Working Paper No.

84

An economic model of work-related stress

by

Alfred Greiner

University of Bielefeld
Department of Economics
Center for Empirical Macroeconomics
P.O. Box 100 131
33501 Bielefeld, Germany
 An economic model of work-related stress

Alfred Greiner∗

Abstract

In this paper we present an economic model of optimal consumption and labour

supply where we assume that working may generate stress which affects the well-
being of the representative individual. As to stress we posit that it is influenced by
cumulated past labour and capital. The latter reflects the fact that work-related
stress evolves gradually over time and that it is more likely to occur in modern
societies. Using optimal control theory we demonstrate that sustained cycles may
result. Further, we numerically compute the global optimal value function and give
a representation of the limit cycle.

Keywords: Work-related stress, intertemporal optimization, limit cycles

JEL: C61, D91, J22

∗
Department of Business Administration and Economics, Bielefeld University, P.O. Box 100131, 33501
Bielefeld, Germany.
1 Introduction
According to the European Agency for Safety and Health at Work, work-related stress
affects 28% of workers in the European Union (EU) and is the second most common
health problem related to work, after back pain ([3], p. 10). On the individual level,
the consequences of stress are that the person’s general quality of life as well as his well-
being are detracted. For some people who experience work-related stress the consequences
may be more drastic implying that stress negatively affects their health. Typical health
problems caused by work-related stress are for example insomnia, constant tiredness, high
blood pressure, nervous twitches, just to mention a few.
In addition to these individual problems work-related stress also causes costs for soci-
ety. So, the European Commission estimates that costs due to work-related stress in the
EU amount to at least 20 billion Euro annually ([7]). For France, [1] have estimated that
work-related stress cost between 830 and 1656 million Euro in 2000, which represented
13% to 26% of total spending of social security occupational illnesses and work injuries
branch. Besides direct health costs work-related stress leads to costs due to absenteeism
and raising individuals’ quitting behaviour causing costs for firms. [14] find that individ-
uals experiencing stress are 25% more likely to hold intentions to quit or being absent
from work than those without work-related stress.
As concerns stress, one has to point out that stress can also be favourable to a person’s
well-being and a certain amount of stress is even needed in order to remain healthy and
alert. Therefore, in the psychological literature one finds the distinction between two forms
of stress, between the so-called eustress and the distress. Eustress is the positive form of
stress which is beneficial to an individual. This kind of stress or pressure is stimulating
and enhances performance. However, when stress or pressure becomes too large such that
the individual perceives himself unable to cope successfully with a situation, he is subject
to distress, the negative form of work-related stress (as to the distinction between eustress
and distress see [16] or [5]). The latter form of stress, distress, is perceived as negative by

1
a person and may lead to that sort of individual health problems mentioned above.
In general, work situations are experienced as stressful when the demands made on
the person do not match the resources available (in the individual or provided by the
organization) or do not meet the person’s needs and motivation. This can also serve as
a definition for (di-)stress. As concerns the causes of work-related stress, [15] summa-
rizes the factors under four headings, which are then differentiated further: quantitative
overload, qualitative underload, lack of control over work and lack of social support.
In this paper, we will focus on the first and the third factor. So, one of our assumptions
will be that stress arises as individuals simply have too much work, a fact which seems
to be of relevance particularly for Japan (see [14]). But we refrain from modelling the
second factor, qualitative under- or overload as a possible source of stress. Qualitative
underload means that the individual’s work is not demanding so that he may be bored
by his work, overload simply means that the work is too difficult. The other factor
generating stress is the lack of control over work which is perceived as a threat to individual
freedom, autonomy and identity. We take account of this factor because there is strong
evidence that machine- and systems-paced work, especially of high rate, is detrimental to
psychological and physical health (see e.g. [4] or [2]). It should also be pointed out that
workload has to be considered in relation to work pace such that it is in particular the
interrelation between these two factors which generates stress.
The rest of the paper is organized as follows. In the next section we present our model
and our modelling of work-related stress. Section 3 studies the dynamics of the model
and section 4, finally, concludes.

2 The structure of the model

Our model consists of a representative household with a utility function which positively
depends on consumption at time t, C(t), and negatively on labour, L(t). The latter models
the preference for leisure, as usual in economics. In addition, we make the assumption

2
that utility depends on work-related stress, S(t). As to the effect of stress on utility we
posit that utility rises with stress when stress is below a certain threshold, S ⋆ , and declines
when stress exceeds this threshold. Thus, we take into consideration that for small values
of stress a rise in stress may well have a positive effect on a person’s well-being which
raises his work performance (see e.g. [5], pp. 6-8). In this case we speak of eustress. If
stress becomes too large, i.e. if it exceeds the threshold S ⋆ , well-being declines with stress.
In this case, we speak of distress as already mentioned in the Introduction.
As to the utility function U (·) we assume1 that it is separable in C, L and S and a
root function of C, linear in L and quadratic in S. Thus, the utility function is given by
√
U (C, L, S) = C + (L̄ − L) − a(S − S ⋆ )2 , (1)

with S ⋆ > 0, L̄ the maximum available labour supply and a > 0 a constant.
As to the constraints the first is the usual budget constraint stating that production
is spent for consumption and saving, i.e.

K̇ = K α Lβ − C − δ K. (2)

K is capital and equals cumulated past investment and δ > 0 is the depreciation rate.
K α Lβ is the production function with α ∈ (0, 1) the capital share and β ∈ (0, 1) the
labour share.
The second constraint describes the stress variable S. Stress is a function of cumulated
past labour. Thus, we take into account that work-related stress is caused by labour and,
second, that it is the cumulative effect of labour which leads to stress. The latter seems
to be important because working overtime one or two times a month does not necessarily
lead to stress. However, if this occurs more often stress is likely to occur. In addition, we
assume that the effect of labour on stress is the stronger the higher the capital stock is.
We do this because, as mentioned in the Introduction, machine- and system-paced work
is detrimental to health since it forces people to work in accordance with the machines
1
In the following we delete the time argument t.

3
depriving them of their control over work, which implies that their personal needs may
be left behind. Further, work-related stress is a phenomenon which occurs primarily in
developed countries which dispose of higher capital stocks compared to countries say 100
hundred years ago. With these assumptions the change in stress can be described by the
following differential equation,
Ṡ = L K − η S, (3)

with η > 0 reflecting the ability to recover.

The intertemporal optimization problem, then, is to choose consumption and labour
such that the discounted stream of utility over an infinite time horizon is maximized
subject to the two constraints (2) and (3). Denoting by ρ > 0 the subjective discount
rate, the formal problem is to
Z ∞ √ 
max e−ρt C + (L̄ − L) − a(S − S ⋆ )2 dt, (4)
C,L 0

subject to (2) and (3).

Optimality conditions are derived form the current-value Hamiltonian which is given
by
√
H = C + (L̄ − L) − a(S − S ⋆ )2 + λ1 (K α Lβ − C − δK) + λ2 (L K − η S) , (5)

with λ1 and λ2 denoting costate variables or shadow prices of K and S.

The necessary optimality conditions are given by

∂H
= 0.5 C −0.5 − λ1 = 0 (6)
∂C
∂H
= −1 + λ1 βLβ−1 K α + λ2 K = 0 (7)
∂L
λ̇1 = (ρ + δ)λ1 − λ1 αK α−1 Lβ − λ2 L (8)

λ̇2 = (ρ + η)λ2 + 2 a (S − S ⋆ ). (9)

In addition we require that the transversality condition limt→∞ e−ρt (λ1 K + λ2 S) = 0 must
be fulfilled.

4
 From (6) and (7) we get optimal consumption and labour supply as functions of the
costate variables and of state variables as
1/(1−β)
βλ1 K α

C= 0.25 λ−2
1 , L= . (10)
1 − λ2 K

Inserting (10) in K̇, Ṡ, λ̇1 and λ̇2 gives an autonomous system of differential equations in
the state variables K and S and in the costate variables λ1 and λ2 . This system is given
by

K̇ = K α L(K, λ1 , λ2 , ·)β − C(λ1 ) − δK (11)

Ṡ = L(K, λ1 , λ2 , ·) K − η S (12)

λ̇1 = (ρ + δ)λ1 − λ1 αK α−1 L(K, λ1 , λ2 , ·)β − λ2 L(K, λ1 , λ2 , ·) (13)

λ̇2 = (ρ + η)λ2 + 2 a (S − S ⋆ ). (14)

3 The dynamics of the model

3.1 The analytical model

Equations (11)-(14) completely describe the dynamic behaviour of our model. We are
interested in the dynamics around a rest point or stationary point of this system, in
particular in the question of whether the model converges to the rest point or whether it
may generate cycles for example. To do so we first assume that a unique rest point exists
for the analytical model and compute the Jacobian matrix evaluated at the rest point.
The Jacobian is given by
 
α−1 β α β−1 α β−1 α β−1
αK L + K βL LK − δ 0 βK L Lλ1 − Cλ1 K βL Lλ2
 
L + K LK K Lλ1 K Lλ2
 
 −η 
J = ,
0 a33

 −a31 −a34 
 
0 2a 0 ρ+η

5
with

a31 = λ2 LK + λ1 (α − 1)αK α−2 Lβ + αβK α−1 Lβ−1 LK

a33 = ρ + δ − αK α−1 Lβ − λ2 Lλ1 − αβK α−1 λ1 Lβ−1 Lλ1

a34 = λ2 Lλ2 + αβK α−1 λ1 Lβ−1 Lλ2 + L . (15)

Lk , Ck denote the derivative of L and C with respect to variable k, k = K, λ1 , λ2 . From

(10) it can easily be seen that the derivatives have the following signs:

Cλ1 < 0, Lλ1 > 0, Lλ2 > 0, LK > = < 0 for λ2 > = < −αK −2α−1 (1 − λ2 K). (16)

The eigenvalues of that matrix are given by

v s 
u
2
r u  r 2 W W
µ1,2,3,4 = ± − ± − det J,
t
2 2 2 2

with W defined as
     
 a11 a13   a22 a24  0 a14
     
 
W =  +  + 2 
 a31 a33   a42 a44  0 a34
    
 

where aij is the element of the i−th row and j−th column (see [6]).
Looking at the formula for the eigenvalues, one immediately realizes that the eigen-
values are symmetric around ρ/2. Since ρ > 0 holds this implies that the system is never
completely stable (in the sense that all eigenvalues have negative real parts) but it can
only be saddle point stable. From an economic point of view, convergence to the station-
ary state means that all variables are constant in the long run. That is there are constant
levels of consumption and labour supply and, as a consequence, a constant capital stock
and a constant level of stress. The transitional behaviour of the variables in case of saddle
point stability is characterized by unimodal time paths if the eigenvalues are real. If the
eigenvalues are complex conjugate, however, the variables are characterized by cyclical os-
cillations until the stationary point is reached. This means that both the capital stock as

6
well as the level of stress shows oscillations over time, however, with declining amplitudes
until the stationary point is reached asymptotically.
Besides convergence to the stationary state in the long run, the system may show
persistent endogenous cycles. This behaviour can be observed if the dynamic system (11)-
(14) undergoes a Hopf bifurcation. A Hopf bifurcation states the following (for a complete
statement of the Hopf bifurcation theorem see e.g. [12]): Assume that we continuously
vary a parameter, say the discount rate, and that for a certain critical value of that
(bifurcation) parameter two eigenvalues become purely imaginary. Assume in addition
that the crossing speed of the eigenvalues is non-zero as the bifurcation parameter is
varied. Then, there exist stable or unstable limit cycles which occur for values of the
bifurcation parameter which are larger or smaller than the critical parameter value for
which two eigenvalues are purely imaginary.
Let us find out whether persistent cycles may occur in our model. From the formula
of the eigenvalues (see, e.g. [6]) we know that W > 0 is a necessary condition for two
purely imaginary eigenvalues and, thus, for the emergence of a Hopf bifurcation which
leads to stable limit cycles. Looking at the constant W we see that only the expression
a11 a33 − a31 a13 may become positive. Using the fact that a11 + a33 = ρ holds (cf. [8], p.
134) we may write a11 a33 − a31 a13 as

a11 a33 − a31 a13 =

αK α−1 Lβ − λ2 Lλ1 + αβK α−1 λ1 Lβ−1 Lλ1 − δ ·

ρ + δ − αK α−1 Lβ − λ2 Lλ1 − αβK α−1 λ1 Lβ−1 Lλ1 +

λ2 LK + λ1 (α − 1)αK α−2 Lβ + αβK α−1 Lβ−1 LK βK α Lβ−1 Lλ1 − Cλ1 . (17)

For δ ≥ αK α−1 Lβ − λ2 Lλ1 + αβK α−1 λ1 Lβ−1 Lλ1 the first term in equation (17), a11 a33 ,
is negative while it is difficult to make a clear statement for the second term, −a31 a13 .
However, it is seen that a positive λ2 makes a positive sign of the second term more likely.2
2
Recall that the signs of Li and Ci are given in (16) and that λ2 > 0 implies LK > 0, since 1−λ2 K > 0
must hold for L to be real.

7
Further, a positive λ2 together with a high subjective discount rate make it more likely
that the first term is positive, too. So, a Hopf bifurcation leading to limit cycles is more
likely for a positive value of λ2 together with a high discount rate. It should be noted
that λ2 at the stationary point is positive (negative) if the level of stress is lower (higher)
than S ⋆ .
From an economic point of view, the conditions leading to persistent cycles can be
interpreted such that these oscillations may occur when the individual’s stress level is
smaller than S ⋆ , that is when the individual experiences eustress in his work, and when he
is impatient, the latter being reflected by a high subjective discount rate. These conditions
state that a rise in the level of stress raises the individual’s well-being suggesting that he
identifies himself with his work and may even be enthusiastic in his work. Since the
individual is impatient, he works a lot at nearby time periods, thus raising the level of
stress and his well-being as he is in the eustress range. However, with a rising level of
stress the marginal value of additional stress, its shadow price, declines. A declining
shadow price of stress leads the individual to reduce his work supply generating a decline
in the stress level. This goes on until the shadow price of stress rises again, due to the
decline in the level of stress, thus, leading to persistent cycles. It should be mentioned
that cyclical labour supply implies oscillations in the income and also in consumption.
In order to gain additional insight into our model and to prove the existence of per-
sistent cycles we next present a numerical example.

3.2 A numerical example

To study our model numerically, we assume a constant returns to scale production function
with a capital share of 30% and a labour share of 70%, i.e. α = 0.3 and β = 0.7. δ and n
are set to δ = 0.075 and n = 0.05. S ⋆ is set to S ⋆ = 3, a = 0.1 and the subjective discount
rate ρ serves as bifurcation parameter.
Before we study the time paths of the variables of our model we address the question

8
of existence and uniqueness of a stationary state. To do so, we first set ρ = 0.05. With this
parameter value we solve (12) = 0 with respect to S and insert the resulting3 S̄(K, λ2 , λ1 , ·)
in (11), (13) and (14) and, then, solve (14) = 0 with respect to λ1 yielding λ̄1 (K, λ2 , ·).
Inserting λ̄1 (K, λ2 , ·) in (11) and (13) and solving (11) = 0 and (13) = 0 with respect to
K and λ2 gives the rest point of the dynamic system. Figure (1) shows the K̇ = 0 and
λ̇1 = 0 curves in the (K − λ2 ) plane demonstrating that there exists a unique rest point
for our model. Varying the discount rate with ρ = 0 as lower bound and ρ = 0.35 as
upper bound does not change the qualitative outcome, i.e. there always exists a unique
rest point. Further, it should be noted that for about ρ = 0.054 we get λ̄2 = 0 and
S̄ = S ⋆ = 3 whereas for ρ < (>) 0.054, λ̄2 is negative (positive) and S̄ is larger (smaller)
than S ⋆ = 3.

-1

-2
0 1 2 3 4 5
Figure 1: K̇ = 0 (monotonously falling) and λ̇1 = 0 (first rising, then declining) curves in
the (K − λ2 ) plane.

Next, we analyze the local dynamics at the rest point for different values of the discount
3
The¯denotes values at the rest point.

9
rate by computing the eigenvalues of the Jacobian matrix. It turns out that for ρ ∈
(0, 0.3083) the eigenvalues are complex conjugate with two having negative real parts
and two having positive real parts. This implies that the model is characterized by
saddle point stability with a two dimensional stable manifold. For ρcrit = 0.3083432 the
differential equation system undergoes a supercritical Hopf bifurcation giving rise to stable
limit cycles.4 The limit cycles occur for an interval with strictly positive measure of the
discount rate where the discount rates are slightly larger than the critical value ρcrit . If
ρ is increased further all real parts of the eigenvalues become positive implying that the
system becomes unstable.
Our analysis so far has used necessary optimality conditions and characterized the
local dynamics around the stationary state. To get an idea about the global dynamics
of the optimally controlled system we numerically compute the optimal value function
by solving the Hamilton-Jacobi-Bellman equation. This method gives the full global
information about the optimal value function which, for its part, yields the optimal control
in feedback form (a detailed description of the algorithm we use is given in [10], [11]).
In particular, we are interested in the question of whether persistent cycles may turn
out to be the optimal solution. Therefore, we set the subjective discount rate ρ to ρ =
0.3084 which is slightly larger than the critical value ρcrit where a Hopf bifurcation was
detected. The calculation of the optimal controls confirm that persistent cycles turn out
to be the optimal solution. Figure (2) shows the convergence of the optimal path to the
limit cycle in the (K − S) plane.
4
The bifurcation analysis was done with LOCBIF (see [13]).

10
 2.1
2.0
S

1.9
1.8

0.35 0.40 0.45

Figure 2: Limit cycle in the (K − S) plane.

4 Conclusion
This paper has presented a simple model of work-related stress. Assuming that the change
in stress depends on labour and capital reflecting the fact that stress is built up gradually
and that it is more likely to occur in modern societies where machine- and systems-paced
work occurs, we could show that persistent cycles may turn out to be the optimal solution.
While there exist quite a lot of empirical papers dealing with work-related stress there
are no formal economic models investigating this phenomenon. The reason may be that
it is difficult to model the variable stress and the forces affecting it. So, our model is just
one possibility to formalize stress and other formulations may be relevant and feasible as
well. In particular, since lack of social support is also a factor which can lead to stress (cf.
[15] and [9]), possible interactions between working individuals, e.g. between subordinate
and superior, should be taken into account in future studies.

11
References
[1] Bejean, Sophie, Helen Sultan-Taib, Nathalie Wolff, 2003, “Modelling the social cost
of work-related stress.” LEG-Document de travail, No. 2003-11, Universite de Bour-
gogne.

[2] Bradley, Gunilla, 1989, Computers and the Psychological Work Environment. Taylor
and Francis, London.

[3] Cox, Tom, Amanda Griffiths, Eusebio Rial-Gonzalez, 2000, Research on Work-related
Stress. European Agency for Safety and Health at Work, Luxembourg.

[4] Cox, Tom, 1985, “The nature and measurement of stress.” Ergonomics, Vol. 28:
1155-63.

[5] Cooper, Cary, Sue Cartwright, 1996, Mental Health and Stress in the Workplace. A
Guide for Employers. HMSO, London.

[6] Dockner, Engelbert J., Gustav Feichtinger, 1991, “On the optimality of limit cycles
in dynamic economic systems.” Journal of Economics, Vol. 53, no. 1: 31-50.

[7] European Commission, 2002, Guidance on Work-related Stress - Spice of Life or Kiss
of Death. Luxembourg.

[8] Feichtinger, Gustav, Richard F. Hartl, 1986, Optimale Kontrolle ökonomischer

Prozesse: Anwendungen des Maximumprinzips in den Wirtschaftswissenschaften. De
Gruyter, Berlin.

[9] Giebels, Ellen, Janssen, Onne, 2004, “Conflict stress and reduced wellbeing at work:
The buffering effect of third-party help.” IACM 17th Annual Conference Paper.
http://ssrn.com/abstract=602063

12
[10] Grüne, Lars, 1997, “An adaptive grid scheme for the discrete Hamilton-Jacobi-
Bellman equation.” Numerische Mathematik, Vol. 75, no. 3: 319-337.

[11] Grüne, Lars, 2004, “Error estimation and adaptive discretization for the discrete
stochastic Ham,ilton-Jacobi-Bellman equation.” Numerische Mathematik, Vol. 99,
no. 1: 85-112.

[12] Hassard, Brian D., Nicholas D. Kazarinoff, Y.-H. Wan, 1981, Theory and Applica-
tions of Hopf Bifurcation. Cambridge University Press, Cambridge.

[13] Khibnik, Alexander I., Yuri A. Kuznetsov, Victor V. Levitin, Eugene V. Nikolaev,
1993, “Continuation techniques and interactive software for bifurcation analysis of
ODE’s and iterated maps.” Physica D (Nonlinear Phenomena), Vol. 62, no. 4: 360-
371.

[14] Leotaridi, Rannia, M., Melanie E. Ward, 2002, “Dying to work? An investigation
into work-related stress, quitting intentions and absenteeism.” IZA Discussion Paper,
No. 493, Bonn.

[15] Levi, Lennart, 1984, Stress in Industry: Causes, Consequences and Prevention. Oc-
cupational Safety and Health Series, no. 51, International Labour Office, Genevea.

[16] Selye, Hans, 1974, Stress, Bewältigung und Lebensgewinn. Piper & Co., München.

13