OQC and Trust Base test quantum methods for financial risk

OQC and Trust Base, a digital subsidiary of Sumitomo Mitsui Trust Group, directly compared classical and quantum approaches to financial modeling for Value-at-Risk, Credit Valuation Adjustment, and derivative pricing. The project assessed classical Monte Carlo baselines against quantum-enhanced Monte Carlo, Physics-Informed Neural Networks, and hybrid Quantum Physics-Informed Neural Networks, seeking to advance quantum finance. Researchers found near-term value lies in “disciplined benchmarking, hybrid architectures and hardware-aware research” rather than solely pursuing fully quantum solutions. “The real question is not simply whether quantum algorithms are interesting in theory,” the team states. OQC and Trust Base Explore Quantum Methods for Financial Risk This partnership focused on practical application, assessing how quantum methods could integrate into existing financial pipelines rather than simply demonstrating algorithmic potential. The project’s scope included a direct comparison of classical and quantum approaches to these critical financial calculations, a methodology designed to establish a clear benchmark for performance. Classical Physics-Informed Neural Networks currently offer a balance of speed, accuracy, and ease of integration into existing systems, serving as a present-day solution while quantum methods mature. This acknowledgement highlights a pragmatic approach, recognizing that quantum computing will likely augment, rather than immediately replace, established classical techniques in the financial sector.

The team evaluated these neural networks alongside both classical Monte Carlo simulations and their quantum-enhanced counterparts, as well as hybrid Quantum Physics-Informed Neural Networks, to create a comprehensive performance landscape. This rigorous, side-by-side evaluation aimed to identify where quantum methods could genuinely offer an advantage, not just in theory, but in real-world financial operations. Establishing robust classical baselines is paramount to accurately measuring the value of any quantum enhancement, according to the research. Without reproducible benchmarks for established methods like Monte Carlo and finite-difference techniques, it becomes impossible to determine if a quantum approach truly delivers an improvement. The work with Trust Base emphasizes a staged process for financial institutions, beginning with capability building and workload identification, before integrating quantum components selectively into existing risk and pricing pipelines. The project’s findings suggest a pragmatic roadmap for quantum adoption in finance, prioritizing hybrid workflows where quantum systems contribute as part of a broader classical stack. This approach allows quantum components to be utilized for specific tasks where they may excel in representation, sampling, or optimization, maximizing their impact within the limitations of current hardware, Trust Base says. Evaluating the complete workflow, including pricing, sensitivities, exposure, and risk calculations, is also important because runtime, calibration cost, stability, and integration all play a significant role in determining a model’s usefulness. As quantum hardware continues to improve in qubit quality, count, latency, and error correction, the balance between classical and hybrid methods will inevitably shift, and this research aims to prepare for that transition. Classical PINNs Balance Accuracy and Runtime for Derivative Pricing Experiments revealed these networks were fast, flexible across various financial products, and particularly well-suited for repeated calculations essential for processes like Exposure at Potential Default and Credit Valuation Adjustment. This acknowledgement signals a pragmatic approach, focusing on near-term applicability rather than solely pursuing distant quantum advantages. The project deliberately contrasted classical PINNs with more complex quantum-enhanced methods, including Quantum Physics-Informed Neural Networks and quantum-enhanced Monte Carlo simulations, to establish a clear performance baseline. While Quantum PINNs demonstrated potential in reducing parameter counts and improving certain scalar pricing metrics, the researchers found the repeated evaluation and differentiation required by the quantum feature map significantly slowed down overall processing time. This finding highlights that gains in algorithmic efficiency must be weighed against the computational cost of utilizing quantum resources. The evaluation extended beyond simple pricing to encompass a full financial workflow, including sensitivity analysis, exposure calculations, and risk assessments. This detailed approach is crucial, as a useful model must perform reliably across all stages of the process, not just deliver a single price point. Runtime, calibration costs, model stability, and seamless integration with existing systems were all considered key metrics, demonstrating a focus on operational feasibility.

The team’s work suggests that practical quantum advantage will not emerge from algorithm design alone, but from improvements across the entire technological stack. Classical PINNs, with their established performance and ease of implementation, provide a valuable stepping stone towards more advanced hybrid models. The research being conducted now, the team believes, will help determine which workloads will be best positioned to benefit from these future hardware advancements. QPINNs and Quantum Monte Carlo as Hybrid Research Directions Quantum-compressed Physics-Informed Neural Networks and Quantum Monte Carlo are not isolated pursuits, but areas where practical gains hinge on hardware improvements, efficient compilation, and seamless integration with existing classical systems. Financial services presents a clear application space, with some quantum applications already delivering limited commercial benefits due to the sector’s reliance on complex models, rapid decision-making, and the inherent uncertainty surrounding derivatives pricing, market risk, and credit exposure. This focus on operational utility, rather than purely algorithmic novelty, defines a pragmatic approach to quantum finance. Derivative pricing serves as a useful testbed because a pricing model must determine a value for a future payoff dependent on uncertain market movements; established methods like Black-Scholes and Monte Carlo simulation are well understood for simpler products. Quantum Physics-Informed Neural Networks extend this architecture by incorporating a parameterised quantum circuit as a trainable quantum feature map, not to replace classical workflows entirely, but to explore whether quantum features can enhance expressivity or parameter efficiency within a hybrid model.

Quantum Monte Carlo explores amplitude-estimation-based methods for estimating payoff expectations, offering an alternative to directly solving the pricing partial differential equation. However, near-term performance of Quantum Monte Carlo is heavily influenced by practical considerations, including reflector choice, compilation strategy, qubit placement, and the impact of hardware noise.

The team found that lower-cost circuit designs could sometimes yield smoother empirical results than more precise, but deeper, constructions, according to Trust Base. “Today, QPINNs are best understood as a platform for studying hybrid expressivity and future quantum-classical interfaces, rather than as a production replacement for classical solvers,” the researchers state.

Financial Workflows Demand Surface Reliability Beyond Price Accuracy Instruments with complex features, barrier options, local-volatility models, and swaptions, posed particular challenges, demanding models capable of delivering stable sensitivities beyond a single price point. “For financial institutions, that distinction matters as a model that gets the price right but produces unstable sensitivities is not yet ready to be used for real risk management,” the researchers state. Physics-Informed Neural Networks learn pricing surfaces by directly addressing the governing equations, boundary conditions, and payoff conditions, offering potential speedups over traditional grid-based methods. However, realizing these benefits requires addressing practical hurdles including circuit depth, qubit layout, noise, and calibration.

The team found that while several models could match benchmark prices at specific points, the overall pricing surface and its derivatives revealed a more complex picture, the company says. Delta hedging could often be improved with targeted training, but Gamma, dependent on second derivatives, remained fragile, amplifying even minor surface roughness. This sensitivity to surface roughness is critical because financial workflows rely on consistent behavior across the entire pricing surface, not just at a single market point. Hedging, exposure analysis, Credit Valuation Adjustment calculations, and stress testing all demand models that provide reliable results across a range of inputs. “Across the Black-Scholes/Garman-Kohlhagen, Dupire local-volatility and Hull-White model families, several models were able to match a benchmark price at a reference point,” the team reports. But the ability to accurately price an instrument at a single point does not guarantee the stability of its derivatives, which are essential for effective risk management. The project examined two primary approaches: quantum-enhanced learning for pricing surfaces and quantum-enhanced simulation for expectation estimation, ultimately highlighting the need for a holistic assessment of model reliability beyond simple price accuracy. Source: https://oqc.tech/resources/quantum-finance-beyond-the-benchmark More like thisQuantum AlgorithmsParityQC’s new optimizer cuts gate count for quantum problemsQuantum Research NewsNTU-IBM Quantum Hub shows shallow circuits beat language models on key tasksQuantum AlgorithmsResearchers Bound Error Weight Controlling Quantum Memory FailuresQuantum Research NewsGerman scientists cut Toffoli gate count for sparse quantum statesStay currentSee today’s quantum computing news on Quantum Zeitgeist for the latest breakthroughs in qubits, hardware, algorithms, and industry deals. Tags: