Scholarlink

This paper explores the use of Neural Networks (NNs) and Reinforcement Learning (RL) to optimize financial decision-making in corporate treasury management, particularly for forecasting the cash in and out of the cash pool. This is due to the fact that precise forecasting of the cash flow and liquidity will promote effective decision-making during dynamic and complex environments. A three-modular combination based on Long Short-Term Memory (LSTM), Deep Q-Networks (DQNs), and Autoencoders is applied using the dataset about historical and actual financial information towards the estimation of future cash flow and liquidating levels. There are inflow and outflows of cash flows, accounts for receivable, payable, detailed transactional files, interest rate, and changes in currency fluctuation along with stock markets together with indicators and measures related to the general environment. The models’ accuracy, precision, recall, F1 score, and AUC-ROC are used in the evaluation criteria. DQN follows closely thereafter, showing superior decision-making and Autoencoders contribute by having the ability of anomaly detection thus ensuring data quality. The findings depict the complementary strengths of these models in enhancing corporate treasury operations. The best model for liquidity forecasting is LSTM, and DQN is the dynamic decision-making support model. Altogether, this research presents a robust and efficient solution for financial forecasting in corporate treasury management through advanced machine learning techniques.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Optimizing Financial Decision-Making in Corporate Treasury Management Using Reinforcement Learning and Neural Networks for Liquidity Forecasting

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Moghal Yaseen Pasha,
  • Ankur Maurya,
  • Jagendra Singh,
  • Lakhan Patidar,
  • Harnit Saini,
  • Anshita Tyagi

摘要

This paper explores the use of Neural Networks (NNs) and Reinforcement Learning (RL) to optimize financial decision-making in corporate treasury management, particularly for forecasting the cash in and out of the cash pool. This is due to the fact that precise forecasting of the cash flow and liquidity will promote effective decision-making during dynamic and complex environments. A three-modular combination based on Long Short-Term Memory (LSTM), Deep Q-Networks (DQNs), and Autoencoders is applied using the dataset about historical and actual financial information towards the estimation of future cash flow and liquidating levels. There are inflow and outflows of cash flows, accounts for receivable, payable, detailed transactional files, interest rate, and changes in currency fluctuation along with stock markets together with indicators and measures related to the general environment. The models’ accuracy, precision, recall, F1 score, and AUC-ROC are used in the evaluation criteria. DQN follows closely thereafter, showing superior decision-making and Autoencoders contribute by having the ability of anomaly detection thus ensuring data quality. The findings depict the complementary strengths of these models in enhancing corporate treasury operations. The best model for liquidity forecasting is LSTM, and DQN is the dynamic decision-making support model. Altogether, this research presents a robust and efficient solution for financial forecasting in corporate treasury management through advanced machine learning techniques.