Reproduce “Forecasting Company Fundamentals” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Felix Divo, Eric Endress, Kevin Endler, Kristian Kersting, Devendra Singh Dhami
This research systematically evaluates 24 diverse forecasting models for company fundamentals, comparing classical statistical and modern deep learning approaches on real-world financial data. It demonstrates that automated forecasts can achieve accuracy comparable to human analysts and, when integrated into factor investment strategies, can yield competitive returns over market benchmarks.
Key insights
- Deep learning models, particularly RNNs like LSTM and GRU, provided superior probabilistic forecasts, effectively learning common representations across many companies and generalizing well.
- The inherent characteristics of financial data, including its non-Gaussian nature, heavy tails, and high non-stationarity, pose significant challenges for model robustness, leading to varied performance across different metrics and models.
- Automated forecasts can match or exceed the accuracy of human expert analysts for certain CFs and, when used in factor models, can lead to outperformance against market indices during stable periods.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.