Can LLMs jointly reason over fundamentals and trading signals?

Determine whether current large language models can correctly answer financial questions that require joint reasoning over company fundamentals extracted from SEC 10-K/10-Q filings and trading signals computed from historical price and volume data, rather than relying on either information source in isolation.

Background

Most prior financial QA benchmarks focus on textual fundamentals from regulatory filings and tables, with limited or no evaluation of trading-signal reasoning based on historical price dynamics. As a result, model capability on tasks that demand integrating these heterogeneous signals remains uncertain.

FinTradeBench is introduced specifically to evaluate this joint reasoning capability by combining questions grounded in fundamentals, trading signals, and their hybrid interactions across NASDAQ-100 companies over a decade.