Detection Abnormal Changes in Financial Markets · 2020. 3. 14. · Time scale of financial markets...

48
Detection of Abnormal Changes in Financial Markets Hideki Takayasu ADAM, Tokyo Institute of Technology (Advance Data Analysis and Modeling Unit) and Sony Computer Science Laboratories., Japan Foreign Exchange Market as a complex system Universal mechanisms of random walks Detection of abnormal changes

Transcript of Detection Abnormal Changes in Financial Markets · 2020. 3. 14. · Time scale of financial markets...

Page 1: Detection Abnormal Changes in Financial Markets · 2020. 3. 14. · Time scale of financial markets 10-3 1 102 105 107 1010 sec min hours day month year dealers' strategy fractal

Detection of Abnormal Changes in Financial Markets

Hideki TakayasuADAM, Tokyo Institute of Technology

(Advance Data Analysis and Modeling Unit)and 

Sony Computer Science Laboratories., Japan

Foreign Exchange Market as a complex systemUniversal mechanisms of random walks

Detection of abnormal changes 

Page 2: Detection Abnormal Changes in Financial Markets · 2020. 3. 14. · Time scale of financial markets 10-3 1 102 105 107 1010 sec min hours day month year dealers' strategy fractal

What are financial markets ?“The coming together of buyers (borrowers) and sellers (lenders) to trade financial securities”

cf. Wikipedia

There are more than 1 million markets as a whole in the world.

Page 3: Detection Abnormal Changes in Financial Markets · 2020. 3. 14. · Time scale of financial markets 10-3 1 102 105 107 1010 sec min hours day month year dealers' strategy fractal

FX market data is very interesting because: 

●Total volume of transaction is the largest in the world

400x1010 USD (4 Trillion) /day

(The whole world trade (materials) is 4x1010 USD /day)

Foreign exchange market: currency marketEx. Buy 1 million US‐dollar by JP‐yen by a rate 123.456 JPY/USD

●The market is open 24 hours from Monday to Friday

(Continuity of data is better than any other markets) 

●Time resolution is very high, 1 msec

●All orders (buy, sell, cancel) are visible 

(except user’s names) 

Mostly for speculation

Page 4: Detection Abnormal Changes in Financial Markets · 2020. 3. 14. · Time scale of financial markets 10-3 1 102 105 107 1010 sec min hours day month year dealers' strategy fractal

From the viewpoint of complex systems

Financial markets are consisted of millions of  different components (individuals, companies)

Millions of markets are interacting

Any news can affect the markets 

Sometimes rules change

Human desire and emotion also affects(the ultimate intangible complex systems)

Changing every time, no steady state ever existed ! 

Scientific approaches seem very difficult, but they are extremely important for our society 

News

Page 5: Detection Abnormal Changes in Financial Markets · 2020. 3. 14. · Time scale of financial markets 10-3 1 102 105 107 1010 sec min hours day month year dealers' strategy fractal

Good properties for scientific study

The structure of one market is rather simple:Assembly of buy and sell orders on 1 dimension (price)(1‐dim data with time stamps)

The aim of each dealer can be simple,             buy at a low price and sell at a high price to earn(Agent‐based modeling is available) 

We can get high quality data for scientific research that enable us detail analysis (Empirical laws are derived from the data)(Number of direct dealers are limited, about 1000) 

Page 6: Detection Abnormal Changes in Financial Markets · 2020. 3. 14. · Time scale of financial markets 10-3 1 102 105 107 1010 sec min hours day month year dealers' strategy fractal

Fractal property = Random walk

Random walk property of market pricesDollar-Yen rate for 13 years

zig-zag Tick level

T.Mizuno, S.Kurihara, M.Takayasu, and H. Takayasu[Physica A324(2003), 296‐302]

Mandelbrot

For shorter time scale the fractal property breaks down

Page 7: Detection Abnormal Changes in Financial Markets · 2020. 3. 14. · Time scale of financial markets 10-3 1 102 105 107 1010 sec min hours day month year dealers' strategy fractal

Time scale of financial markets

10-3 1 102 105 107 1010

sec min hours day month year

dealers' strategy

fractal properties

Financial technology

"10-minutes is an epoch"

millisec

Algorithm trader’s time scale

Langevin model and PUCK model

Page 8: Detection Abnormal Changes in Financial Markets · 2020. 3. 14. · Time scale of financial markets 10-3 1 102 105 107 1010 sec min hours day month year dealers' strategy fractal

Why Random Walk ?Buy and Sell orders occur randomly 

This is roughly correct, but the mechanism is not that simpleDeeper understanding necessary

A naïve question :Does the randomness comes from out‐side or is it an intrinsic property of the market ?Exogenous?  Endogenous?

Exogenous effects exist, of course,  also an endogenous mechanism of chaos exists due to the market rule

Page 9: Detection Abnormal Changes in Financial Markets · 2020. 3. 14. · Time scale of financial markets 10-3 1 102 105 107 1010 sec min hours day month year dealers' strategy fractal

9

Every dealer hopes to increase the money by suitable exchange

buy

sell

gain

1: Try to buy at a low price           

and to sell at a high price

pricebid ask2: Follow the trend 

x spread (greediness)

expectationTrend?

Common strategies …

Yamada et al, Phys. Rev. E 79 (2009), 051120 

The dealer model H.Takayasu, H.Miura, T.Hirabayashi, and K.Hamada [Physica A, 184 (1992),127‐134](The brain of Abe‐Cabinet)   An oldest agent‐based model 

of artificial market

Page 10: Detection Abnormal Changes in Financial Markets · 2020. 3. 14. · Time scale of financial markets 10-3 1 102 105 107 1010 sec min hours day month year dealers' strategy fractal

10

Dealer model (artificial markets, agent models)Each dealer has both buying and selling pricesA trade occurs when buying and selling prices meet

price

Buy     Sell

bids     asks

x(t) market price

Min{Ask(j)}≦Max{Bid(i)}Trades have attractive interactions for the dealers in the price space

PlatinumCO

O

OCO

Similar to catalytic reaction

O O OCO

CO

Or matter‐antimatter reaction

Bid(i) < Ask(i) for all i

Page 11: Detection Abnormal Changes in Financial Markets · 2020. 3. 14. · Time scale of financial markets 10-3 1 102 105 107 1010 sec min hours day month year dealers' strategy fractal

11

Trend‐following action?

?

time

price

Appearance of trend ・・・

depends on dealers

A trend continues or turns ・・・depends on dealers strategy

We can assume a term proportional to latest price change( )

nd x t

Coefficient: positive means trend‐follow  

Negative means anti‐trend action (contrarian)

Page 12: Detection Abnormal Changes in Financial Markets · 2020. 3. 14. · Time scale of financial markets 10-3 1 102 105 107 1010 sec min hours day month year dealers' strategy fractal

12

Compromise:A seller (or buyer) gradually decreases (or increases) the price

In the simplest deterministic dealer model, we assume that after the deal, the seller becomes a buyer, the buyer becomes a seller

price

Ask (seller)

Bid (buyer)

time

Trade occurs

A hasty dealer has a steeper slope

A slow dealer has a gentle slope

(these values are given initially)

nnsa s=1:buyers=‐1:seller

Page 13: Detection Abnormal Changes in Financial Markets · 2020. 3. 14. · Time scale of financial markets 10-3 1 102 105 107 1010 sec min hours day month year dealers' strategy fractal

13

( ) ( ) ( ) ( )n n n n nx t t x t a s t d x t

Evolution of dealers (deterministic, no randomness added)

n‐th dealer’s buying price

(selling price is given by adding the spread L )

A market simulation with 3 deterministic dealers

The mechanism of chaos makes seemingly random fluctuations

Trend‐followBuyer or seller

Page 14: Detection Abnormal Changes in Financial Markets · 2020. 3. 14. · Time scale of financial markets 10-3 1 102 105 107 1010 sec min hours day month year dealers' strategy fractal

What is the chaos in this model ?Chaos is the mechanism of enhancing small difference by nonlinear effect

price

Dealer 0Dealer 1Dealer 2

Dealer 1 deals Dealer 2 deals

Deals are highly nonlinear (step‐function type) irreversible interaction

Dealer 1 and 2 are very different

All dealers are quite similar

“Randomness” is produced by this market mechanism intrinsically

Page 15: Detection Abnormal Changes in Financial Markets · 2020. 3. 14. · Time scale of financial markets 10-3 1 102 105 107 1010 sec min hours day month year dealers' strategy fractal

Without the trend‐follow (d=0),     the market price fluctuates around an equilibrium price.  

Trend‐follow (d≠0) is essential for realization of random walk

( )x t

( )x tTrend‐follow (d≠0) is also essential for volatility clustering

Volatility clustering occurs automatically by this model

t

( 1) ( )x x t x tD º + -

Real Market Simulated MarketxD

Page 16: Detection Abnormal Changes in Financial Markets · 2020. 3. 14. · Time scale of financial markets 10-3 1 102 105 107 1010 sec min hours day month year dealers' strategy fractal

( ) ( )X

P X p x d x

X a

¥

D-

³D = D D

µD

ò

Price difference distribution follows a power law just like the real market 

K.Yamada, H.Takayasu, T.Ito,  M.Takayasu,  Phys. Rev. E79, 051120 (2009)

( ) ( )x x t dt x tD º + -

Our latest data analysis clarified that there exist real dealers who behave quite similar to this model, pair‐wise ordering (buy‐sell) and trend‐follow behaviors  (preparing a paper) 

Page 17: Detection Abnormal Changes in Financial Markets · 2020. 3. 14. · Time scale of financial markets 10-3 1 102 105 107 1010 sec min hours day month year dealers' strategy fractal

17

Theoretically we can show that the motion of mass center is approximated by Langevin eq. in the case d>0  

2

2

1( ) ( ) ( ) , ( ) ( )

nn

d dx t x t F t x t x t

dt Ndt

T.Hirabayashi, H.Takayasu, H.Miura,  K.Hamada,  Fractals, 1(1993), 29‐

Langevin eq. is the key concept for Brownian motion

Therefore, the dealer model produces price motion similar to Brownian motion

In 1908 Langevin introduced the equation as a model of 

Brownian motion of a fine particle suspended in water. 

Page 18: Detection Abnormal Changes in Financial Markets · 2020. 3. 14. · Time scale of financial markets 10-3 1 102 105 107 1010 sec min hours day month year dealers' strategy fractal

Open Access 

Recently, market price dynamics in very short time scale is shown to follow Langevin equation with non‐white noise by analyzing high quality market data, "order‐book data".

( ) ( ) ( ) ( )dv t t v t G t

dtm=- +

Page 19: Detection Abnormal Changes in Financial Markets · 2020. 3. 14. · Time scale of financial markets 10-3 1 102 105 107 1010 sec min hours day month year dealers' strategy fractal

In the case of a colloidal particle motion in water molecules, we expect the following relationsAssume that the colloidal particle is moving to the rightThe density of molecules will changes around the particle 

Financial Brownian particleCore size = bid‐ask spreadCenter at the mid price=

(best‐bid + best‐ask)/2

In material:Water molecules are invisible, however, In market: Financial molecules are visible.  

- + - +

Inner layer

Density changes 

outer layer outer layer

Inner layer diameter = core size + 2 γc

Page 20: Detection Abnormal Changes in Financial Markets · 2020. 3. 14. · Time scale of financial markets 10-3 1 102 105 107 1010 sec min hours day month year dealers' strategy fractal

Density change profile when price goes up (right)

Push by moleculedensity increase

Pull by moleculedensity decrease

ɤc =18 pips (0.001 yen)No correlation for ɤ>100

No correlation for ɤ>100

:[ , ] :inner

( ) { ( ) ( ) ( ( ) ( ))}i x x x x

s t t t x

F t c s a s c s a s- - + +

+D

= - - -å åNumber of particles increased in the inner layer bid side

Number of particles increased in the inner layer ask side

Summing up all changes in the inner layer for a time interval

ɤ : the depth from the best bid (ask)Density change

Page 21: Detection Abnormal Changes in Financial Markets · 2020. 3. 14. · Time scale of financial markets 10-3 1 102 105 107 1010 sec min hours day month year dealers' strategy fractal

( )/slope L t t= D D

Velocity of the colloidal particle is approximated by the following quantity

( )( ) ( ) ( )

i

L tv t F t t

th

D= +

D

( )L tD : Price change per order change→ mean free path

( )th : Noise term

: Time window 100 deals ≒160 sec

tD

( ) :iF t

The total particle number change in the inner layer  (anti‐particles are counted with negative sign)  is highly correlated with market price velocity.

Layered order flow

Page 22: Detection Abnormal Changes in Financial Markets · 2020. 3. 14. · Time scale of financial markets 10-3 1 102 105 107 1010 sec min hours day month year dealers' strategy fractal

The correlation is much higher than the conventional order flow

( ) ( )i

v t F t< >Correlation becomes higher by decomposition of the inner layer

Layered order flow            vs. conventional order flow( )iF t

( ) { ( , ) ( , ) ( , ) ( , )}x

F t c x t c x t a x t a x t¥

- + - +

=-¥

= - - +å( )F t

(sum of increment of all buy orders – sum of increment of all sell orders) 

( ) ( )v t F t< >Correlation is lower because opposite contributions are added

The inner layer orders work as driving force accelerating price motion,and the outer layer work oppositely braking the movement

Page 23: Detection Abnormal Changes in Financial Markets · 2020. 3. 14. · Time scale of financial markets 10-3 1 102 105 107 1010 sec min hours day month year dealers' strategy fractal

Conventionally, it has been believed that 

Price change  ∝ demand  - supply(all buy orders)    (all sell orders)

The fact is Price change  ∝ change of (buy-sell orders) 

only in the inner layer

Outer layer molecules work in an opposite way 

driving force 

drag force (resistive force)

Page 24: Detection Abnormal Changes in Financial Markets · 2020. 3. 14. · Time scale of financial markets 10-3 1 102 105 107 1010 sec min hours day month year dealers' strategy fractal

Space‐time composition of particlesParticle created in the inner layer

Annihilation in the inner, outer

Number of particles coming from outer and annihilated in the inner (origin of drag force)

Particle created in the outer layer

Creation in the inner and outer

Particle annihilated in the inner layer

Particle annihilated in the outer layer

Number of particles created in the inner and annihilated in the inner, never being in the outer   (driving force)

Page 25: Detection Abnormal Changes in Financial Markets · 2020. 3. 14. · Time scale of financial markets 10-3 1 102 105 107 1010 sec min hours day month year dealers' strategy fractal

( )( ) ( ) ( )

i

L tv t F t t

th

D= +

D

( ) ( ) ( ) ( ) ( )i i i i if t c t c t a t a t- + - += - - +

( ) ( ) ( ) ( )oia others

v t v t v t th= + +

0

( )( ) { ( ) ( )}

oi

t

a oi ois

L tv t a t s a t s

tD - +=

D= - + + +

D å

0

( )( ) { ( ) ( ) ( ) ( )}

t

others i i ii iis

L tv t c t s c t s a t s a t s

tD - + - +=

D= + - + - + + +

D å

( ) ( ) ( )oi

t

av t t s v s dsf

= -ò

Molecules coming from outer to inner

Linear Response relation  between            and        

Decomposition of velocity

2 2 2| ( ) | | ( ) | | ( ) |oiav vw f w w=

2( ) ( )oi oi

i ta av v t e dtp ww

¥-

= òFourier Transform

( )oiav t ( )v t

Response function

Velocity can be decomposed into           originated and the others.( )oia t

Page 26: Detection Abnormal Changes in Financial Markets · 2020. 3. 14. · Time scale of financial markets 10-3 1 102 105 107 1010 sec min hours day month year dealers' strategy fractal

Estimation of the response function

2

22

| ( ) || ( ) |

| ( ) |oiav

v

wf w

w< >=< >

2| ( ) |oiav w< >

2

0.00725

0.0764 w+

Average is taken over 200 realizations for window size 28.

0( ) tt e df f -=

00.08, 0.27f d= =

From the data we calculate the power spectrum of          and   

2| ( ) |v w< >

( )v t ( )oiav t

By taking their ratio,                  the spectrum of response function is estimated   

The functional form  of  

is approximated by  a Lorentzian.  This means that the response function is an exponential function:  

2

2

| ( ) |

| ( ) |oiav

v

w

w< >

2

22

| ( ) || ( ) |

| ( ) |oiav

v

wf w

w< >=< >

Page 27: Detection Abnormal Changes in Financial Markets · 2020. 3. 14. · Time scale of financial markets 10-3 1 102 105 107 1010 sec min hours day month year dealers' strategy fractal

( )0

( ) ( )oi

tt s

av t e v s dsdf - -

= ò

Derivation of Langevin Equation and resistivity

( ) ( ) ( ) ( )oia others

v t v t v t th= + +

( ) ( ) ( ) ( )oia others

d d d dv t v t v t t

dt dt dt dth= + +

0

0

( ) ( ) ( )

{ ( ) ( )} ( )oi oia a

F

dv t v t v t

dtv t v t v t

d f

d h f

=- +

=- - - +

0( ) ( ) ( ){ ( ) ( )}

( ) ( )others

dv t v t t

dtv t G t

d f d h

m

=- - + + +

=- +

00.27 0.08 0.19m d f= - = - =

Drag coefficient (viscosity):

In the PRL, time is normalized by φ0 and   0 0' ( ) / 2.2m d f f= - =

contributes as the resistivity force( )

oiav t

Page 28: Detection Abnormal Changes in Financial Markets · 2020. 3. 14. · Time scale of financial markets 10-3 1 102 105 107 1010 sec min hours day month year dealers' strategy fractal

( ) ( ) ( )d

m v t v t f tdt

n=- +

Langevin equation is naturally accepted for colloid in physics

Conventional finance theory neglects the inertia term

0 ( ) ( )v t f tn=- + ( ) ( ) /v t f t n=

mass         resistivity   random force  

We proved existence of the inertia term for price motion

( ) ( ) ( ) ( )dv t t v t G t

dtm=- +

Resistivity is time‐dependentNoise term is not white but highly correlated(to vanish the autocorrelation of v)

( ) /t mm n=

Page 29: Detection Abnormal Changes in Financial Markets · 2020. 3. 14. · Time scale of financial markets 10-3 1 102 105 107 1010 sec min hours day month year dealers' strategy fractal

Brownian motion of a colloid particle 

Random walk of Market prices

time

trajectory

Surrounding molecules are observable

“Particle” is not directly observable

From motion of surrounding molecules the dynamics of “Particle” is estimated

Price 

Water molecules are invisible“Colloid particle” is observableFrom trajectory of a colloid particleThermal motion of water molecules are estimated.

price

The biggest difference is that

Molecule number  1023

Ingenhousz 1785Brown 1827Enstein 1905Perin 1907

Bachelier 1900

A financial molecule costs 1 million USD! 

A lot of similarities are confirmed

Page 30: Detection Abnormal Changes in Financial Markets · 2020. 3. 14. · Time scale of financial markets 10-3 1 102 105 107 1010 sec min hours day month year dealers' strategy fractal

( ) ( ) ( ) ( )dv t t v t G t

dtm=- +

Langevin equation with random resistivity produces power law distributions (known as Kestin‐process)

( ) (1 ( ) ) ( ) ( )v t t t t v t f tm+D = - D +

( ) ( ) ( ) ( )v t t b t v t f t+D = +

( | |) | |P v v a-³ µ ( ) 1b t a< >=

Power law distributions of financial markets can be explained by temporal fluctuation of resistivity of markets

Page 31: Detection Abnormal Changes in Financial Markets · 2020. 3. 14. · Time scale of financial markets 10-3 1 102 105 107 1010 sec min hours day month year dealers' strategy fractal

PUCK model

Deviations from RWDealer model

Potential forceTrend‐follow

Time dependent Langevin eq.

Validity of continuum description

ARCH・GARCH model

Viewing only variance

Inflation Eq.

Macro limit

Crash alarm

Change‐point detection

New PhysicsFinancial Technology Macro Economy

Practical application

Time series analysisArtificial Markets

Our data analyses and models of financial markets

Order‐Book analysis

Layered OB model

Statistical properties of OB and correlation between changes in number of order‐book and velocity

Slow response of outer‐layer contributes as "viscosity"

Knudsen number of markets

1992‐2014‐

1997‐

2014‐

2015‐

2005‐

2002‐

2009‐

2009‐

2012‐

2009‐

2013‐

All empirical laws fulfilled

Page 32: Detection Abnormal Changes in Financial Markets · 2020. 3. 14. · Time scale of financial markets 10-3 1 102 105 107 1010 sec min hours day month year dealers' strategy fractal

PUCK model bridges physics and finance 

2

( )( ) ( 1) ( ) ( ) ( ) ( )

1( )

( ) ( ) ( )2( 1)

M

M

b tv t x t x t x t x t f t

Mb t

x t x t f tx M

1

0

1( ) ( )

M

Mk

x t x t kM

Potential of Unbalanced Complex Kinetics Misako Takayasu et alPhysica A370 (2006) , 91PRE 79(2009), 051120PRE 80 (2009), 056110 

Moving average of price

( ) ( ) Mx t x t

Non‐trivial relation

Random walk

: Moving average of market price

( 1) ( )x t x t

potential

Page 33: Detection Abnormal Changes in Financial Markets · 2020. 3. 14. · Time scale of financial markets 10-3 1 102 105 107 1010 sec min hours day month year dealers' strategy fractal

Price chart

Market Potential

Potential coefficient

An example of PUCK analysis for a typical day

b>0b<0

stable unstable

Page 34: Detection Abnormal Changes in Financial Markets · 2020. 3. 14. · Time scale of financial markets 10-3 1 102 105 107 1010 sec min hours day month year dealers' strategy fractal

Microscopic anomalies are reproduced by PUCK model

Real data

PUCK model using measured {b(t)} with random noise f(t)

Time series Diffusion

Auto-correlation

PUCK potential

Page 35: Detection Abnormal Changes in Financial Markets · 2020. 3. 14. · Time scale of financial markets 10-3 1 102 105 107 1010 sec min hours day month year dealers' strategy fractal

Market potential force in the dealer model

Prediction termContrarian (d<0)

Stable potential (b>0)

Trend‐follower (d>0)

Unstable potential(b<0)

Dealers’ strategy is gradually changed from contrarian to trend‐follower(d<0)

(d>0)

K. Yamada et al,  EPJ B 63(2008), 529,    PRE 79(2009)051120

Page 36: Detection Abnormal Changes in Financial Markets · 2020. 3. 14. · Time scale of financial markets 10-3 1 102 105 107 1010 sec min hours day month year dealers' strategy fractal

The first attack

The second attack

News Effects in Real Markets  9.11.2001

Market potential changed drastically by the news

The market was stable. It became very unstable. It gradually stabilized.

President Bush appeared

Pentagon attacked

A WTC collapsed

22.0

0

‐2.0

1.0

‐1.0

Stable

Unstable

( )b t

Page 37: Detection Abnormal Changes in Financial Markets · 2020. 3. 14. · Time scale of financial markets 10-3 1 102 105 107 1010 sec min hours day month year dealers' strategy fractal

Observation of asymmetric potentials before large crash  (1998 for a month) Yen‐Dollar Physical Review E 80 (2009), 074911 

Stable attractive potential

Asymmetric potentialSymptom of a crash

Unstable but symmetric random walk

Asymmetric potential: onset of crash

3

1

( 1) ( ) (( ) ) ( ) ( )kM

kkP t P t P t Pb t tt f

Page 38: Detection Abnormal Changes in Financial Markets · 2020. 3. 14. · Time scale of financial markets 10-3 1 102 105 107 1010 sec min hours day month year dealers' strategy fractal

PUCK model can be used by financial practitioners via Bloomberg App Portal:

“Order Book Tracer”

Page 39: Detection Abnormal Changes in Financial Markets · 2020. 3. 14. · Time scale of financial markets 10-3 1 102 105 107 1010 sec min hours day month year dealers' strategy fractal

Derivation of time evolution of variance by PUCK model( )( ) ( ) { ( ) ( )} ( )2

b tx t t x t x t x t t f t

In the continuum limit, PUCK model becomes Langevin Eq.

( ) ( ) ( ) ( )dv t t v t F t

dt

( ) ( )

( )x t x t t

v tt

1 ( ) / 2( )

b tt

t

( )( )

f tF t

t

2

2( , ) { ( ) ( , )} ( , ) 0vw v t t vw v t D w v t

t v v

2 2( ) 2 ( ) ( ) 2v

dt t t D

dt

2 ( ') ( ) ( ')vD t t f t f t

2 2( ) 2 {1 2 ( ) } ( )v

t t D t t t t Time evolution of varianceARCH‐GARCH model

1. The mass is normalized to be unity2. Viscosity changes with time 

Fokker‐Planck equation holds for the probability density of velocity

2M

2 2( ) ( ) ( )t v t w t dt

( , )w v t : Probability density of velocity

Page 40: Detection Abnormal Changes in Financial Markets · 2020. 3. 14. · Time scale of financial markets 10-3 1 102 105 107 1010 sec min hours day month year dealers' strategy fractal

By applying the Particle filter version of PUCK model, PUCK parameters, b(t) (the potential coefficient) and M (the size of moving average) are estimated in very short time                                          

Particle filter method:104 PUCK models with different parameters are run in parallel, and they evolve like species of life

A test: sudden change of parameters

Particle filter version

Conventional method

Particle filter PUCK is suitable for real time data analysis 

Y.Yura, H.Takayasu, K.Nakamura, M.TakayasuJour. Stat. Compu. and Simu. Vo.84, 2073‐, 2014

Page 41: Detection Abnormal Changes in Financial Markets · 2020. 3. 14. · Time scale of financial markets 10-3 1 102 105 107 1010 sec min hours day month year dealers' strategy fractal

stable

unstable

10 sec

M: scale of moving average shortens

In this example the change of market condition is detected within 10 seconds

Page 42: Detection Abnormal Changes in Financial Markets · 2020. 3. 14. · Time scale of financial markets 10-3 1 102 105 107 1010 sec min hours day month year dealers' strategy fractal

Example: USD-JPY for a week (Mar 2011)

Intervention

Flash Crash

zooming

1 hourFlash Crash takes more than 10 minutes continuous unstable states with exponential growth

( ) * ( ) ( )dv t v t G t

dtm=+ +

If the market is too much unstable, the viscosity becomes negative, and price difference grow exponentially !

( ) exp( * )v t tmµ

Page 43: Detection Abnormal Changes in Financial Markets · 2020. 3. 14. · Time scale of financial markets 10-3 1 102 105 107 1010 sec min hours day month year dealers' strategy fractal

11 years dayly data of Australia Dollar / Japanese YenNon‐stationary time series analysis・change‐point detection

By human eye we may segment the time series 

A‐h. Sato and H.Takayasu,  arXiv:1309.0602

Page 44: Detection Abnormal Changes in Financial Markets · 2020. 3. 14. · Time scale of financial markets 10-3 1 102 105 107 1010 sec min hours day month year dealers' strategy fractal

Here, there is bias from left down to right topWe apply this method for price changes

0    left           s         right        t

x

There are many methods for change point detection, here, weIntroduce a most basic non‐parametric method (not assuming specific functional form of statistical model)

We apply Fisher’s exact probability test for the change‐point detection.Non‐stationary time series is segmented into stationary periods by the p‐value ( the probability that steady independent random fluctuation can realize the bias)(Fisher's test is very popular in medicine)

Left Right

UP a b a+bDown c d c+d

a+c b+d n

Page 45: Detection Abnormal Changes in Financial Markets · 2020. 3. 14. · Time scale of financial markets 10-3 1 102 105 107 1010 sec min hours day month year dealers' strategy fractal

Assuming that the whole interval is statistically stationary,then this level of deviation (a,b,c,d) occurs with probability less than10‐25

We consider that the minimum point is the strongest change‐point

Page 46: Detection Abnormal Changes in Financial Markets · 2020. 3. 14. · Time scale of financial markets 10-3 1 102 105 107 1010 sec min hours day month year dealers' strategy fractal

The time series is segmented automatically by the statistical analysisFor p‐value smaller than 10 ‐ 4

Price difference

p=10‐10

p=3x10‐27

p=6x10‐5 p=9x10‐9

p=3x10‐8

Page 47: Detection Abnormal Changes in Financial Markets · 2020. 3. 14. · Time scale of financial markets 10-3 1 102 105 107 1010 sec min hours day month year dealers' strategy fractal

PUCK model

Deviations from RWDealer model

Potential forceTrend‐follow

Time dependent Langevin eq.

Validity of continuum description

ARCH・GARCH model

Viewing only variance

Inflation Eq.

Macro limit

Crash alarm

Change‐point detection

New PhysicsFinancial Technology Macro Economy

Practical application

Time series analysisArtificial Markets

Spending about 25 years, we are ready for the next step 

Order‐Book analysis

Layered OB model

Statistical properties of OB and correlation between changes in number of order‐book and velocity

Slow response of outer‐layer contributes as "viscosity"

Knudsen number of markets

1992‐2014‐

1997‐

2014‐

2015‐

2005‐

2002‐

2009‐

2009‐

2012‐

2009‐

2013‐

All empirical laws fulfilled

Page 48: Detection Abnormal Changes in Financial Markets · 2020. 3. 14. · Time scale of financial markets 10-3 1 102 105 107 1010 sec min hours day month year dealers' strategy fractal

Financial  institutions

General public

Firms

StorageAnalysis Simulation

Countries

Observatory

bondscurrencies stocks

Markets

Investors

Change‐pointsPUCK StabilityMalfunctions・・・

StabilizationRule changesInterventions・・・

“Observing the markets as it is” is the first step for establishment of science of financial markets

The world situation

The natural environmentcommodities

derivatives

Financial Observatory of Real‐time Markets