LLMs vs. Time-Series Models: Surprising Results in Japanese Stock Predictions

research#llm📝 Blog|Analyzed: Apr 15, 2026 22:44
Published: Apr 15, 2026 09:48
1 min read
Zenn ML

Analysis

This fascinating study reveals an exciting shift in how we approach financial forecasting by testing Large Language Models (LLMs) against dedicated time-series models for Japanese stock predictions. Surprisingly, models like Claude Opus demonstrated clear superiority in practical trading scenarios, showing the incredible potential of LLMs beyond traditional text generation. This innovative application of language models to complex quantitative tasks opens up thrilling new possibilities for the future of AI-driven trading strategies.
Reference / Citation
View Original
"A simple question was tested with a 100-day, 10-stock backtest: "Which is stronger for short-term prediction of Japanese stocks: a time-series foundation model (Kronos) or a Large Language Model (Claude Sonnet/Opus)?" The result was the somewhat surprising conclusion that the LLM showed a clear advantage over the dedicated time-series model."
Z
Zenn MLApr 15, 2026 09:48
* Cited for critical analysis under Article 32.