Conclusion
When scaling to larger portfolios, computing the optimal result might be too expensive compared to quantum approaches. Still, as we have seen, even when quantum-computing those large combinatorial problems, they come at the cost of needing a complete certainty of the outcome.
It is important to understand that these techniques require, as happens in traditional machine learning approaches, a good understanding of how the best architecture for our ansatz plays in our favor. And in many cases, this will come from the experience of fitting against different types of portfolios and stock combinations. Not all assets show similar behaviors. This will require exploring the vast extension of potential ansatzes, repetitions of schemes in those ansatzes, and optimization techniques that require fewer iterations to find the best parameters.
Even though gate-based quantum devices may offer a generalist approach to quantum computation, it is undeniable that, nowadays, quantum...