The International Journal of Business and Finance Research ♦ Volume 2 ♦ Number 1 ♦ 2009

LONG MEMORY IN EXCHANGE RATES:

INTERNATIONAL EVIDENCE
BSc, Msc, Phd Candidate Ilaman Shagylyjov
Graduated student from Queen`s University of Belfast

ABSTRACT
In this paper we test for the presence of fractional integration, or long memory, in the daily returns of
exchange rates using ARFIMA(p,d,q) models. We consider 34 exchange rates against the US dollar
(USD) covering the period April 1991 to April 2006. The results suggest that 17 exchange rates show
strong evidence of long memory. This indicates that shocks to the exchange rates persist over a long
period of time (this is stronger in emerging market economies). This also indicates that these markets are
not stable, and hence they offer an opportunity to investors and traders to add some risk to their
strategies. The lack of long memory in the daily returns of exchange rates supports the efficient market
hypothesis (EMH). These findings are helpful to traders dealing with long dated assets.

JEL: C22, C52, G14

INTRODUCTION

I t is widely believed that the logs of financial prices contain a unit root. However, some series
evidently do not possess a further unit root, while they show signs of dependence. Such series are
argued to possess long memory. Long memory in time series can be defined as autocorrelation at
long lags, of up to hundreds of time periods (Tolvi, 2003). Under this phenomenon, systems are
characterized by their ability to ‘remember’ events in the long history of time series data and their ability
to make decisions on the basis of such memories. According to long memory hypothesis, what happens
today affects the future.

Knowledge of the time series properties of exchange rates has important economic implications. The
empirical analysis of exchange rates provides useful information that can be used to evaluate the
performance of exchange rate models. The objectives of this paper are twofold: (i) to examine and justify
the presence of long memory (via fractional integration) in the prices of foreign exchange rates, and (ii) to
test the validity of time series models on the presence of long memory in exchange rates. Time series
models are important for the determination of international trade flows, prices of tradable goods, prices of
foreign exchange futures and options and international asset portfolios (Cheung, 1993).

A number of studies have tested the long memory hypothesis using data from mature and emerging
financial markets. If asset returns display long memory, they exhibit significant autocorrelation between
observations widely separated in time. The absence/presence of long-term dependence on the mean of
financial-asset returns is used as a proxy for analyzing market efficiency. The presence of long memory in
asset returns contradicts the weak form of the efficient market hypothesis (EMH), which states that,
conditioning on historical returns, future asset returns are unpredictable (Barkoulas et al., 2000).

Previous studies on this issue have relied on ARIMA models and have examined only a few countries.
We test for the presence of long memory, or fractional dynamics, using the framework of Autoregressive
Fractionally Integrated Moving Average (ARFIMA) allowing for a more precise specification of the order
of integration. We use daily data from 34 exchange rates over the US dollar. Our findings are important
since no previous work has examined this hypothesis using daily data from a large sample of countries
covering all five continents or has used the ARFIMA methodology. Thus this paper extends the literature
on two fronts. The remainder of this paper is organized as follows: In the next section a Literature

31
C. Floros ⎪ The International Journal of Business and Finance Research ♦Vol. 2 ♦ No. 1 ♦ 2009

Review is provided. This section is followed by a Methodology Section that presents the long memory
methodology through ARFIMA (p,d,q) models. The following section discusses the data used in this
study. Next, the empirical results are examined. The paper closes with some concluding comments.

LITERATURE REVIEW

A number of researchers examine the long memory hypothesis in exchange rates (Bhar, 2000; Gil-Alana,
2000; Laurini and Portugal, 2003). The exchange rate between two currencies specifies how much one
currency is worth in terms of the other. Traditionally, it has been assumed that the exchange rates have a
unit root implying that shocks have permanent effects on the series (Taylor, 1995). However, some
authors suggest that the exchange rates are mean-reverting (Gil-Alana, 2000).

Lo (1991) tests for long-run memory in daily and monthly stock market indices and finds no evidence of
long-range dependence. Cheung and Lai (1995) examine long memory in international stock market
returns using the Morgan Stanley Capital International stock index data for eighteen countries. Their
results provide little support for long memory. According to Cheung and Lai (1995), the findings are not
sensitive to inflation adjustments in stock returns, data sources, and statistical methods used.
Furthermore, Ding and Granger (1996) examine the long memory property for various speculative
returns. They report evidence of long memory for S&P 500 returns and four other speculative returns.
Recently, Gil-Alana (2006) uses parametric and semi-parametric methods to test for the order of
integration (and fractional integration) in daily stock market indices: EOE, DAX, Hang Seng, FTSE100,
S&P 500, CAC 40, Singapore All shares and the Japanese Nikkei. He reports that the order of integration
of the Singapore All Shares and the Hang Seng is much higher than one. Furthermore, Gil-Alana (2006)
finds conclusive evidence against mean reversion, but long memory on their returns. For S&P 500, the
fractional integration (long-memory) parameter is below one, while for the remaining series the values
oscillate around the unit root. Barkoulas et al. (2000) test for the presence of long memory in the return
series for the Greek stock market (an emerging capital market). They find significant and robust evidence
of positive long-term persistence. In addition, Tolvi (2003) tests the long memory hypothesis in Finnish
stock market, while Vougas (2004) extends the work of Barkoulas et al. (2000). He analyzes long
memory and volatility of returns in the Athens Stock Exchange and finds weaker evidence in favour of
long memory.

Cheung (1993) finds evidence of long memory in exchange rate data. Bhar (2000) examines long memory
in the Yen/dollar rate. He finds no evidence of long memory indicating efficient pricing by the market
participants. Cheng (2001) examines the long memory dynamics in the daily and weekly rates of six Asia
Pacific countries. His findings reveal strong evidence of long memory in the weekly series. Laurini and
Portugal (2003) show that the evidence of long memory for the daily R$/US$ series after the
implementation of the Rate Plan is not robust. Furthermore, Gil-Alana (2004) shows that exchange rates
series have a component of long memory behaviour. Recently, Soofi et al. (2006) use the plug-in and
Whittle methods (spectral regression analysis) to test for the long memory property in 12 Asian/dollar
daily exchange rates. Their findings, based on the two different methods, are mixed.

METHODOLOGY

Previous studies used the standard Integrated Autoregressive Moving Average (ARIMA) model to study
the intertemporal dynamics of exchange rates. An integrated series of order d must be differenced d times
before it can be represented by a stationary and invertible ARMA process. If this ARMA representation is
of order (p,q) then the original, undifferenced series is following an ARIMA(p,d,q) representation.

This paper examines the dynamics of exchange rates in the fractionally integrated autoregressive moving
average (ARFIMA) framework which is a generalization of the ARIMA(p,d,q) model. The ARFIMA

32
The International Journal of Business and Finance Research ♦ Volume 2 ♦ Number 1 ♦ 2009

model is the generalization of the concept of the order of integration used in ARIMA models, thus
allowing the order of integration to be a fractional number (Laurini and Portugal, 2004). We test
ARFIMA(p,d,q) models via conditional maximum likelihood (ML), following the recent work of Vougas
(2004). For the returns series rt , where t = 1,...,T , the ARFIMA model is given by:

φ ( L)(1 − L) d (rt − μ ) = θ ( L)ε t (1)

r t −φ j( L)(1 − L) ( rt − μ ) = θ ( L)ε t is the autoregressive polynomial, and

d
L is the lag operator ( L j r t = ),
θ ( L ) = 1 + θ 1 L + ... + θ q L q is the moving average polynomial. The differencing parameter d is not
necessarily an integer (it takes real values), but integer values of d lead to the traditional ARIMA model.

Therefore, the fractional differencing parameter (1−L)d can be defined for non-integer values by the
following binomial expansion:
∞
⎛d ⎞
(1 − L ) d = ∑ ⎜⎜ j ⎟⎟ ( − L ) j (2)
j=0 ⎝ ⎠
εt ~ NID(0,σt2 )
We also make the assumptions that (i) the residuals and (ii) the roots of the AR and
MA parameters fall outside the unit circle and do not have common roots. Significance of d parameter is
evidence of long memory. When d parameter has values greater or equal to 0.5, the series does not have
stationary covariance, and consequently it has infinite covariance as shown by Baillie et al. (1996). When
d is between 0 and 0.5, the lag length increases the autocorrelations decay hyperbolically to zero, while
when d = 0, decays exponentially to zero. If d is between -0.5 and 0, then it is usually identified as
having intermediate memory, since autocorrelations are always negative. Further, we need to select a
parsimonious ARFIMA(p,d,q) model using two information criteria: the Akaike (AIC) and Schwarz
(SBC). The information criteria are given by:
AIC = −2(ˆA / n) + (2( p + q + 2)) / n (3)
SBC = −2(ˆA / n) + (( p + q + 2) ln(n)) / n

where  is the value of the maximized likelihood. The best (selected) model has the smallest AIC or SBC
value. It is known that AIC always selects a generously parameterized model, while SBC selects a less
generously parameterized model (Vougas, 2004). In general, the AIC is one of the most commonly used
in time series analysis. The selected ARFIMA model is a parsimonious and flexible model that can be
used to study long memory and short-run dynamics simultaneously. Fractional integration is a more
general way to describe long-range dependence than the unit-root specification and provides an
alternative perspective to examine the unit-root hypothesis (Cheung, 1993).

Data Description

Foreign exchange markets are rather different from most financial markets. The vast bulk of trading takes
place between professional foreign exchange dealers of banks. They do not meet the people they are
trading with face to face, but they do their transactions over the phone or electronically. Note that foreign
exchange markets are particularly rich on noise traders.

The structure of this market has two interesting implications: (i) exchange rates are moved by news. Since
news is random and unpredictable, exchange rates will tend to move in a random way, and (ii) in the
foreign exchange markets, there are not active cross-markets between all parts of currencies, but this does

33
C. Floros ⎪ The International Journal of Business and Finance Research ♦Vol. 2 ♦ No. 1 ♦ 2009

not matter, because all currencies that have a market at all have one with the US dollar. Hence, the US
dollar can be thought of as the medium of exchange of the foreign exchange markets.

The data covers the period from April 1991 to April 2006, and the main source is the Datastream
International. We consider daily exchange rates from 34 countries (the total number of observations vary
from 1397 to 3915), as follows: Europe (Czech Republic, Cyprus, Denmark, Hungary, Norway, Poland
and UK), Asia (Bahrain, Hong Kong, India, Israel, Japan, Kuwait, Lebanon, Malaysia, Oman,
Philippines, Russia, Singapore, South Korea, Sri Lanka and Turkey), Africa (Botswana, Egypt, Kenya,
Mauritius, Morocco and Tunisia), America (Argentina, Canada, Mexico, Uruguay and Venezuela) and
Australia. The empirical analysis is based on both developed and developing (or emerging) markets.
Note that emerging countries constitute approximately 80% of the global population, representing about
20% of the world’s economies (worldbank.com).

Table 1 presents the descriptive statistics for the log-exchange rates. Most series show positive skewness
(the distribution is skewed to the right), while the distribution is peaked (leptokurtic) for Bahrain,
Lebanon and Oman and flat (platykurtic) relative to normal for the rest. We also reject the hypothesis of
normal distribution at the 5% level (for all series). The ADF tests show evidence of non-stationarity, I
(1), for all series (the results are not reported here).

Table 1: Descriptive Statistics for Log-exchange Rates over the US Dollar

Country Obs. Mean S.d. Skewness Kurtosis Normality

Argentina 3915 -0.3 0.49 -1.01 -0.914 7358 (0.00*)
Australia 3915 -0.38 0.13 -0.78 -0.53 1540 (0.00*)
Bahrain 3915 0.97 0.0012 -9.6276 467.35 43410 (0.0*)
Botswana 3200 -1.45 0.27 0.44 -0.87 508 (0.00*)
Canada 3915 -0.313 0.094 0.298 -0.815 323 (0.00*)
Cyprus 1397 0.628 0.139 -0.3178 -1.4159 386 (0.00*)
Czech Re. 3740 -3.39 0.147 -0.389 -0.769 409 (0.00*)
Denmark 3915 -1.88 0.123 -0.633 -0.393 713 (0.00*)
Egypt 1397 -1.655 0.169 0.578 -1.209 620 (0.00*)
Hong Kong 3915 -2.0492 0.0035 -0.269 -1.4925 1120 (0.00*)
Hungary 3740 -5.16 0.412 0.683 -0.787 1347 (0.00*)
India 3915 -3.64 0.207 0.731 -0.344 1013 (0.00*)
Israel 3915 -1.279 0.2266 0.50474 -0.98699 944 (0.00*)
Japan 3915 -4.741 0.1021 0.29175 0.32928 54.5 (0.00*)
Kenya 1397 -4.3481 0.0308 0.9155 0.6655 306 (0.00*)
Kuwait 3915 1.205 0.01884 0.4014 -0.64159 363 (0.00*)
Lebanon 3915 -7.3244 0.15394 1.8736 6.6586 1824 (0.00*)
Malaysia 3915 -1.1674 0.19495 0.28252 -1.8101 2570 (0.00*)
Mauritius 1397 -3.3619 0.04427 0.77599 0.34759 220 (0.00*)
Mexico 3915 -1.9514 0.48526 0.89259 -0.89499 3684 (0.00*)
Morocco 1397 -2.2861 0.10878 -0.32073 -1.3311 330 (0.00*)
Norway 3915 -1.9694 0.11723 -0.63968 -0.35334 698 (0.00*)
Oman 1397 0.95477 0.00147 2.3933 6.7964 2582 (0.00*)
Philippines 3915 -3.6109 0.32409 0.012577 -1.7227 1372 (0.00*)
Poland 3740 -1.0961 0.35652 1.1464 0.4519 2342 (0.00*)
Russia 1397 -3.3856 0.043161 -0.3789 -1.1515 283 (0.00*)
Singapore 3915 -0.48963 0.078495 0.71816 -0.59781 1277 (0.00*)
South Korea 3915 -6.9066 0.21657 -0.0916 -1.301 539 (0.00*)
Sri Lanka 1397 -4.5698 0.05347 0.93817 0.82954 297 (0.00*)
Tunisia 1397 -0.2929 0.0634 -0.16369 -1.1314 147 (0.00*)
Turkey 3915 1.7043 1.99 0.5 -1.1419 1185 (0.00*)
UK 3915 0.48072 0.078546 0.36253 -0.53742 253 (0.00*)
Uruguay 3915 -2.2933 0.76255 0.27532 -0.79199 280 (0.00*)
Venezuela 3915 -6.0992 1.1483 0.37894 -1.0415 655 (0.00*)
This Table shows the summary statistics for log-exchange rates. Skewness is a measure of asymmetry of the distribution
of the series around its mean. Kurtosis measures the peakedness or flatness of the distribution of the series. Normality
(under Jarque-Bera test) is a test statistic for testing whether the series is normally distributed (probability value is in parentheses).
* indicates significance at the 5% level.

34
The International Journal of Business and Finance Research ♦ Volume 2 ♦ Number 1 ♦ 2009

EMPIRICAL RESULTS

We use the ARFIMA framework which allows for long memory in the data. ARFIMA(p,d.q) models are
estimated via conditional Maximum Likelihood (ML) using the Ox language (PcGive software). We run
various ARFIMA(p,d,q) specifications with p+q≤2 (not reported here). There is evidence of unit roots,
and the selected ARFIMA model, for all samples, is ARFIMA(1,d,1). The selected ARFIMA(1,d,1)
model corresponds to the smallest AIC and SBC information criteria.

The results from Europe are reported in Table 2 (Panel A). Accordingly, all models show insignificant d
parameter. Hence, the results from the European/$US rates show weak evidence of long memory. The
results for Africa (Table 2 – Panel B) show that four rates (Botswana, Egypt, Kenya and Mauritius)
support the long memory hypothesis. Similarly, the results for Asian countries (Table 2 – Panel C) show
evidence of long memory for ten countries (Bahrain, Hong Kong, India, Israel, Lebanon, Oman,
Philippines, Russia, South Korea and Turkey). However the evidence for America and Australia is
mixed. Table 2 – Panel D shows that only three rates (Argentina, Mexico and Uruguay) have long
memory properties, with the remaining countries not showing the presence of long memory.

Furthermore, Figure 1 presents the variation of d parameter for all countries which support the long
memory hypothesis. The results from Africa and Asian countries (Figures 1.1-1.2) show that -0.1643<
d africa < 0.3741 and -0.947< d aasian <0.3649, while for America (Figure 1.3) 0.144< d america <0.2816.
More specific, for American exchange rates (Argentina, Mexico and Uruguay), empirical evidence shows
that the lag length increases the autocorrelations decay hyperbolically to zero. This is also true for
African exchange rates (Botswana, Egypt and Mauritius). The only exception is Kenya; there is evidence
of intermediate memory, since autocorrelations are always negative. Finally, for the Asian/$US rates, the
empirical evidence is mixed. Five exchange rates (India, Israel, Lebanon, Russia and Turkey) support the
property that the lag length increases the autocorrelations decay hyperbolically to zero. In addition, five
Asian/$US exchange rates (Bahrain, Hong Kong, Oman, Philippines and South Korea) show evidence of
intermediate memory, since autocorrelations are always negative.
Table 2: Empirical Results
PANEL A. ML Estimation of ARFIMA(1,d,1) Models: EUROPE

COUNTRY φ1  θ1  d AIC LL

Czech Rep -0.138 (-0.9) 0.013 (0.08) 0.006 (0.247) -6.9930 13078.7000
Cyprus -0.49 (-1.22) 0.41 (0.94) -0.029 (-0.9) -7.1880 5021.3000
Denmark 0.04 (0.0534) -0.07 (-0.09) -0.00012 (-0.0048) -7.2260 14145.4000
Hungary -0.1866 (-1.43) 0.037 (0.264) -0.0036 (-0.184) -6.9790 13051.9000
Norway 0.2318 (0.1) -0.239 (-0.104) -0.0305 (-1.08) -7.1205 13938.9000
Poland -0.0971 (-2.21)* -0.35 (-5.88)* 0.043 (1.61) -6.3411 11858.7400
UK 0.5597 (0.6) -0.5143 (-0.576) -0.020 (-0.249) -7.5624 14803.5200
* Significant at 5% Level

35
C. Floros ⎪ The International Journal of Business and Finance Research ♦Vol. 2 ♦ No. 1 ♦ 2009

PANEL B. ML Estimation of ARFIMA(1,d,1) Models: AFRICA

COUNTRY φ1  θ1  d AIC LL

Botswana 0.398 (4.27)* -0.56 (-5.15)* 0.085 (1.90)* -6.9700 11159.7000

Egypt 0.182 (2.22)* -0.486 (-4.63)* 0.13 (2.3)* -7.2899 5092.3600
Kenya 0.1478 (1.1) 0.04 (0.346) -0.1643 (-4.56)* -8.0474 5621.1300
Mauritius -0.1615 (-4.4)* -0.737 (-18.1)* 0.3741 (6.63)* -8.1154 5668.5700
Morocco -0.4215 (-1.34) 0.3466 (1.03) -0.00323 (-0.11) -7.7020 5380.3500
Tunisia 0.1354 (1.21) -0.3642 (-2.65)* 0.01316 (0.241) -7.5351 5263.5500

* Significant at 5% Level

PANEL C. ML Estimation of ARFIMA(1,d,1) Models: ASIA

COUNTRY  φ1  θ1  d AIC LL

Bahrain 0.175 (2.56)* 0.139 (2.61)* -0.947 (-37.5)* -10.6000 20754.5000

Hong Kong -0.153 (-1.14) -0.0196 (-0.127) -0.156 (-5.7)* -12.5390 24544.5000
India 0.45 (5.00)* -0.61 (-5.81)* 0.138 (3.07)* -8.1910 16033.9000
Israel 0.0474 (1.55) -0.563 (-11.2)* 0.116 (2.8)* -7.2710 14233.9000
Japan 0.0825 (0.0951) -0.1 (-0.115) 0.0016 (0.0679) -7.1052 13909.0700
Kuwait 0.0106 (0.199) -0.3249 (-4.61)* -0.0354 (-1.1) -9.4924 18580.7000
Lebanon 0.3059 (9.40)* -0.7177 (-20.5)* 0.3649 (7.59)* -7.3175 14324.4500
Malaysia -0.3849 (-3.91)* 0.2914 (2.74)* 0.0277 (1.54) -7.4419 14567.8000
Oman 0.33192 (2.32)* -0.465 (-3.0)* -0.569 (-10.4)* -10.5862 7393.1900
Philippines -0.07 (-0.349) 0.158 (0.831) -0.052 (-2.85)* -7.2156 14125.1000
Russia -0.0962 (-1.03) -0.255 (-2.12)* 0.1062 (2.41)* -9.3367 6521.0490
Singapore -0.04689 (-0.26) -0.0544 (-0.285) 0.0099 (0.410) -8.4362 16513.6600
South Korea 0.1343 (2.1)* 0.087 (1.57) -0.083 (-4.46)* -6.6835 13083.7500
Sri Lanka 0.384 (2.39)* -0.515 (-2.85)* 0.066 (1.05) -8.5790 5992.1500

Turkey 0.432 (11.6)* -0.8132 (-29.6)* 0.266 (5.3)* -5.4 10571.96

* Significant at 5% Level

PANEL D. ML Estimation of ARFIMA(1,d,1) models: AMERICA & AUSTRALIA

COUNTRY φ1  θ1  d AIC LL

Argentina 0.6604 (14.2)* -0.80 (-25.9)* 0.179687 (3.97)* -6.1700 12097.6700

Canada 0.079 (0.138) -0.1 (-0.173) -0.006 (-0.271) -8.3590 16363.9000
Mexico 0.3946 (7.86)* -0.605 (-10.1)* 0.144 (3.36)* -6.5502 12822.7600
Uruguay 0.4754 (12.7)* -0.8326 (-34.1)* 0.2816 (5.59)* -6.7462 13206.3200
Venezuela -0.00502 (-0.1) -0.4157 (-6.49)* 0.0357 (1.18) -4.4919 8794.7800
Australia -0.296 (-0.76) 0.3179 (0.807) -0.024 (-1.52) -7.3514 14390.8500

* Significant at 5% Level
This Table shows the estimation of ARFIMA (1,d,q) models. LL indicates the log-likelihoods, d indicates the long memory parameter, and AIC is
the Akaike information criterion. T-statistics are in parentheses.

36
The International Journal of Business and Finance Research ♦ Volume 2 ♦ Number 1 ♦ 2009

Figure 1: The Variation of Long Memory Parameter (d) Across Countries

Figure 1.1: Africa (d) Parameter

0.5

0.4

0.3

0.2

0.1 d

0
Botswana Egypt Kenya Mautitius
-0.1

-0.2

Figure 1.2: Asia (d) Parameter

0.6

0.4

0.2

0
Philippines
Bahraihn

Israel

Lebanon

Oman

Turkey
India

Russia

South Korea
Hong Kong

-0.2
d
-0.4

-0.6

-0.8

-1

-1.2

Figure 1.3: America (d) Parameter

0.3

0.25

0.2

0.15
d
0.1

0.05

0
Argentina Mexico Uruguay

Figure 1 shows the variation of significant long memory parameter (d) across countries.

37
C. Floros ⎪ The International Journal of Business and Finance Research ♦Vol. 2 ♦ No. 1 ♦ 2009

CONCLUSIONS

In this paper, conditional maximum likelihood is employed to estimate ARFIMA models and test for long
memory in exchange rates. We examine the evidence of long memory in the daily exchange rates over
the US dollar. Since currency trading always involves buying one currency and selling another, we
consider data from 34 countries (from all five continents) covering the period April 1991 to April 2006.

The results from the selected ARFIMA(1,d,1) models show evidence of long memory in African and
Asian countries. The results from the European rates show weak evidence of long memory (d is not
significant), while the evidence from America and Australia is mixed. In general, we find that 17
exchange rates show strong evidence of long memory. This indicates that shocks to these rates persist
over a long period of time. Furthermore, the long memory hypothesis is stronger in emerging market
economies (economies with low-to-middle per capital income). This indicates that these transition
(emerging) markets are not stable, and hence they offer an opportunity to investors and traders to add
some risk to their strategies.

As an emerging market, a country should embark on an economic reform program that will lead it to
stronger and more responsible economic performance levels, as well as transparency and efficiency in the
capital market. An emerging market economy should reform its exchange rate system because a stable
local currency builds confidence in an economy, especially when foreigners are considering investing.
Exchange rate reforms also reduce the desire for local investors to send their capital abroad.

These findings are helpful to financial managers, traders and investors dealing with foreign exchange
rates. Further research should (i) investigate the predictability of foreign exchange rates using ARFIMA
time-series methods, and (ii) run a sensitivity analysis to ascertain the robustness and temporal stability of
the long memory.

REFERENCES

Baillie, R., C. Chung, and M. Tieslau (1996) “Analysing Inflation by the Fractionally Integrated
ARFIMA-GARCH Model”, Journal of Applied Econometrics, vol. 11, p. 23-40.

Barkoulas, J., C. Baum, and N. Travlos (2000) “Long Memory in the Greek Stock Market”, Applied
Financial Economics, vol. 10, p. 177-184.

Bhar, R. (2000) “Testing for long-term memory in Yen/Dollar exchange rate”, Working Paper,
Department of Economics and Finance, University of Western Sydney.

Cheng, T. C. K. (2001) “Long memory features in the exchange rates of Asia-Pacific countries”, Working
Paper, Department of Economics, National University of Singapore.

Cheung, Y-W. (1993) “Long memory in Foreign exchange rates” Journal of Business and Economic
Statistics, vol. 11, p. 93-101.

Cheung, Y-W., and K. Lai (1995) “A Search for Long Memory in International Stock Market Returns”,
Journal of International Money and Finance, vol. 14, p. 597-615.

Ding, Z., and C. Granger (1996) “Modeling Volatility Persistence of Speculative Returns: A New
Approach”, Journal of Econometrics, vol. 73, p. 185-215.

38
The International Journal of Business and Finance Research ♦ Volume 2 ♦ Number 1 ♦ 2009

Gil-Alana, L. A. (2000) “Mean reversion in the real exchange rates”, Economics Letters, vol. 69, p. 285-
288.

Gil-Alana, L. A. (2004) “Fractional integration and cointegration in the Japanese exchange rates”,
Japanese Economy, vol. 32, p. 12-35.

Gil-Alana, L. A. (2006) “Fractional integration in daily stock market indexes”, Review of Financial
Economics, vol. 15, p. 28-48.

Hiemstra, C., and D. Jones (1997) “Another Look at Long Memory in Common Stock Returns”, Journal
of Empirical Finance, vol. 4, p. 373-401.

Laurini, M. P. and M. S. Portugal (2003) “Long memory in the R$/US$ exchange rate: A Robust
analysis”, FINANCELAB Working paper, IBMEC Sao Paulo, FLWP-03-2003.

Lo, A. (1991) “Long-Term Memory in Stock Market Prices”, Econometrica, vol. 59, p. 1279-1313.
Soofi, A. S., S. Wang, and Y. Zhang (2006) “Testing for long memory in the Asian foreign exchange
rates”, Journal of Systems Science and Complexity, vol. 19, p. 182-190.

Taylor, M. P. (1995) “The economics of exchange rates”, Journal of Economic Literature, vol. 33, p. 13-
47.

Tolvi, J. (2003) “Long memory and outliers in stock market returns”, Applied Financial Economics, vol.
13, p. 495–502.

Vougas, D. (2004) “Analysing Long Memory and Volatility of Returns in the Athens Stock Exchange”,
Applied Financial Economics, vol. 14, p. 457-460.

BIOGRAPHY
BSc,Msc,Phd Candidate Ilaman Shagylyjov completed his first degree in Textile engineering at International
Turkmen-Turkish University, holds Bsc (Textile) and MSc (Finance) degrees from Queen`s University of
Belfast . He is currently phd degree student at Frankfurt Geothe University and his research interests include financial
banking and weak efficiency.

39