Machine Learning Approaches to Credit Scoring and Portfolio Optimization in Wealth Management

Abstract

There is a wealth of information available to wealth managers for analyzing clients’ profiles and their financial needs. Machine learning techniques can help to extract better insights from this data. Using machine learning, wealth managers can make better predictions about clients’ life events and financial preferences, resulting in better financial advice and improved revenues. Besides client profiling and advising, machine learning can also be used to optimize the asset management process. Machine learning techniques can now outperform other statistical models traditionally used in finance in both credit scoring and portfolio returns prediction, while also offering properties such as dimensionality reduction, transfer learning, and semi- or unsupervised learning that are not available in most competing models. Wealth managers can benefit from using machine learning to better predict default and portfolio returns, resulting in increased performance from better portfolio construction and rebalancing.

Although machine learning has received more and more attention in finance over the past years and has been successfully applied to a wide variety of areas, some other aspects have only recently started attracting attention. The majority of papers in the asset management field for instance still rely on classifiers traditionally used in finance, such as linear models. Moreover, even though they highlight the appealing properties of machine learning methods, there are still too many asset managers that remain skeptical about the usefulness of machine learning models and the risks surrounding their implementation. We explore these issues by applying and comparing several machine learning algorithms over real-world financial data for the problems of credit scoring and portfolio optimization. We show that most algorithms can indeed lead to better predictive performances, leading to well-founded financial decisions.