Thoughts about the next stage research

• 1. Benchmarks for (Financial) LLMs
  • 1.1 Multi-round Benchmark (Ongoing)
  • 1.2 Multimodal Benchmark (Would Begin Soon)
• 2. Research on Online Social Media Platforms like Xueqiu (Ongoing)
• 3. Integrating Large Language Models with Financial Applications (Conceptual Stage)
  • 3.1 Using Satellite Imagery in Financial Analysis
  • 3.2 Assisting in Calculating Company Innovation Indices
  • 3.3 Assessing Population Aging for Financial and Social Impact

After two months as a Research Assistant at CUHKSZ, it feels like the right time to plan ahead and set goals for the next stage of my research and professional development. The following are key areas I want to focus on, including ongoing projects and new initiatives.

1. Benchmarks for (Financial) LLMs

LLMs have been making groundbreaking advancements in various fields, and financial technology is no exception. However, to unlock their full potential in this domain, the development of specialized benchmarks that account for the unique requirements of financial systems is crucial. This project focuses on designing and implementing benchmarks specifically tailored to evaluate and improve the performance of financial LLMs.

1.1 Multi-round Benchmark (Ongoing)

The complexity of financial decision-making often requires multi-step reasoning and interactions. Therefore, a single-turn evaluation would not sufficiently capture the nuanced performance needed for real-world financial tasks. A multi-round benchmark, by contrast, would evaluate LLMs across extended dialogues, simulating real-world interactions where financial models are engaged in iterative decision-making processes.

1.2 Multimodal Benchmark (Would Begin Soon)

In the financial world, language is not the only input format; financial analysis often involves interpreting graphs, tables, charts, and other non-textual data. Therefore, a comprehensive benchmark should include multimodal inputs, such as combining textual financial news with accompanying stock charts.

2. Research on Online Social Media Platforms like Xueqiu (Ongoing)

Online social media platforms like Xueqiu (雪球) have become valuable hubs where influential voices discuss market trends, companies, and their own investment strategies. The goal of this research is to analyze the semantic association between influential figures and the companies they mention, while also investigating the alignment between their publicly stated opinions and actual trading behavior (both in real and simulated markets).

3. Integrating Large Language Models with Financial Applications (Conceptual Stage)

To truly revolutionize financial technology, integrating LLMs into practical financial applications is essential. This section explores how LLMs can be embedded into various tools to perform sophisticated financial analysis and decision-making, with a focus on three key areas: 1) leveraging LLMs to interpret satellite imagery for financial insights, 2) assisting in calculating company innovation indices, and 3) helping assess the degree of aging in populations.

3.1 Using Satellite Imagery in Financial Analysis

Satellite imagery provides an untapped source of real-time data that can offer unique insights into market movements, especially in sectors like agriculture, infrastructure, and transportation. The goal is to leverage LLMs to analyze this imagery, uncover trends, and translate the findings into meaningful financial interpretations.

3.2 Assisting in Calculating Company Innovation Indices

Innovation is a critical driver of company growth and market valuation. LLMs can play a significant role in quantifying a company’s innovation capabilities by analyzing diverse data sources.

3.3 Assessing Population Aging for Financial and Social Impact

Aging populations pose significant challenges for both economies and financial systems, especially in regions where pension and healthcare costs are rising. LLMs can assist in modeling the aging degree of populations and assessing the financial impact.