Price Forecast of Treasury Bond Market Yield: Optimize Method Based on Deep Learning Model

Abstract

Accurate forecasting of the treasury bond market is beneficial for financial institutions to formulate investment research strategies and for national managers to build a modern financial system. This paper integrates the ideas of improved multivariate time series sampling and deep learning prediction model structure optimization, and proposes an optimized deep learning model framework under the LASSO-SMLR-PCA machine learning method. Through the LASSO and SMLR methods, the multicollinearity of the multivariate time series is reduced and the variables with insignificant correlation coefficients are eliminated. Then, the PCA method is used for dimensionality reduction and reconstruction, and finally, the LSTM deep learning model with Bayesian optimized hyperparameters is used to achieve rolling time prediction of the treasury bond market yield price. The empirical results show that the optimized deep learning model performs excellently in terms of evaluation indicators for treasury bond yield price forecasting, with accurate curve fitting, efficient model structure, and stable and effective practical application.