September's Halving Isn't a Base Effect — the High Base Was the Anomaly, and This Was the Weakest September in Ten Years AI Generated
Romanian business registrations fell 53.34% year on year in September 2026, from 14,815 to 6,912. The obvious defence is that September 2025 was an unusually elevated comparison point. That defence is correct on the arithmetic and useless as an explanation: once the base is stripped out, the number rebased against the calendar and set beside lifetime exits, the decline survives every test. What the high-base framing gets backwards is which month was the outlier. September 2026’s 6,912 registrations are the lowest September on record since 2017 at least, and the weakest single month in the entire current series — weaker than December 2025, a month that includes the year-end tail.
The base is real, but smaller than the headline
September 2025 was genuinely inflated. At 14,815 registrations it sat 30% above the ten-year September mean of 11,385 and 55% above September 2024’s 9,585. It was also the only month in the current window in which registrations exceeded total exits, producing a net gain of 1,864 businesses in a single month. Nothing about September’s seasonal position supports that level: September ranks ninth of twelve calendar months by mean volume.
But correcting for an elevated base cuts the decline roughly in half, not to nothing. Against the ten-year September average of 11,385, September 2026’s 6,912 is down 39%. Against the twelve-month moving average of 11,968 — which by construction already absorbs last September’s peak — it is down 42%. A base effect can explain a comparison point; it cannot explain why a month lands 42% below its own trailing trend.
Rebasing every way available, the conclusion holds
Test the number three ways.
Against the calendar. September 2026 is the lowest September in the ten years from 2017 to 2026. The full September series reads: 9,705 (2017), 9,585 (2018), 9,674 (2019), 12,138 (2020), 13,140 (2021), 15,951 (2022), 12,343 (2023), 9,585 (2024), 14,815 (2025), 6,912 (2026). The previous floor was 9,585, reached twice — and this September is 28% below it. The 2020 and 2021 figures should be treated as structurally distorted by pandemic restrictions on registration activity, not as economic signals; removing them changes nothing, because every clean-year September from 2017 to 2019 and 2023 to 2025 sits between 9,585 and 15,951.
Against the trend. The twelve-month moving average fell from 12,627 to 11,968, a drop of 659. Expressed per working day, September 2026’s 6,912 over roughly 22 working days is about 314 registrations a day, against roughly 674 a day in September 2025 — the same halving, because September has the same working-day count in both years. The per-day measure adds nothing beyond the headline; there is no calendar distortion to remove.
Against the intra-year path. September was not a one-month stumble. The 2026 sequence runs 14,438 in March, 12,639 in April, 12,427 in May, 11,277 in June, 11,049 in July, 8,103 in August, then 6,912 in September. That is seven consecutive monthly declines from March, and the two largest drops — from July to August and August to September — came right at the point where a genuine seasonal recovery would normally begin.
Exits fell too — and that is the real finding
The most important number in this release is not the registration figure. Exits fell even faster. Total lifecycle events dropped from 12,951 in September 2025 to 7,629 in September 2026: suspensions down 46.75%, dissolutions down 46.15%, deregistrations down 35.12%. Because exits fell harder than entries, the churn rate actually rose from 87.42% to 110.37% — but only because the ratio is exits over a collapsing registration base.
The net position tells the more honest story. September 2026 recorded a net loss of 717 businesses, taking the ecosystem health ratio to 0.91. That extends a run of net-negative months to ten, beginning in December 2025, with a cumulative net loss of 17,473 businesses. September 2026 is the third-mildest month in that run — the streak’s severity has been easing. The exit side of the ledger, in other words, is not deteriorating; the entry side is not replenishing. September 2025 was the last month in the current data window in which entries beat exits at all.
What kind of slowdown this is
The decline is not uniform, which matters for interpretation. Every major sector contracted, but by very different amounts. Manufacturing registrations fell 68.74%, other services 64.51%, education 59.81%, administrative support 58.46%, construction 53.55%, retail 52.62%, professional and technical activities 50.48%, real estate 46.1% and transport 45.1%. Information and communications was the clear outlier at −29.88%, roughly half the rate of the national average — the only sector to hold anywhere near its prior level.
By entity type, incorporated SRL formations fell 58.49% to 3,878 while sole-trader PFA registrations fell 44.29% to 2,751. The SRL share of new registrations, 56.1%, is the lowest in the series, continuing the shift identified since January 2026 toward unincorporated forms. This is consistent with the mid-2026 debate over whether the cooling is SRL-specific rather than broad: both halves of the market are contracting, but the incorporated, capital-committed half is contracting considerably faster.
Geographically the pattern is consistent rather than localized. Bucharest accounted for 25.7% of national registrations, with Ilfov, Timiș, Cluj and Iași following. Every county in the bottom-growth list fell — the least-bad performer among them, Ialomița, was still down 35.37% — so this is not a story of regional rotation away from the capital.
Why the base cannot carry the explanation
The base was elevated; that is documented and material. But a base adjustment is a one-time correction, and this decline does not behave like a one-time correction. If an inflated comparison point were the whole story, the twelve-month moving average would hold roughly flat and September 2026 would land near its seasonal norm. Instead the trend line fell 659, the month landed 39% below its seasonal norm, and it set a new ten-year floor for the month by a wide margin. The high base explains why the year-on-year percentage looks as dramatic as it does. It does not explain why the month happened.
There is also a compositional clue that the base story cannot accommodate. In September 2025 both entries and exits were running high — 14,815 registrations against 12,951 exits. That is the signature of a busy, high-turnover month, not a quiet one. In September 2026, activity on both sides of the ledger has fallen, but entries have fallen further. A base effect would depress the numerator of a single comparison. This is showing up in every denominator available: calendar-adjusted, trend-adjusted, working-day-adjusted and lifecycle-adjusted.
The honest summary is narrower than “structural collapse” and stronger than “base effect.” The Romanian business register is now in its tenth consecutive month of net contraction, and September 2026 was its weakest single month on record — but it was also one of the milder months within that run, and the exit side is improving in absolute terms. The constraint is formation, concentrated in incorporated companies and in manufacturing, construction and the trades, with information and communications the one sector holding its ground. That is a specific problem, not a general one, and it is not a trick of the calendar.