Physica A 512 (2018) 222–229

Contents lists available at ScienceDirect

Physica A
journal homepage: www.elsevier.com/locate/physa

Stochastic process with multiplicative structure for the

dynamic behavior of the financial market
∗
Leonardo S. Lima , Greicy K.C. Santos
Departamento de Física, Centro Federal de Educação Tecnológica de Minas Gerais, 30510-000, Belo Horizonte, MG, Brazil
Centro Federal de Educação Tecnológica de Minas Gerais, 30510-000, Belo Horizonte, MG, Brazil

highlights

• The temporal series of the volatility and the return for the model is analyzed.
• The distribution of the returns is verified and the power of the long tail distribution gotten.
• The Hurst index is calculated using the rescaled range analysis.

article info a b s t r a c t

Article history: A stochastic model with multiplicative noise has been proposed as a mathematical model
Received 14 December 2017 for the prices dynamics of the financial market. We have presented a model which allows us
Received in revised form 18 June 2018 to test within the same framework the comparative explanatory power of rational agents
Available online xxxx
versus irrational agents with respect to facts of the financial market. We calculate the
Keywords: long range memory of the model and studied the behavior of the long tail distribution of
stochastic model the cumulative distribution of probabilities for the model with additive and multiplicative
noise.
© 2018 Elsevier B.V. All rights reserved.

1. Introduction

Microscopic dynamics containing multiplicative noise may be encountered in several dynamical systems. In general, the
dynamical behavior is a common feature of many non-equilibrium systems which can be also characterized by power-law
probability density functions [1]. The dynamics of prices of the market has been studied since some decades ago by Black
and Scholes with their famous equation for the price dynamics of the European market [2]. Being in the last decades, there
has been a large amount of data on this subject [3]. In general, the mathematical modeling of the stock market must simulate
the market structure, trading mechanism and price dynamics. Mike and Farmer have made an empirical behavioral model
(MF model) in an order-driven market to simulate the dynamics of stock price formation [4]. In following, Gu and Zhou [5]
have modified the MF model by incorporating long memory into aggressiveness of incoming orders [6].
In a general way, socioeconomic systems are complex systems in which extreme events occur more frequently and
exhibit complex behaviors [7,8]. They have been a great field for the application of concepts and mathematical methods
for theoretical physics, used to treat complex systems [9–11]. An important model that has been used for the modeling of
the financial market is the two-dimensional Ising model and its extensions [12–17]. An important quantity to study the
dynamics of the financial market is the volatility. The study of the volatility in time series is not only crucial for revealing the
underlined mechanism of the financial markets dynamics, but also useful for traders because it can help them to estimate

∗ Corresponding author at: Centro Federal de Educação Tecnológica de Minas Gerais, 30510-000, Belo Horizonte, MG, Brazil.
E-mail address: lslima@cefetmg.br (L.S. Lima).

https://doi.org/10.1016/j.physa.2018.08.049
0378-4371/© 2018 Elsevier B.V. All rights reserved.
 L.S. Lima, G.K.C. Santos / Physica A 512 (2018) 222–229 223

the risk and optimize the portfolio [18]. It is well known that financial time series look unpredictable and their future values
are essentially impossible to predict [3]. Another way for modeling the financial market is through stochastic differential
equations [19–21]. Moreover, the modeling through a set of linear ordinary differential equations also has been employed
in the literature for modeling the market [22].
An important thing in finances is the observation of scaling laws in financial markets that are the widespread power-law
behavior exhibited by large price changes. This is corroborated for practically all types of financial data and markets [23–
26]. The quantity of interest here is the relative price change or return. The statistical analysis of daily returns indicates
a strong evidence for a hyperbolic behavior of the tails of the distribution, were is well known that the distribution of
returns is given as r(t) ≡ ln S(t + ∆t) − ln S(t) [3] and the long tail cumulative probabilities distribution obeys to an
inverse cubic-law P(|r |) ∼ |r |−γ , where γ ∼ 3 is the tail exponent. The volatility is defined as the modulus of the return,
g(t) = |r(t)| [5,7,27–32]. In contrast, empirical analyses for other stock markets have unveiled power-law tail exponents
other than the Lévy regime and the inverse cubic-law. Moreover, the fluctuating character of the volatility in financial
markets has been considered as major responsible for change of the price dynamics [1,25].
In this work we intend to study the model of price dynamics given by Itô’s stochastic differential equation with a
multiplicative noise and a nonlinear polynomial interaction potential of the type φ 4 . In general, the stock price is governed by
stochastic differential equations with a linear interaction potential (i.e. without a term of type φ 4 in the potential) [19]. Here,
we intend to study the more general case with a nonlinearity of the type φ 4 included. The case of this model with additive
noise has been considered recently in [33]. Moreover, nonlinear terms in the stochastic equations have been proposed as a
model for stock market fluctuations and crashes [34]. Where the linear case represents ‘‘risk neutral’’ case. Terms of type φ 4
included can represent the market when it is unstable, with an exponential rise or decay of the stock value, corresponding
to a speculative bubble.
The plan of this paper is the following. In Section 2, we describe the stochastic model. In Section 3, we present the
numerical results and in Section 4, we present our conclusions and final remarks.

2. The model

The model is described by the following Itô stochastic differential equation with a multiplicative white noise

dX (t) = α X (t) − β (X (t))3 dt + γ (X (t))δ dW (t),

[ ]
(1)

with δ ∈ [0, 1] and dW (t) being the Winner increment. For β = 0.0, is well known that the model above gives a good fit
to option prices crossing different strikes at a single expiration date [19]. In this case, γ (X (t))δ−1 can be interpreted as the
volatility that is a decreasing function of the stock price. When one wishes to account for different volatilities implied by
options expiring at different dates as well as different strikes, one needs to allow γ to depend on t as well as x. The function
γ (t , x) is called the volatility surface [19].
The associated Fokker–Planck equation to the above model is given as
∂ P(x, t) ∂[(α x − β x3 )P(x, t)] γ ∂ 2 [xδ P(x, t)]
= + . (2)
∂t ∂x 2 ∂t2
The potential of type φ 4 makes the probability density solution of the steady state in a non-Gaussian form, for β ̸ = 0.0. For
β = 0.0, it assumes the Gaussian form. Moreover, Eq. (1) can be used to describe the behavior of a particle under action of
an force given by f (x) = −β x3 + α x plus a dissipative force −bẋ, being b a damping constant plus a stochastic noise of the
environment δξ (t), where we can take the limit of the inertial term negligible (m ≃ 0). Such model can represent a financial
market upon different extreme conditions on the process, where the price tends to go for two minimums, represented by
the double well. The case δ and ς = 0 has been considered in the Ref. [34] and is nominated as Bouchaud–Cont–Langevin
model for stock market fluctuations and crashes.
For β = 0.0, we can write the Eq. (1) as [21]
dXt
= (α − β Xt2 )dt + γ Xtδ−1 dW (t). (3)
Xt
Using the Itô’s differential given as
( )
1 1 1
d (ln Xt ) = dXt + − (dXt )2 (4)
Xt 2 Xt2
and
dXt 1
= d(ln Xt ) + γ 2 Xt2δ dt , (5)
Xt 2Xt2
we have
( )
1
d(ln Xt ) = α − β Xt2 − γ 2 Xt2δ−2 dt + γ Xtδ−1 dWt .
2
224 L.S. Lima, G.K.C. Santos / Physica A 512 (2018) 222–229

Fig. 1. Evolution of the time series of the return r(t) = ln X (t + ∆t) − ln X (t), for values of the parameter α = 1.0, β = 0.0 and δ = 1.0.

Thus
[( ∫ t
X (t) = X (0) exp α − β (X (t ′ ))2 dt ′
0
∫ t ) ∫ t ]
1
− γ2 (X (t ′ ))2δ−2 dt ′ + γ (X (t ′ ))δ−1 W (t ′ )dt ′ , (6)
2 0 0

where in the limit dt → 0 we can replace X (t ′ ) → X (t) in the first two integrals in the equation above. However, the same
procedure cannot be made in the last term because of the stochastic process dW , which becomes each point of the integrand
uncorrelated of each another point. Instead, we do the replacement
X (t) + X (t + dt)
X (t ′ ) → , (7)
2
and we find that
[( )
1
X (t) = X (0) exp α − β (X (t))2 − γ 2 (X (t))2δ−2 +
2
( )δ−1 ]
dXt
γ X (t) + W (t) . (8)
2
dXt δ−1
Where in the limit dX → 0, we have (X (t) + 2
) ≈ (X (t))δ−1 + O(...) and finally
[( )
1 2 2δ−2
X (t) = X (0) exp α − β (X (t)) − γ (X (t))
2
+
2
]
γ (X (t))δ−1 W (t) . (9)

For β = 0.0 and δ = 1.0 the solution of Eq. (1) is given as

[ ]
1
X (t) = X (0) exp (α − γ 2 )t + γ W (t) . (10)
2

3. Numerical results
√ √
We solve numerically the equation above making σw = ∆t and writing it as γ dW (t) ∼ γ dtRG . Where RG is an aleatory
generator number with a Gaussian distribution of mean zero and variance 1. In general, solutions of stochastic differential
equations have the Markovian property and if relate with each other, where there are two partial differential equations, one
that includes a discounting and the another not [19].
The time series of the return r(t) = ln X (t + ∆t) − ln X (t) is plotted in Fig. 1. We have gotten the time series oscillating
quickly where the width between the maximum value and minimum value of the time series r(t) do not vary a lot of with
the parameters α , β , γ and δ in the Eq. (1), as is showed in the Figure.
In Fig. 2, we show the Log–Log fit of the long-tail distribution of the volatilities which is well known to obey to law
of power given by P(|r |) = 1/|r |α [27–30]. We have gotten the exponent of the fitted curve of the long tail distribution
 L.S. Lima, G.K.C. Santos / Physica A 512 (2018) 222–229 225

Fig. 2. Graphic Log–Log of the cumulative probability distribution of the volatilities P(|r |) vs. |r |. The fit of least squares of the power law is given as
P(|r |) ∼ |r |−α . Where we have gotten the value of the exponent α as α = 12.20(3), for the long tail distribution of the volatilities. The solid line is the best
power-law fit to the data in the long tail limit.

Fig. 3. Log–Log graphic using the rescaled range (RS) method, log (R/s) vs. l = log N for the calculation of the Hurst index of the volatility and return (b),
for values α = 1.0 and γ = 1.0 and γ = 10.0 respectively. Furthermore, we use also δ = 0.0 and β = 0.0.
226 L.S. Lima, G.K.C. Santos / Physica A 512 (2018) 222–229

Fig. 4. Graphic Log–Log of the cumulative distribution of the volatility P(|r |) vs. |r | for the model with δ < 1 (δ = 0.25). The fit of least squares of power
law is given as P(|r |) ∼ |r |−α , where we have gotten the value of the exponent α as α = 17.22(2) for the long tail distribution of volatilities. The solid line
is the best power-law fit to the data in the long tail limit.

Fig. 5. Log–Log graphic using the rescaled range (RS) method, log (R/s) vs. l = log N for the calculation of the Hurst index of the volatility, for the model
with δ ̸ = 1, for values of α = 1.0, β = 0.0 and γ = 1.0. The value δ used is δ = 0.25.

as α = 12.20(3). However, as the so-called ‘‘power-law tails’’ have only about 0.1 orders of magnitude in any curve
like of Fig. 3, we can obtain any values of ‘‘power-law’’ tail exponent. Hence, we cannot think there are any "power-
law tails", in this case. For the R/S analysis figures, in Fig. 3, we have calculated the Hurst index using the rescaled
range (RS) method for the volatility and return (b) respectively, log(R/s) vs. l = log n, for the time series. We have
determined they for values of parameters β = 0.0 and α = 1.0 in Eq. (1). We obtain the Hurst index for the time
series of the returns, using the RS method as H = 0.5014(2) for the value γ = 1.0. The value near H ∼ 0.5
obtained for the Hurst index is due to exponential nature given for the solution in the Eq. (1) obtained for the cases
analyzed. However, the negative slopes obtained in the RS plots may be due to the artifact of the RS method for the model
studied.
Hence, the results obtained for the power-law with the tail exponent being as large as 12 or 17 and the Hurst index of
returns and volatility close to 0.5 indicates that the model is not a model for financial markets because the results are far
from well-known stylized facts of statistical regularities of financial returns and volatilities.
In Fig. 4, we show the fit Log–Log of the long tail distribution of the volatilities for the model with δ ̸ = 1.0
(δ = 0.25). We obtain a large change of the power-law fit to the data, for the behavior of long tail of the cumulative
probability distribution. In Fig. 5, we present the Hurst index for the model with δ ̸ = 0.0. As we have obtained the
Hurst index near H = 0.5018(3), we have not found any significant change in the Hurst index with the parameter
δ.
 L.S. Lima, G.K.C. Santos / Physica A 512 (2018) 222–229 227

Fig. 6. Log–Log graphic using the detrended fluctuation analysis (DFA) of F (n) ∝ nα of the volatility and return (b), for values α = 1.0 and γ = 1.0 and
γ = 10.0 respectively. Furthermore, we use also δ = 0.0 and β = 0.0.

3.1. Detrended fluctuation analysis (DFA)

We can also employ the detrended fluctuation analysis (DFA) to analyze the behavior of time series that appear to be long-
memory processes (diverging correlation time, e.g. power-law decaying autocorrelation function). The obtained exponent is
similar to the Hurst exponent, except that DFA may also be applied to signals whose underlying statistics (such as mean and
variance) or dynamics are not stationary and if relate to measure based upon spectral techniques such as autocorrelation
and Fourier transform.
In Fig. 6, we obtain the log–log graph of F (n) ∝ nα where n are different window sizes and F (n) is the fluctuation [35]. The
scaling exponent α is calculated as the slope of a straight line fit to the log–log graph of n against F (n) using least-squares.
This exponent is a generalization
√ of the Hurst exponent. Because the expected displacement in an uncorrelated random walk
of length N grows like N, an exponent of 12 would correspond to uncorrelated white noise. When the exponent is between
0 and 1, the result is Fractional Brownian motion, with the precise value giving information about the series self-correlations.
For the returns we have found a value of α exponent α ≈ 0.48 near to the value of Hurst index obtained using the RS method.
However, for the volatility we have obtained a larger value (α ≈ 0.66) from that obtained using the RS method. However, as
the RS method is rarely adopted in recent years and the lines obtained using the DFA method are not perfect straight lines in
log–log scales, we have the result of the DFA method must be more realistic. Moreover, the slopes obtained in the DFA plots
are not negative as that obtained in the RS plots, what shows that the results of the DFA method are more reliable.
228 L.S. Lima, G.K.C. Santos / Physica A 512 (2018) 222–229

4. Conclusions and final remarks

In summary, we study a Itô’s diffusion model with a multiplicative noise as a mathematical model for the financial market.
We have calculated the Hurst index for the model and verified the behavior of the long tail distribution of the volatilities,
which must obeys the well known power law [27–30]. We obtain the time series behavior for dynamics of prices X (t), and
the long tail power distribution of the volatility. Moreover, we have determined the Hurst index for this model that is used
as a measure of the long-term memory of the time series which if relates with the autocorrelations of the time series and the
rate at which this decreases as the lag between pairs of values increases. We have employed the rescaled range (RS) method
to estimate the Hurst indexes as H = 0.5014(2) for the value γ = 1.0. For γ ̸ = 0.0 such as γ = 0.25 we have not found a
significative change in the Hurst index found. However, results using the DFA method have given results a few different for
the Hurst index of the return and volatility. As the DFA method is more adopted in recent years, the results obtained using this
method are more reliable. Experimental results can give support to our theoretical results. How we have found a value of the
exponent of the long tail distribution > 3.0, the model Eq. (1) must not describe well the behavior of the American market but
this power law is only an empirical law and hence we do not can know the behavior of the market with certainly. However,
with relation to the real financial markets and price behaviors which is not only produced by a stochastic equation, we can
consider employ one of the real assets’ prices series, such as a stock, from the real financial markets to do some empirical
works for identifying the theoretical model proposed in this paper. The empirical works may strength the equation’s power
proposed here, given by Eq. (1). As for instance, employing SP500 index to compare with the numerical results obtained and
checking whether the stochastic equation can describe SP500 index’ behaviors well. Analyzing the behavior of time series
of some indexes SP500 such as Dow Jones Industrial Average, VIX, we can see that they obey a behavior look like the times
series obtained for the model Eq. (1), given in Fig. 1. What could revel the power of the theoretical model proposed in this
article.
In general way, The φ 4 potential V (x) has a minimum in x = x∗ , which is positive and a maximum in x = x∗∗ . The price
increment ∆x oscillates around a positive value, which means that there is a non zero trend not based on true growth but
entirely induced by the fact that a price increase motivates more people to buy. This is called a speculative bubble. Thus, the
stronger the φ 4 term β in Eq. (1) pulling the price back towards the fundamental value x0 , the shorter must be the duration
of speculative bubbles. In the literature there are some works for the study of the dynamic behavior of prices such as the
energy price [36–38]. However, until where I know there is none model of type studied here.

Acknowledgments

The Brazilian Agencies CNPq (Grant Nr. 1592874587824075) and FAPEMIG.

References

[1] Silvio M. Duarte Queiros, Celia Anteneodo, Constantino Tsallis, Power-law distributions in economics: a nonextensive statistical approach (Invited
Paper), in: Proc. SPIE 5848, Noise and Fluctuations in Econophysics and Finance, 2005.
[2] F. Black, M. Scholes, J. Polit. Econ. 81 (1973) 637.
[3] R.S. Mantegna, H.E. Stanley, An Introduction to Econophysics, Correlations and Complexity in Finace, fourth ed., Cambridge, New York, 2007.
[4] S. Mike, J.D. Farmer, J. Econom. Dynam. Control 32 (2008) 200.
[5] G.-F. Gu, W.-X. Zhou, Europhys. Lett. 86 (2009) 48002.
[6] X. Zhang, J. Ping, T. Zhu, Y. Li, X. Xiong, PLoS One 11 (8) (2016) e0160406.
[7] H. Meng, F. Ren, G.-F. Gu, X. Xiong, Y.-J. Zhang, W.-X. Zhou, W. Zhang, Europhys. Lett. 98 (2012) 38003.
[8] D. Sornette, Critical Phenomena in Natural Sicences - Chaos, Fractals, Self-organization and Disorder: Concepts and Tools, second ed., Springer, Berlin,
2004.
[9] T. Assenzaa, D.D. Gattia, J. Grazzinia, J. Econ. Dyn. Control 50 (2015) 5.
[10] C. Diks, J.X. Wang, J. Econ. Dyn. Control 69 (2016) 68.
[11] Daan in t Veld, J. Econ. Dyn. Control 69 (2016) 45.
[12] L.S. Lima, Physica A 482 (2017) 544.
[13] W.-X. Zhou, D. Sornette, Eur. Phys. J. B 55 (2007) 175.
[14] K. Sznajd-Weron, R. Weron, Internat. J. Modern Phys. C 13 (2002) 115.
[15] E. Callen, D. Shapero, Phys. Today 27 (1974) 23.
[16] E.W. Montroll, W.W. Badger, Introduction to Quantitative Aspects of Social Phenomena, Gordon and Breach, New York, 1974.
[17] A. Orléan, J. Econ. Behav. Org. 28 (1995) 274.
[18] F.W.K. Yamasaki, S. Havin, H.E. Stanley, Phys. Rev. E 77 (2008) 016109.
[19] Steven E. Shreve, Stochastic Calculus for Finance II Continuous-Time Models, Springer, New York, 2004.
[20] C. Gardiner, Stochastic Methods, A Handbook for the Natural and Social Sciences, fourth ed., Springer, New Zealand, 2009.
[21] B. Oksendal, Stochastic Differential Equations An Introduction with Applications, sixth ed., Springer-Verlag, Oslo, 2013.
[22] Paloma O.C. Xavier, A.P.F. Atman, A.R. Bosco de Magalhães, Phys. Rev. E 96 (2017) 032305.
[23] B. Mandelbrot, J. Bus. 35 (1963) 394.
[24] E.F. Fama, E. Mandelbrot, J. Bus. 35 (1963) 420.
[25] E.F. Fama, J. Bus. 38 (1965) 34.
[26] Thomas Lux, Michele Marchesi, Nature 397 (1999) 498.
[27] P. Gopikrishnan, M. Meyer, L.A.N. Amaral, H.E. Stanley, Eur. Phys. J. B 3 (1998) 139.
[28] P. Gopikrishnan, V. Plerou, Lus A.N. Amaral, M. Meyer, H.E. Stanley, Phys. Rev. E 60 (1999) 5305.
[29] V. Plerou, P. Gopikrishnan, Lus A.N. Amaral, M. Meyer, H.E. Stanley, Phys. Rev. E 60 (1999) 6519.
 L.S. Lima, G.K.C. Santos / Physica A 512 (2018) 222–229 229

[30] F. Botta, H.S. Moat, H.E. Stanley, Tobias Preis, PLoS One 10 (2015) e0135600.
[31] G.-F. Gu, W.-X. Zhou, Eur. Phys. J. B 67 (4) (2009) 585–592.
[32] J. Zhou, G.-F. Gu, Z.-Q. Jiang, X. Xiong, W. Chen, W. Zhang, W.-X. Zhou, Comput. Econ. 50 (2017) 579.
[33] L.S. Lima, L.L.B. Miranda, Physica A 490 (2018) 828.
[34] J.-P. Bouchaud, R. Cont, Eur. Phys. J. B 6 (1998) 543.
[35] C.-K. Peng, S. Havlin, H.E. Stanley, A.L. Goldberger, Quantification of scaling exponents and crossover phenomena in nonstationary heartbeat time
series, Chaos 5 (1995) 82.
[36] Fred Espen Benth, J.S. Benth, S. Koekebakker, Stochastic Modeling of Electricity and Related Markets, University of Oslo, Norway, 2008.
[37] G. Gallego, G. van Ryzin, Manag. Sci. 40 (1994) 999.
[38] O. Rusanov, M. Laskin, AIP Proc. 1836 (2017) 020087.