[Paper Review] From Natural Language to Executable Option Strategies via Large Language Models

Haochen Luo, Zhengzhao Lai+5 more|arXiv (Cornell University)|Mar 17, 2026

Stock Market Forecasting Methods0 citations

TL;DR

The paper introduces Option Query Language (OQL), a domain-specific intermediate representation that converts natural-language trading intents into executable option strategies via a neuro-symbolic pipeline, improving accuracy and reliability over direct generation baselines.

ABSTRACT

Large Language Models (LLMs) excel at general code generation, yet translating natural-language trading intents into correct option strategies remains challenging. Real-world option design requires reasoning over massive, multi-dimensional option chain data with strict constraints, which often overwhelms direct generation methods. We introduce the Option Query Language (OQL), a domain-specific intermediate representation that abstracts option markets into high-level primitives under grammatical rules, enabling LLMs to function as reliable semantic parsers rather than free-form programmers. OQL queries are then validated and executed deterministically by an engine to instantiate executable strategies. We also present a new dataset for this task and demonstrate that our neuro-symbolic pipeline significantly improves execution accuracy and logical consistency over direct baselines.

Motivation & Objective

• Bridge natural-language intents and executable option strategies using a constrained intermediate representation.
• Reduce context explosion and hallucinations when LLMs interact with high-dimensional option data.
• Provide a benchmark and dataset for natural-language to option-strategy translation in 2025 market conditions.
• Extend Text-to-SQL paradigms to a financial domain with structured option-chain querying.

Proposed method

• Introduce Option Query Language (OQL) as a declarative, role-based, constraint-driven language for option strategies.
• Use a two-stage pipeline: semantic parsing (LLM to OQL) and deterministic execution (OQL engine over option-chain data).
• Formulate problem as y|x,D = sum_z P_theta(z|x) * P_phi(y|z,D), decoupling linguistic parsing from execution.
• Define OQL principles: role-based abstraction, scoped filtering, and semantic soft matching with approximate operators.
• Describe neuro-symbolic execution: AST parsing, role-constraint verification, vectorized filtering, and Cartesian assembly to produce executable strategies.
• Provide a dataset of 200 NL instructions constrained to 2025 market data, and evaluation protocols for query-level and strategy-level performance.

Experimental results

Research questions

• RQ1 Can LLMs reliably generate executable OQL queries that yield non-empty and correct option strategies?
• RQ2 How do different LLMs perform in generating valid and faithful OQL queries compared to baselines?
• RQ3 Do OQL-derived strategies outperform baselines that operate on raw option data, and how do they perform in backtests?

Key findings

• All large models achieved Validity Rates (VR) above 0.87 for generated OQL queries.
• Specialized coding models (e.g., DeepSeek-Coder-6.7B) can outperform larger general models on this task.
• OQL outperforms unstructured baselines (FFLG, PCG) and standard Text-to-SQL in reliability and risk management.
• OQL reduces dangerous hallucinations compared to SQL-based baselines and achieves favorable win rates and profitability.
• OQL offers efficient token usage with a high cache hit rate, making it cost-effective for strategy retrieval.
• Across assets and market regimes, OQL enables models to tailor strategies to conditions rather than producing uniform outputs.