Understanding these results
Both portfolios above are derived entirely from past price data over the selected window. They show what would have been most risk-efficient historically — not a prediction of what will perform best going forward, and not investment advice.
Why the “optimal” sometimes looks counterintuitive: Markowitz optimization rewards recent past winners and penalizes recent losers — even if those losers may be sounder going forward. If an asset had a great year, the model overweights it. If it had a rough year, the model avoids it. Past returns don’t predict future returns. This is the fundamental limitation of optimizing purely on historical data.
What IS reliably useful: Volatility and correlation are more stable over time than raw return rankings. The risk breakdown, the correlation matrix, and the diversification structure are the genuinely valuable output — they show how your assets move together and where your risk is concentrated. Use those insights; treat the specific return numbers as one historical snapshot.
40% cap: We limit each asset to 40% across all 12,000 simulations to prevent the model from concentrating everything in one past winner. Unconstrained Markowitz commonly allocates 90%+ to a single asset.
