Sara Garcia

Le maillon faible là-dedans est l'extrapolation d'un banc d'essai à 250 actifs vers une application réelle.
Testé sur un sous-ensemble du S&P 500 avec 64 qubits, c'est une démo, pas une preuve de concept opérationnelle face aux contraintes de liquidité et coûts de transaction.
Les marchés exigent des milliers d'actifs, des instruments complexes et une tolérance à l'erreur proche de zéro, ce qu'un prototype ne peut pas gérer.
Par exemple, la gestion des positions "0DTE" (options à expiration journalière) où la latence est cruciale montre que la puissance brute ne suffit pas sans gestion des risques et des coûts de glissement.

Where does this fail in the Swiss markets? A benchmarking conducted on only 250 S&P 500 stocks is not sufficient evidence to claim large-scale portfolio optimization.
The weak link here is that this test ignores overall diversification and the asset complexity typically found in Swiss mandates, for example.
Our portfolios include much more than U.S. stocks: we manage thousands of instruments, from trust bonds to commodities, including exotic currencies and derivative products across numerous jurisdictions.
Confirming the robustness of such a system would require, at a minimum, a simulation including liquidity risks specific to each market and local regulatory constraints.
For a Swiss private bank, a portfolio of 250 stocks is anecdotal, whereas operational reality demands dynamic optimization across tens of thousands of global positions.

The single point of failure of this statement is the sample size used for validation. A universe of 250 assets, even from the S&P 500, is an excessive simplification that does not reflect the complexity and granularity required for large-scale portfolio optimization. Swiss hedge funds manage investment universes much larger, including thousands of securities and derivative instruments, making this scope too limited. Extrapolating the performance of a 64-qubit system on such a restricted sample to portfolios of several billion ignores scalability risks and real operational constraints. For example, a pension fund managing hundreds of diversified strategies could never rely on such a small-scale benchmark for its critical investment decisions.

The single point of failure of this approach is the transferability from a laboratory environment to a real market. A test bench of 250 assets from the S&P 500, although admirable academically, does not account for the daily noise of markets. The stability and error correction necessary for the portfolio management of a Swiss pension fund, with its mandates of capital preservation, are far from being achieved with a 64-qubit system. This type of quantum optimization is a classic case where the attack surface of practical constraints is ignored.