Physica A 388 (2009) 455–461

Contents lists available at ScienceDirect

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

A wave function for stock market returns

Ali Ataullah, Ian Davidson, Mark Tippett ∗
Business School, Loughborough University, Leicestershire, LE11 3TU, United Kingdom

article info a b s t r a c t

Article history: The instantaneous return on the Financial Times-Stock Exchange (FTSE) All Share Index
Received 14 April 2008 is viewed as a frictionless particle moving in a one-dimensional square well but where
Available online 26 October 2008 there is a non-trivial probability of the particle tunneling into the well’s retaining walls.
Our analysis demonstrates how the complementarity principle from quantum mechanics
Keywords: applies to stock market prices and of how the wave function presented by it leads to a
Complementarity principle
probability density which exhibits strong compatibility with returns earned on the FTSE
Econophysics
Probability density
All Share Index. In particular, our analysis shows that the probability density for stock
Quantum tunneling market returns is highly leptokurtic with slight (though not significant) negative skewness.
Schrödinger equation Moreover, the moments of the probability density determined under the complementarity
principle employed here are all convergent — in contrast to many of the probability density
functions on which the received theory of finance is based.
© 2008 Elsevier B.V. All rights reserved.

1. Introduction

It has often been argued, and there is corroborating empirical evidence which suggests, that stock market prices exhibit
wave like properties [1,2]. This raises the important question of whether the complementarity principle of quantum
mechanics, applied with varying degrees of success to a number of areas outside of physics including the connection between
consciousness and the brain, the problem of free will in human decision making and the physiological properties of biological
organisms, can also be applied to the evolution of stock market prices [3,4]. Our purpose here is to demonstrate that the
complementarity principle does in fact apply to stock market prices and of how the wave function presented by it leads
to a probability density function which is strongly compatible with the returns earned on the United Kingdom’s Financial
Times-Stock Exchange (FTSE) All Share Index. Our analysis views the instantaneous return on the FTSE All Share Index as
a frictionless particle moving in a one-dimensional square well but where there is a non-trivial probability of the particle
tunneling into the well’s retaining walls [5]. When the particle is constrained to move within the well’s retaining walls, only
‘‘regular’’ returns are earned by the Index. When, however, the particle tunnels into the retaining walls, ‘‘irregular’’ returns
arise on the Index. The next section develops the wave function for stock market returns under the quantum tunneling model
briefly articulated here. Section 3 uses returns computed from the FTSE All Share Index over the fourteen year period from
1994 until 2007 to demonstrate how the empirical evidence is strongly compatible with the probability density implied by
the given wave function. Section 4 contains our summary conclusions.

2. Stock market return as a one-dimensional motion in a square potential well

Consider the square well pictured in Fig. 1. The well is (b − a) in width whilst U is the energy required for the particle
to escape from the well. Note, however, that the particle has kinetic energy of E < U and that this is not sufficient for

∗ Corresponding author. Tel.: +44 0 1509 228829; fax: +44 0 1509 223960.
E-mail address: [email protected] (M. Tippett).

0378-4371/$ – see front matter © 2008 Elsevier B.V. All rights reserved.
doi:10.1016/j.physa.2008.10.035
456 A. Ataullah et al. / Physica A 388 (2009) 455–461

Fig. 1. Motion of a particle representing the return on the FTSE All Share Index in a square potential well. The point marked ‘‘a’’ representing the left-hand
barrier of the well defines the point at which negative ‘‘tunneling’’ occurs and negative irregular returns are earned by the Index. The point marked ‘‘b’’
representing the right-hand barrier of the well defines the point at which positive ‘‘tunneling’’ occurs and positive irregular returns are earned by the Index.
Within the well (Region II) regular returns are earned by the Index.

the particle to escape from the well. Now, in classical mechanics the particle bounces off the walls of the well and thus,
cannot enter regions I and III. In quantum mechanics, however, there is a non-trivial probability the particle will penetrate
into regions I and III. Moreover, suppose region III is comprised of more porous material than is the case with region I. This
means the particle is likely to tunnel further into region III than it will into region I.
One can apply this model to the evolution of stock market prices by supposing that returns earned from investing on the
stock market will ordinarily fall between the walls of the well; that is between the points a and b. There will, however, be
irregular occasions when the return tunnels into regions I and III. Interpreting the well’s parameters in this way allows one
to determine the wave function responsible for the propagation of stock market returns. In particular, in region I the steady
state version of the Schrödinger equation turns out to be:

d2 ψ
− θψ = 0 (1)
dx2
where x is the particle’s position (correspondingly, the stock market’s instantaneous return) and θ is a positive constant
which hinges on the material comprising the well in region I as well as the difference between the particle’s kinetic energy,
E, and the energy, U, required for the particle to escape from the well [5]. It follows immediately that:
√ √
θ(x−a) θ(x−a)
ψI = C e + De− (2)
is the general solution of the Schrödinger equation in region I and C and D are constants which have to be determined. Now,
if ψI is to remain finite over its entire domain, −∞ < x < a, then D will have to be zero. It follows from this that the wave
function for stock market returns in region I will have to be:
√
θ(x−a)
ψI = C e . (3)
Similar considerations show that in region III the steady state version of the Schrödinger equation will take the form:

d2 ψ
− γψ = 0 (4)
dx2
where x is the instantaneous return on the stock market and γ is a positive constant. The general solution to this equation
is:
√ √
γ (x−b) γ (x−b)
ψIII = F e + Ge− (5)
where F and G are constants and ψIII indicates that the solution applies only in region III of the well. Now, if ψIII is to remain
finite over its entire domain, ∞ > x > b, then F will have to be zero. It follows from this that in region III the wave function
for stock market returns will take the form:
√
γ (x−b)
ψIII = Ge− . (6)
Note how the wave functions ψI and ψIII decrease exponentially beyond the barriers defining the irregular stock market
returns. However, within the well – where only regular returns are earned – the steady state Schrödinger equation takes
the form:
d2 ψ
+ λψ = 0 (7)
dx2
 A. Ataullah et al. / Physica A 388 (2009) 455–461 457

where λ is a positive constant that hinges on the particle’s kinetic energy, E, alone [5]. The general solution to this equation
is given by:
√ √
ψII = A · sin( λ(x − a)) + B · cos( λ(x − a)) (8)
where A and B are constants. Now since ψI = C when x = a and ψIII = G when x = b it necessarily follows that both
trigonometric terms in ψII will make a non-trivial contribution to the wave function. In particular, the wave functions in
sections I and III describing irregular stock market returns must connect in a continuously differentiable way with the wave
function describing the evolution of regular returns in section II [5]. This in turn means that at the point x = a the following
‘‘value matching’’ criterion will have to be satisfied:
ψI = C = B = ψII . (9)
Likewise, at the point x = b the ‘‘value matching’’ criterion will be:
√ √
ψIII = G = A · sin( λ(b − a)) + B · cos( λ(b − a)) = ψII . (10)
Moreover, the wave function must be ‘‘smooth’’ at the junction points between the regular and irregular returns. This will
mean that at the point x = a the following ‘‘smooth pasting’’ condition will have to be satisfied:

dψII √ √ dψI
=A λ=C θ = (11)
dx dx
q
θ
or equivalently, A = C · λ
. Similarly, at the point x = b the smooth pasting condition will be:

dψII √ √ √ √ √ dψIII
= A λ · cos( λ(b − a)) − B λ · sin( λ(b − a)) = −G γ = . (12)
dx dx
Now, from Eqs. (9)–(11) it follows:

θ
r
√ √
C· · sin( λ(b − a)) + C · cos( λ(b − a)) = G.
λ
Likewise, from Eqs. (9), (11) and (12) we have:
s s
θ √ λ √
C· · cos( λ(b − a)) − C · · sin( λ(b − a)) = −G.
γ γ
One can add these latter two equations and thereby show that the wave function for stock market returns must satisfy the
following condition:
√ √
θ + θγ √ λ − θγ √
C· √ · cos( λ(b − a)) − C · √ · sin( λ(b − a)) = 0.
θγ γλ
Now here one can let C = 0 in which case A, B and G will also be zero and the wave function for stock market returns is
trivially zero. Alternatively, one can set:
√ √
θ + θγ √ λ − θγ √
√ cos( λ(b − a)) = √ · sin( λ(b − a))
θγ γλ
or equivalently:
√ !
λ θ + θγ
r
1
(b − a) = √ · tan −1
· √ . (13)
λ θ λ − θγ
Under this latter scenario the size of the well, (b − a), depends on the period, λ, of the wave function describing the evolution
of stock market returns within the well in addition to the rate of decay in the wave function, θ , γ , outside the well. Moreover,
the above results show that the complete wave function for stock market returns will be:
√
"r C e θ(x−a) # x≤a
θ √ √
ψ(x) = C · sin( λ(x − a)) + cos( λ(x − a)) a<x≤b
λ (14)
"s s #
λ √ θ √ √
C · sin( λ(b − a)) − · cos( λ(b − a)) e− γ (x−b) b<x
γ γ
458 A. Ataullah et al. / Physica A 388 (2009) 455–461

where the constant, C , has yet to be determined. Now, whilst the evolution of stock market returns is completely described
by the wave function, ψ(x), Copenhagen orthodoxy interprets the square of the wave function, ψ 2 (x) as the probability
density of the stock market’s instantaneous return, x. One can thus square the wave function (14) and thereby show that
the affected probability density will be:
√
" C 2 e2 θ (x−a) # x≤a
θ θ
r
√ 1 √ √
ψ 2 ( x) = C2 (1 − cos(2 λ(x − a))) + (1 + cos(2 λ(x − a))) + · sin(2 λ(x − a)) a<x≤b
2λ 2 λ
λ √ θ √

(15)
C2 (1 − cos(2 λ(b − a))) + (1 + cos(2 λ(b − a)))
2γ 2γ #
√
θλ √ √
− · sin(2 λ(b − a)) e−2 γ (x−b) b<x
γ

where:
" √
1 (λ + θ ) (λ + θ )(b − a) (λ − θ ) √ θ √
C =2
√ + √ + + √ · sin(2 λ(b − a)) + (1 − cos(2 λ(b − a)))
2 θ 4γ γ 2λ 4λ λ 2λ
√ #−1
(θ − λ) √ 2 λθ √
+ √ cos(2 λ(b − a)) − √ · sin(2 λ(b − a))
4γ γ 4γ γ

is the normalising constant that insures a unit probability mass. Although the particle (that is, the instantaneous stock
market return) may tunnel into the sides of the well the probability mass will nonetheless be preserved; that is, if the
particle penetrates into the sides of the well it will eventually be reflected and is not absorbed.

3. United Kingdom stock index data and analysis

The previous section demonstrates how stock markets can be viewed as having wave like properties. To do this, one
thinks of the instantaneous stock market return as a particle evolving in a square potential well. When the particle remains
inside the well the stock market earns only regular returns. When, however, the particle tunnels into the barriers comprising
the well the stock market earns irregular returns. Moreover, this means the wave (or probability amplitude) function for
stock market returns must satisfy two ‘‘value matching’’ and two ‘‘smooth pasting’’ criteria. However, linear dependence
means that only three of these equations can be used in conjunction with the fact that stock market returns must have a unit
probability mass, to determine the wave function’s four constants (A, B, C and G). There are, however, four other parameters
that must also be determined before the model can be implemented; namely, the ‘‘energy’’ constants, θ , λ, and γ , associated
with the Schrödinger equation in regions I, II and III respectively as well as the level, a, at which negative irregular returns
arise; that is, the return where ‘‘negative’’ quantum tunnelling effects first arise. We now use the continuously compounded
daily returns on the FTSE All Share Index over the period from 1 January, 1994 until 30 June, 2007 to illustrate the procedures
we used to estimate the parameters of the wave function (14) and its implied probability density (15).
We begin our analysis by noting that prior to this period UK index compilers ‘‘underestimated’’ the importance of
dividends in the returns calculation and it was not until mid-1993 that the FTSE All Share Index was first computed on
a ‘‘total returns’’ basis [6]. Hence, it is only after this point that one can have confidence in the returns implied by the
FTSE All Share Index and this explains why we have used this period as the basis of our empirical analysis. There were a
total of n = 2816 individual daily returns over the fourteen year period on which our analysis is based and these were
‘‘annualised’’ in order to re-state them on a ‘‘per unit time’’ basis. This procedure renders our return series compatible with
the probability density function (15) which, it will be recalled, describes returns that evolve on a per unit time (or annualised)
basis. As an example of how the per unit time returns were calculated, we use the fact that the FTSE All Share Index stood at
1204.24 at the close of business on 7 February, 1995. At the close of business on 8 February, 1995 the Index had increased to
1204.72. This means the continuously compounded return for the day ending with the close of business on 8 February, 1995
.72
is log( 1204
1204.24
) ≈ 0.000399. If we multiply this daily return by 365 we re-state the return on a per unit time or annualised
basis. Hence, in this instance the per unit time return amounts to 0.000399 × 365 = 0.145457 or 14.5457% (per annum).
The returns employed in our empirical analysis were all stated on a per unit time basis and were all calculated in this way.
The first thing to note about the returns computed from the FTSE All Share Index is that they are incompatible with
the normal distribution. One can confirm this by computing the standardised skewness and excess kurtosis statistics. The
standardised skewness statistic for our returns data is [7]:

n  3
n X xj − x̄
= −0.2905 (16)
(n − 1)(n − 2) j=1
s
 A. Ataullah et al. / Physica A 388 (2009) 455–461 459
q
(xj − x̄)2 = 3.3617
1
Pn
where n = 2816 is the total number of returns on which our empirical analysis is based, s = (n−1) j =1

j=1 xj = 0.104248 is the simple average of the returns

1
Pn
is the standard deviation of the returns series and x̄ = n
series. Whilst the standardised skewness statistic given here is not significantly different from zero, the standardised excess
kurtosis measure [7]:

n(n + 1) 3(n − 1)2

n  4
X xj − x̄
− = 3.0759 (17)
(n − 1)(n − 2)(n − 3) j=1
s (n − 2)(n − 3)
is different from zero at any reasonable level of significance. Recall that under the hypothesis the return series is drawn
from a normal distribution one would expect the standardised excess kurtosis measure to be zero on average. Moreover,
the standardised excess kurtosis statistic given above is itself asymptotically distributed as a unit normal variate under this
hypothesis [7]. This in turn means that the probability of obtaining a standardised excess kurtosis score in excess of 3.0759
by chance alone is much less than 1%. Given this, one can have reasonable confidence that the underlying return series is
not drawn from a normal distribution. Having disposed of the possibility that our returns data may have been drawn from a
normal distribution, our attention now focuses on whether the wave function (14) and its implied probability density (15)
determined in the previous section might be compatible with our returns data. We begin by demonstrating the procedures
used to estimate the parameters of the given wave function. Pn
As previously noted the average for the n = 2816 returns comprising our data set turns out to be x̄ = 1n j =1 x j =
0.104248 or 10.4248% per annum. This average was subtracted from all n = 2816 returns so that the wave function is based
on the centred returns associated with the FTSE All Share Index. We then ordered the centred returns from the smallest to
the largest return. It then follows that r1 = x1 − x̄ = −19.6446 is the lowest centred return, r2 = x2 − x̄ = −15.7966 is the
second lowest centred return and so on, right up to the largest centred return which is r2816 = x2816 − x̄ = 18.4905. The
ordered centred returns were then used in conjunction with the Cramér–von Mises goodness of fit statistic, T3 , where [8]:
" #2
n 1
1 X j−
T3 = + F (rj ) − 2
(18)
12n j =1
n

and F (rj ) is the accumulated area under the probability density below the jth ordered centred return in our sample, to
estimate the parameters of the wave function (14) implied by our returns data. In particular, seed values were randomly
allocated to the ‘‘energy’’ parameters, λ, θ and γ , as well as the parameter, a, signifying the return at which ‘‘negative’’
quantum tunnelling begins. The Newton–Raphson algorithm [9] was then used to determine the values of the parameters
that minimise the Cramér–von Mises statistic, T3 . This procedure returned a minimised Cramér–von Mises test statistic of
T3 = 0.0404 with estimated parameters of θ̂ = 0.0438, γ̂ = 0.0601, λ̂ = 0.1091 and â = −1.4648. These results in
turn imply that b̂ = 2.1775, Â = 0.2141, B̂ = 0.3379, Ĉ = 0.3379 and Ĝ = 0.3212. Hence, our estimates suggest that
regular centred returns on the FTSE All Share Index vary from a low of â = −1.4648 (or about −0.40 of 1% per day) up to
b̂ = 2.1775 (or about 0.6 of 1% per day). Irregular returns are earned outside of this interval. Fig. 2 plots the wave function
(14) for centred stock market returns based on these parameter values together with its associated probability density (15)
as well as the normal distribution with a mean of zero and the standard deviation implied by our returns data which, as
previously noted, is s = 3.3617. The normal distribution is the symmetric and less peaked of the two distributions. These
graphs show that there are significant differences between the probability distribution for stock market returns implied by
the wave function and the normal distribution as computed from our returns data. The wave function leads to a leptokurtic
probability distribution with slight (though not significant negative) skewness whilst the normal distribution is a smooth
symmetric mesokurtic probability density function.
Here, Anderson and Darling [10] have shown that if the ordered data series, rj is generated by the hypothesised probability
density then the Cramér–von Mises test is distribution free and the test statistic, T3 , has an expected value of E (T3 ) = 61 ≈
0.1667. Furthermore, Anderson and Darling [10] have also tabulated percentiles of the asymptotic distribution function
for the Cramér–von Mises test statistic, T3 , given that the data, rj , on which T3 is based are generated by the hypothesised
probability distribution. These tables show that the probability of the Cramér–von Mises test statistic, T3 , exceeding 0.3473
is 10%, the probability of T3 exceeding 0.4614 is 5% whilst the probability of T3 exceeding 0.7435 is 1%. Hence, the return
series on which our empirical analysis is based appears to be strongly consistent with the hypothesised wave function (14)
since the observed test statistic, T3 = 0.0404, is both well below its expected value and any of the generally accepted
levels of significance for T3 tabulated by Anderson and Darling [10]. One must exercise care with this conclusion, however,
since the Cramér–von Mises test statistic, T3 , is derived on the premise that the hypothesised probability density function
is completely specified [11]. Unfortunately, this is not the case in the present instance since there are four parameters –
λ, θ, γ and a – which have all been estimated from the returns data.
Given this we also used the fitted probability density obtained from minimising the Cramér–von Mises test statistic, T3 ,
to compute the Cramér goodness of fit test statistic, T0 , [12] for our returns data. This test makes allowances in its testing
procedure for the fact that the hypothesised probability density may not have been completely specified; that is, some of
the probability density’s parameters have had to be estimated from the underlying data. The Cramér goodness of fit test
was implemented using the probability density (15) with the parameter estimates summarised earlier (that is, θ̂ = 0.0438,
460 A. Ataullah et al. / Physica A 388 (2009) 455–461

Fig. 2. The upper graph is the wave function (14) for centred stock market returns with parameter values  = 0.2141, B̂ = 0.3379, Ĉ = 0.3379,
Ĝ = 0.3212, θ̂ = 0.0438, γ̂ = 0.0601, λ̂ = 0.1091, â = −1.4648 and b̂ = 2.1775. The lower graph is the probability density (15) implied by the wave
function (14) and the given parameter values superimposed on the normal distribution with a mean of zero and standard deviation of s = 3.3617 as
computed from our returns data. The normal distribution is the symmetric and less peaked of the two distributions.

γ̂ = 0.0601, λ̂ = 0.1091, â = −1.4648, etc.) to determine the 5th, 10th, 15th, — up to the 100th percentile returns. We then
used the actual number of returns falling between each of the n = 20 percentile returns to compute the Cramér goodness of
fit statistic, T0 . If the return series is described by the wave function leading to the probability density (15), then the Cramér
goodness of fit statistic, T0 , is asymptotically distributed as a Chi-square variate with n − k − 1 degrees of freedom, where
k is the number of parameters estimated from the data [12]. Since k = 4 parameters are estimated from our returns data,
this means the Cramér goodness of fit statistic has n − k − 1 = 20 − 4 − 1 = 15 degrees of freedom. Tabulated values
of the Chi-square distribution with 15 degrees of freedom show that the probability of the Cramér goodness of fit statistic,
T0 , exceeding 22.31 is 10%, the probability of T0 exceeding 25.00 is 5% whilst the probability of T0 exceeding 30.58 is 1%. In
our case the Cramér goodness of fit statistic turns out to be T0 = 14.58 which is both marginally lower than its expected
value of E (T0 ) = 15 and well within any of the generally accepted levels of significance for the Chi-square distribution given
earlier. This again confirms that the return series on which our empirical analysis is based is strongly consistent with the
hypothesised wave function (14) and the probability density (15) implied by it.
Fig. 3 plots the difference between the fitted distribution function and the empirical distribution function, or:
rj 1
j−
Z
ψ 2 (x)dx − 2
(19)
−∞ 2816

where ψ 2 (x) is the probability density function (15), rj is the jth ordered centred return and j = 1, 2, 3, . . . , 2816 represents
the return series on which our empirical analysis is based. This graph confirms that the probability density (15) provides
a more than adequate description of our centred returns data since the difference between the fitted distribution function
and the empirical distribution function is generally quite small and well within limits imposed by several other goodness of
fit tests which because of space considerations, we do not report here [11].

4. Summary conclusions

Our purpose here is to demonstrate how the instantaneous return on the FTSE All Share Index can be viewed as a
frictionless particle moving in a one-dimensional square well but where there is a non-trivial probability of the particle
tunneling into the well’s retaining walls. Our analysis shows how the complementarity principle from quantum mechanics
can then be applied to stock market prices and of how the wave function (14) presented by it leads to a probability density
 A. Ataullah et al. / Physica A 388 (2009) 455–461 461

Fig. 3. This graph gives the difference between the probability distribution function with parameter values of  = 0.2141, B̂ = 0.3379, Ĉ = 0.3379,
Ĝ = 0.3212, θ̂ = 0.0438, γ̂ = 0.0601, λ̂ = 0.1091, â = −1.4648 and b̂ = 2.1775 and the empirical probability distribution function for the n = 2816
Rr j− 21
j
centred returns on which our empirical analysis is based: −∞ ψ 2 (x)dx − 2816
. Here rj is the jth ordered centred return and j = 1, 2, 3, . . . , 2816.

function (15) which is strongly compatible with the historical returns earned on the FTSE All Share Index. In particular, our
analysis shows that the density function for equity returns is highly leptokurtic with slight (though not significant) negative
skewness. Moreover, all moments are convergent for the probability density (15), unlike most density functions used in the
literature to model the stock market returns process [13,14]. There are two important extensions which can be made to
our analysis. The first involves determining the wave function for a portfolio comprised of several assets (as compared to
the wave function for the single asset considered here). The wave function for a portfolio of assets will be a combination of
the wave functions of the individual assets comprising the portfolio. One can then take the square of the combined wave
function to determine the probability density of the return on the given portfolio of assets. A second important extension
involves comparing the valuation expressions for derivative securities such as options, futures, etc. implied by the density
function (15) with the valuation expressions for these securities obtained from the standard distributions appearing in the
literature [13,14] and on which much of the received theory of finance is based.

References

[1] A. Ataullah, M Tippett, Physica A 382 (2007) 557.

[2] K. Yuan, L. Zheng, Q. Zhu, J. Financ. Econ. 13 (2006) 1.
[3] M. Born, Physics in My Generation, Pergamon, London, 1956.
[4] M. Born, Physics and Politics, Oliver and Boyd, Edinburgh, 1962.
[5] A. Beiser, Concepts of Modern Physics, McGraw-Hill, Boston, 2003.
[6] E. Dimson, P. Marsh, J. Bus. 74 (2001) 1.
[7] M. Kendall, A. Stuart, The Advanced Theory of Statistics: Volume I, Charles Griffin, London.
[8] M. Kendall, A. Stuart, The Advanced Theory of Statistics: Volume II, Charles Griffin, London.
[9] B. Carnahan, H. Luther, J. Wilkes, Applied Numerical Methods, Wiley, New York, 1969.
[10] T. Anderson, D. Darling, Ann. Math. Stat. 23 (1952) 193.
[11] W. Conover, Practical Nonparametric Statistics, Wiley, New York, 1999.
[12] H. Cramér, Mathematical Methods of Statistics, Princeton University Press, New Jersey, 1946.
[13] G. Premaratne, A. Bera, J. Financ. Econ. 3 (2005) 169.
[14] D. Ashton, M. Tippett, J. Bus Financ. Account. 33 (2006) 1586.