Combating insurance fraud: a structural econometrics approach

Read time: 8 minutes

News & Blog Blog Combating insurance fraud: a structural econometrics approach

Combating insurance fraud: a structural econometrics approach

Read time: 8 minutes
Prof. Dr. Roger Laeven and Yuan Yue MSc of the University of Amsterdam give an insight into the aspects of research into insurance fraud. This article discusses the aspects of ‘adverse selection’ and ‘moral hazard’ and explains the merits of a structural econometrics approach.
While it is the era of big data and information, the problem of information asymmetry, arising when one party has more information than another, and its adverse consequences have not ceased to exist. A key example of a problem of information asymmetry is insurance fraud, a phenomenon that has regrettably become increasingly relevant. In 2013, an appalling 1.3 billion pounds of fraudulent claims were uncovered in UK. This is believed to be only the tip of the iceberg.

Insurance companies strive to combat insurance fraud not just for their own benefit but also for the rights of their honest law-abiding customers. They develop auditing and monitoring strategies, design new screening and detection mechanisms, and set up investigation units. Yet such efforts are costly and can be inefficient. It is believed that there are important advantages to approaching the problem of insurance fraud in a more structural way, which we will briefly outline below.

Adverse selection and moral hazard
What exactly is insurance fraud? The Federal Bureau of Investigation (FBI) maintains a long list of various insurance fraud schemes, including premium division, asset churning etc., but they all boil down to policyholders having an “informational edge” on the insurers: they are much more aware of their own risk profile, the actual circumstances of accident occurrences, and their claiming behaviour. Besides outright criminal acts of fraud, such as making fictitious claims or claim build-up (majorization), there are other less pronounced acts arising from information asymmetry that might harm the equitability and effectiveness of the insurance market. Jointly, these phenomena are generally referred to as “adverse selection” and “moral hazard”.


Starting with the seminal work by Kenneth Arrow in the 1960’s, the problems of adverse selection and moral hazard have gained much attention in insurance economics. Here, adverse selection refers to the phenomenon that people with characteristics that are relatively unfavourable to the insurer have relatively high demand for insurance. Moral hazard refers to the phenomenon that actions of policyholders change once they are protected by insurance coverage and no longer bear the potential losses themselves.

In the academic world, researchers have been trying to identify and measure the potential problems of information asymmetry among insurers and policyholders for a long time. A common approach is to examine whether there is a positive correlation between choice of coverage and claim occurrence. Adverse selection suggests that policyholders often involved in accidents are likely to buy more insurance. In the case of moral hazard, those who have bought more insurance are assumed to act more carelessly, or even fraudulently, having more accidents and submitting larger claims. These phenomena should give rise to a statistically discernible claim-coverage correlation.

This simple test, however appealing, has major limitations, in part because of the amount of relevant information that is disregarded by this test. A more promising approach is to model the claim intensity structurally and to develop econometric tests and techniques that exploit not only the claim rates but also the times and extent of the claims.

No claims bonus distorts risk class picture
Let us consider car insurance for example. The bonus/malus (or no claims bonus) system has been designed to assign policyholders to the correct risk class. However, the differential premium structure sometimes makes policyholders refrain from claiming small losses, in order to stay in a more favourable bonus/malus class. As a result the genuine risk profile of policyholders is concealed: they may generate larger future losses than expected on the basis of their bonus/malus class. To detect such behaviour, we take advantage of the theoretical prediction that policyholders’ claiming incentives change immediately after a claim, if they have this so-called “hunger for bonus”. Consequently, claim rates and the extent of claims become dependent on past claims.

More generally speaking, by specifying a structural model that trades off the policyholders’ incentives and their impact  on the claiming process, formal econometric tests can be developed, based on the distribution of the claim rates, times, and extent of the claims.

‘Ex post moral hazard’ and insurance fraud
Whereas in the academic literature “hunger for bonus” is commonly referred to as a form of “ex post moral hazard”, this behaviour has little to do with (im)morality and insurance fraud: it is just an economic phenomenon driven by rational people balancing the costs and benefits of notifying an insurance claim.

More pressing concerns are about actual insurance fraud, ranging from simple claim buildup and exaggeration (majorization) to falsification and making fictitious claims.  But the incentives to commit actual insurance fraud also vary with the insurance policy and risk profile, the actual circumstances of accident occurrences, and the claims history. Their dynamic relationship as well as associated econometrics tests for fraud detection call for a structural econometrics analysis and further research, which we are currently conducting at the University of Amsterdam.

This is just our starting point and much more analysis is to follow. Our aim is that one day the ‘iceberg’ will be exposed in its entirety, and will eventually be largely eliminated. This is the stance of an econometrician in the fight against insurance fraud.

[3] Arrow, Kenneth J. (1963). Uncertainty and the welfare economics of medical care, American Economic Review 53, 941–969.
[4] Abbring, J. H., Chiappori, P.-A., and Zavadil, T. (2008). Better safe than sorry? Ex ante and ex post moral hazard in dynamic insurance data. Mimeo.



Share This