\ssr. HD28 .M414 ALFRED P. WORKING PAPER SLOAN SCHOOL OF MANAGEMENT Optimal Consumption and Portfolio Policies When Markets are Incomplete by Henri Pages WP //1883 April 1987 MASSACHUSETTS INSTITUTE OF TECHNOLOGY 50 MEMORIAL DRIVE CAMBRIDGE, MASSACHUSETTS 02139 Optimal Consumption and Portfolio Policies When Markets are Incomplete by Henri Pages WP #1883 April 1987 M.I.T. LIBRARIES Preliminary, Goniincnts Invited OPTIMAL CONSUMPTION AND PORTFOLIO POLICIES WHEN MARKETS ARE INCOMPLETE by Henri PaRrs February 1987 Abstract We eonsider a conhnntnis time versinn of the eonsnniptinn jiortfolio where an agent invests sumption and possibly his initial endnwiiK^nt final wealth the general existence cjuestion ovei- a on markets rajfital hxal horizon. Usiup; when markets tli(^ to i5rol)leni under luirertainty, maximize his utility of con- martiiif^ale apjiroarh. are flynamically incomi)lete. of the diffusion process followed by the assets prices motion that describes Ihe uncertain environment. The less is tlian tli(< i.(\. sufficient conditions and arc exjiressed primitives of the model. \\c slinw in paitirular that even when in the economy, only that which is the dimension dimension of the Brownian identical to those j^ut forth in the complete markets case information when wc address we require arc virtually entirely in terms of the the invcsinr ha^ access to the whole reflected in the securities price system sti'ictly is relevant to manufacture his liest consumption jtlan. fnr lii' nuliiniing fsoarrli a'^^istniirc I would likr to tliank (."^lii-fu Hiiniis. SInnn Srlinnl nf Maiiacmf nf and imtirnrc Inm aNn r«])prially jndo}>trd in D A^' Stroork fr"in i]\r Nfatlirinaf ire Drjinrtiii'^iit NflT. witli whnni iiiaii^ if not nil of tlif* proofs worked owf Fiii'tu'id 'iiipporf finm flio n:iiik of Fi uir*' is jiUo pr-'^tofnlh ill tlii*; iiajt'^r w rro «iirrn««;fiill\ know If^dii'^d. All ( . . . ;tt prrois r^'niain of rrnir<in rm" own. Introduction 1. Optimal consumption portfolio been stvidied work in the Merton of Karatzas undn- dynaiiiir [1909]. et al. [198G). and found one [1971], A common as a two prol)lems with mathematics feature of means tlie contributions in this consumption The Brownian motion. one first For instance can easily jioiffdliri jiolicies In' on first to solve a nnn-linpar partinl differential ecjuation, sn that Another possibility is that to invoke directly it is already emphasized, this assunqitif)n is However, when financial Ijut that An it that stochastic tlu^ Cox and Huang and Huang [1987] in securities is dynamic ])rogramming Im^cii ])ut furward This aiii)roarh takes behave Krei)s [1979], then extended by Harri-^on is that there must exist some So the proof under Cox and Huang [198G] as frictionless. hideec' no constraints on the amount clearly iniboiinded is in ))lare of and thus a very jjowerfid technicnie the dynamic jirogramming: like maitingales was Chaml)erlain [1985] advantage of scune tools develojied full martingale jirobability theory. That a nnilti])eriod seciuities market its and to check in control theory, [19SG]. (1987] aiul Tliska [1084] in i.nrtf.,lin theory, or general efjuilibnuiii the discoinited prices of One general consunqition jiortfolio ])roblem iiosed by economists. alternative aiiprdarh has recently notably. in i? not best suiterl td is ])ut will construct a control in markets are invest in each security, so that the admissible set Our conelusinn nnrt-rnw]i(irt many economists unfortunately one has this some existence theorem tnipalatable given asstimed to is iiulcrfl ojjtimal. To do markets with no transactions costs nor restrictions on short-selling money one can is it vci'v flifficult to a compactness assunqition of the set of admissible strategies. of in the last reference to ran^tnirt a control is characterization. its when an issue under general circumstances. afterwards— with the vrriBrati<<n tlimmii of control theory of the existence of a solution hinjies There seem Seconflly, the very existence inteii^st. dynamic i)rogrammin<j Ijy the use of stochastic field is therefore uncertain whether It is paper a recent inherent difficulty, especially its is most situations of practical in to solve the existence lunblcni in general in quite elaborate, even though the secin'ities price process continue to use this technique way rontinnon? time have long to derive the oi')timal ])olicy for an individual. approach. this is follow a simple geometric of ojjtimal in achievements of its latest non-negativity constraints on constnnj)tion are involved. aljove, the and )iiir(Mtainty economic? literature. Tlie traditional approach originated from the pioneering dynamic programming to be jiolirics first arbitiage free whenever is foreshadowed by Harrison and and Tliska [1981] and Duffie and Huang nu^asme under which the jirobability [1985]. The point risky s<'cuiitie? earn in average the rate of return on the riskless bond. (This would certainly be true under the original probability lieliefs if agents were risk neutral. What risk aversion.) However a ])ar with dynamic inograiiimiiig financial comjilete its markets are when the as it staiuls. tins <lyn:iini<-;illy number the change in jn'obability really does new strand relative, roiiiphlr. because Roughly of risky securities is of literature cannot it rests sjieakiiu',. ecjual to the ' on the ]>v is that it subsumes (pute considered on critical assumjition that markets are said to be dynamically diniensioii of uncertainty. The goal of tlii? i)aper i? to fill pap hy thi? dynamic programming, and more sporifirally to show how an approach is oxistrnr*^ jiroof can bo derived when Needless to say, we l)elieve that the martingale markets are dynamically incomplete. financial hoped thn martingale apjiroarli as an alternative to stochastic extriidiiip; more appropriate than the dynamic i>rogranuning one and in this ajiplicatiou, it is that this paper will contribute to the idea that the former dominates the latter, both in terms of their respective tractability and of their int\iitive appeal. Before we delve into the arcana of probability theory, we jnesent a lirief summary. This pa- per takes the continuous version of the canonical consumption portfolio problem and examines the behavior of an investor who endeavors preferred consumption plan. His prol)lem to design a best financial strategy to achieve his maximize to is his exjiected utility of most consumption over time, subject to a budget constraint which equates the agent's revenue (capital gains on assets, div- somewhat idends earnings, endowment) and expenses (consumiitidu. net trading nn stocks). To be more explicit single about good — taken oiu- framework, rerall that there to be the numeraire -available for consumjition; there the ternunal date too, all reasons we Init for ex]iositionaI confingrnt rjaini or mnf^uinptinn pinu across T a finite horiz.on is states of nature and at all A times the space of niarkrtri] in assets. income all means wdiole consuiiq)tion sjiace or. over time that the investor if one niiglif sjian that the space of jirefcrs. at Then a consumption good a decision regarding the continuous There are M risky securities, posited is riskless. are either consumed or reinvested. and Finally, the contuigent claims that can be manufactured by withdrawals from the portfolio durin? the time f/yiiajjiiVai/y inrniiiplcte just ]iresently. entin^ly generated by the capital gains is The proceeds consists of /'iin'//es is economy and a and a bond, whose instantaneous rate of return Exce]>t for his initial enflownient, the agent "s dividends on his investments stratof^y it a distribution of the to time T. trading of the existing securities from time to follow a vector diffusion juocess, liy this may be some bequest not bf)ther about will characterized is in that [n.T]. Then to say that capital markets are marketed bundles strictly is included are distributions of the consum]ition tlier(^ want to jnck but which cannot l)e in the good delivered by an apjiropriate trading strategy. We first will M call question we have to address incomplete, when i.e., literature obtains lying the space of maiketed binidles and A' when is uiidei- whirh circumstances .M does not coincide^ with the number Brownian motion which representation theorem tli.-\t M X. of risky securities generat(\= unrertainty any The It is iiiarki^ts liajipen to ty])iral tli(Mi is trnr when A/ consumption is less than sjjare so that A', .M N of the inider- dimension be shown M be dynamically so far in the ~ X martingale usinp; the attainable tlirou<rh trading strategy so that markets are dynamically compK^te and some This result ap]iioi)riate is hi this srroiid case, there are too fi-w secui'ities to si)an" becomes strictly inrlufled in A. The model considercvl etjual to the ran contiiip;ent rlaiiii in A' whole consumjition space. thi' no longer the entire Not surprisiuRly. cnnsofinpurrs in tniiip of the infoiination ptnirtuir a? well. tlicrr aro couveniciirr the infomiation set (n-alpehra) peneiate'l With as the "priee information set". information tin liy system securities jirire rlynaniirally eomjilete markets, it is For referred to turns out that the price coincides with the whole information in the eronomy. so that in particular any set process observed at the current time will he a function of the past and present realizations of the securities price process. become incomplete. This The surely a strong re(iuirement, that is relation more subtle {S stands here little T inclusions contained M C is it can hv shown (though this M becomes then a Namely, one has the strict which are outside the price information not so easy to j)rove) that is Hence oiu' first all i)rice is set. the consumption ol)servation has to bo that maximization i)rogram ha])])ens to be borrow fiom ('ox and Huang [198C) shall Their idea ]irog;iam. to the sjiace restrictefl M of is to marketed to bear martingale theory to transform already been studied by mathematirians prices in units of the riskless interest rate if a measurable, then ex])lf)de. is j)ositive The bond This their solution roncejit with regard to the end)ed the investoi's i)ioblem. wliirh l)undles. dynamir problem a ccunplex Before we do this we find is without it intec:;i-aterl financial gains tlie efpial to tlie / sum c'{f)ilt = W'(0) + that manufactures r The fhscounted in sum / financial r. The of the (discoiniteil) cajiital stains for the entire jirice this set-uji manufactures all of the initial wealth and of the For any consumjition plan gam- 7r(c*) , Jo denotes the discounted flow of consumi)tion. n"(0) the of financial gains the because convenient to normalize write accordingly .'n r* one that has liudfet constraint then stat(\s the familiar condition that the present on the risky securities where into a static and bounded from above, so that the juice of the bond can neither vanish present discounted value of we can by assumjition because we assume that the loss of generality, is Al. is above the whole eonsumjition space, and to bring in discountcii value of the stream of consumption G markets automatically marketed. maximization c T set letting Because investors have access to more information than that which the j^rice information set are market(>d. in We nor and the juice information for the vector of securities prices). jiartirular solution to the investors it M system, there are bundles in in the price Conversely however, goods C X. between we can relax by initial wealth, and the integral and dividends accriuuf: to the strategy time sjian (n.T]. of the above consnm])tion sti-eam the initial investment the only aiiwnuit is at time f) is ecjual required by the strategy that agent's maximization jirotri'am can thcicforc be foiniulated as max/? / ^i(r{t).f],lf s.t, 4 ;r(c*) to H'*(n), = n"(n) where u time at some is We now 0. and n the linear utility funrtion extend maximization jirogram this mi^xE where 7r over X stands of all as before for the X. When markets I u(r(t).t]dt in the following 4,(r' s.t. prices should one ascrilie to a stream of (ii) What ensures that a solution to the extended program, question arises from the fact that commodities have their tliat is extend w over all jirice of A', system, i.e.. determined jiy it consumption that M when ])riori such that the shadow pijre of consunqition this jiarticidar valuation, always be chosen tn it turns out that a solution lir jii'icc c expectation unrler some probability. Let then of the price it preferred as Q marketed? in A', an abundance of But c. c is it to the r>xtended when Thus, as far as the ojitimization juolihun be the this, we have to recall a result Q under with respect that it is In as least as should be clear that the two solutions it concerned, with our choice measurable by construction. jirice is r thei-c^ is is strictly concnv(\ nothing more than the securities prices themselves that the agent needs to oliserve to devisi^ his ojitimal strati^gy. enrlowed with a very rich information structure, only that which is With maximization program can ])robal)ility associated the utility function is 4> of any contingent claim can he written as optimal by assumption, so that are in fact indifferent, or even identical functiouals the jirice information set. itself in is same budget constraint and satisfies the ]irice iiieasiuable with rep])ect to the price is The new consumption plan can be shown that only the marketed any one of them. However, one candidate measurable valuation, and take the conditional exjiectation of to the price information set. addition, is included and thus marki^ted. To see ineasuralilc, from option juicing theory which states that the price its two questions: not marketed? is any, can he stictly is the (unitiue) one which is if There arl)itrage. and one could choose a of sjiecial interest to us: (0), a linear functional that extends is <f> way are incoini)Iete, the ahove emliedding procedine raises What first W ^ ) whole eonsnmption space and (i) The marketed goods fniirtional that gives the priee of relevant to manufacture his ojitimal consiuuption jilan i.-^ contained The crux Even in a in thi' capital of the argument is world markets that an investor can loose nothing in terms of his expected utility by j^rojecting the whole jiroblem onto the information generaterl by juici-s. An many given as follows. Since there are infinitely claims outside M. the investor may l)e consumi">tion spare subject to an infinite valuation which price is the consti-aints are slack. Hence c is But all indeed a solution to the valuations which give the seen as maximizing his exjx^ctiMl number in the pi'ice inf'>i'iiiation set measurable and thus maiketed works can tentatively be intuitive accoinit of jiow this Now of budget constraints iitility the contingent over the whole Simi)Iy jiiek the jiarticular and observe that the corresponding sohition exi)(~cterl utility the valuations agier on inv(^stor"s jiriccs of maximization can M In- improved ujmn only so that ])r()blem. all if c is some of constaints are saturated. Til'' dynaiiiir rost of this jiajirr ronsumption markets. Wliat is is orKanizcd as follow?. jiortfolio prolilcm new about onr rrs\unrfl aiui is ajijjroarli In scrtinn is that 2. siiiijiiy flir framework fraditioiial cxtriKlcd to allow for iuroniplcte we rhararterizc the set of marketed consuinption biuidles entirely in terms of the orip^inal ])rol)ahility beliefs lather tlian relatively to reference mcasur(\ The specifying the null sets. nice fact about this The other is that beliefs do jirovide something features of the morlel are standard. martingale representation theorem to show that This is all jirice some arbitrary more than merely In section 3 measurable juocesses are pose the investor's maximization in-oblem and directly extend the The last section we use the in fact important since otherwise our existence proof would not go through. In section our economy. for a Cox and Huang 4, marketed. we formally [198G] results to concludes witli a very simjile example carried out for the class of constant relative risk aversion utility functions. The Model 2. 111 Before we can define the dynamic maximization forniuiated. be useful to characterize the economy and the trading securities, strategies terms of in its process will be witli diffiisiou price faced by an investor, i)r()l)lem it will information structure, the price system of its agents can afford. its The Information Structure 2.1. common Agents share a future information wliich is revealed continuously thiough time They have no uncertainty about the present surprise. gradually resolved over time; at is < = T or before time Taken number = Ti The following mathematical model as a primitive ate a jirobaliility space (Q. positive real we take t. (t[)J'(,'') t < : ,« < to <] We 7t- use the restriction of Z to [0. f : ] 0) t PM X n there is no about the their uncertainty We may think can occur at gives these notions a j)recise meaning. 7 P) . and a time span T]. where [O. T Brownian motion defined on liy 11' uji to time is it some where Furthermore t. to denote the rr-field generated by the progressively X /? and — as the collection of events that the rr-algebi'a generated lie (A funrtion Z measuiable processes for each = Let {W{t). Ti.P) be an N-dimensional 7 = assumed that is {t they learn the true state of nature. of the information structure at an interniediate time it time this section a iiiodrl of scciiritirp inaikots in fontiiiuons is — 17 > /? is said to be 7t-prnf^rrsfiivrly nicasrira/»/e X 7)-measurable. where Bi Bi is the rr-field if of Borel subsets of \0.t].) All the processes to processes appear proiiosition addresses this to be interpreted as < ,' < f]. be progressively measurable, hi our framewoik, the ^/-adapted Chung and Williams (cf. Proposition will 2.1. If a ){' in ratliiM' our As.'JiijHr- prnrrss Proof. See Stroock and is are [1983], p. 8) /^-]irogressiv(~Iy pednntir measural)ility <iui-stioii. measural)le. The following whin'e the generic A' ]irocess is setuj). A'f .w) ftilnptrij /.<; ti< ri^lif rontiiiunu:^ 7i . thrill it i.s Varadhan (1979), Exercise ,t/s(i f<'i rn<-]i w E H rwJ 'IrBiir 7i — oIX{k) : J)li<f::irssivrly njea.siira/i/e. l.G.G. I Another whose imjiortaiit descrijition is infniiiiation given next. stiuctuie is that genernteil by the secui-ifies ]irice system, The 2.2. Wr ron?i(lcr liy rn = System Securities Price 0, ?\ market with A/ seruritir? The market A/ 1 and no transaction costs (hke dividends at rate (n(0 pays no divi(h'nd and riskless, and r(t) and r(.T. f ) /? : X [0. is T — sells at T?"*" > ] precisely, is it = ^^p{Jo D{f) foi- i{S{.'').s)d.'^ / = they : P^' X h(S{.o].s)ds+ S({))^ T \(\ I of full rank for r and all \n(x.t) n is x.y E P a matrix, then that S{-.ui) The it au'l all is \rr\ (. We G [n. - V T and ] : \h{x.t) sub-rr-alpeliia generated by all ])ri('es is < chosen to \\c jiriee terms of the current and possibly We now use Ci{i\ to (locally) that both in{f) t) R be bounded from economy are as.'sumed jirocess. More fl(>iiote {a.s..P) (1) 7^' K \V< We liavc usivi the following notation: nn measui-able jiast the l)y is a countable p. 94). all are adajited to i)iic(-s show of y^^'^'''" Jiieasiirahie 7[. Let if it continuous] with respect at time tinu'.': / at can be expressed or before tiiiu^ i. that the best consumiition ])lan for the ag(-nt Thus, his entire stiate<:y ])rice system. t Sit] 4 lie riplit number if Tliese conditions imply . said to is 5 since Jtlnrrs!^ at) = Since Jf measurable variable values of the ^:\)]i'^ '' ) Stroock [198G]. a jnocess ada])t('(l jirice will (rf i/|. peiKMal strictly contained in the orip'inal one. the following definition: Stroock (1979). Exercise 1.5.C) ])e is P^' X [0,Tl —^ : Vte[O.T]. < K\x - h{y.i)\ d<Mioted in is Mathematically, we have that any can always - measiue zero as a fuii'tion of the ]incr ])roc(-ss evaluated at (cf, — pays assume that denotes the Hilbert-Schmirlt norm (Ti'ace progressively measurable (or. etjnivalently. 7f Asrents in our risky, P^' X [0,7] -> n^'""^' are Borel measurable, and ?f)me constant follows that this second information set to We assume rT(.S'(..).,9),/H'(.^) I shall butlierinore n[y.i)\ unifiue uj) to a set of P- us adopt once and for is f bond i? Jo -^ 7?^' and n{x.t) ] indexed satisfying; \n{t.O)\ V\h{t.Q)\ all — the on the law of the price all afcree , a ripht continuous, /f-pro^ressiveiy measurable, (a.s.,P) is .'n where h{j.t) for A/ I, 2 Security ''(') '''} away from zero lielow in the sense that Jo is t — Borel measuralile. Finally, we rerjuire that r{x. assumed that S E P' continuous diffusion process rT{x.f) time < N no constraint on short sclhnp Security n for S„(t) ex-dividends. t bounded above and have rational expectations 5(04 time i? can be respectively written as /„(.'>(0-0 and r{S(t),t) with i„{x.t) above. Thus, D{t) to will lie frictimile?? in tliat there "^^ whore A/ lonsj-livcd scnuitirs traclrd. 1 l)rokera;^e fees for instance). ^^"^ '^^'^^ 4- / li.o]'!.^ will bi- defined entii-ely in . and rewrite (1) in unit? of the O-th sernrity. = roirespondiiiRly G'{t) J^ /*(.«) (/.i. G t (a..«., To be a reasonable probability measure Q Q assumption on n aiul market model seeuritie? we need two before that liut 2.1. ]rt n the element o{ of to /? honiomorphisms from P' Lemma }ir Lrt n Next, suppose a. G S'^{P' N ] N X and Range , dnnotr ), S (P' tliFit rr measure when mFiitinf^a.lr the set of symmetric non-negative definite of P it E such that We Harrison and Kreps [1979]). (cf. there must exist a in the arliitraf^e free sense, [{dQ /dP)^] an f^ntififyiuf^ G Hom{P' . We technical lemmas. Horn (/?' some qualifying P } = use S~^{n , /?' = na unci that n n„. Then n. to denote the set R /?' ). , onto nangr(a). and n >—» a"^ /s a //leastiraWe Then = n ) denote to ) G Hom(7?' to denote the range of a w„ flm nytliognnnl jirojrrtidn of l<y We satisfies those three conditions. real matrices. {rr) < oo and say that a probabil- will the absolutely continuous inartingalr measure? under all h. ' P). absolutely rontinnous with respect to momentarily will elicit ^ Jo an ahsolntply continuous is Z?(.f) Jn that (G'(t), Ti.Q) be a martingale ity i{t)/D(t) and f h'{S{s).D{s),s)d!^+ f n'{S(s),D{K)..9)d'W(s) S'(0)+ [0,r], = i'(t) Ito's fnniiul.i imi)lies that Jo for all S{t)/B{t). B(s) Jo = + S'{t) = DefiniiiR S'(f) 7?aiifp(a) = Range function {rr ) and (T(n'^n) + n'^. Proof Stroork We G Pio(\f. II1.2..S. introduce the notation Lemma rri] lemma (19801. 2.2. } Define h , anJ = hituitivcly. . G Hom{P^' P^' Let n Panf^e(rr rr m-rtj (n = a)'^ »; Then k^. is simj)ly tlir inverse of Then there ). n^t] fni a// rr rr G for /?' M<'Vei'\'<'V any n exists a uni(]ue G r/ P' rr i—> is rr rr ojierating on G Eom{P^^ .P^' its ) range. such that a inensural>le function of there exist? some ( E P' such that ^oV = ^^- Thxis rrt] by lemma 2.1 = (n rr)'^ Null (rr) = t„t„( = = rT{n rr) rr 7r„t] follows from thr fart that fnr two {n wliei-e ^ 7r„T{ G Range (rr''") hi aflditjnn. rrrrr; The uniqueness n rr i? - the null sjinrr of rr.rr G (Null rr]^ H The iiicasui-al )ilit n)ri rr = n^n^i] G Hnm Null f^) = K„ti p-^') (7?''^ , = and for all t] G P^ . 0. y follows alsii diivrtly rr from lemma 2 1. Remark. Whrn hccomrs Wo rrrr = a in < rr\S(if) 7^/, thr hf-rJimriisinuHl iijrntity to find ])o?iti(>ii s < (i.e., t] assumption on Assumption X = onto, n^ matrix, sn that above relation tlir thr absolutely rontinunii? niaitinj^ale measures. of whirh a(lai)trd is We to the a-field All our results will be subject to the following price measurable). and rr all many such measures, one that there are infinitely (}ualifyinf; 7?'^ is >]. now arr shall see T,^ = r] n h. 2.1. Lrt K,(x,t) = this assumption is — —rr{j.l){l)(x,t) Then k r{x.t)j). bounded js for all {x.t) G io,rj. Note that Brownian motion a? in the model Proposition 2.2. Suppose Nikodym derivative R = = f, satisfied in particular — Q when 5 follows a multii)licative geometric considered by Mertnn [1971]. first Thm an al^sohitrly mntinunus niaitini^alr niras[irr. is its Radon- dQ v{T) = is equal to tiN where cxp I I k{S{.]. dWi.") ..) - 1/2 |«(.S'(,s), I ,)f d." I and N = N{T) = exp / r{S{s). dW{s) - 1/2 f .0 |r(,9(,0. .')|^ d.o \ . I fnr snmr In this Prnnf, R{f] Dy E ,>{t) Null rrif) siirh theorem = E /n^i ''(.*'(''), ')|^'/.'' I .7J ) is a inartiii<r^)r nnrler the martins^ale rejiresentation thmi-t^in (rf 4) there exists < as well as in tlir scnjuel. the notation 7? ( E^ that a G PML'^iX X T; /?"•') P OO. y will .t whirh can lie stand for the inner product taken to for instanr(- Kunitrv lie continuous (a..'.. P). and Watanabr (19C7). Section such that R{1) = 1 + ^f') / dWi.-^), (3) Jf) where PML Lebespue (cf. [X X P) denotes the But iiieasurr. Dellacherie and d{R(t)(r{f)) Hence liy {f^'{t). Ji Meyer JJ-iiin^ressively . [1982], Q]\< R(t)dCr'{t) = R{t)h'(S{i).t)dt arpumnif martingale Lemma = a standarrl ^ VII 48) L (A X P] and A (R(t](V(i]. J,. P] measurable jirocesses if and only if By Itns formula of is is a martinfrale then, we have + (r{t)dR{f) + dR{t)dn'{t) it -¥ must R(t)<7'{S{t).t]dM'it) 4 (r(()dR{t) lie that R{t)h'{S{t).t) -+n'{S(t).t)n{1) 10 = + the n'(.^(t).t)a{t)dt. (cf. Jacod (1979), Proposition 1.43), or cqnivalently R{t)(h{f)-r(t)S(t)) where rr(t)n(t) = 0. for brevity h{t) stand? for h(S{().t) anri similarly for other varialiles. To lemma solve the last equation, note that by r{t)S{t)) e Range a''"{«) n(t)K(t)) = such that -(h{t) some i/(<) {R{t)K{t) + G Null rr(<) nrA/L^(A x Now dW{t]. i'(0) formula to logr7(/) shows that let = r(t)S(t)) rT{t)K.{t). = = solve dt](t) —a(t){h{t) then follows that (T(a{t) It — - 1/(0 a fuid dR{t) d)V(t). K{t) T]{t) we in (3), An statement of the theorem with as in the again K{S(t).() 2 2 + R(t) K(t) P). Substituting for t}(t) t]{t) is Note that by lemma in [0,r]. - = unique k(() 2.2 there exists a so that the general solution for a disjilays the following orthogonal decomposition n{t) for + a{t) dlVit) = application of Ito's T irjilaced by any t meas\u'able and that by the jirogi-essiveiy is = boTUidedness assumption of k we have K{S{t).t]\ < dt oo; (4) /: hence the integral defining Now put iV(() = t]{f) /?(0 We . well-defined is have (rf. Stroock [198C], Section to sjiow that ti{f) is II. 3). strictly positive {n.s., X X P). For this it n(t) suffices to demonstrate that rT fT / K{S(t].t) Jo But dM'it) - {n.f.. P] (4) implies s(t).t)dW(t: cf. > -oo 1/2 / \K{S{t).t)f dt Jo Lijister ('I..".. 7^) and Shiryayev (1977). Theorem By ItnV formula dN(t.] = R{f) This, toj^ether with (4). ensures that t]{T) IRlt) -^rr'^nU) + > + -) = - dW dM' - N 1 , i/2^{dn{t)f - t]^{f) riii) K oo: again. dRU) — i- - = {N 7.1. < -^j-'ir^if) n^(t) tr(t] 1' t: dW + N\K\'dt - K + ( N JMf) dt dt • Thus we can write dN{t) since r and k are orthogonal. The = N{t)i''{t) dWit). exi)ression for A'(') can in turn ajiplied to log A'^(/). We now jji'ocerd to show tjiat r;(/) is jince mea<=urab 11 (5) bi^ derived from It6"s formula Proposition Proof. By The 2.3. process^ ri{t) is prirr inrHsurahlc. proposition 2.2. / logr;(t)= dW - »c(.S(..),.s) LemiiiH 2.2 phnw? that the srroiid intrpral of the approximated sum as the K{S(t)j) hand riplit it |^/,s. ;.) measurable. $\(]c is ])riro Consider the of i)rice measvu'aljle functions.) E Range definition K{S{t).t) 1/2 I |/c(5(,.). integral. first (It can be Since by (<), dWU) = dWit) = K(S{t)j)n„Td'W{t). 7r„T«.(5(0,<) But = n^(<TrT^]^ndW(t) = a^ (rrrr^ )^ n„jdM'(t) and since S(t) + I i(S(.'>). s) df — h(S (,<<), I i^) dit is (^dS{t) + ,{S(t).t.]dt - h(S (t) . t) dt^, certainly price measurable, the result follows. Jo Jo I From now on be denoted by Q. sets. It the jirnbability Note that since remains to check that i] whose Raflon-Nikodym r; is strictly positive, T-square integrable. is diu'ivative {(i..<> . P P). t] i? and jirice Q measurable will have the same null In fact, a stronger result holds imder our assumptions. Lemma 2.3. Prnnf. Let 7 ajci r;~' arc t] > 0. Tj" We jjj L'^{P] fnr any q > have =cxpl I qr -/ir (V = exp - '/H' 1/2 I 7|K|'r;/| - I a martingale Since k is bounded (say by denotes the norm of finitr. rj K). E'^[if] in L'^(P) ^ ly? 1 ^li^ jy\^ dt. I P under < max| -xp '/' expf "^ ^^" The same argument works T A'' j for r;^ \ <n \\i]\\^,r < 00 where ||r?||,,r . I Wc turn now our attenti'in to tradiiif' strategies. 12 2.3. The Trading Strategies The ronsumjition Fpacc for the ngcnt^ is PML^X X P) X L^(r) = L^\O.T] where PML'^{\ X P) A alt) t. ^,,,(0 are the bond B{t) and an is x P) x L^[^.7,P). M+ 1-vector process ('>,") = {(it(0. numl^er of shares of the 0-th and the m-tli Thus, the vahie of the of the U.PM.X the space of con^umptinn rafe processes and L^{P) the space of final wenlfh. ip trading strategy and X stratep;y (a. {fi,„{t)\ = rti at /?) time M) I W{t) = 1. , seciu'ity. resj^ectively, Af}, where held at time shares of the risky securities S{t). or + a{t)B{t) For a progressively measiualile process ^ ^ = that nf the portfoho consisting; of q(() shares is ( '',ii('))- E /?' (!(t)- S{t). we . make use constantly will of the following notation: where as already mentionned ||()||,,,r t^ia. P) = Note that C^(G P) . A (i). (ii). is |/? \r{t]\'dt) / v-r r-r denotes the L''{P)- e PM{X X Accordinf:;ly noi'iii. let lll^^^llkr < oo}. P: R^') entirely defined in terms of the prol)abi]ity P. trading strategy is said tn hr .Tifniissi/i/e if fteC'(rr.P) There (7) phn (r.W) G PML^{X x exists a rnnfiuinj:fion a(t}D(t) + 0{t] S[t) + P) x L'^{P] such tliat r {.'<), is I Jo (8) = a(n)i?(0) + ^(0) S(0) + + (^n(,'^)dD(s) 0(s} r/(7(,9 I for all t e [0. T] {a..o..P) and ir The consumption plan [rt.fl.c.W) is is = a(T)D(T) + (c.lV) of (S) and (9) S(T) 0(T) is said to he a srH-finnnrinf; sfrafep-j said to (rf. ]io (a.s.. finnurrij liy (n.O) anrl the quadruple Cox and Huang just the natural intej^ratrd budsret constraint. Fiisf <i{t)D(t\ W(^altli valui'd at the current ])rice of the existing securiti(-s. of consuni])tion from D to t. Of coiuse rt(f))Z?(()) + 13 P) ^(0) S{i^] + "(/) The = W [198G]. .S'(/) = integral /„ (i)) is ji. W 5). Condition [t] is ((.•>] fh (ii) the financial is the stream the initial wealth that the . a^cnt invrst? at tiinf' = t The qnanfity 0. + J^(n(.*) dD{.'') capital gain? of the portfolio (n. 0) raniod from to the final wealth wliirh ran he viow(-fl a? a l)e(inrst. is W wliile t. ^ a{T)D{T) + Condititm strategy depends on the state only through the information availalilc constraint — and that J^ ({(''{/>) fl(,i) is can bo interpreted as the dG(.'^)) 0{.<>) an informational time at that a well-defiiied stochastic integral, as is the trading that iinplir? (i) = W{T) 0{T)- S(T) shown in the next lemma. Lemma 2.4. Let Thru satisfy (7). L E an L'(Q)-j/iarfuifa/r for any q is Ptoof lit 4- h' — rr{h = Define X{t) dW — n' = —k rS) nIC' is — + rS)dt we have hoiiiidrfi, -ft . We (.>]. r/lj'j. rr I By [1.2). d(r /o'^i") da'{s) (i(s) h The first pmve stochastic integi'al — rS = —nn. and {h^rS)dt = - X{T] G that is .1 n" the Cauchy-Schwar?, inefpiality and the fart that ^ and sum thus the term tlie first I L'^iP). of By (2), dC = two terms. Since is Kdf. ft D nvv lioinided, < h" L'^{P)-nonn its is such that < K dt ' (t' K for some constants A' rr So X{T) 7 e € L^(r) By I>rt ^ stochastic lie D • '/^"(.«). and lemma T,.Q] is fact u/ijf.s B at tiiiir f oo. < oo. r 2.3 tins implies that A'(r) of the nnriii, we {>i . . r \^' . see that ] ^ G [1.2 By the . is G L''(Q) for any ). linearity of the a linear sjiace. the juiec-s in units nf the O-th seenrity (the Ixnid), all about strategies of n^d an L''((V)-martin'Ta!r fnr any 7 (cf. ("ox and Huang Lemma 2.5. Suj^pnsr (n.ft.r.W) G ^ and H r' {t) = r{t)/D(t) drnntc tlir nnrnializrd wraith {r.,W) in K 12. and the suli-additivity convenient tn normalize well-known < II denot<' the sjiare consistinp^ of tlie qiiadiMijilrs intefriv'il < 2.r ?.r H.'.ldrrs inociiuvlity (1.2) so that (/„'^(.') a^O 7.r ./o Fur the s(>cnnd term and K' ./n Kdt e 2.r .(- ..),/.« + \V'{t) = \V{t]/B(t] ami rDn^niujitinn ir 14 will record a (198G). rro])osition 3.2). is EQ We It T, -^"'•(0. at = n{t] tiitir i + 0{t] S'{t) and Then the vahir of fl.cW) E ^- Proof. Lot (n. - dW'it) But + d(n(t) and + + S'{t) dfl(t) S' {t)r{t) sinre dfl{t) = S'(t)] ^(<) dW'(t) = = dt de{t) we + da{t) rf^(/) + S'{t) • + dS'it) 0(t) dS' dO(t) {t) D(t) one find? l)y + idS'(t) + S'(t)r(t) dt) = c'(t) dt 0(t.) /*(() dt get (dS'(t) ()(t) that iini)liop and dividing through differentiating (8) dn(t] fonimlH Ito's + - i'(t) dt) = c'(t) dt dG'(t) 0(t) - c'(t) dt. Eqnivalently, W(t)+ for ail under t € Q i^.r). (a .«., /r*(.s)f/,'. = + [ 0{.^)da'(s) iy*(n) P). Writing the al)ove equality at with resjiert to we 7J. = T and / taking eonditional expectations onre we have taken get the desired result (10) and leinnra (9) 2.4 into areount. 3. Marketed goods and the price information marketed consuinjition In general, a to notier that in the defiiiitiriu of a strategy ineasiu'abie with respect to T," information in the Tlii? . need not he ]>lan (r. \V) economy. However, nowhere we ass^imed that natural is if tlie if we insist that wage income flow measural)Ie consumjition bundle (c.ir) can be manufactun-d This result tools is aimed 1-ie at showing that the at f = set of marketed bundles a by-in-ndnrt, wi' will also price measurable. Writing (10) T, one sees that, foi- is liy is t]ic r'(t)dt I suffices to be progressively agents have access to the whole itself i)rice measnral:)le, any price an ajiiuojiriate trading strategy. 4. we present the hi this section sufficiently large to contain that k''' it inartinp;ale mi'asure all Q is the price the only in section 4. a strategy, = W; + .'n W^ is had this, This certainly restricts the chnice of solutions considered W+ wdiere y essential for the existence proof develojied in section measurable bundle?. As one to measurable. To see jiriee ()(t) drr'(t). I ./n the initial wealth in units the f)f 0-tli security held by the agent. This prompts the following definition; M = Ixit.w) By lemma th(^ ])rice 2.4. : X = A'n we know that it measurable functions consumiition goods will + is <>{ then follow / ^(0 ' dn'{t) for some a subsjiace n{ L'^(P) L'^{P). The relation easily. 15 We X,, G /? and will first between ]ii'ice fl G show £-('.'. that M T) contains all measurable and marketed Lemma 3.1. Fiupposn constant (^ surh X € L^iP) and H X, = E'^\X\T,\ € t T], (0. Tlmn there exists a t}iat < \\X,\\2.r From the Bayrs Pronf. f«i a// rule Jarod (1979). Lenuiia for iustanrc (rf. WXlU.p. (' 7.9) and the fart that r]{t) is adapted X, = -^E^inx E^\-^x = 7,] I But — ;;]. I n(t] nit) rT = exp( K / dW - 1/2 / Kl, tl{t) dt] \k\'^ ./( •• ' so that (-^) = By exactly some K tlie same argument exp{ f 2k - f/H' • / |K|^/f} exp / 2 E a? in leiiiiiia 2.3, (i]/t]{t)] Thu? hy the (laufliy-Srhwarz majoriz-ing k. < i?^ (r;/r,(())' < r or. takinp: < ,(r-OA' < T, ,r/v- ^ ^ f^^ iiuMjuality 1/2 |A',| |K|'fi«. ll/2 /?' 7, X' r, l'/2 A'^ e:'^ r, norm?. \X,\\2.r <('\\X\\2.i Lemma For each 3.2. X G L^lf) thnr is Xq G a luiiqur 7? and a imiqitr px € PML'^{\ X T; T?''^') such that A' farf J/i An = E^\X] awl tlinr = A-o4 / px{t)d]r{t) is a C < oo sudi III III Ilk.v|||2.r Pn^nf Su]>pose that A„ G By is p P aii<i p (ii.x.. To jirr)ve A X P). /'' ^ S ' . a rontinurtus martiiifrale and so A'd = tliat vll liA||2,rII G PML^(X X P P^ the Caurhy-Srhwarz inequality |||H||i.r^ Thi.= jiroves the = < /„ (a....P). ) are .^ucli ||'7||2.r|||/'|||2.r. i'" p{t) dM''{t) = uniqueness. the existenre. consider S{.>']..9)d.<' Jn 16 that A'n (a,, 0. + j^ p(i) + /p%(0 Henre dM''(t) = '/li'*(0. 7,, |||/'|||i.g = 0. g) and so r the Cameron-Martin-Girsanov throiciii, W*(t) By law Q = r o W'^ Q = Li = Q Q. But i/^Yli-i III.4.C). Thus But (1,2), so if if is to the - = j \,(S{s)...)\' ds] 1/2 flit) the iniiqne sohition to the martingale problem associated with the operator is -ajj Q Brownian motion with respect itself a defined by ^=exp^.(5( [j k(S{s),.) dW so is ~ extremal for initiating froju 7,ero at ^""^^ Yli=i '^ig^ = < "U^*. e L^(r) then by Holder's inequality and lemma Xi = E^[X\T,] we have that (A',. Ti.Q) is an //'(Q)A" representation theorem Jacod (1979), Theorem (cf. Stroock [1986], Theorem (rf. 11.2 X 2.3. By martingale. and Corollary X, = E'^{X]+ [ Pxit) G L'(Q) for any q G the martingale 11.4), dW'(t) •'0 where px G PML'^iX X Q). It follows rl + A,+, rl + /< ti -x,= dW rx = therefore IrxT' <nv rx J I r 1+^ - rx K llfi I (if 2,r 7.r rf+i " X( ^f+A + A \rx\o I i+f < Xi^f — + AS 1/2 A'/ Irxl'd.f' 12. ('hoosf- f > 2.r such that AiS^^' < 1/2 and assume that 2,,„^l/2 < \rx\'<i 2 |||/'.Y|||2.r — Xi Xi^,^ lemma 3.1. Finally, choose ^ > 1 CO- Then < AC 2.r 2.r 1)y < so that \/k < Tliei /> 1/2 rx 2.r < Irxl'd.^) /* IG/rC' 2.r 2.r or rx where C' < ('• A 7.r (12) 2.r = 4C^/k. For the general rase, we aiiiirnxiinatr A' liy bninided fnurtions the Leiiesgne dominatrrl cnuvergenrr theorem. A''"' 17 — X ' in L'^{P). A''"' = Now A'lin.nld A'|). let From as Ix'fore A''"' = + Xj"' dW /o^/'.y'"' X anrl = Xo + dW J^r.x he roproscutations for X'"' and tlir iiiartiu!^al<- resppctivcly Tlir Caiirhy-Schwarz inequality iniplirs that |||/'a'"'|||2 BurkhnldrrV inrqiiality in turn impiiop that D stant Since X^"' bonuded, is |||/'.\-'"'||kQ follows that it < D q Ik~'ll2,c?- xl"^\\i_Q for some con- III^.V*"'|||4 - ||X'"' We may now Hb.r < CO. |||/'.\'' — r X, apply (12) to get _ j^i") <(;' ..v'"'-..v""' x^"'^ 2.r 2,r when n and any X m — 6 L^(P) by Thus oo. converges in the ^.\ • n'^nn |||2.r ||| p\ and we obtain t^o (12) for n -» oo. letting I The Lemma t e lemma following last Let 3.3. X e L'(r). X Tltni 6 M M <=^ and the PxU) 1 null space of Null{rj(t)) cr. P) (a.,«.. for cacL \0.T]. Xe rwof. Say .Y = PML and = (i(r{t) orthogonal rr*"'"/i_v(f). X= Then flit] + / = zero. rr''^ + / 1 Null(r7(()). a''^ pxit) rr'^(t)0(i) < •^'^ < ||k''^''il|2.r oo. = («..»., 7r^jpx{t) fl{t) = P) for each ^^inre pxi^) / G is dW'(t). Tlius px = px(t) rT'{t).lM''(t) = [O.T], and t*"""^ G 0(t) = let € Range rT"^(0- Hence Ao+ / .'n ^(0 dG'(t). ./n Ijounded lik^'^llkr some constant A'n of n. .'n D = ./o dW'it) = A'n+ / A:o+ / n'^iDflit) Moreover, since a' ,nV'(t) (}{f) |||rT*"'^fl|||2,r tjio null sjiace tf> Conversely, suppose px{t] <j''^ A'n .'n bounded away from is is M. Then + I Hf) .'0 A'o B But since for between gives the relation K by < /^llk'^^llkr = A'lllrvllkr lemma 3 2. Thus A' < /^^^' i|A||2.r < oo. G MI Returning to our main concern, wn can iinw state the fnjlnwing Proposition 3.1. rrnn{. We n'idlV -4 If X E L^iP] '^ j'lirr ;neasiira/'/e. have seen that (H'V 7,.Q) K dt) = n' d'W' so that C is is a fiien A' lesult: E M. Brnwniau motion. But by (2). - = h' dt + n' dW = a ri^ht continuous. TJ-P'ogressively measural)Ie and (n.n..r) continuous process satisfying (^[t] d(V(i] ,S'-(n) 4 I n'iDdirH) .In 18 in .. .P) . r Since from L, Q and are eqnivalont. thosr statmirnt? hold also uiidoi- Q. zero, heiire = ifrV* On Q aud ' i? from 5*(0) the other hand, A', if = at = < Q Tims 0. E'^\X\Jf\. (X,. T,^,Q) extremal for H' is an L'(Q)-martingale for any q is Hence by Jacods martingale representation theorem and the X We 1^' sums If < we define A'' = '"*("') d." /^ + W we know . / r'it)df = Xn+ lemma by define \V'(t) in accordance with = lemma l)y 3.3, t = T shows that IV = Finally gives H'*(0) W'[t) + A'd- / we 2.5 W+ E"^ = \V(T] Al- is is m I price measurable so that (9) is is also satisfied + \V dK c / 7, satisfied. On the other ft. r.W) G conclude tins sectinn by Proposition 3.2. /?(/) = rj(?) if we »"(o)+ 7, hi t) -dC'lt). cIkjosc rt{t) so that H''"(t) = a{t) +f)(t] S' (t), .W. jn'ovinrr a icsult of ).<; hand the choice Jo then clear that It is as d(r{t). r*(.')f/.' / price see that r'[!^)d then the corresponding {n. We G G L^{P) and tliat A' n so that (10) X approximate the integral defining X). Therefore we have (13) which wc rewrite to choice price measurable (13) Wit] = The is apply proj)Opitiou 3.1 to the case where the consumption good (c.W) W+ Now X (1,2). Jn orthogonal to the null space of n. so that i? now can measuralile. (use ^ fart that G = At = An + Jo But px = away l)ouiidcd is to the maitingalo i)rol)Iein associated with the operator tlir niiiqiir polntioii iiaitiatiug B But thr unii]'!/^ indejjcndant niterest. jnjrr mea.-^uj-n/Je ,T/i.-;o/ute/_v r<intiimnuA niHitingHlc mea.'Jine. Proof. Recall from (5) that dN, r(0 0. since j^ /' {n.s..r). and r: lie in We have = /'(f) N, some dli'it) for i'(f] dM"{t) = r{t) i'{1) {dM'lt] orthogonal subsjiaces. so that N(t) is a - G (NulU)-'-. and suppose that t:{i] dt) = rlt) dVJH) = dMt) martingale mider both P and Q. In particular. A' = 1+ / N(t]ry{t) dM''{t) Jo with ^.v(/) resjiect to = tlir i'{t] G Null {n{t)). price ])roces? By lemma and the 3.3 then. A' result follows. 19 ^ M Therefore. A' is not measurable with 4. Optimization under incomplete markets wr In this section finally address the maximization program fared a utility funrtidu on ronsuniiitinn lates final \' wealth /?+ : a strategy {n, 0, — » r,W) U {— oo}. /? to maximize i; : x 7?^ His initial wealth expected his problem a(0)5(0) + The problem H''(0). is on to find V{W\ -f S{()] fl(0) < \V(0) [14 Cnx and Huang |198G]. we eml)ed this dynamic problem in the following static variational E' SUJ) rML\{\ anrl a utility function (n..^..r). that there exists a snlntjnn to (14), as in (c.w) e endowed witb ..'0 W>{) To prnve U { — 00} denoted hy is c{t]A)dt (nJ.c.W) e M s.t. /? is utility E sup — [('.r] an agent. He liy u[r(t).i),it / + V{W] X r) X L\[r + t]di s.t. H'* (15) .'n where Q is the jirobability defiiieri in section 2.2. By lemma constraint of (14) (15) can i)e We it = / it is immediate that satisfies also tliat of (15). associated with an [n. . c .\V ) So G M we onc(- a trading strategy satisfies the biidget if sr>lution (cH') of same budget constraint, we arc done. liave jjroved tliat satisfying the any record the following useful lemma. Lemma ])lnrrRfi with 2 5 ap])lird 4.1. Y(t) Lrt X{i) hr Sljr/] th;it pi Thru juniitivr pinrrsfi fni ryrry Tf -stcij<j<inf^ an T/' •j>rogrrssivrly niea.si!ra/)/e T tillir' E"^ A'r/,-<: tlirir rxistfi = 7} (IG) y,j,. Afr)renver, E'^ I X(i],U J Proof Dellacherie and M^'ver a rlefinition However, we will E^[X{t)\7,^\ = y{f]<U / [1982]. tln-orcnis VI. 43 and 57. ourselves to the case where Thr process Y [i] is and the Chung and Williams call(><l th-- 20 fact tliat all oiitional jirocesses [1983], Section 3.4). (^huug and Williams [1983], Section of stojiiiing times, sec for instanrr ie,'--tiict Y{i]. £•'-' .'0 are necessarily progressively measuinblc (rf For = l\ r = ^ A T, so that (IG) o]>tional luojection of A'(0- 1.7. becomes simply Now of c W) lot (r, lemma as in c(t)/D(t)). a solution of (15), define lie 4.1. Consider we can rewrite it first W = |^) and £^*^[iy the budget ronstraint (15). the oi)tional projection let r l)e By lemma 4.1 (choose X{t) = as T E^ where we have used the now (c,H') initial cost is at W'(0). fact that the optional projection of r* most H'*(0). It remains to check that is the discoxinted value of we know that price measurable so that from section 3, is By Jensens < + W' I c'{t)dt Jo at least as is it But marketed and that is it c. good its as the original (c.W''). inequality, u{i{t).t)>E%{r{l).t)\ r,'] and similarly V(W) > E'^lViW) I Tr Thu Ht).t)>ii E'^l + V(\V + V[W] u{r{t).t)dt .•'0 V(W 1 E' by proposition since r; = 2.4 (choose A' U.{r{t].t)df + —ti{r{t).t] .'^-measurable (proposition 2.3) is V{\V + E"^ by lemma 4.1 again. = T n{c{t).()df E' + V{\V ..'n ThuF, tuider assumjitinn two problem? yield the solution concave, is onr- mrasurablr in fart ecjuivalent solutions. unique so that may have good (c.H) where any so]ntif)n of (15) 2.1. W it will \\V (]iro]iosition 2.3), (c,H') is jiric(^ than one sfilutmn 'y\r](t) \ mafchcd by : ^ < ' < a solution of (14), so that the the utility function u automatirally be of course luori- = E When is T]\ When measiu'abje ("onsidcr for instance and similarly certainly marki-ted, A for verification in the one abov(^ shows that (r,n') satisH'S the budgif (rmstraint anci that 21 strictly is it tlir is all in c, not strictly consiunjition Sinre q(t) r. is t; concave is jirice resjiect similar to at least as good as the oriRiiial (c.W). Titility Wr may and rhoosr the tlm? have two (lirrr, nu^asnrHliIr) on inost ronvrni'^iif df^iiendinK Tlie riear advantage of (15) that is under whirh there exists a solution to (15) can he found commodity is either a where there agent's discount rate utility fimctions of jiractical interest finite inniiiont luider our (oj). cit.. theorems 2 = is Suj>jui<ir Sufipnsr further thnt thrrr Hsymptofirnlly fnr aowo rroof. Cox and Huang Thr ex/sf.s > [198C], last jn'ojuisitinn a sjiecial case. 2.3. : G /; >—+ 7?"*^ t; and When when the the utility Note that Theorem > 0. price this last condition r]{t) is has a certain always satisfied For the sake of conijileteness, we report two sets of /?"*" (0. ]] .surh Df, shadow i.s bounded ronfinuou.'>. that for rvrry f or unliounded from below. ronr^vr hwI strictly increasing. > Tiien thi^rr rxif^ta a sohition to (15). 2 4 includes the utility functions that arc of constant relative risk aversion as For utility functions that are uubouiuled from below, we have the following. Proposition 4.2. Sujtjinfir Af, tli^t [1985], the final date. at that the invrrs(- of the sufficient conditions for utility fiuictions that are either 4.1. and Htiang e~'^''u(T), 9 aiul 2 10) own assumptions by lemma Proposition ('ox Ceneral conditions hoiuided from below, then a sufficient condition for almost is /?, than (14) form fvuictiou displays a time- additive structure of the all in consumption rate process or cons\imption «(j,<) same expected ajiiilirations. ninrli rasirr to solve is it solutions yielding tbo .'?n;'/in.'jr further that tlirrr that n rxif^ts }< : /?^\{n} > sucli that \l'{x) asywptoticpilly fnr snnir Kf. rronf Cox and Huang > [198C]. n i—» 7? i> fi<r < Ky (Jiffcimtinl'lc. rvrry A > concave and d -ft+(* 7 Tlirii t}irrr rxists a snlutidll to (15). Theorem 2.C 22 f^trictly incrcaf^inc. 5. Wo we An example conclude this paper by giving an apjiliration to a conpinni)tion portfolio problem. Since piini)lc are primarily interested in cxistmcr rather than characterization, example. in this We where markets are incomplete. consumption space = B(t) G{t) 1. The ft is a i{S{s),s)d.^ 1 vector of constants, n Al = |h- at [ d.« -^ The the final date. constant and equal to zero so that is Brownian motion V(G[0,T) Is{s)rTdW(.'^) {a.s.,r), Jo is MX an N constant matrix of e(jual t" Si{t). is i = M 1 full rank The (M < N), j^rocess ri[t) of becomes the lognormal variable 0{t) e where £.^(G,P) S(t) L'^iF) is = cxp{-nh W{t) - -l^'f }. A take a particularly simple form hi this case strategies -f Is{s)h ?-tli eiitiy ,,(0 a(t] interest rate Jo A/ X jiro])nsition 2.2 = The = S(0)+ [ and 75(f) a diagonal matrix wlinsr W{t) consumption occurs only Siipjiose that then be L^{P). will can be extended to the situation easily its solution risky securities gain i)rocess follows a nmltiplicative geometric = S(t)+ f Jo where how only want to illustrate we do not seek completeness ; = W(0) + W self-financing strateg)' is defined by dG(t) and the space of marketed commodities jlj(t) = W(0) + / (17) dG(t) 0(1] for some H'(0) G the set of 7,-prngressiveIy measural)])' such that /? |||rT and is G C^(G,P) 7s(0'^(0lil2,r < }, O^i ^^ before. We now choose a utility functinu of constant relative risk aversion: 1 u{x] = ^ 1 l/i ~ ~ X if T G (n,oo) and 7 > 0; if 3- = and -y > 1; if 3: = and 7 G (0. 1). I '^ = nrit defined where we have used the convention that The j)rnl)lem faced by an investfir .t max E^ulW) su]i log .t. endowcMl with one unit of the Cdnsumjition good s.t. new Consider the extended maximization = /n ])rop;raiii \e^u{X) : W > and n'(n) = at time is 1. coi-respondinj-'; to (15) X> 23 n and E'^ X = l] . (IS) whrrr t^ = {e^X^~^ P.ip Now defiup When as bofore. rj. X Y = : X >0 X X <^ '^ > 7 Ca\irhy-Schwarz inequality 1 wr have and V = X E'^ = where 7 E L^(Q), l] = sui' Y E L"^ E^\X\'~-^ conjnpatc (i aiul 7' arc then { : i; -f- E'^IXI = l] = Since by the 1). (Q) so the above siipreinum is . no greater than i}= {e'\Y\:\\Y\U.,,= sup {e'^-Y:\\Y\U.,,= sup If? VeL->'(g) YeL-^'iQ] that supreniuni tlic is i,Q -»-i Y = acliieved for V ) ->-i One can check - i\= which yields 7.Q (19) -).<? The cape D be solved < 7 < 1 can treated in the l)e same way with " replacing "sup", while 7=1 can diiectly. Since X G L^(P) lemma by 2.3, A' indeed the solution of (18). is measurable by proposition 2.3 so that by projiosition strategy that manufactures H'. Substituting (17) for W where iuf ' as a result of the = exp{7^'' 3.1, = in (19), rj M''{T] W A' marketed. is we hi addition. A' get after We now jjrice is elicit the some manipulations - ^I'^H'p ('ameron-Maitin transformation li'* is given by Tl'*(f) = Wit) + aht. Defining W{i) = exp{7'^'' ItfVs formula - —\^^>?]- ^''(O imi)lies that \\\i\ = e"^ W 7,1 = + 1 I 7 \V{.'<)rrh,nr(.'>) Jo = + 7 1 / \V{-'')n^n^nl,- ,/H'*(.,) ./n = since da{t) = Is(t)ndV,''(1). + 1 Hence by lemma 7 / n'(.') i^^{.-<)<^nh 3.2, Oit) = ,/r;(..) il'^ (f)n'^nh\V (t). But hh G Range rr^ so that (7"^ because S'l 7r„ = /\/ when n is nh = {rrrr )" nrrli of full rank. thai, a= exix'ctefl. thr piirtfnlui shares We ar'- = {frrr tlien = [nn )~ h have rciustant 24 )~ n^h and indejiendan* of the level of wralfli 6. Concluding remarks In this pajjer incomplete. of we have oxtrndrd the martinsalr apprnarh We obtained essentially two results. First, it is still One hedging the risk associated with holding positions in it is consistent with the has, for instance, a constiiictive technique any ojjtion written on stocks. Second, also possible to solve a very general version of the intertemporal jiortfolio problem. not seek to characterize the optimal individual pohcies, existence of a solution, which are virtually the same dynamically possible to duplicate the pay-offs any contingent claim by using an ajipropriate trading strategy, jirovided information generated by the securities price system. for to the rase wliere iiiaikets are we provided as in the it is Though we did sufficient conditions for the complete markets case. Furthermore, those conditions were entirely defined in terms of the primitives of the model. The model we have dealt with agree on the law of motion i')rices are necessarily a rational expectations one in the sense that by the seciuitics pnrc satisfi<vi have shown that when markets are fjy is incojiijilete, all ''marketivl" , in proc(\>;s. its agents all Under those conditions, we actions cfinsistent with the information revealed the sense that they can be imjilemented by trading the few long-lived securities continuously through time, and this result has jirovided the clue to the derivation of the ojitimal strategies. 25 References 1. G. Chanihrrlaiii, Asset piiring in iiuiltiixMiod srniritics markets, Miineo, University of Wis- consin. 1985. 2. K. Chung and R. Williams, An Int induction to StocliHstic bitogrHtinn, Birkhauser, Boston, 1983. 3. .1. Cox and C. Huang. A variational pro])Iem arising in finanrial eronomirs with an applica- tion to a portfolio turnpike theorem. .Sloan School of Management Working Paper ^^1751-86, Massachusetts histitute of Technology. 198G. 4. .1. Cox and C. Huang, Optimal consumption and portfolio i)olicies Management, Massaclnisetts diffusion process, Sloan School of 5. C Hollanrl C. P Mryer. Dellacherie and Piohnhiliticf: when asset prices follow a Institute of Technology, 1987. and PofrntinI D: Theory of Mnrtingales, North- PuMishing Company, New York, 1982. D. Duffie and C. Huang. Imjilementing Arrow-Debreu equilil)iia hy continuous trading of few long-lived securities. Ernnonirfrirn. 1985. 7. M. Harrison and D. ofErnnnmir Tlwory. 20, 8. M. Harrison and Martingales and arbitrage Krejis. S. in muUiiieriod s(^curities markets. Journal 1981, 381 408. Pliska, Mai'tingalep anrl stochastic integrals in the theory of continuous trading, ftfnrlinsfir Prorr!if;rf^ an'/ TijejV App/iVaf/o/i.s. 11, 1981. 215 2G0. 9. C. Huang, An intertemjioral general etjuilibrium asset jiricing model; The ca.e of diffusion information, Eronojnef;j>a 55. 1987, 117 142. in. I. Karatzas, Lehnrzky. S .1. tion/investment 11 H Setlii and S. Shrevc. Exjilirit solution of a general consump- MnthriiiRtirs of Oprrntimis Brtirnrrli 11, No. ])robl(-iii. Kunitii and S. Wata.nabe. On s(iuar('-integral)Ir martingales. Nngny;^ 2. 198C. Mnthrmnticfi Journal 30. 19G7. 209 245. 12. R. Lii)ts(i- Verlag. 13 and A. Shiryayev. New R. Mi-rton. 0])timniii ''onsuiii]itinn and i)ortfolio S Pliska. A 3. 1071. ri(<<T.<sr.-i /. (Tcncinl Theory. Sjninger- str)rhastic calrulu? D. Stroock, Lerfiire.s on I'ules in a contiinious time model, Joiirnai of 413 37,'5 mode] of ()prr;iti'nis Brfirnirli 11, No. 15. of nnndoni York. 1977. Eronxniir Tlimry 14 St:ttisticfi 2, of continuous tia'iing; 0])timal ixntfolios, MafhrniHtics 198G. .Sforiia.sf/V Ana/y.sjs. Diffusion Theory, (;and)ridge University Press, 1987, IG D. Stroock and S. Vararlhan. Mtilti'iiuirnsionH} Difftn^ion 1979. h53k ] [00 26 Prorr.s.sr.s, Springer- Verlag, New York. Date Due WJMK - Lib-26-67 Mn 3 ^DflD LIBRARIES DOS 131 MB?