AI research
LLM Investment Signal Research
Studying how financial inputs change LLM investment signals.
Overview
An experiment on Korean equities, comparing how market data, financial indicators, analyst reports, and disclosures influence LLM investment signals.
Experiment design
- Reproducible experiments. A shared pipeline compares data-input conditions and runs backtests.
- Separate forward evaluation. A forward test examines whether backtest results carry over to a different evaluation period.
- Baseline comparisons. Results are evaluated against baselines, with the limitations of each condition documented.
What the evaluation showed
Some backtest performance did not carry over to the forward test. The project documents those limits alongside the comparisons, rather than treating historical results as evidence of reliable future returns.