Primary Submission Category: Machine Learning and Causal Inference
THE BIOPHILIA PREMIUM: VISUAL PLANT IMAGERY AND PROPERTY VALUATION
Authors: Yuqian Chang, Nathan Fong, Ning Ye, Maureen Morrin,
Presenting Author: Yuqian Chang*
Homebuyers’ reliance on real estate websites has increased the importance of optimizing a home’s online profile. We investigate the influence of biophilic imagery (i.e., images depicting plant life) on property valuation, and demonstrate the biophilia premium with a large unstructured dataset of 40,294 sold homes with 569,828 digital images. We employ a three-step framework with multiple machine learning tools to attribute price differences to biophilic images as opposed to other correlated features of the listings, including 1) a highly accurate biophilic image classifier founded upon a two-stage deep learning algorithm (Faster R-CNN) to minimize measurement error, 2) assembling extensive covariates diagnosed as effective to address the most plausible confounders, and 3) using double machine learning to maximize the predictive power of the covariates. Homes with more biophilic images garnered more likes on the website and higher intent to visit, ultimately exhibiting transaction values 0.13% higher per biophilic image. We further reveal the heterogeneity of the effect identified, showing that the biophilia premium is stronger in areas with lower prevalence of plant imagery usage and higher property homogeneity. These areas tend to have lower socioeconomic status, property valuation, and image quality, suggesting resource constraints may be the reason behind the limited usage of biophilic imagery in certain markets, and thus providing direct implications to practitioners.