Arbitrage and Control Problems in Finance. Presentation.

The theory of asset pricing takes its roots in the Arrow-Debreu model (see,for instance, Debreu 1959, Chap. 7), the Black and Scholes (1973) formula,and the Cox and Ross (1976) linear pricing model. This theory and its link to arbitrage has been formalized in a general framework by Harrison and Kreps (1979), Harrison and Pliska (1981, 1983), and Du¢e and Huang (1986). In these models, security markets are assumed to be frictionless: securities can be sold short in unlimited amounts, the borrowing and lending rates are equal, and there is no transaction cost. The main result is that the price process of traded securities is arbitrage free if and only if there exists some equivalent probability measure that transforms it into a martingale, when normalized by the numeraire. Contingent claims can then be priced by taking the expected value of their (normalized) payoff with respect to any equivalent martingale measure. If this value is unique, the claim is said to be priced by arbitrage and it can be perfectly hedged (i.e. duplicated) by dynamic trading. When the markets are dynamically complete, there is only one such a and any contingent claim is priced by arbitrage. The of each state of the world for this probability measure can be interpreted as the state price of the economy (the prices of $1 tomorrow in that state of the world) as well as the marginal utilities (for consumption in that state of the world) of rational agents maximizing their expected utility.

The theory of asset pricing takes its roots in the Arrow-Debreu model (see, for instance, Debreu 1959, Chap. 7), the Black and Scholes (1973) formula, and the Cox and Ross (1976) linear pricing model. This theory and its link to arbitrage has been formalized in a general framework by Harrison and Kreps (1979), Pliska (1981, 1983), and Duffie and Huang (1986). In these models, security markets are assumed to be frictionless: securities can be sold short in unlimited amounts, the borrowing and lending rates are equal, and there is no transaction cost. The main result is that the price process of traded securities is arbitrage free if and only if there exists some equivalent probability measure that transforms it into a martingale, when normalized by the numeraire. Contingent claims can then be priced by taking the expected value of their (normalized) payoff with respect to any equivalent martingale measure. If this value is unique, the claim is said to be priced by arbitrage and it can be perfectly hedged (i.e. duplicated) by dynamic trading. When the markets are dynamically complete, there is only one such a and any contingent claim is priced by arbitrage. The of each state of the world for this probability measure can be interpreted as the state price of the economy (the prices of $1 tomorrow in that state of the world) as well as the marginal utilities (for consumption in that state of the world) of rational agents maximizing their expected utility.
When there are frictions, including dynamic market incompleteness, the characterization of the no-arbitrage condition is nomore equivalent to the existence of a unique equivalent martingale measure. More precisely, for each equivalent martingale-measure condition equivalent supermartingale-measure condition, equivalent submartingale-measure condition, absolutely continuous martingale-measure condition kind of imperfection, the is replaced by a weaker one : , etc. Besides, we generally have, more than one measure satisfying these conditions. Furthermore, when there are frictions, even if a contingent claim can be duplicated by dynamic trading, it is not necessarily possible to price it by arbitrage. However arbitrage bounds can be computed, for arbitrary contingent claims, taking the expected value of their (normalized) payoff with respect to all the measures that characterize the absence of arbitrage opportunities. These bounds are the minimum amount it costs to hedge the claim and the maximum amount that can be borrowed against it using dynamic strategies. These are the tightest bounds that can be inferred on the price of a contingent claim without knowing the agent s preferences. The determination of these bounds in a dynamic setting leads to a maximization (and/or minimization) program, and, in a dynamic setting, is often transformed into a stochastic optimal control problem. The main assumption in these models is, in fact, a necessary condition for the existence of an equilibrium: the no-arbitrage condition. These preference-free theories give results of great generality without specifying the equilibrium in its full details.
Another important class of valuation theories makes assumptions on preferences and derives more speci c pricing restrictions than the preference-free theory does, even in the presence of imperfections. The price of a given contingent claim, for these theories, is just the expected value of its (normalized) terminal payoff with respect to a probability measure, whose density is proportional to the marginal utility (for consumption) of the considered agent. From a mathematical point of view, starting with a given utility function, the problem is to write the rst order conditions of the agent s utility maximization program, taking into account the potential imperfections in the description of the budget constraints and/or of the strategies in order to characterize the marginal utility for consumption at the nal date. In a multi-period setting, this maximization problem is a stochastic optimal control problem. The main advantage of this approach is that it leads to a unique price for a given contingent claim. The main drawback is that this price depends on the choice of the utility function and on the agent endowment. If the utility function belongs to a given connected class of utility functions, we will obtain an interval of possible prices for that claim. More speci cally, if the considered class is the set of all von Neumann-Morgenstern (VNM) increasing and concave utility functions, the set of possible prices is exactly the set obtained with the arbitrage approach as shown by Jouini and Kallal (1999). The unique way to obtain tighter bounds with the utilitymaximization approach seems then to consider speci c functions, or speci c sets of functions, smaller than the set of all the VNM ones.
In fact, there is an interesting link between the two approaches. Since the arbitrage upper bound for a given contingent claim is equal to the minimum amount it costs to hedge it, taking the market frictions into account, the agent s problem (maximization of the utility provided by the terminal payoff among the strategies satisfying a given dynamic budget constraint) can be transformed The no-arbitrage condition We say that the semi-martingale satis es the condition of noarbitrage, (NA), if into a static problem where we maximize the utility among the set of contingent claims satisfying a budget constraint where the classical price functional is replaced by the functional. If we have, for instance, an explicit formula for the arbitrage upper bound, it suffices then to solve a static maximization problem instead of the initial stochastic dynamic control problem. The characterization of the no-arbitrage assumption is therefore crucial in order to solve the contingent claim pricing problem as well as to solve the individual utility maximization problem of each agent in the economy. The last step, if we want to explore all the implications of the Arrow-Debreu model in this nancial setting is then to write the equilibrium conditions in order to ensure that all the individual solutions are compatible .
Let be a probability space and be a ltration which models our information structure. This ltration is supposed to satisfy the usual conditions, i.e. the ltration is right continuous and contains all negligible sets (if and then We also suppose that the sigmaalgebra , and we consider a real valued semimartingale which models the price process for the marketed claims. In the next, we will denote by real line and by the set of nonnegative ones. Let us de ne, as in Delbaen and Schachermayer (1994), an admissible strategy as follows: This admissibility condition can be interpreted as a condition for strategies with a zero initial investment.
We consider, as in Stricker (1990), the convex cone in the space of equivalent classes of measurable functions, de ned up to equality almost everywhere, given by admissible and exists a.s.
The set is then the set of all terminal payoffs obtained through some admissible strategy.

Since
represents the set of all admissible terminal payoffs the no-arbitrage condition amounts to say that it is impossible to obtain a non-negative, non-zero payoffs with a zero initial investment.
Assume that it is possible to separate and in the sense that there exists a non-zero linear functional and a real number such that and and assume that the separating functional admits a representation as an expectation operator with respect to a probability measure , then under the (NA) condition, is equivalent to and for each in In particular, for each we have , therefore, and is a martingale measure for Unfortunately, the (NA) condition is seldom sufficient to apply a separation theorem. In the case where is locally bounded we have the following: This last condition is called No Free-Lunch with Vanishing Risk (NFLVR) and deals with sequences of strategies such that the negative parts of their terminal payoff tends to zero uniformly instead of strategies with non-negative terminal payoff as in the (NA) condition. Remark that the main difference between these two conditions lies in the fact that we have to consider a closure. Indeed, the condition is weaker than which is equivalent to the (NA) condition If, in the previous condition, we replace the norm-topology closure by the topology closure (where is the space of all integrable measurable functions) we obtain a version of the No Free-Lunch (NFL) condition introduced by Kreps (1981), and the existence of an equivalent local martingale measure is obtained for a bounded càdlàg, and adapted process Other intermediary concepts, as the No Free-Lunch with Bounded Risk (NFLBR) condition (where the closure is de ned as the set of weak*-limits and where the negative parts of the terminal payoffs tend to zero in probability and remains uniformly bounded), have been introduced in the literature and permit to obtain results similar to the previous theorem in different contexts: nite time set and (NA) condition with Dalang et al. (1989) (see Schachermayer (1992), Kabanov and Kramkov (1994a) and Rogers (1995) for elementary proofs), innite but discrete time set with Schachermayer (1994), continuous and bounded processes in continuous time with Delbaen (1992). Harrison and Kreps (1979) and Harrison and Pliska (1981) used the concept of simple strategies. Kreps (1981) used a concept of no free-lunch involving the convergence of nets or generalised sequences. Duffie and Huang (1986) and Stricker (1990) used convergence. Lakner (1993) used convergence in Orlicz spaces. Furthermore, it appears that the class of semi-martingales is the most general one compatible with this kind of results. Indeed, from the work of Föllmer and Schweizer (1991) and Ansel and Stricker (1993) we know that No Free-Lunch conditions imply, in some sense, that is a semi-martingale. Conversely, the existence of an equivalent-martingale measure for implies, by Girsanov s theorem, that is a semi-martingale.
As in Harrison and Kreps (1979) and Kreps (1981) and for a given contingent claim, we de ne the arbitrage pricing interval as the set of all the prices that are t t easier 0 0 0 where (resp. ) models the long (resp. short) position returns. When there are transaction costs, can be identi ed with admissible, and exists a.s.
where represents the total variation of and the magnitude of the transaction costs. Jouini and Kallal (1995a and b) characterized rst the absence of arbitrage opportunities in these different situations. Other contributions on related subjects are due to Kabanov and Kramkov (1994b), Shirakawa and Konno (1995), Kusuoka (1995), Cvitanic and Karatzas (1996), Cvitanic, Pham and Touzi (1999), Kabanov (1999). The differences between all these references are in the choice of the topology (or no topology) in order to de ne the concept of free-lunch, the choice of a space of admissible strategies (discrete strategies, simple strategies,...) and nally the choice of possible imperfections (or no imperfection). This choice is summarized by the choice of a convex cone contained in instead of itself in order to model the opportunity set. In this context, it is to nd a separating hyperplane between that set (or its closure with respect to some topology) and . Jouini and Kallal (1999) extended all the arbitrage, viability and equilibrium classical results to that setting mainly by assuming that the opportunity set is a convex cone (or even a convex set) and the pricing rule is sublinear. In this issue, Kabanov and Stricker (2001) propose a generalization of Jouini and Kallal s (1995a) result to the important case of a multi-asset market model where the transaction costs are de ned for each kind of transaction between any pair of assets. They use the geometric formalism developed previously by Kabanov (1999) and they characterize the absence of arbitrage opportunities in terms of martingale-like measures. Their result is established in a discrete time and nite set of states of the world framework and they only deal with arbitrages and not with freelunches.
In order to take a large set of possible frictions into account Carassus and Jouini (1997,1998,2000) in discrete time or in a deterministic setting, Jouini and Napp (2000) and Jouini, Napp and Schachermayer (2000) in continuous The utility maximization problem separating kill the combined problem of optimal portfolio selection and consumption rules for an individual in a continuous-time model where his income is generated by returns on assets and these returns or instantaneous growth rates are stochastic time propose to deal directly with the space of possible cash-ows instead of the space of terminal payoffs and they provide a characterization of the No Free-Lunch assumption in terms of the existence of a functional. Napp (2001) develops an arbitrage pricing theory and a super-replication concept in this cash-ow space.
However, all these results are obtained under a convexity condition on the space of attainable payoffs. This last assumption is not satis ed in economies with xed costs, i.e. with transaction costs which are not proportional to the size of the transactions. In this framework, the terminal payoff of a strategy is where is a bounded non-linear function of the strategy instead of as in the classical case. Therefore, it is easy to understand that large scale transactions will the transaction cost effect and that the characterization of the no-arbitrage condition should be an asymptotic version of the classical one. Jouini, Kallal and Napp (2001) prove that this characterization is in terms of absolutely continuous martingale measures and show that the existence of such a measure is necessary but not sufficient and that we need the existence of a family of such measures each one associated with a given date and a given event at that date in order to characterize the absence of free-lunches.
Before Merton s (1969) paper, most models of portfolio selection only considered one period. Furthermore, the investment decision by households was viewed in two parts : (a) the consumption-saving choice where the individual decides how much income and wealth to allocate into current consumption and how much to save for future consumption; and (b) the portfolio-selection choice where the investor decides how to allocate savings among the available investment opportunities. Merton (1969) The original analysis of Merton s model is based on the Hamilton-Jacobi-Bellman equation and requires an underlying Markov state process. After the papers of Harrison and Kreps (1979) and Pliska (1981, 1983), and their characterization of the no-arbitrage assumption in terms of the existence of martingale-measures, Pliska (1986), Huang (1989, 1991) and Karatzas, Lehoczky and Shreve (1987) used this methodology in order to analyze this consumption-investment problem. This new approach is based on duality arguments and permits to transform the initial dynamic problem into a static one and to solve it without assuming any Markov condition.
Let us now introduce the main results related to this problem. Let be a xed probability space and denote the interval on which we are going to treat our problem : corresponds to the terminal date for all economic activity under consideration. All processes that we shall encounter in this section are de ned on .
All vectors are column vectors and transposition is denoted by the superscript . We denote by the nonnegative real number . See for instance Karatzas-Shreve As usual, denotes the -dimensional vector whose component are equal to one.
The real-valued interest rate process , the dimensional process , the dimensional dividend yield process as well as the volatility -matrix-valued process are supposed to be progressively measurable with respect to and bounded uniformly in in For all in the volatility matrix has full rank and the norm of is uniformly bounded.
We consider a market consisting in one bond and assets. More precisely, the primitive market model is the same as in Karatzas , except that we consider here dividends paying assets. We adopt a model for the market consisting of one bond with price at time denoted by satisfying the differential equations , and stocks with prices at time denoted by the dimensional vector satisfying (2) Here, is a -dimensional Brownian motion on a probability space and denotes the -augmentation of the natural ltration generated by . We assume that the sample paths of specify completely all the distinguishable events, which mathematically entail . Since standard Brownian motions start from zero with probability one, is trivial. We will denote by the set of -progressively measurable processes taking values in such that a.s.
Under this assumption , Equation admits a unique real-valued, adapted, continuous solution , satisfying .
Therefore, a -dimensional process can be de ned by : a.s.
With the above assumptions, is -progressively measurable and uniformly bounded. We shall also introduce the discounted price process de ned by for all in . Using Itô s Lemma, we easily get that is the unique solution of the following stochastic differential equation: For any -valued process in let the real-valued process denote the exponential local martingale given for each in by If denotes a vector in then denotes the diagonal matrix whose diagonal entries are the components of Notice that assets prices can uctuate in an almost arbitrary, not necessarily Markovian way.
We know that in such a model, there exists a unique equivalent probability measure de ned on that makes the full process a martingale for . It is given by We then have , where is the Brownian motion for de ned by for all in (see Girsanov s Theorem) We shall denote in the following, the martingale process by . In the context of the above market-model, consider an agent who starts out with an initial capital and can decide of the amounts that he invests at time in the different assets, and of the rate at which he withdraws funds for consumption. Assuming that at each time , sales and dividends must nance purchases and consumption, the corresponding wealth process, denoted by , satis es the following stochastic differential equation which can be rewritten ( Self-nancing condition) The set of investment-consumption strategies satisfying the previous self-nancing condition and the following no-bankruptcy condition is called the admissible strategies set and denoted by : This last condition amounts to saying that at each time , the investor must be able to cover his debts -see e.g. Karatzas-Lehoczky- Shreve (1987) or Duffie (1992) where the same assumption is made.   Under the self-nancing condition, the process consisting in the current discounted wealth plus the total discounted consumption is a -supermartingale (Fatou s Lemma). It is then easy to see that the market excludes any arbitrage opportunity which turns out to be characterized in our context by the existence of a pair in such that ).

Let denote the set of pairs
where is an adapted nonnegative consumption rate process and is a nonnegative -measurable random variable describing the terminal wealth. An agent is represented by a utility function given by where and satisfy the following assumption. , , i.e., and ) Under Assumption , we shall denote by the derivative of and by the inverse function of , which is a strictly decreasing continuous function on in . We shall also denote by the inverse function of . The considered agent has an initial endowment and tries to maximize his utility on both his consumption over the time-interval and his terminal wealth. The optimal demand of the agent in the consumption commodity as well as his optimal portfolio are determined by the optimization problem .
Adapting the proofs of Duffie (1994) and Karatzas (1989), we get that This last proposition permits to solve explicitly the agent s optimization program. Huang and Pagès (1992) extended this methodology to the in nite horizon framework. Karatzas, Lehoczky, Sethi and Shreve (1986) provided explicit computations in that framework assuming constant coefficients in the price evolution equations. He and Pearson (1991 a and b) and Karatzas, Lehoczky, Shreve and Xu (1991) extended the methodology to incomplete markets and proved that the optimal investment/consumption plan is given, as in the classical case, by the inverse of the marginal utility evaluated at the random variable which is optimal for a well-de ned dual problem. Cvitanic and Karatzas (1992) used the same approach in order to solve the problem when there is convex constraints on the strategies (short-sales constraints, borrowing constraints,...). Fleming and Zariphopoulou (1991) solved the problem assuming different borrowing and lending rates. Cuoco (1997) and El Karoui and Jeanblanc-Picqué (1997) considered random endowment streams. Cvitanic and Ma (1996) and Cuoco and Cvitanic (1998) generalized these results in the context of a large investor . In that context, the strategy of the investor has a direct nonlinear impact on the price dynamics. The main technique in all these references consists in embedding the original problem into a family of perfect (linear) ctitious markets, where security prices dynamics are modi ed and agents receive an additional stochastic endowment re ecting the nonlinearity in the market price of risk. The ctitious markets are designed in such a way that the optimal policy in one of them coincides with that in the actual, nonlinear market. Using the partial differential equations (PDE) approach, Dumas and Luciano (1989) rst formulated the problem in the presence of transaction costs. The main contributions in this context are Davis and Norman (1990), Fleming, Grossman, Vila and Zariphopoulou (1990) and Shreve and Soner (1994).
When there are imperfections, the utility maximization approach can be used in order to provide pricing formulas for new contingent claims. There are mainly two methods. The rst one, initiated by Hodges and Neuberger (1989) in the transaction costs setting, consists in using the marginal utility of the considered agent at his optimal consumption-investment plan as a state-price density. The second method, initiated by Davis (1994) consists in comparing the optimal utility levels with a deterministic initial endowment and with a stochastic endowment equal to the payoff of the considered claim. The fair price is then de ned by the equation Papers along these lines include Constantinides (1986), Panas (1993), Davis, Panas and Zariphopoulou (1993), Davis and Panas (1994), Davis and Zariphopoulou (1995), Cvitanic and Karatzas (1996), Constantinides and Zariphopoulou (1997) and Barles and Soner (1998).
. d R R

The equilibrium
In this issue, Cvitanic and Wang (2001) show that the martingale/duality approach adopted in the frictionless model works also in the transaction costs framework and prove that the optimal terminal wealth is given as the inverse of marginal utility evaluated at the random variable which is optimal for an appropriately de ned dual problem. They prove the existence of a solution for this dual problem and doing so they resolve a question left open by Cvitanic and Karatzas (1996). A similar problem is studied by Deelstra, Pham and Touzi (2000) where the utility functions are de ned on instead of Framstad, Øksendal and Sulem (2001) considers also the transaction costs framework but in a jump diffusion market. Using a viscosity solution approach, they show that the solution of the problem in that context has the same form as in the pure diffusion case : there is a no-transaction cone such that it is optimal to make no transactions as long as the wealth position remains in that cone and to trade on the boundary. Bellamy (2001) solves the same problem but assuming market incompleteness instead of the presence of transaction costs and using a Hamilton-Jacobi-Bellman (HJB) approach.
Using ltering techniques, Lakner (1995) considers utility maximization problems where the agent must estimate the mean rate of return of the assets. In this issue, Dokuchaev and Zhou (2001) considers the case where the stock appreciation rates are not observable and where the strategies depend only on the known distribution of these rates and on the current prices. Furthermore, they use general utility/loss functions (including mean-variance criteria and goal achieving problems) and they consider some lower and upper constraints on the terminal wealth. The problem is solved by means of backward stochastic differential equations as well as a dual formulation.
Finally, we want to mention another family of optimization problems related to the contingent claims pricing : the hedging problems. These problems are not represented in this special issue but are studied by many papers in the recent literature. The main problem in all these paper is to compute the hedging price of a given contingent claim with respect to a given hedging criterion. If we assume that the agents want to minimize the downside risk, then the hedging price is equal to the super-replication price and its computation leads to solve a stochastic-control-based problem as in the pionneering paper of El Quenez (1991, 1995). If we assume that the agents want to minimize the quadratic risk then we have to solve the mean-variance hedging problem and we refer to Föllmer and Schweizer (1991) or Schweizer (1993) for a survey about related results.
Models of competitive equilibrium go back to Walras (1874). The rst complete proof for the existence of an equilibrium in an economy with nitely many commodities was given by Arrow and Debreu (1954). In the chapter 7 of Debreu (1959), the author explains how this model permits to take into account dynamic markets with uncertainty. Bewley (1972) studied the competitive equi- All the previous models does not take explicitely into account dynamic security trading. Models where the agents achieve equilibrium allocations by trading in securities like the capital asset pricing model (CAPM) or the consumption based capital asset pricing model (CCAPM) can be found in the literature going back to Merton (1971), Cox, Ingersoll and Ross (1985), Duffie and Huang (1985), Huang (1987) and Karatzas, Lehoczky and Shreve (1990).
The link between these two approaches is made by Duffie and Huang (1985) where the authors explain how an Arrow-Debreu equilibrium can be implemented by trading in securities. This role of securities was, in fact, already recognized by Arrow (1952). The difference between the two approaches is illustrated by Cuoco (1997) where the budget constraints are associated to all the possible equilibrium prices (all the risk-neutral measures) instead of a unique budget constraint associated to the equilibrium price as in the classical general equilibrium model.
In Karatzas, Lehoczky and Shreve (1990), all agents are endowed in units of the same perishable commodity, which arrives at some time-varying random rate. Agents may consume their endowment as it arrives, they may sell some portion of it to other agents, or they may buy extra endowment from other agents. The endowment, however, cannot be stored, and agents wish to hedge the variability in their endowment process by trading with one another.
In this model all the prices are in term of a unique consumption good. When the market is complete it is equivalent to assume that the agents receive their endowment initially rather than over time. In that case and in order to have a stochastic total wealth, we assume that the consumption good is produced by the rms and distributed as dividends among the shareholders. The equilibrium condition imposes then a total consumption equal to the total supply of the consumption good and a total investment in each rm equal to the total value in term of consumption good of that rm.
With the notations of Jouini and Napp (1998), the mathematical description of the model is the following. Let be an economy with agents indexed by and let us assume that the agent has an initial wealth and a utility function given by where and satisfy our assumption . As previously each agent maximizes In this framework an equilibrium consists in a dimensional price processes and trading-consumption choices which are optimal for the agents, i.e. and such that for all in , the following market clearing conditions hold almost surely: where is the number of rm outstanding shares Note that the last condition is redundant with the two previous ones by the self-nancing condition.
In Karatzas, Lehoczky and Shreve (1990) it is shown that under mild conditions a unique equilibrium exists. In this issue, Chiarolla and Haussmann (2001) specializes and extends the Karatzas and al. (1990) model to a situation where the endowment streams of the agents are denominated in money, not in goods, and are not exogenous. The labor provided by the agents to a rm produces the consumable good through a production function. The agents have then to choose a consumption and a leisure levels in order to maximize their utility function. Furthermore, the rm de nes the level of employment by a pro t maximization program. The utility functions of the agents depend then on two control variables and the main contribution of this paper is to extend the classical one-dimensional approach to this framework. The authors provide rst order necessary conditions for equilibrium, and derive from there the existence of such an equilibrium. They also solve explicitly two examples.
Basak and Croitoru (2001) exploit the equilibrium conditions in order to analyze the taxation impact on the asset prices. They consider a simple two agents model and use the ctitious market techniques described in the previous section in order to solve the individual utility maximization problem. The main difficulty is due to the presence of two redundant assets but with different taxation rules. The redundancy adds an extra step in the agent s problem : once he has chosen his risk exposure, he must decide how to allocate that risk between the two securities. The authors establish general necessary conditions for equilibrium and show, in particular, that arbitrage opportunities still exist at the optimum. They characterize this mispricing and they provide an analysis of its equilibrium role.
Some of the papers gathered here have been presented at the International Conference on Mathematical Finance, Hammamet, Tunisia, 14-18 juin 1999 organized by Nizar Touzi and the author. We wish to thank all the participants at this conference for helpful discussions as well as the members of the scienti c committee (Gérard Debreu (President), Freddy Delbaen, Ivar Ekeland, Ioannis Karatzas, Pierre-Louis Lions, Stanley Pliska and Dieter Sondermann), the members of the organizing committee (Guillaume Bernis, Abdelhamid Bizid, Laurence Carassus, Kaïs Hamza, Clotilde Napp, Abdelhamid Trad and Amel Zenaïdi) and our sponsors (UNESCO, Tunisian authorities, French authorities, ADRES, Tunisie-Valeurs group, Amen Invest and the Tunis Stock Market). We are also most grateful to the and to its Editor, Bernard Cornet, for welcoming us in print. I have also bene tted from comments and discussions with Guillaume Bernis, Abdelhamid Bizid, Bruno Bouchard, Martino Grasselli, Clotilde Napp and Nizar Touzi in the writing of this presentation. I am also indebted to CREST for the very stimulating scienti c atmosphere during the 1997-2000 period and to the Stern Business School where this paper was written for inviting me during the 1999-2000 period.
[11] Bellamy N. (2001), Wealth optimization in an incomplete market driven by a jump-diffusion process. , this issue.
[20] Constantinides G.M.and Th. Zariphopoulou (1997), Bounds on option prices in an intertemporal economy with proportional transaction costs and general preferences. Preprint.