Random forests reduce variance compared to a single decision tree by averaging B trees, each trained on a bootstrap sample with random feature subsets. If individual trees have variance σ2 and pairwise correlation ρ, what is the variance of the forest?