Stereoscopic Portfolio Optimization with Machine Learning Ensembles

Summary

The article presents Stereoscopic Portfolio Optimization (SPO), a framework that combines traditional mean-variance allocation with bottom-up analysis of asset-level market microstructure. It describes volatility as one possible input, while noting that liquidity and order arrival rates could also be used. The proposed process uses K-means clustering to identify stock groups, Gaussian mixture models to represent uncertain group membership, and random forests as part of an ensemble intended to reduce a portfolio risk objective. It also reviews how these machine-learning methods work.

The article compares equal-weighted, efficient-frontier, bottom-up, and SPO portfolio approaches for an intraday strategy, and discusses relative performance. Its central claim is that portfolio behavior depends on both component microstructures and their relationships. The account provided here does not include the full numerical results, so it does not support an independent assessment of performance or statistical significance. SPO is presented as a flexible framework rather than a fixed recipe, and its usefulness depends on the chosen data, models, and validation process.

Key ideas

• SPO combines top-down asset allocation with bottom-up machine-learning analysis of market microstructure.
• Market microstructure inputs can include volatility, liquidity, or order arrival rates.
• K-means assigns observations to discrete groups, while Gaussian mixture models estimate probabilities of group membership.
• Random forests combine decision trees and can reduce dependence on any single feature.
• The article frames portfolio risk minimization as an ensemble objective but provides limited performance detail in the supplied text.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.