Quantifying the advantages of applying quantum approximate algorithms to portfolio optimisation

Investors have long sought a way to build a portfolio of assets that minimizes risk while adhering to the messy realities of the market. In the ideal world of theory, one could buy any fraction of a stock or bond to achieve a perfect balance. In the real world, however, assets are sold in discrete chunks; you cannot buy half a share of a company or a fraction of a futures contract. This requirement for whole numbers turns the search for the safest possible investment mix into a notoriously difficult mathematical puzzle. Solving this puzzle, known as the discrete global minimum variance portfolio problem, is essential for quantitative analysts and fund managers, but it is so complex that even the most powerful classical computers struggle to find the absolute best solution quickly. As the number of assets grows, the number of possible combinations explodes, making the search for the perfect portfolio feel like looking for a single specific grain of sand on a beach that keeps growing.

A team of researchers at the University of Cambridge has explored whether quantum computers, which operate on the principles of quantum mechanics, can solve this specific financial puzzle more effectively than traditional machines. They developed a complete method using a quantum algorithm called the Quantum Approximate Optimization Algorithm, or QAOA. This approach does not guarantee a perfect answer every time, but it is designed to find a very good approximation by navigating the landscape of possible solutions in a way that classical computers cannot. The researchers built a full pipeline for this method, starting with how to translate the financial problem into a language a quantum computer understands, designing the specific steps the computer must take, and testing how well the system performs under various conditions. Their work provides a clear, end-to-end blueprint for how a quantum computer might one day handle the discrete nature of real-world trading.

The researchers began by figuring out how to represent the weight of each asset in a portfolio using binary variables, essentially converting the problem into a series of on and off switches. They then designed a specific set of instructions, known as an ansatz, which guides the quantum computer through a sequence of operations. This process involves preparing an initial guess for the portfolio, applying a cost function that measures how risky a particular mix is, and using a mixing operator to shuffle the possibilities around in search of a better solution. Crucially, they designed a "hard-constraint" mixing operator that ensures the computer never considers invalid portfolios, such as those that do not add up to the total budget or violate the rule that assets must be bought in whole units. This constraint is vital because it keeps the search focused on realistic investment strategies rather than wasting time on impossible scenarios.

To test their method, the team ran extensive numerical simulations on models of financial markets with varying numbers of assets and different levels of precision. They compared several strategies for finding the best settings for their quantum circuit, testing different types of initial guesses and optimization routines. They found that starting with a "warm-started" state—an initial guess derived from a simplified, continuous version of the problem—often led to better results than starting with a completely random guess. Furthermore, they discovered that a specific optimization technique called dual annealing, combined with a layer-by-layer approach to building the quantum circuit, provided the most robust performance. This combination allowed the algorithm to navigate the complex landscape of solutions more effectively, even when the data was noisy.

The simulations revealed a promising trend regarding the efficiency of the quantum approach, though with important caveats. When the researchers looked at how the number of measurements required to find the best solution scaled as the problem grew larger, they observed a favorable pattern specifically when using the warm-started initial state. In this regime, the number of measurements needed to find the global minimum—the absolute safest portfolio—grew much more slowly for their quantum algorithm than it would for a standard method that simply samples random valid portfolios. This suggests that for large, complex portfolios, the quantum method could eventually require far fewer attempts to find the optimal solution than constrained uniform sampling methods. However, the researchers noted that this scaling advantage was not observed with the max-bias initial state, and that constrained uniform sampling only provides a lower bound for classical sampling complexity, leaving comparisons to improved classical algorithms for future work.

However, the study also delivered a sobering reality check regarding the current state of hardware. When the researchers introduced realistic levels of noise, specifically thermal relaxation which mimics the way quantum states decay over time, the performance of the algorithm dropped sharply. The noise was strong enough to obscure the signal, making it impossible to see any advantage over classical methods with current technology. The researchers found that even with techniques to filter out invalid results, the noise required so many additional measurements that it negated any speed benefit. This indicates that for quantum finance to become a practical reality, the error rates in quantum hardware must improve by several orders of magnitude. Until then, the theoretical advantages remain just that: theoretical.

Despite the hardware limitations, the work offers a significant step forward in understanding how quantum algorithms can be applied to finance. The researchers demonstrated that a complete, end-to-end solution for the discrete portfolio problem is possible and identified the specific components, such as the hard-constraint mixing operator and the dual annealing optimizer, that make it work best in simulation. They showed that while current machines are too noisy to be useful for this task, the path to a future advantage is clear. If hardware improves to the point where thermal noise is no longer a dominant factor, the stochastic measurement noise will become the primary challenge. In that future regime, the favorable scaling observed in their simulations—specifically when utilizing the warm-started initial state—suggests that quantum computers could indeed provide a powerful tool for managing risk in complex financial markets, offering a way to navigate the discrete constraints of the real world with a speed that classical sampling methods cannot match.