Option-Implied Return Distribution Guided Option Trading Agent Framework
ORIGINAL / SOTA: Stock Options Trading Agents Guided by Option-Implied Return Distributions
This work introduces SOTA, a framework that abstracts thousands of option contracts into strategy-level decisions, with deterministic resolvers handling portfolio implementation to avoid fixed strategy structures. Its novelty lies in using option-implied return distributions to guide a language-model agent in structured option-strategy selection, evaluated against rule-based and machine-learning selectors in the same trading environment.
01 ABSTRACT
The authors present SOTA, an agentic trading framework for structured option-strategy selection that abstracts the large option universe into strategy-level decisions while deterministic resolvers handle portfolio implementation. They post-train Qwen3.8-27B with supervised fine-tuning and reinforcement learning, and evaluate on options on nine large-cap U.S. equities and SPY. The authors report an 18.3% total return, Sharpe ratio of 1.60, and maximum drawdown of 8.96% over a six-month out-of-sample period, and document an asymmetric role of news: news improves frontier-teacher trajectories, but retaining news during reinforcement learning reduces out-of-sample return from 18.3% to -2.7%. These are author-reported results requiring independent verification.
02 KEY FINDINGS
- SOTA abstracts the large option universe into strategy-level decisions, with deterministic resolvers handling portfolio implementation.
- The agent is post-trained from Qwen3.8-27B using supervised fine-tuning followed by reinforcement learning.
- The authors report 18.3% total return, Sharpe ratio of 1.60, and maximum drawdown of 8.96% over a six-month out-of-sample period.
AI GENERATED SUMMARY / DISCOVERED BY ARXIV CS.LG · ARXIV Q-FIN