The one-step-ahead AR(1) forecast-error ratio is always strictly positive

Let $\{X_t\}$ be a stationary AR(1) time series given by the recursive equation $X_t=\phi X_{t-1}+Z_t$, where $Z_t\sim WN(0,\sigma^2)$. Let $P_3X_4$ represent the forecast of $X_4$ given the observations $(X_1,X_2,X_3)$. If $C(\phi)$ is defined as $C(\phi)=\dfrac{E(X_4-P_3X_4)^2}{E(X_4-E(X_4))^2}$, then which of the following is not true?