emerging markets ETFs - IEB

6 downloads 369 Views 2MB Size Report
Jun 3, 2016 - ... Str, Peristeri, Athens,. Greece, 12131. E-mail: [email protected] ... más que un fenómeno universal
RESEARCH ARTICLE

AESTIMATIO, THE IEB INTERNATIONAL JOURNAL OF FINANCE, 2017. 14: 2-31 © 2017 AESTIMATIO, THE IEB INTERNATIONAL JOURNAL OF FINANCE

DOI:10.5605/IEB.14.2

Evaluating the weak-form efficiency of emerging markets ETFs

Rompotis, Gerasismos 왘 RECEIVED : 왘 ACCEPTED :

23

APRIL

2016

3 JUNE 2016

Abstract This paper examines the weak-form efficiency of emerging markets ETFs and to that end several parametric and non-parametric empirical tests are applied. In particular, the autocorrelation and the serial correlation in ETF returns are tested and the randomness in the series of ETF returns is then evaluated by applying runs tests. Finally, three alternative types of variance ratio tests are used to evaluate whether the prices of ETFs follow a random walk, that is, whether the market in question is efficient in the weak form. Overall, the results of the tests reveal that weak-form efficiency is a fundspecific rather than a universal phenomenon. The majority of serial correlation tests used demonstrate that the pricing of most ETFs in the sample is efficient. On the other hand, the autocorrelation, runs and variance ratio tests provide evidence of inefficiency for some of the ETFs examined. Keywords: ETFs, Market efficiency, Weak-form, Emerging markets. JEL classification: G14.

Rompotis, G. Department of Economics, National and Kapodistrian University of Athens, Greece. 25 Ypsilantou Str, Peristeri, Athens, Greece, 12131. E-mail: [email protected]

2

AESTI

M AT I O

AESTIMATIO, THE IEB INTERNATIONAL JOURNAL OF FINANCE, 2017. 14: 2-31 © 2017 AESTIMATIO, THE IEB INTERNATIONAL JOURNAL OF FINANCE

Evaluación de la eficiencia débil de los fondos cotizados de mercados emergentes

Rompotis, Gerasismos

Resumen El objetivo de este artículo es el análisis de la eficiencia débil de los fondos cotizados de mercados emergentes. Para ello se utiliza una serie de contrastes empíricos de carácter tanto paramétrico como no paramétrico. En concreto, se aborda la cuestión de la autocorrelación y correlación serial de los rendimientos de los fondos cotizados objeto de este artículo. Posteriormente se contrasta la hipótesis de aleatoriedad en los rendimientos de dichos fondos mediante contrastes de rachas. Finalmente, se utilizan tres tipos alternativos de contrastes de ratio de varianza para determinar si los precios de los fondos cotizados de mercados emergentes siguen un paseo aleatorio o, en otros términos, para contrastar la hipótesis de eficiencia débil. En general, los resultados de los contrastes realizados revelan que la eficiencia, en su forma débil, en los fondos cotizados de mercados emergentes, es una cuestión específica de determinados fondos más que un fenómeno universal. En cuanto a resultados particulares, de los contrastes de correlación serial se deduce que la valoración de la mayoría de los fondos cotizados considerados en este artículo es eficiente, mientras que los contrastes de autocorrelación, rachas y ratio de varianzas proporcionan evidencia de ineficiencia en algunos de los fondos examinados. Palabras clave: Fondos cotizados, eficiencia de mercado, forma débil, mercados emergentes.

AESTI

M AT I O

3

n 1. Introduction Given the international investment community’s increasing interest in investing in stock markets of emerging economies in recent years, this paper aims to provide reliable empirical answers to the key question raised by Exchange Traded Funds (ETFs) investors about the pricing efficiency of these markets and the potential for gaining any substantial abnormal returns by exploiting any possible pricing inefficiencies.

AESTIMATIO, THE IEB INTERNATIONAL JOURNAL OF FINANCE, 2017 14: 2-31

Evaluating the weak-form efficiency of emerging markets ETFs. Rompotis, G.

When it comes to the efficiency of small and emerging markets, the relevant literature demonstrates that these markets can be easily controlled and suffer from thin trading; they are thus basically informationally inefficient. Numerous studies, such as those of Barnes (1986), Buttler and Malaikah (1992), Dickinson and Muragu (1994), and Omran and Farrar (2006), attest to the inefficiency in developing markets. LagoardeSegot and Lucey (2008) reveal heterogeneous levels of efficiency in the Middle Eastern and North African stock markets. The efficiency in these markets is affected by market depth and corporate governance factors but not by economic liberalization. Other explanations for the informational inefficiency of emerging markets include the illiquidity which affects the market’s capacity to accommodate orders (Chordia et al., 2005), the low degree of competition and the dominance of a few key players who can cause stock prices to deviate from their intrinsic value (Mobarek and Keasey, 2000), the lack of market transparency and the scarce corporate information, the limited auditing experience and weak regulation (Blavy, 2002), and the fragmentation of capital markets and the presence of political and economic uncertainties (El-Erian and Kumar, 1995).

4

This paper is an extension to our previous work on the pricing efficiency of developed markets ETFs. In particular, Rompotis (2011) examined the validity of weak-form market efficiency using data from US-listed ETFs, which mainly track indices from developed economies as well as some indices from emerging markets, such as Hong Kong, Taiwan and Singapore. In this paper, we exclusively focus on emerging markets using a sample of US-listed ETFs, which cover a wide range of country and regional emerging economies from South Africa, Asia, the Americas and Europe. Another difference from our previous work relates to the techniques used to assess market efficiency. From a methodological perspective, the main difference concerns the use of the runs test and variance ratio tests rather than the Augmented Dickey-Fuller (ADF) and the Phillips-Peron (PP) unit root tests. Overall, the results of the tests reveal that weak-form efficiency is a fund-specific rather than a universal phenomenon. In particular, the majority of serial correlation AESTI

M AT I O

tests used demonstrate that the pricing of most ETFs in the sample is efficient. On the other hand, the autocorrelation, runs and variance ratio tests provide evidence of inefficiency for some of the ETFs examined. In essence, these results are weaker than those in Rompotis (2011), who revealed that weak-form efficiency is a valid hypothesis when analysing developed markets ETFs. The rest of the paper is organized as follows: Section 2 describes the methodology employed to test the weak-form efficiency of emerging markets ETFs. Section 3 describes the sample used in our study and Section 4 discusses the results of the tests applied to examine the efficiency of the sample of ETFs under study. Lastly, a summary and conclusions are presented in Section 5.

n 2. Methodology This study seeks to assess the weak-form informational efficiency of trading for USlisted ETFs investing in emerging markets. The null hypothesis is that the prices of ETFs follow a random walk and, thus, the ETF market is efficient in the weak form. We examine efficiency using various types of parametric and non-parametric tests, which have been extensively used in the literature.

The next method we use to examine the weak form of market efficiency is the socalled “Wald-Wolfowitz Runs Test”. This is a non-parametric test which checks a AESTI

M AT I O

AESTIMATIO, THE IEB INTERNATIONAL JOURNAL OF FINANCE, 2017 14: 2-31

We then perform serial correlation testing. A common finding in time series regression analysis is that the residuals are correlated with their own lagged values. For the purposes of our investigation, the existence of statistically significant estimates of serial correlation implies that the daily returns of ETFs are not independent of their lagged values and, therefore, this market cannot be considered efficient in the weak form. The lack of significant serial correlation coefficients indicates that the null hypothesis of the random walk is valid and therefore that the market is efficient.

Evaluating the weak-form efficiency of emerging markets ETFs. Rompotis, G.

First, we perform a similar autocorrelation test to that described in Rompotis (2011). Autocorrelation is a non-parametric test for serial dependence in the time series of returns. Statistically, the absence of significance in autocorrelation coefficients in returns implies that the price series follow a random walk, which in turn means that the market is efficient in the weak form. The null hypothesis is that the autocorrelation coefficients are equal to zero (the market is efficient) whereas the alternative is that they deviate from zero (the market is inefficient). Following the approach of Awad and Daraghma (2009), we estimate autocorrelation for n1,2,3 lagged return estimates.

5

randomness hypothesis for a two-valued data sequence. More specifically, it is used to test whether the elements of the sequence are serially independent. In other words, the runs test examines whether the value of one observation influences the values taken by later observations. If there is no impact, the return generating process is random and, thus, the observations are independent.1 As a final step, we apply three alternative types of variance ratio testing in order to establish whether or not the returns of emerging markets ETFs are predictable. If returns can be predicted, market is inefficient in the weak form. The three types of variance ratio tests are used to examine the Random Walk Hypothesis (RWH) in ETF prices. These tests are the Lo and MacKinlay (1988) variance ratio test, the ranksbased and signs-based variance ratio of Wright (2000) and the Chow and Denning (1993) variance ratio test.2

AESTIMATIO, THE IEB INTERNATIONAL JOURNAL OF FINANCE, 2017 14: 2-31

Evaluating the weak-form efficiency of emerging markets ETFs. Rompotis, G.

n 3.The sample

6

This paper employs a sample of 40 iShares ETFs which are listed on the New York Stock Exchange but track country or regional indices from the Americas, Europe, Asia and South Africa. Table 1 presents the profiles of the sample with the symbol, name and benchmark name for each ETF. Moreover, each ETF’s inception date is provided along with its expense ratio, average daily trading volume, assets (in $000s) as of 30th April 2015, average intraday volatility calculated as the percentage difference between the highest and the lowest trading price on day t divided by the closing price on the same day, and trading frequency calculated as the number of trading days with non-zero volume as a percentage of total trading history (in days) for each fund. Nasdaq.com and us.iShares.com provided us with the necessary information. Another feature of the table (and of the tables to follow) is that it displays the data with ETFs grouped geographically. More specifically, four groups of emerging markets are considered. The first group, the South African one, contains only one fund. The second group, the Asian one, contains the most funds with 27 ETFs.3

1 2 3

More information on runs test can be found on: www.itl.nist.gov/div898/handbook/eda/section3/eda35d.htm.

The methodology applied in performing these three tests has been based on information found in Chen (2008).

It should be noted that the Asian group includes some ETFs which are not solely invested in stocks listed in Asian stock exchanges. An example is the iShares MSCI BRIC ETF (BKF), which invests in the MSCI BRIC Index, an index which covers stocks from Brazil, Russia, India and China. For this ETF, however, the exposure to Asia (i.e. India, China and Kong Kong) counts for almost the 59% of the total portfolio (according to information provided on the us.ishares.com). In this as well as other similar obscure cases where the total exposure to Asia dominates, we classify ETFs within the Asian group. A similar approach has been followed for the iShares MSCI Emerging Markets EMEA ETF (EEME), which covers several countries such as South Africa, Russia, Poland,Turkey, Greece and Hungary. The portion of European stocks in the portfolio is more than 58% and, therefore, this ETF has been included in the European group.

AESTI

M AT I O

The third group, which is the European one, contains five funds. The last group, the Americas, includes seven ETFs. With respect to ETF inception dates, Table 1 displays a wide range of “ages”. The oldest ETFs are those invested in the MSCI indices of Malaysia and Mexico. These funds, initially known as WEBs (World Equity Benchmark Shares), were created in March 1993 by Barclays Global Investors. The second-oldest ETFs are also invested in single country indices, namely the MSCI indices of Taiwan, Thailand and Brazil. On the other extreme, the youngest funds in the sample were created sometime in 2012. The majority of the new listings invest in indices which cover emerging Asian markets. The increasing interest in Asian stocks reflected in the listing of new ETFs is probably due to the robust growth prospects of these economies, at least at the time when the ETFs in question were listed. This is presumably also the case for the iShares MSCI Emerging Markets Latin America ETF (EEML), which is heavily invested (more than 55%) in Brazilian stocks.

AESTI

M AT I O

AESTIMATIO, THE IEB INTERNATIONAL JOURNAL OF FINANCE, 2017 14: 2-31

Going further, Table 1 presents an average daily volume of 2.4 million shares for the ETFs in the sample. The average volume of the South African ETF amounts to 298,000 shares, the average Asian volume is 2.95 million shares, the average European ETF is the least traded with a volume of 162,000 shares and, finally, the group of American ETFs has an average volume of 2.4 million shares per day. The above listed figures are very striking but they somehow misrepresent the actual outlook of these emerging markets ETFs’ trading activity. More specifically, if we exclude the three ETFs whose average daily volume exceeds 10 million shares and are sample outliers, and calculate the average volume of the rest of the sample, we obtain a figure of 553,000 shares per day (not clearly reported in the table). This figure is much lower than the average volume of the total sample and the mean of two out of the four groups. Consequently,

Evaluating the weak-form efficiency of emerging markets ETFs. Rompotis, G.

Regarding expenses, Table 1 reports an average sample expense ratio of 0.60%. The group means are also around 60 basis points (bps). In particular, the expense ratio of the South African ETF is 0.61% while for Asian and European ETFs it is equal to 0.61% and 0.60%, respectively. The average fee for the Americas ETFs is 0.55%. It should be noted here that there is a modest dispersion in the expense ratios of the European and the American groups but the corresponding variation in the Asian group is more marked. Specifically, the lowest expense ratio in the group is 0.18%, comparable to many US-listed ETFs tracking domestic indices. On the other hand, the highest expense ratio is almost 1% (0.93%), which is extremely high for an ETF. One last comment on expenses is that, on average and unsurprisingly, the expense ratios charged by emerging markets ETFs are much greater than those charged by domestically invested ETFs. Investors, especially those who are expense-averse, should take this into consideration when deciding on which ETFs to go with.

7

8

AESTI

M AT I O

iShares MSCI Emerg Markets Financials ETF

iShares MSCI Emerging Markets Materials

iShares MSCI Emer Markets Consumer Discr

iShares Emerging Markets Infrastructure ETF

iShares Emerging Markets Dividend ETF

iShares MSCI China ETF

iShares FTSE China ETF

iShares China Large-Cap ETF

iShares MSCI China Small-Cap ETF

iShares MSCI India ETF

EMFN

EMMT

EMDI

EMIF

DVYE

MCHI

FCHI

FXI

ECNS

INDA

MSCI BRIC Index

iShares MSCI BRIC ETF

iShares MSCI Emerging Markets Energy Cap

iShares MSCI Emerging Markets Asia ETF

EEMA

BKF

iShares MSCI Emerging Markets Small-Cap E

EEMS

EMEY

MSCI Emerging Markets Asia Index

iShares MSCI Emerging Markets Value ETF

EVAL

MSCI India Total Return Index(SM)

MSCI China Small Cap Index

FTSE China 25 Index

FTSE China (HK Listed) Index

MSCI China Index

Dow Jones Emerging Markets Select Dividend

S&P Emerging Markets Infrastructure Index

MSCI Emerging Markets Consumer Discretion

MSCI Emerging Markets Materials Index

MSCI Emerging Markets Financials Index

MSCI EM Energy 25-50 Index

MSCI Emerging Markets Small Cap Index

MSCI Emerging Markets Value Index

MSCI Emerging Markets Growth Index

iShares MSCI Emerging Markets Growth ETF

EGRW

MSCI Emerging Markets Index(SM)

MSCI Emerging Markets Minimum Volatility

iShares MSCI Emerging Markets ETF

iShares MSCI Emerging Markets Minim Vol.

EEMV

MSCI Emerging Markets Investable Market In

MSCI South Africa Index

Benchmark

EEM

iShares Core MSCI Emerging Markets ETF

iShares MSCI South Africa ETF

Name

IEMG

ASIA

EZA

Symbol

2/2/2012

28/9/2010

5/10/2004

24/6/2008

29/3/2011

23/2/2012

16/6/2009

8/2/2012

20/1/2010

20/1/2010

8/2/2012

12/11/2007

8/2/2012

16/8/2011

8/2/2012

8/2/2012

7/4/2003

18/10/2011

18/10/2012

3/2/2003

Inception Date

0.67%

0.61%

0.74%

0.74%

0.61%

0.49%

0.75%

0.67%

0.67%

0.67%

0.67%

0.67%

0.49%

0.67%

0.49%

0.49%

0.67%

0.25%

0.18%

0.61%

Expense Ratio

402,307

8,683

16,585,437

6,887

380,610

38,946

18,242

1,322

3,175

3,051

4,000

145,202

27,602

9,241

2,034

944

47,965,097

298,843

1,123,609

298,260

Volume

3,594,896

46,852

7,591,369

38,858

2,460,854

212,419

82,470

5,666

6,960

5,604

3,238

307,732

169,612

81,447

23,355

5,837

33,309,982

2,569,281

7,850,084

467,241

Assets ($000s)

1.022%

1.032%

1.769%

1.291%

1.015%

1.044%

1.204%

0.270%

0.859%

0.695%

0.399%

2.004%

0.742%

1.021%

1.288%

0.392%

1.745%

0.885%

0.927%

1.955%

Intraday Volatility

96.93%

98.44%

100.00%

96.58%

99.32%

99.88%

100.00%

45.56%

89.63%

72.81%

47.78%

100.00%

93.33%

92.26%

86.79%

50.99%

100.00%

99.55%

100.00%

99.89%

Trading Frequen.

This table presents the profiles of emerging markets ETFs, with their symbol, name, benchmark, inception date, expense ratio, average volume (in shares), assets as of 30th April 2015, average intraday volatility calculated as the percentage difference between the intraday highest minus the intraday lowest price divided by the closing price on day t, and the percentage trading frequency calculated as the days with non-zero volume as a percentage of the total trading history of each ETF.

l Table 1. Profiles of ETFs

AESTIMATIO, THE IEB INTERNATIONAL JOURNAL OF FINANCE, 2017 14: 2-31

Evaluating the weak-form efficiency of emerging markets ETFs. Rompotis, G.

iShares MSCI Thailand Capped ETF

iShares MSCI Turkey ETF

TUR

iShares MSCI Mexico Capped ETF

iShares MSCI All Peru Capped ETF

EWW

EPU

Grand Mean

Mean

iShares MSCI Brazil Small-Cap ETF

iShares MSCI Chile Capped ETF

iShares MSCI Brazil Capped ETF

EWZ

EWZS

iShares Latin America 40 ETF

ILF

ECH

iShares MSCI Emerg Markets Latin America

EEML

AMERICA

Mean

iShares MSCI Poland Capped ETF

iShares MSCI Russia Capped ETF

ERUS

iShares MSCI Emerging Markets EMEA ETF

EEME

EPOL

iShares MSCI Emerg Markets Eastern Europ

ESR

EUROPE

Mean

THD

MSCI Korea 25/50 Index

iShares MSCI South Korea Capped ETF

iShares MSCI Taiwan ETF

EWY

MSCI Philippines Investable Market Index

EWT

MSCI Malaysia Index

iShares MSCI Malaysia ETF

iShares MSCI Philippines ETF

EWM

EPHE

28/9/2010 12/11/2007 12/3/1996 19/6/2009

MSCI Brazil Small Cap Index MSCI Chile IMI 25/50 Index MSCI Mexico IMI 25/50 Index MSCI All Peru Capped Index

10/7/2000

MSCI Brazil 25/50 Index

25/10/2001

18/1/2012

MSCI Emerging Markets Latin America Index S&P Latin America 40 Index

26/3/2008

9/11/2010

25/5/2010

18/1/2012

30/9/2009

26/3/2008

20/6/2000

9/5/2000

28/9/2010

12/3/1996

5/5/2010

8/2/2012

18/11/2009

MSCI Turkey Investable Market Index

MSCI Russia 25 / 50 Index

MSCI Poland IMI 25/50 Index

MSCI Emerging Markets EMEA

MSCI Emerging Markets Eastern Europe Index

MSCI Thailand IMI 25/50 Index

MSCI Taiwan Index(SM)

MSCI India Small Cap Index MSCI Indonesia Investable Market Index

iShares MSCI India Small-Cap

iShares MSCI Indonesia ETF

SMIN

EIDO

CNX Nifty Index

iShares India 50 ETF

INDY

AESTIMATIO, THE IEB INTERNATIONAL JOURNAL OF FINANCE, 2017 14: 2-31

M AT I O

Evaluating the weak-form efficiency of emerging markets ETFs. Rompotis, G.

AESTI

9

191,414 2,420,585 2,442,726

0.51%

0.60%

2,467,357

189,489

28,166

12,704,833

1,360,691

2,146

0.55%

0.49%

0.61%

0.61%

0.61%

0.50%

0.49%

277,180 161,564

0.61%

336,593

168,481

1,901

23,665

2,950,329

212,079

7,600,647

2,339,314

205,673

1,712,395

391,807

8,691

163,041

0.60%

0.61%

0.61%

0.49%

0.67%

0.61%

0.61%

0.61%

0.61%

0.61%

0.49%

0.61%

0.74%

0.93%

1,952,344

931,280

188,563

1,893,493

274,660

47,096

3,347,980

756,273

10,893

201,948

469,594

294,720

207,335

9,291

28,801

2,596,215

429,336

4,370,345

4,408,973

517,168

459,335

518,126

54,888

973,114

1.327%

1.790%

1.781%

2.048%

2.023%

1.511%

2.539%

2.054%

0.577%

1.500%

2.196%

1.709%

1.520%

0.737%

1.336%

1.151%

1.573%

1.595%

1.642%

1.240%

1.405%

1.584%

1.260%

1.180%

92.69%

96.99%

100.00%

100.00%

100.00%

100.00%

100.00%

100.00%

78.91%

94.93%

100.00%

100.00%

100.00%

75.88%

98.79%

90.90%

100.00%

100.00%

100.00%

100.00%

100.00%

99.76%

84.69%

100.00%

we can infer that the high average volume figures are skewed by the trading intensity of specific ETFs such as that of the iShares MSCI Emerging Markets ETF (EEM), which has an average volume of 48 million shares, and the iShares China Large-Cap ETF (FXI), which has an average daily volume of 16.6 million shares. The next trading variable concerned is ETF assets as of 30th April 2015. The sample average is $1.95 billion, while the group means are $467.2 million, $2.60 billion, $201.9 million and $931.3 million for the South African, Asian, European and Americas group, respectively. As is the case with volumes, the averages do not tell the whole truth about the assets held by the ETFs examined in this study. In particular, the median of the entire sample is $284.7 million (not reported in the table) and shows that, as is the case with trading activity, the assets invested in emerging markets ETFs are mainly channeled to the specific funds mentioned above, which dominate the trading activity in the field.

AESTIMATIO, THE IEB INTERNATIONAL JOURNAL OF FINANCE, 2017 14: 2-31

Evaluating the weak-form efficiency of emerging markets ETFs. Rompotis, G.

On the question of intraday volatility, Table 1 provides an average figure for the whole sample of 1.327%. With respect to the individual groups, the South African ETF shows higher intraday volatility than the average of the entire sample, the Asian ETF group is slightly less volatile whereas the corresponding European and American ETF groups are more volatile than the sample average. A general comment on intraday volatility is that there is no specific trend within each individual group. For instance, in the case of Asian ETFs, there are funds which present modest or even low volatility while there are also funds showing a highly volatile intraday behaviour. This is undoubtedly linked to the special features of each underlying market and it is therefore hard to form particular expectations about volatility for a specific region or market.

10

When it comes to trading frequency, the average term presented in Table 1 is 92.69%. The group mean values are 99.89%, 90.90%, 94.93% and 96.99% for the South African, Asian, European and Americas group, respectively. These percentages indicate that, on average, there are just a few days on which investors do not trade with emerging markets ETFs. Nevertheless, by scanning through the individual ETFs, we observe that there are funds whose trading frequency is lower than 50%, while other funds present an absolute trading activity with no zero volume days. The descriptive statistics of ETF and index returns for the sample of ETFs examined in this study are provided in Table 2. This table presents the average daily return of each ETF and benchmark, the risk of each fund and index expressed in terms of the standard deviation of the returns, the extreme scores, and the coefficients of skewness and kurtosis over a period ending 30th April, 2015.4 4

Returns have been calculated in percentage terms with daily closing price data found on Bloomberg.

AESTI

M AT I O

l Table 2. Descriptive statistics of returns

This table presents the descriptive statistics of emerging markets ETFs and benchmarks’ daily returns (average return), standard deviation, minimum and maximum return records, skewness and kurtosis. ETF returns are calculated with closing trade prices. Symbol

Average ETF Index

EZA

0.058

0.065

IEMG

0.014

EEMV

0.032

EEM EGRW

Standard deviation

Minimum

ETF Index

ETF Index

Maximum ETF Index

2.253

1.895 -20.082 -12.686

0.020

0.997

0.736

-4.360

0.041

0.948

0.678

-4.895

0.050

0.048

2.045

1.324 -16.166

-9.484

0.021

0.023

1.425

0.833

-7.286

-4.099

EVAL

0.001

0.001

1.573

0.831

-6.044

-3.948

7.118

3.522

EEMS

0.012

0.021

1.200

0.811

-5.235

-5.085

5.803

3.934

-4.765

-3.416

Skewness ETF Index 0.122

-0.114

Kurtosis ETF Index

22.922

13.148

11.225

4.337

-3.920

4.341

2.868

-0.142

-0.185

1.145

1.474

-3.494

5.124

2.720

-0.017

-0.021

3.065

1.704

22.770 10.598

0.708

-0.380

16.270

8.672

0.175

-0.010

7.381

1.386

0.289

0.024

3.160

1.346

-0.092

-0.745

2.711

4.270

0.046

2.671

1.201

-0.002 10.946

10.819

ASIA

8.093

3.292

EEMA

0.027

0.033

1.071

0.845

5.187

3.540

0.298

BKF

0.006

0.006

2.252

1.731 -14.216 -11.230 20.630

14.477

0.381

EMEY

-0.042

-0.038

1.676

1.150 -13.692

-4.634 12.054

4.643

-0.393

0.078

14.566

1.816

EMFN

0.020

0.019

1.767

1.209

-6.579

7.767

5.759

-0.103

-0.206

3.097

2.538

EMMT

-0.016

-0.010

1.678

1.320 -10.968

-7.035

7.436

5.833

-0.268

-0.213

3.476

2.336

EMDI

0.022

0.023

1.211

0.921

-6.523

-4.058

7.003

3.428

0.297

-0.048

8.424

0.522

0.035

0.044

1.293

1.027

-8.271

-6.232

7.174

5.605

-0.133

-0.272

3.456

3.270

-0.019 -0.004

0.994

0.726

-4.161

-3.775

4.359

2.527

0.034

-0.099

0.834

1.252

1.489

1.337

-8.211

-6.106

9.368

6.765

0.183

0.097

4.070

3.133

EMIF DVYE

0.040

-8.808

MCHI

0.029

FCHI

0.041

0.045

2.291

1.885 -15.529 -12.332 20.393

15.146

0.725

0.447

13.623

9.151

FXI

0.067

0.069

2.341

1.943 -14.845 -17.919 20.270 15.450

0.589

0.082 10.240

10.124

ECNS

0.025

0.030

1.399

1.153

-7.932

-5.217

7.295

7.294

-0.030

-0.200

2.706

3.907

INDA

0.023

0.028

1.501

1.272

-6.253

-5.521

6.465

5.998

-0.020

0.024

1.327

2.232

INDY

0.026

0.026

1.667

1.376

-6.709

-5.905

13.452

6.630

0.274

0.084

4.126

1.612

SMIN

0.048

0.049

1.696

1.358

-5.237

-4.466

6.764

5.208

0.112

-0.305

1.120

1.112

EIDO

0.027

0.027

1.863

1.505 -11.958

-9.123

9.708

8.255

-0.083

-0.241

4.183

4.868

EWM

0.035

0.045

1.355

0.992 -11.622 -10.654

8.988

5.954

-0.149

-0.499

6.254

8.449

EPHE

0.051

0.055

1.387

1.220

-7.511

7.198

6.484

-0.234

-0.532

3.405

4.232

EWY

0.052

0.054

2.166

1.928 -13.823 -18.675

22.422

28.385

0.909

0.369 16.300

23.479

-8.046

0.034

1.815

1.454 -11.067

-7.881

14.155

8.580

0.294

-0.155

7.207

0.046

0.050

2.035

1.631 -11.657 -13.259

15.458

9.650

-0.041

-0.443

5.808

3.709 6.713

Mean

0.024

0.029

1.598

1.229 -9.196

-7.465 10.622

7.502

0.132 -0.122

5.984

4.642 3.070

EUROPE ESR

0.005

0.012

1.971

1.740

-9.884

-9.816

10.732

8.032

-0.242

-0.262

3.082

EEME

0.002

0.008

1.479

1.233

-8.650

-5.840

6.714

5.051

-0.251

-0.032

4.115

1.784

EPOL

0.021

0.031

1.922

1.687 -10.965 -10.281

9.013

7.407

-0.341

-0.362

3.571

3.437

ERUS

-0.021

-0.010

2.170

1.952 -12.006 -12.212

9.917

14.581

-0.427

-0.174

3.367

6.657

TUR

0.034

0.034

2.656

2.255 -14.950 -13.618 20.586 16.924

0.161

-0.018

6.582

5.748

Mean

0.008

0.015

2.040

1.773 -11.291 -10.353 11.392 10.399 -0.220 -0.170

4.144

4.139

-0.039

-0.028

1.394

1.262

5.002

-0.063

0.078

3.182

1.186

0.059

0.066

2.175

2.053 -19.467 -16.800 26.246 24.025

0.313

0.243

14.817

13.703

0.058

0.069

2.438

2.252 -19.628 -16.742

25.581

18.079

0.094

0.007

9.543

7.940

-0.052

-0.042

1.633

1.524

-8.406

6.747

6.799

-0.279

-0.158

2.209

2.303

AMERICA EEML ILF EWZ EWZS

-7.252

-8.880

-5.101

6.676

ECH

0.007

0.011

1.732

1.522 -12.072 -11.071

15.686

16.764

-0.025

0.060

EWW

0.057

0.058

1.840

1.674 -10.986 -10.194

21.473

16.176

0.500

0.124

11.201

8.116

EPU

0.025

0.036

1.421

1.292 -13.784 -14.110

6.532

7.165

-0.650

-0.971

7.329

12.152

Mean

0.016

0.025

1.805

1.654 -13.153 -11.775 15.563 13.430 -0.016 -0.088

8.307

8.459

Grand Mean

0.022

0.027

1.706

1.388 -10.422

6.292

5.239

AESTI

-8.711 11.890

M AT I O

9.042

0.062 -0.122

9.867 13.809

AESTIMATIO, THE IEB INTERNATIONAL JOURNAL OF FINANCE, 2017 14: 2-31

0.027

THD

Evaluating the weak-form efficiency of emerging markets ETFs. Rompotis, G.

EWT

11

The average daily ETF return of the sample is 0.022%, while the corresponding return for the benchmarks is slightly better at 0.029%. Moreover, the South African ETF performed better than the sample average (average return of 5.8 bps), the group of Asian ETFs slightly outperformed the sample average (2.4 bps average return), while the European and Americas groups performed slightly worse than the sample average. The same pattern applies to the corresponding benchmarks. When it comes to risk, Table 2 reports an average sample variance of 1.706 whereas the average risk figure of the tracking indices is 1.388. At the group mean level, the South African ETF is more volatile than the sample average, as are the European and American ETFs. On the other hand, the average Asian ETF is less volatile than the sample average. The same behaviour is displayed by the ETF benchmarks.5 Going further, the extreme scores in Table 2, i.e., the minimum and maximum returns, are indicative of the great volatility involved in trading with ETFs covering emerging markets. Finally, the table shows that the return series do not suffer from any skewness bias but they are subject to a kurtosis impact. In particular, the series seem to be leptokurtic.

n 4. Empirical results

AESTIMATIO, THE IEB INTERNATIONAL JOURNAL OF FINANCE, 2017 14: 2-31

Evaluating the weak-form efficiency of emerging markets ETFs. Rompotis, G.

4.1. Autocorrelation tests

12

Table 3 reports the results of the time series autocorrelation tests used to examine the efficiency of US-listed emerging markets ETFs. Presented in the table are the autocorrelation coefficient of each ETF in the sample, the corresponding Ljung-Box-Q statistics and the p-values which indicate the significance of autocorrelations. Autocorrelations are calculated by successively taking into account one, two and three lagged return observations. In regards to the 1st-order autocorrelation, the average coefficient of the entire sample is 0.086. Moreover, the Asian ETFs have a higher 1st-order autocorrelation than the sample average whereas the South African, the European and the American ETFs present lower autocorrelation estimates than the sample average. Going further, the autocorrelation for the majority of the individual ETFs is significant at 1% or better but the European ETFs present non-significant 1st-order autocorrelation (with the exception of one fund). Finally, with just a few exceptions, the individual autocorrelations are positive. Based on these results, we may conclude that the pricing of the majority of the examined ETFs is not efficient in the weak form, when the 1st-order autocorrelation is taken into consideration. In other words, the ETF returns as a whole cannot be 5

It should be noted that the comparison performed in return and risk among groups is just for information purposes without being absolutely valid given that the trading history of the various ETFs examined is not similar to each other and, thus, a reliable comparison of the groups’ performance and risk cannot be performed.

AESTI

M AT I O

l Table 3. Autocorrelation of ETF daily returns

This table presents the calculations of autocorrelations in ETFs’ daily returns along with the relevant q-statistics and the probabilities (p-values) indicating of the statistical significance of calculations. Symbol EZA

Autocorrelation (1 lag) Coef.

Q-Stat

Autocorrelation (2 lags)

p-value 0.090

Coef. -0.042

Q-Stat

p-value

7.678

0.022

0.032

2.883

IEMG

0.225

32.204

0.000

0.017

35.027

EEMV

0.163

23.508

0.000

0.054

29.054

EEM

0.152

65.182

0.000

-0.054

67.730

EGRW

0.160

20.689

0.000

0.039

EVAL

0.183

27.142

0.000

EEMS

0.149

20.701

EEMA

0.071

4.115

BKF

0.154

EMEY EMFN

Autocorrelation (3 lags) Coef.

Q-Stat

p-value

-0.033

11.210

0.011

0.000

-0.027

35.056

0.000

0.000

-0.033

29.163

0.000

0.000

0.014

67.732

0.000

23.996

0.000

-0.063

25.655

0.000

0.047

32.221

0.000

-0.051

32.808

0.000

0.000

0.070

28.306

0.000

-0.001

28.767

0.000

0.042

0.056

7.089

0.029

-0.061

9.305

0.025

44.599

0.000

-0.064

47.430

0.000

-0.012

48.916

0.000

0.190

29.233

0.000

0.039

33.620

0.000

-0.035

33.756

0.000

0.183

33.297

0.000

-0.008

33.977

0.000

-0.057

36.676

0.000

EMMT

0.185

33.975

0.000

-0.017

34.274

0.000

-0.037

35.521

0.000

EMDI

0.116

10.871

0.001

0.024

11.986

0.002

-0.068

14.964

0.002

EMIF

0.172

43.917

0.000

0.008

45.992

0.000

-0.046

48.027

0.000

DVYE

0.154

18.932

0.000

0.044

22.498

0.000

-0.010

22.531

0.000

MCHI

0.025

0.667

0.414

0.051

3.368

0.186

-0.047

5.372

0.146

FCHI

0.003

0.014

0.908

0.028

1.383

0.501

-0.069

9.457

0.024

ASIA

FXI

0.669

0.414

0.008

0.831

0.660

-0.031

3.446

0.328

0.098

11.145

0.001

0.091

22.597

0.000

-0.014

22.613

0.000

INDA

0.062

3.088

0.079

-0.028

3.572

0.168

0.005

3.574

0.311

INDY

0.083

9.372

0.002

-0.004

9.388

0.009

0.005

9.423

0.024

SMIN

0.133

14.294

0.000

0.000

14.547

0.001

0.057

17.305

0.001

EIDO

0.039

1.902

0.168

0.026

2.871

0.238

-0.093

13.116

0.004

EWM

0.099

27.672

0.000

0.006

28.378

0.000

0.005

28.512

0.000

EPHE

0.078

7.111

0.008

0.017

7.748

0.021

-0.073

13.281

0.004

EWY

-0.206

120.490

0.000

-0.029

121.120

0.000

-0.028

122.730

0.000

EWT

0.030

2.561

0.110

0.020

3.772

0.152

-0.030

6.057

0.109

THD

0.024

1.036

0.309

0.047

5.050

0.080

0.023

6.214

0.102

Mean

0.100

22.533

0.091

0.018

25.105

0.076

-0.029

27.036

0.040

ESR

0.033

1.554

0.213

0.012

1.804

0.406

-0.028

2.867

0.413

EEME

0.029

0.678

0.410

-0.020

0.976

0.614

-0.013

1.132

0.769

EPOL

0.017

0.379

0.538

-0.004

0.395

0.821

-0.018

0.810

0.847

ERUS

0.089

8.914

0.003

0.006

9.133

0.010

-0.048

11.479

0.009

TUR

0.038

2.571

0.109

0.032

4.554

0.103

0.023

5.704

0.127

Mean

0.041

2.819

0.255

0.005

3.373

0.391

-0.017

4.398

0.433

AMERICA EEML

0.073

4.443

0.035

0.033

5.673

0.059

-0.017

5.780

0.123

ILF

0.006

0.099

0.753

-0.074

15.543

0.000

0.016

16.202

0.001

EWZ

0.005

0.075

0.785

-0.051

7.299

0.026

0.011

7.618

0.055

EWZS

0.079

7.166

0.007

0.072

14.119

0.001

-0.013

14.123

0.003

ECH

0.125

29.443

0.000

-0.016

29.443

0.000

-0.023

30.563

0.000

EWW

0.102

29.394

0.000

-0.029

30.394

0.000

-0.024

32.745

0.000

EPU

0.084

10.490

0.001

-0.028

11.105

0.004

-0.013

11.554

0.009

Mean

0.068

11.587

0.226

-0.013

16.225

0.013

-0.009

16.941

0.027

Grand Mean

0.086

17.662

0.135

0.009

20.399

0.103

-0.024

22.044

0.086

AESTI

M AT I O

AESTIMATIO, THE IEB INTERNATIONAL JOURNAL OF FINANCE, 2017 14: 2-31

EUROPE

Evaluating the weak-form efficiency of emerging markets ETFs. Rompotis, G.

-0.016

ECNS

13

considered independent of their lagged values and, thus, the random walk hypothesis tends to be rejected. This finding implies that past returns do affect future returns, mostly in a positive fashion as indicated by the majority of 1st-order autocorrelation. As to the question of the 2nd-order autocorrelation, Table 3 reports an average autocorrelation coefficient of 0.009. The average estimated p-value is 0.103 while there are 28 ETFs with significant 2nd-order autocorrelations at 5% or better, indicating that the efficiency hypothesis is rejected for the majority of the ETFs in the sample. When it comes to the behaviour of individual groups, it is similar to that shown in the case of the 1st-order autocorrelation, namely the Asian ETFs are more serially dependent than the average fund of the sample while the remaining groups are less dependent, with the European ETFs being the most efficient group given that the returns of all but one European ETFs present non-significant 2nd-order autocorrelation. In essence, the results concerning the 3rd-order autocorrelation are similar to those of the 1st- and 2nd-orders. Specifically, the relevant average autocorrelation coefficient is equal to -0.024 while the corresponding average p-value is 0.086. In addition, there are 29 ETFs with significant 3rd-order autocorrelations at the 5% level or better. These estimates indicate that the null hypothesis, which assumes that the returns of ETFs are not affected by their lagged values, is a fund-specific phenomenon. As with the previous autocorrelation orders, the European ETFs are the most informationally efficient group among the various groups examined.

AESTIMATIO, THE IEB INTERNATIONAL JOURNAL OF FINANCE, 2017 14: 2-31

Evaluating the weak-form efficiency of emerging markets ETFs. Rompotis, G.

4.2. Serial correlation testing

14

The results of serial correlation testing are described in this section. Table 4 reports the results of the relevant correlogram prepared. The table contains the correlation coefficients, the corresponding Q-statistics for the statistical significance of the estimates and the p-values, which indicate the significance of estimates. Following the presentation of autocorrelations in Table 3, the correlogram estimates are presented for the 1st-, 2nd- and 3rd-order serial correlation and for the four ETF groups examined in this paper. Regarding the correlogram and the relevant Q-statistics estimates, the main inference that can be drawn from the results in Table 4 is that no matter what the order of the serial correlation may be, all 1st-, 2nd-and 3rd- order serial correlation estimates are insignificant at any acceptable level (the relevant p-values of the estimated Q-statistics of all funds are greater than 0.95, approximating unity) whereas the average serial correlations do not materially differ from zero. This is also the case for all the individual estimates. No meaningful differences are to be noted among the individual groups. Based on the results in Table 4, we can reach the conclusion that the ETF market under examination is efficient in the weak form. AESTI

M AT I O

l Table 4. Serial correlation of ETF daily returns (Q-statistics)

This table presents the calculations of serial correlations in ETF daily returns along with the relevant q-statistics and the probabilities (Ps) of the statistical significance of calculations. Symbol

Autocorrelation (1 lag)

Autocorrelation (2 lags)

Coef.

Q-Stat

p-value

Coef.

Q-Stat

p-value

0.000

0.000

0.994

-0.002

0.010

0.995

IEMG

0.000

0.000

0.999

-0.001

0.001

EEMV

-0.002

0.003

0.960

-0.001

0.003

EZA

Autocorrelation (3 lags) Coef.

Q-Stat

p-value

-0.003

0.035

0.998

1.000

0.010

0.070

0.995

0.998

-0.003

0.012

1.000

ASIA

0.000

0.000

0.984

-0.001

0.002

0.999

0.003

0.026

0.999

EGRW

-0.003

0.009

0.923

-0.004

0.020

0.990

0.021

0.375

0.945

EVAL

-0.001

0.001

0.972

-0.004

0.017

0.992

0.020

0.345

0.951

EEMS

0.002

0.005

0.943

0.005

0.024

0.988

0.003

0.030

0.999

EEMA

EEM

-0.003

0.008

0.928

-0.001

0.010

0.995

0.013

0.139

0.987

BKF

0.000

0.000

0.987

-0.001

0.002

0.999

-0.002

0.007

1.000

EMEY

0.000

0.000

0.996

-0.003

0.009

0.996

0.013

0.145

0.986

EMFN

0.001

0.001

0.978

-0.003

0.010

0.995

-0.003

0.021

0.999

EMMT

-0.001

0.000

0.987

-0.005

0.028

0.986

-0.002

0.033

0.998

EMDI

-0.003

0.008

0.928

-0.005

0.025

0.987

0.015

0.205

0.977

EMIF

-0.002

0.005

0.944

-0.004

0.029

0.985

0.007

0.111

0.991

DVYE

0.001

0.000

0.988

0.001

0.002

0.999

0.005

0.022

0.999

MCHI

-0.003

0.008

0.928

0.001

0.009

0.995

0.002

0.016

0.999

FCHI

-0.004

0.025

0.874

0.002

0.035

0.983

-0.006

0.093

0.993

FXI

-0.001

0.004

0.948

-0.001

0.006

0.997

-0.002

0.015

1.000

ECNS

0.000

0.000

0.992

-0.001

0.001

0.999

0.003

0.013

1.000

INDA

0.000

0.000

0.995

0.001

0.001

1.000

-0.002

0.004

1.000

INDY

0.002

0.004

0.949

-0.002

0.012

0.994

0.000

0.012

1.000 0.996

0.001

0.001

0.976

-0.003

0.008

0.996

0.008

0.059

EIDO

-0.009

0.095

0.758

0.005

0.122

0.941

0.000

0.122

0.989

EWM

0.000

0.000

1.000

0.001

0.001

0.999

0.003

0.026

0.999

EPHE

-0.002

0.004

0.950

-0.005

0.031

0.985

0.005

0.057

0.996

EWY

-0.001

0.001

0.976

-0.001

0.004

0.998

-0.004

0.057

0.996

EWT

-0.002

0.012

0.914

0.001

0.015

0.992

0.001

0.021

0.999

THD

0.000

0.000

0.993

0.001

0.001

0.999

0.003

0.013

1.000

Mean

-0.001

0.007

0.954

-0.001

0.016

0.992

0.004

0.076

0.992

ESR

0.000

0.000

0.986

-0.003

0.015

0.993

-0.002

0.019

0.999

-0.001

0.000

0.988

0.002

0.003

0.999

-0.004

0.018

0.999

EPOL

0.000

0.000

0.994

0.000

0.000

1.000

-0.004

0.018

0.999

ERUS

-0.001

0.001

0.980

-0.003

0.011

0.995

0.002

0.014

1.000

TUR

0.001

0.001

0.972

0.001

0.003

0.999

-0.004

0.029

0.999

Mean

0.000

0.000

0.984

-0.001

0.006

0.997

-0.002

0.020

0.999

-0.001

0.000

0.986

0.002

0.002

0.999

0.003

0.010

1.000

0.001

0.005

0.944

-0.003

0.039

0.981

0.000

0.039

0.998

EEME

AMERICA EEML ILF EWZ

0.001

0.003

0.956

-0.002

0.010

0.995

0.000

0.010

1.000

EWZS

0.000

0.000

0.992

0.001

0.001

0.999

0.004

0.017

0.999

ECH

0.001

0.002

0.965

-0.002

0.011

0.995

-0.004

0.041

0.998

EWW

0.000

0.000

0.996

-0.002

0.009

0.995

0.002

0.024

0.999

EPU

0.001

0.003

0.957

0.001

0.004

0.998

-0.006

0.053

0.997

Mean

0.000

0.002

0.971

-0.001

0.011

0.995

0.000

0.028

0.999

-0.001

0.005

0.962

-0.001

0.014

0.993

0.002

0.059

0.994

Grand Mean

AESTI

M AT I O

AESTIMATIO, THE IEB INTERNATIONAL JOURNAL OF FINANCE, 2017 14: 2-31

EUROPE

Evaluating the weak-form efficiency of emerging markets ETFs. Rompotis, G.

SMIN

15

l Table 5. Breusch-Godfrey serial correlation LM Test

This table presents the calculations of the Breusch-Godfrey serial correlation LM Test statistics along with the relevant probabilities (p-values) of the statistical significance of calculations. In addition, the F-statistic for the significance of the estimated AR(3) model is presented along with the corresponding p-values. Symbol

1st Order serial correlation

2nd Order serial correlation

3rd Order serial correlation

LM statistic

p-value

LM statistic

p-value

LM statistic

p-value

F-statistic

p-value

0.036

0.849

5.896

0.052

8.654

0.034

2.628

0.112

IEMG

0.001

0.971

0.669

0.716

6.340

0.096

1.437

0.302

EEMV

1.788

0.181

1.810

0.405

5.305

0.151

8.671

0.000

EEM

1.850

0.174

6.688

0.035

8.023

0.046

25.222

0.000

EGRW

1.992

0.158

7.620

0.022

9.736

0.021

8.949

0.000

EVAL

0.320

0.572

6.513

0.039

8.166

0.043

11.239

0.000

EEMS

0.827

0.363

3.335

0.189

3.573

0.311

8.991

0.000

EEMA

2.110

0.146

6.060

0.048

8.199

0.042

2.258

0.210

BKF

0.079

0.778

0.234

0.890

1.037

0.792

18.311

0.000

EMEY

0.006

0.937

2.066

0.356

3.101

0.376

11.673

0.000

EMFN

0.211

0.646

2.204

0.332

2.602

0.457

12.379

0.000

EMMT

0.197

0.658

6.517

0.038

7.048

0.070

12.112

0.000

EMDI

1.588

0.208

5.482

0.064

6.986

0.072

0.429

0.898

EMIF

1.106

0.293

8.222

0.016

8.296

0.040

16.261

0.000

DVYE

0.096

0.756

0.258

0.879

1.362

0.714

6.958

0.000

MCHI

3.738

0.053

5.189

0.075

5.201

0.158

1.829

0.140

FCHI

5.213

0.022

5.342

0.069

15.499

0.001

3.009

0.123

FXI

3.453

0.063

4.626

0.099

8.772

0.032

1.134

0.334

ECNS

0.096

0.757

0.144

0.931

1.512

0.680

6.973

0.000

INDA

0.038

0.845

0.406

0.816

0.461

0.927

1.275

0.282

INDY

9.718

0.002

15.114

0.001

19.415

0.000

2.185

0.231

SMIN

0.242

0.623

2.060

0.357

6.062

0.109

5.645

0.001

EIDO

10.518

0.001

10.836

0.004

10.843

0.013

4.705

0.003

EWM

0.000

0.994

1.468

0.480

3.002

0.391

9.391

0.000

EPHE

0.450

0.502

3.919

0.141

4.815

0.186

4.552

0.004

EWY

1.155

0.283

2.025

0.363

2.418

0.490

43.384

0.000

EWT

13.332

0.000

13.569

0.001

15.469

0.001

2.048

0.105

THD

0.153

0.695

0.374

0.829

4.258

0.235

1.981

0.115

Mean

2.233

0.433

4.546

0.304

6.574

0.239

8.630

0.102

ESR

0.308

0.579

3.770

0.152

7.297

0.063

0.981

0.401

EEME

0.322

0.571

0.903

0.637

5.956

0.114

0.397

0.755

EPOL

0.030

0.863

0.050

0.975

9.851

0.020

0.270

0.847

ERUS

0.265

0.607

4.481

0.106

4.493

0.213

2.828

0.119

TUR

0.834

0.361

1.314

0.518

3.160

0.368

1.857

0.135

Mean

0.352

0.596

2.104

0.478

6.151

0.155

1.267

0.451

EZA

AESTIMATIO, THE IEB INTERNATIONAL JOURNAL OF FINANCE, 2017 14: 2-31

Evaluating the weak-form efficiency of emerging markets ETFs. Rompotis, G.

ASIA

16

EUROPE

AMERICA EEML

0.457

0.499

0.858

0.651

1.774

0.621

1.872

0.133

ILF

7.743

0.005

8.886

0.012

9.109

0.028

5.436

0.001

EWZ

3.841

0.050

4.807

0.090

4.915

0.178

2.097

0.158

EWZS

0.178

0.673

2.183

0.336

4.448

0.217

4.468

0.004

ECH

1.201

0.273

10.887

0.004

10.894

0.012

10.173

0.000

EWW

0.021

0.885

5.283

0.071

5.634

0.131

11.280

0.000

EPU

1.275

0.259

1.301

0.522

131.807

0.000

3.791

0.010

Mean

2.102

0.378

4.886

0.241

24.083

0.170

5.588

0.044

Grand Mean

1.920

0.454

4.334

0.308

9.637

0.211

7.027

0.136

AESTI

M AT I O

The results of the Breusch-Godfrey Lagrange Multiplier serial correlation test used are shown in Table 5. This table presents the coefficients of LM test and the corresponding p-values indicating of the statistical significance of the estimated Breusch-Godfrey statistics. Furthermore, the table presents the 1st-, 2nd- and 3rd-order serial correlations. With respect to the serial correlation of first order, the average estimate is 1.920 while the corresponding average p-value is 0.454. The average term implies that, on average, there is no 1st-order serial correlation in residuals. The 2nd- and 3rd-order serial correlations behave similarly to the 1st-order serial correlation. The average LM statistics of the entire sample are equal to 4.334 and 9.637, respectively, and the corresponding average p-values are 0.308 and 0.211. There are, however, 11 and 14 out of 40 single serial correlations of 2nd- and 3rd-order respectively that are statistically significant at the 5% level or better. Finally, the estimates of the F-statistic which assesses the hypothesis that coefficients of the relevant autoregressive AR(3) model used are jointly equal to zero are also presented in Table 5. We should recall here that a rejection of the null hypothesis is indicative of serial correlation in ETF returns and consequently the efficient market hypothesis should be rejected. According to the results in Table 5, the hypothesis of no serial correlation in returns is rejected for a sufficient number of ETFs. More specifically, in 22 cases, the estimated F-statistics are statistically significant and, thus, informational efficiency is not the case for the respective ETFs. 4.3. Wald-Wolfowitz runs test

Overall, the results of the runs tests suggest that the returns of a significant number of funds in the sample are affected by their lagged values. In other words, for the 23 ETFs AESTI

M AT I O

AESTIMATIO, THE IEB INTERNATIONAL JOURNAL OF FINANCE, 2017 14: 2-31

We attempt to answer the key question about the randomness of the ETF return-generating process by applying the runs test, and the relevant z-statistics reveal that the returns of the majority of funds are not the outcome of a random process. In particular, there are 23 ETFs which present a statistically significant z-statistic at the 5% level or better. The majority of these significant z-statistics are found in the Asian group while the European group seems to be the most efficient based on the outcomes of the runs tests. These findings resemble those obtained by analysing the estimations of autocorrelation in Table 3.

Evaluating the weak-form efficiency of emerging markets ETFs. Rompotis, G.

The results of the Wald-Wolfowitz runs test, which assesses whether the returns of emerging markets ETFs are the result of a random process, are presented in Table 6. The table reports the test-statistic value, which is the median return for each ETF in the sample, the number of return observations (cases) which are lower than the test value, the number of cases which are greater than the test value, the number of runs, the estimated z-statistics and the relevant p-values corresponding to the z-statistics.

17

in question, the lagged prices can convey some information about the trends in concurrent returns of these ETFs, and, consequently, the pricing of these ETFs seems to be inefficient in the weak form.

l Table 6. Wald-Wolfowitz runs test

This table presents the calculations of the Wald-Wolfowitz Runs Test which is used to determine whether or not the time series of ETF returns are related to a random process. Symbol EZA

Test Value (median return)

Cases< Test Value

Cases>= Test Value

Total cases

No of runs

z-stat

p-value

0.140

1,415

1,415

2,830

1,410

-0.226

0.890 0.000

ASIA IEMG

0.031

316

316

632

254

-5.016

EEMV

0.037

443

442

885

402

-2.792

0.003

EEM

0.096

1,415

1,415

2,830

1,205

-7.934

0.000

EGRW

0.041

405

404

809

352

-3.764

0.000

EVAL

0.003

405

404

809

344

-4.327

0.000 0.000

EEMS

0.063

465

464

929

400

-4.300

EEMA

-0.015

405

404

809

371

-2.427

0.008

0.025

937

937

1,874

785

-7.071

0.000

AESTIMATIO, THE IEB INTERNATIONAL JOURNAL OF FINANCE, 2017 14: 2-31

Evaluating the weak-form efficiency of emerging markets ETFs. Rompotis, G.

BKF

18

EMEY

-0.037

405

404

809

357

-3.412

0.000

EMFN

0.067

496

496

992

440

-3.621

0.000

EMMT

-0.047

496

496

992

427

-4.447

0.000

EMDI

0.061

405

404

809

385

-1.442

0.074

EMIF

0.058

738

737

1,475

663

-3.933

0.000 0.001

DVYE

0.004

400

399

799

356

-3.151

MCHI

-0.004

512

513

1,025

522

0.531

0.760

FCHI

-0.002

862

861

1,723

865

0.121

0.943

FXI

0.014

1,329

1,328

2,657

1,328

-0.058

0.978

ECNS

0.062

577

576

1,153

523

-3.211

0.001

INDA

0.000

407

406

813

403

-0.316

0.931

INDY

0.047

684

683

1,367

642

-2.300

0.010

SMIN

0.082

405

404

809

372

-2.357

0.011

EIDO

0.043

627

626

1,253

632

0.254

0.902

EWM

0.004

1,415

1,415

2,830

1,315

-3.798

0.000

EPHE

0.022

577

576

1,153

546

-1.856

0.028

EWY

0.042

1,415

1,415

2,830

1,413

-0.113

0.947

EWT

0.009

1,415

1,415

2,830

1,425

0.338

0.942

THD

0.004

893

892

1,785

871

-1.065

0.147

Mean

0.026

698

697

1,396

652

-2.647

0.248

ESR

0.013

701

701

1,402

690

-0.641

0.683

EEME

0.000

412

412

824

408

-0.349

0.929

EPOL

0.005

620

620

1,240

646

1.420

0.071

ERUS

-0.040

562

561

1,123

547

-0.925

0.089

EUROPE

TUR

0.118

893

892

1,785

871

-1.065

0.147

Mean

0.019

638

637

1,275

632

-0.312

0.384

AMERICA EEML

-0.031

412

412

824

405

-0.558

0.755

ILF

0.090

1,415

1,415

2,830

1,312

-3.911

0.000

EWZ

0.092

1,415

1,415

2,830

1,364

-1.955

0.025

EWZS

0.012

577

576

1,153

548

-1.738

0.038

ECH

0.001

937

937

1,874

831

-4.945

0.000

EWW

0.109

1,415

1,415

2,830

1,317

-3.723

0.000

EPU

0.062

737

737

1,474

683

-2.866

0.002

Mean

0.048

987

987

1,974

923

-2.814

0.117

Grand Mean

0.032

759

759

1,518

716

-2.324

0.258

AESTI

M AT I O

4.4. Variance ratio tests 4.4.1 Lo and MacKinlay (1988) variance ratio test The results of the Lo and MacKinlay (1988) variance ratio test on whether the ETF prices follow a random walk are presented in Table 7. The table has two panels: Panel A contains the results from the first version of the variance ratio, which assesses the random walk hypothesis by assuming homoskedasticity-consistent standard errors; and Panel B reports on the second version of the ratio, which examines the random walk hypothesis by assuming heteroskedasticity-consistent standard errors. Table 7 lists the variance ratios and the z-statistics, and indicates their significance level. The test periods used are q1=2, q2=4, q3=8 and q4=16. As far as the first version of the test is concerned, i.e., the homoskedastic one, the results in Table 7 indicate that the individual statistics for the majority of ETFs reject the null hypothesis that the variance ratio is not statistically different from 1 at any q considered. There is a stronger rejection of the null hypothesis at the low test periods, namely at q1=2 and q2=4, for which 25 and 23 variance ratio estimates, respectively, are statistically significant at 5% or better. The rejection of the null hypothesis weakens as we move to the next test periods, that is, to q3=8 and q4=16. In these cases, the statistically significant variance ratios are equal to 16 and 11, respectively.

AESTI

M AT I O

AESTIMATIO, THE IEB INTERNATIONAL JOURNAL OF FINANCE, 2017 14: 2-31

Overall, the main conclusion that can be drawn from the analysis of the results of the Lo and MacKinlay (1988) conventional variance ratio is that the pricing of about half of the US-listed ETFs tracking stock indices from emerging economies considered is not a random walk and, consequently, for those specific funds the market cannot be considered efficient in the weak form. These results are in line with the results of the Wald-Wolfowitz Runs Test in the previous section, which also showed that informational efficiency is not universal in the ETF market examined but rather is a fund-specific phenomenon.

Evaluating the weak-form efficiency of emerging markets ETFs. Rompotis, G.

The results of the second version of the Lo and MacKinlay (1988) variance ratio in Panel B, i.e., the heteroskedastic one, resemble the results in Panel A. This means that the null hypothesis of a random walk is rejected for the majority of the funds as the variance ratios calculated are statistically different from unity in 23, 21, 12 and 8 cases at q1=2, q2=4, q3=8 and q4=16, respectively. A minor difference in results between the homoskedastic and heteroskedastic versions of the variance ratio is that the second one produces slightly less statistically significant estimates than the first one.

19

20

AESTI

M AT I O

1.108a

1.146

1.023

EMDI

EMIF

DVYE

MCHI

1.265

5.709

1.181

EMMT

a

1.287

4.120

a

1.275

3.755

1.111

ECNS

a

a

1.004

0.946

0.239

-1.802

1.006

0.965

FXI

1.066

FCHI

0.738

1.277a

6.604

1.172a

a

1.171b

3.069

a

1.277a

5.708

1.181a

EMFN

1.213a

1.288a

1.147a

EMEY

7.087

4.193

1.164a

1.138b

2.047

BKF

1.303

4.184

1.137

1.072b

EEMA

a

a

EEMS

1.315a

4.739

1.167a

EVAL

1.272a

4.363

1.153a

1.217a

7.972

EGRW

1.297

4.705

1.158

1.150a

EEMV

a

EEM

1.387a

5.428

0.968

Var. Ratio

a

0.797

z-stat

1.216a

1.015

Var. Ratio

z-stat

4.987

-1.499

0.086

1.131

4.340

5.691

2.595

4.459

4.663

4.371

4.939

2.095

4.941

4.788

4.131

6.160

4.724

5.196

-0.923

q2=4

1.360

a

0.917

0.936

1.013

1.331 a

1.245a

1.098

1.162

1.221b

1.279b

1.197b

1.062

1.409 a

1.297b

1.193

1.199a

1.260 b

1.396a

0.887 b

Var. Ratio

4.136

-1.454

-0.904

0.145

3.161

3.185

0.943

1.720

2.358

2.680

2.890

0.599

4.215

2.852

1.852

3.574

2.618

3.366

0.819

-2.041

1.570

a

0.935

0.924

1.081

1.428

b

1.278b

1.118

1.164

1.239

1.403b

1.182

1.101

1.594 a

1.416b

1.213

1.216b

1.355 b

1.537a

b

Var. Ratio

0.591

2.750

2.422

0.761

1.176

1.714

2.604

1.787

0.652

4.113

2.686

1.377

2.606

2.401

3.070

-2.184

z-stat

4.394

-0.763

-0.717

q4=16 z-stat

q3=8

Panel A: Homoskedastic random walk hypothesis

IEMG

ASIA

EZA

Symbol

q1=2

1.140

0.228

0.661

4.282

5.711

3.025

5.071

5.183

3.971

5.059

2.330

3.020

4.868

4.648

5.512

4.644

5.717

0.700

z-stat

1.113

b

2.517

0.966 -0.785

1.007

1.025

1.149

a

1.174a

1.111a

1.184

a

1.184a

1.150a

1.165a

1.075b

a

1.170a

1.156a

1.151a

1.161 a

1.220a

1.016

Var. Ratio

q1=2

1.312

1.281

a

0.948

1.007

1.072

1.297

a

1.282a

1.179b

1.273

a

1.285a

1.297a

1.217a

1.146b

a

1.325a

1.281a

1.219a

1.306

a

1.400a

0.970

Var. Ratio

z-stat

3.330

-0.664

0.121

1.016

4.513

4.806

2.619

3.772

4.189

4.203

3.294

2.256

3.480

4.864

4.269

4.014

4.505

5.287

-0.706

q2=4

1.281

1.377

b

0.921

0.943

1.027

1.354

a

1.257b

1.117

1.178

1.239b

1.301b

1.206

1.081

1.431

a

1.319a

1.214

1.205b

b

1.427a

0.891

Var. Ratio

z-stat

2.793

-0.648

-0.569

0.233

3.288

2.743

1.072

1.504

2.145

2.673

1.879

0.749

2.986

2.924

1.967

2.289

2.504

3.463

-1.584

q3=8

1.611

a

0.946

0.940

1.114

1.483

b

1.304b

1.160

1.200

1.278

1.457b

1.201

1.143

1.647

a

1.470b

1.259

1.229

1.402

b

1.613a

0.828

Var. Ratio

3.161

-0.320

-0.404

0.648

2.915

2.221

0.985

1.132

1.663

2.695

1.201

0.865

3.032

2.823

1.562

1.681

2.359

3.257

-1.671

z-stat

q4=16

Panel B: Heteroskedastic random walk hypothesis

This table presents the variance ratio estimates and test z-statistics of the Random Walk Hypothesis (RWH) based on the conventional variance ratio of Lo and MacKinlay (1988). Two versions of the test are applied. The first version assesses the homoskedastic random walk hypothesis, which assumes homoskedasticity-consistent standard errors while the second version assesses the heteroskedastic random walk hypothesis, which assumes heteroskedasticity-consistent standard errors. The test periods used are q1=2, q2=4, q3=8 and q4=16. a and b reflect statistical significance at 1 and 5% level, respectively.

l Table 7. Lo-MacKinlay variance ratio test

AESTIMATIO, THE IEB INTERNATIONAL JOURNAL OF FINANCE, 2017 14: 2-31

Evaluating the weak-form efficiency of emerging markets ETFs. Rompotis, G.

1.134

1.239a

3.042

3.799

1.367

3.422

a

1.082

1.134a

1.037

1.060

0.999

1.027

1.032

1.103

INDY

SMIN

EIDO

EWM

EPHE

EWY

EWT

THD

Mean

1.029

1.038

TUR

Mean

1.037

0.957

-0.208

3.233

2.614

0.996

b

1.078

1.142a

1.091a

1.084a

1.064

1.086

EWZS

ECH

EWW

EPU

Mean

Grand Mean

4.858

2.953

1.109a

6.158

1.142

1.086

1.076

1.180

1.209a

2.643

a

0.954

EWZ

-0.065

1.058

0.999

1.115

1.056

1.096b

1.126b

1.016

0.992

1.051

1.180

1.070

ILF

1.679

1.328

1.213

2.901

0.830

0.465

1.230

1.421

EEML

AMERICA

1.024

1.087a

ERUS

1.016

EEME

EPOL

1.033

ESR

EUROPE

1.093

2.022

b

0.994

1.141a

4.577

1.086a

-0.042

1.045

1.312

b

1.098

2.042

1.072b

INDA

1.761

2.532

1.714

1.562

3.106

4.840

3.263

-1.232

-1.306 -1.719 1.959 1.857 0.559 1.140 0.296 1.181

0.904 1.171 1.127 1.031 1.088 1.036 1.112

1.141

1.038

1.141

1.025

1.151

1.200

0.874

1.038 0.834b

0.548 -2.271

1.056

1.021

1.149

1.122

0.963

0.899

0.973

1.202

1.134

0.966

0.934

1.092

1.156

0.873

1.263

1.204

1.066

0.874b

0.375

1.652

1.004

-0.003

-0.694

-0.086

1.679

1.166

-0.521

-0.966

0.225

2.555

-1.274

2.513

2.373

1.322

1.025

1.116

2.157

1.122

1.089

1.000

0.928

0.993

1.156

1.082

0.971

0.946

1.020

1.142b

0.894

1.261b

1.190 b

1.137

2.253

0.306

-0.126

1.021

3.133

1.574

1.063

-0.175

1.680

4.022

0.859

3.636

2.639

1.488

AESTIMATIO, THE IEB INTERNATIONAL JOURNAL OF FINANCE, 2017 14: 2-31

M AT I O

Evaluating the weak-form efficiency of emerging markets ETFs. Rompotis, G.

AESTI

21

0.984

0.183

1.234

0.300

1.481

1.542

-1.520

-2.002

0.248

0.233

1.427

0.925

-0.299

-0.657

-0.230

1.449

1.282

-0.415

-0.796

0.711

1.887

-1.019

1.698

1.712

0.427

1.534

3.316

0.977

3.466

3.011

2.041

1.291

1.693

1.107

0.994

2.342

0.722

0.521

0.959

3.056

1.226

1.139

1.186

1.088

1.065

1.086

1.092a

2.543

1.854

1.735

3.652

3.650

1.148

1.090

1.081

1.112b

1.213a

b

2.376 1.143a

1.080

b

0.959

0.956

1.123

1.061

1.099

1.132

1.021

0.999

1.056

1.186

1.073

1.040

0.996

1.098

1.144b

1.050

1.248a

b

1.106

0.997 -0.111

0.999 -0.017

1.061

1.039

1.030

1.088b

1.025

1.019

1.034

1.105

1.034

1.027

1.000 -0.003

1.061

1.087a

1.039

1.136a

1.084 a

1.074b

2.182

1.300

0.818

2.257

2.974

2.689

-0.711

-0.718

1.791

0.954

1.761

1.863

0.323

-0.014

0.837

2.745

1.316

0.928

-0.085

1.176

2.824

0.677

3.303

2.604

1.489

1.125

1.045

1.098

1.036

1.135

1.185

0.909

0.878

1.075

1.037

1.124

1.102

1.011

0.944

1.003

1.171

1.090

0.976

0.951

1.032

1.148

0.904

1.283b

1.202 b

1.157

1.106

0.358

0.654

0.453

1.240

1.675

-0.965

-1.225

0.675

0.388

1.391

0.928

0.106

-0.517

0.031

1.532

0.982

-0.344

-0.625

0.236

1.841

-0.821

2.369

2.368

1.365

1.171

1.058

1.165

1.036

1.169

1.232

0.884

0.843

1.077

1.047

1.168

1.152

0.986

0.933

0.994

1.236

1.153

0.976

0.944

1.121

1.169

0.895

1.311

1.231

1.106

0.997

0.325

0.833

0.300

1.088

1.442

-0.814

-1.037

0.462

0.339

1.287

0.947

-0.086

-0.415

-0.039

1.391

1.123

-0.230

-0.469

0.604

1.478

-0.613

1.789

1.796

0.613

22

AESTI

M AT I O

1.261

1.267a

4.368

9.243

1.147

1.174a

1.160a

1.177a

1.079

EEMV

EEM

EGRW

EVAL

EEMS

EEMA

1.098b

1.171a

1.150a

0.998

1.008

1.010

1.094a

EMDI

EMIF

DVYE

MCHI

FCHI

FXI

ECNS

1.294

6.025

1.191

a

EMMT

1.291a

5.674

1.180a

1.281a

4.248

1.011

1.201a

3.182

1.018

0.498

0.323

1.010

1.271a

-0.051

1.151b

2.789

6.566

a

1.324a

EMFN

1.310a

8.314

4.827

1.192a

1.170a

1.130

2.234

b

BKF

1.350a

5.083

1.167a

EMEY

1.329a

5.021

1.279a

a

4.564

1.395a

5.695

0.973

Var. Ratio

a

0.808

z-stat

1.227a

1.015

Var. Ratio

3.641

0.311

0.395

0.169

4.250

5.558

2.294

4.952

4.901

4.925

7.184

1.979

5.699

4.997

4.239

7.584

4.150

5.306

-0.769

z-stat

1.210b

0.983

0.993

0.970

1.324a

1.252a

1.084

1.259 b

1.306a

1.364a

1.333a

1.062

1.472a

1.353a

1.231b

1.256a

1.266b

1.440a

0.890

Var. Ratio

z-stat

2.408

-0.298

-0.098

-0.326

3.099

3.267

0.808

2.757

3.260

3.501

4.871

0.597

4.859

3.397

2.224

4.611

2.680

3.742

-1.975

q3=8

Panel A: Ranks test

q2=4

IEMG

ASIA

EZA

Symbol

q1=2

1.319b

1.005

0.996

1.045

1.400b

1.275b

1.063

1.239

1.318b

1.458a

1.320a

1.070

1.602a

1.472a

1.204

1.252a

1.319 b

1.535a

0.805 b

Var. Ratio

2.153

0.329

2.567

2.397

0.410

1.710

2.276

2.957

3.151

0.450

4.169

3.047

1.318

3.048

2.457

0.064

1.153

1.099a

1.005

1.006

1.012

1.150a

1.171a

1.105a

1.191

a

1.183a

1.162a

1.187a

1.074

b

1.159a

1.171a

1.158a

1.165a

a

1.222a

1.014

-2.355

3.054

Var. Ratio

q1=2 z-stat

-0.039

q4=16

3.347

0.247

0.268

0.400

4.248

6.566

2.974

6.019

5.762

4.615

8.090

2.110

4.861

4.853

4.504

8.803

4.559

5.590

0.736

z-stat

1.217a

1.006

1.013

1.039

1.281a

1.271a

1.163b

1.283

a

1.285a

1.309a

1.276a

1.132

b

1.336a

1.320a

1.275a

1.247a

1.279

a

1.391a

0.964

Var. Ratio

z-stat

3.931

0.178

0.294

0.665

4.250

5.558

2.478

4.771

4.798

4.696

6.387

2.002

5.478

4.862

4.187

7.023

4.439

5.258

-1.030

q2=4

1.247b

0.972

0.973

0.991

1.324a

1.252a

1.096

1.209b

1.250b

1.323a

1.276a

1.057

1.452a

1.322a

1.208b

1.231a

1.265

b

1.415a

0.872

b

Var. Ratio

z-stat

2.840

-0.489

-0.379

-0.101

3.099

3.267

0.922

2.227

2.663

3.103

4.045

0.548

4.659

3.098

2.005

4.161

2.666

3.524

-2.306

q3=8

Panel B: Rank scores test

1.389a

0.995

0.977

1.063

1.400

1.275b

1.111

1.196

1.248

1.440b

1.264b

1.086

1.596a

1.449b

1.213

1.237b

1.346

b

1.535a

0.786

b

Var. Ratio

2.999

-0.063

-0.214

0.457

2.567

2.397

0.717

1.403

1.775

2.840

2.596

0.556

4.125

2.901

1.379

2.869

2.337

3.055

-2.589

z-stat

q4=16

This table presents the ranks-based and the signs-based variance ratio estimates and test z-statistics of the Random Walk Hypothesis (RWH) based on Wright’s (2000) non-parametric variance ratio. In general, the Wright (2000) variance ratio assesses the homoskedastic random walk hypothesis, which assumes homoskedasticity-consistent standard errors. Three versions of the test are applied, namely the ranks-based, the rank scores-based and the signs-based version of the test. The test periods used are q1=2, q2=4, q3=8 and q4=16. a, b and c reflect statistical significance at 1, 5 and 10% level, respectively.

l Table 8. Wright’s non-parametric variance ratio test

AESTIMATIO, THE IEB INTERNATIONAL JOURNAL OF FINANCE, 2017 14: 2-31

Evaluating the weak-form efficiency of emerging markets ETFs. Rompotis, G.

1.179

3.506

1.095

1.165a

INDY

SMIN

1.186

3.682

0.990

1.013

1.028

1.110

EWT

THD

1.056

1.035

1.032

TUR

Mean

1.032

1.167

2.081

2.711

4.611

3.563

1.039b

b

1.080

1.158a

1.095a

1.120a

1.084

1.093

EWZS

ECH

EWW

EPU

Mean

Grand Mean

3.269

5.046

6.842

1.153

1.122

1.175a

1.100b

1.243a

a

1.039

EWZ

2.063

1.055

1.039b

1.101

1.041

1.078

1.072

1.003

1.011

1.042

1.189

1.080

1.011

ILF

1.588

1.121

1.459

1.871

0.307

0.921

1.049

0.666

EEML

AMERICA

1.009

ERUS

1.032

EEME

EPOL

1.028

ESR

EUROPE

Mean

1.069

1.069

EPHE

EWY

1.006

2.351

b

-0.508

0.994

1.183a

1.026

6.001

1.029

1.113a

EWM

1.302a

EIDO

4.700

a

a

1.097

1.877

1.066

INDA

2.820

2.666

3.595

2.852

5.616

3.035

0.897

1.111

1.555

0.823

1.772

1.282

0.051

0.166

0.844

3.362

1.797

0.304

0.174

1.255

5.215

-0.107

4.591

3.537

1.473

AESTIMATIO, THE IEB INTERNATIONAL JOURNAL OF FINANCE, 2017 14: 2-31

M AT I O

1.257

1.949 3.036 0.486 2.339 1.098 1.573

1.170

1.027 1.180b 1.088 1.140

-0.357

0.980

1.207a

0.827 -0.591

1.085 0.967

0.265

1.272

0.567

-0.412

-0.304

0.201

2.069

1.666

-0.562

-0.338

-0.501

3.577

-1.344

3.590

3.214

1.201

1.018

1.089

1.050

0.965

0.969

1.016

1.186

1.117

0.969

0.981

0.956

1.199a

0.888

1.373a

a

1.125

Evaluating the weak-form efficiency of emerging markets ETFs. Rompotis, G.

AESTI

23

1.257

1.156

1.094

1.243b

1.013

1.235b

1.202

0.957

0.930

1.078

0.992

1.106

1.048

0.865

0.947

0.993

1.216

1.189

0.963

1.001

0.919

1.277a

0.877

1.423b

b

1.034

1.164

0.757

2.119

0.162

2.312

1.561

-0.518

-0.846

0.506

-0.021

1.019

0.363

-1.083

-0.349

-0.057

1.620

1.810

-0.453

0.009

-0.627

3.353

-0.992

2.736

2.161

0.218

2.413

5.798

1.176

4.378

3.257

2.090

1.167

1.093

1.083

1.120a

1.096a

3.255

3.503

4.600

5.084

7.035

2.711 1.163a

1.080

1.151

1.119

1.167a

1.103a

1.240a

a

1.007

1.373

1.039

1.109

1.048

1.089

b

2.063

1.657

1.281

1.554

1.012 1.096

0.526

1.004

1.040

1.185

1.071

1.014

2.466

0.731

1.127

3.650

1.205

0.870

0.999

1.074

1.179a

1.009

1.274a

1.150 a

1.103

1.026

1.039b

1.058

1.036

1.037

1.074b

1.015

1.025

1.030

1.109

1.029

1.016

0.992 -0.451

1.071 b

1.109a

1.033

1.154a

1.088 a

1.073b

2.757

2.561

3.428

2.917

5.565

3.035

0.196

1.111

1.678

0.965

2.018

1.716

0.218

0.065

0.809

3.279

1.594

0.410

-0.041

1.350

5.077

0.177

4.169

2.972

1.568 1.211

1.124

1.074

1.168b

1.018

1.176b

1.170

0.954

0.967

1.068

1.014

1.094

1.053

0.984

0.945

0.992

1.167

1.085

0.951

0.956

0.954

1.175a

0.870

1.317a

b

1.134

1.363

0.895

2.181

0.331

2.573

1.949

-0.836

-0.591

0.658

0.224

1.348

0.605

-0.195

-0.534

-0.106

1.832

1.213

-0.881

-0.784

-0.524

3.150

-1.550

3.044

2.638

1.290

1.142

1.074

1.218a

0.996

1.187a

1.202

0.925

0.930

1.058

0.989

1.104

1.059

0.899

0.916

0.966

1.202

1.126

0.930

0.944

0.944

1.222b

0.841

1.356b

1.215

1.047

1.022

0.555

1.903

-0.047

1.843

1.561

-0.903

-0.846

0.377

-0.037

1.003

0.453

-0.807

-0.546

-0.287

1.472

1.206

-0.847

-0.682

-0.432

2.688

-1.283

2.300

1.803

0.306

Panel C: Signs test q1=2 Symbol EZA

q2=4

q3=8

q4=16

Var. Ratio z-stat Var. Ratio z-stat Var. Ratio z-stat Var. Ratio z-stat 0.989 -0.564

0.969 -0.884

0.952 -0.871

0.944 -0.672

IEMG

1.193a

4.853

1.364a

4.890

1.448a

3.806

1.523a

EEMV

1.114a

3.395

1.235a

3.737

1.246b

2.477

1.231

1.560

EEM

1.145a

7.707

1.261a

7.425

1.321a

5.767

1.388a

4.693

EGRW

1.117a

3.340

1.214a

3.251

1.159

1.527

1.039

0.250

EVAL

a

1.152

4.324

1.277a

4.210

1.358a

3.447

1.448b

2.893

EEMS

1.132a

4.035

1.281a

4.577

1.379a

3.904

1.426a

2.952

EEMA

1.088b

2.496

1.167b

2.537

1.166

1.599

1.201

1.300

BKF

1.164a

7.115

1.289a

6.680

1.335a

4.908

1.358a

3.519

EMEY

a

1.122

3.481

a

1.201

3.063

b

1.219

2.110

1.293

1.891

EMFN

1.125a

3.937

1.232a

3.903

1.327a

3.478

1.361b

2.586

EMMT

1.131a

4.128

1.233a

3.920

1.198b

2.109

1.093

0.665

EMDI

1.048

1.371

1.078

1.184

0.981 -0.178

0.925 -0.487

EMIF

1.096a

3.671

1.144a

2.964

1.179b

2.319

1.264b

2.303

DVYE

1.089b

2.512

1.185b

2.799

1.210b

2.009

1.128

0.824

AESTIMATIO, THE IEB INTERNATIONAL JOURNAL OF FINANCE, 2017 14: 2-31

Evaluating the weak-form efficiency of emerging markets ETFs. Rompotis, G.

ASIA

24

2.984

MCHI

0.971 -0.937

0.984 -0.267

0.962 -0.406

1.026

0.188

FCHI

1.001

1.014

1.001

0.016

1.016

0.149

0.999 -0.021

1.003

0.056

1.024

0.279

1.300a

1.371a

4.261

1.555a

4.282

0.024

0.309

FXI

0.997 -0.136

ECNS

1.134a

4.565

INDA

1.009

0.246

0.974 -0.394

0.950 -0.486

INDY

1.058b

2.137

1.127b

2.516

1.196b

2.450

1.231

1.939

SMIN

1.105

2.988

1.173

2.631

1.214b

2.062

1.270

1.743

EIDO

0.998 -0.085

EWM

1.068a

EPHE

1.060

b

EWY

1.004

0.226

0.458

1.119

1.434

EWT

0.989 -0.602

0.982 -0.512

0.986 -0.254

1.053

0.635

THD

1.020

1.080

1.797

1.170b

2.428

1.319a

3.065

Mean

1.079 2.639

1.147 2.631

1.167

1.918

1.201 1.630

1.035

0.447

a

b

5.447

0.865 -0.876

0.980 -0.378

0.967 -0.396

1.050 0.403

3.609

1.093b

2.633

1.137b

1.266a

2.032

1.070

1.275

0.987 -0.149

1.030

0.854

0.828

1.025

2.466

3.218

0.949 -0.393

EUROPE ESR

1.010

0.374

1.026

EEME

1.007

0.209

0.962 -0.577

0.528

0.942 -0.559

EPOL

0.955 -1.590

0.916 -1.579

0.831b -2.016

ERUS

1.019

0.627

1.049

0.877

1.067

0.757

1.039

0.298

TUR

1.044

1.870

1.100b

2.265

1.145b

2.064

1.144

1.383

Mean

1.007 0.298

1.011 0.303

1.004 0.138

1.039

0.328

0.945 -0.356 0.762 -1.905

0.986 -0.050

AMERICA 1.058

1.657

1.109

1.678

1.068

0.658

1.058

ILF

EEML

1.066a

3.496

1.088b

2.512

1.055

0.985

1.084

1.011

EWZ

1.047b

2.519

1.057

1.608

1.037

0.661

1.022

0.265

EWZS

1.063b

2.150

1.101

1.826

1.114

1.309

1.134

1.035

ECH

1.107a

4.620

1.189a

4.371

1.195b

2.850

1.228b

2.241

EWW

1.083a

4.436

1.097b

2.763

1.086

1.551

1.147

1.780

EPU

1.068b

2.605

1.129b

2.645

1.165b

2.140

1.194

1.689

Mean

1.070 3.069

1.110 2.486

1.103

1.451

1.124 1.200

Grand Mean

1.066 2.342

1.119 2.227

1.130 1.544

1.154 1.287

AESTI

M AT I O

0.377

4.4.2 Wright (2000) ranks and signs variance ratio test The results of the non-parametric Wright (2000) ranks and signs variance ratio test are presented in Table 8. The table has three panels. The first panel contains the results of the test which are based on the ranks-based version of Wright’s variance ratio test. The second panel relates to the results obtained from the rank scoresbased version of the test. Finally, the third panel presents the results of the signs-based test. All the alternative versions of Wright’s variance ratio test are assumed to be homoskedasticity-consistent in contrast to the Lo and MacKinlay (1988) variance ratio test, which assumes both a homoskedasticity — and a heteroskedasticity-consistent test. Given that the homoskedasticity — and heteroskedasticity-consistent Lo and MacKinlay versions of variance ratio are similar, any heteroskedasticity there may be is minimal. Reported in the table are the variance ratios and the relevant z-statistics on the significance of estimates. As with the approach in the previous section, the test periods used are q1=2, q2=4, q3=8 and q4=16. ETFs at q1=2, q2=4, q3=8 and q4=16, respectively. Based on these estimates, we can conclude that market efficiency is not a valid hypothesis for the ETFs, which present Wright’s variance ratios that statistically differ from unity. By comparing these results to the Lo and MacKinlay (1988) variance ratios in the previous section, it is evident that the rejection of the null hypothesis by both tests becomes weaker as we move from the low to the high test periods, that is from q1=2 to q4=16.

The rejection of the efficient market hypothesis for a number of ETFs means that for these funds the past information which is incorporated in their historical returns AESTI

M AT I O

AESTIMATIO, THE IEB INTERNATIONAL JOURNAL OF FINANCE, 2017 14: 2-31

Overall, the main inference that can be drawn from analysing the results of Wright’s (2000) non-parametric variance ratio is that the pricing of a significant number of emerging markets ETFs is not a random walk and, consequently, the weak-form efficiency hypothesis can be rejected for those specific funds. We should recall that a similar conclusion was drawn in the previous section when the results of the conventional Lo and MacKinlay (1988) test were broken down.

Evaluating the weak-form efficiency of emerging markets ETFs. Rompotis, G.

The rank scores-based version of Wright’s test does not convey any new information about the validity of the random walk hypothesis with respect to the ETFs examined. This means that the results are quite similar to the ranks-based type of the test, both as far as the number of statistically significant estimates and the power of the RWH test as we move from the low to the high test periods are concerned. This is also the case for the signs-based version of the test, and so this too leads to the rejection of the random walk hypothesis for a sufficient number of ETFs in the sample.

25

may be valuable to investors seeking abnormal profits via trading with ETFs. However, such opportunities are rather fund-specific and, in any case, abnormal profits are not ensured and should be assessed against the costs of monitoring the whole ETF market to spot those funds whose pricing is not a random walk and may offer some chace of above-average market returns. 4.4.3 Chow and Denning (1993) multiple variance ratio test The results of the Chow and Denning (1993) multiple variance ratio test are displayed in Table 9. In particular, the table presents five alternative estimates of the Chow-Denning joint variance ratio test, two of which are based on the framework of the homoskedasticity-consistent and the heteroskedasticity-consistent Lo and MacKinlay (1988) variance ratio test, whereas the other three are based on the ranks-based, rank scores-based and signs-based versions of the Wright (2000) homoskedastic variance ratio test. Reported in the table are the relevant maximum absolute z-statistics and the corresponding p-values of the test statistics.

AESTIMATIO, THE IEB INTERNATIONAL JOURNAL OF FINANCE, 2017 14: 2-31

Evaluating the weak-form efficiency of emerging markets ETFs. Rompotis, G.

The results in Table 9 verify the inference about the rejection of the random walk hypothesis for a sufficient number of ETFs in the sample drawn by analysing the Lo and MacKinlay (1988) and the Wright (2000) variance ratio tests. For about half of the ETFs the RWH is rejected. More specifically, in 22, 20, 22 and 23 cases the alternative types of the Chow-Denning joint variance ratio test statistics are significant at 5% or better, thus failing to verify the weak-form efficient market hypothesis which assumes that the ETF prices follow a random walk.

26

As is the case with the results in the previous two sections, the statistical significance of a significant number of the Chow and Denning (1993) multiple variance ratio tests indicate that investors stand some chance of making a profit based on the historical prices of ETFs covering emerging stock markets, but these opportunities are fund-specific rather than abundant and universal.

AESTI

M AT I O

l Table 9. Chow-Denning variance ratio test

This table presents the maximum test |z|-statistics and the respective p-values of the Random Walk Hypothesis (RWH) based on the methodology of the multiple variance ratio test developed by Chow and Denning (1993). Five versions of the test are applied, which are successively based on the Lo-MacKinlay (1988) homoskedastic variance ratio test, the Lo-MacKinlay (1988) heteroskedastic variance ratio test, the Wright (2000) ranks-based variance ratio test, the Wright (2000) rank scores-based variance ratio test and the Wright (2000) signs-based variance ratio test. Lo-MacKinlay (1988)-based test Symbol EZA

Wright (2000)-based test

homoskedastic RWH heteroskedastic RWH homoskedastic RWH

homoskedastic RWH

homoskedastic RWH

Max |z|

P-value

Max |z|

P-value

Max |z|

P-value

Max |z|

P-value

Max |z|

P-value

0.694

0.931

1.761

0.278

0.395

0.956

0.379

0.965

2.645

0.028

ASIA IEMG

2.184

0.111

1.671

0.328

0.329

0.982

0.807

0.729

0.279

0.990

EEMV

5.428

0.000

5.717

0.000

1.083

0.544

2.589

0.026

0.876

0.694

EEM

7.972

0.000

4.644

0.000

2.355

0.055

5.590

0.000

0.403

0.962

EGRW

4.363

0.000

5.512

0.000

1.810

0.178

4.559

0.000

1.434

0.340

EVAL

1.230

0.627

4.648

0.000

5.695

0.000

8.803

0.000

0.635

0.847

EEMS

7.087

0.000

4.868

0.000

4.368

0.000

4.504

0.000

3.065

0.007

EEMA

5.708

0.000

2.330

0.077

9.243

0.000

2.110

0.071

7.707

0.000

BKF

1.802

0.257

5.071

0.000

0.921

0.656

5.762

0.000

3.481

0.001

EMEY

2.042

0.155

3.025

0.010

1.588

0.249

6.019

0.000

3.937

0.000

EMFN

3.042

0.009

5.711

0.000

2.063

0.122

2.974

0.007

4.128

0.000

EMMT

3.799

0.001

3.650

0.001

8.314

0.000

6.566

0.000

1.371

0.362

EMDI

1.312

0.569

0.785

0.896

5.674

0.000

1.373

0.343

3.671

0.002

EMIF

4.577

0.000

2.041

0.155

6.025

0.000

7.035

0.000

2.519

0.041

DVYE

4.858

0.000

3.011

0.010

2.789

0.011

2.090

0.085

2.150

0.070

MCHI

2.901

0.015

3.652

0.001

0.498

0.920

5.084

0.000

3.609

0.000

FCHI

1.421

0.491

1.735

0.292

1.877

0.146

4.600

0.000

4.436

0.001

FXI

4.724

0.000

1.534

0.414

4.700

0.000

2.413

0.054

2.032

0.099 0.000

ECNS

4.788

0.000

0.722

0.921

6.001

0.000

2.466

0.025

4.890

INDA

4.941

0.000

2.342

0.075

5.046

0.000

4.862

0.000

3.737

0.001

INDY

2.095

0.137

1.291

0.584

4.611

0.000

5.478

0.000

4.577

0.000

0.278

3.480

0.002

2.351

0.043

1.678

0.234

2.537

0.034

4.371

0.000

1.791

0.263

1.871

0.140

4.696

0.000

0.528

0.898

EWM

4.340

0.000

4.203

0.000

0.508

0.910

4.250

0.000

0.577

0.867

EPHE

4.987

0.000

1.016

0.773

4.925

0.000

3.931

0.000

0.309

0.982

EWY

2.271

0.089

1.225

0.631

3.641

0.001

0.489

0.927

0.877

0.677

EWT

1.719

0.301

0.965

0.804

3.537

0.001

1.550

0.283

0.884

0.741

THD

0.904

0.838

0.569

0.966

1.772

0.176

0.784

0.756

2.265

0.064

Mean

3.579

0.144

2.860

0.267

3.466

0.190

3.817

0.131

2.478

0.321

EUROPE ESR

5.709

0.000

0.959

0.807

4.564

0.000

1.127

0.534

3.340

0.000

EEME

3.069

0.009

0.521

0.975

5.021

0.000

0.731

0.786

4.324

0.000

EPOL

1.574

0.388

3.330

0.004

4.250

0.000

1.594

0.258

5.447

0.000

ERUS

2.157

0.118

1.316

0.566

3.035

0.007

2.018

0.125

2.516

0.023

TUR

0.966

0.803

0.625

0.952

1.344

0.371

0.881

0.673

2.016

0.109

Mean

2.695

0.264

1.350

0.661

3.643

0.076

1.270

0.475

3.529

0.026

AMERICA EEML

6.604

0.000

5.059

0.000

2.234

0.069

2.063

0.091

3.496

0.004

ILF

6.158

0.000

5.183

0.000

1.049

0.568

8.090

0.000

7.115

0.000

EWZ

3.233

0.005

3.466

0.002

6.566

0.000

3.257

0.006

4.620

0.000

EWZS

2.022

0.162

0.977

0.797

2.081

0.084

4.378

0.001

0.937

0.640

ECH

0.830

0.876

3.316

0.004

6.842

0.000

5.798

0.000

2.988

0.006

EWW

3.263

0.004

4.513

0.000

0.666

0.837

3.035

0.010

1.678

0.212

EPU

1.131

0.697

2.689

0.028

5.699

0.000

0.665

0.827

2.799

0.012

Mean

3.320

0.249

3.600

0.119

3.591

0.223

3.898

0.134

3.376

0.125

Grand Mean

3.351

0.197

2.773

0.290

3.433

0.201

3.427

0.195

2.771

0.243

AESTI

M AT I O

AESTIMATIO, THE IEB INTERNATIONAL JOURNAL OF FINANCE, 2017 14: 2-31

1.761

EIDO

Evaluating the weak-form efficiency of emerging markets ETFs. Rompotis, G.

SMIN

27

n 5. Conclusion This paper expands the existing literature on the efficient capital markets framework by investigating weak-form efficiency of ETF returns. More specifically, we use historical daily return data for a sample of 40 US-listed ETFs invested in emerging market indices. The data used covers the period from the inception of each ETF till 30th April 2015. In this study, we use a series of parametric and non-parametric tests provided by the finance literature and statistical economics to assess whether all the previously publiclyavailable information is reflected in the prices of ETFs or, alternatively, whether an ETF investor is likely to obtain above-market returns based on the released information. First, we apply various types of serial correlation testing. The estimated autocorrelations provide evidence for the rejection of the efficient market hypothesis. The efficient market hypothesis is likewise rejected for several funds in the sample via the other serial correlation tests used. Therefore, we could say that the serial correlation testing suggests that the efficiency of emerging markets ETFs is fund-specific rather than universal.

AESTIMATIO, THE IEB INTERNATIONAL JOURNAL OF FINANCE, 2017 14: 2-31

Evaluating the weak-form efficiency of emerging markets ETFs. Rompotis, G.

Another non-parametric test used in this study is the Wald-Wolfowitz Runs test. This tests assesses whether the ETF returns are the result of a random process. The findings obtained from this test indicate that the returns of a significant number of funds in the sample are affected by their lagged values. In other words, the lagged returns can convey some information about the trends in concurrent returns for the ETFs in question, and, consequently, the pricing of these ETFs seems to be inefficient in the weak form.

28

In the next step, we apply three alternative and supplementary variance ratio tests. The first method, a parametric one, is the well-known Lo and MacKinlay (1988) variance ratio test, the results of which demonstrate that there is a significant number of ETFs whose pricing is not a random walk. This pattern implies that for those particular funds the pricing is not efficient in the weak form. The second test applied is the Wright (2000) non-parametric variance ratio test. This tests provides similar results to the Lo and MacKinlay (1988) test. The last test applied to test the random walk hypothesis is the Chow and Denning (1993) multiple variance ratio test. The results of this test are in line with the results of the previous two tests. Overall, the results of the tests reveal that weak-form efficiency is a fund-specific rather than a universal phenomenon. In particular, the majority of serial correlation tests used demonstrate that the pricing of most ETFs in the sample is efficient. On the other hand, the autocorrelation, runs and variance ratio tests provide evidence of inefficiency for some of the ETFs examined. AESTI

M AT I O

As a concluding note we would add that along with informational efficiency there may be other forms of efficiency applicable in the case of ETFs. One obvious such example would be the tracking efficiency of ETFs, namely, their ability to perfectly replicate the performance of their underlying benchmarks. Another form of efficiency may concern the tracking ability of ETFs in comparison to the respective ability of their immediate index funds counterparts. A final type of efficiency, which probably relates more to the concept of the current study, concerns the efficient arbitrage execution on behalf of large institutional investors. Efficient arbitrage means that any gaps between the trading and net asset values of ETFs are just temporary and rapidly eliminated. On the contrary, non-efficient arbitrage execution implies that there are opportunities for informed investors to gain sufficient above-average returns in violation of the efficient capital market hypothesis. These issues could be dealt with in future research.

n References n Awad, I. and Daraghma, Z. (2009). Testing the Weak-Form Efficiency of the Palestinian Securities Market, International Research Journal of Finance and Economics, 32, pp. 7-17. n Barnes P. (1986). Thin Trading and Stock Market Efficiency: A Case of the Kuala Lumpur Stock Exchange, Journal of Business Finance and Accounting, 13(4), pp. 609-617. n Blavy, R. (2002). Changing Volatility in Emerging Markets: A Case Study of Two Middle Eastern Stock Exchanges, Revue Entente Cordiale Autumn-Winter, pp. 1-35. n Butler, K.C. and Malaikah, S.J. (1992). Efficiency and Inefficiency in Thinly Traded Stock Markets: Kuwait and Saudi

Business and Economics Research Journal, 7(12), pp. 97-106. n Chordia, T., Roll, R. and Subrahmanyam, A. (2005). Evidence on the Speed of Convergence to Market Efficiency, Journal of Financial Economics, 76, pp. 271-292. n Chow, K.V. and Denning, K.C. (1993). A Simple Multiple Variance Ratio Test, Journal of Econometrics, 58(3), pp. 385-401. n Dickinson, J.P. and Muragu, K. (1994). Market Efficiency in Developing Countries: A Case Study of the Nairobi Stock Exchange, Journal of Business Finance and Accounting, 21(1), pp. 133-150. n El-Erian, M.A. and Kumar, M.S. (1995). Emerging Equity Markets in Middle Eastern Countries, International Monetary Fund Staff Paper, 42, pp. 313-343. n Lagoarde-Segot, T. and Lucey, B. (2008). Efficiency in Emerging Markets-Evidence from the MENA Region, International Financial Markets, Institutions and Money, 18, pp. 94-105.

AESTI

M AT I O

AESTIMATIO, THE IEB INTERNATIONAL JOURNAL OF FINANCE, 2017 14: 2-31

n Chen, J.H. (2008). Variance Ratio Tests Of Random Walk Hypothesis Of The Euro Exchange Rate, International

Evaluating the weak-form efficiency of emerging markets ETFs. Rompotis, G.

Arabia, Journal of Banking and Finance, 16, pp. 197-210.

29

n Lo, A.W. and MacKinlay, A.C. (1988). Stock Market Prices Do Not Follow Random Walks: Evidence from a Simple Specification Test, Review of Financial Studies, 1(1), pp. 41-66. n Mobarek, A. and Keasey, K. (2000). Weak Form Efficiency of an Emerging Market: Evidence from Dhaka Stock Market of Bangladesh, Working Paper presented at the ENBS Conference held on Oslo, May 2000. n Omran, M. and Farrar, S. (2006). Tests of Weak Form Efficiency in the Middle East Emerging Markets, Studies in Economics and Finance, 23, pp. 13-26. n Rompotis, G.G. (2011). Testing Weak-Form Efficiency of Exchange Traded Funds Market, The IEB International Journal of Finance, 2, pp. 2-23.1 n Wright, J.H. (2000). Alternative Variance-Ratio Tests Using Ranks and Signs, Journal of Business & Economic

AESTIMATIO, THE IEB INTERNATIONAL JOURNAL OF FINANCE, 2017 14: 2-31

Evaluating the weak-form efficiency of emerging markets ETFs. Rompotis, G.

Statistics, 18(1), pp. 1-9.

30

AESTI

M AT I O

n

AESTIMATIO, THE IEB INTERNATIONAL JOURNAL OF FINANCE, 2017 14: 2-31

Evaluating the weak-form efficiency of emerging markets ETFs. Rompotis, G.

31

M AT I O

AESTI

32

AA EE ST II ST

MM AA TT IO IO

AESTIMATIO, THE IEB INTERNATIONAL JOURNAL OF FINANCE, 2017 14: 2-31

Evaluating the weak-form efficiency of emerging markets ETFs. Rompotis, G.