The Bond Language for Interest Rates Prediction: A Transformer Model Approach

Considering the development of bond market, effective analysis of the interest rate could help bond investment managers obtain higher profits and avoid potential loss. However, prior literature that using natural language process to discuss the interest rate is quite limited. Our study builds bond language and adopts a Transformer model to infer the tendency of 1/5/10 year Treasury bonds' interest rates. Our sample is interest rates data and economic data from 2016.1.20 to 2023.9.15. We find that the difference between predicting results and true tendency of interest rates is relatively similar. Moreover, the application of correcting mechanism shows that considering the impact of the release of economic data could improve the accuracy of our model. This study provides evidence that researchers could consider the weight of disparate indicators during the analysis on bond market.