Quantum Trial Lifts Bond Fill-Rate Predictions at HSBC
HSBC says a hybrid quantum–classical model with IBM’s Heron chip improved fill-probability predictions by up to 34% on real European corporate bond data, a first-known empirical result on production-scale datasets.
News
- Intel Courts Apple, TSMC After Nvidia’s $5 Billion Lifeline
- Google Introduces Gemini Robotics 1.5 for Safer, Smarter Robots
- Alibaba Takes Aim at Rivals With Qwen3 Max at 1 Trillion Parameters
- OpenAI, Oracle, and SoftBank Partner for 5 Data Centers Under Stargate
- Towards Superintelligence: Nvidia Puts $100B in OpenAI
- India’s Next GCC Hubs May Rise Outside Big Cities
HSBC Holdings Plc said Thursday that a test with IBM’s quantum computer made its bond-trading predictions up to 34% more accurate, calling it the first known real-world result on production-scale data, though it did not involve live trades.
The test combined quantum and classical techniques on a large real-world data set from the European corporate bond market, a part of finance where milliseconds and small model gains decide who wins an order.
The bank ran the experiment with IBM’s latest Heron quantum processor, applying quantum-produced data transformations to an anonymized set of real request-for-quote (RFQ) interactions and then measuring the improvement against common classical methods used in the industry.
HSBC calls it the “world’s first-known empirical evidence” that current quantum hardware can surface useful pricing signals in noisy, real-market conditions.
The immediate benefit for customers of these markets—bond investors, issuers, and dealers’ clients—is simple.
If a desk can estimate the chance of a fill more accurately and a touch faster, it can respond with tighter, more confident quotes more of the time. That raises the chance clients get the trade done at the first ask, lowers frustrating repeated quote requests, and can, at scale, encourage dealers to show better prices because their risk model is less uncertain.
In dealer-to-dealer and client markets where trades are between two parties without a central exchange (an RFQ market where liquidity is fragmented and price discovery is partly bilateral), smaller percentage gains in model accuracy compound across thousands of daily inquiries into measurable improvements in execution quality.
HSBC stressed that the trial was not a live trade but was conducted on “real and production-scale trading data,” a caveat that keeps this in the realm of rigorous back-testing rather than deployed profit and loss.
Still, the bank said the results show what a near-term, combined approach—classical systems augmented by quantum steps—might unlock as hardware matures.
“This is a ground-breaking world-first in bond trading,” said Philip Intallura, HSBC’s group head of quantum technologies. “It means we now have a tangible example of how today’s quantum computers could solve a real-world business problem at scale and offer a competitive edge, which will only continue to grow as quantum computers advance.”
IBM made the same case from the technology side. “This exciting exploration shows what becomes possible when deep domain expertise is integrated with cutting-edge algorithm research, and the strengths of classical approaches are combined with the rich computational space offered by quantum computers,” said Jay Gambetta, Vice-President of IBM Quantum. “Such work is essential to unlock new algorithms and applications that are poised to transform industries as quantum computers scale, and the future of computing takes shape.”
HSBC’s credit algorithmic trading team framed the commercial significance in trader’s terms. “We spend all day looking for single-digit improvements, because when you repeat that thousands of times a day, it can really make a difference,” said Josh Freeland, the bank’s global head of algorithmic credit trading. He added that achieving gains of this order with consistency “would be quite something.”
Under the hood, the team validated the method on multiple IBM quantum systems and on simulators.
A technical paper posted this week detailed how quantum-hardware-generated transformations embedded as an offline, queryable component inside fast-response trading workflows, reduced fill-prediction error for standard machine-learning models on data the model had not seen before.
Notably, the authors report that the inherent noise in today’s quantum hardware appears to contribute to the observed improvement, a counterintuitive finding that suggests noisy quantum devices can add useful structure to real-world, time-based financial data.
What would this mean for clients if it moves from trial to production?
First, better hit-rate prediction helps a dealer choose when to price aggressively versus defensively, improving the odds a client’s trade executes on the first quote instead of cycling through repeated quote requests.
Second, sharper probability estimates support smarter inventory and risk-limit allocation across simultaneous inquiries, which can free up balance sheet for more client flow.
Third, if more dealers run with tighter, better-calibrated quotes, investors should see narrower bid-ask spreads in routine sizes and faster responses in busy markets—quality-of-execution benefits that asset managers can measure against their policies for getting the best execution.
None of this changes the fact that market conditions, liquidity, and trade size still dominate outcomes; it does mean the model behind the quote wastes less information and time.
Important limits remain. The reported improvement is “up to 34%,” not a guaranteed constant; it was observed on HSBC’s data in European corporate credit, and results may differ by asset class, market regime, or trading venue.
Today’s approach inserts a quantum step into an otherwise classical pipeline rather than replacing it, and any live rollout would have to clear speed, reliability, oversight, and model-risk hurdles.
HSBC and IBM are explicit that this is a step toward practical use, not a claim that quantum systems are broadly superior to classical supercomputers across finance.
Even with those caveats, the trial matters because corporate bond RFQ is messy, high-dimensional, and noisy, precisely the kind of pattern-discovery problem where a larger computational search space can help.
If banks can safely productize combined quantum-classical features in pricing engines, clients should feel the impact not as “quantum” but as more dependable quotes and higher first-time fill rates. As Intallura put it, “We have great confidence we are on the cusp of a new frontier of computing in financial services, rather than something that is far away in the future.”
