Post on 01-Nov-2021
International Equity Market Correlations
presented at Northfield Research Conference 4 December, 2000
Rosemary MacedoBailard, Biehl & Kaiser
950 Tower Lane #1900, Foster City, CA 94404rmacedo@bailard.com 650-483-7953
2
Discussion Topics
• Is our ability to diversify international portfolios deteriorating?
• Does diversification disappear right when you need it most?
• Should the country approach be replaced with a sector approach?
3
Why the Answers Matter
Stress-testing--assessment of effects of high volatility on portfolio should include expected changes in the correlation.
Choice of long-term or recent history to estimate covariance.
Better assessment of risk and opportunity.
Optimal allocation of assets.
Hedging strategies.
4
Data
Datastream indices:
Daily market returns ( Jan 1973 through Oct 2000, US$)60 day cross-country correlations ( 231 pairs of countries)
Weekly individual country sector returns(9cty X 10sect, Jan 1990 - Oct 2000, E)eg: French telecom, German telecom . . .
French banks, German banks . . .
Weekly regional sector returns ( Jan 1973 through Oct 2000, E)Weekly country returns ( Jan 1973 through Oct 2000, E)
5
Deteriorating Diversification
Globalization--countries, companies and markets becoming more and more alike.
Euroland--move to single currency, single market.
Cross border mergers and acquisitions.
Multiple listings, expanded trading hours.
Institutional portfolio manager herding.
6
Average Correlations With USA1.0
.9
.8
.7
.6
.5
.4
.3
.2
.1
0.0
-.1
0007
1300
0127
9908
1299
0225
9809
0898
0324
9710
0397
0418
9610
3096
0515
9511
2795
0612
9412
2394
0708
9401
2193
0806
9302
1992
0902
9203
1891
0930
9104
1590
1025
9005
1089
1121
8906
0688
1219
8807
0488
0118
8707
3087
0212
8608
2686
0311
8509
2 085
040 5
8410
1 784
050 2
8311
1 583
053 1
8212
1 482
062 9
8201
1 281
072 8
8102
1 080
082 2
8003
0 779
091 9
7904
0 478
101 6
7805
0 177
111 0
7705
2 676
1209
7606
2476
0108
7507
2275
0204
7408
1674
0301
7309
1273
0328
across country pairs: ρ(USA,UK),ρ(USA,Jpn),ρ(USA,Ger) . . .
Source: BB&K
Trailing 60days (yymmdd)
7
0007
1300
0127
9908
1299
0225
9809
0898
0324
9710
0397
0418
9610
3096
0515
9511
2795
0612
9412
2394
0708
9401
2193
0806
9302
1992
0902
9203
1891
0930
9104
1590
1025
9005
1089
1121
8906
0688
1219
8807
0488
0118
8707
3087
0212
8608
2686
0311
8509
2 085
040 5
8410
1 784
050 2
8311
1 583
053 1
8212
1 482
062 9
8201
1 281
072 8
8102
1 080
082 2
8003
0 779
091 9
7904
0 478
101 6
7805
0 177
111 0
7705
2 676
1209
7606
2476
0108
7507
2275
0204
7408
1674
0301
7309
1273
0328
1.0
.9
.8
.7
.6
.5
.4
.3
.2
.1
0.0
across country pairs: ρ(Fra,Ger),ρ(Fra,Ita),ρ(Ger,Ita) . . .
Average Correlations Within Euroland
Source: BB&K
Trailing 60days (yymmdd)
8
1.0
.9
.8
.7
.6
.5
.4
.3
.2
.1
0.0
0007
1300
0127
9908
1299
0225
9809
0898
0324
9710
0397
0418
9610
3096
0515
9511
2795
0612
9412
2394
0708
9401
2193
0806
9302
1992
0902
9203
1891
0930
9104
1590
1025
9005
1089
1121
8906
0688
1219
8807
0488
0118
8707
3087
0212
8608
2686
0311
8509
2 085
040 5
8410
1 784
050 2
8311
1 583
053 1
8212
1 482
062 9
8201
1 281
072 8
8102
1 080
082 2
8003
0 779
091 9
7904
0 478
101 6
7805
0 177
111 0
7705
2 676
1209
7606
2476
0108
7507
2275
0204
7408
1674
0301
7309
1273
0328
across country pairs: ρ(Jpn,HKg),ρ(Jpn,Sng),ρ(HKg,Sng) . . .
Average Correlations Within Pacific Rim
Source: BB&K
Trailing 60days (yymmdd)
9
Regional Comparison
73-81 82-91 92-10/00avg st dev avg st dev avg st dev
Euro-Other Eur 0.28 0.22 0.44 0.21 0.48 0.18within Euroland 0.27 0.22 0.43 0.21 0.47 0.19within Other Eur 0.21 0.22 0.43 0.21 0.45 0.16within Pacific Rim 0.17 0.20 0.30 0.22 0.31 0.20Pac Rim-Other Eur0.13 0.18 0.27 0.19 0.22 0.16Pac Rim-Euro 0.14 0.18 0.25 0.20 0.21 0.16
with USA 0.19 .21 0.23 0.21 0.28 0.18
10
Correlation StabilityLESS
STABLE
MORESTABLE
MOREDIVERSIFYING
LESSDIVERSIFYING
Average of 60day Correlations
NZE,BEL
NZE,UK
.7.6.5.4.3.2.10.0
.3
.2
.1
ITA,POR
USA,CAN
AUS,NZE
GER,NET
GER,SWI
ITA,FRAITA,GER
ITA,NET
CAN,POR
CAN,ITA
CAN,UK
CAN,FRANZE,NET
USA,FRAUSA,SWE
USA,NET
FRA,SPA
UK,SWI
USA,POR
USA,AUT
USA,DEN
USA,JPN
SPA,PORFRA,GER
GER,AUT
FRA,UK
Sta
ndar
d D
evia
tion
of 6
0day
Cor
rela
tions
11
Phantom Diversificationvanishes when most needed
Higher correlations when markets fall.
Higher correlations when markets are volatile.
Does this reflect a real change in the relationships between asset returns?
How inconsistent is this with constant correlation assumption? Normality assumption? Compare with theoretical conditional correlations.
What about “contagion”?
12
Extreme Market Moves
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
bottomdecilereturns
topdecilereturnsC
ondi
tiona
l Cor
rela
tion
Unconditional Correlation
13
Constant Correlation Mandates Higher Conditional Correlation in the Tails
ρ2 + (1-ρ2) Var(x) / Var(x|A)
ρρA=
ρA = f( ρ, )Var(x|A)Var(x)
Significant differences between the conditional correlations are caused by the choice of subsamples alone; they do not necessarily indicate any change in the parameters of the data generating process, merely the time-varying market volatility.
14Expected Increase Conditional Correlation vs.
Unconditional
0.00
0.05
0.10
0.15
0.20
0.25
0.30
0.35
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Two-sided event probability
5%10%20%
top&bottom decile
50%top&bottom quartile
Exp
ecte
d In
crea
se (b
ivar
iate
nor
mal
rand
om v
aria
bles
)C
ondi
tiona
l Cor
rela
tion
vs. U
ncon
ditio
nal C
orre
latio
n(e
ffect
s w
ould
be
mor
e pr
onou
nced
for d
istri
butio
ns w
ith fa
tter t
ails
)
Source: BB&K
Unconditional Correlation
15
Observations
Expect correlations to increase significantly, especially for unconditional correlations between 0.4-0.5, and especially for more extreme conditions:
.4 --> .5+ conditioned on top&bottom quartile returns.4 --> .6+ conditioned on top&bottom decile returns
Correlations do not increase uniformly. Optimal portfolios based on the conditional variance-covariance matrix therefore will differ from those based on the unconditional.
Market returns exhibit significant excess kurtosis & negative skewness--affects will be worse than shown on previous slide.
16Average Correlations With USA vs. Volatility
54321.9.8.7.6.5.4.3
1.0
.9
.8
.7
.6
.5
.4
.3
.2
.1
0.0
-.1Ave
rage
60d
ay C
orre
latio
n ac
ross
cou
ntry
pai
rs:
ρ(U
SA
,UK
),ρ(U
SA
,Jpn
),ρ(U
SA
,Ger
) . .
.
60day Volatility
17Average Correlations Within Euroland vs. Volatility
54321.9.8.7.6.5.4.3
1.0
.9
.8
.7
.6
.5
.4
.3
.2
.1
0.0Ave
rage
60d
ay C
orre
latio
n ac
ross
cou
ntry
pai
rs: ρ
(Fra
,Ger
),ρ(F
ra,It
a),ρ
(Ger
,Ita)
. . .
60day Volatility
18Average Correlations Within Pacific Rim
vs. Volatility
54321.9.8.7.6.5.4.3
1.0
.9
.8
.7
.6
.5
.4
.3
.2
.1
0.0Ave
rage
60d
ay C
orre
latio
n ac
ross
cou
ntry
pai
rs: ρ
(Fra
,Ger
),ρ(F
ra,It
a),ρ
(Ger
,Ita)
. . .
60day Volatility
19
Worse in Euroland?
Does the steeper slope (correlation vs volatility) in Euroland indicate some effect beyond what statistics predicts?
Check by comparing actual increase in conditional correlation for top quartile of volatility with expected. (from Monte Carlo Simulation)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Probably not. The unconditionalcorrelations within Euroland are higher than for other regions, closer to peak impact from sampling effects.
20
WithinEuroland
others
-0.10
-0.05
0.00
0.05
0.10
0.15
0.20
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7Incr
ease
:C
ondi
tiona
l Cor
rela
tion
vs. U
ncon
ditio
nal
Unconditional Correlation
Monte Carlo Simulationmedian
Top Quartile of Volatility
21
Summary
Very easy to make compelling, in some cases alarming, pictures.
Especially post mortem.
Autopsy reveals statistics not pathology.
22
2 Aspirin
Use a conditional variance-covariance matrix: • if downside risk control is paramount• when stress-testing portfolios
Significant serial correlation in volatility: around 0.5 for larger markets, 0.3-0.4 for most other markets.• ARCH, GARCH, SWARCH, M ARCH etc.
prescribed by some researchers (see references).
c
23
Sectors vs Countries
Many of the same arguments as for deteriorating ability to diversify.
Premise: stocks more correlated with other stocks in same sector in any market than they are with other stocks in their own market but in different sectors.
Self-fulfilling prophesy?
24
Counterarguments
Structural differences among countries-- taxes, government, regulation, labor, language, accounting,
Asynchronous business cycles,
Asymmetric impact of single monetary policy,
Currency alone does not a single market make (Hong Kong ),
Sell-side ploy.
25
Countries vs. Sectors
Are correlations stronger:• within the same sector across countries?• within the same country across sectors?
I.E. is the German Financial sector more correlated with:• French financials, Swiss Financials, Dutch financials …• or German Utilities, German Consumer Cyclicals …
Use a scatterplot to compare the average correlation for each country--sector index with the indices for:• other countries--same sector• same country--other sectors
26
.7.6.5.4.3.2.10.0
.7
.6
.5
.4
.3
.2
.1
0.0
1/90 - 10/00
countriesmore
important
sectorsmore
important
Average correlation across countries in same sectorρ(GerFinl,FraFinl), ρ(GerFinl,ItaFinl), ρ(GerFinl,NetFinl) ...
ρ(G
erFi
nl,G
erU
til),
ρ(G
erFi
nl,G
erC
onsC
yc),.
..A
vera
ge c
orre
latio
n ac
ross
sec
tors
in s
ame
coun
try
Source: BB&K
Countries vs. Sectors
ITresourcesbasic industriesgeneral industrialsconsumer cyclicalsconsumer noncyclicalscyclical servicesnoncyclical servicesfinancialsutilities
AustriaBelgiumFranceGermanyItalyNetherlandsSpainFinlandIreland
27
Countries vs. Sectors1/98 - 10/00
.7.6.5.4.3.2.10.0
.7
.6
.5
.4
.3
.2
.1
0.0
countriesmore
important
sectorsmore
important
Average correlation across countries in same sectorρ(GerFinl,FraFinl), ρ(GerFinl,ItaFinl), ρ(GerFinl,NetFinl) ...
ρ(G
erFi
nl,G
erU
til),
ρ(G
erFi
nl,G
erC
onsC
yc),.
..A
vera
ge c
orre
latio
n ac
ross
sec
tors
in s
ame
coun
try
Source: BB&K
ITresourcesbasic industriesgeneral industrialsconsumer cyclicalsconsumer noncyclicalscyclical servicesnoncyclical servicesfinancialsutilities
AustriaBelgiumFranceGermanyItalyNetherlandsSpainFinlandIreland
28
“Risk Map”
Plot each series as a point (x,y) so that proximity on the map corresponds to strength of correlation.
Effectively a least squares fit withρAB = correlation of series A and B zAB = (xA-xB)2+(yA-yB)2
Code the points on the plot & look for clustersby color for countryby shape for sector
Specifically which indices are most/least alike?
29
“Risk Map” 1/90-10/00
2.52.01.51.0.50.0-.5-1.0-1.5-2.0
2.0
1.5
1.0
.5
0.0
-.5
-1.0
-1.5
-2.0
-2.5
Proximity on plot indicates correlation
Source: BB&K
ITresourcesbasic industriesgeneral industrialsconsumer cyclicalsconsumer noncyclicalscyclical servicesnoncyclical servicesfinancialsutilities
AustriaBelgiumFranceGermanyItalyNetherlandsSpainFinlandIreland
30
“Risk Map” 1/98-10/00
2.01.51.0.50.0-.5-1.0-1.5-2.0-2.5
2.5
2.0
1.5
1.0
.5
0.0
-.5
-1.0
-1.5
-2.0
Proximity on plot indicates correlation
Source: BB&K
ITresourcesbasic industriesgeneral industrialsconsumer cyclicalsconsumer noncyclicalscyclical servicesnoncyclical servicesfinancialsutilities
AustriaBelgiumFranceGermanyItalyNetherlandsSpainFinlandIreland
31
Opportunity
991210981211
971212961213
951215941216
931217921218
911220901221
891222881223
871225861226
851227841228
831230821231
820101810102
800104790105
780106770107
760109750110
740111730112
108
6
4
2
1.8
.6
.4
.2
.1
Ratio of weekly cross-sectional dispersion in European country indicesto weekly cross-sectional dispersion in European sector indices.
Country approach more favorable
Sector approach more favorableOverall c70 i30 pre ‘85 c57 i43) Source: BB&K
32
Observations
Evidence in the Countries vs. Sectors debate fluctuates over time.
Last two years, the balance of opportunity (as well as sentiment), has shifted to sectors.
Overall, countries have represented the greater opportunity (70/30),
Even pre-1985 when countries were not as dominant. (57/43)
33
References, Further Reading“Pitfalls in Tests for Changes in Correlations,” Boyer, Brian H, Michael S Gibson and Mico Loretan, Board of Governors of the Federal Reserve System International Finance Discussion Papers Number 597, December 1997.• how conditioning on events affects correlations (selection bias)
“Evaluating ‘Correlation Breakdowns’ During Periods of Market Volatility,” Loretan, Mico and William B. English, Board of Governors of the Federal Reserve System International Finance Discussion Papers Number 658, February 2000.• If “contagion” is defined as an elevation in correlations between asset returns,
then contagion is a natural by-product of temporal variation in volatilities.
“Is the Correlation in International Equity Returns Constant: 1960-1990?” Longin, Francois, BrunoSolnik, CEPR Financial Markets Paper, RePEc:cpr:ceprfm:0037, October 1993
“Covariance and Correlation in International Equity Returns: A Value-at-Risk Approach,” Campbell, Rachel, Kees Koedijk, Paul Kaufman May 2000• fat tails in return distribution are better fit by student-t distribution than normal distribution.
“Correlation in International Equity and Currency Markets: A Risk Adjusted Perspective,” Sheedy, Elizabeth, Centre for Studies in Money, Banking and Finance [CMBF] Paper No. 17, June 1997• accounting for volatility clustering effectively eliminates structure in return correlation
GARCH-CC, BEKK, several other models evaluated• daily data significantly improve portfolio efficiency (Sharpe ratio), even when portfolio
adjustments are made only monthly.
34
. . .
“Do World Markets Still Serve as a Hedge?” Erb, Claude B., Campbell R. Harvey, Tadas E. Viskanta, Journal of Investing, Fall 1995 pp26-42• developed and emerging market stocks & bonds• correlations higher for extreme moves, especially negative moves• currency hedging increases correlation
“International Market Correlation and Volatility,” Solnik, Bruno, Cyril Boucrelle, Yann Le Fur. Financial Analysts Journal, Sept/Oct 1996• r(US,XXX) monthly 1959-95 correlations increasing slightly• r(US,XXX) weekly 1982-95 correlations not increasing• correlations fluctuate widely over time• correlation increases during volatile markets
“European Equity Markets and the EMU,” Rouwenhorst, K Geert, Financial Analysts Journal, May/June 1999 pp 57-64.• “no evidence supports the disappearance of differences between EMU countries’ equity returns.”
For very interesting work on emerging markets contagion tests, start with MIT Prof Kristin Forbes’ website & papers, with good links, also.
35
Conclusions
• Our ability to diversify international portfolios does not appear to be deteriorating.
• Correlations are inherently higher in extreme periods. – This is consistent with constant correlations. – Even so, worthwhile to diversify.
• Evidence for a permanent switch from countries- to sectors- focus is not compelling.