Overview

Humanity Forecast combines official UN World Population Prospects 2024 data through 2100 with a long-range stochastic extension after 2100. The project is designed as an exploratory scenario tool, not as an official forecast. The key question is not what the UN says through 2100, but what a simplified long-range scenario engine implies if low fertility persists well beyond the period the UN formally projects.

The site shows two related products at once:

• World mode: humanity-wide projections under 100 fertility assumptions
• Country mode: each country's historical path plus one continuous UN-anchored modeled path through 2100 and into the long-range tail

Coverage

The underlying demographic model is built on 237 UN country or area series. The public product focuses on 192 sovereign nations plus a small set of aggregated blocs. That distinction matters:

• The simulation engine uses the broader 237-series UN structure so global totals remain internally consistent.
• The public country table focuses on 192 sovereign nations to keep the main experience understandable.
• Blocs such as the European Union, BRICS, and G7 are aggregated views layered on top of the underlying country data.

Inputs

The model stitches together several source families, each used for a different purpose.

• UN World Population Prospects 2024: population, fertility, life expectancy, crude birth rates, crude death rates, migration, and survival structure through 2100.
• Historical population data: OWID / Gapminder-style pre-1950 series used to show the long historical arc in the charts.
• Language and religion metadata: used for country table context, not for the demographic simulation itself.

Through 2100, Humanity Forecast is anchored directly to the UN data. Beyond 2100, the model becomes a scenario engine built on that base rather than an attempt to reproduce the full UN internal methodology. In other words: UN-anchored, but not an official UN extension.

Projection model

The core structure is a cohort-component model, which is the same general family of method used by the UN. Population is tracked in 21 five-year age groups by sex. Each step advances the population five years.

1. Each age cohort survives forward using survival ratios derived from UN life tables.
2. Births are estimated from crude birth rates and added into the youngest cohort with infant survival applied.
3. Net migration is distributed across age structure rather than treated as a single lump sum.
4. Country totals roll up into world totals and bloc totals for display.

After 2100, the model transitions into a stochastic extension. Birth and death rates evolve under an AR(1) style process with age-structure feedback. Fertility is represented in the public interface as Total Fertility Rate (TFR), but the long-range engine uses a simplified crude-birth-rate approximation rather than the UN's full bayesTFR machinery. Births are applied from crude rates to total population rather than from projected age-specific fertility schedules for women of reproductive age. That is a real limitation and should be read that way.

Scenarios

The main sweep computes 100 fertility assumptions from 0.1 to 10.0 in steps of 0.1. Each world scenario is run with 10 million simulated trajectories. The public presets are simply named anchors inside that wider sweep.

Country mode

The country hero and country pages now use a dedicated per-country sweep file, not just a blue overlay on top of the world chart. Each country has a full historical path and then one internally consistent, UN-anchored modeled median through the handoff into the post-2100 long-range tail under 100 fertility assumptions.

There is one important conceptual caveat: the selected TFR is applied globally in the model and then read out for the country. So “Japan at TFR 1.4” should be interpreted as “Japan under a world that converges toward 1.4,” not “only Japan changes while the rest of the world stays fixed.”

For very small countries, long-range medians can become quantized or visually noisy because the GPU sweep uses histogram-based aggregation. The country-specific files still exist for these nations, but the main table may suppress some extremely small values to avoid implying more precision than the bins can support.

Validation and uncertainty

The near-term requirement is simple: the model should stay close to the UN through the period the UN actually projects. In internal checks, the medium-path implementation matches the UN medium variant within roughly 1.6% at tested reference points. That is a useful implementation check, but it does not validate the post-2100 structural assumptions.

Uncertainty is represented through Monte Carlo trajectories and summarized as percentile bands, typically the 5th to 95th percentile range. Those bands are useful as scenario spread, but they should not be read as formal confidence intervals in the strict statistical sense.

Economic projections

Each country page includes an Economic Forecast tab showing dependency ratios, GDP projections, and fiscal pressure estimates. These are derived from the population projections, not independently simulated.

• Age structure estimation:Youth (0-14), working-age (15-64), and elderly (65+) shares are estimated from median age using calibrated empirical relationships, not from the simulation's age-group outputs. This is an approximation — a future GPU upgrade could output age-structured data directly.
• Dependency ratios: Old-age dependency = pop(65+) / pop(15-64). Youth dependency = pop(0-14) / pop(15-64). These are standard demographic measures.
• Fiscal pressure index: A weighted dependency ratio where elderly dependents are counted at 3× the cost of youth dependents, reflecting the higher per-capita cost of pensions and healthcare relative to education. Values above 120 indicate severe fiscal strain.
• GDP projections: Use a Barro-style conditional convergence model. Countries with lower HDI grow faster (up to 3% real/year), converging toward a moving frontier (currently ~$80K, growing at 1.5%/year). GDP = population × GDP per capita. This is a mechanical illustration, not a prediction — it does not model institutional quality, conflict, technology disruption, or policy.
• Migration: Not modeled. All projections assume zero net migration after the UN horizon ends at 2100. This significantly understates the population of high-income countries that would realistically attract immigrants over centuries.

Near-term economic projections (2050) are more credible than far-future values. By 2200+, all countries converge toward similar GDP per capita levels, which reflects model mechanics rather than a genuine forecast. These projections should be read as scenario illustrations showing how demographics drive economic structure.

Data sources

Country profiles draw from multiple international sources. All are publicly available.

• UN WPP 2024: Population, TFR, life expectancy, CBR, CDR, net migration (1950-2100)
• World Bank (WDI): GDP per capita, Gini index, literacy, internet penetration, urbanization, physicians, electricity/water/sanitation access, education and health spending, female labor participation, adolescent fertility, contraceptive prevalence, maternal mortality, governance indicators (corruption, rule of law, political stability — percentile ranks)
• UNDP HDR 2023: Human Development Index (HDI), Inequality-adjusted HDI (IHDI), Gender Inequality Index (GII)
• ILO (ILOSTAT): Average monthly earnings in PPP dollars
• Our World in Data / UN: Median age (UN WPP 2024 via OWID API)
• Institute for Economics & Peace: Global Peace Index rank and score
• UNODC: Intentional homicide rate per 100,000

Coverage varies by indicator. Wage data covers 130 of 192 countries (ILO does not report for all nations). The Global Peace Index ranks 158 countries. Where data is unavailable, the field is omitted rather than estimated. Literacy rates for developed nations where the World Bank does not report (assumed universal) are set to 99%.

Compute pipeline

The large world sweep was computed on an 8-GPU A100 cluster. The country sweep was also run on GPU with adaptive per-country histogram accumulation. The purpose of the GPU setup is not marketing theater; it is simply what makes the high-trajectory Monte Carlo runs feasible at interactive scale.

• World sweep: 100 TFR values × 10M trajectories each
• Country sweep: 100 TFR values × 237 country readouts × 1M trajectories
• Frontend: Next.js, TypeScript, Tailwind, Canvas 2D
• Data store: SQLite for the public product plus JSON sweep files for interactive country mode

Limitations

• Anything beyond 2100 is substantially more speculative than the UN horizon and should be read as scenario analysis, not official forecast output.
• The post-2100 engine uses a simplified CBR-based extension rather than the UN's full fertility modeling framework, and births are not projected from age-specific fertility schedules.
• Net migration is assumed to decay toward zero after 2100, which can materially affect country-level long-range results.
• Very small-country long-range values are lower-resolution than large-country or world totals.
• Blocs are useful summaries, but bloc uncertainty is more approximate than country or world medians.
• Percentile bands are model-conditional scenario ranges, not formal confidence intervals over all plausible futures.
• The farther the chart extends into the future, the more the results should be read directionally rather than literally, especially once dates are extrapolated beyond 2500.
• Upstream UN source data also carry uncertainty, especially where recent census and register data are limited.

This is the right way to read Humanity Forecast: a rigorous, transparent scenario engine built on real UN data, with explicit caveats, not a claim that anyone can know the exact world population centuries ahead.