Paper: arXiv 2510.05533
Authors: Weilong Fu
Abstract
Large language models are reshaping quantitative investing by turning unstructured financial information into evidence-grounded signals and executable decisions. This survey synthesizes research with a focus on equity return prediction and trading, consolidating insights from domain surveys and more than fifty primary studies. We propose a task-centered taxonomy that spans sentiment and event extraction, numerical and economic reasoning, multimodal understanding, retrieval-augmented generation, time series prompting, and agentic systems that coordinate tools for research, backtesting, and execution. We review empirical evidence for predictability, highlight design patterns that improve faithfulness such as retrieval first prompting and tool-verified numerics, and explain how signals feed portfolio construction under exposure, turnover, and capacity controls. We assess benchmarks and datasets for prediction and trading and outline desiderata-for time safe and economically meaningful evaluation that reports costs, latency, and capacity. We analyze challenges that matter in production, including temporal leakage, hallucination, data coverage and structure, deployment economics, interpretability, governance, and safety. The survey closes with recommendations for standardizing evaluation, building auditable pipelines, and advancing multilingual and cross-market research so that language-driven systems deliver robust and risk-controlled performance in practice.
Complexity vs Empirical Score
- Math Complexity: 6.5/10
- Empirical Rigor: 4.0/10
- Quadrant: Lab Rats — high math complexity, limited empirical testing
Why this score: The paper discusses sophisticated mathematical and computational concepts like transformer architectures, efficient tuning (LoRA, quantization), and multimodal reasoning, but its summary and excerpt focus on surveying research, proposing taxonomies, and identifying challenges (e.g., temporal leakage, governance) rather than presenting specific backtests, statistical metrics, or implementation details.
Research Flowchart
flowchart TD A["Research Goal:<br>Assess LLMs for<br>Equity Prediction & Trading"] --> B["Data & Inputs:<br>Unstructured Info<br>e.g., News, Filings"] B --> C["Core Methodology:<br>Task-Centered Taxonomy"] C --> D["Computational Processes:<br>RAG, Time Series Prompting, Agentic Systems"] D --> E["Key Findings & Outcomes:<br>Standardized Benchmarks, Auditable Pipelines, Risk-Controlled Signals"]
Evidence Card
What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card
Proposes
We propose a task‑centered taxonomy that spans sentiment and event extraction, numerical and economic reasoning, multimodal understanding, retrieval‑augmented generation, time series prompting, and agentic systems that coordinate tools for research, backtesting, and execution.
Source sentence
We propose a task-centered taxonomy that spans sentiment and event extraction, numerical and economic reasoning, multimodal understanding, retrieval-augmented generation, time series prompting, and agentic systems that coordinate tools for research, backtesting, and execution.
Assumes
No explicit assumptions found in the text.
Data
No dataset named in the text.
Costs
Not applicable
Finds
No quantified findings extracted.
Checked by others
No other paper in the archive has been found to support or contradict these findings. Coverage is partial, so this is neither replication nor refutation.
3 papers in the archive
Code
No repository by the authors found in the paper.
Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.