= Finite-sample innovations of an MA(1) process
{title2=$\lambda_t=\gamma_1/\varphi_{t-1},\quad\varphi_t=\gamma_0-\gamma_1\lambda_t$}
Starting at time one, the orthogonal residuals satisfy $U_t=X_t-\lambda_tU_{t-1}$, with $\lambda_t=\gamma_1/\varphi_{t-1}$ and $\varphi_t=\gamma_0-\gamma_1\lambda_t$. Earlier innovation directions have zero <covariance> with the new observation. Explicitly, with $S_m=\sum_{j=0}^m\theta^{2j}$, $\varphi_t=\sigma^2S_t/S_{t-1}$ and $\lambda_t=\theta S_{t-2}/S_{t-1}$ for $t\ge2$. These formulas follow by induction from $S_t=(1+\theta^2)S_{t-1}-\theta^2S_{t-2}$, with $S_0=1$ and $S_1=1+\theta^2$. This includes $\theta=\pm1$, where $\varphi_t=\sigma^2(t+1)/t$.
Back to article page
