Black-Litterman: incorporating your own views cleanly

Advanced8 min read · updated 4 September 2026 · 6 sources

The Black-Litterman model solves the biggest problem of classic optimisation: implausible weights from uncertain return estimates. Instead of starting from historical returns, it starts from the returns the market prices in through its weights, and blends in your own views only as strongly as you trust them. Fischer Black and Robert Litterman developed it at Goldman Sachs around 1990.

Why a new starting point?

Markowitz optimisation needs an expected return for every asset. If you use historical averages, you often get extreme weights: everything in the asset that happened to do well recently, and nothing, or even short sales, in the rest. Michaud (1989) called this error maximisation (see efficient frontier). Black and Litterman (1992) turned the question around: which expected returns would have to hold for today's market weights to be optimal? These implied returns are a neutral starting point: without any views of your own, the model leads exactly to the market portfolio.

Step 1: implied returns

The formula is the reverse of the optimisation: an investor with risk aversion δ chooses the weights w = (δΣ)⁻¹ × μ. If you insert the market weights and solve for μ, you get Π. Equities receive a higher implied return here because they contribute more to the portfolio's risk. The market only holds them if they pay for it.

Where does δ come from? From the excess return and the variance of the market portfolio: δ = (expected market return − risk-free rate) / market variance. If you expect a 4% excess return at 13% volatility for a global portfolio, that gives 0.04 / 0.13² = 2.37. Values between 2 and 3 are common; a higher δ raises all implied returns in the same proportion but does not change the weights in the absence of views.

Step 2: your own view, with a confidence level

Suppose you expect an excess return of 5% for equities instead of 3.92% and are 50% sure of it. Black-Litterman blends the two values according to their uncertainty. If the market and the view are equally uncertain, the equity return ends up in the middle:

Formally, the new “posterior” returns are:

With a single view and a confidence c following Idzorek (2007), the formula becomes intuitive: μ_BL = Π + c × Σpᵀ × (Q − pΠ) / (pΣpᵀ). For equities, that means 3.92 + 0.5 × (5.00 − 3.92) = 4.46%. For bonds, the view acts through the covariance: 0.37 + 0.5 × 0.03125 × 1.08 = 0.39%. The factor 0.03125 is the covariance of bonds and equities divided by the equity variance.

Step 3: new weights

The posterior returns go through the same optimisation as the implied ones. Because the starting returns produce exactly the market weights, the weights shift only in the direction of the views, and only as far as the confidence allows.

The relative view in the last example is: equities return 4 percentage points more than bonds. The market prices in only 3.92 − 0.37 = 3.55 percentage points. With 50% confidence, the expected excess returns rise to 4.13% for equities and fall to 0.36% for bonds. The gap widens to around 3.8 percentage points. Relative views are often more natural than absolute ones: “Europe beats the US by 2 percentage points” is easier to justify than a precise return for Europe.

Advanced: Bayes, τ and Ω

Black-Litterman is a Bayesian calculation. The implied returns are the prior distribution with mean Π and covariance τΣ; the views are observations with error covariance Ω. The result is the weighted average of the two, weighted by their precisions. Satchell and Scowcroft (2000) derive this formally and present variants; He and Litterman (1999) use examples to explain how each view shifts the weights only in the assets concerned and their offsetting positions.

The most contentious parameter is τ: how uncertain are the implied returns? The literature cites values from 0.01 to 1. One way out is the method of Idzorek (2007): you state a confidence level in percent for each view and set the uncertainty so that it matches that confidence:

Because ω_k is then proportional to τ, τ cancels out of the formula for the posterior returns. Confidence alone controls how far the returns move. At 100% confidence, the model adopts the view exactly (5.00% for equities in the example); at 50%, the result lies in the middle; at low confidence, it stays close to the market.

Besides the mean, the uncertainty changes as well: the posterior covariance is Σ + τΣ − τΣPᵀ(PτΣPᵀ + Ω)⁻¹PτΣ. Some implementations optimise with this matrix; without views, they then do not arrive exactly at the market weights but at somewhat more cautious ones. Litterman (2003) describes the model as part of a broader equilibrium approach, together with risk budgets.

In practice: formulating good views

1. Few views: two or three well-founded statements are enough. Each additional one shifts weights, even if it is only half thought through.
2. Reasoning before numbers: a view should follow from an argument, such as a low valuation measured by the Shiller CAPE, or higher yields for bonds. The mere return of recent years is not an argument.
3. Relative rather than absolute: whether Europe will beat the US is often easier to judge than the precise return of a market.
4. Choose your confidence honestly: 100% means you would adopt the number without any discount. For market forecasts, 20 to 50% is usually more realistic.
5. Check the result: compare implied, historical and posterior returns side by side and test the weights in the backtest. If they look implausible, a view is usually too strong.

When Black-Litterman helps and when it does not

• You have a well-founded opinion on a few assets and want to leave the rest market-neutral.
• You want to use optimisation without historical accidents determining the weights.
• Without any views of your own, the model simply delivers the market weights. That is a sensible default.
• It is not helpful if the market weights themselves are unclear (commodities, alternative investments) or if views are derived only from past returns. Then the estimation error comes back in through the back door.
• The quality of the result depends on the quality of the views. The model only makes them consistent, not correct.

How Portfolio Lab calculates

In Optimisation, you choose a portfolio from your grid or equal weighting as the market weights. δ is fixed (default 2.5) or is estimated from the excess return and variance of the market portfolio and then capped to between 0.5 and 10. Views are absolute (expected total return p.a., from which the risk-free rate is deducted) or relative (outperformance of one asset over another), each with a confidence of 1 to 100% (default 50%); the uncertainty follows Idzorek's method with τ = 0.05. The weights maximise w × μ_BL − δ/2 × wᵀΣw subject to the budget and your weight limits, using the estimated covariance matrix (sample or Ledoit-Wolf). Without views, this produces exactly the market weights. Implied, historical and posterior returns are shown side by side, and the result can be checked directly in the backtest. More in the optimisation methodology.

This article explains a method; it is not investment or tax advice. Worked examples are simplified and rounded. All terms in the glossary.