The EC Project Extreme events: Causes and consequences E2C2

1
The EC Project Extreme events Causes and
consequences (E2-C2)

• WP7 Research Activities
• Overview

2
WP7 Socio-Economic Barometer

• Participants
• CDCS J.Kurths
• CEPS G.Schaber, P.Liégeois, P.Van Kerm
• ENS M.Ghil
• MITPAN I.Kuznetsov, M.Rodkin, A.Soloviev,
  F.Winberg
• ULG M.Ausloos
• IGPP V.Keilis-Borok

• Objective
• Development of a socio-economic barometer that
  is the methodology and relevant algorithms and
  software for forecasting crises (or stable
  development) in socio-economic systems.
• The forecast will be based on analysis of
  relevant socio-economic indicators, which are the
  input for the algorithms
• The output of the algorithms is a forecast
  whether a crisis is or is not approaching the
  forecasts will include probabilities of false
  alarms and failures to predict
• Different algorithms will be needed for
  prediction of different crisis types, but for a
  given crisis type we plan to develop a
  self-adaptive algorithm and relevant software
  that can be applied without readaptation in
  different territories.

3
WP7 Approaches

• At each moment an algorithm indicates whether a
  crisis should or should not be expected within
  subsequent ? months, ? being duration of alarm.

4
WP7 Predicting the End of an Economic Recession

• The problem of predicting the end of an American
  economic recession by analysis of macroeconomic
  indicators within the recession period has been
  considered. The study is a technical analysis
  that is a heuristic search of phenomena preceding
  the recession end. The methodology of pattern
  recognition of infrequent events is used. The
  goal is to identify by an analysis of
  macroeconomic indicators a robust and rigidly
  defined prediction algorithm of the yes or no
  variety indicating at any time moment, whether
  the recession end should be expected or not
  within the subsequent months. A specific
  premonitory pattern of six macroeconomic
  indicators that may predict algorithmically the
  recession end has been found for six economic
  recessions in the U.S. between 1960 and 2000. The
  ends of all six recessions under consideration
  are preceded within 5 months by this pattern that
  appears at no other time. The end of the last
  recession occurred in 2001 has been also
  predicted.

5
WP7 Analysis of Time Series Behavior

• The problems that arise in data processing (i)
  what kind of change of the examined time series
  should be treated as a non-random and (ii) what
  kind of change (events) could be tried to be
  forecasted were studied. The approaches to
  development of some formal procedure(s) to
  distinguish random and non-random features and
  random and non-random time-frequency domains in
  the regime of the time series under examination
  have been analyzed.

6

• Difference in the regime of large and ordinary
  events was studied by using the distribution law
  (for the case of discrete events) and the
  spectral approach (in the case of regular data
  series). Methods of the morphological analysis of
  time series were applied for the formal (not
  expert decision based only) determination of
  specific patterns connected with extreme events.
  For example the periods of an increase in crime
  activity and the fuzzy extremes in crime
  statistics series are formally identified by this
  way.

7
The model similar with the Brownian motion model
but with an additional component simulating a
long and random switching in memory has been
used to imitate the main features of the real
dynamic systems and to examine the effectiveness
of methods used for the investigation of such
systems. This model is rather natural and simple,
and it gives the results similar of natural
dynamic systems. The time rows produced by the
model are similar visually with that of typical
natural systems and have a spectrum similar with
the typical spectra of the natural dynamic
systems (time rows of crime statistics
including).
8
Distinguishing the cases of quasi-equilibrium
behavior of the studied systems, and the time
intervals of their highly non-equilibrium
behavior The method of morphologic analysis is
used for recognition of intervals of activation
in both synthetic and natural time series.
Application of the morphologic analysis to the
synthetic series shows its effectiveness in
recognition of time intervals of system activity
increase. Application of the morphologic analysis
to crimes statistics and heavy social incident
statistics have been carried out and preliminary
results have been obtained.
9
Prediction of a strong change in the regime of a
dynamic system using the effect of critical
deceleration This approach was applied firstly
and verified in the case of the long synthetic
time series and the promising results were
obtained. Then the similar method was applied to
natural systems the seismic regime in
California, time variation of the share cost (the
General Motors, the Microsoft, and the General
Electric), and dynamics of the number of crimes.
The preliminary results testify for the
effectiveness of this approach.
10
WP7 Analysis of Background Activity in Complex
Systems

• The patterns of background activity are
  introduced for complex socio-economic systems.
• The analysis of the background activity shows an
  analogy with background seismicity.
• It has been found that the background activity
  can be used for prediction of extreme events in
  the economy (the U.S. economy) and the megacity
  (Los Angeles) is studied.
• The precursory pattern reflects increasing the
  background activity before an extreme event.

11
WP7 Current Prediction of the Increase in the
Unemployment Rate in the U.S.

• The algorithm for prediction a specific
  phenomenon in the dynamics of unemployment
  episodes of a sharp increase in the unemployment
  rate, called Fast Acceleration of Unemployment
  (FAU) was applied for advance prediction of FAUs
  in the U.S. It is based on analysis of trends of
  monthly series of three macroeconomic indicators
  industrial production, long-term interest rate on
  10-year government bonds, and short-term interest
  rate on 3-month bills. The values of the trends
  are considered on the lowest level of resolution,
  distinguishing only the values above and below a
  certain threshold (large and low values). The
  algorithm declares an alarm for FAU after a month
  when all three trends are large. In 1999 an
  experiment in prediction in advance for the U.S.
  was launched. The first prediction for early 2000
  has been correct. The next alarm should be
  declared after April 2006 when all three trends
  for the first time after 2000 are large.

12
WP7 Analysis of Crime Dynamics
Crime dynamics has been studied on the basis of
weekly crime statistics for Yaroslavl (Russia). A
hypothetical algorithm for predicting sharp
surges of homicides (SSH) has been suggested. For
the retrospective data the algorithm predicts
about 70 of SSHs within 30 of alarm time and
gave 4 of false alarms. The mean alarm time is
about 3 weeks.