Thanks CFAdreams. It makes sense now.
The unconstrained model is unstable and can result in big shifts/heavy allocations to, for example, commercial real estate (because of the high sharpe ratio byproduct of the smoothed volatility from stale / infrequent pricing). Resampling uses historical data so is more diversified but relies purely on the historical piece. Bl uses expectations, which is one of the biggest advantages.
The book says that Black-Litterman(BL) model reverse engineers the expected return…a reverse optimization. But the unconstrainted Black-Litterman(UBL) model is unlikely a reverse optimization process…it mentions something called Bayesian procedure. (V3, p270). – this looks unusual to me.
Anyone have a clearer view on this now? my understanding; BL Reverse engineers E® based on Market Portfolio and use of tweaked variances and covariances to solve for E® which are produced from a optimisation/solving process. We then use these E® and the Variance/Covariances we tweaked and run a normal Mean Variance Optimisation to determine the optimal asset allocation to arrive at the optimal portfolio that we should combine with the Risk Free depending on asset allocations. Unconstrained is not the focus of CFAI curriculum but allows for the non-negative weight constraints to be removed. sound right?