Quantum Computing for Asset Allocation and Dynamic Redemption in Wealth Management: A QUBO-Based Study

1. Asset Allocation and Dynamic Redemption Scenario

Recent years have seen Chinese wealth‑management products evolve from closed‑ended to open‑ended, net‑value‑based structures, creating multi‑asset, multi‑layer holdings such as direct assets and asset‑management plans. High‑frequency subscription/redemption actions increase uncertainty and require frequent fund scheduling and position adjustments to keep net‑asset value stable, meet return targets and ensure liquidity. The core problem is a high‑dimensional resource‑allocation task with multiple objectives and constraints.

Using a fixed‑income product as an example, the authors model the objective of maximizing returns while satisfying constraints on fund balance, concentration limits, liquidity, and operational feasibility. Decision variables include daily purchase amount x_i, redemption amount y_i, and end‑of‑day idle cash s. Additional parameters such as expected return r_i, market value V_i, total net assets A, net subscription amount N, and initial idle cash C are defined.

2. Quantum Computing Model Construction

Four groups of constraints are formulated:

Fund‑balance constraint : ensures no cash shortfall or excessive idle cash after all operations; expressed as a penalty term with exponential growth when violated.

Concentration constraint : limits any single underlying asset’s market‑value proportion to ≤10%; penalty grows with the square of the excess.

Liquidity constraint : requires high‑liquidity assets (cash, short‑term government bonds, etc.) to constitute at least 5% of total assets; penalty increases quadratically with shortfall.

Operational and scenario constraints : include redemption‑limit (redeem amount ≤ current market value), mutual‑exclusivity of purchase and redemption for the same asset on the same day, and scenario‑driven directionality based on net subscription/ redemption sign.

All constraints are transformed into penalty terms and combined with the maximization objective (converted to a minimization of negative return) to obtain a unified QUBO expression. Penalty coefficients λ_i are tuned according to constraint priority: redemption‑mutual‑exclusivity > concentration > fund‑balance > liquidity.

3. Quantum Computing Model Application

The case study considers a fixed‑income product with a daily net asset value of CNY 1.06 billion, a net subscription of CNY 10 million, a target annualized return ≥2.75 %, idle‑cash ratio ≤1 %, and the constraints described above.

The problem is solved on a coherent‑light quantum computer (coherent Ising machine) and compared with classical Simulated Annealing (SA) and Tabu Search (TS) algorithms. Results:

SA: 31.14 s, return 3.0283 %.

TS: 12.00 s, return 3.0224 %.

Quantum computer: 16.97 ms, return 3.0325 %.

Speed‑up factors: SA is 1 835× slower, TS is 707× slower than the quantum solution. Energy consumption for the pure computation phase is 15 570 J (SA), 6 000 J (TS) and 16.99 J (quantum). Considering data transfer and conversion, end‑to‑end energy is 1 000.017 J for the quantum workflow, making SA about 16× and TS about 6× more energy‑intensive. The authors argue that, given the large number of wealth‑management products and frequent optimization needs, the quantum approach offers significant speed and energy advantages, supporting greener fintech.

4. Conclusion

The study demonstrates that translating wealth‑management asset‑allocation and dynamic‑redemption problems into QUBO form and solving them on a coherent‑light quantum computer yields feasible solutions that meet all business constraints while dramatically reducing solving time and energy consumption. Future work may incorporate macro‑economic variables and investor‑behavior data to enrich constraints and extend the approach to securities firms and insurers.

References: Markowitz (2012), Rockafellar & Uryasev (1999), Fabozzi & Drake (2011), Dong et al. (2019), etc.