A Study on Japanese Public Pension System using Multi Agent Based

Document Sample
A Study on Japanese Public Pension System using Multi Agent Based Powered By Docstoc
					A Study on Japanese Public Pension System using Multi Agent Based Simulation

Masatoshi MURAKAMI Graduate School of Economics, Kansai University Noriyuki TANIDA Department of Economics, Kansai University June 13-15, 2005 at the University of Essex, United Kingdom
1 M. MURAKAMI & N. TANIDA WEHIA 2005

Today’s Talk
1.

Background

(1) Population Projections in Japan The number of people who did not pay pension premium (2) Improvements in this study

2.

Simulation Settings
(1) (2) (3) (4)

3.

Simulations

Basic design Data Calculation formula of pension benefit Model Visualization(1), (2)

(1) Common rule, (2) Rules of each model

4.

Simulation Results
(1) Pension premium fund (2) Monthly pension premium – benefit difference (3) Result of Friedman’s test

5.

Conclusion and Future works
WEHIA 2005

M. MURAKAMI & N. TANIDA

2

Background (1)
(%) 80. 00

P opul on P roj on i Japan ati ecti n The num ber of P eopl w ho di not P ay P ensi P rem i e d on um

Thousand 3500

70. 00

3000

60. 00 2500 50. 00 C hid Popul on (Age 0 to 14) l ati 2000 Popul on of W orki Ages (Age 15 to 64) ati ng 40. 00 Agi Popul on ( Age over 65) ng ati Peopl W ho di not pay pensi prem i e d on um 30. 00 1000 20. 00 1500

10. 00

500

1980 1985 1990 1995 1998 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010

0

(Source) Social Insurance Agency (2001), National Pension Insured Survey, 2001
National Institute of Population and Social Security Research, Population Statistics of Japan 2003. M. MURAKAMI & N. TANIDA WEHIA 2005 3

Background (2)
1. The problem of unpaid pension premiums in Japan
In Japan, Institutional complexity, negligence on the part of the competent authorities and distrust in the system, are caused of the problem of unpaid pension premiums.

2.

The cause of unpaid pension premiums in Japan
In fact, the phenomenon of nonpayment of pension premiums has arisen due to a combination of these factors.

3.

Improvements in this study
We implemented the factor of distrust in public pension system and its propagation into multi agent based simulation model.

M. MURAKAMI & N. TANIDA

WEHIA 2005

4

Today’s Talk
1.

Background
(1) Population Projections in Japan The number of people who did not pay pension premium (2) Improvements in this study

2.

Simulation Settings
(1) (2) (3) (4)

3.
4.

Simulations

Basic design Data Calculation formula of pension benefit Model Visualization(1), (2)

(1) Common rule, (2) Rules of each model

Simulation Results
(1) Pension premium fund (2) Monthly pension premium – benefit difference (3) Result of Friedman’s test

5.

Conclusion and Future works
WEHIA 2005 5

M. MURAKAMI & N. TANIDA

Simulation Settings (1) - Basic Design 17 properties are assigned to each agent and they are divided into 5 categories

(A) People’s environment
(a) (b) (c) (d) Basic property ( sex, age etc.) Income of a worker (career, type of job etc.) Money-margin (resource, spending etc.) The relationship of the agent to public pension system (duration of employment, No.1 to 3 insured)

(B) People’s Consciousness
(e) Whether or not pay premiums (distrust, social norm etc.)
M. MURAKAMI & N. TANIDA WEHIA 2005 6

Simulation Settings (2) - Data 

Marriage

Age-specific marriage rate == Ministry of Health, Labour and Welfare, Vital Statistics 2003


Birth

Fertility rate from 1990 to 2003 == National Institute of Population and Social Security Research, Population


Death

Statistics of Japan 2003




Unemployment
Income
5%

Ministry of Health, Labour and Welfare, Abridged Life Table

Data categorized by academic record, and the number of people at every income level. Weighted average == Ministry of Internal Affairs and Communications Statistics Bureau,

2002 Employment Status Survey

M. MURAKAMI & N. TANIDA

WEHIA 2005

7

Simulation Settings (3) – Pension Benefit
Calculation Formula of Pension Benefit
  premium paid period   



(Source) Social Insurance Agency (1999)

794,500yen 

period of half exemption  2  period of total exemption 1      3  from premium  3 from premium     40 the maximum participat year   12 ion

Ex. Premium Paid Period = 420 (35 years) Income per Month under 90 thousand yen period = 36 ( 3 years) Income per Month under 70 thousand yen period = 24 ( 2 years)

= 750,008 yen ( One Year )
= 62,500 yen ( per month)
M. MURAKAMI & N. TANIDA WEHIA 2005 8

Model Visualization (1)


House Man Woman Boy Girl

To translate the real world to the artificial model

Man

Woman

Child

  

 

Create Man, Woman, Child, House Agent A family has nine cells Implement house agent to get the information and position of each agent in family Define one block ( 27 × 27 cells) as one family line Simulate with 9 family lines

Child House

Family

Family Line

×9
M. MURAKAMI & N. TANIDA WEHIA 2005 9

Model Visualization (2) - Ex. Marriage of Agent
Family Line 1 Family Line 2

•Move Block •Take over Properties

Rest on the cast of the dice

  

Independent child agent (boy) create new house in same family line. Independent child agent (girl) looks for man agent in the other blocks to marry. The cell which she should go is decided on probability.



To translate real world to artificial model by creating agent, agent’s move. - Agent’s birth,Agent’s death, Agent’s divorce etc.
M. MURAKAMI & N. TANIDA WEHIA 2005 10

Today’s Talk
1.

Background
(1) Population Projections in Japan The number of people who did not pay pension premium (2) Improvements in this study

2.

Simulation Settings
(1) (2) (3) (4)

3.
4.

Simulations

Basic design Data Calculation formula of pension benefit Model Visualization(1), (2)

(1) Common rule, (2) Rules of each model

Simulation Results
(1) Pension premium fund (2) Monthly pension premium – benefit difference (3) Result of Friedman’s test

5.

Conclusion and Future works
WEHIA 2005 11

M. MURAKAMI & N. TANIDA

Simulations (1) - Common Rule  Define one turn as one month  Employment or unemployment is decided from the data and occupation are randomly decided every six months (every 6 turn)  Demographic composition in Japan in 1990 (at simulation start)  Fertility rate is changed:
From From From From From 0 to 120 turn 121 to 132 turn 133 to 144 turn 145 to 156 turn 157 to end of simulation  1.54  1.36  1.34  1.32  1.29

 Agent who does not or can not pay premiums

a) Agents who lose their jobs,whose income is below a certain standard = add period of half or total exemption from premium b) Woman agents whose partner is salaried employee c) Man agents who are parents pay the premium for child agents who are in school and not independent
WEHIA 2005 12

M. MURAKAMI & N. TANIDA

Simulation (2) - Rules of Each Model 

Distrust in public pension system and its propagation

(Source) Social Insurance Agency (1999), National Pension Insured Survey, 1996, 1999
1. No distrust model

: Nonpayment of pension premiums arisen from agent’s unemployment or income level.
2. Low level of distrust model

:No distrust model + Man agents distrust in public pension system (based on 1996 survey). His distrust is not propagated.
3. High level of distrust model

:No distrust model + Man agents distrust in public pension system (based on 1999 survey). His distrust is propagated the woman agent, who is a partner.
M. MURAKAMI & N. TANIDA WEHIA 2005 13

Simulations (3)

  

500 turns (about 40 years) Ten simulations in every model Average is calculated for analysis Friedman’s test
Whether is the difference between three models statistically significant ?

M. MURAKAMI & N. TANIDA

WEHIA 2005

14

Today’s Talk
1.

Background
(1) Population Projections in Japan The number of people who did not pay pension premium (3) Improvements in this study

2.

Simulation Settings
(1) (2) (3) (4)

3.
4.

Simulations

Basic design Data Calculation formula of pension benefit Model Visualization(1), (2)

(1) Common rule, (2) Rules of each model

Simulation Results
(1) Pension premium fund (2) Monthly pension premium – benefit difference (3) Result of Friedman’s test

5.

Conclusion and Future works
WEHIA 2005 15

M. MURAKAMI & N. TANIDA

Simulation Results (1) - Pension Premium Fund (Average) C om par son of T hr i ee m odel( ensi P r i P on em um Fund )
T en T housand Y en 45000 40000 35000 30000 25000 20000 15000 10000 5000 0 No Di strust Low Levelof D i strust H i Levelof D i gh strust

Agent’s distrust in public pension system and its propagation has a decisive influence on the amount of the pension premium fund.
M. MURAKAMI & N. TANIDA WEHIA 2005 16

0 13 26 39 52 65 78 91 104 117 130 143 156 169 182 195 208 221 234 247 260 273 286 299 312 325 338 351 364 377 390 403 416 429 442 455 468 481 494
N um ber of T urns

Simulation Results (2) - Premium - Benefit Difference (Average) C om pari son of T hree M odel( rem i P um - B eneft D if i f erence)
T en T housand Y en 120 No Di strust Low Levelof D i strust H i Levelof D i gh strust

100

80

60

40

20

0

13 26 39

52 65 78

91 104

0

117 130 143 156 169 182 195 208

221 234 247

260 273 286

299 312

325 338 351

364 377 390

403 416 429

442 455

Downward trend Agent’s distrust in public pension system and its propagation has a decisive influence on the amount of the pension premium – benefit difference.
M. MURAKAMI & N. TANIDA WEHIA 2005 17

N um ber of T urns

468 481 494

Simulation Results (3) The Result of the Statistical Test
The Result of the Statistical Test Performed on Premium Fund

Number of Turns Rank Average 2 Χ Degrees of Freedom Significance Probability

No Distrust 501 3

Low level of Distrust High Level of Distrust 501 501 1.99 1.01 991.036 2 6.3×10-216 Low level of Distrust High Level of Distrust 501 501 2.18 1.18 556.756 2 1.26×10-121

The Result of the Statistical Test Performed on the Monthly Premium-and-Benefit Difference

Number of Turns Rank Average 2 Χ Degrees of Freedom Significance Probability

No Distrust 501 2.63

These results lead us to the conclusion that the people’s distrust in public pension system and its propagation has a decisive influence on the amount of the pension premium fund and on the amount of the monthly premium-and-benefit difference.
M. MURAKAMI & N. TANIDA WEHIA 2005 18

Today’s Talk
1.

Background
(1) Population Projections in Japan The number of people who did not pay pension premium (2) Improvements in this study

2.

Simulation Settings
(1) (2) (3) (4)

3.
4.

Simulations

Basic design Data Calculation formula of pension benefit Model Visualization(1), (2)

(1) Common rule, (2) Rules of each model

Simulation Results
(1) Pension premium fund (2) Monthly pension premium – benefit difference (3) Result of Friedman’s test

5.
M. MURAKAMI & N. TANIDA

Conclusion and Future works
WEHIA 2005 19

Conclusion and Future Works
Conclusion


People’s distrust in public pension system and its propagation has a decisive influence on the amount of the pension premium fund and the amount of monthly premium-and-benefit difference.

Future works
     

Examine the relationship of each property (Ex. The process that agent distrusts in public pension system) Implement the property other than distrust Closely examine statistical data Analyze the continuous change in each agent’s environment and consciousness ‘Verification’, ‘Validation and Test’, ‘Sensitivity Analysis’ Increase family line

M. MURAKAMI & N. TANIDA

WEHIA 2005

20