Scholarlink

A correspondent bank serves as a financial institution that offers its services to foreign financial institutions by eliminating the need for physical branches abroad. Efficiently managing the funds allocated to correspondent banks is important for profitability and customer satisfaction. In this study, we focus on predicting daily demand for a domestic bank’s correspondent bank branches in Turkey. For this purpose, we apply statistical and deep learning-based time series forecasting techniques using transaction count data. The experiments are performed on data involving transactions between 2015 and 2021 belonging to 20 correspondent banks. The obtained results indicate that hybrid CNN-LSTM-based model gives promising results for accurate correspondent bank transaction forecasting. It is also observed that the model performance is highly influenced by the specific characteristics of the transaction count time series data. The proposed system offers practical applications, including optimizing fund allocation, risk management, improving customer service through timely transaction processing, and foreign exchange management for both domestic and correspondent banks.

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

Transaction Count Prediction in Correspondent Banking Using a Hybrid CNN-LSTM Neural Network

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Fazil Cuneyt Cucu,
  • C. Okan Sakar

摘要

A correspondent bank serves as a financial institution that offers its services to foreign financial institutions by eliminating the need for physical branches abroad. Efficiently managing the funds allocated to correspondent banks is important for profitability and customer satisfaction. In this study, we focus on predicting daily demand for a domestic bank’s correspondent bank branches in Turkey. For this purpose, we apply statistical and deep learning-based time series forecasting techniques using transaction count data. The experiments are performed on data involving transactions between 2015 and 2021 belonging to 20 correspondent banks. The obtained results indicate that hybrid CNN-LSTM-based model gives promising results for accurate correspondent bank transaction forecasting. It is also observed that the model performance is highly influenced by the specific characteristics of the transaction count time series data. The proposed system offers practical applications, including optimizing fund allocation, risk management, improving customer service through timely transaction processing, and foreign exchange management for both domestic and correspondent banks.