VCIx Methodology

1. What the number means

VCIx is a chain-linked composite of five pillars. Each constituent series is rebased to 100 at its own first month in the panel, and the composite is published against the base below.

It means: where the measurable parts of the Venezuelan economy stand relative to the base month, given everything we could actually measure.

It does not mean Venezuela is that percentage of any pre-collapse peak, and it is not a like-for-like comparison against the base year — see section 4, which quantifies exactly how far from like-for-like it is. Read the level as a direction of travel, not as a multiple.

An earlier version of this page said the index was benchmarked to 2012, the pre-collapse reference year. It was not, and it cannot be. Every working collector begins in 2021 or later, and when this was audited on 2026-08-07 only a fifth of the pillar-assigned series carried any 2012 observation at all. Publishing a 2012 base would mean inventing values for the rest, which the provenance rule in section 5 forbids outright. The base is late, and that is the honest option.

2. How it is built

Six steps, in order. Each one exists because the version without it produced a number that was wrong in a specific, documented way.

1. 1

   Eligibility

   A series enters only if it is active, assigned to a pillar, has at least six months of panel history, and has a non-zero base value. Nominal exchange-rate levels are excluded entirely: a rate is a price, it moves one way under sustained inflation, and its depreciation is already captured by step three. Charging it twice would make recovery arithmetically impossible while inflation persists.

2. 2

   Monthly resampling

   Every series is resampled to one value per calendar month — the last observation in that month is that month's reading. This happens before any aggregation, so that a daily FX series and an annual GDP print contribute like-for-like monthly observations. The predecessor averaged a daily tick's two-year change against an annual print's two-year change as though they were the same quantity.

3. 3

   Carry-forward, with a budget

   Between releases a series holds its last value, but only for a bounded number of months: two for daily, weekly and monthly series, six for quarterly, twenty-four for annual. Past that budget the series is treated as absent, not as flat. Publishing a stale value as though it were current is the quiet way an index goes wrong.

4. 4

   Deflation and direction

   Bolívar-denominated quantities — bank credit, deposits, the consumer basket, the local equity index — are converted to USD at the official rate before entering the panel. Without this the index measures currency debasement and reports it as growth: the first build of this engine printed a banking pillar of 157,269. Series where a falling number is an improvement (inflation, depreciation, the FX trust gap) are inverted, so narrowing reads as progress.

5. 5

   Rebasing, and a bound on any one series

   Each series is set to 100 at its own first panel month. Because an inverted series is divided by its current value, a rate trending toward zero has an unbounded score — a trust gap narrowing from 30% to 2% scores 1500 on a base of 100, and at a 30% pillar weight that single ratio can set the headline. Each constituent score is therefore bounded to the range below. The band is symmetric, so it cannot bias the composite in either direction, and wide enough that real multi-year moves pass through untouched: a series doubling still reads 200, one halving still reads 50.

6. 6

   Aggregation and chain-linking

   Within a pillar, constituents are combined with a geometric mean — the standard for index numbers, and the only aggregation under which a doubling and a halving offset exactly. Pillars are then combined on the weights below, renormalised across whichever pillars have data. When the constituent set changes, both the old and new baskets are valued on the overlap month and the new one is spliced onto the old at their ratio, so the published level moves when the data moves rather than when a collector starts or stops.

2.5 The panel

The index does not auto-admit anything with enough history: the constituents are an explicit, versioned editorial decision, named here with the cadence and source grade behind each one.

3. What is in it right now

Live, from the most recent computation. A pillar with no qualifying series is published as no signal rather than as a neutral value, and its weight is redistributed across the pillars that do have data.

4. What this number does not support

Everything below is a known weakness of the current index, published because a reader cannot evaluate the number without it. None of it is hypothetical; each item is measured.

• Much of the level is chain arithmetic, not measured change

  Every time the constituent set changes, the new basket is spliced onto the old one to keep the series continuous. Each splice is individually correct — that is what chain-linking is for — but they accumulate, and the figures below show how much of the published level comes from splicing rather than from series that moved. A level built this way is not a like-for-like comparison against the base year. Widening the panel is what fixes this; it is the open work.

• The panel is thin

  The count below is the number of series actually contributing in the latest month. Some pillars are currently carried by a single constituent. A composite this narrow will move on single-series news, and month-to-month volatility should be read in that light rather than as economic signal.

• The panel does not cover the whole economy

  Panel v1.3 is a deliberately limited editorial basket. Admission now requires real observations at each series' own cadence, not carried-forward months; that fixes the earlier admission defect, but it does not make a small basket comprehensive. An economic change not represented by one of the panel's signals may not appear in the index.

• The base year is near the bottom

  The base month sits close to the trough of the collapse and of the hyperinflation. Ratios measured against a trough flatter by construction. This is a consequence of when collection began, not a framing choice, but it should temper how the level is read.

• Sources are uneven and some are brittle

5. Provenance

The rule this project runs on: every observation must be traceable to a real, checkable origin. A number that cannot be traced is fabricated, however plausible it looks. A series has provenance only if all three of the following hold.

1. 1A real, named, reachable source — a genuine institution or site with a working link. A placeholder or catch-all source name is not provenance.
2. 2A collector that can be re-run, or a documented manual entry that can be re-checked, reproducing the value.
3. 3A recorded methodology, or derivation from other provenanced series by code in the repository. Derived metrics inherit provenance from their inputs plus the formula, and are attributed to a derived source — never to an external institution they did not come from.

Prohibited outright: inventing values to fill a chart or an empty page; interpolating, extrapolating or backfilling a reasonable estimate and presenting it as an observation; attributing a number to an institution that did not publish it; and adding jitter to make generated data look observed. The last two are not hypothetical — both have happened here and are recorded below.

An empty pillar, a null value, or not currently measured is always the correct output when data is absent. The index publishes nulls rather than neutral placeholders, which is why some pillars above may read no signal.

What has been deleted

Two purges of fabricated or unsourced data, both on 2026-08-07. Backups of every deleted row are retained in the repository so the deletions are auditable rather than merely asserted.

6. Source grading

Every observation carries a source class and a confidence class. Higher classes take precedence when values conflict; lower classes are shown with explicit caveats rather than hidden.

• Class A

  Official statistical agency or regulator

• Class B

  Company filings and reputable secondary aggregators

• Class C

  Credible estimates and market monitors

• Class D

  Proxies and thinly documented signals

Confidence classes are assigned per observation, alongside a review status. What each actually claims — and what it does not — is in section 6.5.

6.5 What “verified” means here

Every observation also carries a review status — verified or pending. Read on its own, “verified” sounds like a person checked the number. That is not what it means on this site, and the gap between those two readings is worth stating outright rather than leaving a reader to assume the friendlier one.

Some data on this site is entered and checked by a person, and says so — the funding-round records behind the VC pillar, for instance, are a documented manual research pass, not a collector. Reading every verified row as unreviewed would be its own kind of overstatement. What changed is only that the verified label stopped being read as a promise that this happened on every row it appears on.

Confidence — high, medium or low — tracks the same thing as review status: how well the collection path is documented, not a forecast of how close the number will turn out to be. A high-confidence observation is fully documented and can still be revised if its source revises it; a medium- or low-confidence one is thinner on paper, not necessarily more wrong.

7. Collection cadence

Counted from the live catalogue, not from a list maintained by hand.

• daily43 series — Venezuela-Exposure Equity Basket, Banco de Venezuela — Close (BVL), Bolívar Daily Depreciation…
• weekly3 series — Venezuela 2031 Bond, PDVSA 2027 Bond, P2P Crypto Volume (LocalBitcoins, weekly)
• monthly20 series — Oil Exports (volume), Oil Production, Brent Crude Price…
• quarterly3 series — VC Capital Deployed, Mobile Subscribers, VC Rounds (quarterly)
• annual15 series — Air Passengers, Mirror Imports — Machinery, Mirror Imports — Electrical/Telecom…

8. Sources

Every source currently feeding an active series, counted from the catalogue. If a source appears here it has at least one live series behind it.

• un-comtrade

  feeds 5 series

  Class A

• bcv

  feeds 4 series

  Class A

• opec-momr

  feeds 3 series

  Class A

• world-bank

  feeds 3 series

  Class A

• cavenez

  feeds 2 series

  Class A

• world-bank-pink-sheet

  feeds 2 series

  Class A

• baker-hughes

  feeds series