Quant-Arena: Benchmarking LLM Agents for Quantitative Trading Research

Quant-Arena. Benchmarking LLM Agents for Quantitative Trading Research. By Houde Qian and eleven others, from Navers Lab at Einsia AI. Published on the twenty-first of October. No code repository is linked to it.

Quantitative research is an iterative loop. A researcher proposes a market hypothesis, implements it as an executable signal, tests it on historical data, and decides what to keep. Then the loop begins again. Existing benchmarks score pieces of that loop: formulas, code, or individual trading decisions. Left open is whether large language models can complete the whole loop, and produce strategies that keep working beyond the research period. That second part is the hard one, since a strategy tuned to history can look excellent on the very data it was built from.

Quant-Arena is a benchmark and public leaderboard for end-to-end quantitative research on real A-share market data. It asks a model to run the whole loop, and then judges the result on a period the model never saw.

A model constructs features and stock-scoring code. It receives feedback only on two thousand twenty-two through two thousand twenty-three. It then freezes a single strategy, which is assessed on the sealed year two thousand twenty-four. A cash-account evaluator applies trading constraints and costs, including market impact estimated from a simulated order book. Scores are dominated by realized excess return, and are complemented by signal quality and by the diversity of features the strategy uses.

Nine frontier models were evaluated. GPT-6 Astra scores highest among them. Yet no model beats the equal-weight market in two thousand twenty-four.

The headline is a leaderboard with no one above the line that matters. The best model ranks first among its peers and still trails a portfolio that simply holds the market in equal weights. The sealed year is what makes that visible. Quant-Arena stays open to submissions from any model or agent harness, so the line remains there to be crossed.