[Paper Review] Opportunities and Challenges of Generative-AI in Finance

Akshar Prabhu Desai, Ganesh Mallya+4 more|arXiv (Cornell University)|Oct 21, 2024

Stock Market Forecasting MethodsDecision Sciences3 citations

TL;DR

This paper provides a comprehensive analysis of generative AI (Gen-AI) applications in finance, outlining key opportunities in interactive, assistive, educative, and advisory systems, while identifying critical challenges such as data scarcity, hallucination, latency, and bias. It presents training methodologies including fine-tuning and retrieval-augmented generation (RAG), and demonstrates Gen-AI's impact in areas like fraud detection, risk modeling, and portfolio management, with studies showing up to 30% outperformance in backtested strategies.

ABSTRACT

Gen-AI techniques are able to improve understanding of context and nuances in language modeling, translation between languages, handle large volumes of data, provide fast, low-latency responses and can be fine-tuned for various tasks and domains. In this manuscript, we present a comprehensive overview of the applications of Gen-AI techniques in the finance domain. In particular, we present the opportunities and challenges associated with the usage of Gen-AI techniques. We also illustrate the various methodologies which can be used to train Gen-AI techniques and present the various application areas of Gen-AI technologies in the finance ecosystem. To the best of our knowledge, this work represents the most comprehensive summarization of Gen-AI techniques within the financial domain. The analysis is designed for a deep overview of areas marked for substantial advancement while simultaneously pin-point those warranting future prioritization. We also hope that this work would serve as a conduit between finance and other domains, thus fostering the cross-pollination of innovative concepts and practices.

Motivation & Objective

• To provide a holistic overview of generative AI applications across the financial ecosystem.
• To identify and analyze key opportunities in interactive, assistive, educative, and advisory Gen-AI applications in finance.
• To examine major challenges such as data quality, hallucination, inference latency, and model bias in financial deployment.
• To evaluate training methodologies including prompt engineering, fine-tuning, and RAG for domain-specific financial Gen-AI models.
• To map concrete use cases and quantify performance outcomes in risk management, trading, and compliance.

Proposed method

• Systematic review and synthesis of Gen-AI applications in finance, focusing on large language models (LLMs) and their adaptation to financial tasks.
• Categorization of Gen-AI use cases into four domains: interactive (e.g., chatbots), assistive (e.g., form filling), educative (e.g., financial literacy), and advisory (e.g., trading assistants).
• Evaluation of training techniques: zero-shot/few-shot prompting, instruction tuning, task-specific fine-tuning, and parameter-efficient fine-tuning (PEFT).
• Application of Retrieval-Augmented Generation (RAG) to enhance factual consistency and access to up-to-date financial data in LLMs.
• Analysis of domain-specific models such as BloombergGPT (50B parameters) and StockGPT for financial NLP tasks.
• Synthesis of empirical results from studies on Gen-AI in risk modeling, portfolio rebalancing, and document processing.

Experimental results

Research questions

• RQ1 How can generative AI enhance customer service and support in financial institutions through interactive and assistive applications?
• RQ2 In what ways can Gen-AI improve financial decision-making in investment and risk management, and what performance gains are achievable?
• RQ3 What are the primary technical and ethical challenges in deploying Gen-AI models in regulated financial environments?
• RQ4 How do fine-tuning and RAG techniques improve factual accuracy and domain relevance in financial LLMs?
• RQ5 What are the most promising application areas for Gen-AI in finance, and which require further research?

Key findings

• Gen-AI can reduce human-serviced customer contacts by up to 50%, as estimated by McKinsey, through advanced chatbots and dialog systems.
• BloombergGPT demonstrated strong performance in financial question answering and sentiment analysis while maintaining general language capabilities.
• StockGPT and similar models using attention mechanisms on token sequences showed substantial predictive power in quantitative trading research.
• Portfolio rebalancing using Gen-AI-generated signals outperformed passive indices by 10–30% in backtested studies.
• Gen-AI models enhanced fraud detection, with Mastercard reporting up to a 20% average improvement in detection rates.
• RAG-enhanced models significantly improved factual consistency and up-to-date information access in financial query systems.