Liquidity in the Used Car Market: an Application of Survival Analysis

Absztrakt (kivonat)

Buying a car is a major item in the household consumption basket, so participants in this market make their decisions based on carefully considered arguments. After a few years of use, it is very common to sell a vehicle on the used car market, where it is an important question how long it will take to hand over. In this paper, I build a survival model based on a large dataset of used car ads which I collected from the Internet by webscraping, revealing what makes a used car liquid, which is valuable knowledge for all participants in this market. Clearly, one of the most important causes is the price-to-value ratio, which can be derived from the offer price and the price estimated by machine learning models (Ordinary Least Squares, Regression Tree, Random Forest, eXtreme Gradient Boosting, K-Nearest Neighbor, Neural Network, Support Vector Machine). These can provide reliable estimates in a dataset that contains a large number of variables. For computational reasons, I reduced the dimension of the data by Multiple Correspondence Analysis. In summary, the results show that: (1) eXtreme Gradient Boosting significantly outperforms other prediction methods in estimating the value for money; (2) the value for money is indeed the most important determinant of the time to sell and (3) and some other features of cars can be identified to influence selling prices, such as whether they were American brands, exhibit cars, or had an amplifier output.