Special Issue on Dimensionality Reduction, Learning, and Machines
本特刊收录了降维与机器学习在金融计量中的前沿研究，涵盖高维协方差估计、美式期权定价、非参数定价核、抵押贷款风险建模及零售机器人投顾，对金融从业者和研究者有参考价值。
New important research related to dimensionality reduction, learning, and machines is progressively emerging across key areas of Finance and Financial Econometrics. This special issue collects a selected choice of some of this research. Gianluca de Nard, Olivier Ledoit, and Michael Wolf introduce a new methodology for high-dimensional covariance matrix estimation, which combines factor structures and time-varying conditional heteroskedasticity to improve dynamic asset allocation and the detection of cross-sectional stock anomalies. Simon Scheidegger and Adrien Treccani propose a novel numerical framework based on adaptive sparse grids, which enables the efficient solution of American option pricing problems in very high dimensions. Francesco Audrino, Robert Huitema, and Markus Ludwig consider a nonparametric pricing kernel estimator for solving an empirical Ross recovery problem, based on a state transition matrix that is recovered with a neural network. Justin Sirignano, Apaar Sadhwani, and Kay Giesecke employ a deep learning model of multiperiod mortgage risk to analyze an unprecedented dataset of origination and monthly performance records, gaining important implications for mortgage-backed security investors, rating agencies, and housing finance policymakers. Humoud Alsabah, Agostino Capponi, Octavio Ruiz Lacedelli, and Matt Stern introduce a new theoretical reinforcement learning framework for retail robo-advising, in which the robo-advisor learns investor’s risk preferences from her portfolio choices under different market environments, finally improving on the stand-alone investor policy.