Assignment of National Pension Beneficiaries and Amounts Received to Synthetic Population Data

Japan’s public pension system is based on a levy system in which insurance premiums paid by the working-age population are used to finance pension benefits. As a result, various problems have arisen due to factors such as the falling birthrate and aging population, and the system has been revised each time to maintain or improve it. In addition, the ratio of pensioners to the total population is increasing, and it is expected that social simulations and other analyses will continue to be conducted in order to revise the pension system. In creating models for social simulations, synthetic population data is used to set attributes, but at present, attributes related to pensions do not exist. Therefore, this paper proposes a method for assigning national pension beneficiaries and their benefit amounts to synthetic population data. This approach combines three novel innovations: defining beneficiaries as multi-attribute agents, a robust SA optimization framework that simultaneously achieves population consistency and attribute optimization, and an implementation using technically advanced multi-temperature design capable of handling multi-scale error landscapes. Furthermore, the proposed method is applied to Tottori Prefecture and other regions, and the results are compared with statistical data. For Tottori Prefecture, the allocation achieved zero error in the number of beneficiaries and successfully limited the error in benefit amounts to within 3.65% of the average benefit amount.