study of the correlations between stocks of different marketscoelhor/apfa_presentation_v2.0.pdf ·...
TRANSCRIPT
Study of the correlations between stocks ofdifferent markets
Ricardo Coelho
School of PhysicsTrinity College Dublin
APFA 6 - Presentation, 5th June 2007
Outline
Introduction
DefinitionsLogarithmic ReturnCorrelationsRandom Matrix TheoryDistances
ResultsIndustrial clusteringGeographical clusteringIndustrial vs Geographical clustering
Summary
Outline
Introduction
DefinitionsLogarithmic ReturnCorrelationsRandom Matrix TheoryDistances
ResultsIndustrial clusteringGeographical clusteringIndustrial vs Geographical clustering
Summary
Outline
Introduction
DefinitionsLogarithmic ReturnCorrelationsRandom Matrix TheoryDistances
ResultsIndustrial clusteringGeographical clusteringIndustrial vs Geographical clustering
Summary
Outline
Introduction
DefinitionsLogarithmic ReturnCorrelationsRandom Matrix TheoryDistances
ResultsIndustrial clusteringGeographical clusteringIndustrial vs Geographical clustering
Summary
Introduction
We studied three different sets of data.I A portfolio from the London Stock Exchange FTSE100 index:
I 67 stocksI Daily closing priceI From 2nd August 1996 until 27th June 2005
I Different World IndicesI 53 countries’ equity marketsI Wednesday closing price (weekly returns)I From 8th January 1997 until 1st February 2006
I More than 6000 stocks from markets around the worldI Daily closing priceI From 30th December 1994 until 1st January 2007
Log-return of stocks shows extreme events (e.g. HSBC)
Ri (t) = lnPi (t)− lnPi (t − 1) (1)
10-1995 03-1997 07-1998 12-1999 04-2001 08-2002 01-2004 05-2005 10-2006
time (month - year)
-0.1
-0.05
0
0.05
0.1
0.15
Ret
urn
- R
( t
)
11th
Sept. 2001
Distribution of the Log-return is non-Gaussian
k ∼ 2.90 and q ∼ 1.34
-0.2 -0.1 0 0.1 0.2R(t)
0.01
0.1
1
10
100P
( R
(t)
)
Correlation between stocks i and j
ρij =〈RiRj〉 − 〈Ri 〉〈Rj〉√(
〈R2i 〉 − 〈Ri 〉2
) (〈R2
j 〉 − 〈Rj〉2) (2)
I ρij form a symmetric N × N matrix; −1 ≤ ρij ≤ 1; ρii = 1.
0 0.2 0.4 0.6 0.8ρ
ij
0
5
10
15
20
P (
ρ ij )
67 stocks from FTSE100
67 random time series
Matrix of correlations show non-randomness
Random Matrix Theory for N →∞ and T →∞, where Q = T/Nis fixed and bigger than 1.
PRM(λ) =Q
2πσ2
√(λmax − λ)(λ− λmin)
λ(3)
0.6 0.8 1 1.2 1.4 1.6 1.8 2
Eigenvalue λ0
0.5
1
1.5
2
2.5
3
P RM
( λ
) 67 random time series
0 5 10 15
Eigenvalue λ0
0.5
1
1.5
2
P real
( λ
)
67 stocks from FTSE100
highest eigenvalue
Distance between stocks i and j
dij =√
2(1− ρij) (4)
I 0 ≤ dij ≤ 2, small values imply strong correlations.I From distances we construct the Minimal Spanning Tree
(MST).I Choose the minimum distance between 2 stocks - one link
between these 2.
I Find the next minimum distance. If the 2 stocks are notlinked or both linked, choose the next minimum distance. Ifjust 1 of the stocks has already links, link them.
Distance between stocks i and j
dij =√
2(1− ρij) (4)
I 0 ≤ dij ≤ 2, small values imply strong correlations.I From distances we construct the Minimal Spanning Tree
(MST).I Choose the minimum distance between 2 stocks - one link
between these 2.I Find the next minimum distance. If the 2 stocks are not
linked or both linked, choose the next minimum distance. Ifjust 1 of the stocks has already links, link them.
Distance between stocks i and j
dij =√
2(1− ρij) (4)
I 0 ≤ dij ≤ 2, small values imply strong correlations.I From distances we construct the Minimal Spanning Tree
(MST).I Choose the minimum distance between 2 stocks - one link
between these 2.I Find the next minimum distance. If the 2 stocks are not
linked or both linked, choose the next minimum distance. Ifjust 1 of the stocks has already links, link them.
Distance between stocks i and j
dij =√
2(1− ρij) (4)
I 0 ≤ dij ≤ 2, small values imply strong correlations.I From distances we construct the Minimal Spanning Tree
(MST).I Choose the minimum distance between 2 stocks - one link
between these 2.I Find the next minimum distance. If the 2 stocks are not
linked or both linked, choose the next minimum distance. Ifjust 1 of the stocks has already links, link them.
Distance between stocks i and j
dij =√
2(1− ρij) (4)
I 0 ≤ dij ≤ 2, small values imply strong correlations.I From distances we construct the Minimal Spanning Tree
(MST).I Choose the minimum distance between 2 stocks - one link
between these 2.I Find the next minimum distance. If the 2 stocks are not
linked or both linked, choose the next minimum distance. Ifjust 1 of the stocks has already links, link them.
Stocks cluster in industrial sectors (ICB classification)
Full time series, T = 2322 days, for the FTSE100 stocks.
Physica A 373 (2007) 615-626
Markets organised by geographical location
Full time series, T = 475 weeks, for the 53 World Indices.
Physica A 376 (2007) 455-466
What about stocks from different countries???
Full time series, T = 3127 days, for 2500 stocks from differentmarkets.
NYSE vs LON - Correlation between stocks of differentmarkets
322 stocks from each market.
0 0.2 0.4 0.6 0.8 1ρ
ij
0
500
1000
1500
2000
2500
3000
# o
f ρ
ij
global correlation matrix
NYSE-NYSE
LON-LON
LON-NYSE
NYSE vs LON - Eigensystem information
The two largest eigenvalues could represent each market.
0 1 2 3 40
0.5
1
1.5
2
0 20 40 60 800
highest eigenvalue
NYSE vs LON - Highest eigenvector information
NYSE vs LON - Highest eigenvector information
Sum of elements from each sector/country -∑
i λi
indu
stria
ls
finan
cial
s
heal
th_c
are
tech
nolo
gy
oil_
and_
gas
utili
ties
basi
c_m
ater
ials
tele
com
mun
icat
ions
cons
umer
_goo
ds
cons
umer
_ser
vice
s0.000
0.005
0.010
0.015
0.020
0.025
0.030
0.035
0.040
0.045
0.050
LON NYSE
NYSE vs LON - Clustering of stocks in market
NYSE vs LON - Clustering in sector for NYSE
Different choice of stocks gives different clustering for LON.Oil & Gas - black / Basic Materials - blue / Industrials - gray / Consumer Goods - yellow / Health Care - greenConsumer Services - red / Telecommunications - pink / Utilities - purple / Financials - white / Technology - orange
Pajek
NYSE vs LON - Correlation (after filtering market mode)
Filtering affects NYSE more than LON.
-0.4 -0.2 0 0.2 0.4 0.6 0.8 1ρ
ij
0
500
1000
1500
2000
2500
3000#
of
ρij
global correlation matrix
NYSE-NYSE
LON-LON
LON-NYSE
NYSE vs LON - Highest eigenvector (after filtering)
Sum of elements from each sector/country -∑
i λi
indu
stria
ls
finan
cial
s
heal
th_c
are
tech
nolo
gy
oil_
and_
gas
utili
ties
basi
c_m
ater
ials
tele
com
mun
icat
ions
cons
umer
_goo
ds
cons
umer
_ser
vice
s-0.04
-0.03
-0.02
-0.01
0
0.01
0.02
0.03
0.04
0.05
0.06
LON NYSE
And if LON is one day ahead?
Correlation between NYSE stocks in time t and LON stocks intime t + 1.
0 0.1 0.2 0.3 0.4 0.5ρ
ij
0
500
1000
1500
2000
2500
3000
# o
f ρ
ij
LON (t) - NYSE (t)
LON (t+1) - NYSE (t)
NYSE (t) vs LON (t+1) - Highest eigenvector
Sum of elements from each sector/country -∑
i λi
indu
stria
ls
finan
cial
s
heal
th_c
are
tech
nolo
gy
oil_
and_
gas
utili
ties
basi
c_m
ater
ials
tele
com
mun
icat
ions
cons
umer
_goo
ds
cons
umer
_ser
vice
s0.000
0.005
0.010
0.015
0.020
0.025
0.030
0.035
0.040
0.045
0.050
LON NYSE
NYSE (t) vs LON (t+1) - Highest (after filtering)
Sum of elements from each sector/country -∑
i λi
indu
stria
ls
finan
cial
s
heal
th_c
are
tech
nolo
gy
oil_
and_
gas
utili
ties
basi
c_m
ater
ials
tele
com
mun
icat
ions
cons
umer
_goo
ds
cons
umer
_ser
vice
s0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
LON NYSE
NYSE (t) vs LON (t+1) - 2th highest (after filtering)
Sum of elements from each sector/country -∑
i λi
indu
stria
ls
finan
cial
s
heal
th_c
are
tech
nolo
gy
oil_
and_
gas
utili
ties
basi
c_m
ater
ials
tele
com
mun
icat
ions
cons
umer
_goo
ds
cons
umer
_ser
vice
s-0.055-0.050-0.045-0.040-0.035-0.030-0.025-0.020-0.015-0.010-0.0050.0000.0050.0100.0150.020
LON NYSE
What about two markets from the same continent?PAR vs LON - Correlation
125 stocks from each market.
0 0.2 0.4 0.6 0.8ρ
ij
0
100
200
300
400
500
# o
f ρ
ij
global correlation matrix
PAR-PAR
LON-LON
LON-PAR
PAR vs LON - Eigensystem
One major eigenvalue in contrast with the NYSE-LON situation.
0 1 2 3 40
0.5
1
1.5
2
0 10 20 30 400
0.5
1
highest eigenvalue
PAR vs LON - Highest eigenvector
Sum of elements from each sector/country -∑
i λi
finan
cial
s
oil_
and_
gas
cons
umer
_goo
ds
cons
umer
_ser
vice
s
indu
stria
ls
tech
nolo
gy
basi
c_m
ater
ials
heal
th_c
are
utili
ties0.00
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
0.10
LON PAR
PAR vs LON - No clustering of stocks in market
Circles for LON and boxes for PAR.
Pajek
PAR vs LON - No clustering of stocks in sectors
Oil & Gas - black / Basic Materials - blue / Industrials - gray / Consumer Goods - yellow / Health Care - greenConsumer Services - red / Telecommunications - pink / Utilities - purple / Financials - white / Technology - orange
Pajek
PAR vs LON - Highest eigenvector (after filtering)
Sum of elements from each sector/country -∑
i λi
finan
cial
s
oil_
and_
gas
cons
umer
_goo
ds
cons
umer
_ser
vice
s
indu
stria
ls
tech
nolo
gy
basi
c_m
ater
ials
heal
th_c
are
utili
ties-0.09
-0.08-0.07-0.06-0.05-0.04-0.03-0.02-0.01
00.010.020.030.040.050.06
LON PAR
PAR vs LON - Clustering of stocks in market (a.f.)
PAR vs LON - No clustering of stocks in sectors (a.f.)
Oil & Gas - black / Basic Materials - blue / Industrials - gray / Consumer Goods - yellow / Health Care - greenConsumer Services - red / Telecommunications - pink / Utilities - purple / Financials - white / Technology - orange
Pajek
PAR vs LON - 2th highest eigenvector (after filtering)
Sum of elements from each sector/country -∑
i λi
finan
cial
s
oil_
and_
gas
cons
umer
_goo
ds
cons
umer
_ser
vice
s
indu
stria
ls
tech
nolo
gy
basi
c_m
ater
ials
heal
th_c
are
utili
ties-0.09
-0.07
-0.05
-0.03
-0.01
0.01
0.03
0.05
0.07
LON PAR
Conclusions
I Two different methodologies of analysing large amount ofdata give same results
I Similar behaviour for both the FTSE100 stocks and the WorldIndices:
I MST show different clusters (Industrial sectors / Geography)
I The choice of the stocks can influence the clustering in sectors
I The stocks cluster in market before clustering in sectors
I The 3rd eigenvalue for the PAR-LON andNYSE(t)-LON(t + 1) show the segregation of stocks inmarkets.
Future work
I Results for more than 2 markets from the same continent
Econophysics group of the Trinity College Dublin:I Prof. Peter RichmondI Dr. Stefan HutzlerI Ricardo CoelhoI Joseph Barry
Collaborations:I Prof. Brian Lucey (Business School of Trinity College Dublin)I Prof. Claire Gilmore (McGowan School of Business of King’s
College, Pennsylvania)
Part of this work was published in Physica A:
Physica A 373 (2007) 615-626
Physica A 376 (2007) 455-466
AcknowledgmentsI SFI - Science Foundation IrelandI FCT - Fundacao para a Ciencia e a Tecnologia