AI analyzing stock market data

Decoding the Market: Can AI Predict Stock Prices and Explain Why?

Lena Kashyap in Business & Economy December 2025 • 4 min read.

"Unveiling the potential of self-reflective AI in generating explainable stock market predictions."

For years, the stock market has been a complex puzzle, challenging even the most seasoned investors. Traditional methods often fall short of providing clear, actionable insights, leaving many to wonder if there's a better way to understand market movements. The rise of deep learning offered some solutions, but it also created new problems. While it could do some analysis it remained as black box models that lacked the transparency and explainability needed for practical application.

Enter Large Language Models (LLMs), a new breed of AI known for their ability to generate human-readable explanations. With their language skills, LLMs present a potential solution to this challenge. However, stock prediction is difficult for these types of AI because the varying chaotic impacts of social texts on stock prices are difficult to weigh and explain verbally.

A groundbreaking study from the National University of Singapore and Eastspring Investments introduces a novel framework called Summarize-Explain-Predict (SEP). This innovative approach utilizes a self-reflective AI agent and Proximal Policy Optimization (PPO) to enable LLMs to autonomously learn how to generate explainable stock predictions, promising a new era of transparency and understanding in the financial world.

How Self-Reflective AI Learns to Predict and Explain Stock Movements

The SEP framework tackles the challenge of creating explainable stock predictions by dividing the process into three key components:

Here’s how each component contributes to the overall goal:

Summarize: This module leverages the summarization capabilities of LLMs to condense vast amounts of text data into concise summaries of factual information.
Explain: A self-reflective agent iteratively refines its predictions and explanations through verbal self-reflection, learning from past mistakes to improve future accuracy.
Predict: A specialized LLM, fine-tuned using Proximal Policy Optimization (PPO), generates confidence-based predictions and explanations based on the insights gained during the reflective process.

This innovative approach allows the AI to teach itself how to make explainable stock predictions without relying on human experts for annotated data. The self-reflective agent learns to explain past stock movements through a self-reasoning process, while the PPO trainer optimizes the model to generate the most likely explanations given the input texts.

The Future of AI in Stock Market Analysis

The SEP framework represents a significant step forward in the quest for explainable AI in stock market analysis. By combining the power of LLMs with self-reflection and reinforcement learning, this approach offers a promising path towards more transparent and understandable investment insights. As AI continues to evolve, we can expect even more sophisticated models that not only predict stock movements but also provide clear, concise explanations for their decisions, empowering investors to make more informed choices.

About this Article -

This article was crafted using a human-AI hybrid and collaborative approach. AI assisted our team with initial drafting, research insights, identifying key questions, and image generation. Our human editors guided topic selection, defined the angle, structured the content, ensured factual accuracy and relevance, refined the tone, and conducted thorough editing to deliver helpful, high-quality information.See our About page for more information.

This article is based on research published under:

DOI-LINK: 10.1145/3589334.3645611,

Title: Learning To Generate Explainable Stock Predictions Using Self-Reflective Large Language Models

Subject: cs.lg cs.cl q-fin.st

Authors: Kelvin J. L. Koa, Yunshan Ma, Ritchie Ng, Tat-Seng Chua

Published: 05-02-2024

Everything You Need To Know

What is the Summarize-Explain-Predict (SEP) framework and how does it work?

The Summarize-Explain-Predict (SEP) framework is a novel approach developed by researchers from the National University of Singapore and Eastspring Investments. It's designed to create explainable stock market predictions using a self-reflective AI agent. The framework operates in three key stages: Summarize, Explain, and Predict. The Summarize module uses the summarization capabilities of Large Language Models (LLMs) to condense vast amounts of text data into concise summaries. The Explain module employs a self-reflective agent that iteratively refines its predictions and explanations through verbal self-reflection, learning from past mistakes. Finally, the Predict module utilizes a specialized LLM, fine-tuned with Proximal Policy Optimization (PPO), to generate confidence-based predictions and explanations based on the insights gained during the reflective process.

How does the SEP framework's 'Explain' component contribute to creating explainable AI in stock market analysis?

The 'Explain' component within the Summarize-Explain-Predict (SEP) framework is crucial for achieving explainable AI. This module leverages a self-reflective agent. This agent iteratively refines its predictions and explanations. It does so through verbal self-reflection, where the AI learns from its past predictions and the corresponding outcomes. The self-reflective process allows the AI to understand the relationships between different factors and stock movements, enabling it to provide clear, human-readable explanations for its predictions. This is a significant departure from traditional 'black box' AI models that offer predictions without any insights into the decision-making process.

What role does Proximal Policy Optimization (PPO) play in the SEP framework for stock market prediction?

Proximal Policy Optimization (PPO) is used within the Predict module of the Summarize-Explain-Predict (SEP) framework. PPO serves as a training method that optimizes the specialized Large Language Model (LLM) to generate the most accurate and reliable stock market predictions. The PPO algorithm helps the LLM learn from its mistakes by refining its prediction strategy. This ensures that the AI can make better predictions, providing confidence-based outcomes along with the corresponding explanations. In essence, PPO is a reinforcement learning technique that helps the AI learn to make better predictions over time by rewarding it for accurate insights.

What are the advantages of using Large Language Models (LLMs) in the SEP framework for stock market analysis compared to traditional methods?

The use of Large Language Models (LLMs) in the Summarize-Explain-Predict (SEP) framework offers several advantages over traditional stock market analysis methods. LLMs excel at processing and understanding natural language, which is vital for interpreting the complex and unstructured data found in financial news and social media. Unlike many traditional AI models, LLMs can generate human-readable explanations for their predictions. This increases transparency and trust. LLMs can summarize information from diverse sources, providing a comprehensive view of market dynamics. Traditional methods often struggle with these tasks, relying on static data and lacking the ability to offer clear, understandable insights.

How does the self-reflective AI agent in the SEP framework improve stock market prediction accuracy?

The self-reflective AI agent within the Summarize-Explain-Predict (SEP) framework enhances stock market prediction accuracy through a continuous learning process. The agent repeatedly evaluates its past predictions, comparing them with actual market outcomes. This reflection process allows the agent to identify patterns, correlations, and factors that influence stock movements. By analyzing its mistakes, the AI refines its understanding of market dynamics. Through this process of self-reflection, the AI agent gradually improves its ability to generate accurate predictions and provide clear explanations for its decisions, leading to more reliable insights for investors.