Computing Equilibrium Prices in Exchange Economies with Tax Distortions
|
|
- Philip Simon
- 5 years ago
- Views:
Transcription
1 Computng Equlbrum Prces n Exchange Economes wth Tax Dstortons Bruno Codenott 1 Lus Rademacher 2 Kastur Varadaraan 3 Abstract We consder the computaton of equlbrum prces n market settngs where purchases of goods are subect to taxaton. Whle ths scenaro s the standard one n appled computatonal work, so far t has not been an obect of study n theoretcal computer scence. Taxes ntroduce sgnfcant dstortons: equlbra are no longer Pareto optmal, suffcent condtons for unqueness do not contnue to guarantee t, exstence tself must be revsted. We analyze the effects of these dstortons on scenaros whch, n the absence of taxes, admt polynomal tme algorthms, and prove a number of results: For Fsher s model wth homothetc preferences: f consumers are subect to a unform tax regme, the model loses the representatve consumer and becomes prone to multple dsconnected equlbra; however we show that the dstorton has a structure whch leads to two (as opposed to one) representatve consumers. We take advantage of ths property to develop polynomal tme algorthms, n spte of the presence of multple dsconnected equlbra. To obtan the result above, we develop a technque to estmate the senstvty of Fsher s equlbrum prces wth respect to the consumers ncome, whch mght fnd wder applcatons n the feld. For the exchange model wth Cobb-Douglas utlty functons: f consumers are subect to a dfferentated tax regme, we show that the model s equvalent to a Cobb-Douglas exchange economy wth an extra good. Ths mples that the polynomal tme computablty s preserved. For the exchange model wth utlty functons satsfyng weak gross substtutablty: f consumers are subect to a dfferentated tax regme, the technques based on convex programmng seem to break down. However we are able to modfy certan combnatoral algorthms and prce adustment methods to obtan polynomal tme approxmaton schemes n two specal cases: an economy wth lnear utltes and an economy where the demands do not change too rapdly as a functon of the prces. 1 IIT-CNR, Psa, Italy. E-mal: bruno.codenott@t.cnr.t. Part of ths work has been done whle vstng the Toyota Technologcal Insttute at Chcago. 2 Department of Mathematcs, MIT. Work done whle vstng Toyota Technologcal Insttute at Chcago, Chcago IL E-mal: lrademac@math.mt.edu. 3 Department of Computer Scence, The Unversty of Iowa, Iowa Cty IA E-mal: kvaradar@cs.uowa.edu. 1
2 1 Introducton The equlbrum problem for a pure exchange economy amounts to fndng a set of prces and allocatons of goods to economc agents such that each agent maxmzes her utlty, subect to her budget constrants, and the market clears. The equlbrum depends only on the agents utlty functons and ntal endowments of goods. If one ams at analyzng equlbrum problems arsng from real world applcatons, the scenaro outlned above has often to be extended. Indeed one needs to take nto account the presence of sutable dstortons, whch mght be, dependng on the specfc applcaton, transacton costs, transportaton costs, tarffs, and/or taxes. In these frameworks, whch are the standard ones for appled computatonal work, one has to deal wth equlbrum condtons nfluenced by addtonal parameters whch often change the mathematcal propertes of the problem. For nstance, n models wth taxes, () the equlbrum allocatons mght lose ther Pareto optmalty; () restrctons whch mply, n the absence of taxes, the unqueness of equlbrum prces mght become compatble wth multple dsconnected equlbra. In ths paper, we consder exchange economes wth ether unform or dfferentated ad valorem taxes (see Secton 2 for the approprate defntons). We explore the effects of such tax dstortons on models whch admt - n the absence of taxes - polynomal tme algorthms. In spte of the loss of certan structural propertes (ncludng unqueness), we are able to obtan polynomal tme algorthms or approxmaton schemes n several nstances where the model wthout taxes admtted them. Background. We now descrbe the model of an exchange economy, and provde some basc defntons. Let us consder m economc agents whch represent traders of n goods. Let R n + denote the subset of R n wth all nonnegatve coordnates. The -th coordnate n R n wll stand for good. Each trader has a concave utlty functon u : R n + R +, whch represents her preferences for the dfferent bundles of goods, and an ntal endowment of goods w = (w 1,...,w n ) R n +. At gven prces π R n +, trader wll demand a bundle of goods x = (x 1,...,x n ) R n + whch maxmzes u (x) subect to the budget constrant π x π w. Let W = w denote the total amount of good n the market. An equlbrum s a vector of prces π = (π 1,...,π n ) R n + at whch, for each trader, there s a bundle x = ( x 1,..., x n ) R n + of goods such that the followng two condtons hold: () for each trader, the vector x maxmzes u (x) subect to the constrants π x π w and x R n +; () for each good, x W. The celebrated result of Arrow and Debreu [2] states that, under qute mld assumptons, such an equlbrum exsts. We now gve the defnton of approxmate equlbrum. We assume that all the utlty functons u() dscussed n ths paper satsfy u(0) = 0. A bundle x R n + s an ε-approxmate demand, for 0 < ε < 1, of trader at prces π f u (x ) (1 ε)u and π x (1 + ε)π w, where u = max{u (x) x R n +,π x π w }. A prce vector π R n + s an ε-approxmate equlbrum f there s a bundle x for each such that (1) for each trader, x s an ε-approxmate demand of trader at prces π, and (2) x (1 + ε) w for each good. An mportant specal case of an exchange economy s the dstrbutonal economy, where the ntal endowments are all collnear,.e., w = δ w, δ > 0, so that the relatve ncomes of the traders are ndependent of the prces. Ths specal case s equvalent to Fsher model, whch s 1
3 a market of n goods desred by m utlty maxmzng buyers wth fxed ncomes. For any prce vector π, the vector x (π) that maxmzes u (x) subect to the constrants π x π w and x R n + s called the demand of the -th trader. By addng up the traders demands, one gets the market demand. The utlty functon of an ndvdual trader s sad to satsfy weak gross substtutablty f ncreasng the prces of some of the goods whle keepng some others and her ncome fxed cannot cause a decrease n demand for the goods whose prce s fxed. It s well known and easy to see that the market demand satsfes weak gross substtutablty f the utlty functon of each ndvdual trader does. A utlty functon u( ) s homogeneous of degree one f t satsfes u(αx) = αu(x), for all α > 0, whle t s log-homogeneous f t satsfes u(αx) = u(x) + log α, for all α > 0. Both homogeneous and log-homogeneous utlty functons represent consumers wth homothetc preferences,.e., a bundle x s preferred to a bundle y f and only f the bundle αx s preferred to the bundle αy, for all α > 0. A lnear utlty functon has the form u (x) = a x. A CES (constant elastcty of substtuton) utlty functon has the form u(x ) = ( (a x ) ρ ) 1/ρ, where < ρ < 1, ρ 0. The Cobb-Douglas utlty functon has the form u (x) = (x ) a, where a 0 and a = 1. Related Work. Substantal work has been done on extendng equlbrum models to handle scenaros where good purchases are subect to taxaton [19, 20, 25, 26, 27]. Such efforts have provded exstental results [25, 26], evdence of tax-nduced multplcty [27] and of the loss of Pareto-optmalty of equlbra [19, 20]. Buldng upon ths body of results, appled models have been desgned to explctly take nto account tax dstortons (see for nstance the popular GAMS-MPSGE programmng envronment). Prevous work wthn theoretcal computer scence, whch was ntated by [9], has been focussed on restrctons under whch the market equlbrum problem, n ts verson wthout taxes, can be solved n polynomal tme. These restrctons nclude () dstrbutonal economes (the Fsher settng) where the traders have homogeneous utlty functons [10, 18, 8, 13], () exchange economes whch satsfy weak gross substtutablty [16, 15, 7, 4], and () exchange economes wth some famles of CES and nested CES utlty functons [3, 17]. For a more comprehensve lst of results, see [6]. Our Results. In ths paper we analyze the effects of tax dstortons on scenaros whch, n the absence of taxes, admt polynomal tme algorthms. Frst of all, an exchange economy wth unform ad valorem taxes can be effcently transformed nto an equvalent exchange economy wthout taxes, wth the same utlty functons and dfferent ntal endowments, obtaned by redstrbutng the orgnal ones (Secton 2.1). Therefore algorthms for the exchange model that do not depend on the dstrbuton of endowments carry through the model wth unform taxes. We prove that a dstrbutonal economy (whch s equvalent to Fsher s model) wth unform ad valorem taxes and homothetc consumers can be effcently transformed nto an equvalent two-trader exchange economy wthout taxes (Secton 3.1). We then develop a polynomal-tme algorthm for approxmatng the equlbrum (Secton 3.2). To analyze some parameters related to the accuracy of the algorthm, we use the tool of mplct dfferentaton, whch we beleve wll fnd more general applcatons (Secton 3.4). Note that a two-trader exchange economy wth homothetc consumers admts multple dsconnected equlbra [14]. The example n [14] can be modfed to model an exchange economy that s equvalent to a dstrbutonal economy wth unform ad valorem taxes. Therefore our algorthm provdes the frst sgnfcant example 2
4 of polynomal tme computaton of equlbra n a settng wth multple dsconnected equlbra. We then show that an n-good exchange economy wth dfferentated ad valorem taxes and m Cobb-Douglas consumers can be effcently transformed nto an equvalent (n+1)-good exchange economy wthout taxes, wth m Cobb-Douglas consumers (Secton 2.2). Snce the equlbrum for a Cobb-Douglas exchange economy can be computed n polynomal tme [11], our reducton shows that the same s possble for the model wth non-unform taxaton. We beleve that ths reducton s not possble for more general utlty functons, ncludng some of the commonly studed ones. Consequently, we do not at present have polynomal tme algorthms or approxmaton schemes for the model wth non-unform taxes when the traders have CES functons wth 1 ρ < 0, even though there are explct convex programs and polynomal tme algorthms for the correspondng pure exchange model [3]. The same happens for certan nested CES functons [17]. The stuaton s notably better when the utlty functons satsfy weak gross substtutablty. Here too, methods based on explct or mplct convex programs seem to be unavalable, n contrast wth the pure exchange model. Ths s true even for the case of lnear utlty functons, where explct convex programs are known for the pure exchange model [23, 15]. Interestngly, n spte of ths, we are able to show that the combnatoral algorthm of Garg and Kapoor [13] for lnear utltes can be adapted to handle the dfferentated tax scenaro (Secton 4). We also consder the case of economes wth weak gross substtutablty and demand functons that do not change too rapdly wth prces. In ths case, t s possble to obtan a poly-tme approxmaton scheme va a smple prce adustment (tâtonnement) scheme. The man tool s Lemma 7 from [1], whch guarantees convergence of a normalzed verson of tâtonnement, where the prce of one good (called the numérare) s kept fxed. By expressng the role of taxaton n terms of a fcttous good (the tax good of Secton 2.2) whch acts as the numérare, we can then analyze the convergence of the normalzed process. The lemma above becomes thus relevant because t does not requre gross substtutablty wth respect to the good whose prce s not part of the dynamcs. For lack of space, ths last result s omtted from the extended abstract. 2 Exchange Economes wth Taxes on Consumpton We descrbe the model of an exchange economy wth ad valorem taxes as presented by Kehoe ([19], pp ), and dstngush between the unform case, where the tax rate s unform across consumers, and the non-unform case, where the tax rate s dfferentated among consumers. 2.1 Unform Ad Valorem Taxes Consder a trader wth utlty functon u (x ) and ntal endowment w. Let τ 0 be the unform ad valorem tax assocated wth the consumpton of good. Ths means that f consumer purchases x unts of good at prce π, she wll spend on ths good the amount π x + τ π x. We postulate the presence of a specal actor, the government, whch wll rebate the tax revenues to consumers. Let θ 0, wth θ = 1, be the share of total tax revenues rebated to consumer as a lump sum. Then the classcal consumer s maxmzaton problem gets modfed as follows: 3
5 max u (x ) (1) s.t. π (1 + τ )x π w + θ R, (2) where x R n +, and R s the total amount of revenues dstrbuted by the government. In ths context, the market equlbrum problem conssts of fndng ( π, x, R) such that at prces π, x solves (1) (2), (optmalty and budget constrant are satsfed for all consumers); x = w, (the market clears all the goods); R = π τ x (the amount of taxes dstrbuted s equal to the amount of taxes collected). We now show that the equlbra for such an economy are n a one-to-one correspondence wth the equlbra of an exchange economy wthout taxes, where the traders have a dfferent set of ntal endowments, obtaned by a sutable redstrbuton of the orgnal ones. Ths correspondence has been establshed n [5], where models of taxaton were one of the targets of some expermental work. Whenever the equlbra of two economes are n a one-to-one correspondence, and can be mmedately computed one from the other, we say that the two economes are equvalent. Proposton 1 [5] Let E = E(u ( ),w,τ,θ) be an exchange economy wth unform ad valorem taxes τ = (τ 1,...,τ n ), and tax shares θ = (θ 1,...,θ m ). E s equvalent to an exchange economy wthout taxes E = E (u ( ),w ) where w = w τ 1+τ + θ 1+τ w. Proof : The proof s n Appendx A. 2.2 Dfferentated Taxes We now consder a more general model n whch the taxes on purchases are dfferent for each trader. We call ths scheme specfc taxaton or dfferentated ad valorem taxaton. In an exchange economy wth specfc taxes, τ 0 s the tax rate mputed to trader on purchase of good. The settng for an exchange economy wth specfc taxes dffers from that wth unform taxes n the budget constrant, whch s now gven by π (1 + τ )x π w + θ R. (3) At equlbrum, we must have R = π τ x. The lack of unformty of ths model, whch dfferentates between consumers, prevents the possblty of a drect reducton to a pure exchange economy, obtaned by redstrbutng the ndvdual endowments, as n Proposton 1. Nevertheless, ths model can be made smlar to a pure exchange economy wth an extra good. Indeed, note that the budget constrant can be rewrtten as π x + Rx,n+1 π w + Rθ (4) 4
6 P where x,n+1 = π τ x R. If we nterpret R as the prce of an addtonal (fcttous) good, that we call the tax good, and θ and x,n+1 as the -th trader s ntal endowment and demand of such good, then nequalty (4) corresponds to the budget constrant of a consumer n a pure exchange economy wth an extra good. To get a reducton to an exchange economy wthout taxes, one would now need to exhbt a utlty functon, defned on n + 1 goods, whch, combned wth the budget constrant (4), gves P π τ x R a demand of x (π,r) for the frst n goods, and x,n+1 (π,r) = for the tax good. We do not know f such a reducton s possble n general. However we show below (Proposton 2) that t can be done n the case of exchange economes where the traders have Cobb- Douglas utlty functons. Proposton 2 Let n and m be the number of goods and the number of traders, respectvely. Let u ( ), = 1,...,m, be Cobb-Douglas utlty functons. We denote by E n,m = E n,m (u ( ),w,τ,θ) a Cobb-Douglas exchange economy wth dfferentated ad valorem taxes τ = (τ 1,...,τ n ), and tax shares θ = (θ 1,...,θ m ). E n,m s equvalent to an exchange economy wthout taxes E n+1,m = E n+1,m (v ( ),w ) where the v ( ) s are Cobb-Douglas utlty functons, and w = (w 1,...,w n,θ ). Proof : The proof s n Appendx B. 3 Collnear endowments dstorted by unform taxaton The general reducton of Proposton 1 shows that unform taxaton does not affect algorthms whch compute equlbrum prces for exchange economes wthout explotng any partcular property of the ntal endowments. Therefore, several results for pure exchange economes extend to the model wth unform taxaton. One nterestng case where the redstrbuton of ntal endowments potentally carres negatve computatonal consequences s that of exchange economes wth collnear endowments and homothetc preferences. In ths settng an equlbrum (wthout taxes) can be computed n polynomal tme by convex programmng [8], based on certan aggregaton propertes of the economy whch mply the exstence of a representatve consumer [12]. The redstrbuton of endowments assocated wth taxaton clearly destroys the collnearty of endowments, and thus the collapse to a sngle consumer s problem. We show that n an exchange economy wth collnear endowments and homothetc consumers, the model wth unform taxaton s equvalent to an exchange economy wth two representatve consumers. Buldng upon ths property, we then show how to compute an approxmate equlbrum n polynomal tme for a wde famly of problems. 3.1 Reducton to two representatve consumers Recall that a dstrbutonal economy s an exchange economy where the ntal endowments of the traders are collnear. In other words, the k-th trader has endowments of the form w k = γ k w, for k = 1,...,m, where w = (w 1,...,w n ) descrbes the overall amount of each good n the market, and γ k s a postve constant less then one. We have k γ k = 1. In ths scenaro, the relatve ncomes of the traders are constants, so that the model s equvalent to Fsher s model. 5
7 If we specalze the model of Secton 2.1 to an economy wth collnear ntal endowments, then the redstrbuton descrbed by Proposton 1, gves w = γ w τ 1+τ + θ 1+τ w. Notce that the matrx whose columns represent the new ntal endowments of the traders has rank at most two. Indeed all the columns are lnear combnatons of the vectors z and s, whose -th entres are z = w 1+τ, and s = τ 1+τ w, respectvely. Note that w = γ z + θ s. (Note also that w = z + s and verfy that w = w.) Thus the effect of unform taxaton on economes wth proportonal endowments amounts to an ncrease of the rank of the endowment matrx from one to two. Whenever the consumers are homothetc, exchange economes wth a rank two endowment matrx can be reduced to a two-trader economy, accordng to the followng scheme: 1. Let z be the n-vector whose -th component s s wτ 1+τ. w 1+τ, and s be the n-vector whose -th component 2. For all k, splt the k-th trader nto two traders, whch have the same utlty functon of the orgnal trader and ntal endowments γ k z, and θ k s, respectvely. Ths procedure produces two groups of m traders each, where the traders n each group have proportonal endowments. 3. Based on the propertes n [12], aggregate all the consumers from each group nto one representatve consumer, wth endowment gven by the sum of ther endowments and utlty functon obtaned by aggregatng the utlty functons as n [12]. Ths gves a two-trader economy. These arguments lead to the followng result. Proposton 3 Let u ( ), = 1,...,m, be log-homogeneous utlty functons. Let E m = E(u ( ),w,τ,θ,γ) be an m-trader dstrbutonal economy wth unform ad valorem taxes τ = (τ 1,...,τ n ), tax shares θ = (θ 1,...,θ m ), and ncome shares γ = (γ 1,...,γ m ). E m s equvalent to a twotraders exchange economy wthout taxes E 2 = E 2 (v 1( ),v 2 ( ),z,s), where v 1 and v 2 are the log-homogeneous utlty functons of the two consumers, and z and s are ther ntal endowment vectors. Here, v 1 (x) (resp. v 2 (x)) s defned to be the maxmum of γ u (x ) (resp. θ u (x )) over all x 1,...,x m R n + such that x = x. 3.2 The algorthm The reducton summarzed n Proposton 3, combned wth some results of Mantel [21] on two-trader economes, suggest the followng algorthm, whch we call RSR (Reduce-Solve- Reconstruct), for the computaton of an approxmate equlbrum. For the analyss, we assume that τ > 0 for each. Let τ mn = mn τ and τ max = max τ. Algorthm RSR 1. The nput s gven n terms of m log-homogeneous utlty functons u, = 1,...,m, and vectors w, γ, θ, τ. 2. Apply the transformaton of Proposton 3, whch returns an economy wth two homothetc consumers (wth utlty functons v 1 and v 2, and ntal endowments z and s) and n goods. 3. Consder the followng constraned maxmzaton problem: max αv 1 (x 1 ) + (1 α)v 2 (x 2 ) s.t. x 1 + x 2 = w x 1, x 2 0 6
8 For a gven 0 α 1, let x 1 (α) and x 2 (α) be maxmzng allocatons, and let π(α) be the vector of shadow prces (Lagrange multplers). It can be shown that π(α) x 1 (α) = α, π(α) x 2 (α) = 1 α, and thus π(α) w = 1. Moreover, x 1 (α) and x 2 (α) have the rght shape they are proportonal to the optmal bundles demanded by the two traders at the prce π(α). Let B 1 (α) = π(α) (z x 1 (α)) and B 2 (α) = π(α) (s x 2 (α)) be the functons expressng the (postve or negatve) savngs of consumer 1 and 2, respectvely. Note that B 1 (α)+b 2 (α) = 0, snce z+s = w. Thus, f B 1 (α) = 0, then we have π(α) x 1 (α) = π(α) z, and π(α) x 2 (α) = π(α) s. Thus, x 1 (α) and x 2 (α) are not merely proportonal to the optmal bundles, but they are the optmal bundles of the two traders at prce π(α). Thus π(α) s an equlbrum for the two trader economy [22]. We therefore fnd an approxmate equlbrum for the two-trader economy by fndng a value of α such that B 1 (α) and B 2 (α) are suffcently close to zero. In such a case, π(α), x 1 (α) and x 2 (α) form an approxmate equlbrum, provded that the functons B (α) are smooth enough (see the extensve dscusson below). The search for an approprate value of α can be done by the bsecton method,.e., bnary search, guded by the value of B (α), computed from the values x (α) and π(α) returned by the soluton of the maxmzaton problem above. The applcablty of bsecton method bulds upon some results by Mantel [21] on the global convergence of the welfare adustment process when appled to two-trader economes. 4. From the soluton to the two-trader problem, reconstruct the soluton to the 2m-trader problem,.e., the correspondng allocatons, and then to the m-trader problem wthout taxes. 5. Compute approxmate equlbrum prces π, = 1,..., for the orgnal economy wth taxes, by scalng prces π (α),.e., π = π(α) 1+τ. 3.3 Analyss of algorthm RSR For any prce vector π, t s easy to see that π z π w les n the nterval [α 1 mn = 1+τ max,α max = 1 1+τ mn ]. Thus f B 1 (ᾱ) = 0, then ᾱ = π(ᾱ) x 1(ᾱ) π(ᾱ) w = π(ᾱ) z π(ᾱ) w les n the range [α mn,α max ]. It s also easy to verfy that B 1 (α mn ) 0 and B 1 (α max ) 0. So we perform our bnary search n the nterval [α mn,α max ]. The bnary search s descrbed and analyzed n Appendx C, where we show that Algorthm RSR computes an ε-approxmate equlbrum n tme of the order of T(n)(log M + log 1 ε log α mn + log (1 α max) ), where T(n) s the polynomal bound on the tme requred to solve the convex program, and M s an upper bound on the absolute value of the dervatve of B 1 (α) n the nterval [α mn,α max ]. The next secton takes a close look at the parameter M that nfluences the runnng tme. 3.4 Senstvty of the welfare maxmzaton problem The runnng tme of algorthm RSR depends on the logarthm of M, where M s an upper bound on B 1 (α) for α n [α mn,α max ]. We now show how to estmate M for a wde famly of utlty functons. The two-trader maxmzaton problem occurrng n step 3 of Algorthm RSR, when wrtten n ts expanded form, becomes the 2m-trader maxmzaton problem: max α γ u 1 (x 1 ) + (1 α) θ u 2 (x 2 ) s.t. x x1 m + x x2 m = w x 1, x2 0 7
9 where u 1 () = u 2 () = u (). Let x l (α) denote the soluton of ths problem, and π(α) the correspondng shadow prces. We wll smply denote x l (α) by xl and π(α) by p. Snce B 1(α) = p z α, we can upper bound B 1 p (α) by boundng the elements of the vector α. Let us consder the frst order optmalty condtons for the problem above, whch we wll denote by y {}}{ H( x,p,α) = 0. These equatons can be explctly wrtten as: αγ u 1 (x 1 ) p = 0; (1 α)θ u 2 (x 2 ) p = 0; x x1 m + x x2 m w = 0. By Implct Dfferentaton, from the frst order condtons we obtan the followng equaton: y H(x,p,α) α y(α) + H(x,p,α) = 0. (5) α Let A and B denote the Hessan αγ 2 u 1 (x 1 ), and (1 α)θ 2 u 2 (x 2 ), respectvely. Equaton 5 takes the form: A 1 I A m B 1 I I B m I I... I I... I x 1 1 α... x 1 m α x 2 1 α... x 2 m α p α γ 1 u 11(x 1 1). γ m u 1m(x 1 m) θ 1 u 21(x 2 1). θ m u 2m(x 2 m) = Assume that the A s and B s be nonsngular,.e., that they are negatve defnte. In ths case the matrx of the lnear system above has an nverse, whch s 2 A 1 1 A 1 1 KA 1 1 A 1 1 KA 1 2 A 1 1 KB 1 m A 1 1 K , 7 4 Bm 1 KA Bm 1 Bm 1 KBm 1 Bm 1 K 5 KA KBm 1 K ) 1. where K = ( A A 1 m + B Bm 1 Let now d 1 = γ u 1 (x 1 ), and d 2 = θ u 2 (x 2 ). We obtan the expresson p α = K( A 1 d B 1 d 2 ), (6) from whch we can upper bound the absolute value of any element of p α n terms of A 1, B 1, K, and u l (x l ), where denotes the spectral norm of a matrx, whch, n the case of sem-defnte matrces, concdes wth the spectral radus. In order to bound the spectral norm of K n terms of those of A and B, we now use a classcal result from lnear algebra. The theorem we need (see [24], p.192) states that f C 1 and 8
10 C 2 are real symmetrc matrces wth egenvalues λ 1 λ 2... λ n and µ 1 µ 2... µ n, respectvely, then the egenvalues s 1 s 2... s n of C 1 + C 2 satsfy λ k + µ n s k λ k + µ 1. For any matrx Q, let λ n (Q) denote ts smallest egenvalue. Then, f Q s postve defnte, 1 the spectral norm of ts nverse s gven by λ n(q). We apply the theorem to the postve defnte matrx K, and obtan (the calculaton s shown for m = 1, for the sake of clarty) ( A 1 1 B 1 1 ) 1 = λmax ( A 1 1 B 1 1 ) 1 = 1 λ mn ( A 1 1 B 1 1 ) 1 λ mn ( A 1 1 ) + λ mn( B1 1 ) = 1 1 A + 1 mn{ A 1, B 1 }. 1 B 1 Let now A m = mn A, B m = mn B, A 1 M = max A 1, and B 1 M = max B 1. From Equaton 6, we then get the crude upper bound p α 2mdmn{ A m, B m }max{ A 1 M, B 1 M }, = 1,...,m, where d s the maxmum of the d l s. We have shown how to bound B 1 (α) n terms of the Hessans of the utlty functons u l evaluated at x l. For utlty functons wth negatve defnte Hessans, ths gves an approach for boundng B 1 (α). We show how we can proceed even further for an nterestng class of such functons. (Incdentally, recall that negatve defnteness of the Hessan s a suffcent condton for strct concavty of the utlty functon, but t s not necessary.) Let f(x 1,...,x n ) be a concave log-homogeneous utlty functon. Its Hessan F(x 1,...,x n ) s negatve sem-defnte. Consder the functon g(x 1,...,x n ) = 1 n (log x log x n ). Its Hessan G(x 1,...,x n ) s the dagonal matrx whose (,)-th entry s 1, so that the maxmum nx 2 1 egenvalue of G s at most, where c s the mnmum consumpton of any good n x, and nc 2 1 ts mnmum egenvalue s at least, where C s the maxmum consumpton. nc 2 Let now consder the functon h(x 1,...,x n ) = (1 σ)f(x 1,...,x n ) + σg(x 1,...,x n ), for a small postve constant σ less than one. Ths functon s log-homogeneous, and ts Hessan H(x 1,...,x n ) s negatve defnte (agan by applyng the result n [24]). Let f max = max 2 f. x x We obtan: H (1 σ)nf max + σ nc 2, H 1 1 (1 σ)λ mn ( F) + σλ mn ( G) nc2 σ. (7) Therefore the functon h (whch s ntended as an approxmaton of the concave log-homogeneous functon f) s such that: () the Hessan H s a negatve defnte matrx (f there s postve consumpton of each of the goods);() the spectral norm of H can be bounded n terms of the mnmum consumpton, and the absolute value of the maxmum entry of F, whle the spectral norm of ts nverse can be bounded n terms of the maxmum consumpton. In order to lower bound the mnmum consumpton c we can proceed as follows. Let us consder the maxmzaton problem of a consumer wth a utlty functon of the form h(x) = (1 σ)f(x) + σg(x), where f and g are as above, and budget constrant π x = I. The frst order condtons for ths problem are of the form (1 σ) f(x) x k + σ 1 nx k = λπ k. 9
11 If we multply both sdes of ths equaton by x k, and then sum over k, we obtan that λ = 1 I. Snce f(x) x k 0, we get x k I σ nπ k. Snce α [α mn,α max ], the mnmum expendture I mn over all traders at α s bounded below by mn mn{γ α mn,θ (1 α max )}. Snce the total expendture of the 2m traders n the welfare maxmzaton problem above s one, then p k w k 1, from whch we must have x k I σw k n. Therefore, the mnmum consumpton s lower bounded by σi mnw mn where w mn s the quantty of the most scarce good. The maxmum consumpton C can be upper bounded by observng that π x 1. Now notce that at least one of the 2m traders has an expendture of at least 1/2m. Ths trader wll σ 2π nm n, σ demand at least unts of good. If π s to be an equlbrum, we must have 2π nm w. Therefore we obtan C 2nmwmax σ, where w max the quantty of the most abundant good. Fnally, for a wde famly of utlty functons, we can upper bound f max n terms of mnmum and maxmum consumptons. The reader s nvted to check ths when f s the log of a CES functon. 4 Lnear Exchange Economes wth Dfferentated Taxes We consder an economy wth m traders and n goods where each trader has a lnear utlty functon. Let u = a x denote the utlty functon of the -th trader. Each trader has the ntal endowment w R n +. Let τ 0 denote the tax rate of the -th trader for the consumpton of the -th good. And let θ denote the share of the -th trader n the overall tax collected. We have θ = 1. An equlbrum s a prce vector π = (π 1,...,π n ) and a number R 0 at whch there are bundles x R n + for each trader such that (1) x maxmzes u (x) over all x R n + such that the cost π (1 + τ )x of bundle x s at most the ncome θ R + π w ; (2) x w for each good ; and (3) T (x,π) = R, where T (x,π) s defned to be τ π x, the tax that has to pay to consume x at prce π. For ε > 0, we defne an ε-approxmate equlbrum to be a prce vector π = (π 1,...,π n ) and a number R 0 at whch there are bundles x R n + for each trader such that (1) π x + T(x,π) (1 + ε)(θ R + π w ), and u (x ) (1 ε)v (π,r), where v (π,r) s the maxmum value of u (x) over all x R n + such that π (1 + τ )x θ R + π w ; (2) x (1 + ε) w for each good ; and (3) (1 ε)r T (x,π) (1 + ε)r. We now descrbe our algorthm, an adaptaton of the aucton based algorthm of Garg and Kapoor [13], for computng an approxmate equlbrum of the model. The analyss of the algorthm assumes that a > 0 for each and. Let τ max denote max, τ. For smplfyng some expressons, we also assume, wthout loss of generalty, that w = 1 for each. Let w mn = mn, w. Our analyss also assumes that w mn > 0. Let κ = 1/w mn. ε Let δ = 90n(1+τ. The algorthm has varables π max)κ for the prces, and a varable R that stands for the tax money that s dstrbuted to the buyers. The algorthm starts wth all prces set to 1. From tme to tme, t ncreases the prce of some good by a multplcatve factor of 1 + δ. It has varables y and h correspondng to the amounts of good allocated to at the current prce π and the prevous prce π /(1 + δ), respectvely. Let x = y + h. Let T (y,h,π) = τ (π y + π 1+δ h ), the tax that pays n consumng y and h. Let a D (π) = { (1 + τ )π a k (1 + τ k )π k for 1 k n}. Intalze. Let π = 1 for 1 n, R = 1, y = 0 and h = 0 for each and. Phase 1: 10
12 1. We make a call to the procedure Allocatemore(), descrbed below. 2. If T (y, h, π) = R, let R R(1 + δ) and go to Step 1 of Phase If for each, we have (π y + π 1+δ h ) + T (y, h, π) = θ R + π w, the algorthm ends. 4. If for some, we have (π y + π 1+δ h )+T (y, h, π) < θ R+ π w, consder any D (π). An nspecton of Allocatemore() tells us that we must have h = 0 for every trader and y = 1. We call Raseprce(). If π k > 0 for every good k, we ump to Step 1 of Phase 2. Otherwse, return to Step 1 of Phase 1. Phase 2: 1. If R δ π w for each, let R T (y, h, π); the algorthm ends. 2. Make a call to Allocatemore(). 3. If T (y, h, π) = R, the algorthm ends. 4. If for each, we have (π y + π 1+δ h ) + T (y, h, π) = θ R + π w, the algorthm ends. 5. If for some, we have (π y + π 1+δ h )+T (y, h, π) < θ R+ π w, consder any D (π). We must have h = 0 for every trader and y = 1. We call Raseprce() and return to Step 1 of Phase 2. To complete the descrpton of the algorthm, we need to specfy the two procedures Allocatemore and Raseprce. The procedure Allocatemore. In ths procedure, we frst solve a lnear program, whch uses as data the current values of the varables π, y, h, and R. The lnear program has varables y and h for each trader and good. The lnear program s: X Maxmze X, y Subect to X (y + h ) 1 for each X X (y + h ) = 1 for each such that π > 1 X T (y, h, π) R π y + π 1 + δ h + T (y, h, π) θ R + X π w for each y = 0 for each and D (π). h = 0 for each and such that h = 0. y 0 for each, h 0 for each, As we argue below, ths lnear program wll always be feasble. Thus the maxmzaton s well defned. After solvng the lnear program, we set y = y and h = h for each and. Ths completes the descrpton of the procedure Allocatemore. The procedure Raseprce(). In ths procedure, we set π π (1 + δ), h y for each, and y 0 for each. Ths completes the descrpton of Raseprce. 11
13 Phases 1 and 2 of our algorthm are smlar to the basc algorthm of Garg and Kapoor [13] - a call to Allocatemore() replaces the sequence of steps n ther algorthm occurrng between two prce rases. Our algorthm needs to track the relaton of T (y,h,π) the tax pad as a consequence of consumpton to R, the tax that s dstrbuted as ncome. The reason we have Phase 2 s that at the end of Phase 1, T (y,h,π) can be sgnfcantly smaller than R. Theorem 4 For any ε > 0, our algorthm computes an ε-approxmate equlbrum for the model n tme that s polynomal n the nput sze, 1/ε, and κ. For lack of space, the proof of ths theorem s gven n Appendx D. In the process, we state some nvarants that help n understandng the algorthm. References [1] K. Arrow; H. Block; L. Hurwcz, On the Stablty of the Compettve Equlbrum, II, Econometrca, Vol. 27, No. 1. (Jan., 1959), pp [2] K.J. Arrow and G. Debreu, Exstence of an Equlbrum for a Compettve Economy, Econometrca 22 (3), pp (1954). [3] B. Codenott, B. McCune, S. Penumatcha, and K. Varadaraan, Exstence, Multplcty and Computaton of Equlbra for CES Exchange Economes, FSTTCS [4] B.Codenott, B. McCune and K. Varadaraan, Market Equlbrum va the Excess Demand Functon. Proc. STOC [5] B.Codenott, B. McCune and K. Varadaraan, Computng Equlbrum Prces: Does Theory Meet Practce? Proc. ESA [6] B.Codenott, S. Pemmarau and K. Varadaraan, Algorthms Column: The Computaton of Market Equlbra, ACM SIGACT News 35(4) December [7] B. Codenott, S. Pemmarau, K. Varadaraan, On the Polynomal Tme Computaton of Equlbra for Certan Exchange Economes. Proc. SODA [8] B. Codenott and K. Varadaraan, Effcent Computaton of Equlbrum Prces for Markets wth Leontef Utltes. ICALP [9] X. Deng, C. Papadmtrou, and S. Safra. On the complexty of equlbra. Proc. STOC [10] N. R. Devanur, C. H. Papadmtrou, A. Saber, V. V. Vazran, Market Equlbrum va a Prmal-Dual-Type Algorthm. FOCS 2002, pp [11] B. C. Eaves, Fnte Soluton of Pure Trade Markets wth Cobb-Douglas Utltes, Mathematcal Programmng Study 23, pp (1985). [12] E. Esenberg, Aggregaton of Utlty Functons. Management Scences, Vol. 7 (4), (1961). [13] R. Garg and S. Kapoor, Aucton Algorthms for Market Equlbrum. In Proc. STOC, [14] S. Gerstad, Multple Equlbra n Exchange Economes wth Homothetc, Nearly Identcal Preference, Unversty of Mnnesota, Center for Economc Research, Dscusson Paper 288, [15] K. Jan, A polynomal tme algorthm for computng the Arrow-Debreu market equlbrum for lnear utltes. FOCS
14 [16] K. Jan, M. Mahdan, and A. Saber, Approxmatng Market Equlbra, Proc. APPROX [17] K. Jan and K. Varadaraan. Equlbra for Economes wth Producton: Constant-Returns Technologes and Producton Plannng Constrants. To appear n SODA 06. [18] K. Jan, V. V. Vazran and Y. Ye, Market Equlbra for Homothetc, Quas-Concave Utltes and Economes of Scale n Producton, SODA [19] T.J. Kehoe, Computaton and Multplcty of Equlbra, n Handbook of Mathematcal Economcs Vol IV, pp , Noth Holland (1991). [20] T.J. Kehoe, The Comparatve Statcs Propertes of Tax Models, The Canadan Journal of Economcs 18(2), (1985). [21] R. R. Mantel, The welfare adustment process: ts stablty propertes. Internatonal Economc Revew 12, (1971). [22] T. Negsh, Welfare Economcs and Exstence of an Equlbrum for a Compettve Economy, Metroeconomca 12, (1960). [23] E. I. Nenakov and M. E. Prmak. One algorthm for fndng solutons of the Arrow-Debreu model, Kbernetca 3, (1983). [24] B.N. Parlett, The Symmetrc Egenvalue Problem, Prentce Hall (1980). [25] J.B. Shoven, J. Whalley, Appled General Equlbrum Models of Taxaton and Internatonal Trade, Journal of Economc Lterature 22, (1984). [26] J.B. Shoven, J. Whalley, Applyng General Equlbrum, Cambrdge Unversty Press (1992). [27] J. Whalley, S. Zhang, Tax nduced multple equlbra. Workng paper, Unversty of Western Ontaro (2002). 13
15 Appendx A: Proof of Proposton 1 From the market clearance condton, t follows that the tax revenue at equlbrum must be R = π τ x = π τ w. We can therefore elmnate the equlbrum requrement R = π τ x by substtutng π τ w for R n the rght hand sde of nequalty (2), whch thus becomes π (w + θ τ w ). After dvdng and multplyng each term of ths expresson by 1 + τ, the budget constrant can be rewrtten as π (1 + τ )x π (1 + τ )w, where w = w τ + θ w. 1 + τ 1 + τ We have therefore reduced the equlbrum n the orgnal economy to an equlbrum n a pure exchange economy, where the -th trader now has an ntal endowment w. It s easy to check that w = w for each, so that the two economes have the same quantty of each good. Fnally we have that (π 1,...,π n ) s an equlbrum for the orgnal economy E f and only f ((1 + τ 1 )π 1,...,(1 + τ n )π n ) s an equlbrum for the pure exchange economy E. Appendx B: Proof of Proposton 2 Suppose that consumer has a Cobb-Douglas utlty functon of the form n u (x ) = where α 0, n =1 α = 1. It s known (and easy to see) that the resultng demand s such that consumer wll spend on good an α fracton of her ncome. In vew of the budget constrant (3), the -th consumer ncome s π w + θ R, wth apparent prces π (1 + τ ). Thus, α x (π,r) = π (1 + τ ) (π w + θ R), = 1,...,n. We want to derve a consumer s problem that looks lke the consumer s problem of an economy wthout taxes, and that nduces the same demand as the orgnal one. The new problem wll have an addtonal good, the tax good, a utlty functon v( ) defned on n + 1 goods, and a budget constrant as n (4). The demand for the tax good has to be π τ x (π,r) R =1 x α α τ 1+τ x r (π,r) = = (π w + θ R). R Observe that ths demand looks exactly lke the demand generated by a Cobb-Douglas utlty functon: a constant dvded by the prce and multpled by the ncome. Thus, f we set α r = α τ 1+τ, α = α /(1 + τ ), the utlty functon v that we want s the Cobb-Douglas functon v(x 1,...,x n,x r ) = x α r x α r. 14
16 Appendx C: Analyss of algorthm RSR The correctness of the aggregaton and dsaggregaton processes executed n steps 2 and 4 of Algorthm RSR readly follows from Esenberg s results [12]. Consder now the two-trader problem of step 3, where we approxmately compute a zero of B 1 by means of the bsecton method, startng from the nterval [α mn,α max ]. Each bsecton step depends on the sgn of B 1 and s thus guded by the soluton to an nstance of the constraned maxmzaton problem. The next Lemma gves a stoppng crteron whch guarantees an ε-equlbrum: Lemma 5 If for some α we have 0 B 1 (α) α(1 α)ε then the prces π(α) and the allocatons x 1 (α) and (1 ε)x 2 (α) are an ε-equlbrum. Proof : The hypothess of the Lemma mples that B 1 (α) εmn{α,1 α}. The ε-optmalty of the frst consumer s guaranteed when B 1 (α) εα: snce α > 0, we have that 0 B 1 (α) εα s equvalent to 0 π(α) z π(α) x 1 (α) αε and, usng that α = π(α) x 1 (α), t s equvalent to ε π(α) z π(α) x 1(α) π(α) z. (8) On the other hand, f we want a smlar nequalty for the other trader, we observe that s equvalent to 0 B 2 (α) ε(1 α) (9) π(α) s π(α) x 2 (α) 1 π(α) s. (10) 1 ε Thus the bundles x 1 (α) and x 2 (α) nearly exhaust but do not sgnfcantly exceed the respectve budgets π(α) z and π(α) s. Snce they have the rght shape, we are at an O(ε)- equlbrum. Corollary 6 Suppose that B 1 ( ) s dfferentable n (0,1) and that there exsts M > 0 such that for all α [α mn,α max ] we have d dα B 1(α) M. If the bsecton method appled to B 1 n [α mn,α max ] stops when t determnes an nterval [α,β] such that β α α(1 α)ε M and B 1 (α) 0, B 1 (β) 0, then the concluson of Lemma 5 follows. Proof : Under the hypotheses, B 1 (α) B 1 (β) α(1 α)ε. In partcular, B 1 (α) α(1 α)ε and the concluson of Lemma 5 follows. Now we need to argue that from an ε-approxmate equlbrum to the two trader economy we can recover an ε-approxmate equlbrum to the orgnal economy wth taxes. We wll consder the mplcatons of undong the aggregatons and dsaggregatons nto equvalent economes one by one, n reverse order. 15
17 Suppose that the algorthm reaches the concluson of Lemma 5 at α = α 1. Let π = π(α 1 ), x 1 = x 1 (α 1 ), x 2 = x 2 (α 1 ), and α 2 = 1 α 1. Wrte the maxmzaton problem n Step 5 of the algorthm n ts extensve form: max s.t. α 1 ( γ u (x )) + (1 α 1)( x = x 1 x = x 2 x 1 + x 2 = w x, x 0 θ u (x )) Let x = x (α 1) and x = x (α 1). Obvously, we have x = x 1 and x = x 2. It can be shown, usng the log-homogenety of the utlty functons, that π x = γ α 1 and π x = θ α 2. For the dsaggregaton nto 2m traders, we further note that x and x have the rght shape they are proportonal to the bundles demanded by the two nstances of the -th trader. Snce α 1 = π x 1 roughly equals π z (recall the proof of Lemma 5), t follows that π x = γ α 1 roughly equals γ π z. Smlarly, π x = θ α 2 roughly equals θ π s. So we are at an approxmate equlbrum of the 2m trader economy. For the reaggregaton of pars of traders, t s mmedate that π s an approxmate equlbrum of the m trader pure exchange economy. The approxmate utlty maxmzng bundles are x = x + x. The natural canddate equlbrum prces for the orgnal economy wth taxes are π = π /(1 + τ ). We set R = π τ w. Snce w = x, R = π τ x. Obvously, we have x = w. We can easly verfy that the x are approxmate utlty maxmzng bundles for each trader at prces π and R. The runnng tme of Algorthm RSR s domnated by step (3), whch can be executed n tme proportonal to T(n)(log M +log 1 ε +log 1 1 α mn +log ), where T(n) s the polynomal bound on the tme requred to solve the convex program. (1 α max) Appendx D: Proof of Theorem 4 Invarants. We begn by statng some mportant nvarants mantaned by the algorthm. Proposton 7 At any stage n the algorthm, we have (1) y > 0 = D (π); (2) h > 0 = D (( π 1+δ,π )), where (r,π ) s defned to be the vector that dffers from π only n ts -th component, whch s r; (3) π > 1 = x = 1; (4) The lnear program n Allocatemore() s always feasble. Proof : The varables y, h, and π, are only changed by calls to Allocatemore() or Raseprce(). Note that the lnear program n Allocatemore() s feasble f (1) and (3) hold: y = y and h = h yelds a feasble soluton. On the other hand, t s evdent that Allocatemore() preserves (1), (2), and (3). 16
18 It s also easy to check that a call to Raseprce() preserves (1), (2), and (3). Arguments are smlar to those n [13]. Proposton 8 At any stage of the algorthm, we have: (1) x 1 for each ; (2) T (y,h,π) R. Proof : Easly follows by nspectng Allocatemore() and Raseprce(). Proposton 9 At any stage of the algorthm, except possbly after the executon of the (termnal) assgnment statement n Step 1 of Phase 2, we have, for each, that π y + π 1+δ h + T (y,h,π) θ R + π w. Proposton 10 Let π max denote the largest component of π and π mn the smallest component of π. At any stage of the algorthm, we have (1) πmax π mn 2amax(1+τmax) a mn ; (2) R max{1,2nτ max π max }. Proof : To prove (1), t s suffcent to consder the case when π max > 1. Suppose π max = π > 1. Then by statement (3) of Proposton 7, we must have x > 0 for some. By Proposton 7, we ether have D (π) or D (( π 1+δ,π )). In ether case, t s true that D (( π 1+δ,π )). Thus a (1+δ) (1+τ )π whch completes the proof of (1). a k (1+τ k )π k for every k. Rearrangng, we have π π k a (1+τ k )(1+δ) a k (1+τ ), For (2), observe that f at some stage R > 1, then R = (1+δ)R, where R = T (ŷ,ĥ, ˆπ), where ŷ,ĥ, and ˆπ are the values at some prevous stage of y, h, and π. Usng statement (1) of Proposton 8, we see that T (ŷ,ĥ, ˆπ) T (ˆx, ˆπ) = τ ˆx ˆπ τ maxˆx ˆπ = τ max ˆπ τ max ˆπ. Thus R (1 + δ)τ max ˆπ (1 + δ)nτ max π max, snce π max π ˆπ for every (the algorthm never decreases a prce). Ths completes the proof. Runnng Tme. For each phase, we bound the number of tmes we call Raseprce and the number of tmes we rase the value of R. Frst observe that the moment π mn exceeds 1 n Phase 1, we ump to Phase 2 va Step 4 of phase 1. Thus π mn (1 + δ) 2 n Phase 1, and by statement (1) n Proposton 10, we have π max 4amax(1+τmax) a mn n Phase 1. Thus, by statement (2) n Proposton 10, we have R max{1, 8nτmax(1+τmax)amax a mn } 8n(1+τmax)2 a max a mn n Phase 1. Snce R s never rased n Phase 2, we have that ˆx R α = 8n(1 + τ max) 2 a max a mn (11) 17
19 throughout the algorthm. We now use ths bound to bound π max n Phase 2. Phase 2 stops when R δ π w for each, due to the stoppng rule n Step 1 of Phase 2. To have R > δ π w for some, we must have R > δπ mn w mn, whch mples that π mn < R δw mn. Thus we must have π mn < (1+δ)R δw mn 2R δw mn throughout Phase 2. Usng statement (1) n Proposton 10, and Equaton 11, we have π max 16n(1+τmax)2 a max a mn δw mn. Snce ths bound s larger than the bound we establshed for π max n Phase 1, we conclude that π max β = 16n(1 + τ max) 2 a max a mn δw mn (12) It follows that the number of calls to Raseprce plus the number of tmes R s rased s bounded by n log 1+δ α+log 1+δ β, whch s a polynomal n the nput sze, 1/ε, and κ. It follows that the runnng tme of the algorthm s bounded by a polynomal n the nput sze, 1/ε, and κ. Correctness. We now argue that the algorthm termnates at an ε-approxmate equlbrum. Suppose that the algorthm termnates ether va Step 3 of Phase 1 or Step 4 of Phase 2. Then, we have for each, T (y,h,π) + π x T (y,h,π) + (π y + π 1+δ h ) = θ R + π w. Addng over all, we get T (y,h,π) + π x R + π w. (13) Multplyng the -th nequalty n statement (1) of Proposton 8 by π, addng these nequaltes and the nequalty n statement (2) of Proposton 8, we get T (y,h,π) + π x R + π w. (14) Thus for Equaton 13 to hold, the nequaltes n Proposton 8 must be equaltes. So we have for each that x = w. Also, we have T (x,π) (1 + δ)t (y,h,π) = (1 + δ)r, whereas T (x,π) T (y,h,π) = R, so we have R T (x,π) (1 + δ)r. For each trader, we have T (x,π) + π x T (y,h,π) + π x θ R + π w. Thus the bundle x costs at least her ncome. But by Proposton 7, we have that f x > 0 then ether D (π) or D (( π 1+δ,π )). Thus only spends money on goods that nearly optmze her bang for the buck. It follows va standard arguments that u (x ) v (π,r) 1+δ. On the other hand, T (x,π) + π x (1 + δ)(t (y,h,π) + (π y + π 1+δ h )) = (1 + δ)(θ R + π w ), so the bundle x does not cost more than 1 + δ tmes ts ncome. Thus we are at a δ-approxmate equlbrum. Now suppose that the algorthm termnates at ether Step 1 of Phase 2 or Step 3 of Phase 2. Note that we enter Phase 2 when π k > 1 for every good k, so by statement (3) of Proposton 7 we have x k = 1 for every good k. Moreover, termnaton at ether of these steps mples that T (y,h,π ) = R, whch mples that R T(x,π) (1 + δ)r. We now reason about the budget constrants of the traders. By Proposton 9, we have, for each T (x,π) + π x (1 + δ)(t (y,h,π) + π y + π 1 + δ h ) (1 + δ)(θ R + π w ), 18
20 f we termnate va Step 3 of Phase 2. If we termnate after makng the assgnment to R n Step 1 of Phase 2, then we have R δ π w for each ust before the assgnment, whch mples that θ R + π w (1 + δ) π w ust before the assgnment. So the assgnment decreases the ncome of each trader by at most a multplcatve factor of (1+δ). Snce the nequalty n Proposton 9 holds ust pror to the assgnment, we have that for each, T (y,h,π) + π y + π 1 + δ h (1 + δ)(θ R + π w ) after the assgnment. It follows that on termnaton, we have for each T (x,π) + π x (1 + δ) 2 (θ R + π w ) (1 + 3δ)(θ R + π w ). (15) Note that ths nequalty also holds f we termnate va Step 3 of Phase 2. Havng argued that no trader exceeds hs budget by much, we now argue that each trader almost exhausts hs budget. We have (θ R+ π w π x T (x,π)) = π ( w x )+R T (x,π) 0. (16) It follows that for any trader, we have θ R + π w π x T (x,π) max{0, π x + T (x,π) θ R π w }. Usng Equaton 15, and Proposton 10, we have θ R + π w π x T (x,π) 3δ(θ R + π w ) = 3δ(R + π ) 3δ(2nπ max + 2nπ max τ max + π ) 3δ3nπ max (1 + τ max ) ε 10 π kw k ε 10 ( π w + θ R). Here k s the good for whch π max = π k. We have argued that each trader spends at least a fracton 1 1+ε of her budget on bundle x. But by Proposton 7, we have that f x > 0 then ether D (π) or D (( π 1+δ,π )). Thus only spends money on goods that nearly optmze her bang for the buck. It follows that u (x ) (1 cε)v (π,r), where c 1 s some constant. Thus we are at a cε-approxmate equlbrum. 19
Economics 101. Lecture 4 - Equilibrium and Efficiency
Economcs 0 Lecture 4 - Equlbrum and Effcency Intro As dscussed n the prevous lecture, we wll now move from an envronment where we looed at consumers mang decsons n solaton to analyzng economes full of
More informationCS294 Topics in Algorithmic Game Theory October 11, Lecture 7
CS294 Topcs n Algorthmc Game Theory October 11, 2011 Lecture 7 Lecturer: Chrstos Papadmtrou Scrbe: Wald Krchene, Vjay Kamble 1 Exchange economy We consder an exchange market wth m agents and n goods. Agent
More informationComputing Correlated Equilibria in Multi-Player Games
Computng Correlated Equlbra n Mult-Player Games Chrstos H. Papadmtrou Presented by Zhanxang Huang December 7th, 2005 1 The Author Dr. Chrstos H. Papadmtrou CS professor at UC Berkley (taught at Harvard,
More informationprinceton univ. F 17 cos 521: Advanced Algorithm Design Lecture 7: LP Duality Lecturer: Matt Weinberg
prnceton unv. F 17 cos 521: Advanced Algorthm Desgn Lecture 7: LP Dualty Lecturer: Matt Wenberg Scrbe: LP Dualty s an extremely useful tool for analyzng structural propertes of lnear programs. Whle there
More informationPerfect Competition and the Nash Bargaining Solution
Perfect Competton and the Nash Barganng Soluton Renhard John Department of Economcs Unversty of Bonn Adenauerallee 24-42 53113 Bonn, Germany emal: rohn@un-bonn.de May 2005 Abstract For a lnear exchange
More informationModule 9. Lecture 6. Duality in Assignment Problems
Module 9 1 Lecture 6 Dualty n Assgnment Problems In ths lecture we attempt to answer few other mportant questons posed n earler lecture for (AP) and see how some of them can be explaned through the concept
More informationCopyright (C) 2008 David K. Levine This document is an open textbook; you can redistribute it and/or modify it under the terms of the Creative
Copyrght (C) 008 Davd K. Levne Ths document s an open textbook; you can redstrbute t and/or modfy t under the terms of the Creatve Commons Attrbuton Lcense. Compettve Equlbrum wth Pure Exchange n traders
More informationA NOTE ON CES FUNCTIONS Drago Bergholt, BI Norwegian Business School 2011
A NOTE ON CES FUNCTIONS Drago Bergholt, BI Norwegan Busness School 2011 Functons featurng constant elastcty of substtuton CES are wdely used n appled economcs and fnance. In ths note, I do two thngs. Frst,
More informationCOS 521: Advanced Algorithms Game Theory and Linear Programming
COS 521: Advanced Algorthms Game Theory and Lnear Programmng Moses Charkar February 27, 2013 In these notes, we ntroduce some basc concepts n game theory and lnear programmng (LP). We show a connecton
More informationThe Minimum Universal Cost Flow in an Infeasible Flow Network
Journal of Scences, Islamc Republc of Iran 17(2): 175-180 (2006) Unversty of Tehran, ISSN 1016-1104 http://jscencesutacr The Mnmum Unversal Cost Flow n an Infeasble Flow Network H Saleh Fathabad * M Bagheran
More informationWelfare Properties of General Equilibrium. What can be said about optimality properties of resource allocation implied by general equilibrium?
APPLIED WELFARE ECONOMICS AND POLICY ANALYSIS Welfare Propertes of General Equlbrum What can be sad about optmalty propertes of resource allocaton mpled by general equlbrum? Any crteron used to compare
More informationChapter 5. Solution of System of Linear Equations. Module No. 6. Solution of Inconsistent and Ill Conditioned Systems
Numercal Analyss by Dr. Anta Pal Assstant Professor Department of Mathematcs Natonal Insttute of Technology Durgapur Durgapur-713209 emal: anta.bue@gmal.com 1 . Chapter 5 Soluton of System of Lnear Equatons
More informationMMA and GCMMA two methods for nonlinear optimization
MMA and GCMMA two methods for nonlnear optmzaton Krster Svanberg Optmzaton and Systems Theory, KTH, Stockholm, Sweden. krlle@math.kth.se Ths note descrbes the algorthms used n the author s 2007 mplementatons
More informationDifference Equations
Dfference Equatons c Jan Vrbk 1 Bascs Suppose a sequence of numbers, say a 0,a 1,a,a 3,... s defned by a certan general relatonshp between, say, three consecutve values of the sequence, e.g. a + +3a +1
More informationEquilibrium with Complete Markets. Instructor: Dmytro Hryshko
Equlbrum wth Complete Markets Instructor: Dmytro Hryshko 1 / 33 Readngs Ljungqvst and Sargent. Recursve Macroeconomc Theory. MIT Press. Chapter 8. 2 / 33 Equlbrum n pure exchange, nfnte horzon economes,
More informationAPPENDIX A Some Linear Algebra
APPENDIX A Some Lnear Algebra The collecton of m, n matrces A.1 Matrces a 1,1,..., a 1,n A = a m,1,..., a m,n wth real elements a,j s denoted by R m,n. If n = 1 then A s called a column vector. Smlarly,
More informationLecture Notes, January 11, 2010
Economcs 200B UCSD Wnter 2010 Lecture otes, January 11, 2010 Partal equlbrum comparatve statcs Partal equlbrum: Market for one good only wth supply and demand as a functon of prce. Prce s defned as the
More informationEconomics 2450A: Public Economics Section 10: Education Policies and Simpler Theory of Capital Taxation
Economcs 2450A: Publc Economcs Secton 10: Educaton Polces and Smpler Theory of Captal Taxaton Matteo Parads November 14, 2016 In ths secton we study educaton polces n a smplfed verson of framework analyzed
More informationMixed Taxation and Production Efficiency
Floran Scheuer 2/23/2016 Mxed Taxaton and Producton Effcency 1 Overvew 1. Unform commodty taxaton under non-lnear ncome taxaton Atknson-Stgltz (JPubE 1976) Theorem Applcaton to captal taxaton 2. Unform
More informationLectures - Week 4 Matrix norms, Conditioning, Vector Spaces, Linear Independence, Spanning sets and Basis, Null space and Range of a Matrix
Lectures - Week 4 Matrx norms, Condtonng, Vector Spaces, Lnear Independence, Spannng sets and Bass, Null space and Range of a Matrx Matrx Norms Now we turn to assocatng a number to each matrx. We could
More informationProblem Set 9 Solutions
Desgn and Analyss of Algorthms May 4, 2015 Massachusetts Insttute of Technology 6.046J/18.410J Profs. Erk Demane, Srn Devadas, and Nancy Lynch Problem Set 9 Solutons Problem Set 9 Solutons Ths problem
More informationAssortment Optimization under MNL
Assortment Optmzaton under MNL Haotan Song Aprl 30, 2017 1 Introducton The assortment optmzaton problem ams to fnd the revenue-maxmzng assortment of products to offer when the prces of products are fxed.
More informationThe Order Relation and Trace Inequalities for. Hermitian Operators
Internatonal Mathematcal Forum, Vol 3, 08, no, 507-57 HIKARI Ltd, wwwm-hkarcom https://doorg/0988/mf088055 The Order Relaton and Trace Inequaltes for Hermtan Operators Y Huang School of Informaton Scence
More informationCollege of Computer & Information Science Fall 2009 Northeastern University 20 October 2009
College of Computer & Informaton Scence Fall 2009 Northeastern Unversty 20 October 2009 CS7880: Algorthmc Power Tools Scrbe: Jan Wen and Laura Poplawsk Lecture Outlne: Prmal-dual schema Network Desgn:
More informationMore metrics on cartesian products
More metrcs on cartesan products If (X, d ) are metrc spaces for 1 n, then n Secton II4 of the lecture notes we defned three metrcs on X whose underlyng topologes are the product topology The purpose of
More informationAdditional Codes using Finite Difference Method. 1 HJB Equation for Consumption-Saving Problem Without Uncertainty
Addtonal Codes usng Fnte Dfference Method Benamn Moll 1 HJB Equaton for Consumpton-Savng Problem Wthout Uncertanty Before consderng the case wth stochastc ncome n http://www.prnceton.edu/~moll/ HACTproect/HACT_Numercal_Appendx.pdf,
More informationBOUNDEDNESS OF THE RIESZ TRANSFORM WITH MATRIX A 2 WEIGHTS
BOUNDEDNESS OF THE IESZ TANSFOM WITH MATIX A WEIGHTS Introducton Let L = L ( n, be the functon space wth norm (ˆ f L = f(x C dx d < For a d d matrx valued functon W : wth W (x postve sem-defnte for all
More informationPROBLEM SET 7 GENERAL EQUILIBRIUM
PROBLEM SET 7 GENERAL EQUILIBRIUM Queston a Defnton: An Arrow-Debreu Compettve Equlbrum s a vector of prces {p t } and allocatons {c t, c 2 t } whch satsfes ( Gven {p t }, c t maxmzes βt ln c t subject
More informationU.C. Berkeley CS294: Beyond Worst-Case Analysis Luca Trevisan September 5, 2017
U.C. Berkeley CS94: Beyond Worst-Case Analyss Handout 4s Luca Trevsan September 5, 07 Summary of Lecture 4 In whch we ntroduce semdefnte programmng and apply t to Max Cut. Semdefnte Programmng Recall that
More informationThe Second Anti-Mathima on Game Theory
The Second Ant-Mathma on Game Theory Ath. Kehagas December 1 2006 1 Introducton In ths note we wll examne the noton of game equlbrum for three types of games 1. 2-player 2-acton zero-sum games 2. 2-player
More informationCSci 6974 and ECSE 6966 Math. Tech. for Vision, Graphics and Robotics Lecture 21, April 17, 2006 Estimating A Plane Homography
CSc 6974 and ECSE 6966 Math. Tech. for Vson, Graphcs and Robotcs Lecture 21, Aprl 17, 2006 Estmatng A Plane Homography Overvew We contnue wth a dscusson of the major ssues, usng estmaton of plane projectve
More informationErrors for Linear Systems
Errors for Lnear Systems When we solve a lnear system Ax b we often do not know A and b exactly, but have only approxmatons  and ˆb avalable. Then the best thng we can do s to solve ˆx ˆb exactly whch
More informationLinear Approximation with Regularization and Moving Least Squares
Lnear Approxmaton wth Regularzaton and Movng Least Squares Igor Grešovn May 007 Revson 4.6 (Revson : March 004). 5 4 3 0.5 3 3.5 4 Contents: Lnear Fttng...4. Weghted Least Squares n Functon Approxmaton...
More informationLecture 21: Numerical methods for pricing American type derivatives
Lecture 21: Numercal methods for prcng Amercan type dervatves Xaoguang Wang STAT 598W Aprl 10th, 2014 (STAT 598W) Lecture 21 1 / 26 Outlne 1 Fnte Dfference Method Explct Method Penalty Method (STAT 598W)
More informationTHE CHINESE REMAINDER THEOREM. We should thank the Chinese for their wonderful remainder theorem. Glenn Stevens
THE CHINESE REMAINDER THEOREM KEITH CONRAD We should thank the Chnese for ther wonderful remander theorem. Glenn Stevens 1. Introducton The Chnese remander theorem says we can unquely solve any par of
More informationModule 3 LOSSY IMAGE COMPRESSION SYSTEMS. Version 2 ECE IIT, Kharagpur
Module 3 LOSSY IMAGE COMPRESSION SYSTEMS Verson ECE IIT, Kharagpur Lesson 6 Theory of Quantzaton Verson ECE IIT, Kharagpur Instructonal Objectves At the end of ths lesson, the students should be able to:
More informationMaximizing the number of nonnegative subsets
Maxmzng the number of nonnegatve subsets Noga Alon Hao Huang December 1, 213 Abstract Gven a set of n real numbers, f the sum of elements of every subset of sze larger than k s negatve, what s the maxmum
More informationNorm Bounds for a Transformed Activity Level. Vector in Sraffian Systems: A Dual Exercise
ppled Mathematcal Scences, Vol. 4, 200, no. 60, 2955-296 Norm Bounds for a ransformed ctvty Level Vector n Sraffan Systems: Dual Exercse Nkolaos Rodousaks Department of Publc dmnstraton, Panteon Unversty
More informationMarket Equilibrium in Exchange Economies with Some Families of Concave Utility Functions
Maret Equlbrum n Exchange Economes wth Some Famles of Concave Utlty Functons Bruno Codenott Kastur Varadarajan Abstract We consder the problem of computng equlbrum prces for exchange economes. We tacle
More informationU.C. Berkeley CS294: Spectral Methods and Expanders Handout 8 Luca Trevisan February 17, 2016
U.C. Berkeley CS94: Spectral Methods and Expanders Handout 8 Luca Trevsan February 7, 06 Lecture 8: Spectral Algorthms Wrap-up In whch we talk about even more generalzatons of Cheeger s nequaltes, and
More informationUniversity of California, Davis Date: June 22, 2009 Department of Agricultural and Resource Economics. PRELIMINARY EXAMINATION FOR THE Ph.D.
Unversty of Calforna, Davs Date: June 22, 29 Department of Agrcultural and Resource Economcs Department of Economcs Tme: 5 hours Mcroeconomcs Readng Tme: 2 mnutes PRELIMIARY EXAMIATIO FOR THE Ph.D. DEGREE
More informationHow Strong Are Weak Patents? Joseph Farrell and Carl Shapiro. Supplementary Material Licensing Probabilistic Patents to Cournot Oligopolists *
How Strong Are Weak Patents? Joseph Farrell and Carl Shapro Supplementary Materal Lcensng Probablstc Patents to Cournot Olgopolsts * September 007 We study here the specal case n whch downstream competton
More information,, MRTS is the marginal rate of technical substitution
Mscellaneous Notes on roducton Economcs ompled by eter F Orazem September 9, 00 I Implcatons of conve soquants Two nput case, along an soquant 0 along an soquant Slope of the soquant,, MRTS s the margnal
More informationStanford University CS359G: Graph Partitioning and Expanders Handout 4 Luca Trevisan January 13, 2011
Stanford Unversty CS359G: Graph Parttonng and Expanders Handout 4 Luca Trevsan January 3, 0 Lecture 4 In whch we prove the dffcult drecton of Cheeger s nequalty. As n the past lectures, consder an undrected
More informationSTAT 309: MATHEMATICAL COMPUTATIONS I FALL 2018 LECTURE 16
STAT 39: MATHEMATICAL COMPUTATIONS I FALL 218 LECTURE 16 1 why teratve methods f we have a lnear system Ax = b where A s very, very large but s ether sparse or structured (eg, banded, Toepltz, banded plus
More information6) Derivatives, gradients and Hessian matrices
30C00300 Mathematcal Methods for Economsts (6 cr) 6) Dervatves, gradents and Hessan matrces Smon & Blume chapters: 14, 15 Sldes by: Tmo Kuosmanen 1 Outlne Defnton of dervatve functon Dervatve notatons
More informationEdge Isoperimetric Inequalities
November 7, 2005 Ross M. Rchardson Edge Isopermetrc Inequaltes 1 Four Questons Recall that n the last lecture we looked at the problem of sopermetrc nequaltes n the hypercube, Q n. Our noton of boundary
More information1. relation between exp. function and IUF
Dualty Dualty n consumer theory II. relaton between exp. functon and IUF - straghtforward: have m( p, u mn'd value of expendture requred to attan a gven level of utlty, gven a prce vector; u ( p, M max'd
More information2.3 Nilpotent endomorphisms
s a block dagonal matrx, wth A Mat dm U (C) In fact, we can assume that B = B 1 B k, wth B an ordered bass of U, and that A = [f U ] B, where f U : U U s the restrcton of f to U 40 23 Nlpotent endomorphsms
More information3.1 Expectation of Functions of Several Random Variables. )' be a k-dimensional discrete or continuous random vector, with joint PMF p (, E X E X1 E X
Statstcs 1: Probablty Theory II 37 3 EPECTATION OF SEVERAL RANDOM VARIABLES As n Probablty Theory I, the nterest n most stuatons les not on the actual dstrbuton of a random vector, but rather on a number
More information6.854J / J Advanced Algorithms Fall 2008
MIT OpenCourseWare http://ocw.mt.edu 6.854J / 18.415J Advanced Algorthms Fall 2008 For nformaton about ctng these materals or our Terms of Use, vst: http://ocw.mt.edu/terms. 18.415/6.854 Advanced Algorthms
More informationOnline Appendix: Reciprocity with Many Goods
T D T A : O A Kyle Bagwell Stanford Unversty and NBER Robert W. Stager Dartmouth College and NBER March 2016 Abstract Ths onlne Appendx extends to a many-good settng the man features of recprocty emphaszed
More informationKernel Methods and SVMs Extension
Kernel Methods and SVMs Extenson The purpose of ths document s to revew materal covered n Machne Learnng 1 Supervsed Learnng regardng support vector machnes (SVMs). Ths document also provdes a general
More informationCHAPTER 5 NUMERICAL EVALUATION OF DYNAMIC RESPONSE
CHAPTER 5 NUMERICAL EVALUATION OF DYNAMIC RESPONSE Analytcal soluton s usually not possble when exctaton vares arbtrarly wth tme or f the system s nonlnear. Such problems can be solved by numercal tmesteppng
More informationSL n (F ) Equals its Own Derived Group
Internatonal Journal of Algebra, Vol. 2, 2008, no. 12, 585-594 SL n (F ) Equals ts Own Derved Group Jorge Macel BMCC-The Cty Unversty of New York, CUNY 199 Chambers street, New York, NY 10007, USA macel@cms.nyu.edu
More informationPerron Vectors of an Irreducible Nonnegative Interval Matrix
Perron Vectors of an Irreducble Nonnegatve Interval Matrx Jr Rohn August 4 2005 Abstract As s well known an rreducble nonnegatve matrx possesses a unquely determned Perron vector. As the man result of
More informationFoundations of Arithmetic
Foundatons of Arthmetc Notaton We shall denote the sum and product of numbers n the usual notaton as a 2 + a 2 + a 3 + + a = a, a 1 a 2 a 3 a = a The notaton a b means a dvdes b,.e. ac = b where c s an
More information( ) 2 ( ) ( ) Problem Set 4 Suggested Solutions. Problem 1
Problem Set 4 Suggested Solutons Problem (A) The market demand functon s the soluton to the followng utlty-maxmzaton roblem (UMP): The Lagrangean: ( x, x, x ) = + max U x, x, x x x x st.. x + x + x y x,
More informationf(x,y) = (4(x 2 4)x,2y) = 0 H(x,y) =
Problem Set 3: Unconstraned mzaton n R N. () Fnd all crtcal ponts of f(x,y) (x 4) +y and show whch are ma and whch are mnma. () Fnd all crtcal ponts of f(x,y) (y x ) x and show whch are ma and whch are
More informationSection 8.3 Polar Form of Complex Numbers
80 Chapter 8 Secton 8 Polar Form of Complex Numbers From prevous classes, you may have encountered magnary numbers the square roots of negatve numbers and, more generally, complex numbers whch are the
More informationAppendix B. The Finite Difference Scheme
140 APPENDIXES Appendx B. The Fnte Dfference Scheme In ths appendx we present numercal technques whch are used to approxmate solutons of system 3.1 3.3. A comprehensve treatment of theoretcal and mplementaton
More informationVARIATION OF CONSTANT SUM CONSTRAINT FOR INTEGER MODEL WITH NON UNIFORM VARIABLES
VARIATION OF CONSTANT SUM CONSTRAINT FOR INTEGER MODEL WITH NON UNIFORM VARIABLES BÂRZĂ, Slvu Faculty of Mathematcs-Informatcs Spru Haret Unversty barza_slvu@yahoo.com Abstract Ths paper wants to contnue
More informationOn the Multicriteria Integer Network Flow Problem
BULGARIAN ACADEMY OF SCIENCES CYBERNETICS AND INFORMATION TECHNOLOGIES Volume 5, No 2 Sofa 2005 On the Multcrtera Integer Network Flow Problem Vassl Vasslev, Marana Nkolova, Maryana Vassleva Insttute of
More informationCS286r Assign One. Answer Key
CS286r Assgn One Answer Key 1 Game theory 1.1 1.1.1 Let off-equlbrum strateges also be that people contnue to play n Nash equlbrum. Devatng from any Nash equlbrum s a weakly domnated strategy. That s,
More informationLecture 10 Support Vector Machines II
Lecture 10 Support Vector Machnes II 22 February 2016 Taylor B. Arnold Yale Statstcs STAT 365/665 1/28 Notes: Problem 3 s posted and due ths upcomng Frday There was an early bug n the fake-test data; fxed
More informationYong Joon Ryang. 1. Introduction Consider the multicommodity transportation problem with convex quadratic cost function. 1 2 (x x0 ) T Q(x x 0 )
Kangweon-Kyungk Math. Jour. 4 1996), No. 1, pp. 7 16 AN ITERATIVE ROW-ACTION METHOD FOR MULTICOMMODITY TRANSPORTATION PROBLEMS Yong Joon Ryang Abstract. The optmzaton problems wth quadratc constrants often
More informationFinding Primitive Roots Pseudo-Deterministically
Electronc Colloquum on Computatonal Complexty, Report No 207 (205) Fndng Prmtve Roots Pseudo-Determnstcally Ofer Grossman December 22, 205 Abstract Pseudo-determnstc algorthms are randomzed search algorthms
More informationSolutions HW #2. minimize. Ax = b. Give the dual problem, and make the implicit equality constraints explicit. Solution.
Solutons HW #2 Dual of general LP. Fnd the dual functon of the LP mnmze subject to c T x Gx h Ax = b. Gve the dual problem, and make the mplct equalty constrants explct. Soluton. 1. The Lagrangan s L(x,
More informationECE559VV Project Report
ECE559VV Project Report (Supplementary Notes Loc Xuan Bu I. MAX SUM-RATE SCHEDULING: THE UPLINK CASE We have seen (n the presentaton that, for downlnk (broadcast channels, the strategy maxmzng the sum-rate
More informationFormulas for the Determinant
page 224 224 CHAPTER 3 Determnants e t te t e 2t 38 A = e t 2te t e 2t e t te t 2e 2t 39 If 123 A = 345, 456 compute the matrx product A adj(a) What can you conclude about det(a)? For Problems 40 43, use
More informationInner Product. Euclidean Space. Orthonormal Basis. Orthogonal
Inner Product Defnton 1 () A Eucldean space s a fnte-dmensonal vector space over the reals R, wth an nner product,. Defnton 2 (Inner Product) An nner product, on a real vector space X s a symmetrc, blnear,
More informationx i1 =1 for all i (the constant ).
Chapter 5 The Multple Regresson Model Consder an economc model where the dependent varable s a functon of K explanatory varables. The economc model has the form: y = f ( x,x,..., ) xk Approxmate ths by
More informationLecture 12: Discrete Laplacian
Lecture 12: Dscrete Laplacan Scrbe: Tanye Lu Our goal s to come up wth a dscrete verson of Laplacan operator for trangulated surfaces, so that we can use t n practce to solve related problems We are mostly
More informationSimultaneous Optimization of Berth Allocation, Quay Crane Assignment and Quay Crane Scheduling Problems in Container Terminals
Smultaneous Optmzaton of Berth Allocaton, Quay Crane Assgnment and Quay Crane Schedulng Problems n Contaner Termnals Necat Aras, Yavuz Türkoğulları, Z. Caner Taşkın, Kuban Altınel Abstract In ths work,
More informationLECTURE 9 CANONICAL CORRELATION ANALYSIS
LECURE 9 CANONICAL CORRELAION ANALYSIS Introducton he concept of canoncal correlaton arses when we want to quantfy the assocatons between two sets of varables. For example, suppose that the frst set of
More informationSolutions to exam in SF1811 Optimization, Jan 14, 2015
Solutons to exam n SF8 Optmzaton, Jan 4, 25 3 3 O------O -4 \ / \ / The network: \/ where all lnks go from left to rght. /\ / \ / \ 6 O------O -5 2 4.(a) Let x = ( x 3, x 4, x 23, x 24 ) T, where the varable
More informationThe Geometry of Logit and Probit
The Geometry of Logt and Probt Ths short note s meant as a supplement to Chapters and 3 of Spatal Models of Parlamentary Votng and the notaton and reference to fgures n the text below s to those two chapters.
More informatione - c o m p a n i o n
OPERATIONS RESEARCH http://dxdoorg/0287/opre007ec e - c o m p a n o n ONLY AVAILABLE IN ELECTRONIC FORM 202 INFORMS Electronc Companon Generalzed Quantty Competton for Multple Products and Loss of Effcency
More informationEcon107 Applied Econometrics Topic 3: Classical Model (Studenmund, Chapter 4)
I. Classcal Assumptons Econ7 Appled Econometrcs Topc 3: Classcal Model (Studenmund, Chapter 4) We have defned OLS and studed some algebrac propertes of OLS. In ths topc we wll study statstcal propertes
More informationSome modelling aspects for the Matlab implementation of MMA
Some modellng aspects for the Matlab mplementaton of MMA Krster Svanberg krlle@math.kth.se Optmzaton and Systems Theory Department of Mathematcs KTH, SE 10044 Stockholm September 2004 1. Consdered optmzaton
More informationLecture 2: Gram-Schmidt Vectors and the LLL Algorithm
NYU, Fall 2016 Lattces Mn Course Lecture 2: Gram-Schmdt Vectors and the LLL Algorthm Lecturer: Noah Stephens-Davdowtz 2.1 The Shortest Vector Problem In our last lecture, we consdered short solutons to
More informationEconomics 8105 Macroeconomic Theory Recitation 1
Economcs 8105 Macroeconomc Theory Rectaton 1 Outlne: Conor Ryan September 6th, 2016 Adapted From Anh Thu (Monca) Tran Xuan s Notes Last Updated September 20th, 2016 Dynamc Economc Envronment Arrow-Debreu
More informationNotes on Frequency Estimation in Data Streams
Notes on Frequency Estmaton n Data Streams In (one of) the data streamng model(s), the data s a sequence of arrvals a 1, a 2,..., a m of the form a j = (, v) where s the dentty of the tem and belongs to
More informationIntroduction to Vapor/Liquid Equilibrium, part 2. Raoult s Law:
CE304, Sprng 2004 Lecture 4 Introducton to Vapor/Lqud Equlbrum, part 2 Raoult s Law: The smplest model that allows us do VLE calculatons s obtaned when we assume that the vapor phase s an deal gas, and
More informationLecture 4. Instructor: Haipeng Luo
Lecture 4 Instructor: Hapeng Luo In the followng lectures, we focus on the expert problem and study more adaptve algorthms. Although Hedge s proven to be worst-case optmal, one may wonder how well t would
More informationFinding Dense Subgraphs in G(n, 1/2)
Fndng Dense Subgraphs n Gn, 1/ Atsh Das Sarma 1, Amt Deshpande, and Rav Kannan 1 Georga Insttute of Technology,atsh@cc.gatech.edu Mcrosoft Research-Bangalore,amtdesh,annan@mcrosoft.com Abstract. Fndng
More informationFeature Selection: Part 1
CSE 546: Machne Learnng Lecture 5 Feature Selecton: Part 1 Instructor: Sham Kakade 1 Regresson n the hgh dmensonal settng How do we learn when the number of features d s greater than the sample sze n?
More informationk t+1 + c t A t k t, t=0
Macro II (UC3M, MA/PhD Econ) Professor: Matthas Kredler Fnal Exam 6 May 208 You have 50 mnutes to complete the exam There are 80 ponts n total The exam has 4 pages If somethng n the queston s unclear,
More informationAppendix B. Criterion of Riemann-Stieltjes Integrability
Appendx B. Crteron of Remann-Steltes Integrablty Ths note s complementary to [R, Ch. 6] and [T, Sec. 3.5]. The man result of ths note s Theorem B.3, whch provdes the necessary and suffcent condtons for
More informationRemarks on the Properties of a Quasi-Fibonacci-like Polynomial Sequence
Remarks on the Propertes of a Quas-Fbonacc-lke Polynomal Sequence Brce Merwne LIU Brooklyn Ilan Wenschelbaum Wesleyan Unversty Abstract Consder the Quas-Fbonacc-lke Polynomal Sequence gven by F 0 = 1,
More informationPhysics 5153 Classical Mechanics. D Alembert s Principle and The Lagrangian-1
P. Guterrez Physcs 5153 Classcal Mechancs D Alembert s Prncple and The Lagrangan 1 Introducton The prncple of vrtual work provdes a method of solvng problems of statc equlbrum wthout havng to consder the
More informationLecture 5 Decoding Binary BCH Codes
Lecture 5 Decodng Bnary BCH Codes In ths class, we wll ntroduce dfferent methods for decodng BCH codes 51 Decodng the [15, 7, 5] 2 -BCH Code Consder the [15, 7, 5] 2 -code C we ntroduced n the last lecture
More informationCanonical transformations
Canoncal transformatons November 23, 2014 Recall that we have defned a symplectc transformaton to be any lnear transformaton M A B leavng the symplectc form nvarant, Ω AB M A CM B DΩ CD Coordnate transformatons,
More informationA Combinatorial Polynomial Algorithm for the Linear Arrow-Debreu Market
A Combnatoral Polynomal Algorthm for the Lnear Arrow-Debreu Market Ran Duan Max-Planck-Insttut für Informatk Saarbrücken, Germany duanran@mp-nf.mpg.de Kurt Mehlhorn Max-Planck-Insttut für Informatk Saarbrücken,
More informationLINEAR REGRESSION ANALYSIS. MODULE IX Lecture Multicollinearity
LINEAR REGRESSION ANALYSIS MODULE IX Lecture - 30 Multcollnearty Dr. Shalabh Department of Mathematcs and Statstcs Indan Insttute of Technology Kanpur 2 Remedes for multcollnearty Varous technques have
More informationTransfer Functions. Convenient representation of a linear, dynamic model. A transfer function (TF) relates one input and one output: ( ) system
Transfer Functons Convenent representaton of a lnear, dynamc model. A transfer functon (TF) relates one nput and one output: x t X s y t system Y s The followng termnology s used: x y nput output forcng
More informationInexact Newton Methods for Inverse Eigenvalue Problems
Inexact Newton Methods for Inverse Egenvalue Problems Zheng-jan Ba Abstract In ths paper, we survey some of the latest development n usng nexact Newton-lke methods for solvng nverse egenvalue problems.
More informationAn efficient algorithm for multivariate Maclaurin Newton transformation
Annales UMCS Informatca AI VIII, 2 2008) 5 14 DOI: 10.2478/v10065-008-0020-6 An effcent algorthm for multvarate Maclaurn Newton transformaton Joanna Kapusta Insttute of Mathematcs and Computer Scence,
More informationCalculation of time complexity (3%)
Problem 1. (30%) Calculaton of tme complexty (3%) Gven n ctes, usng exhaust search to see every result takes O(n!). Calculaton of tme needed to solve the problem (2%) 40 ctes:40! dfferent tours 40 add
More informationLecture 20: Lift and Project, SDP Duality. Today we will study the Lift and Project method. Then we will prove the SDP duality theorem.
prnceton u. sp 02 cos 598B: algorthms and complexty Lecture 20: Lft and Project, SDP Dualty Lecturer: Sanjeev Arora Scrbe:Yury Makarychev Today we wll study the Lft and Project method. Then we wll prove
More information