Professional Documents
Culture Documents
: APPENDIX
Polynomials and
Difference Equations
A time-series model is one which postulates a relationship amongst a number of temporal sequences or time series. For example, we have the regression
model
(1)
where x(t) is an observable sequence indexed by the time subscript t and (t) is
an unobservable sequence of independently and identically distributed random
variables.
A more general model, which we shall call the general temporal regression model, is one which postulates a relationship comprising any number of
consecutive elements of x(t), y(t) and (t). Thus we have
(2)
p
X
i y(t i) =
i=0
k
X
i x(t i) +
i=0
q
X
i (t i)
i=0
Any of the sums in this expression can be infinite, but if the model is to be
viable, the sequences of coefficients {i }, {i } and {i } can depend on only a
limited number of parameters.
We are particularly interested in a number of specialisations of the model.
The model represented by the equation
(3)
y(t) = x(t i) +
q
X
i (t i)
i=0
is described asP
a regression model with a serially correlated disturbance sequence (t) =
i (t i). If this sum is finite, then (t) is called a moving
average process. The model represented by
(4)
p
X
i=0
y(t) =
i x(t i) + (t)
i=0
y(t) =
q
X
i (t i),
i=0
(7)
i y(t i) = (t),
i=0
p
X
i y(t i) =
i=0
q
X
i (t i)
i=0
Sx(t) =
x(t i).
i=0
(12)
The advantage which comes from defining polynomials in the lag operator
stems from the fact that they are isomorphic to the set of ordinary algebraic
polynomials. Thus we can rely upon what we know about ordinary polynomials
to treat problems concerning lag-operator polynomials.
Algebraic Polynomials
Consider the equation 0 + 1 z + 2 z 2 = 0. On dividing the equation by
2 , we can factorise it as (z 1 )(z 2 ) where 1 , 2 are the roots of the
equation that are given by the formula
(13)
p
12 42 0
.
22
= + i = (cos + i sin ) = ei ,
= i = (cos i sin ) = ei ,
where
(15)
(2 + 2 )
and
3
= tan1
Im
Re
+
2!
4!
6!
i3
i5
i7
i sin = i
+
+
3!
5!
7!
cos = 1
(16)
(17)
cos + i sin = 1 + i
i3
4
2
+
+
2!
3!
4!
= ei .
Likewise, by subtraction, we get
(18)
cos i sin = ei .
cos =
ei + ei
2
4
sin =
(20)
(21)
(22)
where some of the roots may be real and others may be complex. The complex
roots come in conjugate pairs, so that, if = + i is a complex root, then
there is a corresponding root = i such that the product (z)(z ) =
z 2 + 2z + (2 + 2 ) is real and quadratic. When we multiply the n factors
together, we obtain the expansion
X
XX
i z n1 +
i j z n2 (1)n 1 2 n .
(23)
0 = zn
i
n = 0
n
Y
(i )1 .
i=1
Y
z
.
1
= 0
i
P
We should also note that, if is a P
root of the equation
i z i = 0,
then = 1/ is a root of the equation
i z ni = 0. This follows since
5
(L)
x(t).
(L)
Q
Imagine that (z) = (1 z/i ). Then repeated applications of this basic
result enables us to write
2
n
1
+
+ +
.
(27)
(z) =
1 z/1
1 z/2
1 z/n
By adding the terms on the RHS, we find an expression with a numerator of
order n 1. By equating the terms of the numerator with the terms of (z),
we can find the values 1 , 2 , . . . , n .
Consider, for example,
(28)
3x
3x
=
1 + x 2x2
(1 x)(1 + 2x)
2
2
=
+
1 x 1 + 2x
1 (1 + 2x) + 2 (1 x)
.
=
(1 x)(1 + 2x)
6
3x = (21 2 )x + (1 + 2 ),
(30)
= 1 + z/ + (z/)2 +
(1 z/)
to converge when |z| 1, it is necessary and sufficient that || > 1.
Linear Difference Equations
An nth-order linear difference equation is a relationship amongst n + 1
consecutive elements of a sequence x(t) of the form
(31)
(L)x(t) = u(t),
(L)
1
j
t
= (0 + 1 L + + n Ln )
= 0
(33)
1
j
+ 1
1
j
t1
= (j )
1
j
+ + n
= (0 + 1 j + + n nj )
1
j
1
j
1
j
tn
Q
Alternatively, consider the factorisation (L) = 0 i (1 L/j ). Within this
product, we have the term 1 L/j and, clearly, since
L
1
j
1
j
t
=
1
j
1
j
t
= 0,
(34)
y(t; c) = c1
1
1
+ c2
1
2
+ + cn
1
n
t
,
where c1 , c2 , . . . , cn are the constants which are determined by the initial conditions.
In the case where two roots coincide at a value of , the equation (L)y(t)
= 0 has the solutions y1 (t) = (1/)t and y2 (t) = t(1/)t . To show
this, let
Q
us extract the term (1 L/)2 from the factorisation (L) = 0 i (1 L/j )
given under (25). Then, according to the previous argument, we have (1
L/)2 (1/)t = 0, but, also, we have
(35)
L
1
2 t
t
2L L2
1
1
+ 2 t
t
= 1
t
t
1
1
1
=t
+ (t 2)
= 0.
t 2(t 1)
= + i = (cos + i sin ) = ei ,
= i = (cos i sin ) = ei .
These will appear in a general solution of the difference equation of the form of
(37)
y(t) = ct + c ( )t .
This is a real-valued sequence; and, since a real term must equal its own conjugate, we require c and c to be conjugate numbers of the form
(38)
c = (cos + i sin ) = ei ,
c = (cos i sin ) = ei .
Thus we have
(39)
ct + c ( )t = ei (ei )t + ei (ei )t
n
o
= t ei(t) + ei(t)
= 2t cos(t ).
To analyse the final expression, consider first the factor cos(t ). This
is a displaced cosine wave. The value , which is a number of radians per unit
period, is called the angular velocity or the angular frequency of the wave. The
value f = /2 is its frequency in cycles per unit period. The duration of one
cycle, also called the period, is r = 2/.
The term is called the phase displacement of the cosine wave, and it
serves to shift the cosine function along the axis of t so that the peak occurs at
the value of t = / instead of at t = 0.