Heterogeneity of the Accident Externality from Driving

• Author: Larry Y. Tzeng,Rachel J. Huang,Kili C. Wang
• Published date: 01 December 2014
• Date: 01 December 2014
• DOI: http://doi.org/10.1111/j.1539-6975.2013.01517.x

©

DOI: 10.1111/j.1539-6975.2013.01517.x

735

HETEROGENEITY OF THE ACCIDENT EXTERNALITY FROM

DRIVING

Rachel J. Huang

Larry Y. Tzeng

Kili C. Wang

ABSTRACT

This article examines the accident externality from driving in terms of loss

probability and severity by using a unique individual-level data set with

more than 3 million observations from Taiwan.Two types of accident exter-

nality are, respectively, measured: the average number of kilometers driven

per month per vehicle and the total number of speeding tickets per month.

For both variables, we find significant evidence to support the existence of

the accident externality. Moreover, we find that the accident externality is

heterogeneous in terms of the vehicles’ characteristics.

INTRODUCTION

The risk of a specific driver is affected by other drivers’ driving behavior. This is

referred to as the accident externality from driving. Such an externality could be

very costly to a society and has received much attention in the literature. For exam-

ple, by using aggregate panel data for the United States, Edlin and Karaca-Mandic

(2006) provide intriguing evidence to support the existence of an accident external-

ity from driving. They find that to correct the substantial accident externalities, a

Pigouvian tax could raise over $220 billion per year nationally.By adopting Edlin and

Karaca-Mandic’s methodology,Saito, Kato, and Shimane (2007) also find evidence of

a positive and significant externality in Japan. The estimated nationwide Pigouvian

tax is about $16–$51 billion in Japan.

In complementing the above literature that estimates the total size of the accident

externality, this article studies two important questions that have so far not been

explored to any significant extent in the literature.First, how does the externality affect

Rachel J. Huang is an Associate Professor, Graduate Institute of Finance, National Taiwan

University of Science and Technology, Taiwan; Research Fellow, Risk and Insurance Research

Center, College of Commerce, National Chengchi University. Larry Y. Tzeng is a Professor,

Department of Finance, National Taiwan University, Taiwan;Research Fellow, Risk and Insur-

ance Research Center, College of Commerce,National Chengchi University. Kili C. Wang is an

Associate Professor, Department of Insurance, Tamkang University, Taiwan; Research Fellow,

Risk and Insurance Research Center,College of Commerce, National Chengchi University. Kili

Wang can be contacted via e-mail: kili@mail.tku.edu.tw.

The Journal of Risk and Insurance, 2013, Vol. 81, No. 4, 735–756

736 THE JOURNAL OF RISK AND INSURANCE

the individual’s loss probability and loss severity? Second, who suffers more due to

other people’s driving? In other words, this article seeks to examine the heterogeneity

of the accident externality. The answers to these two questions can guide a social-

welfare maximizing government in delicately coping with the externality.It is because

information on the accident externality both in terms of frequency and severity is

necessary for the government to push the private optimum to the social optimum

when drivers are risk averse.1In addition, the government could directly compensate

the identified victims to improve the social welfare.

Since the heterogeneity of the accident externality from driving cannot be analyzed

through aggregate data, we adopt individual-level data. Wehand-collect our data by

integrating data from a vehicle manufacturer with data from an insurance company

in Taiwan. Our insurance data include both the occurrenceand the amount of money

involved in the accident.2We are therefore in a position to investigate the impact of

the accident externality on the frequency and severity separately.Our insurance data

also contain the individuals’ demographic variables that can be used to analyze the

heterogeneity of the accident externality.

Weuse two variables to measure the accident externality.3One is the average number

of kilometers driven per vehicle since the more kilometers that other drivers cover,

the greater the potential for each driver to risk causing an accident. Our data from

the vehicle manufacturer contain the kilometers driven for each vehicle. Thus, we

could estimate the accident externality conditional on the individual’s own driving.

Furthermore, even if the average number of kilometers driven per vehicle is high, it

might not necessarily mean that the risk is higher if others drive at a reasonable speed.4

Since speed is one of the major risk factors associated with driving, we further adopt

the total speeding tickets per month in Taiwan as another variable for the accident

externality.

The major findings are as follows. First, we confirm the existence of the accident

externality arising from driving in Taiwan both on the average number of kilometers

driven per month per vehicle and on the total number of speeding tickets per month.

Moreover, we find that the accident externality exists in terms of both the accident

probability and accident severity. With respect to loss probability, an individual will

increase her accident probability by 0.2937 percent per month when the average

number of kilometers driven per vehicle per month increases by 100 km. Furthermore,

the probability will increase by 0.0081 percent when the total number of speeding

tickets increases by 1,000 per month. With respect to loss severity, the loss severity

will increase by $8.2 per vehicle per month when the average number of kilometers

driven per vehicle per month increases by 100 km. We also find that it will increase

by $1.8 per vehicle per month when the total number of speeding tickets increases by

1Please see the Appendix for the details.

2We use insurance claim data as the proxy for accidents.

3Edlin and Karaca-Mandic (2006) propose the use of the square of traffic density as the proxy

for “the likelihood that two other vehicles are in the same location at the same time.” Other

papers in the literature (e.g., Belmont, 1953; Lundy, 1965; Turner and Thomas 1986) consider

the impact of traffic volume on the accident rate.

4We would like to thank an anonymous refereefor pointing this out.