Researchers on both sides of the Atlantic—including a trio of Dutch banks—believe quantum computing can help banks produce quicker, more accurate results in regulatory stress tests.

In the Netherlands, ABN Amro, ING and Rabobank are jointly exploring quantum technology, with stress testing one of the possible use cases. US-based quantum developer Zapata Computing, meanwhile, is seeking a patent that applies the technology to the Federal Reserve’s twin-track domestic testing regime: the Comprehensive Capital Analysis and Review (CCAR), and the Dodd-Frank Act Stress Tests (DFAST).

“If you talk to chief information officers of banks in the US, one of the major demands on high-performance computing—and on the computing budget—comes from CCAR and DFAST,” says Christophe Savoie, chief executive at Zapata Computing.

He says quantum machines will be able to deliver test results “much more quickly and much more accurately”.

The Dutch banks have the same conviction.

“Currently, it takes a long time to run the required number of Monte Carlo simulations. The more we can do, the better our predictions will be,” says Dimitri van Esch, quantum lead at ABN Amro.

Other banks are exploring the use of quantum technology for a range of tasks, all requiring a vast amount of conventional computing power, from derivatives valuation and options pricing to settlement problems and fraud detection. A technical paper published by WatersTechnology’s sibling publication Risk.net last year showed portfolio optimization could be carried out up to 1,000 times faster when using quantum machines.

Bit of a problem

The key to the technology’s leap in performance is the quantum bit, or qubit. While a bit within a classical computer can be in one of two states—storing either a one or a zero—a qubit can exist in both states simultaneously, a phenomenon known as superposition. This means quantum computers have the capacity to run more calculations at once, which could be a big help in stress testing.

Testing regimes vary between nations, but typically require banks to take a prescribed set of shifts in macroeconomic and market factors, and then estimate the impact on their existing portfolios. Estimations are often generated by running large numbers of time-consuming Monte Carlo simulations.

As an illustration of the computational lift, the European Union-wide stress testing exercise carried out in 2018 required banks to complete 38 different spreadsheets. One of these spreadsheets—covering forecast credit risk losses—had more than 9,000 rows and 78 columns.

If you talk to chief information officers of banks in the US, one of the major demands on high-performance computing—and on the computing budget—comes from CCAR and DFAST

Christophe Savoie, Zapata Computing

It will be some years before this kind of work can be handed over to quantum technology in its entirety, experts believe. Quantum computers cannot be bought today. The only machines that can be rented for research—from manufacturers such as IBM—are an embryonic form of the technology, known as noisy intermediate-scale quantum, or Nisq. Each contains just a few dozen qubits—as more are added, the machines become less stable and their error rate increases.

Despite these problems, researchers believe stress testing is one application that is attainable in the relatively near term.

“The speed-up depends on the number of qubits, but at least you can run it with a limited number of qubits,” says ABN Amro’s van Esch.

Savoie says Zapata has found a way to get help from even today’s immature quantum machines. 

“We figured out how to do [CCAR] on a Nisq computer,” Savoie 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, he adds: “Some of the subroutines can have a quantum equivalent that can do these highly complex multivariate analyses quicker and with greater accuracy.”

“Four or five years”

Banks may have to confront more than technological obstacles if they want to use quantum computers to perform stress test calculations. The results of the exercises are used to set capital requirements and, in the US, to sign off on dividend payouts and share buybacks. The models used to generate the test results are also heavily scrutinized, and regulators may not be comfortable if the modeling is carried out on brand-new, possibly wobbly technology.

The Dutch banks have already briefed domestic regulators on their plans, in an attempt to address possible objections before the project reaches fruition.

ABN Amro’s Van Esch says: “We believed it would be helpful for them to be connected to this project and know what we’re going to do, so they can prepare for any kind of law change or regulatory change that might be needed. Otherwise, it might take us four or five years to build an algorithm, and then we’ll have a delay for another four or five years because a law needs to be changed.”

At Zapata, Savoie says regulators ought to take a close look if a new form of stress simulation had been designed specifically for a quantum computer: “If there were a new quantum algorithm or technique, we would want that to be subjected to some regulatory testing.”

However, if the same calculation was just speeded up by handing off the subroutines, then it might not require the same type of scrutiny, he argues: “You’re doing the same math to get to your answer.”