Paper: arXiv 2411.05791
Authors: Felix Divo, Eric Endress, Kevin Endler, Kristian Kersting, Devendra Singh Dhami
Abstract
Company fundamentals are key to assessing companies’ financial and overall success and stability. Forecasting them is important in multiple fields, including investing and econometrics. While statistical and contemporary machine learning methods have been applied to many time series tasks, there is a lack of comparison of these approaches on this particularly challenging data regime. To this end, we try to bridge this gap and thoroughly evaluate the theoretical properties and practical performance of 24 deterministic and probabilistic company fundamentals forecasting models on real company data. We observe that deep learning models provide superior forecasting performance to classical models, in particular when considering uncertainty estimation. To validate the findings, we compare them to human analyst expectations and find that their accuracy is comparable to the automatic forecasts. We further show how these high-quality forecasts can benefit automated stock allocation. We close by presenting possible ways of integrating domain experts to further improve performance and increase reliability.
Complexity vs Empirical Score
- Math Complexity: 4.5/10
- Empirical Rigor: 8.0/10
- Quadrant: Street Traders — empirical and applied, lower math complexity
Why this score: The paper focuses on applying and comparing existing deep learning and statistical models to real financial data, with a strong emphasis on backtesting and practical investment outcomes rather than developing new complex mathematical theory. It presents empirical results on 24 models and validates findings against human analyst expectations and portfolio allocation strategies.
Research Flowchart
flowchart TD
A["Research Goal<br>Compare Forecasting Models<br>for Company Fundamentals"] --> B{"Data Input<br>Real Company Fundamentals<br>Time Series Data"}
B --> C["Methodology<br>Evaluate 24 Models<br>Statistical, ML, & Deep Learning"]
C --> D["Computational Process<br>Probabilistic & Deterministic Forecasting<br>with Uncertainty Estimation"]
D --> E["Key Findings<br>Deep Learning Superior to Classical Models"]
D --> F["Validation<br>Accuracy Comparable to Human Analysts"]
D --> G["Outcome<br>Improved Automated Stock Allocation"]
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 thorough evaluation of 24 deterministic and probabilistic company fundamentals forecasting models, highlighting that deep learning models outperform classical ones, especially for uncertainty estimation.
Source sentence
we try to bridge this gap and thoroughly evaluate the theoretical properties and practical performance of 24 deterministic and probabilistic company fundamentals forecasting models on real company data.
Assumes 2
- Prophet model assumes daily business time series with typical weekend and holiday effects. model
Source sentence
the Prophet model (Taylor & Letham, 2018) assumes “business time series”, i.e., daily samples with the typical effects of weekends and holidays, to be applied most effectively.
- The dataset contains a relatively small number of companies (2578), implying limited sample size. data
Source sentence
only 2578 companies fit our selection criterion explained in Section 4.1.
Data
MSCI World · equities · monthly
- company fundamentals dataset fundamentals unknown
Source sentence
only 2578 companies fit our selection criterion explained in Section 4.1.
Refinitiv / LSEG (Datastream, Eikon, Tick History)
Costs
Unclear - sources disagree
Finds 5
annual return 15.1%
- Deep learning models provide superior forecasting performance to classical models, especially for uncertainty estimation. performance
Source sentence
deep learning models provide superior forecasting performance to classical models, in particular when considering uncertainty estimation.
- Recurrent models achieve the overall best performance across multiple error metrics. performance
Source sentence
Recurrent models provide the overall best performance.
- Automatic forecasts achieve accuracy comparable to human analyst expectations. predictive-accuracy
Source sentence
their accuracy is comparable to the automatic forecasts.
- Probabilistic forecasts can be obtained by AutoARIMA or any deep learning model, offering more helpful predictions than deterministic baselines. predictive-accuracy
Source sentence
the much more helpful probabilistic forecasts can be obtained by either AutoARIMA or, equivalently, by any of the deep learning models.
- The best automated forecasts can outperform the reference index over ten years, especially during periods of higher financial stability. performance
Source sentence
We find that they can outperform the reference index over ten years and observe the benefit to be greatest in times of higher financial stability.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 31 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.