6/09/2011

The Demand for Non-Life Insurance in Taiwan

The Demand for Non-Life Insurance in Taiwan

Min-Sun Horng

Department of Risk Management and Insurance,

National Kaohsiung First University of Science and Technology, Taiwan.

Yung-Wang Chang

National Kaohsiung First University of Science and Technology

and Meiho Institute of Technology,

No. 2, Jhuoyue Rd., Nanzih District, Kaohsiung City 811, Taiwan,

E-mail: u9127907@ccms.nkfust.edu.tw,

Tel: +886-7-351-7964.

Abstract

Several theories have been developed to explain the motives for purchasing non-life insurance (i.e., property-casualty insurance), but there have been few empirical tests conducted by emerging insurance markets. This study examines the determinants of non-life insurance consumption in Taiwan between 1970 and 2005, focusing on fire insurance and automobile insurance. Using econometric regression models, the analysis indicates that economic conditions affect the demand for insurance differently across lines of coverage. The results show that income has a far greater effect on automobile insurance demand, than on fire insurance demand. Moreover, the results reveal that the purchase of non-life insurance is significantly and positively related to income and risk aversion, in addition to providing weaker evidence of a negative relationship with price. Managerial implications are identified based on the empirical findings.

Keywords: Insurance demand; Non-life Insurance; Insurance Penetration; Taiwan

JEL classification codes: G22; O16

1. Introduction

Despite the financial crisis in Asia, the Taiwan insurance market continues to grow. Taiwan’s average annual economic growth rate over the past two decades was relatively high (6.06 percent).[1] Not only has the high-tech electronics industry made remarkable progress, but so has the financial industry. Interestingly, Sigma (No. 5/2006) reported that insurance spending per capita in Taiwan is currently higher than the average in Europe. [2] Table 1 reveals that Taiwan had the highest insurance premium penetration (i.e., premiums as a share of GDP) with 14.11 percent in 2005, followed closely by South Africa (13.87 percent) and the United Kingdom (12.45 percent). Since the premium penetration in Taiwan (14.11 percent) was almost double the world average (7.52 percent) in 2005, it is worth investigating the development factors of insurance markets in Taiwan.

Another insurance indicator, insurance density, is calculated by dividing direct gross premiums by the population. This represents the average insurance spending per capita in a given country. Table 1 also shows that, in 2005, Switzerland ranked first in insurance density ($5,558), with the United Kingdom ranking second ($4,599). During the same year, the European insurance density was $1,514 and the world insurance density was only $519. However, Taiwan’s insurance density was $2,146, ranking twentieth in the world.

Table 1

2005 Ranks in Insurance Penetration and Insurance Density

Insurance Penetration (%)

Insurance Density (USD)

Country

Rank

Total

Life

Non-life

Rank

Total

Life

Non-life

Taiwan

1

14.11

11.17

2.93

20

2,146

1,699

446

South Africa

2

13.87

10.84

3.03

32

715

558

156

United Kingdom

3

12.45

8.90

3.55

2

4,599

3,287

1,312

Switzerland

4

11.19

6.20

4.99

1

5,558

3,078

2,480

Belgium

5

11.15

8.36

2.79

4

3,986

2,989

997

Japan

6

10.54

8.32

2.22

7

3,747

2,956

790

South Korea

7

10.25

7.27

2.98

22

1,706

1,211

496

France

8

10.21

7.08

2.77

9

3,569

2,475

1,094

Hong Kong

9

9.52

6.38

3.13

15

2,545

2,213

332

Netherlands

10

9.79

5.12

4.67

8

3,740

1,954

1,786

Europe

7.78

4.69

3.10

1,514

912

602

World

7.52

4.34

3.18

519

300

219

Source: Swiss Reinsurance Company, Sigma Publications, No. 5/2006

Note: 1. Insurance Density: premiums per capita.

2. Insurance Penetration: premiums as share of GDP (%).

In Table 2, it can easily be seen that in recent years, Taiwan’s insurance industry has grown. It grew at an average annual growth rate of 21 percent between 1996 and 2006 (Swiss Reinsurance Company, Sigma No. 4/1998 and 4/2007). Data also shows that during the same period, the non-life insurance business grew at a rate of 10 percent annually, while the life insurance business grew at a rate of 25 percent annually. In 2006, Taiwan’s insurance premium volume was approximately $52 billion and accounted for about 14.5 percent of gross domestic product (GDP), i.e., insurance penetration, compared to 5.8 percent in 1996. Taiwan’s insurance density in 1996 was only $740. However, in 2006, it rose to $2,250. These statistics indicate that the insurance business has become increasingly important to Taiwan’s economy.

Table 2: 1996~2006 Total Gross Premium, Premium Density, and Insurance Penetration in Taiwan

Year

1996

2006

Gross Premium (USD million)

Non-life Insurance

4,833

10,318

Life Insurance

10,994

41,245

Total

15,827

51,563

Premium density (USD)

740

2,250

Insurance penetration (%)

5.8

14.5

Source: Swiss Reinsurance Company, Sigma Publications, No. 4/1998 and No. 4/2007.

Note: 1. Insurance Density: premiums per capita.

2. Insurance Penetration: premiums as share of GDP (%).

Although the insurance industry has grown significantly in Taiwan, insurance researchers have not paid much attention to the empirical assessment of contributions regarding the insurance sector of the Taiwan economy. Furthermore, previous studies have focused on cross-country studies or well-established markets in developed countries (Ma and Pope, 2003; Esho et al. 2004), while research into the Taiwan market is seemingly scarce. Therefore, this study takes the lead in conducting research on different lines of non-life insurance in Taiwan. According to the premium volume in 2005, automobile insurance and fire insurance are in first and second place in non-life insurance, with densities of 49.7% and 18.6%, respectively. In other words, this study focuses on two lines of insurance: fire and motor vehicle.

The remainder of the study is divided into four sections. The following section presents the factors affecting non-life insurance. Next, this data utilized and the empirical framework of the study is described. Finally, the conclusions and implications of the study are discussed.

2. Determinants of non-life insurance demand

From an economic viewpoint, the demands for insurance are based on the expected utility paradigm (e.g. Mossin, 1968Szpiro, 1985; Browne et al. 2000), and suggests that different factors influence insurance purchases of an individual. These factors include the individual‘s income and wealth, the price of insurance, the probability of loss, and the individual‘s degree of risk aversion. MacMinn and Han (1990) and Garven and MacMinn (1993) suggest that corporate insurance purchases could decrease the possibility of financial crisis. Basically, corporate demand for insurance is driven by the maximization of the current shareholder value (e.g. MacMinn, 1987; Regan and Hur, 2007). However, in a corporate insurance market with transaction costs, the same factors as individual purchase of insurance—income, the price of insurance, the probability of loss, risk aversion—are hypothesized to be important determinants of insurance demand by businesses.

2.1 Income

Income level is hypothesized to positively influence insurance demand. Beenstock et al. (1988) point out a positive relationship in industrialized countries between national income and non-life insurance spending. Browne et al. (2000) analyzes general liability and motor vehicle insurance in OECD countries, and finds a significant positive relationship between premium density and GNP per capita. Additionally, Esho et al. (2004) examines developed and developing countries between 1984 and 1998, and finds a strong positive relationship between national income and the non-life insurance premium. Outreville (1990) and Ward and Zurbruegg (2002) strongly emphasize that the insurance industry, through risk transfer, financial intermediation, and employment can generate externalities and economic growth.

2.2 Price of Insurance

The demand for any product or service is affected by its price. In this study, the inverse of the loss ratio, i.e. defined as premiums divided by claims, is taken to measure the price of insurance and is expected to be negatively correlated with the insurance demand. Cummin and Danzon (1997) and Esho et al. (2004) use a similar measure in their study of price determination. Browne et al. (2000), based on the economic theory, considered that if the market of a country excludes the competition of foreign business, it will lead to merchandise of low quality and high prices. And the ratio of the premium volume of foreign companies in the market will be taken as the measure of price, i.e. the negative correlation between the shares of the market of foreign insurance companies and the price of insurance. In fact, considering the price with the probability of loss (i.e., claims ¸ premiums) and profit (i.e., premiums – claims) would be more tightly related.

2.3 Risk Aversion

The primary motive for purchasing insurance is risk aversion to avoid loss. Therefore, the level of risk aversion is hypothesized to be positively correlated with insurance demand. Schlesinger (1981) demonstrates that an individual with a higher loss probability, a higher degree of risk aversion, or a lower level of initial wealth, will purchase more insurance. Mayers and Smith (1990) believe that closely held firms are more likely to purchase insurance than firms with less-concentrated ownership for the same reason that an individual purchases insurance—risk aversion. Mayers and Smith (1990) further indicate a supposition that a company does not exhibit proper risk aversion, because risk aversion is not so obvious to the corporate purchasers of insurance. As stated previously, even though risk aversion could not perfectly explain why consumers would buy insurance, it is still an important indicator.

Although risk aversion is a “rational” motive for an individual’s purchase of insurance, unfortunately, it is difficult to measure. According to the discussion of Browne and Kim (1993), in general, a higher level of education may lead to a greater degree of risk aversion and greater awareness of the necessity of insurance. Nonetheless, Szpiro and Outreville (1988) proved the negative correlation between the level of education and risk aversion. They deemed that higher education leads to lower risk aversion, and that, in turn, leads to more risk-taking by skilled and well-educated people. When Browne et al. (2000) and Esho et al. (2004) were discussing non-life insurance; they also took the level of education as a proxy for risk aversion.

2.4 Others

The factors that influence the demand for insurance also can be compartmentalized by the different law systems in various countries and areas. For instance, Esho et al. (2004) pointed out that, other thing equal, automobile insurance consumption is greater in common-law countries than in statutory-law countries. Other factors, like the degree of economic development and market structure, would also influence the demand of non-life insurance. However, the purchase of insurance could help with tax saving and clients’ demands for insurance (Ma and Pope, 2003). These factors would influence the demand of non-life insurance, more or less, but would not do much harm.

3. Empirical Analysis

3.1 Data

This study follows the recommendations of previously studies and uses real GDP per capita to measure the value of income (Ward and Zurbruegg, 2002). Annual GDP data was obtained from the AREMOS database of the Taiwan Ministry of Education for the period of 1970 through 2005. Insurance density (i.e., premiums per capita) was utilized as a proxy of insurance demand (Browne et al., 2000; Li et al., 2007). Annual premium data was acquired from the Taiwan Statistical Yearbook for the period of 1970 through 2005. The level of education refers to the Statistics Yearbook published by the Ministry of the Interior (Taiwan). The content includes information from all people over the age of 15. Nominal GDP per capita and nominal insurance density data in Taiwanese currency were converted into real GDP per capita and real insurance density in 2001 prices, using the consumer price index (2001=100). Real GDP per capita, real insurance density (fire and automobile insurance), price of insurance, and risk aversion data were also transformed by the use of natural logarithms (LINCOME, LFIRE, LCAR, LPRICE and LEDU) to ease interpretation of coefficients. Coefficients in the log function indicate a percentage change in a dependent variable given as a percentage change in an independent variable.

3.2 Unit root tests

When the variables are non-stationary or exhibit a unit root, the procedures of conventional regressions may not be appropriate (Engle and Granger, 1987; Enders, 2003). The augmented Dickey–Fuller (1981, ADF) method was applied to assess the existence of unit roots and identify the order of integration for each variable. ADF tests were represented by the following regression equations:[3]

(1)

(2)

(3)

Where is the level of the variable, is a drift term, and is the time trend. is a normally distributed random error term with mean zero and constant variance. Δ denotes the series in first difference. is the number of lags necessary to obtain white noise.

Results of the ADF test for stationarity are reported in Table 3. The ADF statistics for all series levels except LPRICE were not significant at the 5% significance level, implying that the null hypothesis of a unit root test cannot be rejected. In addition, the corresponding statistics for their first-order differences are significant at the 5% significance level and thus suggest rejecting the null hypothesis. The results concluded that ΔLFIRE, ΔLCAR, ΔLINCOME, ΔLPRICE and ΔEDU variables were stationary.

Table 3

Unit Root Test

(None)

(Constant, no trend)

(Constant , trend)

Variables

ADF statistic

95% Critical values

p value

ADF statistic

95% Critical values

p value

ADF statistic

95% Critical values

p value

LFIRE

3.90

-1.95

1.00

-2.17

-2.95

0.22

-2.36

-3.55

0.39

LCAR

1.35

-1.95

0.95

-2.46

-2.95

0.13

-0.07

-3.55

0.99

LINCOME

2.07

-1.95

0.99

-1.50

-2.95

0.52

-1.25

-3.55

0.88

LPRICE

-1.46

-1.95

0.13

-3.98

-2.95

0.00

-3.84

-3.54

0.03

LEDU

2.86

-1.95

1.00

-0.13

-2.95

0.94

-2.22

-3.54

0.47

ΔLFIRE

-3.36

-1.95

0.00

-4.34

-2.95

0.00

-4.64

-3.55

0.00

ΔLCAR

-1.95

-1.95

0.05

-2.80

-2.95

0.07

-4.48

-3.55

0.01

ΔLINCOME

-3.20

-1.95

0.00

-4.05

-2.95

0.00

-4.25

-3.55

0.01

ΔLPRICE

-4.33

-1.95

0.00

-4.25

-2.96

0.00

-4.34

-3.56

0.01

ΔLEDU

-5.57

-1.95

0.00

-6.94

-2.95

0.00

-6.91

-3.55

0.00

Note: Δ denotes series in first difference. The lag parameters are selected based on the SIC. The critical values of the ADF test and one sided p-values are based on MacKinnon (1991, 1996). EViews 5.0 was used as the statistical software package for all tests.

3.3 Model Estimation

This study utilizes the disaggregated data and specifications of the regression model based on theory, in order to examine the relationship between the premium density and the independent variables. In this way, we might understand more about the decision of insurance demand. The regression specification is:

(4)

where is the dependent variable, is the parameter estimator of the intercept, stand for slope parameters, and is the random deviation. The model was estimated separately for fire and automobile premium densities as dependent variables.

If the time series of the variable is nonstationary, the regression equation can be written as:

(5)

where denotes the series in first difference. is the parameter estimator of the intercept. stand for slope parameters and is the random deviation.

Table 4 reports the results from the ordinary least square regression in levels. In the presence of nonstationary variables, this regression output (as Eqs. 4) looks good, but the results lack any economic meaning.

Table 4

Empirical Model Estimation in levels

LFIREt

LCARt

Variables

Coefficients

t-statistic

p-value

Coefficients

t-statistic

p-value

Intercept

-0.48

-2.81

0.01

-3.74

-10.05

0.00

LINCOMEt

0.70

11.26

0.00

1.61

11.84

0.00

LPRICEt

-0.05

-0.53

0.60

-0.08

-0.37

0.72

LEDUt

0.40

4.60

0.00

0.37

1.96

0.06

R2

0.98

0.97

DWa

0.90

0.24

LB – Qb

χ2 (6) = 24.25[0.00]

χ2 (6) = 61.66[0.00]

ARCHc

χ2 (6) = 5.75[0.45]

χ2 (6) = 22.15[0.00]

Normald

χ2 (2) = 1.86[0.40]

χ2 (2) = 1.12[0.57]

FFe

χ2 (1) = 2.84[0.09]

χ2 (1) = 28.13[0.00]

Notes: The number of lags is chosen by making the SIC statistic as small as possible; it should also be large enough to remove any serial correlation in the residuals. a Durbin–Watson test. b Lagrange multiplier test for residual serial correlation. c Based on the regression of squared residuals on squared fitted values. d Based on the test of the skewness and kurtosis of residuals. e Ramsey’s RESET test using the square of fitted values.

Table 4 also presents the empirical model in first differences. Thus, the discussion focuses on results obtained from first-differenced model (as Eqs. 5).

Table 4

Empirical Model Estimation in first differences

∆LFIREt

∆LCARt

Variables

Coefficients

t-statistic

p-value

Coefficients

t-statistic

p-value

Intercept

-0.44

-2.59

0.01

-3.67

-9.74

0.00

∆LINCOMEt-1

0.68

10.62

0.00

1.57

11.17

0.00

∆LPRICE t-1

-0.06

-0.61

0.55

-0.10

-0.44

0.66

∆LEDU t-1

0.44

4.69

0.00

0.46

2.24

0.03

R2

0.98

0.97

DWa

0.95

0.24

LB – Qb

χ2 (6) = 23.28[0.00]

χ2 (6) = 57.05[0.00]

ARCHc

χ2 (6) = 5.04[0.54]

χ2 (6) = 20.44[0.00]

Normald

χ2 (2) = 1.47[0.48]

χ2 (2) = 1.14[0.56]

FFe

χ2 (1) = 7.74[0.01]

χ2 (1) = 12.87[0.00]

Notes: The number of lags is chosen by making the SIC statistic as small as possible; it should also be large enough to remove any serial correlation in the residuals. a Durbin–Watson test. b Lagrange multiplier test for residual serial correlation. c Based on the regression of squared residuals on squared fitted values. d Based on the test of the skewness and kurtosis of residuals. e Ramsey’s RESET test using the square of fitted values.

3.3.1 Income

As reported in Table 5, the relationship between the dependent variable premium density and the independent variable LINCOME (i.e., real GDP per capita) is positive and statistically significant in both the fire and automobile insurance model. Our data is consistent with the findings from prior studies in that income is positively correlated with insurance demand. Therefore, the results suggest that the higher the income is, the more they would purchase insurance. Although income is statistically significant in both models, the regression coefficient of 0.68 and 1.57 (Table 5), the coefficient estimates suggest that changes in income have a more pronounced effect on automobile insurance demand than on fire insurance demand. As a result, the proof shows that income has a far greater effect on automobile insurance demand than on fire insurance demand.

3.3.2 Price of Insurance

As reported in Table 5, the regression coefficients of the premium density of automobile insurance and the independent variable of insurance price is -0.10. This is consistent with the negative relationship in economical implications, though the p-value is statistically insignificant. At the same time, the regression coefficients of the premium density for fire insurance and the independent variable of insurance price is -0.06. This is also consistent with the negative relationship in economical implication, though the p-value is also statistically insignificant.

The domestic insurance industry, in its early ages, was strictly limited by government agencies. This enabled it to become stabilized. However, in later years, even with the openness for foreign insurance companies or domestic companies, the insurance industry has turned into an oligopoly market. As a result, the price variation of insurance products is not obvious; nevertheless, the relationship between insurance price and premium density is statistically insignificant.

3.3.3 Risk Aversion

The risk-aversion measure used in this study, for people over the age of 15, is statistically significant in all models. This result means that the higher the level of education, the greater the demand for insurance. For fire insurance, the regression coefficient was strong statistically significant, and that of automobile insurance is barely statistically significant. This corresponds to the prior research of Browne and Kim (1993), which showed that a high level of education leads to high risk aversion and these people would buy more insurance.

4. Conclusions

Since the 1970s, the global insurance market has developed quickly, and Taiwan is no exception. But the factors affecting the demand of insurance in Taiwan are different. Several theories have been developed to explain the motives to purchase non-life insurance (i.e., property-casualty insurance), but there have been few empirical tests conducted by emerging insurance markets.

This study examines the determinants of non-life insurance consumption in Taiwan between 1970 and 2005, focusing on fire insurance and automobile insurance. Using econometric regression models, the analysis indicates that economic conditions affect the demand for insurance differently across lines of coverage (i.e., fire insurance and automobile insurance). The results suggest that income has a far greater effect on automobile insurance demand, than on fire insurance demand. Moreover, the results show the purchase of non-life insurance is significantly and positively related to income and risk aversion, as well as providing weaker evidence of a negative relationship with price.

Managerial implications will now be suggested. The income elasticity of demand for insurance can assist insurance companies in precisely determining the consumption of insurance products. Furthermore, the insurance industry should consider different marketing strategies for the marketing mix between fire insurance and automobile insurance. In future studies, we recommend insurance researchers compare multiple emerging markets using the same variables we have used in this study. Based on the results, the insurer can carry out this model to predict future insurance demand and decision-making to enter the market or promote insurance products.


Reference

[1] Beenstock, M., Dickinson, G. and Khajuria S. (1988) the he Relationship between Property-Liability Insurance Premiums and Income: An International Analysis, Journal of Risk and Insurance, 55: 259-272.

[2] Browne, M. J., and Kim K. (1993) An International Analysis of Life Insurance Demand. Journal of Risk and Insurance, vol. 60, pp. 617-634.

[3] Browne, M.J., Chung J., and Frees E. W. (2000) International Property-Liability Insurance Consumption Journal of Risk and Insurance, 67(1): 73-90.

[4] Cummins, D. J., and P. M. Danzon, (1997) Price, Financial Quality, and Capital Flows in Insurance Markets, Journal of Financial Intermediation, 6:3-38.

[5] Dickey, D. A. and Fuller, W. A. 1981, Likelihood ratio statistics for autoregressive time series with a unit root, Econometrica, 49(4),1057–1071.

[6] Enders, W., 2003, Applied Time Series Econometrics (New York: John Wiley and Sons).

[7] Engle, R., and Granger, C. W., 1987, Co-Integration and Error Correction: Representation, Estimation, and Testing, Econometrica, 55: 251-276.

[8] Esho, N., A. Kirievsky, D. Ward, and Zurbruegg, R., 2004, Law and the Determinants of Property-Casualty Insurance, Journal of Risk and Insurance, 71(2): 265-283.

[9] Garven, J. R., and MacMinn R. D. (1993) The Underinvestment Problem, Bond Covenants, and Insurance, Journal of Risk and Insurance, 60: 635-646.

[10] Hamilton, J.D., 1994, Time Series Analysis, Princeton University Press, Princeton, NJ.

[11] Li, D., Moshirian, F., Nguyen, P., and Wee, T., 2007, The Demand for Life Insurance in OECD Countries, Journal of Risk and Insurance, Vol. 74, No. 3, pp. 637-652.

[12] Ma, Y. L., and Pope, N., 2003, Determinants of International Insurers Participation in Foreign Non-Life Markets, Journal of Risk and Insurance, Vol. 70, No. 2, 235-248.

[13] MacKinnon, J. G., 1991, Critical Values for Cointegration Tests in Long-Run Econometric Relationships, in: R. F. Engle and C. W. J. Granger, eds., Readings in Cointegration (New York: Oxford Press).

[14] MacKinnon, J. G., 1996, Numerical Distribution Functions for Unit Root and Cointegration Test, Journal of Applied Econometrics, 11, 601-618.

[15] MacMinn, R. D. (1987) Insurance and Corporate Risk Management Journal of Risk and Insurance, vol. 54, no. 4, pp.658-677.

[16] MacMinn, R. D. and Han L. M. (1990) Limited Liability, Corporate Value, and the Demand for Liability Insurance, Journal of Risk and Insurance, 57:581-607.

[17] Mayers, D. and C. W. Smith, Jr. (1990) On the Corporate Demand for Insurance: Evidence from the Reinsurance, Journal of Business, Vol. 63, pp.19-40.

[18] Mossin J. (1968) Aspects of Rational insurance Purchasing, Journal of Political Economy, 553-56

[19] Outreville, J. F. (1990) The Economic Significance of Insurance Markets in Developing Countries, Journal of Risk and Insurance, 63(2): 263-278.

[20] Regan L. and Hur Y. (2007) On the Corporate Demand for Insurance: the Case of Korean Nonfinancial Firms Journal of Risk and Insurance, 74: 829-850.

[21] Schlesinger, H. (1981) the Optimal Level of Deductibility in Insurance Contracts, Journal of Risk and Insurance, vol. 48, pp.465-481

[22] Szpiro, G., and Outreville, J. F. (1988) Relative Risk Aversion around the World, Studies in Banking and Finance, 6: 123-129.

[23] Szpiro, G. G. (1985) Optimal Insurance Coverage, Journal of Risk and Insurance. Vol. 52, 88.705-710. (4)

[24] Ward, D. and Zurbruegg, R., 2002, Law, Politics and Life Insurance Consumption in Asia, The Geneva Papers on Risk and Insurance, 27(3), 395-412.



[1] For more detailed information on the economic growth rate, refer to the following website: http://www.dgbas.gov.tw

[2] Zurich: Swiss Reinsurance Company, Sigma Publications, No. 5/2006

[3] There are three cases for the ADF unit root test discussed in Hamilton (1994) and Enders (2003). To select the optimal lag-length for each model, we select a model with the lowest value of the SIC.

No comments:

Post a Comment

Search for content in this blog.

Life Insurance