Let's say I've coin where P(head on first toss) = P(tail on first toss) = 0.5, and P(head | previous toss was a head) = 0.8 and P(tail | previous toss was a tail) = 0.8.
Suppose I have a given realization of these coin tosses, say H, H, H, H, H, H, T, T, T. It is clear that there is serial correlation, since the first six tosses were heads and the next three tosses were tails. In other words, given that a toss was a head, I can predict whether or not the next toss was a head or tail better than chance.
If I were to randomly rearrange the order of the flips for a given realization, will there still be serial correlation (on average)?
My intuition is that there won't be serial correlation due to the randomness, but I'm not sure how to prove it.