State Space Models

State Space Representation

A state space model is a dynamic system of equations

\[\begin{split}\begin{align} \boldsymbol{\xi}_{t+1} & = F\boldsymbol{\xi}_{t} + \boldsymbol{v}_{t+1} \\ \boldsymbol{Y}_{t} & = A^{'}\boldsymbol{x}_{t} + H^{'}\boldsymbol{\xi}_{t} + \boldsymbol{w}_{t} \\ E[\boldsymbol{v}_{t}\boldsymbol{v}_{\tau}^{'}] & = \bigg\{\begin{array}{c} Q \hspace{10pt} t = \tau \\ \boldsymbol{0} \hspace{10pt} \text{o/w} \\ \end{array} \\ E[\boldsymbol{w}_{t}\boldsymbol{w}_{\tau}^{'}] & = \bigg\{\begin{array}{c} R \hspace{10pt} t = \tau \\ \boldsymbol{0} \hspace{10pt} \text{o/w} \\ \end{array} \\ E[\boldsymbol{v}_{t}\boldsymbol{w}_{\tau}^{'}] & = \boldsymbol{0} \hspace{4pt} \forall \hspace{4pt} t,\tau. \end{align}\end{split}\]

State Space Representation

  • \(\smash{\boldsymbol{Y}_{t}}\) is a vector of \(\smash{n}\) variables observed at \(\smash{t}\).
  • \(\smash{\boldsymbol{\xi}_{t}}\) is a vector of \(\smash{r}\) unobserved variables at \(\smash{t}\).
  • \(\smash{\boldsymbol{x}_{t}}\) is a vector of exogenous or predetermined variables at \(\smash{t}\).
  • The first equation of the system is the state equation.
  • The second equation of the system is the observation equation.
  • \(\smash{\boldsymbol{v}_{t}}\) and \(\smash{\boldsymbol{w}_{t}}\) are vector WN processes and mutually uncorrelated at all lags.

Mutually Uncorrelated Errors

If we assume \(\smash{E[\boldsymbol{v}_{t}\boldsymbol{\xi}_{1}^{'}] = E[\boldsymbol{w}_{t}\boldsymbol{\xi}_{1}^{'}] = \boldsymbol{0} } \,\, \forall t > 1\):

\[\begin{split}\begin{align} E[\boldsymbol{v}_{t} \boldsymbol{\xi}_{\tau}^{'}] & = E[\boldsymbol{v}_{t}(\boldsymbol{v}_{\tau}^{'} + \boldsymbol{v}_{\tau - 1}^{'}F^{'} + \ldots + \boldsymbol{\xi}_{1}^{'}F^{\tau-1 \hspace{2pt} '})] = \boldsymbol{0} \hspace{4pt} \forall \hspace{4pt} \tau < t \\ E[\boldsymbol{v}_{t} \boldsymbol{Y}_{\tau}^{'}] & = E[\boldsymbol{v}_{t}(A^{'} \boldsymbol{x}_{\tau} + H^{'}\boldsymbol{\xi}_{\tau} + \boldsymbol{w}_{\tau})^{'}] = \boldsymbol{0} \hspace{4pt} \forall \hspace{4pt} \tau < t. \end{align}\end{split}\]

Similarly,

\[\begin{split}\begin{align} E[\boldsymbol{w}_{t} \boldsymbol{\xi}_{\tau}^{'}] & = \boldsymbol{0} \hspace{4pt} \forall \hspace{4pt} \tau < t \\ E[\boldsymbol{w}_{t} \boldsymbol{Y}_{\tau}^{'}] & = \boldsymbol{0} \hspace{4pt} \forall \hspace{4pt} \tau < t. \end{align}\end{split}\]

Example \(\smash{AR(p)}\)

The standard form of an \(\smash{AR(p)}\):

\[\smash{Y_{t+1} - \mu = \phi_{1}(Y_{t} - \mu) + \phi_{2}(Y_{t-1} - \mu) + \ldots + \phi_{p}(Y_{t-p+1} - \mu) + \varepsilon_{t+1}}.\]

Define:

\[\begin{split}\smash{\boldsymbol{\xi}_{t} = \left[\begin{array}{c} Y_{t} - \mu \\ Y_{t-1} - \mu \\ \vdots \\ Y_{t-p+1} - \mu \\ \end{array} \right], \,\,\,\,\, F = \left[\begin{array}{ccccc} \phi_{1} & \phi_{2} & \ldots & \phi_{p-1} & \phi_{p} \\ 1 & 0 & \ldots & 0 & 0 \\ \vdots & \vdots & \ddots & \vdots & \vdots \\ 0 & 0 & \ldots & 1 & 0 \\ \end{array} \right], \,\,\,\,\, \boldsymbol{v}_{t} = \left[\begin{array}{c} \varepsilon_{t} \\ 0 \\ 0 \\ \vdots \\ 0 \\ \end{array} \right],}\end{split}\]
\[\,\,\]

Example \(\smash{AR(p)}\)

\[\begin{split}\begin{gather} \boldsymbol{Y}_{t} = Y_{t}, \hspace{10pt} A^{'} = \mu, \hspace{10pt} \boldsymbol{x}_{t} = 1, \hspace{10pt} H^{'} = [1 \hspace{3pt} 0 \hspace{3pt} \ldots \hspace{3pt} 0], \hspace{10pt} \boldsymbol{w}_{t} = 0, \hspace{10pt} R =0, \\ Q = \left[\begin{array}{cccc} \sigma^{2} & 0& \ldots & 0 \\ 0 & 0 & \ldots & 0 \\ \vdots & \vdots & \ddots & \vdots \\ 0 & 0 & \ldots & 0\\ \end{array} \right]. \end{gather}\end{split}\]

Example \(\smash{ARMA(p,q)}\)

The standard form of an \(\smash{ARMA(p,q)}\):

\[\begin{split}\begin{align} Y_{t+1} - \mu & = \phi_{1}(Y_{t}- \mu) + \ldots + \phi_{r}(Y_{t-r+1}- \mu) \\ & \hspace{2in} + \varepsilon_{t+1} + \theta_{1}\varepsilon_{t} + \ldots + \theta_{r-1}\varepsilon_{t-r+2} \end{align}\end{split}\]

where \(\smash{r = \text{max}\{p,q+1\}}\) and \(\smash{\phi_{j} = 0}\) for \(\smash{j>p}\) and \(\smash{\theta_{j} = 0}\) for \(\smash{j > q}\).

Example \(\smash{ARMA(p,q)}\)

Define:

\[\begin{split}\boldsymbol{\xi}_{t} = \left[\begin{array}{c} Y_{t}- \mu \\ \phi_{2}(Y_{t-1}-\mu) + \ldots + \phi_{r}(Y_{t-r+1} - \mu) + \theta_{1}\varepsilon_{t} + \ldots + \theta_{r-1}\varepsilon_{t-r+2} \\ \phi_{3}(Y_{t-1}-\mu) + \ldots + \phi_{r}(Y_{t-r+2} - \mu) + \theta_{2}\varepsilon_{t} + \ldots + \theta_{r-1}\varepsilon_{t-r+3} \\ \vdots \\ \phi_{r}(Y_{t-1} - \mu) + \theta_{r-1}\varepsilon_{t} \\ \end{array} \right],\end{split}\]
\[\,\,\]

Example \(\smash{ARMA(p,q)}\)

\[\begin{split}\begin{gather} F = \left[\begin{array}{ccccc} \phi_{1} & 1 & 0 & \ldots & 0 \\ \phi_{2} & 0 & 1 & \ldots & 0 \\ \vdots & \vdots & \ddots & \ddots & \vdots \\ \phi_{r-1} & 0 & 0 & \ldots & 1 \\ \phi_{r} & 0 & 0 & \ldots & 0 \\ \end{array} \right], \,\,\,\,\, \boldsymbol{v}_{t} = \left[\begin{array}{c} \varepsilon_{t} \\ \theta_{1}\varepsilon_{t} \\ \vdots \\ \theta_{r-2}\varepsilon_{t} \\ \theta_{r-1}\varepsilon_{t} \\ \end{array}\right], \\ \boldsymbol{Y}_{t} = Y_{t}, \hspace{10pt} A^{'} = \mu, \hspace{10pt} \boldsymbol{x}_{t} = 1, \hspace{10pt} H^{'} = [1 \hspace{3pt} 0 \hspace{3pt} \ldots \hspace{3pt} 0], \hspace{10pt} \boldsymbol{w}_{t} = 0, \hspace{10pt} R = 0. \end{gather}\end{split}\]

Example \(\smash{ARMA(p,q)}\)

Alternatively, define \(\smash{\boldsymbol{\xi}_t = (\xi_t, \xi_{t-1}, \ldots, \xi_{t-r+1})^{'}}\) and

\[\begin{split}\begin{gather} F = \left[\begin{array}{ccccc} \phi_1 & \phi_2 & \ldots & \phi_{r-1} & \phi_r \\ 1 & 0 & \ldots & 0 & 0 \\ 0 & 1 & \ldots & 0 & 0 \\ \vdots & \vdots & \ddots & \vdots & \vdots \\ 0 & 0 & \ldots & 1 & 0 \end{array} \right], \,\,\,\,\, \boldsymbol{v}_t = \left[\begin{array}{c} \varepsilon_t \\ 0 \\ 0 \\ \vdots \\ 0 \end{array} \right], \\ \boldsymbol{Y}_{t} = Y_{t}, \hspace{10pt} A^{'} = \mu, \hspace{10pt} \boldsymbol{x}_{t} = 1, \hspace{10pt} H^{'} = [1 \hspace{3pt} \theta_1 \hspace{3pt} \ldots \hspace{3pt} \theta_{r-1}], \hspace{10pt} \boldsymbol{w}_{t} = 0, \hspace{10pt} R =0, \\ Q = \left[\begin{array}{cccc} \sigma^{2} & 0& \ldots & 0 \\ 0 & 0 & \ldots & 0 \\ \vdots & \vdots & \ddots & \vdots \\ 0 & 0 & \ldots & 0\\ \end{array} \right]. \end{gather}\end{split}\]

Example \(\smash{ARMA(p,q)}\)

Then, by the state equation

\[\begin{split}\begin{gather} \boldsymbol{\xi}_t = F \boldsymbol{\xi}_{t-1} + \boldsymbol{v}_t \\ \implies \phi_r(L) \xi_t = \varepsilon_t, \end{gather}\end{split}\]

and the observation equation

\[\begin{split}\begin{gather} \boldsymbol{Y}_t = A^{'} \boldsymbol{x}_t + H^{'} \boldsymbol{\xi}_t + \boldsymbol{w}_t \\ \implies (Y_t - \mu) = \theta_r(L) \xi_t. \end{gather}\end{split}\]

Example \(\smash{ARMA(p,q)}\)

Combining these two equations,

\[\begin{split}\begin{align} \phi_r(L) (Y_t-\mu) & = \theta_r(L) \phi_r(L) \xi_t \\ & = \theta_r(L) \varepsilon_t, \end{align}\end{split}\]

which is the equation for an \(\smash{ARMA(p,q)}\).