Fat Tails and the Physics of !Collecve!Dynamics!in!a...
Transcript of Fat Tails and the Physics of !Collecve!Dynamics!in!a...
![Page 1: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/1.jpg)
Lisa Borland Integral Dev. Corp., CA, USA
Synerge:cs Symposium in Honor of Professor Haken On the Occasion of his 85th Birthday Delmenhorst, November 2012
The Physics of Finance: Collec:ve Dynamics in a Complex World Fat Tails and the Physics of
Finance
Lisa BorlandEvnine and Associates
Kavli Frontiers of Science 2007
![Page 2: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/2.jpg)
The price forma-on process: mul-tude of complex global interac-ons Not possible to run experiments: just one single history, one realiza-on
![Page 3: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/3.jpg)
The price forma-on process: mul-tude of complex global interac-ons Not possible to run experiments: just one single history, one realiza-on
Tsalllis (q = 1.4)
![Page 4: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/4.jpg)
•For q = 1.4, a 10-sigma event has probabilty
p = 0.028%
•This is equally probable as a 3.45-sigma event
based on a Gaussian distribution
•For a Gaussian, a 10-sigma event has probability
p = 0%
(about 1 in a decade)
![Page 5: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/5.jpg)
What do we know about volatility?
- Across time for a given stock Part 1 - Across stocks at a given time Part 2
Exploring the joint stochastic process
![Page 6: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/6.jpg)
Some more proper-es of financial -me-‐series Universal features: Across -me, Across the globe
For stocks (IBM), Commodi-es (Oil), Currencies (EUR/USD) etc
• Power Law distribu-ons, persistent
• Slow decay to Gaussian
• Vola-lity clustering
• Close-‐to log-‐normal distribu-on of vola-lity
• Time-‐reversal asymmetry
• Leverage effect (Skew: Nega-ve returns -‐-‐> Higher vola-lity)
![Page 7: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/7.jpg)
Vast number of anomalous sta-s-cs (stylized facts)
Across Time:
![Page 8: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/8.jpg)
Challenge: A somewhat realis-c model of price varia-ons Important for real-‐world reasons: -‐ Risk control
-‐ Hedging
-‐ Development of trading strategies
-‐ Op-on pricing
-‐ Pricing of credit risk
![Page 9: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/9.jpg)
Century long quest: First random walk model by Bachelier in 1900
Gaussian model, basis for celebrated 1974 Black-‐Scholes op-on pricing formula
!
"
µ
S Price Rate of return Vola-lity Gaussian noise
!"µ SdSdtdS +=
![Page 10: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/10.jpg)
o Empirical -‐-‐-‐ Gaussian q=1.43 Tsallis Distribu-on
100 SP stocks, daily
Real log-‐returns not Gaussian
![Page 11: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/11.jpg)
The ideal model: -‐ Intui-ve
-‐ Analy-cally tractable
-‐ Parsimonious
Popular models: -‐Stochas-c Vola-lity (Heston 1993) -‐Levy noise -‐GARCH -‐ Mul--‐fractal model (Bacry,Delour Muzy PRE 64 2001)
![Page 12: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/12.jpg)
Exploring the Joint Stochas-c Process:
Across -me ….
1. A non-‐Gaussian model of returns
2. Op-ons pricing incorpora-ng fat-‐tails 3. Applica-ons
![Page 13: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/13.jpg)
!+= SdSdtdS "µ
Non-‐Gaussian model based on sta-s-cal feedback
![Page 14: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/14.jpg)
!+= SdSdtdS "µ
!dPdq21
)("
#=#
For stocks q = 1.4
Borland L, Phys. Rev.Lee 89 (2002)
Non-‐Gaussian model based on sta-s-cal feedback
![Page 15: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/15.jpg)
!dPdq21
)("
#=#
!+= SdSdtdS "µ
Tsallis Distribu-on (Student-‐t)
Nonlinear Fokker-‐Planck 2
22
21
!=
"
dPd
dtdP q
qtqtZ
P !"!!= 11
2 ))()1(1()(1
#
Mathema-cal details: the non-‐Gaussian model
![Page 16: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/16.jpg)
ttttt dbqad !21
20 ]()1([ "#"#+="
)(S!=!
Ager inser-ng the expression for P -‐-‐>
a state dependent determinis-c model
Work with
as a computa-onal tool allowing us to find the solu-on
In other words:
Averaging and condi-oning w.r.t variable 0!Genraliza-on: M.Vellekoop and H. Nieuwenhuis, QF, 2007
![Page 17: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/17.jpg)
Not a perfect model of returns:
Well-‐defined star-ng price and -me
Nevertheless: Reproduces fat-‐tails and vola-lity clustering Closed form op-on-‐pricing formulae Success for op-ons and credit (CDS) pricing
![Page 18: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/18.jpg)
-0.07 -0.02 0.03 -0.05 0.00 0.05100
101
102
103
104
105
106
107
1 3 5 7time
-0.10
-0.05
0.00
0.05
0.10
1 2 3 4 5 6time
-0.01
0.03
1 2 3 4 5 6time
-0.02
-0.01
0.00
0.01
0.02
1 2 3 4 5 6time
-0.10
-0.05
0.00
0.05
SP500 q =1.5
SP500
q=1.5
Reproduces fat-‐tails and vola-lity clustering
![Page 19: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/19.jpg)
With a model of stock behavior, we want to study more complicated deriva-ve markets Some deriva-ves: -‐ Op-ons -‐ Credit default swaps
Like stocks, these are traded in large volumes, across the world How to price them? The price must be related to the underlying instrument, the stock
![Page 20: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/20.jpg)
Op-ons The right to buy or sell a stock at a certain price (the
strike) at a certain -me in the future.
In Holland in the 1600s, tulip op-ons were traded quite a bit by speculators prior to the famous tulip bubble.
1973 -‐ the Nobel-‐prize winning Black-‐Scholes op-on pricing formula.
Greek mathema-cian Thales used call op-ons on olives to make a huge profit when he foresaw a good harvest.
![Page 21: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/21.jpg)
Black-‐Scholes PDE for solving F (1973, Nobel prize 1997)
Black-‐Scholes Op-on Pricing Paradigm
the risky stock the risky op-on on S Risk-‐less porkolio: Return on must be risk-‐free rate r, due to no-‐arbitrage This leads to the famous
!
"F"t
+ rS "F"S
+12"2F"S2
# 2S2 $ rF = 0
SF(S)
!
" = S + nF
!
"
![Page 22: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/22.jpg)
Black-‐Scholes Op-on Pricing Paradigm
Equivalently: Price op-ons as expecta-ons with respect to a (risk-‐neutral) equivalent mar-ngale measure
(Whole field of mathema-cal finance based on these no-ons)
F = e!rTh(S)Q
where h = the payoff of the op-on
![Page 23: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/23.jpg)
Example European Call Op-on (F=C): the right to buy at strike K at -me T
Problem: Black-‐Scholes model must use a different for each K.
Vola-lity Smile: the plot of versus K.
!
Q
rT KTSec ]0,)(max[ != !
!
The price of the call op-on depends on S(0), K, T, r and . !
![Page 24: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/24.jpg)
Op-ons: Real-‐-me pricing and hedging
![Page 25: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/25.jpg)
Vola-lity smile in the example of previous slide
Traders intui-vely correct for the tails by bumping up the Black-‐Scholes vola-lity at far strikes
![Page 26: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/26.jpg)
Part 1: Across -me ….
1. A non-‐Gaussian model of returns
2. Op-ons pricing incorpora-ng fat-‐tails 3. Applica-ons
![Page 27: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/27.jpg)
Non-Gaussian option pricing with statistical feedback model:
There is a unique equivalent martingale measure can obtain closed-form option pricing formulae
Note: This is not the case for Levy noise or stochastic volatility
Parsimony:
Because the model incorporates tails for q = 1.4, we only need onevalue of sigma to capture all option prices across strikes
![Page 28: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/28.jpg)
1) Exploit PDE’s implied by arbitrage-‐free porkolios
Solve PDE to get op-on price
2) Convert prices of assets into mar-ngales
Take expecta-ons to get op-on price
Generaliza6on: M.Vellekoop and H. Nieuwenhuis, 2007
!
c = e"rT h(S(T)Q
?Detail
![Page 29: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/29.jpg)
Use non-‐Gaussian model as basis for stock price dynamics
!"#
$%&
'2
( + '= ) (2 T q
tT dtPrTSTS0
1)(exp)0()(*
!
" #(T)$T2
Integrate using generalized Feynman-Kac
- Approximation allows for closed-form solutions
Mar-ngale “breaks down” but valid for
12 <<T! - Due to the approximation:
?Detail
![Page 30: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/30.jpg)
})({ KTS > 21 dd T !"!à
! "# $$# =
2
1
)())((d
dTq
rT dPKTSec
Stock Price
Payoff if
European Call
q = 1: P is Gaussian q >1 : P is fat tailed Tsallis dist.
Q
rT KTSec ]0,)(max[ != !
!"#
$%&
'((2
( + '=2
2)1(exp)0()(TTTT gqrTSTS )
**
![Page 31: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/31.jpg)
T=0.6
T=0.05
Call Price Difference
S(0) = $50, r= 6%, ! =0.3
$
![Page 32: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/32.jpg)
T=0.1
T=0.4
This is very similar to the vola-li-es that traders use
Need only one sigma across all strikes if q = 1.4
Implied Black-Scholes Volatility (from q = 1.4 model with sigma = .3)
![Page 33: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/33.jpg)
Example (Japanese Yen futures)
![Page 34: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/34.jpg)
72 76 80 84Strike
9
10
11
12
13
14
Vol
Implied Volatility JY Futures 16 May 2002 T=17 days
![Page 35: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/35.jpg)
72 76 80 84Strike
9
10
11
12
13
14
Vol
Implied Volatility JY Futures 16 May 2002 T=37 days
![Page 36: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/36.jpg)
72 76 80 84Strike
9
10
11
12
13
14
Vol
Implied Volatility JY Futures 16 May 2002 T=62 days
![Page 37: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/37.jpg)
72 76 80 84Strike
9
10
11
12
13
14
Vol
Implied Volatility JY Futures 16 May 2002 T=82 days
![Page 38: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/38.jpg)
72 76 80 84Strike
9
10
11
12
13
14
Vol
Implied Volatility JY Futures 16 May 2002 T=147 days
![Page 39: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/39.jpg)
Extension to include skew (asymmetry) in the model [Borland and Bouchaud, 2004] Price crashes induce higher vola-lity than price rallies
!" # dPSq21$
Leverage effect
! Controls the skew
! Vola-lity parameter
q Controls the tail
We obtain closed-‐form formulae for call op-ons
Fluctua-ons modeled as 1!"
Pade expansion, Feynman-‐Kac
![Page 40: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/40.jpg)
Strike K Strike K
T=.03
T=0.1
T=0.2
T=0.3
T=0.55
SP500 OX q=1.4, with skew
![Page 41: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/41.jpg)
Part 1: Across -me ….
1. A non-‐Gaussian model of returns
2. Op-ons pricing incorpora-ng fat-‐tails 3. Applica-ons
![Page 42: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/42.jpg)
-‐ Can fit the model to historical distribu-on à calculate op-on prices -‐ Can imply the parameters from the op-on -‐ market
Results for stocks (top 100 stocks in S&P index) In both stock and op-on markets: -‐ converging slowly to Gaussian
-‐ -‐
3.0!"
%30!"
4.1=q
![Page 43: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/43.jpg)
Back test of trading strategy (with A. Chris-an Silva)
Exploit slight devia-ons which should converge
0
50
100
150
200
250
300
350
1 9 17 25 33 41 49 57 65 73 81 89 97 105 113 121 129 137 145 153 161 169
Series1
Live returns 2009
![Page 44: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/44.jpg)
Credit Default Swaps:
Protec-on Buyer
Protec-on Seller
Payment if the company defaults
Premium (CDS price)
CDS price depends on probability to default. We get pricing formula from the non-‐Gaussian model.
Op-ons look good … what about pricing credit?
! !qWe implied from empirical CDS prices:
4.12.1 !=q5.02.0 !="
%30=!
![Page 45: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/45.jpg)
• Key ingredient of success for statistical feedback model: long-range memory (with respect to a single time)
• Random increments are uncorrelated but not independent
Non-‐Gaussian model well describes many features of:
Stock Markets Op-on Markets
Debt and Credit Markets
![Page 46: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/46.jpg)
Mo-va-on of sta-s-cal feedback model: Intui-on -‐ traders react to extreme events Generaliza-on: Mul- -mescale sta-s-cal feedback model: In-ui-on -‐ different classes of traders react on different -me-‐scales
For example: HARCH (Muller et al) FIGARCH (Baillie et al) Mul--‐-mescale sta-s-cal feedbacl (Borland and Bouchaud)
![Page 47: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/47.jpg)
!
" t2 =" 0
2 1+ g 1l#
yt $ y(t$ l )( )2l=1
%
&'
( )
*
+ ,
!
"yt =# t"$ t
Reproduces “all” stylized facts across -me
A mul--‐-me scale non-‐Gaussian model of stock return [Borland and Bouchaud 2012]
![Page 48: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/48.jpg)
Reproduces “all” stylized facts across -me
y = log(Stock Price)
!
" t Gaussian noise uncorrelated in time
!
" t Volatility
!
g," Parameters
g controls the feedback controls the memory
!
"
A mul--‐-me scale non-‐Gaussian model of stock return [Borland and Bouchaud 2012]
Mo-va-on: Traders act on all different -me horizons
GARCH HARCH (Muller et al) FIGARCH (Baillie et al)
For t-‐ l = 0 recovers Sta-s-cal feedback model
!
" t2 =" 0
2 1+ g 1l#
yt $ y(t$ l )( )2l=1
%
&'
( )
*
+ ,
!
"yt =# t"$ t
![Page 49: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/49.jpg)
In our study: Simula-ons Some analy-c calcula-ons Calibra-on to stylized facts
Other typical calibra-on methods: -‐ Method of moment -‐ Maximum likelihood
![Page 50: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/50.jpg)
Distribu-on of Vola-lity
Mul--‐fractal scaling
Vola-lity Variogram (autocorrela-on)
Condi-onal Vola-lity
Calibra-on to stylized facts 15.1,85.0 == !z
+ US stocks o z = 0.85
Squares: =1.15 Diamonds: = 1.3 Triangles: US stocks
!!
Triangles: US stocks Lines: z = 0.6-‐0.85
vs. ln( ) )|ln( 220
2 se!!
s=1
s=0.5
s=0
s= -‐0.5
n
! n
)ln(!
)ln(P
!
!
!!
“jis_bouchaud” — 2011/12/5 — 14:52 — page 21 — #21 !!
!!
!!
On a multi-timescale statistical feedback model for volatility fluctuations 21
with ! ! .3 " q/=.q " 1/. Using these relations, one can infer, from part (a) ofFigure 5 on the facing page, the value of q that one should use for different timescalesin order to get an approximate functional form for the distribution of returns. This isuseful for option pricing (see, for example, Borland (2002)).
4.3 MultifractalityA property related to the systematic change of the distribution of returns with ` ismultifractality, which means that different moments of price changes scale as a powerof time, but with different scaling exponents. More precisely, multifractal scaling isthe following property:
Mn.`/ D hjxiC` " xi jni D An`!n ; ` # L (4.9)
where An are constants and "n is an n-dependent exponent. In the monofractal case,where the distribution is the same on all timescales up to a rescaling of the returns,then "n D n=2 $ "2. The simplest example is obviously the (geometric) Brownianmotion, for which "n D 1
2n. Any deviation from a linear behavior of "n is calledmultifractality, for which several explicit models have been proposed recently (see,for example, Lux (2001, 2012); Calvet and Fisher (2001, 2002); Muzy et al (2001);and Muzy and Bacry (2002)).
One example is the Bacry–Muzy–Delour (BMD) stochastic volatility model, whichmakes the following assumptions (Muzy et al (2001) and Muzy and Bacry (2002)):
% the log-volatilities ln #i are multivariate Gaussian variables (or, more generally,infinitely divisible (Muzy and Bacry (2002)));
% the log-volatility variogram is given by (4.5);
% the volatilities #i are independent from the (Gaussian) noises $i .
From these assumptions, we can compute exactly the moments of the return distribu-tion on different timescales. We find that these are indeed given by (4.9), with:
"n D 12nŒ1 " %2.n " 2/& (4.10)
whenever n < 1=%2, beyond which the moments are infinite. (All An can also beexactly computed (Muzy et al (2001)).) These assumptions and predictions werefound to account rather well for some aspects of empirical data.
In this section, we show that, although our model is, strictly speaking, not multifrac-tal, many of the multifractal predictions actually hold numerically quite accurately.This means that our model can account very well for apparent multifractal propertiesof financial time series, and in fact cures some of the deficiencies of standard mul-tifractal models (see Section 4.5). First, we note that our model is not multifractal,
Research Paper www.thejournalofinvestmentstrategies.com
![Page 51: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/51.jpg)
Interes-ng analogy between mul--‐-mescale models and mul--‐fractal (cascade) models
Turbulent velocity increments Financial returns
Small scales (top), larger scales (boIom)
For finance, mul--‐-mescale makes more sense because of -me-‐reversal asymmetry
![Page 52: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/52.jpg)
So far, we have studied models of stock prices across :me Now let us look at the dynamics across stocks [Correla:ons]
![Page 53: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/53.jpg)
What do we know about volatility?
- Across time for a given stock Part 1 - Across stocks at a given time Part 2
Exploring the joint stochastic process
![Page 54: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/54.jpg)
Across stocks …. 1. Cross-‐sec-onal dispersion
2. Cross-‐sec-onal kurtosis
3. Correla-ons 4. Do markets exhibit a phase-‐transi-on in -mes of panic?
5. A simple model
See also “Borland L, “Statistical signatures in times of panic: markets as a self-organizing system”, (2009)
![Page 55: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/55.jpg)
![Page 56: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/56.jpg)
![Page 57: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/57.jpg)
Even by eye you can see the very high nega-ve correla-on between cross-‐sec-onal dispersion and kurtosis
<Disp(t) Kurt(t)> = -‐ 30%
![Page 58: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/58.jpg)
In times of market panic: c-s kurtosis LOW but c-s dispersion HIGH
Defining Market Panic
kurtosis
SP 500
Some panic periods: Fall 2008, Sep 08 - April 09, 2002
![Page 59: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/59.jpg)
This implies that the distribution of cross-sectional returns is different in panic or normal states
![Page 60: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/60.jpg)
Next thing to look at are correla-ons - Perform rolling Principal Components Analysis Look at percentage of variance captured by first eigenvector
Larger percentage -> More of a “market model” -> Higher cross-sectional co-movement of stocks
![Page 61: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/61.jpg)
Distribution becomes more Gaussian in panic times
Two possible explanations:
1) Distribution of individual stock volatilities narrows
2) Correlations among random stock moves increases
![Page 62: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/62.jpg)
Distribution becomes more Gaussian in panic times
Two possible explanations:
1) Distribution of individual stock volatilities narrows
2) Correlations among random stock moves increases
![Page 63: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/63.jpg)
Define on each day the quan-ty
!
s =s" # s$
s" + s$
Where
!
s" is the number of stocks that had posi-ve moves
!
s" is the number of stocks that had nega-ve moves
![Page 64: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/64.jpg)
Correlations: Define on each day the quantity
!
s =s" # s$
s" + s$
Where
!
s" is the number of stocks that had positive moves
!
s" is the number of stocks that had negative moves
If no correlation, if correlation among stocks
!
s = 0
!
s " 0s can be seen as an Order Parameter
![Page 65: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/65.jpg)
Analogy:
s is the Order Parameter
(external volatility perception) is the Control Parameter
!
"
!
V (s) = as2 + bs4
!
s = 0
s = ±"ab
!
a = "(# c $#)
!
" >"c
!
" <"c
|s| = correlations
0 0
If
!
" <" c s is in the disordered state - low correlation across stocks
!
" >" c s is in the ordered state - high correlation Symmetry breaking (can be positive or negative)
![Page 66: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/66.jpg)
If this is plausible we expect to see - unimodal distribution of s in normal times
- bimodal distribution of s in panic times
![Page 67: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/67.jpg)
In times of panic, correlations are high ==> bimodal distribution of s
!
s =s" # s$
s" + s$
!
s" is the number of stocks that had positive moves
!
s" is the number of stocks that had negative moves
![Page 68: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/68.jpg)
-‐ Log returns follow mul- -mescale model across -me -‐ Log returns correlated across stocks with corr =|s| (disorder to order) -‐ The transi-on is triggered by a vola-lity shock We get: Kurtosis down Dispersion up Distribu-on of s bimodal
A simple phase transi-on model: Sta-s-cal signatures in -mes of panic [L. Bprland. Quan-ta-ve Finance 2011]
![Page 69: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/69.jpg)
Putting it all together
!
" ti2 =" 2 1+ g 1
l#yti $ yt$ l
i( )2
l=1
%
&'
( )
*
+ ,
!
dyti =" t
id# ti
For each stock i across time t y = log(stock price)
!
dsdt
= "as" bs3 + Ft
If : disordered state, correlations
!
a > 0" s = 0
!
a < 0" s # 0 If : ordered state, correlations <!ti!t
j >i! j= s<!t
i!tj >i! j= 0
Across stocks for each time
a =! c !!M
!
" c = critical volatility
Market vola-lity can increase due to: i) Exogenous jumps (news) ii) Endogenous jumps (mul- -me-‐scale dynamics)
![Page 70: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/70.jpg)
!
a ="c #"M
!
"c = 2 standard deviations of recent market returns
!
"M = Current market volatility
Panic: a < 0
!
"ti2 ="0
2 1+gl"2l
yti # yt# l
i( )2 +$gl" l
yti # yt# l
i( )l=1
N
%l=1
N
%&
' (
)
* +
!
dyti =" t
id# ti
Extension to include skew [Borland and Hasaad 2010
SIMULATIONS :
![Page 71: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/71.jpg)
RESULTS Histogram of s: Normal Period
Histogram of s: Panic Period
![Page 72: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/72.jpg)
Correlation -15%
![Page 73: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/73.jpg)
Summary and Open Ques-ons • A realis-c model of the joint stochas-c process of stock prices
• Dynamics across -me reproduced
• Market vola-lity drives correla-ons across stocks
• In -mes of panic, fluctua-ons become more correlated
• In -mes of panic, cross-‐sec-onal distribu-on becomes more Gaussian . All stocks experience higher vola-lity.
• All our models have been extended to incorporate the asymmetry that nega-ve returns lead to more panic than posi-ve ones.
• Calibra-on ?
• Predic-on of cri-cal sigma ?
• Analy-c solu-ons (e.q. for op-on pricing, trading strategies)?
![Page 74: Fat Tails and the Physics of !Collecve!Dynamics!in!a ...users.physik.fu-berlin.de/~pelster/Haken/Talks/Borland.pdf · useful for option pricing (see, for example, Borland (2002)).](https://reader033.fdocuments.in/reader033/viewer/2022041417/5e1c40d2e1b5ff1f7d0f749c/html5/thumbnails/74.jpg)
References Borland L, Phys. Rev.Lee 89 (2002) Borland, L. (2002). A theory of non-‐Gaussian op-on pricing. Quan:ta:ve Finance 2, 415– 431. Borland, L. (2004). A mul--‐-me scale non-‐Gaussian model of stock returns. Preprint (cond-‐ mat/0412526). Borland, L., and Bouchaud, J.-‐P. (2004). A non-‐Gaussian op-on pricing model with skew. Quan:ta:ve Finance 4, 499–514. Borland, L., Bouchaud, J.-‐P., Muzy, J.-‐F., and Zumbach, G. (2005). The dynamics of financial markets: Mandelbrot’s mul-fractal cascades, and beyond. WilmoS Magazine (March). Borland, L. (2011). Sta-s-cal signatures in -mes of panic: markets as a self-‐organizing system Quan-ta-ve Finance Borland, L. and Bouchaud, J.-‐P.. (2012) On a mul--‐-mescale sta-s-cal feedback model for vola-lity fluctua-ons. The Journal of Investment Strategies (1–40) Borland, L. & Hassid. (2010) Y. Market panic on different -me-‐scales. ArXiv e-‐prints 1010.4917 ). Preis T. , Kenee D., Stanley E.,Helbing D. and Ben-‐Jacob, W,E. Quan-fying the Behavior of Stock Correla-ons Under Market Stress . Nature, October 2012