HomeArticlesServicePriceAbout

A recent study, ChatGPT and DeepSeek: Can They Predict the Stock Market and Macroeconomy?, investigates whether large language models (LLMs) such as ChatGPT-3.5 and DeepSeek-R1 can predict stock market returns and macroeconomic trends by analyzing Wall Street Journal (WSJ) headlines.

Key findings reveal that ChatGPT-3.5 demonstrates predictive power by capturing delayed investor reactions to positive news, whereas DeepSeek and smaller models like BERT show weaker forecasting ability due to their training limitations.

Methodology

• Data Collection: Headlines from WSJ front-page articles (1996–2022).

• Sentiment Analysis: ChatGPT-3.5 and DeepSeek-R1 classify headlines as positive, neutral, or negative.

• Regression Analysis:

Monthly positive news ratio (NRG) and negative news ratio (NRB) are used as independent variables.

Stock market returns and macroeconomic indicators are dependent variables.

• Robustness Testing: Uses ChatGPT-4, fine-tuning, and alternative prompts to validate results.

Key Findings

1. ChatGPT-3.5’s Market Predictive Ability

Positive News Ratio (NRG) as a Predictor:

• Higher NRG predicts higher stock returns in the following months.

• Regression analysis shows NRG has a 0.53% coefficient on next-month stock returns (statistically significant at 5% level).

• R-squared increases to 8.52% for 12-month predictions, indicating stronger long-term effects.

Delayed Market Reaction to Positive News:

• Investors fail to immediately incorporate positive news, leading to a gradual price adjustment.

Limited Impact of Negative News (NRB):

• Negative news is priced in immediately, showing no predictive power for future returns.

2. Economic Conditions & News Impact

NRG is More Predictive During Recessions:

• During economic downturns, positive news is often undervalued, making it a stronger signal for future stock movements.

Novelty Effect:

• ChatGPT identifies more unique, impactful news—reinforcing its ability to detect high-value information.

Economic Policy Uncertainty (EPU) Interaction:

• In times of high policy uncertainty, ChatGPT’s predictions improve as investors struggle to interpret complex economic signals.

3. Comparing ChatGPT to Other AI Models

DeepSeek’s Limitations:

• DeepSeek, despite being structurally similar to ChatGPT, performs worse due to less training on English financial news.

BERT’s Weak Performance:

• Traditional models like BERT fail to capture nuanced financial signals, making them less effective in market prediction.

4. Methodological Considerations

Prompt Engineering Effects:

• Different prompts for sentiment classification can affect results.

Data Period Dependency:

• Findings are based on 1996–2022 data, meaning different periods could yield different results.

[Compliance Note]

• All posts by Sellsmart are for informational purposes only. Final investment decisions should be made with careful judgment and at the investor’s own risk.

• The content of this post may be inaccurate, and any profits or losses resulting from trades are solely the responsibility of the investor.

• Core16 may hold positions in the stocks mentioned in this post and may buy or sell them at any time.