Vrancea earthquakes, 1700–2026

The big one isn’t overdue

40 yearssince the last big one. Of the 12 gaps before it, 2 were longer.

Romania’s last earthquake of magnitude 7 or more struck on 30 August 1986, under the Vrancea mountains. People often say the next one is “due”, that it comes “every 40 years”. On the drum, each row is a decade since 1700 and each red scratch is a big quake. A clock would line them up in a column. They don’t line up.

#The gaps

From 1700 to today, the ROMPLUS catalogue of Romania’s National Institute for Earth Physics (INFP) lists 13 Vrancea earthquakes of magnitude 7 or more. Between them lie 12 gaps: the shortest one year (1893–1894), the longest 56 years (1838–1893).

Today’s gap, 40 years, has been beaten 2 times: 1740–1790 (50 years) and 1838–1893 (56 years). Only 2 of the 12 gaps came out near 40 years, between 35 and 45: 1701–1738 and 1940–1977.

All the gaps

Between	Years
1701–1738	37.0
1738–1740	1.8
1740–1790	50.0
1790–1802	12.6
1802–1829	27.1
1829–1838	8.2
1838–1893	55.6
1893–1894	1.0
1894–1908	14.1
1908–1940	32.1
1940–1977	36.3
1977–1986	9.5
1986–today	40.2

#What a clock would look like

On a drum, anything that keeps time lines up. The circles are the “40-year cycle”, set as kindly as possible: started from 4 March 1977 and run backwards and forwards. On the drum they fall in one column, every fourth row. The real quakes don’t sit on it, and 2017 came and went without one.

The usual measure of regularity is the coefficient of variation of the gaps: 0 for a perfect clock, around 1 for chance. Vrancea’s is 0.75. A series drawn at random with the same number of quakes comes out this regular or more in 25% of cases. The test could have found a clock: a gamma renewal clock whose gaps vary by half the average gap would have shown up in 89% of simulations.

#Which one is real?

Three of the drums hold 13 quakes thrown at random between 1700 and today, with the same magnitudes. One is Vrancea. If you can’t tell which, Vrancea looks just like chance.

#A test on years it never saw

A clock can be put to the test. A clock model (the Brownian Passage Time renewal process used in hazard maps) was fitted only to the big quakes of 1700–1899. For every year of 1900–2025 it then gave the chance of a big quake, from the time since the last one. What it learned from 1700–1899 was mainly that big quakes sometimes come in pairs (1738 and 1740, 1893 and 1894), so it gives high chances right after each one. Beside it, the model with no clock: the same chance every year, 4.1%.

Over the 126 unseen years, with the 4 quakes of 1908, 1940, 1977 and 1986, the clock did worse than no clock on both scores: Brier 0.0325 against 0.0308 (lower is better) and log score −0.1568 against −0.1418 (higher is better). A stricter clock, its spread fixed at half the average gap (dotted line, added after the test was written), came out mixed: slightly worse on Brier (0.0312), slightly better on log score (−0.1409). On these years, the clock fitted to the old data guessed no better than having no clock.

#The same at magnitude 5

Big quakes are few, so the test above has limits. Quakes of magnitude 5 or more are more numerous and well recorded since 1973: 29 in the USGS catalogue, about one every 1.8 years, with gaps from 36 days to 4.5 years. Each row is now one year.

With 29 quakes, a gamma renewal clock whose gaps vary by half the average gap would have shown up in 99.8% of simulations; a weaker one could slip through. The test finds no clock: the coefficient of variation is 0.76, and the gaps do not differ from gaps drawn at random (Kolmogorov–Smirnov test, with the mean re-estimated in every simulation: p 0.76). No seasons show either: the 114 main quakes of magnitude 4 or more since 1940 spread across the months about as chance would spread them (χ² test on months, p 0.91).

#Romania keeps a clock

Same drum, different pen. The blue line is how much the Romanian Wikipedia page “Cutremurul din 1977” (the 1977 earthquake) was read each day since July 2015: views per million views of all Romanian Wikipedia, on a log scale. The column at 4 March is the anniversary. In 7 of 11 years it was the page’s busiest day. In years without a felt magnitude-5 quake, 4 of 5; it lost only in 2023, to the earthquake in Turkey.

On 4 March the page is read 43 times more than on an ordinary day of the year. The red scratches are quakes of magnitude 5 or more: on each of those days the blue pen jumps too.

#Interest does not grow with the quiet

If “it’s due” were a feeling that builds up, interest in earthquakes would climb as the quiet years pass. Nothing like that shows. Interest jumps when a quake comes, and in the 7 measured cases the extra reading above ordinary days halved within a day of its peak.

Top: weekly views of the Romanian Wikipedia page “Cutremur” (earthquake) per million views of all Romanian Wikipedia. Bottom: the share of Romanian media articles that are about earthquakes, from the GDELT archive. Weeks around 4 March, after quakes and in January, when the pages are read for school, were left out. In the gaps between magnitude-5 quakes, interest does not climb with time. The median correlation between interest and time elapsed is −0.14 on Wikipedia and −0.09 in the press. What is measured here is reading, not fear; but no rise in reading during quiet years shows up.

#Your day

Pick a day, your birthday for example. The drum shows how long it had been since the last big quake, and where that gap would sit among the others.

Day

#What others say

The clock idea has serious sources. In 1979 the seismologist Gheorghe Purcaru described “quasi-cycles” in Vrancea’s quakes. On them, in 2006 and 2007, the Japanese seismologists Nobuo Hurukawa and Masajiro Imoto, the first together with Mihaela Popa and Mircea Radulian of INFP, built a model in which big quakes climb periodically in depth. A 2006 abstract called a magnitude-7 quake “by 2010” very likely; a 2007 one expected it in 2007, give or take 6 years. It did not come. Their model also used quakes from the 1500s and 1600s that newer European catalogues move out of Vrancea, as smaller, shallow quakes.

Others reached the same place as this page. Imoto and Hurukawa (2006) found that clock models fit the data no better than chance. Mircea Radulian, Mihaela Popa and Raluca Dinescu of INFP wrote in 2023 that big quakes come in clusters, that the gaps between clusters look random, and that no forecast can be made from statistics alone. They count gaps between clusters, not between quakes, which is why in their count the gap at the time, 37 years, had been beaten only once: by the 55-year gap of the 19th century.

The number 40 turns up mostly in the press, where particular gaps and years have been quoted, though no article found speaks of a strict 40-year cycle. Gheorghe Mărmureanu, honorary director of INFP, spoke in 2015 of 36–37 years between the very strong, less deep quakes (and 65–100 years for the deep ones), and in 2017 placed the next big one in 2040 or in 2077–2078. In October 2026 he said that “40 can no longer be it”.

#Questions

So a big quake isn’t coming?

It is, just not on a timetable. Vrancea has averaged one quake of magnitude 7 or more about every {n(m7["mean"])} years, but that is an average, not a schedule. It could come tomorrow or in 50 years. The data do not show that a long gap brings it closer.

Then why do people talk about a “40-year cycle”?

Because two easily remembered quakes fell that far apart: 1940 and 1977 (36 years), then 1977 and 1986 only 9. Only 2 of the 12 gaps since 1700 were between 35 and 45 years.

Don’t small quakes “let off pressure”?

Not in that way. Energy grows about 32 times per step of magnitude; it would take tens of thousands of magnitude-4 quakes to match one of 7. The page does not use this idea.

So how likely is a big quake this year?

The page gives no probability for this year or for a place: that needs a hazard model. What it shows is simpler: nothing in these tests says that a long gap makes a big quake more imminent. How hard the ground could shake where you live is on the “Cât de tare” page; how many seconds of warning you get, on “Câte secunde ai”.

Why 4 March?

On 4 March 1977 at 21:22, the magnitude-7.4 Vrancea earthquake killed 1,578 people, most of them in Bucharest. The anniversary is in the news every year, and people read about it then.

On the same subject

#Method

Big quakes: INFP’s ROMPLUS catalogue (downloaded 9 November 2026, years 984 to February 2026), quakes at least 50 km deep, within 45.2–46.1° N, 25.9–27.0° E. Two quakes less than 30 days apart count once, at the stronger magnitude. The threshold is moment magnitude 7.0, from 1700; before 1700 the catalogue has dates set conventionally to 1 January and gaps of over a century, which means missing records, not quiet.

The result was rechecked with other thresholds and start years (6.9 adds the quake of 30 May 1990; 7.2 keeps only the largest; starting in 1600 or 1800). In no case does the series come out more regular than chance (lowest p: 0.25). At 7.2 only 7 quakes remain, too few for a strong test; the page says so.

The European catalogues (SHEEC, EPICA, EMEC) take their Vrancea values from the same Romanian catalogue, so they are not an independent check; from 1700 on they give the same list of quakes of 7 or more. Where an independent check exists, the result holds: starting the list in 1750, the year from which SHEEC counts big Vrancea quakes as complete, p is 0.33; without the 1908 quake, which the instrumental ISC-GEM catalogue puts at only 6.6, p is 0.26, and today’s gap would have been beaten 3 times. Magnitudes on the drum are ROMPLUS values; instrumental measurements put 1977 at 7.5 and 1986 at 7.2.

The test was written before it was run, with the rules above. The results for magnitude-5 quakes (USGS, 1973–2026) and for the 4 March anniversary had already been seen; the page reports them as such. The strict clock in the 1900–2025 test (spread fixed at 0.5) was added afterwards and is marked so.

Wikipedia: daily views of the pages “Cutremurul din 1977” and “Cutremur” (people only, not bots), from the Wikimedia API, divided by all views of Romanian Wikipedia. Press: daily count of articles from Romanian sources about earthquakes, divided by all articles from Romanian sources, from GDELT DOC 2.0, by week.

What the page does not say: it gives no probability for next year; it does not claim that big Vrancea quakes have no long-term regularity at all (some papers argue for one, and 13 quakes cannot rule out a weak one); it says nothing about quakes larger than those in the catalogue.

#Sources

On INFP’s site the ROMPLUS catalogue carries no written licence; the version INFP deposited at Mendeley Data in 2022 is CC BY 4.0. The page publishes only figures derived from it and cites it.