiener process. Using simulation, they obtain a 5% critical value of 1.5073 for a sample size of 100.
test that compares the supremum statistics with the right-tail critical values obtained from the limit distribution where W is a standard Wiener process, and is a demeaned Wiener process. Using simulation, they obtain a 5% critical value of 1.5073 for a sample size of 100.To explore the time line of the bubble, they suggest that the start of the bubble can be identified by 
:
:
is the right-sided 100 βn% critical value of the limit distribution of 
statistic based on τs = ⌊ns⌋ observations and βn the size of the one-sided test. Similarly, assuming the existence of 
, they date the collapse of the bubble by 
where:
γ ln(n)/n is used so that the duration of the bubble is nonnegligible. For practical implementation, they set the critical value sequence for using an expansion rule: 
. The value of the constant C can be varied to make the test more or less conservative. In Phillips and Yu (2011) a value C = 5 is used to give a conservative test; they suggest that a value of about 100 gives the asymptotic 5% level.
using an expansion rule: . The value of the constant C can be varied to make the test more or less conservative. In Phillips and Yu (2011) a value C = 5 is used to give a conservative test; they suggest that a value of about 100 gives the asymptotic 5% level.5. Overview of Eastern European Markets and Sample Selection
5.1. Eastern European Markets
As many of the prior studies of frontier markets have found, access to quality data can be problematic. This arises for a number of reasons, including nascent stock markets, questions as to the veracity of the data, and issues of incomplete data, often due to remoteness of the market. In an attempt to minimize these issues, this study focuses on Eastern European frontier markets, which came into existence after the fall of the Berlin Wall in 1989 and the subsequent collapse of the Communist system in Eastern Europe. These markets have changed radically over the subsequent 25 years, in terms of their political and economic systems, geopolitical relations, living standards, and degree of foreign direct investment (FDI). Several of these former Communist countries are now members of the European Union (EU), a move which has accelerated their integration into the broader capitalist system. These countries are Bulgaria, Croatia, the Czech Republic, Estonia, Hungary, Latvia, Lithuania, Poland, Romania, Slovakia, and Slovenia. Three of these former Communist countries, the Czech Republic, Hungary, and Poland, are defined as emerging markets by the principal index providers (Dow Jones, Morgan Stanley, FTSE, and Russell) as of 2015, and are therefore excluded from this study. The majority of the former Communist countries have attempted to build their comparative advantage primarily in manufacturing, and secondarily in services. All of them have well-educated labor forces, with relatively low pay scales compared to other Western European economies. One country not mentioned but also classified as a frontier market by all of the index providers is Ukraine, which has had—and sadly, continues to have—a troubled relationship with its former neighboring Communist partner, Russia. Ukraine is therefore also included within this study, along with two other European frontier markets that never fell within the domain of the Communist system, namely Cyprus and Malta. This, therefore, brings the total sample size to 12 countries, 10 of which are former Communist countries (Bulgaria, Croatia, Estonia, Latvia, Lithuania, Romania, Serbia, Slovakia, Slovenia, and Ukraine), and 2 noncommunist countries (Cyprus and Malta). The study includes the two former noncommunist states for comparative purposes.
5.2. Sample Selection
The sample for the study is taken from the widely available Bloomberg financial database. Bloomberg provides coverage of all developed, emerging, and frontier markets globally, with data from a number of verified sources. For the Eastern European frontier markets, data are available for most countries from the host stock market, Dow Jones, Morgan Stanley (MSCI), Nomura, and/or FTSE. In selecting our sample, we strove to obtain the data source with the longest period of complete coverage. To minimize issues of thin trading in individual companies, the study examines broad market indices for each of the countries, comprising large-cap, mid-cap, and small-cap companies. The final sample and periods covered are illustrated in Table 8.1.
Table 8.1
Sample Derived From Bloomberg Financial Database
| Country | Series | Bloomberg code | Number of firms in index | Start | Finish |
| Bulgaria | MSCI Bulgaria local | MSEIBGLP | 12 | 05/31/2002 | 06/22/2015 |
| Croatia | MSCI Croatia daily net TR USD | MSEICRUN | 15 | 05/31/2002 | 06/22/2015 |
| Cyprus | FTSE/Cyprus Stock Exchange 20 | CYSMFTSE | 20 | 12/01/2000 | 06/23/2015 |
| Estonia | MSCI Estonia daily gross TR USD | MSEIESUG | 7 | 05/31/2002 | 06/22/2015 |
| Latvia | NOMURA Central Eastern local | NCEELAL | 9 | 01/16/1996 | 07/22/2013 |
| Lithuania | NOMURA Central Eastern local | NCEELITD | 15 | 01/01/1996 | 07/22/2013 |
| Malta | Malta Stock Exchange | MALTEX | 43 | 01/02/1996 | 06/22/2015 |
| Romania | MSCI Romania daily gross TR USD | MSEIROUG | 15 | 11/30/2005 | 06/20/2015 |
| Serbia | SRX SERBAIN traded Intex EUR | SRXEUR | 7 | 12/08/2005 | 06/23/2015 |
| Slovakia | Dow Jones total stock market | DWSKD | 8 | 11/30/2006 | 06/22/2015 |
| Slovenia | MSCI daily gross TR USD | MSEISVUG | 10 | 05/31/2002 | 06/22/2015 |
| Ukraine | PFTS Stock Exchange Index | PFTS | 18 | 01/12/1998 | 06/22/2015 |
Data availability for the each of the Eastern European frontier markets studied varies. The maximum amount of continuous data is available for Latvia, Lithuania, and Ukraine, with 17 years each. Following these markets, the remaining Eastern European frontier markets predominantly have data availability beginning in the early 2000s until 2015. For Malta and Cyprus, the two non–former Communist countries, the time series available run from 1996 to 2015 (Malta) and from 2000 to 2015 (Cyprus). Therefore, all of these times series, for all of the frontier markets examined, cover the period running up to, during, and after the financial crisis that began in 2007.
6. Results
The time series plots in Fig. 8.1a and b suggest that the series have all exhibited either bubble behavior or major shocks in the period pre-2010. It was therefore decided to carry out the analysis for the total time periods and for the period from Jun. 2010 to 2015. Table 8.2 gives the results of the ADF–GLS and KPSS tests for the total time period of each series. As explained in the methodology, the sample distribution is available for the ADF–GLS test only without trend. For the other test statistics, only the 10, 5, and 1% critical values are available. The tables all use the following notation to highlight significant results: *, **, ***, meaning that given the current test statistic the null hypothesis cannot be supported at the 10, 5, and 1% levels, respectively. Therefore, for the level series, all except Romania seem to be clearly I(1) series, as the ADF–GLS test statistic is not significant, meaning that the null of d = 1 cannot be rejected; and the KPSS test statistic is highly significant, meaning that the null of d = 0 cannot be accepted. This result would suggest that all the markets, except perhaps Romania, are efficient. However, Romania could also be seen as efficient if it is accepted that it is trend stationary. Conducting the same inference on the returns series basically provides similar conclusions. At the 5% level of significance, the KPSS test statistic cannot reject the null hypothesis of stationary returns except for Malta, Slovenia, and Ukraine. The ADF–GLS test statistic rejects the null hypothesis that the returns are I(1) for every series except for Bulgaria and Cyprus.
Figure 8.1 Time series plots of the logged values of the indices. (a) Bulgaria to Lithuania; (b) Malta to Ukraine
Table 8.2
Integer d Level of Integration Tests for Full Period
| Country | Levels | Returns | ||||||
| ADF–GLS | KPSS | ADF–GLS | KPSS | |||||
| Trend | Trend | Trend | Trend | |||||
| Bulgaria | −0.01 (0.68) |
−1.63 (>0.1) |
19.28*** (0.00) |
2.98*** (0.00) |
−0.92 (0.32) |
−2.26 (>0.10) |
0.27 (>0.10) |
2.98*** (0.00) |
| Croatia | −0.36 (0.55) |
−1.2 (>0.1) |
14.80*** (0.00) |
4.37*** (0.00) |
−5.87*** (0.00) |
−7.91*** (0.00) |
0.21 (>0.10) |
0.07 (>0.10) |
| Cyprus | 0.90 (0.90) |
−1.03 (>0.1) |
19.83*** (0.00) |
7.10*** (0.00) |
−0.89 (0.33) |
−2.24 (>0.10) |
0.32 (>0.10) |
0.21** (0.02) |
| Estonia | 0.19 (0.74) |
−0.66 (>0.1) |
8.08*** (0.00) |
3.18*** (0.00) |
−4.00*** (0.00) |
−5.91*** (0.00) |
0.44* (0.06) |
0.16* (0.05) |
| Latvia | −0.07 (0.66) |
−1.06 (>0.1) |
11.02*** (0.00) |
3.22*** (0.00) |
−15.69*** (0.00) |
−15.4*** (0.00) |
0.25 (>0.10) |
0.11 (>0.10) |
| Lithuania | −0.35 (0.56) |
−1.59 (>0.1) |
27.42*** (0.00) |
3.87*** (0.00) |
−9.42*** (0.00) |
−11.4*** (0.00) |
0.18 (>0.10) |
0.19** (0.03) |
| Malta | 033 (0.78) |
−1.05 (>0.1) |
15.69*** (0.00) |
4.28*** (0.00) |
−6.85*** (0.00) |
−8.65*** (0.00) |
0.57** (0.03) |
0.21* (0.02) |
| Romania | −1.96** (0.05) |
−1.99 (>0.1) |
6.00*** (0.00) |
3.28*** (0.00) |
−4.92*** (0.00) |
−7.69*** (0.00) |
0.13 (>0.10) |
0.11 (>0.10) |
| Serbia | −0.65 (0.43) |
−1.57 (>0.1) |
19.69*** (0.00) |
2.43*** (0.00) |
−4.41*** (0.00) |
−6.13*** (0.00) |
0.29 (>0.10) |
0.28*** (0.00) |
| Slovakia | −1.07 (0.26) |
−2.40 (>0.1) |
7.81*** (0.00) |
2.91*** (0.00) |
−2.79*** (0.00) |
−5.19*** (0.00) |
0.04 (>0.10) |
0.03 (>0.10) |
| Slovenia | −0.27 (0.59) |
0.91 (>0.1) |
6.54*** (0.00) |
5.66*** (0.00) |
−3.24*** (0.00) |
−4.16*** (0.00) |
0.47** (0.05) |
0.16** (0.04) |
| Ukraine | −0.01 (0.68) |
−1.23 (>0.1) |
13.54*** (0.00) |
6.21*** (0.00) |
−6.05*** (0.00) |
−9.74*** (0.00) |
0.64** (0.03) |
0.28*** (0.00) |
Table 8.3 gives the results of applying the simple z- and t-tests for long memory using the local Whittle and GPH estimates of d, respectively. As the test is valid only if d < 0.5, only the return series are tested. The more powerful z-test using the local Whittle estimate suggests that all series, except that for Slovakia, could have long memory. The t-test based on the GPH estimator gives similar results but also suggest that the Latvian series doesn’t have long memory.
Table 8.3
Simple Fractional Integration Tests on the Returns Series for Full Period
| Country | z-test (local Whittle estimate) | t-test (GPH estimator) |
| Bulgaria | 5.39*** (0.00) |
4.02*** (0.00) |
| Croatia | 3.02*** (0.00) |
2.33** (0.02) |
| Cyprus | 2.31** (0.02) |
1.72* (0.08) |
| Estonia | 2.56** (0.01) |
2.16** (0.03) |
| Latvia | 1.96** (0.05) |
0.67 (0.49) |
| Lithuania | 5.05*** (0.00) |
3.48*** (0.00) |
| Malta | 6.23*** (0.00) |
5.45*** (0.00) |
| Romania | 5.08*** (0.00) |
3.75*** (0.00) |
| Serbia | 6.23*** (0.00) |
5.35*** (0.00) |
| Slovakia | −0.57 (0.56) |
−0.04 (0.97) |
| Slovenia | 4.12*** (0.00) |
3.54*** (0.00) |
| Ukraine | 4.92*** (0.00) |
4.84*** (0.00) |
The results of applying the Phillips–Yu recursive DF test, with =5, are given in Table 8.4 and Fig. 8.2a and b. Apart from Croatia and Latvia, these results suggest that all the other series have experienced bubble activity during the full time period. For all except Cyprus, this activity was around the time of the financial crisis during 2007–09. With this in mind, the I(1)/I(0) and long memory analysis was conducted on the series for the past 5 years only.
Table 8.4
Phillips and Yu Recursive ADF Test for Full Period
| Country | Max
|
Start
|
Finish
|
| Bulgaria | 4.4 11/20/2008 |
10/01/2008 | 07/27/2009 |
| Croatia | n/a | n/a | n/a |
| Cyprus | 2.38 09/04/2012 |
06/26/2012 | 10/05/2012 |
| Estonia | 1.39 06/04/2003 |
05/09/2003 | 06/18/2003 |
| Latvia | n/a | n/a | n/a |
| Lithuania | 5.10 1/20/1997 |
03/11/2004 04/11/2005 |
03/18/2005 04/27/2005 |
| Malta | 5.08 12/21/1999 |
05/14/1999 | 07/26/1999 |
| Romania | 3.62 02/18/2009 |
01/08/2009 | 04/02/2009 |
| Serbia | 2.89 04/19/2007 |
01/19/2007 | 05/21/2007 |
| Slovakia | 1.46 11/12/2008 |
11/07/2008 | 11/14/2008 |
| Slovenia | 1.39 08/09/2007 |
08/09/2007 | 08/10/2007 |
| Ukraine | 2.34 07/26/2007 |
08/22/2005 01/22/2007 |
09/14/2005 03/20/2008 |
Figure 8.2 Time series plots of the Phillips–Yu test statistic. (a) Bulgaria to Lithuania; (b) Malta to Ukraine
Tables 8.5 and 8.6 give the results of the I(1)/I(0) analysis and the long memory analysis for the period 2010–15. These results are quite different from those for the full data period given in Tables 8.2–8.4. The I(1)/I(0) analysis suggests that all the level series are I(1) and the corresponding returns are I(0), with a slight probability that Cyprus and Serbia might be more complex. The fractional integration analysis of the returns suggests that apart from Bulgaria and Serbia there is little probability that the markets have long memory.
Table 8.5
Integer d Level of Integration Tests for 2010–15
| Country | Levels | Returns | ||||||
| GLS–ADF | KPSS | GLS–ADF | KPSS Cv 0.46 |
|||||
| Trend | Trend | Trend | Trend | |||||
| Bulgaria | −0.40 (0.54) |
−1.63 (>0.1) |
3.06*** (0.00) |
2.98*** (0.00) |
−3.17*** (0.00) |
−2.86* (0.09) |
−0.32 (>0.10) |
0.21** (0.13) |
| Croatia | −1.42 (0.14) |
−1.24 (>0.1) |
7.55*** (0.00) |
4.38*** (0.00) |
−0.64 (0.44) |
−7.91*** (0.00) |
0.20 (>0.10) |
0.07 (>0.10) |
| Cyprus | 0.88 (0.90) |
−0.80 (>0.1) |
11.62*** (0.00) |
3.53*** (0.00) |
−3.80*** (0.00) |
−4.31*** (0.00) |
0.36* (0.09) |
0.10 (>0.10) |
| Estonia | −1.12 (0.23) |
−1.18 (>0.1) |
2.74*** (0.00) |
1.55*** (0.00) |
−0.74 (0.40) |
−4.22*** (0.00) |
0.20 (>0.10) |
0.10 (>0.10) |
| Latvia* | −1.32 (0.17) |
−1.49 (>0.1) |
2.32*** (0.00) |
1.03*** (0.00) |
−0.48 (0.51) |
−2.46 (>0.10) |
0.15 (>0.10) |
0.08 (>0.10) |
| Lithuania | −0.66 (0.43) |
−1.26 (0.00) |
2.53*** (0.00) |
1.13*** (0.00) |
−0.77 (0.38) |
−1.98 (>0.10) |
0.17 (>0.10) |
0.16** (0.04) |
| Malta | −0.60 (0.45) |
−0.47 (>0.1) |
3.19*** (0.00) |
1.96*** (0.00) |
−2.4* (0.04) |
−6.36*** (0.00) |
0.08* (0.08) |
−6.36*** (0.00) |
| Romania | −0.62 (0.45) |
−2.01 (>0.1) |
9.51*** (0.00) |
1.54*** (0.00) |
−0.35 (0.56) |
−2.09 (>0.10) |
0.06 (>0.10) |
0.06 (>0.10) |
| Serbia | −0.48 (0.51) |
−1.36 (>0.1) |
5.07*** (0.00) |
3.00 (0.00) |
−4.53*** (0.00) |
−1.90 (>0.10) |
0.41* (0.07) |
0.18** (0.03) |
| Slovakia | −2.29** (0.02) |
−2.28 (0.00) |
1.16*** (0.00) |
0.96 (0.00) |
−0.15 (0.63) |
−2.02 (>0.10) |
0.31 (>0.10) |
0.08 (>0.10) |
| Slovenia | −1.07 (0.26) |
−1.29 (0.00) |
2.76*** (0.00) |
2.56*** (0.00) |
−3.67*** (0.00) |
−1.97 (>0.10) |
0.14 (>0.10) |
0.09 (>0.10) |
| Ukraine | 0.14 (0.72) |
−1.30 (>0.1) |
10.31*** (0.00) |
2.96*** (0.00) |
−0.37 (0.55) |
−4.70*** (0.00) |
0.22 (>0.10) |
0.13** (0.09) |
*Only to 07/22/2013.
Table 8.6
Simple Fractional Integration Tests on the Returns Series for 2010–15
| Country | z-test (local Whittle estimate) | t-test (GPH estimator) |
| Bulgaria | 2.76** (0.01) |
1.44 (0.15) |
| Croatia | 0.60 (0.55) |
0.62 (0.54) |
| Cyprus | 0.51 (0.61) |
0.50 (0.62) |
| Estonia | 0.78 (0.43) |
0.55 (0.58) |
| Latvia* | −0.72 (0.47) |
−0.51 (0.61) |
| Lithuania* | 0.07 (0.94) |
0.73 (0.47) |
| Malta | 1.87* (0.06) |
1.22 (0.23) |
| Romania | 0.79 (0.43) |
0.08 (0.93) |
| Serbia | 2.45** (0.01) |
0.50 (0.61) |
| Slovakia | −0.31 (0.75) |
−0.91 (0.37) |
| Slovenia | 1.39 (0.16) |
0.68 (0.50) |
| Ukraine | 1.99** (0.05) |
1.96* (0.05) |
*Only to 07/22/2013.
7. Conclusions and Recommendations
In this study, 12 frontier financial markets, primarily in Eastern Europe, are investigated to determine whether they are informationally efficient and/or demonstrate evidence of “bubble” activity between the 1990s and 2015. Since the collapse of communism in the early 1990s, all of these markets have adopted a free market agenda and have fully embraced the capitalist system.
The evidence initially presented would appear to indicate that all of the market indices examined follow a random walk process, with the exception of Romania, and can therefore be deemed to be informationally efficient, if one assumes a simple definition of efficiency (ADF–GLS and KPSS tests). This result stands for both the level and returns data. However, when one examines the data for each of the markets from a long memory perspective, it would appear that, with the exception of Slovakia, all of the markets exhibit some evidence of long memory, and thus could be deemed to be showing signs of inefficiency (local Whittle and GPH estimators). A question then arises as to whether this long memory result is actually a result of “bubble” behavior within each of the markets. To test for this, the study applies the Phillips–Yu test and finds evidence of bubble behavior for all of the markets, with the exception Croatia and Latvia. With the exception of Cyprus, this bubble behavior is predominantly to be found around the time of the financial crisis of 2007–09. To determine if the financial crisis has adversely affected the results, the data from 2010 to 2015 are examined in isolation, with results quite different from those of the complete data series. The fractional integration tests for these data indicate that, with the exceptions of Bulgaria and Serbia, the frontier markets examined have little evidence of long memory, but that the initial findings have been strongly influenced by the bubble behavior detected in the 2000s. These findings are more consistent with the tenets of the AMH rather the EMH as espoused by the likes of Fama.
In order to examine the issue more fully, one suggestion for a subsequent study would be to examine the time series for individual share prices in each of the respective frontier markets—although issues of data availability and thin trading in such markets might inhibit such an analysis.
..................Content has been hidden....................
You can't read the all page of ebook, please click here login for view all page.