Research lab · Portfolio construction
A Bayesian way to build a portfolio. Start from the returns the market portfolio implies, add your own views with a stated confidence, and read the expected returns and weights that come out.
Risk aversion sets the level of all equilibrium returns: the implied Sharpe ratio of the market is δ times its volatility. If the number above looks too high or too low for the asset mix, move δ.
Each row is one view. The weights define a portfolio. Weights that sum to 1 make an absolute view (that portfolio earns Q). Weights that sum to 0 make a relative view (the long leg beats the short leg by Q). The standard error says how wide your uncertainty around Q is: 0 means certain, a large number means the view is ignored. Under the He and Litterman option the standard error is not an input: the field is switched off and the view is used whatever it holds.
Weight on view is the share of the gap between equilibrium and the view that this view would close if it were the only one: PτΣP′ / (PτΣP′ + ω). With several correlated views the posterior can land elsewhere, because the views inform each other.
The weights are the unconstrained mean-variance optimum and are not forced to add to 100%. The balance is cash when positive and borrowing when negative. Short positions are allowed. The standard error columns show how uncertain the model is about each mean before and after the views.
The two curves are the best return available at each level of risk from the risky assets alone, fully invested, under equilibrium and under posterior returns. The straight line adds cash or borrowing under posterior returns. The square is the market portfolio at equilibrium returns. The circle is the Black-Litterman portfolio at posterior returns: it lies on the straight line.
Reverse optimisation. An investor with risk aversion δ who maximises w′μ − (δ/2)·w′Σw holds w = (δΣ)−1μ. Setting w to the market weights and solving for μ gives the equilibrium excess returns:
Π = δ Σ wmkt
Prior. The true mean is uncertain around equilibrium, with covariance proportional to the asset covariance. τ is small when the equilibrium is trusted.
μ ~ N(Π, τΣ)
Views. K views on N assets. P is the K by N matrix of view portfolios, Q the K view returns, Ω the diagonal matrix of view error variances. The page takes a standard error per view and squares it. The He and Litterman option sets each variance equal to the prior variance of the view portfolio instead, in which case τ cancels out of the posterior mean.
Pμ = Q + ε, ε ~ N(0, Ω) ωk = sk2 or ωk = pkτΣpk′
Posterior. Bayes' rule for normal distributions gives the mean and the covariance of the mean. The page computes the first form, which stays valid when a view is certain (ω = 0). The second form is the same thing written as a precision-weighted average.
μBL = Π + τΣP′ (PτΣP′ + Ω)−1 (Q − PΠ)
M = τΣ − τΣP′ (PτΣP′ + Ω)−1 PτΣ
μBL = [(τΣ)−1 + P′Ω−1P]−1 [(τΣ)−1Π + P′Ω−1Q]
Weights. The same investor, now using posterior returns. The second equality shows why the result is stable: the portfolio is the market plus a multiple of each view portfolio, nothing else.
wBL = (δΣ)−1 μBL = wmkt + (τ/δ) P′ (PτΣP′ + Ω)−1 (Q − PΠ)
Choices made here. The optimisation uses Σ, so no views returns the market portfolio exactly. He and Litterman use the posterior predictive covariance Σ + M, which scales the no-view portfolio by 1/(1+τ). The difference is small for small τ. Volatility, active risk and the frontier in Exhibit 3 are all measured with Σ.
Limits. The output is only as good as Σ, which is taken as known. Returns are assumed normal. The equilibrium argument needs the market weights to describe the whole investable universe, and five blocks do not. There are no constraints: no long-only rule, no budget, no transaction costs, no limit on leverage. The calibration of τ and Ω has no agreed answer, and the weights are sensitive to both. The correlation matrix must be positive definite, and the page refuses to compute when it is not.
Sources: Black and Litterman (1992), Global Portfolio Optimization. He and Litterman (1999), The Intuition Behind Black-Litterman Model Portfolios. Idzorek (2005), A Step-by-Step Guide to the Black-Litterman Model.
Educational tool. Illustrative inputs, not live data. Not investment advice.