Synthesizing Real-Scale Populations with Location and Income Attributes of Whole Japan using Birth Cohort

This study proposes a method to synthesize nationwide, third-party-shareable microdata (synthetic households and individuals) for Japan with geographic information and income-related attributes using only publicly available statistics. Prior simulated-annealing (SA)-based approaches typically fit all parent–child relationships to age-at-birth distributions observed in a single year, which can fail to capture cohort-specific fertility behaviors such as shifts in maternal age at childbirth. To address this, we incorporate birth-cohort tables (parent birth year × age at childbirth × number of births) as constraints so that cohort-dependent birth patterns are reflected in the synthesized parent–child age structure. Because birth-cohort statistics are published only at the national level, we generate the synthetic population for the entire country in one run, and adjust the birth-cohort constraints to be consistent with estimated counts of fathers and mothers by birth year derived from census tabulations. The SA optimization then minimizes discrepancies across multiple statistical targets, including spousal age differences, distributions by family type, and sex–age population distributions. Experiments at the scale of the 2015 Japanese census (≈50.96 million households; ≈115.55 million individuals) show that increasing search iterations reduces overall error, but improvements become limited when the birth-cohort adjustment is insufficient, highlighting the need for more robust cohort-adjustment procedures.