# How can a correlation structure for a stochastic process not correspond to a Toeplitz matrix?

I was reading the following question: A stochastic process that is 1st and 2nd order (strictly) stationary, but not 3rd order stationary

The following matrix is given representing the correlation structure of a stochastic process $$X[t], X[t+1], X[t+2]$$:

$$\left[\begin{array}{cc} \sigma^2 & a & b \newline a & \sigma^2 & c \newline b & c& \sigma^2 \end{array}\right]$$

I noticed that the correlation structure matrix is not a Toeplitz matrix. However, this seems to result in a contradiction: if you take $$t=k$$ then you find that the correlation between $$X(k+1)$$ and $$X(k+2)$$ is $$c$$, whereas if you take $$t=k+1$$, then you find that the correlation between $$X(k+1)$$ and $$X(k+2)$$ is $$a$$. This seems to mean that the matrix is not a valid correlation structure unless it is a Toeplitz matrix.

Is my reasoning valid?

This is not a stationary process, so there is no requirement for $$X(t)$$ and $$X(t+1)$$ to have the same correlation as $$X(t+1)$$ and $$X(t+2)$$.
• "First order stationary" just says the distributions of the individual $X(t)$ are all the same. For example, they might be normal with mean $0$ and variance $\sigma^2$. Robby didn't mean to have the covariance of $X(t)$ and $X(t+1)$ be $a$ for all $t$, just for some particular $t$. You might, for example, have this alternate between $a$ and $c$ depending on whether $t$ is odd or even. – Robert Israel Feb 12 '19 at 17:05
• Ah ok. I didn't realise that the variable $t$ was meant to refer to only one time. In that case, it doesn't seem like a proof that first order stationary processes need not be secondary in Robby's question when he says that the matrix is not a Toeplitz matrix. It doesn't show how the first order statistics can be the same but the second order ones can differ... – Requiens Feb 13 '19 at 2:23