Macroscope
RU

Migration: UN forecasts vs actual

Pick a country and switch indicators on: they all line up on one time axis. You can see what changed together, what came first, where official forecasts point and how far past ones missed. Under the chart: observations computed from what is on screen.

Net migrationpersons per year610k a year 2023
01M
Refugees hostedpersons, end of year2.66M 2025
01M2M
Migrant share% of population19.8% 2024
101520
actualcurrent UN forecastUN low variantUN 95% interval

Without JavaScript you see a ready view: Germany — Net migration, Refugees hosted, International migrants, share, 1990–2035. Every country has its own page that works without JavaScript (list below); the data is in the API.

How the indicators changed

  • Net migration: on average 311k a year in the 1990s, 48.2k a year in the 2000s, 455k a year in the 2010s, 526k a year in the 2020s.
  • Refugees hosted: 816k in 1990 → 2.66M in 2025 (×3.3). Low: 188k in 2013; ×14 since then. The trend broke around 1996 and 2015: +38.2k → −50.6k → +250k a year.
  • Migrant share: 8.7% in 1990 → 19.8% in 2024 (+11 pp). Threshold: above 10% since 1995.

How the indicators relate

  • Net migration and refugees hosted (1990–2023, 34 years): levels are barely related (r = 0.22); year-to-year changes show no reliable link (r = 0.07).
  • Net migration and migrant share: too few common years with data (7) to speak of a relationship.

Forecasts vs actual

  • Net migration: past UN projections. The outcome was above the projection in 12 of 12 revisions (1998–2022) and below in none; median difference +422k a year. The biggest miss: the 2012 revision expected 90k a year for 2020–2025; the outcome was 572k a year (for the years of the period so far).
  • Net migration: UN 2024 projection (medium): 2030 — 163k a year, 2050 — 179k a year, 2100 — 141k a year.
  • Net migration: Eurostat projections (EUROPOP). For the latest year with data the outcome was below the projection in 1 of 6 editions (2013–2025) and above in 5; median gap +167k a year. The biggest miss: EUROPOP2013 expected 231k a year for 2024; the outcome was 494k a year.

Moving together in time does not prove a cause: incomes, cities, education and the era itself act on many indicators at once. These observations suggest where to look, not what causes what.

Key findings

  • UN projection of 2024: in 33 of 42 countries that already report 2024–2025 data, births are below the medium variant; in 8, below the lower bound of the 95% interval.
  • Japan: 709k births in 2024 (including children of foreign nationals). Even the latest official projection (IPSS 2023, medium) expected this level only in 2042, 18 years later. An earlier projection (IPSS 2002, medium) expected it in 2045.
  • Taiwan: 108k births in 2025. Even the latest official projection (NDC 2024, medium) expected this level only in 2036, 11 years later. An earlier projection (NDC 2016, medium) expected it in 2057.
  • China: 7.92M births in 2025. Even the latest official projection (UN WPP 2024, medium) expected this level only in 2050, 25 years later. An earlier projection (UN WPP 2022, medium) expected it in 2054.
  • United States: 1.60 children per woman in 2024. The SSA 2015 projection assumed 2.04 for that year, a gap of −22%.
  • South Korea: 0.80 children per woman in 2025. Projections made before 2020 expected 1.00–1.38 for 2025, while the 2021 and 2023 projections turned out lower than reality: fertility rose slightly in 2024–2025. Errors go both ways.
  • World: 67% of people live in countries with fertility below replacement (2.1), 31% below 1.5. World population is still growing: the UN expects a peak of about 10.3B in 2084.

All forecast-vs-actual checks →

How it works

  • A source for every number. Links go to the primary publication: the statistical office, the UN, the projection archive.
  • Forecast against fact. Next to a figure: what official projections expected for it and how far reality departed.
  • Association is not cause. Moving together in time suggests where to look; the observations under the chart always say so and check a link not only in levels but also in year-to-year changes.
  • Open data. Every table is available as CSV and JSON to check, recompute and reuse.
  • For programs and AI agents. The same data through an API and an MCP server, no page scraping: how to connect.

Data as of 7 October 2026

Country pages

Each country has a page that works without JavaScript: key indicators, observations, past UN and IMF forecasts against actual data, every series with source links. Countries →