Source: Black, F. & Litterman, R. (1992) · Financial Analysts Journal 48(5), 28–43 · DOI 10.2469/faj.v48.n5.28
TL;DR
Introduces the Black-Litterman model, a Bayesian fix for the fragility of mean-variance optimization. Instead of feeding in raw expected-return estimates (which produce extreme, unstable portfolios), it starts from the market-equilibrium implied returns (reverse-engineered from market-cap weights) as a prior, then blends in the investor's views with confidence weights — yielding stable, intuitive, well-diversified portfolios.
Problem it solves
Markowitz mean-variance optimization is hypersensitive to expected-return inputs: tiny changes in assumed returns produce wildly different, often extreme and corner-solution weights. Practitioners distrust the output. Black-Litterman tames this by anchoring the inputs to equilibrium rather than to noisy raw estimates.
The method
Reverse-optimize market-cap (equilibrium) weights to obtain the CAPM-implied expected returns — the neutral prior that, if used alone, reproduces the market portfolio.
Express views ("asset A will outperform B by x%," or absolute return views) as a set of linear combinations of assets, each with an attached uncertainty.
Combine the equilibrium prior and the views via Bayes' rule to produce blended (posterior) expected returns, then run standard mean-variance optimization on those.
The resulting portfolio tilts away from market weights only in the directions of the views, scaled by how confident those views are.
Assumptions & inputs
A prior covariance matrix of returns.
A risk-aversion scalar (used in the reverse optimization to back out implied returns).
A specification of views and the confidence (uncertainty) on each, plus a scalar weighting the overall confidence in the equilibrium prior.
How to use it
Use the global market portfolio (or a relevant benchmark) as the equilibrium starting point.
Add only the views you actually hold; assets without views stay near their market weights.
Calibrate view confidences deliberately — they, the risk-aversion scalar, and the prior covariance drive the output.
Limitations & pitfalls
Requires several judgment-call inputs (prior covariance, risk-aversion scalar, view confidences) that materially shape results.
Garbage views in, garbage portfolio out: it stabilizes optimization but does not create forecasting skill.
It mitigates, but does not eliminate, estimation error — a companion lesson to DeMiguel et al. on naive diversification.
Key references
Black, F. & Litterman, R. (1992) — Global Portfolio Optimization — Financial Analysts Journal
Markowitz, H. (1952) — Portfolio Selection — Journal of Finance
DeMiguel, V., Garlappi, L. & Uppal, R. (2009) — Optimal Versus Naive Diversification — Review of Financial Studies
Provenance: generated from the paper's abstract and metadata, not full text; sample periods and replication notes are indicative — verify against the source.
