Social Network Under Stress · Social Network Temporal Dynamics Temporal dynamics of networks:...

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