EPOCH: analysis code, data and posteriors for the population-output elasticity

Complete analysis pipeline, input data, posterior samples and result set for “Population ageing raises the bar for depopulation to lift living standards” by Tomoki Kawahara, Shuhei Terada, Nobutoshi Nawa and Takeo Fujiwara. Let ε be the elasticity of aggregate real output with respect to the working-age population. Output per person is output divided by total population, so demographic change adds to output per person only where ε lies below the ratio of the proportional fall in population to the proportional fall in the working-age population. Because workforces shrink faster than populations, that threshold is itself below one: across the 75 economies projected to depopulate by 2100 its median is 0.65. On 896 non-overlapping country-decades covering 179 countries between 1961 and 2020, a three-level hierarchical Bayesian long-difference model returns ε = 0.805 (95% credible interval 0.577–1.029), below one with posterior probability 0.961 under a prior centred at proportionality. The estimate is indistinguishable across seven World Bank regions and stays within roughly 0.8 to 1.0 under widened priors, a doubled independent sample and a specification curve. A separate levels specification shows that super-unitary estimates are an artifact of measuring output in current prices. A single projection engine carries the fitted elasticities through United Nations World Population Prospects 2024 medium-variant demography to 2100 for 175 economies. Output per person rises in every one of them, including all 75 whose population falls; the demographic channel is negative in 143 and smaller in absolute value than the underlying trend in all 175. Before writing any result, the engine asserts that aggregate growth minus per-capita growth equals population growth for every country and every year, and a test suite exercises each assertion. This record holds the code repository (tag v2.2.0) and seven posterior sample files (about 10 GB). The repository alone reproduces every figure and every reported number; the posteriors are needed only to re-derive a quantity not already summarised in results/. All results use a fixed random seed of 42, and README.md gives the exact command order.