Abstract:
|
Estimations in Vector Auto-regressive (VAR) processes are extensively studied when the process is stationary. But, little is known when the transition matrix of the process is not stable. In this case, the VAR can not be stationary, and beyond this, the process has no asymptotic limit distribution.
Estimation of the transition matrix, observing the VAR, is still of interest in many applications. We present high probability results relating the sample size and estimation error for an extensive class of noise distributions. For example, presented results are valid for noise vectors with no moment generating function, as well as those with a support of zero Lebesgue-measure.
|