OQC Study Maps What Quantum Finance Still Needs
A study published on September 17, 2026, by Oxford Quantum Circuits (OQC) and Trust Base addressed a crucial question often overlooked in quantum finance research: not just if a quantum algorithm can price a derivative, but whether it can do so effectively enough for integration into a bank’s operational systems. The current answer, according to the study, is no. Instead, the study provides a valuable roadmap of requirements, which proves more beneficial than another benchmark victory.
The results are detailed in a technical blog post from OQC.
Study Participants and Scope
OQC contributed the quantum expertise. Trust Base, their research partner and a digital subsidiary of Sumitomo Mitsui Trust Group, grounded the work in a real financial institution, moving beyond purely academic exploration. They examined three key financial workloads: derivative pricing, Value-at-Risk, and Credit Valuation Adjustment (CVA).
The study compared four methodologies: classical Monte Carlo, classical physics-informed neural networks (PINNs), hybrid quantum-classical neural networks, and quantum-enhanced Monte Carlo. Testing encompassed various models, including Black-Scholes, Garman-Kohlhagen, Dupire, and Hull-White, covering stock and currency options, as well as scenarios with changing volatility and interest rates, as reported in the OQC post.
Defining “Advantage” in Practice
The study’s core insight is that accurate pricing alone is insufficient. Banks require more than just the instantaneous value of a contract. They need sensitivities, which indicate how value changes with market movements, and these sensitivities must remain robust across diverse conditions.
OQC reported that delta, which measures a derivative’s price response to its underlying asset, could often be improved with targeted training. Gamma, however, presented a challenge. Gamma tracks the rate of change of sensitivity itself, and even minor pricing errors amplified into significant gamma errors. A model that accurately prices but yields unstable gamma is nearly useless for a trading desk, as hedging strategies depend on this metric.
Therefore, OQC argues that any claim of advantage must encompass full pricing surfaces, sensitivities, exposure profiles, CVA, and stress calculations. A method might succeed in an isolated test but still fail when integrated into a complete workflow.
Quantum’s Role: Successes and Limitations
The team explored two main approaches. The first involved physics-informed neural networks (PINNs), which learn a continuous representation of how a product’s price should behave under varying conditions. The hybrid version, quantum-compressed PINNs (QPINNs), embeds a parameterized quantum circuit into the network as a trainable feature map.
QPINNs sometimes matched classical models in parameter count and occasionally improved a single price or exposure figure. However, they were slower. Repeatedly executing the quantum circuit and computing its gradients significantly increased total runtime. Fewer parameters did not translate to less computation once circuit execution, optimization, and the handoff between quantum and classical systems were accounted for.
Consequently, OQC recommended classical PINNs as the most practical method tested, viewing QPINNs as a research tool for studying hybrid interfaces rather than a production-ready replacement.
A related lesson emerged: adaptive sampling, or directing training towards areas where a model exhibits the most errors, proved more beneficial than simply adding parameters. Understanding the financial problem effectively outweighed brute-force computation. This also underscores why classical baselines must be meticulously tuned, otherwise a quantum “win” might merely reflect an inadequate comparison.
Hardware Assumptions and Future Timeline
The second approach investigated quantum amplitude estimation, which promises desired precision with fewer samples than traditional Monte Carlo methods. In practice, OQC found that performance hinged on circuit construction, compilation, qubit placement, and noise. Deeper, more exact circuits sometimes performed worse than simpler ones when noise accumulated.
The resource estimates warrant careful scrutiny. OQC modeled that increasing the assumed error-correction threshold from 1% to 5% could reduce the physical-qubit requirement for a specific 32-qubit configuration from approximately 404,000 to about 130,000.
It also projected that quantum signal processing might reduce T-gate needs by up to 16 times and logical qubits by four. These are modeled figures from an interested vendor, lacking independent peer review or competing studies for verification. Achieving that 5% threshold presupposes future hardware like dual-rail or erasure qubits, not technology currently available.
Such error-correction work is evolving within the broader ecosystem. In mid-September 2026, Altera and Riverlane announced a validated QEC interface on Agilex 7 FPGAs, indicating maturing decoding capabilities. Meanwhile, IonQ has set a target of 200,000 physical qubits by 2028, a roadmap item that represents a future goal.
Software and Commercial Context
The financial technology stack intended for these methods is beginning to take shape. Fujitsu released OpenQARP v0.1.0 in September 2026 under Apache 2.0, open-sourcing a quantum application package to facilitate algorithm reuse rather than constant rebuilding.
Security, though not directly addressed in this study, is an adjacent concern. The OQC work focuses on computational pricing and risk, thus quantum key distribution (QKD) and post-quantum cryptography (PQC) are outside its scope for advantage mapping. However, banks planning for quantum will also need to navigate a security migration.
Q*Bird rebranded as Falqon Systems in September 2026, offering QKD as operational quantum secure networks. PQC and QKD address the “harvest-now-decrypt-later” risk differently, with NIST’s standards FIPS 203, 204, and 205 covering PQC algorithms. As of now, no cryptographically relevant quantum computer exists.
Commercial momentum can sometimes obscure the full picture. IonQ reported record Q2 2026 revenue of $80.1 million, a 287% year-on-year increase, alongside a net loss of $1.87 billion. Revenue growth and demonstrated advantage in a live financial application are distinct metrics.
Validating the Roadmap
The OQC and Trust Base study functions as a requirements document. Its value lies in its concrete identification of workloads, resource thresholds, and conditions. It concludes that classical PINNs are currently superior, hybrid models do not yet justify their runtime, and amplitude estimation depends on hardware that has not yet been built. All forward-looking projections originate from an interested vendor and lack third-party verification.
To validate this roadmap, we would need: peer-reviewed error-correction results at the scales OQC assumes, an independent benchmark of the entire pricing-and-sensitivity workflow, and a bank successfully deploying a quantum component in production alongside a well-tuned classical system. Until then, the study should be regarded for what it clearly is: a useful specification of the work that remains, and a clear statement that quantum advantage in finance has not yet arrived.