Introducing Deep Fundamental Factor Models for after-work quants

Takeaway. Deep fundamental factor models push fundamental characteristics through deep architectures to capture nonlinear cross-sectional structure—still grounded in fundamental inputs, not vibes.

Deep Fundamental Factor Models (Matthew Dixon, Nicholas Polson, 2019, arXiv:1903.07677) sits next to other deep-factor open PDFs as a fundamentals-first deep learning read.

Fit for learners: Deep nets on fundamental characteristics for cross-sectional equity modelling—another open-PDF deep-factor companion.

This Hub note is a plain-language intro. Not investment advice. Read with Nakagawa deep factor and deep multi-factor.

The problem

Fundamental factor models (earnings yields, leverage, profitability, …) are linear by default. Markets may price nonlinear combinations. Deep nets can approximate those—but financial panels are noisy, biased, and riddled with missingness.

The idea / core intuition

• Start from fundamental characteristics familiar to quant fundamentalists.
• Allow deep layers to learn interactions / nonlinear maps into return-relevant predictions or factors.
• Emphasise statistical learning discipline: regularisation, validation design, comparison to linear fundamental models.
• Interpretability is harder; treat outputs as research objects, not oracles.

Paper details beat any Hub paraphrase—use this only as a map.

Why after-work Python learners care

If you already scrape or join fundamentals tables in pandas, this paper asks the next question: what happens when the map from fundamentals → return is allowed to bend? That is a controlled ML homework theme.

Tiny evening exercise: pick 3–5 toy fundamental z-scores; fit ridge linear vs a tiny MLP to next-month toy returns on a fake panel; compare out-of-sample R² and say which you trust more—and why trust ≠ trade.

One-evening reading checklist

1. Name what makes the model “fundamental” vs pure price-based deep learning.
2. List data issues (missing, look-ahead, survivorship) you must watch.
3. Write one reason linear fundamentals remain a mandatory baseline.
4. Note how this differs in spirit from IPCA’s instrumented loadings.

Download PDF

Download PDF

Landing: arXiv abs

Important disclaimer

Not investment advice. This Hub note is educational only. Dixon–Polson is academic research—not a fundamental long/short product sheet.

What this is not

• Not CEF / subsidy framing
• Not Quant or Macro course CTA
• Not an earnings-call tip service

Next step

Single course CTA: SoloQuant Academy’s Introduction to Python Programming—clean CSV handling and functions first; deep fundamental notebooks come later.