11-FinEcon-Time Series Page 289 Wednesday, February 4, 2004 12:58 PM
Financial Econometrics: Time Series Concepts, Representations, and Models 289
where the hi are coefficients and εt–i is a one-dimensional zero-mean
white-noise process. This is a causal time series as the present value of
the series depends only on the present and past values of the noise pro-
cess. A more general infinite moving-average representation would
involve a summation which extends from –∞to +∞. Because this repre-
sentation would not make sense from an economic point of view, we
will restrict ourselves only to causal time series.
A sufficient condition for the above series to be stationary is that the
coefficients hi are absolutely summable:
∞
∑
h^2 < ∞
i
i = 0
Also, in general it can be demonstrated that given any stationary pro-
cess xi, if the sequence of coefficients hi is absolutely summable, then the
process
∞
yi = ∑hixi
i = 1
is stationary.
The Lag Operator L
Let’s now simplify the notation by introducing the lag operator L. The
lag operator L is an operator that acts on an infinite series and produces
another infinite series shifted one place to the left. In other words, the
lag operator replaces every element of a series with the one delayed by
one time lag:
Lx()t = xt – 1
The n-th power of the lag operator shifts a series by n places:
L
n
()xt =xtn–
Negative powers of the lag operator yield the forward operator F,
which shifts places to the right. The lag operator can be multiplied by a
scalar and different powers can be added. In this way, linear functions
of different powers of the lag operator can be formed as follows: