Learn more about causal portfolio construction and how it is different to traditional approaches.
0:00: Intro & multiple uses of portfolio construction
0:54 Why we need something better than traditional approaches (Random Matrix Theory, Concept Drift, Causality is not correlation)
4:30 Demonstration of mean-variance overfitting to in-sample correlations using artificial data
10:26 Demonstration of mean-variance overfitting to in-sample correlations using S&P500 constituents
12:01 Spectral decomposition, eigenvalues and eigenvectors
14:05 Mean-variance and hierarchical clustering techniques: similarities and differences
20:57 Mean-variance replacing the correlation matrix with a causal graph
25:30 Practical benefits: less overfitting, more robust, less turnover,
28:25 Hierarchical clustering replacing the correlation matrix with a causal graph
29:25 Practical benefits: finding the true causal drivers are more stable
31:34: Predicting future correlations is hard. Building portfolios on correlations is sub-optimal. Causal approaches deliver an alternative.
