A Study of Market Efficiency in Emerging Markets Using Improved Statistical Techniques

ABSTRACT

This article helps resolve the current unsatisfying and inclusive studies covering the efficiency of stock markets in developing countries. Previous studies have used limited data and partial statistical tests. We use a large, unique data set, across 12 countries, and a comprehensive set of traditional and recent statistical methods as well as powerful multiple-break unit root and spectral analysis tests, many of which have never been used to evaluate the efficient market hypothesis (EMH) in emerging markets. Our results confirm the rejection of the EMH for emerging markets. Our findings have important implications for investors and policy makers, suggesting the possibility for excess profits in these markets.

KEYWORDS:

• Developing markets
• efficient market hypothesis
• BDS test
• multi-break unit root tests
• spectral analysis

JEL:

• C32
• G14
• G15
• O16

Notes

1. The 12 countries are: Brazil, Chile, Columbia, Hong Kong, India, Indonesia, Malaysia, Mexico, Philippines, Taiwan, Thailand. Selection of these countries is discussed in the data section.

2. These tests allow for more than one structural break in the data and, if not accounted for, can lead to misleading results (Lumsdaine and Papell 1997; Lee and Stracivich, 2003; Baltagi, 2008).

3. In addition, we looked at using GARCH-style unit root tests. After careful examination of our data, the lack of volatility clustering in our return series and formal tests of conditional heteroskedasticity suggest that this newer method is not appropriate for our data.

4. Our data spans a relatively long period of time using daily frequency. For most countries, the data start from the late 1980s and end in 2015.

5. The correlation coefficients among the WRDS indices and those of COMPUSTAT are between 0.95 and 0.98 for the countries in our sample, according to WRDS.

6. We also performed the single-break Zivot–Andrews test and found that most of the answers are similar to those of the multiple unit root tests. Due to its shortcomings and to converse space, we do not report the results which are available from the authors upon request.

7. To conserve space, we report the results for 2 lags since the results are essentially the same any of these methods.

8. We thank an anonymous subject editor for the suggestion to include this test.

9. We thank an anonymous referee for his/her suggestion to include this analysis into the article, along with other suggestions that improve the quality of our study.

10. We first examine the periodogram for each series and, when relevant, the smoothed (log) version of the data which has the advantage of stabilizing the variance and providing a closer look of the spectrum.

11. Results using other lags (0,4) are qualitatively similar and are thus not reported, to conserve space. Results are also invariant when using the AIC criterion.

12. When employing 3 breaks with the LP test, it should be noted that the critical values for Chile cannot be computed. This is not the case with most other countries.

13. To conserve space, we do not show the regular periodograms as they are strikingly similar. Specifically, they do not suggest the presence of a fixed cycle. Rather, they seem to indicate a stochastic cycle or a pseudo-cyclical pattern, similar to those of an autoregressive moving average model. The log periodograms with significantly higher signal to noise ratios (SNR) – about 3 times larger – provide a much closer look at the series and are shown in Table 6 instead.

14. These values can be seen from the graphical outputs or computed from the data. In many cases, the highest peaks for each series tend to happen in the periods following the Asian Financial Crisis, not surprisingly.