Does asymmetric information matter in the early insurance market? Evidence from the auto insurance market

Abstract

In the late 1960s, the performance of automobile insurance declined dramatically in Japan in spite of rapid growth in the diffusion rate, and the premiums were sharply raised several times in order to improve the situation. This observation indicates the possible presence of adverse selection (death spiral), and provides an ideal situation for assessing informational asymmetry. Using bodily injury liability (BIL) insurance data from 46 Japanese prefectures over the period 1966 to 1975, this article tests two hypotheses of adverse selection: (i) high-risk drivers were more likely to join the BIL insurance market and (ii) sharp premium increases drove low-risk policyholders away. Various empirical analyses show that there is little evidence for either type of adverse selection. We also test whether a risk-misperception hypothesis can explain our results, and find some evidence that the population density have a significantly positive impact on the demand for BIL insurance.

† An earlier version of this article was circulated under the title, ‘Does Asymmetric Information Matter in the Early Insurance Market? Evidence from the Japanese bodily injury liability insurance market, 1966–1975

Acknowledgements

For helpful comments, I would like to thank Fumio Akiyoshi, Yoshiro Miwa, Akio Torii, Daisuke Tsuruta and the seminar participants at the University of Tokyo as well as the conference participants at the 2003 Spring Meetings of the Japanese Economic Association and I am grateful to Reiko Yabe at the Library of Non-Life Insurance Institute of Japan for collecting the data set. Any errors are my own responsibility.

Notes

† An earlier version of this article was circulated under the title, ‘Does Asymmetric Information Matter in the Early Insurance Market? Evidence from the Japanese bodily injury liability insurance market, 1966–1975

¹ Also, Saito (2006) provides negative evidence of asymmetric information using an individual data set from the Japanese automobile insurance market where premiums are strictly regulated. Important exceptions include Cohen (2005) and Israel (2004). Cohen (2005) uses a 5 year panel data from an Israelite automobile insurance market and finds strong evidence of adverse selection for the group of drivers with comparatively long driving experience. Also, Israel (2004) finds small but statistically significant evidence of moral hazard in the US insurance market. However, negative evidence of asymmetric information seems to be common in the automobile insurance market. See Chiappori and Salanié (2003) and Siegelman (2004) for broader surveys.

² See Bond and Crocker (1991) and Crocker and Snow (2000) for theoretical discussions about categorical discrimination in the insurance industry, and Rubinstein and Yaari (1983) and Cooper and Hayes (1987) about experience ratings.

³ ‘Adverse selection death spiral’ is often used to point out the situation in which coverage is reduced by a community rating, that imposes the same or similar rate to all individuals or small groups regardless of various attributes of the policyholders. While this terminology is used mainly in the health insurance market, the same restrictions on premium setting are also observed in many other insurance markets. See Buchmueller and DiNardo (2002) for the relationship between community rating and the ‘adverse selection death spiral.’

⁴ Our methodology cannot distinguish adverse selection from moral hazard, although disentangling adverse selection from moral hazard is one of the most important empirical issues in this field.

⁵ See AIRO (2002) and Dionne (2002) for more information about the institutional aspect of the Japanese automobile insurance market.

⁶ In 1970, premiums were divided into three categories: covered over 26, covered over 21, and all ages covered.

⁷ Premium rates were partially deregulated in 1997.

⁸ Figures are calculated as ([total amount of loss paid for compulsory automobile liability insurance] + [total amount of loss paid for voluntary bodily injury liability insurance])/(number of claims paid for compulsory automobile liability insurance). As a reference, the general price index, excluding imputed rent, changed from 100 in 1960, to 133 in 1965 and to 174 in 1970.

⁹ Focusing on BIL insurance has an additional benefit as follows. In most cases, traffic accidents involving bodily injury are reported to insurance companies or police because they usually involve more than two parties. But in the case of other types of auto insurance, an accident does not necessarily become a claim because of the Bonus-Malus system. In the case of collision insurance, for example, whether an accident becomes a claim or not largely depends on the policyholder's strategic behavior, and he/she might not file a claim if the cost of premium increases in following years exceeds the benefit of the claim.

¹⁰ Chiappori et al. (2006) calls this hypothesis ‘the positive correlation property’ and discusses under what assumptions it can be used as a test for asymmetric information.

¹¹ This situation is sometimes called an adverse selection ‘death spiral’ and is used mostly in the health insurance market. While this aspect of adverse selection has not been fully considered, one of a few exceptions includes Buchmueller and DiNardo (2002), which found no evidence that the community rating induces an adverse selection death spiral.

¹² We do not use the data before 1965 because it was not available. Also, we do not use data after 1975 because the introduction of the package contract in 1976 makes it difficult to focus exclusively on the BIL insurance contract. Okinawa is excluded because the retrocession was in 1972.

¹³ Fixed-effects model is recommended by Hausman specification test.

¹⁴ Dividing the sample into two periods also is preferable because the number of accidents began to decrease in 1970, and the performance of insurance companies improved after 1970.

¹⁵ See Cameron and Trivedi (1991) and Propper (1989) in the health insurance market Browne and Hoyt (2000) in the flood insurance market and Smith and Wright (1992) in the auto insurance market for example. “Kuniyoshi Saito”

¹⁶ Lichtenstein et al. (1978) is one of the earliest literature discussing risk perception biases. See Viscusi et al. (2005) for the relationship between perceived risks and actual risks of various kinds of events.

¹⁷ However, Cohen and Einav (2007) show that, in the automobile insurance market, risk averse drivers are likely to have more accidents.