The third quarter of 2026 leaves two verdicts that pull in opposite directions. Measured by what companies did — exits versus registrations — the quarter is the steadiest stretch in ten months: the net gap narrowed from −1,030 in July to −641 in August and −717 in September. Measured by how many companies were created, the quarter’s last two months were the two weakest of the entire downturn, with September’s 6,912 registrations the lowest single month of 2026 so far. Both readings are accurate. They describe different halves of the same ecosystem, and the tension between them is the story of the quarter.

The exit gap improved even as the streak lengthened

The most important structural finding of Q3 does not fit the usual framing. Net business formation stayed negative for a tenth consecutive month, a run that began in December 2025 and has now accumulated to −17,473. That streak is measured only from July 2024, the earliest month with reliable exit data, and cannot be compared to longer historical records.

But within the quarter, the size of the monthly deficit shrank sharply. July’s net gap of −1,030 marked an improvement on each of the four preceding months — April at −3,018, May at −2,051, June at −1,748. All three Q3 months sit in the milder half of the ten-month run, and September’s deficit ranks third-mildest of the ten. August’s −641 was the mildest month of the run so far.

Crucially, that improvement in the net figure did not come from more businesses being formed. It came from fewer businesses leaving. Total exits fell from 12,079 in July to 8,744 in August and 7,629 in September. Every exit category contracted on both a monthly and a year-over-year basis: September’s dissolutions of 2,875 were down 46.2% from 5,339 a year earlier, suspensions fell 46.8% and deregistrations 35.1%. That pattern held across the whole quarter — dissolution declines of 18.0% in July and 31.0% in August.

The business churn ratio tells the same story from a different angle. July’s ratio of exits to registrations stood at 0.91, August at 0.93 and September at 0.91 — tightly clustered, and a marked improvement on June, when the register recorded 1.16 exits for every registration. By comparison, September 2025 posted 0.87 exits per registration, which was better on the ratio measure precisely because registrations were far higher.

Registrations: a slide, not a plateau

On the formation side, the quarter deteriorated month by month. The series ran 11,049 in July, 8,103 in August and 6,912 in September — a 14.7% decline from August to September. This is not a flattening; it is a steepening descent.

To be precise about how low September sits: it is the weakest registration month of 2026, below January’s 11,075 and below the previous low of 8,103 set in August. Registration volume and the net-growth deficit do not move together, though: within the current ten-month run, the mildest net deficit came in February, when registrations ran at 13,139 — nearly double September’s level — because exits that month were unusually low too. September’s 6,912 is also close to the 6,968 recorded in July 2024, the first month of the trustworthy exit window, and is the second-lowest registration month in that entire window — trailing only August 2024’s 5,288.

The 12-month moving average, which had already begun to sag, dropped from 12,904 in July to 12,627 in August and 11,968 in September — a fall of 659 units in September alone, the largest single-month drop of the three.

The year-over-year contrast is severe. Every month of Q3 was down by more than a quarter: July’s 11,049 was 27.2% below July 2025’s 15,168, August’s 29.1% below and September’s 53.3% below. September’s gap is not a marginal miss: 6,912 against 14,815 a year earlier is a fall of nearly 7,900 registrations in a single month.

Is the summer-to-autumn path normal? Partly — but not to this degree

Seasonality does explain a good deal of the intra-quarter shape. By long-run average registrations, August ranks tenth of twelve at a mean of 10,681 — the third-weakest calendar month, not the weakest. September ranks ninth at 11,385 and July seventh at 11,926. A July-to-August slump is therefore an expected feature of the pattern, not an anomaly, though August’s weakness is milder than the raw numbers sometimes suggest.

What is not ordinary is the magnitude and the persistence of the decline into September. In 2025, the same three months ran 15,168, 11,435 and 14,815 — and September actually recovered, ending 2.3% below July. In 2026, September finished 37.4% below July. The autumn rebound that the seasonal profile would predict did not materialise. September’s 6,912 is roughly 39% below the long-run September mean of 11,385.

The quarter-level comparison confirms this. Q3 2026 totalled 26,064 registrations against 41,418 in Q3 2025 — a shortfall of 37.1%.

SRL weakness explains more than broad decline

The composition of the drop is the most diagnostically useful part of the quarter. Since the site’s ongoing debate has turned on whether mid-2026 weakness is SRL-specific or broad-based, the Q3 sector data is worth stating carefully.

In July, two sectors actually grew year over year: transport and storage rose 4.4% to 2,307, and information and communications rose 2.5% to 971. By August, transport was down only 6.0% and IT down 2.7%.

Then September broke the pattern. The entire top of the sector table fell by roughly half: transport 1,181, down 45.1%; retail and vehicle repair 1,004, down 52.6%; professional, scientific and technical activities 722, down 50.5%; construction 628, down 53.6%. Even information and communications, the quarter’s most resilient large sector, fell 29.9% in September. The two sectors that had held up through July and August did not hold up through September — though IT’s September drop was notably shallower than the 45–54% declines around it.

Entity types split along similar lines. In September, SRLs fell 58.5% year over year to 3,878, while authorised natural persons (PFA) fell 44.3% to 2,751. Over the quarter, SRL registrations ran 5,963 in July, 4,419 in August and 3,878 in September — a 35.0% decline across three months. The earlier “SRL-specific” diagnosis fits July and helps explain the summer; it fits September less well, because by then the pullback had spread across essentially every large sector and both principal entity forms, with SRLs simply falling furthest.

Where the quarter leaves the ecosystem

Two things can be said with confidence and they point in different directions. The exit side of the ledger genuinely stabilised: exits contracted faster than registrations did, the churn ratio held near 0.91 for three straight months, and Q3 was a clear improvement on the second quarter, when net formation ran −1,748 in June alone.

The formation side did not stabilise. It worsened steadily through the quarter and ended at the lowest monthly registration figure of 2026. The geographic concentration also sharpened: Bucharest accounted for 25.7% of September’s national total, up from 22.7% a year earlier. Meanwhile the counties that fared least badly in September still fell sharply: of the month’s five smallest year-over-year declines, all were negative, from Ialomița at −35.4% to Covasna at −75.6%.

The honest summary of Q3 2026 is therefore not that conditions held steady, and it is not that the market is shedding companies at a faster rate. It is that the stock of new businesses is shrinking markedly — and now shrinking across the board rather than in one segment — while the outflow of existing businesses has simultaneously slowed, narrowing the net gap from the exit side rather than the entry side. That combination produces a flatter net line over a declining base, and it makes the formation trend, not the dissolution trend, the number to watch over the closing months of the year.