Every instinct says the July–September quarter was the rupture — three straight months of more businesses dying than being born, registrations down more than a third year-on-year. But run the quarter against its own history and the story inverts. On the metric that has been quietly deteriorating all year, Q3 was the mildest stretch of 2026. That is not reassurance. It is a warning about which number is actually breaking.

Let us start with the damage, because the damage is real and it is large. Romania registered 11,049 companies in July, 8,103 in August and just 6,912 in September. That is 26,064 new businesses across the quarter, against 12,079 exits in July alone, 8,744 in August and 7,629 in September.

Net the two sides out and the quarter lost 2,388 businesses. The count is not close. The direction is not contested.

The Outlier Isn’t There

Plot net growth month by month through 2026 and a shape emerges that the panic over September has obscured:

• December 2025: −4,292
• January 2026: −2,686
• February 2026: −119
• March 2026: −1,171
• April 2026: −3,018
• May 2026: −2,051
• June 2026: −1,748
• July 2026: −1,030
• August 2026: −641
• September 2026: −717

Q3’s combined loss of 2,388 is smaller than April and May alone in Q2, which together lost 5,069. It is smaller than December 2025 by nearly two thousand. The authoritative streak reading confirms it: of the ten months in the current run, only February and August were milder than September — September ranks third by mildness, a position of 3 of 10. The cumulative damage for the whole ten-month run, from December 2025 through September, stands at −17,473 — and seven of those ten months were logged before this quarter began, when the running total was already −15,085.

Read plainly: this quarter added barely 14% to a hole that was dug in the first half of the year. If July–September were a genuine break — a step-change, a new regime — it should have accelerated the erosion. It decelerated it.

The ratio test says the same thing. The registration-to-exit health ratio ran at 0.91 in July, 0.93 in August and 0.91 in September. Compare that to June’s 0.87, May’s 0.86 and April’s 0.81. Q3 is the best quarter of 2026 on this measure — the only quarter of the year whose three months all cleared 0.90.

So no: on its own history, this quarter is not an outlier. It is an improvement. Anyone calling it a breach is looking at the wrong column.

Why the Quarter Looks So Bad — And It’s Not the Reason You Think

So why does September feel like a cliff? Because of the other side of the ledger, and because of last year.

Registrations fell 27.2% in July, 29.1% in August and a staggering 53.3% in September. The quarter as a whole registered 26,064 businesses against 41,418 a year earlier — a collapse of 37%.

But exits fell even faster. Dissolutions dropped 46.2% year-on-year in September, suspensions 46.8% and deregistrations 35.1%. The business register is not processing a wave of deaths. It has gone quiet on both sides. That is the signature of a register that has stopped recording, not an economy that has stopped working — and it is precisely why the health ratio improved.

Which brings us to the number that should actually alarm you.

The Twelve-Month Average Turns Negative

Strip out the monthly noise and the picture sharpens into something uglier than any single month. The 12-month moving average of registrations has rolled over: it now stands at 11,968, down 659 from the prior reading. A year ago at the same point, the same measure was rising by 436.

That is a swing of 1,095 between an expanding trend and a contracting one. Individual months can be blamed on holidays, on filing calendars, on the seasonal trough. A moving average cannot. The trend line has changed sign, and it changed it before this quarter, not during it.

This is the honest version of the outlier story: the quarter isn’t the anomaly. The quarter is where a pre-existing anomaly became impossible to ignore.

Seasonal Cover Doesn’t Hold

The reflex is to call July–September a summer lull. The seasonality data refuses. Across ten years of observations, September ranks 9th of twelve calendar months — fourth-weakest — and August ranks 10th, third-weakest. Neither is the trough. July, by contrast, ranks 7th, placing it in the upper half of the calendar, with a ten-year mean of 11,926 registrations.

And September’s historical mean of 11,385 is higher than August’s 10,681. Yet this September (6,912) came in below this August (8,103) — the wrong way round relative to a decade of pattern.

Seasonality can explain roughly a tenth off a July. It cannot explain subtracting 4,119 from July, or 7,903 from September. The seasonal defence is dead.

Where It Hurts: SRLs, and Everything Real

The damage is concentrated where it matters most. Company-form (SRL) registrations fell 58.49% year-on-year in September, from 9,343 to 3,878. Sole-trader authorisations (PFA) fell 44.3%.

Industry-level readings for September are uniformly brutal, and uniformly so — which is itself a diagnostic clue. Transport and storage: −45.1%. Retail and wholesale trade: −52.62%. Professional, scientific and technical activities: −50.48%. Construction: −53.55%. Other service activities: −64.51%. Administrative and support services: −58.46%.

Real economies do not shed half their formation across unrelated sectors simultaneously. Registers do. When hauliers, shopkeepers, consultants, builders and educators all fall by half in the same month, the common factor is not demand — it is the pipeline.

Geographically the same signature. Bucharest alone supplied 1,779 of September’s 6,912 registrations, 25.7% of the national total. The steepest county declines are in the small, rural, low-volume counties — Covasna at −75.58%, Mehedinți at −62.5% — where a handful of unprocessed files moves the percentage by fifty points. Small counties are where administrative backlogs show up first.

The Verdict

Does the July–September quarter stand out as an extreme outlier within the trustworthy exits window?

No — not on net loss, and not on the registration-to-exit ratio. On both, it is the least extreme quarter of 2026. The streak it belongs to began in December 2025, and its worst month was the first one, nine months before this quarter opened. Anyone who tells you the summer was the break is not reading the series.

But the shape of the quarter is a break in a different dimension. The health ratio stabilised precisely because exits collapsed at a similar or faster rate than registrations. That is not a market healing — it is a market and a register both seizing up. And the 12-month moving average, which nothing seasonal can distort, has now turned from +436 to −659 in a single year. The trend, unlike the quarter, is a genuine break.

So the alarming conclusion is not that Q3 was the worst three months Romanian business formation has ever had. It is that Q3 was merely average for 2026 — and that “average for 2026” now means bleeding more companies than you create, every month, for ten months running, while the annual trend line crosses zero going down. The crisis is not a spike. It is a plateau, and plateaus are harder to notice and much harder to reverse.

The single most useful thing to watch next is not the September registration count. It is whether October’s moving average falls by more than September’s 659. If it does, the seasonal explanation is finished for good.