A Noise-Aware Quantum Algorithm for Credit Valuation Adjustments on Real Quantum Hardware

Battle Count: 50

The paper addresses Credit Valuation Adjustment (CVA), which is a critical component of derivative portfolio valuation and counterparty credit risk management in institutional banking. While not directly about trading strategies or market prediction, CVA computation is essential for pricing derivatives, managing counterparty risk, and regulatory capital requirements. The quantum amplitude estimation methodology could eventually accelerate Monte Carlo-based risk calculations used in trading desks. However, the current implementation is far from production-ready, with dominant discretisation and encoding errors, and no demonstrated end-to-end quantum advantage. The work is more relevant to quantitative risk management and derivative valuation than to algorithmic trading or market-making strategies.

Extremely high complexity. Requires: (1) quantum hardware access (IBM Quantum via BasQ); (2) QCBM training with variational optimization for state preparation; (3) CRCA training for three separate controlled-rotation blocks (exposure, discount, default); (4) implementation of CABIQAE with Bayesian inference, contrast calibration, Fisher-information-based scheduling, and posterior transport; (5) hardware-replay methodology with readout mitigation and contrast-model fitting; (6) Q-CTRL Performance Management integration; (7) heavy-hex topology-aware circuit design; (8) classical Monte Carlo benchmarking with multi-asset lognormal dynamics, CDS bootstrapping, and Black-Scholes pricing. The full CVA circuit at k=1 has 1122 transpiled two-qubit gates and depth 2691. Multiple software components (Qiskit, Q-CTRL, custom Bayesian code) must be integrated.

The paper provides a GitHub repository (https://github.com/gborras10/Quantum-CVA) with code for generating finite-grid tensors and running the pipeline. Detailed numerical controls, hyperparameters, market data specifications, and circuit architectures are reported in appendices. Hardware access was through Basque Quantum (BasQ) on ibm_basquecountry. However, market data from LSEG Workspace is proprietary, and hardware-replay results depend on specific backend calibration. The hardware-replay methodology (not live adaptive execution) limits full reproducibility without equivalent hardware access. AI tools were used for manuscript refinement but not for scientific content.