Abstract
This paper is concerned with an infinite-horizon problem of optimal investment and consumption with proportional transaction costs in continuous-time regime-switching models. An investor distributes his/her wealth between a stock and a bond and consumes at a non-negative rate from the bond account. The market parameters (the interest rate, the appreciation rate, and the volatility rate of the stock) are assumed to depend on a continuous-time Markov chain with a finite number of states (also known as regimes). The objective of the optimization problem is to maximize the expected discounted total utility of consumption. We first show that for a class of hyperbolic absolute risk aversion utility functions, the value function is a viscosity solution of the Hamilton–Jacobi–Bellman equation associated with the optimization problem. We then treat a power utility function and generalize the existing results to the regime-switching case.
Similar content being viewed by others
References
Magill, M.J.P., Constantinides, G.M.: Portfolio selection with transaction costs. J. Econ. Theor. 13, 245–263 (1976)
Davis, M.H.A., Norman, A.R.: Portfolio selection with transaction costs. Math. Oper. Res. 15, 676–713 (1990)
Shreve, S.E., Soner, H.M.: Optimal investment and consumption with transaction costs. Ann. Appl. Probab. 4, 609–692 (1994)
Zariphopoulou, T.: Investment-consumption models with transaction fees and Markov-chain parameters. SIAM J. Control Optim. 30, 613–636 (1992)
Crandall, M.G., Ishii, H., Lions, P.L.: User’s guide to viscosity solutions of second order partial differential equations. Bull. Am. Math. Soc. 27, 1–67 (1992)
Rishel, R.: Whether to sell or hold a stock. Commun. Inf. Syst. 6, 193–202 (2006)
Dai, M., Zhang, Q., Zhu, Q.J.: Trend following trading under a regime switching model. SIAM J. Financ. Math. 1, 780–810 (2010)
Acknowledgements
The author would like to thank the two anonymous referees and the editors for their valuable comments, which helped to improve the exposition of this paper.
Author information
Authors and Affiliations
Corresponding author
Additional information
Communicated by Vladimir Veliov.
Appendix
Appendix
Proof of Theorem 3.1
In view of Remark 2.1 and Corollary 3.1, it suffices to show that for each \(i\in \mathcal {M}\), V(x,y,i) has limit 0 at any point (x 0,y 0)∈∂Π. Using (6), we obtain V≤K 1 V p where V p denotes the value function for the power utility function U(c)=c γ/γ where γ is specified in (6) and K 1 is some positive constant. In view of the upper bound property established in Proposition 5.3, we have
for some constants K 2,K 3>0, where D 0 and E 0 are defined in (32) and (43), respectively.
Given (x 0,y 0)∈∂ 1 Π and (x 0,y 0)≠(0,0). Consider an δ-neighborhood of (x 0,y 0): B δ (x 0,y 0)={(x,y):|(x,y)−(x 0,y 0)|<δ} for some δ>0. Then for any point \((x,y)\in B_{\delta}(x_{0},y_{0})\cap\overline{D}_{0}\), we have 0≤V(x,y,i)≤K 2[x+(1−μ)y]γ. Sending (x,y)→(x 0,y 0), we obtain
Similarly, we can show that (83) holds at any point (x 0,y 0)∈∂ 2 Π∖{(0,0)}.
Finally, we consider the origin (0,0). For any point \((x,y)\in B_{\delta}(0, 0)\cap\overline{\varPi}\) where B δ (0,0)={(x,y):|(x,y)|<δ} for some δ>0, we consider three cases:
(I) If \((x,y)\in\overline{D}_{0}\), then
(II) If \((x,y)\in\overline{E}_{0}\), then
(III) If \((x,y)\in\varPi\setminus(\overline{D}_{0}\cup\overline{E}_{0})\), then in view of the homotheticity property (25) of the function V p , we have
Note that \(( x/\sqrt{x^{2}+y^{2}}, y/\sqrt{x^{2}+y^{2}})\) is on the arc segment of the unit circle r=1, −arctan(1/ξ 0)≤θ≤π−arctan(1/δ 0), which is a closed subset in Π. In view of Corollary 3.1, V p is continuous on this arc segment and therefore bounded. It follows that
for some constant K 4>0. Sending (x,y)→(0,0) in (84), (85), and (87), we obtain
This completes the proof. □
Proof of the ≤ part of Proposition 5.4
We only show the inequality for x≤0 since that for x≥0 can be proved analogously and therefore omitted. For this purpose, we introduce a sequence of closed subsets in Π. For n=1,2,… , define
Given (x 0,y 0)∈Π, \(i\in \mathcal {M}\), and \((C,M,N)\in \mathcal {A}(x_{0},y_{0},i)\), define
and
where (x(t),y(t)) are given by (3). It can be shown as in [3, Lemma 5.6] that ν n ↑ν a.s. as n→∞.
We consider φ defined in (55). It is readily seen that φ(x,y,i)=V(x,y,i), ∀(x,y)∈∂D, \(i\in \mathcal {M}\). We extend the definition of φ to \(\overline{\varPi}\) by letting φ=V on \(\overline{\varPi }\setminus \overline{D}\).
Next, we introduce a sequence of closed subsets in \(\overline{D}\). For n=1,2,… , define
Let (x 0,y 0)∈D. Define
and
Then τ n ↑τ a.s. as n→∞. Moreover, similar to [3, Proposition 5.7], we have
Now consider 0≤t<n∧ν n ∧τ n . Note that (57) implies that for each \(i\in \mathcal {M}\), the function φ i is non-increasing in the jump directions (1−μ,−1) and (−(1+λ),1) at any point in D. Thus,
Applying the Itô formula to e −βt φ(x(t),y(t),α t ) for 0≤t≤n∧ν n ∧τ n , noting that φ(x(t),y(t),α t ) and y(t)φ y (x(t),y(t),α t ) are bounded on [0,n∧ν n ∧τ n [, we obtain
In view of (14) and (58), we have, for t∈]0,n∧ν n ∧τ n [,
Using (98) and the fact that φ=V on \(\overline {\varPi}\setminus\overline{D}\), we get from (97)
Note that n∧ν n ∧τ n →ν∧τ as n→∞. We have two cases. (I) If τ(ω)<∞, then we must have τ(ω)<ν(ω). Otherwise the state (x(t,ω),y(t,ω)) would reach the origin (0,0) and stay there forever and hence never exit \(\overline{D}\), resulting in τ(ω)=∞, a contradiction! (II) If τ(ω)=∞, then ν(ω)∧τ(ω)=ν(ω). For this case, we have
where the first equality is due to the fact that C(t,ω)=0 for t>ν(ω). It follows by using the monotone convergence theorem that
On the other hand, we can show that \(I_{\{\nu_{n}\wedge\tau_{n}=\tau \leq n\}} \uparrow I_{\{\tau<\infty\}}\) as n→∞. Again by the monotone convergence theorem we have
Therefore we obtain
Taking the supremum of the RHS of (100) over \((C, M, N)\in \mathcal {A}(x_{0}, y_{0}, i)\) and using (19), we obtain φ(x 0,y 0,i)≥V(x 0,y 0,i). □
Partition of the Region Π by Convex Analysis
For each (x,y)∈Π and each \(i\in \mathcal {M}\), define the sub-differential of V by
Since V i (x,y) is concave and finite on Π, ∂V(x,y,i) is a nonempty, compact and convex subset of \(\mathbb{R}^{2}\). The function V i (x,y) is differentiable at the point (x,y)∈Π iff ∂V(x,y,i) is a singleton. In this case we have ∂V(x,y,i)={V x (x,y,i),V y (x,y,i)}.
The following two results can be established by applying [3, Lemma 6.1 and Proposition 6.2] to each function V i , \(i\in \mathcal {M}\).
Lemma A.1
Given \(i\in \mathcal {M}\). Let {(x n ,y n ),n≥1} be a sequence of points in Π with limit (x 0,y 0)∈Π. If \((\delta^{i,n}_{x}, \delta^{i,n}_{y})\in \partial V(x_{n},y_{n},i)\) for every n≥1, then the sequence \(\{ (\delta ^{i,n}_{x}, \delta^{i,n}_{y}), n\geq1\}\) is bounded and its every limit point is in ∂V(x 0,y 0,i).
Proposition A.1
Let O be an open subset of Π. Then V i is of C 1 in O if and only if ∂V(x,y,i) is a singleton for every (x,y)∈O.
Given \(i\in \mathcal {M}\), (x,y)∈Π, and \((\delta^{i}_{x}, \delta^{i}_{y})\in \partial V(x,y,i)\). We define function \(\phi(\tilde{x}, \tilde{y},i)= V(x,y,i)+\delta^{i}_{x}(\tilde{x}-x)+\delta^{i}_{y}(\tilde{y}-y), (\tilde {x}, \tilde{y})\in\varPi\). Then \(\phi(\tilde{x}, \tilde{y},i)\geq V(\tilde{x}, \tilde{y},i), \forall(\tilde{x}, \tilde{y})\in\varPi\). For each h>0 satisfying (x−(1−μ)h,y+h)∈Π from which (x,y) can be reached by a transaction, by Proposition 3.3, we have
It follows that
Similarly, we can show
Since for (x,y)∈Π, x+(1−μ)y>0 and x+(1+λ)y>0, we have V(x,y,i)>0 in view of Proposition 5.2. It follows from Corollary 5.1 that \(\delta^{i}_{x}=\phi_{x}(x,y)>0\), and then from (102) that \(\delta^{i}_{y}>0\). To summarize, we have
For θ sufficiently close to 1, using the homotheticity property (25), we have
Dividing both sides of (105) by θ−1, sending θ→1+ and θ→1−, respectively, we obtain
Now, for \((\tilde{x}, \tilde{y})\in\varPi\), θ>0, we have
It follows that \((\theta^{\gamma-1}\delta^{i}_{x}, \theta^{\gamma -1}\delta ^{i}_{y}) \in\partial V(\theta x, \theta y, i)\). Thus
The reverse set containment can be established by simply replacing θ by \(\frac{1}{\theta}\), x by θx and y by θy. Therefore we have
Using the sub-differentials, we can partition the solvency region Π into three convex cones. First, for \(i\in \mathcal {M}\), (x,y)∈Π, \((\bar {x},\bar{y})\in\varPi\), \((\delta^{i}_{x}, \delta^{i}_{y}) \in\partial V(x,y,i)\) and \((\delta^{i}_{\bar{x}}, \delta^{i}_{\bar{y}}) \in\partial V(\bar {x},\bar{y},i)\), in view of (101), we have
It follows that
For \(i\in \mathcal {M}\), (x,y)∈Π, define
Note that the maxima and minima in (109) are attained since ∂V(x,y,i) is compact. In addition, ϑ +(x,y,i)≥ϑ −(x,y,i)≥0 due to (102).
Consider the parametric equations of the half line L 1 originating at the point (1+λ,−1) on ∂ 2 Π and parallel to ∂ 1 Π, defined by
Define
where \(\rho_{0}^{i}=\infty\) if the above set is empty.
Lemma A.2
Given \(i\in \mathcal {M}\). For \(0<\rho<\bar{\rho}<\infty\),
If \(0<\rho_{0}^{i}<\infty\), then \(\vartheta^{-}(x(\rho_{0}^{i}),y(\rho _{0}^{i}),i)=0\) and
Let \(i\in \mathcal {M}\) be fixed. We partition Π into two open, convex (possibly empty) cones as follows:
Proposition A.2
Given \(i\in \mathcal {M}\). We have
In a similar way, we introduce the parametric equations of the half line L 2 originating at the point (−(1−μ),1) on ∂ 1 Π and parallel to ∂ 2 Π:
Define
Let \(i\in \mathcal {M}\) be fixed. We partition Π into two open, convex (possibly empty) cones as follows:
Proposition A.3
Given \(i\in \mathcal {M}\). We have
Corollary A.1
SA i∩BU i=∅, \(\forall i \in \mathcal {M}\).
Corollary A.2
For each \(i \in \mathcal {M}\), the value function V i is C 1 in SA i∪BU i.
Note that Corollary A.2 implies that −(1−μ)V x (x,y,i)+V y (x,y,i)=0 for (x,y)∈SA i and (1+λ)V x (x,y,i)−V y (x,y,i)=0 for (x,y)∈BU i. In view of Proposition 5.4, we conclude that both SA i and BU i are nonempty, since the two equations are satisfied by V in D and E, respectively, where D and E are defined by (54) and (60), respectively. Clearly, SA i⊃D and BU i⊃E. Moreover, it follows from these equations, the homotheticity of the value function, and the functions presented in Proposition 5.4 that
where A i , B i are given by (56) and (62), respectively.
For each \(i\in \mathcal {M}\), let \(NT^{i}=\varPi\setminus\overline{SA^{i}\cup BU^{i}}\). Then we have
Proposition A.4
Given \(i\in \mathcal {M}\). we have
Rights and permissions
About this article
Cite this article
Liu, R. Optimal Investment and Consumption with Proportional Transaction Costs in Regime-Switching Model. J Optim Theory Appl 163, 614–641 (2014). https://doi.org/10.1007/s10957-013-0445-y
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10957-013-0445-y