In September 2026, Romania recorded 6,912 new business registrations against 7,629 exits — a net loss of 717 firms. That is a larger hole than August’s 641, and it looks like the tenth month of a crisis. It is not easing. The gap narrowed from July’s -1,030 in absolute terms, but because the entire business population collapsed on both sides at once — 6,912 registrations in September versus 8,103 in August. Exit activity did not calm down; it shrank alongside the intake. The ratio that actually measures the relationship tells a different story: the health ratio — registrations divided by exits — came in at 0.91, below August’s 0.93 and only marginally better than July’s 0.91.

The critical number is not 717. It is 110.37 — September’s churn rate, the total exits measured against registrations, expressed as a percentage. Read that plainly: for every 100 enterprises that appeared in the register this month, 110 left it. And when we place September inside its own recent distribution rather than just comparing it to August, the “improvement” narrative disintegrates.

The ratio is stuck in the middle of a bad range — not at the good end

Across the ten months of the current negative run, from December 2025 through September 2026, the exit-to-registration ratio has moved in a band between 100.91 and 136.9:

• December 2025: 136.9
• January 2026: 124.25
• February 2026: 100.91
• March 2026: 108.11
• April 2026: 123.88
• May 2026: 116.5
• June 2026: 115.5
• July 2026: 109.32
• August 2026: 107.91
• September 2026: 110.37

September’s 110.37 is not at the benign edge of that distribution. It sits above the run’s low of 100.91 in February and above both of the two preceding months, and it is worse than the run’s rough midpoint — putting it squarely in the bad half of the range. The deceleration in the absolute gap from July (-1,030) to September (-717) is real arithmetic, but it was bought almost entirely on the exits side of the equation, not by a recovery in formation.

Notice what the line does after June: it drifts down, then ticks back up. September is the first month since April in which the ratio moved in the wrong direction. If exits were genuinely normalising, the ratio would continue falling while registration volume held steady or recovered. Neither happened.

Both sides of the ledger deflated together

The scale of the simultaneous contraction is what should alarm, not the size of the net figure. On the exits side, September’s 7,629 total compares with 8,744 in August, 12,079 in July and 15,657 in April — a 51% fall in exit activity in five months. On the registrations side, 6,912 new firms in September compares with 8,103 in August and 14,438 in March — a 52% fall over six months.

Two series falling by almost identical percentages will, mathematically, produce a shrinking absolute gap while the ratio between them barely moves — and that is precisely the pattern on the table. The deficit is getting smaller because the entire economy is doing less of everything. Romania is not retaining more firms. It is churning through a smaller stock of them, which is arguably worse: the same failure rate applied to a thinning base.

Worse than a year ago — and worse than its own seasonal baseline

The year-over-year picture removes any lingering doubt. September 2026 produced 6,912 registrations against 14,815 in September 2025 — a fall of 53.34%. The decline is broad across legal forms: SRLs down 58.49% to 3,878, PFAs down 44.29% to 2,751, and individual enterprises (II) down 52.36% to 232. The month’s 12-month moving average is 11,968 against a rolling average of 12,627 the month before — a decline of 659.

And September cannot be excused as a seasonal trough. It ranks 9th of 12 calendar months by mean registrations, with a mean of 11,385 — making it the fourth-weakest of the twelve. September’s actual 6,912 is roughly 39% below even that unremarkable baseline. August, which ranks 10th of 12 and so is the third-weakest month, should by seasonal logic have underperformed September; instead it produced 8,103 registrations, and September fell 14.7% behind it.

Exits fell too, but not enough to change the verdict: suspensions down 46.75%, dissolutions down 46.15%, deregistrations down 35.12% year-over-year. Deregistrations — the harshest category, removal from the register — remain the single largest exit channel at 3,909, roughly 51% of all exits and nearly four times the number of dissolutions.

Who is most exposed

The registered-agent and sole-trader end of the market is absorbing the worst of it in absolute terms. SRLs at 3,878 and PFAs at 2,751 account for 96% of all September registrations, and both are contracting sharply. The near-invisible forms — SNC, SA, SC, SCS, GIE — together registered just 21 entities for the entire month. Capital companies (CA) and general partnerships (SNC) both tripled from trivial bases, from 5 to 15 and 3 to 9 respectively, which is arithmetic noise dressed as growth.

Sectorally, every one of the top ten industries is down between 30% and 69% year-over-year. The steepest collapses are in manufacturing (-68.74% to 246 registrations), other services (-64.51% to 456) and education (-59.81% to 377). Construction and trade, two of the largest and most labour-intensive sectors, are both down more than 52%. Even the relatively resilient information and communications sector fell 29.88% to 737 — the smallest decline in the top ten, and still a collapse of nearly a third.

Geographically, deregistrations are concentrated where the economy is concentrated: Bucharest 747, Cluj 203, Ilfov 173, Iași 161, Constanța 160. This is not a regional accident in one depressed county; the exits are tracking the country’s productive core.

The conclusion: continuity, not recovery

September 2026 is the tenth consecutive month in which exits exceeded registrations. The run began in December 2025 and has cumulatively removed 17,473 more firms than it added, with the worst single month, December 2025, at -4,292. September’s -717 ranks third-mildest of those ten months — an improvement in rank, not in substance, and far from the mildest month of the run, which remains February 2026 at -119.

The honest reading is that the smaller gap is an artefact of proportional deflation on both sides. Exits are still 1.10 times registrations, the ratio is above its own ten-month low and above the two months that preceded it, and the trend in the health ratio is negative at -0.02 while the net-growth trend sits at -76. There is no month in this entire series where the ecosystem created more businesses than it lost. September does not break that pattern; it repackages it in smaller numbers, which is easier to mistake for good news than a large deficit would be.

Anyone presenting -717 as evidence of stabilisation should be asked a single question: stabilisation of what? The base is contracting by half year-over-year, the exit ratio has turned back upward, every major sector is shrinking by a third or more, and the count of months with positive net formation in this dataset remains zero throughout the entire ten-month streak. That is not a turning point. That is continuity — and the smaller the denominators get, the harder the next month’s arithmetic will be.