Given a multidimensional stochastic process $X_t$ which satisfies a stochastic differential equation of the form
$$dX_t = A(X_t-m)dt + SdB_t$$
with matrices $A,B$ and a vector $m$ (Ornstein-Uhlenbeck process). I'd like to estimate the eigenvalues of $A$ using a discrete sample. If I use an estimator for $A$ (MLE for example) and its eigenvalues I might need a lot of data. Also, a lot of research here is based on symmetric matrices or at least matrices with independent entries, which might not be true for estimators of $A$.
Is it possible to estimate the eigenvalues without estimating $A$? Which statistical properties would these estimators have?