Asymmetric Information in the Home Insurance Market

• Published date: 01 March 2017
• Date: 01 March 2017
• DOI: http://doi.org/10.1111/jori.12084

ASYMMETRIC INFORMATION IN THE HOME INSURANCE

MARKET

Karl Ove Aarbu

ABSTRACT

We test for the presence of asymmetric information in the home insurance

market on a data set containing about half a million home insurance

contracts, applying several different specifications of the conditional

correlation test. Unlike earlier studies, we control for private information

about risk aversion by having access to detailed administrative register

information at the policyholder level. We find robust evidence of asymmetric

information. Asymmetric information may stem from adverse selection or

moral hazard. To disentangle moral hazard and adverse selection, we utilize

an exogenous law reform that had an effect on the insurance price. Our test

shows no indication of moral hazard.

INTRODUCTION

This article investigates the presence of asymmetric information in the market for

home insurance using a unique data set containing almost 500,000 home insurance

contracts obtained from a Norwegian insurer. In addition to detailed data on each

insurance contract, we also link the insurance data to administrative register data at

the policyholder level. We use these data in combination with the insurance data to

test for both adverse and advantageous selection.

From standard textbook models, we know that in insurance markets characterized by

asymmetric information, a theoretically robust equilibrium result is that high-risk

agents acquire more insurance thanlow-risk agents (Rothschild and Stiglitz, 1976). In

Karl Ove Aarbu is at the Norwegian School of Economics. Aarbu can be contacted via e-mail:

karl.aarbu@nhh.no. I would like to thank two anonymous referees for insightful comments that

significantly improved the article. Moreover, I am grateful for extremely valuable comments

from Fred Schroyen, Frode Steen, Bruno Jullien, Kjetil Gramstad, Andreas Fagereng, Tobias

Berg, Gaute Torsvik, Bertil Tungodden, and Miles Kimball on earlier versions of this article. I

will also express gratitude to staff seminar participants at the University of Stavanger in 2011,

participants at the Norwegian Economic Researchers’ Annual Meeting 2011, and participants at

the 12th Symposium in Banking, Finance, and Insurance in Karlsruhe 2011. Moreover, I am

deeply grateful to the company that provided the insurance data and to Statistics Norway that

linked the insurance data to public administrative registers. Finally, I thank Mikis

Stasinopoulos for providing me with the gamlss.rsm package in R and Roger Bivand for

helping me to understand the basics of R.

© 2015 The Journal of Risk and Insurance. Vol. 84, No. 1, 35–72 (2017).

DOI: 10.1111/jori.12084

35

recent years, several studies have empirically investigated this prediction using data

from various insurance marketsincluding, for example, Chiappori and Salanie

´(2000),

Finkelstein and Poterba (2004), Cohen (2005), Chiappori, Jullien, Salanie

´, and Salanie

´

(2006), Finkelstein and McGarry (2006),

1

Bolhaar, Lindeboom, and van der Klaauw

(2012), Dumm, Eckles, and Halek (2013), Spindler, Winter, and Hagmayer (2013), Su

and Spindler (2013),Olivella and Vera-Herna

´ndez (2013), andZavadil (Forthcoming).

2

The empirical method for measuring the presence of asymmetric information—often

denoted as “the positive correlation test”—was originally developed by Chiappori

and Salanie

´(2000), later refined by Chiappori et al. (2006), and extended by

Finkelstein and McGarry (2006). The basic premise is that a positive correlation

between insurance cover and claims—conditional on the tariff variables used for

setting the insurance premium—indicates the presence of asymmetric information.

Thus, the empirical test is a mirror image of the central prediction of the Rothschild–

Stiglitz model. Unfortunately, implementing this particular test is not straightfor-

ward, as it requires specific and detailed information on each component of the

insurance contract (Finkelstein and McGarry, 2006).

3

Finkelstein and McGarry (2006) propose a comprehensive version of the conditional

correlation test. Their basic insight is that risk type is not always sufficient to explain

insurance purchase. In addition, private information about risk aversion may play a

role, especiallyif there is reason to believe that policyholders with high risk aversionare

more willing to purchase insurance. If these policyholders also are lower risk, the

correlation betweeninsurance cover and risk could then take any sign—positive,zero,

or negative—depending on whether risk preferences or risk type is the dominating

force.

4

For example, if risk aversion dominates, the correlation between claims and

insurance demand may be negative, thereby indicating advantageous selection.

5

1

As the positive correlation test really is a test of conditional independence between insurance

cover and subsequent claims, where the null hypothesis is zero correlation, the term

“conditional correlation test” will be used in the remainder of the article.

2

Cohen and Siegelman (2010) and Einav and Finkelstein (2011) provide overviews of this

literature up until 2010.

3

The positive correlation test requires robust and detailed data on each tariff parameter.

Moreover, to take account of heterogeneous preferences one should also have access to some

kind of survey or administrative data. Often such data are unavailable. In an interesting article,

Gan, Huang, and Mayer (2011) propose a variant of the positive correlation test that can be

employed within insurance markets where this type of information is not fully available.

4

Donder and Hindriks (2009) present a theoretical model that does not entail negative

correlation in equilibrium, even though more risk-averse agents acquire more insurance and,

at the same time, are more cautious.

5

Hemenway (1990) introduces the concept of propitious selection to explain the positive

correlation between caution and risk aversion. Hemenway (1992) presents survey evidence

confirming propitious/advantageous selection among vehicle drivers in the United States.

Elsewhere, de Meza and Webb (2001) specifically discuss how advantageous selection may

work in insurance markets. In a study that focuses on the sources of advantageous selection,

Fang, Keane, and Silverman (2008) show that cognitive ability can explain the existence of

advantageous selection in the U.S. MediGap market.

36 THE JOURNAL OF RISK AND INSURANCE

Another implication is that if risk preferences play a role but empirical information

about this dimension is unavailable, there will be bias in the standard conditional

correlation test. Therefore, in empirical investigations of asymmetric information, we

should ideally have access to full information about each contract, along with

information on risk preferences.

6

Unlike most existing studies, we have access to extremely detailed socioeconomic

information at an individual level assembled from administrative public registers

that allows us to include sources of private information in the conditional correlation

test. It is important to emphasize that this administrative information has not been

used in pricing of the insurance policies investigated in this article. The reason is

simply that this type of administrative information is restricted to researchers and

research institutions. As this information is unavailable for insurers, we can use

these administrative data in combination with insurance data and directly test

for effects of private information about the degree of risk aversion on contract choice

and risk (cf. Finkelstein and Poterba, Forthcoming).

7

Based on the earlier literature, we construct three risk aversion proxies that are

included as covariates in the conditional correlation test. We use information from

individual tax returns to calculate the share of income from self-employment,

detailed information on working sector to identify public sector employees, and

information from individual income and wealth registers to calculate the share of

6

The empirical literature has indeed produced a diversity of results concerning the existence

of selection problems. One of the first is Puelz and Snow (1994), who find evidence on

asymmetric information in the U.S. car insurance market. However, their result may arise

from misspecification of the empirical model (see Dionne, Gourie

´roux, and Vanasse, 2001;

Chiappori and Salanie

´, 2000; and Chiappori, 2000). Cawley and Philipson (1999),

employing data from a U.S. life insurer, find no evidence of asymmetric information

within the life insurance market. Chiappori and Salanie

´(2000) cannot provide any evidence

of asymmetric infor mation using French car insurance data , while Finkelstein and Poterba

(2004) find evidence of asymmetric information in data gathered from a large British

annuity provider. Cohen (2005) finds evidence of asymmetric information in a data set she

collected from an Israeli car insurer, as do Chiappori et al. (2006). Bolhaar, Lindeboom, and

van der Klaauw (2012) and Olivella and Vera-Herna

´ndez (2013) both document

asymmetric information in the Irish and British health insurance mar kets, respectively.

Zavadil (Forthco ming), however, doe s not find any evidence of a symmetric inform ation

among senior drivers in a data set containing insurance policies from a Dutch car insurer.

Finally, Finkelstein and Poterba (Forthcoming) find evidence that an insurer may choose

not to utilize relevant information in setting their tariffs, which ultimately may lead to

selection problem s.

7

We obtained the administrative data from Statistics Norway. See “Data Collections for

Research Purposes” (http://www.ssb.no/english/mikrodata_en/). For this specific project,

we used Income Register data, data from the National Education Database, and data from the

register-based employment file. The process required approval by Statistics Norway, approval

by the Data Protection Official for Research at Norwegian Social Science Data Services, and

finally from the Norwegian Data Protection Authority. These databases cover the full

population and do not contain self-reported survey data.

ASYMMETRIC INFORMATION IN THE HOME INSURANCE MARKET 37