Quantum Trading: How HSBC and IBM Investigated Quantum to Predict Bond Markets

HSBC has claimed the world’s first known evidence of Quantum Computing’s real-world value in algorithmic bond trading, following a collaboration with IBM.

By combining Quantum and classical computing techniques, the collaborative trial claimed up to a 34% improvement in predicting the probability of winning customer inquiries in Europe’s corporate bond market.

It’s a result that aims to signal the growing practical potential of Quantum technology in finance. HSBC and IBM claimed that the resulting models delivered up to 34% higher probability of accurately predicting whether trades would be filled.

Quantum Computing in Finance

In a highly competitive landscape, every conceivable advantage could make massive differences to the bottom line. These experiments were conducted with production-scale Request for Quote (RFQ) data and IBM's Heron processors.

"It means we now have a tangible example of how today's Quantum Computers could solve a real-world business problem at scale," said HSBC's group head of Quantum technologies, Philip Intallura.

Interestingly, their research indicates that the performance boost did not appear when researchers tested the system on a perfect, simulated Quantum Computer. It only showed up when they used real Quantum machines, which are still prone to random errors known as “noise.”

In this experiment, which you can read all about in this study here, the researchers claimed that the noise seemed to help the model make better predictions. It’s an unexpected result that would imply that the hardware imperfections might sometimes aid learning from complex financial data.

Currently, it’s too early to confirm this effect and these results are being validated, and peer-reviewed (both formally and informally) by the scientific community.

The Science of Knowing If a Deal Will Stick

In bond trading, many transactions happen through an RFQ, or Request for Quote. It’s a system where investors ask multiple dealers to quote a price for buying or selling a bond, then choose the best offer.

These RFQ-based markets demand fast, accurate predictions about whether a quoted price will actually lead to a completed trade. In ruthless, high-frequency environments like corporate bonds, cryptocurrencies, stock trading, and forex markets, a bad estimate can cause poor spread management or costly inventory risks. In short, it’s bad for business.

Standard models have plateaued somewhat (see Cadez & Smyth’s 2013 study on diminishing returns in model complexity).

Even when analysts try tricks like adding more features, combining different models, or using deeper neural networks, the gains can be minimal. The problem is the data itself: it’s patchy, price movements are irregular, and the inputs are messy and complex.

Most traditional fill prediction systems rely on algorithms optimized for classical computing infrastructure. These algorithms are typically constrained by the structure of the data and the assumptions built into classical methods.

Hybrid Quantum-Classical Architecture at Scale

The HSBC-IBM team applied a structured methodology. Classical trade event vectors were cleaned up and normalized, and then passed through what they call a Projected Quantum Feature Maps (PQFM), a Quantum embedding and measurement scheme.

The aim of this step is to reframe the data in a way that highlights patterns classical methods might miss.

The transformed data was then used to train traditional ML models: logistic regression (a simple probability-based model), XGBoost (a boosting algorithm for accuracy), random forests (many decision trees voting together.), and neural networks.

Key outcomes claimed by the collaboration:

• 34% median predictive accuracy improvement using hardware-generated Quantum features
• Consistent outperformance across four model types
• The natural “noise” in today’s Quantum Computers seemed to act like a filter, smoothing out the data and making the model’s patterns clearer.

This architecture was tested on over 1 million RFQs and 9 million trading signals across 294 trading days. The researchers then tested the system by running it through detailed “what if” scenarios that replay real bond trading patterns, to see how it would have performed in practice.

The process takes trading data and applies core Quantum physics concepts to them: encoding, measurement, and probabilistic outcomes, applied directly to financial data.

Quantum Observations for Traders

Better fill prediction would lead directly to tighter spreads and higher-quality trade execution. The hybrid pipeline introduced by HSBC and IBM requires no change to existing model architecture. Quantum-generated features are modular and can be slotted into existing statistical learning systems.

Even if it can’t run trades in real-time yet, this approach would still be useful for day-to-day trading analysis.

An improvement in predicting the fill probability would help traders and liquidity providers make sharper decisions during the day or when closing out positions; especially in fast, punishing timeframes in unpredictable markets where small mistakes can quickly eat into profits.

This methodology could then be adapted to other markets such as cryptocurrencies, forex trading, and stock trading. It also opens the door to expanded applications using TradingView, Ninjatrader, Metatrader, and other trading tools designed for fast-moving environments.

Traders working with volume indicators and volume price analysis (VPA) could get an extra edge if the data is cleaned and structured better before it goes into their models. More robust signal extraction at the input level can complement existing trading indicators.

Think of it like cleaning the lens before taking picture: the clearer the input, the more reliable the signals traders see from their usual tools.

What This Means for the Quantum Industry

This project claims to be one of the first real-world examples showing that Quantum Computers can actually outperform traditional methods in a business setting. The researchers claimed that the experiment worked only on an actual Quantum Computer, where natural hardware noise is present, and not in clean, simulated tests.

Therefore, the collaboration implies that the result suggests noise itself may function as a statistical smoothing force in complex, unpredictable systems like financial markets.

Projected Quantum Feature Maps functioned as standalone components within a modular pipeline.

IBM Heron’s cloud-based architecture enabled reproducible access and consistent results across three separate hardware systems. The study argues that these results make a strong case for pushing Quantum Computing deeper into real-world financial modelling.

Applications could include improved signal design for trading algorithms, as well as offline modeling for regulatory strategy simulations, such as responses to potential changes inCFTC rule implementations.

“Quantum computers are starting to become useful tools to explore modeling in financial markets, and in particular in the quantitative investment space,” Dr Manuel Proissl, IBM Quantum Industry Applications Lead, said of the discovery.

Friction and Fractures in the Narrative

The most controversial aspect of this research is the dependence on noise. Noiseless Quantum simulations showed no advantage. Only noisy hardware delivered improved results. It’s not clear if this can be replicated reliably, or if the observed gains are artifacts of hardware idiosyncrasies.

Because the methodology, code, and data are proprietary, only HSBC and IBM have access to the full details of how this experiment was done. That means outsiders can’t yet fully review or replicate the results independently.

For any forex trader or quant desk evaluating Quantum approaches, the lack of reproducibility remains a critical barrier. These findings are preliminary and need to be properly tested to their limit and validated across other assets and market environments before the true sea change occurs.

Quantum Applications in Retail and Institutional Trading Stacks

The HSBC-IBM framework aligns with workflows familiar to traders using TradingView, Ninjatrader, and Metatrader. The system can be configured to integrate with Volume Price Analysis (VPA) systems, volume indicators, and a variety of trading indicators and trading tools.

Quantum processing could create new kinds of signals from the data, giving traders extra layers of insight for predicting price movements. When paired with data feeds from social media, Google Analytics, and live order flow, these signals may augment how traders interact with timeframes, RFQ responses, and predictive analytics.

Quantum Computing Hoping to Find a Real Edge in Finance

The HSBC-IBM Quantum bond trading experiment aims to move Quantum Computing further into practical territory for institutional finance. The collaboration claims that the results offer early evidence of utility in forecasting, even without fault-tolerant Quantum systems.

While the findings are bounded by hardware and methodology, they suggest that Quantum processors may no longer restricted to theoretical research.

Traders working across forex markets, cryptocurrencies, and corporate bonds should monitor this space closely for more research updates, given the inherent risks of being left behind in such a cutthroat space. As Hemingway put it: “Gradually, then suddenly.”

To find out how QuantumBasel can help your organization navigate the Quantum and AI future, contact us here.