The power spectrum P,(u) is defined as the squared
amplitude spectrum
Poux dx».
The spectral representation of an ARMA process can easily
be drived by evaluating the z-transformation on the complex
unit circle |z| = 1 where z - exp(j2Iu) /10/. An estimate
of the power spectrum of an ARMA process is obtained from
q
ae ra
k=1
u) = S
jac +102
k=}
b, exp (-j2Iku) |?
P
arma
a, exp (-j2IHku) |?
k
3.3 Autocorrelation Function
The autocorrelation function R,(k) often provides the basis
for analysing stationary random processes. The power
spectrum and the autocorrelation function of a time series
x(t) are related by the Wiener-Khinchin theorem. It states
that the power spectrum P (u) is the Fourier transform of
the autocorrelation function R_(k)
Pu) = F { R,(k} }
The parameters of ARMA processes are connected with the
autocorrelation function by the Yule-Walker equations. For
an AR(p) process they are given by
- » a,
AR(p) : R. (k)- ii
"i a, Ra(-1) + oR for k = 0
p
R_(k-1) for k > 0
The extended Yule-Walker equations for ARMA processes are:
p
ARMA(p,q): R_(k) = « D a
|=1
q
1 R,(k-1) + E b, R,, (k-1)
1
where R,, (k) - E (e(t):x(t-k)) for k x Q0
R k) = 0 fork »-0
ex(
4. Parameter Estimation
A time series x(t) may be described as an ARMA process of
known order p and q
P q
x(t).- -:Z a, x(t-i1) tie(t) 4 I b,e(t-j)
iz] j=
In practical application it is a major problem to estimate
the parameters a, and b,. Due to the non-linear relations
of the unknown parameters b, with the unknown innovations
e(t), the parameter estimation of the ARMA and MA models
requires more effort than the AR model. For this reason the
- 430 -