-
Did four European stock markets get more volatile together? A changepoint check
The same changepoint detector on four European indices over 1,859 days: DAX and SMI show the volatility rise within 7 days, FTSE 60 later, CAC not at all.
-
Does a seasonal-drift warning cry wolf on a clean series?
We tested the changepoint detector's drift warning on 195 synthetic months with no shock. It stays quiet on white noise and fires on autocorrelation alone.
-
A longer series did not help the changepoint detector tell a shock from a season
195 months of euro-area factory orders through the 2008 crash: the detector found the collapse, warned about a drifting season, and cut 16 to 22 steps in all.
-
Does allowing for autocorrelation hide a drifting season? On UK gas, yes
The changepoint detector's --ar1 flag cuts false drift warnings on sticky noise. On real series it keeps the airline warning but loses UK gas.
-
Shape or shock? A change-point warning I could not sharpen
The changepoint-detector's drift warning fires on a real shock as well as on a growing seasonal shape. A cancellation test looked promising and failed.
-
Shape or shock? A second try that mostly failed
Can a changepoint detector tell a growing seasonal pattern from a real shock? A level-scaled refit helped on one of three real series.
-
Telling a growing season from a real shock: a third try that mostly failed
Refitting on the log clears one growing seasonal series but not the airline one, so the changepoint drift warning stays as it is.
-
Does a change detector find a hurricane? Puerto Rico's daily deaths
The changepoint detector on 1,201 real days of Puerto Rico deaths dates Hurricane Maria to the day in five of five setups, with nine or more other changes.
-
Testing a changepoint warning on 9/11 and SARS
The free changepoint detector's seasonal-drift warning fired in 7 of 8 runs on real Australian visitor data. What that tells us, and what it doesn't.
-
Thirty real airline series, none of them quiet: testing a changepoint detector on the 1989 pilots' strike
The changepoint detector's date ranges held the start of the 1989 Australian pilots' strike in 29 of 30 real passenger series (count model). None was quiet.
-
When noise alone looks like a drifting season
A changepoint tool's drift warning fired on 75% to 88% of series with no change at all. Allowing for autocorrelation cut that to about 1%.
-
Is it a burst or two level changes? A test that says no on real data
changepoint-detector now flags when two count changes are one temporary burst. On the classic great-discoveries series it declines, rightly.
-
A five-year burst is not two level shifts: three ways to read the great-discoveries counts
Real yearly counts of great inventions, 1860-1959, run through the changepoint-detector's count options. Two decline the burst, one half-fixes it.
-
Why a changepoint detector found 16 changes in a company's growing earnings
Johnson & Johnson's quarterly earnings, fed raw to a changepoint detector, come out as 12 to 16 steps. Logging the series first leaves 2 to 5.
-
Does the 'burst or two steps?' note work? 100 simulated series per case
On simulated Poisson counts the note names 84 to 97% of true bursts and wrongly calls 1 in 100 lasting steps a burst.
-
A --log flag for series that grow by a percentage
The changepoint detector cut J&J earnings into 12 steps. A new --log flag logs the series first, leaves 2 to 5 changes, and reports in original units.
-
We tried a narrower drift warning in the changepoint detector, and kept the old one
A shorter window cut false alarms on monthly data by 4 series in 200 but lost 71% of detections on packed steps, so the rule stayed at two cycles.
-
One burst or two changes? Reading the 1884 discoveries spike
Plain Poisson change detection cuts a five-year burst into a rise and a fall. A new --episodes flag labels it as one burst; we measured where that goes wrong.
-
The changepoint page can now take logs first, so growth stops looking like a staircase
On Johnson & Johnson's quarterly earnings, a level-change detector cuts 12 steps from smooth growth; a new checkbox on the page logs the series first.
-
A seasonal-drift warning, tested on five real series: right four times, wrong once
The detector warns when steps pile up inside two seasonal cycles. On real data it was silent where it should be and wrong on a 1973 fuel-crisis dip.
-
A changepoint detector reads two odd months as four steps
On 20 years of Nottingham temperatures with no real step, the default mean model cuts four changepoints; a longer minimum segment or meanvar cuts two or none.
-
A five-year burst is not two level changes: what a changepoint detector said about great discoveries
On 100 years of counted inventions, plain Poisson changepoint detection finds three confident changes; a bootstrap penalty finds none. Which one to believe.
-
An interval that admits where it is optimistic
The changepoint detector's growth-change range holds the true bend 94% of the time on clumpy counts, but 82% on steep bends. The CLI and JSON now say so.
-
A multiplicative season did not fix the airline passengers series
changepoint-detector on the classic monthly airline passengers data: 21 false steps with an additive season, 22 with a multiplicative one, 8 after taking logs.
-
Heavy-tailed noise fooled the Bayesian changepoint view too: clipping fixed it
On t3 noise with no change, changepoint-detector's probability view flagged a change in 47% of 200-point series. Clipping at 2 robust sds cut that to 1%.
-
Fewer false alarms on volatile data: a bootstrap penalty for changepoint-detector
changepoint-detector's --bootstrap sets the meanvar penalty from block-resampled residuals, cutting false alarms on GARCH-style noise from 51% to 16%.
-
Two ways to keep the tail in a bootstrap penalty, and neither worked on the franc
changepoint-detector tried a robust scale for its bootstrap penalty on real daily FX returns: one changed nothing, the other erased the 2015 Swiss franc shock.
-
Changepoint detector: the count bootstrap stops working when you add a trend
A measurement of --bootstrap on clustered counts combined with --trend, a weekly season and negative-binomial noise.
-
Changepoint detector warns when drifting-season steps pile up
With a drifting seasonal shape, a cluster of steps usually means the shape is still changing, not that regimes did. The detector now says so.
-
How often does the changepoint detector's drift note cry wolf?
On 1,600 simulated seasonal series the drift note fires on 2-3.5% with no change, but on 11.5% of monthly series with two real steps.
-
A GARCH filter fixes false alarms in changepoint detection, then breaks real ones
Dividing a series by a fitted GARCH volatility cut false change alarms from about 50% to 1% but hid a real variance break 94 times in 100. A negative result.
-
Holidays do not break the count bootstrap, a growth trend does
changepoint-detector on simulated daily counts with known holidays: the block bootstrap cuts false alarms from 85% to 21% on clustered counts.
-
A season that grows with the level: changepoint-detector gets --multiplicative
A seasonal swing that grows with the level reads as steps when removed as a fixed amount. A multiplicative option helps monthly data and hurts quarterly.
-
A changepoint tool on a series where the answer depends on the scale
UK quarterly gas 1960-1986 through changepoint-detector: a slope bend near 1970 on the raw series, none on the log series, and a seasonal swing cut into steps.
-
Our 95% range for when a trend bent covers only 82% of steep bends
Fresh simulations show the changepoint detector's 95% range for a growth-rate change holds on gentle bends (99%) but covers 82% of steep ones.
-
Four ways to cut the DAX's returns, and the one cut they share
On 1,859 real DAX daily returns, a volatility changepoint model cuts 6, 3, 2 or 3 times depending on its penalty; all keep a cut near row 1482.
-
Does scanning for autocorrelation help find real level shifts? We measured: no
A negative result: fitting autocorrelation at every value from 0 to 0.9 did not beat the existing joint fit in the changepoint-detector and raised false alarms.
-
Is the AR(1) option hiding Lake Huron's drop? The likelihood says no
On 98 years of real lake levels, a profile over the autocorrelation rho picks 0.8 and no change point; the 1923 drop appears only with near-white noise.
-
Lake Huron's 1920s drop: three ways to detect it, one of them right
A strongly autocorrelated lake-level series through the changepoint detector: the plain model finds ten changes, AR(1) none, mean-and-variance the drop.
-
Does forgetting a holiday break a changepoint detector? Barely
On simulated daily counts, telling the changepoint-detector none, half or all of the true holidays barely changed false alarms or detection.
-
Changepoint detector: a bootstrap penalty for clumpy counts
Counts that come in busy and quiet runs fool a Poisson changepoint test; a block-bootstrap penalty cut false alarms from 89% to 11% on synthetic series.
-
Checking the changepoint detector on 112 years of coal-mining disasters
The textbook Poisson changepoint series, run through changepoint-detector: it finds the fall near 1892 and its 95% range holds the published 1890.
-
Making a volatility-break detector less jumpy on messy noise
changepoint-detector's
meanvarmodel looks for changes in a series' level or volatility. It assumes independent Gaussian noise, and on heavy-tailed or... -
A bootstrap penalty for the mean model cuts false alarms on autocorrelated noise
changepoint-detector --bootstrap now covers the mean model: on AR(1) noise at 0.3, false alarms fall from 80% to 13%; a 2 sd step is still found 98% of runs.
-
Changepoint detector on real currency shocks: which volatility option helps?
Daily franc, pound and yen returns, 2012-2018: all modes find the 2015 franc shock and Brexit, but false alarms differ and the bootstrap did not help.
-
Finding the UK seat-belt law in 192 months of road casualties
The changepoint detector dates a documented 1983 drop in British driver casualties to the right month, and shows why count data needs the overdispersion option.
-
Why the bootstrap penalty missed the Swiss franc shock
On real franc returns the changepoint-detector bootstrap barely raised its penalty. The pilot fit standardised the 2015 shock away, kurtosis 111 down to 4.3.
-
A 95% range for a bend in clumpy counts, checked on seeds we never tuned on
On two seeds never used for tuning, the tempered 95% range for a growth-rate bend covered 101 of 107 bends (94.4%); steep bends run optimistic.
-
A second round of the dispersion penalty doesn't fix GARCH false alarms
Re-measuring the dispersion factor for a second round cuts GARCH false alarms from 30% to 27%: a negative result.
-
The Nile's 1898 dam break, found four ways
Four changepoint settings on the real Nile flow series all date the dam break to 1899; only the autocorrelation setting is honest that the date is less certain.