slides geotop
TRANSCRIPT
![Page 1: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/1.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
Bivariate Count Processesfor Earthquake Frequency
Mathieu Boudreault & Arthur Charpentier
Université du Québec à Montréal
http ://freakonometrics.blog.free.fr/
Séminaire GeoTop, January 2012
1
![Page 2: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/2.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
MotivationâLarge earthquakes are known to trigger earthquakes elsewhere. Damaging largeaftershocks occur close to the mainshock and microearthquakes are triggered bypassing seismic waves at significant distances from the mainshock. It is unclear,however, whether bigger, more damaging earthquakes are routinely triggered atdistances far from the mainshock, heightening the global seismic hazard afterevery large earthquake. Here we assemble a catalogue of all possible earthquakesgreater than M5 that might have been triggered by every M7 or larger mainshockduring the past 30 years. [...] We observe a significant increase in the rate ofseismic activity at distances confined to within two to three rupture lengths of themainshock. Thus, we conclude that the regional hazard of larger earthquakes isincreased after a mainshock, but the global hazard is not.â Parsons & Velasco(2011)
Figures : Number of earthquakes (magnitude exceeding 2.0, per 15 sec.)following a large earthquake (of magnitude 6.5), normalized so that the expectednumber of earthquakes before and after is 100.
2
![Page 3: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/3.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
â
ââ
â
â
ââ
â
â
â
â
â
â
â
ââ
â
â
â
â
â
â
ââ
â
ââ
ââ
â
â
â
â
â
â
ââ
â
â
ââ
â
â
â
â
â
â
â
â
ââ
ââ
âââ
â
â
â
â
â
âââ
â
â
ââââââ
âââ
âââ
â
ââ
â
ââââââ
â
â
â
â
â
â
â
â
ââ
â
â
â
â
ââ
â
â
â
ââââ
âââ
â
â
ââ
â
â
ââ
â
â
â
â
âââ
ââ
â
â
â
ââ
â
â
â
â
ââââ
âââ
â
â
ââ
â
â
ââ
â
â
â
ââ
â
â
ââ
â
â
â
â
â
ââ
â
â
â
â
â
â
â
â
â
â
â
â
â
âââââââ
ââ
â
â
â
â
â
â
â
â
â
â
â
â
â
ââ
â
ââ
â
â
â
â
â
â
â
â
â
â
â
ââ
â
â
â
â
ââ
âââ
â
âââ
ââ
â
â
â
â
â
â
ââ
ââ
â
â
â
â
âââ
â
â
â
â
â
â
âââââ
â
â
â
â
â
â
ââ
â
â
â
â
â
â
â
â
ââââ
â
â
â
âââ
â
â
â
â
â
â
â
ââ
â
ââ
â
â
â
âââ
ââ
â
â
â
â
â
â
â
ââ
â
â
ââ
â
ââ
âââ
ââââ
â
â
â
â
â
â
â
â
â
ââ
â
ââ
â
ââ
â
â
âââ
â
â
â
â
â
ââ
â
â
â
ââ
â
ââ
â
â
â
ââ
ââ
â
â
ââ
â
ââ
ââ
ââ
âââ
ââ
â
â
â
ââ
â
â
â
â
â
ââ
â
â
ââ
ââ
â
â
â
â
â
â
â
â
ââââ
â
ââ
â
â
â
ââ
â
â
â
â
â
â
â
â
â
â
â
â
ââ
â
â
â
â
âââââ
â
ââ
â
â
â
â
â
â
â
â
ââ
ââ
â
âââ
ââ
â
ââ
ââ
â
â
ââââ
â
ââââ
â
â
â
â
â
ââ
â
ââ
ââ
â
â
â
â
â
â
â
â
â
ââ
âââ
â
â
â
â
â
âââ
âââ
â
ââ
â
ââ
â
â
â
â
â
â
â
â
â
ââ
â
â
â
ââââ
â
â
â
â
â
â
â
ââ
âââ
ââ
â
ââ
âââ
âââ
â
â
âââ
âââ
âââ
â
â
â
â
ââ
ââ
â
â
â
â
â
â
ââ
â
â
ââ
â
â
âââ
ââââ
ââ
â
ââ
â
â
ââ
â
â
â
âââ
â
â
ââ
â
â
â
ââ
â
â
â
ââ
ââ
â
â
â
ââ
â
ââ
â
â
â
â
â
ââ
â
â
ââ
â
ââ
â
â
â
â
â
â
ââââ
ââ
â
â
â
â
ââ
â
âââ
â
ââ
ââââââ
â
â
â
ââ
â
â
ââ
â
â
â
â
â
â
â
â
â
â
ââââ
ââ
ââ
â
â
â
â
â
â
â
â
â
â
â
â
â
â
â
â
â
â
â
â
ââ
â
âââ
â
â
ââ
ââ
â
ââ
â
â
â
â
â
â
â
ââââ
â
â
â
â
â
â
â
â
â
â
ââ
â
â
â
â
â
ââ
ââââ
â
ââ
ââââ
â
ââ
ââ
â
â
â
â
ââââââ
â
â
â
ââ
ââ
âââ
âââ
â
â
â
â
âââ
â
ââââ
â
â
â
â
â
â
ââ
ââ
ââ
â
â
â
ââ
â
ââ
â
â
â
ââ
â
â
â
â
â
â
ââ
â
âââ
â
ââââ
ââ
âââ
â
â
âââ
â
ââ
â
â
âââââ
ââ
â
â
â
â
â
ââ
âââ
â
â
â
â
âââ
ââ
â
ââ
âââ
ââ
âââââ
â
â
ââ
â
â
âââ
â
â
â
â
â
ââ
â
â
â
â
â
ââ
ââ
â
ââââ
â
â
â
â
â
â
â
â
âââ
â
âââââ
âââ
ââ
â
â
ââ
â
ââ
â
â
â
ââ
ââ
âââ
â
ââ
â
ââ
â
â
â
â
â
â
âââ
âââ
ââ
â
â
â
ââ
â
ââ
â
â
â
ââ
â
â
â
â
â
â
ââ
â
âââ
â
â
â
â
ââ
ââ
â
ââ
â
ââ
â
â
â
âââ
â
â
â
ââ
â
â
â
â
ââ
â
âââ
âââ
â
ââ
â
ââ
ââ
â
â
ââ
â
â
â
â
â
â
â
ââ
â
â
â
â
ââââââââ
ââ
â
â
â
ââ
â
â
â
ââ
â
â
ââ
â
âââ
â
âââââ
â
â
â
ââ
â
â
â
â
â
â
â
â
ââ
ââââ
â
â
â
ââ
â
â
â
â
â
â
ââ
â
ââ
â
â
â
â
ââ
â
â
â
â
ââ
â
â
âââ
â
â
âââ
â
ââ
â
ââ
â
ââ
ââ
â
âââ
â
â
ââ
ââ
â
âââ
ââââ
âââ
â
â
âââ
â
âââ
â
â
â
â
âââ
âââ
ââ
ââ
ââ
â
â
âââ
â
â
â
â
â
âââââ
ââ
â
â
â
â
â
â
â
â
â
â
â
â
â
ââ
â
â
ââ
â
ââ
â
â
â
â
â
ââ
â
â
â
â
â
â
â
â
â
ââ
ââ
â
ââ
â
â
ââ
â
â
â
â
â
â
â
â
ââ
â
ââ
â
â
â
âââ
â
â
â
â
â
â
â
â
ââ
â
â
â
ââ
â
â
â
â
ââ
â
â
â
â
â
â
â
â
â
ââ
â
â
â
â
â
âââ
â
â
â
ââ
â
â
ââââ
ââ
ââ
ââ
ââââ
ââ
â
ââ
ââ
ââ
â
â
âââââ
ââ
â
â
ââ
â
â
ââ
â
â
â
ââ
ââ
â
â
â
ââ
ââ
â
â
ââ
â
â
â
â
ââ
â
â
â
ââ
ââ
ââ
â
âââ
â
â
â
âââ
â
â
â
â
â
ââ
â
â
â
â
ââ
â
ââ
âââ
â
â
â
â
â
â
â
â
âââ
ââ
â
â
â
âââ
ââââ
â
ââââ
â
â
â
ââ
â
â
ââââ
ââ
â
ââââ
â
â
â
â
â
â
â
ââ
ââ
â
â
â
â
ââ
â
â
â
â
â
â
ââ
â
â
â
â
â
â
â
â
ââ
â
â
â
ââ
â
ââ
ââââ
â
â
â
â
â
ââââ
ââ
â
â
â
â
ââ
â
ââ
â
âââ
â
â
â
â
â
â
âââââ
â
ââ
â
â
â
â
â
â
â
âââ
ââ
â
â
ââ
ââ
â
â
â
ââ
â
ââ
âââââ
â
â
âââ
âââ
â
âââ
ââ
â
â
â
â
â
â
ââ
â
â
ââ
ââ
â
â
ââ
â
â
ââ
âââ
ââ
â
â
â
â
â
â
â
â
â
â
ââ
â
â
ââ
â
â
â
ââ
â
ââââââ
â
â
â
â
ââ
â
ââ
ââ
âââ
â
â
â
ââ
â
â
ââ
â
ââââ
â
â
â
âââ
ââââ
â
â
â
ââââ
â
ââââ
â
ââ
â
â
âââ
â
â
â
â
â
â
ââ
âââ
ââ
â
ââ
ââ
â
ââââââ
â
â
â
ââ
â
â
â
â
â
â
ââ
â
â
â
ââââ
â
ââ
ââ
ââ
â
ââ
â
ââ
âââ
â
â
â
â
â
â
ââ
â
ââ
â
âââ
â
ââ
âââ
â
â
â
â
â
â
â
â
ââ
ââ
â
â
â
â
â
â
â
â
â
â
â
â
â
â
â
â
â
â
ââ
â
ââ
â
â
â
ââââ
ââ
â
â
âââââ
â
â
â
â
â
â
â
â
â
ââ
ââ
â
â
â
â
ââ
âââ
â
â
â
â
â
â
â
â
â
â
ââ
â
â
ââ
â
â
â
ââ
â
ââ
â
â
â
âââ
â
â
ââ
â
â
â
â
â
â
âââ
â
â
â
â
â
ââ
â
â
â
â
â
â
ââ
â
â
â
â
â
â
â
â
â
ââ
ââ
â
ââââ
â
â
â
ââ
ââ
â
â
â
â
ââ
â
â
â
ââââ
â
â
â
â
â
â
â
ââ
â
â
â
ââââ
âââââ
â
â
ââ
â
âââ
ââ
ââââ
ââ
â
â
â
ââ
â
â
ââ
â
â
ââ
â
â
â
â
â
â
âââ
â
ââ
ââ
â
ââ
â
ââ
â
â
â
â
â
â
â
â
â
â
â
â
â
â
ââ
â
â
â
â
ââ
ââ
â
â
â
â
ââ
ââ
â
ââ
â
âââ
â
âââ
â
â
â
ââ
ââ
âââ
ââ
â
â
ââ
â
â
â
â
â
ââ
â
â
ââ
ââ
ââ
â
â
ââ
â
â
â
â
â
ââ
â
â
â
â
â
â
â
â
â
ââ
â
â
ââ
â
ââ
â
â
â
â
ââ
â
âââ
â
â
â
â
â
ââ
â
â
â
âââââ
â
â
â
â
âââ
ââ
ââââ
ââââââ
â
â
ââââ
â
â
â
â
â
â
â
â
ââ
â
ââ
â
â
âââââ
â
ââ
â
â
â
ââ
â
â
â
â
â
â
â
â
ââ
â
â
â
â
â
â
â
â
â
â
ââ
â
â
â
ââââ
â
âââ
â
ââ
âââ
â
âââââ
â
â
â
â
â
â
â
â
ââ
â
â
â
â
âââ
â
â
â
ââ
â
â
ââ
â
â
â
â
â
â
â
â
ââ
â
â
â
âââ
â
â
â
â
âââ
ââ
â
â
â
â
â
â
ââ
â
â
â
â
âââ
â
â
ââ
â
ââ
ââ
â
ââ
ââââ
ââ
â
âââ
ââ
ââââ
â
âââ
ââ
ââ
â
â
ââ
â
â
â
â
â
â
â
ââ
â
ââ
âââ
âââ
â
ââ
ââ
â
â
â
â
ââ
â
â
â
â
âââ
â
â
â
ââ
â
â
â
â
â
â
â
â
â
âââââ
â
â
â
â
â
â
âââ
â
ââ
â
âââ
ââ
â
ââ
ââ
â
â
âââ
ââââ
â
â
ââ
â
â
â
â
â
â
ââ
ââ
â
â
âââ
â
ââ
â
â
â
â
ââ
â
â
ââ
ââ
ââââ
âââ
â
ââ
â
ââ
âââ
â
ââ
ââââ
â
ââââ
â
â
â
ââ
â
ââ
ââ
â
â
âââ
â
â
âââ
â
â
â
â
âââ
â
ââ
â
â
â
â
â
â
â
ââ
â
â
â
â
ââ
â
â
ââ
â
ââââ
ââ
â
â
ââ
â
âââ
ââ
ââ
âââ
â
â
â
âââ
â
ââ
ââ
â
ââââ
â
â
â
ââ
ââ
â
ââââ
â
âââ
ââ
â
â
ââ
â
â
â
ââ
â
â
â
â
â
âââ
ââ
ââ
â
â
â
ââ
â
â
â
â
â
ââ
â
ââ
â
â
â
â
â
âââ
ââââ
â
â
âââ
â
â
â
â
â
â
ââââââ
â
ââ
ââ
â
â
ââ
âââ
ââ
â
â
â
â
â
â
â
â
â
ââ
â
â
â
â
ââ
â
â
â
â
â
â
â
âââ
ââ
â
â
â
â
â
â
â
â
â
â
âââ
â
â
â
â
â
ââ
â
â
âââ
ââââ
ââââ
ââ
â
â
â
â
â
â
ââ
ââ
ââââ
ââ
ââ
â
â
â
â
ââ
â
ââ
âââââ
â
â
ââ
ââ
ââ
Time before and after a major eathquake (magnitude >6.5) in days
Num
ber
of e
arth
quak
es (
mag
nitu
de >
2) p
er 1
5 se
c., a
vera
ge b
efor
e=10
0
â15 â10 â5 0 5 10 15
8010
012
014
016
018
0
3
![Page 4: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/4.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
ââ
â
â
â
â
â
â
â
â
â
ââ
â
ââââ
âââ
â
â
â
â
ââââ
â
ââ
â
â
â
â
â
ââ
âââ
â
ââ
ââ
â
â
â
â
âââ
â
ââ
â
ââ
â
ââ
ââ
â
â
ââ
â
â
â
â
â
â
â
ââ
â
â
ââ
â
â
ââââ
â
â
â
â
â
â
ââ
â
â
ââ
â
â
â
ââ
â
ââ
â
â
â
â
ââ
ââ
â
â
ââ
â
â
â
â
â
â
â
â
â
â
â
ââ
â
âââ
â
â
â
â
â
â
â
â
ââââ
â
â
â
â
â
âââ
â
ââ
â
â
â
â
ââ
â
â
â
ââ
â
â
â
â
â
â
â
â
ââ
â
â
â
ââ
â
â
ââââ
âââ
â
â
ââ
â
â
â
â
â
â
â
â
â
â
â
â
â
â
âââ
â
â
âââ
â
â
â
â
â
â
â
â
ââââ
â
â
â
â
â
â
â
â
â
â
â
ââ
â
ââ
â
â
â
â
â
â
âââ
âââ
â
â
ââ
âââââ
ââ
â
â
ââ
â
â
â
â
â
â
â
â
â
â
â
â
ââââ
â
â
â
â
â
ââ
â
ââ
ââ
ââ
â
ââ
â
â
â
â
â
â
â
â
â
â
â
â
â
â
â
â
â
ââ
â
â
ââ
â
â
â
ââââ
â
â
â
â
ââ
â
â
â
â
ââââ
â
â
âââ
ââ
ââ
â
â
â
â
â
â
â
â
â
â
â
â
â
â
â
â
â
â
âââ
â
ââ
â
â
â
â
ââ
ââ
â
ââ
â
â
â
â
â
â
ââ
âââ
â
â
â
â
â
â
â
â
ââ
â
â
â
â
ââ
â
â
â
â
â
â
ââ
â
â
â
â
ââ
â
â
â
â
â
â
â
â
â
â
â
â
ââ
â
â
â
â
ââ
â
â
â
ââ
â
â
ââ
ââ
â
ââ
â
ââ
â
â
ââ
â
â
â
ââ
ââ
ââ
â
â
â
âââ
â
â
â
â
â
ââ
ââ
â
â
â
â
â
â
â
â
âââ
â
â
â
â
â
â
â
ââ
â
ââ
ââ
ââ
âââ
â
â
â
â
â
âââ
â
â
â
â
ââââ
â
â
â
â
â
â
â
â
â
â
â
â
â
â
â
â
â
âââ
â
â
âââ
â
â
ââââ
â
â
â
â
â
ââ
â
â
ââ
â
â
â
â
â
ââ
â
â
â
â
â
âââ
â
â
ââ
â
ââ
ââ
â
â
ââ
â
â
â
â
â
âââ
â
â
ââ
ââ
â
â
ââ
â
â
âââ
â
â
â
â
â
ââ
â
â
â
â
â
â
â
â
â
â
â
ââ
â
â
â
ââ
â
â
â
â
ââ
â
â
â
ââ
ââ
â
â
â
â
â
â
â
ââ
â
ââ
â
â
â
ââââ
â
â
â
âââ
ââ
â
â
â
â
â
â
â
â
ââ
â
â
â
ââ
â
â
â
ââ
ââ
â
â
â
â
â
â
ââââ
â
â
âââ
â
â
ââ
â
â
â
ââ
â
âââ
ââ
â
â
âââ
ââ
â
â
â
â
â
â
âââ
â
â
â
ââ
ââ
â
â
ââ
â
â
â
â
ââ
âââ
ââââ
â
ââ
â
â
â
â
â
â
â
â
â
â
ââ
ââ
â
â
ââ
â
â
â
âââ
â
â
ââ
â
ââ
â
â
â
â
â
â
ââ
â
â
ââ
â
â
â
ââ
â
â
â
âââ
â
â
â
ââ
â
â
â
â
â
â
ââ
â
ââ
â
â
â
â
â
â
â
ââ
ââ
â
â
ââ
â
â
â
â
â
â
âââ
â
â
ââââ
â
â
â
ââ
ââââ
â
â
â
âââ
â
ââ
â
â
â
â
â
â
â
â
â
â
âââ
â
ââ
â
â
â
ââ
â
â
â
â
â
â
ââ
â
â
â
â
â
â
ââ
â
â
â
â
ââ
â
â
âââ
â
â
â
ââ
âââ
â
â
â
ââ
â
â
ââ
â
ââ
â
â
â
â
â
â
ââ
â
â
â
â
â
â
ââ
â
â
â
â
â
â
â
â
ââ
ââ
â
â
â
â
â
â
âââ
â
â
ââ
â
â
â
ââ
âââ
â
â
â
â
â
â
â
â
â
â
â
â
ââ
â
â
âââ
â
â
ââ
â
â
â
â
â
ââ
â
â
ââ
â
â
â
â
â
â
â
â
â
â
â
â
ââ
â
â
â
ââ
â
â
â
â
â
âââ
â
â
â
â
ââ
â
â
ââ
ââ
â
â
ââ
â
â
â
â
â
â
â
â
â
â
ââ
ââ
â
â
â
â
ââ
â
â
â
â
ââ
â
â
â
ââ
â
â
â
â
ââ
â
â
â
â
â
ââ
â
â
âââ
â
â
â
â
â
â
ââ
â
ââ
â
ââ
ââ
âââ
â
â
âââ
â
ââ
â
â
ââ
ââ
â
â
â
âââ
âââ
â
â
â
â
ââ
â
ââ
â
â
â
ââ
â
âââ
â
â
â
â
â
â
â
â
ââ
âââ
ââ
â
ââ
â
â
â
â
âââ
âââ
â
â
â
â
â
â
ââ
â
â
âââââ
â
â
â
â
â
ââ
â
â
â
â
â
â
â
â
ââ
â
â
â
â
â
â
â
â
â
âââ
ââ
â
â
â
â
â
â
â
â
â
âââ
â
â
â
â
âââ
â
ââââââ
ââ
â
â
â
â
ââ
â
â
â
âââ
âââ
ââ
â
â
â
â
â
â
ââ
ââ
â
ââ
â
â
ââ
â
ââ
â
â
ââ
âââ
â
â
â
âââ
ââ
â
â
â
ââ
â
â
â
ââââââ
ââ
â
â
ââââ
â
ââââ
ââ
ââ
âââ
ââ
ââ
ââ
â
â
â
â
ââ
ââ
â
âââ
â
â
â
â
â
â
â
ââ
â
â
â
â
â
â
â
ââ
â
â
ââ
â
â
ââ
â
â
ââ
â
â
â
â
â
ââ
â
â
â
ââ
â
â
â
â
ââ
â
âââ
â
â
ââ
â
â
â
âââ
â
â
â
â
â
ââ
ââ
â
â
â
â
ââ
â
â
â
â
ââ
â
â
â
â
â
ââ
â
â
â
â
â
â
ââ
â
â
ââ
ââ
â
â
â
â
âââââ
â
â
â
â
â
ââ
â
â
â
â
ââ
â
â
â
â
â
â
ââ
â
â
â
ââ
ââ
â
âââ
â
ââ
â
â
â
â
â
ââ
â
ââ
â
âââ
âââ
â
ââââ
â
ââ
â
âââ
â
â
â
âââ
â
â
â
ââ
â
ââ
â
â
â
ââ
â
â
â
ââââ
â
ââ
â
â
â
â
â
ââââ
â
â
â
ââ
â
ââ
ââ
â
âââââ
â
ââ
â
â
â
ââ
â
â
âââ
â
ââ
ââ
ââ
â
âââ
ââ
â
â
â
â
ââ
ââââ
âââ
ââ
â
â
ââââ
â
â
â
â
â
ââ
ââââ
ââ
â
â
ââ
â
â
ââ
ââ
ââ
â
â
â
ââ
â
âââââ
â
ââ
ââ
â
â
â
â
â
ââââ
â
â
ââ
â
â
â
â
â
âââ
â
ââââ
â
â
â
âââ
â
ââ
â
â
ââ
â
â
â
ââ
â
â
â
â
â
â
â
â
â
ââ
â
âââ
â
â
ââ
â
â
â
â
â
â
â
â
â
â
â
â
ââ
â
ââ
â
ââââ
â
ââ
â
â
â
â
â
â
â
â
â
â
ââ
â
â
â
â
â
â
ââ
â
â
â
â
â
âââ
â
â
â
â
â
â
â
â
â
ââ
â
â
â
â
â
â
âââââ
ââ
â
â
â
â
â
â
â
â
â
ââ
â
ââ
ââ
ââ
â
ââ
ââ
â
âââ
âââ
â
â
â
â
â
â
â
â
ââ
â
â
â
â
â
ââ
â
â
â
â
â
â
â
â
â
â
â
â
â
ââ
ââ
â
â
â
â
â
â
â
â
â
â
ââ
ââ
â
ââ
ââ
ââ
â
â
â
â
â
â
â
ââââ
âââ
â
â
â
â
âââ
ââ
â
â
â
â
â
â
â
â
ââ
â
â
â
â
âââ
â
â
â
â
â
â
â
â
ââ
ââ
ââ
âââ
ââ
â
â
â
â
â
â
â
â
â
ââ
â
â
ââ
â
â
â
â
â
âââ
â
â
ââ
â
â
ââââ
â
ââ
â
â
â
â
â
â
â
â
â
ââ
â
ââ
ââââ
ââ
â
ââ
ââ
ââ
â
ââ
â
â
ââ
â
â
â
ââ
â
â
â
â
â
â
â
â
âââ
â
â
â
â
â
â
â
â
â
â
â
â
âââ
â
â
â
â
ââ
â
â
â
â
â
â
â
â
â
â
ââ
â
â
â
ââ
â
â
â
â
â
â
â
â
â
â
â
â
â
â
ââ
â
ââ
â
â
ââ
â
â
â
âââ
â
ââ
â
â
â
â
ââ
ââ
â
â
â
â
âââ
â
â
â
â
â
â
â
â
â
âââ
ââââ
â
ââ
â
â
â
â
â
ââ
â
â
â
â
â
â
ââ
ââ
â
â
â
âââ
â
â
â
ââ
â
â
â
ââ
â
â
ââ
â
â
ââ
â
â
â
ââ
â
â
â
â
â
â
â
â
â
ââââ
â
â
ââ
â
â
â
â
â
ââ
â
â
â
â
ââââ
â
â
â
â
ââ
â
ââ
â
â
â
â
â
â
â
â
â
â
â
â
â
â
â
âââ
â
ââ
ââ
â
â
â
âââ
â
â
â
ââ
â
ââ
ââ
ââ
â
â
â
â
â
ââ
â
â
â
â
â
â
â
â
â
âââ
â
â
â
â
â
â
â
â
â
ââ
â
â
â
â
â
â
â
â
â
â
â
â
â
â
ââ
â
â
âââ
â
â
â
â
â
â
â
â
â
â
âââ
ââ
âââ
â
âââ
ââ
â
ââ
ââ
â
ââ
â
â
â
â
â
â
â
ââ
â
â
â
â
ââ
â
â
â
â
â
â
â
â
âââ
â
ââ
â
ââ
â
â
â
ââ
â
â
â
â
â
â
â
ââ
â
ââ
â
âââ
â
â
â
â
ââ
ââ
ââ
â
âââ
â
â
â
âââ
â
â
â
â
â
â
â
â
â
â
â
â
â
â
â
ââ
â
âââ
â
ââââ
ââ
ââ
â
â
â
ââ
â
â
â
â
â
â
â
ââ
ââ
â
â
â
â
â
ââ
â
ââââ
â
â
â
â
â
â
â
â
â
â
â
âââ
â
â
â
â
â
ââ
â
â
ââ
â
â
â
â
â
ââ
â
ââ
ââ
ââ
â
â
â
â
â
â
â
â
â
â
ââ
â
â
â
â
â
â
â
â
â
â
â
â
â
âââââ
â
ââ
ââ
â
â
â
â
â
â
ââ
âââ
â
â
â
â
â
ââ
â
â
â
â
â
â
â
ââ
â
â
ââ
â
â
ââ
â
â
â
ââ
â
â
â
â
â
â
ââ
âââ
â
â
ââ
â
â
â
â
â
âââ
â
â
â
ââ
âââ
â
â
ââ
â
â
â
ââ
â
â
ââââ
â
â
â
â
ââ
ââ
â
â
â
ââ
â
â
ââ
â
â
â
ââââ
â
â
ââ
â
â
â
ââ
â
â
â
â
â
ââ
ââ
â
â
ââ
â
ââ
â
â
â
âââ
â
â
â
â
â
ââ
ââ
â
â
âââ
â
â
ââ
â
â
âââ
â
â
â
ââ
â
â
â
â
ââââ
ââ
â
ââ
â
â
â
â
ââ
â
ââ
â
â
â
â
âââ
â
â
â
â
â
â
â
ââ
â
â
â
â
â
â
â
â
âââââ
â
â
â
â
â
â
â
â
â
â
ââ
â
â
â
â
â
â
â
ââ
Time before and after a major eathquake (magnitude >6.5) in days
Num
ber
of e
arth
quak
es (
mag
nitu
de >
4) p
er 1
5 se
c., a
vera
ge b
efor
e=10
0
â15 â10 â5 0 5 10 15
8010
012
014
016
018
020
022
0Main event, magnitude > 6.5Main event, magnitude > 6.8Main event, magnitude > 7.1Main event, magnitude > 7.4
Number of earthquakes before and after a major one, magnitude of the main event, small events more than 4
4
![Page 5: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/5.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
â
âââ
â
â
â
â
â
â
ââââââââââ
ââââââââââ
â
âââââââââââ
âââ
ââââââââââ
âââââââââââ
â
âââââââââââââââââââ
ââ
â
ââââââââââââ
âââââ
âââ
â
ââââ
âââ
â
ââââââââââââ
ââ
â
â
ââââ
ââââ
â
âââ
ââââ
â
ââââââ
âââââââ
ââ
âââ
ââââââ
ââââââââââ
ââ
â
â
â
âââ
â
ââââ
ââââââââ
âââââââââ
ââââ
âââââââââ
âââââ
ââ
â
âââââ
â
âââââ
âââââââââââââ
â
âââââââââââ
â
ââ
ââ
ââââââ
â
âââ
ââââââââââââââ
â
âââââââ
â
â
â
ââ
âââ
ââ
ââââââââ
âââââââââ
â
â
âââ
ââ
â
ââââââââââ
â
ââ
â
â
âââââââââââ
â
âââââââ
âââââ
â
â
â
âââ
ââââ
ââââââââ
âââââ
ââââââ
ââ
â
â
âââ
âââââââââââ
ââââ
âââââââ
ââââ
â
ââ
â
âââ
âââ
â
âââââââ
â
ââ
ââââââ
ââââ
ââââââââââââââââ
âââââââ
â
âââââââââââ
â
ââââ
âââ
â
ââââââââ
â
ââ
âââââââââââââ
â
â
ââ
ââââââ
ââ
âââ
ââââââ
â
ââââââââââââââââ
âââââââââ
âââââââââ
â
â
ââââââââ
âââ
â
â
âââââââ
ââââââââ
â
ââ
â
âââ
â
ââ
âââââââââââ
âââââ
âââââ
â
ââââââââââ
â
â
âââââââ
âââââ
â
â
âââââââââââââ
â
ââââ
ââââ
âââ
ââââââ
â
ââ
ââââ
âââââââ
â
âââââââââââ
â
ââââ
â
ââââ
âââââââ
ââââââ
â
ââ
â
âââ
ââââââ
âââââââââ
ââââ
â
âââââââââââââ
ââââââââââââââ
âââââââââââ
ââââââââââ
â
âââââââââ
ââ
â
â
âââââââââ
âââ
âââ
â
ââââââââââââââââââ
âââââââ
âââââââ
â
âââââ
â
â
ââââââââââ
âââââââ
â
ââââââââââ
â
âââ
â
âââââââ
â
âââââââââ
ââââ
â
âââ
â
âââââ
ââ
â
âââââââ
ââââââââââââââââââââââ
ââ
âââââââ
ââ
âââ
â
â
âââââ
â
âââââ
â
ââââââââ
ââââ
â
â
ââ
â
â
â
â
ââââ
â
âââ
â
âââââ
âââââ
â
âââ
ââ
ââââââ
â
ââââ
ââââââââââââ
âââââ
â
âââ
âââââââââ
ââ
â
ââââââ
â
âââââ
â
âââââââ
â
â
ââ
ââââ
ââââââ
â
ââââ
ââ
âââââââââââ
âââ
âââââââ
ââââââ
âââââââ
âââââââââââââ
âââââââââââââââââââ
ââââââââââââ
âââââ
âââââ
ââââ
âââ
âââââââ
ââââ
âââââââ
âââââââââââââ
â
âââââââââ
ââ
ââ
â
âââââââââââ
ââââââ
âââ
â
ââââ
âââââââââ
ââââââââââââââ
ââââââââ
ââ
â
âââââ
ââââââ
ââ
âââââââââââââ
ââââ
âââ
â
â
â
â
ââââ
â
â
ââ
â
ââ
â
âââ
ââââââ
âââ
âââ
â
ââââââ
â
â
â
ââ
ââ
â
â
â
ââ
â
â
â
â
ââ
â
â
â
ââ
ââ
â
ââ
âââ
â
â
â
âââ
â
â
â
â
âââââ
â
â
â
âââ
â
â
âââ
â
â
â
â
ââ
â
â
â
â
ââ
â
âââââââ
âââ
ââ
â
ââ
â
ââââ
ââ
â
â
ââ
â
âââââââââ
âââ
ââ
â
ââââââââ
ââ
âââ
â
âââ
ââââââ
â
ââ
âââ
âââ
âââââââââââ
âââââ
â
âââ
âââ
â
ââââ
â
â
âââââ
â
â
â
â
â
â
âââââââ
ââââââ
âââââ
ââ
ââââ
âââââ
â
âââââââ
â
â
âââââââ
â
â
âââââââ
ââ
âââââ
ââââââ
â
â
ââ
ââ
ââ
â
ââ
âââ
âââ
ââ
âââââ
âââ
â
â
âââââââââââ
â
âââââââ
â
ââ
â
ââ
â
â
â
ââââ
âââ
ââââââ
â
â
â
ââââââ
ââ
âââââââ
â
ââââââ
â
âââââ
â
â
â
âââââ
âââ
â
ââ
âââââââââ
â
â
â
â
â
ââââââ
ââ
ââââ
â
ââââ
ââ
ââ
ââââ
â
âââ
ââââ
âââââââââââ
â
âââââââ
â
â
ââââââââââ
â
â
ââââ
â
â
âââ
â
ââ
â
âââââ
ââââ
â
âââ
â
ââââââ
ââââââ
â
â
â
â
â
ââââââââââââââââ
â
ââ
âââââ
ââ
âââ
â
â
ââ
âââââ
âââââ
â
ââ
ââ
ââ
ââââââ
ââââââ
ââ
â
ââ
âââââââââââââ
â
â
â
ââ
â
ââ
â
â
ââââââââââ
ââ
ââââ
â
ââ
â
â
ââââ
âââââââââââââââââââ
ââââ
â
âââââââââ
ââââââ
â
â
ââââ
ââââ
â
ââ
ââ
âââââââ
â
â
â
âââ
ââââââ
â
ââââ
â
â
âââââ
â
ââââââ
ââââ
â
âââââââââââââââ
ââââââ
â
â
â
âââââââââ
â
ââââ
â
âââââââ
ââ
â
âââââââââââ
â
â
âââ
âââ
â
âââ
âââ
â
âââââââââ
â
â
ââ
ââââââ
âââ
â
â
â
â
ââ
â
âââ
âââââââââââââ
ââââ
â
â
ââââââ
ââ
âââââââââââ
â
ââ
â
ââââ
â
âââââââââââ
â
ââ
âââ
â
âââââ
ââââ
âââ
ââââââââââââââââââââââââââ
â
âââââââââ
ââ
â
âââââââ
ââââââ
ââ
âââââââââââââââââ
â
ââââââ
â
â
ââââââââââââââ
â
âââ
â
âââââ
â
ââââ
â
ââââââââ
â
âââ
â
âââââââ
â
ââââââ
â
â
âââ
â
ââââ
â
â
âââââ
âââ
ââââââââââââââ
â
ââââââ
â
ââ
â
ââââ
ââââ
âââ
ââââââ
â
ââââââââââââ
â
â
âââ
âââââââ
âââââââ
âââââââââ
âââââââ
â
â
âââââââââââââââââââ
âââââââ
â
ââââââ
â
â
âââââââ
âââââ
ââ
ââââââââââ
ââââ
â
ââ
ââââââ
â
ââââ
âââ
â
ââ
ââ
â
â
ââââââââ
â
ââ
ââ
â
âââââââ
ââââââââ
âââââââââ
âââââ
ââââ
ââ
â
ââ
ââââââââââââââ
â
â
âââ
â
â
â
âââââââ
âââââ
â
ââââââââââ
â
â
ââââ
âââââââââââââ
â
â
â
â
ââââ
â
ââ
â
ââââââââââ
âââââ
â
âââââââ
âââ
ââ
â
âââââââ
â
ââââ
â
ââ
âââ
â
âââââââââââââââââââ
â
â
â
âââââââââââââââ
â
ââ
Time before and after a major eathquake (magnitude >6.5) in days
Num
ber
of e
arth
quak
es (
mag
nitu
de >
2) p
er 1
5 se
c., a
vera
ge b
efor
e=10
0
âââââââââââââââââââââââââââââââââââââââââââââ
âââââââââââââââââââââââââââââââââââââââââââââââ
ââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââ
ââââââââââââââââââââââââââââââââââââââââââââ
ââââââââââââââââââââââââââââââââââââ
âââââââââââââââââââââââââââââââââââââââââââ
ââââââââââââââââââââââââââââââââââââââââââââââââ
âââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââ
âââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââ
ââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââ
ââââââââââââââââââââââââââââ
ââââââââââââââââââââââââââââââââââââââââââââââââââ
âââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââ
âââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââ
ââââââââââââââââââââââââ
â
â
ââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââ
âââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââ
ââââââââââââââââââââââââââââââââââââ
âââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââ
ââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââ
âââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââ
âââââââââââââââââââââââââââââââââââââââââââââââââââââââââââ
âââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââ
âââââââââââââââââââââââââââââââââââââââ
ââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââ
âââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââââ
â15 â10 â5 0 5 10 15
020
040
060
080
010
00
Same techtonic plate as major oneDifferent techtonic plate as major one
5
![Page 6: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/6.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
6
![Page 7: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/7.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
We look at seismic risks with the eyes of actuaries and statisticians...
7
![Page 8: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/8.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
8
![Page 9: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/9.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
Agenda⢠Motivation (Parsons & Velasco (2011))⢠Modeling dynamics⊠AR(1) : Gaussian autoregressive processes (as a starting point)⊠VAR(1) : multiple AR(1) processes, possible correlated⊠INAR(1) : autoregressive processes for counting variates⊠MINAR(1) : multiple counting processes
⢠Application to earthquakes frequency⊠counting earthquakes on tectonic plates⊠causality between different tectonic plates⊠counting earthquakes with different magnitudes
9
![Page 10: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/10.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
Part 1
Modeling dynamics of counts
10
![Page 11: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/11.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
(Gaussian) Auto regressive processes AR(1)Definition A time series (Xt)tâN with values in R is called an AR(1) process if
Xt = Ï0+Ï1Xtâ1 + εt (1)
for all t, for real-valued parameters Ï0 and Ï1, and some i.i.d. random variablesεt with values in R.
It is common to assume that εt are independent variables, with a Gaussiandistribution N (0, Ï2), with density
Ï(ε) = 1â2ÏÏ
exp(â ε2
2Ï2
), ε â R.
Note that we assume also that εt is independent of Xtâ1, i.e. past observationsX0, X1, · · · , Xtâ1. Thus, (εt)tâN is called the innovation process.
Definition A time series (Xt)tâN is said to be (weakly) stationary if⢠E(Xt) is independent of t ( =: µ)⢠cov(Xt, Xtâh) is independent of t (=: γ(h)), called autocovariance function
11
![Page 12: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/12.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
Remark As a consequence, var(Xt) = E([Xt â E(Xt)]2) is independent of t(=: γ(0)). Define the autocorrelation function Ï(·) as
Ï(h) := corr(Xt, Xtâh) = cov(Xt, Xtâh)âvar(Xt)var(Xtâh)
= γ(h)γ(0) , âh â N.
Proposition (Xt)tâN is a stationary AR(1) time series if and only if Ï1 â (â1, 1).
Remark If Ï1 = 1, (Xt)tâN is called a random walk.
Proposition If (Xt)tâN is a stationary AR(1) time series,
Ï(h) = Ïh1 , âh â N.
12
![Page 13: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/13.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
Proposition If (Xt)tâN is a stationary AR(1) time series, (Xt)tâN satisfies
Xt = Ï1[Xtâ1] + εt
= Ï1[Ï1[Xtâ2] + εtâ1] + εt
= Ï1[Ï1[Ï1[Xtâ2] + εtâ2] + εtâ1 + εt
· · ·
= Ït1X0 +tâ
h=0Ïh1εtâh,
which is a MA(â) representation (moving average of the innovation process).
13
![Page 14: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/14.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
Vector (Gaussian) Autoregressive processes V AR(1)Definition A time series (Xt = (X1,t, · · · , Xd,t))tâN with values in Rd is called aVAR(1) process if
X1,t = Ï1,1X1,tâ1 + Ï1,2X2,tâ1 + · · ·+ Ï1,dXd,tâ1 + ε1,t
X2,t = Ï2,1X1,tâ1 + Ï2,2X2,tâ1 + · · ·+ Ï2,dXd,tâ1 + ε2,t
· · ·Xd,t = Ïd,1X1,tâ1 + Ïd,2X2,tâ1 + · · ·+ Ïd,dXd,tâ1 + εd,t
(2)
or equivalentlyX1,t
X2,t...
Xd,t
ïžž ïž·ïž· ïžž
Xt
=
Ï1,1 Ï1,2 · · · Ï1,d
Ï2,1 Ï2,2 · · · Ï2,d...
......
Ïd,1 Ïd,2 · · · Ïd,d
ïžž ïž·ïž· ïžž
Ί
X1,tâ1
X2,tâ1...
Xd,tâ1
ïžž ïž·ïž· ïžž
Xtâ1
+
ε1,t
ε2,t...εd,t
ïžž ïž·ïž· ïžž
εt
14
![Page 15: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/15.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
for all t, for some real-valued dà d matrix Ί, and some i.i.d. random vectors εtwith values in Rd.
It is common to assume that εt are independent variables, with a Gaussiandistribution N (0,Σ), with density
Ï(ε) = 1â(2Ï)d|det Σ|
exp(âεâ²Î£â1ε
2
), âε â Rd.
Thus, independent means time independent, but can be dependentcomponentwise.
Note that we assume also that εt is independent of Xtâ1, i.e. past observationsX0,X1, · · · ,Xtâ1. Thus, (εt)tâN is called the innovation process.
Definition A time series (Xt)tâN is said to be (weakly) stationary if⢠E(Xt) is independent of t (=: µ)⢠cov(Xt,Xtâh) is independent of t (=: γ(h)), called autocovariance matrixRemark As a consequence, var(Xt) = E([Xt â E(Xt)]â²[Xt â E(Xt)]) is
15
![Page 16: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/16.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
independent of t (=: γ(0)). Define finally the autocorrelation matrix,
Ï(h) = ââ1γ(h)ââ1, where â = diag(â
γi,i(0)).
Proposition (Xt)tâN is a stationary AR(1) time series if and only if the deignvalues of Ί should have a norm lower than 1.
Proposition If (Xt)tâN is a stationary AR(1) time series,
Ï(h) = Ίh, h â N.
16
![Page 17: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/17.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
Proposition If (Xt)tâN is a stationary AR(1) time series, (Xt)tâN satisfies
Xt = Ί[Xtâ1] + εt= Ί[Ί[Xtâ2] + εtâ1] + εt= Ί[Ί[Ί[Xtâ2] + εtâ2] + εtâ1 + εt· · ·
= ΊtX0 +tâ
h=0Ίhεtâh,
which is a VMA(â) representation.
17
![Page 18: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/18.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
Statistical inference for AR(1) time seriesConsider a series of observations X1, · · · , Xn. The likelihood is the jointdistribution of the vectors X = (X1, · · · , Xn), which is not the product ofmarginal distribution, since consecutive observations are not independent(cov(Xt, Xtâh) = Ïh). Nevertheless
L(Ï, Ï; (X0,X )) =nât=1
ÏÏ,Ï(Xt|Xtâ1)
where ÏÏ,Ï(·|Xtâ1) is a Gaussian density.
Maximum likelihood estimators are
(ÏÌ, ÏÌ) â argmax logL(Ï, Ï; (X0,X ))
18
![Page 19: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/19.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
Poisson processes and counting modelsEarthquake count models are mostly based upon the Poisson process (see Utsu(1969), Gardner & Knopoff (1974), Lomnitz (1974), Kagan & Jackson(1991)), Cox process (self-exciting, cluster or branching processes, stress-releasemodels (see Rathbun (2004) for a review), or Hidden Markov Models (HMM)(see Zucchini & MacDonald (2009) and Orfanogiannaki et al. (2010)).
See also Vere-Jones (2010) for a summary of statistical and stochastic modelsin seismology.
19
![Page 20: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/20.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
Thinning operator âŠSteutel & van Harn (1979) defined a thinning operator as follows
Definition Define operator ⊠as
p âŠN = Y1 + · · ·+ YN if N 6= 0, and 0 otherwise,
where N is a random variable with values in N, p â [0, 1], and Y1, Y2, · · · are i.i.d.Bernoulli variables, independent of N , with P(Yi = 1) = p = 1â P(Yi = 0). Thusp âŠN is a compound sum of i.i.d. Bernoulli variables.
Hence, given N , p âŠN has a binomial distribution B(N, p).
Note that p ⊠(q âŠN) L= [pq] âŠN for all p, q â [0, 1].
Further
E (p âŠN) = pE(N) and var (p âŠN) = p2var(N) + p(1â p)E(N).
20
![Page 21: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/21.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
(Poisson) Integer Autoregressive processes INAR(1)Based on that thinning operator, Al-Osh & Alzaid (1987) and McKenzie(1985) defined the integer autoregressive process of order 1 :
Definition A time series (Xt)tâN with values in R is called an INAR(1) process if
Xt = p âŠXtâ1 + εt, (3)
where (εt) is a sequence of i.i.d. integer valued random variables, i.e.
Xt =Xtâ1âi=1
Yi + εt, where Y â²i s are i.i.d. B(p).
Such a process can be related to Galton-Watson processes with immigration, orphysical branching model.
21
![Page 22: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/22.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
Xt+1 =Xtâi=1
Yi + εt+1, where Y â²i s are i.i.d. B(p)
22
![Page 23: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/23.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
Proposition E (Xt) = E(εt)1â p , var (Xt) = γ(0) = pE(εt) + var(εt)
1â p2 and
γ(h) = cov(Xt, Xtâh) = ph.
It is common to assume that εt are independent variables, with a Poissondistribution P(λ), with probability function
P(εt = z) = eâλλk
k! , z â N.
Proposition If (εt) are Poisson random variables, then (Nt) will also be asequence of Poisson random variables.
Note that we assume also that εt is independent of Xtâ1, i.e. past observationsX0, X1, · · · , Xtâ1. Thus, (εt)tâN is called the innovation process.
Proposition (Xt)tâN is a stationary INAR(1) time series if and only if p â [0, 1).
23
![Page 24: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/24.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
Proposition If (Xt)tâN is a stationary INAR(1) time series, (Xt)tâN satisfies
Xt = p ⊠[Xtâ1] + εt
= p ⊠[p ⊠[Xtâ2] + εtâ1] + εt
= p ⊠[p ⊠[p ⊠[Xtâ2] + εtâ2] + εtâ1 + εt
· · ·
= pt âŠX0 +tâ
h=0ph ⊠εtâh,
which is an INMA(â) representation.
Proposition If (Xt)tâN is a stationary INAR(1) time series, (Xt)tâN is anhomogeneous Markov chain.
24
![Page 25: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/25.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
Ï(xt, xtâ1) = P(Xt = xt|Xtâ1 = xtâ1) =xtâk=0
P
(xtâ1âi=1
Yi = xt â k
)ïžž ïž·ïž· ïžž
Binomial
·P(ε = k)  Poisson
.
25
![Page 26: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/26.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
Inference of Integer Autoregressive processes INAR(1)Consider a Poisson INAR(1) process, then the likelihood is
L(p, λ;X0,X ) =[nât=1
ft(Xt)]· λX0
(1â p)X0X0! exp(â λ
1â p
)where
ft(y) = exp(âλ)min{Xt,Xtâ1}â
i=0
λyâi
(y â i)!
(Ytâ1
i
)pi(1â p)Ytâ1ây, for t = 1, · · · , n.
Maximum likelihood estimators are
(pÌ, λÌ) â argmax logL(p, λ; (X0,X ))
26
![Page 27: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/27.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
Multivariate Integer Autoregressive processes MINAR(1)Let Nt := (N1,t, · · · , Nd,t), denote a multivariate vector of counts.
Definition Let P := [pi,j ] be a dà d matrix with entries in [0, 1]. IfN = (N1, · · · , Nd) is a random vector with values in Nd, then P âŠN is ad-dimensional random vector, with i-th component
[P âŠN ]i =dâj=1
pi,j âŠNj ,
for all i = 1, · · · , d, where all counting variates Y in pi,j âŠNj âs are assumed to beindependent.
Note that P ⊠(Q âŠN) L= [PQ] âŠN .
Further, E (P âŠN) = PE(N), and
E ((P âŠN)(P âŠN)â²) = PE(NN â²)P â² + â,
with â := diag(V E(N)) where V is the dà d matrix with entries pi,j(1â pi,j).
27
![Page 28: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/28.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
Definition A time series (Xt) with values in Nd is called a d-variate MINAR(1)process if
Xt = P âŠXtâ1 + εt (4)
for all t, for some dà d matrix P with entries in [0, 1], and some i.i.d. randomvectors εt with values in Nd.
(Xt) is a Markov chain with states in Nd with transition probabilities
Ï(xt,xtâ1) = P(Xt = xt|Xtâ1 = xtâ1) (5)
satisfying
Ï(xt,xtâ1) =xtâ
k=0P(P ⊠xtâ1 = xt â k) · P(ε = k).
Proposition Let (Xt) denote a d-variate MINAR(1) process, withautoregressive matrix P with entries in [0, 1], then (Xt) admits a d-variate
28
![Page 29: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/29.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
INMA(â) representation
Xt =ââh=0
P h ⊠εtâh.
Proposition Let P with entries in (0, 1), such that its largest eigenvalue is lessthan 1, and assume that P(εt = 0) â (0, 1) with E(âεtââ) <â, then there existsa stationary d-variate MINAR(1) process satisfying Equation (4).
29
![Page 30: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/30.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
Parameter inference for MINAR(1)Proposition Let (Xt) be a d-variate MINAR(1) process satisfying stationaryconditions, as well as technical assumptions (called C1-C6 in Franke & SubbaRao (1993)), then the conditional maximum likelihood estimate ÎžÌ of Ξ = (P ,Î)is asymptotically normal,
ân(ÎžÌ â Ξ) Lâ N (0,Σâ1(Ξ)), as nââ.
Further,
2[logL(N , ΞÌ|N0)â logL(N ,Ξ|N0)] Lâ Ï2(d2 + dim(λ)), as nââ.
30
![Page 31: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/31.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
Granger causality with BINAR(1)(X1,t) and (X2,t) are instantaneously related if ε is a noncorrelated noise,g g gg g g g g g g g g g g
X1,t
X2,t
ïžž ïž·ïž· ïžž
Xt
=
p1,1 p1,2
p2,1 p2,2
ïžž ïž·ïž· ïžž
P
âŠ
X1,tâ1
X2,tâ1
ïžž ïž·ïž· ïžž
Xtâ1
+
ε1,t
ε2,t
ïžž ïž·ïž· ïžž
εt
, with var
ε1,t
ε2,t
=
λ1 Ï
Ï Î»2
31
![Page 32: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/32.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
Granger causality with BINAR(1)1. (X1) and (X2) are instantaneously related if ε is a noncorrelated noise, g g g g
g g g g g g g g g g
X1,t
X2,t
ïžž ïž·ïž· ïžž
Xt
=
p1,1 p1,2
p2,1 p2,2
ïžž ïž·ïž· ïžž
P
âŠ
X1,tâ1
X2,tâ1
ïžž ïž·ïž· ïžž
Xtâ1
+
ε1,t
ε2,t
ïžž ïž·ïž· ïžž
εt
, with var
ε1,t
ε2,t
=
λ1 ?
? λ2
32
![Page 33: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/33.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
Granger causality with BINAR(1)2. (X1) and (X2) are independent, (X1)â¥(X2) if P is diagonal, i.e.p1,2 = p2,1 = 0, and ε1 and ε2 are independent,
X1,t
X2,t
ïžž ïž·ïž· ïžž
Xt
=
p1,1 00 p2,2
ïžž ïž·ïž· ïžž
P
âŠ
X1,tâ1
X2,tâ1
ïžž ïž·ïž· ïžž
Xtâ1
+
ε1,t
ε2,t
ïžž ïž·ïž· ïžž
εt
, with var
ε1,t
ε2,t
=
λ1 00 λ2
33
![Page 34: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/34.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
Granger causality with BINAR(1)3. (N1) causes (N2) but (N2) does not cause (X1), (X1)â(X2), if P is a lower
triangle matrix, i.e. p2,1 6= 0 while p1,2 = 0,
X1,t
X2,t
ïžž ïž·ïž· ïžž
Xt
=
p1,1 0? p2,2
ïžž ïž·ïž· ïžž
P
âŠ
X1,tâ1
X2,tâ1
ïžž ïž·ïž· ïžž
Xtâ1
+
ε1,t
ε2,t
ïžž ïž·ïž· ïžž
εt
, with var
ε1,t
ε2,t
=
λ1 Ï
Ï Î»2
34
![Page 35: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/35.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
Granger causality with BINAR(1)4. (N2) causes (N1) but (N1,t) does not cause (N2), (N1)â(N2,t), if P is a upper
triangle matrix, i.e. p1,2 6= 0 while p2,1 = 0,
X1,t
X2,t
ïžž ïž·ïž· ïžž
Xt
=
p1,1 ?
0 p2,2
ïžž ïž·ïž· ïžž
P
âŠ
X1,tâ1
X2,tâ1
ïžž ïž·ïž· ïžž
Xtâ1
+
ε1,t
ε2,t
ïžž ïž·ïž· ïžž
εt
, with var
ε1,t
ε2,t
=
λ1 Ï
Ï Î»2
35
![Page 36: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/36.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
Granger causality with BINAR(1)5. (N1) causes (N2) and conversely, i.e. a feedback effect (N1)â(N2), if P is a
full matrix, i.e. p1,2, p2,1 6= 0
X1,t
X2,t
ïžž ïž·ïž· ïžž
Xt
=
p1,1 ?
? p2,2
ïžž ïž·ïž· ïžž
P
âŠ
X1,tâ1
X2,tâ1
ïžž ïž·ïž· ïžž
Xtâ1
+
ε1,t
ε2,t
ïžž ïž·ïž· ïžž
εt
, with var
ε1,t
ε2,t
=
λ1 Ï
Ï Î»2
36
![Page 37: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/37.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
Bivariate Poisson BINAR(1)A classical distribution for εt is the bivariate Poisson distribution, with onecommon shock, i.e. ε1,t = M1,t +M0,t
ε2,t = M2,t +M0,t
where M1,t, M2,t and M0,t are independent Poisson variates, with parametersλ1 â Ï, λ2 â Ï and Ï, respectively. In that case, εt = (ε1,t, ε2,t) has jointprobability function
eâ[λ1+λ2âÏ] (λ1 â Ï)k1
k1!(λ2 â Ï)k2
k2!
min{k1,k2}âi=0
(k1
i
)(k2
i
)i!(
Ï
[λ1 â Ï][λ2 â Ï]
)with λ1, λ2 > 0, Ï â [0,min{λ1, λ2}].
λ =
λ1
λ2
and Î =
λ1 Ï
Ï Î»2
37
![Page 38: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/38.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
Bivariate Poisson BINAR(1) and Granger causalityFor instantaneous causality, we test
H0 : Ï = 0 against H1 : Ï 6= 0
Proposition Let Î»Ì denote the conditional maximum likelihood estimate ofλ = (λ1, λ2, Ï) in the non-constrained MINAR(1) model, and λ⥠denote theconditional maximum likelihood estimate of λ⥠= (λ1, λ2, 0) in the constrainedmodel (when innovation has independent margins), then under suitableconditions,
2[logL(N , λÌ|N0)â logL(N , λÌâ¥|N0)] Lâ Ï2(1), as nââ, under H0.
38
![Page 39: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/39.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
Bivariate Poisson BINAR(1) and Granger causalityFor lagged causality, we test
H0 : P â P against H1 : P /â P,
where P is a set of constrained shaped matrix, e.g. P is the set of dà d diagonalmatrices for lagged independence, or a set of block triangular matrices for laggedcausality.
Proposition Let PÌ denote the conditional maximum likelihood estimate of P inthe non-constrained MINAR(1) model, and PÌ
cdenote the conditional maximum
likelihood estimate of P in the constrained model, then under suitable conditions,
2[logL(N , PÌ |N0)â logL(N , PÌc|N0)] Lâ Ï2(d2âdim(P)), as nââ, under H0.
Example Testing (N1,t)â(N2,t) is testing whether p1,2 = 0, or not.
39
![Page 40: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/40.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
Autocorrelation of MINAR(1) processesProposition Consider a MINAR(1) process with representationXt = P âŠXtâ1 + εt, where (εt) is the innovation process, with λ := E(εt) andÎ := var(εt). Let µ := E(Xt) and γ(h) := cov(Xt,Xtâh). Then µ = [Iâ P ]â1λ
and for all h â Z, γ(h) = P hγ(0) with γ(0) solution ofγ(0) = Pγ(0)P â² + (â + Î).
40
![Page 41: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/41.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
Part 2
Application to earthquakes
41
![Page 42: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/42.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
Multivariate models ?
Shapefiles fromhttp://www.colorado.edu/geography/foote/maps/assign/hotspots/hotspots.html
42
![Page 43: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/43.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
The dataset, and stationarity issuesWe work with 16 (17) tectonic plates,â Japan is at the limit of 4 tectonic plates (Pacific, Okhotsk, Philippine and
Amur),â California is at the limit of the Pacific, North American and Juan de Fuca
plates.Data were extracted from the Advanced National Seismic System database(ANSS) http://www.ncedc.org/cnss/catalog-search.html
â 1965-2011 for magnitude M > 5 earthquakes (70,000 events) ;â 1992-2011 for M > 6 earthquakes (3,000 events) ;â To count the number of earthquakes, used time ranges of 3, 6, 12, 24, 36 and
48 hours ;â Approximately 8,500 to 135,000 periods of observation ;
43
![Page 44: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/44.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
Multivariate models : comparing dynamics
X1,t
X2,t
=
p1,1 p1,2
p2,1 p2,2
âŠX1,tâ1
X2,tâ1
+
ε1,t
ε2,t
with var
ε1,t
ε2,t
=
λ1 Ï
Ï Î»2
Complete model, with full dependence
44
![Page 45: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/45.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
Multivariate models : comparing dynamics
X1,t
X2,t
=
p1,1 00 p2,2
âŠX1,tâ1
X2,tâ1
+
ε1,t
ε2,t
with var
ε1,t
ε2,t
=
λ1 Ï
Ï Î»2
Partial model, with diagonal thinning matrix, no-crossed lag correlation
45
![Page 46: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/46.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
Multivariate models : comparing dynamics
X1,t
X2,t
=
p1,1 00 p2,2
âŠX1,tâ1
X2,tâ1
+
ε1,t
ε2,t
with var
ε1,t
ε2,t
=
λ1 00 λ2
Two independent INAR processes
46
![Page 47: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/47.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
Multivariate models : comparing dynamics
X1,t
X2,t
=
p1,1 00 p2,2
âŠX1,tâ1
X2,tâ1
+
ε1,t
ε2,t
with var
ε1,t
ε2,t
=
λ1 00 λ2
Two independent INAR processes
47
![Page 48: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/48.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
Multivariate models : comparing dynamics
X1,t
X2,t
=
0 00 0
âŠX1,tâ1
X2,tâ1
+
ε1,t
ε2,t
with var
ε1,t
ε2,t
=
λ1 Ï
Ï Î»2
Two (possibly dependent) Poisson processes
48
![Page 49: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/49.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
Multivariate models : tectonic plates interactionsâ For all pairs of tectonic plates, at all frequencies, autoregression in time is
important (very high statistical significance) ;â Long sequence of zeros, then mainshocks and aftershocks ;â Rate of aftershocks decreases exponentially over time (Omoriâs law) ;
â For 7-13% of pairs of tectonic plates, diagonal BINAR has significant better fitthan independent INARs ;â Contribution of dependence in noise ;â Spatial contagion of order 0 (within h hours) ;â Contiguous tectonic plates ;
â For 7-9% of pairs of tectonic plates, proposed BINAR has significant better fitthan diagonal BINAR ;â Contribution of spatial contagion of order 1 (in time interval [h, 2h]) ;â Contiguous tectonic plates ;
â for approximately 90%, there is no significant spatial contagion for M > 5earthquakes
49
![Page 50: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/50.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
Granger causality N1 â N2 or N1 â N2
1. North American Plate, 2. Eurasian Plate,3. Okhotsk Plate, 4. Pacific Plate (East), 5. Pacific Plate (West), 6. Amur
Plate, 7. Indo-Australian Plate, 8. African Plate, 9. Indo-Chinese Plate, 10. Arabian Plate, 11. Philippine Plate, 12.
Coca Plate, 13. Caribbean Plate, 14. Somali Plate, 15. South American Plate, 16. Nasca Plate, 17. Antarctic Plate
â
â
â
â
â
â
â
â
â
â
â
â
â
â
â
â
â
â
17
16
15
14
13
12
11
10
9
8
7
6
5
4
3
2
1
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
Granger Causality test, 3 hours
â
â
â
â
â
â
â
â
â
â
â
â
â
â
â
â
â
â
17
16
15
14
13
12
11
10
9
8
7
6
5
4
3
2
1
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
Granger Causality test, 6 hours
50
![Page 51: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/51.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
Granger causality N1 â N2 or N1 â N2
1. North American Plate, 2. Eurasian Plate,3. Okhotsk Plate, 4. Pacific Plate (East), 5. Pacific Plate (West), 6. Amur
Plate, 7. Indo-Australian Plate, 8. African Plate, 9. Indo-Chinese Plate, 10. Arabian Plate, 11. Philippine Plate, 12.
Coca Plate, 13. Caribbean Plate, 14. Somali Plate, 15. South American Plate, 16. Nasca Plate, 17. Antarctic Plate
â
â
â
â
â
â
â
â
â
â
â
â
â
â
â
â
â
â
17
16
15
14
13
12
11
10
9
8
7
6
5
4
3
2
1
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
Granger Causality test, 12 hours
â
â
â
â
â
â
â
â
â
â
â
â
â
â
â
â
â
â
17
16
15
14
13
12
11
10
9
8
7
6
5
4
3
2
1
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
Granger Causality test, 24 hours
51
![Page 52: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/52.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
Granger causality N1 â N2 or N1 â N2
1. North American Plate, 2. Eurasian Plate,3. Okhotsk Plate, 4. Pacific Plate (East), 5. Pacific Plate (West), 6. Amur
Plate, 7. Indo-Australian Plate, 8. African Plate, 9. Indo-Chinese Plate, 10. Arabian Plate, 11. Philippine Plate, 12.
Coca Plate, 13. Caribbean Plate, 14. Somali Plate, 15. South American Plate, 16. Nasca Plate, 17. Antarctic Plate
â
â
â
â
â
â
â
â
â
â
â
â
â
â
â
â
â
â
17
16
15
14
13
12
11
10
9
8
7
6
5
4
3
2
1
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
Granger Causality test, 36 hours
â
â
â
â
â
â
â
â
â
â
â
â
â
â
â
â
â
â
17
16
15
14
13
12
11
10
9
8
7
6
5
4
3
2
1
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
Granger Causality test, 48 hours
52
![Page 53: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/53.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
Multivariate models : frequency versus magnitudeX1,t =
âi=1
1(Ti â [t, t+ 1),Mi †s) and X2,t =âi=1
1(Ti â [t, t+ 1),Mi > s)
Here we work on two sets of data : medium-size earthquakes (M â (5, 6)) andlarge-size earthquakes (M > 6).
â Investigate direction of relationship (which one causes the other, or both) ;â Pairs of tectonic plates :
â Uni-directional causality : most common for contiguous plates (NorthAmerican causes West Pacific, Okhotsk causes Amur) ;
â Bi-directional causality : Okhotsk and West Pacific, South American andNasca for example ;
â Foreshocks and aftershocks :â Aftershocks much more significant than foreshocks (as expected) ;â Foreshocks announce arrival of larger-size earthquakes ;â Foreshocks significant for Okhotsk, West Pacific, Indo-Australian,
Indo-Chinese, Philippine, South American ;
53
![Page 54: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/54.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
Risk management issuesâ Interested in computing P
(âTt=1 (N1,t +N2,t) ⥠n
â£â£â£F0
)for various values of T
(time horizons) and n (tail risk measure) ;â Total number of earthquakes on a set of two tectonic plates ;
â 100 000 simulated paths of diagonal and proposed BINAR models ;â Use estimated parameters of both models ;â Pair : Okhotsk and West Pacific ;
â Scenario : on a 12-hour period, 23 earthquakes on Okhotsk and 46 earthquakeson West Pacific (second half of March 10th, 2011) ;
54
![Page 55: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/55.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
Diagonal model
n / days 1 day 3 days 7 days 14 days
5 0.9680 0.9869 0.9978 0.999910 0.5650 0.7207 0.8972 0.988415 0.1027 0.2270 0.4978 0.854820 0.0067 0.0277 0.1308 0.4997
Proposed model
n / days 1 day 3 days 7 days 14 days
5 0.9946 0.9977 0.9997 1.000010 0.8344 0.9064 0.9712 0.997015 0.3638 0.5288 0.7548 0.947920 0.0671 0.1573 0.3616 0.7256
55
![Page 56: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/56.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
Some referencesAl-Osh, M.A. & A.A. Alzaid (1987) "First-order integer-valued autoregressiveprocess", Journal of Time Series Analysis 8, 261-275.
Dion, J.-P., G. Gauthier & A. Latour (1995), "Branching processes withimmigration and integer-valued time series", Serdica Mathematical Journal 21,123-136.
Du, J.-G. & Y. Li (1991), "The integer-valued autoregressive (INAR(p))model", Journal of Time Series Analysis 12, 129-142.
Ferland, R.A., Latour, A. & Oraichi, D. (2006), Integer-valued GARCHprocess. Journal of Time Series Analysis 27, 923-942.
Fokianos, K. (2011). Count time series models. to appear in Handbook ofTime Series Analysis.
Franke, J. & Subba Rao, T. (1993). Multivariae first-order integer-valuedautoregressions. Forschung UniversitÀt Kaiserslautern, 95.
56
![Page 57: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/57.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
Gourieroux, C. & J. Jasiak (2004) "Heterogeneous INAR(1) model withapplication to car insurance", Insurance : Mathematics & Economics 34, 177-192.
Gardner, J.K. & I. Knopoff (1974) "Is the sequence of earthquakes inSouthern California, with aftershocks removed, Poissonean ?" Bulletin of theSeismological Society of America 64, 1363-1367.
Joe, H. (1996) Time series models with univariate margins in theconvolution-closed infinitely divisible class. Journal of Time Series Analysis 104,117-133.
Johnson, N.L, Kotz, S. & Balakrishnan, N. (1997). Discrete MultivariateDistributions. Wiley Interscience.
Kagan, Y.Y. & D.D. Jackson (1991) Long-term earthquake clustering,Geophysical Journal International 104, 117-133.
Kocherlakota, S. & Kocherlakota K. (1992). Bivariate DiscreteDistributions. CRC Press.
57
![Page 58: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/58.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
Latour, A. (1997) "The multivariate GINAR(p) process", Advances in AppliedProbability 29, 228-248.
Latour, A. (1998) Existence and stochastic structure of a non-negativeinteger-valued autoregressive process, Journal of Time Series Analysis 19,439-455.
Lomnitz, C. (1974) "Global Tectonic and Earthquake Risk", Elsevier.
LÃŒtkepohl, H. (2005). New Introduction to Multiple Time Series Analysis.Springer Verlag.
Mahamunulu, D. M. (1967). A note on regression in the multivariate Poissondistribution. Journal of the American Statistical Association, 62, 25 1-258.
McKenzie, E. (1985) Some simple models for discrete variate time series, WaterResources Bulletin 21, 645-650.
Orfanogiannaki, K., D. Karlis & G.A. Papadopoulos (2010) "IdentifyingSeismicity Levels via Poisson Hidden Markov Models", Pure and AppliedGeophysics 167, 919-931.
58
![Page 59: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/59.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
Parsons, T., & A.A. Velasco (2011) "Absence of remotely triggered largeearthquakes beyond the mainshock region", Nature Geoscience, March 27th, 2011.
Pedeli, X. & Karlis, D. (2011) "A bivariate Poisson INAR(1) model withapplication", Statistical Modelling 11, 325-349.
Pedeli, X. & Karlis, D. (2011). "On estimation for the bivariate PoissonINAR process", To appear in Communications in Statistics, Simulation andComputation.
Rathbun, S.L. (2004), âSeismological modelingâ, Encyclopedia ofEnvironmetrics.
Rosenblatt, M. (1971). Markov processes, Structure and Asymptotic Behavior.Springer-Verlag.
Steutel, F. & K. van Harn (1979) "Discrete analogues ofself-decomposability and stability", Annals of Probability 7, 893-899.
59
![Page 60: Slides GEOTOP](https://reader034.vdocument.in/reader034/viewer/2022052621/55853632d8b42a86388b528d/html5/thumbnails/60.jpg)
Arthur CHARPENTIER & Mathieu BOUDREAULT, Bivariate count processes for earthquake frequency
Utsu, T. (1969) "Aftershocks and earthquake statistics (I) - Some parameterswhich characterize an aftershock sequence and their interrelations", Journal ofthe Faculty of Science of Hokkaido University, 121-195.
Vere-Jones, D. (2010), "Foundations of Statistical Seismology", Pure andApplied Geophysics 167, 645-653.
Weiβ, C.H. (2008) "Thinning operations for modeling time series of counts : asurvey", Advances in Statistical Analysis 92, 319-341.
Zucchini, W. & I.L. MacDonald (2009) "Hidden Markov Models for TimeSeries : An Introduction Using R", 2nd edition, Chapman & Hall
The article can be downloaded from http ://arxiv.org/abs/1112.0929
60