HRP vs MVO Portfolio Construction
Comparing Hierarchical Risk Parity against Mean-Variance Optimization for cryptocurrency portfolios
Executive Summary
HRP Advantages
- + Lower volatility (23.9% vs 30.2%) - wins 75% of windows
- + Better drawdown protection (worst: 14.4% vs 16.8%)
- + More consistent returns (std: 31.9% vs 71.4%)
- + No matrix inversion - works with singular covariance
MVO Advantages
- + Higher absolute returns (115.3% vs 80.5% average)
- + Slightly higher Sharpe (3.67 vs 3.10)
- + Can capture outlier gains (231.8% in Window 2)
- + Maximum Sharpe portfolio option available
Performance Metrics Comparison
HRP Win Rates vs MVO
Returns Across Test Windows
Volatility Across Test Windows
Maximum Drawdown Comparison
Risk-Return Profile
Methodology
HRP Algorithm
- Tree Clustering: Ward's hierarchical clustering on correlation distance
- Quasi-Diagonalization: Reorder assets so correlated ones are adjacent
- Recursive Bisection: Allocate using inverse variance within clusters
MVO Algorithm
- Covariance Matrix: Build from correlation and volatilities
- Gradient Descent: Minimize portfolio variance
- Simplex Projection: Enforce weight constraints (sum=1, non-negative)
Test Configuration
- Windows: 4 rolling 6-month periods
- Training: 6 months for correlation/weight calculation
- Testing: 6 months out-of-sample performance
- Assets: Top 20 grid-suitable cryptocurrencies
Rating System Integration
- ETS Filter: Only grid-suitable assets (ETS >= 40)
- Weight Boost: Higher ETS = allocation boost within clusters
- Correlation Aware: HRP handles crypto's high correlation naturally
Implementation Recommendation
Conservative Profiles
Use HRP as default. Provides better risk management and more consistent returns.
Aggressive Profiles
Use MVO for return-focused users willing to accept higher volatility and drawdowns.
Hybrid Approach
Offer both methods in UI. Let users choose based on their risk tolerance and goals.
Related Research
* Note on this figure, added 2026-08-05: The 0.886 correlation (and the return/win-rate figures on this page) were measured under an earlier version of the rating formula (dated 2026-01-29). The formula has since been re-optimized three times; internal walk-forward testing as of 2026-05-13 indicates a higher correlation under the current formula. We are re-running this study under the live formula and will republish updated figures. The qualitative finding — that higher-rated assets have historically outperformed lower-rated ones — is not affected by this note; the exact numbers above should not be cited as describing today's system until the re-run is complete.