Abstract
We address a problem of stochastic optimal control drawn from the area of mathematical finance. The goal is to minimize the expected value of a general quadratic loss function of the wealth at close of trade when there is a specified convex constraint on the portfolio over the trading interval, together with a specified almost-sure lower-bound on the wealth at close of trade. We use a variational approach of Rockafellar which leads naturally to an appropriate vector space of dual variables, a dual functional on the space of dual variables such that the dual problem of maximizing the dual functional is guaranteed to have a solution (i.e. a Lagrange multiplier) when a simple and natural Slater condition holds for the terminal wealth constraint, and obtain necessary and sufficient conditions for optimality of a candidate wealth process. The dual variables are pairs, each comprising an Itô process paired with a member of the adjoint of the space of essentially bounded random variables measurable with respect to the event \(\sigma \)-algebra at close of trade. The necessary and sufficient conditions are used to construct an optimal portfolio in terms of the Lagrange multiplier. The dual problem simplifies to maximization of a concave function over the real line when the portfolio is unconstrained but the terminal wealth constraint is maintained.
Similar content being viewed by others
References
Aliprantis, C.D., Border, K.C.: Infinite Dimensional Analysis, 3rd edn. Berlin: Springer (2006)
Aubin, J.P.: Applied Functional Analysis. New York: Wiley (1978)
Bielecki, T.R., Jin, H., Pliska, S.R., Zhou, X.Y.: Continuous-time mean-variance portfolio selection with bankruptcy prohibition. Math Financ 15, 213–244 (2005)
Bismut, J.M.: Conjugate convex functions in optimal stochastic control. J Math Anal Appl 44, 384–404 (1973)
Chow, Y.S., Teicher, H.: Probability Theory: Independence, Interchangeability, Martingales, 2nd edn. New York: Springer (1988)
Coqueret, G.: Diversified minimum variance portfolios. Ann Financ (to appear) (2014)
Cvitanić, J., Karatzas, I.: Convex duality in constrained portfolio optimization. Ann Appl Probab 2, 767–818 (1992)
Dubovitskii, A.Ya., Mil’yutin, A.A.: Necessary conditions for a weak extremum in problems of optimal control with mixed inequality constraints. Zhur Vychislitel Mat i Mat-Fys 8, 725–779 (1968)
Ekeland, I., Témam, R.: Convex Analysis and Variational Problems. Amsterdam: North-Holland (1976)
Hartl, R.F., Sethi, S.P., Vickson, R.G.: A survey of optimal control problems with state constraints. SIAM Rev 37, 181–218 (1995)
Ioffe, A.D., Levin, V.L.: Subdifferentials of convex functions. Trans Moscow Math Soc 26, 1–72 (1972)
Karatzas, I., Shreve, S.E.: Methods of Mathematical Finance. New York: Springer (1998)
Kolmogorov, A.N., Fomin, S.V.: Introductory Real Analysis. New York: Dover (1975)
Korn, R.: Some applications of \(L_{2}\)-hedging with a non-negative wealth process. Appl Math Financ 4, 65–79 (1997)
Korn, R., Trautmann, S.: Continuous-time portfolio optimization under terminal wealth constraints. ZOR Math Methods Oper Res 42, 69–92 (1995)
Labbé, C., Heunis, A.J.: Convex duality in constrained mean-variance portfolio optimization. Adv Appl Probab 39, 77–104 (2007)
Makowski, K., Neustadt, L.W.: Optimal control with mixed control-phase variable equality and inequality constraints. SIAM J Control Optim 12, 184–228 (1974)
Rockafellar, R.T.: Integrals which are convex functionals. Pac J Math 24, 525–539 (1968)
Rockafellar, R.T.: Convex Analysis. Princeton: Princeton University Press (1970)
Rockafellar, R.T.: Integrals which are convex functionals II. Pac J Math 39, 439–469 (1971)
Rockafellar, R.T.: State constraints in convex control problems of Bolza. SIAM J Control Optim 10, 691–715 (1972)
Rockafellar, R.T.: Conjugate Duality and Optimization (CBMS-NSF Ser No 16). Philadelphia: SIAM (1974)
Rockafellar, R.T., Wets, J.B.: Stochastic convex programming: singular multipliers and extended duality. Pac J Math 62, 507–522 (1976)
Schweizer, M.: Mean-variance hedging for general claims. Ann Appl Probab 2, 171–179 (1992)
Schweizer, M.: Approximating random variables by stochastic integrals. Ann Probab 22, 1536–1575 (1994)
Schweizer, M.: Approximation pricing and the variance-optimal martingale measure. Ann Probab 24, 206–236 (1996)
Xu, G.L., Shreve, S.E.: A duality method for optimal consumption and investment under short-selling prohibition. I. General market coefficients. Ann Appl Probab 2, 87–112 (1992)
Yosida, K., Hewitt, E.: Finitely additive measures. Trans Am Math Soc 72, 46–66 (1952)
Zaanen, A.C.: Introduction to Operator Theory in Riesz Spaces. Berlin: Springer (1997)
Acknowledgments
The research reported here was supported by NSERC (National Science and Engineering Research Council of Canada). I thank the editor and referees for suggestions which improved this paper.
Author information
Authors and Affiliations
Corresponding author
Appendix: Proofs
Appendix: Proofs
We collect in this appendix the proofs of several propositions in the main text. We shall need the following result (see Proposition I-1 on p.387 of Bismut (1973)):
Proposition 22
For \(X\equiv (X_0,\dot{X},\Lambda _X)\in \mathbb {B}\) and \(Y\equiv (Y_0,\dot{Y},\Lambda _Y)\in \mathbb {B}\) (recall Remark 4), put
Then \(\{(\mathbb {M}(X,Y)(t),\mathcal {F}_t),\,t\in [0,T]\}\) is a continuous martingale with \(\mathbb {M}(X,Y)(0)=0\).
Proof of Proposition 2
Fix some \(X \in \mathbb {B}- \mathbb {A}_{1}\). From (34) we have either \(X \not \in \mathbb {B}_{2}\) or \(X \not \in \mathbb {A}\). Suppose \(X \not \in \mathbb {B}_{2}\): then, from (34), we have \(P [ X(T) - B< u_{2}] > 0\) for each \(u_{2}\in L_{\infty }\), so that (26) gives \(F(X, u) = +\infty \) for all \(u \in \mathbb {U}\). On the other hand, when \(X \not \in \mathbb {A}\), then (26) again gives \(F(X,u) = +\infty \) for all \(u\in \mathbb {U}\). Since \(\langle (u_1,u_{2}), (Y,Z) \rangle \in \mathbb {R}\) for all \((u_1, u_{2}) \in \mathbb {U}\) and \((Y,Z)\in \mathbb {Y}\), we conclude from (33) that (i) \(K(X,(Y,Z)) = +\infty \), for all \(X \in \mathbb {B}- \mathbb {A}_{1}\) and \((Y,Z) \in \mathbb {Y}\). Now fix \(X \in \mathbb {A}_{1}\): from (33), (32), (26), for each \((Y,Z) \in \mathbb {Y}\) we have
From Remark 12 one has (ii) \(\inf _{\{u_{2}\in L_{\infty }, \; u_{2}\le X(T) - B\}} \left\{ Z(u_{2})\right\} = -\infty \), for all \(Z\in L_{\infty }^{*}\) such that \(Z\not \le 0\). Moreover, for each \((X,Y) \in \mathbb {B}\times \mathbb {B}_{1}\) we also have
and (36) follows from (108), (ii), (107) and (i). Note that the first equality at (108) follows upon defining \(v_1:=X(T) - u_1\), and using \(X(T) \in L_{2}\). The second equality at (108) follows from Thrm. 2 on p.532–533 of Rockafellar (1968), since \(L_{2}\) is decomposable (see p.532 of Rockafellar (1968)), while (35) and (8), together with Condition 2(iii), ensure that \(J(\cdot )\) and \(J^{*}(\cdot )\) are normal convex integrands (see Lemma 2 on p.528 of Rockafellar (1968)), and that \(J(u)\) and \(J^{*}(u)\) are \(P\)-integrable for each \(u\in L_{2}\). \(\square \)
Proof of Proposition 3
Fix some \((X,(Y,Z)) \in \mathbb {B}\times \mathbb {Y}\). Since \(\text {E}[X(T)Y(T)]\) and \(\text {E}J^{*}(Y(T))\) are \(\mathbb {R}\)-valued (recall Remark 16) from (39) and (36) we get
Moreover, from (15) and (36), we have \(f(X) = K(X,(Y,Z))\in \mathbb {R}\) \(\iff \) (1) \(X(T)-B\ge 0\), (2) \(X\in \mathbb {A}_{1}\), (3) \(Z\le 0\), (4) \(\text {E}[ J(X(T)) + J^{*}(Y(T)) - X(T)Y(T) ]\) \(= \inf _{\{u_{2}\in L_{\infty }, u_{2}\le X(T) - B\}} Z(u_{2})\) \(\iff \) \((1^\prime )\) \(X(T)-B\ge 0\), \((2^\prime )\) \(X\in \mathbb {A}_{1}\), \((3^\prime )\) \(Z\le 0\), \((4^\prime )\) \(\text {E}[ J(X(T)) + J^{*}(Y(T)) - X(T)Y(T) ] = 0\), \((5^\prime )\) \(\inf _{\{u_{2}\in L_{\infty }, u_{2}\le X(T) - B\}} Z(u_{2}) = 0\) (note that the second equivalence follows from (35), which ensures \(J(X(T)) + J^{*}(Y(T)) - X(T)Y(T) \ge 0\) \(\text {a.s.}\), together with the fact that \(X(T)-B\ge 0\) ensures \(\inf _{\{u_{2}\in L_{\infty }, u_{2}\le X(T) - B\}} Z(u_{2}) \le 0\)). Moreover, from the non-negativity of the integrand at \((4^\prime )\), we see that \((4^\prime )\) holds if and only if \(J(X(T)) + J^{*}(Y(T)) - X(T)Y(T) = 0\) \(\text {a.s.}\) if and only if \(X(T) = (\partial J^{*})(Y(T))\) (see Propos. I.5.1 and Cor. I.5.2 on p.21-22 of Ekeland and Témam (1976)). We then get
Moreover, from (15), we know that \(f(X) \in ( -\infty , \infty ]\), for all \(X \in \mathbb {B}\). Also, from (38) and \(\mathbb {A}_{1}\not = \emptyset \) (recall (34)) we have \(\varkappa (Y,Z) \in (-\infty ,\infty ]\), for all \((Y,Z)\in \mathbb {Y}\); since \(J^{*}(Y(T))\) is \(P\)-integrable for each \(Y\in \mathbb {B}_{1}\) (see Remark 16), from (39) we obtain \(g(Y,Z) \in [-\infty ,\infty )\), for all \((Y,Z)\in \mathbb {Y}\). Thus, for each \(X\in \mathbb {B}\) and each \((Y,Z)\in \mathbb {Y}\), we obtain from (40) the equivalence that \(f(X) = g(Y,Z)\) \(\iff \) \(f(X) = K(X,(Y,Z)) \in \mathbb {R}\) and \(g(Y,Z) = K(X,(Y,Z)) \in \mathbb {R}\). This equivalence, together with (110) and (109), gives (41). \(\square \)
Proof of Proposition 4
We have “separation of infima” for the regular and singular parts of \(Z\), namely
(recall (34)), as follows immediately from Thm. 1 of Rockafellar (1971). Consider the first term on the right of (111) when \(Z\le 0\), in which case (recall Remark 13) we have \(Z_{r}\le 0\) \(\text {a.s.}\) Fix some \(X \in \mathbb {B}_{2}\); then, from (34) and the \(P\)-essential boundedness of \(B\) (see Condition 2(iii)), both \(Z_{r}(X(T)-B)\) and \(Z_{r}X(T)\) are majorized by the integrable random variable \(\alpha Z_{r}\) (for some \(\alpha \in \mathbb {R}\); recall that \(Z_{r}\in L_{1}\)), thus \(\text {E}[ Z_{r}(X(T)-B)]\) and \(\text {E}[ Z_{r}X(T) ]\) exist in \([-\infty ,\infty )\) (see the remark just prior to Cor. 4.1.1 in Chow and Teicher (1988)), and (i) \(\text {E}[ Z_{r}(X(T)-B)] = \text {E}[ Z_{r}X(T)] - \text {E}[ Z_{r}B]\) (see Thrm. 4.1.1(v) of Chow and Teicher (1988)). Moreover, since \(Z_{r}\le 0\) \(\text {a.s.}\), and \(X(T) -B\) is \(\text {a.s.}\) lower-bounded by some \(\alpha \in \mathbb {R}\) (from (34)), it is easily seen from the monotone convergence theorem that one has (ii) \(\inf _{\{u_{2}\in L_{\infty }, u_{2}\le X(T) - B\}}\text {E}[Z_{r}u_{2}] = \text {E}[Z_{r}(X(T)-B)] = \text {E}[Z_{r}X(T) ] - \text {E}[ Z_{r}B]\) (the second equality following from (i)). From (ii) and (111) we get
(recall Notation 6). Upon combining (112), (39), (38), we obtain (42).
As for (43), this is a matter of showing equivalence of (43)(4) and (41)(4), since we know from Remark 13 that (43)(3) and (41)(3) are equivalent. Fix a pair \((X,(Y,Z))\in \mathbb {B}\times \mathbb {Y}\) such that (iii) \(X(T) - B\ge 0\) and (iv) \(Z\le 0\). From (iii), (iv) and Remark 13, we get that (v) \(\inf _{\{u_{2}\in L_{\infty }, u_{2}\le X(T) - B\}}Z^{\circ }(u_{2}) \le 0\) and \(\text {E}[ Z_{r}(X(T) - B)] \le 0\). From (v) and (112) we have \(\inf _{\{u_{2}\in L_{\infty }, u_{2}\le X(T) - B\}}Z(u_{2}) = 0\) \(\iff \) \(\text {E}[ Z_{r}(X(T) - B)] + \inf _{\{u_{2}\in L_{\infty }, u_{2}\le X(T) - B\}}Z^{\circ }(u_{2}) = 0\) \(\iff \) \(\inf _{\{u_{2}\in L_{\infty }, u_{2}\le X(T) - B\}}Z^{\circ }(u_{2}) = \text {E}[ Z_{r}(X(T) - B)] =0\) \(\iff \) \(\inf _{\{u_{2}\in L_{\infty }, u_{2}\le X(T) - B\}}Z^{\circ }(u_{2}) =0\) and \(Z_{r}(X(T) - B) =0\) \(\text {a.s.}\) (since, from (iii), (iv), and Remark 13, we have \(Z_{r}(X(T) - B) \le 0\)). This gives the equivalence of (43)(4) and (41)(4) as required. \(\square \)
Proof of Proposition 8
We evaluate the supremum at (46). In view of Remark 5(b), for each \(\pi \in \Pi _{2}\) we obtain (i) \(X^{\pi }\in \mathbb {B}\) with (ii) \(X^{\pi }(0) = x_0\), \(\dot{X}^{\pi } = r X^{\pi }+ \pi ^{\prime } \sigma \theta \), and \(\Lambda _{X^{\pi }}^{\prime } = \pi ^{\prime } \sigma \). From (ii) and (18), we get (iii) \(\mathbb {M}(X^{\pi },Y)(T) = X^{\pi }(T)Y(T) - x_0Y_{0} - \int _{0}^{T} \; \pi ^{\prime }(\tau ) \sigma (\tau ) [ \theta (\tau ) Y(\tau ) + \Lambda _{Y}(\tau ) ]\,\text {d}\tau \), for all \(\pi \in \Pi _{2}\) and \(Y\in \mathbb {B}_{1}\), and, from (i) and Proposition 22, we get (iv) \(\text {E}[ \mathbb {M}(X^{\pi },Y)(T) ] = 0\), for all \(\pi \in \Pi _{2}\) and \(Y\in \mathbb {B}_{1}\). It then follows from (iv) and (iii) that
From (7), (14), Notation 7(a), and (113), we get
for each \(Y\in \mathbb {B}_{1}\). We next evaluate the supremum at the right of (114). The indicator function \(\delta _{\mathbb {R}^{N}}(\cdot |K)\) (see Notation 7(a)) is a normal convex integrand by Lemma 1 on p.528 of Rockafellar (1968). From Notation 7(b), and the fact that \(0 \in K\), the functions \(\delta _{\mathbb {R}^{N}}(\vartheta (\cdot )|K)\) and \(\delta ^{*}_{\mathbb {R}^{N}}(\vartheta (\cdot )|K)\) are identically zero (hence \(\lambda \otimes P\)-integrable) on \([0,T]\times \Omega \), when \(\vartheta \) is the zero element of \(\Pi _{2}\). Since \(\Pi _{2}\) is decomposable (see p.532 of Rockafellar (1968)), and \(\delta ^{*}_{\mathbb {R}^{N}}(\cdot |K)\) is the convex conjugate of \(\delta _{\mathbb {R}^{N}}(\cdot |K)\), we have
(as follows upon identifying \(f(\cdot ):=\delta _{\mathbb {R}^{N}}(\cdot |K)\), and \(L = L^{*} :=\Pi _{2}\), in Thrm. 2 on p.532–533 of Rockafellar (1968)). Now fix \(Y\in \mathbb {B}_{1}\), and put \(\vartheta (t,\omega ) :=- \sigma (t,\omega ) [ \theta (t,\omega ) Y(t,\omega ) + \Lambda _{Y}(t,\omega )]\), \((t,\omega )\in [0,T]\times \Omega \); since \(\Lambda _{Y}\in \Pi _{2}\) (see (18), (12)), it follows from the boundedness of \(\theta \) (see Remark 4), the bound given by Remark 1, and Remark 5(a), that \(\vartheta \in \Pi _{2}\). Using this choice of \(\vartheta \) in (115), and then combining with (114), we obtain (46). \(\square \)
Proof of Proposition 9
Put \(\alpha _1 :=\inf _{X\in \mathbb {A}_{1}} \text {E}[X(T) Y(T)]\) and \(\alpha _2 :=\inf _{X\in \mathbb {A}} \text {E}\) \( [X(T) Y(T)]\) for some \(Y \in \mathbb {B}_{1}\). Since \(\mathbb {A}_{1}\subset \mathbb {A}\), (recall (34)), we have \(\alpha _1 \ge \alpha _2\). To show the opposite inequality fix some \(X \in \mathbb {A}\). From (14) and (7) we have (i) \(X = X^{\pi }\) for some \(\pi \in \mathcal {A}\). Define the \(\mathcal {F}_t\)-stopping times \(T_n:=\inf \{ t \in [0,T] \; | \; S_0^{-1}(t) X(t) \le -n\} \wedge T\). Since \(0 \in K\) (see Condition 2(i)) we have (ii) \(\pi _{n}\in \mathcal {A}\) for \(\pi _{n}\) defined by \(\pi _{n}(t) :=\pi (t) I\{ t \le T_n\}\). Now put (iii) \(X_n:=X^{\pi _{n}}\), for each \(n = 1,2,\ldots \) (recall (5)). Now (iv) \(S_0^{-1}(t) X_n(t) = S_0^{-1}(t\wedge T_n) X(t\wedge T_n)\), \(t\in [0,T]\), as follows from (ii), (iii) and (5). From the definition of \(T_n\) we find (v) \(S_0^{-1}(T_n) X(T_n) \ge - n\). Upon combining (iv) and (v), we see that \(S_0^{-1}(T) X_n(T) \ge -n\), and thus (vi) \(X_n(T) - B\ge -n S_0(T) - B\). Now \(S_0(T)\) and \(B\) are \(P\)-essentially bounded (from (2) with Conditions 1 and 2(iii)), so it follows from (vi) and (34) that (vii) \(X_n\in \mathbb {B}_{2}\). Moreover, from (iii) and (ii) we have \(X_n\in \mathbb {A}\), and from this with (vii) and (34) we obtain (viii) \(X_n\in \mathbb {A}_{1}\) for all \(n = 1,2,\ldots \). Now of course \(\lim _{n \rightarrow \infty }T_n= T\), and from this together with (iv) we get (ix) \(\lim _{n \rightarrow \infty }X_n(T) = X(T)\) a.s. and therefore (x) \(\lim _{n \rightarrow \infty }\text {E}[ X_n(T) Y(T) ] = \text {E}[ X(T) Y(T) ]\). To see how (x) follows put
From (iii), (ii) and (5), we have (xi) \(| X_n(T)| \le \zeta \) for all \(n = 1,2,\ldots \), and, from Condition 1, the bounds given by Remark 1, and Doob’s maximal \(L^{2}\)-inequality, it is straightforward to check that (xii) \(\zeta \in L_{2}\). Moreover, (xiii) \(Y(T) \in L_{2}\) (since \(Y\in \mathbb {B}_{1}\)), so it follows from (xii) and (xiii) that (xiv) \(\zeta |Y(T)| \in L_{1}\). In view of (xi) and (xiv), we have (xv) \(| X_n(T) Y(T) | \le \zeta |Y(T) | \in L_{1}\) for all \(n = 1,2,\ldots \), and (x) follows from (ix), (xv), and dominated convergence. Finally, from (viii), (x) and the fact that \(X\in \mathbb {A}\) is arbitrary, we obtain \(\alpha _1 \le \alpha _2\), as required. \(\square \)
Proof of Proposition 11
(a) follows upon expanding \(X(t) = H^{-1}(t) [ X(0) + \int _{0}^{t}\; \psi ^{\prime }(\tau ) \,\text {d}W(\tau )]\) by Itô’s formula and using Remark 25(b). An argument which is identical to that for Lemma 5.1 of Labbé and Heunis (2007) shows that \(\text {E}[ \sup _{t\in [0,T]} |X(t)|^{2}] < \infty \), and this fact, together with an argument identical to that for Lemma 5.2 of Labbé and Heunis (2007), establishes \(\pi \in \Pi _{2}\); now \(X \in \mathbb {B}\) follows from this, together with Condition 1 and the expansion of \(\,\text {d}X(t)\) given by (a). \(\square \)
Proof of Proposition 19
Fix \((Y,Z) \in \mathbb {Y}\) with \(Z\le 0\), \(Z^{\circ }= 0\), and \(Z_{r}\in L_{1}- L_{2}\). Then, from Remark 13, we have (i) \(Z_{r}\le 0\) \(\text {a.s.}\). Now put (ii) \(\zeta _{n}:=(-Z_{r}) \wedge n \ge 0\) \(\text {a.s.}\) for all \(n = 1,2,\ldots \) Then (iii) \(\zeta _{n}\in L_{\infty }\subset L_{2}\), and, since \(Z_{r}\not \in L_{2}\), the monotone convergence theorem gives (iv) \(\lim _{n \rightarrow \infty }\text {E}[\zeta _{n}^{2}] = \infty \). Now put (v) \(\left|| \zeta _{n}\right||_{2} :=( \text {E}[\zeta _{n}^{2}])^{1/2}\) and \(\eta _{n}:=\zeta _{n}/ \left|| \zeta _{n}\right||_{2} \ge 0\), for all \(n=1,2,\ldots \) Then, from (iii), we have that (vi) \(\eta _{n}\in L_{\infty }\subset L_{2}\) with \(\left|| \eta _{n}\right||_{2} :=( \text {E}[\eta _{n}^{2}])^{1/2} = 1\), for all \(n = 1,2,\ldots \) Next, identifying \(\xi \) with \(\eta _{n}\) at (v), use Proposition 11 to define the \(\mathbb {R}\)-valued process \(X_{n}\), and the \(\mathbb {R}^{N}\)-valued processes \(\psi _n \in {\mathcal {F}}^{*}\), \(\pi _n \in {\mathcal {F}}^{*}\). Then (vii) \(X_{n}(T) = \eta _{n}\ge 0\), and (viii) \(X_{n}(0) = \text {E}[\eta _{n}H(T)]\). From Proposition 11(a), we have
Next define the \(\mathbb {R}\)-valued process \(\hat{X}_n\) by
From (117), (116) and (viii), we get (ix) \(\hat{X}_n(T) - X_n(T) = S_0(T) [x_0- \text {E}[\eta _{n}H(T)]]\). In light of (ix) and (vii) we see that (x) \(\hat{X}_n(T) \ge S_0(T) [x_0- \text {E}[\eta _{n}H(T)]]\). Since \(S_0(T)\) and \(B\) are \(P\)-essentially bounded (as follows from Condition 1 and Condition 2(iii)), from (x) we get (xi) \(\hat{X}_n \in \mathbb {B}_{2}\) (recall (34)). Moreover, since \(K = \mathbb {R}^{N}\), we have \(\mathcal {A}= \Pi _{2}\) (recall (7)); but \(\pi _n \in \Pi _{2}\) (from Proposition 11(b)), so it follows from (117) that (xii) \(\hat{X}_n \in \mathbb {A}\) (recall (14)). In view of (xii), (xi) and (34), we obtain \(\hat{X}_n \in \mathbb {A}_{1}\), and therefore
From (vii) and (ix) we find that the right side of (118) is given by
From (vi) and the Cauchy-Schwarz inequality we get (xiii) \(| \text {E}[ \eta _{n}H(T)] | \le (\text {E}[H^{2}(T)])^{1/2} < \infty \) and (xiv) \(| \text {E}[ \eta _{n}Y(T)] | \le ( \text {E}[Y^{2}(T)])^{1/2} < \infty \). Moreover, (xv) \(\text {E}[ \eta _{n}Z_{r}] = \text {E}[ \zeta _{n}Z_{r}] / \left|| \zeta _{n}\right||_{2} \le - \text {E}[ \zeta _{n}^{2} ] / \left||\zeta _{n}\right||_{2} = - \left|| \zeta _{n}\right||_{2}\) (the first equality at (xv) follows from (v), while the inequality follows from (ii)), and then, from (xv) and (iv), we obtain (xvi) \(\lim _{n \rightarrow \infty }\text {E}[ \eta _{n}Z_{r}] = -\infty \). From (xvi), (xiv) and (xiii), one sees that the right side of (119) must converge to \(-\infty \) as \(n \rightarrow \infty \); but the quantity on the right of (119) is an upper-bound for the infimum on the left of (118) for each \(n = 1,2,\ldots \), and thus
Since \(Z^{\circ }= 0\), from (120) and (42) we get \(g(Y,Z) = - \infty \), as required. \(\square \)
Proof of Proposition 20
The function \(\alpha \rightarrow \tilde{h}(\alpha H(T,\omega ),\omega )\) is concave and continuous for \(\alpha \in \mathbb {R}\) (see (89)), thus, from dominated convergence, it is easily seen that (i) \(\alpha \rightarrow \varPsi (\alpha )\) is continuous and clearly concave on \(\alpha \in \mathbb {R}\). From (89) and (86) we have \(\tilde{h}(y,\omega ) \le h((c + a B)(\omega ), \omega )\) for all \(y \in \mathbb {R}\), and from this with Condition 2(iii) and (35), we get (ii) \(\text {E}\tilde{h}(\alpha H(T) ) \le \text {E}h( c + a B) < \infty \) for all \(\alpha \in \mathbb {R}\). From Condition 4, with \(K :=\mathbb {R}^{N}\), there is some \(\tilde{\pi }\in \Pi _{2}\) and \(\epsilon \in (0,\infty )\) such that (iii) \(X^{\tilde{\pi }}(T) H(T) \ge BH(T) + \epsilon H(T)\). But \(\{ X^{\tilde{\pi }}(t) H(t) \}\) is a \(\mathcal {F}_t\)-martingale (from (53), (4) and Itô’s formula), thus \(\text {E}[ X^{\tilde{\pi }}(T) H(T) ] = x_0\), and then from (iii) we get (iv) \(x_0> \text {E}[ BH(T) ]\). Now it follows from (iv), (ii), and (90) that (v) \(\lim _{\alpha \rightarrow - \infty } \varPsi (\alpha ) = - \infty \). Since \(H(T) > 0\) \(\text {a.s.}\) we have (vi) \(P [ \alpha _0 H(T) \ge c + a B] > 0\) for some \(\alpha _{0} > 0\). Moreover, from (89), (86), (35), and Condition 2(iii), it is easily seen that there are some non-negative constants \(k_1, k_2 \in \mathbb {R}\) such that (vii) \(\text {E}\tilde{h}(\alpha H(T)) \le k_1 + | \alpha | k_2 - (\alpha ^{2} / 2) \text {E}[ H^{2}(T) / a ; \; \alpha _0 H(T) \ge c + aB]\) for all \(\alpha \ge \alpha _{0}\). From (vi) we also have (viii) \(\text {E}[ H^{2}(T) / a ; \; \alpha _0 H(T) \ge c + aB] > 0\). Now it follows from (viii), (vii) and (90), that (ix) \(\lim _{\alpha \rightarrow \infty } \varPsi (\alpha ) = - \infty \), and existence of a maximizer \(\bar{\alpha }\in \mathbb {R}\) of \(\varPsi (\cdot )\) follows from (ix), (v), and (i). \(\square \)
Rights and permissions
About this article
Cite this article
Heunis, A.J. Quadratic minimization with portfolio and terminal wealth constraints. Ann Finance 11, 243–282 (2015). https://doi.org/10.1007/s10436-014-0254-9
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10436-014-0254-9
Keywords
- Portfolio optimization
- Stochastic control
- Conjugate duality
- Constraints
- Lagrange multiplier
- Slater condition