From Fama-French to LSTM: Rethinking Risk Factors in the Swedish Stock Market
(2025) NEKH02 20242Department of Economics
- Abstract
- This paper investigates the application of a Long Short-Term Memory (LSTM) model to construct a forward-looking risk factor for asset pricing. The LSTM model's predictions are utilized to create a novel risk factor, derived from long-short portfolios, offering an innovative approach to decomposing asset returns. For comparative analysis, the Fama-French three-factor (FF3) and Carhart four-factor (C4) models are constructed using data from the Swedish stock market. The evaluation framework employs time-series regression on three 5x5 testing portfolios, with an analysis of intercepts conducted through t-values, and GRS tests. Spanning regressions are performed to assess whether the proposed forward-looking risk factor captures returns not... (More)
- This paper investigates the application of a Long Short-Term Memory (LSTM) model to construct a forward-looking risk factor for asset pricing. The LSTM model's predictions are utilized to create a novel risk factor, derived from long-short portfolios, offering an innovative approach to decomposing asset returns. For comparative analysis, the Fama-French three-factor (FF3) and Carhart four-factor (C4) models are constructed using data from the Swedish stock market. The evaluation framework employs time-series regression on three 5x5 testing portfolios, with an analysis of intercepts conducted through t-values, and GRS tests. Spanning regressions are performed to assess whether the proposed forward-looking risk factor captures returns not explained by the FF3 and C4 risk factors. The $R^2$ values increase from smaller to larger portfolios, indicating that the models better explain variation in returns in less volatile, large-cap portfolios. However, the GRS test reveals that while mid-sized portfolios have the lowest unexplained returns, small-cap portfolios struggle with volatility, and large-cap portfolios exhibit persistent unexplained returns despite their higher $R^2$. The predictive accuracy of the LSTM model for stock prices is limited, suggesting areas for further development. However, integrating the LSTM-derived risk factor into the FF3 model slightly enhances explanatory power, as evidenced by lower GRS values. Furthermore, the LSTM-derived factor demonstrates overlap with momentum, indicating a potential improvement when substituting momentum with the forward-looking risk factor. In summary, this study illustrates the potential of a LSTM-derived risk factor to complement traditional models, advancing risk factor frameworks and providing valuable insights for future data-driven financial modeling. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9184754
- author
- Ritseson, Victor LU and Linde, Johan LU
- supervisor
- organization
- course
- NEKH02 20242
- year
- 2025
- type
- M2 - Bachelor Degree
- subject
- keywords
- Asset Pricing Models, Fama French Three-Factor Model, Carhart Four-Factor Model, Factor Model, Machine Learning
- language
- English
- id
- 9184754
- date added to LUP
- 2025-05-16 10:47:17
- date last changed
- 2025-05-16 10:47:17
@misc{9184754, abstract = {{This paper investigates the application of a Long Short-Term Memory (LSTM) model to construct a forward-looking risk factor for asset pricing. The LSTM model's predictions are utilized to create a novel risk factor, derived from long-short portfolios, offering an innovative approach to decomposing asset returns. For comparative analysis, the Fama-French three-factor (FF3) and Carhart four-factor (C4) models are constructed using data from the Swedish stock market. The evaluation framework employs time-series regression on three 5x5 testing portfolios, with an analysis of intercepts conducted through t-values, and GRS tests. Spanning regressions are performed to assess whether the proposed forward-looking risk factor captures returns not explained by the FF3 and C4 risk factors. The $R^2$ values increase from smaller to larger portfolios, indicating that the models better explain variation in returns in less volatile, large-cap portfolios. However, the GRS test reveals that while mid-sized portfolios have the lowest unexplained returns, small-cap portfolios struggle with volatility, and large-cap portfolios exhibit persistent unexplained returns despite their higher $R^2$. The predictive accuracy of the LSTM model for stock prices is limited, suggesting areas for further development. However, integrating the LSTM-derived risk factor into the FF3 model slightly enhances explanatory power, as evidenced by lower GRS values. Furthermore, the LSTM-derived factor demonstrates overlap with momentum, indicating a potential improvement when substituting momentum with the forward-looking risk factor. In summary, this study illustrates the potential of a LSTM-derived risk factor to complement traditional models, advancing risk factor frameworks and providing valuable insights for future data-driven financial modeling.}}, author = {{Ritseson, Victor and Linde, Johan}}, language = {{eng}}, note = {{Student Paper}}, title = {{From Fama-French to LSTM: Rethinking Risk Factors in the Swedish Stock Market}}, year = {{2025}}, }