AI Investor Barometer

AI is increasingly involved in investment decisions — either directly or indirectly through analysis tools, news aggregations and chatbots. The problem is that a single model appears reliable even when it is systematically over- or under-estimating certain stocks.

When five different models estimate the same company on the same day, you immediately see:

• Do the models agree — or do estimates diverge dramatically?
• Is any model consistently higher or lower than others?
• Does any model change its estimates daily, while another remains stable?
• Does any model stay closer to analyst consensus than another?

These questions are essential if you want to understand how different AI models differ. This tool makes model-specific comparison visible.

Most AI outputs are black boxes: the model gives a number directly without you knowing how it arrived at it. In this project, the approach is different.

This approach makes model-specific differences transparent: if GPT and Claude arrive at different estimates, it's due to different growth assumptions — not because one "calculated incorrectly".

• Not investment advice. All content consists of AI models' computational estimates, not investment advice. Draw your own conclusions.
• AI models can be systematically wrong. Models learn from historical data that may be biased or incomplete. A high confidence score does not mean being correct.
• Public data only. Models use only publicly available financial data and market information.
• Experimental tool. This is an experimental AI comparison and measurement tool. Data may be incomplete, the scheduler may fail, results may be incorrect.

A growing body of academic research examines whether large language models can reason about financial data. Most studies test a single model on a single task — sentiment classification, earnings prediction, or stock price forecasting. This project takes a different approach: five LLMs run the same valuation task daily, over months, producing a behavioural dataset that captures how model opinions form, drift, and diverge.

The platform generates three types of empirical data that are difficult to obtain elsewhere: temporal consistency (how much a model changes its estimate for the same stock day to day), cross-model disagreement (where models converge and where they don't), and assumption-level transparency (the exact CAGR, margin, and WACC inputs each model produces, not just the final number).

Selected references from the emerging LLM-finance literature:

• Kim, A., Muhn, M. & Nikolaev, V. (2024). Financial Statement Analysis with Large Language Models. University of Chicago Booth.
• Lopez-Lira, A. & Tang, Y. (2023). Can ChatGPT Forecast Stock Price Movements? SSRN.
• Xie, Q. et al. (2024). FinBen: A Holistic Financial Benchmark for Large Language Models. arXiv.