Assume that we have T measurements over the last T years of n various variables such as $X_1=$ GDP-growth, $X_2$=growth in private consumtion, ... . My question is whether it makes sense to estimate the Covariance-Matrix $Cov(X_1, ...., X_n)$ by calculating the Covariance-Matrix of the sample. Because is there even a "true" covariance Matrix to be estimated? I mean I dont really see why the values of BIP-growth can be modelled as being generated by a random process, so I would argue that the supposed covariance-matrix that ought to be estimated by the sample does not even exist. Is my argument correct or do I miss something?