Default Prediction of Commercial Real Estate Properties Using Machine Learning Techniques

Abstract

Academics and analysts have mostly employed stochastic and statistical default models to project defaults for properties backing commercial mortgage-backed securities. Although over the last few years there has been increased interest in using machine learning models to predict defaults on consumer loans, there has not been a comprehensive study that uses machine learning techniques to predict commercial real estate loan defaults. In this article, the authors investigate the use of machine learning techniques to predict defaults for commercial real estate property loans. The authors assess the performance of classification techniques based on machine learning (support vector machine, random forest, boosting, and classification tree) compared to the performance of the typical statistical technique. The principal findings of this study are that the support vector machine technique for predicting defaults on commercial property loans significantly outperforms other methods, and it has stable performance in imbalanced datasets. Moreover, the boosting technique identified the ratio of the capitalization rate spread to the average capitalization rate spread of property type as the most important driver of defaults in commercial real estate loans.

TOPICS: Real estate, big data/machine learning

Key Findings

• • To predict defaults and future performance of commercial property loans, a model has to be developed that incorporates the property characteristics of the individual properties.

• • Support vector machine technique (a type of machine learning) for predicting defaults on commercial property loans significantly outperforms typical statistical default models.

• • The machine learning technique of boosting identified the ratio of the capitalization rate spread to the average capitalization rate spread of property type as the most important driver of defaults in commercial real estate loans.