What our data measure, and what they don’t

Our sources

Spending is one of our sources, not the only one. Each source has its own page: what we take, coverage, updates at the source and on our side, limits, and attribution wording. These pages are in Arabic.

• General Authority for Statistics (GASTAT)

  Ten categories: prices, population, households, real estate, labour, business, trade, tourism, energy, education and health

  16 cleaned · 5 raw · 29 not yet pulled
• Census 2022

  Population down to 150 governorates and 60 cities

  Year: 2022
• Saudi Central Bank (SAMA)

  Card spending: point-of-sale weekly, monthly and daily, and ATMs

  Last week: 20–26 September 2026 · 9–15 Rabi' al-Akhir 1448 AH
• KAPSARC open-data mirror

  History no longer available from the original source

  11 datasets, 9 of them stopped or late at source
• Calendar

  Umm al-Qura, salary dates and holidays

  Until 30 December 2028 (14 Sha'ban 1450 AH)
• What we calculate

  Cost of living, retail index, real growth, forecast: formulas and versions

  7 published calculations
• Forecast accuracy

  How far our forecast missed on data it had not seen

  On this page
• Independent verification

  Our card-spending figures compared with independent sources

  Available
• What we cannot reach

  The gaps and why: property deals by district, commercial registrations after 2023, building permits after 2019

  9 documented gaps

Update frequency and coverage

Limits and coverage warnings

• Cards only: no cash, transfers or online purchases; cards were about 57% of (point-of-sale payments + cash withdrawals) in 2025. Card share = point-of-sale payments ÷ (point-of-sale payments + ATM cash withdrawals), yearly; an approximation that excludes transfers and online purchases.
• Breaks in SAMA series: 5 activities show a permanent level change at the source with no published cause (Public Utilities & Services, Electronic & Electric Devices, Pharmacies & Medical Supplies, Professional & Business Services, Telecommunication); we show the figures as published, marked “break at source”, and any year-on-year comparison across the break is misleading. In 5 more (Construction & Building Materials, Furniture & Home Supplies, Recreation & Culture, Telecommunication, Transportation) the weekly series alone jumps where it switches from the old to the 2024+ activity classification; that is a classification difference, not spending, and the monthly report is the reference.
• GASTAT geography: apart from population, the finest level is the administrative region (13). Any “by city” view takes the city’s regional figure and says so. Consumer prices are published by region, not city, since February 2025 (Sha'ban 1446 AH).
• Index numbers measure change, not level: a regional consumer price index compares prices with the same region in the base year (2023 = 100); it says prices “rose more”, not that a region is “more expensive”.
• Older surveys: regional education and health data in the GASTAT database come from 2017–2018 surveys, and are shown with their year. 167 of 560 database indexes are broken at the source.
• Census: census city boundaries differ from SAMA’s “city” (the terminal’s location), so we publish the link table. Per-capita spending in Makkah and Al-Madinah includes visitors’ spending, so it looks higher than residents’ spending. The census is not published for download at district level.
• Stopped sources: data that stopped at the source are shown historically, marked with their last year, and never extended with an estimate.
• Estimates: forecasts are marked “estimate” with their historical error range, and are not official figures.

The full list of gaps is on What we cannot reach.

How numbers are checked before publishing

• Every published figure is recalculated by an independent check script from the source files at every build; any difference stops publishing.
• Match with the source: our figures match SAMA’s published figures within rounding. We compared the figures on the central bank’s statistics portal with its weekly bulletins for the same weeks in 3,172 figures: the mean difference was less than 0.1% and the largest less than 0.1%.
• Raw data kept: every pull is stored raw as downloaded, with its date and fingerprint, and never deleted. Cleaning standardises columns and does not change the numbers. When a source revises a figure, the new pull is stored beside the old one and becomes the one shown.
• Overlaps compared: where KAPSARC overlaps the original SAMA bulletins we compare the figures, and take the SAMA bulletin figure where they differ.
• Independent sources: we compare card spending with GASTAT’s wholesale and retail trade survey and publish the result even when it goes against us — see Independent verification.
• Rounding: millions to one decimal, percentages to one decimal, and “less than 0.1%” instead of zero. Any change beyond ±100% is shown with the absolute figure and a “small base” tag.

How we calculate

• Number of transactions and average transaction value: the count as SAMA publishes it for each city, activity and week, shown in “million transactions”; the average = payment value ÷ count for the same city, activity and week, in riyals per transaction, each with its change against the same week last year. Where SAMA publishes no count we write “—” and do not estimate it.
• Year-on-year change: against the same week 364 days earlier (same weekday); monthly against the same month last year.

Formulas for the indicators we derive (cost of living, retail index, spending after inflation, next-week estimate, calendar effects, card share, demand-generator density) are on What we calculate.

Dates: Hijri and Gregorian

• Every Gregorian date on the site carries its Hijri date (Umm al-Qura), taken from Umm al-Qura tables through an open conversion library, never converted by hand.
• Weeks run Sunday to Saturday, as SAMA publishes them.
• Hijri seasons: Ramadan and the two Eids move about 11 days earlier each year, so we flag when they shift between the two years compared. Each week in our calendar knows whether it contains Ramadan, an Eid or a salary date.
• Future dates: upcoming Ramadan and Eid dates are approximate until the official announcement (the actual date depends on the sighting). Government salary dates come from the Ministry of Finance schedule for 2026, and after it from the announced payment rule, marked “estimated”.

Sources and limits of the calendar: Calendar methodology.

Forecast accuracy: next week

Analysis, not official statistics. We measure how far our forecast of next week’s spending missed on past weeks, by comparing it with what SAMA later published.

• Basic method: the same week last year multiplied by the growth of the last 4 weeks.
• Hybrid method: the basic method outside the seasons; inside the Hijri seasons (Ramadan and the Eids) we compare with the same season in the two previous Hijri years, because the seasons move about 11 days earlier each Gregorian year.
• Measurement period: the last 52 weeks, 21 September 2025 – 19 September 2026 · 29 Rabi' al-Awwal 1447 – 8 Rabi' al-Akhir 1448 AH. No week’s forecast sees any data after it.
• Mean error % (MAPE): the average gap between forecast and actual, as a share of the actual. Lower is better.
• Error versus simple repetition (MASE): the method’s error divided by the error of repeating the same week last year; below 1 means more accurate than simple repetition. Each week’s error is divided by the average error of simple repetition over all earlier weeks, then averaged over the measured weeks; it is calculated for each series alone, and series are never pooled into one number.
• Series: one continuous weekly spending line since the start of the history: the whole Kingdom, one city, or one activity at Kingdom level.

Scope of these figures: the accuracy published here applies to forecasting next week only, and not to forecasts further ahead (a month or a quarter). Accuracy at longer horizons is still being measured and reviewed, and has not been approved.

The hybrid was tested on past data: the season rule was chosen on the older period, and accuracy was measured on the last 52 weeks (one Ramadan) and on the last 111 weeks (two Ramadans). It was more accurate than the basic method in all 26 series in both periods. The method itself was designed after seeing the whole history, so the decisive test is the first season it has not seen: Ramadan 1448 AH.

The Kingdom, 10 cities and 15 Kingdom-level activities (last 52 weeks)

The hybrid beat the basic method in 26 of 26 series over the last 52 weeks. Outside the seasons the two methods are identical; the whole difference comes from 12 seasonal weeks (in 20 of 26 series; 4 to 17 weeks in the rest, depending on the season definition chosen for each). Over the last 111 weeks (4 August 2024 – 19 September 2026 · 29 Muharram 1446 – 8 Rabi' al-Akhir 1448 AH), with two Ramadans, the hybrid was more accurate in 26 of 26. See the table for that longer period(Arabic).

Highest and lowest expected figure (the range): designed so that the actual figure falls inside it in about 8 of every 10 weeks. Over the last 52 weeks it held 80% of weeks on average (73% to 88% by series), and less than 80% in 13 of 26 series; the spread between series is within what 52 weeks produce by chance (about ±5 points). Four alternative ways of building the range were tried on the older period and none was chosen.

Caveats

1. A forecast of next week only, for the spending of a whole city or activity, not for any single shop. Figures are before removing inflation.
2. The season rule was chosen on the older period and measured on 52 weeks not used to choose it; the design of the method itself was set after seeing the whole history. The measurement period contains one Ramadan, one Eid al-Fitr and one Eid al-Adha, and the series are related, so they do not count as independent trials.
3. Outside the Hijri seasons the two methods are identical, so the whole difference between them comes from 12 seasonal weeks. The first real test on an unseen season is Ramadan 1448 AH (8 February – 8 March 2027 · 1–29 Ramadan 1448 AH).
4. The range holds about 8 of every 10 weeks on average; in a particular series it may be 5 points lower or higher by chance.
5. The city × activity level is not measured (its history starts 21 September 2025 (29 Rabi' al-Awwal 1447 AH)). Source: Saudi Central Bank (SAMA), weekly point-of-sale data; DataPulse calculations. Last week: 13–19 September 2026 · 2–8 Rabi' al-Akhir 1448 AH.

The table is recalculated automatically every month. Last week in the current calculation: 13–19 September 2026 · 2–8 Rabi' al-Akhir 1448 AH. The Hijri calendar in the season calculation is tabular and may differ from Umm al-Qura by a day.

The forecast is an estimate whose error is measured here; it is not a guarantee of what will happen.

Forecast log: every weekly estimate and what happened next