Data methodology

What counts as a reliable source

National primary sources have priority for macroeconomic data: the National Institute of Statistics, National Bank of Cambodia, Ministry of Economy and Finance, Council for the Development of Cambodia, tax administration and sector ministries. Their mandates differ. The statistics office produces national accounts and indices, the central bank covers money, credit, exchange rates, banks and stability, and sector agencies provide administrative data.

International institutions are used not as replacements for national data but for a comparable framework, current-year estimates and disclosed assumptions. The World Bank, IMF and Asian Development Bank may publish different estimates and forecasts because their data cut-offs and models differ. Every forecast remains attached to the institution that produced it.

Dated reports from major property consultancies are accepted for supply, vacancy, rents and market segmentation. They are sample-based market research, not an official register. Agency blogs, project pages, listings and aggregators without a traceable primary source are not evidence for a market statistic.

Why every number has two dates

The period field identifies the observation: calendar year, quarter, month or half-year. The released field records when the source became available. Neither can be replaced by the page update date. A report released in June 2026 may contain final 2024 data, a 2025 estimate and a 2026 forecast.

The data_checked_at field records the day on which the editorial team reconciled the page against available publications. It does not make an older observation current. If the latest official series ends in December 2025, the page says so even when checked in July 2026.

When a source provides no day or month, we do not invent precision. The release field contains only what can be verified, such as “2025, month not stated on the page.” This distinguishes a confirmed publication date from a file upload timestamp or a date inferred from search results.

Observation, estimate and forecast

An observation is a published result for a completed period in the source’s own classification. It can still be preliminary and later revised. “Observed” does not mean immutable; it means the period has ended and the source is reporting an outcome rather than a future scenario.

An estimate refers to a completed or nearly completed period for which the full data set is not yet available. Growth for 2025 published by an international institution in June 2026 may remain an estimate until final national accounts are released. The card labels it explicitly.

A forecast covers a future or incomplete period and is always attributed to an institution and a release date. An old forecast is not silently overwritten. It is retained as a separate vintage when a report needs to show what was known at the time.

How conflicting figures are handled

When two reliable sources report different values for one indicator, we first check the definition, period, unit, seasonal adjustment, data cut-off and estimate status. A difference often reflects a different data vintage rather than an error. If both figures refer to the same concept, both remain visible.

We do not average World Bank and National Bank estimates, select the number that best supports a narrative, or call a gap “minor” without quantifying it. A note explains why the comparison requires caution and which source will take priority when final national statistics become available.

When a new publication revises a historical series, the table receives the new version and a revision note. A period report may also preserve the value that was available at the time, allowing readers to see both the eventual result and the information set then available.

Units, currencies and scale

Real growth is measured at constant prices, while the nominal size of the economy is measured at current prices. A dollar amount is not a substitute for a real growth rate, and rising nominal revenue does not prove a larger physical volume. Any currency conversion identifies its source and rate; an editorial calculation is never presented as a published figure.

Property area is accompanied by a definition: net saleable area, internal area, gross floor area or net lettable area. A price per square metre without an area convention can differ materially even for the same apartment. Such figures are not merged into one series.

Percent and percentage points are kept separate. An occupancy decline from 61.4% to 58.7% is 2.7 percentage points, not 2.7%. Every monetary amount carries a currency, and million or billion is never used without an explicit scale.

Official statistics and market samples

The National Bank’s official residential property price index and Knight Frank’s supply estimate measure different phenomena. The index tracks price change within its sample and model; the supply estimate counts existing units under the consultant’s definition. Stock growth cannot be used as a denominator for the index or interpreted as absorption.

A new-launch price is not a citywide average transaction price. An advertised rent is not a signed lease rate. Office occupancy is not the share of apartments sold. Each page retains the source’s market term and provides a practical definition in both languages.

When consultancies use different Phnom Penh boundaries, segments or inclusion rules, their series are not automatically spliced. Comparison requires a methodology check. Where the method is not sufficiently disclosed, figures are shown separately with a limiting note.

Rules for calculations and examples

A page calculation exists only to explain mechanics. Every assumption is visible: price, area, rent, vacancy, service charge, repair reserve, management fee, tax and exchange rate. The result is labelled as an example, not as a market-average yield.

We do not fill missing months, interpolate a quarter, convert asking to transaction prices using a universal discount, or treat promised rent as verified income. When inputs are unavailable, the issue belongs in “What we do not know.”

An editorial arithmetic result is acceptable when it can be reproduced in full. A ratio may be calculated from two published values, for example, but both inputs, the formula and the date remain visible. The output is labelled as a NovAsia calculation, not a source figure.

Updates, corrections and archive

A full reconciliation follows major releases from the World Bank, National Bank of Cambodia, National Institute of Statistics and market-report providers. An individual card may be updated earlier when a primary source materially changes the indicator.

A factual error is corrected with a change-log entry: the previous wording, the corrected wording, the confirming source and the correction date. A revised estimate is not called an error when the previous figure accurately reflected the publication available at that time.

Every page displays its last check date, while reports preserve their publication period. This separates a live hub from a historical snapshot and prevents the past from being rewritten using information that market participants did not yet have.

Similar words, different statistical objects

The hub does not substitute a stock for a flow, an approval for execution or a gross measure for a net one. Year-end credit is a stock of claims, while credit growth is its change; FDI is a recorded financial flow, while approved investment is an administrative intention; gross NPLs show impaired claims before provisions, while net NPLs show the residual after provisioning. Each pair answers a different question.

The same rule applies to remittances and the budget. Workers’ inflows are not net remittances or the broader secondary-income category. An approved budget is not cash execution, a signed loan commitment is not a disbursement, and a disbursement is not completed infrastructure. The content names the stage rather than using a convenient umbrella term.

Where two measures have similar everyday labels, the editor checks unit, sector, institutional coverage, point-in-time or period basis and revision status. If comparability cannot be established, the figures remain separate. This protects the reader from false precision and makes updates reproducible.

Inflation is not used as a rent index

CPI measures a consumer basket. It can explain pressure on household and operating budgets, but does not replace signed-rent data. A monthly year-on-year rate, a month-on-month change and a calendar-year average remain separate measures.

Where an agency’s current card does not display a separate release date, we record our verification date rather than inventing one. Earlier forecasts retain their original vintage and are not rewritten to match the eventual outcome.

Rules for tourism statistics

An international arrival is a border crossing, not necessarily a unique person, night or booking. Domestic regional visits can also include repeat trips. The series are not added as unique individuals.

Average stay, hotel occupancy and tourism receipts apply only to their period and segment. Hotel figures are not transferred to private apartments, and a five-month total is not labelled as a quarter.

A stable mirror of an official PDF is used only when authorship is clear and a stable government URL cannot be located; the limitation is recorded in notes.

Census, survey and population projection

A census is a full count at a specified date. A projection is a model based on fertility, mortality and migration. A household survey is a sample estimate with its own uncertainty. The status remains visible beside every number.

The renter-household share is not the condominium share: the survey includes different dwelling types. National population cannot be converted into apartment demand without assumptions about households, location, income and tenure.

Method for the risk page

A risk is described through source, transmission mechanism, affected asset and buffer. An institution’s reference to downside risk is not converted into a numerical probability. We do not create a country score or rank scenarios without a published method.

Different NPL ratios, forecasts or GDP estimates remain separate. If coverage cannot be fully reconciled, the difference becomes part of “what we do not know” rather than an error hidden by averaging.

Quarter-close rule and report statuses

A report becomes final only when a sufficient set of period-specific sources exists and the latest observation is checked inside each release. A First Quarter or Second Quarter label on a cover does not by itself prove that the tables contain that quarter.

Preliminary status is used when the quarter has ended and several independent indicators exist, but part of the key monthly evidence remains missing. Such a page must display data_cutoff, revision_required and a visible missing-data section.

Finalisation does not create a new slug. The period, original cutoff and change list are retained. Numbers are not silently overwritten: source, observation period and new vintage remain auditable.

What we do not know

We cannot fully audit a consultancy’s internal sample when only a summary report is published. The figure can still be used with the firm name, date and limitation, but it is not called official.

Not every government series is machine-readable, and PDF versions may be replaced without a public revision log. We retain a copy, access date and publication details, but cannot guarantee a complete version history.

Missing data do not imply that the underlying phenomenon is absent. If there is no open series for transaction discounts or net yields, we do not infer zero discounts or equate net yield with an advertised figure.

Frequently asked questions

Where this leads next