A Globally Optimal Portfolio for m-Sparse Sharpe Ratio Maximization

The Sharpe ratio is an important and widely-used risk-adjusted return in financial engineering. In modern portfolio management, one may require an m-sparse (no more than m active assets) portfolio to save managerial and financial costs. However, few existing methods can optimize the Sharpe ratio with the m-sparse constraint, due to the nonconvexity and the complexity of this constraint. We propose to convert the m-sparse fractional optimization problem into an equivalent m-sparse quadratic programming problem. The semi-algebraic property of the resulting objective function allows us to exploit the Kurdyka-Lojasiewicz property to develop an efficient Proximal Gradient Algorithm (PGA) that leads to a portfolio which achieves the globally optimal m-sparse Sharpe ratio under certain conditions. The convergence rates of PGA are also provided. To the best of our knowledge, this is the first proposal that achieves a globally optimal m-sparse Sharpe ratio with a theoretically-sound guarantee.

Paper

References (35)

Scroll for more · 23 remaining

Similar papers

Peer review

Reviewer pxYT5/10 · confidence 4/52024-07-11

Summary

This study addresses the problem of Sharpe ratio maximization under a cardinality constraint, referred to in the paper as an m-sparse constraint. Adding a cardinality constraint typically makes optimization problems NP-hard. Existing studies usually approach this problem using heuristic methods or relaxations. The former can be time-consuming and do not guarantee optimality, while the latter struggles to control cardinality accurately. This research proposes transforming the m-sparse fractional optimization problem into an equivalent m-sparse quadratic programming problem, which ensures convergence to a locally optimal solution and allows for more precise control of cardinality. However, as will be discussed in the question section, the study essentially transforms the problem into a mean-variance optimization form. This raises the question of why not simply perform m-sparse mean-variance optimization. Since Sharpe ratio maximization is just one of many Pareto optimal points in mean-variance optimization, addressing the more general problem could be more advantageous. Also, as this paper deals with practical issue (i.e., cardinality constraint), the authors should discuss more about practical implementation of the proposed method (e.g., computational time, adding more practical constraints).

Strengths

- The proposed method appears to be theoretically sound and effectively controls cardinality. - The approach has been tested on various datasets, and the experimental results indicate strong performance. - The writing structure is well-organized and the content is presented in a clear and readable manner.

Weaknesses

- Ultimately, this research focuses on a portfolio optimization problem with a cardinality constraint driven by practical needs, making practical implementation the most critical aspect. However, there is a lack of discussion or analysis on this aspect. - Practical implementation would require consideration of various additional constraints (e.g., lower or upper bound of portfolio weights, turnover constraint). It would be beneficial to demonstrate whether the proposed model can accommodate a range of additional constraints. - There is a need for a comparative analysis of computation time to evaluate the efficiency of the proposed method. - The following paper addresses the exact same problem of cardinality constrained Sharpe ratio maximization. While their approach, based on relaxation, has the advantage of transforming the problem into a convex optimization issue, it fails to control cardinality exactly. [1] Kim, M. J., Lee, Y., Kim, J. H., & Kim, W. C. (2016). Sparse tangent portfolio selection via semi-definite relaxation. Operations Research Letters, 44(4), 540-543.

Questions

The authors emphasize several times in bold that directly maximizing the Sharpe ratio is a key point of this study. However, as shown in equations (3.4) and (3.7), the problem is eventually transformed into a mean-variance optimization form, with Theorem 1 demonstrating how this can be converted back to a solution for the Sharpe ratio maximization problem. This raises the question of why the study is framed as a Sharpe ratio maximization problem. Why not simply present it as m-sparse mean-variance optimization, which is more general? After all, Sharpe ratio maximization is just one of many Pareto optimal points in mean-variance optimization. Is this framing due to overlapping aspects with previous research? Further clarification on this point would be helpful.

Rating

5

Confidence

4

Soundness

3

Presentation

3

Contribution

3

Limitations

The authors mention the inability to directly apply fractional optimization as a limitation. However, I believe that more emphasis should be placed on practical implementation. Since this study defines and solves a problem based on practical needs, more attention should be given to practical implementation rather than purely mathematical formulation. Ensuring that the proposed method can be effectively implemented in real-world scenarios is crucial for its overall utility.

Reviewer 4h7E5/10 · confidence 3/52024-07-12

Summary

In summary, this paper studies Sharpe ratio optimization in portfolio management and contributes the achievement of sparse distribution iterates converging to a local optimum by converting the fractional optimization problem into a quadratic programming.

Strengths

Originality: The task of optimizing SR with sparse distributions is somewhat new. The work incorporates quadratic programming and its well-known techniques into SR optimization in a novel way. Quality: The submission is technically sound. The existing claims are generally supported by theory and experiments. The methods are appropriate. Clarity: The writing of the submission is mostly clearly, well organized and adequately informs the reader. Significance: The work advances the state of the art in a demonstrable way with its theory. Other researchers and practitioners are likely to use the ideas, possibly build on them, considering the experimental results. With a somewhat unique theoretical approach, it provides unique conclusions about existing optimization targets in the form of sparse optimization.

Weaknesses

Originality: It is not clear how this work differs from previous contributions beyond the fact that they did not study this setting. Due to this, it is also possible that the related work is not adequately cited. More explanations are needed in that regard. Quality: This is more of a work in progress. The authors are at times not careful and possibly honest about evaluating their work. Their achievement is only to a local optimum and the feasibility of the scenario for global optimality is questionable. They also do not provide convergence rates, which is important in performance analysis. Significance: The importance of the results is a bit questionable due to the lack of specific comparisons with the literature. Hence, it is also not demonstrated that the submission addresses a difficult task in a better way than the previous works.

Questions

Major Questions: - Why are self-financing and long-only constraints needed? - What does Section 2.1 accomplish? It is too superficial with no heed towards the roles of the introduced models and parameters in their respective optimization scenarios, akin to a laundry list of past works. How are they related to SR maximization? It appears each method aims for something different. - What does "guarantee suboptimality" mean? - Considering the definition of $Q_\epsilon$, when is (i) possible? - What is the convergence rate? Minor Questions: - Page 1 Line 33: How is "market crashes" from [25] a strategy? Suggestions: - Please correct the reference numbering. - Page 6 Line 240: Grammar error.

Rating

5

Confidence

3

Soundness

2

Presentation

2

Contribution

2

Limitations

The authors addressed the limitations. For improvement, they are suggested to update their limitations with lack of convergence rate and global optimality.

Reviewer EtEi5/10 · confidence 3/52024-07-14

Summary

This paper studies the optimization of an m-sparse portfolio, which has an additional sparsity constraint on the portfolio compared to traditional portfolio optimization. Instead of the mean-variance approach, this work proposed to directly optimize the fractional objective which is the Sharpe ratio. The paper shows that the m-sparse SR optimization can be converted into an equivalent m-sparse quadratic programming, which is non-convex. It then proposed to use a proximal gradient algorithm to obtain a locally optimal solution.

Strengths

- this paper studies an important and practical problem of sparse portfolio optimization, and it is presented with a good clarity. - The paper proposed to directly optimize the fraction - Sharpe ratio, which can be turned into a nonconvex fractional optimization under constraints. Such an idea to alternatively formulate the problem is natural and very interpretable. - It is shown theoretically that the proposed proximal gradient algorithm is guaranteed to converge to a locally optimal Sharpe. Experimental results further show that PGA outperforms other baselines in terms of the achieved Sharpe ratios.

Weaknesses

- Although the proposed PGA algorithm is guaranteed to converge to a locally-optimal solution, there is a lack of analysis and results on how suboptimal the converged solution might be in the worst case. - Moreover this is a lack of discussions on the convergence rate for the proposed PGA algorithm.

Questions

- Directly optimizing the fractional objective can be very unstable with real-world noisy financial data. How does the proposed PGA algorithm compared to the traditional optimization in terms of robustness? - What are the computational time and costs for the different baselines?

Rating

5

Confidence

3

Soundness

3

Presentation

3

Contribution

2

Limitations

Yes.

Reviewer iiAu6/10 · confidence 1/52024-07-16

Summary

This paper studies Sharpe ratio optimization with sparsity constraints. The paper transforms the original m-sparse fractional optimization problem into an m-sparse quadratic programming problem and develops a proximal-gradient algorithm to solve it. Numerical experiments show that the proposed method improves the Sharpe ratio compared with existing methods.

Strengths

I am not familiar with this area. This paper first represents the Sharpe ratio by substituting the mean and variance with unbiased estimates, and then rewrites the problem in a quadratic objective. From a purely mathematical perspective, this entire procedure is quite standard. What’s interesting seems to be the convergence guarantee and the experiment performance of the proposed algorithm. I have some questions about them but if they turn out justified, I think this work brings fair contribution.

Weaknesses

While the paper transforms the original problem to (3.4), the $\ell_0$ constraint is non-convex. In (3.7), the objective function is still non-convex and it seems to contradict with standard optimization theory that you can guarantee convergence to global optimum in this non-convex objective. Indeed, this is a combinatorial optimization problem and has computation lower-bounds. In sum, I think the theoretical results need further justification on why they do not contradict with existing hardness results. POST REBUTTAL: the author justified their results in rebuttal.

Questions

See the question in the weakness section.

Rating

6

Confidence

1

Soundness

3

Presentation

3

Contribution

2

Limitations

The authors have addressed the limitations.

Reviewer 4h7E2024-08-09

Thank you for your answers. I am raising my score in consideration of the renewed global optimality claims.

Reviewer pxYT2024-08-11

Thank you for the comments. But I still do not think directly solving 'cardinality constrained Sharpe ratio maximization problem' instead of 'cardinality constrained mean-variance optimization problems' is "novel". I will maintain my evaluation as it is.

Program Chairsdecision2024-09-25

Decision

Accept (poster)

© 2026 NYSGPT2525 LLC