MFE_C09a.

qxd 16/12/2005 10:49 Page 553

section 9.1
Difference equations
Objectives

At the end of this section you should be able to:

Find the complementary function of a difference equation.

Find the particular solution of a difference equation.

Analyse the stability of economic systems.

Solve lagged national income determination models.

Solve single-commodity-market models with lagged supply.

A difference equation (sometimes called a recurrence relation) is an equation that relates con-
secutive terms of a sequence of numbers. For example, the equation
Yt = 2Yt−1
describes sequences in which one number is twice its predecessor. There are obviously many
sequences that satisfy this requirement, including
2, 4, 8, 16, . . .
5, 10, 20, 40, . . .
−1, −2, −4, −8, . . .
In order to determine the sequence uniquely, we need to be given some additional information,
such as the first term. It is conventional to write the first term as Y0, and once this is given a
specific value all remaining terms are known.
MFE_C09a.qxd 16/12/2005 10:49 Page 554

554 Dynamics

Example

Write down the first four terms of the sequence defined by

Y0 = 3 and Yt = 2Yt−1
and obtain a formula for the general term Yt in terms of t.

Solution
If Y0 = 3 then
Y1 = 2Y0 = 2 × 3 = 6
Y2 = 2Y1 = 2 × 6 = 12
Y3 = 2Y2 = 2 × 12 = 24
In order to produce a formula for the general term, we write these as
Y1 = 2Y0 = 21 × 3
Y2 = 2Y1 = 22 × 3
Y3 = 2Y2 = 23 × 3
It is now obvious from this pattern that the general term, given by
Yt = 3(2 t )
is the solution of the difference equation
Yt = 2Yt−1
with initial condition
Y0 = 3

The following problem gives you an opportunity to solve difference equations for yourself.

Practice Problem

1 Starting with the given initial conditions, write down the first four terms of each of the following
sequences. By expressing these as an appropriate power, write down a formula for the general term,
Yt in terms of t.
(1) (a) Yt = 3Yt−1; Y0 = 1 (b) Yt = 3Yt−1; Y0 = 7 (c) Yt = 3Yt−1; Y0 = A
1 1 1
(2) (a) Yt = Yt−1; Y0 = 1 (b) Yt = Yt−1; Y0 = 7 (c) Yt = Yt−1; Y0 = A
2 2 2
(3) Yt = bYt−1; Y0 = A

The result of the last part of Practice Problem 1 shows that the solution of the general equation
Yt = bYt−1 (1)
with initial condition
MFE_C09a.qxd 16/12/2005 10:49 Page 555

9.1 • Difference equations 555

Y0 = A
is given by
Yt = A(bt)
Before we can consider the use of difference equations in economic models, we must exam-
ine the solution of more general equations of the form
Yt = bYt−1 + c (2)
where the right-hand side now includes a non-zero constant, c. We begin by defining some
terminology. The general solution of equation (2) can be written as the sum of two separate
expressions known as the complementary function (CF) and the particular solution (PS). The
complementary function is the name that we give to the solution of equation (2) when the con-
stant, c, is zero. In this case, equation (2) reduces to equation (1), and so
CF = A(bt)
The particular solution is the name that we give to any solution of equation (2) that we are
clever enough to ‘spot’. This turns out to be rather easier to do than might at first appear and
we will see how this can be done in a moment. Finally, once CF and PS have been found, we
can write down the general solution of equation (2) as
Yt = CF + PS = A(bt ) + PS
A proof of this result can be found in Practice Problem 8 at the end of this section. The letter
A is no longer equal to the first term, Y0, although it can easily be calculated, as the following
example demonstrates.

Example

Solve the following difference equations with the specified initial conditions. Comment on the qualitative
behaviour of the solution in each case.
(a) Yt = 4Yt−1 + 21; Y0 = 1
1
(b) Yt = Yt−1 + 8; Y0 = 2
3

Solution
(a) The difference equation
Yt = 4Yt−1 + 21
is of the standard form
Yt = bYt−1 + c
and so can be solved using the complementary function and particular solution. The complementary
function is the general solution of the equation when the constant term on the right-hand side is
replaced by zero: that is, it is the solution of
Yt = 4Yt−1
which is A(4t).
The particular solution is any solution of the original equation 
MFE_C09a.qxd 16/12/2005 10:49 Page 556

556 Dynamics

Yt = 4Yt−1 + 21
that we are able to find. In effect, we need to think of a sequence of numbers, Yt, such that when this is
substituted into
Yt − 4Yt−1
we obtain the constant value of 21. One obvious sequence likely to work is a constant sequence,
Yt = D
for some number, D. If this is substituted into
Yt = 4Yt−1 + 21
we obtain
D = 4D + 21
(Note that Yt = D whatever the value of t so Yt−1 is also equal to D.) This algebraic equation can be re-
arranged to get
−3D = 21
and so D = −7.
We have therefore shown that the complementary function is given by
CF = A(4t )
and that the particular solution is
PS = −7
Hence
Yt = CF + PS = A(4t ) − 7
which is the general solution of the difference equation
Yt = 4Yt−1 + 21
To find the specific solution that satisfies the initial condition
Y0 = 1
we simply put t = 0 in the general solution to get
Y0 = A(40 ) − 7 = 1
that is,
A−7=1
which gives
A=8
The solution is
Yt = 8(4t) − 7
A graphical interpretation of this solution is shown in Figure 9.1 where Yt is plotted against t. It is
tempting to join the points up with a smooth curve. However, this does not make sense because t is
allowed to take only whole-number values. Consequently, we join up the points with horizontal lines
to create the ‘staircase’ which more properly reflects the fact that t is discrete. Figure 9.1 shows that the
values of Yt increase without bound as t increases. This is also apparent from the formula for Yt because
MFE_C09a.qxd 16/12/2005 10:49 Page 557

9.1 • Difference equations 557

Figure 9.1

the numbers 4t get ever larger as t increases. We describe this by saying that the time path diverges uni-
formly or explodes. This sort of behaviour can be expected to occur for any solution
Yt = A(bt ) + PS
where b > 1.
(b) The difference equation
1
Yt = Yt−1 + 8
3
can be solved in a similar way to that of part (a). The complementary function is given by
1 t
CF = A A DF
C3
and for a particular solution we try
Yt = D
for some constant D. Substituting this into the difference equation gives
1
D= D+8
3
which has solution D = 12, so
PS = 12
The general solution is therefore
t
Yt = CF + PS = A AC DF + 12
1
3 
MFE_C09a.qxd 16/12/2005 10:49 Page 558

558 Dynamics

Figure 9.2

Finally, the specific value of A can be found from the initial condition
Y0 = 2
Setting t = 0 in the general solution gives
0
2 = A AC DF + 12 = A + 12
1
3
and so A is −10. The solution is
t
Yt = −10 AC DF + 12
1
3
This solution is sketched in Figure 9.2, which shows that the values of Yt increase but eventually settle
down at 12. We describe this by saying that the time path converges uniformly to the value of 12, which
is referred to as the equilibrium value. This behaviour is also apparent from the formula for Yt because
the number (1/3)t gets ever smaller as t increases. In effect, the complementary function disappears,
leaving just the particular solution. The particular solution is the equilibrium value of Yt whereas the
complementary function measures the deviation from the equilibrium which, in this case, converges to
zero as t increases. This sort of behaviour can be expected to occur for any solution
Yt = A(bt) + PS
when 0 < b < 1.
MFE_C09a.qxd 16/12/2005 10:49 Page 559

9.1 • Difference equations 559

Practice Problem

2 Solve the following difference equations with the specified initial conditions:
(a) Yt = − 1/2Yt−1 + 6; Y0 = 0 (b) Yt = −2Yt + 9; Y0 = 4
In each case, sketch the corresponding ‘staircase’ diagram and comment on the qualitative behaviour
of the solution as t increases.

The results of the previous example and Practice Problem 2 can be summarized:

If b > 1 then Yt displays uniform divergence.

If 0 < b < 1 then Yt displays uniform convergence.
If −1 < b < 0 then Yt displays oscillatory convergence.
If b < −1 then Yt displays oscillatory divergence.

The remaining possibilities, b = 1, b = −1 and b = 0, are considered in Practice Problem 6 at the

end of this section, which shows that Yt converges when b is 0 but diverges when b is 1 or −1.
We conclude that the solution of the difference equation eventually settles down to an equilib-
rium state only when b lies in the range −1 < b < 1.
If convergence does occur in an economic model, the model is said to be stable. If the vari-
ables diverge, it is said to be unstable.
We now investigate two applications of difference equations taken from macroeconomics
and microeconomics, respectively:

national income determination

supply and demand analysis.
We consider each of these in turn.

9.1.1 National income determination

In Section 1.6 we introduced a simple two-sector model with structural equations
Y =C+I
C = aY + b
I = I*
where b and I* denote autonomous consumption and investment, and a is the marginal
propensity to consume, which lies in the range 0 < a < 1. In writing down the equations in this
form, we are implicitly assuming that only one time period is involved, that consumption
depends on national income within this time period and that equilibrium values are attained
instantaneously. In practice, there is a time lag between consumption and national income.
Consumption, Ct, in period t depends on national income, Yt−1, in the previous period, t − 1.
The corresponding consumption function is given by
Ct = aYt−1 + b
MFE_C09a.qxd 16/12/2005 10:49 Page 560

560 Dynamics

If we assume that investment is the same in all time periods then

It = I*
for some constant, I*. Finally, if the flow of money is in balance in each time period, we also
have
Yt = Ct + It
Substituting the expressions for Ct and It into this gives
Yt = aYt−1 + b + I*
which we recognize as a difference equation of the standard form given in this section. This
equation can therefore be solved and the time path analysed.

Example

Consider a two-sector model:

Yt = Ct + It
Ct = 0.8Yt−1 + 100
It = 200
Find an expression for Yt when Y0 = 1700. Is this system stable or unstable?

Solution
Substituting the expressions for Ct and It into
Yt = Ct + It
gives
Yt = (0.8Yt−1 + 100) + 200
= 0.8Yt−1 + 300
The complementary function is given by
CF = A(0.8)t
and for a particular solution we try
Yt = D
for some constant D. Substituting this into the difference equation gives
D = 0.8D + 300
which has solution D = 1500. The general solution is therefore
Yt = A(0.8)t + 1500
The initial condition,
Y0 = 1700
gives
1700 = A(0.8)0 + 1500 = A + 1500
MFE_C09a.qxd 16/12/2005 10:49 Page 561

9.1 • Difference equations 561

and so A is 200. The solution is

Yt = 200(0.8)t + 1500
As t increases, (0.8)t converges to zero and so Yt eventually settles down at the equilibrium level of 1500. The
system is therefore stable. Note also that because 0.8 lies between 0 and 1, the time path displays uniform
convergence.

Practice Problem

3 Consider the two-sector model:

Yt = Ct + It
Ct = 0.9Yt−1 + 250
It = 350
Find an expression for Yt when Y0 = 6500. Is this system stable or unstable?

In the previous example, and again in Practice Problem 3, we noted that the model is stable
and that it displays uniform convergence. If we return to the general equation
Yt = aYt−1 + b + I*
it is easy to see that this is always the case for the simple two-sector model because the coeffici-
ent of Yt−1 is the marginal propensity to consume, which is known to lie between 0 and 1.

9.1.2 Supply and demand analysis

In Section 1.3 we introduced a simple model of supply and demand for a single good in an iso-
lated market. If we assume that the supply and demand functions are both linear then we have
the relations
QS = aP − b
QD = −cP + d
for some positive constants, a, b, c and d. (Previously, we have written P in terms of Q and have
sketched the supply and demand curves with Q on the horizontal axis and P on the vertical axis.
It turns out that it is more convenient in the present context to work the other way round and
to write Q as a function of P.) In writing down these equations, we are implicitly assuming that
only one time period is involved, that supply and demand are dependent only on the price in
this time period, and that equilibrium values are attained instantaneously. However, for cer-
tain goods, there is a time lag between supply and price. For example, a farmer needs to decide
precisely how much of any crop to sow well in advance of the time of sale. This decision is made
on the basis of the price at the time of planting and not on the price prevailing at harvest time,
which is unknown. In other words, the supply, QSt, in period t depends on the price, Pt−1, in the
preceding period t − 1. The corresponding time-dependent supply and demand equations are
QSt = aPt−1 − b
QDt = −cPt + d
MFE_C09a.qxd 16/12/2005 10:49 Page 562

562 Dynamics

If we assume that, within each time period, demand and supply are equal, so that all goods are
sold, then
QDt = QSt
that is,
−cPt + d = aPt−1 − b
This equation can be rearranged as
−cPt = aPt−1 − b − d (subtract d from both sides)
b+d
Pt = AC − DF Pt−1 +
a
(divide both sides by −c)
c c
which is a difference equation of the standard form. The equation can therefore be solved in
the usual way and the time path analysed. Once a formula for Pt is obtained, we can use the
demand equation
Qt = −cPt + d
to deduce a corresponding formula for Qt by substituting the expression for Pt into the right-
hand side.

Example

Consider the supply and demand equations

QSt = 4Pt−1 − 10
QDt = −5Pt + 35
Assuming that the market is in equilibrium, find expressions for Pt and Qt when P0 = 6. Is the system stable
or unstable?

Solution
If
QDt = QSt
then
−5Pt + 35 = 4Pt−1 − 10
which rearranges to give
−5Pt = 4Pt−1 − 45 (subtract 35 from both sides)
Pt = −0.8Pt−1 + 9 (divide both sides by −5)
The complementary function is given by
CF = A(−0.8)t
and for a particular solution we try
Pt = D
for some constant D. Substituting this into the difference equation gives
MFE_C09a.qxd 16/12/2005 10:49 Page 563

9.1 • Difference equations 563

D = −0.8D + 9
which has solution D = 5. The general solution is therefore
Pt = A(−0.8)t + 5
The initial condition, P0 = 6, gives
6 = A(−0.8)0 + 5 = A + 5
and so A is 1. The solution is
Pt = (−0.8)t + 5
An expression for Qt can be found by substituting this into the demand equation
Qt = −5Pt + 35
to get
Qt = −5[(−0.8)t + 5] + 35
= −5(−0.8)t + 10
As t increases, (−0.8)t converges to zero and so Pt and Qt eventually settle down at the equilibrium levels
of 5 and 10 respectively. The system is therefore stable. Note also that because −0.8 lies between −1 and 0,
the time paths display oscillatory convergence.

Practice Problems

4 Consider the supply and demand equations

QSt = Pt−1 − 8
QDt = −2Pt + 22
Assuming equilibrium, find expressions for Pt and Qt when P0 = 11. Is the system stable or unstable?
5 Consider the supply and demand equations
QSt = 3Pt−1 − 20
QDt = −2Pt + 80
Assuming equilibrium, find expressions for Pt and Qt when P0 = 8. Is the system stable or unstable?

Two features emerge from the previous example and Practice Problems 4 and 5. Firstly,
the time paths are always oscillatory. Secondly, the system is not necessarily stable and so equi-
librium might not be attained. These properties can be explained if we return to the general
equation
b+d
Pt = AC − DF Pt−1 +
a
c c
The coefficient of Pt−1 is −a/c. Given that a and c are both positive, it follows that −a/c is
negative and so oscillations will always be present. Moreover,

if a > c then −a/c < −1 and Pt diverges

if a < c then −1 < −a/c < 0 and Pt converges.
MFE_C09a.qxd 16/12/2005 10:49 Page 564

564 Dynamics

We conclude that stability depends on the relative sizes of a and c, which govern the slopes of
the supply and demand curves. Bearing in mind that we have chosen to consider supply and
demand equations in which Q is expressed in terms of P, namely
QS = aP − b
QD = −cP + d
we deduce that the system is stable whenever the supply curve is flatter than the demand curve
when P is plotted on the horizontal axis.
Throughout this section we have concentrated on linear models. An obvious question to ask
is whether we can extend these to cover the case of non-linear relationships. Unfortunately, the
associated mathematics gets hard very quickly, even for mildly non-linear problems. It is usu-
ally impossible to find an explicit formula for the solution of such difference equations. Under
these circumstances, we fall back on the tried and trusted approach of actually calculating the
first few values until we can identify its behaviour. A spreadsheet provides an ideal way of doing
this, since the parameters in the model can be easily changed.

Example EXCEL

Consider the supply and demand equations

QSt = P 0.8
t−1

QDt = 12 − Pt
(a) Assuming that the market is in equilibrium, write down a difference equation for price.
(b) Given that P0 = 1, find the values of the price, Pt for t = 1, 2, . . . , 10 and plot a graph of Pt against t.
Describe the qualitative behaviour of the time path.

Solution
(a) If
QDt = QSt
then
12 − Pt = P 0.8
t−1

which rearranges to give

Pt = 12 − P 0.8
t−1

Notice that this difference equation is not of the form considered in this section, so we cannot obtain
an explicit formula for Pt in terms of t.
(b) We are given that P0 = 1, so we can compute the values of P1, P2, . . . in turn. Setting t = 1 in the differ-
ence equation gives
P1 = 12 − P0.8
0

= 12 − 10.8
= 11
This number can now be substituted into the difference equation, with t = 2, to get
P2 = 12 − P0.8
1 = 12 − 11
0.8
= 5.190 52
and so on.