Exact Maximum Likelihood Estimation of an ARMA(1, 1) Model with Incomplete Data (Q4544838)

Let \(\{X(t)\}\) denote a time series generated by the equation \[ X(t)=\theta X(t-1)+Z(t)-\theta Z(t-1), \] where \(\{Z(t)\}\) is Gaussian white noise. The author deals with the problem: how can one obtain exact likelihood estimates for parameters of the \(ARMA(1,1)\) model with nonconsecutively observed or missing data. For this purpose, the author obtains the Cholesky decomposition for the inverse of the covariance matrix without any assumptions, such as stationarity or invertibility of the process. This results leads to a closed form of the likelihood function. Then a simple method based on the Plackett equation is proposed for finding the exact maximum likelihood estimator of the parameters for the ARMA model in the stationary case.