Calculating covariance of squares of random variables

82 Views Asked by At

The title isn't quite representative of the whole problem but it is where I found myself stuck. For simplicity, I will post the entire problem and then explain where I ended up at.

Let $ \{Z_t\} $ be independent random variables with mean $0$ and variance $σ2$. Let $\{Y_t\}$ be a stationary sequence with a covariance function $γ_Y (h)$. Assume also that the sequences $ \{Z_t\} $ and $\{Y_t\}$ are independent from each other. Define $X_t = Y_tZ_t.$ Verify that for $h ≥ 1$ we have $Cov(X_t, X_{t+h}) = 0 $ and $Cov(X_t^2 , X_{t+h}^2) \neq 0 $ (at least not for all t and h).

The first result was quite easy to find just by writing down the definition of covariance. However, for the second result I get that $Cov(X_t^2 , X_{t+h}^2) = σ^2σ^2Cov(Y_t^2 , Y_{t+h}^2)$ by once again just writing out the definition. But I'm stuck at trying to show $Cov(Y_t^2 , Y_{t+h}^2) \neq 0 $ for all $h$. I think I can assume the cov function $γ_Y (h) \neq 0$ for atleast one $h$ as it is a function of $h$, but I'm not sure how to use this to prove the final result.

Any help would be greatly apreciated. If you need any additional info please ask.

1

There are 1 best solutions below

1
On

In the above conditions, the statement is false. Let us consider the stationary process $Y_t=Y_0$ for all $t$, where $$ \mathbb P(Y_0=1)=\mathbb P(Y_0=-1)=\frac12. $$ Then $\gamma_Y(h)=\text{Var}(Y_0)=1\neq 0$ and $$ \text{Cov}(Y_t^2 , Y_{t+h}^2) = \text{Cov}(1 , 1) = 0. $$