A causal time series is a stochastic process 
that can be represented using only
present and past innovations. For an autoregressive
model this takes the form
 |
where 
and 
, with no dependence
on future innovations 
for 
. For an autoregressive model
with zero-mean, uncorrelated innovations of constant variance, the condition that
every root of the autoregressive
polynomial lie outside the unit circle gives the
unique causal stationary solution.
