Petit Dejeuner de la Finance

Paris, Nov 27, 2002

Empirical Aspects of Dispersion

Trading in U.S. Equity Markets
Marco Avellaneda
Courant Institute of Mathematical
Sciences, New York University
& Gargoyle Strategic Investments

What is Dispersion Trading?

Sell index option, buy options on index components

(sell correlation)

Buy index option, sell options on index components

(buy correlation)

Motivation: to profit from price differences in volatility markets

using index options and options on individual stocks
Opportunities: Market segmentation, temporary shifts in correlations
between assets, idiosyncratic news on individual stocks

Index Arbitrage versus Dispersion

Trading
Stock N

Index Arbitrage:
Reconstruct
an index product (ETF)
using the
component stocks

*
*
*

*
Index
Stock 3

Stock 2

Stock 1

Dispersion Trading:
Reconstruct an index option
using options on the
component stocks

Main U.S. indices and sectors

Major Indices: SPX, DJX, NDX
SPY, DIA, QQQ (Exchange-Traded Funds)
Sector Indices:
Semiconductors: SMH, SOX
Biotech: BBH, BTK
Pharmaceuticals: PPH, DRG
Financials: BKX, XBD, XLF, RKH
Oil & Gas: XNG, XOI, OSX
High Tech, WWW, Boxes: MSH, HHH, XBD, XCI
Retail: RTH

COMS
COMS
ADPT
ADPT
ADCT
ADCT
ADLAC
ADLAC
ADBE
ADBE
ALTR
ALTR
AMZN
AMZN
APCC
APCC
AMGN
AMGN
APOL
APOL
AAPL
AAPL
AMAT
AMAT
AMCC
AMCC
ATHM
ATHM
ATML
ATML
BBBY
BBBY
BGEN
BGEN
BMET
BMET
BMCS
BMCS
BVSN
BVSN
CHIR
CHIR
CIEN
CIEN
CTAS
CTAS
CSCO
CSCO
CTXS
CTXS

CMGI
CMGI
CNET
CNET
CMCSK
CMCSK
CPWR
CPWR
CMVT
CMVT
CEFT
CEFT
CNXT
CNXT
COST
COST
DELL
DELL
DLTR
DLTR
EBAY
EBAY
DISH
DISH
ERTS
ERTS
FISV
FISV
GMST
GMST
GENZ
GENZ
GBLX
GBLX
MLHR
MLHR
ITWO
ITWO
IMNX
IMNX
INTC
INTC
INTU
INTU
JDSU
JDSU
JNPR
JNPR
KLAC
KLAC

LGTO
LGTO
LVLT
LVLT
LLTC
LLTC
ERICY
ERICY
LCOS
LCOS
MXIM
MXIM
MCLD
MCLD
MEDI
MEDI
MFNX
MFNX
MCHP
MCHP
MSFT
MSFT
MOLX
MOLX
NTAP
NTAP
NETA
NETA
NXTL
NXTL
NXLK
NXLK
NWAC
NWAC
NOVL
NOVL
NTLI
NTLI
ORCL
ORCL
PCAR
PCAR
PHSY
PHSY
SPOT
SPOT
PMTC
PMTC
PAYX
PAYX

PSFT
PSFT
PMCS
PMCS
QLGC
QLGC
QCOM
QCOM
QTRN
QTRN
RNWK
RNWK
RFMD
RFMD
SANM
SANM
SDLI
SDLI
SEBL
SEBL
SIAL
SIAL
SSCC
SSCC
SPLS
SPLS
SBUX
SBUX
SUNW
SUNW
SNPS
SNPS
TLAB
TLAB
USAI
USAI
VRSN
VRSN
VRTS
VRTS
VTSS
VTSS
VSTR
VSTR
WCOM
WCOM
XLNX
XLNX
YHOO
YHOO

NASDAQ-100
Index (NDX)
and ETF (QQQ)
QQQ ~ 1/40 * NDX
Capitalization-weighted
QQQ trades as a stock
QQQ options: largest daily
traded volume in U.S.

Sector Exchange Traded Funds

~ 20 - 40 stocks
in same
sector
Weightings by:
capitalization
equal-dollar
equal-stock

SOX

XNG

XOI

ALTR
AMAT
AMD
INTC
KLAC
LLTC
LSCC
LSI
MOT
MU
NSM
NVLS
RMBS
TER
TXN
XLNX

APA
APC
BR
BRR
EEX
ENE
EOG
EPG
KMI
NBL
NFG
OEI
PPP
STR
WMB

AHC
BP
CHV
COC.B
XOM
KMG
OXY
P
REP
RD
SUN
TX
TOT
UCL
MRO

Index Option Arbitrage

(Dispersion Trading)
Takes advantage of differences in implied volatilities of
index options and implied volatilities of individual stock
options
Main source of arbitrage: correlations between asset prices
vary with time due to corporate events, earnings, and
``macro shocks
Full or partial index reconstruction

The trade in pictures

Index

Sell index call

Stock 1

Stock 2

Buy calls on different stocks.

Also, buy index/sell stocks

Profit-loss scenarios for a dispersion

trade in a single day
Scenario 1

Scenario 2
2.5
2
1.5

1.5

0.5

standard move

standard move

0
-0.5
-1
-1.5
-2

1
0.5
0
-0.5
-1
-1.5
-2
-2.5
-3

10

11 12 13 14 15

10

11

12

13

14

15

stock #

stock #

Stock P/L: - 2.30

Index P/L: - 0.01
Total P/L: - 2.41

Stock P/L: +9.41

Index P/L: - 0.22
Total P/L: +9.18

First approximation to hedging:

``Intrinsic Value Hedge
I=

M
i =1

wi = number of shares, scaled by ``divisor' '

wi Si

K=

j =1

max (I K ,0 )

wi Ki

M
j =1

C I (I , K , T )

M
j =1

wi max (Si Ki ,0 )

wiCi (Si , Ki , T )

IVH: use index

weights for option
hedge

IVH:
premium from index
is less than premium
from components
Super-replication
Makes sense for deep-in-the-money options

Intrinsic-Value Hedging is `exact only if

stocks are perfectly correlated
I (T ) =

M
i =1

ij 1

wi Si (T ) =

1
2

i N i i2T

i =1

wi Fi e

N i N = standardiz ed normal

K=

Solve for X in :

Similar to
Jamshidian (1989)
for pricing bond
options in 1-factor
model

1
2

i X i2T

i =1

wi Fi e
1
2

i X i2T

Ki = Fi e

Set :

max (I (T ) K ,0) =

M
i =1

wi max (Si (T ) Ki ,0 ) T

IVH : Hedge with ``equal-delta options

K i = Fie

1
2

i X T i2T

X =
X =

i T
1

i T

ln

Ki
1
+ i T
Fi
2

ln

Fi
1
i T = d2
Ki
2

N (d 2 ) = constant

log - moneyness constant

Deltas constant

What happens after you enter a trade:

Risk/return in hedged option trading

Unhedged call option

Hedged option

 

!




 

 

 

 

 


 

 


 






   
    
   


 
    
    

" # $" # %" & $" & %" ' $" ' %(") $ $" ) $ %") ) $") ) %" ) * $") * %") + $

Profit-loss for a hedged single option position (Black Scholes)

P / L (n 2 1) + NV

S
n=
,
S t

= time - decay (dollars),

NV = normalized Vega =

n ~ standardized move

Gamma P/L for an Index Option

Assume d = 0
Index Gamma P/L = I (n I2 1)
nI =

M
i =1

pi i

pi =

ni

wi Si
M
j =1

I2 =

wjS j

pi p j i j ij

ij=1

Index P/L = I

M
i =1

pi2 i2

2
I

(n

2
i

1) + I

pi p j i j
i j

I2

(n n
i

ij )

Gamma P/L for Dispersion Trade

i th stock P/L i (ni2 1)
Dispersion Trade P/L

i +

pi2 i2

I2

i =1

I (ni2 1) + I

pi p j i j
i j

I2

(n n
i

ij )

off-diagonal term:
realized cross-market
movements vs.
implied correlation

diagonal term:
realized single-stock
movements vs.
implied volatilities

Introducing the Dispersion Statistic

N

D2 =

N
i =1

P/L =

N
i =1

N
i =1

Si
I
, Y=
Si
I

pi i2 ni2 I2 n I2

i =1

Xi =

i =1

D2 =

pi ( X i Y )

N
i =1

i (ni2 1) + I (nI2 1)
i ni2 + I nI2
i ni2 +

I
I2

N
i =1

pi i2 ni2

N
i =1

I
I2

N
i =1

i + I

pi i2 ni2 + I nI2

I pi n

+ i ni2 I2 D 2
2
I
I
2 2
i i

Summary of Gamma P/L for

Dispersion Trade
I pi i2ni2

+ i ni2 I2 D 2
2
I
I

Gamma P/L =

i =1

Idiosyncratic
Gamma

Dispersion
Gamma

Time-Decay

Example: ``Pure long dispersion (zero idiosyncratic Gamma):

2

i = I

pi i2

pi i2

= I

I2

I2

1 I

pi i
i

I2

1 > 0

Payoff function for a trade

with short index/long
options (IVH), 2 stocks

30
25
20
130

15

120
110

10

100

70
130

115

120

125

100

105

110

85

90

80
95

70

75

80

90

25

Value function (B&S) for the

IVH position as a function of
stock prices (2 stocks)

20
15
10
5
130
120
110
10 0

90

70 75 80
85 90

95 100
105 110
115 120
125 130

80
70

In general: short index IVH

is short-Gamma along the
diagonal, long-Gamma for
``transversal moves

Gamma Risk: Negative exposure for parallel shifts, positive

exposure to transverse shifts
+10.84

10.31130 -2.29
125
120
115
110

1 = 30%
2 = 40%
12 = .5

105
100
95
90

5.80

85

20.49
80
75

-6.80 70

75

80

85

90

95

100

105

110

115

120

70
130

125

+7.88

Gamma-Risk for Baskets

1.E+06
8.E+05
6.E+05
4.E+05
2.E+05
0.E+00
-2.E+05
-4.E+05
-6.E+05
-8.E+05
-1.E+06
-1.E+06

I
I

Y=

pi ( X i Y )

i =1

ex

-0.15

D= Dispersion, or cross-sectional move,

D/(Y*Y)= Normalized Dispersion

D=

Si
Si

in d

-0.08

-0.01

0.06

0
0.13

0.07

0.012

1.21
norma lized
dispersion

0.3

Xi =

D /Y 2 =

pi ( X i / Y 1)

i =1

From realistic portfolio

10

Vega Risk
Sensitivity to volatility: move all single-stock implied volatilities
by the same percentage amount
Vega P/L =

Vega j j + Vega I I

j =1

M
j =1

(NV ) j

M
j =1

+ (NV )I

( NV ) j + (NV )I

NV = normalized vega =

100
95
90
85
80
75

vol % multiplier

130%

120%

110%

90%

100%

80%

70%

70

Market level

130%

105

100%

110

market level

115

20
19
18
17
16
15
14
13
12
11
10
9
8
7
6
5
4
3
2
1
0
85%

120

70
70%

125

130
125
120
115
110
105
100
95
90
85
80
75

130

115%

Market/Volatility Risk

Vol % multipler

Short Gamma on a perfectly correlated move

Monotone-increasing dependence on volatility (IVH)

11

``Rega: Sensitivity to correlation

ij ij +
I2

M
ij =1

i j

pi p j i j ij +

i j

pi p j i j

I2 = ( I(1) ) ( I( 0 ) ) ,
2

I(1) =

j =1

p j j ,

I(0 ) =

j =1

p 2j 2j

1 ( I(1) ) ( I(0 ) )

2
I2
2

(1)
( 0)
1
(NV )I ( I ) 2( I )
2
I
2

Correlatio n P/L =

1 ( I(1 ) ) ( I(0 ) )
2
I2
2

Rega =

(NV )I

Market/Correlation Sensitivity
130
125

5.1
4.8
4. 5
4. 2
3.9
3.6
3.3
3
2.7
2.4
2.1
1.8
1.5
1.2
0.9
0. 6
0. 3
0

120
115
110
105
100

market level

95
90
85

130

80

corr change

75
70

0.3

0.2

0.1

-0.1

-0.3

market level
-0.2

0. 2

70
0. 3

0.1

90
-0.1

-0. 3

-0.2

110

corr change

Short Gamma on a perfectly correlated move

Monotone-decreasing dependence on correlation

12

Entering a trade

Valuation Method I: Weighted

Monte Carlo
Simulate scenarios (paths) for the group of stocks that comprise
the index or indices under consideration
Simulate the cash-flows of options on all the stocks and the
index options
Select weights or probabilities on the scenarios in such a way
that all options/forward prices are correctly reproduced by
averaging over the paths
Use ``weighted Monte Carlo to derive fair-value of target
options and compare with market values

Avellaneda, Buff, Friedman, Kruk, Grandchamp: IJTAF, 1999

dX = dW + B dt

time

13

Avellaneda, Buff, Friedman, Kruk, Grandchamp: IJTAF, 1999

dX = dW + B dt

p1
p2

p3
time

Computation of weights:
Max-Entropy Method

Market prices
of single-stock
options

Risk-neutral
pricing probabilities

cash-flow matrix

14

Example of Pricing with WMC

Index Market Vols vs. Model Vols : January 03 expiration
60.00
50.00

impliedvol

40.00

BidVol
AskVol
ModelVol
RHO=1

30.00

20.00
10.00
0.00

360

380

400

420

440

460

Index Strike Price

Another Valuation Example with

WMC (From Aug 2002, front month)

Vol

Implied vol Expiration Sep02

40
35
30
25
20
15
10
5

Bid
Ask
Model

0
440 445 450 455 460 465 470 475 480 485 490 495 500 505

Index Strike

15

Another Valuation Example with

WMC (From Aug 2002, second month)
40
35
30
25
20
15
10
5
0

Bid
Ask

0
52

51

0
50

0
49

0
48

0
47

0
46

0
45

44

43

Model

Vol

Implied vol Expiration Oct02

Index Strike

Another Valuation Example with

WMC (From Aug 2002, third month)
Implied vol Expiration Nov02
35
30

Vol

25
Bid

20

Ask

15

Model

10
5
0
430

440

450

460

470

480

490

500

510

520

530

Index Strike

16

Another Valuation Example with

WMC (From Aug 2002, 4 month)
th

Implied vol Expiration Dec02

35
30

Vol

25
Bid
Ask

20
15

Model

10
5
0
420

440

460

470

480

490

500

510

520

530

540

Index Strike

Valuation Method II: (WKB)

Steepest-Descent Approximation
(Avellaneda, Boyer-Olson, Busca, Friz: RISK 2002, C.R.A.S. Paris 2003)

Improvement on Standard Volatility Formula for Index Options

I2 =

N
j =1

p 2j 2j +

i j

pi p j i j ij

(*)

Assume that the correlation is given

Use markets on single-stock volatilities taking into account
volatility skew
How can we integrate volatility skew information into (*)?

17

Steepest-Descent Approximation
Define a risk-neutral 1-factor model
for the index process

dI
= I (I , t )dW + I (I , t )dt
I

Local index vol= conditional expectation of local variance (rigorous)

I2 (I , t ) = E

j (S j (t ), t ) k (Sk (t ), t ) jk p j pk

jk =1

w j S j (t ) = I

j =1

Approximate this conditional expectation using the most

wi Si (t ) = I
likely stock configuration (S1* ,..., S N* ) given that
i

I2 (I , t )

N
ij=1

pi p j Si*S *j i (Si* , t ) j (S *j , t )

Steepest descent vs. Market vs.

WMC (Aug 20, 2002, front month)
Expiration: Sep 02
40

BidVol
AskVol
W MC vol
Steepest Desc

30
25
20

49
5
50
0
50
5

49
0

48
5

48
0

47
5

47
0

46
5

46
0

45
5

45
0

44
5

15
44
0

implied vol

35

strike

18

Steepest descent vs. Market vs.

WMC (Aug 20, 2002, 2nd month)
Expiration: Nov 02
40

BidVol
AskVol
W MC vol
Steepest Desc

30
25
20

0
52

0
51

50

0
49

0
48

0
47

0
46

45

0
44

15

43

implied vol

35

strike

Gargoyle Dispersion Fund

Joint venture between Gargoyle Strategic Partners and
Marco Avellaneda (manager)
Started Trading: May 2001
Uses proprietary system to detect trades and executes
electronically and through network of brokers in 5 U.S.
exchanges
1 FT junior trader, 3 PT senior traders, 1 FT risk manager

19

ROI May01-Oct02
$1.65
$1.60
$1.55
$1.50
$1.45
$1.40
$1.35
$1.30
$1.25
$1.20
$1.15
$1.10
$1.05
$1.00
$0.95
$0.90
$0.85
$0.80
$0.75
$0.70
$0.65
$0.60
$0.55
$0.50

Gargoyle
Dispersion
Fund

-0
2
ct
O

02

p02

gAu

Se

-0
2

l -0
2
Ju

Ju
n

02

02
M

ay
-

-0
2

pr
A

ar
M

2
Fe
b02

c01

Ja
n0

De

-0
1

v01

ct
O

No

01

-0
1

g-

Au

ep

l -0
1

-0
1

Ju

ay
M

Ju
n

01

$1

Trading History: Monthly Returns

8.80%

O ct-02

-10.87%

Se p-0 2
Au g- 0 2

0.66%

-16.17%
-7.79%
-8.49%
-7.12%

J u l- 0 2
Ju n - 0 2

5.20%
-0.74%
-2.04%

Ma y- 0 2

-6.06%

A pr- 0 2
M a r- 0 2
Fe b-0 2

N o v- 0 1

3.27%
3.76%

-1.02%
-1.93%

S&P 500

6.09%

-1.46%

Ja n - 0 2
Dec-01

0.49%
0.88%

Ga rgoyle
Dispe rsion Fund

3.78%
7.67%

1.90%

O c t- 0 1

-8.07%

S e p-0 1

J u l- 0 1

1.82%
-0.98%

-7.56%

-2.43%

Jun- 0 1
M a y- 0 1

10.10%
0.67%

-1.38%
-15%

-10%

13.97%

9.18%

3.58%

-6.26%

Aug-0 1

-20%

12.54%

-3.17%

-5%

0%

5%

10%

15%

20%

20

Dispersion Fund Performance

Trading Period: 15 months
Cumulative ROI* since inception: 28.33%
Annualized Rate of Return: 22.65%
Annualized Standard Deviation: 26.59%
Worst monthly loss: August 02, -16%
Correlation with S&P 500:

35%

Correlation with VIX Index: - 33%

* After paying brokerage fees and commissions, etc

Volatility

Dow Industrial
Average (DJX)

Correlation
60%

Average Corr
Weighted Corr

50%
40%
30%
20%
10%
0%
De c

Ja n

Fe b

Ma r

Apr

Ma y

Jun

Jul

Aug

Se p

Oct

Nov

21

Volatility

Amex Biotechnology Index

(BTK)

Correlation

0.8
0.7
0.6
0.5
0.4

Average Corr

0.3

Weighted Corr

0.2
0.1
0
Dec

Ja n

Fe b

Mar

Apr

May

Jun

Jul

Aug

Se p

Oct

Nov

DJX Correlation Blowout, July 2002

DJX expiration 9/ 21/ 2002 strike 86

DJX Sep 86 Call

1.2

90

80
1
70

60

ImpliedCorr
Delta

50
0.6
40

BidRho
AskRho
Delta

30

0.4

20
0.2
10

0
02
/2
00
2
8/
2/
20
02
8/
4/
20
02
8/
6/
20
02
8/
8/
20
0
8/
10 2
/2
00
8/
12 2
/2
00
8/
14 2
/2
00
8/
16 2
/2
00
8/
18 2
/2
00
8/
20 2
/2
00
8/
22 2
/2
00
8/
24 2
/2
00
8/
26 2
/2
00
8/
28 2
/2
00
8/
30 2
/2
00
2

02

/2
0

7/
31

02

/2
0

7/
29

02

/2
0

7/
27

02

/2
0

7/
25

02

/2
0

7/
23

02

/2
0

7/
21

02

/2
0

7/
19

02

/2
0

7/
17

/2
0

7/
15

7/
13

/2
0

02

7/
11

Correlation

0.8

22

Conclusions
Dispersion trading: a form of ``statistical correlation arbitrage
Sell correlation by selling index options and buying options
on the components
Buy correlation by buying index options and selling options
on the components
``Convergence trading style.
Price discovery using model and market data on vol skews
Sophisticated trading strategy. Potentially very profitable,
with moderate (but not low) risk profile.

23