A Test of Asymmetric Learning in Competitive Insurance With Partial Information Sharing

• Published date: 01 September 2016
• Author: Peng Shi,Wei Zhang
• Date: 01 September 2016
• DOI: http://doi.org/10.1111/jori.12056

©2014 The Journal of Risk and Insurance. Vol.83, No. 3, 557–578 (2016).

DOI: 10.1111/jori.12056

A Test of Asymmetric Learning in Competitive

Insurance With Partial Information Sharing

Peng Shi

Wei Zhang

Abstract

This article examines whether an insurer could gain advantageous informa-

tion on repeatcustomers over its rivals in the Singapore automobile insurance

market, which is featured by partial information sharing among insurers. We

find that the insurer does update and accumulate more information regard-

ing its policyholders’ riskiness through repeatedobservations and thus make

higher profits with repeat customers especially those of lower risk. We also

show that the higher profit is driven by the fact that low risks tend to stay

longer with the insurer, and in the meanwhile, they are chargeda premium

higher than their actuarial risk level.

Introduction

In economics and contract theory,information asymmetry refers to the situation where

one party possesses information that is not available to another party in a transaction.

Asymmetric learning, a particular form of information asymmetry, arises when a

“seller” could gain advantageous information on its repeat “buyers” over its rivals

through repeated contracting. Asymmetric learning could resultin market power, and

thus lead to higher profits from repeat customers (see Kunreuther and Pauly, 1985).

In this study, we aim to examine asymmetric learning in repeated contracting when

the market lacks a complete information-pooling mechanism among various par-

ticipants. Since the seminal work of Akerlof (1970), Spence (1973), and Rothschild

and Stiglitz (1976), the pervasive effects of information asymmetry in markets have

attracted much attention and have been extensively studied in numerous contexts.

Historically, the insurance industry provides an interesting environment to investi-

gate information asymmetry theoretically and empirically (see Dionne, Doherty, and

Fombaron, 2000, for a comprehensive review). This empirical study differs frommost

of the existing literature in its concerns with the dynamics of repeated contracting. We

Peng Shi is at the Department of Actuarial Science, Risk Management, and Insurance, School

of Business, University of Wisconsin–Madison, Madison, WI 53706. Shi can be contacted via

e-mail: pshi@bus.wisc.edu. Wei Zhang is at the Department of Economics, Northern Illinois

University,DeKalb, IL 60115. Zhang can be contacted via e-mail wzhang1@niu.edu. The authors

wish to express their gratitude to Editor Keith Crocker and an anonymous referee for their

valuable comments that helped improve the article.

557

558 The Journal of Risk and Insurance

use a unique panel data set to test whether repeat customers provide an insurer with

information-based advantages over its rival companies in the Singapore automobile

insurance market.

The concept of asymmetric learning evolves from a dynamic context where private

information develops through time. Since the initial efforts of Rothschild and Stiglitz

(1976) and Wilson (1977) to remedy adverse selection in a competitive insurance mar-

ket, insurance models with information asymmetry have been well developed for

multiperiod contracting. In particular,different predictions on the evolution of insurer

profits over time are obtained under various commitment assumptions between the

contractual parties. Early works focus on the multiperiod models with full commit-

ment on the supply side. One example is Cooper and Hayes (1987), where the contracts

are offered in a way that high-risk agents select short-term contracts and low-risk

agents receive long-term contracts. Thus, with experience rating, this arrangement

results in a highballing profit pattern; that is, the long-term contracts are initially prof-

itable to the insurer but become unprofitable in later periods. Recent studies look more

into the limited commitment relation between an insurerand its insurees. The repeated

insurance model in a competitive market without commitment was first proposed by

Kunreuther and Pauly (1985), in which, through offering pure price contracts, an in-

surer will take a lowballing pricing strategy in that new business is unprofitable but

economic rent will be extracted from repeat business. The same lowballing prediction

could also be derived using the traditional screening device of price-quantity con-

tracts. This is shown and extended in later studies such as Prendergast (1992), Nilssen

(2000), and de Garidel-Thoron (2005). In an intermediate case between full commit-

ment and no commitment, known as commitment with renegotiation, insurers have

the option to recontract with their insurees though the proposition could be rejected

by the insurees. With renegotiation, the observed pricing strategy of insurers could

be one of highballing (Dionne and Doherty,1994) or lowballing when outside options

are endogenous (Dionne, Doherty, and Fombaron, 2000).

Asymmetric learning is consistent with the lowballing prediction in that the con-

tracting insurer possesses an informational advantage over competitors in repeatedly

monitoring its policyholders, and thus has the potential to gain ex post market power

and make higher profits from repeated contracting. The above competing models pro-

vide testable predictions on the temporal pattern of profits and consumer lock-in that

allow us to empirically examine the significance of information asymmetry. How-

ever, despite the large literature on empirical testing of adverse selection (see Cohen

and Siegelman, 2010; Spindler, Winter, and Hagmayer, 2013; Zavadil, 2014, among

others for recent contribution), empirical work on asymmetric information in a dy-

namic context is not emerging in parallel with the theoretical development in mul-

tiperiod models. Some recent examples include Dionne et al. (2011), Cohen (2012),

and Dionne, Michaud, and Dahchour (2013). Empirical testing of asymmetric learn-

ing in repeated contracting is even sparse. Only a few articles investigate the time

trend of profits in insurance markets and not all findings support the hypothesis of

asymmetric learning. Among them, D’Arcy and Doherty (1990) perform a test using

the U.S. automobile insurance data, and consistent with asymmetric learning, they

show a negative relationship between loss ratio and policy age for various cohorts of