Mining Interpretable Fundamental Factors with OpenFE for Chinese Equities

Summary

The report describes applying OpenFE, an enumerate-and-filter framework, to discover interpretable fundamental factors from Chinese companies’ balance sheets, income statements, and cash-flow statements. It first removes fields with substantial missing data, uses LightGBM to narrow the accounting inputs, then combines selected features with simple arithmetic, growth, and cross-sectional ranking operators. The resulting candidate factors are organized into leverage, return, quality, valuation, and growth styles. Successive halving screens candidates on progressively larger samples; a second LightGBM stage ranks survivors by their incremental contribution alongside the original features.

The authors use selected synthetic and basic factors in a monthly stock-ranking model and report positive historical excess returns, with momentum, market capitalization, and industry among the strongest inputs. Results vary across index universes, and quality factors rank relatively weakly in their analysis. The evidence comes from a limited historical backtest and factor-importance analysis. The report cautions that market regimes, sample choices, model settings, randomness, and computing constraints can affect results; the findings do not establish future factor performance.

Key ideas

• OpenFE builds structured fundamental features by combining accounting inputs with simple operators.
• Successive halving reduces the candidate set before a multivariate LightGBM ranking stage.
• The study groups generated features into leverage, return, quality, valuation, and growth styles.
• Momentum, size, and industry rank strongly in the reported stock-selection model.
• Historical backtest results may be sensitive to regime changes, sample choices, and model settings.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.