T h e R e v iew o f F in a n ce a n d B a n k in g

Vo lu m e 0 1 , Issu e 1, Yea r 2 0 0 9 , P a ge s 0 0 7— 0 1 3
S p rint IS S N 2 0 6 7 -2 7 1 3 o n lin e IS S N 2 0 6 7 -3 8 2 5

ADAPTIVE MARKETS HYPOTHESIS: EVIDENCE FROM ASIA-PACIFIC

ALEXANDRU TODEA, MARIA ULICI, AND SIMONA SILAGHI

Abstract. In this paper we investigate the profitability of the moving average strategy
on six Asian capital markets considering the episodic character of linear and/or nonlinear
dependencies, the period under study being 1997-2008. For each market, the most profitable
strategy from 15000 alternatives is selected. The main conclusion is that profitability of
moving average strategies is not constant in time; it is episodic showing when sub-periods of
linear and non-linear correlation appear. Thus, one can thus say that the degree of market
efficiency varies through time in a cyclical fashion over time and these statistical features are
in line with those postulated by Adaptive Markets Hypothesis (AMH) of Lo (2004, 2005).

1. Introduction
Numerous recent studies have used the Hinich—Patterson windowed-test procedure (1995)
to research the temporal persistence of linear and especially nonlinear dependencies on the
emergent stock markets. Thus, Lim and Hinich (2005) in Asian Stock Markets, Bonilla et al.
(2006) in Latin American Stock Markets and Todea and Zoicas-Ienciu (2008) in Central and
Eastern Europe stock markets are emphasizing the existence of different stock price behaviors,
namely long random walk sub periods alternating with short ones characterized by strong linear
and/or nonlinear correlations. All these studies suggest that these serial dependencies have an
episodic nature being also the main cause for the low performance of the forecasting models.
In this context, Lim, Brooks and Hinich (2006) computed the bi-correlation statistics of
Hinich (1996) in fixed-length moving sub-sample windows and found that nonlinear predictabil-
ity for Asian emergent markets follows an evolutionary time path. The results of this test of
evolving efficiencies are in line with those postulated by Adaptive Markets Hypothesis (AMH)
of Lo (2004, 2005) according to which the profit opportunities do exist from time to time. In
another study Lim, Brooks and Hinich (2007), relying on the same test, found that the cross-
country differences in nonlinear departure from market efficiency can be explained by market
size and trading activity, while the transient burst of nonlinear periods in each market can
be attributed to the occurrence of economic and political events. Lim (2007) compute the
bi-correlation statistics in rolling sample windows and ranking market efficiency from the per-
centage of windows in which these statistics reject the random walk hypothesis. By a similar
approach, Cajueiro and Tabak (2004) employed a rolling sample approach to compute the Hurst

Received by the editors May 11, 2009. Accepted by the editors November 23, 2009.
Keywords : episodic dependencies, bicorrelation test, technical analysis, adaptive market hypothesis.
JEL Classification : C15, G11, G14.
Alexandru Todea is Senior Lecturer at "Babes-Bolyai" University, Faculty of Economics Sciences and Business
Administration, Department of Finance.
E-mail: [Link]@[Link].
Maria Ulici is Phd Student at "Babes-Bolyai" University, Faculty of Economics Sciences and Business Ad-
ministration, Department of Finance.
E-mail: [Link]@[Link].
Simona Silaghi is Lecturer at Oradea University, Department of Economics.
E-mail: ssilaghi@[Link].
This paper is in final form and no version of it will be submitted for publication elsewhere.

c 00 9 T h e R ev ie w o f F in a n c e an d B a n k in g
°2

7
8 ALEXANDRU TODEA, M ARIA ULICI, AND SIM ONA SILAGHI

exponent, for testing the presence of long-term predictability in Asian equity markets. These
authors rank the markets using the medians of those computed rolling Hurst exponents. The
rolling bi-correlation test statistics is used by Lim Brooks and Kim (2007) to investigates the
effects of the 1997 financial crisis on the efficiency of eight Asian stock markets and found that
the crisis adversely affected the efficiency of most Asian stock markets, with Hong Kong being
the hardest hit.
The test of random walk of stock price is a test associated to the hypothesis of information
efficiency in weak-form and not a direct test. Thus, the acceptance of a random walk hypothesis
implies information efficiency, while the rejection is not synonym with information inefficiency.
That is why the random walk tests should be accompanied by direct tests of information
efficiency, such as tests of technical analysis. There have been a limited number of studies
conducted on technical trading rules in the Asian-Pacific markets. Bessembinder and Chan
(1995) were the amongst the first to report that the moving average and the trading range break-
out rule have predictive ability in the emerging markets of Malaysia, Thailand and Taiwan but
had less explanatory power in the more developed markets of Hong Kong and Japan. Ratner and
Leal (1999) adapted the methodology of study the specific of emerging markets to take account
of low trading and inflation. Thus, they examine the stock indices on 10 emerging markets
in Asia and Latin America and the SP 500 and Nikkei 225 indices. Profitable strategies are
focused mainly on markets from Mexico, Taiwan, Thailand and the Philippines. Tian, Wan
and Guo (2002) explored the profitability of technical trading rules in U.S. and China stock
markets. In the case of the U.S. the rules are not profitable after 1975, whereas for Chinese
markets are profitable across the 1990’s. Charlie and Keasey (2005) extend the analysis to
other developed markets — namely, the U.K., Hong Kong and Japan. The results show that
after 1990 the profitability of trading rules on the Chinese market decline progressively.
Most studies apply separately random walk and technical analysis tests to assess the degree
of efficiency of financial markets. One of the few exceptions is the study realized by Lagoarde-
Segot and Lucey (2006) who chose their combination path by constructing an index of efficiency
to assess the relative efficiency for seven emerging Middle-Eastern North African stock markets.
In this context it presents an interest the possibility of combined methodology windows, such
as the random walk tests with technical analysis tests. Such a request was made by Todea
(2008) on the Romanian market, as well as Todea et al. (2009) on six developed markets from
Europe. These studies demonstrate that the profitability of technical analysis strategies is an
episodic one and is given by the episodic behaviour of linear dependencies and, in particular,
by non-linear dependencies. According to these results, one can say that the degree of market
efficiency varies in a cyclical fashion over time and these statistical features are in line with
those postulated by the Adaptive Markets Hypothesis (AMH) of Lo (2004). This article aims
to investigate to what extent the obtained results on the European markets can be extrapolated
to the Asian capital markets.
The remainder of the paper is structured as follows. Section 2 discusses the methodology
employed. Section 3 provides a description of the data and Section 4 analyses the empirical
results and assesses their significance. Finally, Section 5 concludes succinctly the implications
of the results and the contribution of the paper in the area of research.

2. Methodology
The methodology applied in this paper is similar to that used by Todea (2008) on the Ro-
manian market or by Todea et al.(2009) on six developed European markets. This consists of
the study of the profitability of the moving average strategy over linear and nonlinear corre-
lation windows, as well as over non-correlation windows. These windows are identified using
an improved variant of the “windowed” methodology recommended by Hinich and Patterson
(1995), in the rolling sample approach. Such a different reading was used by (2007) in ranking
market efficiency for stock markets or by Lim et al. (2007) in investigating the effects of the
1997 financial crisis on the Asian stock markets. As a matter of fact, Todea and Zoicaş-Ienciu
 ADAPTIVE M ARKETS HYPOTHESIS 9

(2008) showed that the original Hinich-Patterson methodology can be suspected of inaccurate
identification of sub-periods exhibiting linear/nonlinear dependencies, because the test results
depend on how the first day of the sample is chosen.
Correlations identification is made using the portmanteau test (C) for linear correlations and
the bi-correlation test (H) for the nonlinear ones. In the rolling sample approach, C and H
statistics are computed for the first window of a specified length (n), and then the sample is
rolled forward eliminating the first observation and including the next one for re-estimation of
C and H statistics. The procedure is repeated until the last observation of the sample is used.
The return sample {R (t)} is considered to be the realization of a stochastic process, where t
(integer) is the time unit. In each window the data is standardized to have a sample mean of
zero and a sample standard deviation of one, as follows:

R (t) − mR
Z (t) = (2.1)
σR
where t takes values from 1 to n and mR , σ R are the mean and standard deviation within each
window. The null hypothesis is that {Z (t)} is the realization of a white noise process with null
correlations and bi-correlations, described by CRR (r) = E [R (t) R (t + r)] and CRRR (r, s) =
E [R (t) R (t + r) R (t + s)], where r and s are integers satisfying 0 < r < s < L with L being
the number of the lags. The correlations and bi-correlations are then given by: CRR (r) =
n−r
X n−s
X
−1/2 −1
(n − r) Z (t) Z (t + r) and CRRR (r, s) = (n − s) Z (t) Z (t + r) Z (t + s) for 0 ≤
t=1 t=1
r ≤ s.
C and H statistics are distributed according to a χ2 law of probability with L respectively
(L − 1) (L/2) degrees of freedom, having the following formulas:
L
X 2
C= [CRR (r)] (2.2)
r=1

and
L X
X s−1
H= G2 (r, s) (2.3)
s=2 r=1
1/2
where G (r, s) = (n − s) CRRR (r, s).
Lim and others (2007) draw attention upon the fact that the determination of H bi-correlation
statistics requires a prior filtration of the linear component, for which reason, this study filters
out the autocorrelation structure by an autoregressive AR(p) fit. The AR fitting is employed
as a pre-whitening operation, and not to obtain a model of best fit. The p order of the AR(p)
model is chosen between 0 and 10 as the smallest value for which the Ljung-Box Q(10) statistic
is insignificant at the 10% level. Brooks and Hinich (2001) showed that it is not necessary
to filter the returns through an AR-GARCH model in order to determine the statistics of the
bi-correlation test, because the presence of GARCH effects will not cause a rejection of the
null hypothesis of pure white noise. C statistics are being computed on the basis of unfiltered
returns, meanwhile H statistics are computed inside each window on the basis of the unfil-
tered residuum resulted in pursuance of the linear component filtration. The null hypothesis of
linear/nonlinear correlation is accepted or rejected in each window at a risk level of 1%.
The number of lags (L) is specified as L = nb , with 0 < b < 0.5. Hinich and Patterson
based on Monte-Carlo simulation recommends the usage of b = 0.4 in order to maximize the
power of the test assuring at the same time a good asymptotical approximation. The window
length must be large enough to offer a robust statistical power and yet small enough for the test
to be able to identify the arrival and disappearance of transient dependencies, as changes in
the variables behavior. Despite the fact that in the previously mentioned studies the window’s
10 ALEXANDRU TODEA, M ARIA ULICI, AND SIM ONA SILAGHI

volume was of 50 or even 35 trading sessions, in this study we will employ windows of 200
observations, a volume that is recommended by Patterson and Ashley (2000) as a result of the
Monte Carlo simulations achieved for six popular nonlinearity tests including the bi-correlation
test.
The use of the moving average strategy is based on the fact that financial series are volatile
and contain certain trends. The crossing of the prices or short term moving average line with
the long term moving average line may be a sign that a trend has been initiated. Following this
assertion a buying signal is generated when the short term moving average is higher than the
long term moving average, and the selling signal is generated when the inequality is reverse.
Around the long term moving average a percentage envelope has been introduced in order to
eliminate contingent « noisy » signals. The majority of studies are limited to different moving
average combinations proposed by Brock and others (1992), but taking into account that this
variant is a restrictive one, a program has been developed that allows, for every market, the
selection of the strategy for which the return in excess of the buy-and-hold strategy is maximum.
This strategy is selected from 15000 strategies, number resulted as a combination of all variants
of short term moving average (between 1 to 10 days) with all variants of long term moving
average (between 50 to 200 days) and of 10 envelops comprised between 0,1% and 1% (multiple
of 0,1%).
The commission for each transaction differs from market to market, being considered every
time a selling or buying signal appears. In this study the investor is considered to be always
on the market in a long or short position.
In each window of volume of 200 observations, the cumulated returns corresponding to the
selected moving average strategy and to the buy-and-hold strategy have been computed, as well
as the cumulated excess return. By dividing these results to 200 the daily average return/excess
has been found and grouped together in 4 sub-samples depending on the results of the C and H
tests. Considering the addition property of the average, the daily average excess was computed
for each sub-sample and then the significance of each result has been tested. The daily average
returns and the daily average excess returns have been annualized and expressed in percentage,
by multiplying them with 250, respectively with 100 in order to offer a clearer image of the
results.

3. The Data
The data consists of daily closing prices for six Asia-Pacific stock market indices from, July
1, 1997 to, April 14, 2008. Specifically, indices are collected for Australia (All Ordinaries
Index), Hong Kong (Hang-Seng Index), India (BSE national Index), Malaysia (Kuala Lumpur
Composite Index), Singapore (Strait Times Index) and Japan (Nikkei 225 Index). All the
closing prices obtained from Datastream are denominated in their respective local currency
units. The data is transformed into a series of continuously compounded percentage returns by
taking 100 times to log price relatives, i.e. rt = ln (pt /pt−1 ) ∗ 100, where pt is the closing price
of the index on day t.

4. Empirical Results
Running the program has allowed the identification for each market of the strategy for which
the excess return from the buy-and-hold strategy to be full, (this finding is reported on the
second column of Table 1). Determining the most profitable strategy was based on the closing
prices of the indices for the period 7/01/1997 - 4/14/2008; but the profitability analysis of
these strategies on linear / non-linear non-correlation and correlation sub-periods will be made
on samples of different volumes depending on the value obtained for the long-term moving
average. As an example, in the case of the Australian market, where the optimal strategy is (9,
62, 0.9%), the first 62 stock exchange meetings are waived from the initially sample. Thus, the
 ADAPTIVE M ARKETS HYPOTHESIS 11

first observation from the new sample corresponds to the 25-th of September 1997. The start
date of the sample for each market is founded on the third column of Table I.
The results of Table I show that, with the exception of the Australian market, for any
other market, the program could identify a strategy of moving average whose return is superior
than buy-and-hold passive strategy. These results are obtained for a medium commission on
transaction of 0.5%. The use of accurate commissions, specific to each market, may change
to some extent the results, but such an approach is not necessarily required to achieve the
objective study. In fact, it presents interest to what extent the profitability is higher in the
sub-periods events of episodic dependencies than in sub-periods of acceptance of the random
walk hypothesis.
With the exception of the Australian market, the annualized excess return of the optimum
strategy comparative to the buy-and-hold strategy, lies between 4-13%. This excess varies over
time depending on the evolutionary process of the stock market prices. One of the fundamental
hypotheses of the moving average strategies is that the past tends to repeat itself. For this
reason, the existence of an episodic behavior of the linear and nonlinear dependencies certainly
influence their profitability over time.
Table II shows that the excess return inside the windows in which the random walk hypothesis
is rejected due to linear and nonlinear correlations is superior to the excess return over the sub
periods of acceptance of the random walk hypothesis, with the exception of the Japanese market.
The percentage of windows in which the random walk hypothesis is rejected at a risk level of
1% varies from one market to another in an interval of 40-60%. This percentage is significantly
influenced by the length of the window, which is 200 trading sessions in this study. Thus, the
use of windows of 50 observations on Asian markets by Hinich (2005), Lim (2007) and Lim and
others (2007) led to a percentage of windows in which are linear and nonlinear correlations of
less than 20%.
Table III shows that the rejection of the random walk hypothesis is especially due to the
presence of the nonlinear dynamics. In the case of markets from Australia, India, Singapore
and Japan the moving average strategies exploit in particular non-linear correlations, a similar
result to that obtained by Todea (2009) on market from Austria, Germany and UK. Nonlinear
dynamics are used on markets from Hong Kong and Malaysia.

5. Conclusion
The contribution this study brings to the literature is the validation of the assumption under
which the existence of episodes of dependency increases the degree of predictability of stock
prices on six Asian equity markets. This ex-post validation is achieved through a combination
of random walk tests (which are tests associated to the hypothesis of information efficiency in
the weak-form), with direct tests of this hypothesis based on technical analysis.
The main conclusion is that the profitability of moving average strategies is not constant in
time, but rather episodic showing when sub-periods of linear and non-linear correlation appear.
Thus, it can be said that the degree of market efficiency varies in a cyclical fashion over time and
these statistical features are in line with those postulated by the Adaptive Markets Hypothesis
(AMH) proposed by Lo (2004, 2005). According to this theory, profit opportunities do exist
from time to time.
The results obtained on the Asian markets consolidate those obtained on the European
markets by Todea et al. (2009). They represent a strong argument for further research towards
trying to anticipate the appearance of sub-periods of linear and non-linear correlation.

References
[1] Bessembinder, H., & Chan., K. (1995). The Profitability of Technical Trading Rules in the Asian Stock
Markets. Pacific-Basin Finance Journal, no.3, pp. 257-284.
[2] Bonilla, C. A., Romero-Meza, R., & Hinich, M. J. (2006). Episodic nonlinearity in Latin American Stock
Market Indices. Applied Economics Letters, 13, pp. 195—9.
12 ALEXANDRU TODEA, M ARIA ULICI, AND SIM ONA SILAGHI

[3] Brock, W., Lakonishok, J., & LeBaron, B. (1992). Simple technical trading rules and the stochastic prop-
erties of stock returns. Journal of Finance, 47, pp. 1731-1764.
[4] Cajueiro, D.O., & Tabak, B.M. (2006). Testing for predictability in equity returns for European transition
markets. Economic Systems, forthcoming.
[5] Hinich, M. J., & Patterson, D.M. (1995). Detecting epochs of transient dependence in white noise. mimeo,
University of Texas, Austin.
[6] Lagoarde-Segot, T., & Lucey, B. (2008). Efficiency in emerging markets: evidence from the MENA region.
Journal of International Financial Markets, Institutions and Money,.18(1), pp. 94-105.
[7] Lim, K. P., Brooks, R. D., & Hinich, M. J. (2006). Testing the assertion that emerging Asian stock markets
are becoming more efficient. SSRN Working Paper Series.
[8] Lim, K.P., & Hinich, M.J. (2005). Cross-temporal universality of non-linear dependencies in Asian stock
markets. Economics Bulletin, 7(1), pp. 1 - 6.
[9] Lim, K.P., Brooks, R.D., & Hinich, M.J. (2007). Nonlinear serial dependence and the weak-form efficiency
of Asian emerging stock markets. Journal of International Financial Markets, Institutions & Money.
[10] Lim, K. P., Brooks, R.D., & Kim, J.H. (2007). Financial crisis and stock market efficiency: Empirical
evidence from Asian countries. International Review of Financial Analysis.
[11] Lim, K. P. (2007). Ranking of efficiency for stock markets: A nonlinear perspective. Physica A, 376, pp.
445-454.
[12] Lo, A. W. (2004). The Adaptive Markets Hypothesis: Market Efficiency from an Evolutionary Perspective.
Journal of Portfolio Management, 30, pp. 15-29.
[13] Lo, A. W. (2005). Reconciling efficient markets with behavioral finance: the Adaptive Markets Hypothesis.
Journal of Investment Consulting 7(2), pp. 21-44.
[14] Patterson, D.M., & Ashley, R.A. (2000). A nonlinear time series workshop: a toolkit for detecting and
identifying nonlinear serial dependence. Kluwer Academic Publishers, Boston.
[15] Ratner, M., & Leal, R.P.C. (1999). Tests of Technical Trading Strategies in the Emerging Equity Markets
of Latin America and Asia. Journal of Banking & Finance, 23(1999), pp. 1887-1905.
[16] Todea, A., & Zoicaş-Ienciu, A. (2008). Episodic Dependencies in Central and Eastern Europe Stock Markets.
Applied Economics Letters, Vol.15, No.13-15, oct-dec 2008, pp.1123-1126.
[17] Todea, A. (2008). Episodic dependencies and the profitability of moving average strategy on Romanian
capital market. Theoretical and Applied Economics, pp. 186-193.
[18] Todea, A., Zoicaş-Ienciu, A., & Filip, A. (2009). Profitability of the Moving Average Strategy and the
Episodic Dependencies: Empirical Evidence from European Stock Markets. European Research Studies,
Vol. XII, Issue 1, 2009, pp. 63-72.

Appendix
Table I. Optimal strategy, start date of sample, and returns
Country Optimal Start date Buy-and- Technical Excess
Strategy of sample Hold return analysis return return
Australia (9, 62, 0.9%) 9/25/1997 0.000338* 0,000165* -0,00017*
(8,45%) (4,125%) (-4,325%)
Hong Kong (3, 189, 0.9%) 4/14/1998 0,00045* 0,000608** 0,000158*
(11,25%) (15,20%) (3,95%)
India (1, 53, 0.7%) 9/16/1997 0,000643 0,00088 0,000237*
(16,075%) (22%) (5,925%)
Malaysia (7, 58, 0.4%) 9/23/1997 0,000331 0,001031 0,000653*
(8,275%) (25,775%) (16,325%)
Singapore (4, 84, 0.8%) 10/27/1997 0,000371* 0,000904* 0,000533*
(9,275%) (22,6%) (13,325%)
Japan (5, 159, 1%) 2/25/1998 0,0000033 0,000538* 0,000535*
(0,084%) (13,45%) (13,375%)
Note: In the brackets are the annualized returns. Numbers marked with * (**) are significant
at the 5% (1%) levels.
 ADAPTIVE M ARKETS HYPOTHESIS 13

Table II. The average excess return generated by the optimum strategy over sub-periods of
acceptance and rejection of the random walk hypothesis
C and H insignificant C or H significant
Country Total number Number of Excess Number of Excess
of windows windows return windows return
Australia 2474 1336 -0,000396** 1138 0,0000895*
(-9,9%) (2,24%)
Hong Kong 2279 1866 0,0000621 413 0,000592**
(1,55%) (14,8%)
India 2412 931 -0,00049** 1481 0,000697**
(-12,35%) (17,425%)
Malaysia 2398 1025 0,000416** 1373 0,00083**
(10,4%) (20,75%)
Singapore 2419 1346 0,000468** 1073 0,000613**
(11,7%) (15,325%)
Japan 2290 1688 0,000531** 602 0,000546**
(13,275%) (13,65%)
Note: In the brackets are the annualized returns. Numbers marked with * (**) are significant
at the 5% (1%) levels.

Table III. The average excess return generated by the optimum strategy in the sub-periods
of the random walk hypothesis rejection depending on the type of correlation
C significant and H significant and C and H
H insignificant C insignificant significant
Country Nb. of Excess Nb. of Excess Nb. of Excess
windows return windows return windows return
Australia - - 1071 0,000104* 67 -0,00015*
(2,6%) (-3,75%)
Hong Kong 51 0,001829** 358 0,000419** 4 0,00027**
(45,725%) (10,475%) (6,775%)
India 49 0,000569** 1263 0,000699** 169 0,000716**
(14,225%) (17,475%) (17,9%)
Malaysia 10 0,001275** 848 0,000658** 515 0,001104**
(31,875%) (16,45%) (27,6%)
Singapore 85 -0,0000329 941 0,000701** 47 0,000021
(-0,823%) (17,525%) (0,53%)
Japan 35 -0,00023** 548 0,00061** 19 0,000145*
(-5,775%) (15,25%) (3,625%)
Note: In the brackets are the annualized returns. Numbers marked with * (**) are significant
at the 5% (1%) levels.