Do the Better Insured Cause More Damage? Testing for Asymmetric Information in Car Insurance

• DOI: http://doi.org/10.1111/jori.12040
• Date: 01 December 2015
• Published date: 01 December 2015

©2014 The Journal of Risk and Insurance. Vol.82, No. 4, 865–889 (2015).

DOI: 10.1111/jori.12040

Do the Better Insured Cause More Damage? Testing

for Asymmetric Information in Car Insurance

Tibor Zavadil

Abstract

This article tests for the presence of asymmetric information in Dutch car

insurance among senior drivers using several nonparametric tests based on

conditional-correlation approach.Since asymmetric information implies that

more comprehensive coverage is associated with higher risk, we examine

whether the better insured have a higher frequency of claims or cause more

severe accidents. Using data on claim occurrences, incurred losses and writ-

ten premiums, and controlling for the insureds’ experience rating, we do not

find any evidence of asymmetric information in this market.

Introduction

The analysis of asymmetric information in insurance markets has become a core topic

in the recent empirical literature on the economics of contracts.1After the seminal

work on moral hazard and adverse selection by Arrow (1963), Pauly (1968, 1974), and

Rothschild and Stiglitz (1976), who show that asymmetric information in competitive

insurance markets may lead to inefficient outcomes and market failure, economic

theorists devoted much effort to the development of adverse selection and moral

hazard models. In the last two decades of the twentieth century, contract theory

developed at a rapid pace, but empirical applications lagged behind. At the turn

of the millennium, this gap was filled with numerous empirical papers analyzing

Tibor Zavadil is at the ResearchDepartment, National Bank of Slovakia, Imricha Karva ˇ

sa 1, 813

25 Bratislava, Slovakia. Zavadil can be contacted via e-mail: tibor.zavadil@nbs.sk. I am very

grateful to Professor Jaap Abbring, my former PhD supervisor, for his helpful comments and

revision of the manuscript. This article was written as part of my PhD at the Department of

Economics, VU University Amsterdam, the Netherlands. The first version appeared in August

2008. I fully acknowledge the financial support by the Netherlands Organization for Scientific

Research (NWO) through a MaGW Free Competition grant (400-03-257). I also thank the par-

ticipants of the 65th European Meeting of the Econometric Society in Oslo, the 2nd Bratislava

Economic Meeting, and various research seminars for helpful discussion. Finally, I would also

like to thank my colleague Stuart Amor for his detailed proofreading. Any remaining errors

are mine.

1To mention just a few recent works: Israel (2004), Ceccarini and Pereira(2004), Cohen (2005),

Dionne et al. (2005), Chiappori et al. (2006), Saito (2006), Pinquet et al. (2008), Dionne et al.

(2013), and Spindler et al. (forthcoming) analyze asymmetric information in car insurance,

and Cardon and Hendel (2001), Hendel and Lizzeri (2003), Fang et al. (2008), Finkelstein and

McGarry (2006), and Cutler et al. (2008) study health and life insurance data.

865

866 The Journal of Risk and Insurance

asymmetric information in various insurance markets; see Chiappori and Salani´

e

(2003) or Cohen and Siegelman (2010) for an excellent overview.

The initial empirical studies of car insurance by Dahlby (1983, 1992) and Puelz and

Snow (1994) suggest the existence of adverse selection. These findings were later

challenged by subsequent research. Particularly, Chiappori and Salani´

e (2000) and

Dionne et al. (2001) question the results of Puelz and Snow’s analysis, claiming that

they used overly constrained functional forms relying on very few variables, and did

not control for the agent’s seniority and driving experience.

Chiappori and Salani´

e (2000) adopt an alternative approach based on a theoretical

conclusion that higher risks are associated with more comprehensive coverage. This

result comes from the works of Rothschild and Stiglitz (1976) and Wilson (1977) who

predict that, under adverse selection, high-risk individuals choose higher insurance

coverage and have more accidents (within risk classes). Under moral hazard, Shavell

(1979) and Holmstr¨

om (1979) predict that those with higher insurance coverage have

weaker incentives for safe driving and are therefore expected to have more accidents.

Consequently, asymmetric information leads to a positive correlation between cov-

erage and frequency of accidents (conditionally on all observables). This prediction

is quite general—it does not require any assumptions on preferences or on the firm’s

pricing policy.Given that it is valid under both adverse selection (bad risks buy more

insurance) and moral hazard (comprehensive coverage decreases incentives to drive

carefully), tests based on this prediction cannot distinguish between the two.2

Any test based on the conditional correlation between the choice of coverage and the

occurrence of claims must control for all variables observed by the insurer, because

these are used to price individual risk. Omitting any relevant characteristic observed

by both parties can lead to spurious informational asymmetry.Chiappori and Salani ´

e

(2000) point out that it is quite problematic to control for the past driving recordthat is

obviously endogenous. They circumvent this problemby focusing on a subpopulation

of young drivers who do not yet have a driving history.They do not find any evidence

of asymmetric information using French car insurance data. This article links directly

to their work and extends it in various ways.

First, we repeated all tests from Chiappori and Salani´

e (2000) on the subsample of

young drivers (below the age of 28) and also did not find any evidence of asymmetric

information.3Then we thought that the asymmetric information, if not present at the

beginning of the contractual relationship, might well arise in the course of time due to

asymmetric learning. This entails that drivers can learn faster about their risk than in-

surers because they accumulate information also on near misses and small accidents,

which they do not report to the insurer. Therefore, we decided to extend the analysis

on senior drivers, because this group has much higher heterogeneity in risk, risk aver-

sion, experience, and learning than the group of inexperienced drivers. Testingfor the

2See Cohen and Siegelman (2010, section V) for a discussion on how to disentangle moral

hazard from adverse selection.

3The results are available in Zavadil (2009).