No. 01Portfolio construction

Portfolio Optimizer.

Five construction methods running on any universe of US tickers, with the efficient frontier rendered alongside your weights.

Run it in browserRead the methodology
Capital allocation lineVTIQQQNVDAMin varianceTangency portfolioσ 17.5% · μ 13.2%5%10%15%20%25%30%2%6%10%14%18%Annualized volatility →Annualized expected return →Markowitz, in browser
Fig. 01An efficient frontier for an AAPL · MSFT · NVDA · SPY universe, 2-year lookback. The red dot marks the mean-variance tangency portfolio; greyed circles are random sample portfolios that lie below it.
What it does

A frontier in your browser.

Type a handful of tickers. Pick a method. The optimizer pulls daily closes through our existing market-data layer, builds the sample covariance matrix, and solves — in milliseconds, in your browser — for the weights that minimize variance, equalize risk contributions, or maximize diversification.

The output is shown the way professionals read it: weights as ranked bars, expected return and annualized volatility as headline stats, and — for mean-variance — the full efficient frontier with your current portfolio plotted on it. Save runs to local storage and compare them later.

Methods

Five solvers.

  1. 01
    Mean-Variance
    min wᵀΣw

    Markowitz, closed-form. Returns the global minimum-variance portfolio plus the full efficient frontier.

  2. 02
    Risk Parity
    RC_i = 1/N

    Equal Risk Contribution via Spinu cyclical coordinate descent. Defaults to inverse-volatility seeding.

  3. 03
    HRP
    d_ij = √(½ (1-ρ_ij))

    Hierarchical Risk Parity (López de Prado, 2016). Avoids covariance inversion entirely.

  4. 04
    Black-Litterman
    μ̂ = [(τΣ)⁻¹ + PᵀΩ⁻¹P]⁻¹·…

    Bayesian posterior blending an equilibrium prior with explicit views. Pass any P and Q.

  5. 05
    Max Diversification
    max wᵀσ / √(wᵀΣw)

    Choueifaty & Coignard. Closed-form solution maximizing the diversification ratio.

  6. 06
    Equal Weight
    w_i = 1/N

    The 1/N baseline. Surprisingly hard to beat on realistic universes — included as a sanity check.

Read more

Frequently asked

Which optimization methods does it support?
Mean-variance (Markowitz minimum-variance and target-return), Equal-Risk-Contribution risk parity solved via Spinu cyclical-coordinate descent, Hierarchical Risk Parity using single-linkage clustering and recursive bisection, Black-Litterman with closed-form posterior weights, Maximum Diversification (Choueifaty & Coignard), and 1/N equal-weight as a baseline.
What are the constraints?
Long-only is on by default — negative weights are clipped to zero and the portfolio is renormalized. An optional per-asset cap iterates clip-and-renormalize until no weight exceeds the cap. Sector caps, turnover penalties, and a true quadratic-program solver are on the roadmap.
Where does the data come from?
Daily closing prices via the same provider stack that powers the rest of Quantle — Yahoo Finance with Tiingo IEX as the price-history fallback. Returns are inner-joined on date so every name in the universe contributes to every covariance entry.
How long a lookback should I use?
Defaults to two years (504 trading days). That tends to capture a full market cycle without over-weighting the distant past. The lookback chip picker offers 6 months, 1 year, 2 years, and 5 years; longer windows shrink covariance variance but blur regime changes.
Is the efficient frontier solved analytically?
Yes. The unconstrained problem has closed-form solutions for the global minimum-variance portfolio (w ∝ Σ⁻¹·1) and for any target-return portfolio via the two-fund separation theorem. The frontier shown on screen is sampled across 20 target returns spanning the min and max asset means.
How does HRP differ from mean-variance?
HRP avoids the inversion of the covariance matrix entirely. It clusters assets by distance derived from their correlations, then allocates risk inversely to cluster variance through recursive bisection. The result is far more stable under sample noise — the price you pay for that stability is no explicit return-targeting.
Try

Run the optimizer on your universe.

Free, in browser, no spreadsheet, no Bloomberg seat.

Open the workspace