It is unbiased for any sample size.

It is biased for small samples, with bias of order $O(n^{-1})$.

The variance is proportional to $S_y^2 + R^2 S_x^2 - 2R \rho S_x S_y$.

The variance is independent of the correlation $\rho$ between $x$ and $y$.