Paper: arXiv 2408.02694
Authors: Tianqi Wang, Shubham Singh
Abstract
Inspired by recent advances in Kolmogorov-Arnold Networks (KANs), we introduce a novel approach to latent factor conditional asset pricing models. While previous machine learning applications in asset pricing have predominantly used Multilayer Perceptrons with ReLU activation functions to model latent factor exposures, our method introduces a KAN-based autoencoder which surpasses MLP models in both accuracy and interpretability. Our model offers enhanced flexibility in approximating exposures as nonlinear functions of asset characteristics, while simultaneously providing users with an intuitive framework for interpreting latent factors. Empirical backtesting demonstrates our model’s superior ability to explain cross-sectional risk exposures. Moreover, long-short portfolios constructed using our model’s predictions achieve higher Sharpe ratios, highlighting its practical value in investment management.
Complexity vs Empirical Score
- Math Complexity: 8.0/10
- Empirical Rigor: 6.0/10 (capped from 7.0: transaction costs not modelled; see penalty flags)
- Quadrant: Holy Grail — high math complexity, high empirical rigor (effective empirical rigor capped from 7.0)
Why this score: The paper employs advanced mathematics, including the Kolmogorov-Arnold theorem and spline parameterizations, indicating high complexity. It also demonstrates strong empirical rigor with a large dataset (CRSP data, 1957-2016), backtesting results, and reported Sharpe ratios, making it backtest-ready.
Research Flowchart
flowchart TD
A["Research Goal:<br>Improve Factor Model Accuracy &<br>Interpretability using Neural Networks"] --> B["Data: Asset Characteristics<br>Historical Returns"]
B --> C["Methodology: KAN-based Autoencoder"]
subgraph C
C1["Encoder: Map Characteristics<br>to Latent Factors"]
C2["Decoder: Map Factors<br>back to Returns"]
end
C1 --> D["Computational Process:<br>Nonlinear Approximation via<br>Kolmogorov-Arnold Networks"]
C2 --> D
D --> E["Key Findings & Outcomes"]
subgraph E
E1["Higher Accuracy:<br>Superior Cross-sectional Explanations"]
E2["Better Interpretability:<br>Intuitive Latent Factor Framework"]
E3["Practical Value:<br>Higher Sharpe Ratios<br>in Long-Short Portfolios"]
end
Evidence Card
What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card
Proposes
The paper proposes a KAN-based autoencoder latent factor conditional asset pricing model that improves accuracy and interpretability over MLP approaches.
Source sentence
our method introduces a KAN-based autoencoder which surpasses MLP models in both accuracy and interpretability.
Assumes 6
- The factor structure follows r_t = β(Z_{t-1}) f_t + ε_t with latent factors and exposures modeled as a function of lagged characteristics. model
Source sentence
where factors are now treated as latent, and factor
- Monthly stock characteristics are delayed by one month, quarterly by four months, and annual by one year to avoid look‑ahead bias. data
Source sentence
We assume that the monthly stock characteristics are delayed for a month, quarterly stock characteristics are delayed for 4 months, and annual characteristics are delayed for a year. This practice avoids look-ahead bias in our simulation.
- No filters are imposed on the stocks included in the simulation, implying all listed stocks are used. data
Source sentence
No filters are imposed on the stocks that are included in our simulation.
- The KAN network can serve as a suitable function approximator for the beta mapping from characteristics to factor exposures. model
Source sentence
Instead of a MLP, we use a KAN network to model the factor exposure.
Show 2 more
- The Treasury bill rate is an appropriate proxy for the risk‑free rate to compute excess returns. data
Source sentence
We use the Treasury bill rate to proxy for the risk-free rate from which we calculate individual excess returns.
- Long‑short portfolio Sharpe ratios reflect the economic utility of the model without accounting for trading costs. execution
Source sentence
ratio of the long‑short portfolio formed using the model's predictions quantifies the economic utility that the model has in actual applications.
Data
- CRSP monthly individual stock returns market-prices commercial
Source sentence
We analyze the same dataset studied in Gu et al. (2019), which contains monthly individual stock returns from the Center for Research in Securities Prices (CRSP) for all firms listed in the three major exchanges: NYSE, AMEX, and NASDAQ.
- Treasury bill rate macro public
Source sentence
We use the Treasury bill rate to proxy for the risk-free rate from which we calculate individual excess returns.
Costs
Not modeled
Finds 3
Sharpe 0.96
- The KAN‑CA model achieves higher total R² scores (11.02%, 11.26%, 11.32%) for 1‑, 3‑, and 6‑factor models compared to FF and CA models. performance
Source sentence
the KAN‑CA (KAN based Conditional Autoencoder) model achieves improved R² scores of 11.02%, 11.26%, and 11.32% for 1-, 3-, and 6-factor models respectively
- Predictive R² improves to 0.203%–0.214% for KAN‑CA across factor specifications, outperforming FF and CA models. predictive-accuracy
Source sentence
the KAN‑CA model exhibits superior predictive R² scores of 0.203%, 0.203%, and 0.214% for 1-, 3-, and 6-factor models, outperforming both FF and CA models
- Long‑short portfolios based on KAN‑CA deliver higher Sharpe ratios (0.86, 0.86, 0.96) than CA (0.84, 0.87, 0.91) for 1‑, 3‑, and 6‑factor models. performance
Source sentence
the KAN‑CA model consistently delivering higher Sharpe ratios (0.86, 0.86, and 0.96 for 1-, 3-, and 6-factor models) compared to the CA model (0.84, 0.87, and 0.91 respectively)
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 40 related papers). Coverage is partial, so this is neither replication nor refutation.
Code
No repository by the authors found in the paper.
Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.