Paper: arXiv 2308.00065

Authors: Yuwei Yin, Yazheng Yang, Jian Yang, Qi Liu

Abstract

Financial risk prediction plays a crucial role in the financial sector. Machine learning methods have been widely applied for automatically detecting potential risks and thus saving the cost of labor. However, the development in this field is lagging behind in recent years by the following two facts: 1) the algorithms used are somewhat outdated, especially in the context of the fast advance of generative AI and large language models (LLMs); 2) the lack of a unified and open-sourced financial benchmark has impeded the related research for years. To tackle these issues, we propose FinPT and FinBench: the former is a novel approach for financial risk prediction that conduct Profile Tuning on large pretrained foundation models, and the latter is a set of high-quality datasets on financial risks such as default, fraud, and churn. In FinPT, we fill the financial tabular data into the pre-defined instruction template, obtain natural-language customer profiles by prompting LLMs, and fine-tune large foundation models with the profile text to make predictions. We demonstrate the effectiveness of the proposed FinPT by experimenting with a range of representative strong baselines on FinBench. The analytical studies further deepen the understanding of LLMs for financial risk prediction.

Complexity vs Empirical Score

  • Math Complexity: 2.5/10
  • Empirical Rigor: 8.0/10
  • Quadrant: Street Traders — empirical and applied, lower math complexity

Why this score: The paper focuses on applying existing LLM techniques (Profile Tuning) to financial tabular data, with limited novel mathematics, but it demonstrates high empirical rigor through a curated benchmark (FinBench), extensive experiments against strong baselines (XGBoost, etc.), and publicly released code and datasets.

Research Flowchart

  flowchart TD
  A["Research Goal:<br>Modernize Financial Risk Prediction<br>using LLMs & Open Benchmark"] --> B["Develop New Methodology:<br>FinPT - Profile Tuning"]
  B --> C["Create New Dataset:<br>FinBench - Unified Open Benchmark"]
  C --> D["Data Transformation:<br>Convert Tabular Data<br>to Natural Language Profiles"]
  D --> E["Computational Process:<br>Fine-Tune Large Foundation Models<br>with Profile Text"]
  E --> F["Key Findings:<br>Demonstrated Effectiveness vs Baselines<br>Deepened LLM Understanding"]
  F --> G["Outcome:<br>Modernized Approach for<br>Financial Risk Management"]

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

The paper proposes FinPT, a novel profile tuning method for financial risk prediction, and FinBench, a benchmark dataset collection.

Source sentence
To tackle these issues, we propose FinPT and FinBench: the former is a novel approach for financial risk prediction that conduct Profile Tuning on large pretrained foundation models, and the latter is a set of high-quality datasets on financial risks such as default, fraud, and churn.

Assumes

No explicit assumptions found in the text.

Data

  • cd1 dataset · Kaggle benchmark public
    Source sentence
    cd1 dataset 4 comprises data on thousands of clients from a financial institution, including their random identification numbers, banking default status, loan types, gender, age, education, income, credit scores, and other related information.
  • cd2 dataset · Kaggle benchmark public
    Source sentence
    cd2 dataset 5 consists of credit card client information from April 2005 to September 2005 in Taiwan, encompassing default payments, demographic factors, credit data, history of payment, and bill statements.
  • ld1 dataset (The Home Equity dataset, HMEQ) · Kaggle benchmark public
    Source sentence
    ld1 dataset 6 (The Home Equity dataset, HMEQ) comprises detailed records for home equity loans, including a binary target variable that indicates whether an applicant defaulted or was seriously delinquent.
  • ld2 dataset · Kaggle benchmark public
    Source sentence
    ld2 dataset 7 contains tens thousands of different types of loan records with features like age, annual income, employment length (in years), home ownership, loan intent, loan amount, interest rate, historical default, and credit history length, etc.
  • ld3 dataset · Kaggle benchmark public
    Source sentence
    ld3 dataset 8 consists of more than 200K records on vehicle loan, providing information regarding the loan, loanee, and loan history.
  • cf1 dataset · Kaggle benchmark public
    Source sentence
    cf1 dataset 9 summarizes the usage patterns exhibited by numerous credit card users over a period of six months.

Recognised sources: Kaggle

Costs

Not applicable The paper reports no trading or execution result.

Finds 3

  • FinPT outperforms tree-based methods and previous neural models on average F1-score across FinBench datasets, especially when fully fine-tuning large foundation models like GPT-2, T5, and Flan‑T5. predictive-accuracy
    Source sentence
    On average, FinPT outperforms tree-based methods and previous neural models by a large margin, especially when we apply FinPT to fully fine-tune large foundation models such as GPT-2, T5, and Flan‑T5.
  • Flan‑T5 is the best backbone for FinPT, achieving the highest F1-scores on eight datasets, demonstrating the benefit of instruction tuning and scaling. performance
    Source sentence
    When fine-tuning all the parameters on foundation models, Flan‑T5 is the best backbone model for our FinPT approach. It performs the best among all models in eight datasets, showing the extraordinary capacity to predict financial risks.
  • GPT‑2 consistently outperforms BERT models by a large margin in FinPT, indicating decoder‑only Transformers can be better classifiers for this task. performance
    Source sentence
    Also, we find GPT‑2 consistently outperforms BERT models by a large margin, demonstrating that the decoder-only Transformer model using last-token hidden-states prediction can be a better classifier than the encoder-only model using last-layer hidden-states…

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 19 related papers). Coverage is partial, so this is neither replication nor refutation.

Code

https://github.com/YuweiYin/FinPT — the authors' own repository

Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.