Where does this fail? Claiming that a test bench of 250 assets, even from the S&P 500, is proof of large-scale portfolio optimization is ignoring the reality of markets.
The weak link is the extrapolation from a closed laboratory environment to the operational and regulatory complexity we have in France.
The practical implementation of this technology faces significant liquidity constraints for real trading volumes, a point often underestimated by academic research.
For example, an asset manager managing funds under the supervision of the AMF cannot simply ignore the market risk related to the liquidity of less common assets outside the S&P, an aspect not covered by this test bench.