Naive optimizers turn tiny return guesses into crazy, all-or-nothing portfolios. Black-Litterman fixes that: it starts from the returns the market itself implies - a neutral, sensible anchor - and nudges them only as far as your views and your confidence justify. The result is stable, intuitive weights.
Assets
Market weights (they'll be normalized), annualized volatility, and pairwise correlations.
Asset
Market wt %
Vol %
Your view (optional)
Express an absolute view: an asset's expected annual return, and how sure you are.
The model runs in two moves. First it reverse-engineers the market's expected returns - the returns that would make today's market-cap weights optimal for a given risk appetite (Π = δΣw). That's your neutral starting point, and with no views the optimal weights come right back to the market's. Then it treats your view as noisy evidence and does a Bayesian blend: the posterior returns move toward your view in proportion to your confidence, and only for the assets your view actually touches (plus whatever correlates with them). Feed those posterior returns into the optimizer and you get weights that tilt sensibly - not the wild swings a raw mean-variance optimizer would produce from the same view. Lower confidence, smaller tilt; higher confidence, bigger.
Absolute single-asset views, τ=0.05, view uncertainty scaled from the prior. A teaching implementation of Black & Litterman (1992) with verified matrix math - not investment advice. All computation is local; no data leaves your browser.