We will not show the derivation here as it can be found in several of the references, but the
standard expression for the Black-Litterman posterior estimated mean and sampling variance is:
(10) Π̂ ∼N([(τ Σ)
−1Π+ P
T Ω
−1Q][(τ Σ)
−1
+ P
T Ω
−1
P]
−1
,[(τ Σ)
−1
+P
T Ω
−1
P]
−1
)
Posterior estimate of the mean returns
Π Prior estimate of the mean returns
Σ Known covariance matrix of return distribution about the unknown mean
P View selection matrix
Covariance of the estimated view mean returns about the actual view mean returns
Q Estimated mean returns for the views
Ω is a term similar to τΣ, representing the uncertainty of the estimated returns of the views. In
this reference model, Ω is not the variance of the distribution of returns of the views.
The discussion of formula (10) is easier in terms of the inverse of the covariance matrix, a term
known as precision in the Bayesian literature. We can summarize the posterior estimated mean in
formula (10) as the precision weighted average of the prior estimate and the view estimates. The
posterior precision is the sum of the prior and view precisions. Both these formulations match
our intuition as we expect the precision of our posterior estimate to be more than the precision of
either the prior or the views. Second, the mixed estimation process should make use of the
precision of the estimates in the weighting of the mixing, e.g. an imprecise estimate should have
less impact on the posterior than a precise estimate.
With a small modification to the covariance term we can rewrite (10) using (6) to be an
expression for the Black-Litterman posterior estimate of the mean and covariance of returns
around the mean.
(11) E(r)∼N ([( τΣ)
−1Π+P
T Ω
−1Q ][(τ Σ)
−1
+ P
T Ω
−1
P]
−1
,(Σ+[( τΣ)
−1
+P
T Ω
−1
P]
−1
))
The updated sampling variance of the mean estimate will be lower than either the prior or
conditional sampling variance of the mean estimate, indicating that the addition of more
information will reduce the uncertainty of the posterior estimates. In Bayesian terms, the
posterior estimate is more precise than either the prior or the view estimates.
The variance of the returns about the mean from formula (10) will never be less than the known
variance of returns about the mean, Σ. This matches our intuition about how the variance of
returns can change. Adding more information should reduce the uncertainty (increase the
precision) of the estimates, but cannot reduce the covariance beyond that limit. Given that there
is some uncertainty in the variance of the returns about the mean, then formula (10) provides a
better estimator of the variance of returns about our estimated mean than the known variance
about the mean from the equilibrium