Uncorrelated Returns
Combining several independent return streams beats picking any single one. This builds the portfolio: cluster a multi-asset universe by correlation, then balance risk across the clusters.
6.65%
portfolio vol (vs 15.3% equal-wt)
1.83
diversification ratio
−16.5%
max drawdown (vs −26.2%)
12 → 6
assets clustered into themes
The recommended portfolio
Inverse-vol within each cluster, risk parity across clusters — so a calm bond sleeve and a wild crypto sleeve each contribute the same risk.
How it works
1 · Cluster
Correlation → distance √(½(1−ρ)) → hierarchical clustering, k chosen by silhouette.
2 · Weight within
Inverse-volatility weights inside each cluster.
3 · Balance across
Risk parity over the cluster streams (SLSQP), equalizing risk contributions.
4 · Evaluate
Train/test split; report vol, drawdown, Sharpe, diversification ratio.
Run it
pip install -e ".[dev]"
uncorrelated --plots docs # clusters, weights, performance + charts
uncorrelated --source live # use real market data (yfinance)