Skip to content
EU Candidates

How we estimate an accession date

This is the plain-language methodology — what the forecast number means, how it's built, and what we do not claim.

Our guiding rule is epistemic honesty: facts and estimates are kept separate, every input is sourced, uncertainty is shown as a range with a confidence level, and the tool is built to admit what it doesn't know. We would rather show a wide, honest range than a precise, false one.

We never claim to know better than the EU or the government

The forecast is anchored on the actors' own targets — what the candidate government and the EU themselves say. Our contribution is not a rival opinion; it's the documented historical regularity that first targets slip, and a structural cross-check. We lead with their number and adjust it with the record.

Two independent methods, reconciled

For each country we run two estimates that don't share assumptions, then combine them:

  1. The outside view — reference-class forecasting. Take the official target (e.g. "membership by 2028") and add the slippage every past enlargement saw against its first firm target. The record: Bulgaria/Romania and the 2004 group hit their firm treaty-stage dates (~0 years); Croatia slipped ~2–3 years from its 2009–11 hope; Spain/Portugal ~3 years; the Western Balkans far more. Slip is small from a firm, late-stage target and large from an early aspirational one — so we scale it by how ready the country actually is. (This is the academically-standard method for de-biasing optimistic timelines.)

  2. The inside view — a structural model. A step-by-step decomposition of the work left (open the remaining clusters, close all 35 chapters with rule-of-law "Fundamentals" closing last, sign the treaty, ratify across 27 parliaments + the European Parliament, enter into force), run thousands of times with realistic durations to produce a range.

The headline is their blend. Where the two methods agree, we're more confident and the range tightens; where they differ, the range widens — we never hide a disagreement inside a single tidy number. For a country at war, the membership date stays open-ended, because membership cannot take effect during active fighting.

A frozen negotiation is treated honestly. Some countries are technically well-prepared but politically frozen — Serbia has not opened a single cluster or chapter since December 2021, held up by the rule-of-law question and the normalisation of relations with Kosovo. We don't pretend that work restarts today, and we don't invent a date for when the freeze ends. Instead the estimate adds an open-ended wait drawn from a steady year-by-year chance of unfreezing — usually a few years, with a real chance of much longer. That's why Serbia, despite being one of the most prepared candidates on paper, lands well into the 2030s with a wide range.

The inputs are sourced, not guessed

The numbers that most affect the result trace to citable indices, not our judgment:

The confidence score

Every estimate carries a confidence level. It is not "the probability the date is right" — it is how much of the picture is sourced, and how well the two methods agree: roughly input completeness × method agreement, capped by irreducible uncertainty (a country at war is capped lower no matter what). A country shows, in plain terms, "N of 23 inputs sourced" and what we'd most need next to sharpen it. As we source more, the estimate sharpens and confidence rises — the tool gets better the more it's fed, and is honest about where it stands today.

Calibration

A brand-new tracker has no track record, and we won't fake one. Instead we backtest the structural model against accessions that have already completed — running it for a comparable starting point and checking whether the actual outcome fell within the range it would have predicted. This lets the method be scored against reality (Croatia, the 2004 enlargement, Bulgaria/Romania) rather than merely asserted. The backtest is shown with its caveats — notably that pre-2020 accessions predate the cluster methodology and map only approximately.

What we do not claim