One Basket, Many Experts: Decorrelated Re-ranking for Financial Asset Recommendation (arXiv (Cornell University))

Summary

• DSER is a plug-in re-ranker that replaces only the lowest-ranked items of an expert-following list with decorrelated assets.p.2
• At m=4, the Sharpe measure rises from 1.794 to 2.119 while nDCG falls from 0.391 to 0.362.p.9
• Across 6 non-overlapping windows, the Sharpe change is positive in 4, while the relevance change is negative in all 6.p.9
• DSER's deterministic 2.12 sits at the 80th percentile of the random-fill distribution.p.7
• The walk-forward protocol uses 61 bi-weekly checkpoints.p.5

Key insights

1. Recommendations that best match user behavior concentrate on a few popular assets whose returns are highly correlated.p.1
2. DSER keeps a fixed relevance core and re-ranks only m satellite slots, filling each with the asset of minimal correlation to the current basket.p.3
3. The minimum-variance fill captures the least of the gain, with DSER +0.24 SR over it at m=4 and +0.06 at m=2.p.7
4. The relevance cost falls harder on HR than on nDCG, because HR is a binary hit metric.p.6
5. The Sharpe premium is market-phase dependent, whereas the relevance cost is not.p.7
6. With X=10, the m=4 gain vanishes at the default 180-day lookback and reverses at 360 days.p.8
7. The m=2 gain survives realistic trading costs at retail scale but erodes as portfolios grow.p.9