# FinGAT: Financial Graph Attention Networks for Recommending Top-K Profitable Stocks

**Authors:** Yi-Ling Hsu, Yu-Che Tsai, Cheng-Te Li  
**arXiv:** [2106.10159](https://arxiv.org/abs/2106.10159)  
**Date:** 2021-06-18  
**Summary page:** https://www.summarizepaper.com/en/paper/2106.10159v1/

## Key points

- Financial technology (FinTech) is a prominent area of interest for investors and companies
- Traditional stock analysis in FinTech focuses on predicting stock prices, but lacks emphasis on recommending profitable stocks
- Existing approaches to modeling time series of stock prices often overlook relationships between stocks and sectors or rely on predefined relationships that fail to capture underlying interactions and influences among stock prices
- The authors propose a novel deep learning-based model called Financial Graph Attention Networks (FinGAT) to address these limitations
- FinGAT aims to recommend the top-K profitable stocks based on return ratio using both time series data of stock prices and sector information
- FinGAT operates under the assumption that no pre-defined relationships between stocks are given
- The FinGAT model consists of three key components: hierarchical learning component, fully-connected graphs representing interactions among stocks and sectors, and a multi-task objective for recommending profitable stocks and predicting their movement
- Experiments using datasets from Taiwan Stock, S&P 500, and NASDAQ markets show that FinGAT outperforms state-of-the-art methods in terms of recommendation performance for profitable stocks
- FinGAT leverages deep learning techniques and graph attention networks to offer valuable insights for investors and companies in the FinTech domain.

## Summary

Financial technology (FinTech) has become a prominent area of interest for investors and companies. While traditional stock analysis in FinTech focuses on predicting stock prices, there is a lack of emphasis on recommending profitable stocks. Additionally, existing approaches to modeling time series of stock prices often overlook the relationships between stocks and sectors or rely on predefined relationships that fail to capture the underlying interactions and influences among stock prices. In this study, the authors aim to address these limitations by proposing a novel deep learning-based model called Financial Graph Attention Networks (FinGAT). The objective of FinGAT is to recommend the top-K profitable stocks based on return ratio using both time series data of stock prices and sector information. Importantly, FinGAT operates under the assumption that no pre-defined relationships between stocks are given. The FinGAT model consists of three key components. First, a hierarchical learning component is designed to capture short-term and long-term sequential patterns from the time series data of stock prices. This allows for a comprehensive understanding of the temporal dynamics in stock price movements. Secondly, fully-connected graphs are constructed both between individual stocks and between sectors. These graphs serve as representations of the latent interactions among stocks and sectors. Graph attention networks are employed to learn these interactions effectively. Lastly, a multi-task objective is formulated to jointly recommend profitable stocks and predict their movement. By integrating these two tasks into a single framework, FinGAT can leverage the shared information between them to enhance its recommendation performance. The authors conducted experiments using datasets from Taiwan Stock, S&P 500, and NASDAQ markets. The results demonstrate that FinGAT outperforms state-of-the-art methods in terms of recommendation performance for profitable stocks. Overall, this study presents an innovative approach to recommending profitable stocks by leveraging deep learning techniques and graph attention networks. By considering both temporal patterns in stock prices and latent interactions among stocks and sectors, FinGAT offers valuable insights for investors and companies in the FinTech domain.