Aufdeckung von Versicherungsbetrug bei Kfz-Schäden mit Hilfe des Benford-Tests

The German Insurance Association estimates a yearly amount of damage of € 1.5?bn to German motor vehicle insurance companies because of systematic fraud by insurance holders. It is supposed that about 10% of submitted claim applications contain manipulated data, therefore insurance companies are forced to complete a detailed and cost intensive case-by-case review of each single application. An alternative method to detect fraud in empiric data is the method of digital analysis based on Benford’s law. The Benford method uses a mathematical law of specific logarithmic distribution attributes of first digits. According to this approach, the data of a Benford set confirm with the expected digit distribution, if the data is not manipulated, whereas fraudulent interventions lead to a deviation from Benford’s law. Hence, until now there has not been any investigation whether the Benford method can also be applied on insurance data. The present article analyses a dataset consisting of more than 120,000 damage claim applications to answer this question as well as to identify the impact of specific characteristics on the probability of fraud contained in claim applications, such as the repair of the vehicle in a franchised or an independent workshop, the vehicle brand or the examination by insurance companies experts. Indeed it could be shown that Benford’s Law is only applicable on second digits of insurance data, but delivers very strong results here: All results of the considered characteristics could be verified by plausible arguments. For this reason insurance companies can benefit from making use of the Benford method to identify those claim applications with a high probability of fraud, which should then be reviewed in more detail so that resources can be allocated in a much more cost efficient way.