SOTA: STOCK OPTION TRADING AGENTS GUIDED BY OPTION-IMPLIED RETURN DISTRIBUTIONS

Battle Count: 51

Highly relevant as it addresses the complex decision space of options trading using modern LLM agents, demonstrating superior performance over traditional rule-based and ML baselines in a realistic trading environment.

High complexity due to the need for a custom trading environment, integration of option pricing models (BSM), deterministic resolvers for contract mapping, and a two-stage training pipeline (SFT + RL/GRPO).

The paper provides detailed hyperparameters, data sources (OPRA, OptionMetrics, CRSP, EDGAR), and algorithmic descriptions. However, the specific frontier teacher model used for SFT generation is not explicitly named beyond 'frontier-model', and the code is not explicitly linked in the provided text.