DeepBonds: A Deep Learning Approach to Predicting United States Treasury Yield

Abstract

United State Treasury Bonds are government bonds issued by the United State Treasury through the Public Debt Bureau. The trades of U.S. Treasury Bonds have a huge influence on global economy. To analysis the trend of global economy, many economists believe U.S. Treasury Yield has the ability to predict the fluence of other financial markets such as stock market, futures market, Option market, etc. However, However, most financial prediction models focus only on predicting stock price, which is a sort of multidimensional time-series prediction. Although U.S. Treasury Yield could be viewed as a multidimensional time-series, the prediction models for predicting stock price are not able to completely satisfy the requirements for predicting U.S. Treasury Yield. Besides, most traditional machine learning methods focus only on estimation of short-term cash flow. As the result, the loss of traditional machine learning methods would significantly be increased while the period of prediction target is fluctuated. In this paper, we propose a Deep-Learning framework, DeepBonds, to build a prediction model to predict U.S. Treasury Yield with different issue period. Meanwhile, the Recurrent Neural Network with Long Short Term Memory (LSTM) architecture is utilized for effectively summarizing U.S. Treasury Yield as characteristic vectors. Based on the produced characteristic vectors, we can precisely predict future U.S. Treasury Yield with different issue period. We conduct a comprehensive experimental study based on a real dataset collected from the website of Resource Center of U.S. Department of The Treasury. The results demonstrate a significantly improved accuracy of our Deep Learning approach compared with the existing works.