SlideShare a Scribd company logo
Complexity, tails & trends ...
Nick Watkins (nww@pks.mpg.de)
Max Planck Institute for the Physics of Complex Systems, Dresden, Germany (2013-14)
MCT, Open University
Centre for the Analysis of Time Series, LSE
Centre for Fusion Space & Astrophysics, University of Warwick
Addendum:
These slides were given as a contributed talk at the Trends
2014 meeting at Clare College, Cambridge, UK on 28th July
2014. I have also included the spare slides which were
prepared in case of technical questions etc.
Nick Watkins
29/7/2014
BIG QUESTION: How big, how long
lasting, and how serially dependent
are real world risks and hazards ?
BIG QUESTION: How big, how long
lasting, and how serially dependent
are real world risks and hazards ?
BIG QUESTION: How big, how long
lasting, and how serially dependent
are real world risks and hazards ?
SUBTOPIC: How do 2 types of
complexity affect trend detection ?
This presentation complements those by Sandra and Christian and discusses
aspects of two problems we face in real world time series:
• Long range dependence, if present in a system, would complicate the
trend detection, because it implies the presence of low
frequency "slow" fluctuations.
• Another effect, the presence of heavier than Gaussian tails in probability
distributions, is, meanwhile, a source of "wild" fluctuations.
• Important to distinguish these two concepts, a process begin in the 60s by
Benoit B Mandelbrot.
I will mainly be surveying ideas today, the detail is here:
[1] Watkins, “Bunched black (and grouped grey) swans”,
Frontiers review article, GRL, 2013,
http://onlinelibrary.wiley.com/doi/10.1002/grl.50103/full
[2] Franzke et al, Phil. Trans. A, 2012, doi:10.1098/rsta.2011.0349
[3] Graves et al, “A brief history of long memory”, submitted 2014
arXiv:1406.6018v1 [stat.OT]
[4] Graves et al, “Efficient Bayesian inference for long memory
processes”, submitted 2014, arXiv:1403.2940v1 [stat.ME]
and will be developed in:
[5] Franzke et al, “Hurst without Joseph”, in prep.
[6] Watkins and Franzke, in prep.
 2
Variance
Correlation length
   2 2
( )x
  
  
  

( ) ( ) ( )
e.g. ( ) ~ exp( )
x t x t
How do we quantify “big”, “correlated”?
 2
Variance
Correlation length
   2 2
( )x
  
  
  

( ) ( ) ( )
e.g. ( ) ~ exp( )
x t x t
  

iid Gaussian
white ~ ( )
(0, )N
 2
Correlation length
Damped Brownian motion
"Physics"mv v   
1
First order autoregressive AR(1)
(1 ) "stats"N N N
x x x 
    
 ,x v
Variance  
But what if fluctuations “wild”
in amplitude. Consider fat tailed
pdf , wildest case of which is
power law with small tail
exponent ?
Light tail
Power law tail
(1 )
~
stable
p
range
df (
: 0 2
) xp x 
 
 
  
Fat tails not just curiosity: a well known topic
in financial risk
[S&P 500] Mantegna &
Stanley, Nature, 1996
Fat tails not just curiosity: a well known topic
in financial risk … lead to erratically
fluctuating moments
[S&P 500] Mantegna &
Stanley, Nature, 1996
1963
We also observe fat(tish) tails in solar wind,
ionosphere and magnetosphere
AE: Chapman, Hnat, Rowlands
& Watkins, NPG, 2005
Polar UVI: Uritsky et al, reviewed in
Freeman & Watkins, Science, 2002
SW Poynting flux: Freeman,
Watkins & Riley, PRE, 2000
See also Chapman, et al, GRL, 1998;
Watkins et al, GRL, 1999;
Lui et al, GRL, 2000 etc
Correlation length  
But what if fluctuations “slow”
in time, consider long range
dependence
Light tail
Power law tail
] ~
( ) ( ) ( )
( )
( ) F[
x t x t
S f f


  




  
 

(1 )
~
stable
p
range
df (
: 0 2
) xp x 
 
 
  
We just saw case
Variance  
Nile river minima
600 700 800 900 1000 1100 1200 1300
9
10
11
12
13
14
15
Annual minimum level of Nile: 622-1284
Annualminimum:
Time in years
Nile minima, 622-1284
Long range dependence contentious topic since
Hurst reported anomalous growth of range in Nile
river minima …
600 700 800 900 1000 1100 1200 1300
9
10
11
12
13
14
15
Annual minimum level of Nile: 622-1284
Annualminimum:
Time in years
Hurst, Nature, 1957
Nile minima, 622-1284
Long range dependence contentious topic since
Hurst reported anomalous growth of range in Nile
river minima … and Mandelbrot proposed LRD as
one possible explanation …d parameter … J=d+1/2
600 700 800 900 1000 1100 1200 1300
9
10
11
12
13
14
15
Annual minimum level of Nile: 622-1284
Annualminimum:
Time in years
Hurst, Nature, 1957
Walk built from noise with d=-1/2
As above for d=1/2
Nile minima, 622-1284
Joseph Effect:
29 July 2014 19
Pharoah’s dream of 7 years of plenty and 7 years of drought. Now shuffle
... there came seven years of great plenty throughout the land of Egypt. And
there shall arise after them seven years of famine ...
Genesis: 41, 29-30.
Joseph Effect:
29 July 2014 20
Pharoah’s dream of 7 years of plenty and 7 years of drought. Now shuffle
Point is that marginal distribution, of sample at least, unaffected by
shuffling, but that the two series represent very different worlds for insurers,
or Pharoahs. Former unlikely to happen in random trend free process without LRD.
... there came seven years of great plenty throughout the land of Egypt. And
there shall arise after them seven years of famine ...
Genesis: 41, 29-30.
We see some classic long range dependence
indicators in our space weather data …
AE power spectrum
Tsurutani et al, GRL, 1991
We see some classic long range dependence
indicators in our space weather data …
AE power spectrum .
Tsurutani et al, GRL, 1991
ACF. Watkins, NPG, 2002
after Takalo and Timonen
( )
( ) F[ ] ~S f f



 

 


So how have I been modelling competing
effects of LRD and heavy tails so far ? LFSM
• Use linear fractional stable motion (LFSM) model
• Self-similarity exponent H depends both on memory parameter d and
tail exponent alpha in this class of models.
1 1
1
( ) ( ) ( ) ( )
H H
H H R
X t C t s s dL s 
  
  
 
    
 
   
1/ d H
Memory kernel: d
measures LRD
α-stable jump:
heavy tails
LFSM model unpicks the different scaling of solar
wind Poynting flux driver & response in AE/U/L
Watkins et al, Space Science
Reviews, 2005
Data
Model
Peak pdf vs. diff. time Std dev vs tau
Need better model with LRD & heavy tails & ability to
add HF damping: α-stable AutoRegressive Fractionally
Integrated Moving Average (ARFIMA(p,d,q)):
29 July 2014 25
Ionospheric index
power spectrum .
Tsurutani et al, GRL,
1991
( )(1 ) ( )d
t tB B X B    
1
( ) 1
p
j
j
j
z z

   
1t tBX X 
Granger (& Joyeux), 1980
Need better model with LRD & heavy tails & ability to
add HF damping: α-stable AutoRegressive Fractionally
Integrated Moving Average (ARFIMA(p,d,q)):
29 July 2014 26
0.15
(1 ) t tB X   1.5 Ionospheric index
power spectrum .
Tsurutani et al, GRL,
1991
( )(1 ) ( )d
t tB B X B    
1
( ) 1
p
j
j
j
z z

   
1t tBX X 
Granger (& Joyeux), 1980
Need better inference: Cambridge-BAS PhD Tim Graves
developed Bayesian method: tested on α-stable
ARFIMA(0,d,0) where heavy tails & LRD co-exist
Graves, Gramacy, Franzke & Watkins, submitted 2014; and in prep.
1.5 0.15d 
BUT …
• Both Mandelbrot (1965, 1967), and his critics (notably Vit
Klemes, WRR, 1974) noted that there were nonstationary
models that also produced 1/f spectra-close relatives of what
is now called Alternating Fractional Renewal Process.
• Mandelbrot was at pains to emphasise that needed to use
our eyes as well as mathematical rigour (e.g. his Selecta
series of annotated collected papers).
• Work in progress with Christian on how best to distinguish
“true” LRD from models which share the power spectral
and/or growth of range features but actually show much
shorter typical lengths for dependent runs …
29 July 2014 28
Why does LRD matter in already fat-tailed
hazards ? [Riley, Space Weather, 2012 ]:
• Knowing relative frequency of a coronal mass ejection of a given
magnitude …
Why does LRD matter in already fat-tailed
hazards ? [Riley, Space Weather, 2012 ]:
… Knowing relative frequency of a coronal mass ejection of a given
magnitude doesn’t fully specify the hazard:
Bunching, whether short range, or full blown LRD, affects overall hazard.
“Grouped grey swan” problem …[Watkins, GRL Frontiers, 2013]
Link to interacting hazards
problem, c.f. UCL-Kings-
Southampton Workshop & EGU
session, 2013
Conclusions:
Bold paradigms,
including some
imported from
outside space
physics
Stimulus to statistical
inference-confrontation
of early claims with
rigour
Critical thinking
about
assumptions
and methods
Risk and hazard questions: how big, how correlated ? Basic
statistical concepts (finite) variance and (finite) correlation
length. How we might relax these limits, and why we might
want to [Watkins, GRL Frontiers, 2013]. Showed you some
examples of work in this area [see also Watkins et al, GRL,
1999; Freeman, Watkins et al, PRE, 2000; Watkins et al,
PRE, 2009 and outcomes of BAS Natural Complexity
project & Warwick collaboration].
Statistical methodology: Importance of confronting these
exciting new results and ideas with statistical inference, and
more flexible models such as ARFIMA: e.g. Bayesian
inference of long range dependence [Graves, Gramacy,
Franzke & Watkins, first 2 papers now submitted].
Closing the loop: Better inference alone is not enough,
though-work is in progress on how to distinguish between
competing models. [Franzke et al; Watkins and Franzke]
SPARES
Subtitle
Ideal reservoir
• Average influx over
years, need to ensure annual released
volume equals mean influx:
Accumulated deviation
of the influx from the
mean:
29 July 2014 33
1
1
( )
t
t

 



   
1
) { ( )( },
t
u
uX t   

   
Range:
Standard deviation:
Form against interval Plot loglog.
White Gaussian noise prediction
(Rescaled) range
29 July 2014 34
R
S
0 0max ) min ( ), ,(t tx t x t     
1
S



1/2
/ ~R S 
AR(1): 1st order AutoRegressive
29 July 2014 35
1 1t t tX X  
0 100 200 300 400 500 600 700 800 900 1000
-8
-6
-4
-2
0
2
4
6
8
Example series of AR(1)
1 0.9 
AR(1): 1st order AutoRegressive
29 July 2014 36
1(1( ) ) t tB B X   
1 1t t tX X  
1ttBX X  0 100 200 300 400 500 600 700 800 900 1000
-8
-6
-4
-2
0
2
4
6
8
Example series of AR(1)
1 0.9 
1
( ) 1
p
j
j
j
z z

   
AutoRegressive Fractionally Integrated
Moving Average [ARFIMA(p,d,q)]
29 July 2014 37
( )(1 ) ( )d
t tB B X B    
Autoregressive
term of order p
Moving average
of order q
1
( ) 1
q
j
j
j
z z

   
Fractional
integration of
order d
Granger (& Joyeux), 1980
(1 )d
t tB X  
Pure LRD
ARFIMA(0,d,0 ):
Exact Bayesian inference on ARFIMA
for d in Gaussian special case
• Have data x, assumed to be a realization from a distribution with
likelihood function L, parameters psi are the objects of interest.
• ARFIMA has parameters μ (location), σ (scale), d (order of fractional
difference), φ (autoregression), θ. All but d, and its sign are
essentially nuisance parameters here.
• First assume Gaussian innovations.
• Assume flat priors for μ, log σ and d …
• Even with this, likelihood for d very complex
• No analytic posterior p --- use MCMC sampling29 July 2014 38
( | ) ( ) ( | )x p L x    
Key features
• Don’t want to assume form of p, q – use reversible
jump MCMC [Green, Biometrika 1995]
• Reparameterisation of model to enforce stationarity
constraints on φ and θ.
• Efficient calculation of Gaussian likelihood (long
memory correlation structure prevents use of
standard quick methods)
• Necessary use of Metropolis-Hastings requires careful
selection of proposal distribution
• Parameter correlation (φ,d) requires blocking
29 July 2014 39
Approximate inference in more general
case
• Drop Gaussianity assumption.
• Go to more general distribution (t,α-stable,…)
• Seek joint inference on d, α
• Approximate long memory process as very high order
AR
• Construct likelihood sequentially
• Use auxiliary variables to integrate out unknown
history
29 July 2014 40
Further developments
29 July 2014 41
~1/d n• In addition to joint inference problem.
have studied how posterior variance
of d depends on sample size n,
Motivated by Kiyani, Chapman & Watkins PRE,
2009 study of how errors in structure functions
depend on sample size. Watkins et al, in prep.
LFSM model allowed initial burst scaling study

H2
1
) '( '
t
t
s x t dt  Prediction: ( ) ~ s 1/ (2 H)p s 
   
Watkins et al,
PRE, 2009

More Related Content

What's hot

SVRS submission archive
SVRS submission archiveSVRS submission archive
SVRS submission archive
Timothy Chan
 
Constraints on the Universe as a Numerical Simulation
Constraints on the Universe as a Numerical SimulationConstraints on the Universe as a Numerical Simulation
Constraints on the Universe as a Numerical Simulation
solodoe
 
CLIM: Transition Workshop - On the Probability Distributions of the Intensity...
CLIM: Transition Workshop - On the Probability Distributions of the Intensity...CLIM: Transition Workshop - On the Probability Distributions of the Intensity...
CLIM: Transition Workshop - On the Probability Distributions of the Intensity...
The Statistical and Applied Mathematical Sciences Institute
 
Huterer_UM_colloq
Huterer_UM_colloqHuterer_UM_colloq
Huterer_UM_colloq
Dragan Huterer
 
Born reciprocity
Born reciprocityBorn reciprocity
Born reciprocity
Rene Kotze
 
Doulbe Occupancy as a Probe of the Mott Transition for Fermions in One-dimens...
Doulbe Occupancy as a Probe of the Mott Transition for Fermions in One-dimens...Doulbe Occupancy as a Probe of the Mott Transition for Fermions in One-dimens...
Doulbe Occupancy as a Probe of the Mott Transition for Fermions in One-dimens...
Jorge Quintanilla
 
Statistical Physics Assignment Help
Statistical Physics Assignment HelpStatistical Physics Assignment Help
Statistical Physics Assignment Help
Statistics Assignment Help
 
SPM Physics Formula List Form4
SPM Physics Formula List Form4SPM Physics Formula List Form4
SPM Physics Formula List Form4
Zhang Ewe
 
Problem for the gravitational field
 Problem for the gravitational field Problem for the gravitational field
Problem for the gravitational field
Alexander Decker
 
Volume_7_avrami
Volume_7_avramiVolume_7_avrami
Volume_7_avrami
John Obuch
 
Inversão com sigmoides
Inversão com sigmoidesInversão com sigmoides
Inversão com sigmoides
juarezsa
 
Soil dyn __2
Soil dyn __2Soil dyn __2
Gravity and time
Gravity and timeGravity and time
Gravity and time
Eran Sinbar
 
Climate Extremes Workshop - Decompositions of Dependence for High-Dimensiona...
Climate Extremes Workshop -  Decompositions of Dependence for High-Dimensiona...Climate Extremes Workshop -  Decompositions of Dependence for High-Dimensiona...
Climate Extremes Workshop - Decompositions of Dependence for High-Dimensiona...
The Statistical and Applied Mathematical Sciences Institute
 
Lap2009c&p105-ode3.5 ht
Lap2009c&p105-ode3.5 htLap2009c&p105-ode3.5 ht
Lap2009c&p105-ode3.5 ht
A Jorge Garcia
 
Exact Exchange in Density Functional Theory
Exact Exchange in Density Functional TheoryExact Exchange in Density Functional Theory
Exact Exchange in Density Functional Theory
ABDERRAHMANE REGGAD
 
N. Bilic - "Hamiltonian Method in the Braneworld" 3/3
N. Bilic - "Hamiltonian Method in the Braneworld" 3/3N. Bilic - "Hamiltonian Method in the Braneworld" 3/3
N. Bilic - "Hamiltonian Method in the Braneworld" 3/3
SEENET-MTP
 
Localized Electrons with Wien2k
Localized Electrons with Wien2kLocalized Electrons with Wien2k
Localized Electrons with Wien2k
ABDERRAHMANE REGGAD
 
ON OPTIMIZATION OF MANUFACTURING OF MULTICHANNEL HETEROTRANSISTORS TO INCREAS...
ON OPTIMIZATION OF MANUFACTURING OF MULTICHANNEL HETEROTRANSISTORS TO INCREAS...ON OPTIMIZATION OF MANUFACTURING OF MULTICHANNEL HETEROTRANSISTORS TO INCREAS...
ON OPTIMIZATION OF MANUFACTURING OF MULTICHANNEL HETEROTRANSISTORS TO INCREAS...
ijrap
 
International Refereed Journal of Engineering and Science (IRJES)
International Refereed Journal of Engineering and Science (IRJES)International Refereed Journal of Engineering and Science (IRJES)
International Refereed Journal of Engineering and Science (IRJES)
irjes
 

What's hot (20)

SVRS submission archive
SVRS submission archiveSVRS submission archive
SVRS submission archive
 
Constraints on the Universe as a Numerical Simulation
Constraints on the Universe as a Numerical SimulationConstraints on the Universe as a Numerical Simulation
Constraints on the Universe as a Numerical Simulation
 
CLIM: Transition Workshop - On the Probability Distributions of the Intensity...
CLIM: Transition Workshop - On the Probability Distributions of the Intensity...CLIM: Transition Workshop - On the Probability Distributions of the Intensity...
CLIM: Transition Workshop - On the Probability Distributions of the Intensity...
 
Huterer_UM_colloq
Huterer_UM_colloqHuterer_UM_colloq
Huterer_UM_colloq
 
Born reciprocity
Born reciprocityBorn reciprocity
Born reciprocity
 
Doulbe Occupancy as a Probe of the Mott Transition for Fermions in One-dimens...
Doulbe Occupancy as a Probe of the Mott Transition for Fermions in One-dimens...Doulbe Occupancy as a Probe of the Mott Transition for Fermions in One-dimens...
Doulbe Occupancy as a Probe of the Mott Transition for Fermions in One-dimens...
 
Statistical Physics Assignment Help
Statistical Physics Assignment HelpStatistical Physics Assignment Help
Statistical Physics Assignment Help
 
SPM Physics Formula List Form4
SPM Physics Formula List Form4SPM Physics Formula List Form4
SPM Physics Formula List Form4
 
Problem for the gravitational field
 Problem for the gravitational field Problem for the gravitational field
Problem for the gravitational field
 
Volume_7_avrami
Volume_7_avramiVolume_7_avrami
Volume_7_avrami
 
Inversão com sigmoides
Inversão com sigmoidesInversão com sigmoides
Inversão com sigmoides
 
Soil dyn __2
Soil dyn __2Soil dyn __2
Soil dyn __2
 
Gravity and time
Gravity and timeGravity and time
Gravity and time
 
Climate Extremes Workshop - Decompositions of Dependence for High-Dimensiona...
Climate Extremes Workshop -  Decompositions of Dependence for High-Dimensiona...Climate Extremes Workshop -  Decompositions of Dependence for High-Dimensiona...
Climate Extremes Workshop - Decompositions of Dependence for High-Dimensiona...
 
Lap2009c&p105-ode3.5 ht
Lap2009c&p105-ode3.5 htLap2009c&p105-ode3.5 ht
Lap2009c&p105-ode3.5 ht
 
Exact Exchange in Density Functional Theory
Exact Exchange in Density Functional TheoryExact Exchange in Density Functional Theory
Exact Exchange in Density Functional Theory
 
N. Bilic - "Hamiltonian Method in the Braneworld" 3/3
N. Bilic - "Hamiltonian Method in the Braneworld" 3/3N. Bilic - "Hamiltonian Method in the Braneworld" 3/3
N. Bilic - "Hamiltonian Method in the Braneworld" 3/3
 
Localized Electrons with Wien2k
Localized Electrons with Wien2kLocalized Electrons with Wien2k
Localized Electrons with Wien2k
 
ON OPTIMIZATION OF MANUFACTURING OF MULTICHANNEL HETEROTRANSISTORS TO INCREAS...
ON OPTIMIZATION OF MANUFACTURING OF MULTICHANNEL HETEROTRANSISTORS TO INCREAS...ON OPTIMIZATION OF MANUFACTURING OF MULTICHANNEL HETEROTRANSISTORS TO INCREAS...
ON OPTIMIZATION OF MANUFACTURING OF MULTICHANNEL HETEROTRANSISTORS TO INCREAS...
 
International Refereed Journal of Engineering and Science (IRJES)
International Refereed Journal of Engineering and Science (IRJES)International Refereed Journal of Engineering and Science (IRJES)
International Refereed Journal of Engineering and Science (IRJES)
 

Viewers also liked

2014 Wroclaw Mandelbrot's Other Model of 1/f Noise
2014 Wroclaw Mandelbrot's Other Model of 1/f Noise2014 Wroclaw Mandelbrot's Other Model of 1/f Noise
2014 Wroclaw Mandelbrot's Other Model of 1/f Noise
Nick Watkins
 
2014 tromso generalised and fractional Langevin equations implications for en...
2014 tromso generalised and fractional Langevin equations implications for en...2014 tromso generalised and fractional Langevin equations implications for en...
2014 tromso generalised and fractional Langevin equations implications for en...
Nick Watkins
 
AGU 2012 Bayesian analysis of non Gaussian LRD processes
AGU 2012 Bayesian analysis of non Gaussian LRD processesAGU 2012 Bayesian analysis of non Gaussian LRD processes
AGU 2012 Bayesian analysis of non Gaussian LRD processes
Nick Watkins
 
Cyprus 2011 complexity extreme bursts and volatility bunching in solar terres...
Cyprus 2011 complexity extreme bursts and volatility bunching in solar terres...Cyprus 2011 complexity extreme bursts and volatility bunching in solar terres...
Cyprus 2011 complexity extreme bursts and volatility bunching in solar terres...
Nick Watkins
 
2014 AGU Balancing fact and formula in complex sytems: the example of 1/f noise
2014 AGU Balancing fact and formula in complex sytems: the example of 1/f noise2014 AGU Balancing fact and formula in complex sytems: the example of 1/f noise
2014 AGU Balancing fact and formula in complex sytems: the example of 1/f noise
Nick Watkins
 
Dresden 2014 A tour of some fractional models and the physics behind them
Dresden 2014 A tour of some fractional models and the physics behind themDresden 2014 A tour of some fractional models and the physics behind them
Dresden 2014 A tour of some fractional models and the physics behind them
Nick Watkins
 
2014 OU Mandelbrot's eyes and 1/f noise
2014 OU Mandelbrot's eyes and 1/f noise2014 OU Mandelbrot's eyes and 1/f noise
2014 OU Mandelbrot's eyes and 1/f noise
Nick Watkins
 
Hyderabad 2010 Distributions of extreme bursts above thresholds in a fraction...
Hyderabad 2010 Distributions of extreme bursts above thresholds in a fraction...Hyderabad 2010 Distributions of extreme bursts above thresholds in a fraction...
Hyderabad 2010 Distributions of extreme bursts above thresholds in a fraction...
Nick Watkins
 
2015 Dresden Mandelbrot's other route to1 over f
2015 Dresden Mandelbrot's other route to1 over f2015 Dresden Mandelbrot's other route to1 over f
2015 Dresden Mandelbrot's other route to1 over f
Nick Watkins
 
Venice 2012 Two topics in the history of complexity: Bunched Black Swans and ...
Venice 2012 Two topics in the history of complexity: Bunched Black Swans and ...Venice 2012 Two topics in the history of complexity: Bunched Black Swans and ...
Venice 2012 Two topics in the history of complexity: Bunched Black Swans and ...
Nick Watkins
 
A Random Walk Through Search Research
A Random Walk Through Search ResearchA Random Walk Through Search Research
A Random Walk Through Search Research
Nick Watkins
 

Viewers also liked (11)

2014 Wroclaw Mandelbrot's Other Model of 1/f Noise
2014 Wroclaw Mandelbrot's Other Model of 1/f Noise2014 Wroclaw Mandelbrot's Other Model of 1/f Noise
2014 Wroclaw Mandelbrot's Other Model of 1/f Noise
 
2014 tromso generalised and fractional Langevin equations implications for en...
2014 tromso generalised and fractional Langevin equations implications for en...2014 tromso generalised and fractional Langevin equations implications for en...
2014 tromso generalised and fractional Langevin equations implications for en...
 
AGU 2012 Bayesian analysis of non Gaussian LRD processes
AGU 2012 Bayesian analysis of non Gaussian LRD processesAGU 2012 Bayesian analysis of non Gaussian LRD processes
AGU 2012 Bayesian analysis of non Gaussian LRD processes
 
Cyprus 2011 complexity extreme bursts and volatility bunching in solar terres...
Cyprus 2011 complexity extreme bursts and volatility bunching in solar terres...Cyprus 2011 complexity extreme bursts and volatility bunching in solar terres...
Cyprus 2011 complexity extreme bursts and volatility bunching in solar terres...
 
2014 AGU Balancing fact and formula in complex sytems: the example of 1/f noise
2014 AGU Balancing fact and formula in complex sytems: the example of 1/f noise2014 AGU Balancing fact and formula in complex sytems: the example of 1/f noise
2014 AGU Balancing fact and formula in complex sytems: the example of 1/f noise
 
Dresden 2014 A tour of some fractional models and the physics behind them
Dresden 2014 A tour of some fractional models and the physics behind themDresden 2014 A tour of some fractional models and the physics behind them
Dresden 2014 A tour of some fractional models and the physics behind them
 
2014 OU Mandelbrot's eyes and 1/f noise
2014 OU Mandelbrot's eyes and 1/f noise2014 OU Mandelbrot's eyes and 1/f noise
2014 OU Mandelbrot's eyes and 1/f noise
 
Hyderabad 2010 Distributions of extreme bursts above thresholds in a fraction...
Hyderabad 2010 Distributions of extreme bursts above thresholds in a fraction...Hyderabad 2010 Distributions of extreme bursts above thresholds in a fraction...
Hyderabad 2010 Distributions of extreme bursts above thresholds in a fraction...
 
2015 Dresden Mandelbrot's other route to1 over f
2015 Dresden Mandelbrot's other route to1 over f2015 Dresden Mandelbrot's other route to1 over f
2015 Dresden Mandelbrot's other route to1 over f
 
Venice 2012 Two topics in the history of complexity: Bunched Black Swans and ...
Venice 2012 Two topics in the history of complexity: Bunched Black Swans and ...Venice 2012 Two topics in the history of complexity: Bunched Black Swans and ...
Venice 2012 Two topics in the history of complexity: Bunched Black Swans and ...
 
A Random Walk Through Search Research
A Random Walk Through Search ResearchA Random Walk Through Search Research
A Random Walk Through Search Research
 

Similar to Cambridge 2014 Complexity, tails and trends

LSE 2012 Extremes, Bursts and Mandelbrot's Eyes ... and Five Ways to Mis-esti...
LSE 2012 Extremes, Bursts and Mandelbrot's Eyes ... and Five Ways to Mis-esti...LSE 2012 Extremes, Bursts and Mandelbrot's Eyes ... and Five Ways to Mis-esti...
LSE 2012 Extremes, Bursts and Mandelbrot's Eyes ... and Five Ways to Mis-esti...
Nick Watkins
 
Lecture 2 of 2 at Uni Parma 061012
Lecture 2 of 2 at Uni Parma 061012Lecture 2 of 2 at Uni Parma 061012
Lecture 2 of 2 at Uni Parma 061012
Frederick Green
 
Goodness Dispersion Curves for Ultrasonic Guided Wave based SHM
Goodness Dispersion Curves for Ultrasonic Guided Wave based SHMGoodness Dispersion Curves for Ultrasonic Guided Wave based SHM
Goodness Dispersion Curves for Ultrasonic Guided Wave based SHM
Innerspec Technologies
 
Spacey random walks and higher-order data analysis
Spacey random walks and higher-order data analysisSpacey random walks and higher-order data analysis
Spacey random walks and higher-order data analysis
David Gleich
 
REU_paper
REU_paperREU_paper
REU_paper
Hunter Gabbard
 
Damage detection in cfrp plates by means of numerical modeling of lamb waves ...
Damage detection in cfrp plates by means of numerical modeling of lamb waves ...Damage detection in cfrp plates by means of numerical modeling of lamb waves ...
Damage detection in cfrp plates by means of numerical modeling of lamb waves ...
eSAT Journals
 
Wavelet neural network conjunction model in flow forecasting of subhimalayan ...
Wavelet neural network conjunction model in flow forecasting of subhimalayan ...Wavelet neural network conjunction model in flow forecasting of subhimalayan ...
Wavelet neural network conjunction model in flow forecasting of subhimalayan ...
iaemedu
 
Pranav Conference Paper
Pranav Conference PaperPranav Conference Paper
Pranav Conference Paper
Pranav Keskar
 
first research paper
first research paperfirst research paper
first research paper
Justin McKennon
 
As pi re2015_abstracts
As pi re2015_abstractsAs pi re2015_abstracts
As pi re2015_abstracts
Joseph Park
 
Resonance space-time
Resonance space-timeResonance space-time
Resonance space-time
Alexander Frolov
 
Gwendolyn Eadie: A New Method for Time Series Analysis in Astronomy with an A...
Gwendolyn Eadie: A New Method for Time Series Analysis in Astronomy with an A...Gwendolyn Eadie: A New Method for Time Series Analysis in Astronomy with an A...
Gwendolyn Eadie: A New Method for Time Series Analysis in Astronomy with an A...
JeremyHeyl
 
Steed Variance
Steed VarianceSteed Variance
Steed Variance
Jun Steed Huang
 
Resonance space-time
Resonance space-timeResonance space-time
Resonance space-time
Alexander Frolov
 
slides_cedric_weber_1.pdf
slides_cedric_weber_1.pdfslides_cedric_weber_1.pdf
slides_cedric_weber_1.pdf
sasdude1
 
Deep chandra observations_of_pictor_a
Deep chandra observations_of_pictor_aDeep chandra observations_of_pictor_a
Deep chandra observations_of_pictor_a
Sérgio Sacani
 
Development, Optimization, and Analysis of Cellular Automaton Algorithms to S...
Development, Optimization, and Analysis of Cellular Automaton Algorithms to S...Development, Optimization, and Analysis of Cellular Automaton Algorithms to S...
Development, Optimization, and Analysis of Cellular Automaton Algorithms to S...
IRJET Journal
 
Axion Dark Matter Experiment Detailed Design Nbsp And Operations
Axion Dark Matter Experiment  Detailed Design Nbsp And OperationsAxion Dark Matter Experiment  Detailed Design Nbsp And Operations
Axion Dark Matter Experiment Detailed Design Nbsp And Operations
Sandra Long
 
Fractals in Small-World Networks With Time Delay
Fractals in Small-World Networks With Time DelayFractals in Small-World Networks With Time Delay
Fractals in Small-World Networks With Time Delay
Xin-She Yang
 
First Observation of the Earth’s Permanent FreeOscillation s on Ocean Bottom ...
First Observation of the Earth’s Permanent FreeOscillation s on Ocean Bottom ...First Observation of the Earth’s Permanent FreeOscillation s on Ocean Bottom ...
First Observation of the Earth’s Permanent FreeOscillation s on Ocean Bottom ...
Sérgio Sacani
 

Similar to Cambridge 2014 Complexity, tails and trends (20)

LSE 2012 Extremes, Bursts and Mandelbrot's Eyes ... and Five Ways to Mis-esti...
LSE 2012 Extremes, Bursts and Mandelbrot's Eyes ... and Five Ways to Mis-esti...LSE 2012 Extremes, Bursts and Mandelbrot's Eyes ... and Five Ways to Mis-esti...
LSE 2012 Extremes, Bursts and Mandelbrot's Eyes ... and Five Ways to Mis-esti...
 
Lecture 2 of 2 at Uni Parma 061012
Lecture 2 of 2 at Uni Parma 061012Lecture 2 of 2 at Uni Parma 061012
Lecture 2 of 2 at Uni Parma 061012
 
Goodness Dispersion Curves for Ultrasonic Guided Wave based SHM
Goodness Dispersion Curves for Ultrasonic Guided Wave based SHMGoodness Dispersion Curves for Ultrasonic Guided Wave based SHM
Goodness Dispersion Curves for Ultrasonic Guided Wave based SHM
 
Spacey random walks and higher-order data analysis
Spacey random walks and higher-order data analysisSpacey random walks and higher-order data analysis
Spacey random walks and higher-order data analysis
 
REU_paper
REU_paperREU_paper
REU_paper
 
Damage detection in cfrp plates by means of numerical modeling of lamb waves ...
Damage detection in cfrp plates by means of numerical modeling of lamb waves ...Damage detection in cfrp plates by means of numerical modeling of lamb waves ...
Damage detection in cfrp plates by means of numerical modeling of lamb waves ...
 
Wavelet neural network conjunction model in flow forecasting of subhimalayan ...
Wavelet neural network conjunction model in flow forecasting of subhimalayan ...Wavelet neural network conjunction model in flow forecasting of subhimalayan ...
Wavelet neural network conjunction model in flow forecasting of subhimalayan ...
 
Pranav Conference Paper
Pranav Conference PaperPranav Conference Paper
Pranav Conference Paper
 
first research paper
first research paperfirst research paper
first research paper
 
As pi re2015_abstracts
As pi re2015_abstractsAs pi re2015_abstracts
As pi re2015_abstracts
 
Resonance space-time
Resonance space-timeResonance space-time
Resonance space-time
 
Gwendolyn Eadie: A New Method for Time Series Analysis in Astronomy with an A...
Gwendolyn Eadie: A New Method for Time Series Analysis in Astronomy with an A...Gwendolyn Eadie: A New Method for Time Series Analysis in Astronomy with an A...
Gwendolyn Eadie: A New Method for Time Series Analysis in Astronomy with an A...
 
Steed Variance
Steed VarianceSteed Variance
Steed Variance
 
Resonance space-time
Resonance space-timeResonance space-time
Resonance space-time
 
slides_cedric_weber_1.pdf
slides_cedric_weber_1.pdfslides_cedric_weber_1.pdf
slides_cedric_weber_1.pdf
 
Deep chandra observations_of_pictor_a
Deep chandra observations_of_pictor_aDeep chandra observations_of_pictor_a
Deep chandra observations_of_pictor_a
 
Development, Optimization, and Analysis of Cellular Automaton Algorithms to S...
Development, Optimization, and Analysis of Cellular Automaton Algorithms to S...Development, Optimization, and Analysis of Cellular Automaton Algorithms to S...
Development, Optimization, and Analysis of Cellular Automaton Algorithms to S...
 
Axion Dark Matter Experiment Detailed Design Nbsp And Operations
Axion Dark Matter Experiment  Detailed Design Nbsp And OperationsAxion Dark Matter Experiment  Detailed Design Nbsp And Operations
Axion Dark Matter Experiment Detailed Design Nbsp And Operations
 
Fractals in Small-World Networks With Time Delay
Fractals in Small-World Networks With Time DelayFractals in Small-World Networks With Time Delay
Fractals in Small-World Networks With Time Delay
 
First Observation of the Earth’s Permanent FreeOscillation s on Ocean Bottom ...
First Observation of the Earth’s Permanent FreeOscillation s on Ocean Bottom ...First Observation of the Earth’s Permanent FreeOscillation s on Ocean Bottom ...
First Observation of the Earth’s Permanent FreeOscillation s on Ocean Bottom ...
 

Recently uploaded

Deep Behavioral Phenotyping in Systems Neuroscience for Functional Atlasing a...
Deep Behavioral Phenotyping in Systems Neuroscience for Functional Atlasing a...Deep Behavioral Phenotyping in Systems Neuroscience for Functional Atlasing a...
Deep Behavioral Phenotyping in Systems Neuroscience for Functional Atlasing a...
Ana Luísa Pinho
 
mô tả các thí nghiệm về đánh giá tác động dòng khí hóa sau đốt
mô tả các thí nghiệm về đánh giá tác động dòng khí hóa sau đốtmô tả các thí nghiệm về đánh giá tác động dòng khí hóa sau đốt
mô tả các thí nghiệm về đánh giá tác động dòng khí hóa sau đốt
HongcNguyn6
 
Sharlene Leurig - Enabling Onsite Water Use with Net Zero Water
Sharlene Leurig - Enabling Onsite Water Use with Net Zero WaterSharlene Leurig - Enabling Onsite Water Use with Net Zero Water
Sharlene Leurig - Enabling Onsite Water Use with Net Zero Water
Texas Alliance of Groundwater Districts
 
Nucleophilic Addition of carbonyl compounds.pptx
Nucleophilic Addition of carbonyl  compounds.pptxNucleophilic Addition of carbonyl  compounds.pptx
Nucleophilic Addition of carbonyl compounds.pptx
SSR02
 
Phenomics assisted breeding in crop improvement
Phenomics assisted breeding in crop improvementPhenomics assisted breeding in crop improvement
Phenomics assisted breeding in crop improvement
IshaGoswami9
 
Topic: SICKLE CELL DISEASE IN CHILDREN-3.pdf
Topic: SICKLE CELL DISEASE IN CHILDREN-3.pdfTopic: SICKLE CELL DISEASE IN CHILDREN-3.pdf
Topic: SICKLE CELL DISEASE IN CHILDREN-3.pdf
TinyAnderson
 
Oedema_types_causes_pathophysiology.pptx
Oedema_types_causes_pathophysiology.pptxOedema_types_causes_pathophysiology.pptx
Oedema_types_causes_pathophysiology.pptx
muralinath2
 
THEMATIC APPERCEPTION TEST(TAT) cognitive abilities, creativity, and critic...
THEMATIC  APPERCEPTION  TEST(TAT) cognitive abilities, creativity, and critic...THEMATIC  APPERCEPTION  TEST(TAT) cognitive abilities, creativity, and critic...
THEMATIC APPERCEPTION TEST(TAT) cognitive abilities, creativity, and critic...
Abdul Wali Khan University Mardan,kP,Pakistan
 
BREEDING METHODS FOR DISEASE RESISTANCE.pptx
BREEDING METHODS FOR DISEASE RESISTANCE.pptxBREEDING METHODS FOR DISEASE RESISTANCE.pptx
BREEDING METHODS FOR DISEASE RESISTANCE.pptx
RASHMI M G
 
Bob Reedy - Nitrate in Texas Groundwater.pdf
Bob Reedy - Nitrate in Texas Groundwater.pdfBob Reedy - Nitrate in Texas Groundwater.pdf
Bob Reedy - Nitrate in Texas Groundwater.pdf
Texas Alliance of Groundwater Districts
 
Travis Hills' Endeavors in Minnesota: Fostering Environmental and Economic Pr...
Travis Hills' Endeavors in Minnesota: Fostering Environmental and Economic Pr...Travis Hills' Endeavors in Minnesota: Fostering Environmental and Economic Pr...
Travis Hills' Endeavors in Minnesota: Fostering Environmental and Economic Pr...
Travis Hills MN
 
8.Isolation of pure cultures and preservation of cultures.pdf
8.Isolation of pure cultures and preservation of cultures.pdf8.Isolation of pure cultures and preservation of cultures.pdf
8.Isolation of pure cultures and preservation of cultures.pdf
by6843629
 
The binding of cosmological structures by massless topological defects
The binding of cosmological structures by massless topological defectsThe binding of cosmological structures by massless topological defects
The binding of cosmological structures by massless topological defects
Sérgio Sacani
 
bordetella pertussis.................................ppt
bordetella pertussis.................................pptbordetella pertussis.................................ppt
bordetella pertussis.................................ppt
kejapriya1
 
Cytokines and their role in immune regulation.pptx
Cytokines and their role in immune regulation.pptxCytokines and their role in immune regulation.pptx
Cytokines and their role in immune regulation.pptx
Hitesh Sikarwar
 
The use of Nauplii and metanauplii artemia in aquaculture (brine shrimp).pptx
The use of Nauplii and metanauplii artemia in aquaculture (brine shrimp).pptxThe use of Nauplii and metanauplii artemia in aquaculture (brine shrimp).pptx
The use of Nauplii and metanauplii artemia in aquaculture (brine shrimp).pptx
MAGOTI ERNEST
 
Medical Orthopedic PowerPoint Templates.pptx
Medical Orthopedic PowerPoint Templates.pptxMedical Orthopedic PowerPoint Templates.pptx
Medical Orthopedic PowerPoint Templates.pptx
terusbelajar5
 
原版制作(carleton毕业证书)卡尔顿大学毕业证硕士文凭原版一模一样
原版制作(carleton毕业证书)卡尔顿大学毕业证硕士文凭原版一模一样原版制作(carleton毕业证书)卡尔顿大学毕业证硕士文凭原版一模一样
原版制作(carleton毕业证书)卡尔顿大学毕业证硕士文凭原版一模一样
yqqaatn0
 
Thornton ESPP slides UK WW Network 4_6_24.pdf
Thornton ESPP slides UK WW Network 4_6_24.pdfThornton ESPP slides UK WW Network 4_6_24.pdf
Thornton ESPP slides UK WW Network 4_6_24.pdf
European Sustainable Phosphorus Platform
 
如何办理(uvic毕业证书)维多利亚大学毕业证本科学位证书原版一模一样
如何办理(uvic毕业证书)维多利亚大学毕业证本科学位证书原版一模一样如何办理(uvic毕业证书)维多利亚大学毕业证本科学位证书原版一模一样
如何办理(uvic毕业证书)维多利亚大学毕业证本科学位证书原版一模一样
yqqaatn0
 

Recently uploaded (20)

Deep Behavioral Phenotyping in Systems Neuroscience for Functional Atlasing a...
Deep Behavioral Phenotyping in Systems Neuroscience for Functional Atlasing a...Deep Behavioral Phenotyping in Systems Neuroscience for Functional Atlasing a...
Deep Behavioral Phenotyping in Systems Neuroscience for Functional Atlasing a...
 
mô tả các thí nghiệm về đánh giá tác động dòng khí hóa sau đốt
mô tả các thí nghiệm về đánh giá tác động dòng khí hóa sau đốtmô tả các thí nghiệm về đánh giá tác động dòng khí hóa sau đốt
mô tả các thí nghiệm về đánh giá tác động dòng khí hóa sau đốt
 
Sharlene Leurig - Enabling Onsite Water Use with Net Zero Water
Sharlene Leurig - Enabling Onsite Water Use with Net Zero WaterSharlene Leurig - Enabling Onsite Water Use with Net Zero Water
Sharlene Leurig - Enabling Onsite Water Use with Net Zero Water
 
Nucleophilic Addition of carbonyl compounds.pptx
Nucleophilic Addition of carbonyl  compounds.pptxNucleophilic Addition of carbonyl  compounds.pptx
Nucleophilic Addition of carbonyl compounds.pptx
 
Phenomics assisted breeding in crop improvement
Phenomics assisted breeding in crop improvementPhenomics assisted breeding in crop improvement
Phenomics assisted breeding in crop improvement
 
Topic: SICKLE CELL DISEASE IN CHILDREN-3.pdf
Topic: SICKLE CELL DISEASE IN CHILDREN-3.pdfTopic: SICKLE CELL DISEASE IN CHILDREN-3.pdf
Topic: SICKLE CELL DISEASE IN CHILDREN-3.pdf
 
Oedema_types_causes_pathophysiology.pptx
Oedema_types_causes_pathophysiology.pptxOedema_types_causes_pathophysiology.pptx
Oedema_types_causes_pathophysiology.pptx
 
THEMATIC APPERCEPTION TEST(TAT) cognitive abilities, creativity, and critic...
THEMATIC  APPERCEPTION  TEST(TAT) cognitive abilities, creativity, and critic...THEMATIC  APPERCEPTION  TEST(TAT) cognitive abilities, creativity, and critic...
THEMATIC APPERCEPTION TEST(TAT) cognitive abilities, creativity, and critic...
 
BREEDING METHODS FOR DISEASE RESISTANCE.pptx
BREEDING METHODS FOR DISEASE RESISTANCE.pptxBREEDING METHODS FOR DISEASE RESISTANCE.pptx
BREEDING METHODS FOR DISEASE RESISTANCE.pptx
 
Bob Reedy - Nitrate in Texas Groundwater.pdf
Bob Reedy - Nitrate in Texas Groundwater.pdfBob Reedy - Nitrate in Texas Groundwater.pdf
Bob Reedy - Nitrate in Texas Groundwater.pdf
 
Travis Hills' Endeavors in Minnesota: Fostering Environmental and Economic Pr...
Travis Hills' Endeavors in Minnesota: Fostering Environmental and Economic Pr...Travis Hills' Endeavors in Minnesota: Fostering Environmental and Economic Pr...
Travis Hills' Endeavors in Minnesota: Fostering Environmental and Economic Pr...
 
8.Isolation of pure cultures and preservation of cultures.pdf
8.Isolation of pure cultures and preservation of cultures.pdf8.Isolation of pure cultures and preservation of cultures.pdf
8.Isolation of pure cultures and preservation of cultures.pdf
 
The binding of cosmological structures by massless topological defects
The binding of cosmological structures by massless topological defectsThe binding of cosmological structures by massless topological defects
The binding of cosmological structures by massless topological defects
 
bordetella pertussis.................................ppt
bordetella pertussis.................................pptbordetella pertussis.................................ppt
bordetella pertussis.................................ppt
 
Cytokines and their role in immune regulation.pptx
Cytokines and their role in immune regulation.pptxCytokines and their role in immune regulation.pptx
Cytokines and their role in immune regulation.pptx
 
The use of Nauplii and metanauplii artemia in aquaculture (brine shrimp).pptx
The use of Nauplii and metanauplii artemia in aquaculture (brine shrimp).pptxThe use of Nauplii and metanauplii artemia in aquaculture (brine shrimp).pptx
The use of Nauplii and metanauplii artemia in aquaculture (brine shrimp).pptx
 
Medical Orthopedic PowerPoint Templates.pptx
Medical Orthopedic PowerPoint Templates.pptxMedical Orthopedic PowerPoint Templates.pptx
Medical Orthopedic PowerPoint Templates.pptx
 
原版制作(carleton毕业证书)卡尔顿大学毕业证硕士文凭原版一模一样
原版制作(carleton毕业证书)卡尔顿大学毕业证硕士文凭原版一模一样原版制作(carleton毕业证书)卡尔顿大学毕业证硕士文凭原版一模一样
原版制作(carleton毕业证书)卡尔顿大学毕业证硕士文凭原版一模一样
 
Thornton ESPP slides UK WW Network 4_6_24.pdf
Thornton ESPP slides UK WW Network 4_6_24.pdfThornton ESPP slides UK WW Network 4_6_24.pdf
Thornton ESPP slides UK WW Network 4_6_24.pdf
 
如何办理(uvic毕业证书)维多利亚大学毕业证本科学位证书原版一模一样
如何办理(uvic毕业证书)维多利亚大学毕业证本科学位证书原版一模一样如何办理(uvic毕业证书)维多利亚大学毕业证本科学位证书原版一模一样
如何办理(uvic毕业证书)维多利亚大学毕业证本科学位证书原版一模一样
 

Cambridge 2014 Complexity, tails and trends

  • 1. Complexity, tails & trends ... Nick Watkins (nww@pks.mpg.de) Max Planck Institute for the Physics of Complex Systems, Dresden, Germany (2013-14) MCT, Open University Centre for the Analysis of Time Series, LSE Centre for Fusion Space & Astrophysics, University of Warwick
  • 2. Addendum: These slides were given as a contributed talk at the Trends 2014 meeting at Clare College, Cambridge, UK on 28th July 2014. I have also included the spare slides which were prepared in case of technical questions etc. Nick Watkins 29/7/2014
  • 3. BIG QUESTION: How big, how long lasting, and how serially dependent are real world risks and hazards ?
  • 4. BIG QUESTION: How big, how long lasting, and how serially dependent are real world risks and hazards ?
  • 5. BIG QUESTION: How big, how long lasting, and how serially dependent are real world risks and hazards ?
  • 6. SUBTOPIC: How do 2 types of complexity affect trend detection ? This presentation complements those by Sandra and Christian and discusses aspects of two problems we face in real world time series: • Long range dependence, if present in a system, would complicate the trend detection, because it implies the presence of low frequency "slow" fluctuations. • Another effect, the presence of heavier than Gaussian tails in probability distributions, is, meanwhile, a source of "wild" fluctuations. • Important to distinguish these two concepts, a process begin in the 60s by Benoit B Mandelbrot.
  • 7. I will mainly be surveying ideas today, the detail is here: [1] Watkins, “Bunched black (and grouped grey) swans”, Frontiers review article, GRL, 2013, http://onlinelibrary.wiley.com/doi/10.1002/grl.50103/full [2] Franzke et al, Phil. Trans. A, 2012, doi:10.1098/rsta.2011.0349 [3] Graves et al, “A brief history of long memory”, submitted 2014 arXiv:1406.6018v1 [stat.OT] [4] Graves et al, “Efficient Bayesian inference for long memory processes”, submitted 2014, arXiv:1403.2940v1 [stat.ME] and will be developed in: [5] Franzke et al, “Hurst without Joseph”, in prep. [6] Watkins and Franzke, in prep.
  • 8.  2 Variance Correlation length    2 2 ( )x           ( ) ( ) ( ) e.g. ( ) ~ exp( ) x t x t How do we quantify “big”, “correlated”?
  • 9.  2 Variance Correlation length    2 2 ( )x           ( ) ( ) ( ) e.g. ( ) ~ exp( ) x t x t     iid Gaussian white ~ ( ) (0, )N
  • 10.  2 Correlation length Damped Brownian motion "Physics"mv v    1 First order autoregressive AR(1) (1 ) "stats"N N N x x x        ,x v
  • 11. Variance   But what if fluctuations “wild” in amplitude. Consider fat tailed pdf , wildest case of which is power law with small tail exponent ? Light tail Power law tail (1 ) ~ stable p range df ( : 0 2 ) xp x        
  • 12. Fat tails not just curiosity: a well known topic in financial risk [S&P 500] Mantegna & Stanley, Nature, 1996
  • 13. Fat tails not just curiosity: a well known topic in financial risk … lead to erratically fluctuating moments [S&P 500] Mantegna & Stanley, Nature, 1996 1963
  • 14. We also observe fat(tish) tails in solar wind, ionosphere and magnetosphere AE: Chapman, Hnat, Rowlands & Watkins, NPG, 2005 Polar UVI: Uritsky et al, reviewed in Freeman & Watkins, Science, 2002 SW Poynting flux: Freeman, Watkins & Riley, PRE, 2000 See also Chapman, et al, GRL, 1998; Watkins et al, GRL, 1999; Lui et al, GRL, 2000 etc
  • 15. Correlation length   But what if fluctuations “slow” in time, consider long range dependence Light tail Power law tail ] ~ ( ) ( ) ( ) ( ) ( ) F[ x t x t S f f                (1 ) ~ stable p range df ( : 0 2 ) xp x         We just saw case Variance  
  • 16. Nile river minima 600 700 800 900 1000 1100 1200 1300 9 10 11 12 13 14 15 Annual minimum level of Nile: 622-1284 Annualminimum: Time in years Nile minima, 622-1284
  • 17. Long range dependence contentious topic since Hurst reported anomalous growth of range in Nile river minima … 600 700 800 900 1000 1100 1200 1300 9 10 11 12 13 14 15 Annual minimum level of Nile: 622-1284 Annualminimum: Time in years Hurst, Nature, 1957 Nile minima, 622-1284
  • 18. Long range dependence contentious topic since Hurst reported anomalous growth of range in Nile river minima … and Mandelbrot proposed LRD as one possible explanation …d parameter … J=d+1/2 600 700 800 900 1000 1100 1200 1300 9 10 11 12 13 14 15 Annual minimum level of Nile: 622-1284 Annualminimum: Time in years Hurst, Nature, 1957 Walk built from noise with d=-1/2 As above for d=1/2 Nile minima, 622-1284
  • 19. Joseph Effect: 29 July 2014 19 Pharoah’s dream of 7 years of plenty and 7 years of drought. Now shuffle ... there came seven years of great plenty throughout the land of Egypt. And there shall arise after them seven years of famine ... Genesis: 41, 29-30.
  • 20. Joseph Effect: 29 July 2014 20 Pharoah’s dream of 7 years of plenty and 7 years of drought. Now shuffle Point is that marginal distribution, of sample at least, unaffected by shuffling, but that the two series represent very different worlds for insurers, or Pharoahs. Former unlikely to happen in random trend free process without LRD. ... there came seven years of great plenty throughout the land of Egypt. And there shall arise after them seven years of famine ... Genesis: 41, 29-30.
  • 21. We see some classic long range dependence indicators in our space weather data … AE power spectrum Tsurutani et al, GRL, 1991
  • 22. We see some classic long range dependence indicators in our space weather data … AE power spectrum . Tsurutani et al, GRL, 1991 ACF. Watkins, NPG, 2002 after Takalo and Timonen ( ) ( ) F[ ] ~S f f          
  • 23. So how have I been modelling competing effects of LRD and heavy tails so far ? LFSM • Use linear fractional stable motion (LFSM) model • Self-similarity exponent H depends both on memory parameter d and tail exponent alpha in this class of models. 1 1 1 ( ) ( ) ( ) ( ) H H H H R X t C t s s dL s                     1/ d H Memory kernel: d measures LRD α-stable jump: heavy tails
  • 24. LFSM model unpicks the different scaling of solar wind Poynting flux driver & response in AE/U/L Watkins et al, Space Science Reviews, 2005 Data Model Peak pdf vs. diff. time Std dev vs tau
  • 25. Need better model with LRD & heavy tails & ability to add HF damping: α-stable AutoRegressive Fractionally Integrated Moving Average (ARFIMA(p,d,q)): 29 July 2014 25 Ionospheric index power spectrum . Tsurutani et al, GRL, 1991 ( )(1 ) ( )d t tB B X B     1 ( ) 1 p j j j z z      1t tBX X  Granger (& Joyeux), 1980
  • 26. Need better model with LRD & heavy tails & ability to add HF damping: α-stable AutoRegressive Fractionally Integrated Moving Average (ARFIMA(p,d,q)): 29 July 2014 26 0.15 (1 ) t tB X   1.5 Ionospheric index power spectrum . Tsurutani et al, GRL, 1991 ( )(1 ) ( )d t tB B X B     1 ( ) 1 p j j j z z      1t tBX X  Granger (& Joyeux), 1980
  • 27. Need better inference: Cambridge-BAS PhD Tim Graves developed Bayesian method: tested on α-stable ARFIMA(0,d,0) where heavy tails & LRD co-exist Graves, Gramacy, Franzke & Watkins, submitted 2014; and in prep. 1.5 0.15d 
  • 28. BUT … • Both Mandelbrot (1965, 1967), and his critics (notably Vit Klemes, WRR, 1974) noted that there were nonstationary models that also produced 1/f spectra-close relatives of what is now called Alternating Fractional Renewal Process. • Mandelbrot was at pains to emphasise that needed to use our eyes as well as mathematical rigour (e.g. his Selecta series of annotated collected papers). • Work in progress with Christian on how best to distinguish “true” LRD from models which share the power spectral and/or growth of range features but actually show much shorter typical lengths for dependent runs … 29 July 2014 28
  • 29. Why does LRD matter in already fat-tailed hazards ? [Riley, Space Weather, 2012 ]: • Knowing relative frequency of a coronal mass ejection of a given magnitude …
  • 30. Why does LRD matter in already fat-tailed hazards ? [Riley, Space Weather, 2012 ]: … Knowing relative frequency of a coronal mass ejection of a given magnitude doesn’t fully specify the hazard: Bunching, whether short range, or full blown LRD, affects overall hazard. “Grouped grey swan” problem …[Watkins, GRL Frontiers, 2013] Link to interacting hazards problem, c.f. UCL-Kings- Southampton Workshop & EGU session, 2013
  • 31. Conclusions: Bold paradigms, including some imported from outside space physics Stimulus to statistical inference-confrontation of early claims with rigour Critical thinking about assumptions and methods Risk and hazard questions: how big, how correlated ? Basic statistical concepts (finite) variance and (finite) correlation length. How we might relax these limits, and why we might want to [Watkins, GRL Frontiers, 2013]. Showed you some examples of work in this area [see also Watkins et al, GRL, 1999; Freeman, Watkins et al, PRE, 2000; Watkins et al, PRE, 2009 and outcomes of BAS Natural Complexity project & Warwick collaboration]. Statistical methodology: Importance of confronting these exciting new results and ideas with statistical inference, and more flexible models such as ARFIMA: e.g. Bayesian inference of long range dependence [Graves, Gramacy, Franzke & Watkins, first 2 papers now submitted]. Closing the loop: Better inference alone is not enough, though-work is in progress on how to distinguish between competing models. [Franzke et al; Watkins and Franzke]
  • 33. Ideal reservoir • Average influx over years, need to ensure annual released volume equals mean influx: Accumulated deviation of the influx from the mean: 29 July 2014 33 1 1 ( ) t t           1 ) { ( )( }, t u uX t        
  • 34. Range: Standard deviation: Form against interval Plot loglog. White Gaussian noise prediction (Rescaled) range 29 July 2014 34 R S 0 0max ) min ( ), ,(t tx t x t      1 S    1/2 / ~R S 
  • 35. AR(1): 1st order AutoRegressive 29 July 2014 35 1 1t t tX X   0 100 200 300 400 500 600 700 800 900 1000 -8 -6 -4 -2 0 2 4 6 8 Example series of AR(1) 1 0.9 
  • 36. AR(1): 1st order AutoRegressive 29 July 2014 36 1(1( ) ) t tB B X    1 1t t tX X   1ttBX X  0 100 200 300 400 500 600 700 800 900 1000 -8 -6 -4 -2 0 2 4 6 8 Example series of AR(1) 1 0.9  1 ( ) 1 p j j j z z     
  • 37. AutoRegressive Fractionally Integrated Moving Average [ARFIMA(p,d,q)] 29 July 2014 37 ( )(1 ) ( )d t tB B X B     Autoregressive term of order p Moving average of order q 1 ( ) 1 q j j j z z      Fractional integration of order d Granger (& Joyeux), 1980 (1 )d t tB X   Pure LRD ARFIMA(0,d,0 ):
  • 38. Exact Bayesian inference on ARFIMA for d in Gaussian special case • Have data x, assumed to be a realization from a distribution with likelihood function L, parameters psi are the objects of interest. • ARFIMA has parameters μ (location), σ (scale), d (order of fractional difference), φ (autoregression), θ. All but d, and its sign are essentially nuisance parameters here. • First assume Gaussian innovations. • Assume flat priors for μ, log σ and d … • Even with this, likelihood for d very complex • No analytic posterior p --- use MCMC sampling29 July 2014 38 ( | ) ( ) ( | )x p L x    
  • 39. Key features • Don’t want to assume form of p, q – use reversible jump MCMC [Green, Biometrika 1995] • Reparameterisation of model to enforce stationarity constraints on φ and θ. • Efficient calculation of Gaussian likelihood (long memory correlation structure prevents use of standard quick methods) • Necessary use of Metropolis-Hastings requires careful selection of proposal distribution • Parameter correlation (φ,d) requires blocking 29 July 2014 39
  • 40. Approximate inference in more general case • Drop Gaussianity assumption. • Go to more general distribution (t,α-stable,…) • Seek joint inference on d, α • Approximate long memory process as very high order AR • Construct likelihood sequentially • Use auxiliary variables to integrate out unknown history 29 July 2014 40
  • 41. Further developments 29 July 2014 41 ~1/d n• In addition to joint inference problem. have studied how posterior variance of d depends on sample size n, Motivated by Kiyani, Chapman & Watkins PRE, 2009 study of how errors in structure functions depend on sample size. Watkins et al, in prep.
  • 42. LFSM model allowed initial burst scaling study  H2 1 ) '( ' t t s x t dt  Prediction: ( ) ~ s 1/ (2 H)p s      Watkins et al, PRE, 2009