ChatGPT Vs. DeepSeek: What 5,000 Chinese Stocks Reveal About AI’s Limits

Key Takeaways

• ChatGPT gives higher price targets for Chinese firms
• DeepSeek forecasts align closer to actual outcomes
• Bias linked to limited US media coverage of China
• Adding Chinese news eliminates prediction gap
• Similar performance observed for US‑listed companies

Summary

Harvard Business School researchers examined how ChatGPT and China‑based DeepSeek evaluate roughly 5,000 Chinese publicly listed firms. The study found a pronounced “foreign bias”: ChatGPT consistently issued higher price targets and more buy recommendations than DeepSeek, yet its forecasts were less accurate. The bias stems from limited U.S. media coverage of Chinese companies, which disappears when Chinese news sources are incorporated. Both models performed similarly on U.S. stocks, highlighting the role of information environments in shaping AI‑driven financial signals.

Pulse Analysis

The rapid adoption of large‑language models in finance promises faster analysis, but the ChatGPT‑DeepSeek comparison underscores a hidden vulnerability: data provenance. When AI tools are trained primarily on Western sources, they inherit a coverage bias that skews sentiment toward optimism for markets with scant exposure. For Chinese equities, this manifests as inflated price targets and bullish recommendations from ChatGPT, despite poorer predictive accuracy. Investors relying on such signals risk overpaying for stocks whose fundamentals are misrepresented.

Understanding the mechanics behind this bias reveals a broader lesson for AI deployment. Information availability, not model sophistication, drives divergent forecasts. By feeding Chinese‑language news and local analyst reports into the models, researchers eliminated the optimism gap, aligning ChatGPT’s outputs with DeepSeek’s more grounded predictions. This suggests that integrating region‑specific data pipelines can dramatically improve model reliability, a practice that should become standard for asset managers seeking global coverage.

For policymakers and market participants, the findings raise questions about the transparency and regulation of AI‑driven analytics. As roughly 40 % of institutional investors already use AI for market insights, ensuring that models incorporate balanced, multilingual data sources is crucial to prevent systemic mispricing. Firms that proactively audit their AI inputs and diversify data feeds will gain a competitive edge, while those overlooking these nuances may expose themselves to hidden risks in an increasingly algorithm‑centric investment landscape.