TLDR: IKNet is a new AI model that predicts stock prices by integrating financial news keywords with technical indicators. Unlike previous models, IKNet uses a technique called SHAP to provide clear, keyword-level explanations for its predictions, showing how specific words in news articles influence stock movements. Evaluated on S&P 500 data from 2015-2024, IKNet outperformed baseline models, reducing prediction errors and improving investment returns, while offering valuable transparency into market sentiment.
Predicting stock prices is a complex challenge, with many factors influencing market movements. While traditional methods often rely on historical price data and technical indicators, the impact of unstructured information, such as financial news articles, has become increasingly significant. However, many existing news-based forecasting models struggle to provide clear, context-aware explanations for how public sentiment, as captured in news, actually affects stock predictions.
Introducing IKNet: A Smarter Way to Predict Stocks
To address this gap, researchers Jinwoong Kim and Sangjin Park have developed an innovative framework called the Interpretable Keyword-Guided Network (IKNet). This new model offers an explainable approach to stock price forecasting by focusing on the semantic connection between individual news keywords and stock price changes. Essentially, IKNet doesn’t just tell you what might happen; it helps explain why, by highlighting the specific words in news articles that drive its predictions.
How IKNet Works Behind the Scenes
IKNet operates through several key steps. First, it uses a specialized language model called FinBERT to identify the most important keywords in daily financial news articles. These aren’t just any words; they are the ones contextually most relevant to financial sentiment. Each of these identified keywords then gets its own unique processing, preventing their meanings from getting mixed up and allowing for individual analysis.
These keyword representations are then integrated with time-series data from traditional technical indicators, such as moving averages and the Relative Strength Index (RSI). By combining these two distinct types of information – the qualitative insights from news and the quantitative data from market trends – IKNet aims to create a more comprehensive picture for forecasting next-day closing prices.
Understanding the ‘Why’ with SHAP
One of IKNet’s most significant contributions is its interpretability. It employs a technique called Shapley Additive Explanations (SHAP) to quantify and visualize the contribution of each keyword to the final prediction. This means you can see exactly how much a specific word, like “layoffs” or “growth,” influenced the model to predict a rise or fall in stock prices. This level of detail goes beyond simply knowing if sentiment is positive or negative; it provides a fine-grained understanding of which specific terms are driving market expectations.
Impressive Performance and Practical Insights
Empirical evaluations using S&P 500 data from 2015 to 2024 showed that IKNet consistently outperformed other advanced models, including recurrent neural networks and transformer models. It achieved a significant reduction in prediction errors (RMSE by up to 32.9%) and improved cumulative returns by 18.5%. This robust performance was maintained even during volatile market periods, demonstrating its stability and accuracy.
Beyond just accuracy, IKNet enhances transparency. For instance, when analyzing a news article about a U.S. employment report, IKNet could clearly show that negative keywords like “tumbled,” “plunged,” and “layoffs” strongly contributed to a predicted decline in stock prices, aligning with actual market reactions. Conversely, positive terms like “boosted” or “growth” would push predictions upward. This contextualized explanation of volatility events driven by public sentiment is invaluable for investors and analysts.
Also Read:
- Benchmarking Financial Knowledge Graphs: Introducing FinReflectKG – EvalBench
- Hybrid AI Model Offers Superior Accuracy and Clearer Insights for Time Series Forecasting
The Future of Financial Forecasting
IKNet represents a significant step forward in making AI-driven stock predictions more understandable and trustworthy. By integrating keyword-level insights from news with technical indicators, it offers both superior predictive accuracy and clear explanations, making it a powerful tool for decision-making in financial markets. Future research may explore even more nuanced ways to capture keyword meanings and dynamically adjust the importance of news versus technical data as market conditions change.
For more detailed information, you can read the full research paper here: IKNet: Interpretable Stock Price Prediction via Keyword-Guided Integration of News and Technical Indicators.
