Public research brief
A stock with a short history should not force us to throw away everyone else's long history.
Bongiorno and Villassero study portfolio risk estimation when assets have unequal or incomplete return histories. Using every pair's available overlap preserves information, but the combined correlation matrix can become mathematically unsuitable for optimization.
Their neural estimator repairs and regularizes that structure while training for realized portfolio risk. The reported gains are substantial. We have not reproduced them; below the paywall is the comparison we would require before adopting the method.
Full analysis on Substack
Follow the evidence further.
The complete review examines the findings, implementation limits, and research questions beyond this public brief. Access and subscriptions are handled on Substack.
Continue reading on Substack ↗View subscription optionsAI assists curation and drafting. The research status distinguishes paper reviews from house tests. Read our research approach.