Abstract
The principal portfolio approach is an emerging method in signal-based trading. However, these principal portfolios may not be diversified to explore the key features of the prediction matrix or robust to different situations. To address this problem, we propose a novel linear trading position with sparse spectrum that can explore a larger spectral region of the prediction matrix. We also develop a Krasnosel’skiĭ-Mann fixed-point algorithm to optimize this trading position, which possesses the descent property and achieves a linear convergence rate in the objective value. This is a new theoretical result for this type of algorithms. Extensive experiments show that the proposed method achieves good and robust performance in various situations.
Introduction: The Rigidity of Principal Portfolios
In quantitative finance, "Principal Portfolios" (PP) are a popular way to exploit cross-predictability—the idea that signals from one asset can help predict the returns of another. Traditionally, traders use a "prediction matrix" to identify these PPs. However, current methods suffer from a significant "rigidity" problem. Most strategies either include a principal portfolio entirely (spectral energy of 1) or exclude it completely (spectral energy of 0). This binary choice often fails to capture the nuanced features of market data and lacks the diversification needed to remain robust across different economic states. Furthermore, the mathematical optimization of these positions is notoriously difficult because it involves a "three-part" complexity: the Frobenius norm, the nuclear norm for sparsity, and the spectral norm for self-financing constraints.
Methodology: Sparse Spectrum and Fixed-Point Optimization
We propose the LTPSS model, which treats the trading position as an optimization problem over a "sparse spectrum." Instead of picking the top $l$ portfolios, LTPSS allows every principal portfolio to have a weight (spectral energy) between 0 and 1. This flexibility allows the model to "dial in" the exact amount of influence each portfolio should have, leading to better diversification.
To solve this, we developed a Krasnosel’skiĭ-Mann (KM) fixed-point algorithm. The technical implementation involves:
- Proximal Mapping: Handling the non-differentiable nuclear norm to encourage a sparse representation, focusing the model on the most important signals.
- Non-expansive Projection: Ensuring the portfolio remains "self-financing" (spectral norm $\le 1$) through a custom projection operator.
- Theoretical Convergence: While most KM algorithms only guarantee slow convergence, we prove that for this specific trading problem, the algorithm achieves a linear convergence rate in the objective value—a significant theoretical contribution to the field of optimization.
Experimental Results
The researchers evaluated LTPSS against seven diverse benchmark datasets to test its "extendability" beyond standard US markets, including Fama-French portfolios, the MSCI global index, EURO Stoxx50, Chinese mutual funds (FOF), and the UK FTSE100.
LTPSS consistently outperformed baseline strategies (like LTP-PP and LTP-CF) across nearly all metrics:
- Sharpe Ratio (SR): LTPSS achieved the highest SR on all datasets. For example, on the FTSE100, it reached an SR of 0.1521, significantly higher than the standard LTP-PP (0.0114).
- Mean Return (MR): It delivered superior returns, particularly in non-US markets. On the FOF (China) dataset, LTPSS returned 0.0030, while the standard principal portfolio approach returned only 0.0012.
- Alpha ($\alpha$): The model showed a high ability to generate "pure" returns not explained by standard market factors. In the FTSE100 test, LTPSS generated an alpha of 0.0264, whereas the traditional PP approach yielded a negative alpha.
Conclusion
The LTPSS model demonstrates that flexibility in spectral energy is superior to rigid selection. By automatically adapting its spectrum to the specific characteristics of the dataset, LTPSS provides robust performance across global markets. Future research may extend this sparse spectrum logic to non-linear architectures or explore its relationship with other conventional financial factors like momentum and volatility.
Summary updated