arXiv:2108.02283v3 Announce Type: replace-cross Abstract: Classification outperforms regression across matched machine learning models in portfolio construction. A stacking ensemble of gradient boosted tree, random forest, and neural network yields a value-weighted annualized Sharpe ratio of 1.83 for classification and 1.11 for regression. This outperformance persists in multiclass settings, across subsamples, and after transaction costs. Spanning tests show that classification retains economically large alphas after we control for regression, whereas regression alphas shrink substantially onc
The proliferation of machine learning techniques in financial services is fostering continuous innovation in quantitative portfolio management strategies, including the refinement of loss functions for better performance.
This research suggests a fundamental improvement in applying machine learning to portfolio construction, potentially leading to significantly enhanced investment returns and risk management for institutions.
The explicit choice of a classification loss function rather than a regression loss function for portfolio construction models is now demonstrated to be a superior approach for achieving higher Sharpe ratios.
- · Quantitative FInancial Institutions
- · Hedge Funds
- · AI/ML Developers
- · Asset Managers
- · Traditional Regression-Based Investment Strategies
- · Less Sophisticated AI/ML Models
Financial institutions will likely re-evaluate and optimize their machine learning models for portfolio construction with a greater emphasis on classification approaches.
Increased adoption of these sophisticated models could intensify competition in quant finance and potentially alter market dynamics through more efficient capital allocation.
The demonstrated outperformance might accelerate the development of specialized hardware and software tailored for classification-based financial modeling, further embedding AI as a critical infrastructure in finance.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.LG