ON THE SEMIMARTINGALE PROPERTY OF DISCOUNTED ASSET-PRICE PROCESSES IN FINANCIAL MODELING CONSTANTINOS KARDARAS AND ECKHARD PLATEN arxiv:83.189v4 [q-fin.pr] 1 Aug 29 This work is dedicated to the memory of our colleague and dear friend Nicola Bruti Liberati, who died tragically on the 28th of August, 27. Abstract. A financial market model where agents trade using realistic combinations of buy-andhold strategies is considered. Minimal assumptions are made on the asset-price process in particular, the semimartingale property is not assumed. Via a natural assumption of limited opportunities for unlimited resulting wealth from trading, coined the No Unbounded Profit with Bounded Risk condition, we establish that asset-prices have to be semimartingales. In a slightly more specialized case, we extend the previous result in a weakened version of the Fundamental Theorem of Asset Pricing that involves supermartingale deflators rather than Equivalent Martingale Measures. 1. Introduction In the process of obtaining a sufficiently general version of the Fundamental Theorem of Asset Pricing (FTAP), semimartingales proved crucial in modeling asset-price processes. The powerful tool of stochastic integration with respect to general predictable integrands, that semimartingales are exactly tailored for, finally lead to the culmination of the theory in [9, 1]. The FTAP connects the economical notion of No Free Lunch with Vanishing Risk (NFLVR) with the mathematical concept of existence of an Equivalent Martingale Measure (EMM), i.e., an auxiliary probability, equivalent to the original (in the sense that they have the same impossibility events), that makes the discounted asset-price processes have some kind of martingale property. For the above approach to work one has to utilize stochastic integration using general predictable integrands, which translates to allowing for continuous-time trading in the market. Even though continuous-time trading is of vast theoretical importance, in practice it is only an ideal approximation; the only feasible way of trading is via simple strategies, i.e., combinations of buy-and-hold strategies. Recently, it has been argued that existence of an EMM is not necessary for viability of the market; to this effect, see [18, 2, 11]. Even in cases where classical arbitrage opportunities are present in the market, credit constraints will not allow for arbitrages to be scaled to any desired degree. It is rather the existence of a strictly positive supermartingale deflator, a concept weaker than existence of an EMM, that allows for a consistent theory to be developed. Our purpose in this work is to provide answers to the following questions: Date: August 1, 29. Key words and phrases. Numéraire portfolio; Semimartingales; Buy-and-hold strategies; No-short-sale constraints; Unbounded Profit with Bounded Risk; Supermartingale deflators. The first author would like to thank the warm hospitality of the School of Finance and Economics of the University of Technology, Sydney, where this work was carried out. 1
2 CONSTANTINOS KARDARAS AND ECKHARD PLATEN (1) Why are semimartingales crucial in modeling discounted asset-price processes? (2) Is there an analogous result to the FTAP that involves weaker (both economic and mathematical) conditions, and only assumes the possibility of simple trading? A partial, but precise, answer to question (1) is already present in [9]; here, a different approach is undertaken. A comparison is carried out later on in the text. In [1, 3, 17], the semimartingale property of the asset-price processes is obtained via a different approach, using the fact that the indirect utility maximization problem has a finite value. (In the present paper, this approach is revisited as a corollary to our main result.) All the above conditions are only sufficient to ensure the asset prices are semimartingales; here, we shall also discuss conditions that are both necessary and sufficient. The weakened version of the FTAP that we shall come up with as an answer to question (2) is a simple, no-short-sale trading version of Theorem 4.12 from [14]. The structure of the paper is as follows. Section 2 introduces the market model, simple trading and no-short-sale constraints, and discusses the condition No Unbounded Profit with Bounded Risk (NUPBR a weakening of condition NFLVR) for simple, no-short-sale trading, as well as the concept of strictly positive supermartingale deflators. After this, our main result, Theorem 2.1, is formulated, which establishes both the importance of semimartingales in frictionless financial modeling, as well as the weak version of the FTAP. Finally, Section 3 deals with proving Theorem 2.1. We note that, though hidden in the background, the proofs of our results depend heavily on the notion of the numéraire portfolio (also called growth-optimal, log-optimal or benchmark portfolio), as it appears in a series of works; [15, 19, 2, 13, 2, 21, 14, 8], to mention a few. 2. The Semimartingale Property of the Discounted Asset-Price Process and a Version of the Fundamental Theorem of Asset Pricing 2.1. The financial market model and trading via simple, no-short-sale strategies. The random movement of d N risky assets in the market is modeled via an arbitrary càdlàg, nonnegative stochastic processes S i, where i {1,...,d}. As is usual in the field of Mathematical Finance, we assume that all wealth processes are discounted by another special asset which is considered a baseline. The above process S = (S i ) i=1,...,d is defined on a filtered probability space (Ω, F, (F t ) t R+, P), where (F t ) t R+ is a filtration satisfying F t F for all t R +, as well as the usual assumptions of right-continuity and saturation by all P-null sets of F. Observe that there is no a priori assumption on S being a semimartingale. This property will come as a consequence of a natural market viability assumption. In the market described above, economic agents can trade in order to reallocate their wealth. Consider a simple predictable process θ := n j=1 ϑ τ j 1 I ]τj 1,τ j ], where n ranges in N, and for j {,...,n}, τ j is a finite stopping time and ϑ τj 1 = (ϑ i τ j 1 ) i=1,...,d is F τj 1 -measurable. Each τ j, j {,...,n}, is an instance when some given economic agent may trade in the market; then, ϑ i τ j 1 is the number of units from the ith risky asset that the agent will hold in the trading interval ]τ j 1,τ j ]. This form of trading is called simple, as it comprises of a finite number of buy-and-hold strategies, in contrast with continuous trading where one is able to change the position in the assets in a continuous fashion. This last form of trading is only of theoretical value, since it cannot be implemented in reality, even if one ignores market frictions, as we do here to keep the exposition
SEMIMARTINGALES IN FINANCIAL MODELING 3 simple. Starting from initial capital x R + and following the strategy described by the simple predictable process θ := n j=1 ϑ τ j 1 I ]τj 1,τ j ], the agent s discounted wealth process is given by (2.1) X x,θ = x + θ t, ds t := x + n j=1 ϑτj 1,S τj S τj 1. (We use, throughout to denote the usual Euclidean inner product on R d.) The wealth process X x,θ of (2.1) is càdlàg and adapted, but could in principle become negative. In real markets, some economic agents, for instance pension funds, face several institution-based constraints when trading. The most important constraint is prevention of having negative positions in the assets; we plainly call this no-short-sales constraints. Consider a wealth process X x,θ as in (2.1), where x R + and θ is simple and predictable. In order to ensure that no short sales are allowed in the risky assets, which also include the baseline asset used for discounting, we define X(x) to be the set of all wealth processes X x;θ given by (2.1) such that ϑ i τ j and ϑ τj,s τj X x,θ τ j hold for all i {1,...,d} and j {,...,n}, where the collection { = τ < τ 1 <... < τ n } ranges through all stopping times τ j, j {,...,n}, and n N. Note that the previous conditions on no short sales, coupled with the nonnegativity of S i, i {1,...,d}, imply the stronger θ i for all i = 1,...,d and θ,s X x,θ. (The subscript is used to denote the left-continuous version of a càdlàg process.) Observe also that X(x) = xx(1) for all x R + \ {}, but it might be the case that X() {} = X(1). Finally, define X := x R + X(x). 2.2. Unbounded Profit with Bounded Risk. We define here the essential no-free-lunch concept to be used in our discussion. We shall say that a market where only simple, no-short-sale trading is allowed satisfies the no unbounded profit with bounded risk (NUPBR) condition if for all x R + and T R +, lim l sup X X(x) P[X T > l] = (in other words, if the collection {X T X X(x)} is bounded in probability). Since X(x) = xx(1) for x >, we only have to check the above condition for x = 1. Note that under condition NUPBR we have {X T X X(x)} being bounded in probability for all x R + and finite stopping times T, something that is not valid for the No Free Lunch with Vanishing Risk (NFLVR) condition of [9]. If condition NUPBR fails, then one can find some financial planning horizon T R +, a sequence (X k ) k N of elements in X(1) and a p > such that P[XT k > k] > p for all k N. This sequence (X k ) k N has bounded risk, that is, no more than unit losses, while with at least some fixed positive probability p > it can make unbounded profit, which explains the appellation of condition NUPBR. Now, a combination of Lemma A.1 of [9] and Lemma 2.3 in [5], shows that NUPBR fails if and only if there exists a financial planning horizon T R + and a sequence (X k ) k N of wealth process in X such that lim k X k = and P-lim k XT k = ξ, where ξ is a [, ]-valued random variable with P[ξ > ] >. Therefore, condition NUPBR coincides with absence of cheap thrills in the market, in the sense of [18], under simple, no-short-sale trading in the market. This fact also exhibits that the NUPBR condition is weaker that the NFLVR condition. 2.3. Supermartingale deflators. We now introduce a concept that is closely related, but weaker, to that of equivalent (super)martingale probability measures. It appears as the natural dual domain in the solution of the utility maximization problem from terminal wealth in [16]. It is also discussed
4 CONSTANTINOS KARDARAS AND ECKHARD PLATEN in [14] in a context close to what will be discussed in this section, but in a general semimartingale setting and using continuous-time trading. Define Y := {Y > Y = 1, and Y X is a supermartingale for all X X } to be the class of strictly positive supermartingale deflators for simple, no-short-sale trading. Note that a process Y > with Y = 1 is an element of Y if and only if it satisfies the following seemingly weaker property: all processes Y and Y S i, i {1,...,d}, are supermartingales. Indeed, assume the latter property and consider X x,θ in the notation of (2.1). Using integration by parts, we obtain ( ) Y X x,θ = x + X x,θ t θ t,s t dy t + θ t, d(y t S t ). If X x,θ X(x), we have X x,θ θ,s, as well as θ i for i {1,...,d}. Then, the supermartingale property of Y and Y S i, i {1,...,d} gives that Y X x,θ is a supermartingale. 2.4. The main result. Condition NUPBR, existence of strictly positive supermartingale deflators and the semimartingale property of S are immensely tied to each other, as will be revealed below. Remember that each S i, i {1,...,d}, is a nonnegative process and define the (first) bankruptcy time of the ith risky asset as ζ i := inf{t R + St i = or Si t = }. For i {1,...,d}, we shall say that S i cannot revive from bankruptcy if St i = holds for all t ζ i on {ζ i < }. Before stating our main Theorem 2.1, whose proof is the content of Section 3, recall that S i, i {1,...,d}, is an exponential semimartibgale if there exists a semimartingale R i with R i =, such that S i = S ie(ri ) where E denotes the stochastic exponential operator. Theorem 2.1. Let S = (S i ) i=1,...,d be a adapted, càdlàg stochastic process such that S i is nonnegative for all i {1,..., d}. Consider the following four statements: (i) The NUPBR condition holds for simple, no-short-sale trading. (ii) Y =. (iii) S is a semimartingale, and S i cannot revive from bankruptcy for all i {1,...,d}. (iv) For all i {1,...,d}, S i is an exponential semimartingale. Then, we have the following: (1) It holds that (i) (ii) (iii), as well as (iv) (i). (2) Assume further that S i ζ i > holds on {ζi < } for all i {1,...,d}. Then, we have the equivalences (i) (ii) (iii) (iv). 2.5. Remarks on Theorem 2.1. We now discuss some topics related to Theorem 2.1. 2.5.1. Comparison with the result of Delbaen and Schachermayer. Theorem 7.2 of the seminal paper [9] of F. Delbaen and W. Schachermayer establishes the semimartingale property of S under the NFLVR condition for simple admissible strategies, coupled with a local boundedness assumption on S (together with the càdlàg property and adaptedness). The assumptions of Theorem 2.1 are different than the ones in [9]. Condition NUPBR for simple, no-short-sale strategies is weaker than NFLVR for simple admissible strategies. Furthermore, local boundedness from above is not required in our context, but we do require that each S i, i {1,...,d}, is nonnegative. If the components of S are unbounded both from above and below, not even condition NFLVR is enough to ensure the semimartingale propery of S; see Example 7.5 in [9].
SEMIMARTINGALES IN FINANCIAL MODELING 5 Interestingly, and in contrast to [9], the proof of Theorem 2.1 provided here does not use the deep Bichteler-Delacherie theorem on the characterization of semimartingales as good integrators (see [4, 22] where one starts by defining semimartingales as good integrators and obtains the classical definition as a byproduct). Actually, statement 2 of Theorem 2.1 can be seen as a multiplicative counterpart of the Bichteler-Delacherie theorem. Its proof exploits two simple facts: (a) positive supermartingales are semimartingales, which follows directly from the Doob-Meyer decomposition; and (b) reciprocals of strictly positive supermartingales are semimartingales, which is a consequence of Itô s formula. Crucial in the proof is also the concept of the numéraire portfolio. 2.5.2. The semimartingale property of discounted asset-price processes via bounded indirect utility. There has been previous work in the literature obtaining the semimartingale property of S is obtained using the finiteness of the value function of a utility maximization problem via the use of simple strategies only. (See [1, 3, 17].) In all case, there has been an assumption of local boundedness (or even continuity) on S. The result below is in the same spirit, droping the local boundedness assumption, only assuming that discounted asset-price processes are nonnegative. Furthermore, only no-short-sale simple strategies are considered, which allows one to obtain a sharper result. Proposition 2.2. Let S = (S i ) i=1,...,d be such that S i is an adapted nonnegative càdlàg process for i {1,...,d}, U : R + R be a nondecreasing function with U( ) =, x >, and T be a finite stopping time. If sup X X(x) E[U(X T )] <, the process (S T t ) t R+ is a semimartingale. Proof. Since we only care about the semimartingale property of (S T t ) t R+, assume without loss of generality that S t = S T t for all t R +. Suppose that condition NUPBR fails. Then, we can find a sequence (X k ) k N of elements in X(x) and p > such that P[XT k > k] p for all k N. But then, sup X X(x) E[U(X T )] lim inf k E[U(XT k)] p lim inf k U(k) =. This is a contradiction to the fact that sup X X(x) E[U(X T )] <. Therefore, we conclude that (S T t ) t R+ is a semimartingale using the implication (i) (iii) mentioned in statement (1) of Theorem 2.1. Remark 2.3. If we do not require the additional assumption on S in statement (2) of Theorem 2.1, implication (iii) (i) might fail. We present below a counterexample where this happens. On (Ω, F, P), let W be a standard, one-dimensional Brownian motion. (with respect to its own natural filtration we have not defined (F t ) t R+ yet). Define the process ξ via ξ t := exp( t/4+w t ) for t R +. Since lim t W t /t =, P-a.s., it is straightforward to check that ξ := lim t ξ t =, and actually that ξ t dt <, both holding P-a.s. Write ξ = A + M for the Doob-Meyer decomposition of the continuous submartingale ξ under its natural filtration, where A = (1/4) ξ t dt and M = ξ t dw t. Due to ξ t dt <, we have A < and [M,M] <, where [M,M] is the quadratic variation process of M. In the terminology of [6], ξ is a semimartingale up to infinity. If we define S via S t = ξ t/(1 t) for t [,1[ and S t = for t [1, [, then S is a nonnegative semimartingale. Define (F t ) t R+ to be the augmentation of the natural filtration of S. Observe that ζ ζ 1 = 1 and S ζ = ; the condition of statement (2) of Theorem 2.1 is not satisfied. We shall show below that NUPBR fails for T = 1. Using continuous-time trading, define a wealth process X for t [,1[, via X = 1 and the dynamics d X t / X t = (1/4)(dS t /S t ) for t [,1[. Then, Xt = exp ( ) (1/16)(t/(1 t)) + (1/4)W t/(1 t) for t [,1[, which implies that, P-a.s., lim t 1 Xt =. The fact that the percentage of investment
6 CONSTANTINOS KARDARAS AND ECKHARD PLATEN is 1/4 [,1] means that X is the result of a no-short-sale strategy. One can find an approximating sequence (X k ) k N such that X k X(1) for all k N, as well as P[ X k 1 X 1 1/k < 1] > 1 1/k. (Approximation results of this sort are discussed in greater generality in [23].) Then, (X k 1 ) k N is unbounded in probability, and NUPBR fails. Of course, in this example we also have (iii) (iv) of Theorem 2.1 failing. 3. Proof of Theorem 2.1 3.1. Proof of statement (1). We split the proof in parts, indicating in every case which implication is being proved. (i) (ii). Define the set of dyadic rational numbers D := {m/2 k k N, m N}, which is dense in R +. Define also, for all k N, the set of trading times T k := {m/2 k m N, m k2 k }. Then, T k T k for k < k and k N Tk = D. In what follows, X k (1) denotes the subset of X(1) consisting of wealth processes where trading only may happen at times contained in T k. We claim that, under condition NUPBR, for each k N one can find a wealth process X k X k (1) with X t k > for all t T k such that, by defining Ỹ k := 1/ X k, E[Ỹ t kx t F s ] Ỹ s kx s holds for all X X k (1), where T k s t T k. Indeed, the previous process X k is nothing more than the numéraire portfolio under no-short-sale constraints in the discrete-time market with discounted asset-prices given by (S t ) t T k. The existence of the numéraire portfolio follows from Theorem 4.12 of [14], as soon as one notices that condition NUPBR for the continuous-time market trivially implies condition NUPBR in each of the aforementioned discrete-time markets where trading is only allowed in times contained in T k, k N. For all k N, every Ỹ k satisfies Ỹ k = 1 and is a positive supermartingale when sampled from times in T k ; therefore, it is easily seen that for any T D, the convex hull of the set {Ỹ T k k N} is bounded in probability. We also claim that, under condition NUPBR, for any T R +, the convex hull of the set {Ỹ T k k N} is bounded away from zero in probability. Indeed, for any collection (α k ) k N such that α k for all k N, having all but a finite number of α k s non-zero and satisfying k=1 αk = 1, we have 1/ ( k=1 αk Ỹ k) k=1 αk (1/Ỹ k ) = k=1 αk Xk X(1). Since {X T X X(1)} is bounded in probability for all T R +, the previous fact proves that the convex hull of the set {Ỹ T k k N} is bounded away from zero in probability. Now, using Lemma A1.1 of [9], one can proceed as in the proof of Lemma 5.2(a) in [12] to infer the existence of a sequence (Ŷ k ) k N and some process (Ŷt) t D such that, for all k N, Ŷ k is a convex combination of Ỹ k,ỹ k+1,..., and lim k Ŷt k = Y t for all t D, P-a.s. The discussion of the preceding paragraph ensures that P[ < Ŷt <, t D] = 1. Let D s t D. Then, s T k and t T k for all k large enough. According to the conditional version of Fatou s Lemma, for all X k=1 X k we have that (3.1) E[ŶtX t F s ] lim inf k E[Ŷ t k X t F s ] lim inf Ŷs k X s = ŶsX s. k It follows that (ŶtX t ) t D is a supermartingale for all X k=1 X k. (Observe here that we sample the process Ŷ X only at times contained in D.) In particular, (Ŷt) t D is a supermartingale. For any t R + define Y t := lim s t,s D Ŷ s the limit is taken in the P-a.s. sense, and exists in view of the supermartingale property of (Ŷt) t D. It is straightforward that Y is a càdlàg process; it
SEMIMARTINGALES IN FINANCIAL MODELING 7 is also adapted because (F t ) t R+ is right-continuous. Now, for t R +, let T D be such that T > t; a combination of the right-continuity of Y, as well as the filtration (F t ) t R+, the supermartingale propery of (Ŷt) t D, and Lévy s martingale convergence Theorem, give E[ŶT F t ] Y t. Since P[ŶT > ] = 1, we obtain P[Y t > ] = 1. Right-continuity of the filtration (F t ) t R+, coupled with (3.1), imply that E[Y t X t F s ] Y s X s for all R + s t R + and X k=1 X k. In particular, Y is a càdlàg nonnegative supermartingale; since P[Y t > ] = 1 holds for all t R +, we conclude that Y >. We have essentially constructed a strictly positive supermartingale deflator for the class k=1 X k. Of course, 1 X k and S i X k hold for all k N and i {1,...,d}. It follows that Y is a supermartingale, as well as that Y S i is a supermartingale for all i {1,...,d}. From the discussion following the definition of Y in 2.3, we get that Y Y, i.e., that Y =. (ii) (i). Let Y Y and fix T R +. Then, sup X X(x) E[Y T X T ] x. In particular, the set {Y T X T X X(x)} is bounded in probability. Since P[Y T > ] = 1, the set {X T X X(x)} is bounded in probability as well. (ii) (iii). Let Y Y. Since S i X, Y S i is a supermartingale, thus a semimartingale, for all i {1,...,d}. Also, the fact that Y > and Itô s formula give that 1/Y is a semimartingale. Therefore, S i = (1/Y )(Y S i ) is a semimartingale for all i {1,...,d}. Furthermore, since Y S i is a nonnegative supermartingale, we have Y t St i = for all t ζ i on { ζ i < }, for i {1,...,d}. Now, using Y > again, we obtain that St i = holds for all t ζi on { ζ i < }. In other words, each S i, i {1,...,d}, cannot revive after bankruptcy. (iv) (i). Since S is a semimartingale, we can talk about continuous-time trading. For x R +, let X(x) be the set of all wealth processes X x,θ := x+ θ t, ds t, where θ is d-dimensional, predictable and S-integrable, θ t, ds t denotes a vector stochastic integral, X x,θ and θs X x,θ. Of course, X(x) X(x). We shall show in the next paragraph that { X T X X(1) } is bounded in probability for all T R +, therefore establishing condition NUPBR. For all i {1,...,d}, write S i = S ie(ri ), where R i is a semimartingale with R i =. Let R := (R i ) i=1,...,d. It is not hard to see that X(1) coincides with the class of all processes of the form E ( π t, dr t ), where π is predictable and take values in the d-dimensional simplex d := { z = (z i ) i=1,...,d R d z i for i = 1,...,d, and d i=1 zi 1 }. Since, for all T R +, ( ( T )) T log E π t, dr t π t, dr t holds for all d -valued and predictable π, it suffices to show the boundedeness in probability of the class of all T π t, dr t, where π ranges in all d -valued and predictable processes. Write R = B + M, where B is a process of finite variation and M is a local martingale with M i 1, i {1,...,d}. Then, T π t, db t d i=1 T db t <. This establishes the boundedness in probability of the class of all T π t, db t, where π ranges in all d -valued and predictable processes. We have to show that the same holds for the class of all T π t, dm t, where π is d - valued and predictable. For k N, let τ k := inf{t R + d i=1 [Mi,M i ] t k} T, Note that [M i,m i ] τ k = [M i,m i ] τ k + M i 2 k + 1 holds for all i {1,...,d}. Therefore, using the τ k
8 CONSTANTINOS KARDARAS AND ECKHARD PLATEN notation η L 2 := E[ η 2 ] for a random variable η, we obtain τ k d τ k π t, dm t πt i d dmi t [M i,m i ] τ k d k + 1 L 2 L 2 i=1 Fix ǫ >. Let k = k(ǫ) be such that P[τ k < T] < ǫ/2, and also let l := d 2(k + 1)/ǫ. Then, [ T ] [ [ ] ] τ k P π t, dm t > l P τ k < T + P π t, dm t > l ǫ 2 + τ k 2 π t, dm t L 2 l ǫ. L 2 The last estimate is uniform over all d -valued and predictable π. We have, therefore, established the boundedness in probability of the class of all T π t, dm t, where π ranges in all d -valued and predictable processes. This completes the proof. 3.2. Proof of statement (2). In view of statement (1) of Theorem 2.1, we only need to show the validity of (iii) (iv) under the extra assumption of statement (2). This equivalence is really Proposition 2.2 in [7], but we present the few details for completeness. For the implication (iii) (iv), simply define R i := (1/Si t )dst i for i {1,...,d}, The latter process is a well-defined semimartingale because, for each i {1,...,d}, S i is a semimartingale, S i is locally bounded away from zero on the stochastic interval [[,ζ i ]], and S = on [[ζ i, [[. Now, for (iv) (iii), it is clear that S is a semimartingale. Furthermore, for all i {1,...,d}, S i cannot revive from bankruptcy; this follows because stochastic exponentials stay at zero once they hit zero. i=1 References [1] S. Ankirchner and P. Imkeller, Finite utility on financial markets with asymmetric information and structure properties of the price dynamics, Ann. Inst. H. Poincaré Probab. Statist., 41 (25), pp. 479 53. [2] D. Becherer, The numeraire portfolio for unbounded semimartingales, Finance Stoch., 5 (21), pp. 327 341. [3] F. Biagini and B. Øksendal, A general stochastic calculus approach to insider trading, Appl. Math. Optim., 52 (25), pp. 167 181. [4] K. Bichteler, Stochastic integration with jumps, vol. 89 of Encyclopedia of Mathematics and its Applications, Cambridge University Press, Cambridge, 22. [5] W. Brannath and W. Schachermayer, A bipolar theorem for L +(Ω, F,P), in Séminaire de Probabilités, XXXIII, vol. 179 of Lecture Notes in Math., Springer, Berlin, 1999, pp. 349 354. [6] A. Cherny and A. Shiryaev, On stochastic integrals up to infinity and predictable criteria for integrability, in Séminaire de Probabilités XXXVIII, vol. 1857 of Lecture Notes in Math., Springer, Berlin, 25, pp. 165 185. [7] T. Choulli, C. Stricker, and J. Li, Minimal Hellinger martingale measures of order q, Finance Stoch., 11 (27), pp. 399 427. [8] M. M. Christensen and K. Larsen, No arbitrage and the growth optimal portfolio, Stoch. Anal. Appl., 25 (27), pp. 255 28. [9] F. Delbaen and W. Schachermayer, A general version of the fundamental theorem of asset pricing, Math. Ann., 3 (1994), pp. 463 52. [1], The fundamental theorem of asset pricing for unbounded stochastic processes, Math. Ann., 312 (1998), pp. 215 25. [11] R. Fernholz, I. Karatzas, and C. Kardaras, Diversity and relative arbitrage in equity markets, Finance Stoch., 9 (25), pp. 1 27.
SEMIMARTINGALES IN FINANCIAL MODELING 9 [12] H. Föllmer and D. Kramkov, Optional decompositions under constraints, Probab. Theory Related Fields, 19 (1997), pp. 1 25. [13] T. Goll and J. Kallsen, A complete explicit solution to the log-optimal portfolio problem, Ann. Appl. Probab., 13 (23), pp. 774 799. [14] I. Karatzas and C. Kardaras, The numéraire portfolio in semimartingale financial models, Finance Stoch., 11 (27), pp. 447 493. [15] J. L. Kelly, Jr, A new interpretation of information rate, Bell System Technical Journal, 35 (1956), pp. 917 926. [16] D. Kramkov and W. Schachermayer, The asymptotic elasticity of utility functions and optimal investment in incomplete markets, Ann. Appl. Probab., 9 (1999), pp. 94 95. [17] K. Larsen and G. Žitković, On the semimartingale property via bounded logarithmic utility, Annals of Finance, 4 (28), pp. 255 268. [18] M. Loewenstein and G. A. Willard, Local martingales, arbitrage, and viability. Free snacks and cheap thrills, Econom. Theory, 16 (2), pp. 135 161. [19] J. B. Long, Jr., The numéraire portfolio, Journal of Financial Economics, 26 (199), pp. 29 69. [2] E. Platen, Arbitrage in continuous complete markets, Adv. in Appl. Probab., 34 (22), pp. 54 558. [21], A benchmark approach to finance, Math. Finance, 16 (26), pp. 131 151. [22] P. E. Protter, Stochastic integration and differential equations, vol. 21 of Stochastic Modelling and Applied Probability, Springer-Verlag, Berlin, 25. Second edition. Version 2.1, Corrected third printing. [23] C. Stricker, Simple strategies in exponential utility maximization, in Séminaire de Probabilités, XXXVI, vol. 181 of Lecture Notes in Math., Springer, Berlin, 23, pp. 415 418. Constantinos Kardaras, Mathematics and Statistics Department, Boston University, 111 Cummington Street, Boston, MA 2215, USA. E-mail address: kardaras@bu.edu Eckhard Platen, School of Finance and Economics & Department of Mathematical Sciences, University of Technology, Sydney, P.O. Box 123, Broadway, NSW 27, Australia. E-mail address: eckhard.platen@uts.edu.au