yuweiyin / finpt

FinPT: Financial Risk Prediction with Profile Tuning on Pretrained Foundation Models

FinPT: Financial Risk Prediction with Profile Tuning on Pretrained Foundation Models

• 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by the following two facts: 1) the algorithms used are somewhat outdated, 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.

Environment

conda create -n finpt python=3.9
conda activate finpt
pip install -r requirements.txt

Data

• FinBench on Hugging Face Datasets: https://huggingface.co/datasets/yuweiyin/FinBench

from datasets import load_dataset

# ds_name_list = ["cd1", "cd2", "ld1", "ld2", "ld3", "cf1", "cf2", "cc1", "cc2", "cc3"]
ds_name = "cd1"  # change the dataset name here
dataset = load_dataset("yuweiyin/FinBench", ds_name)

Experiments

The instructions obtained in Step 1 and customer profiles generated in Step 2 are provided as X_instruction_for_profile and X_profile in FinBench.

Run Tree-based Baselines

SAVE_DIR="./log/baseline_tree/"
mkdir -p "${SAVE_DIR}"

DATASETS=("cd1" "cd2" "cd3" "ld1" "ld2" "cf1" "cc1" "cc2" "cc3")
MODELS=("RandomForestClassifier" "XGBClassifier" "CatBoostClassifier" "LGBMClassifier")
SEEDS=(0 1 42 1234)

for dataset in "${DATASETS[@]}"; do
  for model in "${MODELS[@]}"; do
    for seed in "${SEEDS[@]}"; do
      echo -e "\n\n\n>>> run_step3_baseline_tree.py: dataset: ${dataset}; model: ${model} seed: ${seed}"
      python run_step3_baseline_tree.py --ds_name "${dataset}" --model_name ${model} --seed ${cur_seed} --grid_search \
        > "${SAVE_DIR}/${dataset}-${model}-${seed}.log"
    done
  done
done

Run Neural Network Baselines

SAVE_DIR="./log/baseline_nn/"
mkdir -p "${SAVE_DIR}"

DATASETS=("cd1" "cd2" "cd3" "ld1" "ld2" "cf1" "cc1" "cc2" "cc3")
MODELS=("DeepFM" "STG" "VIME" "TabNet")
SEEDS=(0 1 42 1234)

for dataset in "${DATASETS[@]}"; do
  for model in "${MODELS[@]}"; do
    for seed in "${SEEDS[@]}"; do
      echo -e "\n\n\n>>> run_step3_baseline_nn.py: dataset: ${dataset}; model: ${model} seed: ${seed}"
      python run_step3_baseline_nn.py --cuda "0" --ds_name "${dataset}" --model_name ${model} --seed ${cur_seed} \
        > "${SAVE_DIR}/${dataset}-${model}-${seed}.log"
    done
  done
done

Run FinPT

SAVE_DIR="./log/finpt/"
mkdir -p "${SAVE_DIR}"

DATASETS=("cd1" "cd2" "cd3" "ld1" "ld2" "cf1" "cc1" "cc2" "cc3")
MODELS=("bert" "finbert" "gpt2" "t5-base" "flan-t5-base" "t5-xxl" "flan-t5-xxl" "llama-7b" "llama-13b")

for dataset in "${DATASETS[@]}"; do
  for model in "${MODELS[@]}"; do
      echo -e "\n\n\n>>> run_step3_finpt.py: dataset: ${dataset}; model: ${model} seed: ${seed}"
      python run_step3_finpt.py --cuda "0,1" --ds_name "${dataset}" --model_name ${model} --use_pos_weight \
        > "${SAVE_DIR}/${dataset}-${model}-${seed}.log"
  done
done

License

Please refer to the LICENSE file for more details.

Citation

@article{yin2023finpt,
  title   = {FinPT: Financial Risk Prediction with Profile Tuning on Pretrained Foundation Models},
  author  = {Yin, Yuwei and Yang, Yazheng and Yang, Jian and Liu, Qi},
  journal = {arXiv preprint arXiv:2308.00065},
  year    = {2023},
  url     = {https://arxiv.org/abs/2308.00065},
}

finpt's Issues

Reproducibility

Hello, thank you for releasing your code. However, I have a question.

When I tried to reproduce the result of FinBERT on CF1, the test F1 score was not the same as in your paper.

Paper: 45.1
Reproduced: 34.3
I suspect the problem may stem from unspecified package versions. Although I have spent a lot of time trying to align the package versions with the code, I almost didn't modify the code.

Could you inform me about the specific versions of the packages? If you have any ideas about other possible issues, sharing them would also help me solve this problem.

Reproducibility of the results

Hi,
I have been trying to re-implement the paper but in many of the combinations of the dataset and model that is being for training and evaluation, the F1-scores and accuracies are at a constant value from the very first epoch till the last (F1-score is at 0 and Accuracy at 0.9153) although the loss decreases. I have not changed any of the code and am unable to understand the issue.
For some combination of dataset and model it works well but not for others. Currently for ld1 dataset and bert model this is the issue similarly for many others as well.

Can you please help me sort out the issue?

Model Checkpoint

Limited by computer resources, could you please provide the trained model parameters file? Thanks a lot!!!