1: \documentclass[preprint,aps,showkeys,showpacs]{revtex4}
2: \usepackage{graphicx}% Include figure files
3: \usepackage{dcolumn}
4: % Align table columns on decimal point
5: \usepackage{bm}
6: % bold math
7:
8: %\nofiles
9:
10: \preprint{}
11: \def\para{\par\noindent}
12: \def\sqr#1#2{{\vcenter{\vbox{\hrule height.#2pt
13: \hbox{\vrule width.#2pt height#1pt \kern#1pt
14: \vrule width.#2pt}
15: \hrule height.#2pt}}}}
16: \def\square{\mathchoice\sqr34\sqr34\sqr{2.1}3\sqr{1.5}3}
17: \newcount\notenumber
18: \def\clearnotenumber{\notenumber=0}
19: \def\note{\advance\notenumber by 1
20: \footnote{$^{\the\notenumber}$}} \baselineskip 20pt
21: \begin{document}
22: \title{Persistence in Financial Markets}
23: \author{S.Jain and P.Buckley \\ Information Engineering,\\ The Neural
24: Computing Research Group,\\ School of Engineering and Applied
25: Science,\\ Aston University, \\ Birmingham B4 7ET, \\ U.K.}
26:
27: \begin{abstract}
28:
29: Persistence is studied in a financial context by mapping the time
30: evolution of the values of the shares quoted on the London Financial
31: Times Stock Exchange 100 index (FTSE 100) onto Ising spins.
32: By following the time dependence
33: of the spins, we find evidence for power law decay of the proportion
34: of shares that remain either above or below their \lq starting\rq\
35: values. As a result, we estimate a persistence exponent for the
36: underlying financial market to be $\theta_f\sim 0.5$.
37: \end{abstract}
38:
39: \pacs{05.20-y, 05.50+q, 75.10.Hk, 75.40.Mg, 89.65.Gh, 89.75.-k}
40: \keywords{Ising Model, Financial Markets, Persistence}
41: \maketitle
42: \section{Introduction}
43: \para In its most general form, persistence is concerned
44: with the fraction of space
45: which persists in its initial state up to some later time. The
46: problem has been extensively studied in recent years for pure spin
47: systems at both zero [1-4] and non-zero [5] temperatures.
48: \para For example, in the non-equilibrium dynamics of spin systems
49: at zero-temperature,
50: the system is prepared in a random state at $t=0$ and the fraction of
51: spins, $P(t)$, that persist in the same state as at $t=0$ up to some
52: later time $t$ is studied. For the pure ferromagnetic
53: two-dimensional Ising model the persistence probability has been
54: found to decay algebraically [1-4]
55: $$ P(t)\sim t^{-\theta},\eqno(1)$$
56: where $\theta\sim 0.22$ is the non-trivial persistence exponent
57: [1-3].
58: \para The value of $\theta$ depends on both the spin [6] and
59: spatial [3] dimensionalities; see Ray [7] for a recent review.
60: \para At non-zero temperatures [5], consideration of the global order
61: parameter leads to a value of $\theta_{global}\sim 0.5 $ for the
62: two-dimensional Ising model.
63: \para Very recently, disordered systems [8-10] have also been studied
64: and have been found to exhibit different persistence behaviour to
65: that of pure systems.
66: \para Persistence has also been studied in a wide range of experimental
67: systems and the value of $\theta$ ranges from $0.19$ to $1.02$
68: [11-13].
69: Much of the recent
70: theoretical effort has gone into obtaining the numerical value of
71: $\theta$ for different models.
72:
73: \para In this work we present the first estimate for a persistence
74: exponent extracted from {\it financial} data.
75: \para Long-range correlations in persistent and
76: anti-persistent random walks were first discussed by Mandelbrot
77: [14]. Zhang [15] has presented empirical evidence to support that
78: daily returns in composite indices are not completely randomized.
79: Here, on the other hand, we study the behaviour of the constituent
80: stock prices.
81:
82: \section{Financial Markets}
83: \para A financial market is an example of a complex
84: many-body system exhibiting many of the characteristics found in
85: model systems studied in statistical physics.
86:
87: There is an element of both co-operation and \lq frustration\rq\
88: [16] in the movement of share values. For example, share values of
89: companies in the same sector tend to move in the same direction
90: (either up or down) when subjected to identical external events. A
91: typical case here would be the movement in the value of shares in
92: oil companies on the news of over/under production. On the other
93: hand, there are also companies whose share values move in opposite
94: directions given the same event. Here, typical examples would be the
95: reactions in the share values of companies from the retail and
96: banking sectors on the news of an increase/decrease in interest
97: rates.
98:
99: \para In this work we make no assumptions about any underlying model
100: systems. Rather, we treat the historical share values of the
101: companies over time as the outcomes of some \lq experiment\rq\.
102: \para The financial market we
103: study is the set of companies quoted on the London Financial Times
104: Stock Exchange 100 (FTSE 100) share index.
105: \section{Method}
106: \para The data used in this study was obtained from Datastream [17], a
107: database of financial information, and refers to the end of day
108: prices over the randomly chosen ten year
109: period from 24 February 1995 to 1 February 2005. The data were mapped
110: onto Ising spins using the procedure outlined below.
111: \para The \lq base\rq\ share price, $(P^b_i(t=0), i=1\dots , 100)$, of
112: each of the companies appearing in FTSE 100 at the end of trading on
113: 24 February 1995 was noted to 2 decimal places.
114: All prices
115: for the shares used in this work were taken to be the closing values
116: at the end of trading. At $t=1$ (the end of trading on the next day)
117: the share price of each company,
118: $P_i(t=1), i=1\dots , 100$, was compared with the corresponding base price.
119:
120: \para We allocate a value $S_i(t=0)=+1$ if $P_i(t=1)\ge P^b_i(t=0)$ and
121: a value of $S_i(t=0)=-1$ if $P_i(t=1)<P^b_i(t=0)$.
122: \begin{table}
123: \centering
124: \begin{tabular}{|c|c|c|c|}
125: \hline
126: % after \\: \hline or \cline{col1-col2} \cline{col3-col4} ...
127: {Date} & {t} & {Closing price} & {$S_i(t)$} \\ \hline
128: 24 February 1995 & 0 & 166.00 & Base price \\ \hline
129: 27 February 1995 & 1 & 166.50 & +1 \\ \hline
130: 28 February 1995 & 2 & 172.75 & +1 \\ \hline
131: 1 March 1995 & 3 & 167.00 & +1 \\ \hline
132: 2 March 1995 & 4 & 161.75 & -1 \\
133: \hline
134: \end{tabular}
135: \caption{A mapping onto Ising spins of the end of day closing share
136: prices for a typical company quoted on the FTSE 100.}
137: \label{spinvalues}
138: \end{table}
139: Table 1 gives a typical example of the mapping. Note
140: that the value of the spin is determined with reference to the base
141: price and not the previous closing price. Furthermore, in this work
142: we disregard all fluctuations which may have taken place during the
143: day and simply use the end of closing prices. In the example
144: discussed in Table 1, as the spin has \lq flipped\rq\
145: when $t=4$, the value of $S_i(t\ge 4)=-1$, {\it irrespective} of
146: subsequent closing prices.
147: \para The values $\{S_i(t=0), i=1,\dots 100\}$ form the initial
148: configuration for our \lq spin\rq\ system. All of the subsequent
149: 10 years worth of data was converted into possible values of Ising
150: spins, $S_i(t)$, using the share values at $t=0$ as the base. As a
151: result, we are able to use
152: $S_i(t)$ to track the value of the underlying share relative to its
153: base price. It is worth noting that companies have to satisfy
154: certain qualification criteria
155: before they are included in the FTSE 100 [18]. As a result, in
156: practice, a given company's presence in the FTSE 100 can fluctuate
157: from year to year.
158: In our analysis we restricted ourselves to the core set of
159: companies remaining in the FTSE 100 throughout the time period under
160: consideration.
161:
162: \para Hence, the {\it first time} $S_i(t)\ne S_i(t=0)$ corresponds to
163: the underlying share value either
164: going above $(S_i(t)=+1)$ or below $(S_i(t)=-1)$ the base price also
165: for the first time. This gives us a direct analogy with the
166: persistence
167: problem that has been extensively studied in spin systems.
168: \para At each time step, we count the number of spins that still
169: persist in their initial $(t=0)$ state by evaluating [19]
170: $$n_i(t)=(S_i(t)S_i(0)+1)/2.\eqno(2)$$
171: Initially, $n_i(0)=1$ for all $i$. It changes to zero when a spin
172: flips (that is, the underlying share price goes above/below the base
173: price) for the first time. Note that
174: once $n_i(t) =0$, it remains so for all subsequent calculations.
175: \para The total number, $n(t)$,
176: of spins which have never flipped until time $t$ is then given by
177: $$ n(t)=\sum_{i}n_i(t).\eqno(3)$$
178: A key quantity of interest is $R(t)$, the density of non-flipping
179: spins [1]
180: $$R(t)=n(t)/N,\eqno(4)$$
181: where $N$ is the number of companies monitored (note that $N$ is not
182: necessarily $100$ for the reasons outlined earlier). The actual
183: values of $N$ used are stated below. However, it's worth noting
184: that, in principle, we are dealing with a model system where the
185: spins are interacting with all other spins. As a result, we do not
186: believe that our system of spins is too small.
187: \section{Results}
188: \para We now discuss our results.
189: In this initial study, three different time periods were considered:
190: \begin{description}
191: \item[a)] 24 February 1995 to 1 February 2005 ($N=79$)
192: \item[b)] 9 January 1996 to 28 December 2000 ($N=79$)
193: \item[c)] 3 January 2000 to 3 January 2005 ($N=92$)
194: \end{description}
195: \para These sets were selected at random. For each time period, the
196: initial and subsequent spin configurations
197: were generated as outlined above and the resulting data analysed for
198: persistence behaviour.
199: \begin{figure}
200: \includegraphics{Jainfig1.ps}
201: \caption{A plot of $\ln R(t)$
202: against $\ln t$ for the data commencing 24 February 1995. The two
203: straight lines are guides to the eye and have slopes
204: $-0.4$ and $-1.4$ as indicated.}
205: \label{fig1}
206: \end{figure}
207: \para In Fig. 1 we show a log-log plot of the density of
208: non-flipping spins against time for the first set of data commencing
209: 24 February 1995. There is evidence for an initial power-law decay
210: (slope = -0.4), leading to a faster subsequent decay with slope =
211: -1.4.
212: \para A key feature of Fig. 1 is that nearly all of the
213: spins have flipped after $\ln t\approx 4$. This corresponds to
214: approximately $30-50$ days of trading on the markets.
215:
216: Note that we
217: are not distinguishing between those shares that remain above or
218: below their base values. There are also a few shares (in this case
219: just 2) that still persist in their initial state over the entire
220: observation period of 10 years.
221: \begin{figure}
222: \includegraphics{Jainfig2.ps}
223: \caption{A log-log plot of the data commencing 9 January 1996. The
224: straight line (guide to the eye) has a slope of $-0.33$.}
225: \label{fig2}
226: \end{figure}
227: To investigate the problem further, the same data was partitioned
228: into essentially two 5-year blocks as outlined above. The analysis
229: was repeated on each of the two sets of data. In Fig. 2 we
230: plot $\ln R(t)$ against $\ln t$ for the data commencing 9 January
231: 1996. Note that for this plot the base prices are determined by
232: close of trading on 9 January 1996. Once again, there is clear
233: evidence for a power-law decay. This time, however, the slope of the
234: linear fit is $-0.33$.
235: \begin{figure}
236: \includegraphics{Jainfig3.ps}
237: \caption{A log-log plot of the data commencing 3 January 2000. Once
238: again, the straight line (slope $-0.55$) is a guide to the
239: eye.}
240: \label{fig3}
241: \end{figure}
242: \para Finally Fig. 3 show a similar plot for the data commencing 3 January
243: 2000. once again, we have evidence for initial power-law decay
244: (slope = $-0.55$) and most of the spins appear to have flipped after
245: $\ln t\approx 4$.
246:
247: Although, as expected, there is noise in the data, we see that
248: in all three cases we have clear evidence of an initial power-law
249: decay. It\rq s\ also clear from the plots that nearly all of the
250: spins have flipped after a fairly short period of time,
251: corresponding to approximately
252: $30-50$
253: days of trading on the markets. However, there are a handful of
254: spins which do not flip over the entire time period under
255: consideration.
256:
257: Furthermore, there appears to be a cross-over to a faster power-law
258: decay for longer times. This cross-over is only really evident in
259: Fig. 1 as there is too much noise in the other 2 sets of data. The
260: initial decay in Fig. 1 has a slope of $-0.4$. The faster decay is
261: indicated by the straight line which has a slope of $-1.4$. The
262: initial power law decays are indicated by the straight lines in
263: Figures 2 and 3. The cross-over could be a signature of the market
264: reacting to external events such as significant interest rate
265: variations or political news.
266:
267: From the linear fits,
268: we can extract a value of $\theta_f$ ranging from $0.33$ to $0.55$.
269: As a consequence, we estimate the persistence exponent for the
270: financial data to be $\sim 0.5$. We believe this to be the first
271: estimate of a persistence exponent from financial data.
272: \para Our value for $\theta_f$ is not inconsistent with
273: the value obtained from computer simulations of the $2D$ Ising model
274: at a non-zero temperature [5]. This is an intriguing result as we
275: have made no assumptions whatsover about the underlying model which
276: gives rise to the financial data.
277: Of course, in our analysis, the value of each $S_i(t)$ incorporates the
278: overall performance of the shares of the underlying company relative
279: to the base value.
280: \section{Conclusion}
281: \para To conclude, we have used a novel mapping to map the share values
282: quoted on the London Financial Times Stock Exchange 100 share index
283: onto Ising spins. As a result, we extracted a value of $\sim 0.5$
284: for the persistence exponent. This should be regarded as an initial
285: estimate and further work is required to confirm the value. It
286: should be noted that, out of necessity, we worked with end of day
287: closing prices. Ideally, it would be better to use tick-data. It\rq
288: s remarkable that our value is not inconsistent with the value of
289: the persistence exponent obtained for the $2D$-Ising model at
290: non-zero temperature. This observation justifies further
291: investigation.
292:
293: \section*{References}
294: \begin{description}
295: \item {[1]} B. Derrida, A. J. Bray and C. Godreche, J. Phys. A:
296: Math Gen {\bf 27},
297: L357 (1994).
298: \item {[2]} A. J. Bray, B. Derrida and C. Godreche, Europhys. Lett.
299: {\bf 27},
300: 177 (1994).
301: \item {[3]} D. Stauffer J. Phys. A: Math Gen {\bf 27}, 5029 (1994).
302: \item {[4]} B. Derrida, V. Hakim and V. Pasquier, Phys. Rev. Lett.
303: {\bf 75},
304: 751 (1995); J. Stat. Phys. {\bf 85}, 763 (1996).
305: \item {[5]} S. N. Majumdar, A. J. Bray, S. J. Cornell, C. Sire, Phys.
306: Rev. Lett. {\bf 77}, 3704 (1996).
307: \item {[6]} B. Derrida, P. M. C. de Oliveira and D. Stauffer, Physica
308: {\bf A}224, 604 (1996).
309: \item {[7]} P. Ray, Phase Transitions {\bf 77} (5-7), 563 (2004).
310: \item {[8]} S. Jain, Phys. Rev. E{\bf 59}, R2493 (1999).
311: \item {[9]} S. Jain, Phys. Rev. E{\bf 60}, R2445 (1999).
312: \item {[10]} P. Sen and S. Dasgupta, J. Phys. A: Math Gen {\bf 37},
313: 11949 (2004)
314: \item {[11]} B. Yurke, A. N. Pargellis, S. N. Majumdar
315: and C. Sire, Phys. Rev. E{\bf 56}, R40 (1997).
316: \item {[12]} W. Y.
317: Tam, R. Zeitak, K. Y. Szeto and J. Stavans, Phys. Rev. Lett. {\bf
318: 78}, 1588 (1997).
319: \item {[13]} M. Marcos-Martin, D. Beysens, J-P
320: Bouchaud, C. Godreche and I. Yekutieli, Physica {\bf D}214, 396
321: (1995).
322: \item {[14]} B. Mandelbrot, International Economic Review {\bf 10},
323: 82 (1969).
324: \item {[15]} Yi-C. Zhang, Physica {\bf A}269, 30 (1999).
325: \item {[16]} G. Toulouse, Comm. Phys. {\bf 2}, 115 (1977).
326: \item {[17]} Datastream [Finance, accounting and economics data
327: online] (2005).
328: \item {[18]} Fact File, London Stock Exchange,
329: London, England (1997).
330: \item {[19]} S. Jain and H. Flynn, J. Phys.
331: A: Math Gen {\bf 33}, 8383 (2000).
332: \end{description}
333: \end{document}
334: