1
$\begingroup$

Let us consider the following state-space model $$ z_{t} = x_{t} + v_{t}\\ x_{t} = \phi x_{t-1} + w_{t} $$ where $ \phi< 1$, the errors $v_{t}\sim \mathcal{N}(0,V^{2})$ and $w_{t}\sim \mathcal{N}(0,W^{2})$ are independent

The stationary variance of $x_{t}$ is given be $Var(x_{t}) = \frac{W^2}{1-\phi^2}$, therefore, the stationary variance of $z_{t}$ is given by $$ Var(z_{t}) = V^{2} + \frac{W^2}{1-\phi^2} $$

The state space model above is equivalent to ARMA(1,1) process $$ z_{t} = \phi z_{t-1} + \theta \varepsilon_{t-1} + \varepsilon_{t}, $$ where $\theta = - \phi \frac{V}{\sqrt{V^{2} + W^{2}}}$ and $\varepsilon_{t}\sim \mathcal{N}(0,V^{2} + W^{2})$ are i.i.d. This is actually AR(1) plus noise, which is equivalent to ARMA(1,1). The prof can be found, for example, here http://www.stats.ox.ac.uk/~reinert/time/notesht10short.pdf

Next, let us consider "equivalent" ARMA(1,1) model. Its stationary variance is given by $$ Var(z_{t}) = \frac{1+2\phi\theta +\theta^{2}}{1 - \phi^{2}}(V^2 + W^2), $$ where $\theta = - \phi \frac{V}{\sqrt{V^{2} + W^{2}}}$, see, for example, https://math.unice.fr/~frapetti/CorsoP/chapitre_23_IMEA_1.pdf

I can not see that the variance of equivalent model is equal to the state-space model.

This question is related to

Estimation of ARMA from state-space generated data

where it was shown by simulations, that "equivalence" is not working and

Alternative construction of ARMA(1,1) process

where the equivalence is proved.

$\endgroup$
  • 1
    $\begingroup$ "Any time series textbook" is not all that helpful. A concrete reference could be more helpful. $\endgroup$ – Richard Hardy Nov 13 '19 at 19:28
  • $\begingroup$ Dear @Richard Hardy, I have added the reference. $\endgroup$ – ABK Nov 13 '19 at 21:47
  • $\begingroup$ Thank you very much. $\endgroup$ – Richard Hardy Nov 14 '19 at 7:14
0
$\begingroup$

There is a mistake in the ARMA(1,1) model. Look at Alternative construction of ARMA(1,1) process where the correct ARMA(1,1) was derived.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.