Factors affecting crash risk within the car-sharing market

Abstract

As the sharing economy becomes increasingly more popular, crash risk assessment has become important not only for insurance companies, but also for companies engaged in the car-sharing business. As such, linear regression and machine learning methods, such as regression trees and random forests, were used to model crash risk based on the observations retrieved from car-sharing systems. The evidence shows that the average daily trip duration, the month of the crash event, and the car brand have the greatest impact on crash rates, while holiday, working day or weekend; peak hour; and gender of the driver hold no valuable information for predicting crash risk. After a proper assessment of the risk indicators that have the greatest impact on the occurrence of crashes, companies might be able to enter into personalised car-sharing pricing by developing usage-based or pay-as-you-drive insurance products.