BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Thick-pen transformation for time series - Piotr Fryzlewicz (Londo
 n School of Economics)
DTSTART:20100129T153000Z
DTEND:20100129T163000Z
UID:TALK22578@talks.cam.ac.uk
CONTACT:8047
DESCRIPTION:Traditional visualisation of time series data often consists o
 f\nplotting the time series values against time and ?connecting the\ndots?
 . We propose an alternative\, multiscale visualisation technique\,\nmotiva
 ted by the scale-space approach in computer vision. In brief\,\nour method
  also ?connects the dots?\, but uses a range of pens of\nvarying thickness
 es for this purpose. The resulting multiscale map\,\ntermed the Thick-Pen 
 Transform (TPT) corresponds to viewing the time\nseries from a range of di
 stances. We formally prove that the TPT is a\ndiscriminatory statistic for
  two Gaussian time series with distinct\ncorrelation structures. Further\,
  we show interesting possible\napplications of the TPT to measuring cross-
 dependence in multivariate\ntime series\, and to testing for stationarity.
  In particular\, we derive\nthe asymptotic distribution of our test statis
 tic\, and argue that the\ntest is applicable to both linear and nonlinear 
 processes under low\nmoment assumptions. Various other aspects of the meth
 odology\,\nincluding other possible applications\, are also discussed.\n\n
 http://stats.lse.ac.uk/fryzlewicz/
LOCATION:MR12\, CMS\, Wilberforce Road\, Cambridge\, CB3 0WB
END:VEVENT
END:VCALENDAR