FRM Exam Part II · Portfolio Construction

Black-Litterman Model Explained for FRM Part II

Updated 11 October 2026 · Fact-checked

The Black-Litterman model starts from equilibrium returns implied by market-cap weights (Π = δΣw), then blends in investor views using Bayesian updating. Each view is weighted by its confidence. The result is a posterior return vector that gives more stable, diversified weights than mean-variance optimization on raw forecasts.

Understand Black-Litterman Model and Bayesian Approaches

Mean-variance optimization is very sensitive to its inputs. A small change in expected returns can flip the weights, and the output often shows extreme long and short positions. This is called error maximization: the optimizer loads up on assets whose inputs are most overstated.

Black-Litterman fixes this by changing the starting point. Instead of raw forecasts, you begin with the implied equilibrium returns. These are the returns that would make the market portfolio optimal. You get them by running the optimizer backwards (reverse optimization): take market-cap weights, the covariance matrix and a risk-aversion coefficient, and solve for the returns. With no views, the model gives back the market weights.

Next you add views. A view can be absolute ("Asset A returns 8%") or relative ("A outperforms B by 2%"). Each view has an uncertainty, captured in the matrix Ω. A view with small uncertainty pulls the return estimate strongly towards the view. A view with large uncertainty barely moves it.

This is a Bayesian idea. The equilibrium returns are the prior. The views are new information. The output is the posterior, a precision-weighted average of the two. Only assets involved in views get tilted from equilibrium, so the weights move away from the market portfolio only where you have an opinion. This is also a form of shrinkage: noisy estimates are pulled towards a stable anchor. Shrinkage estimators for means or covariances follow the same logic, blending the sample estimate with a target in proportion to how reliable each is.

The parameter τ scales the uncertainty of the prior (the equilibrium returns). A small τ means you trust equilibrium more. A large τ means you let the views move the result more.

Key formulas to remember

Risk-aversion coefficient: δ = (E(R_m) − R_f) ÷ σ_m²; Market excess return divided by market variance. Use excess returns and the market portfolio's variance.
Implied equilibrium excess returns (reverse optimization): Π = δ × Σ × w_mkt; Σ is the covariance matrix, w_mkt the market-cap weights. Π is a vector of excess returns.
Unconstrained optimal weights: w* = (δΣ)⁻¹ × μ; With μ = Π you recover w_mkt. With no views, Black-Litterman returns the market portfolio.
Views: P × μ = Q + ε, with ε ~ N(0, Ω); P picks the assets in each view, Q is the view return, Ω is the view uncertainty. For a relative view, the entries in a row of P sum to zero.
Posterior expected returns (precision-weighted form): μ_BL = [(τΣ)⁻¹ + P'Ω⁻¹P]⁻¹ × [(τΣ)⁻¹Π + P'Ω⁻¹Q]; Prior and views weighted by their precision (inverse variance).
Posterior expected returns (update form): μ_BL = Π + τΣP' × (P τΣ P' + Ω)⁻¹ × (Q − PΠ); Equilibrium plus an adjustment based on how far the view is from the prior. For one asset with P = 1: μ_BL = Π + [τσ² ÷ (τσ² + Ω)] × (Q − Π).

How to solve Black-Litterman Model and Bayesian Approaches questions

Use this order for any Black-Litterman or Bayesian shrinkage question. Most questions test only one or two of these steps.

1Identify what is asked: implied returns, a posterior return, a weight change, or a conceptual comparison with mean-variance optimization.
2Check the inputs are excess returns and that market weights sum to 1. Note δ, τ, Σ, P, Q and Ω as given.
3For implied returns, compute Σ × w_mkt first, then multiply by δ. If δ is not given, compute it as market excess return ÷ market variance, where market variance = w'Σw.
4For a view, write down P and Q. Absolute view: a 1 on the asset. Relative view: +1 on the outperformer and −1 on the underperformer.
5Compute the weight on the view: τσ² ÷ (τσ² + Ω) in the single-asset case. A larger Ω means a smaller weight on the view.
6Compute the posterior: Π + (view weight) × (Q − Π). It always lies between the prior and the view in the single-view, single-asset case.
7Interpret: say which assets are tilted, in which direction, and that unaffected assets stay near equilibrium. Mention the stability benefit over raw mean-variance.

Quickest way: Prior-view weighted average

When to use it: Use when you have one view on one asset, or a question asks which option is the posterior return or how it changes when confidence changes.

Compute the prior variance τσ² and take Ω as the view variance.
Work out the view weight k = τσ² ÷ (τσ² + Ω). If τσ² = Ω, k = 0.5.
Posterior = Π + k × (Q − Π).
Sanity check: the answer must sit between Π and Q. Reject any option outside that range.
If a question only asks about direction, remember: lower Ω (more confidence) moves the posterior closer to Q.

Common mistakes in Black-Litterman Model and Bayesian Approaches

Treating a larger Ω as higher confidence in a view.
Ω sounds like a confidence level, but it is the variance of the view error.
Fix: Ω is uncertainty. Larger Ω means a weaker view and a posterior closer to equilibrium. Smaller Ω means a stronger view.
Using total returns instead of excess returns when reverse optimizing.
Candidates plug in the market's expected return directly.
Fix: Subtract the risk-free rate first. Π = δΣw_mkt gives excess returns. Add Rf back only if the question asks for total returns.
Saying Black-Litterman always produces the market portfolio.
The no-view result gets remembered as the general result.
Fix: Only with no views (or views equal to Π) do you get market weights. Views tilt the weights, and only the assets linked to the views move materially.
Writing a relative view with a row of P that does not sum to zero.
Candidates mix up absolute and relative views.
Fix: Absolute view: a single 1. Relative view: +1 and −1 (or weights summing to zero). Q is then the return difference.
Describing τ as the confidence in the views.
τ appears next to Ω in the formula.
Fix: τ scales the uncertainty of the prior equilibrium returns. A larger τ shifts weight to the views, but views also have their own Ω.
Claiming Bayesian shrinkage removes estimation error.
Shrinkage is described as a fix for unstable inputs.
Fix: It reduces the variance of the estimate by accepting some bias towards the prior or target. The error is lower, not zero, and the result depends on the quality of the prior.

Worked examples

Example 1

A market has two assets. Market-cap weights are 60% and 40%. Volatilities are 20% and 10%, with correlation 0.5. Using reverse optimization, with a market excess return that gives a market risk premium of 5.2%, what is the implied equilibrium excess return of asset 1? Options: A) 5.6% B) 7.0% C) 2.5% D) 11.2%

Show the solution

Variances: asset 1 = 0.20² = 0.04. Asset 2 = 0.10² = 0.01. Covariance = 0.5 × 0.20 × 0.10 = 0.01.
Σw for asset 1 = 0.04 × 0.6 + 0.01 × 0.4 = 0.024 + 0.004 = 0.028.
Σw for asset 2 = 0.01 × 0.6 + 0.01 × 0.4 = 0.006 + 0.004 = 0.010.
Market variance = w'Σw = 0.6 × 0.028 + 0.4 × 0.010 = 0.0168 + 0.004 = 0.0208.
δ = 0.052 ÷ 0.0208 = 2.5.
Π₁ = δ × (Σw)₁ = 2.5 × 0.028 = 0.07 = 7.0%. (Π₂ = 2.5 × 0.010 = 2.5%.)
Check: weighted Π = 0.6 × 7.0% + 0.4 × 2.5% = 5.2%, which matches the market premium.

Answer: B) 7.0%. Asset 1 has a higher implied return because it is more volatile and more correlated with the market portfolio.

Example 2

For one asset, the equilibrium excess return Π is 6%, volatility is 20% and τ = 0.05. An analyst states a view that the asset will return 10%, with view variance Ω = 0.002. What is the posterior expected excess return? Options: A) 6% B) 8% C) 10% D) 9%

Show the solution

Prior variance τσ² = 0.05 × 0.04 = 0.002.
With P = 1, view weight k = τσ² ÷ (τσ² + Ω) = 0.002 ÷ (0.002 + 0.002) = 0.5.
Posterior = Π + k × (Q − Π) = 6% + 0.5 × (10% − 6%) = 6% + 2% = 8%.
Check: the result lies between 6% and 10%. Prior and view have equal precision, so the midpoint makes sense.
If Ω were smaller, for instance 0.0005, k would be 0.002 ÷ 0.0025 = 0.8 and the posterior would be 9.2%.

Answer: B) 8%. The prior and the view carry equal precision, so the posterior is their simple average.

Exam tips

Reverse optimization is the most testable calculation: Σw first, then multiply by δ. Practise it with two assets until it takes under two minutes.
Expect conceptual comparisons with mean-variance optimization: stability, diversification, no extreme positions, market portfolio as the neutral starting point.
Remember direction rules: lower Ω means stronger view, higher τ means more weight on views, no views means market weights.
In scalar questions, check that the posterior sits between prior and view. This removes wrong options fast.
For Bayesian shrinkage, state the logic: posterior is a precision-weighted blend of a noisy sample estimate and a stable prior or target.

Practice questions from Portfolio Construction

Black-Litterman Model and Bayesian Approaches: frequently asked questions

How is Black-Litterman different from mean-variance optimization?

Mean-variance optimization takes expected return forecasts directly and often gives concentrated, unstable weights. Black-Litterman starts from equilibrium implied returns and adjusts them only where you have views. This gives more stable and diversified weights.

How do you calculate implied equilibrium returns in Black-Litterman?

Use Π = δΣw_mkt. Multiply the covariance matrix by market-cap weights, then multiply by the risk-aversion coefficient δ. δ equals the market excess return divided by market variance. The result is a vector of excess returns.

What does τ do in the Black-Litterman model?

τ scales the uncertainty of the equilibrium returns. A small τ means the prior is tightly held and views move the result less. A larger τ gives the views more influence, all else equal.

How does Bayesian shrinkage relate to Black-Litterman?

Both blend a noisy estimate with a stable anchor, weighted by reliability. In Black-Litterman the anchor is the equilibrium return and the new information is the view. Shrinkage estimators do the same for sample means or covariances, pulling them towards a target.