View
2
Download
0
Category
Preview:
Citation preview
4322 North Quad, 105 S. State St. Ann Arbor, MI 48109-1285
Social Network Under Stress
DanielM.RomeroSchoolofInforma3onUniversityofMichigan
Incollabora3onwithBrianUzziandJonKleinberg
1
SocialNetworkTemporalDynamics
2
t=1 t=2 t=3 t=4
SocialNetworkTemporalDynamics
Temporaldynamicsofnetworks:Shortdiameter,densifica3on,clustering,heavytaildegreedistribu3on,…[Leskovecetal.2007,Barabasietal.1999,Kossinetsetal.2009,…]
3
t=1 t=2 t=3 t=4
SocialNetworkTemporalDynamics
Temporaldynamicsofnetworks:Shortdiameter,densifica3on,clustering,heavytaildegreedistribu3on,…[Leskovecetal.2007,Barabasietal.1999,Kossinetsetal.2009,…]
Usefulfor:• Linkpredic3on• Detec3nginfluen3alnodes• Findingcommuni3es 4
t=1 t=2 t=3 t=4
SocialNetworkTemporalDynamics
5
t=1
SocialNetworkTemporalDynamics
6
t=1 t=2
SocialNetworkTemporalDynamics
7
t=1 t=2
SocialNetworkTemporalDynamics
8
t=1 t=2
?
HedgeFundDataInstantMessages(IM):• FullrecordofIMs:content,sender,recipient,3mestamp• 182internaldecisionmakers,8646outsidecontacts• 22MillionIMs
9
HedgeFundDataInstantMessages(IM):• FullrecordofIMs:content,sender,recipient,3mestamp• 182internaldecisionmakers,8646outsidecontacts• 22MillionIMs
StockTrading:• Fullrecordofalltransac3ons:stock,price,numberofstocks,typeoftransac3on(Buy,Sell),3mestamp• 600Ktrades• 2008–2012
10
MarketMovements(Shocks) SocialNetwork
11
InThisTalk
MarketMovements(Shocks) SocialNetwork
12
Trading
InThisTalk
MarketMovements(Shocks) SocialNetwork
Performance
13
Trading
InThisTalk
MarketMovements(Shocks) SocialNetwork
Emo3onalandCogni3veContent
Performance
14
Trading
InThisTalk
MarketMovements(Shocks) SocialNetwork
Emo3onalandCogni3veContent
Performance
15
Trading
InThisTalk
16
Measures
Shock:Changeinpriceofstocksondayd%change:(closing–opening)/opening
17
Measures
Foreachstocksanddayd,generatenetworkG(s,d)amongemployeeswhomen3ons
Shock:Changeinpriceofstocksondayd%change:(closing–opening)/opening
18
Measures
Foreachstocksanddayd,generatenetworkG(s,d)amongemployeeswhomen3ons
Shock:Changeinpriceofstocksondayd%change:(closing–opening)/opening
Foreachstocksanddayd,generatenetworkG(s,d)amongemployeeswhomen3ons
Shock:Changeinpriceofstocksondayd%change:(closing–opening)/opening
19
Measures
Foreachstocksanddayd,generatenetworkG(s,d)amongemployeeswhomen3ons
Shock:Changeinpriceofstocksondayd%change:(closing–opening)/opening
20
Measures
Network’sfeatures:• Size(Nodes,edges)
Foreachstocksanddayd,generatenetworkG(s,d)amongemployeeswhomen3ons
Shock:Changeinpriceofstocksondayd%change:(closing–opening)/opening
21
Measures
Network’sfeatures:• Size(Nodes,edges)• Density(Clustering
Foreachstocksanddayd,generatenetworkG(s,d)amongemployeeswhomen3ons
Shock:Changeinpriceofstocksondayd%change:(closing–opening)/opening
22
Measures
Network’sfeatures:• Size(Nodes,edges)• Density(Clustering
Foreachstocksanddayd,generatenetworkG(s,d)amongemployeeswhomen3ons
Shock:Changeinpriceofstocksondayd%change:(closing–opening)/opening
23
Measures
Network’sfeatures:• Size(Nodes,edges)• Density(Clustering
Foreachstocksanddayd,generatenetworkG(s,d)amongemployeeswhomen3ons
Shock:Changeinpriceofstocksondayd%change:(closing–opening)/opening
24
Measures
Network’sfeatures:• Size(Nodes,edges)• Density(Clustering,3estrength)
Foreachstocksanddayd,generatenetworkG(s,d)amongemployeeswhomen3ons
Shock:Changeinpriceofstocksondayd%change:(closing–opening)/opening
25
Measures
Network’sfeatures:• Size(Nodes,edges)• Density(Clustering,3estrength)
Foreachstocksanddayd,generatenetworkG(s,d)amongemployeeswhomen3ons
Shock:Changeinpriceofstocksondayd%change:(closing–opening)/opening
26
Measures
Network’sfeatures:• Size(Nodes,edges)• Density(Clustering,3estrength)
Foreachstocksanddayd,generatenetworkG(s,d)amongemployeeswhomen3ons
Shock:Changeinpriceofstocksondayd%change:(closing–opening)/opening
27
Measures
Network’sfeatures:• Size(Nodes,edges)• Density(Clustering,3estrength)• Openness(Borderedges)
Foreachstocksanddayd,generatenetworkG(s,d)amongemployeeswhomen3ons
Shock:Changeinpriceofstocksondayd%change:(closing–opening)/opening
28
Measures
Network’sfeatures:• Size(Nodes,edges)• Density(Clustering,3estrength)• Openness(Borderedges)
Foreachstocksanddayd,generatenetworkG(s,d)amongemployeeswhomen3ons
Shock:Changeinpriceofstocksondayd%change:(closing–opening)/opening
29
Measures
30
Turtled-upnetwork
31
Turtled-upnetwork
Opennetwork
Networksmayturtle-upduringshocks:• Trust(Granovefer1985,Coleman1988)• Exper3seknowledge,repeated
informa3onchannels(Coleman1990)• Threatrigidity(Staw1981)
32
Theore3calExpecta3ons
Turtled-upnetwork
Opennetwork
Networksmayturtle-upduringshocks:• Trust[Granovefer1985,Coleman1988]• Exper3seknowledge,repeated
informa3onchannels[Coleman1990]• Threatrigidity[Staw1981]
33
Theore3calExpecta3ons
Turtled-upnetwork
Opennetwork
Networksmayopen-upduringshocks:• Newinforma3onthroughweak3es
[Granovefer1973]• Diverseinforma3onfromdifferentgroups
(structuralholes)[Burt92]
Numofnodes|Past:Ra3oofnum.nodesinG(s,d)andmeannum.nodesinG(s,d’)ford’<d.
�10 �5 0 5 10
Change in stock price (%)1.4
1.6
1.8
2.0
2.2
2.4
2.6
Num
.nod
es
Num. nodes | past days
Num
.Nod
es
Changeinstockprice(%)Changeinstockprice(%)
-10010
34
Findings:Size
Numofnodes|Past:Ra3oofnum.nodesinG(s,d)andmeannum.nodesinG(s,d’)ford’<d.
�10 �5 0 5 10
Change in stock price (%)1.4
1.6
1.8
2.0
2.2
2.4
2.6
Num
.nod
es
Num. nodes | past days
Num
.Nod
es
Changeinstockprice(%)Changeinstockprice(%)
-10010
ShocksMorenodesandedges
35
Findings:Size
Findings:ClusteringCoefficient
�10 �5 0 5 10
Change in stock price (%)0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
4.5
5.0
Clu
ster
ing
coef
f.
Clustering coeff. | Num. EdgesClustering coeff. | Num. Nodes
AverageClusterin
gCo
efficien
t
Changeinstockprice(%)Changeinstockprice(%)
-10010
36
Clusteringcoefficientofanoden:thera3ooftheexis3ngandpossiblenumberofedgesamongtheneighborsofn.
�10 �5 0 5 10
Change in stock price (%)0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
4.5
5.0
Clu
ster
ing
coef
f.
Clustering coeff. | Num. EdgesClustering coeff. | Num. Nodes
AverageClusterin
gCo
efficien
t
Changeinstockprice(%)
C=4/10
Changeinstockprice(%)-10010
37
Findings:ClusteringCoefficient
Clusteringcoefficientofanoden:thera3ooftheexis3ngandpossiblenumberofedgesamongtheneighborsofn.
�10 �5 0 5 10
Change in stock price (%)0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
4.5
5.0
Clu
ster
ing
coef
f.
Clustering coeff. | Num. EdgesClustering coeff. | Num. Nodes
AverageClusterin
gCo
efficien
t
Changeinstockprice(%)
Clusteringcoefficientofanoden:thera3ooftheexis3ngandpossiblenumberofedgesamongtheneighborsofn.
HigherClusteringcoefficientShocks
C=4/10
Changeinstockprice(%)-10010
38
Findings:ClusteringCoefficient
�10 �5 0 5 10
Change in stock price (%)0.480
0.482
0.484
0.486
0.488
0.490
0.492
0.494
0.496
Stre
ngth
oftie
s
Strength of ties
Percen
tageofStron
gTies
Tiestrength:(x,y)isk-strong,ifyisamongthetopk%mostfrequentconnec3onsofx
Changeinstockprice(%)-10010
39
Findings:TieStrength
�10 �5 0 5 10
Change in stock price (%)0.480
0.482
0.484
0.486
0.488
0.490
0.492
0.494
0.496
Stre
ngth
oftie
s
Strength of ties
Percen
tageofStron
gTies
HigherDestrengthShocks
Tiestrength:(x,y)isk-strong,ifyisamongthetopk%mostfrequentconnec3onsofx
Changeinstockprice(%)-10010
40
Findings:TieStrength
�15 �10 �5 0 5 10 15
Change in stock price (%)0.805
0.810
0.815
0.820
0.825
0.830
0.835
0.840
Perc
enta
geof
bord
ered
ges
Percentage of border edges
Percen
tageofB
orde
rEdges
Borderedges:involveanoutsidecontact
Changeinstockprice(%)-10010
41
Findings:Openness
�15 �10 �5 0 5 10 15
Change in stock price (%)0.805
0.810
0.815
0.820
0.825
0.830
0.835
0.840
Perc
enta
geof
bord
ered
ges
Percentage of border edges
Percen
tageofB
orde
rEdges
MoreborderedgesShocks
Borderedges:involveanoutsidecontact
Changeinstockprice(%)-10010
42
Findings:Openness
Networks“Turtle-up”DuringShocks
Consistentwiththeoriesof:• Trust• Exper3seknowledge,repeatedinforma3onchannels• Threatrididity
• Higherclustering• Strongeredges• Moreinternalcommunica3on
43
Turtled-upnetwork
LIWCCategoriesLinguisDcInquiryWordCount(LIWC):textanalysistool,whichiden3fieswordsthatbelongtovariouscategories.
AffecDveProcessesPosi3ve Love,niceNega3ve Hurt,uglyAnxiety Worried,fearfulAnger Hate,killSadness Crying,sad
CogniDveProcessesInsight Think,Consider
Causa3on Because,HenceDiscrepancy Should,CouldTenta3ve Maybe,GuessCertainty Always,NeverInhibi3on Block,ConstrainInclusive With,IncludeExclusive But,Exclude
44
PriceChangesvs.Emo3ons
Changeinstockprice(%)�8 �6 �4 �2 0 2 4 6 8
Change in stock price (%)0.002
0.003
0.004
0.005
0.006
0.007
0.008
0.009
Perc
enta
geof
wor
dsNega3
veEmo3
ons
Changeinstockprice(%)-10010
Posi3vepricechangesHigherposi3veemo3ons
45
PriceChangesvs.Emo3ons
Changeinstockprice(%)�8 �6 �4 �2 0 2 4 6 8
Change in stock price (%)0.008
0.009
0.010
0.011
0.012
0.013
0.014
Perc
enta
geof
wor
dsPo
si3veEmo3
on
Changeinstockprice(%)-10010
�8 �6 �4 �2 0 2 4 6 8
Change in stock price (%)0.002
0.003
0.004
0.005
0.006
0.007
0.008
0.009
Perc
enta
geof
wor
dsNega3
veEmo3
ons
Changeinstockprice(%)-10010
Nega3vepricechangesHighernega3veemo3ons
Posi3vepricechangesHigherposi3veemo3ons
Emo3onsareasymmetricwithrespecttopricechange.46
PriceChangesvs.Cogni3veProcesses
�8 �6 �4 �2 0 2 4 6 8
Change in stock price (%)0.070
0.072
0.074
0.076
0.078
0.080
Perc
enta
geof
wor
dsCo
gni3veProcesses
Changeinstockprice(%)�8 �6 �4 �2 0 2 4 6 8
Change in stock price (%)0.0065
0.0070
0.0075
0.0080
0.0085
0.0090
Perc
enta
geof
wor
dsInsig
ht
Changeinstockprice(%)
PricechangesHighercogni3velanguage
Cogni3veprocessesareasymmetricwithrespecttopricechange.
Changeinstockprice(%)-10010
Changeinstockprice(%)-10010
47
48
Predic3ngSen3mentandCogni3on
Task:Forafixedstocksanddayd,predictifIMsthatmen3onsondaydcontainmorewordsinthecategorythanaverage.
Affective Cognitive Insight Neg. emo. Pos. emo.0.50
0.55
0.60
0.65
Acc
urac
yCombinedNetworkPrice changes
Accuracy
0.65
0.50
0.60
0.55
49
Predic3ngSen3mentandCogni3on
Task:Forafixedstocksanddayd,predictifIMsthatmen3onsondaydcontainmorewordsinthecategorythanaverage.
Affective Cognitive Insight Neg. emo. Pos. emo.0.50
0.55
0.60
0.65
Acc
urac
yCombinedNetworkPrice changes
Accuracy
0.65
0.50
0.60
0.55
50
Predic3ngSen3mentandCogni3on
Task:Forafixedstocksanddayd,predictifIMsthatmen3onsondaydcontainmorewordsinthecategorythanaverage.
Affective Cognitive Insight Neg. emo. Pos. emo.0.50
0.55
0.60
0.65
Acc
urac
yCombinedNetworkPrice changes
Accuracy
0.65
0.50
0.60
0.55
51
Predic3ngSen3mentandCogni3on
Task:Forafixedstocksanddayd,predictifIMsthatmen3onsondaydcontainmorewordsinthecategorythanaverage.
Affective Cognitive Insight Neg. emo. Pos. emo.0.50
0.55
0.60
0.65
Acc
urac
yCombinedNetworkPrice changes
Accuracy
0.65
0.50
0.60
0.55
52
Predic3ngSen3mentandCogni3on
Task:Forafixedstocksanddayd,predictifIMsthatmen3onsondaydcontainmorewordsinthecategorythanaverage.
Affective Cognitive Insight Neg. emo. Pos. emo.0.50
0.55
0.60
0.65
Acc
urac
yCombinedNetworkPrice changes
Accuracy
0.65
0.50
0.60
0.55
53
Predic3ngSen3mentandCogni3on
Task:Forafixedstocksanddayd,predictifIMsthatmen3onsondaydcontainmorewordsinthecategorythanaverage.
Affective Cognitive Insight Neg. emo. Pos. emo.0.50
0.55
0.60
0.65
Acc
urac
yCombinedNetworkPrice changes
Networkvariablesaremorepredic3veoftypeofcontentthanpricechanges.
Accuracy
0.65
0.50
0.60
0.55
54
Predic3ngSen3mentandCogni3on
55
Predic3ngStockTrading
56
Predic3ngStockTrading
Task:Predictwhetherastockthathasnotbeentradedforkweekswillbetraded.
Networkvariablesaremorepredic3veoftypeofsuddenstocktradingthanpricechanges.
57
Predic3ngStockTrading
Numberofweekswithoutatrade
0 1 2 3 4 5 6 7 8 9
Number of weeks without trade (k)0.50
0.55
0.60
0.65
0.70
0.75
0.80
Acc
urac
y
All features combinedPrevious Trades and NetworkPrevious Trades and PricesPrevious Trades
02468
0.80
0.70
0.60
0.50
Accuracy
Task:Predictwhetherastockthathasnotbeentradedforkweekswillbetraded.
Conclusions• Rela3onshipbetweenstockmarketshocksandsocialnetwork
structure
• Compe3nghypotheses:turtleupvs.opennetworkstructure
• Communica3on“turtles-up”duringshocks.
• Networkstructureispredic3veoftrading,performance,andemo3onalandcogni3vecontent.
• Stockmarketchangesdonotimprovepredic3onaccuracy.
58
Networkvariablesaremorepredic3veofperformancethanpricechanges.
59
Predic3ngPerformance
Num.daysconsecu3vetransac3on0246
0.58
0.54
0.50
Accuracy
0 1 2 3 4 5 6
Minimum number of days of consecutive transactions0.50
0.52
0.54
0.56
0.58
0.60
Acc
urac
y
CombinedNetworkPrice changes
SubopDmaltrade:Worsepricethantheworstpricethenextday.Task:Forafixedstockstradedondayd,predictifit’ssubop3malN-serialtrades:AtradeofstocksthathasoccurredforatleastNconsecu3vedays
Recommended