The Impact of Population Aging on Youth Wages in Japan: Evidence from Japanese Panel Data

Researchers' Day Conference Proceedings · Budapest University of Economics and Business · 2025

Abstract

Japan's rapidly ageing population is reshaping the structure of its labour market. At the same time, traditional employment practices, including age- and tenure-related wage structures and the continued employment of older workers, remain important features of the Japanese employment system.

This study examines whether the growing wage share of older and higher-income workers is associated with weaker wage outcomes for younger cohorts. Using grouped panel data from the Japanese Panel Study of Employment Dynamics (JPSED), the analysis follows changes in wage distributions across age groups between 2015 and 2024.

An Overall Wage Score is constructed to summarize the distribution of each age group across wage brackets. A Bayesian multilevel model is then used to estimate the relationship between changes in the population share of high-income earners and overall wage outcomes. The results indicate that older age groups increasingly occupy higher-income brackets, while younger cohorts show comparatively weaker wage development. In particular, growth in the population share earning between 10 and 12 million JPY is associated with a statistically credible decline in the Overall Wage Score.

The findings provide evidence consistent with the possibility that demographic ageing and the concentration of high-income positions among older workers may create structural wage pressure for younger workers. Because the study relies on grouped data, however, individual-level characteristics cannot be fully incorporated, and further research using individual panel data is necessary.

Keywords: Ageing · Labor · Wages · Population · Japan

JEL Classification: J82

Citation

Article citation:
Csiszár, C. (2025). “The Impact of Population Aging on Youth Wages in Japan: Evidence from Japanese Panel Data.” In K. Szegedi (Ed.), Researchers' Day Conference Proceedings – Budapest University of Economics and Business 2025 , pp. 57–72. Budapest University of Economics and Business.

DOI: 10.29180/978-615-6886-26-2_6
ISBN: 978-615-6886-26-2

Proceedings citation:
Szegedi Krisztina (2026). Kutatói Nap Konferencia kiadvány - Budapesti Gazdaságtudományi Egyetem 2025. Budapesti Gazdaságtudományi Egyetem, Budapest, pp. 1–83. ISBN 978-615-6886-26-2.

Research Overview

Research Question

How does the increasing wage share of elderly workers in the Japanese labour force influence the wage growth of younger workers?

Contribution

The study examines intergenerational wage dynamics in the context of Japan's rapidly ageing labour force. It extends the discussion of retirement slowdown and labour-market bottlenecks to Japan and introduces an Overall Wage Score that summarizes changes in the wage distribution of different age cohorts over time.

Key Findings

Older cohorts increasingly occupy high-income wage brackets, while younger cohorts show weaker wage development. The Bayesian model estimates a negative coefficient of −41.97 for changes in the population share earning 10–12 million JPY, with a 95% credible interval from −68.39 to −16.32.

Research Methodology

The empirical analysis uses data from the Japanese Panel Study of Employment Dynamics (JPSED), conducted by the Recruit Works Institute. The study covers the period from 2015 to 2024 and uses aggregate panel data in which observations represent combinations of year and age group.

Wage information is reported in thirteen annual income brackets. To summarize the complete wage distribution for each age group, an Overall Wage Score is constructed by assigning a numerical value to each wage bracket and calculating a weighted average using the population share within each bracket.

The modelling strategy initially considered linear and robust mixed-effects models with Year as a random intercept. Because these specifications resulted in singularity and convergence problems, the analysis was subsequently estimated using a Bayesian multilevel model implemented in R with the brms package.

Alternative model specifications were compared using Bayesian model evaluation criteria, including expected log predictive density (ELPD) and leave-one-out cross-validation (LOO). The final model includes the standardized change in the population share earning between 10 and 12 million JPY as the principal explanatory variable and a random intercept for Year.

Bayesian estimation was performed using Markov Chain Monte Carlo sampling and the No-U-Turn Sampler (NUTS). Model reliability was evaluated using convergence diagnostics, posterior predictive checks, credible intervals, and Rhat statistics.

Full Paper

The full publication is available below.