If \(R_{j,t}\) are the basically the scale standardized log returns
for \(j = 1,2,\dots,476\) of 476 stocks from S&P 500, as
from SP500, then \(mu_j = E[R_{j,*}]\) somehow
averaged over time; actually as predicted by muSigma() at the
end of the time period, and
\(\Sigma_{j,k} = Cov(R_j, R_k)\)
are estimated covariances.
These are the main “inputs” needed for the CLA algorithm, see
CLA.