Genetic Programming for Alpha Factors: Incremental Value Versus Manual Factors
Summary
This research summary tests whether machine-generated stock-selection factors add value to established manually designed factor libraries. It uses genetic programming to evolve candidate formulas from market data, then applies random forests to turn factor libraries into return forecasts and compares their portfolio-level performance. Technical and financial factors are evaluated separately. The study describes repeated factor mining over a 2010–2020 backtest, refreshing the search every six months and selecting candidates according to their prior three-year average monthly factor returns.
The reported results are mixed. Generated technical factors improved several measures relative to traditional technical factors, and combining them modestly improved long-short returns and stability, but the incremental gains were not statistically significant. Generated financial factors generally underperformed traditional financial factors, though differences also lacked statistical significance. The summary therefore offers limited evidence that expanding an established library with machine-discovered factors pays off. It does not provide detailed implementation, robustness checks, or full portfolio construction details, and its conclusions are bounded by the described data and evaluation design.
Key ideas
- The study uses genetic programming to evolve candidate stock-selection factors from market and financial data.
- Random forests convert factor libraries into return forecasts for portfolio-level comparisons.
- Factor discovery is refreshed every six months using the preceding three years of average monthly factor returns.
- Generated technical factors showed some improvement, but the reported incremental gains were not statistically significant.
- Generated financial factors generally lagged traditional financial factors, with differences also not statistically significant.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.