Designing factor-organization methods that balance information completeness and economic interpretability

Develop and evaluate factor-organizing methodologies for Transformer-based asset pricing that achieve a balance between information completeness and economic interpretability, improving upon firm characteristic-sorted portfolios that enhance interpretability but may discard information embedded in raw firm characteristics.

Background

The paper employs firm characteristic-sorted portfolios as inputs to preserve economic meaning. However, the authors note this choice sacrifices some information originally present in granular firm characteristics.

They call for new factor-organization methods that retain more informational content while still offering economically interpretable factors suitable for Transformer-based modeling.

References

However, some open questions remain for future researchers. Also, the firm characteristic sorted portfolios as factors increase the economic meanings to a degree, but it sacrifices some of the information originally contained in the firm characteristics. Researchers may explore innovative factor-organizing methods which balance information completeness and economic explanations.

— Asset Pricing in Pre-trained Transformer (2505.01575 - Lai, 2 May 2025) in Section 6, Conclusion