Quantum Computing Gains a First Foothold in Investment Banking
IBM Research and JPMorgan Chase use quantum computing to predict the price of financial options
Up, down and up again — stock prices fluctuate wildly every day. Predict accurately the future economy, and you can sell your assets at their highs, or, if things go south, just move your money into tangible assets such as gold. To take predicting stock market wobbles to a whole new level, more and more banks are now turning to quantum computing.
The sums involved are often astronomical: last year alone, investors traded financial assets worth $70 trillion. Keen to gauge how much their investments might be worth in the future, financial institutions strive to get the best technology possible. Their predictions are not random — investors and traders have long used complex algorithms to forecast the movements of financial assets like shares and commodities.
One way to hedge and trade on the future value of financial assets is using options. An option is a financial tool — a contract that gives its owner the right to sell or buy an asset at a certain price, in the expectation that its market price at the time will generate a profit; however, there is no obligation to do so. It means that the owner can walk away with relatively little loss– or wait until the best moment and cash in.
But evaluating exactly when and by how much the price of an asset might shift is tricky. “The price in the future relies on an uncertainty model, a probability distribution usually taken from time series data,” says Christa Zoufal, a physicist at IBM Research.
Investors typically use the so-called Monte Carlo method to determine the probability of the future payoff while assessing potential risks. They analyze how prices of similar assets have been changing historically, based on various events such as the state of the economy, and create a computer model that gives the probability for the asset to have a certain value in the future. The program generates millions of possible outcomes and takes their average as the asset’s predicted future value.
“You want to evaluate whether if you buy an asset, what will be the expectation value of what you lose, in the worst-case scenario,” says Zoufal — in other words, the system is analyzing the maximum acceptable loss. Called the Conditional Value at Risk (CVaR), it’s the expectation value of the worst outcome of the probability distribution.
On a traditional computer, though, these simulations have to run overnight, meaning they don’t happen in real-time and are very costly. That’s where quantum computers could really speed things up — by using a lot fewer samples while achieving the same accuracy.
A standard computer works with silicon transistors that encode bits of memory as an on-off switch. These bits can be either 1, when the voltage in the transistor is on, or 0 — when it’s off. But a quantum computer relies on probabilities rather than certainties. Its memory consists of quantum bits (or qubits), tiny particles of matter that can be in all possible combinations of 0s and 1s. A coin tossed in the air can land as heads or tails; but while in the air, we don’t know what it is — or rather, it’s heads and tails at the same time, a superposition of heads and tails. Similarly, a qubit is in a superposition of all possible values 1s and 0s — so for two qubits, the states would be 00, 01, 11 or 10 simultaneously, so four probabilities in total. You won’t know the value until you measure it, with the measurement bringing the qubit from the quantum realm into the classical world.
That means that — in theory — a quantum computer increases its computational power dramatically with every qubit added to its memory size, which allows it to do a lot more calculations. As a result, finding a solution — or in the case of investments, the possible future value of an asset — can be done much faster than with a traditional computer.
None of the existing quantum computers are there yet, but a number of banks — among them JPMorgan Chase (JPMC) and Wells Fargo — have already started experimenting with the technology. Their aim is to get ahead of the game, so that they are in the starting blocks as soon as quantum machines reach so-called quantum advantage — the moment when they outperform a traditional computer in at least one useful task. Running programs on IBM’s quantum computers via the cloud as part of the IBM Q Network, JPMC aims to “estimate how long it might take for quantum hardware to progress to a stage that can deliver a practical advantage for use cases of value in our business,” says Nikitas Stamatopoulos, a quantum computing researcher at JPMC.
The idea, says Zoufal, is “to get quantum ready, to evaluate what possibilities there are and what applications are useful.”
The first results are already there. In a recent experiment, JPMC and IBM scientists used three qubits in one of IBM’s 20-qubit quantum computers to run a very simple quantum algorithm. They modelled possible stock prices to estimate the price of what’s known as a European call option — an option that allows the owner to buy an asset at a specific moment in the future. “What we developed is kind of a guide to option pricing on a quantum computer,” says Dr. Stefan Woerner, an IBM mathematician. “This quantum algorithm can achieve a quadratic speed up, meaning that it needs significantly fewer samples — so while for the classical Monte Carlo simulation you need millions of samples, for quantum you only need a few thousand.”
Thousands of samples may still sound daunting but, says Woerner, such computations will be much faster than those run today — instead of overnight, they’d be done in near real time. That would make it possible to react quickly to shifts in financial markets and for investors to make instant decisions.
Three qubits, though, only provide a limited number of possibilities, not sufficient in a realistic financial trading setup. But, says Woerner, the key here is that the algorithm ran on real quantum hardware, showing for the first time that it’s possible to price an option on a quantum computer. “We conceptually demonstrate quantum advantage, but at a scale where it’s not yet relevant,” says Woerner.
“If you scale the problem, the runtime increases slower, as it would on a classical computer. And then at some point if it’s large enough, the quantum really takes over.”
For that to happen, a quantum computer would need to have thousands of what’s known as logical qubits. At the moment, qubits are very noisy — meaning that any disturbance from the outside yanks them out of the superposition state they need to be in to keep on computing. Researchers are constantly trying to supress any noise (like vibrations or temperature fluctuations), by isolating the qubits in cryogenic tanks cooled down to 10 millikelvin (-273 degrees Celsius) — just a whisker above absolute zero and colder than outer space. In the future, 1,000 of such very low noise qubits would be used to make up one error-corrected logical qubit — but right now, none exist. “It will take years until we get there,” says Woerner.
The experiment isn’t just proof that it’s possible to use quantum computers for real financial market scenarios. It’s also, says Woerner, a success story of the IBM Q network — getting the banking industry and IBM to work together to solve a real-world problem on a prototype quantum device.
While quantum advantage may still be years away, with trillions of dollars at stake, the research shows that fintech is now set to get a major boost — from the quantum realm.
The paper can be found here: https://arxiv.org/abs/1905.02666
