全球收入坐标 2026｜Global Income Compass

Abstract

Urban cost is a vector, not one scalar: housing; local services and everyday consumption; nationally convergent online goods; and globally priced travel or non-local spending. The model first separates annual after-tax income into current lifestyle consumption and savings, investments or other retained value. It then weights each destination/origin component price relative by the user’s origin-city spending shares. City prices share a New York=100 scale. Direct city evidence provides anchors; missing cities are fitted in log space from geographically, economically and housing-market-comparable peers. Market FX converts currencies and the retained-value layer, while city price relatives apply only to current consumption. Outputs are central estimates with reasonable bands, not exact quotations.

01City price sources→

02Common vintage & scale→

03Observed anchors & fits→

04Personal lifestyle basket→

05Equivalent after-tax income

Research question, unit and scope

The unit is one adult’s annual individual after-tax income, not household gross income. The starting point is not an average local resident but the user’s existing origin-city lifestyle. The model preserves two things: the current consumption basket can be recreated in the destination, and income not used for current consumption retains its market-FX value.

YAnnual after-tax income

Personal net resources available for consumption, saving or investment; the destination output is also after tax.

sCurrent-consumption share

The share of after-tax income funding current life; the remainder, 1−s, is retained value.

wPersonal basket

Origin-city shares for rent, local life, online goods and non-local spending, summing to 100%.

What the model does not answer

It is not a salary-offer engine and does not value career opportunity, education, healthcare, public services, climate, safety or happiness. Those may matter more than prices, but they should not be hidden inside one purchasing-power number.

Data architecture: what each source is used for

No single official dataset covers 100 cities across China, the United States, the euro area and Japan with comparable housing, service and goods components. The model therefore combines a cross-city comparison layer, official constraint layers and a macro scale layer rather than treating any one source as ground truth.

Evidence layerPrimary sourcesRole in the model

City cross-sectionNumbeo 2026Provides same-vintage, New York=100 relatives for non-rent living costs, rent, groceries and restaurants.

Official regional pricesBEA RPP · Japan Retail Price SurveyConstrains US metro and Japanese regional gradients, especially the direction and magnitude of rent differences.

Basket weights & time updateNBS · BLS CE · Eurostat HBS/HICP · Japan FIESCalibrates broad housing, service and daily-consumption shares and updates heterogeneous evidence toward a common vintage.

Cross-country scaleWorld Bank ICP · Federal Reserve G.5APPP checks macro price levels; annual average market FX converts currencies, global consumption and retained value.

Crowdsourced city data offer breadth and component detail but can be biased by contributor mix, neighborhood and submission frequency. Official data provide more stable sampling and definitions but often lack complete city coverage, arrive with a lag, or measure inflation over time rather than absolute price levels across cities. Each source is used only where it is informative.

City price gradients and missing-city fits

Each city becomes a price vector P=(housing, local life, online goods, global consumption). Direct city evidence is aligned to a common vintage and normalized to New York=100. When direct observations are insufficient, the model does not copy a national average. It selects anchor cities in the same country or currency area with similar economic tier, geography and housing-market intensity, interpolates their prices geometrically in log space, and constrains the result with published rent ranges, regional price orderings and official time-series evidence.

Component estimate for a missing cityP̂_c,k = exp[Σ_j∈A(c) λ_j · log(P_j,k)] · H_c,k

k indexes price components; A(c) is city c’s matched anchor set; similarity weights λ sum to one; H adjusts for housing intensity, urban tier and vintage. Log-space interpolation prevents one exceptionally expensive anchor from dominating a linear average.

CN · 44

Chinese cities

Major cities use same-vintage city cost and rent components as direct anchors. Others are fitted by region, urban tier, rent intensity and neighboring anchors. NBS 70-city housing statistics mainly describe price changes, so they check housing direction and anomalies rather than masquerading as cross-city rent levels.

US · 33

US cities

BEA RPP is the main official constraint: US average=100, comparable across states and metros, with housing rent separately identified. City cross-sections add restaurant, grocery and other detail. Where they conflict, RPP governs the broad gradient and city data refine metro ordering.

EU · 15

Major euro-area cities

Same-year city cost and rent cross-sections form the level comparison. Eurostat HICP is used only to align vintages and check national inflation direction. The current set focuses on euro cities to avoid mixing additional pound or Swiss-franc currency effects into the regional layer.

JP · 8

Japanese cities

Tokyo and Osaka form the main direct anchors. Other cities combine the Retail Price Survey, Housing and Land Survey, city rent evidence and FIES spending structures, fitted by housing and everyday-cost intensity relative to Tokyo or Osaka.

Evidence levelDefinitionRecommended interpretation

HIGHERMain components have direct city anchors plus official or secondary constraintsRead around the center with at least ±10% sensitivity

MEDIUMPartial direct components plus regional fittingPrefer an approximately ±15% band

DIRECTIONALPrimarily inferred from tier, region and neighboring citiesUse mainly for broad dearer/cheaper direction

These bands are not statistical confidence intervals

They are model-sensitivity ranges based on source coverage, source agreement and fitting depth. They remind users that within-city variation and personal lifestyle differences often exceed the displayed decimals.

From average prices to a personal lifestyle basket

The same city index does not imply the same result for everyone. Renters and homeowners, frequent diners and home cooks, locally oriented consumers and frequent international travelers face different effective prices. The model therefore asks for origin-city spending shares; system defaults are editable starting points, not facts about the user.

RENTwᵣ

Long-term rental cost; may be set to zero for homeowners

LOCAL LIFEwₗ

Local services, groceries, transport, utilities and ordinary offline spend

ONLINEwₒ

Online goods whose prices converge within a market

GLOBALw_g

Travel, imports and other non-local spending

Local life is further separated into services and everyday goods. As income rises, non-tradable local services such as dining, fitness, personal care and domestic help tend to gain weight, amplifying differences in service-intensive cities. Online goods converge more within a country or currency area, while global spending remains closer to market-FX pricing.

Two layers of after-tax incomeY = Y · s + Y · (1 − s)

Y·s funds current lifestyle and is compared through city price vectors. Y·(1−s) is savings, investments and other retained value, converted at market FX. The system suggests s from income bands and official expenditure evidence, but the user can override it.

Comparison formula: why component price relatives come first

Because users enter nominal origin-city spending shares, the model uses a Laspeyres-style comparison: it holds the origin consumption structure fixed and prices the same needs in the destination. Dividing two separately aggregated city indices would silently introduce two different average baskets.

Relocation price of the origin-city basketR_o→d = Σ_k w_k,o · P_k,d / P_k,o

R=1 means the same basket costs the same; R>1 means the destination is dearer; R<1 means it is cheaper. Weights w sum to one only within current lifestyle consumption.

Equivalent after-tax income required in the destinationY*_d = (Y_o / FX_o) · [(1−s) + s·R_o→d] · FX_d

Origin income is first converted into a common currency at market FX. Retained value stays unchanged, lifestyle consumption receives the relocation multiplier, and the result is converted into the destination currency.

Overall capacity at FX-equivalent incomeCapacity_d / Capacity_o = 1 / [(1−s) + s·R_o→d]

Why the model does not use one PPP exchange rate

World Bank PPP measures the overall price of a representative national basket. It is useful for macro living-standard comparisons but is neither a city index nor a person’s basket. The model uses PPP as a cross-country plausibility boundary; the individual result comes from city components and personal weights. Imports, international travel, global assets and retained value are governed by market FX and cannot all be PPP-adjusted.

CNY / USD7.1875

2025 annual average market FX

EUR / USD0.8845

Derived from 1 EUR=1.1306 USD

JPY / USD149.57

2025 annual average market FX

PPPCHECK

Not a single city conversion rate

Factors that move the result most

FactorMechanismTypical direction

Rent share & housing statusHousing is usually the most dispersed city componentHigher destination rent and a larger rent share raise required income; homeownership can set it to zero

Consumption share sDetermines how much after-tax income is exposed to city pricesA larger s transmits more city-price difference; a smaller s moves the result toward market FX

Local-service shareDining, domestic help and care are non-tradable across citiesMore service use usually makes high-labor-cost cities dearer

Online & global spendingPrices are closer to national or global marketsLarger shares usually damp local city gaps and strengthen FX effects

FX vintageCross-border income and retained value use annual average FXLarge short-term currency moves can shift the real relocation-date result

Within-city locationNeighborhood, housing type and roommates change actual rentPersonal rent variation can exceed the average gap between two cities

Uncertainty, interpretation and model limits

Uncertainty enters through five channels: contributor composition in city samples; different definitions and release dates across official sources; fitted cities; within-year FX and price movement; and the gap between a user’s actual housing or basket and the defaults. For well-anchored cities, inspect at least roughly ±10% around the center; use about ±15% for partially fitted cities and about ±20% where regional inference dominates.

The output is an annual after-tax equivalent, not destination gross salary; users must separately gross it up under local tax rules.
The default is individual-level and does not automatically model spouses, children, education, childcare or family healthcare.
Setting homeownership to zero rent does not imply zero maintenance, mortgage, property-charge or opportunity cost.
Home purchase prices, investment returns, pensions, visas, moving costs and career progression are outside the current formula.
A result of 1.18× versus 1.20× is rarely a substantive distinction; direction, order of magnitude and the sensitivity band are more reliable.

The same life in a different city: a model of after-tax income and lifestyle purchasing power