Identifying odometer fraud in used car market data

•The paper proposes a statistical test to identify odometer fraud.•Monte Carlo simulations support its validity.•The test is implemented in a large dataset of Czech car sale ads.•Suspicious patterns are more frequent among imported cars.•The method can be used to identify potentially suspicious sellers.

This paper investigates the presence of odometer fraud in the used car market using a large dataset of car sale advertisements from the Czech Republic. The strategic aspects of sale decisions and the practice of rounding odometer readings, however, render the standard statistical tests for fabricated data invalid. I therefore develop and employ a modification of the last-digit test, which has been used to detect fraud in election data. Simulations using the data from advertisements and travel survey data from the United States support the validity of this test under alternative distributional assumptions. The results suggest that suspicious patterns are more prevalent in the segment of cars imported from abroad. I also show that this test can be used at the seller level, which may be of interest to authorities and market participants.