A Dynamic Competitive Analysis of Content Production and Link. Formation of Internet Content Developers
|
|
|
- Brendan Booth
- 10 years ago
- Views:
Transcription
1 A Dynamic Compeiive Analysis of Conen Producion and Link Formaion of Inerne Conen Developers Liye Ma Carnegie Mellon Universiy Job Marke Paper 1 Sepember 2010 PRELIMINARY VERSION PLEASE DO NOT CIRCULATE OR CITE WITHOUT AUTHOR S PERMISSION 1 I am indebed o Baohong Sun and Kannan Srinivasan for heir guidance. I benefied from discussions wih Kinshuk Jerah, David Krackhard, Yaroslav Kryukov, and Joachim Vosgerau. I also hank seminar paricipans a Carnegie Mellon Universiy for heir helpful commen. All errors are mine.
2 A Dynamic Compeiive Analysis of Conen Producion and Link Formaion of Inerne Conen Developers Absrac The emergence of hundreds of revenue sharing conen websies has grealy conribued o he proliferaion of Inerne social media. Conen a hese websies is supplied by exernal independen developers, whom he websies arac hrough revenue sharing. This leads o a compeiion among developers, as each ries o arac viewership o her own conen. A feaure recenly inroduced a many sies, namely allowing developers o link o one anoher, leads o inriguing ineracions among he conen developers, and is impac on conen producion and overall websie viewership is lile undersood. In his sudy, we develop a dynamic oligopoly model for he compeiion among conen developers a a websie. Each developer produces conen and forms links o maximize her discouned viewership ne of cos of acions, and heir sraegic ineracion is characerized as a Markov-perfec equilibrium. Applying he wo-sep esimaor of Baari, Benkard, and Levin (2007) o he daa obained from a popular Inerne produc review sie, we invesigae he following issues: (1) why and how do developers form links? (2) Will linking encourage or discourage conen producion? (3) Wha marke srucure will emerge? (4) Will linking increase or decrease he overall websie viewership? We find ha reciprocal links are naurally encouraged by a promoe-he-promoer effec. This in urn induces developers wih more conen o sraegically iniiae links o invie reciprocaion. In addiion, we find ha link formaion affecs he incenive o produce conen developers wih more conen and unfavorable nework posiions are encouraged o produce, while developers in he opposie saes are discouraged. Furhermore, he curren linking policy may impede compeiion by giving compeiive advanage o a subgroup of conen developers, and our simulaion suggess ha limiing links could increase overall viewership by 17%. Our sudy is among he firs o examine he inerdependence beween online link formaion and conenion producion in a dynamic and compeiive seing. Keywords: Inerne conen, social media, nework, producers, dynamic game, empirical IO 1
3 1. Inroducion Conen is he lifeblood of Inerne markeing. The emergence of hundreds of revenue sharing conen websies has grealy conribued o he recen proliferaion of social media. A wide range of conen vial for online business and consumer aciviies is provided a hese websies: produc reviews a Epinions.com faciliae online reailing; video clips a Youube.com generae adverising revenue; aricles a Fool.com arac subscribers, ec. Millions of viewers visi hese websies on a monhly basis, making hem a maor componen of Inerne business (Table 1). Such websies ypically generae revenue hrough adverising or sales referral. Consequenly, heir success depends crucially on he amoun of viewership raffic hey can arac. 2 A key characerisic of such revenue sharing conen sies is he democraizaion of conen: insead of hiring employees o creae conen, companies operae hese websies as plaforms where exernal, independen developers come o supply conen. Since he success of he websies depends crucially on he viewership heir conen aracs, he websies mus encourage he independen conen developers, or producers, o produce acively. 3 To encourage conen producion, websie companies ypically share revenue wih each producer based on he viewership her conen aracs. Ineresingly, his creaes an inra-websie compeiion among he independen producers, as each seeks o maximize he viewership of her own conen, and when viewers come o he websie and choose among differen producers conen, producers effecively compee agains each oher for viewership. To arac viewership, producers naurally 2 Display adverising fee can be charged on a pay-per-impression basis, wih raes quoed in cos per milli, which is he fee for every housand imes he adverisemen is viewed, or on a pay-per-click basis, where a fee is charged every ime an adverising link is clicked. Sales referral commission is ofen charged on a pay-per-acion basis, where a conen sie is paid based on he sales i helps e-commerce sies generae by direcing viewers o hose sies. The amoun of viewership raffic is he key o all hese revenue models. 3 Boh developer and producer are widely acceped erms in he indusry, and hey are used inerchangeably in his sudy. 2
4 need o acively produce conen he more conen a producer provides, he more likely a viewer will find wha she needs from ha producer, and he higher her viewership. 4 [Inser Table 1 Abou Here] Making he compeiion more inriguing is anoher feaure ha is being increasingly inroduced o such websies: iner-producer linking. As Table 1 shows, mos such sies now allow producers o creae links poining o oher producers a he sie. Links may vary by name, such as rus, favorie, follow, ec, bu all serve as a form of endorsemen of he arge by he source, and make he arge s conen easily accessible from he source s. Such links ogeher form a producer nework ha evolves over ime. Since Inerne viewers ofen navigae hrough links o view conen, and search engines also rely on he link srucure o rank search resuls, where a producer is posiioned in his nework significanly influences he viewership of her conen. In general, he more incoming links a producer has, and he beer posiions he sources of he links have, he beer is her posiion in he nework (Brin and Page 1998). This is because incoming links drive viewership raffic o a producer s conen, and a producer wih more and beer incoming links also ges preferenial placemen when search engine displays search resuls. The inroducion of iner-producer linking leads o several inriguing quesions. Markeing research on conen and linking is sill a he early sage. Exising research has shown, in a saic and analyical seing, ha linking can promoe he posiion of he arge, and meanwhile enhance he conen of he source a viewer may visi a producer even if she does no have he desired conen, if she can poin o anoher producer who does (Mayzlin and Yoganarasimhan 2008, Kaona and Sarvary 2008). 5 However, quesions relaed o link formaion in a dynamic conex 4 Oher facors also maer, such as he qualiy and diversiy of conen, and will be accouned for in his sudy. 5 This refers o he exension on reference links in Kaona and Sarvary (2008). The main model of ha paper focuses on adverising links which are price mediaed, which does no apply o he siuaions in our sudy, as he links among conen producers a hese sies are no bough and sold bu esablished by he sources on voliion. 3
5 and he ineracion of linking and conen producion decisions largely remain open. For example, how do producers form links over ime, how do producers adus heir producion decisions under he presence of linking, and how does one respond o ohers decisions? More imporanly, from he perspecive of he websie, would allowing producers o link encourage or discourage conen producion, and would i increase or decrease he overall viewership a he websie? The obecive of he websies inroducing he linking feaure is cerainly o encourage producion and increase raffic. Bu o find ou wheher his obecive is me, we need a deailed undersanding of how conen producers inerac wih one anoher as hey compee for viewership. Considering his, we address he following quesions in our sudy: (1) Wha drives a producer s linking decisions over ime, and when and o whom would she link o? (2) Will he abiliy o form links encourage or discourage a producer o produce conen, and how does his impac vary across producers? (3) Wha marke srucure will emerge from his compeiion hrough conen producion and link formaion under a given linking policy design a a websie? (4) Finally, wha is he overall effec of linking on he viewership a he websie level, and should he websie company regulae linking? Since hese websies rely on he producers producing conen o arac viewers, ye hey can only incenivize bu canno conrol hose producers, answers o he above quesions are crucial o help he websie companies undersand conen producers decision process, draw implicaions from i, and improve heir plaform design. In his sudy, we model he compeiion among conen producers a a websie as a dynamic game. In our model, each producer chooses her acions (produce conen and link o oher producers) over ime o maximize her payoff discouned viewership ne of coss incurred in producing conen and forming links. Producers adop Markov sraegies, and such sraegies ogeher consiue a Markov-perfec equilibrium, or MPE (Maskin and Tirole 1988, Ericson and 4
6 Pakes 1995). The equilibrium characerizes he dynamic ineracions among conen producers and he radeoffs hey face. In making her decisions, a conen producer balances he cos and benefi of her acions, boh immediae and in fuure, and accouns for he sraegic reacions from oher producers, as one s acions can change he compeiive posiions of ohers. We esimae he model using he wo-sep esimaor recenly developed by Baari, Benkard, and Levin (2007). Applying he model and esimaion approach o a daase obained from a popular Inerne produc review websie, we esimae he viewership demand and cos funcions, and analyze he driving forces of producers decisions and heir implicaions. Our sudy leads o several findings. We firs demonsrae ha link formaion is a dynamic sraegic decision. We show ha he naure of he compeiion encourages reciprociy linking o someone who already links back due o a promoe-he-promoer effec. In he dynamic conex, his endency owards reciprociy furher encourages cerain producers o sraegically iniiae non-reciprocal links in anicipaion of he reciprocaion from arges, which increases viewership in fuure hrough improved posiion brough abou by incoming links. We find ha a producer wih higher conen volume is more likely o sraegically iniiae such links o invie reciprocaion. Nex, we find he dynamic effec of linking can eiher encourage or discourage conen producion, depending on he siuaions of he producers: o obain and in anicipaion of fuure rewards hrough incoming links, a producer will produce he mos conen when she has high conen volume bu low nework posiion. Meanwhile, he prospec of linking discourages a producer wih low conen volume bu high nework posiion from producing conen, as she expecs her relaive nework posiion o diminish over ime. Furhermore, our analysis suggess ha he curren linking design overall could impede compeiion. We find ha alhough boh more conen and higher nework posiion lead o higher 5
7 viewership, only he laer leads o higher ne benefi once cos is accouned for. Thus poenial advanage from having more conen is mosly compeed away, ye significan compeiive advanage is accrued o beer nework posiion. Tha a subgroup of producers enoys susainable advanage over ohers may sofen he compeiion, and lead o inefficiency from he websie s perspecive. This is confirmed in our simulaion, which suggess ha alleviaing he imbalance hrough reducing links could lead o higher overall viewership a a websie. We conribue o he lieraure by oinly modeling conen producion and link formaion decisions, invesigaing heir iner-dependence in a dynamic seing, and evaluaing he impac of linking when boh decisions are deermined endogenously. Exising sudies have analyzed he impac of commerce nework on firm profis (Sephen and Toubia 2009) wihou explicily modeling he formaion process of such nework, and modeled he formaion of conen neworks on he web in a saic seing where conen is exogenously given (Kaona and Sarvary 2008). Our sudy exends he lieraure by analyzing how linking and conen producion decisions inerac wih each oher, and we evaluae he impac of linking on websie viewership when is effec on conen producion is accouned for. Furhermore, by sudying he decision process and compeiion in a dynamic conex, we show how iner-emporal radeoffs and he sraegic ineracions among producers drive decisions over ime, which canno be shown in a saic framework, such as he sraegic inviaion of reciprocal links and he conen producion in anicipaion of incoming links from oher producers. We also conribue o he lieraure by providing a raional economic framework for empirically analyzing he formaion of links in a dynamic sraegic seing. Our empirical findings provide much needed recommendaions o indusry managers. 6
8 The res of he paper is organized as follows. In secion 2 we review relevan lieraure. We hen develop he dynamic game model in secion 3. Following ha, we discuss in secion 4 he approach used for esimaing his model. Secion 5 discusses he empirical applicaion, where we explain he daa used in our sudy, analyze he resul, and discuss he simulaion. Finally, we conclude in secion Relevan Lieraure Our work is relaed o he broad lieraure on Inerne conen and on economic neworks. Markeing researchers have shown grea ineres in Inerne conen, specifically on produc reviews and online word-of-mouh (WOM). Chevalier and Mayzlin (2006) invesigae he effec of online book reviews on sales, and find ha improvemen in reviews leads o higher relaive sales. Godes and Mayzlin (2004) find ha he dispersion of conversaion in online communiies has explanaory power on TV raings. Chinaguna e al. (2010) find he valence of online reviews influence he box-office sales of movies. While he effec of online produc reviews has been sudied frequenly, relaively less aenion has been paid o he supply of such reviews, especially when hey are supplied as informaion goods wih profi incenive. Supply-side srucural models have generally only recenly gained aenion in markeing (Srinivasan 2006), and our work fills in his gap in he case of Inerne conen. Our work is also relaed o he formaion of economic neworks and heir impacs. A rich lieraure exiss on he formaion of social and economic neworks. For example, Bala and Goyal (2000) develop a non-cooperaive game model o sudy linking decisions. Jackson (2004) gives an exensive survey on nework formaion lieraure wih emphasis on sabiliy and efficiency. Mos sudies use cerain general value funcions arising from nework; while given he wide variey of neworks, i is reasonable o expec ha he benefi of he nework, and is formaion in 7
9 urn, be siuaion specific. Two sudies in markeing focus on he creaion of links online. Mayzlin and Yoganarasimhan (2008) invesigaes why an auhor of an Inerne blog may link o anoher compeing blog, even hough doing so effecively promoes her rival. They show ha he abiliy o link o informaion is valuable o readers in addiion o he abiliy o produce he informaion if he blog does no have he informaion, readers will sill appreciae a link o anoher blog ha does. The borrowed conen effec in our sudy models his effec. Kaona and Sarvary (2008) sudy he formaion of links among conen sies as a non-cooperaive game, where links are creaed eiher for paid adverising or for reference effec in he exended model. In boh sudies, he conen a he websies is reaed as exogenous. In conras, Sephen and Toubia (2009) sudy he effec of online commercial neworks. They find ha allowing online reailers o link o one anoher creaes economic value, and such value comes from improved accessibiliy. The sudy focuses on he effec of he nework and does no explicily address is formaion process. Our sudy conribues o he lieraure by oinly sudying boh nework formaion and conen producion decisions and highlighing heir ineracion effec in a dynamic seing. Our work draws from he rich lieraure on empirical indusrial organizaions from he mehodology perspecive. Specifically, we adop he concep of Markov perfec equilibrium, or MPE (Maskin and Tirole 1988, Ericson and Pakes 1995, Maskin and Tirole 2001), for modeling dynamic oligopolisic compeiions. Early esimaion mehods for MPE (Pakes and McGuire 1994, Pakes and McGuire 2001) exend he nesed fixed poin approach (Rus 1987) o explicily compue equilibrium sraegies. Bu he high dimensionaliy of ypical dynamic compeiion models resrics he use of such mehods o games wih only few players. Recen advancemen leads o several wo-sep esimaors (Aguirregabiria and Mira 2007, Baari, Benkard and Levin 8
10 2007, Pakes, Osrovsky and Berry 2007) which exend he condiional choice probabiliies approach (Hoz and Miller 1993). Such wo sep esimaors bypass explici compuaion of equilibrium by calculaing coninuaion values hrough forward simulaion, and by doing so enable he esimaion of dynamic games wih many players. Ackerberg e al. (2007) provides a comprehensive survey of hese esimaion mehodologies. We implemen he esimaor developed in Baari, Benkard and Levin (2007), hereafer BBL. The BBL esimaor has been used for sudies in indusrial organizaions (e.g. Ryan 2009), and has been adoped in markeing lieraure recenly (Yao and Mela 2010). 3. Model We discuss he model in his secion. To prepare for he model, we begin wih a brief summary of he key elemens of he indusry seup. We consider a conen websie on he Inerne. Viewers come o he websie o view conen, which is produced by exernal, independen conen producers, whom he websie aracs hrough revenue sharing. Each conen producer seeks o maximize he viewership of her own conen over ime. In addiion o producing conen, a producer can creae links poining o oher producers. Since viewers can easily follow a link o navigae o he arge producer s conen from he source producer s, a link benefis he arge producer by puing her in a good posiion o receive viewership raffic. Furhermore, when viewers search for a specific opic and he conen from muliple producers maches ha search crieria, he search engine ranks he search resuls based on he linking srucure, where producers wih more incoming links and links from oher producers wih good posiions receive preferenial placemen. Links hus again help he arges hrough his posiional benefi. For he source of a link, he benefi is o enhance conen, as a 9
11 producer who links o oher producers makes i convenien for viewers o find he conen hey wan, and will be favored by viewers. This indusry seup leads o a compeiion among conen producers, since each producer cares abou her own viewership only, and viewers choose he conen from muliple producers. 6 To arac viewership effecively, each producer mus make her producion and linking decisions while aking ino accoun her own siuaion, oher producers siuaions, and he sraegic response o her acions by oher producers. She also needs o balance curren and fuure benefis. Such consideraions lead o ineresing dynamic ineracions. For example, more conen aracs higher viewership, bu producing conen also incurs a cos. Depending on a producer s posiion, his cos-benefi radeoff may or may no usify producion. However, having more conen may also arac links from oher producers, which improves her posiion laer on. This addiional benefi could make conen producion worhwhile, even if i does no arac much immediae viewership. Such dynamic ineracions among maximizing agens call for a dynamic oligopoly model, which we use in his sudy. In our model, here are J independen conen producers compeing for viewership. Time is discree and is indexed by, = 1,2,.... In each ime period, each producer decides wheher o produce conen and wheher o link o oher producers. In he following subsecions, we firs describe he viewership demand marke ha clears in each ime period given producers conen saes and he link srucure. We hen discuss producers dynamic conen producion and link formaion decisions, and how conen and link srucure evolve according o such decisions. 6 For example, a viewer may search for a opic, and read only he op wo aricles on he lis rerieved by he search engine. In his case, each producer wans her conen placed in he op wo posiions, and is compeing agains oher producers for ha. 10
12 Finally, we explain he dynamic compeiion and he equilibrium concep, and discuss he radeoffs faced by producers which shape heir sraegies. 3.1 Viewership Demand There are M consumers, or viewers, in each period. 7 Each viewer chooses o view he conen of one conen producer among he J producers a he websie, or chooses o go o an exernal websie, i.e. he ouside opion. This viewership consiues he demand for producers conen. We adop a logi demand model, which has been widely used in modeling oligopolisic compeiions (e.g. Berry 1994, Berry e al 1995, Dube e al 2009), o characerize viewership demand in his per-period marke. 8 The discree-choice framework of he logi demand model reflecs he compeiive naure of he viewership demand, i.e. viewership of one producer s conen may come he cos of anoher s. A viewer i s laen uiliy from reading he conen of producer in period is: (1) u i,, ui,, + ε i, = 0 + ε i,0,, = f ( C,, P,, C b, ; β ) + g( Q, Q i b, ; γ ) + ε i,, = 1.. J = 0 b b In equaion (1), u = f C, P, C ; β ) + g( Q, Q ; ) is he deerminisic componen of he i,, (,,, i, γ uiliy. C, is he conen quaniy of producer a ime, P, is a numeric measure of her nework posiion, and Q is a vecor of qualiy variables of he producer ha remains consan 7 The erms viewer and reader are used inerchangeably in his sudy. 8 The logi demand model is based on a discree-choice framework, ye i is possible ha a reader may read muliple aricles of a producer in a period, e.g., reading he produc reviews of differen producs, or he conen of several producers. An in-deph modeling of such behavior requires deailed clicksream daa of readers which we unforunaely do no have. Insead, we rea each pageview as one single viewer in our model (ha is, if a viewer reads hree produc review aricles in he period, i is couned as hree viewers in he model). This reduced-form reamen of readership demand can be improved by explicily modeling a viewer s navigaion behavior, which we leave for fuure research as richer daa become available. 11
13 over ime. 9 Furhermore, C, measures he oal quaniy of borrowed conen, i.e. conen b derived from linking o oher producers. Similarly, Q b, measures he average qualiy of he producers being linked o. These measures are explained in deail laer when we discuss producer acions and he nework srucure. The funcion f (.; β i) specifies how conen, nework posiion, and borrowed conen ener ino he uiliy funcion, wih β i as he parameer. Since viewer navigaion behavior is no explicily modeled, we esimae muliple specificaions of funcional forms for f (.; β i), wih he bes specificaion chosen hrough model selecion. The funcion g (.;γ ) capures he qualiy differeniaion among producers. Qualiy is used mainly for conrol purpose in our sudy, so we adop a linear specificaion wih γ as he parameer: b b g ( Q, Q ; γ ) = ( Q, Q ) γ. The relaive araciveness of a producer is deermined by he amoun of conen she has, i.e. he conen quaniy, he locaion of he producer in he nework, i.e. he nework posiion, and he qualiy of he producer. Furhermore, he araciveness of a producer is also influenced by he conen of he oher producers she links o. Inuiively, he more conen a producer has, he more viewership she would receive, as viewers are more likely o find he conen hey wan. Similarly, he more prominen a producer s posiion in he nework, he higher viewership demand she would receive, as her conen will receive more preferenial placemen by he search engine, and more viewers may be direced o her conen when hey navigae hrough he links. Borrowed conen should furher enhance a producer s araciveness due o he convenience benefi i affords he viewers. We expec hese o be refleced from he parameer vecor β i in 9 In our model, we rea qualiy as a characerisic of he producers insead of conen. This assumes away poenial variaion of qualiy across differen conen produced by he same producer. This is a reasonable assumpion in he conex of our sudy, since he qualiy of individual conen is no observed before a viewer decides o view he conen. 12
14 accordance wih he specific funcional form. For example, we expec all coefficiens o be posiive if facors ener he uiliy funcion linearly. Finally ε is an i.i.d random componen which follows he ype I Exreme Value i,, disribuion, resuling in he familiar logi probabiliy of viewer i choosing producer a ime : (2) Pr i,, = J 1+ exp{ u ' = 1 i,, exp{ u } i, ', } Noe ha his viewership model is a reduced form one, and assumes away any explici sae-dependence on viewer s side. In realiy, a viewer s behavior in one period may be influenced by her pas behaviors, e.g. she becomes a rouine follower of a conen producer. In our model, his dependence can come indirecly hrough he persisence of a producer s sae: a produc review of an obsolee produc produced earlier may be of no value now, bu i araced viewers a ha ime, some of whom hen coninues o visi he producer s page, and his is refleced in he uiliy funcion where a cumulaive measure of conen is used. 10 Viewers may have differen navigaion paerns and conen requiremens, which resuls in differen relaive emphasis placed on differen componens in he uiliy funcion. 11 This heerogeneiy is capured using a laen class approach (Kamakura and Russell 1989). Tha is, we assume here are N segmens of viewers, each characerized by is own se of coefficiens, β { n } n= 1.. N, and porion of each ype is denoed as λ n, so ha λ n = 1. N n= 1 10 Since he emphasis of our sudy is on producer s producion and linking behavior, srucurally modeling viewer s persisence over ime adds grea complexiy o he model bu migh no provide much added value. I also requires deailed viewer navigaion daa. We leave he oin srucural modeling of producer and consumer behavior for fuure research. 11 In he case of a sequence of page views, cerain page views may be relaed more o he page conen (e.g. following a opic search) while ohers may be relaed more o nework posiions (e.g., navigaing hrough links or using a search engine ha accouns for nework posiions). The heerogeneiy also capures his effec, since a viewer in he model acually corresponds o a viewer-page view pair in he real world, as discussed earlier. 13
15 3.2 Conen Producer In any ime period, a conen producer is characerized by a collecion of variables: b b { C,, P,, C,, Q, Q, }. Conen, nework posiion, and borrowed conen all evolve over ime according o he acions of boh producer and oher producers. A producer can ake wo ypes of acions, conen producion and link formaion. We discuss hese acions below and how he variables evolve according o hese acions Conen Producion A producer s conen quaniy, C,, is deermined solely by her own producion decisions over ime. In each period, a producer decides wheher o produce addiional conen o add o her webpage wrie anoher produc review, break anoher news sory, creae anoher analyical repor, ec and if yes, he amoun of conen o produce. We denoe his acion by producer a ime as a,, where he superscrip p indicae i is he producion decision. Specifically, p (3) a p, 0 = k do no produce conen produce k unis of conen, k {1,2...} In he equaion, k represens he number of unis of conen produced. Each uni of conen may correspond o an aricle in he real world, hus he acion is discree. Producing conen increases he conen quaniy a a producer s webpage, C,. Meanwhile, here is an opposie, depreciaion, force a work: a produc review will become less needed as he reviewed produc becomes obsolee; a news sory will become non-news afer a few days; an analyical repor will become less relevan as he siuaion expires, ec. Similar o exising lieraure modeling capaciy depreciaion (e.g. Besanko and Doraszelski 2004), we assume ha 14
16 he producer s conen a a websie depreciaes wih a cerain raio over ime. Combining he effecs of producion and depreciaion, he conen quaniy a a producer s webpage evolves as: (4) δ p C, = C, 1 + a, In equaion (4), δ (0,1) is he depreciaion rae of he conen. The smaller he value of δ is, he faser is he depreciaion. Producing conen is a cosly aciviy. We denoe he cos of producing k unis of conen prod prod by producer as c ( k, X ; φ), wih c ( 0, X ; φ) = 0, i.e. he producer incurs no cos if she does no produce conen. X is a vecor of characerisics of producer ha may affec cos, and φ is a vecor of parameers for he producion cos funcion. The producion cos is expeced o be an increasing funcion of k, he unis of conen produced. The exac funcional form of prod c (.) used in his sudy is specified in secion 5 where we discuss he empirical applicaion Link Formaion In each ime period, a producer may also creae a link poining o anoher producer, assuming one o ha producer does no already exis. 12 We denoe his acion by producer a ime as a,, where he superscrip l indicae i is he linking decision. Specifically: 13 l (5) a l, 0 = ' do no creae link creae a link o producer ', ' {1.. J}, ' 12 Links are a producer level insead of conen level, e.g. from producer A o B insead of a specific aricle of producer A o ha of producer B. 13 In our model, we consider he case where only creaion bu no removal of links is allowed. This is consisen wih he daase used in he empirical applicaion. In real-world seings, cerain websies allow link removal, while ohers do no. I is sraighforward o exend our model o allow link removal. Also, we assume ha a producer can creae only one link in a period. This assumpion is also made based on he daase used in his sudy, and i is also sraighforward o change i o allow a producer o creae muliple links in a period. 15
17 Link formaion may also be a cosly aciviy. To form a link, a producer needs o spend ime specifying so a he websie. We denoe he cos of creaing a link by producer as c link ( ', X ; ψ ). The cos may vary according o he arge of he link. For example, if reciprociy has inrinsic value, he producer will incur higher cos creaing a non-reciprocal link, i.e. links o a producer ' when ' already links back a her, han creaing a non-reciprocal one. Similar o producion cos, ψ is he vecor of parameers for he linking cos funcion. The exac funcional link form of c (.) used in his sudy is specified in secion Producer Nework and Nework Posiion The links creaed by all producers ogeher form a producer nework, which is formally represened as a direced graph. Each node in he graph corresponds o a producer, and an edge exiss if he producer corresponding o he source node has a link poining o he producer corresponding o he desinaion node. The nework evolves as producers creae links over ime. The nework a ime period is denoed as G. posiion, From he opology of he nework, a numerical measure of each producer s nework P,, can be derived. As discussed earlier, he posiion of a producer in he nework grealy influences he amoun of viewership raffic direced o her conen he more incoming links a producer s has, and from he more prominen posiions hose incoming links come, he more raffic will be direced o he producer. Thus, boh he number of incoming links and he posiions of he sources maer. The PageRank measure (Brin and Page 1998), iniially adoped by Google, eleganly capures boh effecs. Saisically, PageRank represens he probabiliy of reaching each web page in a nework when viewers follow a random walk along he links. 16
18 PageRank is equivalen o he eigenvecor cenraliy of a damped adacency-graph of he nework. Ineresingly, a rich lieraure in sociology has well esablished he imporance of eigenvecor cenraliy in social neworks (e.g. Bonacich 1987, Faus and Wasserman 1992, Wasserman and Faus 1994, Bonacich and Lloyd 2001), where higher cenraliy i is associaed wih higher presige. Recen markeing lieraure (Kaona & Sarvary 2008) has also adoped PageRank in characerizing he nework posiion of players. Following hese, we use he PageRank of each producer in he nework as he measure of her nework posiion: (6) PageRank P, =, The compuaion of PageRank is explained in he Appendix. The higher he PageRank, he more prominen a producer s posiion is in he nework. This is he nework posiion measure ha eners ino he demand funcion as specified in equaion (1). Tha incoming links increase a producer s posiion also means a producer s own posiion will reduce when she creaes a link poining o anoher producer an ougoing link increases he arge s posiion, and since posiion is relaive, i would also reduce ha of he source. This consiues a sraegic cos of link formaion, which mus be balanced wih he benefi of borrowed conen Borrowed Conen When a producer has a link o anoher producer ', he conen of producer ' can be easily accessed when a reader is viewing producer s conen. This augmens he source s conen, making he producer s webpage more appealing (Kaona and Sarvary 2008). This effec is capured in our model using borrowed conen, C,, which is simply he sum of he conen of b all oher producers being linked o a he ime: 17
19 b (7) C = C J, ', I{ ', ' } ' = 1 In he equaion, I {.} is he indicaor funcion which equals 1 if he link exiss and 0 oherwise. Similarly, he borrowed qualiy Q b, is he average of qualiy measures of he producers being linked o: J J,, ' } ' = 1 ' = 1 b (8) Q = Q ' I{ ', '}/ I{ ' 3.3 Dynamic Compeiion The compeiion among conen producers over ime is naurally modeled as a dynamic game. The key characerisic of he compeiion is ha acions aken by producers no only deermine he curren payoff, bu also affec fuure sraegic ineracions. Consequenly, when a producer makes conen producion and link formaion decisions, she needs o accoun for no only he curren benefi, bu also he fuure benefi according o he sraegic response o her acions by oher producers. In each ime period, he sae of he compeiion is fully described by a se of commonly observed sae variables. Producers ake acions o maximize heir respecive discouned payoffs. Such acions are aken based on he curren sae of compeiion and in anicipaion of he sraegic response. The soluion concep for producer s opimizing behavior is ha of Markovperfec equilibrium, or MPE (Ericson and Pakes 1995). In an MPE, he sraegy played by each producer is a Markov sraegy, where acions are fully deermined by he curren sae, and he sraegy of each producer consiues he bes response o oher producers sraegies. 18
20 3.3.1 Sae The sae a ime period, denoed as s, is he collecion of he individual conen saes of all producers and he sae of he producer nework: s = s,..., s, G ), where s = C, Q, X } ( 1, J,, {, characerizes he quaniy of producer s conen in period and he characerisics of he producer relaed o qualiy and cos, and G conains he opology of he producer nework. Noe ha s, does no include P,, as he posiion of each producer in he nework is fully deermined by he opology of he nework, which is encoded in G ; nor does i include C, or b Q,, as he b borrowed conen is deermined oinly by he opology of he nework and he conen of all producers. In anoher word, P,, C, and b Q, are derived from he sae insead of he b primiives of he sae Acion p l In each ime period, producer s acion a = a, a ) is is conen producion and link, (,, formaion decision. Le a denoe he vecor of acions aken by all producers a ime, i.e. a = a,..., a ). ( 1, J, Consisen wih exan lieraure (e.g. Rus 1987, BBL 2007), we assume ha before choosing her acion a ime, each producer receives an acion-specific privae shock ν a ) ha is independen among producers and over ime. Since in our seing he acions are, (, discree, his privae shock is a vecor where each elemen corresponds o a specific acion ha can be aken a he ime. Also consisen wih exan lieraure, we assume he privae shock follows an exreme value disribuion. This privae shock is needed in dynamic game models o 19
21 accoun for he variabiliy in acions ha goes beyond he observed saes. The collecion of acion-specific privae shocks across all producers is denoed as ν = ν,..., ν ) Payoff ( 1, J, In each ime period, according o he viewership marke demand and producer acions, producer s curren-period payoff is: N prod p link l (9) π ( a, s, ν, ) = mr Mλn Prn, ( s ) c ( a,, X ; φ) c ( a,, X ; ψ ) + ν, ( a, ) n= 1 In equaion (9), mr is he marginal benefi associaed wih each viewer visi, and Mλ n is he number of viewers in segmen n. In each period, he payoff of producer is he benefi of viewership demand ne of any cos associaed wih he acion aken by he producer. Each producer is concerned no us wih he payoff of he curren period, bu also he overall payoff over ime. The oal discouned payoff o producer a ime, which he producer seeks o maximize, is: (10) E[ τ β π ( aτ, sτ, ν, τ ) s ] = τ In equaion (10), β [0,1) is he discoun facor. The expecaion is over he privae shock, producers acions in he curren period, as well as fuure saes, acions, and privae shocks. As is shown clearly in he equaion, he payoff o a producer depends on no only her own acions, bu also he acions of oher producers. This leads o sraegic ineracions which are characerized using an MPE. 20
22 3.3.4 Sraegy and Equilibrium We assume all producers follow Markov sraegies. A Markov sraegy profile σ of he dynamic game is he collecion of he sraegies of all producers: σ = ( σ1, σ 2,..., σ J ) where σ is he sraegy played by producer which depends on he sae and he privae shock, σ S ν A : a, where S is he se of all saes, ν is he se of privae shocks and A is he se of all acions producer can ake. Given a sraegy profile, a producer s value funcion is he expeced discouned payoff given he sae, inegraed over privae shocks. I can be wrien recursively as follows: (11) V ( s; σ ) = Eν [ π ( σ ( s, ν ), s, ν ) + β V ( s'; σ ) dp( s' σ ( s, ν ), s) s] When choosing a sraegy, a producer needs o ake ino accoun no only he curren sae, bu also oher producers sraegies. Following convenion in lieraure, we use σ o denoe he sraegies played by all producers oher han producer. A producer s opimizaion problem is: (12) V ( s; σ ) = max{ E [ π (( σ ( s, ν ), σ ( s, ν )), s, ν ) + β V ( s'; σ ) dp( s' ( σ ( s, ν ), σ ( s, ν )), s) s]} σ ( s, ν ) ν The sraegy which is he soluion o equaion (12) for producer is he bes response of * * * * he producer o ohers sraegies. An MPE is a sraegy profile σ = ( σ, σ,..., ) where each producer s sraegy is he bes response o oher producers sraegies. Tha is, in an MPE, when holding he sraegies of oher producers unchanged, no producer can increase is own expeced payoff by unilaerally deviaing o anoher sraegy: 1 2 σ J * * * (13) V ( s; σ, σ ) V ( s; σ, σ ), s, σ 21
23 Wih observaions of viewership demand and producer acions according o he saes over ime, we can esimae he parameers for he viewership demand model and he dynamic srucural parameers, i.e. cos parameers, using he opimaliy condiion implied by he equilibrium, which we discuss in deail in secion Iner-emporal Tradeoffs We now qualiaively discuss he radeoffs conen producers face in heir producion and linking decisions which are incorporaed in he model. When deciding wheher o produce conen, producers obviously face a radeoff beween he cos incurred in producing conen and he viewership such conen aracs over ime. Furhermore, here are several radeoffs induced by linking, which lead o ineresing ineracions among producers. To begin wih, when linking o anoher producer, a producer faces he radeoff beween borrowing he conen of anoher producer and lower nework posiion arising from promoing her compeior. Depending on how much he borrowed conen will help and how severely he link will reduce her own nework posiion, he producer may or may no find i worhwhile o form a link. Ineresingly, when we ake his radeoff a sep furher, o consider no only wheher o form a link bu also whom o link o, we can see his radeoff provides a simple explanaion o a well known phenomenon in neworks: he endency o form reciprocal links. Reciprociy can be explained by social norm in sociology lieraure (Gouldner 1960), and hrough reward and punishmen schemes in repeaed games (Axelrod and Hamilon 1981). In he seing of our sudy, however, reciprociy may arise naurally from he consideraion of nework posiion. To see his, recall ha he source s nework posiion posiively influences he arge s. Suppose producer A wans o creae a link, and producer B already has a link o producer A while producer C does no. Then if A links o B, hereby improving B s posiion, he enhanced posiion of B will be parially carried over o A. 22
24 Whereas if A links o C, who is no A s source, hen A will no ge his indirec benefi. Oher hings equal, his promoe-he-promoer effec would favor reciprocal links over non-reciprocal ones. 14 Tha is, i is beer o promoe one s own promoer insead of anoher producer. More radeoffs come ino play when we consider he ineracions among producers over ime. When making linking decisions, since a reciprocal links are naurally encouraged, a forward-looking producer may inenionally creae a non-reciprocal link, if she expecs ha he producer she links o would reciprocae in he near fuure. Tha is, a producer may sraegically creae a link o invie reciprocaion. The radeoff she faces in his decision is beween lower nework posiion now and higher nework posiion laer on, if and when he arge reciprocaes. Furhermore, he prospec of linking may also encourage or discourage conen producion. A producer may be encouraged o produce more conen han she oherwise would, if she expecs ha by producing more conen, she can arac incoming links from oher producers laer on. The radeoff she faces in his decision is beween he cos of producing conen now and beer nework posiion laer on when she receives incoming links. A he same ime, if a producer expecs her compeiors o receive incoming links, which diminishes her relaive nework posiion over ime, she may produce less han she oherwise would. All hese radeoffs play a cenral role in deermining conen producion and link formaion decisions, and lead o he equilibrium sraegy adoped by conen producers. 4. Esimaion Our esimaion requires ha he conen producion and link formaion decisions of all producers over a number of ime periods are observed, so is he per-period viewership of each producer s 14 A Mone-Carlo simulaion using random graphs will easily show ha, on average, he reducion in nework posiion hrough creaing a reciprocal link is less han ha hrough creaing a non-reciprocal link. 23
25 conen in muliple ime periods. The parameers o be esimaed are he segmen-specific viewership demand coefficiens and he sizes of he segmens, he qualiy coefficiens, he conen depreciaion rae, he marginal benefi o he producer per reader visi, and he cos parameers of conen producion and link formaion, as summarized below: Param = ({ βn, λn} n= 1.. N, γ, δ, mr, φ, ψ ) The marginal benefi and he cos parameers are no oinly idenified. Considering his, we normalize mr = for idenificaion, which implies ha he uni of accoun for cos is he marginal benefi per housand views. 15 The firs half of he parameers, βn, λn} n= N, γ, ), are ({ 1.. δ he parameers governing he viewership marke in each period. The idenificaion ress on he cross-secional and iner-emporal variaion of he conen and nework of producers, ogeher wih he corresponding variaion of viewership. The second half of he parameers, ( φ, ψ ), are he dynamic srucural parameers ha ogeher wih he viewership demand parameers govern he dynamic compeiion, he idenificaion and esimaion of which res on he opimaliy condiion of he equilibrium. Esimaing dynamic games is challenging due o curse of dimensionaliy he sae space has high dimensionaliy as i incorporaes he saes of all players. Early esimaion mehods (e.g. Pakes and Mcguire 1994) rely on explicily solving for equilibrium hrough value-funcion ieraion, and have limied scalabiliy. Recenly developed wo-sep esimaors call for esimaing as many srucural parameers offline as possible, and bypassing he compuaion of equilibrium when esimaing he dynamic srucural parameers. Our esimaion is implemened using one such wo-sep esimaor as specified in BBL (2007). BBL approaches he esimaion ask in wo sages. In he firs sage, we recover he equilibrium sraegy of producers in reduced form, based 15 This follows he indusry sandard on display adverising, where fees are quoed as cos-per-mille, or CPM, which represens he amoun an adveriser needs o pay for every housand imes an adverisemen is displayed o viewers. 24
26 on observed saes and acions. Esimaion of he equilibrium sraegy, also ermed he policy funcion, should srike he righ balance beween flexibiliy and daa availabiliy. A flexible funcional form is desired for accurae represenaion of he equilibrium sraegy, bu i also requires more daa. The second ask for he firs sage is o esimae he ransiion of saes over ime according o producer acions. The viewership demand will also be esimaed in he firs sage. In he second sage, using he knowledge of policy funcion, sae ransiion, and viewership demand esimaion in he firs sage, we perform forward-simulaion of he observed policy versus perurbed policies. As he observed policy consiues an equilibrium, he opimaliy condiion dicaes ha a producer s payoff when she plays he equilibrium sraegy is no less han her payoff under an alernaive perurbed sraegy, while oher producers sill follow he equilibrium sraegy. This opimaliy consrain forms he basis for consrucing he obecive funcion of a GMM esimaor. As is common in research on empirical dynamic games, we focus on symmeric pure sraegy equilibrium. Such resricion allows us o pool daa across all producers, which reduces daa requiremen and improves esimaion efficiency. 4.1 Firs Sage In he firs-sage of he esimaion, we recover he policy funcion, he sae ransiion process, and he viewership marke demand sysem Policy Funcion In equilibrium, each producer chooses her acion based on her own sae as well as he saes of oher producers and he producer nework. In he firs sage of esimaion, we recover his policy funcion, * σ, which maps saes o acions, in reduced form. BBL recommends using flexible 25
27 funcional forms o approximae he equilibrium policy wih precision, which needs o be balanced wih daa availabiliy. Facing his radeoff, we firs ransform he sae space by deriving he vecors of nework posiions and borrowed conen of all producers from he conen sae of each individual producer and he nework srucure hese are he variables ha ener he uiliy funcions direcly. We hen pariion he ransformed sae space of an individual producer ino quiniles along boh he conen dimension and he nework posiion dimension. For each cell in his pariioned sae space, we run a separae se of regressions wih producer acions as dependen variables. The independen variables include he qualiy and cos relaed characerisics of he producer, he borrowed conen and qualiy of he producer, he number of oher producers in each cell of he pariioned sae space, and he average qualiy of oher producers. 16 Since linking acions differ by desinaion, we disinguish he arge on he following four dimensions: reciprociy, conen, nework posiion, and qualiy. We separae a reciprocal link from a non-reciprocal one, and for each of he oher dimensions, we perform a median-spli on he arge: separae a high conen producer (whose conen quaniy is above median) from a low conen one (below median); separae a high nework posiion producer from a low nework posiion one; separae a high qualiy producer from a low qualiy one. 17 There are hus sixeen differen ypes of linking arges, which combined wih an acion of no-link resuls in seveneen possible linking acions. We esimae each regression funcion using generalized linear models, wih log link funcion for conen producion and logi link funcion for link formaion. 16 Essenially we are esimaing he policy funcion nonparamerically on a producer s own sae bu paramerically on oher producers saes. Ideally, he policy funcion should be esimaed nonparamerically over he enire sae space, bu he high dimensionaliy of he sae space makes his impracical, as o do so requires enormous amoun of daa. BBL suggess using local linear regression, which is similar o wha we do here. 17 Since qualiy aribues are consan over ime in our model, he muli-dimensional qualiy measure of a producer can be reduced o a single dimensional number once he viewership demand is esimaed, by weighing based on he esimaed coefficiens. 26
28 The se of regression funcions hrough his esimaion fully describes he sraegy played by each producer in equilibrium. These policy funcions form he basis for forward-simulaion ha is used in he second sage of he esimaion o recover dynamic srucural parameers Sae Transiion Sae ransiion probabiliies are needed for performing forward-simulaions in he second sage of esimaion. In our model, he ransiion of saes given he acions of all producers is deerminisic linking acions deerminisically change he nework srucure, while producion acions ogeher wih depreciaion deerminisically change conen sae. Consequenly, sae ransiion does no need o be esimaed once he policy funcion is recovered. In he second sage forward simulaion, we simply simulae producer acions based on he esimaed policy funcion, and sae ransiion can be calculaed deerminisically once he acions are simulaed Viewership Demand The viewership marke demand in each period can be esimaed raher sraighforwardly wih MLE. Denoe s N m N, ( s;{ n, βn} n= 1, γ ) = λn Prn, ( s; βn, γ ) n= 1 λ as he heoreical marke share of producer a ime given he sae and he parameers, and he acual marke share observed from daa as s ˆ m,. 19 m m Assuming ha he difference η, = log s ˆ, log s, follows an i.i.d. normal N disribuion (Holmes 2009), he parameers {{ λn, βn} n= 1, γ} can be esimaed using maximum likelihood. 20 The marke size, i.e. he oal number of viewers, M, is needed for calculaing 18 Conen producion is similar o invesmen in empirical IO, where sudies also use probabilisic sae ransiion models (e.g. Besanko and Doraszelski 2004). The difference is minor, as a radeoff beween he precision of saes and he precision of sae ransiion. Our model allows for deerminisic sae ransiion because he exac conen sae and he sae of he producer nework are used. 19 The superscrip m represens marke. This is o avoid confusion wih he same symbol s ha represens producer sae. 20 For he case of one viewer segmen only, his is he same as he inversion suggesed in Berry (1994). 27
29 marke share, and is assumed o be observed. 21 The conen depreciaion parameer, δ, could be esimaed eiher oinly wih he oher parameers of he viewership marke demand equaion, or separaely in an offline manner. 4.2 Second Sage We now discuss he second sage esimaion of he dynamic srucural parameers, i.e. cos of producing conen and forming links. The key o he second sage esimaion is he opimaliy * * * * condiion of an equilibrium: given he equilibrium sraegy profile σ = ( σ, σ,..., ), for any alernaive sraegy σ ' for an arbirary producer and a randomly chosen sae s, he equilibrium condiion dicaes ha: * * ' * (14) V ( s, σ, σ ; φ, ψ ) V ( s, σ, σ ; φ, ψ ) Given a specific σ * ', a uple x = {, s, σ } indexes one such equilibrium condiion. Following BBL s noaion, define * * ' * (15) g( x; φ, ψ ) = V ( s, σ, σ ; φ, ψ ) V ( s, σ, σ ; φ, ψ ) And define obecive funcion 2 (16) Q( φ, ψ ) = (min{ g( x; φ, ψ ),0}) dh ( x) where H is a disribuion over he se X of he equilibrium condiions. Then he rue parameer φ, ) saisfies: ( 0 ψ 0 (17) Q( φ, ψ ) = 0 = min Q( φ, ) 0 0 ψ φ Φ, ψ Ψ I The esimaion is he empirical counerpar of his condiion: le { X n k} be a se of k = 1 n I ' randomly chosen opimaliy condiions. For each X = {, s, σ }, we calculae he payoff of k k k k 1 2 σ J 21 Changing he marke size will change only he consan erm of he esimaed demand parameers. 28
30 ˆ * s k k k k * he focal producer k when she follows he equilibrium sraegy, V (, σ, σ ; φ, ψ ), and ha ˆ * s k k k k ' when she follows he alernaive sraegy, V (, σ, σ ; φ, ψ ), for a proposed parameer value ( φ, ψ ). The empirical counerpar of he obecive funcion is hen (18) 1 Qn ( φ, ψ ) = n 1 = n n I I k = 1 n I I k = 1 (min{( Vˆ (min{ gˆ( X k ( s, σ k * k k, σ ; φ, ψ ),0}) * k 2 ; φ, ψ ) Vˆ k ( s, σ k ' k, σ * k ; θ ), φ, ψ }) 2 BBL shows ha Q n(.) can be calculaed hrough forward simulaion, and he parameer ha minimizes he obecive funcion (19) ( ˆ, φ ψˆ ) = arg min ( φ, ψ ) Q n φ Φ, ψ Ψ is a consisen esimae of he rue parameer under mild regulariy condiions. This recovers he mean esimae of he parameer, while he sandard error can be calculaed using re-sampling of hese equilibrium condiions. 5. Empirical Applicaion 5.1 Daa Our daa is obained from a popular online produc review websie, which in recen years consisenly aracs several million visiors on a monhly basis. A produc reviewer can sar wriing produc reviews once she creaes an accoun a he websie. The producs reviewed a he websie range from auomobiles o oys, books, and movies, ec. Such reviews correspond o he conen in our model. In addiion o wriing produc reviews, a reviewer can also link o oher reviewers by puing hem ino her lis of rused reviewers. Creaing a link is solely a he discreion of he source reviewer, wihou he need for consen from he arge reviewer. Such 29
31 links ogeher form a so-called web of rus among reviewers, and his corresponds o he producer nework in our model. Viewers can easily navigae hrough he rus links o go from one reviewer s reviews o he reviews of anoher reviewer whom she russ. Furhermore, produc reviews wrien by reviewers who are rused by many oher reviewers, and rused by reviewers who are hemselves rused by oher reviewers, will receive preferenial placemen when viewers search he websie. The posiion of a reviewer in his web of rus hus heavily influences he likelihood of her reviews being accessed by viewers. 22 Alhough here are housands of reviewers wriing reviews a he websie, in his sudy we focus on a small group of he mos acive ones, known a he websie as he op reviewers. These op reviewers wrie produc reviews frequenly and consisenly over ime, and hey are paid by he websie based on he viewership heir reviews arac. This group of elie reviewers is suiable for he model we developed earlier, as hey are likely dedicaed producers who are driven by profi incenive and who choose heir acions sraegically. 23 This small group of reviewers also is responsible for a significan share of he websie viewership raffic. 24 Focusing on his group also eases he esimaion of he model, as he number of players is kep a a reasonable level, and a long hisory of conen producion and link formaion decisions is available for hese acive producers. Our daa se conains he decisions of wriing reviews and creaing links a he daily level, from June 2008 o March I also conains he viewership informaion saring from November 2009: for each four-day period saring from November 2009, he number of imes 22 In an inerview, he former CTO of he company said of he rus sysem based on anecdoal evidence, hose who have sared using i end up compleely depending on i o navigae he sie. 23 Tha is, as compared o oher occasional users who wrie reviews infrequenly, and who may be driven by oher incenives such as a sponaneous desire o express one s opinion, for which a sraegic framework may no be applicable. 24 Comparing he viewership saisics of his group of reviewers wih he websie level saisics suggess hey are responsible for abou 30% of he overall viewership. 30
32 each reviewer s reviews is visied is recorded. 25 There are a oal of 199 op reviewers a he sie. Among hem, 6 lef he sie during he period, and we exclude hem from he daa se. [Inser Table 2 Abou Here] The summary saisics are repored in Table 2. As is shown in he able, hese op reviewers are highly acive in wriing reviews, averaging one review aricle per reviewer abou every hree days. In addiion o wriing reviews, hey also creaed over wo housand links over he period, alhough he frequency of creaing links is lower han wriing reviews, wih each reviewer adding a link roughly every wo monhs. These op reviewers ogeher arac a large audience, oaling more han six million view couns over a period of abou four monhs. Comparing wih websie level raffic informaion, we know ha his small group is responsible for abou 30% of he oal visi a he websie, a significan share. Based on he informaion available a he websie, we use hree variables for he qualiy facors in our model: diversiy, popular, and advisor. These hree variables and heir summary saisics are described in Table 3. Togeher, hese facors cover hree imporan aspecs which can affec viewership demand in addiion o conen volume and nework posiion: diversiy, populariy, and qualiy. [Inser Table 3 Abou Here] 5.2 Resul Viewership Demand We firs esimae he model of viewership demand as deermined by each reviewer s conen, nework posiion, borrowed conen, and qualiy facors. As discussed earlier, he conen depreciaion parameer, δ, can be esimaed ogeher wih he oher viewership demand 25 The Websie displays he cumulaive view coun a he reviewer level and he informaion is updaed daily. However, he updae is no well synchronized for all reviewers. Thus we aggregae he informaion ino 4-day periods o eliminae he noise creaed by his echnical issue. 31
33 parameers. In our daase, however, he conen producion and link formaion daa covers a much longer period han he daa for viewership marke demand. Furhermore, he overall viewership a he websie has remained fairly sable over he period for which we observe he acions. Therefore, we esimae his depreciaion parameer offline, by reaing i as a discoun facor and finding he value ha bes keeps he conen quaniy sable over ime. We arrive a he esimae δ = his way. The summary saisics of nework posiions, and hose of he effecive discouned conen and borrowed conen, boh calculaed according o he depreciaion parameer, are repored in able 4. The marke size M is se o be wice he average oal visi couns a he websie o allow for subsiuion effec among compeiors websies. Websie level saisics show ha here were on average 5.2 million views per monh, which resuls in M = Changing his marke size will change he consan erm of he uiliy funcion wihou affecing oher parameers. 26 [Inser Table 4 Abou Here] As discussed in secion 3, muliple funcional forms of he funcion f (.; β i) in equaion (1) need o be esimaed, wih he bes model chosen wih cerain model selecion crierion. This flexibiliy is imporan because our reamen of he viewership marke is reduced form, so esimaing muliple funcional forms can give us more robus resuls. We esimae he following four specificaions: (20) I( Linear) II( Linear Quadraic) III( Log) IV ( Log Embedded) f ( C f ( C f ( C f ( C,,,,, P, P, P, P,,,,, C, C, C, C b, b, b, b, ; β ) = β C i i,1 ; β ) = β C i i,1 ; β ) = β log( C i i,1 ; β ) = β log( C i i,1,, + β P i,2, 2 i,2, + β C,, ) + β + β C + β C i,3 + β P i,2 b i,3, b, i,3, + β P log( P ) + β log( C, ) + β log( P i,2 2 i,4, i,3, ) + β C i,5 b, ) b, + β C i,6 b 2, 26 The websie was esablished in 1999 and is a maure sage now. The websie level viewership remained fairly sable over he observaion period, hus we do no consider he growh of marke size in his sudy. 32
34 Specificaion I is he simples funcional form ha accouns for all hree facors, and we expec each coefficien o be posiive o reflec heir posiive impac on viewership demand. Specificaion II exends he firs specificaion by including a quadraic erm for each facor o accoun for poenial diminishing rae of reurn. For example, alhough linking o oher producers provides a convenience benefi o viewers, when here are oo many such links, viewers could also feel annoyed, so he conen borrowing effec could become sauraed. Similarly, alhough having higher nework posiion gives a producer s conen favorable placemen, his benefi may become sauraed beyond a cerain hreshold, if he nework posiion is high enough o disinguish he producer in mos cases. The quadraic erms are used o capure such effecs. Specificaion III explicily accouns for such diminishing reurn by using log ransformaion. Finally, Specificaion IV also uses log ransformaion, bu adds a weighed componen of he borrowed conen o he original conen before applying he log. The qualiy facors are included in our sudy mainly for conrol purposes, and we adop a simple linear funcional form for he qualiy as well as borrowed qualiy: (21) g( Q, Q ; γ ) = γ + γ Diversiy b 0 γ Diversiy 4 1 b + γ Popular + γ Advisor + + γ Popular 5 2 b + γ Advisor 6 3 b The resul of esimaion is presened in Table 5 (covariaes are sandardized). In all four specificaions I-IV, he coefficiens for conen, nework posiion, and borrowed conen are all posiive and saisically significan. This is clear evidence ha all hree are imporan facors in deermining viewership demand, where higher conen volume, more prominen nework posiion, and more borrowed conen all lead o higher viewer uiliy and in urn higher viewership demand for he reviewer s reviews. The coefficiens for he hree qualiy facors are also all 33
35 posiive and saisically significan, suggesing hey posiively influence viewership demand. Among hem, he populariy indicaor has he highes impac on viewer uiliy. [Inser Table 5 Abou Here] Looking a specificaion II, we find ha conen and borrowed conen have similar conribuions o he viewer uiliy, while nework posiion has higher impac han boh conen and borrowed conen. Specificaion II shows he quadraic erms of borrowed conen and nework posiion boh have negaive signs, suggesing ha diminishing reurn exiss for boh facors. The quadraic erm of conen is also negaive. However, is magniude is very small and i is no saisically significan. Thus here is no clear evidence of diminishing reurn on he conen dimension. Specificaions III and IV boh use log ransformaion, where he coefficien magniude corresponds o percenage change. Specificaion V is he laen class version of specificaion II wih wo segmens. In boh segmens, boh nework posiion and borrowed conen posiively influence viewer uiliy and exhibi diminishing reurns. The firs segmen has conen coefficien larger han ha in specificaion II. Ineresingly, he second segmen has a negaive conen coefficien, and he coefficiens for nework posiion and borrowed conen are quie large. This seems o sugges ha his porion of he demand is mainly driven by he posiion in he nework and he borrowed conen, bu no by he producer s own conen. Among he five compeing model specificaions, specificaion II, he Linear-Quadraic specificaion, has he bes model fi afer adusing for number of parameers using BIC. We herefore adop his specificaion as he per-period viewership demand equaion for he esimaion of dynamic srucural parameers. 5.3 Resul Dynamic Compeiion Policy Funcion 34
36 The policy funcion regression, which capures reviewers wriing and linking decisions, is only he inermediae sep for esimaing he dynamic model parameers, and he coefficiens are no inerpreable. Insead, we repor a few paerns of producer acions based on heir conen and nework posiion saes. 27 Noe ha he policies, esimaed in a reduced-form fashion, consiue he equilibrium play resuling from reviewers dynamic compeiion, and encapsulae he concep of bes response. In his secion we simply presen he observed paerns. In he subsequen secion 5.4, we invesigae in deail how incenives and sraegic ineracions lead o such acions. Figure 1 shows he average daily conen producion, condiional on he reviewer s own sae along he conen and nework posiion dimension. As shown in he figure, reviewers wih higher conen volume in general wrie more frequenly, and i is more so for reviewers wih low nework posiions. In fac, reviewers wih high conen volume bu low nework posiion wrie reviews mos frequenly. This could be unexpeced a he firs look he viewership demand equaion, which capures he payoff hrough immediae viewership, shows ha reviewers wih higher nework posiions have higher marginal benefi and hus should have higher propensiy o produce conen. In secion 5.4, we show how his discrepancy is explained wih he dynamic radeoffs faced by reviewers. [Inser Figures 1, 2, and 3 Abou Here] Figure 2 shows he frequency of creaing an ougoing link, condiional on reviewers own sae. Reviewers wih higher conen volume creae links more frequenly. Reviewers wih very high nework posiions (5-h quinile) also creae links wih have higher frequencies, alhough no by much. 27 Acions can be summarized according o oher dimensions, oo, such as qualiy. In his sudy, we focus on he wo dimensions, conen and nework posiion, as hey are he direc resuls of reviewers review wriing and link formaion acions. As specified in secion 4, conen and nework posiions are each pariioned ino quiniles for he policy funcion regression, so we repor he acion paerns based on he quinile pariions along hese wo dimensions. 35
37 Since our analysis in secion 3.4 indicaes ha reciprocal links would be favored by reviewers, we also repor he relaive probabiliy of creaing a non-reciprocal link over ha of a reciprocal one condiional on a reviewer s own sae, as shown in Figure 3. The firs o noe from he figure is ha reviewers of all saes are much more likely o creae reciprocal links han nonreciprocal ones he raios are all much smaller han 1. Furhermore, reviewers wih higher conen volume have higher relaive probabiliy o creae non-reciprocal links. Oher paerns are ha reviewers wih more conen are more likely o receive incoming links, and ha reviewers wih differen nework posiions have similar likelihood of receiving incoming links as long as hey have similar conen, wih higher nework posiions increasing he likelihood bu only slighly. Togeher, hese paerns summarize he decisions made by reviewers as hey inerac wih one anoher in he compeiion, each rying o maximize her own benefi. The incenives behind hese acions are analyzed in deail in secion Cos Esimaion We now discuss he esimae of he dynamic srucural parameers, i.e. cos parameers. To operaionalize he esimaion, we randomly pick 500 saes from he daase. For each sae, we randomly pick one reviewer and performed wo forward-simulaions. In he firs, all reviewers follow he equilibrium sraegy according o he esimaed equilibrium policy, while in he second simulaion, he chosen reviewer follows a perurbed sraegy. Each simulaion is run for 600 periods, and repeaed muliple imes wih he average aken. We se he discoun facor o as our observaion is a daily level, which is similar o he ofen se in dynamic srucural sudies when daa is a weekly level (e.g. Erdem & Keane 1996). We hen run he minimum disance esimaor o find he cos parameers which minimize he deviaion from he 36
38 opimaliy condiion of equilibrium, as specified in equaion (19). The sandard errors of he esimaes were obained hrough re-sampling of he chosen sae-player pairs. For he producion cos funcion, we adop a linear funcional form. 28 Producion cos may depend on he reviewer s qualiy, as a reviewer needs o exer more effor o achieve higher qualiy. The reviewer s enure migh also influence cos, due o learning-by-doing. Considering his, we assume he uni cos of producion is a linear funcion of he reviewer s effecive qualiy and enure wih he websie, as shown in equaion (22). We also assume he cos of linking is a linear funcion of he reviewer s effecive qualiy and enure, plus an indicaor of wheher he link is reciprocal, as shown in equaion (23). This final erm is added o ease ou possible inrinsic value of forming reciprocal links an inrinsic preference for reciprocal links, aside from he consideraion of how i affecs viewership, would imply lower cos of forming reciprocal links ha non-reciprocal links a he model primiive level, and be refleced from a negaive coefficien for his final erm. prod p p (22) c ( a,, X ; φ ) = a, ( φ0 + φ1g ( Q, γ ) + φ2tenure) link l l (23) c a, X ; ψ ) = ψ + ψ g( Q, γ ) + ψ Tenure + ψ I{ a is reciprocal} (, , [Inser Table 6 Abou Here] The resul of he esimaion is repored in Table 6. The consan erm for he producion cos regression is and saisically significan a.95 level. This means he cos of wriing a review aricle is equivalen o he benefi of 148 page views, which is a reasonable number for A sricly convex cos funcion is ofen used in indusrial organizaion lieraure. In our empirical applicaion, however, i is reasonable o assume here is a uni cos for wriing a review aricle, hence he linear form. Equilibrium condiion holds as long as he marke size is finie. We also esimaed he quadraic specificaion of cos funcion, and he resul when averaged for uni cos is similar o he linear specificaion. The resul for quadraic cos funcion is available from he auhor upon reques. 29 In a sligh abuse of noaion, we use he same funcion symbol, g, o represen a reviewer s own qualiy effec: g( Q ; γ ) = γ 0 + γ 1Diversiy + γ 2Popular + γ 3Advisor, excluding he borrowed qualiy effec cos should be deermined by a reviewer s own characerisics. 37
39 uni cos esimae, as he summary saisics show ha on average a review aricle is viewed a lile over 200 imes. The coefficien for reviewer qualiy is posiive and saisically significan. This suggess ha reviewers of higher qualiy pu more effor in wriing produc review aricles and hus incur higher cos per aricle wrien, consisen wih expecaion. 30 The esimae also shows ha a reviewer s enure a he websie does no have a significan impac on her producion cos. The cos of linking is very close o zero, indicaing ha linking iself is no a high effor aciviy. Neiher qualiy nor enure is shown o have a significan effec on he cos of linking. More noable is he coefficien for he reciprociy erm. The posiive sign of he coefficien shows ha he cos of forming a non-reciprocal link is less han ha of forming a reciprocal link, alhough he resul is no saisically significan. As discussed earlier, he exisence of inrinsic value for reciprocal links would be refleced from a negaive coefficien for his erm, hus here is no evidence of such inrinsic value. Tha reciprocal links are more likely o be formed, as observed in he daase, hus should be mainly aribued o he sraegic consideraions, i.e. he promoe-he-promoer effec as discussed in secion Decision Dynamics and Inerdependence Using he esimaed viewership demand equaions, he dynamic cos parameers, and he equilibrium policy, we now invesigae he compeiive dynamics in deail. To address he research quesions raised for his sudy, we analyze hree aspecs of he compeiive dynamics: Firs, we invesigae he incenive o form links and how i depends on link ypes and reviewer 30 A more general model is o assume ha all reviewers are of he same ype, and ha when hey wrie aricles hey can choose o wrie eiher a high or a low qualiy one, wih he former enailing higher cos han he laer, similar for links. However, o esimae such a model requires qualiy informaion a he level of each review aricle, which we do no have. This is beyond he scope of our sudy and is lef for fuure work. Our model can be considered as a resriced model in his broader conex each reviewer is resriced o choose a qualiy ype and hen follow i hroughou he whole period. 38
40 saes. Nex, we analyze how linking influences conen producion decisions. Finally, we evaluae he ne benefi accrued o reviewers a differen saes and he marke srucure ha emerges from he compeiion Dynamics of Link Formaion The decision of wheher o creae a link and whom o link o is driven by boh he radeoff beween borrowed conen and nework posiions, and he dynamic ineracions beween reviewers. To undersand he incenives o form links for reviewers a differen saes, we evaluae how such links impac reviewers viewership demand. To analyze he implicaion of forming links, we firs quanify, using he daase, he average change in nework posiion hrough esablishing an ougoing link and ha hrough receiving an incoming link given a reviewer s sae. Receiving an incoming link normally increases he reviewer s nework posiion noiceably. Creaing an ougoing link, however, reduces he nework posiion, and a reciprocal link ypically leads o smaller reducion han a non-reciprocal link as discussed earlier. We hen use a subse of he daa, covering he hreemonh period from January 2009 o March 2009, o calculae he incremenal benefi of creaing a link for each reviewer in each day. For creaing a reciprocal link, he incremenal benefi is calculaed as he difference in discouned viewership beween wo oherwise idenical scenarios excep ha in he second scenario he focal reviewer creaes a reciprocal link o anoher reviewer who already links o her. Oher facors are held consan in his calculaion. This calculaion capures he effec of creaing a reciprocal link, which can be considered as a close-loop acion. 31 For creaing a non-reciprocal link, however, his calculaion capures only he direc 31 We can consider ha a reciprocal link finishes a round of dynamic ineracion he arge reviewer already has a link poining back and will no furher respond o he reciprocal link. Thus a loop is closed. In conras, a nonreciprocal link sars a round of dynamic sraegic ineracion he arge reviewer will in subsequen periods decide wheher o reciprocae. Thus a loop is opened. 39
41 effec, i.e. he radeoff beween more borrowed conen and lower nework posiion, bu no he sraegic aspec arising from dynamic ineracions, i.e. he arge reviewer may decide o reciprocae in fuure. To accoun for his dynamic ineracion, we calculae he probabiliy of a non-reciprocal link being reciprocaed in fuure and he average days aken o receive he reciprocaion, condiional on he source reviewer s sae, using he equilibrium policy recovered from daa. We hen calculae he change in discouned viewership assuming ha a reciprocal link is esablished wih he corresponding probabiliy and delay. [Inser Figures 4 and 5 Abou Here] The incremenal benefi of creaing a reciprocal link is repored in Figure 4. The resul is summarized along he conen dimension in quiniles. The op figure shows posiive average effecs for all five quiniles, suggesing ha in general he benefi of more borrowed conen ouweighs he cos of reduced nework posiion hrough forming a reciprocal link. Also, he figure shows ha reviewers wih more conen benefi more from a reciprocal link. This is consisen wih he policy funcion where reviewers wih more conen are more likely o creae reciprocal links, as shown in he boom figure of Figure 4. The resul for creaing a non-reciprocal link is repored in Figure 5, also summarized along he conen dimension. Creaing a non-reciprocal link ypically reduces nework posiion more han does a reciprocal one. As shown in he firs series of he op figure, which includes he direc effec bu does no accoun for fuure reciprocaion, he average incremenal benefi is negaive for all five reviewer quiniles, suggesing ha he cos of reduced nework posiion ouweighs he benefi of more borrowed conen. The incremenal benefi is also significanly lower han ha of forming reciprocal links. Recall ha he cos esimae in secion 5.3 shows no evidence of inrinsic value for reciprocal links, we know ha in he conex of his sudy, he 40
42 endency owards reciprociy is mainly explained by he comparaively favorable impacs of reciprocal links on viewership, due o he promoe-he-promoer effec. This is a noable resul. Sociology lieraure has long recognized he prominence of reciprociy in social neworks, and saisical nework models ofen consider ha as model primiives. Our sudy provides an alernaive explanaion in a raional economic raher han social conex, ha reciprociy can be naurally favored by dynamic sraegic consideraions, wihou he need for a social explanaion as model primiive. 32 However, he firs series in he op figure also shows ha he more conen a reviewer has, he lower her incremenal benefi from forming a non-reciprocal link, ye he policy funcion shows ha reviewers wih more conen are more likely o creae non-reciprocal links (he boom figure). Thus a saic perspecive alone does no explain he linking acions well. This discrepancy is resolved once he dynamic sraegic perspecive is aken ino accoun. As he second series in he op figure shows, afer accouning for fuure reciprocaion, he incremenal benefi of forming a non-reciprocal link increases significanly for all five quiniles, and reviewers wih more conen have higher incremenal benefi. This is because a reviewer wih more conen is more confiden o see he arge reviewer reciprocae, and wih shorer delay. Afer all, he arge reviewer also can benefi from borrowed conen, and when she decides o creae a link, she would favor a reciprocal one o mainain her own nework posiion, hus making he source reviewer a favorable arge. This incenive o reciprocae is furher enhanced when he source reviewer has more conen. In essence, a reviewer is inviing reciprocaion when creaing a non-reciprocal link, in anicipaion of he sraegic response from he arge 32 Tha is, an explanaion such as people end o form reciprocal links because by naure hey like reciprociy, i.e. here is an inrinsic value o reciprocae. 41
43 reviewer, and he more conen a reviewer has, he more effecive his sraegy is. Comparing he wo scenarios clearly shows how he dynamic ineracions drive reviewers linking decisions The Impac of Linking on Conen Producion We now analyze how linking influences reviewers conen producion decisions. Similar o he analysis of link formaion, we evaluae he incremenal effec of wriing reviews by reviewers a differen saes, accouning for he dynamic ineracion effecs arising from linking. For his analysis, we use he same subse of he daa as used in analyzing link formaion. We begin wih analyzing he direc effec of wriing reviews: for each day and each reviewer, we calculae he difference in discouned viewership beween wo oherwise idenical scenarios: in he firs, he focal reviewer does no wrie reviews; in he second, she wries one review aricle, which depreciaes a he esimaed depreciaion rae. This difference in viewership approximaes he direc incremenal benefi of producing one uni of conen, from which he producion cos is hen subraced o arrive a he ne incremenal benefi. The resul is shown in Figure 6(A). The direc incremenal benefi is much higher for reviewers wih higher nework posiions, and is posiive only for reviewers in he op wo quiniles of he nework posiion dimension. Reviewers wih more conen also ge higher benefi, bu he difference along he conen dimension is no as large as along he nework posiion dimension. The direc benefi is only one par of he incenive behind conen producion. When deciding wheher o produce conen, a reviewer considers no only he immediae viewership, bu also he fuure linking acions of oher reviewers. For example, if a reviewer in a high conen sae anicipaes oher reviewers o link o her in he near fuure, which leads o higher nework posiion, hen her incenive o produce will be increased, as he addiional benefi from higher nework posiion laer on adds o he direc benefi. Whereas if a reviewer in anoher sae 42
44 expecs her compeiors o receive more incoming links, which reduces her relaive posiion in he nework, hen her incenive o produce will be lower han suggesed by he direc benefi. Thus linking could significanly aler he incenive o wrie reviews depending on he saes of reviewers. [Inser Figure 6 Abou Here] To analyze how linking influences he incenive o produce, we use he equilibrium policy o calculae he average change in nework posiions, arising from linking, corresponding o differen reviewer saes. The calculaion shows ha reviewers a high conen and low nework posiion saes ge highes average increase in nework posiion over ime, while reviewers in he opposie saes see heir nework posiions reduce laer on. We hen incorporae his saedependen change in nework posiion ino he calculaion of he incremenal effec of wriing one more review. The ne incremenal benefi calculaed his way, repored in Figure 6(B), shows ha once he prospec of linking is accouned for, conen level, insead of nework posiion, becomes he main deermining facor of he incremenal benefi. Reviewers in he op wo quiniles of he conen dimension have posiive ne benefi from producing conen, while oher reviewers have negaive benefi, even for hose wih high nework posiions. In oher words, a reviewer wih high conen bu low nework posiion wries reviews because she expecs oher reviewers o link o her laer on, even hough he immediae viewership is no much. Meanwhile, a reviewer a he opposie sae finds i no worhwhile o wrie reviews as she foresees lower nework posiion ahead. This is consisen wih he observed policy funcion, which shows ha conen level influences he frequency of wriing reviews more han nework posiion does. In summary, he resuls demonsrae a close inerdependence beween link formaion and conen producion, and ha linking is a maor driver of reviewers wriing decisions. 43
45 Ineresingly, he prospec of linking encourages he conen producion of reviewers wih high conen and low nework posiions, while discourages he conen producion of reviewers wih low conen and high nework posiions Ne Benefi and Marke Srucure Given he viewership demand and he cos esimaes, we now analyze he ne benefi accrued o reviewers a differen saes. To do so, we calculae he discouned ne benefi over rolling sixmonh windows over he enire period. Ne benefi is simply he viewership minus he cos of producion and linking. Coss are derived from reviewers acual decisions while viewership is inferred from reviewers saes and he esimaed viewership demand equaion. [Inser Figure 7 Abou Here] The resul is presened in Figure 7, where we show he average ne benefi on he conen dimension and nework posiion dimension. The figure shows ha reviewers wih higher nework posiions derive significanly higher benefi han reviewers wih lower nework posiions. In conras, however, reviewers wih more conen do no derive higher benefi han reviewers wih less conen. This may be unexpeced a firs look, as he demand equaion shows ha more conen leads o higher viewership. Bu i is explained by he cos side: alhough reviewers wih more conen can arac higher viewership, hey also incur higher cos as hey wrie more reviews. The resul shows ha addiional viewership demand is mosly offse by he increased cos, making a reviewer wih more conen no beer han one wih less conen in erms of ne benefi. This resul is also reasonable when we consider he compeiive effec: since conen level is deermined solely by a reviewer s own producion decisions, were here o be significanly higher ne benefi wih higher conen level, all reviewers would wrie more and in so doing, he poenial advanage from higher conen level would be largely compeed away. 44
46 In conras, he advanage coming from higher nework posiions canno be compeed away as easily. This is because, even hough desirable, a reviewer canno unilaerally increase her nework posiion. Insead, i akes incoming links from oher reviewers o increase ha. Thus a reviewer may enoy significan advanage from having a high nework posiion, while oher reviewers lack an effecive way o couner ha. Indeed, our calculaion suggess ha higher nework posiions offer significan compeiive advanage and lead o higher ne benefi. The compeiion hus resuls in a marke where reviewers are differeniaed along he nework posiion dimension, while on he conen dimension surplus is mosly compeed away. The conras beween conen and nework posiion should be aken noe by companies operaing such websies. As a websie seeks o maximize is overall viewership, i should seek o encourage conen producion by creaing a compeiive environmen inernally. Any form of sicky compeiive advanage enoyed by a subse of producers, such as ha led o by higher nework posiions in his conex, may creae imbalance in he sysem. This imbalance can poenially lead o differeniaions ha sofen he compeiion and reduce he overall conen level a he sie. Consequenly, he effec of linking o he overall websie viewership is a maer of concern ha is worh furher invesigaion. 5.5 The Effec of Linking on Websie Viewership A Simulaion For markeing managers who operae hose conen websies, i is imporan o know wheher he nework among conen producers increases he overall viewership a he websie, and how he linking feaure should be designed o generae opimal viewership oucome. If conen is exogenously given, hen we would expec a nework superimposed among producers o increase 45
47 overall viewership, as linking enhances conen. 33 When conen producion is deermined endogenously in a dynamic conex, however, he overall effec of nework is no a all clear. Qualiaive analysis of radeoffs also reveals forces owards boh direcions. On one hand, since a producer wih more conen is more likely o receive incoming links, linking provides an incenive for cerain producers o produce more conen. On he oher hand, however, linking could also discourage oher producers from producing conen, as is shown in secion 5.4. A he websie level, he conen enhancemen effec of linking is expeced o increase overall viewership. However, if here are producers wih high conen volume siing a obscure posiions in he nework, while ohers occupy prominen posiions ye do no have much conen, hen he nework may hinder efficiency by no effecively direcing viewer raffic o conen. Given hese facors wih opposie effecs, he nework could eiher increase or decrease overall viewership. A sign of concern, hough, is ha as shown in secion 5.4, reviewers wih more prominen nework posiions enoy significan compeiive advanage, ye compeiive advanage enoyed by a small se of players in general reduces compeiion inensiy. This suggess ha he curren nework may impede he compeiion among conen producers, and ha alernaive policies regulaing link formaion may help he websie improve overall viewership. To evaluae he overall effec of nework, ideally we wan o compare wo siuaions which are oherwise idenical, excep ha in he firs link creaion is allowed, while in he second i is no. Similarly, we wan o compare siuaions under alernaive linking regulaions, such as resricing he oal number of links a developer can creae, o find ou which link regulaion leads o bes viewership oucome for he websie. However, curren mehodological resricions 33 This is consisen wih exising lieraure, which shows ha nework increases overall sales in an online shopping cener environmen (Sephen and Toubia 2009). 46
48 prohibi us from making hese comparisons direcly, as i requires explicily solving for he equilibria of alernaive dynamic games, which are compuaionally infeasible. 34 Considering his, we resor o a second bes approach by performing simulaions which aler iniial saes bu do no aler he exising equilibrium since i is only he iniial saes ha changes, while he srucural parameers and he game remain he same, he exising equilibrium recovered from daa sill applies. To analyze hrough his approach wheher he imbalance induced by he curren nework reduces viewership, we pick a sae from he daa, and for each reviewer, we randomly remove her ougoing links unil she has no more han five ougoing links remaining. 35 Afer his sysem-wide link removal, he nework becomes more sparse and balanced. We hen perform wo forward simulaions for 60 periods, wih he firs saring from he original sae and he second from his new sae afer he link removal. We compare conen producion, link formaion, and overall viewership beween hese wo simulaions o evaluae he overall effec of he nework. [Inser Table 7 and Figure 8 Abou Here] The resul of he simulaion is repored in Table 7. Wih he sysem wide link removal, conen producion, link creaion, and overall viewership demand all increase significanly. The average daily viewership increases by 17.12% for he websie overall, while conen producion and link creaion boh increase by more han 30%. This suggess ha he curren nework among conen producers, alhough providing benefi hrough enhancing conen, also brings oo much 34 BBL recovers srucural parameers wihou explici compuaion of equilibrium, hus bypassing he curse of dimensionaliy issue. In policy simulaion, however, any change in he rule of he game can poenially lead o a new equilibrium, so equilibrium mus be explicily compued. For example, if link formaion is prohibied, all reviewers will adus heir sraegy for wriing reviews accordingly. To evaluae ha change, we mus compue he new equilibrium, he cos of which is prohibiive given he number of reviewers in our sudy. Recen mehodological advancemen, e.g. he concep of oblivious equilibrium developed in Weinraub, Benkard, and Van Roy (2008), can poenially solve his issue, wih he drawback ha he soluion concep iself is an approximaion. We leave he poenial use of oblivious equilibrium for fuure work. 35 We do no remove all links o avoid poenial issues of boundary bias for he esimaed policy funcions. 47
49 marke power o cerain producers hose wih high nework posiions and impedes efficiency. When he field of compeiion is leveled, compeiion inensifies, wih reviewers collecively producing more conen, and he overall viewership a he websie increases. Figure 8 shows he simulaion resul in furher deail along he ime dimension. The viewership demand umps immediaely afer he link removal, likely because reviewers wih high conen volumes bu low nework posiion now become more visible and arac more viewership. Over ime, he demand increase moderaes slighly bu sill holds sably above 15%. This is suppored by susained increase in conen producion wih a leveled playing field, he compeiion is inensified and reviewers collecively have higher incenive, or are forced, o produce more conen. Link creaion also umps iniially, bu his is comparaively shor-lived, as link creaion falls back o he pre-removal rae afer abou hiry periods. In summary, his simulaion provides evidence ha he curren design over ime leads o inefficien inernal compeiion among reviewers. Alernaive policies ha regulae link formaion could poenially lead o overall viewership and should be considered for experimenaion a he websie. 6. Discussion, Limiaion and Conclusion The adven of online social media brings abou many inriguing phenomena. A prominen one is he emergence of a large number of revenue sharing conen websies, which rely on exernal conen producers o supply conen and induce an inernal compeiion for viewership among producers. The linking feaure recenly inroduced o many websies furher leads o complex and inriguing dynamic ineracions among conen producers. Meanwhile, he implicaion of linking on he overall viewership, crucial o he websie plaform builders, remains an open quesion. A deailed undersanding of producers ineracions and heir implicaions hus no only is of 48
50 academic ineres, bu also has imporan managerial implicaions, as his phenomenon is quickly gaining momenum in he indusry. Moivaed by his, we develop a dynamic oligopoly model o sudy he compeiion among conen producers. In our model, producers compee agains one anoher hrough producing conen and forming links, and we characerize heir sraegic ineracions using he soluion concep of Markov-perfec equilibrium. We esimae he model using he daa obained from a popular produc review websie, leveraging he wo-sep esimaion approach developed in Baari, Benkard, Levin (2007), and provide a deailed analysis of he ineracions among reviewers in heir decision process. Our sudy conribues o he lieraure by invesigaing he ineracions of link formaion and conen producion decisions, by analyzing he iner-emporal radeoffs ha drive he ineracions dynamically, and by providing a raional economic framework for empirically sudying he formaion of neworks in a dynamic sraegic seing. Our sudy leads o several findings wih managerial implicaions. We find ha viewership demand is posiively influenced by conen volume and nework posiions, and here is a conen borrowing effec hrough linking. We find ha reciprocal links are more likely o be formed han non-reciprocal ones, and his is encouraged by he naure of he sraegic ineracion a promoe-he-promoer effec. This endency owards reciprociy furher induces producers wih high conen volume o sraegically creae non-reciprocal links, in anicipaion of reciprocaion laer on which will enhance heir nework posiions. We find ha he prospec of linking encourages producers wih high conen volume bu low nework posiion o produce more conen, ye discourages producers a opposie saes. Furhermore, we find ha he producers ne benefi increases wih heir nework posiions bu no wih heir conen volume, as he higher viewership from more conen is offse by he 49
51 higher cos incurred in producing he conen. This suggess ha linking may lead o inefficiency as compeiive advanage is accrued o producers wih high nework posiions. Finally, our simulaion suggess ha limiing he links a he websie may lead o higher conen producion and overall viewership demand. Managers who operae conen websies can consider several alernaive linking policy designs o improve efficiency. They could prohibi linking alogeher by no offering he feaure. This will preven compeiive advanage from being accrued o a subgroup of producers. Beween compleely disabling linking and no regulaing linking a all, an alernaive a he middle ground is o resric he number of links each producer can form. This could alleviae he imbalance over ime, while producers would also become more selecive in forming links. Anoher alernaive is o impose a ime limi on links so ha hey expire afer some ime. This could make he nework srucure less rigid and ease he issue of unbalanced compeiion. Mehodological resricions limi our abiliy o analyze hese alernaive policies in deail, while indusry managers could explore hese and oher policies hrough experimenaion a he websies. A few oher limiaions of our sudy can be addressed in fuure work. Firs, our sudy focuses on he profi moive of conen producers, and we use a group of op producers for our analysis. Alhough mos websies have a significan share of heir viewership generaed by a small group of elie producers, here is also a larger group of more casual conen producers. This mass group of casual producers may have incenives oher han profi, and a richer model is called for o sudy heir behaviors and conribuions o he business. Second, in he social media marke, he line beween consumers and producers is blurred. While our focus on he small group of elie producers allow us o sill follow he radiional supply-side demand-side dichoomy, an exciing opporuniy exiss o advance he lieraure by invesigaing he dual roles he websie 50
52 users may play. Finally, no all conen is he same, and differen conen may be eiher complemens or subsiues. For feasibiliy reasons, our model considers all conen o be of he same ype, while we leave he ineracions induced by differen conen ypes for fuure research. We also hope ha, wih he rapid advancemen in economerics on dynamic game esimaion mehodologies, we will be able o admi more heerogeneiy among producers in fuure, and o explicily evaluae he effecs of alernaive policies when hey lead o differen equilibrium siuaions. Online conen markes, and social media in general, bring much closer and more dynamic ineracions among consumers, beween consumers and producers, and among producers, han he radiional offline marke does. Wih ha, i also opens an exciing fronier for markeing research. Our work is an early sep owards his direcion, and we are confiden ha fuure research will bring furher insighs in his area and offer much needed managerial guidance. 51
53 References Ackerberg, D., C.L. Benkard, S. Berry and A. Pakes, (2007), Economeric Tools for Analyzing Marke Oucomes, Handbook of Economerics, 6, Aguirregabiria, V. and P. Mira, (2007), Sequenial Esimaion of Dynamic Discree Games, Economerica, 75, 1-53 Axelrod, R. and W. Hamilon, (1981), "The Evoluion of Cooperaion," Science, 211, Baari, P., C.L. Benkard and J. Levin, (2007), Esimaing Dynamic Models of Imperfec Compeiion, Economerica, 75, Bala, V. and S. Goyal, (2000), A Non-cooperaive Model of Nework Formaion, Economerica, 68, Berry, S.T., (1994), Esimaing Discree-Choice Models of Produc Differeniaion, The RAND Journal of Economics, 25, Berry, S., J. Levinsohn and A. Pakes, (1995), Auomobile Prices in Marke Equilibrium, Economerica, 63, Besanko D. and U. Doraszelski, (2004), Capaciy Dynamics and Endogenous Asymmeries in Firm Size, The RAND Journal of Economics, 35, Bonacich, P., (1987), Power and Cenraliy: A Family of Measures, The American Journal of Sociology, 92, Bonacich P. and P. Lloyd, (2001), Eigenvecor-like Measures of Cenraliy for Asymmeric Relaions, Social Neworks, 123, Brin, S. and L. Page, (1998), The Anaomy of a Large-scale Hyperexual Web Search Engine, Compuer Neworks and ISDN Sysems, 30, Chevalier, J.A. and D. Mayzlin, (2006), The Effec of Word of Mouh on Sales: Online Book Reviews, Journal of Markeing Research, Vol. XLIII, Chinaguna, P., S. Gopinah and S. Venkaaraman, (2010), The Effecs of Online User- Reviews on Movie Box-Office Performance: Accouning for Sequenial Rollou and Aggregaion Across Local Markes, Markeing Science, forhcoming Dube, J.P., G.J. Hisch and P.E. Rossi, (2009), Do Swiching Coss Make Markes Less Compeiive? Journal of Markeing Research, Vol. XLVI, Erdem, T. and M.P. Keane, (1996), Decision Making Under Uncerainy: Capuring Dynamic Brand Choice Processes in Turbulen Consumer Goods Markes, Markeing Science, 15, 1-20 Ericson, R. and A. Pakes, (1995), Markov-Perfec Indusry Dynamics: A Framework for Empirical Work, The Review of Economic Sudies, 62, Faus, K. and S. Wasserman, (1992), Cenraliy and Presige: A Review and Synhesis, Journal of Quaniaive Anhropology, 4,
54 Godes, D. and Mayzlin, D., (2004), Using Online Conversaions o Sudy Word-of-Mouh Communicaions, Markeing Science, 23, Gouldner, A.W., (1960), The Norm of Reciprociy: A Preliminary Saemen, American Sociological Review, 25, Holmes, T.J., (2009), The Diffusion of Wal-Mar and Economies of Densiy, Working Paper Hoz, V.J. and R.A. Miller (1993), Condiional Choice Probabiliies and he Esimaion of Dynamic Models, The Review of Economic Sudies, 60, Jackson, M.O., (2004), A Survey of Models of Nework Formaion: Sabiliy and Efficiency, in G. Demange and M. Wooders, eds., Group Formaion in Economics: Neworks, Clubs, and Coaliions, Cambridge Universiy Press Kamakura, W.A. and G.J. Russell, (1989), A Probabilisic Choice Model for Marke Segmenaion and Elasiciy Srucure, Journal of Markeing Research, 26, Kaona, Z. and M. Sarvary, (2008), Nework Formaion and he Srucure of he Commercial World Wide Web, Markeing Science, 27, Maskin, E. and J. Tirole, (1988), A Theory of Dynamic Oligopoly: I & II, Economerica, 56, Maskin, E. and J. Tirole, (2001), Markov Perfec Equilibrium, I. Observable Acions, Journal of Economic Theory, 100, Mayzlin, D. and H. Yoganarasimhan, (2008), Link o Success: How Blogs Build an Audience by Promoing Rivals, Working Paper Pakes, A. and P. McGuire, (1994), Compuing Markov-Perfec Nash Equilibria: Numerical Implicaions of a Dynamic Differeniaed Produc Model, The RAND Journal of Economics, 25, Pakes, A. and P. McGuire, (2001), Sochasic Algorihms, Symmeric Markov Perfec Equilibrium, and he 'Curse' of Dimensionaliy, Economerica, 69, Pakes, A., M. Osrovsky and S. Berry, (2007), Simple Esimaors for he Parameers of Discree Dynamic Games (Wih Enry/Exi Examples), The RAND Journal of Economics, 38, Rus, J., (1987), Opimal Replacemen of GMC Bus Engines: An Empirical Model of Harold Zurcher, Economerica, 55, Ryan, S.P., (2009), The Coss of Environmenal Regulaion in a Concenraed Indusry, Working Paper Srinivasan, K., (2006), Invied Commenary: Empirical Analysis of Theory-Based Models in Markeing, Markeing Science, 25, Sephen, A. and O. Toubia, (2009), Deriving Value from Social Commerce Neworks, The Journal of Markeing Research, forhcoming Wasserman, S. and K. Faus, (1994), Social Nework Analysis: Mehods and Applicaions, Cambridge Universiy Press 53
55 Weinraub, G.Y., C.L. Benkard and B. Van Roy, (2008), Markov Perfec Indusry Dynamics wih Many Firms, Economerica, 76, Yao, S. and C. Mela, (2010), A Dynamic Model of Sponsored Search Adverising, Working Paper Appendix A.1 PageRank In his secion we explain he deail of PageRank, he measure of nework posiion used in our sudy. PageRank, firs presened in Brin and Page (1998), is behind he iniial Google search engine. Given a nework of pages, i produces a numerical measure for each page o represen is relaive imporance in he nework. The measure is well documened in lieraure, and is explained here for compleeness: Le p,... 1, p2 pn be n nodes (web pages) which are conneced by direcional links. Le i c be he ou-degree of page p i he number of ougoing links from ha page. Le d be a damping facor which value beween 0 and 1. Denoe A as he modified adacency marix for he graph of he nodes and he links, where: (A-1) [ A] i 1/ ci = 0 i oherwise The PageRank, denoed as PR, is a vecor such ha: (A-2) PR = ( 1 d) PR + d PR A The i -h elemen of PR is he PageRank of node p i. A larger value indicaes higher imporance of he node in he nework. The measure is rooed on a Markov random navigaion model: assume here is a person visiing he pages; a any ime, wih probabiliy d she chooses o follow an ougoing link, wih link chosen randomly wih equal probabiliy when muliple ougoing links exis, and wih probabiliy 1 d she will ump o anoher page, wih each page 54
56 having he same probabiliy o be he desinaion. The PageRank of a page is hen he seadysae probabiliy of ha page being visied. As saed in Brin and Page (1998), Anoher inuiive usificaion is ha a page can have a high PageRank if here are many pages ha poin o i, or if here are some pages ha poin o i and have a high PageRank. This insigh proves crucial for he success of PageRank in capuring he relaive imporance of web pages on he Inerne, and is insrumenal in our sudy. The PageRank is also similar o eigenvecor cenraliy ha is widely used in social nework lieraure, where i is shown o reflec he power or presige of a node in he nework. (Le Ι be an n n marix where [ Ι ] = 1 n, hen A-2 can be wrien as PR = PR(( 1 d) Ι + d A), i.e. he i / PageRank is an eigenvecor of he adacency marix furher modified by he damping facor.) 55
57 Table 1: Revenue Sharing Websies wih Independen Conen Producers 1 Websie Conen Type Linking (Name) Monhly Visiors (in millions) 3 abou.com Advice No 43.9 answers.yahoo.com Quesions & Answers Yes (Fan) 43.7 associaedconen.com 2 General conen Yes (Favorie) 10.7 ehow.com How-o ip Yes (Subscripion) 30.5 epinions.com Produc review Yes (Trus) 4.1 hubpages.com General conen Yes (Follow) 9.9 irepor.com News repor Yes (Follow) 1.1 seekingalpha.com Invesmen advice Yes (Follow) 1.2 squidoo.com General conen Yes (Fan) 6.6 youube.com Video Yes (Subscripion) A lis of more han 100 revenue sharing conen sies can be found a hp://socialmediarader.com/resource-lis-100-revenue-sharing-sies/ 2. Acquired by Yahoo! in May 2010 for abou $100 million. 3. Source: compee.com. June 2010 Table 2: Summary Saisics Mean SD Min Max Reviews Wrien Per Reviewer Links Creaed Per Reviewer Toal View Coun Per Reviewer Number of Reviewers 193 Toal Reviews Wrien Number of Decision Days 646 Toal Links Creaed 2039 Number of View Coun Periods 28 Toal View Couns Toal Non-reciprocal Links 1148 Toal Reciprocal Links 891 Percen of Reciprocaed Nonreciprocal Links 40.70% Average Days Taken To Reciprocae 52.5 Table 3: Reviewer Qualiy Facors Facor Value Descripion Diversiy Ineger The number of produc caegories for which he reviewer wries reviews as op reviewers Popular Binary Indicaor The reviewer was recognized as he op 100 mos popular auhors before Advisor Binary Indicaor The reviewer is recognized as rused source on conen qualiy Average Diversiy 1.53 Number of Popular Reviewers 59 Number of Advisor 112 Reviewers 56
58 Table 4: Summary Saisics - Conen, Nework Posiion, and Borrowed Conen Mean SD Conen Nework Posiion 5.18E E-03 Borrowed Conen Table 5: Viewership Demand Esimaion I II III IV V Model Specificaion Linear Log- Linear Quadraic Log Embedded LaenClass Parameer Segmen 1 Segmen 2 Conen (***) (***) (***) (***) (***) (.) Conen^ (*) (*) BorrowedConen (.) (**) (**) (.) (***) BorrowedConen^ (**) (***) NeworkPosiion (***) (***) (***) (***) 2.112(***) (***) NeworkPosiion^ (***) (***) (***) Diversiy (***) (***) (***) (***) (***) Popular (***) (***) (***) (***) (***) Advisor (***) (*) (***) (**) (***) BorrowedDiversiy (**) (***) (*) BorrowedPopular (***) (***) (***) (***) (**) BorrowedAdvisor (**) (**) (***) (***) (*) Consan (***) (***) (***) (***) (***) Segmen Size LL BIC Signif. codes: 0 *** ** 0.01 * Table 6: Dynamic Cos Parameer Esimaion Esimae Low 95% CI High 95% CI Producion Consan (*) Qualiy (*) Tenure Link Consan 0.025(*) Qualiy Tenure Reciprocal Uni of measure: housand page views 57
59 Table 7: Simulaion Sysem-wide Link Removal Demand Conen Producion Link Formaion Average Increase Per Day Percenage Increase 17.12% 54.40% 31.32% Figure 1: Average Conen Producion by Own Sae nework posiion conen Figure 2: Probabiliy o Form Links by Own Sae nework posiion conen Figure 3: Relaive Link Probabiliy Non-reciprocal over Reciprocal nework posiion conen 58
60 Figure 4: Effec of Creaing Reciprocal Links Effec of Creaing a Reciprocal Link Incremenal Benefi Conen Quinile Acual Frequency of Creaing Reciprocal Links Frequency Conen Quinile Figure 5: Effec of Creaing Non-reciprocal Links Effec of Creaing a Non-reciprocal Link Incremenal Benefi Conen Quinile Wihou Reciprocaion Wih Reciprocaion Acual Frequency of Creaing Non-reciprocal Links Frequency Conen Quinile 59
61 Figure 6: Incremenal Benefi of Conen Producion by Sae incremenal benefi incremenal benefi nework posiion conen nework posiion conen (A) no accoun for change in nework posiion (B) accoun for change in nework posiion Figure 7: Ne Benefi by Conen and Nework Posiion Ne Benefi By Conen 40 Value (Thousand Page View) Conen Quinile Ne Benefi By Nework Posiion Value (Thousand Page View) Nework Posiion Quinile 60
62 Figure 8: Simulaion Sysem-wide Link Removal Change in Viewership Demand 25.00% 20.00% Percen Change 15.00% 10.00% 5.00% 0.00% Day Change in Conen Producion and Link Formaion Percenage Change % % 80.00% 60.00% 40.00% 20.00% link formaion conen producion 0.00% % Day 61
PROFIT TEST MODELLING IN LIFE ASSURANCE USING SPREADSHEETS PART ONE
Profi Tes Modelling in Life Assurance Using Spreadshees PROFIT TEST MODELLING IN LIFE ASSURANCE USING SPREADSHEETS PART ONE Erik Alm Peer Millingon 2004 Profi Tes Modelling in Life Assurance Using Spreadshees
Analysis of Pricing and Efficiency Control Strategy between Internet Retailer and Conventional Retailer
Recen Advances in Business Managemen and Markeing Analysis of Pricing and Efficiency Conrol Sraegy beween Inerne Reailer and Convenional Reailer HYUG RAE CHO 1, SUG MOO BAE and JOG HU PARK 3 Deparmen of
Chapter 8: Regression with Lagged Explanatory Variables
Chaper 8: Regression wih Lagged Explanaory Variables Time series daa: Y for =1,..,T End goal: Regression model relaing a dependen variable o explanaory variables. Wih ime series new issues arise: 1. One
To Sponsor or Not to Sponsor: Sponsored Search Auctions with Organic Links and Firm Dependent Click-Through Rates
To Sponsor or No o Sponsor: Sponsored Search Aucions wih Organic Links and Firm Dependen Click-Through Raes Michael Arnold, Eric Darmon and Thierry Penard June 5, 00 Draf: Preliminary and Incomplee Absrac
Morningstar Investor Return
Morningsar Invesor Reurn Morningsar Mehodology Paper Augus 31, 2010 2010 Morningsar, Inc. All righs reserved. The informaion in his documen is he propery of Morningsar, Inc. Reproducion or ranscripion
Research on Inventory Sharing and Pricing Strategy of Multichannel Retailer with Channel Preference in Internet Environment
Vol. 7, No. 6 (04), pp. 365-374 hp://dx.doi.org/0.457/ijhi.04.7.6.3 Research on Invenory Sharing and Pricing Sraegy of Mulichannel Reailer wih Channel Preference in Inerne Environmen Hanzong Li College
II.1. Debt reduction and fiscal multipliers. dbt da dpbal da dg. bal
Quarerly Repor on he Euro Area 3/202 II.. Deb reducion and fiscal mulipliers The deerioraion of public finances in he firs years of he crisis has led mos Member Saes o adop sizeable consolidaion packages.
Why Did the Demand for Cash Decrease Recently in Korea?
Why Did he Demand for Cash Decrease Recenly in Korea? Byoung Hark Yoo Bank of Korea 26. 5 Absrac We explores why cash demand have decreased recenly in Korea. The raio of cash o consumpion fell o 4.7% in
Software Exclusivity and the Scope of Indirect Network Effects in the U.S. Home Video Game Market
Sofware Exclusiviy and he Scope of Indirec Nework Effecs in he U.S. Home Video Game Marke Kenneh S. Cors Roman School of Managemen, Universiy of Torono Mara Lederman Roman School of Managemen, Universiy
Distributing Human Resources among Software Development Projects 1
Disribuing Human Resources among Sofware Developmen Proecs Macario Polo, María Dolores Maeos, Mario Piaini and rancisco Ruiz Summary This paper presens a mehod for esimaing he disribuion of human resources
ANALYSIS AND COMPARISONS OF SOME SOLUTION CONCEPTS FOR STOCHASTIC PROGRAMMING PROBLEMS
ANALYSIS AND COMPARISONS OF SOME SOLUTION CONCEPTS FOR STOCHASTIC PROGRAMMING PROBLEMS R. Caballero, E. Cerdá, M. M. Muñoz and L. Rey () Deparmen of Applied Economics (Mahemaics), Universiy of Málaga,
The Greek financial crisis: growing imbalances and sovereign spreads. Heather D. Gibson, Stephan G. Hall and George S. Tavlas
The Greek financial crisis: growing imbalances and sovereign spreads Heaher D. Gibson, Sephan G. Hall and George S. Tavlas The enry The enry of Greece ino he Eurozone in 2001 produced a dividend in he
Chapter 1.6 Financial Management
Chaper 1.6 Financial Managemen Par I: Objecive ype quesions and answers 1. Simple pay back period is equal o: a) Raio of Firs cos/ne yearly savings b) Raio of Annual gross cash flow/capial cos n c) = (1
LEASING VERSUSBUYING
LEASNG VERSUSBUYNG Conribued by James D. Blum and LeRoy D. Brooks Assisan Professors of Business Adminisraion Deparmen of Business Adminisraion Universiy of Delaware Newark, Delaware The auhors discuss
UNDERSTANDING THE DEATH BENEFIT SWITCH OPTION IN UNIVERSAL LIFE POLICIES. Nadine Gatzert
UNDERSTANDING THE DEATH BENEFIT SWITCH OPTION IN UNIVERSAL LIFE POLICIES Nadine Gazer Conac (has changed since iniial submission): Chair for Insurance Managemen Universiy of Erlangen-Nuremberg Lange Gasse
Forecasting and Information Sharing in Supply Chains Under Quasi-ARMA Demand
Forecasing and Informaion Sharing in Supply Chains Under Quasi-ARMA Demand Avi Giloni, Clifford Hurvich, Sridhar Seshadri July 9, 2009 Absrac In his paper, we revisi he problem of demand propagaion in
Journal Of Business & Economics Research September 2005 Volume 3, Number 9
Opion Pricing And Mone Carlo Simulaions George M. Jabbour, (Email: [email protected]), George Washingon Universiy Yi-Kang Liu, ([email protected]), George Washingon Universiy ABSTRACT The advanage of Mone Carlo
The Transport Equation
The Transpor Equaion Consider a fluid, flowing wih velociy, V, in a hin sraigh ube whose cross secion will be denoed by A. Suppose he fluid conains a conaminan whose concenraion a posiion a ime will be
11/6/2013. Chapter 14: Dynamic AD-AS. Introduction. Introduction. Keeping track of time. The model s elements
Inroducion Chaper 14: Dynamic D-S dynamic model of aggregae and aggregae supply gives us more insigh ino how he economy works in he shor run. I is a simplified version of a DSGE model, used in cuing-edge
Niche Market or Mass Market?
Niche Marke or Mass Marke? Maxim Ivanov y McMaser Universiy July 2009 Absrac The de niion of a niche or a mass marke is based on he ranking of wo variables: he monopoly price and he produc mean value.
The Real Business Cycle paradigm. The RBC model emphasizes supply (technology) disturbances as the main source of
Prof. Harris Dellas Advanced Macroeconomics Winer 2001/01 The Real Business Cycle paradigm The RBC model emphasizes supply (echnology) disurbances as he main source of macroeconomic flucuaions in a world
DYNAMIC MODELS FOR VALUATION OF WRONGFUL DEATH PAYMENTS
DYNAMIC MODELS FOR VALUATION OF WRONGFUL DEATH PAYMENTS Hong Mao, Shanghai Second Polyechnic Universiy Krzyszof M. Osaszewski, Illinois Sae Universiy Youyu Zhang, Fudan Universiy ABSTRACT Liigaion, exper
Performance Center Overview. Performance Center Overview 1
Performance Cener Overview Performance Cener Overview 1 ODJFS Performance Cener ce Cener New Performance Cener Model Performance Cener Projec Meeings Performance Cener Execuive Meeings Performance Cener
Individual Health Insurance April 30, 2008 Pages 167-170
Individual Healh Insurance April 30, 2008 Pages 167-170 We have received feedback ha his secion of he e is confusing because some of he defined noaion is inconsisen wih comparable life insurance reserve
The Grantor Retained Annuity Trust (GRAT)
WEALTH ADVISORY Esae Planning Sraegies for closely-held, family businesses The Granor Reained Annuiy Trus (GRAT) An efficien wealh ransfer sraegy, paricularly in a low ineres rae environmen Family business
Appendix D Flexibility Factor/Margin of Choice Desktop Research
Appendix D Flexibiliy Facor/Margin of Choice Deskop Research Cheshire Eas Council Cheshire Eas Employmen Land Review Conens D1 Flexibiliy Facor/Margin of Choice Deskop Research 2 Final Ocober 2012 \\GLOBAL.ARUP.COM\EUROPE\MANCHESTER\JOBS\200000\223489-00\4
SPEC model selection algorithm for ARCH models: an options pricing evaluation framework
Applied Financial Economics Leers, 2008, 4, 419 423 SEC model selecion algorihm for ARCH models: an opions pricing evaluaion framework Savros Degiannakis a, * and Evdokia Xekalaki a,b a Deparmen of Saisics,
An Analysis of Adoption of Digital Health Records under Switching Costs
1 An Analysis of Adopion of Digial Healh Records under Swiching Coss November 2010 ZAFER D. OZDEMIR a JOHN M. BARRON b SUBHAJYOTI BANDYOPADHYAY c [email protected]; [email protected]; [email protected]
TEMPORAL PATTERN IDENTIFICATION OF TIME SERIES DATA USING PATTERN WAVELETS AND GENETIC ALGORITHMS
TEMPORAL PATTERN IDENTIFICATION OF TIME SERIES DATA USING PATTERN WAVELETS AND GENETIC ALGORITHMS RICHARD J. POVINELLI AND XIN FENG Deparmen of Elecrical and Compuer Engineering Marquee Universiy, P.O.
Chapter 6: Business Valuation (Income Approach)
Chaper 6: Business Valuaion (Income Approach) Cash flow deerminaion is one of he mos criical elemens o a business valuaion. Everyhing may be secondary. If cash flow is high, hen he value is high; if he
A Note on Using the Svensson procedure to estimate the risk free rate in corporate valuation
A Noe on Using he Svensson procedure o esimae he risk free rae in corporae valuaion By Sven Arnold, Alexander Lahmann and Bernhard Schwezler Ocober 2011 1. The risk free ineres rae in corporae valuaion
USE OF EDUCATION TECHNOLOGY IN ENGLISH CLASSES
USE OF EDUCATION TECHNOLOGY IN ENGLISH CLASSES Mehme Nuri GÖMLEKSİZ Absrac Using educaion echnology in classes helps eachers realize a beer and more effecive learning. In his sudy 150 English eachers were
4. International Parity Conditions
4. Inernaional ariy ondiions 4.1 urchasing ower ariy he urchasing ower ariy ( heory is one of he early heories of exchange rae deerminaion. his heory is based on he concep ha he demand for a counry's currency
Duration and Convexity ( ) 20 = Bond B has a maturity of 5 years and also has a required rate of return of 10%. Its price is $613.
Graduae School of Business Adminisraion Universiy of Virginia UVA-F-38 Duraion and Convexiy he price of a bond is a funcion of he promised paymens and he marke required rae of reurn. Since he promised
Markov Chain Modeling of Policy Holder Behavior in Life Insurance and Pension
Markov Chain Modeling of Policy Holder Behavior in Life Insurance and Pension Lars Frederik Brand Henriksen 1, Jeppe Woemann Nielsen 2, Mogens Seffensen 1, and Chrisian Svensson 2 1 Deparmen of Mahemaical
DOES TRADING VOLUME INFLUENCE GARCH EFFECTS? SOME EVIDENCE FROM THE GREEK MARKET WITH SPECIAL REFERENCE TO BANKING SECTOR
Invesmen Managemen and Financial Innovaions, Volume 4, Issue 3, 7 33 DOES TRADING VOLUME INFLUENCE GARCH EFFECTS? SOME EVIDENCE FROM THE GREEK MARKET WITH SPECIAL REFERENCE TO BANKING SECTOR Ahanasios
Term Structure of Prices of Asian Options
Term Srucure of Prices of Asian Opions Jirô Akahori, Tsuomu Mikami, Kenji Yasuomi and Teruo Yokoa Dep. of Mahemaical Sciences, Risumeikan Universiy 1-1-1 Nojihigashi, Kusasu, Shiga 525-8577, Japan E-mail:
Supplementary Appendix for Depression Babies: Do Macroeconomic Experiences Affect Risk-Taking?
Supplemenary Appendix for Depression Babies: Do Macroeconomic Experiences Affec Risk-Taking? Ulrike Malmendier UC Berkeley and NBER Sefan Nagel Sanford Universiy and NBER Sepember 2009 A. Deails on SCF
Relationships between Stock Prices and Accounting Information: A Review of the Residual Income and Ohlson Models. Scott Pirie* and Malcolm Smith**
Relaionships beween Sock Prices and Accouning Informaion: A Review of he Residual Income and Ohlson Models Sco Pirie* and Malcolm Smih** * Inernaional Graduae School of Managemen, Universiy of Souh Ausralia
Chapter 7. Response of First-Order RL and RC Circuits
Chaper 7. esponse of Firs-Order L and C Circuis 7.1. The Naural esponse of an L Circui 7.2. The Naural esponse of an C Circui 7.3. The ep esponse of L and C Circuis 7.4. A General oluion for ep and Naural
Multiprocessor Systems-on-Chips
Par of: Muliprocessor Sysems-on-Chips Edied by: Ahmed Amine Jerraya and Wayne Wolf Morgan Kaufmann Publishers, 2005 2 Modeling Shared Resources Conex swiching implies overhead. On a processing elemen,
Risk Modelling of Collateralised Lending
Risk Modelling of Collaeralised Lending Dae: 4-11-2008 Number: 8/18 Inroducion This noe explains how i is possible o handle collaeralised lending wihin Risk Conroller. The approach draws on he faciliies
MACROECONOMIC FORECASTS AT THE MOF A LOOK INTO THE REAR VIEW MIRROR
MACROECONOMIC FORECASTS AT THE MOF A LOOK INTO THE REAR VIEW MIRROR The firs experimenal publicaion, which summarised pas and expeced fuure developmen of basic economic indicaors, was published by he Minisry
Mathematics in Pharmacokinetics What and Why (A second attempt to make it clearer)
Mahemaics in Pharmacokineics Wha and Why (A second aemp o make i clearer) We have used equaions for concenraion () as a funcion of ime (). We will coninue o use hese equaions since he plasma concenraions
How To Calculate Price Elasiciy Per Capia Per Capi
Price elasiciy of demand for crude oil: esimaes for 23 counries John C.B. Cooper Absrac This paper uses a muliple regression model derived from an adapaion of Nerlove s parial adjusmen model o esimae boh
Single-machine Scheduling with Periodic Maintenance and both Preemptive and. Non-preemptive jobs in Remanufacturing System 1
Absrac number: 05-0407 Single-machine Scheduling wih Periodic Mainenance and boh Preempive and Non-preempive jobs in Remanufacuring Sysem Liu Biyu hen Weida (School of Economics and Managemen Souheas Universiy
Principal components of stock market dynamics. Methodology and applications in brief (to be updated ) Andrei Bouzaev, bouzaev@ya.
Principal componens of sock marke dynamics Mehodology and applicaions in brief o be updaed Andrei Bouzaev, [email protected] Why principal componens are needed Objecives undersand he evidence of more han one
Measuring macroeconomic volatility Applications to export revenue data, 1970-2005
FONDATION POUR LES ETUDES ET RERS LE DEVELOPPEMENT INTERNATIONAL Measuring macroeconomic volailiy Applicaions o expor revenue daa, 1970-005 by Joël Cariolle Policy brief no. 47 March 01 The FERDI is a
Vector Autoregressions (VARs): Operational Perspectives
Vecor Auoregressions (VARs): Operaional Perspecives Primary Source: Sock, James H., and Mark W. Wason, Vecor Auoregressions, Journal of Economic Perspecives, Vol. 15 No. 4 (Fall 2001), 101-115. Macroeconomericians
Finance and Economics Discussion Series Divisions of Research & Statistics and Monetary Affairs Federal Reserve Board, Washington, D.C.
Finance and Economics Discussion Series Divisions of Research & Saisics and Moneary Affairs Federal Reserve Board, Washingon, D.C. The Effecs of Unemploymen Benefis on Unemploymen and Labor Force Paricipaion:
I. Basic Concepts (Ch. 1-4)
(Ch. 1-4) A. Real vs. Financial Asses (Ch 1.2) Real asses (buildings, machinery, ec.) appear on he asse side of he balance shee. Financial asses (bonds, socks) appear on boh sides of he balance shee. Creaing
Network Effects, Pricing Strategies, and Optimal Upgrade Time in Software Provision.
Nework Effecs, Pricing Sraegies, and Opimal Upgrade Time in Sofware Provision. Yi-Nung Yang* Deparmen of Economics Uah Sae Universiy Logan, UT 84322-353 April 3, 995 (curren version Feb, 996) JEL codes:
Hedging with Forwards and Futures
Hedging wih orwards and uures Hedging in mos cases is sraighforward. You plan o buy 10,000 barrels of oil in six monhs and you wish o eliminae he price risk. If you ake he buy-side of a forward/fuures
How To Predict A Person'S Behavior
Informaion Theoreic Approaches for Predicive Models: Resuls and Analysis Monica Dinculescu Supervised by Doina Precup Absrac Learning he inernal represenaion of parially observable environmens has proven
Does Option Trading Have a Pervasive Impact on Underlying Stock Prices? *
Does Opion Trading Have a Pervasive Impac on Underlying Sock Prices? * Neil D. Pearson Universiy of Illinois a Urbana-Champaign Allen M. Poeshman Universiy of Illinois a Urbana-Champaign Joshua Whie Universiy
Segmentation, Probability of Default and Basel II Capital Measures. for Credit Card Portfolios
Segmenaion, Probabiliy of Defaul and Basel II Capial Measures for Credi Card Porfolios Draf: Aug 3, 2007 *Work compleed while a Federal Reserve Bank of Philadelphia Dennis Ash Federal Reserve Bank of Philadelphia
Market Liquidity and the Impacts of the Computerized Trading System: Evidence from the Stock Exchange of Thailand
36 Invesmen Managemen and Financial Innovaions, 4/4 Marke Liquidiy and he Impacs of he Compuerized Trading Sysem: Evidence from he Sock Exchange of Thailand Sorasar Sukcharoensin 1, Pariyada Srisopisawa,
Economics Honors Exam 2008 Solutions Question 5
Economics Honors Exam 2008 Soluions Quesion 5 (a) (2 poins) Oupu can be decomposed as Y = C + I + G. And we can solve for i by subsiuing in equaions given in he quesion, Y = C + I + G = c 0 + c Y D + I
DEMAND FORECASTING MODELS
DEMAND FORECASTING MODELS Conens E-2. ELECTRIC BILLED SALES AND CUSTOMER COUNTS Sysem-level Model Couny-level Model Easside King Couny-level Model E-6. ELECTRIC PEAK HOUR LOAD FORECASTING Sysem-level Forecas
Double Entry System of Accounting
CHAPTER 2 Double Enry Sysem of Accouning Sysem of Accouning \ The following are he main sysem of accouning for recording he business ransacions: (a) Cash Sysem of Accouning. (b) Mercanile or Accrual Sysem
Chapter 4: Exponential and Logarithmic Functions
Chaper 4: Eponenial and Logarihmic Funcions Secion 4.1 Eponenial Funcions... 15 Secion 4. Graphs of Eponenial Funcions... 3 Secion 4.3 Logarihmic Funcions... 4 Secion 4.4 Logarihmic Properies... 53 Secion
Economic Analysis of 4G Network Upgrade
Economic Analysis of ework Upgrade Lingjie Duan, Jianwei Huang, and Jean Walrand Absrac As he successor o he sandard, provides much higher daa raes o address cellular users ever-increasing demands for
Usefulness of the Forward Curve in Forecasting Oil Prices
Usefulness of he Forward Curve in Forecasing Oil Prices Akira Yanagisawa Leader Energy Demand, Supply and Forecas Analysis Group The Energy Daa and Modelling Cener Summary When people analyse oil prices,
Stochastic Optimal Control Problem for Life Insurance
Sochasic Opimal Conrol Problem for Life Insurance s. Basukh 1, D. Nyamsuren 2 1 Deparmen of Economics and Economerics, Insiue of Finance and Economics, Ulaanbaaar, Mongolia 2 School of Mahemaics, Mongolian
INTEREST RATE FUTURES AND THEIR OPTIONS: SOME PRICING APPROACHES
INTEREST RATE FUTURES AND THEIR OPTIONS: SOME PRICING APPROACHES OPENGAMMA QUANTITATIVE RESEARCH Absrac. Exchange-raded ineres rae fuures and heir opions are described. The fuure opions include hose paying
Stock Trading with Recurrent Reinforcement Learning (RRL) CS229 Application Project Gabriel Molina, SUID 5055783
Sock raing wih Recurren Reinforcemen Learning (RRL) CS9 Applicaion Projec Gabriel Molina, SUID 555783 I. INRODUCION One relaively new approach o financial raing is o use machine learning algorihms o preic
INTRODUCTION TO FORECASTING
INTRODUCTION TO FORECASTING INTRODUCTION: Wha is a forecas? Why do managers need o forecas? A forecas is an esimae of uncerain fuure evens (lierally, o "cas forward" by exrapolaing from pas and curren
A Universal Pricing Framework for Guaranteed Minimum Benefits in Variable Annuities *
A Universal Pricing Framework for Guaraneed Minimum Benefis in Variable Annuiies * Daniel Bauer Deparmen of Risk Managemen and Insurance, Georgia Sae Universiy 35 Broad Sree, Alana, GA 333, USA Phone:
ARCH 2013.1 Proceedings
Aricle from: ARCH 213.1 Proceedings Augus 1-4, 212 Ghislain Leveille, Emmanuel Hamel A renewal model for medical malpracice Ghislain Léveillé École d acuaria Universié Laval, Québec, Canada 47h ARC Conference
Cointegration: The Engle and Granger approach
Coinegraion: The Engle and Granger approach Inroducion Generally one would find mos of he economic variables o be non-saionary I(1) variables. Hence, any equilibrium heories ha involve hese variables require
Measuring the Effects of Monetary Policy: A Factor-Augmented Vector Autoregressive (FAVAR) Approach * Ben S. Bernanke, Federal Reserve Board
Measuring he Effecs of Moneary Policy: A acor-augmened Vecor Auoregressive (AVAR) Approach * Ben S. Bernanke, ederal Reserve Board Jean Boivin, Columbia Universiy and NBER Pior Eliasz, Princeon Universiy
Interpersonal communications have long been recognized as an influential source of information for consumers.
CELEBRATING 30 YEARS Vol. 30, No. 4, July Augus 2011, pp. 702 716 issn 0732-2399 eissn 1526-548X 11 3004 0702 doi 10.1287/mksc.1110.0642 2011 INFORMS A Dynamic Model of he Effec of Online Communicaions
Measuring the Effects of Exchange Rate Changes on Investment. in Australian Manufacturing Industry
Measuring he Effecs of Exchange Rae Changes on Invesmen in Ausralian Manufacuring Indusry Robyn Swif Economics and Business Saisics Deparmen of Accouning, Finance and Economics Griffih Universiy Nahan
Strategic Optimization of a Transportation Distribution Network
Sraegic Opimizaion of a Transporaion Disribuion Nework K. John Sophabmixay, Sco J. Mason, Manuel D. Rossei Deparmen of Indusrial Engineering Universiy of Arkansas 4207 Bell Engineering Cener Fayeeville,
Premium Income of Indian Life Insurance Industry
Premium Income of Indian Life Insurance Indusry A Toal Facor Produciviy Approach Ram Praap Sinha* Subsequen o he passage of he Insurance Regulaory and Developmen Auhoriy (IRDA) Ac, 1999, he life insurance
Optimal Investment and Consumption Decision of Family with Life Insurance
Opimal Invesmen and Consumpion Decision of Family wih Life Insurance Minsuk Kwak 1 2 Yong Hyun Shin 3 U Jin Choi 4 6h World Congress of he Bachelier Finance Sociey Torono, Canada June 25, 2010 1 Speaker
GoRA. For more information on genetics and on Rheumatoid Arthritis: Genetics of Rheumatoid Arthritis. Published work referred to in the results:
For more informaion on geneics and on Rheumaoid Arhriis: Published work referred o in he resuls: The geneics revoluion and he assaul on rheumaoid arhriis. A review by Michael Seldin, Crisopher Amos, Ryk
Optimal Growth for P&C Insurance Companies
Opimal Growh for P&C Insurance Companies by Luyang Fu AbSTRACT I is generally well esablished ha new business produces higher loss and expense raios and lower reenion raios han renewal business. Ironically,
Contrarian insider trading and earnings management around seasoned equity offerings; SEOs
Journal of Finance and Accounancy Conrarian insider rading and earnings managemen around seasoned equiy offerings; SEOs ABSTRACT Lorea Baryeh Towson Universiy This sudy aemps o resolve he differences in
Making a Faster Cryptanalytic Time-Memory Trade-Off
Making a Faser Crypanalyic Time-Memory Trade-Off Philippe Oechslin Laboraoire de Securié e de Crypographie (LASEC) Ecole Polyechnique Fédérale de Lausanne Faculé I&C, 1015 Lausanne, Swizerland [email protected]
DO I FOLLOW MY FRIENDS OR THE CROWD? INFORMATION CASCADES IN ONLINE MOVIE RATING
DO I FOLLOW MY FRIENDS OR THE CROWD? INFORMATION CASCADES IN ONLINE MOVIE RATING Young Jin Lee Yong Tan Karik Hosanagar Ausin E. Cofrin School of Business, Universiy of Wisconsin, Green Bay, WI 54311 Michael
BALANCE OF PAYMENTS. First quarter 2008. Balance of payments
BALANCE OF PAYMENTS DATE: 2008-05-30 PUBLISHER: Balance of Paymens and Financial Markes (BFM) Lena Finn + 46 8 506 944 09, [email protected] Camilla Bergeling +46 8 506 942 06, [email protected]
The option pricing framework
Chaper 2 The opion pricing framework The opion markes based on swap raes or he LIBOR have become he larges fixed income markes, and caps (floors) and swapions are he mos imporan derivaives wihin hese markes.
Time Series Analysis Using SAS R Part I The Augmented Dickey-Fuller (ADF) Test
ABSTRACT Time Series Analysis Using SAS R Par I The Augmened Dickey-Fuller (ADF) Tes By Ismail E. Mohamed The purpose of his series of aricles is o discuss SAS programming echniques specifically designed
Default Risk in Equity Returns
Defaul Risk in Equiy Reurns MRI VSSLOU and YUHNG XING * BSTRCT This is he firs sudy ha uses Meron s (1974) opion pricing model o compue defaul measures for individual firms and assess he effec of defaul
LECTURE: SOCIAL SECURITY HILARY HOYNES UC DAVIS EC230 OUTLINE OF LECTURE:
LECTURE: SOCIAL SECURITY HILARY HOYNES UC DAVIS EC230 OUTLINE OF LECTURE: 1. Inroducion and definiions 2. Insiuional Deails in Social Securiy 3. Social Securiy and Redisribuion 4. Jusificaion for Governmen
Dependent Interest and Transition Rates in Life Insurance
Dependen Ineres and ransiion Raes in Life Insurance Krisian Buchard Universiy of Copenhagen and PFA Pension January 28, 2013 Absrac In order o find marke consisen bes esimaes of life insurance liabiliies
From Generic to Branded: A Model of Spillover in Paid Search Advertising
OLIVER J. RUTZ and RNDOLPH E. BUCKLIN* In Inerne paid search adverising, markeers pay for search engines o serve ex adverisemens in response o keyword searches ha are generic (e.g., hoels ) or branded
