“Regression to the mean” versus serial correlation “Regression to the mean” says that higher pre-test values will have lower post-test values (and vice versa). This phenomenon will decrease the correlation between the pre- and post-values. However, pre- and post-test values are always serially correlated. How can this be?
 A: Its interesting to bring up the relationship between regression to the mean and autocorrelation, but you  miss something important about autocorrelation. For observations $t=2$, such as apre-test-post-test design, the lag-1 autocorrelation coefficient, $r_k=-0.5$, invariably. Nonetheless, there may be some relationship between the two in an abstract sense.
Edit
In response to your question, here is a simple way to think about it.
The formula for autocorrelation is
$$\frac{\sum_{t=L+1}^T(x_t-\bar{x})(x_{t-1}-\bar{x})}{\sum_{t=1}^T(x_t-\bar{x})^2}$$
Thus, in a pre-test-post-test design with number of observations $T=2$, the only meaningful lag is $L=1$, so the formula for autocorrelation becomes
$$\frac{(x_2-\bar{x})(x_1-\bar{x})}{(x_1-\bar{x})^2+(x_2-\bar{x})^2}$$
Convince yourself that this expands to
$$\frac{(x_1-\bar{x})(x_2-\bar{x})}{2\bar{x}^2-2x_1\bar{x}-2x_2\bar{x}+x_1^2+x_2^2}$$
Remember now that $\bar{x}$ is $(x_1+x_2)/2$. Therefore,
$$\frac{(x_1-\frac{x1+x2}{2})(x2-\frac{x1+x2}{2})}{2(\frac{x1+x2}{2})^2-2x_1(\frac{x1+x2}{2})-2x_2(\frac{x1+x2}{2})+x_1^2+x_2^2}=\frac{(\frac{x_1-x_2}{2})(\frac{x_2-x_1}{2})}{\frac{(x_1-x_2)^2}{2}}=\frac{x_1-x_2}{2(x_2-x_1}=-\frac{1}{2}$$
