# Understanding portfolio-level risk models

I have a tremendous amount of experience training supervised machine learning models. However, I recently became a data scientist at a small financial services company, and I've been asked to build some sort of model that assesses portfolio-level risk. I'm not sure how to go about this task. Is this a supervised learning problem? It doesn't sound like one.

I used to work as a software/data engineer at a different financial services company, and while there I was briefly exposed to the concept of Basel credit risk management. Is Basel used for assessing portfolio-level risk? As far as I could tell (I could be wrong), Basel seemed to aggregate charge-off predictions by running every single client through each of the company's relevant credit risk models (which were trained in a supervised fashion). If my memory is right, how does this aggregation work? Is it effective?

To determine the portfolio risk therefore, you'd need to either predict, or use historicals for $\sigma_1$, $\sigma_2$, $Cov_{1,2}$