Chapter 11.

Advanced Differential and Difference

Equations

In this chapter we deal with harder differential and difference equations. We already discussed first-
order equations in which a first-order derivative or difference is involved. Some more sophisticated
cases are second-, third-, or higher-order differential or difference equations. The chapter is split in
two: first we cover more advanced differential equations, and then we turn onto their discrete-time
counterpart, higher-order difference equations.

Second-order Differential Equations

Consider the linear differential equation

y ( n ) (t )  u1 (t ) y ( n 1) (t )  ...  un 1 (t ) y c(t )  un (t ) y (t ) v(t )

Since it contains the nth derivative y n (t ) of the function y (t ) , it is an n -th order differential
equation with variable coefficients. It is easy to notice that when only a first-derivative y c(t ) is
involved, the equation becomes the special case of a first-order differential equation
dy
 u (t ) y v(t )
dt
which we are already familiar with. By analogy with the constant coefficient case, we have the general
linear nth order equation

y ( n ) (t )  a1 y ( n 1) (t )  ...  an 1 y c(t )  an y b

where again the nth derivative y ( n ) (t ) is involved; but this time, the functions ui (t ) and v(t )
( i 1, 2.., n ) correspond to the constants ai and b , respectively. Similar to the first-order equation
case when only the first derivative y c(t ) is involved, we have the familiar equation
dy
 ay b
dt
We found the general solution to this simple first-order differential equation to be
b
y (t ) yc  y p Ae  at 
a
b
where y p is the particular integral giving the intertemporal equilibrium. This implies that we
a
have the simplest possible type of solution for y (t ) ; that is, y (t ) c where the function y is constant
dy
in time and the derivative is zero. Consider now the case
dt
y cc(t )  a1 y c(t )  a2 y b

where the highest derivative is the second-order derivative y cc(t ) . If we again assume the simplest
possible type, that is, y being a constant, we should have

y cc(t ) y c(t ) 0
615
616 Problems Book to Accompany Mathematics for Economists

and the particular integral is

b
yp a2 z 0
a2

Example: Find the particular integral of the equation y cc(t )  y c(t )  2 y

6 . Since a2 2 and
6
b 6 , substituting in the expression for the particular integral yields y p 3.
2
What if a2 0 so the expression for the particular integral is undefined? Then it must be that y is no
longer constant. A simple case to consider is y ct where again c const. Then the differential
equation becomes
y cc(t )  a1 y c(t ) b a2 0
Since y ct , it follows that y c(t ) c and y cc(t ) 0 , which reduces the equation to
b
y c(t )
a1

We find the particular integral by integrating y c(t ) with respect to t , which gives
b
yp t a2 0 a1 z 0
a1

Given that this time y p is a nonconstant function of time, it constitutes a moving equilibrium.

In the case when a1 a2 0 , the second-order differential equation becomes

y cc(t ) b
Integrating y cc(t ) twice with respect to t gives
bt 2
yp a1 a2 0
2

The Complementary Function

In the case of the first-order linear differential equation, its complementary function was the general
solution of the homogeneous (reduced) equation y c(t )  ay (t ) 0 , i.e., y (t ) Ae  at . Generally, an
expression of the form Ae rt fits well into complementary functions. One reason why we can apply
this exponential term to a second-order differential equation is that the latter is a second-order
generalization of the first-order homogeneous equation. If we assume the solution for the function
y (t ) to be of an exponential type y (t ) Aert , then we have

y c(t ) rAert and y cc(t ) r 2 Ae rt

Substituting these values of the derivatives and the parental function in the homogeneous second-order
differential equation yields

r 2 Ae rt  a1rAe rt  a2 Ae rt 0

which gives rise to the characteristic equation r 2  a1r  a2 0 and the two characteristic roots
a1 r a12  4a2
r1,2
2
 Chapter 11. Advanced Differential and Difference Equations 617

where by Viete’s formula1 r1  r2 a1 and r1r2 a2 . These two roots result in two solutions for
rt
y (t ) Ae , respectively

y1 A1e r1t and y2 A2 e r2t

where A1 and A2 are two arbitrary constants and the complementary function of the nonhomogeneous
(complete) equation is yc y1  y2 . Three possible situations exist in relation to the characteristic
roots r1 and r2 .
Case 1. Distinct real roots

If a12 ! 4a2 , then both roots r1 and r2 are distinct real numbers and we can write

yc y1  y2 A1er1t  A2 e r2t r1 z r2

For particular values of the two constants A1 and A2 implied by some initial conditions of y (t ) and
its derivatives, we can find the general solution to the complete equation as the sum of the
complementary function and the particular integral

y (t ) yc  y p y1  y2  y p A1e r1t  A2 e r2t  y p

Example: Solve the differential equation y cc(t )  y c(t )  2 y 6 . We already found the particular
integral of this nonhomogeneous equation to be y p 3 . How to find the complementary function?
We see that the equation fits this first case since a1 1 , a2 2 and a12 ! 4a2 because 1 ! 8 .
Furthermore, the characteristic roots are

1 r 1  8 1 r 3
r1,2
2 2
r1 1 r2 2

y (t ) A1e r1t  A2 e r2t  y p A1et  A2 e 2t  3

To find the particular values of the constants A1 and A2 , we need two initial conditions. Suppose
y (0) 10 and y c(0) 2 where the initial moment is t 0 . Substituting for t 0 we obtain

y (0) A1e0  A2 e 2(0)  3 A1  A2  3 10

Differentiating y (t ) with respect to t , we get

y c(t ) A1et  2 A2 e 2t

Then at t 0,

y c(0) A1e0  2 A2 e 2(0) A1  2 A2 2

which leads to the system of equations
A1  A2 7
A1  2 A2 2
with solutions A1 4 and A2 3 . Substituting to obtain the definite solution of the second-order
differential equation,

1
Named after the French mathematician François Viete (1540-1603).
618 Problems Book to Accompany Mathematics for Economists

y (t ) 4et  3e 2t  3
Case 2. Single real root
a1
If a12 4a2 , there is only one root (also called a coincident or repeated real root) r  . Then the
2
complementary function is

yc A1e rt  A2 h(t )

where h(t ) is a function that cannot be a constant multiple of e rt . Therefore, we set h(t ) te rt , and
the general solution to the second-order differential equation is

y (t ) yc  y p A1e rt  A2te rt  y p

Example: Solve the differential equation y cc(t )  2 y c(t )  y

5 . We can easily notice that a12 4a2
a 2
since a1 2 , a2 1 and 22 4(1) . Thus, the example is one of a sinlge real root r  1  1
2 2
and y p 5 . Therefore, y (t ) yc  y p A1e  t  A2te t  5 .

Case 3. Complex roots

What if a12  4a2 ? Then the roots r1 and r2 contain the square root of a negative number i 1
called an imaginary number. The very roots are called complex numbers as they contain a real part
and an imaginary part, for instance, (5  i ) , where we already defined i . Complex numbers cannot
be ordered along the real line and, therefore, do not belong to the real-number system. They can
generally be represented in the form (m  ni ) where m and n are two real numbers. A complex
number can be represented graphically in the xy -plane where x is the real-number axis and y is the
imaginary-number axis. In this two-dimensional diagram known as the Argand diagram (shown by
Figure 1) m is plotted on the horizontal axis and n on the vertical. Thus when n 0 , the complex
number does not have an imaginary part and reduces to a real one. When m 0 , it is solely an
imaginary number. By Pythagoras theorem the length of the ON line is found as the radius vector
R m2  n2 .
Imaginary
axis

N (m, n)

R m2  n2
n

0 m M Real axis

Figure 1 Argand diagram

When a12  4a2 , the two roots of the characteristic equation are a pair of conjugate complex numbers:
a 4a2  a12
r1,2 m r ni where m  1 n and i 1
2 2
In the complex-root case the complementary function of the differential equation becomes
 Chapter 11. Advanced Differential and Difference Equations 619

yc A1e( m  ni )t  A2 e( m  ni )t e mt ( A1e nit  A2 e nit )

Example: Find the roots of the characteristic equation r 2  r  2 0 . Express the complementary
function for this equation. We obtain a pair of conjugate complex numbers for the two characteristic
roots.
1 r 7 1 7
r1,2  r i
2 2 2
It can easily be checked that, in accordance with Viete’s formula, r1  r2 a1 1 and r1r2 a2 2.
1 7
Since m  and n , the complementary function is
2 2
 §
t 7it 7it ·

yc e 2 ¨ A1e 2  A2 e 2 ¸
¨ ¸
© ¹
where the imaginary number i appears in the exponents of the two expressions of the complementary
function. To understand such imaginary exponential functions better, we should transform them into
circular functions, which requires some discussion of trigonometry.

An Excursion into Trigonometric Functions

In the next part we will briefly revise some basics of trigonometric functions that may be familiar to
you from high school. Trigonometric functions are often connected with complex numbers. Given an
angle T , as shown in Figure 2 depicting a circle with a radius R , the trigonometric functions are
n m
sin T cos T
R R
where m, n and R happen to be sides of the right-angle triangle OMN . Two more trigonometric
functions can be defined on the basis of these two original functions:
sin T n cos T m
tan T cot T
cos T m sin T n
Q

n
O T
S P
m M

T
Figure 2

The angle T is measured in degrees (say, 90$ ) or in radians which allow expressing the derivatives of
trigonometric functions more easily. The size of the angle T is defined by the PN arc. A complete
620 Problems Book to Accompany Mathematics for Economists

circle like PQST involves an angle of 2S radians which is exactly 360$ or S 180$ . Thus, radians
transform into degrees according to the following conversion table (see Table 1).

Degrees 0 30$ 45$ 60$ 90$ 135$ 180$ 270$ 360$

Radians 0 S S S S 3S S 3S 2S
6 4 3 2 4 2
sin T 0 1 2 3 1 2 0 -1 0
2 2 2 2
cosT 1 3 2 1 0 2 -1 0 1

2 2 2 2
Table 1

The sine and the cosine functions are periodic and repeat every 360$ . They both fluctuate between 0
and 1 but differ in their peaks as shown on the following two diagrams in Figures 3a and 3b.

0
S S 3S 2S 5S 3S
-1 2 2 2

sin T
(a)

0
S S 3S 2S 5S 3S
-1 2 2 2

cos T
Figure 3 (b)

The sine and cosine functions have the following properties:

sin(T )  sin T
cos(T ) cosT
sin 2 T  cos 2 T 1
sin(T1 r T 2 ) sin T1 cosT 2 r cosT1 sin T 2
cos(T1 r T 2 ) cos T1 cos T 2 r sin T1 sin T 2
Given that both sin T and cos T are continuous and smooth, they are differentiable. The derivatives
of the functions, applicable to radians only, turn out to be
 Chapter 11. Advanced Differential and Difference Equations 621

d sin T d cosT d tan T 1

cosT  sin T
dT dT dT cos 2 T
In the general case, given that u is a differentiable function of x , the derivatives for sine amd cosine
can be written as
d sin u du d cos u du
cos u  sin u
dx dx dx dx
Example: Find the second derivative of cosT . Applying the formula twice, we get

d 2 cos T d (  sin T )
 cos T
dT 2 dT

Example: Find the derivative of the trigonometric function sin(5 x 2  2) . Using the general formula,

d sin(5 x 2  2)
10 x cos(5 x 2  2)
dx

Transforming Complex Numbers into Trigonometric Functions

As long as the two Cartesian coordinates2 m and n are defined, we can find the angle T and the
radius R , also known as polar coordinates. A basic relationship between Cartesian and polar
coordinates we obtained previously is R m 2  n 2 . In the opposite case knowing the values of R
and T , we can write m R cos T and n R sin T . Thus, the pair of conjugate complex numbers
m r ni becomes
m r ni R cosT r Ri sin T R (cos T r i sin T )
By what is known as Euler’s formula for complex numbers, which we will not prove here,

eiT cosT  i sin T and e iT cos T  i sin T so

m r ni R (cosT r i sin T ) R.e r iT

Abraham De Moivre3 discovered further that

> R(cosT r i sin T )@k R k (cos kT r i sin kT )

This result allows to find a pair of conjugate complex numbers raised to the power k such as
(m r ni ) k where by De Moivre’s theorem

(m r ni ) k R k e r ikT R k (cos kT r i sin kT )

Going back to the complementary function yc e mt ( A1enit  A2 e  nit ) , we let T nt and by Euler’s
formula

enit cos nt  i sin nt and e nit cos nt  i sin nt

and substituting these consequently into the complementary function,

yc e mt > A1 (cos nt  i sin nt )  A2 (cos nt  i sin nt )@ emt ( B1 cos nt  B2 sin nt )

where B1 A1  A2 and B2 ( A1  A2 )i

2
Named after the talented French mathematician Rene Descartes (1596-1650).
3
Except relating complex numbers to trigonometry, the French mathematician Abraham De Moivre (1667-
1754) is credited for the study of normal distribution and probability theory.
622 Problems Book to Accompany Mathematics for Economists

Example: Find the polar and exponential forms of 2  2i . The Cartesian coordinates are

m 2 and n 2 . We can find R by the formula R m2  n2 22 2 . Thus, we can find

the sine and cosine functions as

n 2 m 2
sin T and cos T
R 2 R 2
S
But we know that these are the values for T 45$ . Hence, from the formula
4
m r ni R (cos T r i sin T ) R.e r iT
iS
§ S S·
2  2i 2 ¨ cos  i sin ¸ 2e 4
© 4 4¹
Example: Find the complementary function and the particular integral of the differential equation
y cc(t )  2 y c(t )  10 y 20
for which the initial conditions are y (0) 3 and y c(0) 11 . Here we have a1 2 , a2 10 and
b 20
b 20, so for the particular integral we get y p 2 . Furthermore, since a12  4a2 or
a2 10
4  40 , the characteristic roots are r1,2 m r ni and

a1 4a2  a12 4(10)  22 6

m  1 n 3
2 2 2 2
yc e mt ( B1 cos nt  B2 sin nt ) e t ( B1 cos3t  B2 sin 3t )
Thus, the general solution of the differential equation is

y (t ) yc  y p e t ( B1 cos3t  B2 sin 3t )  2

To definitize the constants B1 and B2 we must use the initial conditions. Substituting for t 0 in
y (t ), we obtain y (0) e ( B1 cos 0  B2 sin 0)  2 3 where we know that cos 0 1 and sin 0 0 , so
0

we have B1  2 3 or B1 1 . Expressing y c(t ) from y (t ) ,

y c(t ) e t ( B1 cos3t  B2 sin 3t )  e t (3B1 sin 3t  3B2 cos3t )

and setting t 0 again
y c(0) e0 ( B1 cos 0  B2 sin 0)  e0 (3B1 sin 0  3B2 cos 0)  B1  3B2 11 or

1  3B2 11
3B2 12
B2 4
So, the differential equation is

y (t ) e t (cos3t  4sin 3t )  2

Dynamic Stability

The time path of the complementary function yc e mt ( B1 cos nt  B2 sin nt ) depends on the sine and
cosine functions as well as on the term emt . Since the period of the trigonometric functions is 2S and
 Chapter 11. Advanced Differential and Difference Equations 623

their amplitude is 1, their graphs will repeat their shape every time the expression nt increases by 2S .
Alternatively,
§ 2S ·
T nt nt  2S n ¨ t 
© n ¸¹
2S
The first term in the parentheses B1 cos nt is a cosine function of t with a period . Similarly, the
n
second term B2 sin nt has the same period and fluctuates between  B2 and B2 . The dynamic
stability of the function y (t ) depends solely on the third term, emt such that for a positive m , as
t o f , the amplitude of ( B1 cos nt  B2 sin nt ) magnifies and causes an explosive fluctuation for
y (t ) . If m 0 , the complementary function has a uniform fluctuation. When m is negative, the time
path of the function y (t ) is dynamically stable (See Figures 4a, 4b, and 4c).

y (t ) m!0

Equilibrium
level

0
t
Explosive fluctuation

(a)

y (t ) m 0

0
t
Uniform fluctuation
(b)

y (t )

m0

0
t
Damped fluctuation

Figure 4 (c)
624 Problems Book to Accompany Mathematics for Economists

For the second-order differential equation

y cc(t )  a1 y c(t )  a2 y b
we found that the time path of the parental function y (t ) in the case of distinct real roots is
y (t ) A1e r1t  A2 e r2t  y p where A1 and A2 are arbitrary constants. The dynamic stability of the
function is insured only if both roots r1 and r2 are negative; so, as t o f , y (t ) converges to its
equilibrium level y p . If even one of the roots is positive, its exponential term becomes infinitely large,
thereby precluding convergence.

With a single root, the solution is y (t ) A1e rt  A2te rt  y p . Here the necessary and sufficient
condition for dynamic stability is that the single root r be negative as t o f . The second
multiplicative term, A2te rt , also approaches zero because with a negative exponent the exponential
term reaches zero faster than t grows.

Finally, in the case of complex roots, the solution is y (t ) e mt ( A1e nit  A2 e  nit )  y p where

a1 4a2  a12
r1,2 m r ni m  and n
2 2
The condition for convergence of the y (t ) path is m  0 , that is, the real part m of the complex roots
to be negative. Then, for the three cases, it is enough to demand that the real part of every
characteristic root be negative to ensure dynamic stability of equilibrium.

The Arrow-Pratt Measure of Risk Aversion

Let a person have wealth in the amount w and u ( w) be the utility function over this wealth. To
measure the concavity of the utility function u ( w) in portfolio choice theory, Kenneth Arrow and
John W. Pratt use the so-called Arrow-Pratt measure of relative (or absolute) risk aversion at wealth
level w generally given by the expression
u cc( w) w
Ewuc 
u c( w)

which is nothing but the elasticity of the marginal utility function u c( w) with respect to the wealth
level w . Since we do not want total utility of wealth to be declining, we require u c( w) ! 0 . If we
assume the individual to have a constant relative risk aversion (say, be either risk averse or risk loving
or risk neutral), we can adopt a constant elasticity of k . Thus, the expression becomes
u cc( w) w
 k
u c( w)
which gives the second-order differential equation
k
u cc( w)  u c( w) 0
w
Furthermore, if we substitute uw u c( w) for the marginal utility of wealth,

k
uwc  uw 0
w
we obtain a first-order differential equation in marginal utility uw . Rearranging and solving by the
separation of variables method,
 Chapter 11. Advanced Differential and Difference Equations 625

uwc k

uw w

uwc k
³u w
dw  ³ w dw
ln uw  k ln w  c and taking the antilog of both sides,

uw e  k ln w ec Cw k
Integrating marginal utility uw u c( w) further to obtain the total utility function, we have

­ C 1 k
° w  c1 k z1
³ uc(w)dw ³
k
u ( w) Cw dw ®1  k for
° C ln w  c2 k 1
¯
A special case of the measure of risk aversion is the Arrow-Pratt measure of absolute risk aversion
representing the percent rate of change of marginal utility of wealth u c( w) at wealth level w . It is
again a measure of the concavity of the total utility function u ( w) and is given by the expression

u cc( w)
 a( w)
u c( w)
thus giving rise to the second-order differential equation
u cc( w)  a( w)u c( w) 0
which can be solved for a specific function a( w) .

Market Equilibrium with Price Expectations

Very often, market participants base their demand and supply decisions on their expectations about the
price and its behavior in the future. Those expectations are often influenced not only by the price
prevailing at the moment, but also by the trends in the price movements. We can apply second-order
differential equations to establish the time path of market price assuming equilibrium in each moment
in time. Let us take, for example, that
qd D  E p  upc  vpcc D,E ! 0
qs J  G p J ,G ! 0
dp d2 p
where pc and pcc . In the context of price trends, a positive pc implies that market price
dt dt 2
is rising and a positive pcc shows that it is rising at an increasing rate.

Then, if u ! 0 , a rising price increases market demand. Buyers, expecting price to rise, would prefer
to increase current consumption. An example of such move is the real estate market in Bulgaria. Prior
to Bulgaria’s joining the European Union, people expected the prices of houses to continue to rise, so
they increased their purchases, thus pushing the prices further up. Conversely, when u  0 , people
expect the price trend to reverse and, therefore, they cut back on their purchases in expectation of a
lower price in the future. Similarly, the Bulgarian real estate market experienced a slowdown in
housing prices after the country was accepted into the EU. Furthermore, the global financial crisis
influenced the decisions of house buyers negatively. Expecting real estate prices to fall, they stopped
buying, which contributed further to the decline of prices. Thus, this continuous-time model illustrates
how people’s expectations of the future shape current prices. In their buying decisions, consumers
626 Problems Book to Accompany Mathematics for Economists

may be driven not only by the direction of change in market price, but also by the rate at which this
change occurs given here by the parameter v .

Equating market demand with market supply, we get

qd qs or
D  E p  upc  vpcc J  G p
which transforms into
vpcc  upc  ( E  G ) p D  J
Normalizing this second-order differential equation,
u (E  G ) D J
pcc  pc  p 
v v v
Note that the present model assumes market clearance at every moment in time. Thus, every price
established in the market at any given moment is an equilibrium price, although this market
equilibrium may not be the intertemporal equilibrium. You can think of the intertemporal equilibrium
price as the normal price that should prevail in the market at any moment. Yet, at different moments
the market might clear at too high or too low levels of price. When Bulgaria experiences a real estate
bubble, prices of apartments are abnormally high – but demand and supply meet, so the market clears
and the price is the market equilibrium price, though not the normal price level or the intertemporal
equilibrium. In times of a deep recession, the market still clears – but at very low price levels, much
below the intertemporal equilibrium.

For the second-order differential equation that obtains, we have

u E G D J
a1 a2  and b 
v v v
In order to solve the differential equation, we need to find the time path of the price function p (t ) .
The particular integral will give us the intertemporal equilibrium price:
b D J
pp
a2 E G
Note that the intertemporal equilibrium price is positive, as it should be, since all the constants in it are
positive. Also, because it is constant, it represents a stationary, not a moving equilibrium. To find the
complementary function, we discuss three cases.

Case 1. Distinct real roots

2
§u· § E G ·
¨ ¸ ! 4 ¨ ¸
©v¹ © v ¹
The complementary function for this case is

pc A1e r1t  A2 e r2t

where to definitize the constants A1 and A2 we need some initial conditions for price.

1 «ª u º
2
§u· § E G ·»
r1,2  r ¨ ¸  4¨ ¸
2« v ©v¹ © v ¹»
¬ ¼
Thus, the general solution is
 Chapter 11. Advanced Differential and Difference Equations 627

D J
p (t ) pc  p p A1e r1t  A2 e r2t 
E G
Case 2. Single real root
2
§u· § E G ·
¨ ¸ 4 ¨ ¸
©v¹ © v ¹
a1 u
The single root is r   . Thus, the general solution becomes
2 2v
D J
ut ut
 
p (t ) A1e 2v  A2te 2v 
E G
Case 3. Complex roots
2
§u· § E G ·
¨ ¸  4 ¨ ¸
©v¹ © v ¹
The characteristic roots are a pair of conjugate complex numbers r1,2 m r ni where

a u 4a2  a12 1 § E G · § u ·
2
m  1  and n 4 ¨ ¸¨ ¸
2 2v 2 2 © v ¹ ©v¹
For the general solution, we have

D J D J
ut

p (t ) e mt ( A1e nit  A2 e  nit )  e 2v ( B1 cos nt  B2 sin nt ) 
E G E G
§ E G ·
If v ! 0 , then 4 ¨ ¸ is always negative, so only the first case of distinct real roots is possible.
© v ¹
2
§u·
Under the square root we get a number bigger than ¨ ¸ , which means that at least one characteristic
©v¹
root is positive. Therefore, the intertemporal equilibrium must be dynamically unstable. If v  0 , all
three cases are possible. In the case of distinct real roots both roots will be negative, given u  0 . This
2
§u· u
is because the expression under the square root is definitely smaller than ¨ ¸ and the free term is
© ¹
v v
positive. Hence, both characteristic roots turn out to be negative. The condition u , v  0 also ensures
that the single root is negative. In the third case of complex numbers when u , v  0 , m turns out to be
negative too. Therefore, the dynamic stability of the price function in each case is ensured when both
parameters u and v are negative.

The Relationship between Inflation and Unemployment

Let us assume that the rate of inflation is negatively related to the level of unemployment and
positively to the expected rate of inflation in a dependence known as the Phillips relation such that
p D  E U  hS 4
D,E ! 0 0  h d1

4
The original idea underlying the model was expressed by A. W. Phillips in a path-breaking paper titled
“The Relationship between Unemployment and the Rate of Change of Money Wage Rates in the United
Kingdom, 1861-1957,” Economica, November 1958, pp. 283-299. The expanded version of the Phillips
relation incorporates the growth rate of money wage w  where the rate of inflation is the difference between
  T . Thus, inflation would
the increase in wage and the increase in labor productivity T , that is, p w
result only when wage increases faster than productivity. Furthermore, wage growth is negatively related to
628 Problems Book to Accompany Mathematics for Economists

pc
where p is the rate of growth of the price level (that is, the inflation rate), U is the rate of
p
unemployment and S denotes the expected rate of inflation. Thus, the expectation of higher inflation
shapes the behavior of firms and individuals in a way that stimulates inflation. Expecting prices to rise,
they might decide to buy more immediately. As people expect inflation to go down (as a result of
appropriate government policies, for example) this brings actual inflation down. This version of the
Phillips relation that accounts for the expected rate of inflation is called the expectations-augmented
Phillips relation. The adaptive expectations hypothesis further shows how inflationary expectations
are formed. The equation
dS
j ( p  S ) 0  j d1
dt
illistrates that when the actual rate of inflation exceeds the expected one, this nurtures people’s
dS
expectations, so ! 0 . In the opposite case, if the actual inflation is below the expected one, this
dt
makes people believe that inflation would go down, so S is reduced. If the projected and the real
inflation turn out to be equal, people do not expect a change in the level of inflation.

There is also the reverse effect – that of inflation on unemployment. Thus, when inflation is high for
too long, for example, this may discourage people from saving, consequently reducing aggregate
investment and increasing the rate of unemployment. We can write that
dU
 k (m  p ) k !0
dt
or unemployment increases proportionally with real money where m  is the rate of growth of nominal
money. Thus, the expression (m  p ) gives the rate of growth of real money, or the difference
between the growth rate of nominal money and the rate of inflation
mc p c
m  p  rm p
m p
where real money is nominal money divided by the average price level in the economy. The model
then becomes
p D  E U  hS D,E ! 0 0  h d1 (expectations-augmented Philips relation)
dS
j ( p  S ) 0  j d 1 (adaptive expectations)
dt
dU
k (m  p ) k !0 (monetary policy)
dt
We can substitute the first equation into the second, which gives
dS
j (D  EU  hS  S )
dt
dS
j (D  EU )  j (h  1)S
dt
Differentiating further with respect to time t ,

unemployment and positively to the expected rate of inflation or w   E U  hS where U is the rate of
unemployment and S is the expected rate of inflation. If inflationary trends persist long enough, people
start forming further inflationary expectations that shape their money-wage demands.
 Chapter 11. Advanced Differential and Difference Equations 629

d 2S dU dS
 jE  j (h  1)
dt 2 dt dt
dU
and substituting for , we obtain
dt
d 2S dS
2
j E k (m  p )  j (h  1)
dt dt
1 dS
where the second equation of the model implies p  S . Substituting this last expression for
j dt
p , we get
d 2S § 1 dS · dS
j E k ¨ m   S ¸  j (h  1)
dt 2 © j dt ¹ dt

which is a second-order differential equation in S . Transforming the equation,

d 2S dS dS
2
j E km  E k  j E kS  j (h  1)
dt dt dt
d S
2
dS
2
 > E k  j (1  h) @  j E kS j E km or more simply
dt dt
S cc  > E k  j (1  h) @S c  j E kS j E km
where, given the properties of second-order differential equations, we have
a1 E k  j (1  h) a2 jE k b j E km

The coefficients a1 and a2 are both positive in view of the signs of the parameters. We can
immediately find the equilibrium rate of expected inflation to be the particular integral
b
Sp m
a2
Thus, the intertemporal equilibrium of the expected rate of inflation is exactly the rate of growth of
nominal money. To find the time path of S we need to find the characteristic roots of the differential
equation by the usual formula

a1 r a12  4a2

r1,2
2
The time path of S would depend on the particular values of the parameters. Once we find this time
path, we migh be able to determine that of unemployment U or the rate of inflation p .

Inflation and Unemployment: An Extended Model

Following Olivier Blanchard’s book Macroeconomics5, we can assume that the rate of change of the
inflation rate p is proportional to the difference between the actual unemployment rate U and the
natural rate of unemployment U n such that

dp
D (U  U n ) D !0
dt

5
Blanchard, Olivier. Macroeconomics. 2nd edition, Prentice Hall, 2000.
630 Problems Book to Accompany Mathematics for Economists

Thus, when U ! U n , that is, the actual rate of unemployment exceeds the natural rate, the inflation
rate decreases and when U  U n , the inflation rate increases. The intuitive logic behind this is that in
bad economic times when many people are laid off, prices tend to fall. At this point, the actual
unemployment would exceed the normal levels. In times of a boom in the business cycle, the rate of
actual unemployment would be rather low, but high aggregate demand would push prices up. We also
assume that U n is constant and that at any given time the actual unemployment rate U is determined
by aggregate demand which, on its own, depends on the real value of money supply given by nominal
money supply M divided by the average price level p . Thus, unemployment is negatively related to
M
real money supply according to the relationship
p
M
U J  E ln E ,J ! 0
p
By differentiating the first equation with respect to t ,

d 2 p dU
D
dt 2 dt
dU
and the second equation to obtain
dt
dU d M § d ln M d ln p ·
E ln E ¨   E (m  p )
dt dt p © dt dt ¸¹
where we assume that the growth rate of nominal money supply m is constant. This could be in
accordance with government planning or systematic monetary policy. Combining the two results
yields

d 2 p dU
D DE (m  p )
dt 2 dt

d 2 p
 DE p DE m
dt 2
which is a second-order differential equation in inflation rate p . Solving the differential equation, we
have a1 0 , a2 DE and b DE m  . Hence, the particular integral is p e m and the characteristic
equation is

r 2  DE 0
r1,2 r DE i where m 0 and n DE
Thus, the general solution takes the form

p (t ) m  e0 B1 cos t DE  B2 sin t DE m  B1 cos t DE  B2 sin t DE

Since the real part is zero, the function of inflation rate displays regular oscillations about the rate of
growth of money supply, which gives the equilibrium level of inflation. To find the time path of
dp
unemployment U , we express :
dt
dp
DE  B1 sin t DE  B2 cos t DE
dt
and substitute it into
 Chapter 11. Advanced Differential and Difference Equations 631

1 dp E
U Un  Un   B1 sin t DE  B2 cos t DE
D dt D
U n  B1 E D sin t DE  B2 E D cos t DE

where the constants B1 and B2 have not been definitized. It follows that the unemployment rate also
displays regular oscillations, similar to the inflation rate, but its equilibrium is the natural rate of
unemployment. Since the real part is zero, again the time path is neither convergent nor divergent.

Higher-order Differential Equations

Recall the nth -order differential equation with constant coefficients

y ( n ) (t )  a1 y ( n 1) (t )  ...  an 1 y c(t )  an y b

To solve such higher-order differential equations, we simply repeat the steps applicable to second-
order differential equations. With a simplest constant function assumed, such as y c, all derivatives
are zero. The particular integral is
b
yp c an z 0
an

If an 0 , we try y ct such that y c(t ) c; but all other derivatives are zero, so the equation
becomes an 1c b . The particular integral is
b
y p ct t an 0 an 1 z 0
an 1
In the case when an an 1 0 , the solution must be of the type y ct 2 . This produces the
derivatives y (t ) 2ct and y cc(t ) 2c and the particular integral
c
b 2
y p ct 2 t an an 1 0 an  2 z 0
an  2
The complementary function is the general solution of the homogeneous equation

y ( n ) (t )  a1 y ( n 1) (t )  ...  an 1 y c(t )  an y 0

If the solution is in the form y Ae rt , the derivatives can be written off as y c(t ) rAe rt ,
y cc(t ) r 2 Ae rt ,…, y ( n ) (t ) r n Ae rt . This gives rise to the nth -degree characteristic equation with n
roots

Ae rt (r n  a1r n 1  ...  an 1r  an ) 0 or

r n  a1r n 1  ...  an 1r  an 0

Thus, the complementary function, with all roots real and distinct, can be written as

yc A1e r1t  A2 e r2t  ...  An ernt

Assuming that the first two roots are repeated, we have r1 r2 , so the first two terms of the general
solution to the differential equation can be written as A1e  A2te r1t . If we assume further that the next
r1t

two roots are complex such that r3,4 m r ni , the general solution of this differential equation can be
written as

y (t ) A1e r1t  A2te r1t  e mt ( B1 cos nt  B2 sin nt )  y p

632 Problems Book to Accompany Mathematics for Economists

To find the values of the four arbitrary constants, we need four initial conditions.
Example: Solve the differential equation y ccc(t )  y cc(t )  3 y c(t )  5 y 15 . Since the highest
derivative is the third-order derivative y ccc(t ) , this is a third-order differential equation with a
15
particular integral y p  3 . Its characteristic equation is r 3  r 2  3r  5 0, which can be
5
factored out into

(r  1)(r 2  2r  5) 0
with one real root r1 1 and a pair of complex conjugate roots r1,2 1 r 2i . The general solution,
therefore, is

y (t ) A1et  e t ( B1 cos 2t  B2 sin 2t )  3

Second-order Difference Equations

We recall that first-order difference equations involve terms like yt 1 and yt where the difference in
each period is given. Thus knowing some initial value yo we can determine the time path of the y
function as the time factor t changes. A simple second-order difference equation is
yt  2  b1 yt 1  b2 yt c

To find the particular integral in the simplest case, we can take a solution of the form yt k where in
every period y is the constant k :

k  b1k  b2 k c and
c
yp k where b1  b2 z 1
1  b1  b2

If b1  b2 1 , we must adopt yt kt . This will result in the terms yt 1 k (t  1) and

yt  2 k (t  2) for the next time periods. Substituting these expressions into the difference equation
gives
c c
k where b1  b2 1
(1  b1  b2 )t  b1  2 b1  2
Hence, the particular integral is
c
y p kt t where b1  b2 1 and b1 z 2
b1  2
Since the particular integral in this second case involves t , it represents a moving equilibrium. If
b1  b2 1 (so that b1 2 and b2 1 ), then we try a solution of the type yt kt 2 so the other two
c
terms are yt 1 k (t  1) 2 and yt  2 k (t  2) 2 . They generate k and
2
c 2
yp kt 2 t where b1 2 and b2 1
2

The Complementary Function

With first-order difference equations, we found that the expression yt Aa t describes well the
general solution of such an equation, and we try it to find the complementary function. This implies
that yt 1 Aa t 1 and yt  2 Aa t  2 , which upon substitution in
 Chapter 11. Advanced Differential and Difference Equations 633

yt  2  b1 yt 1  b2 yt 0 yields

Aa t  2  b1 Aa t 1  b2 Aa t 0 or

a 2  b1a  b2 0
b1 r b12  4b2
This characteristic equation has the roots a1,2 . Hence, for the complementary
2
function, we have three possibilities again:

Case 1. Distinct real roots

If b12 ! 4b2 , then both roots are real and different, so the complementary function is

yc y1  y2 A1a1t  A2 a2t

Example: Solve the second-order difference equation yt  2  3 yt 1  4 yt 15 . Here we have b1 3,

b2 4, and c 12 . We also note that b1  b2 1, so

c 15
yp t t 3t
b1  2 3 2

The characteristic equation is a 2  3a  4 0 with roots a1 1 and a2 4 . Thus, the general

solution is
yt yc  y p A1  A2 ( 4)t  3t

Suppose we are also given that yo 3 and y1 1 for the two periods t 0 and t 1 , respectively.
Substituting these values for t , we obtain
yo A1  A2 3
y1 A1  4 A2  3 1
Thus, the constants are A1 2 and A2 1 and the final solution is

yt 2  (4)t  3t
Case 2. Single real root
b1
If b12 4b2 , there is only one real root a  . Then the complementary function is
2
yc y1  y2 A1a t  A2ta t

Example: Solve the second-order difference equation yt  2  2 yt 1  yt 8 . Here we have b1 2 ,

b2 1, and c 8 . Also, b1  b2 1 implies a particular integral of the type
c 2 8 2
yp t t 4t 2
2 2
The characteristic equation is a 2  2a  1 0 with a single root a 1 . The general solution is,
therefore, yt yc  y p A1  A2t  4t 2 .

Case 3. Complex roots

When b12  4b2 , again a pair of conjugate complex numbers a1,2 m r ni obtains where
634 Problems Book to Accompany Mathematics for Economists

b 4b2  b12
m  1 and n
2 2
The complementary function is

yc A1a1t  A2 a2t A1 (m  ni )t  A2 (m  ni )t

From the De Moivre’s theorem it follows that (m r ni )t R t (cos T t r i sin T t ) where

b12  4b2  b12

R m2  n2 b2
4
m b1 n b12
Here T is measured in radians and cos T  and sin T 1 . Hence, the
R 2 b2 R 4b2
complementary function is

yc A1 R t (cos T t  i sin T t )  A2 R t (cosT t  i sin T t ) R t ( B1 cosT t  B2 sin T t )

where the multiplicative factor R t substitutes the natural exponential term e mt used in differential
equations.
Example: Find the general solution to the equation yt  2  3 yt 1  9 yt 14 . Here we have b1 3 ,
b2 9, and c 14 . For the particular integral,
c 14
yp 2
1  b1  b2 1  3  9

This is the case of complex roots since b12  4b2 , therefore, R b2 9 3.

b1 3 1 b12 9 3
cos T  and sin T 1 1
2 b2 2(3) 2 4b2 4(9) 2
S
From the respective values of the two trigonometric functions, we infer that T . Hence, the
3
general solution is

§ S S ·
yt yc  y p 3t ¨ B1 cos t  B2 sin t ¸  2
© 3 3 ¹
Dynamic Stability

For the second-order difference equation

yt  2  b1 yt 1  b2 yt c

we have the term a t , which may show oscillatory behavior depending on the value of the base a . In
the case of two distinct roots, if a1 ! 1 and a2 ! 1 , both terms in the complementary function
yc A1a1t  A2 a2t will be explosive and the time path of yt is divergent (the time path is oscillatory, if
a  0 ). When a1  1 and a2  1 , both terms will converge to zero as t o f and the time path is
convergent. If either a1 or a2 is greater than 1, the time path is divergent. With a single root, the
function is dynamically stable if a  1 . For the case of complex roots, we found the solution to be

yt R t ( B1 cos T t  B2 sin T t )  y p
 Chapter 11. Advanced Differential and Difference Equations 635

The parenthesized expression shows a fluctuating path because it contains circular functions. The
fluctuation would be a stepped (nonsmooth) fluctuation, rather than oscillation. If R  1 , the time path
would be dynamically stable. Since R is the absolute value of the conjugate complex roots m r ni ,
the condition for convergence is again that the characteristic roots be less than 1. In all cases, the time
path of yt will be dynamically stable if the absolute value of every root is less than 1.

The Multiplier-Accelerator Model

The multiplier-accelerator model shows the interaction between aggregate investment and output.
Usually, in the presence of positive exogenous shocks, increased investment has a multiplying effect
on GDP by the amount of the investment multiplier, but the increase in GDP makes firms believe that
demand for their goods has increased. This stimulates them to invest more in capital stock, a process
known as the accelerator. Thus, investment stimulates GDP through the multiplier process while GDP
further pushes up investment through the accelerator process in an interactive way. Of course, a
downturn in the economy would have an effect opposite to the multiplier-accelerator process or would
force the economy to contract. The model was first advanced by Paul Samuelson, who extended the
Keynesian national-income model of the investment multiplier by the accelerator principle.6 The
model assumes the following three equations:
Yt Ct  I t  Go
Ct E Yt 1 0  E 1
I t D (Ct  Ct 1 ) D !0
People spend based on income earned in the previous period where E shows the share of income that
is consumed, that is, the marginal propensity to consume. Furthermore, investment is positively
related to the increase in aggregate consumption 'Ct 1 D (Ct  Ct 1 ) showing here the accelerator
effect. That is, based on increased consumption, firms expect demand for their product to rise and,
hence, decide to increase investment. Note also that the parameter D is called an accelerator
coefficient and is greater than zero. Substituting the respective terms for Ct in the last equation, we
obtain
I t D ( E Yt 1  E Yt  2 )
and substitute this new result and the second equation into the first one:

Yt E Yt 1  DE Yt 1  DE Yt  2  Go
Yt  2  E (1  D )Yt 1  DE Yt Go

The parameters in this second-order difference equation are b1  E (1  D ), b2 DE , and c Go .

We can easily find the particular integral as
c Go Go
Yp
1  b1  b2 1  E (1  D )  DE 1 E
Since E is less than 1, we could expect a meaningful intertemporal equilibrium for national income
1
that is positively related to exogenous government spending. Furthermore, is the value of the
1 E
multiplier. The characteristic equation of the model is
E (1  D ) r E 2 (1  D ) 2  4DE
a 2  E (1  D )a  DE 0 where a1,2
2

6
Samuelson, Paul A. “Interaction between the Multiplier Analysis and the Principle of Acceleration.”
Review of Economic Statistics, May, 1939, pp. 75-78; reprinted in American Economic Association,
Readings in Business Cycle Theory, Richard D. Irwin, Inc., Homewood, Ill., 1944, pp. 261-269
636 Problems Book to Accompany Mathematics for Economists

where we know the two roots satisfy the conditions

a1  a2 E (1  D ) and a1a2 DE
Thus,
(1  a1 )(1  a2 ) 1  (a1  a2 )  a1a2 1  E (1  D )  DE 1 E
which implies that 0  (1  a1 )(1  a2 )  1 . Furthermore, for the complementary function, there are
three possible cases depending on whether b12 ! 4b2 or not. This first case is equivalent to

E 2 (1  D ) 2 ! 4DE
4D
E!
(1  D ) 2

In this first case of distinct real roots since a1a2 ! 0 and a1  a2 ! 0 , both roots are positive. This
precludes oscillation, and convergence would depend on whether a1 and a2 are smaller or bigger
than 1. Several cases might be considered, but the legitimate ones are presented in Table 2. Similar is
E (1  D )
the case of a single real root a , which is positive again. Oscillation is excluded and the
2
dynamic stability of national income depends on whether a is smaller or bigger than 1. In the case of
conjugate complex roots the presence of R b2 DE determines stepped fluctuation. If R  1 ,
the fluctuation would be narrowed down, while for R t 1 we would have explosive growth. These
conclusions are summarized in the table below

Case Subcase Time path of Yt Dynamic

stability
1. Distinct real roots
4D 0  a1  a2  1 Nonoscillatory Convergence
E!
(1  D ) 2

4D 1  a1  a2 Nonoscillatory Divergence
E!
(1  D ) 2
2. Single real root
4D 0  a 1 Nonoscillatory Convergence
E
(1  D ) 2
4D a !1 Nonoscillatory Divergence
E
(1  D ) 2
3. Complex roots
4D R 1 Stepped Convergence
E fluctuation
(1  D ) 2
4D R t1 Stepped Divergence
E fluctuation
(1  D ) 2

Table 2

In conclusion, the time path of national income is convergent only if DE  1 in all cases. Furthermore,
the model shows that it is possible for national income to have cyclical fluctuations endogenously
without any external shocks present, but merely due to the interactive play between the multiplier and
the accelerator process.