If there’s an industry steeped in computations, it’s the financial services sector. Optimisation problems, for which a whole chorus of variables must be fine-tuned and modulated, routinely plague financial firms, especially when it comes to highly engineered financial products such as those developed through quantitative analysis.
That need for complex mathematical modeling at scale makes the finance industry a perfect candidate for the promise of quantum computing, which makes (extremely) quick work of computations, including complex ones, delivering results in minutes or hours instead of weeks and months.
But beyond speed, quantum’s ability to deliver accurate knowledge in reasonable time frames is what makes it especially valuable, says Benno Broer, chief commercial officer at PASQAL, a full-stack quantum services company. “We want to know how to price something and do it more accurately and we want that answer in the next hour. With the classical computer, that would take two weeks and the trade [would be] gone,” he says.
Doing “computation better” is one of the reasons why a joint team of engineers and financial analysts at Ally Financial turned to quantum. Their focus was traditional Exchange-Traded Funds (ETFs), which comprise hundreds of thousands of equities that make up a certain return over a period of time. The premise is that even if a few individual components underperform, the rest buoy the net result, delivering a fairly predictable return over a fixed amount of time.
But managing and manipulating so many component parts of an ETF is painful: “There’s a lot of variability in terms of stock performance and there’s a lot of buying and selling and related transaction fees,” says Sathish Muthukrishnan, chief information, data, and digital officer at Ally.
Muthukrishnan and the Ally team explored whether they could deliver similar returns on an ETF that had fewer component parts. To do so, they explored the optimisation problem of “cardinality constraints” and developed a hybrid quantum-classical approach to financial index tracking portfolios that maximises returns and minimises risk. Their work earned them a 2023 US CIO 100 Award for IT innovation and leadership.
The Ally team used a method called quantum annealing that helped them settle on a choice few equities. “You’re able to select a smaller number of stocks with predictive return and lower operational and transaction costs, which ultimately means that you can reduce the variability and more accurately predict returns,” Muthukrishnan says.
Because of quantum’s abilities the Ally team could create 50 separate scenarios and back-test the models. Such rigor also highlights flaws in the models used for traditional computing and helps industries develop more robust foundations for data-related research, Muthukrishnan says, a happy byproduct.
Quantum’s unique edge
Ally’s work with ETFs is just one example signaling what could be a sea change for the industry, thanks to quantum’s computing power.
Cirdan Group, a European banking group that offers investment solutions, is another financial organisation putting quantum computing to work, teaming up with quantum-as-a-service company Terra Quantum on a computational challenge for its investment solutions. The specific focus for the partnership was exotic derivatives, which are uniquely challenging because they are represented by mathematical functions with no closed-end formulas.
“There’s no [simple] equation that says X plus Y equals the value of the derivative,” says Antonio de Negri, CEO at the Cirdan Group. As a result, “we have to calculate our derivatives by running thousands and thousands of Monte Carlo simulations,” de Negri says. Traditional high-performance computing (HPC) finds the process to be backbreaking and time-consuming. But financial institutions like Cirdan have been slogging along, conducting these tedious calculations anyway because they help understand asset risk and manage it more efficiently.
Cirdan had been throwing the best computing muscle at the problem but calculations were time-consuming and expensive. Terra conducted the same optimisation calculations with quantum and delivered a 75% reduction in computing time — from 10 minutes to 2 minutes — with the same accuracy. Future iterations are expected to deliver even greater economies, de Negri says.
Eight minutes might not seem like a lot, but it can translate to large gains. “Given the size of the portfolio or the size of the bets that [financial institutions] are making, if you can do a computation in a slightly better way, 0.1% better or faster than you could, that already means a significant upside,” says Broer, who remembers when he used to work with Excel-based models to evaluate portfolio risk.
“We started our models, pressed run, and had to wait until the morning and hope they hadn’t crashed overnight,” Broer says. It was Monday-morning quarterbacking at its best because you could only tell if you had been over the risk limit for past transactions.
“It’s quite easy to run into the limits of classical computing if you have abundant data, many different assets that you’re trading, or many different clients you’re providing a loan to,” Broer says.
Where quantum can provide impact
Optimisation problems like the ones that Ally and Cirdan have tackled, are well suited for quantum computing because classical computing lacks the capacity to deliver meaningful results in a reasonable time frame. But quantum also holds the promise to make machine learning more efficient as well, says Vishal Shete, managing director UK and head of commercialisation at Terra Quantum AG.
Because qubits, the building blocks of quantum, “can learn with much less and noisier data, they’re very efficient at learning,” Shete says. This means quantum can take on machine learning challenges with fewer constraints than traditional HPC demands.
Nilesh Vaidya, executive vice president at research and consulting firm Capgemini, agrees about the value of quantum for machine learning. “Applying machine learning techniques using quantum computing capability prepares the models better and faster,” Vaidya says. “Today, it takes a while to create and deploy models and visualise the outcomes, but with quantum some parts of it can be greatly accelerated.”
In addition to the technical feasibility qualifications, Shete advises enterprises to cherry-pick projects where even a little improvement can lead to good business value. Stakeholder interest is also key. “You could have all the other factors fitting in but if the business unit lead that you’re working with is either resistant or unwilling to change, that’s a stumbling block,” Shete says.
“If you understand the strengths and weaknesses of quantum, then in each field you can find a good niche where you can add large value,” Broer says. “But if you assume it can add value to everything then you’ll be very disappointed. It’s like a hammer looking for a nail, it’s going to be a lot of work to find that nail but once you have it, you can get started.”
Industry-quantum provider partnerships
How exactly to get started? While a few financial institutions are building their quantum teams from the ground up, many are choosing to partner with experts in the field.
Markus Pflitsch, Terra Quantum’s CEO and founder, argues that “it’s just not feasible for banks and other industries to build quantum capabilities in-house given the dearth of talent.”
In addition to providing access to “best-in-breed” quantum hardware, firms such as Terra Quantum can run quantum software on in-house simulators based on classical HPC components, which is how Cirdan addressed its exotic derivative problem. When quantum computers move beyond the Noisy Intermediate Scale Quantum (NISQ) devices they occupy today, the Terra Quantum software can also translate to those platforms.
Shete points out that quantum specialists can also cross-pollinate solutions from different industries. For example, “the simulation work we’re doing in options pricing has got lots of similarities with work that can be done in molecular simulation in chemical companies,” he says. A quantum-only company might seed ideas borrowed from one sector across the board, Shete suggests.
The future with quantum
One machine learning challenge Terra Quantum is currently working on involves understanding customers with time-series prediction models: “It’s about predicting customer behavior, really understanding how customers will react, what is the best grouping of different customers, what the correlations are and how they should be put together, and hence what are the best products customers should be nudged toward,” Shete says.
In markets, time-series predictions help understand how markets will behave and evaluate correlation between different types of assets. And in risk management, quantum can be deployed “for Monte Carlo simulations or understanding anti-money laundering or compliance issues that might be happening within your bank,” Shete says.
For its part, Ally expects to evaluate more quantum-related projects in the future, including credit loss modeling, where one can predict what percentage of loans granted to customers might end up as losses. The proof-of-concept projects Ally has conducted so far are its trial run for when quantum is ready for prime time.
“It’s important for us to test the technology and be ready,” Muthukrishnan says. “It’s like constantly working out and doing your sprints so when the real race happens, you’re ready to go. You can’t sit around and wait for things to happen — it’s all about consistency, preparation, and then being able to rise to the occasion when the time is right.”