Sei sulla pagina 1di 7

Long-term Dependence in Asian Foreign Exchange Markets

Shu-Fan Hsieh
National Kaohsiung First University of Science and Technology
So-De Shyu*
National Sun Yat-Sen University

abstract

W
e investigate the long-term dependency behavior of Asian foreign exchange markets by using rescaled range analysis.
Emerging markets in Korea, Taiwan, India, and Thailand, show evidences of long memory in the exchange rate return series,
while the exchange rate return persistence is not found in more developed and mature markets in Japan, Australia, Hong
Kong, and Singapore. Our results suggest that the return-generating processes and presence of long memory depends on the degree
of market development. In addition, the findings suggest that Asian financial crisis affects long-term dependences of Korean won and
Thai baht in which their economies and currency were hard hit by the crisis.

Keyword: Long-term dependence; Rescaled range analysis; Hurst exponent; Asian markets

INTRODUCTION dictable component in the returns, which in turn, may provide eco-
nomic benefits available to speculators in foreign exchange markets.
The long-range dependence of exchange rate changes has been To explore the long memory, we use rescaled range analysis (R/S
found in the literature (Booth et al. (1982), Hsieh (1989), Cheung analysis) developed by Mandelbrot (1972, 1982), Mandelbrot and
(1993), Peters (1994), Batten and Ellis (1996), Baillie et al. (2004) Wallis (1969) and Lo (1991). The conventional statistical techniques
and Jin et al. (2006)). Cheung (1993) finds evidence of a long-term utilized by many capital market theories and empirical research are
dependence in British pound, Deutsche mark, French franc and based on the assumption of normality and random walk. These con-
Japanese yen. Batten and Ellis (1996) also find evidences of persis- ventional techniques and tests fail to uncover the non-periodic cycles
tent series in Japanese yen. The literature of long memory effects and the nonlinear stochastic process whose characteristics appear to
in foreign exchange markets, however, focuses more on the well- conform more closely to observed financial returns. R/S analysis, a
developed markets. Less is known about the long memory effects in nonparametric statistical technique which needs no assumptions of
other markets. We investigate long-term dependences in the Asian normality and independency, is a good alternative to detect whether
foreign exchange markets. the time series system displays a fractal structure and whether the
The focus on Asian market is appropriate for a number of reasons. system demonstrates the persistence of trends, which is an indicator
First, during the last two decades, the rapid economic growth in the of long memory effects. A number of papers apply R/S analysis to
Asian countries was accompanied by a substantial increase in the various financial markets returns. Some examples include Lo (1991),
size of their foreign exchange markets. According to the “Triennial Huang and Yang (1995), Mckenzie (2001), Kyaw et al. (2006) and
Central Bank Survey of Foreign Exchange and Derivatives Market Assaf (2006), who find evidences of long memory of stock returns.
Activity” from Bank of International Settlement, the percentage of Corazza et al. (1997) applied this approach to futures returns, while
trading activities in Asian currency market is approximately 15 per- Cheung and Lai (1993) applied it to gold prices. Batten and Ellis
cent.1 As stated in Table 1, foreign exchange trading in some Asian (1996) and Cheung (1993) search for long-term dependences of ex-
currencies had more than 100% growth than the global total, such as change rates by using R/S analysis. In this paper, we apply this tech-
Korean won, New Taiwan dollar, Indian rupee, Indonesian rupiah, nique on foreign exchange markets in Asia.
and Chinese renminbi. Some exceptions are the Hong Kong dollar, We use R/S analysis to examine the daily returns of spot exchange
the Singapore dollar, and the Malaysian ringgit, in which their trad- rates across eight heavily traded Asian currencies: Japanese yen/$,
ing activities exhibit slower growth. Second, we conduct our analysis Australian dollar/$, Hong Kong dollar/$, Singapore dollar/$, Korean
using foreign exchange rate returns in emerging markets vis-à-vis de- won/$, New Taiwan dollar/$, Indian rupee/$, and Thai baht/$. In gen-
veloped markets. Thus, we are able to make inference on the impact eral, we find the presences of long-term dependences in emerging
of currency market development on the long memory effects. Third, markets in Korea, Taiwan, India, and Thailand. Contrarily, the return
our study carries interesting practical implication. The presence of persistence is not found in more developed and mature markets in
long memory in foreign exchange markets implies a potential pre- Japan, Australia, Hong Kong, and Singapore. Our results suggest that

*Corresponding author: Department of Finance, National Sun Yat-sen University, #70 Lien-hai Rd., Kaohsiung 804, Taiwan. Tel.: 886-7-5252000 ext. 4814; Fax: 886-7-5254899;
Email: dshyu@cm.nsysu.edu.tw (So-de Shyu), shufan@ccms.nkfust.edu.tw (S.F. Hsieh)

48 Journal of Asia Business Studies FALL 2009


TABLE 1: Traditional Foreign Exchange Market Turnover in Asia-Pacific in April 2004

(Daily averages, in millions of US dollars)

Spot Forward Swap Total Growth rate


(percentage share) since 2001
Japanese yen 130,382 47,135 181,715 359,231 (63.27%) 35%
Australian dollar 28,539 9,788 58,796 97,123 (17.11%) 96%
Hong Kong dollar 6,827 2,221 24,133 33,181 (5.84%) 21%
New Zealand dollar 4,018 1,432 12,181 17,661 (3.11%) 163%
Singapore dollar 5,177 1,242 10,591 17,010 (3.00%) 32%
Korean won 10,510 6,048 4,592 21,151 (3.73%) 117%
Taiwan dollar 3,607 2,798 856 7,261 (1.28%) 129%
Indian rupee 2,877 1,531 1,658 6,066 (1.07%) 114%
Thai baht 1,333 490 1,669 3,492 (0.62%) 88%
Indonesian rupiah 760 267 1,025 2,051 (0.36%) 283%
Chinese renminbi 992 811 9 1,812 (0.32%) 5,303%
Malaysian ringgit 351 237 399 987 (0.17%) 7%
Philippine peso 345 232 188 765 (0.13%) 52%

Major Currencies
US dollar 528,639 170,357 874,083 1,573,080 48%
Euro 272,887 88,243 298,231 659,361 49%
Pound sterling 82,839 31,338 185,241 299,241 93%
Canadian dollar 23,696 8,947 41,930 41,930 43%

Note: The data is collected from “Trading Asian currencies” in 2004, Bank of International Settlement.

the returns-generating processes and presence of long memory de- A*n=N. Each subperiod maybe labeled as Ia, with a=1, 2, 3, …, A,
pends on the degree of market development. In addition, we divide and each element of Ia may be denoted as Nk,a, with k=1,…,n. For
the sample period into two sub-periods: before and after the Asian each Ia of length n, a rescaled data series Yk,a can be computed by
financial crisis. Korean won and Thai baht, in which they were hard subtracting the subperiod mean:
hit during the crisis, display differences in their return processes when
compared with other currencies. The results indicate that the long- Yk,a = (Nk,a - ea), k=1,...,n (1)
term dependence of Korean won mostly happened before the crisis,
while the return persistence of Thai baht is found after the crisis. where ea is estimated as
n
Methodology and Data
Σ Nk,a
k=1
(2)
ea =
R/S analysis is first introduced by Hurst (1951) for studying the n
storage capacity of the resulting reservoir. He presented a new statis-
tic, Hurst exponent (H), which can differentiate three different types For each series Yk,a, we can construct a cumulative series Xk,a by
of series: random series, persistent series and anti-persistent series. cumulatively summing Yk,a. The last value of Xk,a will always be zero
Mandelbrot (1972, 1982), Mandelbrot and Wallis (1969), and Lo since the series Yk,a has a mean of zero. The range of the cumulative
(1991) later revised the R/S analysis method. We follow the proce- series Xk,a is estimated as:
dures stated in Peters (1994, pp.62-63).
Consider a series of returns calculated by logarithmic ratios with R1 = Max (Xk,a) - Min(Yk,a), 1 ≤ k ≤ n (3)
length N. Because the presence of linear dependence in the data
may bias rescaled range analysis and make it look significant when To standardize the measure over time, we divide each range R1aby
no long-memory process exists, we use a first-order auto-regressive the standard deviation (S1a) of the elements of Ia, i.e. =R1a /S1a. This
model (AR(1)) to moderate the linear dependency. Thus, we follow process will result in A rescaled ranges and the average rescaled range
the prewhitening process in Peters (194) by applying the R/S analysis ((R/S)n ) across the whole sample for subperiod length n is estimated as
on the AR(1) residuals of returns.2 In the second step, the sample
(N) is divided into A contiguous subperiods of length n, such that

Journal of Asia Business Studies FALL 2009


49
TABLE 2: Currency Distribution of Reported Foreign Exchange Market Turnover

1989 1992 1995 19998 2001 2004 2007 Average


Asian Currency
Japanese yen (Yen) 27 23.4 24.1 20.2 22.7 20. 16.5 22.0
Australian dollar (A$) 2 2.5 2.7 3.1 4.2 5.5 6.7 3.8
Hong Kong dollar (HK$) ... 1.1 0.9 1.3 2.3 1.9 2.8 1.7
Korean won (W) ... ... ... 0.2 0.8 1.2 1.1 0.8
New Zealand dollar (NZ$) ... 0.2 0.2 0.3 0.6 1 1.9 0.7
Singapore dollar (S$) ... 0.3 0.3 1.2 1.1 1 1.2 0.9
Taiwan dollar (NT) ... ... ... 0.1 0.3 0.4 0.4 0.3
Indian rupee (R) ... ... ... 0.1 0.2 0.3 0.7 0.3
Thai baht (B) ... ... ... 0.2 0.2 0.2 0.2 0.2
Indonesian Rupiah (Ru) ... ... ... 0.1 0 0.1 0.1 0.1
Chinese Renminbi (RMB) ... ... ... 0 0 0.1 0.5 0.2
Malaysian ringgit (Ri) ... ... ... 0 0.1 0.1 0.1 0.1
Philippine Peso (Pe) ... ... ... 0 0 0 0.1 0.0
All Asia currencies 29 27.5 28.2 26.8 32.5 32.1 32.3 29.8
Heavy Traded Currency
US dollar ($) 90 82 83.3 87.3 90.3 88.7 86.3 86.8
Euro (E) ... ... ... ... 37.6 37.2 37 37.3
Deutsche mark (DM) 27 39.6 36.1 30.1 ... ... ... 33.2
French fran (FF) 2 3.8 7.9 5.1 ... ... ... 4.7
Pound sterling (SF) 15 13.6 9.4 11 13.2 16.9 15 13.4
Swiss franc (SF) 10 8.4 7.3 7.1 6.1 6.1 6.8 7.4
Other currencies 28.3 25.2 27.8 32.7 20.1 19 22.9 25.1
All currencies 200 200 200 200 200 200 200 200.0

Note: The data presented in Table 2 is the percentage of average daily turnover. The data is collected from “Triennial Central Bank Survey-- Foreign
exchange and derivatives market activity” in 1989, 1992, 1995, 1998, 2001, 2004 and 2007, Bank of International Settlement. Because two currencies are
involved in each transaction, the sum of the percentage shares of individual currencies totals 200% instead of 100%.

A
(4) In theory, if the series were random, the scaled range, in asymp-
Σa (R /S )
Ia Ia
totic limit, would increase with the square root of time, i.e. H is 0.5.
(R/S)n =
A When H is less than 0.5, the type of series is an antipersistent series,
which implies the dispersion of returns scales at a slower rate than
We choose different length of subperiods (n) and repeat the pro- that of a random walk. When H is greater than 0.5, we have a persis-
cedure. In general, the length of subperiods (n) may start with 10 tent, or trend-reinforcing, series, which implies that the return pro-
and end with N/2, because small value of n may produce unstable cess must be related in some way and display a long-memory effect.
estimates when sample sizes are small. Additionally, we only use in- In addition, the intercept of the regression provides an indication of
tegers that will evenly divide into the sample length. In other words, the data’s distribution. The intercept should equal π/2 if the data is
the time increments will include both the beginning and ending normal or Gaussian distributed (Batten and Ellis, 1996)
points of the series. Peters (1994, pp.109) indicates that having more When the series length (N) is finite, the expected Hurst exponent
R/S values is certainly more desirable than having more data points, will vary depending on the values of n we use to run the regression.
when we are interested in the scaling of R/S. Therefore, we need a To test the significance of R/S, we compare it to E(R/S), which is the
value of N that offers the most divisors in order to maximize the expected value of R/S and implied by a true random walk process
power of R/S analysis.2 (McKenzie, 2001). The modified version of E(R/S) by Anis and Lloyd
For the last step, the Hurst exponent (H) may be estimated using (1976) and Peters (1994) is given as
an OLS regression of the following form
n-1
log(R/S)n = log(c) + H*log(n)+εt (5) E(R/S)n=((n-0.5)/n)*(n*π/2)-0.5*Σ √(n-r)/r (6)
r

50 Journal of Asia Business Studies FALL 2009


TABLE 3: Summary Statistics Daily Spot Exchange Rate Changes

(20 February 1985 to 31 August 2007)

Yen/$ $/A$ HK$/$ S$/$ W$ NT$/$ R/$ B/$


Mean -1.45E-04 3.25E-05 4.24E-08 -6.28E-05 1.98E-05 -2.98E-05 1.97E-04 2.57E-05
Variance 4.89E-05 4.27E-05 1.83E-07 1.09E-05 5.02E-05 7.62E-06 1.83E-05 3.81E-05
Std deviation 6.99E-03 6.54E-03 4.28E-04 3.30E-03 7.09E-03 2.76E-03 4.27E-03 6.18E-03
Skewness -0.58 -0.32 -1.44 -0.63 -0.58 1.18 9.07 3.41
Kurtosis 6.14 3.25 51.19 15.54 212.03 54.36 244.19 139.64
Std. error 9.21E-05 8.16E-05 5.63E-06 4.35E-05 9.34E-05 3.64E-05 5.63E-05 8.14E-05
Sample size 5,760 5,760 5,760 5,760 5,760 5,760 5,760 5,760

Note. $, Yen, A$, HK $, S$, W, NT, R and B denote the nominal exchange rates for US dollar, Japanese yen, Australian dollar, Hong Kong dollar, Singapor-
ean dollar, Korean won, New Taiwan dollar, Indian rupee and Thai baht respectively.

The E(R/S)n values in all successive values of n are estimated. Substi- series by taking AR(1) residuals in order to minimize the short term
tuting E(R/S)n for (R/S)n in Equation (5), the expected value of H (i.e., linear dependency. We find Hurst exponents are greater than the ex-
E(H)) can be estimated by re-running the regression. The determina- pected values in five exchange rate returns, including Japanese Yen
tion of E(H) enables the testing of hypotheses concerning the signifi- (0.576), Korean won (0.69), New Taiwan dollar (0.639), Indian ru-
cance of the estimated H. These hypotheses may be formally stated as: pee (0.622), and Thai baht (0.597), while the remaining exchange
rate returns are smaller than the expected H. In addition, since none
H0 : H= E(H) (7) of the estimated regression intercepts equal, and hence it is unlikely
that the exchange rates in Asian market conforming to a normal dis-
Because the E(R/S) values are normally distributed random vari- tribution. The determination of E(R/S)n by Equation (6) enables the
ables, the E(H) would also be normally distributed. In this case, the testing of hypotheses in Equation (7), which relates the significance
expected standard deviation of Hurst exponent can be estimated as of the Hurst exponents.
σ=√1/N (Peters, 1994, pp.72).3 It is worth noting that the usual sta- We find significant long-term dependencies displayed in Korean
tistical tests such as the t-statistic, R2 and standard error are not ap- won/$, Taiwan dollar/$, Indian rupee/$ and Thai baht/$ exchange
propriate for testing the significance of the Hurst exponent because rates, while we cannot find evidence to reject the random walk hy-
the values of log(R/S) for each sample length n are highly correlated pothesis in Japanese yen/$. On the other hand, the results from $/
(McKenzie, 2001). Australian dollar, Hong Kong dollar/$, and Singapore dollar/$ sug-
The Asian country daily exchange rate data is from Global Fi- gest anti-persistence. The return series of Hong Kong dollar/$ re-
nancial Data. We selected eight foreign exchange returns: Japanese verses itself more often than a random walk significantly.
yen/$, Australian dollar/$, Hong Kong dollar/$, Singapore dollar/$, The results suggest that the returns-generating processes and pres-
Korean won/$, Taiwan dollar/$, Indian rupee/$ and Thai baht/$ from ence of long memory depends on the degree of market development.
January 1985 to September 2007. For the sample period, we have Long-term dependence of exchange returns is not found in more
5,794 observations for each currency. However, for applying R/S developed and mature markets in Japan, Australia, Hong Kong, and
analysis, we need to have more R/S values. Therefore, we use 5760 Singapore, while that is found in emerging markets in Korea, Taiwan,
observations so that the data can offer the most divisors in order to India, and Thailand. This result may be caused by more government
maximize the power of R/S analysis. We shorten the sample period to interfere and international capital flows in emerging markets. In ad-
be from 20 February 1985 to 31 August 2007. The summary statis- dition, the presence of long memory may imply a potentially predict-
tics is presented in Table 3. The emerging market currencies, such as able component in the returns, which in turn, may provide economic
Korea won, Indian rupee, New Taiwan dollar, and Thai baht exhibit benefits available to speculators in these emerging markets.
high skewness and kurtosis relative to the other four currencies. The One concern may exist in the case of the exchange rates of Asian
summary statistics suggest that currencies in emerging market be- countries. The so-called Asian financial crisis started with the crash
have differently from more developed markets. of the value of Thai baht in late 1997 and then spread to all the
countries in this region. The crisis hit Thailand and Korea hard. It
Empirical Results resulted in 100% depreciation in the foreign exchange market of
Thailand and Korea. The neighboring Asian countries were also af-
The results of the R/S analysis for Hurst exponent on spot ex- fected. For example, Japanese yen and Singapore dollar were depre-
change rates are presented in Table 4. We prewhiten these return ciated by 25%. We examine if the exchange rate return processes and

1
According to the data in Table 2, the percentage of average daily turnover in all Asian currencies is around 30%. Because two currencies are involved in each transaction, the sum of the percentage
shares of individual currencies totals 200% instead of 100%. Hence, the percentage of trading activities in Asian currency market is approximately 15 percent.
2
Peters (1994, pp108-109) indicates that the AR(1) residual method does not eliminate all linear dependence. However, it eliminates enough dependence to reduce the effect to insignificant levels,
even if the AR process is level 2 or 3.

Journal of Asia Business Studies FALL 2009


51
TABLE 4: Hurst Exponent Estimation

(All samples, 20 February 1985 to 31 August 2007)

Yen/$ $/A$ HK$/$ S$/$ W$ NT$/$ R/$ B/$


Regression Output:
Constant (intercept) -0.099 -0.064 0.027 -0.04 -0.278 -0.149 -0.171 -0.116
R squared 0.997 0.998 0.994 0.996 0.991 0.999 0.996 0.999
Hurst exponent (H) 0.576 0.553 0.49 0.549 0.690 0.639 0.622 0.597
Expected hurst exp. (E(H)) 0.56 0.56 0.56 0.56 0.56 0.56 0.56 0.56

SD of Hurst exp. (σ) 0.013 0.013 0.013 0.013 0.013 0.013 0.013 0.013

Total number of samples (N) 5760 5760 5760 5760 5760 5760 5760 5760

Number of observations of H 39 39 39 39 39 39 39 39

Significance (z-value) 1.19 -0.54 -5.36** -0.83 9.82** 6.01 4.69** 2.78**

Note. $, Yen, A$, HK $, S$, W, NT, R and B denote the nominal exchange rates for US dollar, Japanese yen, Australian dollar, Hong Kong dollar, Singapor-
ean dollar, Korean won, New Taiwan dollar, Indian rupee and Thai baht respectively. The regression results summarized in Table 4 are the results of
Equation (5): , where c is constant, n is the length of sub-periods and R/S is the average rescaled range. The Expected Hurst exponent (E(H)) is obtained
by substituting for into Equation (5), where is computed by Equation (6). SD of Hurst exponent ( ) is estimated by , where N is the total number of
samples. ** indicates that the statistic is significant at the 1% level.

TABLE 5: Hurst Exponent Estimation

(Before Asian Financial Crisis, 20 February 1985 to 24 April 1997)

Yen/$ $/A$ HK$/$ S$/$ W$ NT$/$ R/$ B/$


Regression output:
Constant (intercept) -0.124 -0.052 0.088 -0.102 -0.329 -0.164 -0.144 -0.054
R squared 0.996 0.998 0.982 0.996 0.980 0.998 0.996 0.998
Hurst exponent (H) 0.593 0.539 0.460 0.586 0.735 0.679 0.595 0.567
Expected hurst exp. (E(H)) 0.569 0.569 0.569 0.569 0.569 0.569 0.569 0.569

SD of Hurst exp. (σ) 0.018 0.018 0.018 0.018 0.018 0.018 0.018 0.018

Total number of samples (N) 3060 3060 3060 3060 3060 3060 3060 3060

Number of observations of H 28 28 28 28 28 28 28 28

Significance (z-value) 1.35 -1.63 -6.05** 0.95 9.19** 6.10** 1.43 -0.11

Note. $, Yen, A$, HK $, S$, W, NT, R and B denote the nominal exchange rates for US dollar, Japanese yen, Australian dollar, Hong Kong dollar, Singapor-
ean dollar, Korean won, New Taiwan dollar, Indian rupee and Thai baht respectively. The regression results summarized in Table 5 are the results of
Equation (5): , where c is constant, n is the length of sub-periods and R/S is the average rescaled range. The Expected Hurst exponent (E(H)) is obtained
by substituting for into Equation (5), where is computed by Equation (6). SD of Hurst exponent ( ) is estimated by , where N is the total number of
samples. ** indicates that the statistic is significant at the 1% level.

Peters (1994) proved these statistical characteristics by Monte Carlo simulation. For the details, please refer to chapter 5 of Peters (1994).
3

52 Journal of Asia Business Studies FALL 2009


TABLE 6: Hurst Exponent Estimation

(After Asian Financial Crisis, 24 May 1999 to 31 August 2007)

Yen/$ $/A$ HK$/$ S$/$ W$ NT$/$ R/$ B/$


Regression output:
Constant (intercept) -0.075 -0.073 -0.133 -0.03 -0.071 -0.158 -0.221 -0.128
R squared 0.997 0.997 0.983 0.995 0.997 0.998 0.996 0.995
Hurst exponent (H) 0.567 0.564 0.561 0.541 0.579 0.624 0.652 0.62
Expected hurst exp. (E(H)) 0.574 0.574 0.574 0.574 0.574 0.574 0.574 0.574

SD of Hurst exp. (σ) 0.022 0.022 0.022 0.022 0.022 0.022 0.022 0.022

Total number of samples (N) 2160 2160 2160 2160 2160 2160 2160 2160
Number of observations of H 31 31 31 31 31 31 31 31
Significance (z-value) -0.33 -0.46 -0.59 -1.52 0.26 2.32** 3.63** 2.14**

Note. $, Yen, A$, HK $, S$, W, NT, R and B denote the nominal exchange rates for US dollar, Japanese yen, Australian dollar, Hong Kong dollar, Singapor-
ean dollar, Korean won, New Taiwan dollar, Indian rupee and Thai baht respectively. The regression results summarized in Table 6 are the results of
Equation (5): , where c is constant, n is the length of sub-periods and R/S is the average rescaled range. The Expected Hurst exponent (E(H)) is obtained
by substituting for into Equation (5), where is computed by Equation (6). SD of Hurst exponent ( ) is estimated by , where N is the total number of
samples. ** indicates that the statistic is significant at the 1% level.

long-term dependences are different during the Asian financial crisis able to speculators in foreign exchange markets of emerging coun-
period. We divide our sample period into two sub-periods: before tries than developed countries because of the significant presence of
the financial crisis (from February 20, 1985 to April 24, 1997) and long memory effects..
after the financial crisis (from May 24, 1999 to August 31, 2007).
We expect there are difference for countries affected by the crisis the
most severe, such as Korea and Thailand. The results of H exponents
before and after Asian financial crisis are presented in Tables 5 and 6.
For neighboring countries, which did not hit hard by the crisis, we
find the results before and after the crisis are similar. For example,
neither before nor after can we find the presence of long-term de-
pendences in Japanese yen, Australian dollar, Hong Kong dollar, and
Singapore dollar. On the other hand, the returns of Taiwan dollar
display high level of persistence in two subperiods. This may reflect
that the Taiwanese foreign exchange market is highly controlled by
the government. Contrarily, the results suggest that the long-term
dependence of Korean won mostly happen before the crisis, while
the return persistence of Thai baht and Indian is found after the cri-
sis. This results maybe due to Korea quickly accepted IMF’s advice to
reform its financial structure while Thailand and India still struggled
on its own after financial crisis.

Conclusion
We investigate the long-term dependency behavior in the returns
of Asian foreign exchange markets of Japan, Australia, Hong Kong,
Singapore, Korea, Taiwan, India, and Thailand. We find the pres-
ences of long-term dependences in emerging markets in Korea, Tai-
wan, India, and Thailand. On the other hand, the return persistence
is not found in more developed and mature markets in Japan, Aus-
tralia, Hong Kong, and Singapore. Our results suggest that the long
memory effect depends on the degree of market development. In
addition, the presence of long memory may imply a potentially pre-
dictable component in the returns, which in turn, may provide eco-
nomic benefits available to speculators in these emerging markets.
Our findings suggest that there may be more economic benefits avail-

Journal of Asia Business Studies FALL 2009


53
References
Anis, A. A. and Lloyd, E. H. (1976) “The expected value of the adjusted
rescaled Hurst range of independent normal summands”, Biometrika,
63, 111-16.
Assaf, A. (2006) “Dependence and mean reversion in stock prices: The case
of the MENA region”, International Business and Finance, 20, 286-304
Baillie, R. T., Cecen, A. A., Erkal, C and Han, Y. W. (2004) “Measuring non-
linearity, long memory and self-similarity in high-frequency European ex-
change rates”, Journal of International Financial Markets, Institutions &
Money, 14, 401-418
Batten, J. and Ellis, C. (1996) “Fractal structures and naïve trading systems:
evidence from the spot US dollar/Japanese yen”, Japan and the World
Economy, 8, 411-21.
Booth, G., Kaen, F., and Koveos, P. (1982) “R/S analysis of foreign exchange
rates under two international monetary regimes”, Journal of Monetary
Economics, 10, 407-15.
Cheung, Y. (1993) “Long memory in foreign exchange rates”, Journal of
Business and Economic Statistics, 11, 93-101.
Cheung, Y. W. and Lai, K. S. (1993) “Do gold market returns have long mem-
ory?” , The Financial Review, 28, 181-202.
Corazza, M., Malliaris, A.G., and Nardelli, C. (1997) “Searching for fractal
structure in Agricultural Futures Markets”, Journal of Futures Market, 17,
433-73.
Hsieh, D. (1989) “Testing for non-linear dependencies in daily foreign ex-
change rates”, Journal of Business, 62, 339-368.
Huang, B. N. and Yang, C.W. (1995) “The fractal Structure in multinational
stock returns”, Applied Economics Letters, 2, 67-71.
Hurst, H. E., (1951) “Long term storage capacity of reservoirs”, Transactions
of the American Society of Engineers, 116, 770-99.
Jin, H. J., Elder, J and Koo, W. W. (2006) “A reexamination of fractional in-
tegrating dynamics in foreign currency markets”, International Review of
Economics and Finance, 15, 120-135
Kyaw, N. A., Los, C. A and Zong, S. (2006) “Persistence characteristics of
Latin American financial markets”, Journal of Multinational Financial Man-
agement, 16, 269-290
Lo, A. W. (1991) “Long-term Memory in Stock Market Prices”, Econometrica,
59, 1279-313
Mandelbrot, B. B. and Wallis, J. R. (1969) “Robustness of the rescaled range
R/S in the measurement of noncyclic long-run statistical dependence”,
Water Resources Research, 5, 967-88.
Mandelbrot, B. B. (1972) “Statistical methodology for non-periodic cycles:
From covariance to R/S analysis”, Annals of Economic and Social Mea-
surement, 1, 259-90.
Mandelbrot, B. B. (1982) The fractal geometry of nature, W.H. Freeman, New
York.
McKenzie, M. D. (2001) “Non-periodic Australian stock market cycles: Evi-
dence from Rescaled range analysis”, Economic Record, 77, 393-406.
Peters, E. E. (1991), Chaos and order in the capital markets, John Wiley and
Sons, New York.
Peters, E. E. (1994) Fractal market analysis: Applying chaos theory to invest-
ment and economics, John Wiley and Sons, New York.

54 Journal of Asia Business Studies FALL 2009

Potrebbero piacerti anche