HSBC teams up with Haiqu to push scalable quantum risk models by tackling data loading bottlenecks in quantum computing for finance.

They demonstrate a breakthrough in preparing financial data for quantum computers using Quantum State Preparation with Matrix Product States to create shallow circuits, enabling efficient encoding of heavy-tailed distributions used in risk modeling.

The approach avoids storing the entire discretised dataset in classical memory by employing a sampling-based workflow, allowing larger encoding circuits and better scalability for quantum risk calculations.

Experts say quantum tech for financial applications is nearing practical use, though it has not yet been deployed in retail banking operations.

In the 25-qubit tests, standard statistical benchmarks were met; 64-qubit tests probed resilience under realistic noise; 156-qubit simulations demonstrated scalability beyond prior limits.

The effort aims to move quantum-ready risk models from lab studies toward real-world banking applications, potentially shaping how banks assess losses in downturns.

HSBC’s Dr. Philip Intallura underscored that efficient preparation of complex distributions in shallower circuits brings practical quantum finance in risk modeling closer to reality.

The method enables shallow quantum circuits, reducing hardware demands while handling complex probability distributions in risk modeling.

Overall, the work reports successful reproduction of probability distributions and benchmark satisfaction on IBM hardware, signaling progress toward practical financial risk modeling.

The team uses Quantum State Preparation with Matrix Product States to encode heavy-tailed distributions efficiently without requiring full classical storage of the dataset.

Tests were run on IBM quantum hardware across 25-, 64-, and 156-qubit scales, successfully reproducing probability distributions and showing robustness to noise and capacity for massive datasets.

Haiqu’s co-founder CTO Mykola Maksymenko notes encoding realistic financial data onto quantum hardware has been a major barrier, and this work offers a scalable workaround.