Estimation of the ex ante Distribution of Returns for a Portfolio of U.S. Treasury Securities via Deep Learning

Foresti, Andrea

World Bank, Washington, DC

关键词: MACHINE LEARNING;    NEURAL NETWORKS;    CONVOLUTION;    LSTM;    MARKET RISK;   

DOI  :  10.1596/1813-9450-8790
RP-ID  :  WPS8790

学科分类：社会科学、人文和艺术（综合）

来源: World Bank Open Knowledge Repository

【 摘 要 】

This paper presents different deepneural network architectures designed to forecast thedistribution of returns on a portfolio of U.S. Treasurysecurities. A long short-term memory model and aconvolutional neural network are tested as the main buildingblocks of each architecture. The models are then augmentedby cross-sectional data and the portfolio's empiricaldistribution. The paper also presents the fit andgeneralization potential of each approach.