AN INVESTIGATION OF THE INSURANCE SECTORS CONTRIBUTION TO ECONOMIC GROWTH

By Haizhi Tong

A DISSERTATION

Presented to the Faculty of The Graduate College at the University of Nebraska In Partial Fulfillment of Requirements For the Degree of Doctor of Philosophy

Major: Economics

Under the Supervision of Professor Hendrik van den Berg

Lincoln, Nebraska

August, 2008

3315878

2008

3315878

AN INVESTIGATION OF THE INSURANCE SECTORS CONTRIBUTION TO ECONOMIC GROWTH

Haizhi Tong, PhD. University of Nebraska, 2008 Adviser: Hendrik van den Berg The importance of the insurance industry in an economy has been well recognized. Yet there is no extensive study on this area. In this dissertation, the economic role of the Insurance sector has been studied through 1) setting up theoretical models to illustrate how insurance growth may contribute to economic growth and 2) conducting a series of empirical studies to find whether there is empirical evidence to support the models developed. Although the insurance sector plays several important roles in economic growth, I only focus on two roles here in the dissertation. First, property and liability insurance serve as risk sharing institutions in an economy. Second, life insurance sector can provide more long run capital into an economy. I develop two theoretical models to illustrate the above two roles separately. In the property and liability insurance model, an economy, which is composed of risk-averse agents, without an insurance sector is set up first. It shows that individual utility maximization can not achieve social optimization. After an insurance sector is introduced into the economy, social optimization can be achieved through individual utility maximization. In the life insurance model, contractual savings institutions, which include life insurance companies, can shift short term savings to long term savings. At the end of this part, a discussion is presented to link insurance sector with the technological progress. Four countries data have been studied to see whether there is empirical evidence for the above models developed. Three empirical methods are employed and compared: OLS, simultaneous equations and fixed effect model. All four countries data show evidence that property and liability insurance promote economic growth. However, two

countries data show that life insurance promote economic growth while the other two countries data have just the opposite conclusion. Some possible reasons for this evidence are presented.

Acknowledgements I would like to express my sincere appreciation to several of the people who have contributed to my graduate study and dissertation: First, to my advisor, Dr. Hendrik van den Berg, for having the patience to let me work at my own pace when the rest of my life tried to take over. Dr. van den Berg supported me with great ideas, provided guidance and feedback on my progress, and encouraged me to continue when I needed it most. To my committee members, Dr. Craig MacPhee, Dr. Mary McGarvey, and Dr. Lilyan Fulginiti, for their great teachings and advices through my PhD study years, valuable suggestions and comments on this dissertation and always being ready in helping me. They have been so patient and considerate to me during these years. To my friends, Doug and Carol Tanner, for their friendly assistance on proofreading the draft of this dissertation. To my Parent, Zemin Tong and Chunyu Yang, for their love and encouragement through my life. To my son, Luke Zhang, for his love and understanding especially when I was too busy to spend time with him while I was working on this dissertation. Most of All, to my beloved husband, Lingfeng Zhang, for his ongoing love and constant support. Words are not sufficient to express the thanks I owe to him. I thank you all from the bottom of my heart. I could not have done this alone without you.

TABLE OF CONTENTS

CHAPTERS.................................................................................................................................... I LIST OF TABLES ....................................................................................................................IVV

CHAPTERS
CHAPTER 1 INTRODUCTION...............................................................................................................1 I. RISK AND INSURANCE ..............................................................................................................................1 i) Risk and Insurable Risk......................................................................................................................1 ii) Insurance and How It Works.............................................................................................................3 iii) Cost of Insurance .............................................................................................................................5 II. THE IMPORTANCE AND THE FUNCTION OF INSURANCE IN AN ECONOMY ................................................7 III. TYPES OF INSURANCE ............................................................................................................................9 IV. AIMS OF THIS DISSERTATION ..............................................................................................................11 CHAPTER 2 THEORETICAL FRAMEWORK ...................................................................................14 I. PROPERTY/CASUALTY/LIABILITY INSURANCE AND ECONOMIC GROWTH ..............................................14 i) The Risk Sharing Role of Property/Casualty and Liability Insurance in an Economy.....................14 ii) Literature Review on the Economic Role of a Financial Sector......................................................15 iii) Theoretical Model Linking Insurance to Economic growth ...........................................................24 II. THE LONG TERM CAPITAL ACCUMULATION ROLE OF INSURANCE IN AN ECONOMY ............................33 i) Individual Savings Behaviors without Contractual Savings Institutions..........................................36 ii) Individual Savings Behaviors with Contractual Savings Institutions..............................................39 iii) Compositions of Savings with and without Contractual Savings Institutions ...............................42 III. DISCUSSION: TECHNOLOGICAL PROGRESS, INSURANCE AND ECONOMIC GROWTH .............................44 i) Factors that Affect Technological Progress .....................................................................................44 ii) Insurance and Technological Progress...........................................................................................49

II
CHAPTER 3 EMPIRICAL STUDY OF INSURANCE GROWTH AND ECONOMIC GROWTH 53 I. LITERATURE REVIEW .............................................................................................................................53 II. DATA AND METHODOLOGY ..................................................................................................................56 i) Data..................................................................................................................................................56 ii) Methodology....................................................................................................................................58 III. DEFINITION OF VARIABLES ..................................................................................................................64 CHAPTER 4 EMPIRICAL RESULT FROM FOUR COUNTRIES ...................................................67 I. ANALYSIS OF THE US DATA ...................................................................................................................67 i) Unit Root Test Results ......................................................................................................................67 ii) OLS Regression ..............................................................................................................................69 iii) Simultaneous Equations .................................................................................................................71 II. ANALYSIS OF KOREAN DATA................................................................................................................80 i) Unit Root Test...................................................................................................................................80 ii) Simple OLS Regression ...................................................................................................................81 iii) Simultaneous Equations .................................................................................................................84 III. ANALYSIS OF SWEDISH DATA .............................................................................................................90 i) Unit Root Test...................................................................................................................................90 ii) Simple OLS Regression ...................................................................................................................91 iii) Simultaneous Equations .................................................................................................................93 IV. ANALYSIS OF THE GERMAN DATA.......................................................................................................99 i) Unit Root Test Results ......................................................................................................................99 ii) Simple OLS Regression .................................................................................................................100 iii) Simultaneous Equations ...............................................................................................................103 V. FIXED EFFECT MODEL ........................................................................................................................109 VI. SUMMARY AND COMPARISON OF THE RESULTS ACROSS THE FOUR COUNTRIES ...............................114 CHAPTER 5 CONCLUSIONS AND RECOMMENDATIONS FOR FUTURE RESEARCH .......118 I. SUMMARY OF THEORETICAL MODELS...................................................................................................119

III
II. SUMMARY OF EMPIRICAL FINDINGS ...................................................................................................121 III. LIMITATION OF THE RESEARCH ..........................................................................................................123 IV. SUGGESTION FOR FUTURE RESEARCH ...............................................................................................125 APPENDIX A ............................................................................................................................................127 APPENDIX B.............................................................................................................................................129 APPENDIX C ............................................................................................................................................132 APPENDIX D ............................................................................................................................................133 BIBLIOGRAPHY .....................................................................................................................................134

IV

LIST OF TABLES
TABLE 2.2.1 ASSUMED LIFE EXPECTANCY PROBABILITY DISTRIBUTION ................................................38 TABLE 2.2.2 INFLOW AND OUTFLOW OF FUNDS (PENSION) ......................................................................40 TABLE 2.2.3 INFLOW AND OUTFLOW OF FUNDS (LIFE INSURANCE).........................................................43 TABLE 4.1.1 UNIT ROOT TEST RESULTS (US DATA).................................................................................68 TABLE 4.1.2 ESTIMATION RESULTS FROM OLS (US DATA) - POPULATION GROWTH IS USED AS A PROXY OF LABOR GROWTH ...............................................................................................................69 TABLE 4.1.3 ESTIMATION RESULTS FROM OLS (US DATA) - LABOR FORCE GROWTH IS USED AS A PROXY OF LABOR GROWTH ...............................................................................................................70 TABLE 4.1.4 ESTIMATION RESULTS FROM SIMULTANEOUS EQUATIONSEQUATION 1 (US DATA) POPULATION GROWTH IS USED AS A PROXY OF LABOR GROWTH ...................................................72 TABLE 4.1.5 ESTIMATION RESULTS FROM SIMULTANEOUS EQUATIONSEQUATION 2 (US DATA) POPULATION GROWTH IS USED AS A PROXY OF LABOR GROWTH ...................................................73 TABLE 4.1.6 ESTIMATION RESULTS FROM SIMULTANEOUS EQUATIONSEQUATION 3 (US DATA) POPULATION GROWTH IS USED AS A PROXY OF LABOR GROWTH ...................................................75 TABLE 4.1.7 ESTIMATION RESULTS FROM SIMULTANEOUS EQUATIONSEQUATION 1 (US DATA) LABOR FORTH GROWTH IS USED AS A PROXY OF LABOR GROWTH ................................................76 TABLE 4.1.8 ESTIMATION RESULTS FROM SIMULTANEOUS EQUATIONSEQUATION 2 (US DATA) LABOR FORTH GROWTH IS USED AS A PROXY OF LABOR GROWTH ................................................78 TABLE 4.1.9 ESTIMATION RESULTS FROM SIMULTANEOUS EQUATIONSEQUATION 3 (US DATA) LABOR FORTH GROWTH IS USED AS A PROXY OF LABOR GROWTH ................................................78 TABLE 4.2.1 UNIT ROOT TEST RESULTS (KOREAN DATA) .......................................................................80 TABLE 4.2.2 ESTIMATION RESULTS FROM OLS (KOREAN DATA) - POPULATION GROWTH IS USED AS A PROXY OF LABOR GROWTH ...............................................................................................................82 TABLE 4.2.3 ESTIMATION RESULTS FROM OLS (KOREAN DATA) - LABOR FORCE GROWTH IS USED AS
A PROXY OF LABOR GROWTH ............................................................................................................83

V
TABLE 4.2.4 ESTIMATION RESULTS FROM SIMULTANEOUS EQUATIONSEQUATION 1 (KOREAN DATA) - LABOR FORCE GROWTH IS USED AS A PROXY OF LABOR GROWTH ..............................................84 TABLE 4.3.5 ESTIMATION RESULTS FROM SIMULTANEOUS EQUATIONSEQUATION 2 (KOREAN DATA) - LABOR FORCE GROWTH IS USED AS A PROXY OF LABOR GROWTH ..............................................85 TABLE 4.2.6 ESTIMATION RESULTS FROM THE SIMULTANEOUS EQUATIONSEQUATION 3 (KOREAN DATA) - LABOR FORCE GROWTH IS USED AS A PROXY OF LABOR GROWTH ..................................86 TABLE 4.2.7 ESTIMATION RESULTS FROM THE SIMULTANEOUS EQUATIONSEQUATION 1 (KOREAN DATA) - POPULATION GROWTH IS USED AS A PROXY OF LABOR GROWTH .....................................88 TABLE 4.2.8 ESTIMATION RESULTS FROM THE SIMULTANEOUS EQUATIONSEQUATION 2 (KOREAN DATA) - POPULATION GROWTH IS USED AS A PROXY OF LABOR GROWTH .....................................88 TABLE 4.2.9 ESTIMATION RESULTS FROM THE SIMULTANEOUS EQUATIONSEQUATION 3 (KOREAN DATA) - POPULATION GROWTH IS USED AS A PROXY OF LABOR GROWTH .....................................89 TABLE 4.3.1 UNIT ROOT TEST RESULTS (SWEDISH DATA).......................................................................90 TABLE 4.3.2 ESTIMATION RESULTS FROM OLS (SWEDISH DATA) - POPULATION GROWTH IS USED AS A PROXY OF LABOR GROWTH ...............................................................................................................91 TABLE 4.3.3 ESTIMATION RESULTS FROM OLS (SWEDISH DATA) - LABOR FORCE GROWTH IS USED AS
A PROXY OF LABOR GROWTH ............................................................................................................92

TABLE 4.3.4 ESTIMATION RESULTS FROM SMEEQUATION 1 (SWEDISH DATA) - LABOR FORCE GROWTH IS USED AS A PROXY OF LABOR GROWTH .........................................................................93 TABLE 4.3.5 ESTIMATION RESULTS FROM SMEEQUATION 2 (SWEDISH DATA) - LABOR FORCE GROWTH IS USED AS A PROXY OF LABOR GROWTH .........................................................................94 TABLE 4.3.6 ESTIMATION RESULTS FROM SMEEQUATION 3 (SWEDISH DATA) - LABOR FORCE GROWTH IS USED AS A PROXY OF LABOR GROWTH .........................................................................95 TABLE 4.3.7 ESTIMATION RESULTS FROM SMEEQUATION 1(SWEDISH DATA) - POPULATION GROWTH IS USED AS A PROXY OF LABOR GROWTH .........................................................................96 TABLE 4.3.8 ESTIMATION RESULTS FROM SMEEQUATION 2 (SWEDISH DATA) - POPULATION GROWTH IS USED AS A PROXY OF LABOR GROWTH .........................................................................96

VI
TABLE 4.3.9 ESTIMATION RESULTS FROM SMEEQUATION 3 (SWEDISH DATA) - POPULATION GROWTH IS USED AS A PROXY OF LABOR GROWTH .........................................................................97 TABLE 4.4.1 UNIT ROOT TEST RESULTS (GERMAN DATA).......................................................................99 TABLE 4.4.2 ESTIMATION RESULT FROM OLS (GERMAN DATA) POPULATION GROWTH IS USED AS A PROXY OF LABOR GROWTH .............................................................................................................101 TABLE 4.4.3 ESTIMATION RESULTS FROM OLS (GERMAN DATA) LABOR FORTH GROWTH IS USED AS
A PROXY OF LABOR GROWTH ..........................................................................................................102

TABLE 4.4.4 ESTIMATION RESULTS FROM SMEEQUATION 1 (GERMAN DATA) - LABOR FORCE GROWTH IS USED AS A PROXY OF LABOR GROWTH .......................................................................103 TABLE 4.4.5 ESTIMATION RESULTS FROM SMEEQUATION 2 (GERMAN DATA) - LABOR FORCE GROWTH IS USED AS THE PROXY OF LABOR GROWTH ...................................................................104 TABLE 4.4.6 ESTIMATION RESULTS FROM SMEEQUATION 3 (GERMAN DATA) - LABOR FORCE GROWTH IS USED AS A PROXY OF LABOR GROWTH .......................................................................105 TABLE 4.4.7 ESTIMATION RESULTS FROM SMEEQUATION 1 (GERMAN DATA) - POPULATION GROWTH IS USED AS A PROXY OF LABOR GROWTH .......................................................................106 TABLE 4.4.8 ESTIMATION RESULTS FROM SMEEQUATION 2 (GERMAN DATA) - POPULATION GROWTH IS USED AS A PROXY OF LABOR GROWTH .......................................................................106 TABLE 4.4.9 ESTIMATION RESULTS FROM SMEEQUATION 3 (GERMAN DATA) - POPULATION GROWTH IS USED AS A PROXY OF LABOR GROWTH .......................................................................107 TABLE 4.5.1 ESTIMATION RESULTS FROM FEM (INCLUDES GERMAN DATA BEFORE THE MERGE) POPULATION GROWTH IS USED AS A PROXY OF LABOR GROWTH .................................................110 TABLE 4.5.2 ESTIMATION RESULTS FROM FEM (INCLUDES GERMAN DATA BEFORE THE MERGE) LABOR FORCE GROWTH IS USED AS THE PROXY OF LABOR GROWTH ..........................................110 TABLE 4.5.3 ESTIMATION RESULTS FROM FEM (INCLUDES ALL GERMAN DATA) - POPULATION GROWTH IS USED AS A PROXY OF LABOR GROWTH .......................................................................111 TABLE 4.5.4 ESTIMATION RESULTS FROM FEM (INCLUDES ALL GERMAN DATA) - LABOR FORCE GROWTH IS USED AS A PROXY OF LABOR GROWTH .......................................................................111 TABLE 4.6.1 EFFECT OF INSURANCE GROWTH ON ECONOMIC GROWTH ..............................................114

VII
TABLE 4.6.2 EFFECT OF ECONOMIC GROWTH ON INSURANCE GROWTH ..............................................115 TABLE AP-1 PROFITS FROM SHORT TERM PROJECTS ............................................................................130 TABLE AP-2 PROFITS FROM LONG TERM PROJECTS .............................................................................130

Chapter 1
I. Risk and Insurance

Introduction

One major feature in the modern society is increasing risks in economic life. When we talk about risk in the economic life here, we are referring to the risk which may lead to the occurrence of loss. Although classic economics acknowledges the difference of risk bearing across different individuals, there is lack of modeling on different economic results due to the existence of risks. The role of insurance sector in an economy is not studied broadly by economists. In this dissertation, I am trying to build some models to illustrate how insurance can affect an individuals economic behavior and therefore affect the whole economy. Before we do that, we will first start with the types of risks and the function of insurance in reducing some types of risks.

i) Risk and Insurable Risk

1. Objective Risk vs. Subjective Risk Objective risk is measurable risk which can be derived from facts and data. On the other hand, subjective risk is based on a persons perspective on a specific uncertain outcome. Because subjective risk is related to a persons opinion, it can not be measured. An example of the objective risk is the variation of car accidents in a particular year in a city. Suppose we know the expected number of car accidents in the long run. But from year to year, the actual number of car accidents can be quite different. The variation from the expected number of car accidents is an example of objective risk.

2 On the other hand, we can think about people with similar situations such as same age, same gender and same marriage status driving a same car in the same city may perceive differently on their involvement of a car accident. This is an example of subjective risk. Only objective risk is insurable. However, not all objective risk is insurable. To be insurable, the risk should also be a pure risk. 2. Pure Risk vs. Speculative Risk There are only two outcomes for pure riskloss or no loss. An example of pure risk is that a person who drives a car facing two possible results, a car accident which will result in loss and no car accident which incurs no loss. Speculative risk is associated with three outcomesloss, no loss and gain. For example, a person holding some stock is facing three outcomesan increase of the price of the stock, a decrease of the price of the stock and no change in the price of the stock after a period of time. Although speculative risk is greatly dealt with in risk management, it is not in the scope of insurance. Insurable risk is pure risk. This means insurance protects against the risk of loss but not as a tool to gain profit. 3. Financial Risk vs. Non Financial Risk Financial risk is measurable in money. The physical loss from the fire of a particular residence can be measured by the amount of dollars. Non financial risk is not compensated by money. For example, loss of precious photographs from a fire is more related with the fond memory of the owner, which we can not measure by money. Insurance deals with financial risk rather than non financial risk.

3 4. Diversifiable Risk vs. Non-Diversifiable Risk Diversifiable risk is the risk that can be spread out in a large group. It is also called unsystematic risk or particular risk. The risk exposures in the group are not correlated with each other theoretically. When more risk exposures are included into the group, there is less variation from the expected number of loss occurrence. Some examples are car accidents, fire, disability, premature death etc. Non diversifiable risk is also called systematic risk or fundamental risk. It is the risk that will affect the large number of units in the group at the same time. Examples of non-diversifiable risk are earthquakes, hurricanes, economic inflation, and unemployment. Insurance commonly deals with diversifiable risk whereas government usually deals with non-diversifiable risk. From what we summarized above, we can see that usually the character of insurable risk is objective, pure, financial and diversifiable risk.

ii) Insurance and How It Works

According to the American Risk and Insurance Association, Insurance is the pooling of fortuitous losses by transfer of such risks to insurers, who agree to indemnify an insured for such losses, to provide other pecuniary benefits on their occurrence, or to render services connected with the risk. 1 For insurable risk, insurance works by pooling of losses based on the law of large numbers.

Bulletin of the commission on Insurance Terminology of the American Risk and Insurance Association, October 1965

4 Pooling is the spreading of losses incurred by a few over the entire group, so that in the process, average loss is substituted for actual loss 2. By pooling a large number of similar exposure units with same perils group together so that the entire group can share the risk with relative accurate predictions. The accurate prediction is based on the law of large numbers. The law of large numbers states that as the number of exposure units increases, the more closely the actual loss will approach the expected loss experience 3. We can use an example to put all those concepts together to see how insurance works. Suppose for a particular city, each year there is 1000 houses that will catch on fire out of, say, 1 million houses. Also, for simplicity, the fire will incur the whole loss of the house and each house has the same value like $100,000. Therefore, each individual will face either no loss when there is no fire or a total loss of $100,000 when there is a fire. The probability of fire is 0.001. The expected loss for that individual is $100. However, this number is not a number this individual will face if there is no insurance. The individual either has a huge loss with very small probability or no loss with a very big probability. When we pool all those individuals who own a house in the city, we get a very large group. We can see that if there are exactly 1000 houses burnt down in a particular year, the average loss of each individual in the group is $100, which is the same as the expected loss faced by an individual. In real life, the real number of houses burnt down will be different from 1000 very obviously. Yet we can expect that the real number of houses burnt down from year to year is close to 1000. And the
2
3

Principles of Risk Management and Insurance, George E. Rejda, 2000, pg. 20. Principles of Risk Management and Insurance, George E. Rejda, 2000, pg. 4.

5 average loss of each individual will be pretty close to $100, which is the expected value that each individual will face. For the group, the average loss for each individual is very close to the expected loss for each individual due to the law of large numbers. Also, what we can see here is that by pooling the exposure units together, the loss is very predictable. The existence of insurance makes individual exchange their risk of loss for a certain amount of money which is called premium. By doing so, individuals will change their risky economic situation to a safe economic situation. Therefore, an economy with or without an insurance sector can affect the economic behavior of the agents in the economy.

iii) Cost of Insurance

Of course, in the real world situation, insurance companies also need to face the dishonesty and carelessness of individuals. Insurance companies need to deal with moral hazard, morale hazard, and adverse selection. Let us look at the meaning of each of the above term. Moral hazard is defined as dishonesty or character defects in an individual that increase the frequency or severity of loss. Morale hazard is defined as carelessness or indifference to a loss because of the existence of insurance4. The idea of adverse selection is that the insured knows better about her/his risk tendency than the insurance company. By not disclosing the information voluntarily, they may obtain a rate which is lower than their expected rate of loss.

Principles of Risk Management and Insurance, George E. Rejda, 2000, pg. 6.

6 Insurance companies can add deductible or coinsurance factor to reduce moral hazard or morale hazard. It can also design the underwriting policy carefully to

reduce the chance for adverse selection. In addition to the above factors, there is administration cost of insurance companies such as the cost of buildings and wages and salaries paid to the employees of insurance companies. From the above discussion, we can see that what insurance charges for an individual should be higher than the expected loss of each individual in the group. However, we will put those factors aside when we analyze how the existence of the insurance sector can affect individual economic decision.

II. The Importance and the Function of Insurance in an Economy

The importance of the insurance industry in an economy has been well recognized. In 1964 the United Nations Conference on Trade and Development (UNCTAD) has stated that a sound national insurance and reinsurance market is an essential characteristic of economic growth. (page 55 ) The following figures also show the important role of the insurance sector in an economy. According to the research paper of Hess (2002), 7.8% of the world GDP was spent on insurance products in the year 2000. However, the development of the insurance industry is different across countries. Insurance seems to have a more important role in developed countries than in developing countries. This is not surprising since the insurance sector is part of the financial institutions, which has been viewed to be related with economic development. According to Hess (2002), the percent GDP spent on insurance products is between 8.6% and 10.9% among developed countries but is only between 2.1% to 3.8% among developing countries. Empirically, there is clear evidence that differences are dependent on the stage of national economic development. (Hess, 2002, p. 1) The statistics further supported that economic development is linked with insurance development. By reviewing some literatures (Arrow 1965, Albouy and Blagoutine 2001, Hess 2002), we have a brief summarization of the economic role of an insurance sector. First, the insurance industry is one of the risk sharing institutions. The insurance industry provides protection to individuals or firms against a large amount of loss with a small probability (Arrow, 1965). In doing so, the insurance sector can provide more

8 financial security to economic agents. This makes individuals or firms willing to engage in risky activities, which are impossible when there isnt such a system. For example, individuals may reconsider their decisions of buying a car if they cannot buy car insurance. People will be reluctant to own a house if there is no property and liability insurance to protect the loss of their house due to fire, flooding etc. Also, we will expect firms will make different investment decisions whether the risks associated with a new project are insurable or non insurable. Therefore the production decisions may be altered when there is such a risk sharing institution. Second, insurance provides long-term capital in the capital market. Savings is very important to economic growth. Both total savings and the structure of savings matter. When there are more long-term savings versus short-term savings, an economy can finance more long-term projects. Insurance and pension funds provide long term savings to the capital market, which broadening and deepening the functions of the capital market. Third, the insurance industry makes risk management more efficient. The insurance industry is a sector devoted in risk management. They hire experts in risk management to price the risks so that the market will give signal to the individuals on which direction to go: Actuaries calculating the price for earthquake or flood risks give important signals as to where to build house or factories and where not. (Hess 2002 p.2)

III. Types of Insurance

We will divide insurance into four groups here according to its dealing with four different groups of loss exposures. The definition is summarized from the book of Insurance Perspectives by Gibbons, Rejda and Elliott, 1992. 1. Property/Casualty Insurance Property/casualty insurance protects the insured against accidental losses due to loss or damage of the property of the insured. One example is that home owners insurance will cover the owners damage to his/her property due to a fire. His/her property here includes all his/her belongs in the house and the house itself. 2. Liability Insurance Liability insurance protects the insured from any third party claims against the insured when the insured is proven to be legally liable for injury or damage to the claimants. The insurer will pay to the claimants directly the damages the insured is liable to the claimants. One example is the auto liability insurance. It covers the loss and injuries of other parties due to an insured drivers allegedly negligent driving. We will combine property and liability insurance into one group and study its risk sharing role in an economy in this dissertation. Property and liability can also be divided into personal line insurance and commercial line insurance. Personal line insurance covers individuals and families against property and liability loss. Commercial line insurance covers businesses and other organizations against property and liability loss. In this dissertation, we do not separate them into groups.

10 3. Health Insurance (Accident and Sickness Insurance) Health insurance protects individuals and families against financial losses due to accidents and sickness. Both medical expense insurance and disability income insurance belong to the health insurance group. Medical expense insurance pays the cost of medical care of the insured due to accidents and sickness. Disability income insurance covers the income loss when the insured is not able to work due to accidents and sickness. In this dissertation, we will not study the role of health insurance in an economy. Carmichael, Benoit and Dissou (2000) have developed a model to illustrate the role of health insurance in an economy. 4. Life insurance Life insurance policy specifies that upon the death of the insured, the insurance company will pay the beneficiary a specified amount named in the policy. It can provide the named beneficiary from the income loss due to the premature death of the insured. It can be divided into term life insurance and whole life insurance. The former one provides a temporary protection while the latter one provides a life long protection. In the dissertation, we will focus on the capital accumulation role of life insurance companies in an economy.

11

IV. Aims of This Dissertation

Although the contribution of the insurance development to an economys development has been well recognized, the research in this area is not extensive (Albouy and Blagoutine, 2001). Several empirical studies have found there is link between insurance
growth and income growth in an economy. However, there is lack of theoretical models

which explain why the insurance industry may contribute to economic growth. To my best knowledge, there is only one paper explaining why health insurance may contribute to economic growth by setting up a theoretical model. Another research which links life insurance with economic growth has been done by Soo (1996), an econ PhD student in his dissertation. I am interested in developing theoretical framework that links economic development and insurance development, followed by empirical examination on the role of the insurance sector in an economy. The dissertation is composed of two sections. Section I gives out a theoretical framework which explains why the insurance industry may contribute to economic growth. In this part we will just focus on the first two economic roles of the insurance sector in an economyrisk sharing and long term capital accumulation. Two theoretical models will be presented in this chapter. These two models are dealing with these two roles separately. The first model focuses on the role of property and liability insurance. It illustrates the risk sharing role of property and liability insurance. We assume economic agents do not live in risk free environment. But to simplify the situation, we only assume two possible outcomes. Economic agents face the possibility of huge loss when they choose a higher productive method. On the other hand, they face no loss when they

12 choose a safer yet lower productive method. We get the conclusion that agents will choose the lower productive method over the higher productive method without an insurance sector. After the introduction of an insurance sector, we can prove that agents will choose the higher productive method. This is due to the risk averse character (which is shown in the assumed individual function) of economic agents. This gives an illustration that under uncertainty, individual maximization choice is not the optimal economic choice for the whole economy. Presence of the insurance sector can improve the social economic choice. The second model focuses on the role of life insurance sector. It shows how the existence of life insurance companies can provide more long term capital to an economy. Pension funds and life insurance companies are called contractual savings institutions. The emergence of contractual savings institutions may affect both the total amount of savings and the saving pattern of individuals. Here the individuals savings patterns change refers to the structure change between liquid savings (used for short term investment) and illiquid savings (used for long term investment). Therefore, the emergence of contractual savings institutions may affect economic growth through providing more long term capital. The arguments go as follows. Compared with banks, contractual savings institutions have special knowledge on population life expanse distribution. This makes the contractual savings institutions able to offer individuals savings plans. These savings plans can increase individuals utility level. Put it another way, there is a demand for contractual savings institutions. This leads to the emergence of contractual savings

13 institutions. Compared with banks, contractual savings institutions can offer more long run funds. At the end of this chapter, a discussion will be presented to show how the insurance sector can be linked to technological progress, which sustains long run economic growth. Following the theoretical illustration is an empirical study of the relationship between insurance development and economic growth. Several available countries data are examined to see the relationship between economic growth and insurance development. The data from the following four countries: US, South Korea, Sweden, and Germany. When I first collected the data, only data from these four countries were available through our library system. In this chapter, the property and liability and life insurance data are collected separately for these four countries. The growth of property and liability insurance premium is used to measure the development of PCL (property/casualty and liability) sector for each country. The growth of life insurance in force is used to measure the development of life insurance sector for each country. We start with a growth model which includes life insurance and PCL insurance as two factors which contribute to economic growth. Due to the fact that economic growth may also affect insurance growth, we used a three-equation simultaneous equation to run the data again and compared with the result from the simple OLS regression. At the end, we pool all countries data together and run a fixed effect model to compare with the above result. The fixed effect model allows for different constant terms for each country.

14

Chapter 2 Theoretical Framework

This section will present two models to illustrate the two important functions of insurance institutions: the risk sharing function and the capital accumulation function.

I. Property/Casualty/Liability Insurance and Economic Growth

i) The Risk Sharing Role of Property/Casualty and Liability Insurance in an Economy
One important contribution of insurance institutions to economic development is that such institutions can diversify the risks firms exposed to in an economy. Individual firms make production plans to maximize their profits. If they face uncertainty in their production, they may make production plans to maximize their expected profits. In general economics theory, individuals are risk averse. Therefore, individuals may not choose the social optimal production level when there is risk present in the production process. This will lead to underproduction problem. The emerging of insurance institutions can help firms to diversify the risks they face and therefore increase production level which is social maximum. Put it another way, the existence of the risk sharing institutions makes the firms undertake some investment they would not engage in if there were no such risk sharing institutions. The undertaking of such risk activities is beneficial to the whole society. As stated by Arrow (1965), a mans capacity for running a business will need not be accompanies by a desire or ability for bearing the accompanying risks, and a series of institutions for shifting risks has evolved (p. 221).

15 At any moment society is faced with a set of possible new projects which are on the average profitable though one cannot know for sure which particular projects will succeed and which will fail. If risks can not be shifted, then very possibly none of the projects will be undertaken; if they can be, then each individual investor, by diversification, can be fairly sure of a positive outcome, and society will be better off by the increase production. (p. 223) When there is uncertainty and individuals are risk averse, the social optimal choice is different from the individual optimal choice. This results in market inefficiency. The introduction of risk sharing institutions can improve market efficiency by allowing risk-averse individuals to make socially optimal production choices. In the following part, we will present a model which illustrates the above idea. Before we do that we will present a literature review of theoretical models that illustrate how financial institutions contribute to economic growth. We are borrowing some ideas from these models to construct our model. I found few specific models which address how insurance sector can contribute to economic growth. At the end of the review, two research studies that looked at the relationship between insurance and economic growth is presented at the last.

ii) Literature Review on the Economic Role of a Financial Sector

A lot of works have been done to examine the contribution of financial intermediations to economic growth. Theoretical models have been set up to explain the role of financial markets. Main streams of works can be divided into the following groups. First, financial institutions can reduce the demand for liquid savings of economic agents and increase investment on illiquid capital which bears higher rate of return. This will

16 increase economic growth. Second, financial institutions can ease transaction of ownership and therefore influence choice of technologies. Third, financial institutions are innovation supportive. The main works can be summarized as follows 1. Financial intermediaries are growth promoting through the capital accumulation channel. Benvicenga and Smith (1991) set up an endogenous growth model to explain the link between economic growth and the development of financial intermediaries. The model is a three period overlapping generation model. Economic agents can choose to invest in production of consumption good (liquid assets) or capital good (illiquid assets). Production of consumption good bears lower rate of return than production of capital good. However, the production of the latter requires two periods and the production of the former only requires one period. So investment in liquid assets has no loss in return if it is liquidated at period 2. On the other hand, investment in illiquid assets will result a loss if the assets are prematurely removed. The loss makes return on investment on illiquid assets lower than return on investment on liquid assets. Economic agents save all their first period income for second and/or third period consumption. The model is constructed such that there is a liquidity demand (Diamond and Dybvig (1983) preference structure), i.e. some agents of the whole population at the end of period one (beginning of period 2) will find they value period 3s consumption as nothing (there is no utility associated with this period consumption). The other agents will value the consumption of period 2s consumption and period 3s consumption the same degree. To maximize their utility, agents who do not value period 3s consumption wish they had invested all in

17 liquid assets because they want to consume all of them at period 2. Agents who do value period 3s consumption wish they had put all investment in illiquid assets because they bear higher rate of return. But they do not know this at the beginning of period 1, when they need to make decision on how much to invest on liquid assets and how much on illiquid assets. At the beginning of period 1, agents only know the possibility of whether their consumption in period 3 will bring them utility. So what they do is to maximize their expected utility. They choose to invest part of their period 1 income in liquid assets and part on illiquid assets. At the end of period 1, their preference type is revealed. Without financial intermediations, some agents find they do not value period 3s consumption. They will have to prematurely liquidate investment in capital good, which induce a loss of return of their investment and a loss in capital accumulation of the whole society. With financial intermediations, investment in capital goods will not be prematurely liquidated. Agents who do not value period 3s consumption will trade through the financial intermediaries with the agents who do value period 3s consumption. The capital stock of the whole society remains. The production function is constructed in such a way that there is an externality associated with capital accumulation. The production function of an individual firm not only depends on individual input in capital and labor, but it also depends on the total amount of capital accumulated in the society. Therefore, financial intermediaries are economic growth promoting by preserving the capital stock of an economy. Levine (1991) constructs an endogenous growth model to illustrate the function of stock market in risk allocation. In his model, physical capital accumulation can positively contribute to the creation of human capital (human capital is a function of total physical

18 capital stock in a firm). And human capital is a production function shifting parameter, which means the bigger the human capital stock, the higher the output. This links capital accumulation with economic growth. In his model, there are two risks, one is liquidity risk, and the other is the productivity risk. The liquidity risk is modeled similarly as in Benvicenga and Smith (1991), which also uses the preference structure created by Diamond and Dybvig (1983). The productivity risk is modeled by introducing a production shock into the production function. Due to the liquidity risk and productivity risk, risk-averse investors will not invest enough in the production, which reduce the accumulation of physical capital. Therefore, it also reduces human capital stock and affects economic growth. Without financial intermediations, liquidity demand will result in premature removal of physical capital. This will lower capital accumulation. The emerging of the stock market helps investors to trade their shares when the liquidity demand is revealed. This will increase the accumulation of physical capital. The other function of stock markets is that it allows investors to diversify away idiosyncratic productivity shocks by holding shares from different firms. This will increase the level of capital investment in firms. Therefore, stock markets are growth promoting. 2. Financial intermediaries are economic growth promoting through the technology choice channel. Saint-Paul (1992) developed a model to illustrate the substitution between financial diversification and technological diversification. The main idea of his paper is that some technologies are more flexible than others. The greater is the division of labor, the less flexible is the technology. Less flexible technologies are more productive due to the greater division of labor. However, there is higher risk associated with less flexible

19 technologies. When there is demand shock, it is hard for the less flexible technologies to change their production. Agents are risk averse. If there are no other institutions to diversify the risk, more flexible technology will be adopted by an economy. This means an economy will choose to use technologies with less productivity. The emergence of financial markets allows economic agents hedge against such risks by holding a diversified portfolio. So in an economy with well developed financial markets, less flexible (higher productivity) technologies will be chosen. On the other hand, in an economy with poor financial markets, more flexible (lower productivity) technologies will be chosen to diversify the risk. This will hinder productivity growth. In his model, a technology flexibility parameter is introduced into the production functions of two goods. Both goods are produced in two villages. However, the technologies in these two villages are just opposite of each other, i.e. one unit capital can produce the same amount of good 1 in village 1 and good 2 in village 2. The model is constructed so that only one good is demanded at the second period. At the beginning of the first period, the demand information is not known. So both goods will be produced. Entrepreneurs chose technologies to produce two goods and sell shares to the consumer. They will maximize the utility of shareholders. If there is no financial market, agents in each village can only buy shares of entrepreneurs in their own village. The result of the model is that more flexible technologies will be chosen. And the more risk averse are the economic agents, the more diversified (flexible) technologies will be employed. The conclusion is that without financial markets, some degree of technology flexibility is required. Then financial markets are implicitly introduced into the model, i.e. economic agents are allowed to buy shares of entrepreneurs in both villages. Still entrepreneurs maximize the

20 expected utility of shareholders. The result is that less flexible technologies (with higher productivity) will be chosen this time. The author also emphasizes the interaction effect between technologies diversification and financial diversification. Once more flexible technologies are chosen, there is not much motive for the development of an advanced financial system. The economy may be trapped in a lower equilibrium. His model therefore explains why developed financial markets are associated with developed real sector. On the other hand, we can also observe that underdeveloped financial markets are associated with underdeveloped real sector. Bencivenga and Smith (1995) set up a model to explain the connections between financial system and economic growth by emphasizing that choice of technologies is influenced by transaction costs of ownership. The main idea of their model is that the transaction costs of financial markets in an economy will determine the technologies adopted for this economy. High productivity technologies may require the investment on highly illiquid capital. Without an efficient financial system to facilitate transaction of ownerships, high productivity technologies can not be implemented even though they are available. Their model is a two period overlapping generations endogenous growth model with production. Their model allows the existence of different production technologies with different gestation period. The higher the productivity is the technology; the longer gestation period is this technology. Efficiency of financial markets is measured by transaction costs. Financial markets influence the economic growth rate through the following factors: technology chosen, transaction costs in financial markets, savings rate and the composition of savings. With more efficient financial markets, long gestation period technologies are more likely chosen because these technologies are more

21 transaction intensive. This will make economy favor those high productivity technologies when financial market is more efficient. This is growth promoting. At the same time, lower transaction cost will enhance the net productivity of all investments, which is also growth promoting. In their model, this will also increase savings rate and therefore increase growth rate. But the effect of the last factor, the composition of savings, may not favor economic growth when financial markets become more efficient. The reason is that lower transaction costs make capital resale markets more efficient too. Therefore the current technologies may not be replaced by more productive, long gestation period technologies. So financial market efficiency enhance may reduce growth rate if the effect of the fourth factor is large enough to counter the first three factors. 3. Financial intermediaries are economic growth promoting through supporting innovations. King and Levine (1993) construct an endogenous growth model to illustrate how financial intermediaries facilitate innovations in real sector and therefore promote economic growth. In their model, they emphasized four functions of financial markets: evaluation of investment projects, pooling of funds for desired projects, risk diversification, and valuation of expected rewards to innovative projects. In their model, entrepreneurship must be identified by outsider other than individuals themselves. Also, the identification (or rating) process is costly. This requires the existence of financial markets to do the job. The success of identified entrepreneurs to manage a project depends on chance. Risk is present in the production process. Risk diversification make implementing projects possible. In addition, an entrepreneur needs investment to initiate a project, whether he is successful or not. The investment is beyond his personal wealth

22 ability. This requires the fund pooling function of financial intermediaries. Projects with successful innovation will enjoy profit in the industry, with a mark up over its rivals costs. The cost of selection of entrepreneurs and profit from successful innovations will decide the equilibrium in the financial intermediation. The equilibrium in the production side sets up the relationship between growth rate and real rate of return, which is ambiguous. Combined with the preference side growth rate and real rate of return relationship, which is unambiguous positive relationship, a general equilibrium is decided. They illustrate that increasing returns in financial efficiency is like a reduction in tax, which is growth promoting. 4. Theoretical model that links insurance and economic growth. Carmichael and Dissou (2000) construct a model to state the growth promoting role of health insurance. Their model shows that the existence of health insurance allows individuals to insure against health risk. In their model, individuals face the possibility of disease on the second period of life, which will reduce their utility level if they are ill. They can spend money for cures to increase their utility on second period or choose to stay sick. Without health insurance, individuals will save an amount of money for treatment of disease for the possibility of catching the disease. This amount of money is liquid assets. This means part of their income can not be used to invest in illiquid assets, which carry a higher rate of return. In the model, the agents either choose to keep their liquid asset equal to treatment expense or keep no liquid asset. The treatment expense is assumed to be a known and fixed amount of money. Probability of sickness, disutility of non treatment of the disease, interest rate and health care expense are four factors that will influence the decision on liquid investment for treatment of future possible illness.

23 Higher probability of sickness and disutility of non treatment of disease will result in investment in liquid assets. Also, lower interest rate and lower treatment expense will lead to investment in liquid assets. In such a case (a liquid asset will be set aside for health care purpose), if there is market insurance, individuals will pay an actuarially fair insurance and not invest in liquid assets. The result is that the total savings of the society will decrease with the introduction of a health insurance sector. But the savings components are different when there is market insurance. Total investment on illiquid investment increases now, which yields higher rate of return. Therefore a health insurance sector is economic growth promoting by allocating more resource on productive sector, although total savings decrease. Soo (1996,) in his dissertation, researched the economic role of life insurance companies. He first developed a model to show that borrowing individuals and lending individuals use insurance to maximize life time utility. He examined the effect of insurance tax on aggregate consumption and savings. He followed his theoretical model with empirical study to show that there is support that life insurance can promote economic growth. Here we can see the basic idea running through most of the above models is that economic agents face several choices. Individuals will choose the one that maximize their own utility. Due to the fact that information is unknown when they make their choices, what is individually optimal is not socially optimal. With the introduction of financial intermediaries, individuals can make a better choice which will maximize their own utility and also achieve social optimal. This is the idea that also underlines our model.

24

iii) Theoretical Model Linking Insurance to Economic growth

1. Assumptions: There are two technologies: the safe one and the risky one. The production function for the safe technology is specified as follows. y = L1 * K The production function for the risky technology is specified as follows. y = AL1 * K with probability of

y = 0 with probability of 1
Where 0 p p 1 Also, the risky technology will have a higher expected output level than the safe technology for the same amount of input in labor and capital. To meet this requirement, we can specify that A f 1 . This infers that A f 1 too. Individual producer will be exposed to risk so that their production may either result in a high output or a zero output. But if all agents in an economy employ the risky technology, there are a proportion of

producers will not incur loss and there are a proportion of 1 producers will incur
loss because we specify as a number known. So for the whole economy there is no risk. This means that if all agents in the economy employ the risky technology, the total average output level will be y r = AL1 * K . If all the agents in the economy employ the safe technology, the total average output level will be y s = L1 * K , which is lower
than the expected output level when all the society employs the risky technology. It is obvious that for the whole economys sake, a risky technology with higher expected output level should be chosen. However, because individuals are exposed to risk of losing all their investment and incurring a loss, they may not choose to invest all in risky

25 technology when there is no insurance sector in the economic system. This leads to the under production for the whole economy. We need some further assumptions to develop the above result. First, we assume that the safe technology defines the competitive wage rate. So the wage rate is equal to the marginal product of labor by the safe technology, which is w = (1 ) L * K . So the return on L units of labor input to the production process is wL = (1 ) L1 * K . The return of capital depends on the technology an individual firm chooses. If the firm chooses to invest in safe technology, then the return on capital is rs = y s (1 ) L1 * K = L1 * K (1 ) L1 * K = * L1 * K . If the firm chooses to invest in risky technology, then the return on capital is
Pr ob(1 ) = y r (1 ) L1 K = 0 (1 ) L1 K = (1 ) L1 K rr 1 1 1 1 = y r (1 ) L K = AL K (1 ) L K = ( A 1 + ) L K Pr ob( )

The expected return on capital invested in risky technology is rr = y r (1 ) L * K = AL1 * K (1 ) L1 * K = ( A 1 + ) L1 * K . Although the expected return on risky technology is higher than the return on safe technology, individual firms may choose to invest in safe technology due to the possibility of capital loss. In the following, we will show that due to individuals risk averse, safe technology may be chosen, although it is good for the whole economy to choose the risky technology to engage in production. 2. The Model We specify an overlapping two generation model. Individuals are endowed with the same amount of labor at the first period of time. We assume labor is supplied

26 inelastically, which means all individuals will work in the first period and earn a market competitive wage, which is decided by the marginal product from the safe technology. Then they will consume part of their wage and invest the other part in either safe technology or in risky technology. Depending on the technology they choose, they will get a different return on their investment. Individuals will maximize their life time utility. We assume the individual utility function takes the following format, with a constant relative risk aversion. U= ~ (C1 + C 2 ) , where > 1 .

~ C 2 is a random variable which is contingent on the realized return on the investment on the second period. Due to the special format of the individual utility function, if safe technology will yield a positive return (we will specify this later), individuals will not value period ones consumption. They will save all to invest either in safe technology or risky technology. Therefore, capital accumulation is not a factor affecting the path of economic growth. Individuals desirability for risky technology is the factor that will affect the path of economic growth. This specification assumes constant risk aversion. The specification of the utility function follows Benvicenga and Smith (1991) and Levine (1991). But there is some difference with their models. They have a three period overlapping generation model. In their model first period of consumption does not go into the utility function. What we take from their model is that they suppress the time effect on different period of consumption. To put it another way, there is no internal and external discount rate. So ~ there is no coefficient before C 2 . This will make derivation of result much easier.

27 Define q1 as the proportion of wage an individual will consume in the first period and q2 as the proportion of wage an individual invest in safe technology. The proportion invested to the risky technology is 1- q1 - q2. Here we need to assume that 0 < q1 < 1 ,
0 < q2 < 1 , and 0 < q1 + q2 < 1 . Therefore the consumption contingent on the production

results will be

~ C2 = q2 *[ L1 K (1 ) L1 K ] (1 q1 q2 ) * (1 ) L1 * K = ( 1 + q1 + q2 q1 ) * ( L1 * K )
of 1 , when investment in risky technology incurs a loss; And

with probability

~ C2 = q2 * [ L1 K (1 ) L1 K ] + (1 q1 q2 ) * ( A * L1 * K (1 ) L1 * K ) = ( A + 1 + q1 + q2 q1 Aq1 Aq2 ) * ( L1 * K ) probability of , when investment in risky technology is successful.

with

The first item on the above equations shows the return on safe technology and the second item shows the return on risky technology. When individual firms incur loss, the output is zero. But firms still need to pay wage to labors employed. This is why the second item in the first equation is negative. An individual will maximize his/her expected utility, which is to maximize Exp[U = ~ s (C1 + C 2 + C r 2 ) ] by choosing q1 and q2 .

Here C1 is the first period consumption, C2s is the consumption from return of

~ investment in safe technology and C 2r is the consumption from return of investment in risky technology, which can be positive or negative contingent on the probability of production output.

28 We assume that each individual in the initial old generation is endowed with K units of capital. At the beginning there is only safe technology. The old generation will invest only in safe technology, which determines how much they could consume in that period. They will employ the young generation each of whom we assume will be endowed with L unit of labor at the beginning of the first period. Each individual in the young generation will supply all L units of labor. Production is achieved through combining the capital of the old generation with the labor of the young generation. We further assume that L1 K > K to guarantee a positive return of investment on the safe technology. Individuals in the young generation will get a wage rate equal to the marginal product of labor, which is w = (1 ) L * K . Here we assume is a positive number and is smaller than 1. Individuals of the young generation will choose to allocate their wage among consumption during the first period, save for investment in the safe technology or save for investment in the risky technology. At the second period, each individual only invest and does not supply labor to earn wage anymore. Their consumptions at the second period totally rely on their return on investment on different technologies they choose. So the total amount of capital invested is equal to w C1 = w q1 w = (1 q1 ) * w , which will be invested in safe or risky technology. Here w is a function of L and K, which are constant numbers. We will keep w in the following derivations. Whether we express w in terms of L or K or not will not influence our derivation result. Depending on the allocation of the capital invested in safe or risky technology, an individual will get different return on their capital investment. (1 q1 ) * w = q 2 * w + (1 q1 q 2 ) * w . The left hand side is the total capital investment. The first item of the right hand side is the

29 investment in the safe technology and the second item is the investment in the risky technology. So the maximization problem becomes maximize the expected utility of an individual by choosing q1 , which decides the proportion of the wage consumed in the first period, q2 , which decides the proportion of wage invested in the safe and risky technologies. The proportion of the wage invested in risky technology, which is equal to (1 q1 q 2 ) , is known after q1 and q2 are chosen. In the following part, we will prove that q2 is not equal to zero if an expected utility maximization individual has no way to hedge the production risk. This means that individuals will choose at least part of their wealth to invest in safe but low return technology when there is no insurance sector 5. Individuals will choose q1 and q2 to ~ (q 1 *w + [q 2 * L1 * w ] + C r 2 ) Max E (1 q1 q 2 )[ A * L(1 ) * w (1 ) * L1 * w ] ~ Here C 2r = { (1 q1 q 2 )[0 (1 ) * L1 * w ] After combining similar terms, we get (1 q1 q 2 )( A 1 + ) * L(1 ) ) * w ~ C 2r = { (1 q1 q 2 )( 1) * L1 * w The expected utility therefore is 6
prob = prob = 1 prob = prob = 1

Here we assume that there is no population growth. Each individual in the old generation will form a firm. Each firm will combine his/her capital investment with the L units of labor of the young generation to carry out production. So individual and firm is the same concept here. We will use these two terms alternately. 6 Note if we have A and as constant numbers and B is a random variable, then E (

A + B 0 A + B1 B Pr ob( ) A + B ) =( ) * + ( ) * (1 ) . Here B = 0 . B1 Pr ob(1 )

30
~ (q 1 *w + [q 2 * L1 * w ] + C r 2 ) E = + [q1 w + q 2L1 w + (1 q1 q 2 )( A 1 + ) L1 w ] * [q1 w + q 2L1 w + (1 q1 q 2 )( 1) L1 w ] * (1 )

We need to choose q1 and q2 to maximize the expected utility of an individual. First we will assume that capital investment in safe technology will give a positive return. This means rs = y s (1 ) L1 * K = L1 * K (1 ) L1 * K = * L1 * K > K . To meet this requirement, we need to assume that * L1 K > K . Due to the special form of the utility function, we can induce that q1 = 0 if the expected utility is maximized. So we will choose q 2 conditional on q1 = 0 to maximize the expected utility. Take first derivative and set it equal to zero, we get
1 +1 ( D 1) , where D = ( ) . Here D can be viewed as a term represents q2 = 1 A D + A 1
1

the productivity difference between safe technology and risky technology. D > 1 and 0< ( D 1) < 1 . Therefore 0 < q 2 < 1 (see the appendix for the solving process and the D + A 1

discussions). It means that individuals will not choose to invest all in risky technology although it has a higher expected product output. So the whole economy faces an underproduction problem. Now lets assume insuring against risk is possible and individuals pay an actuarially fair premium. When production goes well, the individual firm will have an output level of y = AL1 * K . The return on a successful capital investment in risky technology is

31
r = AL1 * K (1 ) L1 * K premium = ( A 1 + ) L1 K premium . When

production incurs loss, insurance company will compensate for the loss. The compensation is y = AL1 * K . So an insured individual, who incurred loss in his /her investment in risky technology, will have a return on capital investment of
r = AL1 * K (1 ) L1 * K premium too.

Now, lets decide how large a premium the insurance sector needs to collect. When we assume an actuarial fair premium, it means that insurance sector will make zero profit. This means the premium collected is equal to the cost paid to individual firms who will incur loss. To calculate the actuarial fair premium, we assume there are n firms in the economy. For the whole economy, there are n firms which will not incur loss in their production process. There are n(1 ) firms which will incur loss. For each firm incurred loss, the insurance company needs to pay them AL1 * K . Therefore the total cost to the insurance company is n(1 ) AL1 * K . The total premium that the insurance company needs to collect is therefore equal to n(1 ) * AL1 * K . Because there are n firms in the economy, the premium collected from each firm by the insurance company is therefore equal to (1 ) * AL1 * K . For proportion of firms, there is no loss and insurance sector does not need to pay anything. For 1 of firms, they need to pay y = AL1 * K for each. At the end, all premiums will be paid out. Insurance sector does not make any profit. In addition, each firm will face the same output level, whether they incur loss or not in their production process. The output level net of premium cost of each insured firm is

32 AL1 * K (1 ) * AL1 * K = AL1 * K , which is same as the expected output level when a firm employs the risky technology. Because we assume A f 1 , an individual firm will definitely choose to invest all in the risky technology when there is an insurance sector. With insurance possible, a risky technology which yields a higher return will dominate the economy. Due to the special feature of the utility function, the individual maximization problem will have a corner solution-- q1 and q2 will be zero: nobody will choose to invest in safe technology. The maximized utility with an insurance sector is higher than the maximized utility without an insurance sector. Insurance sector makes the whole economy on the path of choosing the risky technology, which benefits all agents in the whole economy. The under production problem is solved. Further, we can link this with economic growth. From the production functions of two different technologies, we can see the safe technology result in a lower output level and the risky technology result in a higher output level if it is employed by the whole economy. In the following discussion, we will see how this will relate to economic growth.

33

II. The Long Term Capital Accumulation Role of Insurance in an Economy

In this part, we will discuss how insurance institutions affect the economic growth through the channel of capital accumulation. We will especially concentrate on life insurance and pension funds because they are viewed as institutions that will influence the savings pattern of an economy. Health insurance may also influence the pattern of individual savings. This is illustrated in the model of Carmichael and Dissou (2000). Their model shows that the introduction of health insurance into an economy may not change the total amount of savings. However, it will increase the proportion of long term capital. Therefore, health insurance can contribute to the growth of an economy. In the following discussions, we will see pension funds and life insurance also can shift part of short term savings to long term savings. This will help economic growth. Pension funds and life insurance companies are called contractual savings institutions. The emergence of contractual savings institutions may affect both the total amount of savings and the savings pattern of individuals. When we say individuals savings patterns change, we mean the structure between liquid savings (used for short term investment) and illiquid savings (used for long term investment) changes. Therefore, the emergence of contractual savings institutions may affect economic growth through the capital accumulation channel. The arguments go as follows. Compared with banks, contractual savings institutions have special knowledge on population life expanse distribution. This makes the contractual savings institutions able to offer individuals savings plans. These savings

34 plans can increase individuals utility level. Put it another way, there is a demand for contractual savings institutions. This leads to the emergence of contractual savings institutions. Compared with banks, contractual savings institutions can offer more long run funds. This means contractual savings institutions can provide more illiquid funds to an economy, which is believed to contribute to economic growth. First, we will go through some empirical studies on the effect of contractual savings institutions. Then we will present the above argument. The empirical study of Murphy and Musalem (2004) from 43 industrial and developing countries has shown mandatory pension programs may increase national savings while voluntary pension programs may not affect national savings. The role of contractual savings institution on economic growth may be through deepening of capital market. Some empirical studies have been done to show whether there is relationship between development of contractual savings institutions and financial market development. The empirical study of Impavido, Musalem, and Tressel (2003) found that an increase in the ratio of the assets of contractual savings institutions to domestic financial assets will increase the depth of domestic stock market and bond markets on average. The specific effect is dependent on factors such as financial system of a country (whether it is market based or not), mandatory or voluntary contribution of pension funds. Another empirical study of Impavido, Musalem, and Tressel (2002) is about how the development of contractual savings institutions affect bank sector. Their study showed that the development of contractual savings institutions is positively related to bank efficiency and resilience to credit and liquidity risks.

35 They also examined how the development of contractual savings institutions affects the firm financing patterns, which may lead to the efficiency gain in firms. Impavido and Musalem (1999) conducted an empirical analysis using OECD and developing countries data to show how contractual savings institutions affect the development of stock markets. They concluded that development in contractual savings institutions is contributing to the development of stock markets in its depth and volatility. In summary contractual savings institutions may hold a larger proportion of illiquid assets because they have long term liabilities. Contractual savings institutions are more stable to the economy. They may not face runs like banks do. Contractual savings institutions include pension funds and life insurance companies. The emergence of contractual savings institutions may not increase the total amount of savings but they will increase the amount of illiquid savings, which are essential to economic growth. Banks and open-end mutual funds have short term liabilities while pension funds and life insurance companies have long term liabilities. Banks will finance more investment with short maturity while contractual savings institutions will finance more investment with long maturity. According to the author, the percentage of long term investment for contractual savings institutions is much higher than that of banks. However, we should be very careful about the above conclusion on contractual savings institutions. Although the liabilities that contractual savings hold are mostly in long term, this does not mean they will have long term investment. At least, this may not be the case for all countries. Different countries have different laws governing the operation of life insurance, which may regulate what kind of projects the life insurance

36 companies can invest in. But for the theoretical illustration purpose, we will just assume that life insurance can freely invest in long term projects to achieve higher profit. The following part will provide a theoretical framework to show the effect of contractual savings institutions on accumulation of capital and on economic development. First, we will show why individuals are willing to put their money in the contractual savings institutions.

i) Individual Savings Behaviors without Contractual Savings Institutions

We assume each individual will work at the first period and earn wage. Labor supply is inelastic. Each individual will face the probability of death at either end of period1, 2, or 3. But the specific life expanse for each individual is not known. Each individual needs to choose a savings plan for his whole life. He will not work in period 2 and period 3. So they need to save period 1s wage for possibility of period 2s and 3s consumption. There are no contractual savings institutions. They will put their money in banks and get a market rate of return. We assume banks offer per period interest rate of r. Now we assume a specific form of utility function for each individual. Because individuals do not want to end up with nothing when they were old, they must plan for their longest possible life span. Also, to make things simple, we assume each individuals internal rate of return is the same as the market rate of interest. Particularly, we specify the utility function as follows.
U (C1 , C 2 , C 3 ) = C1 * C2 L 1+ r * C3 L (1 + r ) 2
C3 C2 + 1 + r (1 + r ) 2

Subject to the constraint that w = C1 +

37 In the utility function, (1+r) represent the internal rate of preference of individuals, which is the same as the rate banks will offer individuals if they deposit their money in banks. The second periods and third periods consumptions are discounted to current value by individuals internal rate of preference in the utility function. In the constraint, it shows that current wage is equal to the sum of current consumption and future consumptions discounted by the interest rate that bank offers. The reason for the special form of the utility function is that it rules out the situation when individuals are alive but have nothing to consume. Notice here the utility function is the product of consumptions in three periods, given an individual is alive. The term C 2 L means the consumption in period 2 given the individual is alive at period 2. Similarly, C 3 L means the consumption in period 3 given the individual is alive at period 3. If the individual is not alive at period 2 or period 3, the utility will just be the product of all previous consumption(s). If an individual is alive and has nothing to consume for any period, the total utility of an individual will be zero. Because there is a possibility that an individual will live to the end of period 3, he has to save for that even if he also has a possibility that he will not live that long. Individuals do not know their specific life expanse. So they have to plan for the longest life expanse to avoid dying from hunger before their natural death. The above utility function is not very weird if we recall the following two-period model in most microeconomics textbooks.

38

C1

C2 The convex shape of the utility function shows an individuals utility is the product of the consumption of two periods. The only difference in our model is that the utility is conditional on whether an individual will be alive in the second and third periods. Also, we will illustrate the probability distribution for the death of each individual in Table 2.2.1. Particularly, we will specify a symmetric distribution although it may not be matching the true world situation 7.
Table 2.2.1 Assumed Life Expectancy Probability Distribution Time of Death End of Period 1 End of Period 2 End of Period 3 Probability of Death for Each Individual

1-2

The format of this particular distribution function is just for the convenience of derivation of the final result. It will not affect the theoretical result.

39 Due to the special form of the utility function, we can see if an individual will maximize expected utility function, he must choose to make real consumption same at each period. If the wage he earns during the first period is w, then he will plan to consume w/3 in real term (after discounted by the internal rate of preference) during period 1, 2 and 3. The above situation can be illustrated by a numerical example. Suppose an individual will earn 150 units during the first period. What he will do is to consume 50 units during the first period for sure. Then he will save the other 100 units which will earn per period interest rate of r. If he does not die at the end of period 1, he will draw out 50(1+r) units at period 2 and consume it. Then he leaves 50(1+r) units in the bank. If he does not die at the end of period of 2, he will draw all the money left, which is 50(1+r)2 , and consume it. So the utility of individuals who will die at the end of period 1 is 50; the utility of individuals who will die at the end of period 2 is 50*50; the utility of individuals who will die at the end of period 3 is 50*50*50. This is under the situation when there are no contractual savings institutions. If there are contractual savings institutions which allow individuals to buy policy and draw an annuity as long as the individual is alive, the situation will be different.

ii) Individual Savings Behaviors with Contractual Savings Institutions

Obviously, individuals utility can be improved when there are contractual savings institutions. What individuals will do is that they will put their entire wage in the institutions and exchange for a life long annuity. If we assume these institutions offer actuarial fair annuity (they will make zero profit), individuals will consume more during

40 each period. Of course, we need to assume that the probability distribution is known to the contractual savings institutions, which is not known to banks. This assumption is very reasonable because life insurance companies or pension funds can hire actuaries to calculate the life table while this is not the case for banks. Then we see how contractual savings institutions will manage. Due to the symmetric distribution of the probability function, contractual savings institutions can afford an annuity with an annual payment of w/2, which is higher than individuals previously planned consumption. The reason is that the institutions can use the money of the early death to finance those who live longer than average. The expected life span is two periods. Table 2.2.2 summarizes the inflow and outflow of funds of the contractual savings institutions.
Table 2.2.2 Inflow and Outflow of Funds (Pension) Inflow of Funds Period 1 Period 2 Period 3 Outflow of Funds

wn

(w/2)n (w/2)n(1-2 ) (w/2)n (w/2)n(1-2 )(1+r) (w/2)n (1+r) (w/2)n (1+r)2

Note: we assume there are n individuals who earn wages in the economy. Also we assume the fund in the contractual savings institutions will accrue with an interest rate of r per period. Notice here the present value of the outflow of fund is equal to wn.

Using the above numerical example, we can see each individual can have a real consumption of 75 units per period as long as he is alive. The utility for the individual who will die at the end of period 1 is 75; the utility of individuals who will die at the end of period 2 is 75*75; the utility of individuals who will die at the end of period 3 is

41 75*75*75. All individuals can improve their utility. The role of the contractual savings institutions is to optimize individuals plan which could not be achieved otherwise due to individuals uncertainty of their death dates. So if there are contractual savings institutions, individuals will put all their savings to those institutions rather than in banks. The reason is that the existence of contractual savings institutions can increase individuals utility level. This is a rather simplified illustration. Please note the above argument does not rule out the existence of banks. In the real world situation, we can assume that there are two kinds of savings: one is for life long purpose; the other is for routine daily purpose (We can view the role of the bank as providing liquidity to individuals inside each living period of above model). The emergence of contractual savings institutions is for the first purpose. Savings transfer from banks to contractual savings institutions does not mean banks are of no use anymore. It only means the savings for life long purpose are transferred from banks to contractual savings institutions. The other role of banks is not modeled here. The difference between banks and contractual savings is that individuals can not draw funds except when they are alive. Banks can not do that because they do not know the life expanse distribution of the population. Therefore, funds in contractual savings institutions can engage more in long term production. This engagement may not be direct. It may be through the deepening of the stock markets, which is also ignored here.

42

iii) Compositions of Savings with and without Contractual Savings Institutions

We can see from the above illustration that the emergence of contractual savings institutions will increase long term investment. Suppose individuals will choose to invest in either short term (one period) or long term investment (two periods). If there are no contractual savings institutions, they will invest 1/3 of their wage in short term and 1/3 in long term because they planned for three periods consumption. They will withdraw at the second period and the third period if they are alive. We assume the fund will be withdrawn early if the person dies earlier. The long term investment will be preliquidated. The pre-liquidated long term fund will be counted as short term fund. The total amount of long term investment when there is no contractual savings institutions (which does not include those interrupted due to their death) is (1/3)*w**n. The total amount of short term investment is (1/3)*w**n +(1/3)*w*(1-2)*n. When there are contractual savings institutions, we can see the contractual savings institutions will increase long term and decrease short term capital. This means there is more capital invested in the production process, which may be a source of growth. We can see the contractual savings institutions can invest (1/2)*w**n in the long term production and (1/2)*w**n + (1/2)*w*(1-2)*n in the short term investment. The above discussion is based upon the case which is similar to pension funds. The assumption is that people will draw funds as long as they are alive. Payments stop at death. This is just opposite to the case of life insurance. Life insurance pays to beneficiaries upon policy holders death and does not make any payment when the policy holder is alive. However, the mathematical derivation wont be much different. We can

43 just assume that the original utility function will have a similar form. But now the second period and third period consumption is based on the consumption of the beneficiaries upon the death of the policy holder. The purpose of life insurance is to supply more financial need upon death of policy holder. We may have the following format of outflow of funds in Table 2.2.3.
Table 2.2.3 Inflow and Outflow of Funds (Life Insurance) Inflow of Funds Period 1 Period 2 Period 3 Outflow of Funds

wn

wn wn(1-2 )(1+r) wn (1+r)2

We can assume the savings due to uncertainty of death is kept liquid all the time since the unknown date of death when there is no life insurance available. Therefore, when there is life insurance available, both long term and short term funds increase.

44

III. Discussion: Technological Progress, Insurance and Economic Growth

i) Factors that Affect Technological Progress
In most growth models such as Solows (1956) growth model, economic growth is sustained by technological growth in the long run. Capital investment alone cannot generate permanent economic growth. However, in those models, technological growth is treated as an exogenous rate. How new technology is invented is not illustrated in those models. Recently, more and more economists realized the importance of technological progress and the factors that determine the progress. Shumpeter was among the earliest who gives inspiration of later economists the idea of creative destruction. Basically, capitalism is in the process of generating new products, ideas, etc. to replace the old ones. Firms are in the process of competition for innovations and gain profit from these innovations. Based on the ideas of Shumperian model (1934), many economists have developed models that make technology progress endogenous variable. Among them are Philippe Aghion and Peter Howitt (1992), Gene Grossman and Elhanan helpman (1991), and Paul Romer (1990). In these models, technological progress can be explained by economic incentives rather than being given exogenously. According to the description above, we can see that economists have two different ways to model technological progress. One way is to treat technological progress as an unintentionally by-product, an externality, of some other activities such as investment and or production. (Van den Berg , 2009 forthcoming, p. 37 of chapter 6). However, the other way is to assume that

45 technological progress is motivated by economic incentives to purposely allocate resource in generating new technology. In his development textbook, Van den Berg summarized models which make technological growth an endogenous variable. He also gives out a mathematical version of the Shumpeterian model to show how entrepreneurs are motivated to innovate. This is basically based on Paul Romers model (1990). In the following part, the basic idea of this model is summarized as well as the factors that determine technological progress. In the Shumepeterian technological growth model, we can see that innovations take time and effort. This makes technological invention very costly. Although the production cost may not be high once the new invention is successful (usually new production method will lower production cost), the upfront cost of making that invention is huge. Therefore, unlike traditional producers who face a horizontal demand curve and can not influence market price in a competitive market, entrepreneurs who engage in innovations face a downward demand curve. This makes entrepreneurs charge a higher price than average cost to make a profit, which can recover the huge upfront cost of innovations. The basic idea of the model can be described as follows. Entrepreneurs engage in costly innovation activities, which cost scarce resources, to seek the profit coming from the new products from successful innovations, also at the expectation that the new product can be further replaced by more advanced technology (the creative destruction process). Then the question is that how many resources entrepreneurs will engage in innovation activities. Entrepreneurs need to compare the cost of innovation activities and

46 the expected profit from the innovations to decide how many resources should be allocated to the innovation activities. The following derivations (Van den Berg, 2009) are illustrated here to show how the equilibrium rate of innovation is decided. This is a close version of Paul Romers model (1990). Here I will just give out a very brief summarization of the derivation and the final result. The model assumes that firms continually create new products and the growth rate of new products is g. The model is to determine the equilibrium growth rate of new products, which is technological progress. The model assumes n firms in the economy, each one will produce a different product. Each firm faces a down-ward sloping demand curve, which is the same across different firms; therefore they can charge a price higher than the marginal cost of production (w) to earn a profit. It is assumed that one unit labor is required to produce one unit of good. Therefore, the cost of the product is w, which is the wage rate. is the ratio of price that is equal to production cost of one unit of product. Therefore, 1- is the price mark up that will give firm profit above marginal cost. The profit of each firm is equal to total output of the economy multiplying price mark up then being divided by n. The present value of all the future profit of a firm decides the equity value of the firm. According to the above assumptions, we have the following equations. p= w

Demand function of each firm

= GDP(1 ) / n Profit of each firm

47 PV = GDP(1 ) / n(r + g ) The present value of a firm when the capital market
is competitive. r is the market interest rate and g is the growth rate of the firm, which is the technological progress of the firm. Firms will engage in innovations till the present value of all future profit equals the cost of innovation. Assume that it takes unit of labor to develop a new product. The cost of innovation is w. PV= w. Equilibrium of innovations growth. From the above equations, we have the following relationship

p=

PV

GDP(1 ) / n(r + g )

GDP(1 ) n (r + g )

On the other hand, the entrepreneurs decide how to allocate the total resources R to current product and innovation of new product. This gives

ng + GDP / p = R
Combine the above two equations, we have

GDP GDP(1 ) = R ng n (r + g )
Therefore,

g = [ R(1 ) / n] r
Here we can see that the growth rate of new products is a function of R (total resources), (units of resources taken for each innovation), (the ratio of price which will cover the cost of production), n (the number of firms in the economy) and r (interest rate).

48 Any change in the above parameters will lead to the change of technological progress g. This corresponding effect can be shown in the sign of the partial derivatives of g against the above variables.
g = (1 ) / n R g = R / n r g = r

g [ R(1 )] = 2n
g [ R(1 )] = n n 2 In the later chapter of his textbook, Van den Berg (2009) further discussed the role of savings in technological progress. In neoclassical Solow growth model, higher savings rate will lead to a higher steady state income level. In such kind of models, savings will automatically turn into investment. Two things we need to consider here. First, savings may not turn into investment automatically. Second, even though savings can be turned into capital investment, it can not sustain long run economic growth. Only savings channeled to investment that leads to technological progress will finally result in economic growth. Economic growth needs ongoing shift out of production function which requires an ongoing expanding of capital investment. In the Shumpeterian model, savings effectively turning to finance technological progress is a key role of promoting economic progress. More savings channeled to investment can lead to higher steady state income level. This is corresponding to a move on the production frontier. At the same

49 time, more savings channeled to technological progress is corresponding to a shift out of production function. This leads to further more output in the economy and more savings in the future. Therefore, the role of financial intermediation is critical in the economic development. The financial sector of an economy plays the role of channeling savings to potential investment projects. Innovations usually require huge upfront cost with the expectation of future profit. Therefore, without funds being channeled to entrepreneurs, technological progress seems not possible. A model of financial intermediation and technological progress developed by Robert King and Ross Levine (1993) shows that the cost of financial intermediation is negatively related with technological progress. In their model, the main cost of financial intermediation is to acquire information on promising entrepreneurs. Here we will just present the derivation result from their model. In summary, technological progress is driven by many factors which determine the cost and profit of innovations and further determine the equilibrium innovation rate. At the same time, innovations also require an efficient financial sector to bring in savings. In the following part, we will see how the insurance sector, as part of the financial sector of an economy, can influence technological progress.

ii) Insurance and Technological Progress

1. The Role of Property/Liability Insurance In the beginning of this chapter, we illustrated that the presence of property/liability insurance make economic agents in an economy to choose a risky technology rather than a safe technology. We can view the risky technology as the

50 innovated technology and the safe technology as the current technology. The insurance sector here helps the economy to forsake the old technology and apply the new technology. It opens the door for the new technology. Without such institutions to spread out risk, economic agents are reluctant to take the new technology even though it is already innovated. For example, compared with possessing horse driven buggies, possessing automobiles may be subject to more risks due to the possibility of the loss of the automobiles, although it may bring more productive economic life. Presence of the insurance sector increases the demand for risky technology. Entrepreneurs get a higher profit ratio due to more demand of new technology. This is corresponding to a lower . We have < 0 . From the above mathematical derivation, we have pc lg

g g g = R / n r < 0 . Therefore, = * > 0 , PCL insurance will lead to a pc lg pc lg higher g when it can expand demand for new technology from the consumption side. On the other hand, property insurance on buildings and machines etc. will lower the risk of property loss that entrepreneurs face and therefore decrease their expected cost. This will be corresponding to a decrease of , which is < 0 From the above mathematical pc lg

derivation, we have

g [ R(1 )] g g < 0 . Therefore, = * > 0 , PCL = 2 pc lg pc lg n

insurance will lead to a higher g when it can reduce the cost associated with the producing of new technology. The combination of the two results due to the presence of an insurance sector will reinforce each other. The total effect is corresponding to an increase of g, the technological progress growth rate.

51 2. The Role of the Life Insurance Sector in Technological Progress We have illustrated that life insurance companies can provide more long term funds into the economy. Innovative activities usually demand for huge upfront cost. Also, it may take a long time to have new innovations. Therefore, there is the need of long term funds for innovation activities. Here we can see that more long term funds may increase the total resource available for innovation activities. We can view this as an increase as total available resources devoted to innovation. This is corresponding to an increase of R, which is R g > 0 . From the mathematical derivation, we have = (1 ) / n > 0 . lfg R g g R = > 0 . Life insurance growth may lead to the technological lfg R lfg

Therefore, progress.

Also, we can illustrate this from the perspective of the cost of financial intermediation. In King and Levines (1993) model, lower cost of financial sector will increase technological progress. Although in their model the cost is associated with acquiring information, we can generalize this cost as the general cost of financial intermediation. More long term funds in the life insurance sector can make their provision of necessary long term to entrepreneurs less costly. This will increase technological growth rate. By the above discussion, we can see that the insurance sector may serve as the assisting institutions of technological progress. These are the possible reasons how insurance can contribute to economic growth through promoting technological progress. However, in the real world situation, there are more costs involved with these institutions as well as their cooperation with other similar institutions. The final effect will rely on the

52 situation of specific countries. In the following section, we will investigate whether insurance will contribute to economic growth through the study of four countries data.

53

Chapter 3 Empirical Study of Insurance Growth and Economic Growth

In the last section, we built up the theoretical framework to explain how the insurance sector may contribute to the growth of an economy. In this section, we have collected several countries data that are available to us. We study the data to see whether they give empirical evidence of the theoretical models we developed in the first section of the dissertation. We emphasize the role of the insurance sector in an economy, which is how the insurance sector can help the growth of an economy. However, there is no doubt that economic growth can also help the development of the insurance sector in an economy. Therefore, in the empirical analysis, we allow the interaction between economic growth and insurance growth. The following is a literature review on the empirical studies on the relationship between economic growth and insurance growth. Actually, what we find here is that almost all the studies focus on the factors that affect the development of the insurance sector in an economy. These literatures help us to set up the growth demand function of either life or non life insurance in an economy.

I. Literature Review
One empirical study of the relationship between economic growth and insurance is by Beenstock, Dickinson and Khajuria (1988). They used the international propertyliability insurance premiums and income data. They found that the marginal propensity of insurance is different across countries and property-liability insurance is a superior good and is disproportionately represented in economic growth (pp. 270)

54 Ward and Zurbruegg (2000) used the data of nine OECD countries from 1961 and 1996 to examine the interaction of economic growth and insurance industry development. They used the VAR to test for the Granger causality between economic growth and insurance development. Their results showed that the causality relationship is different across countries. Only three countries have notable causality test results found: Canada, Japan and Italy. In Canada and Japan, insurance seems Granger cause growth, while in Italy there is bi-causal relationship. Browne and Kim (1993) used international data to analyze the factors that may influence the demand of life insurance in an economy. They include the following factors: income, social security, inflation, education, life expectancy, price of insurance and whether the data are from an Islamic Nation. They found that the following factors are statistical significant in life insurance demand function---dependency ratio, national income, government spending on social security, inflation and the price of insurance and whether a country is an Islamic nation. Outreville (1996) studied the role of life insurance in the financial sector of developing countries. He found that the development of the life insurance market in developing countries was highly related with the personal disposable income and the level of financial development. He also found that inflation expectation is negatively related with life insurance premiums. Liu, et al. (2003) found that urbanization in China is a significant source of the increase in health insurance coverage in rural China. They used an urbanization index to measure the degree of urbanization. The index includes the following variables:

55 population size, infrastructure variables and industrialization variables. They also found that income is positively related with the health insurance coverage. Enz (2000) used the international data to examine the relationship between insurance penetration (premium/GDP) and per capita income. He found that the income elasticity of insurance purchase is not linear but presenting an S-curve shape. The data have shown that penetration of life insurance across countries is more diversified than that of non-life insurance. The income elasticity of insurance is around one for high and low level of income and is near two or more for intermediate income levels. The income elasticity is highest around $10000 for non-life and $15000 for life insurance. From the above literature review, we can see that there is evidence that economic development will affect insurance development. Second, development in the insurance sector is different from country to country. Third, there is a difference between life insurance development and non life insurance development. In the following section, we will present the result from four countries data.

56

II. Data and Methodology

i) Data
The data for this empirical study are from four countries: The U.S., Korea, Sweden, and Germany. These four countries data are available from our library resources. For the empirical study, the more countries are included in the study, the better. However, these are all the data on hand for me to do the time series analysis when I started this dissertation. Not a lot of countries provided very good data on their insurance sector. That is part of the reason that there is lack of literature on insurance study. The real GDP is used to measure the economic growth. For property/casualty/liability (PCL in the following part throughout this dissertation or non life) insurance, real written premiums are used as a proxy for the development of PCL insurance sector. Generally speaking, PCL insurance policies are short term policies. Most of the policies are renewed in half or one year. Therefore, the time when premiums are paid is very close to the time period when the risk is under coverage. So the usage of real written premiums for the PCL insurance sector is a good measurement. For the analysis of life insurance growth, there are three choices to measure its growthlife insurance written premiums, life insurance policy reserve and life insurance in force. The following are the definitions of the four terms. 1. Premium: The payment, or one of the periodic payments, that a policyholder makes to own an insurance policy or annuity. 2. Reserve: The amount required to be carried as a liability on an insurers financial statement to provide for future commitments under policies outstanding.

57 3. Life Insurance in Force: The sum of face amounts and dividend additions of life insurance policies outstanding at a given time. Additional amounts payable under accidental death or other special provisions are excluded. 4. Face Amount: The amount stated on the face of a life insurance policy that will be paid upon death or policy maturity. The amount excludes dividend additions or additional amounts payable under accidental death or other special provisions.
(The above definitions are cited from the website of the American Council of Life Insurers)

Most life insurance policies are in much longer term, some of them are in whole life of the policy holder. The policy holders pay periodicallymonthly (most of the time), quarterly, bi-annually or annually to own a life insurance policy. Obviously, the time when the premiums are paid is quite different with the time period when the risk is under coverage. Therefore it is not a good measure of life insurance growth. Life insurance in force and life insurance policy reserve are correlated with each other. We have to decide which one to use at last. Actuaries use life table and other techniques to estimate life insurance policy reserve, which shows their estimate of future claims on the all current life insurance policy in force. I chose life insurance in force rather than life insurance policy reserve as a measure of life insurance growth for two reasons. First, life insurance policy holders will make their economic decision according to the coverage amount on contingency. Second, life insurance in force is a better independent variable for a life insurance demand function. For the U.S., Germany and Sweden, the data are from their statistical yearbooks, Penn World Table and IMF financial data disk. The data of Korea are from the website of the Korean Bureau of Statistics. The U.S. data are from the Statistical Abstract of the

58 United States. The range of the PCL data available to us is from 1963 to 2000. The range of the life insurance data available for us is from 1956 to 2001. Therefore, we will use the data from 1963 to 2000. The data of Germany are from 1970 to 1999. The data of Sweden are from 1972 to 1992. The data of Korea are from the following website: http://www.kosis.kr/eng/index.htm. Both the ranges of the life insurance and property/casualty insurance are from 1980 to 2004. Premiums collected and life insurance in force both are in nominal terms. They are discounted by GDP deflator of each country to get real premiums. The real GDP data are from the publication of IMF. At the end of the dissertation, a detailed source of each variable is described.

ii) Methodology
Simultaneous equations are used to allow the interaction of GDP growth and insurance growth. An ordinary least square model will be run first. We then compare the result from OLS with the results of the simultaneous equation model. It is well known that non stationary of a variable may lead to the spurious result of regression. So the unit root of each variable is tested. If a variable is suspected of being non stationary, the first difference of the data will be used (given the first difference of the data is stationary). The results will be presented individually for each country because there is some difference in the inclusion of variables for different countries (due to data availability)

59 1. Unit Root Test The unit root test will tell us whether a time-series variable is stationary or nonstationary, which will decide the appropriate form of the regression model. Two tests will be used: the augmented Dickey-Fuller test and the Phillips-Perron test. All the illustration of the methodologies in the following part is from Enders (1995). The augmented Dickey-Fuller test assumes a time series y follows a pth-order autoregressive process as follows. y t = a 0 + a1 y t 1 + a 2 yt 2 + ...... + a p y t p + t Rearrange the above equation; we can get the following equivalent equation:
y t = a 0 + y t 1 + i y t i +1 + t
i =2 p

Where

= (1 ai )
i =1

i = a j
j =i

We are interested in testing the coefficient of . If it is 0, then there is a unit root in the time series y. The test statistics is based on the Dickey Fuller statistics which is different with the traditional t test. One problem with the Dickey-Fuller test is that it assumes that the errors are independent and have a constant variance, which may not be the case for the true data generating process. So the Phillips-Perron Test is conducted. In the Phillips-Perron test, the error terms can be weakly dependent and heterogeneously distributed. The criterion used to select the appropriate lag length is the Akaike Information Criterion.

60 In the Phillips-Perron test, a drift and a stationary trend is included into the least square regression. The regression will look as the following equation. 1 y t = + y t 1 + (t T ) + u t 2 In the Phillips-Perron test, the estimated variance used in calculating the t statistic is constructed by the Newey and West method. All tests are conducted through SHAZAM. 2. OLS Regression of the Growth Model Soo (1996) developed a variant of Solow growth model (Solow 1957) to test the relationship between life insurance and economic growth. I will base on what he developed to set up the relationship between insurance and economic growth. This is consistent with the discussion of chapter two that insurance may contribute to technological progress. In this variant of the Solow growth model, productivity growth (technological progress) is related to factors such as economic institutions, openness of an economy, information, knowledge etc. We assume a Cobb-Douglas production function as follows. F ( K , L) = e gt K L(1 ) e gt is the productivity factor that can explain the output growth that can not be explained by the growth of capital (K) and labor (L). Take the log of the production function and differentiate it with respect to time we can get the following

GY = g + GK + (1 )GL

61 We will take g as an endogenous variable which can be a function of openness of an economy, existence of risk shielding institutions and the degree of influence by government activities. We use the export and import growth as the measurement of openness of an economy, the insurance sector growth as the measurement of the effect of risk sharing institutions and government expenditure ratio as the measurement of degree of government influence. Therefore g = 0 + 1 Re xpg + 2 Rimpg + 3 rlifg + 4 rnon lg + 5 gov + Here Rexpg, Rimpg, Rlifg, Rnonlg and gov represent real export growth, real import growth, real life insurance in force growth and real non life insurance in force growth and government expenditure ratio to GDP respectively. Put g back to the production function we have the following. GY = 0 + 1 Re xpg + 2 Rimpg + 3 rlifg + 4 rnon lg + 5 gov + GK + (1 )GL + This will specify our estimated growth function as follows.
RGDPG = 0 + 1 INVEST + 2 LG + 3 RLIFG + 4 RNONLG + 5 RIMPG + 6 REXPG + 7 GOV + 8TIME +

Here we use real GDP growth as a proxy of output growth and investment ratio as a proxy of capital growth. 3. Simultaneous Equations Simultaneous equations are a system of equations, which allow for the interaction of variables in the system. The above theoretical framework shows that the insurance sectors growth may contribute to economic growth. On the other hand, economic growth may also demand for insurance growth. When the influence between economic growth

62 and insurance sector growth are mutual to each other, the simultaneous equation system should give a better estimation. Here we will give out a simple illustration how ordinary least square may lead to biased estimation of coefficients. We will use a system of two equations for illustration purpose, although in our estimation, we will have a system of three equations. Suppose we have the following structural equations. y1t = 0 + 1 y 2t + t y 2t = 0 + 1 y1t + u t Here we assume that E ( t ) = E (u t ) = 0 E ( t2 ) = 2 E (u t2 ) = u2 E ( t u t ) = 0 The equilibrium condition is y1t = y 2t To Estimate the first equation by OLS

a1 =

(y
t =1

2t

y 2 )( y1t y1 )
2

( y 2t y 2 )
t =1

= 1 +

(y
t =1 n

2t

y 2 )( t )
2t

(y
t =1

y2 ) 2
2

p lim(a1 ) = 1 +

( ) Cov( y 2t , t ) = 1 + 1 2 1 2 Var ( y 2t ) +u

Which is a biased estimator. In the following part, both the single equation and simultaneous equation system will be estimated. Results from different estimations will be compared with each other.

63 The single equation is defined as follows. RGDPG = 0 + 1 INVEST + 2 PG + 3 RLIFG + 4 RPCLG + 5 RIMPG + 6 REXPORT + 7 RGOV + 8TIME + Here is assumed to have a normal distribution with mean zero. The simultaneous equation model is defined as follows.

11 rgdpg t1 + 21 rlifg t 2 + 31 rnon lg t 3 + 11 xt1 + ...... + k1 xtk = t1 12 rgdpg t1 + 22 rlifg t 2 + 32 rnon lg t 3 + 12 xt1 + ...... + k 2 xtk = t 2 13 rgdpg t1 + 23 rlifg t 2 + 33 rnon lg t 3 + 13 xt1 + ...... + k 3 xtk = t 3
Here rgdpg, rlifg and rnonlg are endogenous variables and all the other variables are assumed to be exogenous. The set of x variables are different from country to country due to the different source and availability of the data for each country. However, the first equation in the above system is all the same for each country. It is defined the same as the single estimation equation. The relevance test of the instruments used for each country will be presented in Appendix D. Also, in either the single equation or the simultaneous equation system, if a unit root is suspected, its nth order stationary difference should be used to run the regression. In the following parts, each countrys data will be analyzed and presented individually. Then we will compare the results of the four countries. We will give a definition of all the variables that will be included in the study of all countries. Some variables are just country specific.

64

III. Definition of variables

RGDPG--- Real GDP growth, which is nominal GDP deflated by GDP deflator. This variable serves as a proxy for economic growth. INVEST--- Ratio of capital investment to GDP. This variable is a proxy of capital growth. PG--- Population growth. This variable is one of the measures of labor growth in an economy. LG--- Labor force growth. This variable is the other measure of labor growth in an economy. RIMPG--- Real import growth (nominal import is deflated by GDP deflator) REXPG--- Real export growth (nominal export is deflated by GDP deflator) RGOV---the percent of government expenditure in GDP RLIFG--- Real life insurance in force growth (nominal life insurance in force is deflated by GDP deflator) RPCLG--- Real property/casualty/liability insurance growth (nominal term is deflated by GDP deflator) RNONLG--- Real non life insurance growth (nominal term is deflated by GDP deflator). This should be very similar with RPCLG. But when I collect the data for different countries, they just label the data as non life insurance data. INF--- Expected inflation LEXP--- Life expectancy RSSG--- Real social security growth

65 EDR--- Third level education enrollment ratio. This is a variable to measure the education development. This variable is only available for the US economy. ED--- Percentage of high school graduates entering to higher education. This variable is only available for the Korean economy. WPART--- Woman participation ratio in the labor force DRATIO--- Dependant ratio. This is defined as the ratio of the population with age under 15 to the population with age between 15 and 64. PRATIO--- Profit ratio of the property/casualty/liability insurance industry. This is a variable which is only available for the US economy. NLNLR--- Non life insurance industry net loss ratio. This variable is only available for the Korean economy. It is related with PRARIO with the following relationship 1NLNLR--- PRATIO LNLR--- Life insurance industry net loss ratio. This variable is also available just for the Korea economy. URBANR--- Percent of urban population in total population If FD is added before a variable, it means the first difference of the time series is used because a unit root is suspected. Similarly, SD and TD represent the second difference and the third difference of the time series. Before we present the empirical results from these countries, we will discussion how expected inflation is constructed. Consumer price index is collected for each country. The change in consumer price index from one period to the next is calculated as the inflation of each country. Then the different averages of past inflation are calculated as the expected inflation variables. We calculated the averages from the most recent two

66 periods up to the most recent eight periods. Then the unit root tests on those expected inflation variables are conducted. Then the nth order difference which can reject the existence of a unit root at 90% significance level is chosen. An ordinary least square between the nth order of these expected inflation variables and real life insurance in force is run. The one most correlated with real life insurance in force is chosen to be the expected inflation variable for that country. For the US economy, the average of the last eight periods of inflation is chosen as the expected inflation. For Korea, Germany and Sweden, the averages of the last seven, four and seven periods of inflation are chosen as the expected inflation for these three countries respectively. Because both population growth and labor force growth are available for each country, OLS regression is run twice, first with population growth, then with labor force growth as the measure of labor growth for each country. Labor force growth seems to be a better measure as labor growth for Korea, Germany and Sweden, where the estimated coefficients on labor force growth show stronger statistical significance than the ones on population growth for OLS regressions. However, for the US economy, population growth seems a better one to use as a measure of labor growth. Also, these two alternatives as the measure of labor growth are used in simultaneous equation analysis.

67

Chapter 4 Empirical Result from Four Countries

I. Analysis of the US data

i) Unit Root Test Results
Table 4.1.1 presents the unit root test results of all the variables used in the simultaneous equations of the US data. Both the DF test and the PP test results are presented here. The null hypothesis of the test is that there is a unit root in the time series. If the null hypothesis is not rejected in 10% significance level, a unit root is suspected in the time series. In the DF test and PP test, a drift and a time trend is included in the regressions. The regression data range is from 1966 to 2000. However, when the unit root test is conducted for each time series, a longer period of data is used if the series is available for a longer period of time. This may result in a more accurate unit root test. From the Table 4.1.1, we can see for some variables, the DF test and PP test give out different results regarding the existence of a unit root in the time series. We will choose according to the test results of PP test, because the error assumption for the PP test is more flexible.

68

Table 4.1.1 Unit Root Test Results (US Data) Variables WPART PRATIO INF EDR LEXP RSSG RGDPG PG RLIFG RPCLG RGOV INVEST RTRADEG DRATIO LG REXPG RIMPG URBANR PP Test Statistics -0.25 -3.27* -0.35 -5.40* -1.44 -6.83* -4.67* -6.91* -3.57* -3.22* -2.48 -2.88 -4.80* -0.71 -3.28* -3.92* -6.00* 0.50 DF Test Statistics -0.48 -3.06 -1.81 -5.50* -1.39 -2.29 -3.39* -2.21 -3.30* -3.42 -2.42 -2.77 -4.78 -3.64 -2.55* -3.66 -3.23 -1.48

Note: * shows that unit root test can reject the null hypothesis at 90% significant level.

For the variables that can not reject the null hypothesis of the existence of a unit root, the first difference of that variable is taken to do the unit root test. All unit root tests for the first difference of those variables except INF, DRATIO and URBANR reject the null hypothesis of a unit root at 90% significance level. Then the second difference of the above three variables are taken and the unit root test on them are conducted. The null is rejected at 90% significance level for the second difference of the above three variables.

69

ii) OLS Regression

The simple OLS regression will be run first. The results are used to compare with the simultaneous equation results. The following are the estimated equations. 1. RGDPG = 0 + 1 FDINVEST + 2 PG + 3 IMPG + 4 EXPG + 5 RLIFG + 6 RPCLG + 7 RGOV + 8TIME + 2.
RGDPG = 0 + 1 FDINVEST + 2 LG + 3 IMPG + 4 EXPG + 5 RLIFG + 6 RPCLG

+ 7 RGOV + 8TIME +

The Newey West estimator is used to estimate the variance and covariance of the estimated coefficients. The estimated coefficients and the t statistics based on the Newy West estimator are shown in Table 4.1.2 and Table 4.1.3. As stated before, population growth and labor force growth are used as two different measures of labor growth in the model.
Table 4.1.2 Estimation Results from OLS (US data) - Population Growth is Used as a Proxy of Labor Growth
R-square=0.7664 Adj R34 square=0.6916 Observations Dependent Variable= RGDPG ESTIMATED COEFFICIENT T-RATIO 0.697 3.476 7.556 1.010 0.009 0.333 -0.009 -0.155 -1.685 -2.364 0.159 3.223 0.086 3.131 -0.033 -0.715 -5.001 -0.721

VARIABLE NAME FDINVEST PG REXPG RIMPG FDRGOV RLIFG RPCLG TIME CONSTANT

P-VALUE 0.002 0.322 0.742 0.878 0.026 0.004 0.004 0.481 0.477

70 From the above results we can see that both life insurance growth and PCL insurance growth have a positive contribution to GDP growth. Also, both estimated coefficients show strong statistical significance. Export growth and import growth do not show very strong statistical relationship with GDP growth. Export growth shows a positive relationship with GDP growth while import growth shows a negative relationship with GDP growth. Percent of government expenditure to GDP shows negative relationship with economic growth with strong statistically significance. Both capital growth (approximated by investment share in GDP) and labor growth (approximated by population growth) have shown positive contribution to economic growth (approximated by GDP growth). But the estimated coefficient on Investment ratio is statistically significant while the estimated coefficient on population growth is not statistically significant.
Table 4.1.3 Estimation Results from OLS (US Data) - Labor Force Growth is Used as a Proxy of Labor Growth
R-square=0.7564 34 Observations Adj R-square=0.6785 Dependent Variable= RGDPG ESTIMATED COEFFICIENT T-RATIO 0.786 3.258 0.109 0.217 0.003 0.090 -0.021 -0.340 -1.373 -1.727 0.188 3.020 0.071 2.284 0.002 0.072 1.860 1.342

VARIABLE NAME FDINVEST LG REXPG RIMPG FDRGOV RLIFG RPCLG TIME CONSTANT

P-VALUE 0.003 0.830 0.929 0.737 0.097 0.006 0.031 0.943 0.192

When labor force growth is used as a measure of labor growth, the estimated coefficient on labor force growth is much different than the estimated coefficient on

71 population growth. At the same time, the significance level of the estimated coefficient on labor force is lowered significantly. However, the signs and statistical significance level of the estimated coefficients on all the other variables are very similar. Especially, here we can see the change in the measure of the labor growth does not affect much the estimated coefficients on both life insurance growth and PCL insurance growth. In either situation, they showed statistically significant contribution to economic growth. In the following part, the simultaneous equations are used to allow for the interaction between GDP growth and insurance growth. Two sets of simultaneous equations are run with either PG or LG in the first equation.

iii) Simultaneous Equations

The simultaneous equation system includes three equations. 1. RGDPG = 0 + 1 FDINVEST + 2 PG + 3 RLIFG + 4 RPCLG + 5 RIMPG + 6 REXPG + 7 FDRGOV + 8TIME + 1 2. RLIFG = 0 + 1 SDINF + 2 RGDPG + 3 fFDLEXP + 4 RSSG + 5 EDR + 6 SDRATIO + 7 FDWPART + 8 SDURBANR + 9 D + 10TIME + 2 3. RPCLG = 0 + 1 RGDPG + 2 PRATIO + 3 EDR + 4 FDWPART + 5 SDURBANR + 6TIME + 3 The estimated results are listed in the following tables (Table 4.1.4 to Table 4.1.6)

72

Table 4.1.4 Estimation Results from Simultaneous EquationsEquation 1 (US Data) Population Growth is Used as a Proxy of Labor Growth
R-square=0.6109 8 34 Observations Dependent Variable= RGDPG ESTIMATED COEFFICIENT T-RATIO 0.700 2.193 8.896 1.356 0.130 1.198 0.250 3.262 -0.034 -0.673 0.041 0.925 -1.378 -1.829 -0.013 -0.413 -7.157 -1.121

VARIABLE NAME FDINVEST PG RLIFG RPCLG RIMPG REXPG FDRGOV TIME CONSTANT

P-VALUE 0.028 0.175 0.231 0.001 0.501 0.355 0.067 0.679 0.262

Here we will compare the results from the OLS and the results from the first equation in the simultaneous equation system. All the estimated coefficients in the simultaneous equation system remain the same signs as for those in the single equation model. However, the statistical significance for some estimated coefficients changed. The statistical significance of the estimated coefficient on PG increased. While the statistical significance of the estimated coefficient on RPCLG increases, the statistical significance of the estimated coefficient on RLIFG decreases a lot. The statistical significance of the estimated coefficients of both IMPG and EXPG increases. The statistical significance of the estimated coefficient on FDRGOV decreases a little.

When instrument variables are used, the sum of RSS and ESS is not equal to TSS. The definition of R-square is

R2 =

RSS RSS + ESS

. It may not be a good measure of goodness fit of the model.

73

Table 4.1.5 Estimation Results from Simultaneous EquationsEquation 2 (US Data) Population Growth is Used as a Proxy of Labor Growth
R-square=0.4647 34 Observations Dependent Variable= RLIFG ESTIMATED COEFFICIENT T-RATIO -2.923 -3.357 0.061 0.299 0.869 0.575 -0.155 -1.536 -13.721 -2.973 268.660 2.208 -233.140 -2.118 7.631 1.625 1.058 0.986 0.157 1.521 45.331 3.582

VARIABLE NAME SDINF RGDPG FDLEXP RSSG EDR SDRATIO FDWPART SDURBANR D TIME CONSTANT

P-VALUE 0.001 0.765 0.565 0.125 0.003 0.027 0.034 0.104 0.324 0.128 0.000

This is the second equation in the simultaneous equation system. It examines the factors which affect real life insurance in force growth. We can also view this as a demand growth function of life insurance. The negative relationship between expected inflation growth and life insurance in force growth shows that high expected inflation growth harms the purchase of life insurance. Higher GDP growth contributes positively to life insurance purchase growth, although it is not statistically significant. When the income of an economy increases, life insurance may be more affordable to the agents in this economy. This is why the higher GDP growth may lead to higher growth in life insurance in force. Life expectancy is positively related to life insurance purchase but with very low statistical significance. This means, when people expect to live longer, they purchase more life insurance. Social security growth has a negative contribution to life insurance purchase growth. This means

74 when social security grows, it will crowd out the need for life insurance purchase. This is very meaningful. Third level enrollment has a negative contribution to life insurance in force growth. This is perplexing at the first sight. However, our explanation for this is as follows. When people get more education, especially higher education (remember EDR is the third level enrollment rate in proper age group), they tend to have a better paid job with a better benefit package. This may reduce their need for life insurance. Dependant ratio has a positive contribution to life insurance purchase growth. This is meaningful because when there are more dependants for the primary breadwinner, there is a greater need for life insurance to cover the risk of early death. Women participation ratio has a negative contribution to life insurance purchase. The relationship here is also statistically strong. When women join the labor force, it spreads the risk of the loss due to the death of primary breadwinner, usually the husband in a household. A higher ratio of urban population in the whole population contributes to a higher life insurance in force growth. When there are more people living in the urban area, there is less family-tie, which gives familys members in need help. This probably explains why higher urban population ratio the growth of life insurance in force. D is the variable that defines the higher inflation period in the US economy. It is from 1974 to 1984. During the high inflation period, people seemed to have a higher demand for more life insurance.

75

Table 4.1.6 Estimation Results from Simultaneous EquationsEquation 3 (US Data) Population Growth is Used as a Proxy of Labor Growth
R-square=0.2543 34 Observations Dependent Variable= RPCLG ESTIMATED COEFFICIENT T-RATIO 0.887 1.852 -0.261 -1.462 153.340 0.619 -10.236 -1.035 -0.107 -0.016 -0.091 -0.573 3.565 0.183

VARIABLE NAME RGDPG PRATIO FDWPART SDURBANR EDR TIME CONSTANT

P-VALUE 0.064 0.144 0.536 0.301 0.988 0.567 0.855

The third equation explains the factors that contributes to property/casualty/liability insurance growth. First, GDP growth has a positive contribution to PCL insurance growth and the relationship is very strong statistically. More wealth is accumulated as GDP grows. Therefore there is a greater need for protection against the loss. Profit ratio is negatively related to PCL insurance growth. This supports those literatures stating that there are writing cycles in PCL insurance. Here, third level enrollment ratio seems to have no effect on the demand for PCL insurance. Woman participation ratio in labor force is positively related to PCL insurance purchase. When more women join the labor force, the total income of a household may increase. There is a higher demand for PCL insurance to protect the increased wealth. Urban population ratio has a negative contribution to PCL insurance purchase growth.

76 The following part shows the results from the following simultaneous equation system, which replaces PG with LG in the first equation to see whether there is any change in the results. 1. RGDPG = 0 + 1 FDINVEST + 2 LG + 3 RLIFG + 4 RPCLG + 5 RIMPG + 6 REXPG + 7 FDRGOV + 8TIME + 1 2. RLIFG = 0 + 1 SDINF + 2 RGDPG + 3 fFDLEXP + 4 RSSG + 5 EDR + 6 SDRATIO + 7 FDWPART + 8 SDURBANR + 9 D + 10TIME + 2 3. RPCLG = 0 + 1 RGDPG + 2 PRATIO + 3 EDR + 4 FDWPART + 5 SDURBANR + 6TIME + 3
Table 4.1.7 Estimation Results from Simultaneous EquationsEquation 1 (US Data) Labor Forth Growth is Used as a Proxy of Labor Growth
R-square=0.5181 34 Observations Dependent Variable= RGDPG ESTIMATED COEFFICIENT T-RATIO 0.781 2.468 -0.063 -0.134 0.180 1.596 0.271 3.855 -0.048 -0.980 0.040 0.905 -0.933 -1.275 0.026 0.993 1.097 0.903

VARIABLE NAME FDINVEST LG RLIFG RPCLG RIMPG REXPG FDRGOV TIME CONSTANT

P-VALUE 0.014 0.893 0.110 0.000 0.327 0.365 0.202 0.321 0.367

First, we will compare Table 4.1.3 with Table 4.1.7. Table 4.1.3 is the OLS regression results with LG as a proxy of the labor growth. Table 4.1.7 is the regression results from the first equation in the simultaneous equation system also with LG as a proxy of the labor growth. We can see all the estimated coefficients remain the same sign

77 except the one on LG, which changes from a positive sign to a negative sign, although neither one is statistical significant. The statistical significance level of the estimated coefficients on both RIMPG and REXPG increases. Here we can see in the simultaneous equation system, the statistical significance of the estimated coefficient on RLIFG decreases and the statistical significance of the estimated coefficient on RPCLG increases. This is consistent with what we found when we compare the results from OLS with the results from the simultaneous equation system in the case that PG is used as a measure of labor growth. Therefore, we may make the conclusion that the single equation over estimated the coefficient on RLIFG but under estimated the coefficient on RPCLG. Second, we will compare Table 4.1.4 and Table 4.1.7. Both of them are the first equation in the three equation system. However, Table 4.1.4 is the results when PG is used as a proxy of labor growth while Table 4.1.7 is the results when LG is used as a proxy of labor growth. Both tables show that life insurance and PCL insurance contribute positively to economic growth. Also, the statistical significance level for the estimated coefficients of RPCLG is higher than the one for the estimated coefficients of RLIFG. The following two tables (Table 4.1.8 and Table 4.1.9) are the regression results of the second and third equations when LG is used as a proxy of labor growth.

78

Table 4.1.8 Estimation Results from Simultaneous EquationsEquation 2 (US Data) Labor Forth Growth is Used as a Proxy of Labor Growth
R-square=0.4699 VARIABLE NAME SDINF RGDPG FDLEXP RSSG EDR SDRATIO FDWPART SDURBANR D TIME CONSTANT 34 Observations Dependent Variable= RLIFG ESTIMATED COEFFICIENT T-RATIO -2.974 -3.425 0.082 0.420 0.693 0.455 -0.155 -1.526 -13.970 -3.036 272.930 2.231 -220.580 -2.046 7.777 1.687 1.065 0.983 0.162 1.572 45.954 3.645

P-VALUE 0.001 0.674 0.649 0.127 0.002 0.026 0.041 0.092 0.326 0.116 0.000

Table 4.1.9 Estimation Results from Simultaneous EquationsEquation 3 (US Data) Labor Forth Growth is Used as a Proxy of Labor Growth
R-square=0.2547 34 Observations Dependent Variable= RPCLG ESTIMATED COEFFICIENT T-RATIO 1.095 2.361 -0.263 -1.506 161.970 0.700 -8.474 -0.922 -1.789 -0.279 -0.067 -0.451 7.766 0.425

VARIABLE NAME RGDPG PRATIO FDWPART SDURBANR EDR TIME CONSTANT

P-VALUE 0.018 0.132 0.484 0.356 0.780 0.652 0.671

We can see that Table 4.1.8 and Table 4.1.9 are very similar with Table 4.1.5 and Table 4.1.6. The change of PG to LG almost has no effect on the results from the second and the third equations. We can make the following conclusion from the US data. The US data support our theoretical models in the sense that both life insurance and non life insurance

79 contribute to economic growth. Yet, from the simultaneous equation system, the evidence for life insurance is not as strong as the one for non life insurance.

80

II. Analysis of Korean Data

i) Unit Root Test

Table 4.2.1 Unit Root Test Results (Korean Data) Variables WPART NLNLR LNLR INF ED LEXP RGDPG PG RLIFG RNONLG RGOV INVEST DRATIO LG REXPG RIMPG URBANR PP Test Statistics DF Test Statistics -2.8 -2.95 -0.29 -2.81 -1.47 -1.72 -3.19* -1.75 -3.87* -3.58* -2.34 -1.8 -2.88 -4.47* -4.08* -2.43 -0.92

-2.85 -2.95 -1.61 -2.25 -1.61 -1.96 -4.88* -1.94 -5.79* -3.60* -2.28 -1.91 -0.79 -4.47* -4.09* -4.76* 2.54

Note: * shows that unit root test can reject the null hypothesis at 90% significant level.

For the variables that can not reject the null hypothesis of the existence of a unit root at 90% significant level, the first difference of that variable is taken and a unit root test is conducted on it in Table 4.2.1. The results show that the null hypothesis can be rejected at the 90% significant level for the first difference of INVEST, PG, INF, LEXP, ED, NLNLR, LNLR, RGOV, and WPART. So for any equations to include these variables, the first difference of those variables will be used for the regressions. For the

81 variables with non-stationary first difference, the second difference is taken and a unit root test is conducted on it. These include the variables DRATIO, URBANR and INF. The unit root tests for all three variables show that the second difference can reject the null of the existence of unit root at 90% significance level.

ii) Simple OLS Regression

The following is the estimated equations. 1.
RGDPG = 0 + 1 FDINVEST + 2 FDPG + 3 RLIFG + 4 RNONLG + 5 RIMPG

+ 6 REXPG + 7 FDRGOV + 8TIME +

2.

RGDPG = 0 + 1 FDINVEST + 2 LG + 3 RLIFG + 4 RNONLG + 5 RIMPG

+ 6 REXPG + 7 FDRGOV + 8TIME +

The first equation uses PG as a measurement of labor growth and the second equation uses LG as a measurement of labor growth. Table 4.2.2 is the estimated coefficients and the t statistics based on the Newy West estimator.

82

Table 4.2.2 Estimation Results from OLS (Korean Data) - Population Growth is Used as a Proxy of Labor Growth
R-square=0.9362 21 Adj R-square=0.8937 Observations Dependent Variable= RGDPG ESTIMATED COEFFICIENT T-RATIO 0.505 1.398 0.874 0.390 0.006 3.968 0.089 2.702 0.061 0.647 0.041 0.435 -2.211 -2.187 -0.133 -1.642 6.255 6.205

VARIABLE NAME FDINVEST FDPG RLIFG RNONLG REXPG RIMPG FDRGOV TIME CONSTANT

P-VALUE 0.188 0.703 0.002 0.019 0.530 0.671 0.049 0.127 0.000

From the above result we can see that both the estimated coefficients on FDINVEST and FDPG show positive sign, which is what we expect. However, neither one shows strong statistical significance, especially for the estimated coefficient on FDPG. Both life insurance growth and non life insurance growth show positive contribution to economic growth and both of the estimated coefficients are statistically significant. Both real export growth and real import growth show positive contribution to the economic growth, but neither of them shows strong statistical significance. The estimated coefficient on GDRGOV is negative and statistically strong. This means an increase in the percentage of government expenditure will decrease GDP growth. The results in Table 4.2.3 show when we use LG in replace of PG.

83

Table 4.2.3 Estimation Results from OLS (Korean Data) - Labor Force Growth is Used as a Proxy of Labor Growth
R-square=0.9462 VARIABLE NAME FDINVEST LG RLIFG RNONLG REXPG RIMPG FDRGOV TIME CONSTANT 21 Adj R-square=0.9103 Observations Dependent Variable= RGDPG ESTIMATED COEFFICIENT T-RATIO 0.383 1.130 0.495 2.565 0.006 4.558 0.049 1.918 0.023 0.248 0.064 0.726 -2.520 -2.537 -0.113 -1.736 5.686 10.500

P-VALUE 0.281 0.025 0.001 0.079 0.808 0.482 0.026 0.108 0.000

From the above regression result, we can see that for the Korean economy, labor force growth is a better proxy of labor growth than the population growth. The significance level of the estimated coefficient on LG is much higher than the one on PG from the first table. In this regression, life insurance growth and non life insurance growth also show a positive contribution to economic growth with strong statistical significance. At the same time, both regressions show that life insurance growth has a stronger statistical relationship with GDP growth than non life insurance growth. The change of PG to LG does not change the results on other estimated coefficients either. In the following part, we will present the results from simultaneous equations. Both LG and PG are used in the simultaneous equation system to compare the regression result.

84

iii) Simultaneous Equations

1. RGDPG = 0 + 1 FDINVEST + 2 LG (/ FDPG ) + 3 RLIFG + 4 RNONLG + 5 RIMPG + 6 REXPG + 7 FDRGOV + 8TIME + 1 2.
RLIFG = 0 + 1RGDPG + 2 FDLNLR + 3 FDLEXP + 4 SDRATIO + 5 SDURBANR

+ 6 FDWPART + 7 FDED + 8 SDINF + 9TIME + 2

3. RNONLG = 0 + 1 RGDPG + 2 FDNLNLR + 3 SDURBANR + 4 FDED + 5 FDWPART + 6TIME + 3 The regression results are shown in the following tables (Table 4.2.4 to Table 4.2.6).
Table 4.2.4 Estimation Results from Simultaneous EquationsEquation 1 (Korean Data) - Labor Force Growth is Used as a Proxy of Labor Growth
R-square=0.9434 21 Observations Dependent Variable= RGDPG ESTIMATED COEFFICIENT T-RATIO 0.435 1.688 0.486 1.827 0.004 0.974 0.049 0.979 0.028 0.464 0.064 0.912 -2.068 -2.510 -0.128 -2.391 5.833 6.105

VARIABLE NAME FDINVEST LG RLIFG RNONLG REXPG RIMPG FDGOV TIME CONSTANT

P-VALUE 0.091 0.068 0.330 0.328 0.643 0.362 0.012 0.017 0.000

Comparing the result of OLS regression and the one from the first equation of the simultaneous equation system (Table 4.2.4), we can see that all the estimated coefficient remain the same sign. However, the statistical significance of most estimated coefficients changes from one method to the other method. We can notice here, the statistical

85 significance for the estimated coefficients on both life and non life insurance growth is lower than before. The statistical significance for the estimated coefficient on labor force growth is also lower than before, although it still remains at a high statistical significance level. But the statistical significance of the estimated coefficient on FDININVEST increases. The statistical significance of the estimated coefficients on both RIMPG and REXPG increases, yet they are still not statistically significant. All the other estimated coefficients remain at the similar significance level.
Table 4.3.5 Estimation Results from Simultaneous EquationsEquation 2 (Korean Data) - Labor Force Growth is Used as a Proxy of Labor Growth
R-square=0.4193 Dependent Variable= RLIFG ESTIMATED COEFFICIENT -4.068 -2.830 -216.340 178.710 70.325 -10.136 -2.317 -53.141 -6.647 252.440 21 Observations

VARIABLE NAME RGDPG FDLNLR FDLEXP SDRATIO SDURBANR FDWPART FDED SDINF TIME CONSTANT

T-RATIO -0.599 -1.276 -1.372 0.986 1.165 -0.283 -0.637 -3.089 -1.727 1.836

P-VALUE 0.549 0.202 0.170 0.324 0.244 0.777 0.524 0.002 0.084 0.066

The second equation explains the factors that affect the growth of life insurance in force. This is the demand growth function of the life insurance. Interestingly, for the Korean Economy, GDP growth has a negative contribution to life insurance in force growth. Put it another way, when GDP growth increases, the growth for the demand of life insurance decreases. We need some explanation here for this result. Life insurance net loss ratio is negatively correlated with life insurance in force growth. This means there is no counter writing cycle for the life insurance industry of the Korean economy.

86 Expected life expectance is negatively correlated with growth of life insurance in force. When expected life expectancy increases, the demand for life insurance grows slower. Dependant ratio is positively correlated with life insurance in force growth. This means when dependant ratio increases, the demand for life insurance grows faster. This is what we expect. The estimated coefficient on SDURBANR is positive, which means when there are more population living in the urban area, the growth of life insurance in force will increase. The estimated coefficient on women participation ratio is negative. This means when more women participate in the labor force, it will decrease the growth of life insurance in force. This is what we expect. Here, we also see that participation in higher education will lower the growth rate of life insurance demand. This result is similar with the result from the US economy. At last, expected inflation growth will decrease the growth of life insurance in force. It also shows a great statistical significance. This means high expected inflation harms life insurance growth.
Table 4.2.6 Estimation Results from the Simultaneous EquationsEquation 3 (Korean Data) - Labor Force Growth is Used as a Proxy of Labor Growth
R-square=0.7004 21 Observations Dependent Variable= RNONLG ESTIMATED COEFFICIENT T-RATIO 0.834 1.808 -0.177 -1.170 -10.007 -1.897 0.872 2.560 6.822 2.221 -0.710 -2.747 11.703 2.208

VARIABLE NAME RGDPG FDNLNLR SDURBANR FDED FDWPART TIME CONSTANT

P-VALUE 0.071 0.242 0.058 0.010 0.026 0.006 0.027

87 The third equation explains the factors that influence the growth of non life insurance. First, we can see that the estimated coefficient on RGDPG is positive and statistically significant. This means GDP growth in Korean economy contributes to non life insurance growth. Second, the net loss ratio of non life insurance is negatively related to non life insurance growth. This is shown by a negative and statistically significant estimated coefficient on FDNLNLR. This is actually contrary to what we found in the US economy, where the data shows that there is counter writing cycle in non life insurance. Usually, a counter writing cycle signals the competitiveness of the insurance market. Comparing the results from Korean and US data, we may conclude that the US non life insurance is more competitive than the one in the Korean economy. The estimated coefficient on SDURBANR is negative and statistically significant. This means for the Korean economy, when there are more population living in the urban area, there is a slower growth in demand for non life insurance. Both higher education participation and women participation in labor force have positive and significantly strong contribution to the growth of non life insurance demand. These make sense. When more females join the labor force, the income of each household may increase. This requires more need for insurance to protect household wealth from risk of loss. Then LG is replaced by PG in the first equation and we run the whole equation system again. The results are shown in Table 4.2.7 to Table 4.2.9.

88

Table 4.2.7 Estimation Results from the Simultaneous EquationsEquation 1 (Korean Data) - Population Growth is Used as a Proxy of Labor Growth
R-square=0.9301 21 Observations Dependent Variable= RGDPG ESTIMATED COEFFICIENT 0.565 2.055 0.002 0.088 0.060 0.036 -1.695 -0.151 6.655 T-RATIO 2.071 0.636 0.518 1.816 0.970 0.486 -1.943 -2.518 4.644 P-VALUE 0.038 0.525 0.605 0.069 0.332 0.627 0.052 0.012 0.000

VARIABLE NAME FDINVEST FDPG RLIFG RNONLG REXPG RIMPG FDGOV TIME CONSTANT

Table 4.2.8 Estimation Results from the Simultaneous EquationsEquation 2 (Korean Data) - Population Growth is Used as a Proxy of Labor Growth
R-square=0.4189 VARIABLE NAME RGDPG FDLNLR FDLEXP SDRATIO SDURBANR FDWPART FDED1 SDINF TIME CONSTANT 21 Observations Dependent Variable= RLIFG ESTIMATED COEFFICIENT T-RATIO -4.075 -0.596 -2.751 -1.234 -215.320 -1.362 178.320 0.980 70.197 1.158 -10.923 -0.303 -2.259 -0.619 -53.073 -3.074 -6.628 -1.711 251.700 1.820

P-VALUE 0.551 0.217 0.173 0.327 0.247 0.762 0.536 0.002 0.087 0.069

89
Table 4.2.9 Estimation Results from the Simultaneous EquationsEquation 3 (Korean Data) - Population Growth is Used as a Proxy of Labor Growth
R-square=0.6985 VARIABLE NAME RGDPG FDNLNLR SDURBANR FDED1 FDWPART TIME CONSTANT 21 Observations Dependent Variable= RNONLG ESTIMATED COEFFICIENT T-RATIO 0.813 1.765 -0.173 -1.173 -9.617 -1.860 0.844 2.500 7.676 2.553 -0.705 -2.728 11.749 2.216

P-VALUE 0.078 0.241 0.063 0.012 0.011 0.006 0.027

What we can see here is that for Korea, the change from LG to PG almost has no affect on the regression result of the second and third equation in the system. However the results from the first equation (which represent the economic growth function of each country) do have some significant change on estimated coefficients. First, when LG is used, the estimated coefficients on LG are much more significant than those on PG when PG is used, which prove that LG is a more precise proxy of labor growth here. Use of PG or LG does not alter any sign on the estimated coefficients. It only changes the significance level on some estimated coefficients. For the Korean economy, when PG is used in the economic growth function, the significance level on the estimated coefficients of non life insurance growth increased. At the same time, the significance level on the estimated coefficients of life insurance growth decreased. From the Korean data, we can have the following conclusion. Both life insurance and non life insurance growth contribute to the growth of the Korean economy. But the statistical relationship found in the simultaneous equation system is not as strong as the one from the OLS regression. The change of labor growth proxy does not change the above conclusion.

90

III. Analysis of Swedish Data

i) Unit Root Test

Table 4.3.1 Unit Root Test Results (Swedish Data) Variables WPART INF LEXP RGDPG PG RLIFG RNONLG RGOV INVEST REXPG DRATIO LG URBANR PP Test Statistics 1.39 -1.37 -3.76* -3.73* -1.89 -4.71* -5.95* -2.28 -3.38* -4.07* -1.01 -3.43* -4.15* DF Test Statistics 0.99 -2.04 -2.73 -3.63* -1.49 -4.71* -4.24* -2.24 -3.25* -2.18 -1.38 -2.39 -3.25*

Note: * shows that unit root test can reject the null hypothesis at 90% significant level.

For the variables that the unit root test can not reject the null hypothesis at 90% significant level, the first difference is taken and a unit root test on the first difference of the variable is conducted In Table 4.3.1. These include woman participation in the labor force, expected inflation, population growth, the ratio of government expenditure to GDP and dependant ratio. But the unit root test on the first difference of population growth and dependant ratio concludes that a unit root may be present in the first difference of both variables. I conduct the unit root test until I found that the third difference of both variables can reject

91 the null at 90% significance level. Therefore, we include the third difference of both variables into the regression analysis.

ii) Simple OLS Regression

The following is the estimated equations. 1. RGDPG = 0 + 1 INVEST + 2TDPG + 3 RLIFG + 4 RNONLG + 5 REXPG + 6 FDRGOV + 7TIME + 2.
RGDPG = 0 + 1 INVEST + 2 LG + 3 RLIFG + 4 RNONLG + 5 REXPG

+ 6 FDRGOV + 7 TIME +

The estimated coefficients and the t statistics based on the Newy West estimator are shown in Table 4.3.2.
Table 4.3.2 Estimation Results from OLS (Swedish Data) - Population Growth is Used as a Proxy of Labor Growth
R-square=0.8905 21 Observations Adj R-square=0.8316 Dependent Variable= RGDPG ESTIMATED COEFFICIENT T-RATIO 0.500 7.920 1.847 0.292 -0.083 -2.905 0.058 2.042 0.057 4.914 -1.839 -11.070 -0.032 -1.720 -8.979 -5.239

VARIABLE NAME INVEST TDPG RLIFG RNONLG REXPORTG FDRGOV TIME CONSTANT

P-VALUE 0.000 0.775 0.012 0.062 0.000 0.000 0.109 0.000

Here we can see that the estimated coefficient on TDPG is positive but the statistical significance is really low. This shows that PG is not a good proxy of labor growth. For the Swedish economy, the estimated coefficient on real life insurance in

92 force is negative with high statistical significance. It means life insurance growth will hinder economic growth. This finding is different with the US and Korean economy. It is also against the theoretical models prediction in the first part of the dissertation. However, the estimated coefficient on real non life insurance has a positive sign with strong statistical significance. Also the estimated coefficients on both export growth and FDRGOV show the expected signs with high statistical significance. Now lets see the following result when we replace TDPG with LG in Table 4.3.3.
Table 4.3.3 Estimation Results from OLS (Swedish Data) - Labor Force Growth is Used as a Proxy of Labor Growth
R-square=0.8972 VARIABLE NAME INVEST LG RLIFG RNONLG REXPORTG FDRGOV TIME CONSTANT Adj R-square=0.8418 Dependent Variable= RGDPG ESTIMATED COEFFICIENT T-RATIO 0.430 6.218 0.268 1.307 -0.077 -2.782 0.062 2.166 0.057 4.290 -1.791 -9.129 -0.026 -1.222 -7.617 -4.571 21 Observations P-VALUE 0.000 0.214 0.016 0.049 0.001 0.000 0.244 0.001

After we use LG to replace TDPG, the significance level of the estimated coefficient on LG increases a lot. All the signs and statistical significance level of the estimated coefficients on other variables do not change much. In the following part, we will analyze the data using the simultaneous equation method. Both LG and PG are used in the simultaneous equation system to compare the regression result.

93

iii) Simultaneous Equations

1. RGDPG = 0 + 1 INVEST + 2 LG (/ TDPG ) + 3 RLIFG + 4 RNONLG + 5 REXPG + 6 FDRGOV + 7TIME + 1 2. RLIFG = 0 + 1 RGDPG + 2 LEXP + 3URBANR + 4 FDINF + 5 FDWPART + 6TIME + 2 3. RNONLG = 0 + 1 RGDPG + 2URBANR + 3 FDWPART + 4TIME + 3 The results from the simultaneous equation estimation are reported in Table 4.3.4 to Table 4.3.9. Table 4.3.4 to Table4.3.6 show the results when LG is used in the first equation; While Table 4.3.7 to Table 4.3.9 show the results when TDPG is used in the first equation.
Table 4.3.4 Estimation Results from SMEEquation 1 (Swedish Data) - Labor Force Growth is Used as a Proxy of Labor Growth
R-square=0.8951 21 Observations Dependent Variable= RGDPG ESTIMATED COEFFICIENT T-RATIO 0.422 4.392 0.317 1.224 -0.070 -1.060 0.074 1.698 0.059 2.264 -1.794 -6.509 -0.022 -0.689 -7.589 -3.400

VARIABLE NAME INVEST LG RLIFG RNONLG REXPORTG FDRGOV TIME CONSTANT

P-VALUE 0.000 0.221 0.289 0.089 0.024 0.000 0.491 0.001

Table 4.3.4 shows the estimation result of the first equation. It reports the factors that contribute to the growth of the Swedish economy. The estimated coefficients on

94 INVEST and LG are both positive. There are very similar with the results we get from the single equation estimation. The estimated coefficients on RLIFG and RNONLG remain the same signs and similar values with the one from the single equation estimation. However, the statistical significance of the estimated coefficients on both variables is lowered compared with the single equation result. In the simultaneous equation system, we still can get the conclusion that real life insurance growth will hinder real GDP growth while real non life insurance growth will contribute to real GDP growth positively. Export growth positively contributes to economic growth and the estimated coefficient is statistically significant, just as shown in the OLS. From the simultaneous equation system we can also conclude that a higher ratio of government spending to GDP has negative contribution to GDP growth. This is shown by the negative and statistical significant estimated coefficient on FDRGOV.
Table 4.3.5 Estimation Results from SMEEquation 2 (Swedish Data) - Labor Force Growth is Used as a Proxy of Labor Growth
R-square=0.1798 21 Observations Dependent Variable= RLIFG ESTIMATED COEFFICIENT T-RATIO -0.092 -0.088 5.627 0.903 -10.136 -1.937 -0.438 -0.131 982.270 1.347 27.144 0.889 -0.075 -0.057 413.680 0.564

VARIABLE NAME RGDPG LEXP URBANR FDINF FDWPART QDRATIO TIME CONSTANT

P-VALUE 0.930 0.366 0.053 0.896 0.178 0.374 0.954 0.573

Table 4.3.5 shows the factors that influence life insurance in force growth. The estimated coefficient on RGDPG is negative and almost has no statistical significance. This means an increase in real GDP growth has no effect on real life insurance in force

95 growth. The estimated coefficients on LEXP, FDWPART and QDRATIO are all positive. Therefore, increase in life expectancy, higher woman participation ratio in labor force and higher dependant ratio all contribute to life insurance in force growth. The estimated coefficient on URBANR is negative and statistically significant. More people living in the urban area will decrease the growth of life insurance in force growth. This is different with what we found in the US and Korean economy. Although the estimated coefficient on FDINF is negative but it almost has no statistical significance. This means expected inflation does not affect the growth of life insurance in force. This is also different with what we found in the US and Korean economy.
Table 4.3.6 Estimation Results from SMEEquation 3 (Swedish Data) - Labor Force Growth is Used as a Proxy of Labor Growth
R-square=0.1887 21 Observations Dependent Variable= RNONLG ESTIMATED COEFFICIENT T-RATIO -1.188 -1.606 -8.5209 -1.651 621.87 0.8116 0.46563 0.9329 704.12 1.666

VARIABLE NAME RGDPG URBANR FDWPART TIME CONSTANT

P-VALUE 0.108 0.099 0.417 0.351 0.096

Table 4.3.6 is the regression result of the factors that influence the growth of the non life insurance growth. The estimated coefficients on RGDPG and URBANR are both negative and statistically significant. This means both real GDP growth and urban population growth will lead to a decrease in non life insurance growth. The estimated coefficient on FDWPART is positive but not statistically significant. An increase in women participation ratio will increase the growth of non life insurance. This is consistent with what we found in US and Korean data.

96 The following is the regression result when PG is used as a proxy of labor growth in replacement of LG from Table 4.3.7 to Table 4.3.9.
Table 4.3.7 Estimation Results from SMEEquation 1(Swedish Data) Population Growth is Used as a Proxy of Labor Growth
R-square=0.8855 21 Observations Dependent Variable= RGDPG ESTIMATED COEFFICIENT T-RATIO 0.501 6.590 3.231 0.483 -0.102 -1.579 0.055 1.152 0.059 2.234 -1.820 -6.542 -0.028 -0.947 -8.950 -4.348

VARIABLE NAME INVEST TDPG RLIFG RNONLG REXPORTG FDRGOV TIME CONSTANT

P-VALUE 0.000 0.629 0.114 0.249 0.025 0.000 0.344 0.000

Table 4.3.8 Estimation Results from SMEEquation 2 (Swedish Data) Population Growth is Used as a Proxy of Labor Growth
R-square=0.1793 Dependent Variable= RLIFG ESTIMATED COEFFICIENT -0.098 5.192 -10.448 -0.414 975.830 28.024 0.014 471.740 21 Observations

VARIABLE NAME RGDPG LEXP URBANR FDINF6 FDWPART QDRATIO TIME CONSTANT

T-RATIO -0.094 0.844 -2.026 -0.125 1.375 0.926 0.011 0.654

P-VALUE 0.925 0.399 0.043 0.900 0.169 0.354 0.992 0.513

97
Table 4.3.9 Estimation Results from SMEEquation 3 (Swedish Data) Population Growth is Used as a Proxy of Labor Growth
R-square=0.1884 21 Observations Dependent Variable= RNONLG ESTIMATED COEFFICIENT T-RATIO -1.197 -1.614 -8.628 -1.676 609.020 0.806 0.466 0.941 713.050 1.692

VARIABLE NAME RGDPG URBANR FDWPART TIME CONSTANT

P-VALUE 0.106 0.094 0.420 0.347 0.091

What we can see here is that for Sweden, change from LG to PG almost has no affect on the regression result of the second and third equation in the system. However the results from the first equation (which represents the economic growth function of each country) do have some significant change on estimated coefficients. First, when LG is used, the estimated coefficient on LG is much more significant than that on PG when PG is used, which shows that LG is a better proxy of labor growth. Change from LG to PG does not alter any sign on the estimated coefficients. It only changes the statistically significance on some estimated coefficients. This is similar with the result we found in the Korean data. However, when PG is used in the economic growth function, the statistical significance on the estimated coefficient of life insurance growth increased. But the statistical significance on the estimated coefficients of non life insurance growth decreased. This is different with what we have seen in the Korean data. Here we have to acknowledge that the second and third equations show very low explanation power. However, I have included all the variables which are available.

98 From the analysis of the Swedish data, we find that life insurance and non life insurance have opposite effect on the development of the Swedish economy. The development of life insurance harms the economic growth but the development of the non life insurance helps the economic growth. Also, the statistical support for non life insurances role in helping the economic growth does not change by the empirical methods we employed.

99

IV. Analysis of the German Data

i) Unit Root Test Results
Table 4.4.1 Unit Root Test Results (German Data)

Variables WPART INF LEXP RGDPG PG LG RLIFG RNONLG RGOV INVEST DRATIO URBANR REXPG RIMPG

PP Test Statistics -1.79 -2.19 -2.92 -3.97* -1.53 -4.03* -4.76* -4.35* -3.33* -3.03 -0.12 -1.34 -5.24* -5.27*

DF Test Statistics -1.64 -2.62 -2.84 -2.62 -2.01 -3.48* -4.77* -3.90* -2.8 -2.59 -0.7 -2.34 -3.02 -3.27*

Note: * shows that unit root test can reject the null hypothesis at 90% significant level.

For the variables that the unit root test can not reject the null hypothesis at 90% significant level, the first difference is taken and a unit root test on the first difference of the variable is conducted in Table 4.4.1. These include the following variables: population growth, dependant ratio, woman participation in the labor force, expected inflation, life expectancy ratio, urban population ratio in the whole population and investment ratio. But the unit root test on the first difference of population growth, dependant ratio, urban population ratio and expected inflation concludes that a unit root may still be

100 present in the first difference of the above variables. Then the unit root tests on the second difference of the above variables are conducted. But the unit root test of the second difference of population growth and dependant ratio still can not reject the presence of unit root at 90% significance level. Therefore the third difference of both variables is taken and the unit root test is conducted on them. The test shows that both can reject the null at 90% significance level.

ii) Simple OLS Regression

The following is the estimated equations. 1.
RGDPG = 0 + 1TDPG + 2 FDINVEST + 3 RLIFG + 4 RNONLG + 5 RIMPG

+ 6 REXPG + 7 RGOV + 8 D + 8TIME +

2. RGDPG = 0 + 1 LG + 2 FDINVEST + 3 RLIFG + 4 RNONLG + 5 RIMPG + 6 REXPG + 7 RGOV + 8 D + 8TIME + The estimated coefficients and the t statistics based on the Newey West estimator are demonstrated in Table 4.4.2.

101

Table 4.4.2 Estimation result from OLS (German Data) Population Growth is Used as a Proxy of Labor Growth
R-square=0.7962 28 Adj R-square=0.6943 Observations Dependant Variable=RGDPG ESTIMATED COEFFICIENT T-RATIO P-VALUE 0.139 0.526 0.605 -9.341 -1.038 0.313 -0.632 -3.927 0.001 0.319 4.077 0.001 0.745 3.649 0.002 -0.725 -3.635 0.002 -2.270 -6.292 0.000 -1.354 -1.254 0.226 -0.058 -0.988 0.336 52.448 6.244 0.000

VARIABLE NAME FDINVEST TDPG RLIFG RNONLG RIMPG REXPG RGOV D TIME CONSTANT

The estimated coefficient on TDPG is negative, although it is not statistically significant. This is contrary to economic theory. This shows that population growth is really not a proper proxy of labor growth for German data. Although the estimated coefficient on FDINVEST shows the expected sign, its statistical significance level is really low. The estimated coefficient on RLIFG is negative and the estimated coefficient on RNONLG is positive. Both of them show strong statistical significance. This is similar to what we found in the Swedish data--real life insurance shows a negative contribution to the real GDP growth. The estimated coefficient on RIMPG is positive and is statistically significant. But the estimated coefficient on REXPG is negative and statistically significant. This means import growth contribute to economic growth but export growth harms economic growth in German economy. As in all the other three countries, higher government expenditure ratio harms economic growth, which is shown by the negative and statistically significant estimated coefficient on RGOV. D is a

102 dummy variable which define the period after the merge of East and West Germany. The estimated coefficient on D is negative, which shows the merge may require a period of adjustment and hinders economic growth.
Table 4.4.3 Estimation Results from OLS (German Data) Labor Forth Growth is Used as a Proxy of Labor Growth
R-square=0.8935 28 Adj R-square=0.8402 Observations Dependant variable=RGDPG ESTIMATED COEFFICIENT T-RATIO P-VALUE 0.786 3.224 0.005 0.323 5.311 0.000 -0.092 -0.583 0.567 0.097 1.042 0.311 0.151 0.942 0.359 -0.111 -0.719 0.481 -1.074 -3.571 0.002 -1.030 -1.370 0.188 -0.015 -0.332 0.744 24.630 3.711 0.002

VARIABLE NAME FDINVEST LG RLIFG RNONLG RIMPG REXPG RGOV D TIME CONSTANT

In Table 4.4.3, the estimated coefficients on both FDINVEST and LG are positive and statistically significant when TDPG is replaced by LG. This is consistent with economic theory. Therefore, LG is a much better proxy of labor growth than PG. Like in the first regression, life insurance showed a negative contribution to GDP growth while non life insurance showed a positive contribution to GDP growth. But neither of them has shown strong statistical significance. The signs of the estimated coefficients on both RIMPG and REXPG remain the same but with much lower statistical significance level. The estimated coefficient on RGOV remains the same sign and similar statistical significance level as shown in the first equation.

103 The following part shows the results from the simultaneous equation system. Table 4.4.4 to Table 4.4.6 show the results when LG is used in the first equation; while Table 4.4.7 to Table 4.4.9 show the results when TDPG is used in the first equation.

iii) Simultaneous Equations

1. RGDPG = 0 + 1 FDINVEST + 2 LG (/ TDPG ) + 3 RLIFG + 4 RNONLG +

5 RIMPG + 6 REXPG + 7 RGOV + 8 D + 9 time + 1

2. RLIFG = 0 + 1 RGDPG + 2 FDLEXP + 3 SDURBANR + 4 SDINF +

5 FDWPART + 6TIME + 2
3. RNONLG = 0 + 1 RGDPG + 2 SDURBANR + 3 FDWPART + 4TIME + 3

Table 4.4.4 Estimation Results from SMEEquation 1 (German Data) - Labor Force Growth is Used as a Proxy of Labor Growth
R-square=0.8884 28 Observations Dependent Variable= RGDPG ESTIMATED COEFFICIENT T-RATIO 0.731 2.538 0.307 3.398 -0.207 -0.900 0.152 0.959 0.156 0.835 -0.118 -0.635 -1.095 -2.203 -0.602 -0.577 -0.046 -0.690 25.894 2.238

VARIABLE NAME FDINVEST LG RLIFG RNONLG RIMPG REXPG RGOV D TIME CONSTANT

P-VALUE 0.011 0.001 0.368 0.338 0.404 0.526 0.028 0.564 0.490 0.025

Compared the results in Table 4.4.4 with the result from the single OLS regression, all the signs of the estimated coefficients remain the same, although the statistical significance levels change for some variables. The statistical significance of the

104 estimated coefficient on RLIFG increases compared with the OLS result. We still have the same conclusion -- expansion in the life insurance sector shows negative effect on the GDP growth, although the expansion in the non life sector shows positive contribution to the economy growth. This is similar with what we found in the Swedish economy. The statistical significance of the estimated coefficient on D decreases a lot compared with the OLS result.
Table 4.4.5 Estimation Results from SMEEquation 2 (German Data) - Labor Force Growth is Used as the Proxy of Labor Growth
R-square=0.6126 28 Observations Dependent Variable= RLIFG ESTIMATED COEFFICIENT T-RATIO 0.080 0.465 -0.213 -0.074 -3.860 -0.986 1.595 2.488 -58.779 -0.477 3.255 2.646 -8.756 -0.776 -0.316 -4.399 9.652 9.665

VARIABLE NAME RGDPG FDLEXP SDURBANR SDINF FDWPART D QDRATIO TIME CONSTANT

P-VALUE 0.642 0.941 0.324 0.013 0.633 0.008 0.438 0.000 0.000

Table 4.4.5 shows the regression result of the factors that affect the growth of life insurance in force. The estimated coefficient on RGDPG is positive. This means an increase in real GDP growth will lead to an increase in the growth of life insurance in force. But the statistical relationship is not significant here. Life expectancy almost has no influence on the growth of life insurance in force, which is shown by a negative yet not statistically significant estimated coefficient on FDLEXP. The estimated coefficient on SDURBANR is negative. This means more population living in the urban area decrease the growth of life insurance in force. Strangely, here we see that high expected

105 inflation does not decrease the growth of life insurance in force but rather increases it. This is reflected by the positive estimated coefficient on SDINF with strong statistical significance. When there are more women joining in the labor force, the growth of life insurance in force decreases. This is consistent with our expectation. After the East and West Germany merged, there is higher growth in life insurance in force growth. This is reflected in the positive sign before D with a very strong statistical significance. Probably, there is much potential demand in East Germany which was not met before the merge. But dependant ratio shows a negative relationship with the growth of life insurance in force. This is not consistent with the expectation of economic theory.
Table 4.4.6 Estimation Results from SMEEquation 3 (German Data) - Labor Force Growth is Used as a Proxy of Labor Growth
R-square=0.3565 28 Observations Dependent Variable= RNONLG ESTIMATED COEFFICIENT T-RATIO 0.788 2.336 -0.757 -0.096 -67.131 -0.282 -1.804 -0.789 -0.041 -0.308 2.765 1.746

VARIABLE NAME RGDPG SDURBANR FDWPART D TIME CONSTANT

P-VALUE 0.019 0.923 0.778 0.430 0.758 0.081

Table 4.4.6 is the regression result from the third equation. It shows the factors that influence the growth of non life insurance. The estimated coefficient on RGDPG is positive and statistically strong. This means economic growth increases non life insurance growth for German economy. The estimated coefficients on SDURBANR, FDWPART and D are all negative but not statistically significant. More population living in urban area and more woman participation in labor force will lead to a decrease in non life insurance growth. The negative sign of the estimated coefficient on D shows that

106 after the merge of two economies, the growth for the demand of non life insurance decreases. Table 4.4.7 to Table 4.4.9 are the regression results when PG is used as a proxy of labor growth.
Table 4.4.7 Estimation Results from SMEEquation 1 (German Data) Population Growth is Used as a Proxy of Labor Growth
R-square=0.5141 Dependent Variable= RGDPG ESTIMATED COEFFICIENT 0.105 0.348 -0.547 0.773 0.553 -0.508 -1.580 -0.108 -0.023 35.445 28 Observations

VARIABLE NAME FDINVEST TDPG RLIFG RNONLG RIMPG REXPG RGOV D TIME CONSTANT

T-RATIO 0.328 0.027 -2.463 4.112 3.782 -3.500 -2.917 -0.069 -0.242 2.820

P-VALUE 0.743 0.978 0.014 0.000 0.000 0.000 0.004 0.945 0.809 0.005

Table 4.4.8 Estimation Results from SMEEquation 2 (German Data) Population Growth is Used as a Proxy of Labor Growth
R-square=0.5797 Dependent Variable= RLIFG ESTIMATED COEFFICIENT 0.348 0.287 -5.146 1.770 -177.200 3.206 -16.817 -0.314 9.156 28 Observations

VARIABLE NAME RGDPG FDLEXP SDURBANR SDINF FDWPART D QDRATIO TIME CONSTANT

T-RATIO 1.570 0.091 -1.227 2.555 -1.237 2.465 -1.355 -4.131 8.411

P-VALUE 0.117 0.927 0.220 0.011 0.216 0.014 0.175 0.000 0.000

107
Table 4.4.9 Estimation Results from SMEEquation 3 (German Data) Population Growth is Used as a Proxy of Labor Growth
R-square=0.3069 28 Observations Dependent Variable= RNONLG ESTIMATED COEFFICIENT T-RATIO 0.39699 0.9851 5.4046 0.8113 174.87 0.6769 -1.2815 -0.552 -6.37E-02 -0.4719 3.2233 1.953

VARIABLE NAME RGDPG SDURBANR FDWPART D TIME CONSTANT

P-VALUE 0.325 0.417 0.498 0.581 0.637 0.051

For the German economy, when PG is used in the economic growth function in stead of LG, the results for all three equations have great change. Although there is no sign change in the estimated coefficients in the first equation, there is some sign change in the estimated coefficients of the second and the third equation. When PG is used in the economic growth equation, the regression results of the first equation shows that there is almost no statistical relationship between PG and GDP growth. However, when LG is used in the equation, its estimated coefficient has very strong statistical significance. Therefore, LG is a better proxy of labor growth than PG. Very interestingly, when PG is used in the economic growth equation, the statistical significance level on the estimated coefficients of both life and non life insurance growth increased. However, this may not be a true picture. It is due to the inability of PG to explain economic growth. We find very similar empirical results from the German data as from the Swedish data concerning the role of the insurance sectors on economic growth. From the German data, life insurance development also harms economic growth while non life insurance helps economic growth. But the statistical significance for the evidences in either life insurance or non life insurance is not very strong.

108 In the following part, we will use a fixed effect model to pool all four countries data together. Then we will compare the result from the fixed effect model with what we have from OLS and simultaneous equations.

109

V. Fixed Effect Model

The fixed effect model allows country differences which are reflected in the group specific constant. Because I have an unbalanced panel data set, I used OLS estimation with four dummy variables to count for the difference in the four countries. In the original data, there is a dummy variable for the German economy to show the different periods before and after the merge of East and West Germany. To make the fixed effect model workable, I choose two ways to estimate. First, I just ignore the period after merge. This gives a smaller panel data set. Both PG and LG are used as a proxy of labor growth. The estimated results for this one are shown in Table 4.5.1 and Table 4.5.2. Second, I ignore the two different periods for Germany and just put all available data of Germany in the panel data. Of course, in this way, the two periods of German data are treated as the same. The estimated results are shown in Table 4.5.3 and Table 4.5.4.

110

Table 4.5.1 Estimation Results from FEM (Includes German Data before the Merge) - Population Growth is Used as a Proxy of Labor Growth
R-square=0.8596 VARIABLE NAME PG INVEST RNONLG RLIFG REXPG RGOV TIME US KOREA GERMANY SWEDEN 95 Adj R-square=0.8429 Observations Dependent Variable= RGDPG ESTIMATED COEFFICIENT T-RATIO -0.009 -0.096 0.071 7.923 0.095 4.010 0.004 1.142 0.045 2.359 -0.177 -6.254 -0.037 -2.009 2.998 6.578 5.576 9.324 5.838 8.078 0.574 1.333

P-VALUE 0.924 0.000 0.000 0.257 0.021 0.000 0.048 0.000 0.000 0.000 0.186

Table 4.5.2 Estimation Results from FEM (Includes German Data before the Merge) - Labor Force Growth is Used as the Proxy of Labor Growth
R-square=0.8620 VARIABLE NAME LG INVEST RNONLG RLIFG REXPG RGOV TIME US KOREA GERMANY SWEDEN 95 Observations Adj R-square=0.8456 Dependent Variable= RGDPG ESTIMATED COEFFICIENT T-RATIO 0.213 1.226 0.068 7.435 0.087 3.540 0.004 1.149 0.040 2.106 -0.178 -6.360 -0.030 -1.563 2.526 4.364 5.233 7.989 5.752 7.991 0.487 1.128

P-VALUE 0.224 0.000 0.001 0.254 0.038 0.000 0.122 0.000 0.000 0.000 0.263

111

Table 4.5.3 Estimation Results from FEM (Includes all German Data) - Population Growth is Used as a Proxy of Labor Growth
R-square=0.7656 VARIABLE NAME PG INVEST RNONLG RLIFG REXPG RGOV TIME US KOREA GERMANY SWEDEN 106 Adj R-square=0.7409 Observations Dependent Variable= RGDPG ESTIMATED COEFFICIENT 0.079 0.069 0.113 0.004 0.023 -0.190 -0.025 2.763 5.563 6.472 0.573 T-RATIO 0.740 6.005 3.717 0.761 0.995 -5.215 -1.180 5.023 7.327 7.047 1.051 P-VALUE 0.461 0.000 0.000 0.449 0.322 0.000 0.241 0.000 0.000 0.000 0.296

Table 4.5.4 Estimation Results from FEM (Includes all German Data) - Labor Force Growth is Used as a Proxy of Labor Growth
R-square=0.8675 VARIABLE NAME LG INVEST RNONLG RLIFG REXPG RGOV TIME US KOREA GERMANY SWEDEN 106 Adj R-square=0.8535 Observations Dependent Variable= RGDPG ESTIMATED COEFFICIENT 0.375 0.068 0.074 0.005 0.035 -0.178 -0.025 2.213 5.062 5.759 0.396 T-RATIO 8.600 7.904 3.202 1.185 2.013 -6.510 -1.568 5.341 8.890 8.277 0.972 P-VALUE 0.000 0.000 0.002 0.239 0.047 0.000 0.120 0.000 0.000 0.000 0.334

From the fixed effect model, we can see that in both case the estimated coefficients on LG have strong statistical significance. But the estimated coefficients on PG do not show strong statistical significance. Therefore, LG is a better proxy of labor growth than PG.

112 In all the cases, all the estimated coefficients on RNONLG are positive and have strong statistical significance. This is consistent with the results from simultaneous equationnon life insurance growth contributes to economic growth. Although the estimated coefficients on all RLIFG are positive too but there is no strong statistical significance. This means that life insurance also shows positive contribution to economic growth from the fixed effect model. This is different with the result from the simultaneous equation method. The simultaneous equation method shows that US and Korean data support the idea that life insurance growth contribute to economic growth and Swedish and German data shows that life insurance growth hinders economic growth. The fixed effect model hides the differences here by forcing a same estimated coefficient on life insurance growth. Actually, because for the German economy life insurance growth has a negative contribution to economic growth, when more German data are included in the fixed effect model, the significance level on estimated coefficient of life insurance growth decreased. Also, the constant terms, which accounts for country differences, have estimated coefficients with strong statistical significance. This shows that there are differences in the economic growth equation across countries. What we can conclude here is that the fixed effect model supports the conclusion that non life insurance has positive contribution to economic growth in these four countries. Yet it does not reveal the difference of non life insurances role from country to country. The difference of growth equation across countries shows that individual estimation of simultaneous equation for each country is a better econometric method.

113 Note: some time series are non stationary for some countries but are stationary for other countries. But for the fixed effect model, we have to stack the same time series together from different countries. For example, PG is stationary for US but not for all the other three countries. FDPG is stationary for Korea and TDPG is stationary for Germany and Sweden. To solve the problem, I just stack PG, FDPG and TDPG for the corresponding countries, although in the regression results it uses the term PG in the table. I also multiply FDPG and TDPG with 10 and 100 to make the numbers closer to those in PG.

114

VI. Summary and Comparison of the Results across the Four Countries

Table 4.6.1 Effect of Insurance Growth on Economic Growth

OLS COUNTRY US(PG) US(LG) Korea (FDPG) Korea (LG) Sweden (TDPG) Sweden (LG) Germany (TDPG) Germany (LG) VARIABLE NAME RLIFG RPCLG RLIFG RPCLG RLIFG RNONLG RLIFG RNONLG RLIFG RNONLG RLIFG RNONLG RLIFG RNONLG RLIFG RNONLG ESTIMATED COEFFICIENT 0.159 0.086 0.188 0.071 0.006 0.089 0.006 0.049 -0.083 0.058 -0.077 0.062 -0.632 0.319 -0.092 0.097 TRATIO 3.223 3.131 3.020 2.284 3.968 2.702 4.558 1.918 -2.905 2.042 -2.782 2.166 -3.927 4.077 -0.583 1.042 PVALUE 0.004 0.004 0.006 0.031 0.002 0.019 0.001 0.079 0.012 0.062 0.016 0.049 0.001 0.001 0.567 0.311 Simultaneous Equations ESTIMATED COEFFICIENT 0.130 0.250 0.180 0.271 0.002 0.088 0.004 0.049 -0.102 0.055 -0.070 0.074 -0.547 0.773 -0.207 0.152 TRATIO 1.198 3.262 1.596 3.855 0.518 1.816 0.974 0.979 -1.579 1.152 -1.060 1.698 -2.463 4.112 -0.900 0.959 PVALUE 0.231 0.001 0.110 0.000 0.605 0.069 0.330 0.328 0.114 0.249 0.289 0.089 0.014 0.000 0.368 0.338

Table 4.6.1 summarizes the results from both OLS and simultaneous equation model for the data of all four countries. The left hand side gives us the result from OLS (with both PG and LG as labor growth). We can see that all four countries data support the theoretical model that non life insurance (or PCL) insurance growth can promote economic growth. But not all countries data support the theoretical model that life insurance growth may contribute to economic growth. US and Korean data show that life insurance can contribute to economic growth. However, Swedish and German data show that life insurance growth may hinder economic growth.

115 Comparing the results from OLS and simultaneous equations, we can see that OLS may over estimate or under estimated the coefficients on RLIFG or PCLG/NONLG, if we assume that the simultaneous equation method provides a more precise estimation. Here we can also see that the sizes of the contribution of the insurance sector to the economic growth are different across countries. The US insurance sector seems having the biggest influence on economic growth in these four countries. For example, from the estimated coefficient on life insurance growth in the output growth function, 1% increase in life insurance growth rate will lead to around .15% increase in real GDP growth for the US economy. The German data and the Swedish data show a little bit smaller influence than the US data. The Korean data shows the lowest level of influence. For example, 1% increase in life insurance growth only leads to .006% increase in real GDP growth. The size of the contribution of the insurance sector seems related with the degree of the development of an economy. We will compare the effects of economic growth on insurance growth for all four countries. It is shown in Table 4.6.2.
Table 4.6.2 Effect of Economic Growth on Insurance Growth
VARIABLE NAME US Korea Sweden Germany US Korea Sweden Germany RGDPG RGDPG RGDPG RGDPG RGDPG RGDPG RGDPG RGDPG ESTIMATED COEFFICIENT T-RATIO DEP=RLIFG 0.061 0.299 -4.068 0.599 -0.092 0.088 0.080 0.465 DEP=PCLG/NONLG 0.887 1.852 0.834 1.808 -1.188 1.606 0.788 2.336 P-VALUE 0.765 0.549 0.930 0.642 0.064 0.071 0.108 0.019

116 The results in Table 4.6.2 are based on the second and the third equation of the simultaneous equations. Here we only show the results from the estimation when the better proxy of labor growth is used. For the life insurance sector, two countries data (US and Germany) show that economic growth may promote life insurance growth. One countrys data (Korea) show that economic growth may hinder life insurance growth. There is almost no statistical relationship between economic growth and insurance growth for the Swedish data. We can see here that none of the estimated coefficients on RGDPG is statistical significant in the life insurance demand growth function. As for the non life insurance sector, three countries data support the idea that economic growth may contribute to non life insurance growth. Only Swedish data show that economic growth may lower non life insurance growth. Also, we can see all the estimated coefficients on RGDP in the non life insurance growth function show strong statistical significance. From the four countrys data, it seems that non life insurance more rely on economic growth than life insurance. Here we may provide a couple of reasons why life insurance growth does not promote economic growth as we illustrated in the theoretical framework. First, in the theoretical model, we assume that when there is more long term fund in an economy, there is more resource available for technological innovation. However, it also depends on the institution itself to channel those long term savings to effective technological innovation. If life insurance companies only invest those savings in short term, it can not fulfill the role we described in the theoretical model. As we mentioned earlier that different countries may have different laws governing the operation of

117 contractual savings institutions. The development of contractual savings institution can not promote economic growth if they are only allowed to invest in short term securities or government bond. One way to see whether this is true is to implore the assets holding components of life insurance companies in those countries. Second, it seems that the two countries where life insurance does not contribute to economic growth are two European countries. The social welfare system of European countries may be different than other countries. If they already provide protection like life insurance, life insurances effect may not work as predicted. If this is the case, we can infer that private life insurance may not be a better choice over public social welfare system which also provides the similar function. Both of the above reasons require further research work and we will not do that in this dissertation.

118

Chapter 5 Conclusions and Recommendations for Future Research

The importance of the insurance sector in an economy has been well recognized. This is reflected by the fact that the total world insurance premiums have a significant portion in total world GDP. In this dissertation, we start with the hypothesis that insurance industry can promote economic growth by 1) setting up theoretical models to illustrate how insurance can help economic growth and 2) testing the empirical relationship between insurance growth and economic growth by using four countries data. In the theoretical framework, we focus on the risk sharing role of property/liability insurance and long term capital accumulation role of life insurance. By providing the instrument for risk sharing, property/liability insurance sector can help economic agents making a social optimal production choice. By providing savings plans to increase the utility of the economic agents, contracting savings institution can provide more long term capital. Both life and non life insurance can contribute to economic growth by promoting technological progress. Then we set up empirical models to test our theoretical models. Data from four countries have been usedUS, Korea, Sweden, and Germany. Three empirical methods are employedordinary least square, simultaneous equations and fixed effect model. Empirical study suggests that property/liability insurance have positive contribution to economic growth but life insurance may have positive or negative contribution to economic growth depending on each countrys situation. The following part is the

119 summary of the research. Then we discuss the limitation of the research and make suggestions for future research.

I. Summary of theoretical models

First, we illustrate the economic role of property/liability insurance. We start the model by assuming two different types of technology are available. Risk is introduced into the model by assuming one type of technology is risky technology in the sense that investment in this technology may lead to zero output. The other type of technology is called safe technology in the sense that it will produce a certain output. However, the risky technology is more productive than the safe technology because it yields a higher expected output given the same amount of labor and capital. Then we specify an overlapping two generation model with individuals endowed with a specific amount of labor to decide which technology to invest in. Individuals are risk averse and maximize their utility function. When there are no institutions to help diversifying the risk, utility-maximizing-individuals will choose the safe technology. Because the risky technology has a higher expected output than the safe technology. The risky technology should be a social optimal choice. This illustrates that when there is uncertainty associated with economic outcome and economic agents are risk averse at the same time, individual optimization does not lead to social optimization. Therefore, there is a need for risk sharing institutions. The insurance industry is such a kind of institution. Then we introduce the insurance sector into the economy. This allows economic agents to insure against the loss when the risky technology is not successful

120 by paying a certain amount of money called premium. The model shows that now individuals will choose the risky technology by maximizing their utility. Now the individual optimization is the same as social optimization. Here insurance is illustrated as a door, the open of which allows for a choice which is non-obtainable or appealing to economic agents because of the risk they have to bear. In economic growth model, we know that long run economic growth comes from technology growth. Rather than being innovations themselves, insurance serves as an institution to help those innovations being carried out into economic life. Second, the life insurance sector is studied. Pension funds and life insurance companies are called contractual savings institutions. They are different with banks because they can provide long term savings plans or protection against financial loss against premature death through contracts. Here we illustrate the role of life insurance in economic growth by showing how contractual savings institutions can shift short term capital to long term capital. The main illustration here is to compare banking with contractual savings institutions. The latter one holds more illiquid assets than the former one. Illiquid assets can be used to finance long term project which can yield higher return. In this part, we start with the argument that there is a need for contractual savings institutions. Economic agents facing the uncertainty of life expanse are assumed to work and save for the longest possible life expanse. Their savings decisions are based on a three-period utility function. Individuals maximize their utility function by allocating the same amount of real consumption on each period. Then a contractual savings institution is introduced into the economy. The contractual

121 savings institution has knowledge of the life expectancy probability distribution. This makes economic agents can buy policy to exchange their wage for life long annuity. By doing so, they can increase their maximized utility. Then the savings structure of the economy without contractual savings institutions are compared with the ones with contractual savings institutions. The presence of contractual savings institution in the economy increase both long term and short term investment in the economy. Then a discussion about insurance growth and technological progress is presented. It illustrates how the insurance sector can promote technological innovations by providing either more resources or lowering the cost of innovation.

II. Summary of Empirical Findings

We examined four countrys data to see whether there is empirical support of the theoretical illustration about the role of insurance in economic growth in the second chapter. The four countrys data available at the beginning of this research are from US, Korea, Sweden, and Germany. First, we set up the growth model with insurance and other factors to explain productivity growth (or technological growth). Based on this model, the ordinary least square is used to estimate the effect of insurance growth on economic growth. As illustrated in chapter 2, there are two different effects of the insurance sector on economic growthrisk-diversifying and long term capital accumulation. These two effects are separately associated with property/liability insurance and life insurance. Therefore, property/liability insurance growth and life insurance growth are separated and put into the growth function individually. The result shows that all countries data

122 support that property/liability insurance can contribute to economic growth. On the other hand, life insurance growth has mixed results from these four countrys data. The data of two countries show that life insurance growth has contribution to economic growth while the data of the other two countries show that life insurance growth has hindered economic growth. The two countries with negative correlation are European countries. It is suspected that the social welfare system in European countries may already provide the role of life insurance companies. The other possible reason is that life insurance companies in these two countries may not channel long term capital into technological innovations effectively as the theoretical model illustrated. But further study is needed to make a specific conclusion on this. Then we set up simultaneous equations for each country to further study the effect of insurance growth on economic growth. In Chapter 2 we argued that insurance growth will contribute to economic growth. On the other hand, economic growth may also demand for insurance growth. There are interactions between insurance growth and economic growth. If this is true, ordinary least square estimation will lead to biased estimation. Simultaneous equations allow for the interaction between economic growth and insurance growth. Therefore, it will be a more proper empirical model to use. The result from simultaneous equation is compared with the one from single equation estimation. The result concerning the role of insurance growth on economic growth remains unchanged. This means all the estimated coefficients on RLIFG and RPCLG/RNONLG in the growth function remain the same signs as in the OLS estimation. However, the statistical significance level of most estimated coefficients on RLIFG and RPCLG/RNONLG is lowered.

123 Also, we found that economic growth has more effect on property/liability insurance growth than on life insurance growth. Through simultaneous equation analysis, the factors that affect property/liability insurance and life insurance are identified. The significance of those factors is different across countries. Then the results from a fixed effect model is presented. This model combined all countries data together and estimate a single equation growth model. It allows for different constant terms for different countries. The result also supports the fact that property/liability insurance can promote economic growth. But combination of the data forced a same estimated coefficient on life insurance growth. Therefore, the result can not show the difference of life insurance actual effect in different countries. However, by employing different econometric methods, we can test the robustness of the model.

III. Limitation of the research

In this research, I try to give some theoretical explanation to support the argument that insurance growth will promote economic growth. In addition to the theoretical explanation, I hope to support the argument with empirical study. However, I have to acknowledge that there are some deficiencies in my research. I will discuss them in this section. First, we assume that the insurance market is in a very ideal state. There is no loaded premium to cover the extra cost of insurance companies other than pure loss. Also, the model does not account for those issues of moral hazard and adverse selection, which may also add more cost to administration cost of the insurance

124 industry. When these issues are significant, the theoretical results developed in the second chapter may be altered. Developing countries with under-developed insurance market may face such issues. Second, when we study the role of life insurance on economic growth, we only focus on its role as contractual savings institutions. Life insurance can also protect against interruption on human capital investment due to premature death of the main breadwinner. Life insurance may provide uninterrupted human capital investment and therefore increase the quality of human capital stock of an economy. Actually, this is a very important role of life insurance, which can be examined in the context of the growth model but ignored in this dissertation. Third, the technological progress model is a very simplified model. For example, is a constant term which represents the unit cost of innovation. However, in the real world situation, the innovation process is much complicated. It is hard to use a single term to represent the innovation cost. This will also affect the conclusion on the insurance sectors contribution to economic growth, which is linked to . Fourth, when I started this dissertation, there were only four countries data available for this research. This can not represent all cases in general. Insurance development may be very country-specific. Examining a broad range of data can help test the robustness of the theoretical models. Fifth, only three variables are treated as endogenous in the simultaneous equation systemGDP growth, Life insurance growth, PCL/Non life insurance growth. All the other variables are treated as exogenous. Capital growth should be another important endogenous variable but was treated as exogenous in the

125 simultaneous equations. The reason is that I could not obtain the data of capital growth in each country. Investment to GDP ratio is used as a proxy of capital growth in the GDP growth function. Sixth, there are several reasons are proposed for the observations that life insurance does not contribute to economic growth for two European countries. There is no further exploration to support the reasons proposed.

IV. Suggestion for Future Research

Based on the above limitations, I have the following suggestions for future research. In future models, extra cost can be loaded to the pure premium charged by the insurance sector. By allowing extra cost above pure premium, the limit of the benefit of an insurance industry can be developed. For some developing countries, I suspect there would be empirical evidence to support that property/liability insurance can not promote economic growth. In those markets, the insurance industry may not be very efficient. And its role in economic growth may be undermined. It would be a very interesting topic to study. Second, more countries data should be included in the empirical study. As more and more countries data are studied, we may observe more diversified results from different countries. Studying the difference may lead to some revision of the theoretical arguments presented here. We observe a wide range of development in the development of the insurance institutions in the world. For the developed countries, such as US, there is well developed insurance sector. In contrast, there is very under developed insurance market in developing countries, such as in some African

126 countries. When more countrys data are available, the reason for the difference we observed may be a very interesting topic to study on. In addition to that, in this model we do not examine the cost of the insurance institutions. Policy makers may be concerned whether it is good to subsidize the insurance companies to promote its development. If there are some positive externalities from the development of the insurance sector, it may be worthwhile for the government to initiate the establishment of such institutions. Third, examine how life insurance could help develop the human capital in an economy. It seems the model would be more robust if we include this into the growth model. Fourth, obtain the capital growth time series for each country and put this into the simultaneous equations as an endogenous variable to see whether this will change the result we have here.

127

Appendix A
Solve for q2
Max ~ (q 1 *w + [ * L1 * q 2 * w] + C r 2 )

Conditional on [ q w + q 2 L1 w + ( A 1 + ) L1 (1 q1 q 2 ) w ] = Max{ 1 *
+ [q1 w + q 2 L1 w + ( 1) L1 (1 q1 q 2 ) w ] * (1 )}

q1 = 0

So the maximization problem becomes

Max [q 2 L1 + ( A 1 + ) L1 (1 q 2 )] w [q 2 L1 + ( 1) L1 (1 q 2 )] w * + * (1 )

Take first order derivative and set it equal to zero.

1 1 1 1 1 FOC [q 2 L + ( A 1 + ) L (1 q 2 )] * [L + (1 A ) L ] *

+ [q 2 L1 + ( 1) L1 (1 q 2 )] 1 * [L1 + (1 ) L1 ] * (1 ) = 0 [q 2 L1 + ( A 1 + ) L1 (1 q 2 )] 1 * [L1 + ( A + 1)) L1 ] * = [q 2 L1 + ( 1) L1 (1 q 2 )] 1 * [L1 + (1 ) L1 ] * (1 ) [q 2 L1 + ( A 1 + ) L1 (1 q 2 )] 1 * [( A 1)) L1 ] * = [q 2 L1 + ( 1) L1 (1 q 2 )] 1 * L1 * (1 ) [q 2 L1 + ( A 1 + ) L1 (1 q 2 )] 1 1 = A [q 2 L1 + ( 1) L1 (1 q 2 )] 1

1

q L1 + ( A 1 + ) L1 (1 q 2 ) 1 +1 ) 2 1 =( 1 A q 2 L + ( 1) L (1 q 2 )

1 +1 q 2 + ( A 1 + )(1 q 2 ) 1 +1 Set D=( ) ) =( A q 2 + ( 1)(1 q 2 ) A

q 2 + ( A 1 + )(1 q 2 ) =D q 2 + ( 1)(1 q 2 )

128
A 1 + ( A 1)q 2 =D 1 + q2

D D + Dq 2 = A 1 + ( A 1)q 2
Dq 2 + ( A 1)q 2 = A 1 + + D D

q2 =

( A 1) + (1 D) + D (1 D) ( D 1) = 1+ = 1 ( D + A 1) D + A 1 D + A 1

Here we can approve that 0 < q2 < 1 .

Note that D = ( 1 ) +1 = ( A ) +1 1 1

Also, we have specified that A > 1 so that the expected output level from the risky technology is higher that the output level from the safe technology.
A > 1 1 <1 A

and > 1 1 > 0

+1

D >1 0 <

( D 1) ( D 1) < 1 0 < 1 < 1 0 < q2 < 1 D + A 1 D + A 1

129

Appendix B
The following is a reproduced illustration why banks (which are subject to runs) may choose only to invest in short term project while contractual savings institutions will choose to invest in long term project. According to Impavido and Musalem (2001), contractual savings institutions have comparative advantages in financing long term projects compared with banks because they are not subject to bank runs. They give out a numerical example. Here I will borrow the numerical example to illustrate why contractual savings institutions are more willing to finance long term projects than banks. First, according to Diamond and Bybvig (1983), banks are subject to runs. But due to the special character of contractual institutions, they are not subject to runs. There are two kinds of projects to invest for banks--long term projects which give a higher rate of return and short term projects which give a lower rate of return. Let i d , i s , i l denote the rate of return required by depositors, the rate of return of short term projects and the rate of return of long term projects specifically. Banks are subject to runs with probability p. Banks will choose to invest either in short term projects and long term projects depending on the expected profits. The expected profits from short term projects and long term projects are illustrated in Table AP-1 and Table AP-2.

130

Table AP-1 Profits from Short Term Projects Investment in Short Term project No bank runs With bank runs Expected profit Profit Probability

(1 + i s ) 2 (1 + i d ) 2 (1 + i s ) (1 + i d )
(1 + i s ) 2 (1 + i d ) 2 * (1 p ) + (i s i d ) * p

1-p p

Table AP-2 Profits from Long Term Projects Investment in Long Term Project No Bank Runs With Bank Runs Expected Profit Profit Probability

(1 + i l ) 2 (1 + i d ) 2 -C (1 + i l ) 2 (1 + i d ) 2 * (1 p) C * p

1-p p

Here C denotes the cost of banks runs which lead to bankruptcy. Banks choose to invest in long term projects if only the expected profit from long term projects are higher than profit from short term projects. The condition is that (1 + i l ) 2 (1 + i d ) 2 * (1 p) C * p > (1 + i s ) 2 (1 + i d ) 2 * (1 p ) + (i s i d ) * p To expand to terms on both sides, the condition becomes

131 [2 * i l + (i l ) 2 2 * i d (i d ) 2 ] * (1 p) C * p > [2 * i s + (i s ) 2 2 * i d (i d ) 2 ] * (1 p ) + (i s i d ) * p Which is equivalent to [2 * i l + (i l ) 2 ] * (1 p) C * p > [2 * i s + (i s ) 2 ] * (1 p ) + (i s i d ) * p [2 * i l + (i l ) 2 2 * i s (i s ) 2 ] * (1 p ) C * p (i s i p ) * p f 0 If the above condition is not met, banks will choose to invest in short term projects instead of long term projects. We can see that the higher the cost of bankruptcy (larger C), the higher the probability of bank runs (larger p) and the smaller the difference between long term and short term projects rates of return, the less willingness banks will invest in long term projects. But for contractual savings institutions, they are not subject to runs so that they will always to choose to invest in long term projects. The reason is that (1 + i l ) 2 (1 + i d ) 2 > (1 + i s ) 2 (1 + i d ) 2 .

132

Appendix C
Source of Data Data from the Statistical Abstract of the United States (US): LG, LIFG, PCLG, EDR, RSSG, INF, PRATIO, WPART, LEXP, URBANR, DRATIO The data of Korea are from the following website: http://www.kosis.kr/eng/index.htm except URBANR, LEXP and WPART World Bank Data: URBANR, LEXP, WPART AND DRATIO (for Korea, Germany and Sweden) Penn World Table: Population, INVEST IMF CD: GDP, EXPORT, IMPORT, GOVERNMENT EXPENDITURE (US) OECD STATISTICS WEB: GDP, EXPORT, IMPORT, COMSUMER PRICE, GOVERNMENT EXPENDITURE (GERMANY AND SWEDEN) Statistical Yearbook of Sweden: LIFG, NONLG Statistical Yearbook of Germany: LIFG, NONLG

133

Appendix D
Test the Relevance of the Instruments Regress life insurance growth on all exogenous variables and test the joint significance of the excluded instruments. If the F test can reject the null, the instruments are relevant. Otherwise they are not. The results for each country are presented in the following table. F test statistics US F PValue RLIFG Pg Lg RNONLG Pg Lg 0.532 0.413 0.711 1.066 0.777 0.861 0.704 0.434 0.541 0.497 5.439 7.269 Korea F pValue 0.770 0.742 0.018 0.008 1.152 1.026 1.653 1.222 Sweden F pValue 0.387 0.440 0.227 0.373 3.522 4.295 0.580 1.590 Germany F pValue 0.026 0.012 0.715 0.223

134

Bibliography
Aghion, Philippe, and Peter Howitt. A Model of Growth through Creative Destruction. Econometrica, 60: 323-351. Albouy, Francois-Xavier and Dimitri Blagoutine, 2001, Insurance and Transition Economics: The Insurance Market in Russia, The Geneva Papers on Risk and Insurance, 26(3): 467-479. Arrow, Kenneth J., 1965, Insurance, Risk and Resource Allocation, Reprinted in Dionne, Georges and Scott E. Harrington 1991, eds, Foundations of Insurance Economics: Readings in Economics and Finance, (Boston: Kluwer Academic Publishers). 220-229. Beenstock, Michael, Gerry Dickinson and Sajay Khajuria, 1988, The Relationship between Property-Liability Insurance Premiums and Income: An International Analysis, Journal of Risk and Insurance, 55(2): 259-272. Bencivenga, Valerie R. and Bruch D. Smith. Financial Intermediation and Endogenous Growth. The Review of Economic Studies, Apr. 1991, 58(2): 195-209. Bencivenga, Valerie R. and Bruch D. Smith. Transactions Costs, Technological Choice, and Endogenous Growth, Journal of Economic Theory, 1995, 67:153-177. Carmichael, Benoit and Yazid Dissou. Health Insurance, Liquidity and Growth, Scandinavia Journal of Economics, 2000, 102(2):269-284. Diamond, Douglas W. and Philip H. Dybvig. Bank Runs, Deposit Insurance, and Liquidity, Journal of Political Economy, 1983, 85: 191-206.

135 Enz, Rudolf, 2000, The S-Curve Relation Between per-Capita Income and Insurance Penetration, The Geneva Papers on Risk and Insurance, 25(3): 396-406. Grossman, Gene M., and Elhanan Helpman, 1991, Innvoation and Growth in the Global Economy, (Cambridge, Mass: MIT Press). Impavido, Gregorio and Alberto R. Musalem, Contractual Savings, Stock, and Asset Markets. The World Bank, Policy Research Working Paper Series: 2490, 1999. Impavido, Gregorio, Alberto R. Musalem and Thierry Tressel, Contractual Savings, Capital Market, and Firms Financing Choices. The World Bank, Policy Research Working Paper Series: 2612, 2001. Impavido, Gregorio, Alberto R. Musalem and Thierry Tressel, Contractual Savings Institutions and Banks Stability and Efficiency. The World Bank, Policy Research Working Paper Series: 2751, 2002. Impavido, Gregorio, Alberto R. Musalem and Thierry Tressel, The Impact of Contractual Savings Institutions on Securities Markets. The World Bank, Policy Research Working Paper Series: 2948, 2003. King, Robert G. and Ross Levine, Finance and Growth: Shumpeter Might Be Right, The Quarterly Journal of Economics, 1993, 108: 718-737. King, Robert G. and Ross Levine. Finance, Entrepreneurship, and Growth, Journal of Monetary Economics, 1993, 32: 513-542. Levine, Ross. Financial Development and Economic Growth: Views and Agenda, Journal of Economic Literature, June 1997, 35 (2): 688-726. Levine, Ross. Stock Markets Growth, and Tax Policy, The Journal of Finance, Sep. 1991, 46(4): 1445-1465.

136 Liu, Gordon G., Xiaodong Wu, Chaoyang Peng and Alex Z. Fu, 2003, Urbanization and Health Care in Rural China, Contemporary Economic Policy, 21(1): 11-24. Murphy, Pablo Lopez and Alberto R. Musalem, Pension Funds and National Savings. The World Bank, Policy Research Working Paper Series: 3410, 2004. Musalem Alberto R., Gregorio Impavido, and Mario Catalan, Contractual Savings or Stock Market Development Which Leads?. The World Bank, Policy Research Working Paper Series: 2421, 2000. Outreville, J. Francois, 1996, Life Insurance Markets in Developing Countries, Journal of Risk and Insurance, 63(2): 263-278. Rebelo, Sergio. Long-run Policy Analysis and Long-run Growth. Journal of Political Economy, June 1991, 99 (3): 500-521. Romer, Paul M. Endogenous Technological Change, Journal of Political Economy, 98(5): S71-S102. Saint-Paul, Gilles. Technological Choice, Financial Markets and Economic Development, European Economic Review, 1992, 36: 763-781. Shumpeter, Joseph, 1934, The Theory of Economic Development (Cambridge, Mass: Harvard University Press). Solow, Robert, A Contribution to the Theory of Economic Growth, Quarterly Journal of Economics, 70(1): 65-94. Solow, Robert, Technical Change and the Aggregate Production Function, Review of Economics and Statistics, 39: 312-320. Soo, Hak Hong, 1996, Life Insurance and Economic Growth: Theoretical and Empirical Investigation, PhD Dissertation.

137 Van den Berg, Hendrik, 2009 forthcoming, Economic Growth and Development (McGraw Hill). Ward, Damian and Ralf Zurbruegg, 2000, Does Insurance Promote Economic Growth? Evidence from OECD Countries, Journal of Risk and Insurance, 67(4): 489-506.