Long-Term Interbank Bond Rate Prediction Based on ICEEMDAN and Machine Learning

Abstract

The application of time series forecasting utilizing historical data has become increasingly essential across a variety of industries including finance, healthcare, meteorology, and industrial sectors. The assessment of bond transaction rates in the interbank bond market serves as a crucial indicator for assessing bank risk. In this paper, we proposed a composite model to forecast the transaction interest rates of China’s interbank bonds over a long period. Specifically, our model integrates an intrinsic complete ensemble empirical mode decomposition with adaptive noise (ICEEMDAN) model along with various long-term prediction models including long short-term memory network, temporal convolutional network, transformer, and autoformer. Our findings reveal that: 1) predictive performance of different long-term prediction models varies across different frequencies of single time series data; 2) predictive efficacy of diverse model combinations differs across varying prediction time lengths; 3) best results can be realized by using different prediction model combinations for high-frequency, medium-frequency and low-frequency data under different time steps.

Keywords

• Long-term forecasting
• ICEEMDAN
• machine learning
• interbank bond rate