Short-term Sparse Portfolio Optimization Based on Alternating Direction Method of Multipliers

Zhao-Rong Lai, Pei-Yi Yang, Liangda Fang, Xiaotian WuJMLR 2018 · OpenReview

TL;DR

This paper introduces SSPO, a novel short-term portfolio optimization system that leverages the Alternating Direction Method of Multipliers (ADMM) to construct sparse portfolios. By focusing on assets with high growth potential and proving the theoretical soundness of their ADMM approach, SSPO significantly outperforms existing methods in cumulative wealth, risk-adjusted returns, and robustness to transaction costs.

What Problem Does This Paper Address?

Traditional portfolio optimization (PO) often struggles with the dynamic and fast-paced nature of short-term trading. While many existing methods focus on long-term strategies or simply limit portfolio changes, they often fail to construct truly sparse portfolios for short-term horizons. This means they might invest in too many assets, diluting potential gains and increasing transaction costs. The challenge is to build a system that can efficiently identify and invest in a concentrated set of high-potential assets for short-term gains, while also adhering to practical constraints like self-financing and managing transaction costs.

Methods

The core of this work is the Short-term Sparse Portfolio Optimization (SSPO) system, built upon the Alternating Direction Method of Multipliers (ADMM).

  • SSPO Model: Instead of traditional risk minimization, SSPO aims to maximize the portfolio's increasing factor by focusing on assets with high recent price potential. This potential is estimated using the highest price in a recent window (PMAX) and then transformed into a generalized logarithmic return (Rt). The optimization problem incorporates an $\ell^1$-regularization term to enforce sparsity (investing in fewer assets) and a self-financing constraint (no external borrowing or lending).
  • ADMM Solver: The authors propose a novel ADMM algorithm to solve this complex optimization problem. A key contribution is their proof that the augmented Lagrangian of their ADMM formulation has a saddle point. This is crucial because the existence of a saddle point guarantees the convergence of the ADMM iterative steps, a theoretical aspect often overlooked in other sparsity-based methods.
  • Sparsity Mechanism: The $\ell^1$-regularization term directly encourages the portfolio weights to be zero for many assets, leading to a sparse portfolio. The algorithm is designed to actively rebalance this sparse portfolio based on changing market conditions.

Datasets

The authors evaluated their SSPO system on five benchmark datasets from real-world stock markets:

  • NYSE(O) (New York Stock Exchange - Original)
  • NYSE(N) (New York Stock Exchange - New)
  • DJIA (Dow Jones Industrial Average)
  • SP500 (Standard & Poor's 500)
  • TSE (Toronto Stock Exchange)

These datasets consist of daily price relatives and cover a wide range of assets and time spans, providing a robust testbed for evaluating portfolio optimization strategies.

Key Insights and Ideas

  • Focus on Growth Potential: The shift from traditional risk minimization to maximizing growth potential for short-term gains is a key conceptual departure. This aligns with empirical observations of investor behavior (e.g., overreaction to rising prices).
  • Theoretical Rigor for ADMM: Proving the existence of a saddle point for the augmented Lagrangian in their ADMM formulation is a significant theoretical contribution. This provides a strong foundation for the algorithm's convergence and reliability, which is often assumed rather than proven in similar works.
  • Active Sparse Portfolios: Unlike "lazy update" strategies that merely limit changes, SSPO actively rebalances a sparse portfolio, adapting to market dynamics and seeking aggressive growth opportunities.
  • Robustness to Transaction Costs: The system demonstrates strong performance even with non-trivial transaction costs, making it more practical for real-world trading.

Key Results

  • Superior Cumulative Wealth (CW): SSPO consistently achieved the highest cumulative wealth across all five benchmark datasets, significantly outperforming state-of-the-art systems like OLMAR and RMR. For example, on DJIA, SSPO achieved a CW of 3.68, compared to OLMAR's 2.54 and RMR's 2.67.
  • High Sparsity: The SSPO algorithm produced portfolios with average sparsities near or exceeding 90% on all datasets, meaning only a small fraction of assets held significant weights.
  • Strong Risk-Adjusted Performance: SSPO achieved the highest Sharpe Ratios (SR) on NYSE(N) and DJIA, and competitive SRs on other datasets. It also achieved the highest Information Ratios (IR) on NYSE(N) and demonstrated robust performance across various risk metrics.
  • Credible Returns: Order statistics analysis showed that SSPO's returns were not concentrated on the highest-growth assets, indicating that its performance is not due to simply picking "lucky" assets but rather a robust strategy.
  • Fast and Efficient: SSPO demonstrated fast running times (e.g., ~0.045s per trade on DJIA), making it suitable for high-frequency trading environments.
  • Transaction Cost Resilience: SSPO maintained strong performance even with transaction cost rates up to 0.5%, indicating its practical applicability.

Limitations and Future Work

  • Parameter Sensitivity: While the paper explores parameter settings, the performance of SSPO, like any optimization method, is dependent on choosing appropriate parameters (λ, γ, η, ζ, w). The paper provides guidance on suitable values based on empirical testing.
  • Future Work: The authors suggest continuing to establish more complex SSPO systems to further improve investing performance and robustness to risk. This could involve exploring different asset selection criteria, more sophisticated price prediction models, or incorporating other market factors.

Summary updated

Abstract

We propose a short-term sparse portfolio optimization (SSPO) system based on alternating direction method of multipliers (ADMM). Although some existing strategies have also exploited sparsity, they either constrain the quantity of the portfolio change or aim at the long-term portfolio optimization. Very few of them are dedicated to constructing sparse portfolios for the short-term portfolio optimization, which will be complemented by the proposed SSPO. SSPO concentrates wealth on a small proportion of assets that have good increasing potential according to some empirical financial principles, so as to maximize the cumulative wealth for the whole investment. We also propose a solving algorithm based on ADMM to handle the $\ell^1$-regularization term and the self-financing constraint simultaneously. As a significant improvement in the proposed ADMM, we have proven that its augmented Lagrangian has a saddle point, which is the foundation of the iterative formulae of ADMM but is seldom addressed by other sparsity strategies. Extensive experiments on $5$ benchmark data sets from real-world stock markets show that SSPO outperforms other state-of-the-art systems in thorough evaluations, withstands reasonable transaction costs and runs fast. Thus it is suitable for real-world financial environments.
BibTeX
@article{lai2018short,
  author = {Zhao-Rong Lai and Pei-Yi Yang and Liangda Fang and Xiaotian Wu},
  title = {{Short-term Sparse Portfolio Optimization Based on Alternating Direction Method of Multipliers}},
  journal = {J. Mach. Learn. Res.},
  year = {2018},
  url = {https://openreview.net/forum?id=xJBwDKm3EH6},
}
Figure