If qubits—the building blocks of a quantum computer—were humans, no-one would employ them. They would be unreliable. They would forget everything they learned. Only trivial jobs could be entrusted to them, and then only after painstakingly marshalling them into teams. On top of all that, they would insist on meticulously prepared and maintained workspaces.

The work itself, though, would happen at lightning speed.

That promise—of a literal quantum leap in computing power—has moved beyond the purely theoretical to the halfway practical, with banks and specialist developers now testing its use in everything from derivatives valuation to the management of liquidity buffers. Using the technology in the real world, though, will require fixes for all those qubit niggles—the behavioral quirks that make them so frustrating to work with.

Practical quantum computers can’t be used for a lot of tasks that we typically face in 98% of financial domain applications

Jezri Mohideen, Nomura

Lee Braine, director of research and engineering at Barclays, casts this as “a voyage of discovery that could lead to radically novel solutions for some of today’s greatest computational challenges”.

Others see it as a slog. Some banks are now reining in their expectations—and potentially their investment of time and money.

The head of digital at one investment bank says: “It always feels like we’re two years away from a real quantum use case. The order of magnitude of the efficacy of the processing just wasn’t where everybody thought it would be.”

Jezri Mohideen, global chief digital officer for Nomura’s wholesale business has also become more sceptical, at least in the short term. He says the bank may wait before becoming an adopter: “Two years ago, I was a lot more bullish, and I felt the evolution would come through a lot faster. Practical quantum computers can’t be used for a lot of tasks that we typically face in 98% of financial domain applications.”

To tell this story—of quantum’s huge potential in the risk management and risk transfer markets, and of the barriers to it—WatersTechnology's sibling publication Risk.net has spoken to more than 20 sources, including banks that have tested the technology, hardware manufacturers of the kit and software developers devising uses for it. Everyone is in on the act, from domestic and regional players such as Spain’s BBVA, to global titans such as JP Morgan.

The choice each of them has to make is how and when to engage with technology that may still be a decade away from being useful to their businesses.

Qubit quirks

Understanding the practical obstacles in quantum computing requires some understanding of the qubit.

While a bit within a classical computer can be in one of two states—storing either a one or a zero—a quantum bit, or qubit, can exist in both states simultaneously, a phenomenon known as superposition. This means quantum computers can handle a vast number of calculations. The behavior of qubits can also be correlated—or ‘entangled’—so observing one qubit can tell you something about the state of others, cutting the time required to collect information from the system.

These qualities make quantum computers ideal for handling complex calculations with large numbers of variables. Or, they would, if qubits could be trusted.

Classical bits retain their states—zero or one—because the materials and electronics in classical computers provide stability. If a bit changes state in error, it can be detected and fixed. In quantum computing, errors can be caused if qubits get too warm, or as a result of stray microwaves and photons, or manufacturing defects. As a result, they have to be refrigerated and shielded from the outside world.

They also have to be knitted together—multiple ‘physical qubits’ are required to produce one ‘logical qubit’ that has a low error rate and can be reliably programmed.

IBM’s current machines max out at 65 physical qubits, in its Hummingbird device, debuted this year. In September, the IT giant laid out ambitious plans to build a 127-qubit processor next year, a 433-qubit device in 2022 and a 1,121-qubit machine by the end of 2023.

This would be dramatic progress, but still not enough for quantum machines to handle tasks that are beyond their classical cousins—for this, thousands of qubits will be required.

In addition, more robust quantum random access memory is required, to permit better storage of data during applications—essential for data-heavy use cases such as machine learning.

Until these upgrades arrive, today’s generation of machines—catchily known as noisy intermediate-scale quantum (Nisq)—have some pretty serious limitations.

“The algorithms we devise are still running on fictitious data,” says Marco Pistoia, head of research and engineering at JP Morgan.

“We’re not really analyzing the data of the bank, because the computer would not be able to perform a computation on such a large input. We have to wait for the hardware to make progress.”

These proofs of concept may be limited, but they can still be valuable.

“It doesn’t mean the result is trivial—far from it,” says Braine at Barclays, which has run its own proofs in the area of netting and settlement, with work on front-office machine learning to come. “You’re actually performing often novel research with original contributions to knowledge. But moving from that to deploying it in industry is a very big step.” The hardware necessary to scale such solutions is five to 10 years away, he reckons.

This creates a conundrum: in order to capitalize on quantum technology, banks need some level of expertise—as well as answers to some basic questions about which applications to prioritize. But how much energy should they expend before the technology reaches its tipping point?

Dutch bank ABN Amro is pooling resources with two of its compatriots, ING and Rabobank, as a way to explore quantum—but limit the cost—as the technology develops.

“The amount of resources that you get for doing quantum projects are not sky-high—and, due to Covid-19, are even lower. So we are finding common ground and working together,” says Dimitri van Esch, quantum lead at ABN Amro. The trio has been looking for a quantum short-cut to regulatory stress-testing.

At the other end of the spectrum, JP Morgan created a new ‘lab’ late last year to explore quantum and other emerging technologies, poaching Pistoia to lead it. The 24-year IBM veteran left his former employer with the title ‘master inventor’ because of his record of producing valuable patents. At the US bank, he has been let loose on cutting-edge quantum hardware from Honeywell that seeks to make today’s little knots of qubits more productive—a technique known as mid-circuit measurement and reuse.

Pistoia says he was surprised, after arriving from IBM, that JP Morgan had access to such advanced technology. One of the early fruits of his labors is a co-authored research paper about how to write algorithms for use on the Honeywell device, published last month.

“JPMorgan Chase has been the first research organization worldwide, as far as we know, to devise an algorithm that takes advantage of this,” he says.

The advance means the bank “can now measure a qubit in the middle of a circuit and then reuse it. In the previous hardware environment, the measurement had to be done all at the end. So the qubits can yield more power than before, because they can be reused”.

Pistoia says the research, published on September 14, is particularly important in quantum algorithms for tasks ranging from options pricing to machine learning.

Inside the lab

With so many banks conducting research, the number of future applications for quantum computing is piling up. In most of these cases, banks are renting Nisq machines from a manufacturer—IBM has around two dozen—accessing them through the cloud, and applying them to toy datasets.

Some of the early results have been eye-popping.

NatWest discovered in one experiment that it was 300 times faster to optimize its £120 billion ($155.2 billion) high-quality liquid asset (HQLA) portfolio using a quantum algorithm. But the test could only be applied to a test portfolio of around 90 different assets, rather than the real-world HQLA book, containing in the region of 500.

Jas Oberoi, director of machine learning and quantum-inspired solutions at 1Qbit, NatWest’s Canada-based software developer partner, says: “We are waiting for the hardware to scale to a point where a practically useful problem can be solved better than traditional computing offers today.” Like Braine at Barclays, he believes this tipping point is five to 10 years away.

There are many other possible uses for the technology. Monte Carlo simulations appear throughout finance whenever a firm needs to derive a probability distribution from an array of possible paths—in stress-testing, derivatives pricing and portfolio risk models, for example. The equivalent of these calculations can be carried out on a quantum machine via what’s called an amplitude estimation algorithm, only far more rapidly.

JP Morgan’s Pistoia describes this as “low-hanging fruit”: “Imagine if you have to search through a million elements. With a quadratic improvement in complexity, the number of iterations of the algorithm is reduced to only 1,000.”

Imagine if you have to search through a million elements. With a quadratic improvement in complexity, the number of iterations of the algorithm is reduced to only 1,000

Marco Pistoia, JP Morgan

And time is money. If banks are able to obtain quicker answers to complex questions, then the resulting information can inform their decision-making, says Stefan Woerner, quantum applications research and software lead at IBM Quantum, a manufacturer of quantum computers.

He gives the example of economic capital calculations for a large portfolio of assets. A paper co-authored by Woerner last year showed that a yet-to-be-built, fault-tolerant quantum machine would estimate the risk and the required capital for a portfolio more precisely than a classical computer, while delivering results in less than a minute, rather than hours or days.

He tells Risk.net: “If a bank can do this risk analysis for its whole portfolio in real time, it can provide much more competitive prices.”

CogniFrame, a Canada-based algorithm developer that leverages quantum computing in a hybrid model with classical methods, has worked on a host of research projects and applications, including an attempt to improve margins on an energy trading desk, optimization of a hedging portfolio for a currency trader and exotic options pricing for a large information services provider.

CogniFrame founder and chief executive Vish Ramakrishnan says: “Pricing exotic options was] too much for the hardware that’s available today, so we’re only able to do it more as a proof of concept. Even a basic options basket would require hundreds of [logical] qubits to have reasonable accuracy. That’s definitely not going to be happening in the next three or four years. It’s probably 10 years away.”

CCAR, CVA and VAR

While quantum-powered options pricing may be out of reach for another decade, some experts argue Nisq devices can start picking up simpler work in the much shorter term.

US-based software developer Zapata Computing has applied for a patent on the application of its quantum software to the US Federal Reserve’s annual stress test of large banks, the Comprehensive Capital Analysis and Review (CCAR). This is essentially the same use case that is being explored by the trio of Dutch banks.

The developer’s chief executive, Christopher Savoie, claims Zapata can complete the tests “much more quickly and much more accurately” on today’s quantum devices.

“We figured out how to do [CCAR] on a Nisq computer,” he says. “And that’s significant because if you can do that, even a constant factor speed-up will significantly reduce the time that’s spent doing those simulations and the money that’s spent on compute.”

The key is to hand over some of the underlying components of the testing to Nisq machines, Savoie adds: “Some of the subroutines can have a quantum equivalent that can do these highly complex multivariate analyses quicker and with greater accuracy.”

In a separate experiment, Zapata has worked with Spanish bank BBVA, using quantum versions of Monte Carlo simulations to calculate derivatives prices that incorporate a counterparty risk adjustment—one of six separate applications the bank began exploring last year.

Savoie admits more qubits are needed to scale that particular use case. However, describing its work in July, BBVA’s global head of research and patents, Carlos Kuchkovsky, claimed some quantum applications could be mature enough to use within as little as two years.

As well as testing particular applications, banks can prepare for a quantum future in other ways. Algorithms that run on classical computers need to be rewritten for the radically different physics of a quantum machine. France-based firm QuantFi has trained quants at banks including BNP Paribas to write quantum algorithms that will solve risk management problems such as value-at-risk calculations.

Such sums can be completed in 100 steps on quantum technology, rather than 10,000 steps on a classical computer, claims QuantFi co-founder Paul Hiriart, “but none of the banks we know have decided yet to launch a project for go-live”.

Why? The integration of quantum technology into a bank’s internal systems alone would take at least two years, he says, adding that the pandemic has also had an impact on bank spending priorities: their investment in quantum technology has been postponed by a year in some cases.