How We Project to 2026 — and When We Refuse

Every 2026 figure on this site is a projection, and every projection says so. That is the whole policy in one sentence, but it hides a number of deliberate choices: which series we extend, how far we trust a trend, and — just as important — when we decline to project at all. This article walks through the mechanics, with real examples from the data, including the cases where the method visibly strains.

The scale first: of the 6,075 country-topic pages currently published, 1,901 carry a labeled 2026 projection. The rest show observed data only, each value tagged with its actual source year.

The method: a capped trend, nothing fancier

For each eligible series we take a window of up to the last eight observations, compute the compound annual growth rate between the first and last point of that window, and extend the latest observed value forward to 2026 at that rate. Two guardrails apply:

  • The growth rate is capped at ±15% per year. No series is assumed to grow or shrink faster than that, however steep its recent history.
  • Shares of a population are clamped. A projected percentage of adults can never leave the 0–100 range and is never shown above 99.5%, because “everyone” is not a projection we are willing to print.

Each projection states the window and rate it used — for example, “Projection based on 7-year trend (CAGR 15%); not an official figure.” It is an endpoint-to-endpoint trend, not a regression and not a forecast model. It cannot see a regulatory change, a currency crisis, or a product launch coming, and we say so on the methodology page.

Where the guardrails actually bind

The caps are not decorative. In the current dataset, 446 projections run at the +15% ceiling and 22 at the −15% floor — together roughly one in four. A further 29 percentage-of-adults projections hit the 99.5% clamp. Ranked from fastest assumed growth to fastest assumed decline:

Trend usedCountry and topicLatest observedLabeled 2026 projection
+15%/yr (capped)Albania, mobile money agents (IMF FAS)1,592.1 per 100,000 adults (2024)2,105.6 per 100,000 adults
+15%/yr (capped)Angola, registered mobile money accounts (IMF FAS)339.3 per 1,000 adults (2024)448.7 per 1,000 adults
+15%/yr (capped, clamped)Iraq, internet adoption (World Bank WDI)81.5% (2024)99.5%
+15%/yr (capped, clamped)Myanmar, internet adoption (World Bank WDI)45.4% (2020)99.5%
−15%/yr (capped)Sri Lanka, remittance outflows (World Bank WDI)$166.1 million (2024)$120 million
−15%/yr (capped)Russia, remittance inflows (World Bank WDI)$1.9 billion (2024)$1.4 billion

The Myanmar row is the honest weak point of the table. The last observation is from 2020, so the projection extrapolates six years at the capped rate and slams into the 99.5% clamp. We publish it anyway — labeled, with its window and rate — because hiding it would be a quieter form of dishonesty. But it is a good illustration of why a projection built on stale data deserves extra skepticism.

Refusal one: survey topics never get projected

Of the 47 published topics, only 15 carry projections at all. Every topic built on the World Bank Global Findex — account ownership, digital payments, mobile money accounts among adults — is excluded by design. Findex is a household survey run in waves, not an annual registry. Nigeria’s mobile money account series has exactly four points: 2014, 2017, 2021 and 2024, rising from 2.3% of adults to 32.8%. Fitting a smooth annual growth rate through survey waves years apart, then extending it, would manufacture precision the source does not contain. The same logic excludes interest rates, remittance costs and balance-sheet ratios, where a trend extension answers no useful question.

Refusal two: the series itself disqualifies

Even within the 15 projectable topics, 1,905 series were eligible and 4 were refused. The rules are mechanical: at least two observations in the window, both endpoint values strictly positive, and the latest observation dated before 2026. The failures are all zero endpoints. A zero in source data sometimes means zero and sometimes means not-reported; the zeros that are really not-reported markers — a remittance flow of exactly $0 wedged between billion-dollar years, say — are removed by the source-cleaning rules before any derivation, with every removal logged to the public audit file. The plausible genuine zeros that survive still refuse a growth rate, because a trend computed from zero is either undefined or absurd: Palau’s internet adoption trend window opens at 0% in 1990, and Eritrea’s secure internet servers window opens at exactly zero per million people in 2017. So those pages show the observed series and no 2026 figure — a blank we consider more informative than a guess.

How to read all of this

Treat every 2026 number here as a direction-of-travel estimate with stated assumptions, not a forecast. The observed values — with their real years — are the load-bearing part of the site; the projections are a clearly fenced extension of them. When the fence fails, we leave the field empty.

Sources & method

All figures above were computed from FinStatGlobe’s derived per-country datasets, which are built from snapshots of the World Bank World Development Indicators, the World Bank Global Findex, and the IMF Financial Access Survey. Counts of projections, caps, clamps and refusals come from scanning those derived files directly. The full projection rules, including the capped-CAGR definition and the data-quality checks, are documented on the methodology page. All figures are estimates and projections; nothing on this site is investment advice.