Stock Volatility Prediction Using Recurrent Neural Networks with Sentiment Analysis (1705.02447v1)

Abstract: In this paper, we propose a model to analyze sentiment of online stock forum and use the information to predict the stock volatility in the Chinese market. We have labeled the sentiment of the online financial posts and make the dataset public available for research. By generating a sentimental dictionary based on financial terms, we develop a model to compute the sentimental score of each online post related to a particular stock. Such sentimental information is represented by two sentiment indicators, which are fused to market data for stock volatility prediction by using the Recurrent Neural Networks (RNNs). Empirical study shows that, comparing to using RNN only, the model performs significantly better with sentimental indicators.

• The paper presents an RNN model enhanced with sentiment indicators derived from Chinese stock forum posts to predict stock market volatility.
• Integrating sentiment analysis significantly improves prediction accuracy, achieving up to 69.85% for one stock compared to 57.33% with RNN alone.
• The study demonstrates the value of sentiment analysis for financial prediction, particularly in emerging markets, and provides a dataset for further research.

The paper "Stock Volatility Prediction Using Recurrent Neural Networks with Sentiment Analysis" presents a model that integrates sentiment analysis of online stock forum posts as a component in predicting stock volatility within the Chinese market. The authors perform sentiment analysis on posts from the East Money Forum, a prominent Chinese stock discussion platform, to extract sentimental scores that are then used in conjunction with Recurrent Neural Networks (RNNs) for forecasting stock market volatility.

Key Contributions and Methodology

1. Sentiment Analysis Dataset:
   • The authors compiled a publicly available dataset by manually labeling the sentiment of financial posts as positive or negative, resulting in 3,427 annotated samples.
   • They applied a segmentation tool, "Jieba", to process the original Chinese texts into terms suitable for sentiment analysis.
2. Sentimental Weight Dictionary:
   • Utilizing a machine learning approach, a unique sentimental dictionary was generated which includes financial terms specific to Chinese markets.
   • Sentimental weights for terms are calculated via logistic regression, providing a polarity model that dictates whether posts convey a positive or negative sentiment.
3. Sentiment Indicators:
   • Two sentiment indicators were introduced: Bullishness Index ($B_t$) and Post Volume ($N_t$). The Bullishness Index is derived from the logarithmic ratio of positive to negative sentiment scores, while the Post Volume reflects the sheer quantity of discourse, both normalized to z-scores.
   • These indicators are theorized to correlate with price movement and volatility—more optimistic posts are expected to correlate with price increases, and high post volume with higher volatility.
4. Integration with RNN:
   • The sentiment indicators are integrated into an RNN model, which includes previous stock volatilities and sentimental indicators as inputs to predict future stock volatility.
   • The use of RNN is justified by its ability to manage sequence data and temporal dependencies, suitable for capturing the dynamics of stock price movements.

Empirical Evaluation

• Experiments conducted on data for ten selected Chinese stocks demonstrate that incorporating sentimental indicators into the RNN model significantly enhances prediction accuracy compared to using RNN alone.
• The RNN model with sentiment indicators achieved an accuracy of 69.85% for stock number 000573, an improvement from 57.33% without sentiment integration.
• Overall, the RNN with sentimental indicators outperformed other models, including Multi-Layer Perception (MLP) and Support Vector Machine (SVM).

Conclusion and Future Work

The paper concludes that sentiment derived from stock forum posts can effectively enhance the prediction of stock market volatility in the Chinese context. The authors point out that their models do not account for fraudulent or misleading posts within forums, indicating a potential avenue for future research. Additionally, they provide access to their manually labeled dataset to encourage further research.

Limitations

• The paper focuses exclusively on a linguistic and sentiment-based analysis, which may neglect other critical financial indicators or macroeconomic factors.
• The reliance on data from a single source (East Money Forum) may limit the generalizability of results across other forums or markets with varied investor behavior and market dynamics.

Overall, the paper provides a robust framework for integrating sentiment analysis into financial prediction models, demonstrating a significant correlation between sentiment and stock volatility particularly in emerging markets like China.