AR(1) Forecasting Error Variance and MLE
Problem statement
Consider the AR(1) process $Y_t = \phi Y_{t-1} + \varepsilon_t$, where $|\phi| < 1$ and $\varepsilon_t \overset{\text{i.i.d.}}{\sim} N(0, \sigma^2)$.
(1) Derive the $h$-step-ahead point forecast $\hat{Y}_{t+h|t} = E[Y_{t+h} \mid Y_t, Y_{t-1}, \ldots]$ and the corresponding forecast error variance $\text{Var}(Y_{t+h} - \hat{Y}_{t+h|t})$ as explicit functions of $\phi$, $\sigma^2$, and $h$.
(2) Given an observed sample $\{Y_t\}_{t=1}^{T}$, write down the MLEs $\hat{\phi}$ and $\hat{\sigma}^2$ (conditioning on $Y_1$). Then derive the asymptotic variance of $\hat{\phi}$ as $T \to \infty$.
Show hint
What happens if you substitute the recursion into itself repeatedly, and which terms are unknown at the time you forecast?
Relevant roles: Data Scientist, Applied Scientist and Research Scientist.