Hierarchical Risk Parity (HRP), developed by Marcos López de Prado in 2016, is a modern alternative to Mean Variance Optimization that uses machine learning clustering to build more robust portfolios. Unlike MVO, HRP does not require estimating expected returns — it only uses the covariance matrix, dramatically reducing forecast error.
The algorithm works in three steps: (1) build a hierarchical clustering tree of assets based on their correlation structure; (2) traverse the tree top-down to assign risk budgets to clusters; (3) within each cluster, allocate based on inverse variance (each asset gets a weight inversely proportional to its risk). This produces a naturally diversified portfolio that respects the correlation structure of assets.
HRP portfolios tend to be more stable out-of-sample than MVO portfolios. They avoid the 'error maximization' problem and rarely produce extreme concentration in a single asset. Research shows HRP outperforms MVO, equal-weight, and risk parity allocations in most real-world backtests.
On StressTest.pro, HRP optimization is available under the Optimizations tool. We recommend it as the default starting point for most users.