Portfolio optimization: concepts, constraints, and practical steps
Optimization is only useful if it produces a portfolio you can actually hold. The real work is choosing inputs and adding the right constraints.
What optimization is (in plain English)
You choose an objective (e.g., maximize expected return for a given risk) and a set of constraints (e.g., weight caps). The optimizer finds weights that best satisfy both.
Common objectives
- Mean-variance: maximize return for given variance (classic “efficient frontier”)
- Min variance: focus on stability and diversification
- Risk parity: equalize risk contribution per asset/sector
- Target volatility: scale exposure to a chosen risk level
Constraints that matter in practice
- Long-only vs. allowing short positions
- Max weight per asset, sector caps, country caps
- Turnover limit (avoid over-trading)
- Liquidity: avoid weights that can’t be implemented
Inputs: where models usually fail
- Expected returns are noisy; use robust estimates or conservative priors.
- Correlations change during stress; test correlation spikes.
- Use multiple windows and stress scenarios, not a single backtest window.
Validation checklist
- Out-of-sample testing and walk-forward checks
- Transaction costs and slippage assumptions
- Max drawdown and downside risk metrics
- Concentration risk (top 1, top 5 weights)