Professional Documents
Culture Documents
While most traditional science deals with supposedly predictable phenomena like gravity, electricity, or chemical
reactions, Chaos Theory deals with nonlinear things that are effectively impossible to predict or control, like turbulence,
weather, the stock market, our brain states, and so on. Hence, such processes focus on non-randomness, nonlinearity
and chaotic characteristics. In recent times, such nonlinear dynamics and chaotic dynamics have augmented in the
field of financial analysis. This paper studies the extent to which the daily return data from the Indian Stock Exchange
indices (Nifty and Sensex) exhibit these nonlinear, non-random characteristics. The Hurst exponent in rescaled
range analysis rejects the hypothesis that the index return series are random, independent and identically distributed.
The BDS test provides evidence for nonlinearity. The results confirm the existence of fractal structure (i.e., self-
similarity across different scales) in Nifty and Sensex. Basing on these results, technical analysis theory, i.e., Elliot
Wave Theory, can also be justifiable.
Introduction
It is difficult to tell from the data whether an observed process is random or chaotic, because
in practice no time series consists of a pure ‘signal’. Thus, any real time series, even if typically
deterministic, contains some randomness. Recent research impetus in the capital market
arena is to develop theories that are capable of explaining the random movements in asset
prices and returns. An explanation of such stock return behavior has drawn on the field of
nonlinear dynamics. Nonlinear dynamics is ideally explained by the Chaos Theory which is
based on the assumption that the underlying system is a nonlinear and a deterministic
process.
Chaotic systems have attracted the attention of many researchers from different fields
because of two reasons. First, it is relatively easy to tune the parameters of dynamic economic
models so that they generate complex dynamics (Benhabib and Nishimura, 1979). Second, it
is easy to construct examples of nonlinear dynamics that appear random in linear tests (Sakai
and Tokumaru, 1980). The existence of chaotic behaviors can be detected by the following
indications:
• Chaotic process is not random but Independent and Identically Distributed (IID).
• Chaotic process is nonlinear.
* Assistant Professor, Xavier University, Bhubaneswar, Odisha, India; and is the corresponding author.
E-mail: sibanjan@gmail.com
** Reader (Retd.), Orissa Education Service-I (OES), Government of Odisha, Odisha, India.
E-mail: bimalmishra1955@yahoo.in
© 2015
Test IUP. All Rights
of Persistence Reserved.
in Indian Stock Market: A Rescaled Range Analysis 5
• Chaotic process is a deterministic process which retains its dimensionality when
it is placed in a higher embedding dimension.
• Chaotic process is sensitive to initial conditions.
As the Efficient Market Hypothesis (EMH) and the random walk model fail to capture
the behavior in many stock markets, some researchers like Brock et al. (1991); and Hsieh
(1991) and (1993) have pointed out that nonlinear dynamics is appropriate to explain the
complexity in many stock returns series. Fama (1965) admits that linear modeling techniques
are limited in capturing the complicated patterns which chartists claim to see in stock prices.
Papaioannou and Karytinos (1995) indicated that a non-efficient market is very suitable for
chaotic analysis, because the number of underlying moving forces (degree of freedoms in
statistical terms) is fewer than those in developed and more efficient markets.
Thus, the objective of this paper is to examine whether the time series of the indices
returns in Indian stock exchanges is generated by nonlinear dynamical system. If so, the
predictive ability of the system is strongly limited, especially for long-run predictions.
Furthermore, we focus on examining whether chaos theory can capture the complex and
random behavior that may not be obtained or derived from a stochastic approach or the
traditional random walk model.
Literature Review
Test of persistence and its presence in the financial time series has been discussed in several
past and recent papers. Peters (1992) shows that the S&P 500 Index monthly return series
has long-term dependence and a cycle length of 48 months. Peters (1994) finds that the same
phenomenon appears in Dow Jones Industrial Index. Hampton (1996) reports a strong and
continuing dependence on the S&P 500 returns series before the market crashed in 1987 and
1990. Opong et al. (1999) find the same evidence of non-IID in London Financial Times
Stock Exchange (FTSE) returns series. Same results are also reported in the other European
and some emerging markets. Errunza et al. (1994) find evidence of fractal dynamics in equity
returns of Germany, Japan, Argentina, Brazil, Chile, India and Mexico. Papaioannou and
Karytinos (1995) find evidence of dependence in the Athens Stock Market Index returns
series. Lin (1997) also reports same evidence of nonlinear dependence in the Dow Jones
Industrial Average return series. Dependence is more significant in 1995 when the US equity
markets made the start of a strong bull run. Pandey et al. (1997) report evidence of nonlinear
dependence in the index returns of Hong Kong, Japan and the US.
Regarding the emerging markets, Hamill and Opong (1997) report nonlinear dependence
in Irish stock returns, while Papaioannou and Karytinos (1995) find that the Athens Stock
Exchange Index returns are also nonlinearly dependent. Peters (1994), Vandewalle et al.
(1997), Carbone et al. (2004), Grech and Mazur (2004), Di Matteo et al. (2005), Di Matteo
(2007), Alvarez-Ramirez et al. (2008), Czarnecki et al. (2008) and Matos et al. (2008) estimate
Hurst exponent H with the theoretical value for an independent process of 0.5. Hurst
exponent of 0.5 indicates two possible processes: either independent (Beran, 1994) or short-
Methodology
Rescaled Range (R/S) Analysis
Whether the stock price movement follows a random walk or not can be detected by the
rescaled range analysis or R/S analysis. The R/S analysis is an ideal statistical tool for analyzing
the occurrence of rare events and is robust to possible nonlinear process that normality
assumption may not be needed. The result of the R/S analysis is the Hurst exponent, which
is a measure of the bias or trend in a time series.
N
The dataset is portioned into a sequential non-overlapping blocks, as A , where A is
n
the number of partition, N is the amount of data in the series and n is the amount of data in
each partition. The data in each block is x( t,a ). The mean of x ( t ,a ) ( x a ) for the ath block of data
is defined as
n
1
x a x t,a
n t 1
...(1)
The accumulated differences, X a , between each x ( t ,a ) and the mean for each block of
data, x a is given by
n
Xa ( x
t 1
( t ,a ) xa ) ...(2)
The range, Ra, is defined as the difference between the minimum and the maximum
cumulative deviation for each block of data, given by
Ra = max (Xa) – min (Xa) ...(3)
The standard deviation, Sa, for each block of data is determined, given by
( x
t 1
( t ,a ) x a )2
...(4)
Sa
n
The average rescaled range (R/S)n for length n, which is computed for different lengths
until n = N/2, is defined as
A
1 R
( R / S)n a
A a 1 Sa
...(5)
The final step is to apply an OLS regression with ln(R/S)n as the dependent variable and
ln (n) as the independent variable through the plot. The regression coefficient of the regression
Equation (6) provides the estimation of H, the Hurst exponent:
ln(R/S)n = H . ln(n) + C ...(6)
The value of H can be interpreted in the following ways:
• H = 0.50 denotes a random and statistically independent (uncorrelated) series—
a random walk. The present does not influence the future. The correlation
coefficient is 0. Its probability density function is normal. Such process increases
with the square root of time.
• 0.50 < H 1.00 denotes a ‘persistent’ or trend-reinforcing series. That is, the data
contains long-term memory and has a tendency to follow the current trend in the
next period. This process is said to be mean-averting.
• H < 0.50 denotes an ‘anti-persistent’ or ergodic series. That is, the data has a
tendency to reverse the current trend. This process is said to be mean-reverting.
BDS Test
BDS test was suggested by Brock et al. (1987). This hypothesis testing uses the test statistics
in which the mechanism is based on the correlation integrals. The BDS test is a powerful tool
for detecting serial dependence in time series. It tests the null hypothesis of IID against an
unspecified alternative. The null and alternative hypotheses are as follows:
H0: The data are Independently and Identically Distributed (IID).
H1: The data are not IID; this implies that there may be some serial dependence. If the linear
dependence has been removed in the time series, the serial dependence is thus nonlinear.
However, BDS test is unable to distinguish between nonlinear deterministic chaos and
nonlinear stochastic systems. BDS test cannot test chaos directly but only nonlinearity,
provided that any linear dependence has been removed from the data (e.g., using traditional
ARIMA-type models or taking a first difference of natural logarithms). Nevertheless, since a
nonlinear process is one of the indications of chaos, we may use the BDS test to detect such
indication. The BDS test statistics is defined by:
This test will be repeated at different values of ‘’ and ‘m’. Lin (1997) suggested that the
appropriate values of range from 0.5 to 2. Brock et al. (1987) pointed out the appropriate
values of m to be between 2 and 5.
2
Fluctuation FDFA ( n, l) which is defined as
n
1
2
FDFA ( n, l)
n Y
t 1
2
n ,l ( t ) ...(9)
Scales as
2
FDFA ( n, l) ~ cn 2H(l ) ...(10)
where c is a constant independent of n (Weron, 2002). We again run an ordinary least squares
regression on logarithms of (10) and estimate Hurst exponent H(l) for set l-degree of
polynomial trend in the same way as for R/S as
2
log b FDFA ( n, l) ~ log b c H(l)log b n
DFA can be adjusted and various filtering functions Xn,l can be used.
Tables 4 and 5 indicate that all the test statistics of logarithmic index returns series are
greater than the critical value of 1.96 significantly under different embedding dimension and
ratio of tolerance to standard deviation. Thus, the null hypothesis of IID is rejected for both
series. The results strongly suggest that both the series are nonlinearly dependent at the 5%
level of significance, a characteristic of chaotic behavior. We have shuffled the logarithmic
index returns series 10 times and subjected every shuffled series to the BDS test. The results
are significant at 5% level. It implies a strong evidence of nonlinear dependence on the
logarithmic index returns time series in both the indices.
0.0e+00 0.0e+00
0 1000 2000 3000 0 500 1000 1500 2000
Time Time
RMSE RMSE
125
115.5
log(RMSE)
log(RMSE)
H=0.618 H=0.655
123
113.5
2 3 4 5 6 7 2 3 4 5 6 7
log(scale) log(scale)
The results of DFA-1 coincide with results of R/S test with Hurst coefficient greater than
0.5 in both the series. The Sensex reported H value of 0.618, whereas the Nifty reported H
value of 0.655 (Table 6). Both the series are persistent at the scale ratio of 2 to 12. Scale ratio
is the ratio of successive scales. Thus, the traditional R/S method is a strong measure of
persistence even when compared with the recent technique of DFA.
Conclusion
In this study, we have examined the behavior of the Indian Stock Exchange indices returns
series using the R/S analysis, the BDS test and DFA. The Hurst exponents found in the rescaled
range analysis are greater than 0.5 in both series and reveal that both series are persistent,
fractal and self-similar. These results are confirmed using DFA for both the series. However, the
values of the Hurst exponents are relatively low (around 0.63), which indicates that both series
have strong noisy components. Nonlinearity was found in our time series using the BDS test.
The test statistics are very significant, which indicates a strong evidence of nonlinearity. The
findings of nonlinear serial correlation in both series are consistent with the other studies that
the BDS test has been applied to detect nonlinear structure in the financial data. These studies
include foreign exchange rate data (Hsieh, 1989), NYSE weekly stock return (Scheinkman and
LeBaron, 1989), and US daily return of futures contracts (Yang and Brorsen, 1993). Nonlinearity
is one of the indications of chaotic behavior. Thus, the results confirm that the indices exhibit
chaotic behavior as the returns series is characterized by non-randomness, nonlinear dependence
and deterministic.
The existence of fractal structure implies that the technical analysis tools like the Elliot
Wave Principles can be utilized when analyzing the price movements in any stock markets.
The chaotic behavior found in our time series indicates that the predictions of the price
movements in the Indian stock market are very difficult by the traditional econometric
methods, especially in the long term. The limitations of long-term prediction strongly suggest
that the techniques of charting (technical analysis) which try to infer the short-term future
behavior of the asset prices based on observed up and downturns in the past, may be applied
when analyzing the price movements in the Indian stock markets.
References
1. Alvarez-Ramirez J, Alvarez J, Rodriguez E and Fernandez-Anaya G (2008), “Time-Varying
Hurst Exponent for US Stock Markets”, Physica A, Vol. 387, No. 27, pp. 6159-6169.
2. Assaf A (2006), “Persistence and Long-Range Dependence in the Emerging Stock Market
of Kuwait”, The Middle East Business and Economic Review, Vol. 18, No. 1, pp. 1-17.
3. Barkoulas J, Baum Ch and Travlos N (2000), “Long Memory in the Greek Stock Market”,
Applied Financial Economics, Vol. 10, No. 2, pp. 177-184.
4. Benhabib J and Nishimura K (1979), “The Hopf Bifurcation and the Existence and Stability
of Closed Orbits in Multi-Sector Models of Optimal Economic Growth”, Journal of Economic
Theory, Vol. 21, pp. 421-444.
5. Beran J (1994), Statistics for Long Memory Processes, Monographs on Statistics and Applied
Probability, Vol. 61, Chapman & Hall, New York.
Reference # 01J-2015-10-01-01