Abstract:

Instant interest rates offered to customers are among the most critical topics in the banking sector. It is very important that the offers are at a suitable level for both the bank and the customer. In this study, approximately 8 months of data were used and forecast models were designed for the estimation of the price offers given to the customer. A total of 18 different works have been carried out. In the study, dependent variables were determined as customer segment, instant currency value, day information, time information and volatility value. Independent variable is the margin. The training was carried out with daily data and using random trees (RF), the Gradyan Increasing Machine (GBM), artificial nerve networks (ANN), deep nerve networks (DNN), evolutionary nerve networks (CNN) and Elman nerve networks algorithms. The random search algorithm was used to find the hyper-parameters of the algorithms and the models with the lowest error value were selected to be used in the forecast stage by comparing the results of the model training. For the measurement of success, the MSE (Middle Quadratic Error) and the MSE (Middle Absolute Error) error functions have been used. According to the training conducted through three different models, artificial nervous networks and evolutionary nervous networks algorithms have been observed to give better results than other algorithms. The forecast time for 18 currencies is about 3s.

Keywords: