A family of time indexed r.v. Z(w,t) where w belongs to a sample space and t belongs to an index set.
For a fixed t, Z(w,t) is a r.v. For a given w, Z(w,t) is a function of t, and is sample function or realization.
Strongly stationary
Consider a finite set of r.v. {Zt1,...,Ztn} from a stochastic process {Z(w,t):t=0,±1,±2,...}. The n-dimensional distribution function is defined by
The time series {Xt,t∈Z} is said to be stationary if
var(Xt)=z<∞ for all t∈Z
E(Xt)=m for all t∈Z
yX(r,s)=yX(t+r,t+s) for all t,r,s∈Z
the covariance being functions of the time difference alone.
Classical decomposition model
decompose a time series into
trend: loosely defined as "long-term change in the mean level"
seasonal variation: exhibit variation that is annual in period.
cyclic variation: exhibit variation at a fixed period due to some other cause.
irregular fluctuations
Steps to time series modeling
plot the time series and check for trend, seasonal and other cyclic components, any apparent sharp changes in behavior, as well as any outlying observations.
remove trend and seasonal components to get residuals
choose a model to fit residuals
forecasting can be carried out by forecasting residual and then inverting the transformation carried out in step 2.
For a random time series, ρ^k∼N(0,1/n) for k=0,n is the length of time series. Thus, if a time series is random, we can expect 19/20 of ρ^k to lie in [−2/n,2/n]
Test Zero cross-correlation
H0:∀h.ρx(h)=ρy(h)=0∧ρxy(h)=0 Both Xt,Yt contain no serial correlation and are mutually independent.