1. Category - Based Tail Comovement
Christophe Villa
AUDENCIA - Nantes and CREST
February, 25 2010
Christophe Villa () EM Lyon February, 25 2010 1 / 22
2. Co-authors
Arthur Charpentier
University of Rennes and Ecole Polytechnique
Emilios Galariotis
Audencia Nantes School of Management
Christophe Villa () EM Lyon February, 25 2010 2 / 22
3. Comovement
there are numerous patterns of comovement in the data
Christophe Villa () EM Lyon February, 25 2010 3 / 22
4. Comovement
there are numerous patterns of comovement in the data
common factors in the returns of certain groups of assets
Christophe Villa () EM Lyon February, 25 2010 3 / 22
5. Comovement
there are numerous patterns of comovement in the data
common factors in the returns of certain groups of assets
stocks within the same industry
Christophe Villa () EM Lyon February, 25 2010 3 / 22
6. Comovement
there are numerous patterns of comovement in the data
common factors in the returns of certain groups of assets
stocks within the same industry
small stocks
Christophe Villa () EM Lyon February, 25 2010 3 / 22
7. Comovement
there are numerous patterns of comovement in the data
common factors in the returns of certain groups of assets
stocks within the same industry
small stocks
value stocks
Christophe Villa () EM Lyon February, 25 2010 3 / 22
8. Comovement
there are numerous patterns of comovement in the data
common factors in the returns of certain groups of assets
stocks within the same industry
small stocks
value stocks
closed-end funds
Christophe Villa () EM Lyon February, 25 2010 3 / 22
9. Comovement
there are numerous patterns of comovement in the data
common factors in the returns of certain groups of assets
stocks within the same industry
small stocks
value stocks
closed-end funds
what is the source of this comovement ?
Christophe Villa () EM Lyon February, 25 2010 3 / 22
10. Comovement
there are numerous patterns of comovement in the data
common factors in the returns of certain groups of assets
stocks within the same industry
small stocks
value stocks
closed-end funds
what is the source of this comovement ?
why do certain assets comove while others do not ?
Christophe Villa () EM Lyon February, 25 2010 3 / 22
11. Traditional view : Fundamentals - based comovement
derived from economies without frictions and with rational investors
Christophe Villa () EM Lyon February, 25 2010 4 / 22
12. Traditional view : Fundamentals - based comovement
derived from economies without frictions and with rational investors
assets comove because their "fundamental values" comove
Christophe Villa () EM Lyon February, 25 2010 4 / 22
13. Traditional view : Fundamentals - based comovement
derived from economies without frictions and with rational investors
assets comove because their "fundamental values" comove
fundamental value = rational forecast of future cash ‡ows discounted
at rate appropriate for risk
Christophe Villa () EM Lyon February, 25 2010 4 / 22
14. Traditional view : Fundamentals - based comovement
derived from economies without frictions and with rational investors
assets comove because their "fundamental values" comove
fundamental value = rational forecast of future cash ‡ows discounted
at rate appropriate for risk
in this view, assets comove because of :
Christophe Villa () EM Lyon February, 25 2010 4 / 22
15. Traditional view : Fundamentals - based comovement
derived from economies without frictions and with rational investors
assets comove because their "fundamental values" comove
fundamental value = rational forecast of future cash ‡ows discounted
at rate appropriate for risk
in this view, assets comove because of :
correlated news about their cash ‡ows
Christophe Villa () EM Lyon February, 25 2010 4 / 22
16. Traditional view : Fundamentals - based comovement
derived from economies without frictions and with rational investors
assets comove because their "fundamental values" comove
fundamental value = rational forecast of future cash ‡ows discounted
at rate appropriate for risk
in this view, assets comove because of :
correlated news about their cash ‡ows
correlated changes in their discount rates
Christophe Villa () EM Lyon February, 25 2010 4 / 22
17. Traditional view : Fundamentals - based comovement
derived from economies without frictions and with rational investors
assets comove because their "fundamental values" comove
fundamental value = rational forecast of future cash ‡ows discounted
at rate appropriate for risk
in this view, assets comove because of :
correlated news about their cash ‡ows
correlated changes in their discount rates
changes in interest rates
Christophe Villa () EM Lyon February, 25 2010 4 / 22
18. Traditional view : Fundamentals - based comovement
derived from economies without frictions and with rational investors
assets comove because their "fundamental values" comove
fundamental value = rational forecast of future cash ‡ows discounted
at rate appropriate for risk
in this view, assets comove because of :
correlated news about their cash ‡ows
correlated changes in their discount rates
changes in interest rates
changes in risk aversion
Christophe Villa () EM Lyon February, 25 2010 4 / 22
19. Traditional view : Fundamentals - based comovement
derived from economies without frictions and with rational investors
assets comove because their "fundamental values" comove
fundamental value = rational forecast of future cash ‡ows discounted
at rate appropriate for risk
in this view, assets comove because of :
correlated news about their cash ‡ows
correlated changes in their discount rates
changes in interest rates
changes in risk aversion
correlated changes in rational perception of risk
Christophe Villa () EM Lyon February, 25 2010 4 / 22
20. Traditional view : Fundamentals - based comovement
derived from economies without frictions and with rational investors
assets comove because their "fundamental values" comove
fundamental value = rational forecast of future cash ‡ows discounted
at rate appropriate for risk
in this view, assets comove because of :
correlated news about their cash ‡ows
correlated changes in their discount rates
changes in interest rates
changes in risk aversion
correlated changes in rational perception of risk
useful framework for understanding many types of comovement
Christophe Villa () EM Lyon February, 25 2010 4 / 22
21. Evidence on comovement
Twin stocks (Froot/Dabora [1999])
Christophe Villa () EM Lyon February, 25 2010 5 / 22
22. Evidence on comovement
Twin stocks (Froot/Dabora [1999])
claims to same cash-‡ow stream, but traded in di¤erent locations (eg
Royal Dutch / Shell)
Christophe Villa () EM Lyon February, 25 2010 5 / 22
23. Evidence on comovement
Twin stocks (Froot/Dabora [1999])
claims to same cash-‡ow stream, but traded in di¤erent locations (eg
Royal Dutch / Shell)
Royal Dutch, traded primarily in New York, is a claim to 60% of the
cash ‡ow
Christophe Villa () EM Lyon February, 25 2010 5 / 22
24. Evidence on comovement
Twin stocks (Froot/Dabora [1999])
claims to same cash-‡ow stream, but traded in di¤erent locations (eg
Royal Dutch / Shell)
Royal Dutch, traded primarily in New York, is a claim to 60% of the
cash ‡ow
Shell, traded primarily in London, is a claim to the remaining 40%
Christophe Villa () EM Lyon February, 25 2010 5 / 22
25. Evidence on comovement
Twin stocks (Froot/Dabora [1999])
claims to same cash-‡ow stream, but traded in di¤erent locations (eg
Royal Dutch / Shell)
Royal Dutch, traded primarily in New York, is a claim to 60% of the
cash ‡ow
Shell, traded primarily in London, is a claim to the remaining 40%
under traditional view of comovement, expect them to move in lockstep
Christophe Villa () EM Lyon February, 25 2010 5 / 22
26. Evidence on comovement
Twin stocks (Froot/Dabora [1999])
claims to same cash-‡ow stream, but traded in di¤erent locations (eg
Royal Dutch / Shell)
Royal Dutch, traded primarily in New York, is a claim to 60% of the
cash ‡ow
Shell, traded primarily in London, is a claim to the remaining 40%
under traditional view of comovement, expect them to move in lockstep
in fact, Royal Dutch comoves more with the U.S. stock market, Shell
with the U.K. market
rRD,t rSH,t = α + 0.207 rS&P,t 0.428 rFTSE ,t + εt
Christophe Villa () EM Lyon February, 25 2010 5 / 22
27. Evidence on comovement ctd.
Closed-end country funds (Hardouvelis et al. [1994], Bodurtha et al.
[1995])
Christophe Villa () EM Lyon February, 25 2010 6 / 22
28. Evidence on comovement ctd.
Closed-end country funds (Hardouvelis et al. [1994], Bodurtha et al.
[1995])
funds traded in one location, fund assets in another
Christophe Villa () EM Lyon February, 25 2010 6 / 22
29. Evidence on comovement ctd.
Closed-end country funds (Hardouvelis et al. [1994], Bodurtha et al.
[1995])
funds traded in one location, fund assets in another
under traditional view of comovement, expect fund returns and NAV
returns to move together closely
Christophe Villa () EM Lyon February, 25 2010 6 / 22
30. Evidence on comovement ctd.
Closed-end country funds (Hardouvelis et al. [1994], Bodurtha et al.
[1995])
funds traded in one location, fund assets in another
under traditional view of comovement, expect fund returns and NAV
returns to move together closely
in fact, fund returns comove as much with market where fund is traded
as with market where assets are traded
Christophe Villa () EM Lyon February, 25 2010 6 / 22
31. Evidence on comovement ctd.
Closed-end country funds (Hardouvelis et al. [1994], Bodurtha et al.
[1995])
funds traded in one location, fund assets in another
under traditional view of comovement, expect fund returns and NAV
returns to move together closely
in fact, fund returns comove as much with market where fund is traded
as with market where assets are traded
Domestic closed-end funds (Lee et al. [1991])
Christophe Villa () EM Lyon February, 25 2010 6 / 22
32. Evidence on comovement ctd.
Closed-end country funds (Hardouvelis et al. [1994], Bodurtha et al.
[1995])
funds traded in one location, fund assets in another
under traditional view of comovement, expect fund returns and NAV
returns to move together closely
in fact, fund returns comove as much with market where fund is traded
as with market where assets are traded
Domestic closed-end funds (Lee et al. [1991])
closed-end funds invested in large stocks often comove with small
stocks
Christophe Villa () EM Lyon February, 25 2010 6 / 22
33. Evidence on comovement ctd.
Small stocks and value stocks (Fama/French [1995])
Christophe Villa () EM Lyon February, 25 2010 7 / 22
34. Evidence on comovement ctd.
Small stocks and value stocks (Fama/French [1995])
there is a strong common factor in returns of small stocks and in
returns of value stocks (Fama/French [1993])
Christophe Villa () EM Lyon February, 25 2010 7 / 22
35. Evidence on comovement ctd.
Small stocks and value stocks (Fama/French [1995])
there is a strong common factor in returns of small stocks and in
returns of value stocks (Fama/French [1993])
Fama/French [1995] test whether these factors are due to cash-‡ow
news
Christophe Villa () EM Lyon February, 25 2010 7 / 22
36. Evidence on comovement ctd.
Small stocks and value stocks (Fama/French [1995])
there is a strong common factor in returns of small stocks and in
returns of value stocks (Fama/French [1993])
Fama/French [1995] test whether these factors are due to cash-‡ow
news
do …nd cash-‡ow factors but they line up poorly with the return factors
Christophe Villa () EM Lyon February, 25 2010 7 / 22
37. Evidence on comovement ctd.
Small stocks and value stocks (Fama/French [1995])
there is a strong common factor in returns of small stocks and in
returns of value stocks (Fama/French [1993])
Fama/French [1995] test whether these factors are due to cash-‡ow
news
do …nd cash-‡ow factors but they line up poorly with the return factors
Commodities (Pindyck/Rotemberg [1990])
Christophe Villa () EM Lyon February, 25 2010 7 / 22
38. Evidence on comovement ctd.
Small stocks and value stocks (Fama/French [1995])
there is a strong common factor in returns of small stocks and in
returns of value stocks (Fama/French [1993])
Fama/French [1995] test whether these factors are due to cash-‡ow
news
do …nd cash-‡ow factors but they line up poorly with the return factors
Commodities (Pindyck/Rotemberg [1990])
…nd strong comovement in price changes of seven commodities
Christophe Villa () EM Lyon February, 25 2010 7 / 22
39. Evidence on comovement ctd.
Small stocks and value stocks (Fama/French [1995])
there is a strong common factor in returns of small stocks and in
returns of value stocks (Fama/French [1993])
Fama/French [1995] test whether these factors are due to cash-‡ow
news
do …nd cash-‡ow factors but they line up poorly with the return factors
Commodities (Pindyck/Rotemberg [1990])
…nd strong comovement in price changes of seven commodities
hard to explain comovement through news about aggregate demand
Christophe Villa () EM Lyon February, 25 2010 7 / 22
43. Category-based comovement
Barberis/Shleifer [2003]
many investors allocate funds at the level of an asset category
simpli…es the portfolio problem
makes it easier to evaluate money managers
Christophe Villa () EM Lyon February, 25 2010 8 / 22
44. Category-based comovement
Barberis/Shleifer [2003]
many investors allocate funds at the level of an asset category
simpli…es the portfolio problem
makes it easier to evaluate money managers
categories are based on a salient common characteristic
Christophe Villa () EM Lyon February, 25 2010 8 / 22
45. Category-based comovement
Barberis/Shleifer [2003]
many investors allocate funds at the level of an asset category
simpli…es the portfolio problem
makes it easier to evaluate money managers
categories are based on a salient common characteristic
impressive past performance often spurs category formation
Christophe Villa () EM Lyon February, 25 2010 8 / 22
46. Category-based comovement
Barberis/Shleifer [2003]
many investors allocate funds at the level of an asset category
simpli…es the portfolio problem
makes it easier to evaluate money managers
categories are based on a salient common characteristic
impressive past performance often spurs category formation
categories can be identi…ed by looking at labels on money managers’
products
Christophe Villa () EM Lyon February, 25 2010 8 / 22
47. Category-based comovement
Barberis/Shleifer [2003]
many investors allocate funds at the level of an asset category
simpli…es the portfolio problem
makes it easier to evaluate money managers
categories are based on a salient common characteristic
impressive past performance often spurs category formation
categories can be identi…ed by looking at labels on money managers’
products
e.g. small-cap, large-cap, growth, index
Christophe Villa () EM Lyon February, 25 2010 8 / 22
48. Category-based comovement ctd.
suppose categories are adopted by noise traders with correlated
sentiment
Christophe Villa () EM Lyon February, 25 2010 9 / 22
49. Category-based comovement ctd.
suppose categories are adopted by noise traders with correlated
sentiment
if the noise traders a¤ect prices
) assets will comove simply because they are classi…ed into the same
category
Christophe Villa () EM Lyon February, 25 2010 9 / 22
50. Category-based comovement ctd.
suppose categories are adopted by noise traders with correlated
sentiment
if the noise traders a¤ect prices
) assets will comove simply because they are classi…ed into the same
category
even if fundamentals are uncorrelated
Christophe Villa () EM Lyon February, 25 2010 9 / 22
52. Category-based comovement ctd.
Consider an simple economy with a riskless asset and 2n risky assets
risky asset i is a claim to a single liquidating dividend Di,T to be paid
at some later time T:
Di,T = Di,0 + ei,1 + ... + ei,T
where
et = (e1,t , ..., e2n,t )0
~ N (0, ΣD ) , i.i.d. over time
Christophe Villa () EM Lyon February, 25 2010 10 / 22
53. Category-based comovement ctd.
Consider an simple economy with a riskless asset and 2n risky assets
risky asset i is a claim to a single liquidating dividend Di,T to be paid
at some later time T:
Di,T = Di,0 + ei,1 + ... + ei,T
where
et = (e1,t , ..., e2n,t )0
~ N (0, ΣD ) , i.i.d. over time
Denote the price of asset i at time t as Pi,t and the return as
∆Pi,t = Pi,t Pi,t 1
Christophe Villa () EM Lyon February, 25 2010 10 / 22
54. Category-based comovement ctd.
Consider an simple economy with a riskless asset and 2n risky assets
risky asset i is a claim to a single liquidating dividend Di,T to be paid
at some later time T:
Di,T = Di,0 + ei,1 + ... + ei,T
where
et = (e1,t , ..., e2n,t )0
~ N (0, ΣD ) , i.i.d. over time
Denote the price of asset i at time t as Pi,t and the return as
∆Pi,t = Pi,t Pi,t 1
Some investors group the risky assets into two categories, X and Y
Christophe Villa () EM Lyon February, 25 2010 10 / 22
55. Category-based comovement ctd.
Consider an simple economy with a riskless asset and 2n risky assets
risky asset i is a claim to a single liquidating dividend Di,T to be paid
at some later time T:
Di,T = Di,0 + ei,1 + ... + ei,T
where
et = (e1,t , ..., e2n,t )0
~ N (0, ΣD ) , i.i.d. over time
Denote the price of asset i at time t as Pi,t and the return as
∆Pi,t = Pi,t Pi,t 1
Some investors group the risky assets into two categories, X and Y
suppose assets 1 through n in category X, assets n + 1 through 2n in Y
Christophe Villa () EM Lyon February, 25 2010 10 / 22
57. Category-based comovement ctd.
assume the following cash-‡ow structure:
for an asset i 2 X
ei,t = ψM fM,t + ψS fX ,t +
q
1 ψ2
S ψ2
M εi,t
Christophe Villa () EM Lyon February, 25 2010 11 / 22
58. Category-based comovement ctd.
assume the following cash-‡ow structure:
for an asset i 2 X
ei,t = ψM fM,t + ψS fX ,t +
q
1 ψ2
S ψ2
M εi,t
for an asset j 2 Y
ej,t = ψM fM,t + ψS fY ,t +
q
1 ψ2
S ψ2
M εj,t
Christophe Villa () EM Lyon February, 25 2010 11 / 22
59. Category-based comovement ctd.
assume the following cash-‡ow structure:
for an asset i 2 X
ei,t = ψM fM,t + ψS fX ,t +
q
1 ψ2
S ψ2
M εi,t
for an asset j 2 Y
ej,t = ψM fM,t + ψS fY ,t +
q
1 ψ2
S ψ2
M εj,t
fM,t is the market-wide shock
Christophe Villa () EM Lyon February, 25 2010 11 / 22
60. Category-based comovement ctd.
assume the following cash-‡ow structure:
for an asset i 2 X
ei,t = ψM fM,t + ψS fX ,t +
q
1 ψ2
S ψ2
M εi,t
for an asset j 2 Y
ej,t = ψM fM,t + ψS fY ,t +
q
1 ψ2
S ψ2
M εj,t
fM,t is the market-wide shock
fX ,t and fY ,t are the group-speci…c shocks
Christophe Villa () EM Lyon February, 25 2010 11 / 22
61. Category-based comovement ctd.
assume the following cash-‡ow structure:
for an asset i 2 X
ei,t = ψM fM,t + ψS fX ,t +
q
1 ψ2
S ψ2
M εi,t
for an asset j 2 Y
ej,t = ψM fM,t + ψS fY ,t +
q
1 ψ2
S ψ2
M εj,t
fM,t is the market-wide shock
fX ,t and fY ,t are the group-speci…c shocks
εi,t and εj,t are idiosyncratic shocks
Christophe Villa () EM Lyon February, 25 2010 11 / 22
62. Category-based comovement ctd.
assume the following cash-‡ow structure:
for an asset i 2 X
ei,t = ψM fM,t + ψS fX ,t +
q
1 ψ2
S ψ2
M εi,t
for an asset j 2 Y
ej,t = ψM fM,t + ψS fY ,t +
q
1 ψ2
S ψ2
M εj,t
fM,t is the market-wide shock
fX ,t and fY ,t are the group-speci…c shocks
εi,t and εj,t are idiosyncratic shocks
ψM and ψS are constants that control the relative importance of the
three components
Christophe Villa () EM Lyon February, 25 2010 11 / 22
63. Category-based comovement ctd.
assume the following cash-‡ow structure:
for an asset i 2 X
ei,t = ψM fM,t + ψS fX ,t +
q
1 ψ2
S ψ2
M εi,t
for an asset j 2 Y
ej,t = ψM fM,t + ψS fY ,t +
q
1 ψ2
S ψ2
M εj,t
fM,t is the market-wide shock
fX ,t and fY ,t are the group-speci…c shocks
εi,t and εj,t are idiosyncratic shocks
ψM and ψS are constants that control the relative importance of the
three components
Each shock has unit variance and is orthogonal to the other shocks
Christophe Villa () EM Lyon February, 25 2010 11 / 22
65. Category-based comovement ctd.
Suppose that
category-based investors are noise traders, who channel funds in and
out of the categories depending on their sentiment
Christophe Villa () EM Lyon February, 25 2010 12 / 22
66. Category-based comovement ctd.
Suppose that
category-based investors are noise traders, who channel funds in and
out of the categories depending on their sentiment
the economy also contains "fundamental traders" who invest at the
level of individual assets and act as arbitrageurs
Christophe Villa () EM Lyon February, 25 2010 12 / 22
67. Category-based comovement ctd.
Suppose that
category-based investors are noise traders, who channel funds in and
out of the categories depending on their sentiment
the economy also contains "fundamental traders" who invest at the
level of individual assets and act as arbitrageurs
A simple representation for asset returns is then
∆Pi,t = ei,t + ∆uX ,t , for i = 1, ..., n
∆Pi,t = ei,t + ∆uY ,t , for i = n + 1, ..., 2n
where
uX ,t
uY ,t
N
0
0
, σ2
u
1 ρu
ρu 1
i.i.d. over time
Christophe Villa () EM Lyon February, 25 2010 12 / 22
68. Category-based comovement ctd.
Suppose that
category-based investors are noise traders, who channel funds in and
out of the categories depending on their sentiment
the economy also contains "fundamental traders" who invest at the
level of individual assets and act as arbitrageurs
A simple representation for asset returns is then
∆Pi,t = ei,t + ∆uX ,t , for i = 1, ..., n
∆Pi,t = ei,t + ∆uY ,t , for i = n + 1, ..., 2n
where
uX ,t
uY ,t
N
0
0
, σ2
u
1 ρu
ρu 1
i.i.d. over time
uX ,t (uY ,t ) can be thought of as time t noise trader sentiment about
the securities in category X (Y )
Christophe Villa () EM Lyon February, 25 2010 12 / 22
69. Category-based comovement ctd.
so long as arbitrage is limited in some way, two assets in same
category comove not only because of correlated cash-‡ow news, but
because of a correlated sentiment shock
Christophe Villa () EM Lyon February, 25 2010 13 / 22
70. Category-based comovement ctd.
so long as arbitrage is limited in some way, two assets in same
category comove not only because of correlated cash-‡ow news, but
because of a correlated sentiment shock
assets in the same category comove too much
Christophe Villa () EM Lyon February, 25 2010 13 / 22
71. Category-based comovement ctd.
so long as arbitrage is limited in some way, two assets in same
category comove not only because of correlated cash-‡ow news, but
because of a correlated sentiment shock
assets in the same category comove too much
assets in di¤erent categories comove too little
Christophe Villa () EM Lyon February, 25 2010 13 / 22
72. Category-based comovement ctd.
so long as arbitrage is limited in some way, two assets in same
category comove not only because of correlated cash-‡ow news, but
because of a correlated sentiment shock
assets in the same category comove too much
assets in di¤erent categories comove too little
and reclassifying an asset into a new style raises its correlation with
that category
Christophe Villa () EM Lyon February, 25 2010 13 / 22
73. Category-based comovement ctd.
so long as arbitrage is limited in some way, two assets in same
category comove not only because of correlated cash-‡ow news, but
because of a correlated sentiment shock
assets in the same category comove too much
assets in di¤erent categories comove too little
and reclassifying an asset into a new style raises its correlation with
that category
category-based view suggests that there is excess comovement in
stock prices, with stocks in the same category moving together more
than their intrinsic values say that they should.
Christophe Villa () EM Lyon February, 25 2010 13 / 22
74. Category-based comovement ctd.
so long as arbitrage is limited in some way, two assets in same
category comove not only because of correlated cash-‡ow news, but
because of a correlated sentiment shock
assets in the same category comove too much
assets in di¤erent categories comove too little
and reclassifying an asset into a new style raises its correlation with
that category
category-based view suggests that there is excess comovement in
stock prices, with stocks in the same category moving together more
than their intrinsic values say that they should.
comovement = correlation
Christophe Villa () EM Lyon February, 25 2010 13 / 22
75. Evidence on extreme comovement
Mr. A. Greenspan: "work that characterizes the distribution of
extreme events would be useful as well."
Christophe Villa () EM Lyon February, 25 2010 14 / 22
76. Evidence on extreme comovement
Mr. A. Greenspan: "work that characterizes the distribution of
extreme events would be useful as well."
During the unfolding …nancial crisis stocks have experienced both
extreme movement and extreme comovement in returns.
) Category-based tail comovement
Christophe Villa () EM Lyon February, 25 2010 14 / 22
77. Evidence on extreme comovement
No simple analogue of the correlation coe¢ cient for extreme
dependencies ) various statistical measures.
Christophe Villa () EM Lyon February, 25 2010 15 / 22
78. Evidence on extreme comovement
No simple analogue of the correlation coe¢ cient for extreme
dependencies ) various statistical measures.
Poon, Rockinger and Tawn (2004) derive a general multivariate
framework with two types of extreme dependence structures that
allow for both asymptotic dependence and independence.
Christophe Villa () EM Lyon February, 25 2010 15 / 22
79. Evidence on extreme comovement
No simple analogue of the correlation coe¢ cient for extreme
dependencies ) various statistical measures.
Poon, Rockinger and Tawn (2004) derive a general multivariate
framework with two types of extreme dependence structures that
allow for both asymptotic dependence and independence.
four types of dependence: perfect dependent, independent,
asymptotically dependent and asymptotically independent
Christophe Villa () EM Lyon February, 25 2010 15 / 22
80. Quantile dependence
Quantile dependence is a measure of the dependence in the tails of
the distribution and describes the limiting proportion that
Christophe Villa () EM Lyon February, 25 2010 16 / 22
81. Quantile dependence
Quantile dependence is a measure of the dependence in the tails of
the distribution and describes the limiting proportion that
one margin exceeds a certain threshold
Christophe Villa () EM Lyon February, 25 2010 16 / 22
82. Quantile dependence
Quantile dependence is a measure of the dependence in the tails of
the distribution and describes the limiting proportion that
one margin exceeds a certain threshold
given that the other margin has already exceeded that threshold.
Christophe Villa () EM Lyon February, 25 2010 16 / 22
83. Quantile dependence
Quantile dependence is a measure of the dependence in the tails of
the distribution and describes the limiting proportion that
one margin exceeds a certain threshold
given that the other margin has already exceeded that threshold.
If Z1 and Z2 are random variables with distribution functions F1 and
F2,
) there is quantile dependence in the lower tail at threshold α,
whenever
P Z2 F 1
2 (α) j Z1 F 1
1 (α) 6= 0
) there is quantile dependence in the upper tail at threshold α,
whenever
P Z2 F 1
2 (α) j Z1 F 1
1 (α) 6= 0
Christophe Villa () EM Lyon February, 25 2010 16 / 22
84. Asymptotic dependence
Upper tail dependence index between Z1 and Z2 is de…ned as
λU = lim
α!1
P(Z1 F 1
1 (α) j Z2 F 1
2 (α))
= lim
α!1
P(Z2 F 1
2 (α) j Z1 F 1
1 (α)),
Christophe Villa () EM Lyon February, 25 2010 17 / 22
85. Asymptotic dependence
Upper tail dependence index between Z1 and Z2 is de…ned as
λU = lim
α!1
P(Z1 F 1
1 (α) j Z2 F 1
2 (α))
= lim
α!1
P(Z2 F 1
2 (α) j Z1 F 1
1 (α)),
Lower tail dependence index is
λL = lim
α!0
P(Z1 F 1
1 (α) j Z2 F 1
2 (α))
= lim
α!0
P(Z2 F 1
2 (α) j Z1 F 1
1 (α)).
Christophe Villa () EM Lyon February, 25 2010 17 / 22
86. Asymptotic dependence
Upper tail dependence index between Z1 and Z2 is de…ned as
λU = lim
α!1
P(Z1 F 1
1 (α) j Z2 F 1
2 (α))
= lim
α!1
P(Z2 F 1
2 (α) j Z1 F 1
1 (α)),
Lower tail dependence index is
λL = lim
α!0
P(Z1 F 1
1 (α) j Z2 F 1
2 (α))
= lim
α!0
P(Z2 F 1
2 (α) j Z1 F 1
1 (α)).
0 λ 1
Christophe Villa () EM Lyon February, 25 2010 17 / 22
88. Asymptotic dependence
asymptotically dependent if λ > 0
) using multivariate EVT implicitly assume asymptotic dependence
asymptotically independent if λ = 0
) a range of extremal dependence models describe dependence but
have λ = 0
Christophe Villa () EM Lyon February, 25 2010 18 / 22
89. Asymptotic dependence
asymptotically dependent if λ > 0
) using multivariate EVT implicitly assume asymptotic dependence
asymptotically independent if λ = 0
) a range of extremal dependence models describe dependence but
have λ = 0
perfectly dependent if λ = 1
Christophe Villa () EM Lyon February, 25 2010 18 / 22
90. Asymptotic independence
Although the random variables are asymptotically independent,
di¤erent degrees of dependence are attainable at …nite levels of α.
Christophe Villa () EM Lyon February, 25 2010 19 / 22
91. Asymptotic independence
Although the random variables are asymptotically independent,
di¤erent degrees of dependence are attainable at …nite levels of α.
To this end, PRT introduce another (weak) tail dependence index
Christophe Villa () EM Lyon February, 25 2010 19 / 22
92. Asymptotic independence
Although the random variables are asymptotically independent,
di¤erent degrees of dependence are attainable at …nite levels of α.
To this end, PRT introduce another (weak) tail dependence index
upper tail dependence index between Z1 and Z2 is de…ned as
ηU = lim
α!1
log(1 α)
log P(Z1 > F 1
1 (α), Z2 > F 1
2 (α))
,
Christophe Villa () EM Lyon February, 25 2010 19 / 22
93. Asymptotic independence
Although the random variables are asymptotically independent,
di¤erent degrees of dependence are attainable at …nite levels of α.
To this end, PRT introduce another (weak) tail dependence index
upper tail dependence index between Z1 and Z2 is de…ned as
ηU = lim
α!1
log(1 α)
log P(Z1 > F 1
1 (α), Z2 > F 1
2 (α))
,
lower tail dependence index is
ηL = lim
α!0
log(α)
log P(Z1 F 1
1 (α), Z2 F 1
2 (α))
.
Christophe Villa () EM Lyon February, 25 2010 19 / 22
94. Asymptotic independence
Although the random variables are asymptotically independent,
di¤erent degrees of dependence are attainable at …nite levels of α.
To this end, PRT introduce another (weak) tail dependence index
upper tail dependence index between Z1 and Z2 is de…ned as
ηU = lim
α!1
log(1 α)
log P(Z1 > F 1
1 (α), Z2 > F 1
2 (α))
,
lower tail dependence index is
ηL = lim
α!0
log(α)
log P(Z1 F 1
1 (α), Z2 F 1
2 (α))
.
η > 0, η = 1
2 and η < 1 loosely correspond respectively to when Z1
and Z2 are positively associated in the extremes, exactly independent,
and negatively associated.
Christophe Villa () EM Lyon February, 25 2010 19 / 22
95. Asymptotic independence
Although the random variables are asymptotically independent,
di¤erent degrees of dependence are attainable at …nite levels of α.
To this end, PRT introduce another (weak) tail dependence index
upper tail dependence index between Z1 and Z2 is de…ned as
ηU = lim
α!1
log(1 α)
log P(Z1 > F 1
1 (α), Z2 > F 1
2 (α))
,
lower tail dependence index is
ηL = lim
α!0
log(α)
log P(Z1 F 1
1 (α), Z2 F 1
2 (α))
.
η > 0, η = 1
2 and η < 1 loosely correspond respectively to when Z1
and Z2 are positively associated in the extremes, exactly independent,
and negatively associated.
λ = 0 and η 2 (0, 1) signi…es asymptotic independence, in which case
the value of η determines the strength of dependence within this class
(also known as dependence in independence).
Christophe Villa () EM Lyon February, 25 2010 19 / 22
97. Results
The Category-based comovement is also a Category-based
(weak) tail comovement
Under the category-based comovement model, for two assets i and j
we have λ ∆Pi , ∆Pj = λ ei , ej = 0
Christophe Villa () EM Lyon February, 25 2010 20 / 22
98. Results
The Category-based comovement is also a Category-based
(weak) tail comovement
Under the category-based comovement model, for two assets i and j
we have λ ∆Pi , ∆Pj = λ ei , ej = 0
if two assets i and j, i 6= j, belong to the same category, i, j 2 X or Y ,
then
η ∆Pi,t , ∆Pj,t > η ei,t , ej,t
Christophe Villa () EM Lyon February, 25 2010 20 / 22
99. Results
The Category-based comovement is also a Category-based
(weak) tail comovement
Under the category-based comovement model, for two assets i and j
we have λ ∆Pi , ∆Pj = λ ei , ej = 0
if two assets i and j, i 6= j, belong to the same category, i, j 2 X or Y ,
then
η ∆Pi,t , ∆Pj,t > η ei,t , ej,t
suppose that asset j, previously a member of category Y , is reclassi…ed
as belonging to X. Then η ∆Pj,t , ∆PX ,t increases after j is added to
category X where ∆PX ,t = 1
n ∑
l2X
∆Pl,t .
Christophe Villa () EM Lyon February, 25 2010 20 / 22
100. Results ctd.
we assume that uX has heavy tails
Christophe Villa () EM Lyon February, 25 2010 21 / 22
101. Results ctd.
we assume that uX has heavy tails
if two assets i and j, i 6= j, belong to the same category,
i, j 2 X or Y , then
Christophe Villa () EM Lyon February, 25 2010 21 / 22
102. Results ctd.
we assume that uX has heavy tails
if two assets i and j, i 6= j, belong to the same category,
i, j 2 X or Y , then
if uX has right-heavy tails
1 = λU ∆Pi,t , ∆Pj,t > λU ei,t , ej,t = 0
Christophe Villa () EM Lyon February, 25 2010 21 / 22
103. Results ctd.
we assume that uX has heavy tails
if two assets i and j, i 6= j, belong to the same category,
i, j 2 X or Y , then
if uX has right-heavy tails
1 = λU ∆Pi,t , ∆Pj,t > λU ei,t , ej,t = 0
if uX has left-heavy tails
1 = λL ∆Pi,t , ∆Pj,t > λL ei,t , ej,t = 0
We thus observe that even if cash-‡ow shocks are asympotically
independent, ∆Pi and ∆Pj are perfectly dependent.
Christophe Villa () EM Lyon February, 25 2010 21 / 22
104. Results ctd.
we assume that uX has heavy tails
Christophe Villa () EM Lyon February, 25 2010 22 / 22
105. Results ctd.
we assume that uX has heavy tails
Suppose that asset j, previously a member of category Y , is
reclassi…ed as belonging to X. Then
Christophe Villa () EM Lyon February, 25 2010 22 / 22
106. Results ctd.
we assume that uX has heavy tails
Suppose that asset j, previously a member of category Y , is
reclassi…ed as belonging to X. Then
if uX has right-heavy tails, λU ∆Pj,t , ∆PX ,t increases after j is
added to category X where ∆PX ,t = 1
n ∑
l2X
∆Pl,t .
Christophe Villa () EM Lyon February, 25 2010 22 / 22
107. Results ctd.
we assume that uX has heavy tails
Suppose that asset j, previously a member of category Y , is
reclassi…ed as belonging to X. Then
if uX has right-heavy tails, λU ∆Pj,t , ∆PX ,t increases after j is
added to category X where ∆PX ,t = 1
n ∑
l2X
∆Pl,t .
if uX has left-heavy tails, λL ∆Pj,t , ∆PX ,t increases after j is added
to category X where ∆PX ,t = 1
n ∑
l2X
∆Pl,t .
Christophe Villa () EM Lyon February, 25 2010 22 / 22