1: \documentclass{article}
2: \pagestyle{plain}
3: \renewcommand{\baselinestretch}{1.5}
4: \sloppy
5: \begin{document}
6:
7: \noindent {\bf Scaling in the Bombay Stock Exchange Index }
8: \vskip 1.5 cm
9:
10: Ashok Razdan\\
11: Nuclear Research Laboratory\\
12: Bhabha Atomic Research Centre\\
13: Trombay, Mumbai-400085\\
14: \vskip 2.0 cm
15: \noindent{\bf Abstract:}
16:
17: In this paper we study BSE Index financial time series for fractal
18: and multifractal beahiour. We show that Bombay
19: Stock Exchange (BSE) Index time series
20: is mono-fractal and can be represented by a fractional Brownian motion.
21: \vfill\eject
22: \noindent{\bf Introduction:}
23:
24: A lot of activity has been witnessed in recent times to study the nature
25: of finanicial time series. Econophysics [1-4] is an interdisciplinary
26: field of research in which methods of Physics and Mathematics are applied
27: to analyse economic systems. Applying Physics to financial problems
28: offers fresh look to existing theories of finance.
29: On the other hand, the availabity of long
30: term financial data and its behaviour on both at short and long
31: time scale offers new interesting challanges to physicsts.
32: Indeed ,attempts have been made to look for long range and
33: [3,5] and short range power-law corelation [6].
34: In this paper we will study scaling behaviour of Bombay Stock Exchange
35: (BSE) Index of last one year from Ist January 2000 to 31st December 2000
36: (shown in figure 1).
37: We will use R/S analysis to look for future trends
38: and search for multifractal
39: or Bownian motion features in its time dependence.\\
40:
41: Fractals have been applied in many fields of sciences like physics,
42: Biology,Chemical Sciences, Astrophysics and Engineering Sciences.
43: However ,it is less known that the concept of fractals was born
44: in the field of economics when Mandelbrot
45: in 1963 was investigating the price
46: changes dynamics of an open market in the year 1963
47: [ 7]. He found similarity between various
48: charts of market price changes (of cotton price) with different time resolution.
49: He came to conclusion that such scale invariance could help to characterise
50: many complex phenomena seen in physical sciences. Mandelbrot [2] observed
51: that price movements follow a family of distributions which have high peaks
52: and fat tails. Such distributions are known as Stable Paretian which
53: have infinite or undefined variance.\\
54:
55: \noindent{\bf Stock Market Returns: A new paradigm}
56:
57: Earlier it has been argued that genuinely competitive
58: stock market, returns ,follow
59: random walk model and are normally distributed [ 8 ].
60: Since stock market is a large system and has large
61: degrees of freedom (or investors),there is a underlyning
62: assumption that today's change in price is caused only
63: by today's unexpected "new" information. This means that
64: today's returns have nothig to do with yesterday's behaviour
65: and there are no "memory" effects i.e. the returns are
66: independent. This leads to the argument that the data
67: of stock prices and returns should follow normal distribution
68: with stable mean and finite variance. Because of this
69: arguement, capital market efficiency theories are mainly based
70: on random walk model or classical Brownian motion concepts.
71: This approach means that information arrives to an investor
72: linearly and reaction of an investor to "linear" information
73: is instantanous. This is based on assumption that
74: yesterday's information has been already folded into yesterday's
75: price.
76:
77: However, the actual market data shows that returns are
78: not normally distributed but have higher peak than theoritically
79: predicted around the mean
80: and have fatter tails.
81: Dow Jones Industrial Index from 1963 to 1993 shows lepokurtic
82: distrbution [8]. Apart from Dow Jones other stock exchanges
83: of western countries also show non-normal behaviour [8-11].
84: The presence of fatter tails indicate
85: "memory" effects which arise due to non-linear stochastic processes.
86: Actually the information flow to an investor is
87: clustered and its arrival is
88: irregular rather than continuous and
89: smooth in nature. This clustered and/or irregular arrival of
90: "new' information results in periods of low and high volatility [2]
91: which results in 'leptokurtic' distribution instead of normal.
92: This brings in a new paradigm in which reaction of investor or
93: trader to new information is "non-linear".
94: To investigate the
95: validity of this new paradigm the concepts of choas theory and fractals
96: have been used extensively. Models
97: like ARCH and GARCH [12,13] have been used to include memory
98: effects but these models
99: are not popular (mainly from Physics point of view [ 3 ])
100: because they donot take into account scaling property
101: of the procress.
102:
103: \noindent{\bf Brownian Motion and Fractional Brownian motion concepts:}
104:
105: In economics concept of classical Brownian motion [ 3,9]
106: have been widely used
107: to take into account 'memory effects' which get revealed in
108: the power law behaviour based on random walk model.
109:
110: Let x(t) be the position of particle
111: (which is random function of time), than for Brownian motion
112: \begin{equation}
113: x(t)-x(0) =\zeta\|t-t_0|^ h
114: \end{equation}
115: The position x(t) is obtained when $x(t_0)$ is
116: known and by choosing a random
117: number $\zeta$ from a Gaussian distribution.
118: Here, h=$\frac{1}{2}$ for classical Brownian motion (bm).
119: In Brownian motion it is not the position of the particle
120: which is independent of its position at another time but
121: it is displacement of a given particle at one time which is
122: independent of its displacement at another time interval.
123: Brownian motion is 'self-affine' by nature [ 2 ].
124: A transformation that scales time
125: and distance by different
126: factors is called affine and behaviour that reproduces itself under
127: affine transformation is called self-affine [ 2].
128: Again it was Mandelbrot [ 2 and refrences therein]
129: who intoduced the concept of fractional Brownian motion (fbm).
130: Exponent h varies from 0 to 1 in
131: the above equation for fractional Brownian motion.
132: For h=$\frac{1}{2}$, the time series is
133: independent and uncorrelated but the distribution may not be Gaussian.
134:
135: \noindent{\bf Hurst Analysis: Search for fractal behaviour}
136:
137: Hurst invented a new statistical method called, Rescaled range analysis
138: (R/S analysis) [2,14,15]. He was interested in developing the design of an ideal
139: reserviour based upon the given record of observed water discharges. Hurst
140: developed a new exponent called Hurst exponent (H) which can classify time
141: series into random or non-random . Hurst exponent is also related to fractal
142: dimension . A measure of a signal "roughness" is also given by Hurst exponent.
143: The "roughness" of a profile can be defined by observing how signal amplitudes
144: vary in time (and in space if necessary) ,in particular the correlation
145: between various amplitude fluctuations. R/S analysis is a method for
146: distinguishing completely random time series from a correlated time series.
147: The analysis begins by finding an average over a chosen time period say $\tau$
148: \begin{equation}
149: <Z>_{\tau} =\frac{1}{\tau} \sum_{t=1}^{\tau} Z(t)
150: \end{equation}
151: Let X(t) be the accumlated deparature of the influx Z(t) from the mean
152: $<z>_{\tau}$.
153: \begin{equation}
154: X(t,{\tau})=\sum_{u=1}^{t} (z(u)-<z>_{\tau})
155: \end{equation}
156: The difference between the maximum and the minimum accumlated influx X
157: is the range R which is given as
158: \begin{equation}
159: R({\tau})=max X(t,{\tau})-min X(t,{\tau})
160: \end{equation}
161: Hurst used the dimensionless ratio $\frac{R}{S}$, where S is the
162: standard deviation which is given as
163: \begin{equation}
164: S=(\frac{1}{\tau} \sum_{t=1}^{\tau} (Z(t)-<z>_{\tau})^2)^{\frac{1}{2}}
165: \end{equation}
166: Hurst found that the observed rescaled range, $\frac{R}{S}$ ratio
167: for a time series is given as
168: \begin{equation}
169: \frac{R}{S}= (\frac{\tau}{2})^H
170: \end{equation}
171: where H is the Hurst exponent.
172: Using the above equation we can find exponent H.
173: If H is between 0.5 and 1 , the trend is persistent which indicates
174: long memory effects . This also means that the increasing trend in
175: the past implies increasing trend in the future also or decreasing
176: trend in the past implies decreasing trend in the future also.
177: In contrast to this ,if H is between 0 and 0.5 than an increasing
178: trend in the past implies an decreasing trend in future and decreasing
179: trend in past implies increasing trend in future. It is important to
180: note that persistent stochastic processes have little noise whereas
181: anti-persistent processes show, presence of high frequency noise.
182:
183: The relationship
184: between fractal dimensions $D_f$ and hurst exponent H can expressed
185: as [11]
186: \begin{equation}
187: D_f= 2-H
188: \end{equation}
189:
190: So by finding Hurst exponent of a financial time series,we can find
191: out the fractal dimension of the time series.When $D_f$=1.5 ,there is
192: normal scaling. When $D_f$ is between 1.5 and 2,
193: time series is anti-persistent
194: and when $D-f$ is between 1 and 1.5 the time series is
195: persistent. For $D_f$=1,time series is smooth curve and purely
196: deterministic in nature and for $D_f$=1.5 time series is purely random.
197: Long term correlations of indexes in developed and
198: emerging markets have been studied by using Hurst analysis and
199: detrended fluctuation analysis (DFA) as investigating tools [24].
200: However, it has been argued by Vandewalle and Ausloos [25]
201: that DFA analysis is better than Hurst scaling for
202: short term time series.
203:
204:
205: \noindent {\bf Global Hurst Exponent: Search for multifractal nature}
206:
207: In general box counting method is being used for studying
208: multifractal features [15,16,17,18] but this method is mainly
209: prefered for problems of spatial nature. However,
210: to study multifractal nature of financial time
211: series, alternative methods have been suggested [19].
212: This technique involves calculating qth order height-height correlation
213: function or qth order structure function [19,20] of a normalized time
214: series y($t_i$)
215: \begin{equation}
216: c_q(\tau)=<|y(t_{i+r})-y(t_i)|^{q}>
217: \end{equation}
218: where only non-zero terms are considered in the average,taken over
219: all pairs ($t_{i+r},t_i$) such that
220:
221: \begin{equation}
222: \tau=|t_{i+r} -t_i|
223: \end{equation}
224: and
225: \begin{equation}
226: c_q(\tau) \sim {\tau}^ {\eta(q)}
227: \end{equation}
228:
229: where q $\ge$ 0 is the order of moment and $\eta(q)$ is the scale
230: invarient structure function exponent. For q=1, H=$\eta(1)$ is the
231: "Hurst" exponent characterizing the scaling non connservation of
232: mean. For q=2, we obtain $\eta(2)$=$\beta$-1,where $\beta$ is the
233: slope of the fourrier power spectrum. In general $\eta(q)$ is given
234: by
235: \begin{equation}
236: \eta(q)= qH -{\frac{C_1}{\alpha-1}} (q_{\alpha}-q)
237: \end{equation}
238:
239: where $C_{1}\le$ d is an intermittency parameter, d is the dimension of
240: space( here d=1) and $\alpha$ varies between 0 and 2. $\alpha$ is the
241: Levy index.
242: A multifractal process is characterized by a non-linear behaviour
243: of $\eta(q)$ [21] because of multiplicative cascades where as
244: those processes which are additive in nature, $\eta(q)$ is linear
245: or bi-linear. For Brownian motion (bm), $\eta(q)$=$\frac{q}{2}$ and
246: for fractional Brownian motion (fbm), $\eta(q)$=qH [22,23].Thus for a purely
247: bm of fbm ,$\eta(q)$ is linear, whereas for multifractal nature $\eta(q)$
248: is non-linear.\\
249:
250: \noindent{\bf Data Analysis and Results:}
251:
252: Following the above discussion , we analyse BSE index data of year 2000
253: from Ist january to 31st December which gives us 245 data points. Data
254: from time financial series are one dimensional and more simple to anlyze
255: than spatial one. For a financial time series there are no holidays or
256: weekends. In order to do R/S analysis data has been divided into 20,40,
257: .... 220,240 parts . The next step is to calculate R/S statistics and
258: plot it against the corresponding sample length on double lograithmic
259: plot as shown in the figure 2.
260: The Gaussian asymptotic behaviour of R/S which represents independent
261: random process with finite variances is given in line marked 'b' in figure 2
262: and can be written as
263: \begin{equation}
264: \frac{R}{S}=(\frac{\pi \tau}{2})^{\frac{1}{2}}
265: \end{equation}
266: It is clear from this figure that data fits
267: to the Hurst exponent H = 0.915 .
268: This value has been obtained by fitting $\frac{R}{S}$=$(a\tau)^{H}$
269: to the observed R/S value . The parameter a= 0.61 for the fit
270: given in the line marked a in figure 2.
271: This value of H is an example of persistent
272: behaviour. Fractal dimension for BSE index time series is $D_f$=1.085.
273: Large values of H , have been obtained for many naturally
274: occuring phenomena like monthly sunspot activity.
275: In figure 3 we have plotted $\eta(q)$ =qH(q), with q which
276: has obtained by calculating structure function.
277: We find that $\eta(q)$ has a linear relationaship with q which
278: as discussed above shows that BSE index does not have multifractal
279: nature but it represents a fractional Brownian motion.\\
280:
281: \noindent{\bf Conclusions:}
282:
283: By using $\frac{R}{S}$ analysis ,we are able to find fractal dimension
284: of of BSE Index. We also observed that trend in BSE index is Perisitent
285: for the Year 2000. This means if market is not doing well in the year
286: 2000, persistent trend will continue i.e. in future, market will
287: continue to give low returns. If we see behaviour of BSE from January
288: 2001 onwards, market perfomance has not been good.
289: The linear behaviour of qH(q) values of BSE Index
290: with q shows that signal is mono-fractal and data follows simple
291: scaling values for these values of q. But it important to note that
292: we have used data of only one year comprising 240 data points. It will be
293: intresting to look for multifractal features in Short term ( single day
294: data, but intraday behaviour)
295: or long term (may be decade or more) data of BSE Index.In other stock
296: exchanges such studies have shown multifractal features [24]. It is
297: important to make such studies because market returns have been
298: correlated to [24] multifractal features in the Index data,more so
299: when functioning of BSE will be more transparent from July 2001 onwards,
300: with the introduction of new market mechanisms like "OPITIONS" and
301: "FUTURES".\\
302: \noindent{\bf Acknowledgement:} I am thankful to Dr S.K.Chiragi for
303: his inputs, which helped me in the improvements of this paper.
304:
305: \vfill\eject
306: \noindent{\bf Figure Captions:}
307:
308: Figure1: It shows BSE index for the Year 2000 from Ist January 2000 to
309: 31 st December 2000 [26]. \\
310:
311: Figure2: It shows Log( R/S) with log($\tau$). Curve 'a' corresponds
312: to actual data and curve 'b' to asymptotoc theoritical expectation.\\
313:
314: Figure3: $\eta(q)$ =q H(q), has been plotted against q for BSE index.
315: It shows linear behaviour indicating fractional Brownian motion.\\
316:
317: \vfill\eject
318: \noindent{\bf References:}
319: \begin{enumerate}
320: \item J.M. Pimbly , Phys. Today 50(1997)42.
321: \item B.B. Mandelbrot, Fractals and Scaling in Finance
322: (Springer,New York ,1997 edition)
323: \item N.VandeWalle, M. Ausloos in Econophysics -an Emerging Sciences
324: Edited by J.Kertesz,I.Kondor (Kluwer,Dordrecht 1999 Edition)
325: \item R.N.Mantenga, H.E.Stanely , Nature 376(1995)46;ibid.,
326: Physica A 239(1997)255;ibid.,Physica A 254(1998)77
327: \item B.Chopard, R.Chatangny in Scale Invariance and Beyond, Edited by
328: B.Dubrulle, F.Graner, D.Sornette (EDP Sciences 1997)
329: \item K.Ivanova, M.Ausloos, Physica A 265(1999)279
330: \item B.B. Mandelbrot, J. of Business (Chicago)36 (1963)394
331: \item E.F.Femma, J.Finance (1970) 383-417
332: \item R.N.Mantegne, H.E. Stanely , An Introduction to Econophysics
333: Cambridge University Press ,2000.
334: \item E.E.Peters, Chaos and Order in Capital Markets, Wiley,New York,1991
335: \item E.E.Peters, Fractal Market Analysis,Wiley,New York,1994
336: \item J.D.Hamilton, Time Series Analysis,Princeton University Press,NJ,1994
337: \item R.Engle, Econometrica 50(1982)987
338: \item H.E.Hurst, Long Term Storage of Reseviours, Trans. Amer. Soc.
339: civil Eng. 116(1950)770-808
340: \item R. Weron,B.Przybylowicz, Physica A 283(2000)462
341: \item A.B.Chhabra,C.Meneveau, R.V.Jensen and K.R.Sreenivasan,
342: Physical Review A 40(1989)5284
343: \item A.Haungs,A.Razdan,C.L.Bhat,R.C.Rannot,H.Rebel, Astroparticle
344: Physics 12(1999)145
345: \item A.Razdan, A.Haungs,H.Rebel and C.L.Bhat, Astroparticle Physics(submitted)
346: \item A.L.Barabasi, T.Vicsek, Phys.Rev. A 44(1991)2730
347: \item F.Schmitt,D.Schertzer,S.Lovejoy , Int. J.Theor. Appl. Fin.,3(2000)361
348: \item G.Parisi and U.Frisch in Turbulance and Predicibility in Geophysical
349: fluid dynamics and Climate dynamics,eds M. Ghil et. al.(North-Holland)
350: 1955,pp84-92
351: \item B.J.West, The Lure of Modern Science:Fractal Thinking (World Scientific
352: 1985)
353: \item M.V.Berry ,Z.V.Lewis,Proc. R. Soc. London A370(1980)459
354: \item M.Beben and A.Orlowski, Euro. Phys. J. B 20(2001)527
355: \item N.Vandewalle and M.Ausloos, Physica A 246(1997)454
356: \item www.bseindia.com
357: \end{enumerate}
358:
359:
360: \end{document}
361:
362:
363: