I am trying to reduce correlation matrices into sub portfolios.
For example, I have a covariance matrix $\Sigma$ and weight-vector $w$ of two line items which I blend together into a sub-portfolio $\rho$, then use $$\sigma(\rho) = \sqrt{w'\Sigma w}$$ to get the sub-portfolio standard deviation. Then to find the correlation between a given asset $\alpha$ and the sub-portfolio, I am using $$\frac{cov(\alpha, \rho)}{\sigma(\alpha)\sigma(\rho)}$$ where $$cov(\alpha, \rho) = w_{\alpha}'\Sigma w$$ where $w_{\alpha}$ is a vector with a 1 in the entry corresponding to the row of asset $\alpha$. I've triple checked my code, and for some reason am continuing to get a correlation above 1. Not sure why this is but any help would be greatly appreciated.