Interpretable Machine Learning for Diversified Portfolio Construction

In this article, the authors construct a pipeline to benchmark hierarchical risk parity (HRP) relative equal contribution (ERC) as examples of diversification strategies allocating liquid multi-asset futures markets with dynamic leverage (volatility target). The use interpretable machine learning concepts (explainable AI) compare robustness and back out implicit rules for decision-making. empirical dataset consists 17 equity index, government bond, commodity across 20 years. two are tested about 100,000 bootstrapped datasets. XGBoost is used regress Calmar ratio spread between against features Compared ERC, HRP shows higher ratios better matches volatility target. Using Shapley values, can be attributed especially univariate drawdown measures asset classes. TOPICS:Quantitative methods, statistical big data/machine learning, portfolio construction, performance measurement Key Findings ▪ introduce procedure rule-based investment explain differences in path-dependent risk-adjusted using learning. They apply allocations find have superior performance. return datasets SHAP framework by Lundberg Lee (2017) discuss local global feature importance.