Asymmetric Information in the Market for Automobile Insurance: Evidence From Germany

• Date: 01 December 2014
• DOI: http://doi.org/10.1111/j.1539-6975.2013.12006.x
• Author: Joachim Winter,Martin Spindler,Steffen Hagmayer
• Published date: 01 December 2014

DOI: 10.1111/j.1539-6975.2013.12006.x

Asymmetric Information in the Market for

Automobile Insurance: Evidence From Germany

Martin Spindler

Joachim Winter

Steffen Hagmayer

Abstract

Asymmetric information is an important phenomenon in insurance markets,

but the empirical evidence on the extent of adverse selection and moral haz-

ard is mixed. Because of its implications for pricing, contract design, and

regulation, it is crucial to test for asymmetric information in specific insur-

ance markets. In this article, we analyze a recent data set on automobile

insurance in Germany, the largest such market in Europe. We present and

compare a variety of statistical testing procedures. We find that the extent

of asymmetric information depends on coverage levels and on the specific

risks covered, which enhances the previous literature. Withinthe framework

of Chiappori et al. (2006), we also test whether drivers have realistic expecta-

tions concerning their loss distribution, and we analyze the market structure.

Introduction

Since Akerlof (1970), the consequences of asymmetric information, in particular, ad-

verse selection and moral hazard, have been explored in a vast body of research. The

initial gap between the theoretical developments and empirical studies of asymmetric

information has recently become narrower. In particular, insurance markets have

proved a fruitful and productive field for empirical studies, for two reasons. First, the

Martin Spindler is at the Max Planck Institute for Social Law and Social Policy, Munich, Ger-

many. Joachim Winter is at the University of Munich, Economics Department, Munich, Ger-

many. Steffen Hagmayer, is at the HUK Coburg Insurance Group, Coburg, Germany. The

authors can be contacted via e-mail: spindler@mea.mpisoc.mpg.de, joachim.winter@lrz.uni-

muenchen.de, and steffen.hagmayer@huk-coburg.de.Martin Spindler gratefully acknowledges

financial support from the Deutsche Forschungsgemeinschaft through GRK 801 and in partic-

ular from The Geneva Association through a research grant. We thank Bernard Salani´

e and

Andreas Richter for valuable comments and discussions, and Thomas Yeefor assistance in ap-

plying the R package VGAM. Weare grateful to participants of the EEA/ESEM annual meeting

2011in Oslo, of the annual conference of the German Economic Association (VfS) 2011 in Frank-

furt am Main and of the annual seminar of the EGRIE 2011 in Vienna.Parts of this paper were

written while Martin Spindler was visiting Columbia University, which he thanks for its hos-

pitality.We thank Editors Georges Dionne and Keith Crocker and an anonymous reviewer for

valuable comments that helped to improve the paper.

781

©The Journal of Risk and Insurance, 2013, Vol.81, No. 4, 781–801

782 The Journal of Risk and Insurance

data are well structured: insurance contracts areusually highly standardized, they can

be described completely by a relatively small set of variables, and data on the insured

person’s claim history, that is, the occurrence of claims and the associated costs, are

stored in the database of an insurance company. Second, insurance companies have

hundreds of thousands or even millions of clients and therefore the samples are

sufficiently large to conduct powerful statistical tests. The markets for automobile

insurance, annuities and life insurance, crop insurance, as well as long-term care

and health insurance provide large samples of standardized contracts for which

performances are recorded and are well suited for testing the theoretical predictions

of insurance theory. Chiappori and Salani´

e (1997) provide a detailed justification for

using insurance data to test contract theory. Cohen and Siegelman (2010) present a

comprehensive overview of approaches for testing for adverse selection in insurance

markets, covering a large number of empirical studies in differentinsurance branches.

In statistical terms, the notion of asymmetric information implies a positive (condi-

tional) correlation between coverage and risk. Several different methods that explain

how to test for asymmetric information have been proposed in the literature. In this

article, we apply an array of such tests to detailed contract-level data from the German

car insurance market.

Our study contributes to the existing literature in several respects. First, we present

the first study analyzing the German car insurance market. The German car insurance

market is the largest in Europe and therefore for many insurance companies the most

important sales market for their insurance policies. We had unique access to the data

set of one of the largest insurance companies in the field of automobile insurance in

Germany.

Second, the empirical literature has reached an almost complete consensus that asym-

metric information is not prevalent in automobile insurance concerning coverage

choices (e.g., deductible level).1Our analysis shows that this finding does not hold

in general; in particular, we show that the institutional arrangements of a market

and the structure of the contracts have a great influence on whether the insureds

have an informational advantage that the can possibly exploit. Because of a special

arrangement that holds in the car insurance in Germany,we can show that the extent

of asymmetric information depends on the specific kinds of risks that are covered.

Third, we compare several tests for asymmetric information that have been proposed

in the literature. Chiappori and Salani´

e (2000) propose tests for the positive correlation

property, Dionne, Gourieroux, and Vanasse(2001) use a two-stage approach, and Kim

et al. (2009) modify a multinomial approach. Most studies only apply a selection of

these tests. We apply all tests on the same data set. We find that they give consistent

results, not in a statistical but in a qualitative sense: they all give the same answer to

1Strictly speaking, this refers to a zero correlation between risk and coverage. While most

theoretical models focus only one source of adverse selection, namely, different knowledge of

the risk, there are possibly more sourcesof informational asymmetries, for example, related to

risk preferences and risk aversion, which introduceheterogeneity may complicate the analysis

and interpretation of results.