Transcript Slides Lillo

BigData@SNS:
Air traffic & Financial Markets
Fabrizio Lillo
Scuola Normale Superiore di Pisa
Quantitative Finance group
•
•
•
•
Faculty: Giacomo Bormetti, Fabrizio Lillo, Stefano Marmi
7 PostDocs
13 PhD students
2 Master students
Vision:
– Mixture of theoretical and empirical approach
– Interdisciplinary approach to finance and economics
(math, computer science, physics)
– Big data
www.crisis-economics.eu
Empirically grounded agent based
models for the future Air Traffic
Management scenario
Collaborations
• Oxford, Ecole Polytechnique, Princeton, ETH,
CUNY, Central European University Budapest,
Perm State University,…
• S. Anna, Bologna, IMT, Venezia, Palermo,
Ancona,…
Interaction with industry and regulators
• JP Morgan London
• Unicredit Milan (x2)
– Dynamics and Information Research Institute
•
•
•
•
•
HSBC London
Yahoo Barcelona
Capital Fund Management (Paris)
Banca d’Italia
List (Pisa)
Research areas
• Systemic risk and financial instabilities
• Financial and socio-economic networks
• Econometrics, stochastic processes, and option
pricing
• High frequency finance and market
microstructure
• Data mining and data clustering
• Long-horizon predictability of the market and
value investing
• Firm’s growth
• Transport networks
• Agent based models in economics and finance
Air Traffic
• Through funding from EUROCONTROL (the
European agency for air traffic control) we have
access to the database of all the planned and
actual 4D trajectories of all the flights in Europe
for more than one year
– Stylized facts of the air traffic networks
– Optimal design of air spaces for traffic control
– Identification of “hot spots” in the airspace, i.e.
points where flights are rerouted more frequently)
– Agent based models of air traffic control: toward the
new SESAR scenario
Air Traffic Management and networks
From traffic data to design of air traffic control by using community detection in networks
Whole European air traffic over more than one year
Design of sectors from
traffic data
Design of airspace from traffic between
sectors
Average number of communities
Multiresolution community detection
10
3
May 60, 2010
Nc
90
Nc
FABs
NA
2
10
1
10
(c)
0
10
10
0
g
Ecology of real agents in financial markets
Systemic instability at high frequency in financial market
2000
2010
May 6, 2010 Flash Crash
News, sentiment, and finance
Users clicking data for measuring the
importance of a news
• In financial markets, news should help predicting stock prices
• Sentiment analysis typically performs badly
• Access to clicking data of
Yahoo Finance
• We claim this is due to
very heterogeneous importance
of news
• # of clicks to weight importance
• Granger causality
• Coupling news sentiment with web browsing data predicts intraday stock prices
Systemic risk in the interbank market
• Multiplex representation of the interbank network
• What are the most “systemically important” financial institution?
• Statistical models for the interbank networks: Maximum Entropy approach
Statistical inference of high dimensional data (Maximum Entropy, Belief Propagation)