The idea worth investigating
In a 2025 equity backtest, a hybrid model returned 51.26%, compared with 33.61% for XGBoost alone and 18.01% for the paper’s S&P 500 benchmark. The hybrid’s reported maximum drawdown was 7.76%. A stronger combined signal is one explanation; the universe and execution assumptions deserve just as much attention.
Authors’ single-year simulation with 2.2 bp per trading leg. Uses end-of-sample constituents and omits stock-borrow costs and market impact.
This is a source-based research review. We have not completed a house replication.
Paid analysis on Substack
See what changes the investment case.
The headline is the starting point. The subscriber analysis takes you through:
- Why the ensemble differs from either individual model
- How universe selection and shorting assumptions affect the comparison
- What a faithful independent evaluation would require
AI assists curation and drafting. The research status distinguishes paper reviews from house tests. Read our research approach.