Pension simulation with a huge number of agents

In this paper, we describe an agent-based approach for pension simulation in Japan. As one of industrially advanced countries, Japan faces an aging population and a decreasing birth rate. In such a depopulating society, how to support retirees by national pension program is really important. The government tries to show the effectiveness of the current pension system using a model case of a family with a husband, who works for the same company for 40 years, and a wife, who is a full-time housewife for 40 years. Although they used to employ this model case for their explanation, employment patterns widely diversify nowadays. In order to show a pension plan for each employment pattern, an agent-based approach is one of promising way. However, implementing agent-based simulation with a huge number of agents, we should solve how to keep the internal information properties for each agent. In this paper, we employed a memory-mapped file I/O system call to keep the information of the huge number of agents. Using this method, we come to see pension plans for various employment patterns.