arXiv:2609.30195
Loan Portfolio Optimization with Variational Quantum Algorithms
Balaganchi A. Bhargava, Kamlesh Kumar, Gouranga Dinda, Aniruddha Biswas, Vijay S. Rao, Debarag Banerjee, Arun Sehrawat
Abstract
Loan portfolio optimization (LPO) seeks portfolios that minimize credit risk while satisfying practical selection constraints. Accounting for portfolio-level risk requires modeling not only expected losses but also loss variability and borrower-default correlations, leading to a large-scale combinatorial optimization problem whose complexity grows rapidly with portfolio size. In this work, we formulate LPO as a Quadratic Constrained Binary Optimization (QCBO) problem that jointly captures expected and unexpected losses and transform the resulting formulation into a Quadratic Unconstrained Binary Optimization (QUBO) model suitable for variational quantum optimization. To address the limited qubit resources of current quantum hardware, we employ Pauli Correlation Encoding (PCE), which enables a large number of portfolio-selection variables to be represented using comparatively few qubits. The proposed framework is evaluated on a real-world microfinance dataset comprising 2012 borrowers. For problem instances with up to 1000 portfolio-selection variables, we benchmark classical simulations of a variational quantum algorithm (VQA) against Google OR-Tools under matched time budgets, obtaining feasible portfolios with an objective-value gap of about 40% across all sizes. We further implement the framework on superconducting quantum hardware for instances containing up to 1500 variables. Across these instances, the experimentally obtained objective values are within 10% of the corresponding simulation results. These findings demonstrate the feasibility of applying PCE-based variational quantum optimization to large-scale loan portfolio optimization under current quantum hardware limitations.
AI Summary
The paper studies whether Pauli Correlation Encoding (PCE) can support variational quantum optimization of loan portfolios with substantially more selection variables than qubits. It constructs a binary model from expected losses and borrower-loss covariances, using a microfinance dataset of 2,012 borrowers with 26 monthly observations. To retain a quadratic objective while matching monetary units, it replaces each covariance entry by its sign-preserving square root; this modified term is not a standard risk measure. Borrower-selection scores are encoded in Pauli expectation values and optimized with a hardware-efficient circuit and COBYLA. The implemented objective omits the cardinality penalty; selecting the top $K$ scores enforces portfolio size afterward. A hardware-oriented encoding restricts measurements to three commuting Pauli families, requiring $\lceil\log_2(N/3+1)\rceil$ qubits. For $K=N/2$ and instances up to $N=1000$, classical VQA simulations are compared with an auxiliary-variable CP-SAT formulation in Google OR-Tools under matched time limits, averaging five sampled instances per size. VQA portfolio losses are approximately $40\%$ higher; simulation runtime rises from about 5 seconds at $N=20$ to 800 seconds at $N=1000$. Hardware experiments at $N=20,400,1000,1500$ produce losses within $10\%$ of corresponding simulations. Reported losses use the original expected-loss-plus-standard-deviation objective, rather than the modified training objective. The paper does not establish computational advantage. Stated limitations include simplified default-probability and recovery assumptions, omitted operational and regulatory constraints, and unresolved questions about compression, expressivity, trainability, solution quality, and classical simulability at larger scales.
Central claim may not hold: The quantum stage omits the cardinality penalty and imposes feasibility afterward by top-K ranking, without establishing that its unconstrained scoring objective is a faithful surrogate for constrained portfolio optimization. The reported approximately 40% comparison evaluates original total loss even though both solvers optimize a different, modified total-loss objective. Temporal covariance of estimated expected losses omits default randomness and therefore does not establish the paper's claimed unexpected-loss and borrower-default-correlation interpretation. The entrywise signed-square-root transformation can turn a valid covariance matrix into an indefinite matrix with negative portfolio-risk values, without a demonstrated connection to minimizing original risk.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- 14 pages, 10 figures