Dynamic Cost-Per-Action Mechanisms and Applications to Online Advertising
|
|
- Chester Bell
- 8 years ago
- Views:
Transcription
1 Dynamic Cos-Per-Acion Mechanisms and Applicaions o Online Adverising Hamid Nazerzadeh Sanford Universiy Sanford, CA hamidnz@sanford.edu Amin Saberi Sanford Universiy Sanford, CA saberi@sanford.edu Rakesh Vohra Norhwesern Universiy Evanson, IL r-vohra@kellogg.nwu.edu ABSTRACT We sudy he Cos-Per-Acion or Cos-Per-Acquisiion (CPA) charging scheme in online adverising. In his scheme, insead of paying per click, he adverisers pay only when a user akes a specific acion (e.g. fills ou a form) or complees a ransacion on heir websies. We focus on designing efficien and incenive compaible mechanisms ha use his charging scheme. We describe a mechanism based on a sampling-based learning algorihm ha under suiable assumpions is asympoically individually raional, asympoically Bayesian incenive compaible and asympoically ex-ane efficien. In paricular, we demonsrae our mechanism for he case where he uiliy funcions of he adverisers are independen and idenically-disribued random variables as well as he case where hey evolve like independen refleced Brownian moions. Caegories and Subjec Descripors J.4 [Social and Behavioral Sciences]: Economics; F.2.0 [Analysis of Algorihms and Problem Complexiy]: General; I.2.6 [Arificial Inelligence]: Learning General Terms Economics, Algorihm, Theory Keywords Mechanism Design, Cos-Per-Acion, Inerne Adverising. INTRODUCTION Currenly, he main wo charging models in he online adverising indusry are cos-per-impression (CPM) and cosper-click (CPC). In he CPM model, he adverisers pay he publisher for he impression of heir ads. CPM is commonly used in radiional media (e.g. magazines and elevision) or banner adverising and is more suiable when he goal of he adveriser is o increase brand awareness. A more aracive and more popular charging model in online adverising is he CPC model in which he adverisers pay he publisher only when a user clicks on heir ads. In he las few years, here has been a remendous shif owards he CPC charging model. CPC is adoped by search engines Copyrigh is held by he Inernaional World Wide Web Conference Commiee (IW3C2). Disribuion of hese papers is limied o classroom use, and personal use by ohers. WWW 2008, April 2 25, 2008, Beijing, China. ACM /08/04. such as Google or Yahoo! for he placemen of ads nex o search resuls (also known as sponsored search) and on he websie of hird-pary publishers. In his paper we will focus on anoher naural and widely advocaed charging scheme known as he Cos-Per-Acion or Cos-Per-Acquisiion (CPA) model. In his model, insead of paying per click, he adveriser pays only when a user akes a specific acion (e.g. fills ou a form) or complees a ransacion. Recenly, several companies like Google, ebay, Amazon, Adverising.com, and Snap.com have sared o sell adverising in his way. CPA models can be he ideal charging scheme, especially for small and risk averse adverisers. We will briefly describe a few advanages of his charging scheme over CPC and refer he reader o [8] for a more deailed discussion. One of he drawbacks of he CPC scheme is ha i requires he adverisers o submi heir bids before observing he profis generaed by he users clicking on heir ads. Learning he expeced value of each click, and herefore he righ bid for he ad, is a prohibiively difficul ask especially in he conex of sponsored search in which he adverisers ypically bid for housands of keywords. CPA eliminaes his problem because i allows he adverisers o repor heir payoff afer observing he user s acion. Anoher drawback of he CPC scheme is is vulnerabiliy o click fraud. Click fraud refers o clicks generaed by someone or somehing wih no genuine ineres in he adverisemen. Such clicks can be generaed by he publisher of he conen who has an ineres in receiving a share of he revenue of he ad or by a rival who wishes o increase he cos of adverising for he adveriser. Click fraud is considered by many expers o be he bigges challenge facing he online adverising indusry [3, 0, 23, 20]. CPA schemes are less vulnerable because generaing a fraudulen acion is ypically more cosly han generaing a fraudulen click. For example, an adveriser can define he acion as a sale and pay he publisher only when he ad yields profi. On he oher hand, here is a fundamenal difference beween CPA and CPC charging models. A click on he ad can be observed by boh adveriser and publisher. However, he acion of he user is hidden from he publisher and is observable only by he adveriser. Alhough he publisher can require he adverisers o insall a sofware ha will monior acions ha ake place on heir web sie, even moderaely sophisicaed adverisers can find a way o manipulae he sofware if hey find i sufficienly profiable. CPA makes generaing a fraudulen acion a more cosly enerprize, bu no impossible (e.g., using a solen credi). 79
2 Are he publishers exposed o he manipulaion or misreporing of he adverisers in he CPA scheme? Does CPA creae an incenive for he adverisers o misrepor he number of acions or heir payoffs for he acions? The main resul of his paper is o give a negaive answer o hese quesions. We design a mechanism ha, asympoically and under reasonable assumpions, removes he incenives of he adverisers o misrepor heir payoffs. A he same ime, our mechanism has he same asympoic efficiency and hence revenue as he currenly used CPC mechanisms. We will use echniques in learning and mechanism design o obain his resul. In he nex secion, we will formally describe our model in mechanism design erminology (see [2].) We will refer o adverisers as agens and o he impression of an ad as an iem. For simpliciy of exposiion only, we assume only one adverisemen slo per page. In secion 6 we ouline how o exend our resuls o he case where more han one adverisemen can be displayed in each page. Alhough our work is essenially moivaed by online adverising, we believe ha he applicaion of our mechanism is no limied his domain.. Model We sudy he following problem: here are a number of self-ineresed agens compeing for idenical iems sold repeaedly a imes =, 2,. A each ime, a mechanism allocaes he iem o one of he agens. Agens discover heir uiliy for he good only if i is allocaed o hem. If agen i receives he good a ime, she discovers uiliy u i (denominaed in money) for i and repors (no necessarily ruhfully) he realized uiliy o he mechanism. Then, he mechanism deermines how much he agen has o pay for receiving he iem. We allow he uiliy of an agen o change over ime. For his environmen we are ineresed in aucion mechanisms which have he following four properies.. The mechanism is individually raional in each period. 2. Agens have an incenive o ruhfully repor heir realized uiliies. 3. The efficiency (and revenue) is, in an appropriae sense, no oo small compared o a second price aucion. 4. The correcness of he mechanism does no depend on an a-priori knowledge of he disribuion of u i s. This feaure is moivaed by he Wilson docrine [24] 2. The precise manner in which hese properies are formalized is described in secion 2. We will build our mechanisms on a sampling-based learning algorihm. The learning algorihm is used o esimae he expeced uiliy of he agens, and consiss of wo alernaing phases: exploraion and exploiaion. During an exploraion phase, he iem is allocaed for free o a randomly chosen agen. During an exploiaion phase, he mechanism allocaes he iem o he agen wih he highes esimaed expeced uiliy. Afer each allocaion, he agen who has received he iem, discovers her uiliy and repors i o he mechanism. Subsequenly, he mechanism updaes he esimae of uiliies and deermines he paymen. 2 Wilson criicizes relying oo much on common-knowledge assumpions. We characerize a class of learning algorihms ha ensure ha he corresponding mechanism has he four desired properies. The main difficuly in obaining his resul is he following: since here is uncerainy abou he uiliies, i is possible ha in some periods he iem is allocaed o an agen who does no have he highes uiliy in ha period. Hence, he naural second-highes price paymen rule would violae individual raionaliy. On he oher hand, if he mechanism does no charge an agen because her repored uiliy afer he allocaion is low, i gives her an incenive o shade her repored uiliy down. Our mechanism solves hese problems by using an adapive, cumulaive pricing scheme. We illusrae our resuls by idenifying simple mechanisms ha have he desired properies. We demonsrae hese mechanisms for he case in which he u i s are independen and idenically-disribued random variables as well as he case where heir expeced values evolve like independen refleced Brownian moions. In hese cases he mechanism is acually ex-pos individually raional. In our proposed mechanism, he agens do no have o bid for he iems. This is advanageous when he bidders hemselves are unaware of heir uiliy values. However, in some cases, an agen migh have a beer esimae of her uiliy for he iem han our mechanism. For his reason, we describe how we can slighly modify our mechanism o allow hose agens o bid direcly..2 Relaed Work There is a large number of ineresing resuls on using machine learning echniques in mechanism design. We only briefly survey he main echniques and ideas and compare hem wih he approach of his paper. Mos of hese works, like [5, 8,?, 7], consider one-sho games or repeaed aucions in which he agens leave he environmen afer hey received an iem. In our seing we may allocaes iems o an agen several imes and hence, we need o consider he sraegic behavior of he agens over ime. There is also a big lieraure on regre minimizaion or exper algorihms. In our conex, hese algorihms are applicable even if he uiliies of he agens are changing arbirarily. However, he efficiency (and herefore he revenue) of hese algorihms is comparable o he mechanisms ha allocaes he iem o he single bes agen (exper) (e.g. see [6]). Our goal is more ambiious: our efficiency is close he mos efficien allocaion which migh allocae he iem o differen agens a differen imes. On he oher hand, we focus on uiliy values ha change smoohly (e.g. like a Brownian moion). In a finiely repeaed version of he environmen considered here, Ahey and Segal [2] consruc an efficien, budge balanced, mechanism where ruhful revelaion in each period is Bayesian incenive compaible. Bapna and Weber [4] consider he infinie horizon version of [2] and propose a class of incenive compaible mechanisms based on he Giins index (see []). Taking a differen approach, Bergemann and Välimäki [6] and Cavallo e al. [9] propose an incenive compaible generalizaion of he Vickrey-Clark-Groves mechanism based on he marginal conribuion of each agen for his environmen. All hese mechanisms need he exac soluion of he underlying opimizaion problems, and herefore require complee informaion abou he prior of he uiliies 80
3 of he agens; also, hey do no apply when he evoluion of he uiliies of he agens is no saionary over ime. This violaes he las of our desideraa. For a comprehensive survey in dynamic mechanism design lieraure see [22]. In he conex of sponsored search, aenion has focused on ways of esimaing click hrough raes. Gonen and Pavlov [2] give a mechanism which learns he click-hrough raes via sampling and show ha ruhful bidding is, wih high probabiliy, a (weakly) dominan sraegy in his mechanism. Along his line, Worman e al. [25] inroduced an exploraion scheme for learning adverisers click-hrough raes in sponsored search which mainains he equilibrium of he sysem. In hese works, unlike ours, he disribuion of he uiliies of agens are assumed o be fixed over ime. Immorlica e al. [4], and laer Mahdian and Tomak [8], examine he vulnerabiliy of various procedures for esimaing click hrough, and idenify a class of click hrough learning algorihms in which fraudulen clicks canno increase he expeced paymen per impression by more han o(). This is under he assumpion ha he slo of an agen is fixed and he bids of oher agens remain consan overime. In conras, we sudy condiions which guaranee incenive compaibiliy and efficiency, while he uiliy of (all) agens may evolve over ime. 2. DEFINITIONS AND NOTATION Suppose n agens compeing in each period for a single iem. The iem is sold repeaedly a ime =, 2,. Denoe by u i he nonnegaive uiliy of agen i for he iem a ime. Uiliies are denominaed in a common moneary scale. The uiliies of agens may evolve over ime according o a sochasic process. We assume ha for i j, he evoluion of u i and u j are independen sochasic processes. We also define µ i = E[u i u i,, u i,]. Throughou his paper, expecaions are aken condiioned on he complee hisory. For simpliciy of noaion, we now omi hose erms ha denoe such a condiioning. Wih noaional convenion, i follows, for example, ha E[u i] = E[µ i]. Here he second expecaion is aken over all possible hisories. Le M be a mechanism used o sell he iems. A each ime, M allocaes he iem o one of he agens. Le i be he agen who has received he iem a ime. Define x i o be he variable indicaing he allocaion of he iem o i a ime. Afer he allocaion, agen i observes her uiliy, u i, and hen repors r i, as her uiliy for he iem, o he mechanism. Noe ha we do no require an agen o know her uiliy for possessing he iem in advance of acquiring i. The mechanism hen deermines he paymen, denoed by p i. Definiion. An agen i is ruhful if r i = u i, for all ime x i =, > 0. Our goal is o design a mechanism which has he following properies. We assume n, he number of agens, is consan. Individual Raionaliy: A mechanism is ex-pos individually raional if for any ime T > 0 and any agen i n, he oal paymen of agen i does no exceed he sum of her repors: x ir i p i > 0. M is asympoically ex-ane individually raional if: lim inf E[ T x iµ i p i] 0. T Incenive Compaibiliy: This propery implies ha ruhfulness defines an asympoic Bayesian Nash equilibrium. Consider agen i and suppose all agens excep i are ruhful. Le U i(t) be he expeced oal profi of agen i, if agen i is ruhful beween ime and T. Also, le e U i(t) be he maximum of expeced profi of agen i under any oher sraegy. Asympoic incenive compaibiliy requires ha eu i(t) U i(t) = o(u i(t)). Efficiency: An ex-ane efficien mechanism allocaes he iem o an agen in argmax i {µ i} a each ime (and for each hisory). The oal social welfare obained by an ex-ane efficien mechanism up o ime T is maxi{µi}]. Le W(T) be he expeced welfare of mechanism M beween ime and T, when all agens are ruhful, i.e., E[ P T n W(T) = E[ i= x iµ i] Then, M is asympoically ex-ane efficien if: E[ max{µ i}] W(T) = o(w(t)). i 3. PROPOSED MECHANISM We build our mechanism on op of a learning algorihm ha esimaes he expeced uiliy of he agens. We refrain from an explici descripion of he learning algorihm. Raher, we describe sufficien condiions for a learning algorihm ha can be exended o a mechanism wih all he properies we seek (see secion 3.). In secion 4 and 5 we give wo examples of environmens where learning algorihms saisfying hese sufficien condiions exis. The mechanism consiss of wo phases: explore and exploi. During he explore phase, wih probabiliy η(), η : N [0,], he iem is allocaed for free o a randomly chosen agen. During he exploi phase, he mechanism allocaes he iem o he agen wih he highes esimaed expeced uiliy. Aferwards, he agen repors her uiliy o he mechanism and he mechanism deermines he paymen. We firs formalize our assumpions abou he learning algorihm and hen we discuss he paymen scheme. The mechanism is given in Figure. The learning algorihm, samples u i s a rae η(), and based on he hisory of he repors of agen i, reurns an esimae of µ i. Le bµ i(t) be he esimae of he algorihm for µ i condiional on he hisory of he repors up o ime T. The hisory of he repors of agen i up o ime T is he sequence of he repored values and imes of observaion of u i up o bu no including ime T. Noe ha we allow T >. Thus, informaion a ime T > can be used o revise an esimae of µ i made a some earlier ime. We assume ha increasing he number of samples only increases he 8
4 For =,2,... Wih probabiliy η(), explore: Uniformly a random, allocae he iem o an agen i, i n. p i 0 Wih probabiliy η(), exploi: Randomly allocae he iem o an agen i argmax i {bµ i()}. p i P y ik min{bγ k (), bµ ik (k)} P p ik r i he repor of agen i. p j 0, j i Figure : Mechanism M accuracy of he esimaions, i.e. for any ruhful agen i, and imes T T 2: E[ bµ i(t ) µ i ] E[ bµ i(t 2) µ i ]. () In he inequaliy above, and in he res of he paper, he expecaions of bµ i are aken over he evoluion of u i s and he random choices of he mechanism. For simpliciy of noaion, we omi hose erms ha denoe such a condiioning. To describe he paymens recall ha γ is he second highes µ i and le bγ (T) = max j i {bµ j(t)}, where i is he agen who received he iem a ime. We define y i o be he indicaor variable of he allocaion of he iem o agen i during an exploi phase. The paymen of agen i a ime, denoed p i, is deermined so ha: p ik = y ik min{bγ k (), bµ ik (k)}. An agen only pays for iems ha are allocaed o her during he exploi phase, up o bu no including ime. A ime, he paymen of agen i for he iem she received a ime k < is min{bγ k (), bµ ik (k)}. The firs erm is he reminiscence of he second highes pricing scheme. The second erm, under some reasonable condiions, leads o individually raionaliy. Since he esimaions of learning algorihm for he uiliies of agens become more precise over ime, our adapive cumulaive paymen scheme allows i o correc for errors in he pas. 3. Sufficien Condiions We sar wih a condiion ha guaranees asympoic exane individual raionaliy and asympoic incenive compaibiliy. Le l i be he las ime up o ime ha he iem is allocaed o agen i wihin an exploi phase. If i has no been allocaed any iem ye, l i is defined o be zero. Also, define = max i{ bµ i() µ i }, assuming all agens were ruhful up o ime. Theorem. If for he learning algorihm, for all i n, and T > 0: (C) E[max T {µ i} + P T ] = o(e[p T η()µi]) hen mechanism M is asympoically ex-ane individually raional and incenive compaible. We ouline he proof firs. As we prove in Lemma 2, by condiion (C), he expeced profi of a ruhful agen up o ime T is a leas ( o())e[p T n η()µi]. Also, he expeced oal error in he esimaes of he paymens up o ime T is bounded by O(E[ P T ]). We prove ha he oal uiliy an agen could obain by deviaing from he ruhful sraegy, beween ime and T, is bounded by O(max T {µ i} + E[ P T ]). Hence, he claim follows by condiion (C). Similar o oher applicaions of learning algorihms, we can observe a naural rade-off beween exploiaion and exploraion raes in our conex: higher exploraion raes lead o more accurae esimaes of he uiliies of he agens, a he cos of efficiency. Condiion (C) provides us wih a lower bound on he exploraion rae. Lemma 2. If condiion (C) holds, hen he expeced profi of a ruhful agen i up o ime T, U i(t), is a leas: ( T n o())e[ η()µ i]. Proof. The iems ha agen i receives during he explore phase are free. The expeced oal uiliy of agen i from hese iems up o ime T is E[P T n η()µi]. Le C T = { < l it y i =, if i is ruhful} be he subse of periods ha agen i is charged for he iem she received wihin he period. U i(t) = E[ x iu i p i] = E[ x iu i] + E[ u i p i] / C T C T T n E[ η()µ i] +E[ C T (µ i min{bγ (T), bµ i()})] (2) 82
5 For C T: E[(µ i min{bγ (T), bµ i()})i( C T)] E[(µ i bµ i())i( C T)] E[ µ i bµ i() ] E[ ] Subsiuing ino inequaliy (2), by condiion (C): U i(t) T n E[ η()µ i] E[ ] = T n E[ η()µ i] o(e[ η()µ i]) (3) Proof of Theorem : Lemma 2 yields asympoic exane individual raionaliy. We show ha ruhfulness is asympoically a bes response when all oher agens are ruhful. Fix an agen i inending o deviae and le S be he sraegy she deviaes o. Fixing he evoluion of all u j s, j n, and all random choices of he mechanism, i.e. he seps in he explore phase and he randomly chosen agens, le D T be he imes ha i receives he iem under sraegy S during he exploi phase before ime l it, i.e. D T = { < l it y i =, if he sraegy of i is S}. Similarly, le C T = { < l it y i =, if i is ruhful}. Also, le bµ i, and bγ correspond o he esimaes of he mechanism when he sraegy of i is S. We firs bound he expeced profi of i, under sraegy S, during he exploi phase: E[ y iu i p i] E[ D T µ i min{bγ (T), bµ i()}] + (4) E[max {µi}] T = E[ µ i min{bγ (T), bµ i()}] + D T \C T E[ µ i min{bγ (T), bµ i()}] + D T C T E[max{µi}] (5) T The erm E[max T {µ i}] bounds he ousanding paymen of agen i; recall ha he agen has no paid for he las allocaed iem. For ime, we examine wo cases:. If D T C T, hen agen i, in expecaion, canno decrease he curren price, min{bγ (T), bµ i()}, by more han O( ): min{bγ (T), bµ i()} where (z) + = max{z,0}. min{bγ (T), bγ ()} γ max{γ bγ (T),γ bγ ()} γ (γ bγ (T)) + (γ bγ ()) + Recall ha bγ (T) = max j i {bµ i(t)} and all oher agen are ruhful. Hence, aking expecaion from boh sides, by (): E[min{bγ (T), bµ i()}i( D T C T)] E[(γ (γ bγ (T)) + (γ bγ ()) + )I( D T C T)] E[γ I( D T C T)] E[2 ] (6) 2. If D T \ C T, agen i canno increase her expeced profi, µ i min{bγ (T), bµ i()}, by more han O( ): µ i min{bγ (T), bµ i()} µ i min{bγ (T), bγ ()} (µ i bµ i()) + (bµ i() γ ) + max{γ bγ (T),γ bγ ()} 2 + (γ bγ (T)) + +(γ bγ ()) + Taking expecaion from boh sides, by (): E[(µ i min{bγ (T), bµ i()})i( D T C T)] E[2 I( D T C T)] + E[((γ bγ (T)) + + (γ bγ ()) + )I( D T C T)] E[4 ] (7) Subsiuing inequaliies (6) and (7) ino (5): E[ y iu i p i] E[ 6 ] + E[max {µi}] T +E[ µ i γ ] D T C T E[ 6 ] + E[max{µi}] + (8) T E[ µ i γ ] E[ µ i γ ] C T C T \D T For C T, since bµ i() bγ (), we have: Subsiuing ino (8): E[γ µ i] E[2 ] E[ y iu i p i] 8E[ ] + E[max {µi}] T +E[ C T µ i γ ] Wih algebraic manipulaion, using (), we ge: E[ y iu i p i] O(E[ ] + E[max {µi}]) T +E[ µ i min{bγ (T), bµ i()}] C T By condiion (C), we ge he inequaliy below which complees he proof: E[ y iu i p i] o(e[ η()µ i]) +E[ C T µ i min{bγ (T), bµ i()}] and he las inequaliy follows by (C). The expeced uiliy of he ruhful sraegy and S during he explore phase 83
6 is equal. Therefore, by Lemma 2, he mechanism is asympoically incenive compaible. In he nex heorem we show if he loss in efficiency during exploraion asympoically goes o zero, hen by Condiion (C) he mechanism is asympoically ex-ane efficien. Theorem 3. If for he learning algorihm, in addiion o (C), he following condiion holds (C2) E[ P T η()maxi{µi}] = o(e[p T maxi{µi}]) hen, M is asympoically ex-ane efficien. Proof. M may fail o be ex-ane efficien for wo reasons. Firs one is he loss in welfare during he exploraion when he iem is allocaed randomly o one of adverisers. The expeced loss in his case is equal o E[ P T η() maxi{µi}]. Anoher reason is he misakes during exploiaion. The error in esimaion can lead o allocaion o an agen who does no value he iem he mos. A ime, in he wors case, he iem migh be allocaed o an agen whose expeced uiliy is a mos 2 less han he highes expeced uiliy. Therefore, he expeced efficiency loss during exploraion is bounded by O(E[ P T ]). Since, for he expeced welfare of M beween ime and T, denoed by W(T), we have: E[ max{µ i i}] W(T) = O(E[ ( + η()max{µ i})]) (9) i Bu, condiion (C) implies: n E[ ] = o(e[ η()µ i}]) Plugging ino (9): i= = θ(e[ η() max{µ i i}]) E[ max{µ i}] W(T) = O(E[ η() max{µ i}]) i i = o(w(t)) The las equaliy is followed by (C2) and implies asympoic ex-ane efficiency. While Condiion (C) gives a lower bound on he exploraion rae, Condiion (C2) gives an upper bound. In he nex secion, we will show wih wo examples how condiions (C) and (C2) can be used o adjus he exploraion rae of a learning algorihm in order o obain efficiency and incenive compaibiliy. Remark. In Theorem 3 we showed ha under some assumpions, he welfare obained by he mechanism is asympoically equivalen o efficien mechanism ha every ime allocaes he iem o he agen wih he highes expeced uiliy. We can give similar condiions o (C2) o guaranee ha he revenue of he mechanism is also asympoically equal o he revenue of he efficien mechanism ha every ime charges he winning agen he second highes expeced uiliy. To avoid repeiion, we refrain from explaining his condiion in deails. 3.2 Allowing agens o bid In mechanism M no agen explicily bids for an iem. Wheher an agen receives an iem or no depends on he hisory of heir repored uiliies and he esimaes ha M forms from hem. This may be advanageous when he bidders hemselves are unaware of wha heir uiliies will be. However, when agens may posses a beer esimae of heir uiliies we would like o make use of ha. For his reason we describe how o modify M so as o allow agens o bid for an iem. If ime occurs during an exploi phase le B be he se of he agens who bid a his ime. The mechanism bids on he behalf of all agen i / B. Denoe by b i he bid of agen i B for he iem a ime. The modificaion of M ses b i = bµ i(), for i / B. Then, he iem is allocaed a random o one of he agens in arg max i b i. If i is he agen who received he iem a ime, le A = {b j j B } {µ j, j / B }. Define γ as he second highes value in A. Le bγ (T) o be equal o max j i b jk. The paymen of agen i will be p i y ik min{bγ k (), b ik } p ik. To incorporae he fac ha bidders can bid for an iem, we mus modify he definiion of ruhfulness. Definiion 2. Agen i is ruhful if:. r i = u i, for all ime x i =,. 2. If i bids a ime, hen E[ b i µ i ] E[ bµ i µ i ]. Noe ha iem 2 does no require ha agen i bid heir acual uiliy only ha heir bid be closer o he mark han he esimae. Wih his modificaion in definiion, Theorems and 3 coninue o hold. 4. INDEPENDENT AND IDENTICALLY DISTRIBUTED UTILITIES In his secion, we assume ha for each i, u i s are independen and idenically-disribued random variables. For simpliciy, we define µ i = E[u i], > 0. Wihou loss of generaliy, we also assume 0 < µ i. In his environmen, he learning algorihm we use is an ε- greedy algorihm for he muli-armed bandi problem 3. Le n i = P xi. For ɛ (0,), we define: n i = x i η ɛ() = min{, n ɛ ln +ɛ } ( ( P T bµ i(t) = x ikr ik )/n it, n it > 0 0, n it = 0 Call he mechanism based on his learning algorihm M ɛ(iid). Lemma 4. If all agens are ruhful, hen, under M ɛ(iid) E[ ] = O( ). ɛ 3 See [3] for a similar algorihm. 84
7 The proof of his lemma is given in appendix A. We show ha M ɛ(iid), for ε, saisfies all he desired 3 properies we discussed in he previous secion. Moreover, i saisfies a sronger noion of individual raionaliy. M ɛ(iid) saisfies ex-pos individual raionaliy if for any agen i, and for all T : p i x ir i Theorem 5. M ɛ(iid) is ex-pos individually raional. Also, for 0 ɛ, Mɛ(iid) is asympoically incenive compaible and ex-ane 3 efficien. Proof. We firs prove ex-pos individual raionaliy. I is sufficien o prove i only for he periods ha agen i has received he iem wihin an exploi phase. For T, such ha y it =, we have: p i = y i min{bγ (T), bµ i()} y ibγ (T) y ibµ it (T) n ibµ it (T) = x ir i The hird inequaliy follows because he iem is allocaed o i a ime T which implies bµ it(t) bγ (T). We complee he proof by showing ha condiions (C) and (C2) hold. Noe ha µ i. By lemma 4, for ɛ 3 : E[+ ] = O(T +ɛ 2 ) = o(t ɛ ln +ɛ T) = O( η ɛ()µ i). Therefore, (C) holds. The welfare of any mechanism beween ime and T is bounded by T. For any ɛ > 0, E[ + P T + η] = o(t) which implies (C2). 5. BROWNIAN MOTION In his secion, we assume for each i, i n, he evoluion of µ i is a refleced Brownian moion wih mean zero and variance σ 2 i ; he reflecion barrier is 0. In addiion, we assume µ i0 = 0, and σ 2 i σ 2, for some consan σ. The mechanism observes he values of µ i a discree imes =, 2,. In his environmen our learning algorihm esimaes he refleced Brownian moion using a mean zero maringale. We define l i is defined as he las ime up o ime ha he iem is allocaed o agen i. This includes boh explore and exploi phases. If i has no been allocaed any iem ye, l i is zero. η ɛ() = min{, n ɛ ln 2+2ɛ } (0) 8 >< r ili < T bµ i(t) = r ili, = T () >: r ili,t > T Call his mechanism M ɛ(b). For simpliciy, we assume ha he adveriser repors he exac value of µ i. I is no difficul o verify ha he resuls in his secion hold as long as he expeced value of he error of hese esimaes a ime is o( 6 ). We begin analyzing he mechanism by saing some wellknown properies of refleced Brownian moions (see [7]). Proposiion 6. Le [W, 0] be a refleced Brownian moion wih mean zero and variance σ 2 ; he reflecion barrier is 0. Assume he value of W a ime is equal o y: E[y] = θ( σ 2 ) (2) For T > 0, le z = W +T. For he probabiliy densiy funcion of z y we have: r 2 Pr[(z y) dx] πtσ e x 2 2 2T σ 2 (3) r 8Tσ 2 Pr[ z y x] π x e x 2 2T σ 2 (4) r 8Tσ 2 E[ z y I( z y x)] π e x2 2T σ 2 (5) Corollary 7. The expeced value of he maximum of µ it, i n, is θ( T). Noe ha in he corollary above n and σ are consan. Now, similar o Lemma 4, we bound E[ T]. The proof is given in appendix B. Lemma 8. Suppose under M ɛ(b) all agens are ruhful unil ime T, hen, E[ T] = O(T 2 ɛ ). Now we are ready o prove he main heorem of his secion: Theorem 9. M ɛ(b) is ex-pos individually raional. Also, for 0 ɛ, Mɛ(B) is asympoically incenive compaible 3 and ex-ane efficien. Proof. We firs prove ex-pos individual raionaliy. I is sufficien o prove i only for he periods ha agen i has received he iem wihin an exploi phase. For T, such ha y it =, we have: p i = y i min{bγ (T), bµ i()} T y ibµ i() = x ir i. y ir ili, We complee he proof by showing he condiions (C) and (C2) hold. By (2), he expeced uiliy of each agen a ime from random exploraion is θ( σ 2 ɛ ln +ɛ ) = θ( 2 ɛ ln +ɛ ). Therefore, he expeced uiliy up o ime T from exploraion is θ(t 2 3 ɛ ln +ɛ T). By Lemma (8) and Corollary 7: T E[max T {µit } + ] = O(T + 2 ɛ ). For ɛ, 3 ɛ + ɛ his yields Condiion(C)
8 By Corollary 7, he expeced value of max i{µ it } and γ T are θ( T). Therefore, he expeced welfare of an efficien mechanism beween ime and T is θ(t 3 2 ). For any 0 < ɛ <, we have: θ(t 3 2 ) = ω(t 3 2 ɛ ln +ɛ + T + ɛ 2 ) By condiion (C2), M ɛ(b) is asympoically ex-ane efficien. To apply his model o sponsored search we rea each iem as a bundle of search queries. Each ime sep is defined by he arrival of m queries. The mechanism allocaes all m queries o an adveriser and afer ha, he adveriser repors he average uiliy for hese queries. The paymen p i is now he price per iem, i.e. he adveriser pays mp i for he bundle of queries. The value of m is chosen such ha µ i can be esimaed wih high accuracy. 6. DISCUSSION AND OPEN PROBLEMS In his secion we discuss some exensions of he mechanisms. Muliple Slos. To modify M so ha i can accommodae muliple slos we borrow from Gonen and Pavlov [2], who assume here exis a se of condiional disribuions which deermine he condiional probabiliy ha he ad in slo j is clicked condiional on he ad in slo j 2 being clicked. During he exploi phase, M allocaes he slos o he adverisers wih he highes expeced uiliy, and he prices are deermined according o Holmsrom s lemma ([9], see also []) The esimaes of he uiliies are updaed based on he repors, using he condiional disribuion. Delayed Repors. In some applicaions, he value of receiving he iem is realized a some laer dae. For example, a user clicks on an ad and visis he websie of he adveriser. A couple of days laer, she reurns o he websie and complees a ransacion. I is no difficul o adjus he mechanism o accommodae his seing by allowing he adveriser o repor wih a delay or change her repor laer. Creaing Muliple Ideniies. When a new adveriser joins he sysem, in order o learn her uiliy value our mechanism gives i a few iems for free in he explore phase. Therefore our mechanism is vulnerable o adverisers who can creae several ideniies and join he sysem. I is no clear wheher creaing a new ideniy is cheap in our conex because he raffic generaed by adverising should evenually be roued o a legiimae business. Sill, one way o avoid his problem is o charge users wihou a reliable hisory using CPC. Acknowledgmen. We would like o hank Arash Asadpour, Peer Glynn, Ashish Goel, Ramesh Johari, and Thomas Weber for fruiful discussions. The second auhor acknowledges he suppor from NSF and a gif from Google. 7. REFERENCES [] G. Aggarwal, A. Goel, and R. Mowani. Truhful aucions for pricing search keywords. Proceedings of ACM conference on Elecronic Commerce, [2] S. Ahey, and I. Segal. An Efficien Dynamic Mechanism. manuscrip, [3] P. Auer, N. Cesa-Bianchi, and P. Fischer. Finie-ime Analysis of he Muliarmed Bandi Problem. Machine Learning archive, Volume 47, Issue 2-3, , [4] A. Bapna, and T. Weber. Efficien Dynamic Allocaion wih Uncerain Valuaions. Working Paper, [5] M. Balcan, A. Blum, J. Harline, and Y. Mansour. Mechanism Design via Machine Learning. Proceedings of 46h Annual IEEE Symposium on Foundaions of Compuer Science, [6] D. Bergemann, and J. Välimäki. Efficien Dynamic Aucions. Proceedings of Third Workshop on Sponsored Search Aucions, [7] A. Borodin, and P. Salminen. Handbook of Brownian Moion: Facs and Formulae. Springer, [8] A. Blum, V. Kumar, A. Rudra, and F. Wu. Online Learning in Online Aucions. Proceedings of he foureenh annual ACM-SIAM symposium on Discree Algorihms, [9] R. Cavallo, D. Parkes, and S. Singh, Efficien Online Mechanism for Persisen, Periodically Inaccessible Self-Ineresed Agens. Working Paper, [0] K. Crawford. Google CFO: Fraud A Big Threa. CNN/Money, December 2, [] J. Giins. Muli-Armed Bandi Allocaion Indices. Wiley, New York, NY, 989. [2] R. Gonen, and E. Pavlov. An Incenive-Compaible Muli-Armed Bandi Mechanism. Proceedings of he Tweny-Sixh Annual ACM Symposium on Principles of Disribued Compuing, [3] B. Grow, B. Elgin, and M. Herbs. Click Fraud: The dark side of online adverising. BusinessWeek. Cover Sory, Ocober 2, [4] N. Immorlica, K. Jain, M. Mahdian, and K. Talwar. Click Fraud Resisan Mehods for Learning Click-Through Raes. Proceedings of he s Workshop on Inerne and Nework Economics, [5] B. Kis, P. Laxminarayan, B. LeBlanc, and R. Meech. A Formal Analysis of Search Aucions Including Predicions on Click Fraud and Bidding Tacics. Workshop on Sponsored Search Aucions, [6] R. Kleinberg. Online Decision Problems Wih Large Sraegy Ses. Ph.D. Thesis, MIT, [7] S. Lahaie, and D. Parkes. Applying Learning Algorihms o Preference Eliciaion. Proceedings of he 5h ACM conference on Elecronic Commerce, [8] M. Mahdian, and K. Tomak. Pay-per-acion model for online adverising. Proceedings of he 3rd Inernaional Workshop on Inerne and Nework Economics, , [9] P. Milgrom, Puing Aucion Theory o Work. Cambridge Universiy Press, [20] D. Michell. Click Fraud and Halli-bloggers. New York Times, July 6, [2] N. Nisan, T. Roughgarden, E. Tardos, and V. Vazirani, ediors. Algorihmic Game Theory, Cambridge Universiy Press, [22] D. Parkes. Online Mechanisms Algorihmic Game Theory (Nisan e al. eds.),
9 [23] B. Sone. When Mice Aack: Inerne Scammers Seal Money wih Click Fraud. Newsweek, January 24, [24] R. Wilson. Game-Theoreic Approaches o Trading Processes. Economic Theory: Fifh World Congress, ed. by T. Bewley, chap. 2, pp , Cambridge Universiy Press, Cambridge, 987. [25] J. Worman, Y. Vorobeychik, L. Li, and J. Langford. Mainaining Equilibria During Exploraion in Sponsored Search Aucions. Proceedings of he 3rd Inernaional Workshop on Inerne and Nework Economics, APPENDI A. PROOF OF LEMMA 4 Proof. We prove he lemma by showing ha for any agen i, Pr[ µ i bµ i() µi] = o( ), c > 0. ɛ c Firs, we esimae E[n i]. There exiss a consan d such ha: E[n i] η ɛ(k) n = min{ n, k ɛ ln +ɛ k} > d ɛ ln +ɛ By he Chernoff-Hoeffding bound: Pr[n i E[ni] ] e ɛ ln +ɛ 8d. 2 Inequaliy () and he Chernoff-Hoeffding bound imply: Pr[ µ i bµ i() ɛ µi] = = Pr[ µ i bµ i() E[ni] µi ni ] ɛ 2 + Pr[ µ i bµ i() E[ni] µi ni < ] ɛ 2 2e ɛ ɛ ln +ɛ µ i 2d + e ɛ ln +ɛ 8d = o( ), c > 0 c Therefore, wih probabiliy o( ), for all agens,. Since he maximum value of ui is, E[ ] = ɛ O( ɛ ). B. PROOF OF LEMMA 8 Proof. Define i = µ i,t µ i,t. We firs prove Pr[ i > T 2 ɛ ] = o( ), c > 0. There exiss a consan T c T d such ha for any ime T T d, he probabiliy ha i has no been randomly allocaed he iem in he las < T d sep is a mos: Pr[T l i,t > ] < ( T ɛ ln 2+2ɛ T) e ln2+2ɛ T T ɛ. (6) Le = ln +ɛ T T ɛ. By equaion (4) and (6), Pr[ i > T ɛ 2 ] = Pr[ i > T ɛ 2 T l i,t ] + Pr[ i > T 2 ɛ T l i,t > ] = o( ), c > 0. T c Hence, wih high probabiliy, for all he n agens, i T ɛ 2. If for some of he agens i T ɛ 2, hen, by Corollary 7, he expeced value of he maximum of µ i over hese agens is θ( T). Therefore, E[max i{ i}] = O(T ɛ 2 ). The lemma follows because E[ T] E[max i{ i}]. 87
Dynamic Cost-Per-Action Mechanisms and Applications to Online Advertising
Dynamic Cost-Per-Action Mechanisms and Applications to Online Advertising Hamid Nazerzadeh Amin Saberi Rakesh Vohra July 13, 2007 Abstract We examine the problem of allocating a resource repeatedly over
More informationThe Application of Multi Shifts and Break Windows in Employees Scheduling
The Applicaion of Muli Shifs and Brea Windows in Employees Scheduling Evy Herowai Indusrial Engineering Deparmen, Universiy of Surabaya, Indonesia Absrac. One mehod for increasing company s performance
More informationMorningstar 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
More informationMTH6121 Introduction to Mathematical Finance Lesson 5
26 MTH6121 Inroducion o Mahemaical Finance Lesson 5 Conens 2.3 Brownian moion wih drif........................... 27 2.4 Geomeric Brownian moion........................... 28 2.5 Convergence of random
More informationPROFIT 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
More informationChapter 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
More informationOptimal 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
More informationRandom Walk in 1-D. 3 possible paths x vs n. -5 For our random walk, we assume the probabilities p,q do not depend on time (n) - stationary
Random Walk in -D Random walks appear in many cones: diffusion is a random walk process undersanding buffering, waiing imes, queuing more generally he heory of sochasic processes gambling choosing he bes
More informationTerm 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:
More informationOptimal Stock Selling/Buying Strategy with reference to the Ultimate Average
Opimal Sock Selling/Buying Sraegy wih reference o he Ulimae Average Min Dai Dep of Mah, Naional Universiy of Singapore, Singapore Yifei Zhong Dep of Mah, Naional Universiy of Singapore, Singapore July
More informationOn the degrees of irreducible factors of higher order Bernoulli polynomials
ACTA ARITHMETICA LXII.4 (1992 On he degrees of irreducible facors of higher order Bernoulli polynomials by Arnold Adelberg (Grinnell, Ia. 1. Inroducion. In his paper, we generalize he curren resuls on
More informationTo 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
More informationStochastic 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
More informationMarkit Excess Return Credit Indices Guide for price based indices
Marki Excess Reurn Credi Indices Guide for price based indices Sepember 2011 Marki Excess Reurn Credi Indices Guide for price based indices Conens Inroducion...3 Index Calculaion Mehodology...4 Semi-annual
More informationMultiprocessor 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,
More informationOption Put-Call Parity Relations When the Underlying Security Pays Dividends
Inernaional Journal of Business and conomics, 26, Vol. 5, No. 3, 225-23 Opion Pu-all Pariy Relaions When he Underlying Securiy Pays Dividends Weiyu Guo Deparmen of Finance, Universiy of Nebraska Omaha,
More informationThe 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
More informationI. 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
More informationDYNAMIC 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
More informationMACROECONOMIC 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
More informationConstant Data Length Retrieval for Video Servers with Variable Bit Rate Streams
IEEE Inernaional Conference on Mulimedia Compuing & Sysems, June 17-3, 1996, in Hiroshima, Japan, p. 151-155 Consan Lengh Rerieval for Video Servers wih Variable Bi Rae Sreams Erns Biersack, Frédéric Thiesse,
More informationChapter 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
More informationEfficient Risk Sharing with Limited Commitment and Hidden Storage
Efficien Risk Sharing wih Limied Commimen and Hidden Sorage Árpád Ábrahám and Sarola Laczó March 30, 2012 Absrac We exend he model of risk sharing wih limied commimen e.g. Kocherlakoa, 1996) by inroducing
More informationTowards Optimal Capacity Segmentation with Hybrid Cloud Pricing
Towards Opimal Capaciy Segmenaion wih Hybrid Cloud Pricing Wei Wang, Baochun Li, and Ben Liang Deparmen of Elecrical and Compuer Engineering Universiy of Torono Absrac Cloud resources are usually priced
More informationTowards Optimal Capacity Segmentation with Hybrid Cloud Pricing
Towards Opimal Capaciy Segmenaion wih Hybrid Cloud Pricing Wei Wang, Baochun Li, and Ben Liang Deparmen of Elecrical and Compuer Engineering Universiy of Torono Torono, ON M5S 3G4, Canada weiwang@eecg.orono.edu,
More informationChapter 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
More informationDETERMINISTIC INVENTORY MODEL FOR ITEMS WITH TIME VARYING DEMAND, WEIBULL DISTRIBUTION DETERIORATION AND SHORTAGES KUN-SHAN WU
Yugoslav Journal of Operaions Research 2 (22), Number, 6-7 DEERMINISIC INVENORY MODEL FOR IEMS WIH IME VARYING DEMAND, WEIBULL DISRIBUION DEERIORAION AND SHORAGES KUN-SHAN WU Deparmen of Bussines Adminisraion
More informationCredit Index Options: the no-armageddon pricing measure and the role of correlation after the subprime crisis
Second Conference on The Mahemaics of Credi Risk, Princeon May 23-24, 2008 Credi Index Opions: he no-armageddon pricing measure and he role of correlaion afer he subprime crisis Damiano Brigo - Join work
More informationTowards Incentive-Compatible Reputation Management
Towards Incenive-Compaible Repuaion Managemen Radu Jurca, Boi Falings Arificial Inelligence Laboraory Swiss Federal Insiue of Technology (EPFL) IN-Ecublens, 115 Lausanne, Swizerland radu.jurca@epfl.ch,
More informationNiche 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.
More informationANALYSIS 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,
More informationTowards Optimal Capacity Segmentation with Hybrid Cloud Pricing
Towards Opimal Capaciy Segmenaion wih Hybrid Cloud Pricing Wei Wang, Baochun Li, and Ben Liang Deparmen of Elecrical and Compuer Engineering Universiy of Torono Torono, ON M5S 3G4, Canada weiwang@eecg.orono.edu,
More informationSPEC 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,
More informationSTABILITY OF LOAD BALANCING ALGORITHMS IN DYNAMIC ADVERSARIAL SYSTEMS
STABILITY OF LOAD BALANCING ALGORITHMS IN DYNAMIC ADVERSARIAL SYSTEMS ELLIOT ANSHELEVICH, DAVID KEMPE, AND JON KLEINBERG Absrac. In he dynamic load balancing problem, we seek o keep he job load roughly
More informationSingle-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
More informationThe 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
More informationNikkei Stock Average Volatility Index Real-time Version Index Guidebook
Nikkei Sock Average Volailiy Index Real-ime Version Index Guidebook Nikkei Inc. Wih he modificaion of he mehodology of he Nikkei Sock Average Volailiy Index as Nikkei Inc. (Nikkei) sars calculaing and
More informationCooperation with Network Monitoring
Cooperaion wih Nework Monioring Alexander Wolizky Microsof Research and Sanford Universiy November 2011 Absrac This paper sudies he maximum level of cooperaion ha can be susained in perfec Bayesian equilibrium
More information4. 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
More informationThe 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
More informationTEMPORAL 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.
More informationAnalysis 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
More informationA 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
More informationWhy 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
More informationDuration 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
More informationSTABILITY OF LOAD BALANCING ALGORITHMS IN DYNAMIC ADVERSARIAL SYSTEMS
SIAM J. COMPUT. Vol. 37, No. 5, pp. 1656 1673 c 2008 Sociey for Indusrial and Applied Mahemaics STABILITY OF LOAD BALANCING ALGORITHMS IN DYNAMIC ADVERSARIAL SYSTEMS ELLIOT ANSHELEVICH, DAVID KEMPE, AND
More informationChapter 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
More informationMathematics 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
More informationThe 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.
More informationReturn Calculation of U.S. Treasury Constant Maturity Indices
Reurn Calculaion of US Treasur Consan Mauri Indices Morningsar Mehodolog Paper Sepeber 30 008 008 Morningsar Inc All righs reserved The inforaion in his docuen is he proper of Morningsar Inc Reproducion
More informationINTEREST 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
More informationInventory Planning with Forecast Updates: Approximate Solutions and Cost Error Bounds
OPERATIONS RESEARCH Vol. 54, No. 6, November December 2006, pp. 1079 1097 issn 0030-364X eissn 1526-5463 06 5406 1079 informs doi 10.1287/opre.1060.0338 2006 INFORMS Invenory Planning wih Forecas Updaes:
More informationOnline Convex Programming and Generalized Infinitesimal Gradient Ascent
Online Convex Programming and Generalized Infiniesimal Gradien Ascen Marin Zinkevich Carnegie Mellon Universiy, 5000 Forbes Avenue, Pisburgh, PA 1513 USA maz@cs.cmu.edu Absrac Convex programming involves
More informationMeasuring 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
More informationCointegration: 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
More informationThe naive method discussed in Lecture 1 uses the most recent observations to forecast future values. That is, Y ˆ t + 1
Business Condiions & Forecasing Exponenial Smoohing LECTURE 2 MOVING AVERAGES AND EXPONENTIAL SMOOTHING OVERVIEW This lecure inroduces ime-series smoohing forecasing mehods. Various models are discussed,
More informationTrading on Short-Term Information
428 Trading on Shor-Term Informaion by ALEXANDER GÜMBEL This paper shows ha invesors may wan fund managers o acquire and rade on shor-erm insead of more profiable long-erm informaion. This improves learning
More informationUSE 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
More information11/6/2013. 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
More informationWorking Paper No. 482. Net Intergenerational Transfers from an Increase in Social Security Benefits
Working Paper No. 482 Ne Inergeneraional Transfers from an Increase in Social Securiy Benefis By Li Gan Texas A&M and NBER Guan Gong Shanghai Universiy of Finance and Economics Michael Hurd RAND Corporaion
More informationIndividual 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
More informationHow To Maximize Cooperaion In A Game Of Nework Monioring
Cooperaion wih Nework Monioring Alexander Wolizky Microsof Research and Sanford Universiy February 2012 Absrac This paper sudies he maximum level of cooperaion ha can be susained in perfec Bayesian equilibrium
More informationAnalogue and Digital Signal Processing. First Term Third Year CS Engineering By Dr Mukhtiar Ali Unar
Analogue and Digial Signal Processing Firs Term Third Year CS Engineering By Dr Mukhiar Ali Unar Recommended Books Haykin S. and Van Veen B.; Signals and Sysems, John Wiley& Sons Inc. ISBN: 0-7-380-7 Ifeachor
More informationTask is a schedulable entity, i.e., a thread
Real-Time Scheduling Sysem Model Task is a schedulable eniy, i.e., a hread Time consrains of periodic ask T: - s: saring poin - e: processing ime of T - d: deadline of T - p: period of T Periodic ask T
More informationJump-Diffusion Option Valuation Without a Representative Investor: a Stochastic Dominance Approach
ump-diffusion Opion Valuaion Wihou a Represenaive Invesor: a Sochasic Doance Approach By Ioan Mihai Oancea and Sylianos Perrakis This version February 00 Naional Bank of Canada, 30 King Sree Wes, Torono,
More informationReal Time Bid Optimization with Smooth Budget Delivery in Online Advertising
Real Time Bid Opimizaion wih Smooh Budge Delivery in Online Adverising Kuang-Chih Lee Ali Jalali Ali Dasdan Turn Inc. 835 Main Sree, Redwood Ciy, CA 94063 {klee,ajalali,adasdan}@urn.com ABSTRACT Today,
More informationA general decomposition formula for derivative prices in stochastic volatility models
A general decomposiion formula for derivaive prices in sochasic volailiy models Elisa Alòs Universia Pompeu Fabra C/ Ramón rias Fargas, 5-7 85 Barcelona Absrac We see ha he price of an european call opion
More informationNetwork 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:
More informationDependent 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
More informationOption Pricing Under Stochastic Interest Rates
I.J. Engineering and Manufacuring, 0,3, 8-89 ublished Online June 0 in MECS (hp://www.mecs-press.ne) DOI: 0.585/ijem.0.03. Available online a hp://www.mecs-press.ne/ijem Opion ricing Under Sochasic Ineres
More informationDynamic programming models and algorithms for the mutual fund cash balance problem
Submied o Managemen Science manuscrip Dynamic programming models and algorihms for he muual fund cash balance problem Juliana Nascimeno Deparmen of Operaions Research and Financial Engineering, Princeon
More informationWorking Paper On the timing option in a futures contract. SSE/EFI Working Paper Series in Economics and Finance, No. 619
econsor www.econsor.eu Der Open-Access-Publikaionsserver der ZBW Leibniz-Informaionszenrum Wirschaf The Open Access Publicaion Server of he ZBW Leibniz Informaion Cenre for Economics Biagini, Francesca;
More informationAs widely accepted performance measures in supply chain management practice, frequency-based service
MANUFACTURING & SERVICE OPERATIONS MANAGEMENT Vol. 6, No., Winer 2004, pp. 53 72 issn 523-464 eissn 526-5498 04 060 0053 informs doi 0.287/msom.030.0029 2004 INFORMS On Measuring Supplier Performance Under
More informationEconomics 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
More informationTable of contents Chapter 1 Interest rates and factors Chapter 2 Level annuities Chapter 3 Varying annuities
Table of conens Chaper 1 Ineres raes and facors 1 1.1 Ineres 2 1.2 Simple ineres 4 1.3 Compound ineres 6 1.4 Accumulaed value 10 1.5 Presen value 11 1.6 Rae of discoun 13 1.7 Consan force of ineres 17
More informationRelationships 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
More informationTrading on Short-Term Information
Trading on Shor-Term Informaion Preliminary version. Commens welcome Alexander Gümbel * Deparmen of Economics European Universiy Insiue Badia Fiesolana 5006 San Domenico di Fiesole (FI) Ialy e-mail: guembel@daacomm.iue.i
More informationReal-time Particle Filters
Real-ime Paricle Filers Cody Kwok Dieer Fox Marina Meilă Dep. of Compuer Science & Engineering, Dep. of Saisics Universiy of Washingon Seale, WA 9895 ckwok,fox @cs.washingon.edu, mmp@sa.washingon.edu Absrac
More informationMarket 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,
More informationA Distributed Multiple-Target Identity Management Algorithm in Sensor Networks
A Disribued Muliple-Targe Ideniy Managemen Algorihm in Sensor Neworks Inseok Hwang, Kaushik Roy, Hamsa Balakrishnan, and Claire Tomlin Dep. of Aeronauics and Asronauics, Sanford Universiy, CA 94305 Elecrical
More informationMaking 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 philippe.oechslin@epfl.ch
More informationadaptive control; stochastic systems; certainty equivalence principle; long-term
COMMUICATIOS I IFORMATIO AD SYSTEMS c 2006 Inernaional Press Vol. 6, o. 4, pp. 299-320, 2006 003 ADAPTIVE COTROL OF LIEAR TIME IVARIAT SYSTEMS: THE BET O THE BEST PRICIPLE S. BITTATI AD M. C. CAMPI Absrac.
More informationTSG-RAN Working Group 1 (Radio Layer 1) meeting #3 Nynashamn, Sweden 22 nd 26 th March 1999
TSG-RAN Working Group 1 (Radio Layer 1) meeing #3 Nynashamn, Sweden 22 nd 26 h March 1999 RAN TSGW1#3(99)196 Agenda Iem: 9.1 Source: Tile: Documen for: Moorola Macro-diversiy for he PRACH Discussion/Decision
More informationHedging 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
More informationPresent Value Methodology
Presen Value Mehodology Econ 422 Invesmen, Capial & Finance Universiy of Washingon Eric Zivo Las updaed: April 11, 2010 Presen Value Concep Wealh in Fisher Model: W = Y 0 + Y 1 /(1+r) The consumer/producer
More informationAppendix 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
More informationHow To Find Opimal Conracs In A Continuous Time Model
Appl Mah Opim (9) 59: 99 46 DOI.7/s45-8-95- OpimalCompensaionwihHiddenAcion and Lump-Sum Paymen in a Coninuous-Time Model Jakša Cvianić Xuhu Wan Jianfeng Zhang Published online: 6 June 8 Springer Science+Business
More informationAnalysis of Tailored Base-Surge Policies in Dual Sourcing Inventory Systems
Analysis of Tailored Base-Surge Policies in Dual Sourcing Invenory Sysems Ganesh Janakiraman, 1 Sridhar Seshadri, 2, Anshul Sheopuri. 3 Absrac We sudy a model of a firm managing is invenory of a single
More informationOptimal Time to Sell in Real Estate Portfolio Management
Opimal ime o Sell in Real Esae Porfolio Managemen Fabrice Barhélémy and Jean-Luc Prigen hema, Universiy of Cergy-Ponoise, Cergy-Ponoise, France E-mails: fabricebarhelemy@u-cergyfr; jean-lucprigen@u-cergyfr
More informationStatistical Analysis with Little s Law. Supplementary Material: More on the Call Center Data. by Song-Hee Kim and Ward Whitt
Saisical Analysis wih Lile s Law Supplemenary Maerial: More on he Call Cener Daa by Song-Hee Kim and Ward Whi Deparmen of Indusrial Engineering and Operaions Research Columbia Universiy, New York, NY 17-99
More informationTHE FIRM'S INVESTMENT DECISION UNDER CERTAINTY: CAPITAL BUDGETING AND RANKING OF NEW INVESTMENT PROJECTS
VII. THE FIRM'S INVESTMENT DECISION UNDER CERTAINTY: CAPITAL BUDGETING AND RANKING OF NEW INVESTMENT PROJECTS The mos imporan decisions for a firm's managemen are is invesmen decisions. While i is surely
More informationOptimal Life Insurance Purchase, Consumption and Investment
Opimal Life Insurance Purchase, Consumpion and Invesmen Jinchun Ye a, Sanley R. Pliska b, a Dep. of Mahemaics, Saisics and Compuer Science, Universiy of Illinois a Chicago, Chicago, IL 667, USA b Dep.
More informationSampling Time-Based Sliding Windows in Bounded Space
Sampling Time-Based Sliding Windows in Bounded Space Rainer Gemulla Technische Universiä Dresden 01062 Dresden, Germany gemulla@inf.u-dresden.de Wolfgang Lehner Technische Universiä Dresden 01062 Dresden,
More informationCALCULATION OF OMX TALLINN
CALCULATION OF OMX TALLINN CALCULATION OF OMX TALLINN 1. OMX Tallinn index...3 2. Terms in use...3 3. Comuaion rules of OMX Tallinn...3 3.1. Oening, real-ime and closing value of he Index...3 3.2. Index
More informationDynamic Contracting: An Irrelevance Result
Dynamic Conracing: An Irrelevance Resul Péer Eső and Balázs Szenes Sepember 5, 2013 Absrac his paper considers a general, dynamic conracing problem wih adverse selecion and moral hazard, in which he agen
More informationDOES 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
More informationBALANCE 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, lena.finn@scb.se Camilla Bergeling +46 8 506 942 06, camilla.bergeling@scb.se
More information17 Laplace transform. Solving linear ODE with piecewise continuous right hand sides
7 Laplace ransform. Solving linear ODE wih piecewise coninuous righ hand sides In his lecure I will show how o apply he Laplace ransform o he ODE Ly = f wih piecewise coninuous f. Definiion. A funcion
More informationJournal Of Business & Economics Research September 2005 Volume 3, Number 9
Opion Pricing And Mone Carlo Simulaions George M. Jabbour, (Email: jabbour@gwu.edu), George Washingon Universiy Yi-Kang Liu, (yikang@gwu.edu), George Washingon Universiy ABSTRACT The advanage of Mone Carlo
More informationHow 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
More information