Moving Average (MA) Models


Moving average models are expressed as depended of lagged white noise ($a_t$) values.

$$ X_t = \mu + a_t - \theta_1 a_{t-1} - \theta_2 a_{t-2} - ... - \theta_q a_{t-q} $$

The operator form (zero-mean) is

$$ X_t = \left( 1- \theta_1 B - ... - \theta_{q} B^q \right) a_t $$

Which provides the characteristic

$$ 1- \theta_1 z - ... - \theta_{q} z^q = 0 $$

Properties


  • An MA(q) is a finite GLP that is always stationary
  • $E[X_t] = \mu$
  • $\rho_0 = 1$
  • $\rho_k = 0$ for $k > q$

MA(1) Autocorrelations

The autocorrelations for MA(1) are as follows

$$ \rho_1 = \frac{-\theta_1}{1 + \theta_1^2} $$$$ \rho_k = 0 |_{k > 1} $$

Note: The max absolute value of $\rho_1 = 0.5$ for MA(1).

MA(2) Autocorrelations

The autocorrelations for MA(2) are as follows

$$ \rho_1 = \frac{-\theta_1 + \theta_1 \theta_2}{1 + \theta_1^2 + \theta_2^2} $$$$ \rho_1 = \frac{-\theta_2}{1 + \theta_1^2 + \theta_2^2} $$$$ \rho_k = 0 |_{k > 2} $$

Invertability


Due to model multiplicity in MA models, MA models are restricted to have all roots outside the unit circle. This is to make models identifible i.e. a singlar mapping between parameters and models.

More details

References


  • [1] W. Woodward and B. Salder, "Moving Average (MA(q)) and ARMA(p,q) Models", SMU, 2019