Quality scale

A low score indicates stronger direct evidence, but even official sources can have sampling, coverage, timing or unit-of-analysis limitations.

Important: “Income” does not always mean wages. National tracks can use tax income, household income or harmonized economic income. Definitions are documented in Methodology.

77 country and territory models + 8 regional residuals

The harmonized model covers 77 explicit country/territory components. Regional residuals capture the remaining population and income so the global totals reconcile without double counting. National source tracks are kept separate from the harmonized benchmark.

How the model is built

Country macro levels are combined with the best available distribution shape. Where direct national thresholds exist, they anchor the curve. Where they do not, donor-country shapes are used transparently and receive a weaker data-quality score. Wealth levels can also use source-specific anchors such as UBS median, mean and millionaire counts.

Regional residuals are calculated after explicit countries are removed. The model is designed so population and national-income components reconcile globally; wealth is partly model-based and therefore carries more uncertainty.

For income, use the annual flow that matches the selected source definition. For net worth, use assets and liabilities measured at the same point in time. Do not double count an asset and the gain embedded in its current value.

2. Cross-country comparison

The default cross-country view uses a harmonized economic measure. For each country, the model calculates the share of the population above the same amount. A smaller top share means a higher relative rank. Absolute rank is also shown, but it can favor small countries simply because they have fewer people.

Currency conversion changes the display amount, not purchasing power, local costs, taxes or the economic effect of moving country.

3. Measures and units

Harmonized: intended for cross-country comparison. Personal: follows a national person-level source where available. Household/family: follows the source household unit and should not be divided automatically unless the model explicitly says so.

4. Population and age coverage

The main benchmark generally uses adults aged 20+, but source populations differ. Some national tracks cover taxpayers, households, families or residents aged 16/17+. Where exact 2025 age counts are unavailable, historical age ratios or explicit assumptions are carried forward and labelled as such.

5. Observation year and currency

Source years differ across datasets. The harmonized main track is modelled around a 2025 reference level, while national source tracks retain their own reference years. Display FX rates are fixed conversion rates, not live market rates and not PPP adjustments.

6. Distribution curves

Observed thresholds are connected with monotonic interpolation. Where the source stops before the extreme tail, the model uses an explicit tail assumption rather than pretending the source contains observations that do not exist. More numerical grid points do not mean more source observations.

Top of the wealth distribution (v10). Above the median, every wealth curve follows a smooth generalized-Pareto shape fitted to the published top-50%, top-10% and top-1% wealth shares, the national mean and the median. Where published millionaire counts exist, they are used as an additional calibration point, subject to a rule that rules out flat stretches in the curve. The extreme-tail exponents share one common scaling factor so that the global number of billionaires stays consistent with published annual counts.

7. Donor-country models

When a country lacks a comparable distribution, the model borrows shape information from a weighted set of donor countries. Donor choices are tested using leave-one-country-out procedures where possible. A donor model is clearly labelled and receives a weaker data-quality score than direct national data.

8. The extreme wealth tail

Survey and register sources rarely observe the very richest. Above the top 0.01%, each country curve therefore follows an explicit Pareto-type tail whose parameters are calibrated so that the global total is consistent with published annual billionaire counts (roughly 3,000–3,500 people above USD 1 billion).

No person-level list is used, and no individual can be identified from the model. Ranks in the extreme tail are approximate by construction.

9. Global aggregation

The world curve is the sum of explicit country components plus regional residuals. EU is a filter across member states, not an extra population component. Hong Kong is modelled separately and is removed only from the relevant regional residual, not from mainland China’s curve.

10. Uncertainty

Scenario ranges are sensitivity envelopes, not statistical confidence intervals. They vary curve shape, level and selected demographic assumptions. Definition differences can be larger than the numerical uncertainty shown by any single scenario.

11. Zero and negative wealth

Some sources explicitly contain zero or negative net worth; others do not. Where the source cannot support a precise rank around zero, the interface avoids showing false precision.

12. Validation and audit

Three questions are kept separate: does the math reconcile, can the method reconstruct held-out source points, and does the result agree with external reference data? Passing one does not imply passing the others. The Audit tab exposes the relevant tests.

Sources

Primary and legacy source links used by the model. Source definitions and dates can differ, so links should be read together with the track metadata.