# Pre-registration — marfa-care-nu-a-ajuns

Written 10 Nov 2026, 19:20 GMT, before opening monthly data, peer-country data (BG, HU, PL, CZ) or the export-side mirror.

## Already seen (so not a test)
Annual Comext DS-045409, Romania vs 26 EU partners, 2007–2025, chapters (HS2) and HS4:
- Total: Romania's recorded arrivals exceed partners' recorded dispatches every year (gap −1% to −13%). The brief's hook "partners report billions more than Romania records" is false in total.
- By weight, partners' dispatches exceeded Romanian arrivals for fruit (ch. 08) by 52–62% in 2008–2010, 48% in 2011, 21% in 2012, 8% in 2013; vegetables (07) 27–48% in 2008–2011; cereals (10) 17–26% in 2010–2014; all negative from 2015–2016.

Definition: gap_kg = (X − M) / X, X = kg the partners recorded as dispatched to the country, M = kg the country recorded as arrived from them. Same for value.

## Hypotheses
H1 (Romania-specific). For fresh fruit and vegetables (HS 07+08, kg), the 2008–2011 mean gap of Romania is larger than that of each of BG, HU, PL, CZ, and at least 20 percentage points above the mean of the four. Fails if any peer equals or exceeds Romania or the difference is under 20 points.

H2 (timing vs the food VAT cut). Monthly data, HS 07+08, kg: the 12-month mean gap had already fallen by at least two thirds of its total 2010→2016 decline before June 2015 (the month VAT on food fell from 24% to 9%). Fails if less than two thirds of the decline precedes June 2015.

H3 (the VAT cut itself). Difference-in-differences on annual HS4 gaps (kg), food products that moved to 9% on 1 June 2015 (HS 02–04, 07–12, 15–21, excluding alcohol) vs non-food products (HS 25–97 excluding 27, 71, 87, 93, 99), 2013–2014 vs 2016–2017, value-weighted, bootstrap over HS4 products (2,000 draws). Prediction: the food gap falls relative to non-food by at least 5 percentage points with a 95% interval excluding zero. Placebo: the same contrast in the four peers, where no such cut happened; the Romanian estimate should exceed every peer estimate. Fails otherwise. Cereals (HS 10) reported with and without, because of the 2011 reverse charge.

H4 (the VAT rise). July 2010, VAT 19% → 24%: the gap for all products (kg-free, value) and for food rose in the 12 months after vs the 12 months before by at least 3 points. Fails otherwise. (Fruit was already at 52% in 2008, so the prior is weak.)

H5 (exports). Romania's recorded dispatches of food to the EU vs partners' recorded arrivals: no gap comparable to the import side (|gap| under 10% for HS 07+08 in 2008–2011). Fails if the export side shows a gap over 10% of the same sign pattern.

## Rules
- A failed test stays on the page with its number.
- A statistical gap is a difference between two records. It becomes "undeclared trade" only with the alternative explanations stated beside it: reporting thresholds, estimates for non-response, confidentiality, country of consignment, triangular trade, timing.
- No firm, person or partner country is accused. Quantities are "recorded by X and not by Y".
- VAT amounts are illustrations (gap value × rate), labelled as such, never "stolen".

## Addendum: results and deviations (12 Nov 2026)
- H1 pass. Romania 47.2%, Hungary 31.2%, Bulgaria 14.0%, Czechia −1.2%, Poland −5.0%; Romania first of 27.
- H2 pass. 84% of the 2010→2016 decline of the 12-month mean gap precedes June 2015.
- H3 pass on the Romanian estimate, weak overall. HS4 difference-in-differences −14.1 points [−24.8, −4.0]; without cereals −8.4 [−16.7, −0.4]. Deviation: the peer placebo was run on chapters (HS2) and not on HS4 headings, because the full HS4 download for other countries was throttled. At chapter level Romania −9.4 against Bulgaria +17.4, Hungary +12.4, Poland +4.4, Czechia −1.4, so Romania exceeds each of the four as predicted; but among all 26 other member states four (Germany, Slovakia, Netherlands, Luxembourg) show a change as large. The page reports this.
- H4 fail. Total +2.3 points, food +5.2, fruit and vegetables −8.1; the threshold was 3 points for both total and food.
- H5 fail. The export-side gap for HS 07+08 in 2008–2011 is 27.7%. Deviation: computed from the EU-wide chapter pull, since the dedicated export download was stopped.
- Not pre-registered, found in the monthly data: the break in spring 2012 and the ANAF sealing procedure (Order 506/2012). The page presents the timing and the documents; it does not claim a tested causal effect.