Challenge
The determination of residual values is a central component in the leasing business of automotive manufacturers. Since the lessor remains the owner of the vehicle and thus incurs high financial risks, accurate forecasting is of great importance. Our customer’s challenge was to abolish the decentralized calculation of residual values in the individual markets and to develop a central analytics tool for all markets worldwide. In the future, a precise forecast was to be made based on historical data using a machine learning model. The predicted residual values should also be constantly updated over a leasing period in order to incorporate changing market environments into the model.
Approach
The first phase of the project initially involved building up a data history that could ensure a sufficiently long training period for the machine learning models. Since vehicle- and market-specific data were used in addition to contractual information, multiple data sources had to be merged and cleaned. After the creation and exploration of the data set, different statistical models per vehicle class were trained to predict residual values. In this process, a wide variety of factors influencing the residual value of the vehicle were modeled. Primarily linear models were applied to ensure that the results could be interpreted by the company’s specialist department. The models were then made available to the end users with the development of an application housing the trained models.
Results
The model and data preparation were implemented on our customer’s central data science platform and rolled out to three of the largest leasing markets to date. In addition to more precise estimates, this has above all resulted in less manual effort in determining residual values. This has reduced the accounting risk of inaccurate residual value estimates and streamlined the process. The rollout to other leasing markets is currently planned.