Literature Review // 001

Quantum Portfolio Optimization — From Markowitz to VQE

Semester

Spring 2024–2025

Pages

23 Pages

I. The Problem — Why Quantum?

Portfolio optimization, introduced by Markowitz in 1952, is a cornerstone of quantitative finance. The classical mean-variance model seeks to maximize returns while minimizing risk through an optimal allocation of weights across N assets. For the continuous relaxation, this is a well-behaved quadratic programming problem solvable in O(N³·⁵) time.

However, real-world constraints — discrete trading units, transaction costs, and regulatory limits — transform it into an integer programming problem that is NP-hard. For binary selection alone, the search space is 2ᴺ: 20 assets yield ~1 million combinations (solvable in minutes), but 100 assets yield ~10³⁰ combinations (computationally intractable). This exponential scaling is exactly where quantum computing promises advantage.

II. From Classical to Quantum — The Transformation Pipeline

The review traces a rigorous mathematical pipeline that converts the integer portfolio problem into a format executable on quantum hardware. The process has three key stages:

tag
Binary Encoding

Integer allocation variables nᵢ are encoded into binary using nᵢ = Σ 2ʲ bᵢ,ⱼ. For N assets with max nₘₐₓ units, this requires K = N⌈log₂(nₘₐₓ+1)⌉ qubits.

functions
QUBO Formulation

Constraints are absorbed into the objective via penalty terms with Lagrange multipliers, yielding an unconstrained binary optimization: f(b) = bᵀQb + cᵀb.

memory
Ising Hamiltonian

Binary variables are mapped to spin variables (bᵢ = (1+sᵢ)/2), producing a Hamiltonian H = Σ hᵢZᵢ + Σ JᵢⱼZᵢZⱼ whose ground state encodes the optimal portfolio.

For the 3-asset example used throughout the review (max 7 units each), this yields a 9-qubit Hamiltonian with up to 9 local field terms and 36 coupling terms — small enough for near-term quantum hardware, yet rich enough to demonstrate the full pipeline.

III. The VQE Algorithm — A Hybrid Approach

The Variational Quantum Eigensolver is a hybrid quantum-classical algorithm that finds the ground state of the portfolio Hamiltonian through iterative optimization. The quantum computer prepares a parameterized state |ψ(θ)⟩ using a TwoLocal ansatz circuit — alternating layers of Ry rotations and CZ entangling gates — and measures the Hamiltonian expectation value. A classical optimizer (COBYLA) then updates the circuit parameters to minimize this energy.

Key Result:

VQE achieves convergence to within 1% of optimal portfolio utility with 10⁴–10⁵ shots per iteration. For the 3-asset example, after 100 iterations the decoded allocation yielded n = (5, 2, 3) with expected return 8.9%, volatility 14.2%, and Sharpe ratio 0.627.

The review also addresses the barren plateau problem — gradient variance decays as 2⁻ⁿ with system size, making optimization infeasible beyond ~15–20 qubits with current techniques. This fundamental limitation constrains near-term quantum portfolio optimization to small-scale demonstrations, though layer-wise training and careful parameter initialization offer partial mitigation.

IV. Key Findings & Outlook

check_circle
Mathematical Foundation

The binary encoding method preserves the optimization structure while enabling quantum algorithms to exploit superposition and entanglement.

speed
Quantum Advantage Potential

Classical integer optimization scales as O(2ᴺ). Quantum approaches promise polynomial speedup, though current hardware limits demonstrations to 10–15 assets.

tune
VQE Resilience

The hybrid quantum-classical structure shows resilience against quantum noise, with the TwoLocal ansatz balancing expressibility and trainability.

trending_up
Future Outlook

As quantum devices approach the 1,000-qubit threshold with improved error rates, portfolio optimization may become one of the first commercially valuable quantum applications.

Key References

  1. Markowitz, H. (1952). "Portfolio Selection." The Journal of Finance, 7(1), 77–91.
  2. Tilly, J. et al. (2022). "The Variational Quantum Eigensolver: A Review." Physics Reports, 986, 1–128.
  3. Date, P. et al. (2019). "Efficiently Embedding QUBO Problems on Adiabatic Quantum Computers." Quantum Information Processing, 18(4).
  4. McMahon, D. (2007). "Quantum Computing Explained." Wiley.
  5. Nature (2023). "Quantum Finance Optimization Using VQE-Like Techniques."
arrow_back
Back to All Projects & Reviews