I'd never before bothered to work it out before, but the derivation of least squares regression is actually not that bad. The usual matrix formula, then, is the same idea but with much more bookkeeping for the multiple variables

linear algebra