KernelNet
A market-neutral trading strategy that generalizes pairs trading by replacing static correlations with nonlinear causality networks.
Itaú Asset Quant AI Challenge 2025
Itaú Asset Management, Dec 2025
- Sharpe ratio
- 1.29
- Annualized return
- 54.85%
- Annualized return of the benchmark
- 22.78%
- Of ~1,000 teams and 2,500+ participants
- 2nd
Why pairs trading breaks
Pairs trading bets that two historically correlated assets will return to their usual relationship. That correlation is static and linear, so when the market regime changes, a pair can stop behaving as it did and the trade stops working.
Causality instead of correlation
KernelNet generalizes the idea. Instead of a static correlation between two assets, it builds a nonlinear causality network across assets, which captures which assets drive others rather than which ones happened to move together. The strategy trades those relationships while staying market-neutral.
The result
The strategy reached a Sharpe ratio of 1.29 and a 54.85% annualized return, against 22.78% for the benchmark, and held up across different market regimes.
We placed 2nd (Silver Medal) among nearly 1,000 teams and 2,500+ participants in Brazil's largest challenge for undergraduate students. The field also included Brazilian competitors from MIT, Stanford and Berkeley.
