Doubly elastic net regularized online portfolio optimization with transaction costs.

Area of Science:

• Computational Finance
• Machine Learning
• Optimization Theory

Background:

• Online portfolio optimization presents significant challenges due to market volatility and transaction costs.
• Existing methods struggle with large-scale computations and dynamic cost structures.

Purpose of the Study:

• To develop an efficient machine learning system for online portfolio optimization.
• To address the complexities of transaction costs in dynamic financial markets.

Main Methods:

• Reformulated the optimization problem as a minimization over a simplex.
• Applied linearized augmented Lagrangian method (LALM) and alternating direction method of multipliers (ADMM).
• Derived closed-form solutions for subproblems and guaranteed convergence.

Main Results:

• The proposed algorithms demonstrated superior performance compared to state-of-the-art strategies.
• Experiments were conducted on five real-market benchmark datasets.
• The methods showed effectiveness across six key performance dimensions.

Conclusions:

• The developed machine learning system offers an efficient and effective solution for online portfolio optimization.
• LALM and ADMM provide a robust framework for handling transaction costs in large-scale financial computing.