BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//RLASKEY//CALENDEROUS//EN
CALSCALE:GREGORIAN
METHOD:PUBLISH
BEGIN:VEVENT
DTSTAMP:20261021T115516Z
LAST-MODIFIED:20150324T132449Z
DTSTART:20150409T200000Z
DTEND:20150409T210000Z
UID:event1431@bu.edu
URL:http://physics.bu.edu/internal/events/show/1431
SUMMARY:Lagged Correlation Networks
DESCRIPTION:Featuring Chester Curme\n\nPart of the PhD Final Oral Exams.\n\
	nExamining Committee: H.E. Stanley\, Irena Vodenska\, Shyam Erramilli\, Rob
	ert Carey\, Mark Kon\n\nAbstract: \n\nTechnological advances have provided 
	scientists with large high-dimensional datasets that describe the behaviors
	 of complex systems: from the statistics of energy levels in complex quantu
	m systems\, to the time-dependent transcription of genes\, to price fluctua
	tions among assets in a financial market. In this environment\, where it ma
	y be difficult to infer the joint distribution of the data\, network scienc
	e has flourished as a way to gain insight into the structure and organizati
	on of such systems by focusing on pairwise interactions.\n\nThis work focus
	es on a particular setting\, in which a system is described by multivariate
	 time series data. We consider time-lagged correlations among elements in t
	his system\, in such a way that the measured interactions among elements ar
	e asymmetric. Finally\, we allow these interactions to be characteristicall
	y weak\, so that statistical uncertainties may be important to consider whe
	n inferring the structure of the system. We introduce a methodology for con
	structing statistically validated networks to describe such a system\, exte
	nd the methodology to accommodate interactions with a periodic component\, 
	and show how consideration of bipartite community structures in these netwo
	rks can aid in the construction of robust statistical models.\n\nAn example
	 of such a system is a financial market\, in which high frequency returns d
	ata may be used to describe contagion\, or the spreading of shocks in price
	 among assets. These data provide the experimental testing ground for our m
	ethodology. We study NYSE data from both the present day and one decade ago
	\, examine the time scales over which the validated lagged correlation netw
	orks exist\, and relate differences in the topological properties of the ne
	tworks to an increasing economic efficiency. We uncover daily periodicities
	 in the validated interactions\, and relate our findings to explanations of
	 the Epps Effect\, an empirical phenomenon of financial time series. We als
	o study bipartite community structures in networks composed of market retur
	ns and news sentiment signals for 40 countries. We compare the degrees to w
	hich markets anticipate news\, and news anticipate markets\, and use the co
	mmunity structures to construct a recommender system for inputs to predicti
	on models. Finally\, we complement this work with novel investigations of t
	he exogenous news items that may drive the financial system using topic mod
	els. This includes an analysis of how investors and the general public may 
	interact with these news items using Internet search data\, and how the div
	ersity of stories in the news both responds to and influences market moveme
	nts.\n\n![Curme](/resources/event-image/1431/64e1649_small)
LOCATION:SCI 352\, 590 Commonwealth Avenue\, 02215
STATUS:CONFIRMED
CLASS:PUBLIC
END:VEVENT
END:VCALENDAR