This is a real mean-variance optimization — objective, covariance, constraints, and a constrained solve — executed live in your browser on a small universe. It's the same problem class PRISM runs on 100,004 real assets in ~2.16 s on GPU. Pick a dataset, set your risk appetite, and solve.
max wᵀμ − λ⁄2·wᵀΣw subject to budget, bounds, and direction — the canonical Markowitz program — by projected-gradient descent to convergence. Toy scale here; production scale on the benchmarks page.
Same universe, same risk-free rate. An illustrative 24-month simulation driven by each portfolio's expected return and volatility — it re-runs live every time you re-optimize.
Illustrative simulation from expected return/volatility on a small universe — not a backtest of historical prices. The point: PRISM's constrained, risk-aware portfolio targets a stronger risk-adjusted outcome than a concentration-prone conventional solve or a static equal-weight hold. Production PRISM runs this constrained, tax-aware optimization on 100,004 real assets in ~2.16 s on GPU.
Mean-variance on 11 names is a browser toy. Tax-aware, constrained optimization across 100,000 personalized accounts before the open is the product. See the measured evidence — or run it on your own book.
See the benchmarks →