Abstract
We consider an individual or household endowed with an initial wealth, having an income and consuming goods and services. The wealth development rate is assumed to be a deterministic continuous function of time. The objective is to maximize the discounted consumption over a finite time horizon. Via the Hamilton–Jacobi–Bellman approach, we prove the existence and the uniqueness of the solution to the considered problem in the viscosity sense. Furthermore, we derive an algorithm for explicit calculation of the value function and optimal strategy. It turns out that the value function is in general not continuous. The method is illustrated by two examples.
Similar content being viewed by others
References
Fleming, W.H., Rishel, R.W.: Deterministic and Stochastic Optimal Control. Springer, New York (1975)
Bertsekas, D.P.: Dynamic Programming: Deterministic and Stochastic Models. Prentice Hall, New York (1987)
Bardi, M., Capuzzo-Dolcetta, I.: Optimal Control and Viscosity Solutions of Hamilton–Jacobi–Bellman Equations. Birkhäuser, Boston (1997)
Schmidli, H.: Stochastic Control in Insurance. Springer, London (2008)
Giannessi, F., Maugeri, A. (eds.): Variational Analysis and Applications. Springer, New York (2005)
Aoki, M.: Optimal Control and System Theory in Dynamic Economic Analysis. North Holland, New York (1976)
Bryson, A.E. Jr., Ho, Y.: Applied Optimal Control. Blaisdell, Waltham (1969)
Chow, G.C.: Analysis and Control of Dynamic Economic Systems. Wiley, New York (1975)
Dixit, A.K.: Optimization in Economic Theory. Oxford University Press, Oxford (1990)
Dorfman, R.: An economic interpretation of optimal control theory. Am. Econ. Rev. 59(5), 817–831 (1969)
Macki, J.W., Strauss, A.: Introduction to Optimal Control Theory. Springer, New York (1982)
Intriligator, M.D.: Applications of optimal control theory in economics. Synthese 31, 271–288 (1975)
Kendrick, D.A.: Stochastic Control for Economic Models. McGraw-Hill, New York (1981)
Kendrick, D.A.: Stochastic control for economic models: past, present and the paths ahead. J. Econ. Dyn. Control 29, 3–30 (2005)
Murray, R.M. (ed.): Control in an Information Rich World: Report of the Panel on Future Directions in Control, Dynamics, and Systems. SIAM, Philadelphia (2003)
Neck, R.: Control theory and economic policy: balance and perspectives. Annu. Rev. Control 33(1), 79–88 (2009)
Moyer, H.G.: Deterministic Optimal Control: an Introduction for Scientists. Trafford, Victoria (2004)
Seierstad, A., Sydsæter, K.: Optimal Control Theory with Economic Applications. North-Holland, Amsterdam (1987)
Weber, T.A.: Optimal Control Theory with Applications in Economics. MIT Press, Cambridge (2011). Foreword by A.V. Kryazhimskiy
Karatzas, I., Lehoczky, J.P., Sethi, S.P., Shreve, S.E.: Explicit solution of a general consumption/investment problem. Math. Oper. Res. 11, 261–294 (1986)
Karatzas, I., Lehoczky, J.P., Shreve, S.E.: Optimal portfolio and consumption decisions for a “small investor” on a finite horizon. SIAM J. Control Optim. 25, 1157–1586 (1987)
Cox, J.C., Huang, C.F.: Optimal consumption and portfolio policies when asset prices follow a diffusion process. J. Econ. Theory 49, 33–83 (1989)
Avanzi, B.: Strategies for dividend distribution: a review. N. Am. Actuar. J. 13, 217–251 (2009)
Albrecher, H., Thonhauser, S.: Optimality results for dividend problems in insurance. RACSAM Rev. R. Acad. Cienc. Exactas Fis. Nat. Ser. A Mat. 103(2), 295–320 (2009)
Grandits, P.: Optimal consumption in a Brownian model with absorption and finite time horizon. Appl. Math. Optim. 67(2), 197–241 (2013)
Barron, E.N., Jensen, R.: Semicontinuous viscosity solutions for Hamilton–Jacobi equations with convex Hamiltonians. Commun. Partial Differ. Equ. 15(12), 293–309 (1990)
Evans, L.C.: Partial Differential Equations. Am. Math. Soc., Providence (1998)
Fleming, W.H., Soner, H.M.: Controlled Markov Processes and Viscosity Solutions. Springer, New York (1993)
Sontag, E.D.: Mathematical Control Theory. Springer, New York (1990)
Zabczyk, J.: Mathematical Control Theory: an Introduction. Birkhäuser, Boston (1992)
Sulem, A.: Explicit solution of a two-dimensional deterministic inventory problem. Math. Oper. Res. 11(1), 134–146 (1986)
Sulem, A.: Résolution explicite d’Inéquations quasi-variationelles associées à des problèmes de gestion de stock. Thèse 3e cycle. Paris IX-Dauphine (1983)
Sulem, A.: A solvable one-dimensional model of a diffusion inventory system. Math. Oper. Res. 11, 126–133 (1984)
Amman, H.: Numerical methods for linear-quadratic models. In: Handbook of Computational Economics, vol. 1, pp. 587–618. North-Holland, Amsterdam (1996)
Bellman, R.: An Introduction to the Theory of Dynamic Programming. Rand, Santa Monica (1953)
Crandall, M.G., Ishii, H., Lions, P.: User’s guide to viscosity solutions of second order partial differential equations. Bull. Am. Math. Soc. 27(1), 1–67 (1992)
Paxton, C.: Consumption and income seasonality in Thailand. J. Polit. Econ. 101(1), 39–72 (1993)
Lutz, G.A.: Business Cycle Theory. Oxford University Press, Oxford (2002)
Mnif, M., Sulem, A.: Optimal risk control and dividend policies under excess of loss reinsurance. Stochastics 77(5), 455–476 (2005)
Acknowledgements
The authors thank anonymous referees for their comments and suggestions which helped significantly to improve the presentation of the paper.
The research of the first two authors was supported by the Austrian Science Fund, grant P22449-N18.
Author information
Authors and Affiliations
Corresponding author
Additional information
Communicated by Vladimir Veliov.
Appendix
Appendix
1.1 HJB Equation—Viscosity Solution
Proof of Theorem 3.1
The proof of Theorem 3.1 consists of two steps: the super- and subsolution proofs. The following lemma presents the subsolution direction; the supersolution proof does not include subtleties and follows standard techniques, see Mnif and Sulem [39]. □
Lemma A.1
The function V(t,x) given by (1) is a viscosity subsolution to (13).
Proof
We need to show, that for every \(\bar{\psi}\in C^{(1,1)}[0,T]\times[0,\infty[\):
at every \((\bar{t},\bar{x})\in\,]0,T[\times]0,\infty[\) which is a (strict) maximizer of \(V^{*}-\bar{\psi}\) on [0,T]×[0,∞[ with \(V^{*}(\bar {t},\bar{x})=\bar{\psi}(\bar{t},\bar{x})\).
As usual, the subsolution proof is done by way of contradiction. Suppose there are some \((\bar{t},\bar{x})\) and \(\bar{\psi}\) with the properties stated before but with
for some ξ>0. Consider the function \(\psi(t,x)=\bar{\psi}(t,x)+\frac{(x-\bar{x})^{2}+(t-\bar {t})^{2}}{\bar{t}^{2}+\bar{x}^{2}}\xi\).
Then it holds that \(V^{*}(\bar{t},\bar{x})=\bar{\psi}(\bar{t},\bar {x})=\psi(\bar{t},\bar{x}),\bar{\psi}_{x}(\bar{t},\bar{x})=\psi _{x}(\bar{t},\bar{x})\) and \(\bar{\psi}_{t}(\bar{t},\bar{x})=\psi _{t}(\bar{t},\bar{x})\), which gives
Since ψ(t,x) is continuously differentiable in both t and x and μ t is continuous, there is \(\delta\in\,]0,\frac{\sqrt{\bar{t}^{2}+\bar{x}^{2}}}{2}[\) such that
for \((t,x)\in B_{\delta}(\bar{t},\bar{x})\). We obtain \(V^{*}(t,x)\leq\bar{\psi}(t,x)=\psi(t,x)-\frac{\delta^{2}}{\bar {t}^{2}+\bar{x}^{2}}\xi\) for \((t,x)\in\partial B_{\delta}(\bar{t},\bar {x})\).
Let now \(\varepsilon=\frac{1}{2}\frac{\delta^{2}}{\bar{t}^{2}+\bar {x}^{2}}\xi\), then on \(B_{\delta}(\bar{t},\bar{x})\) we have
while for \((t,x)\notin B_{\delta}(\bar{t},\bar{x})\) we have V ∗(t,x)≤ψ(t,x)−2ε.
Now let \((t_{n},x_{n})\to(\bar{t},\bar{x})\) such that \(V(t_{n},x_{n})\to V^{*}(\bar{t},\bar{x})\) and assume (w.l.o.g.) that \((t_{n},x_{n})\in B_{\delta}(\bar{t},\bar{x})\) for all \(n\in\mathbb{N}\).
Let \(C^{n}\in\mathcal{C}(t_{n},x_{n})\), \(X^{C^{n}}\) be the corresponding wealth starting in (t n ,x n ) and \(\tau^{*}=\tau_{n}\wedge\bar{T}\) (with some \(\bar{T}\) such that \(t_{n}<\bar{T}\leq T\) for all n) where
At first we observe that \(X^{C^{n}}\) can only have downward jumps in the x direction and that V(t,x) is increasing in x. Because of continuity of μ t , jumps in the wealth process are due to jumps in the consumption process, and we have \(X_{s}^{C^{n}}-X_{s-}^{C^{n}}=-\Delta C^{n}_{s}\).
Suppose τ ∗=τ n , i.e. stopping because of leaving \(B_{\delta }(\bar{t},\bar{x})\). Then either we hit the boundary continuously or leave the ball due to a jump at time τ ∗ in which case \(X_{\tau ^{*}-}\in B_{\delta}(\bar{t},\bar{x})\). From the above estimates we get \(V(\tau^{*},X_{\tau^{*}-}^{C_{n}})\leq\psi (\tau^{*},X_{\tau^{*}-}^{C_{n}})-2\varepsilon I_{\{X_{\tau^{*}-}=X_{\tau ^{*}}\}}\).
If \(\tau^{*}=\bar{T}\) then \(X_{\tau^{*}}^{C_{n}}\), as well as \(X_{\tau ^{*}-}^{C_{n}}\), is still inside the ball and we have \(V(\tau^{*},X_{\tau^{*}-}^{C_{n}})\leq\psi(\tau^{*},X_{\tau^{*}-}^{C_{n}})\). In total, we arrive at
In the above formula C n,c denotes the continuous part of strategy C n.
If \(X_{s}^{C_{n}}\neq X_{s-}^{C^{n}}\), we have that \(X_{s}^{C^{n}}- X_{s-}^{C^{n}}=-\Delta C_{s}^{n}\) and we can write
Combining the last expression with the continuous part of C n and using e −βt≤ψ x (t,x) on \(B_{\delta}(\bar{t},\bar {x})\), we arrive at
Finally, using \(\psi_{t}(s,X_{s}^{C^{n}})+\mu_{s} \psi_{x}(s,X_{s}^{C^{n}})\leq -\varepsilon\) from (15) we get
which is the same as
Since also \(\psi(t_{n},x_{n})\to V^{*}(\bar{t},\bar{x})\) if n→∞, we can choose n large enough such that \(0\leq\psi(t_{n},x_{n})-V(t_{n},x_{n})\leq\frac{(\tau^{*}-t_{n}) \varepsilon +2\varepsilon I_{\{\tau_{n}=\tau^{*}\wedge X_{\tau^{*}-}=X_{\tau^{*}}\}}}{2}\). We get
Now, before we can state that (16) is a contradiction to (12), we have to discuss the case τ ∗=τ n =t n , where an immediate lump-sum consumption leads to
We notice that property (P6) and the construction of the value function show that in this case V is continuously differentiable around the point \((\bar{t},\bar{x})\) and \(V_{x}(\bar{t},\bar{x})=e^{-\beta \bar{t}}\). Therefore V ∗=V around \((\bar{t},\bar{x})\), which furthermore yields that \(\psi_{x}(\bar{t},\bar{x})=e^{-\beta \bar{t}}\). Since the test function ψ is continuously differentiable in x, inequality (15) cannot be true and states a contradiction to (14).
Thus we have, when stating (14), that there exists an area around \((\bar{t},\bar{x})\) inside which it is not optimal to consume a lump sum from the wealth. Consequently a strategy C n, for playing a role in the dynamic programming principle for n large enough such that (t n ,x n ) are inside this non-paying area, has the feature that \(X_{t}^{C^{n}}\) can leave \(B_{\delta}(\bar{t},\bar {x})\) through a jump not before leaving the non-paying area continuously.
Therefore τ ∗>t n , which completes the proof and we can conclude that V(t,x) is a viscosity solution to (13). □
Proof of Lemma 3.1
Since V(t,x)≥e −βt x for (t,x)∈[0,T]×[0,∞[ (you can always pay out everything and quit by consuming a small constant rate such that X t+<0), we also have V ∗(t,x)≥e −βt x and V ∗(t,x)≥e −βt x. From e −βT x=V(T,x)≥V ∗(T,x) we get V ∗(T,x)=e −βT x.
Assume that V ∗(T,x)>e −βT x, then there exists η>0 with V ∗(T,x)≥2η+e −βT x.
Now choose a sequence (t n ,x n )→(T,x) such that V(t n ,x n )→V ∗(T,x). There is some n 0>0 such that for n≥n 0 we have
Let \(C^{n}\in\mathcal{C}(t_{n},x_{n})\) and define \(\tau_{n}=\inf\{t\geq t_{n}\mid X_{t}^{C^{n}}<0\}\wedge T\). Since C n is admissible, we have \(\Delta C_{t_{n}}^{n}\leq x_{n}\), there is no lump-sum payment leading to ruin. Furthermore, by the definition of t n , we have τ n −t n →0 if n→∞. Now, fix some ε>0 and choose n large enough such that
Taking a supremum over strategies C n we get \(V(t_{n},x_{n})\leq e^{-\beta t_{n}} x_{n}+\varepsilon\), which contradicts (17) since ε is arbitrary and if (t n ,x n )→(T,x) we have \(e^{-\beta t_{n}} x_{n}\to e^{-\beta T} x\). □
Proof of Theorem 3.2
Assume there is \((\hat{t},\hat{x})\in[0,T]\times\mathbb{R}_{+}\) such that \(u(\hat{t},\hat{x})-v(\hat{t},\hat{x})>0\). We assume, w.l.o.g., \((\hat{t},\hat{x})\in\{(t,x){:}\ t_{j}< t <t_{j-1}, \gamma_{j,i}(t)< x<\gamma_{j,i+1}(t)\}=:A\) with t j <t j−1, γ j,i <γ j,i+1<∞ defined as in Sect. 2 and in Remark 2.2. Note that because of (P5) we do not need to consider {(t,γ j,i (t)): t j <t<t j−1}∪{(t j ,x): γ j,i (t j )<x<γ j,i+1(t j )}. If u(t,γ j,i (t))−v(t,γ j,i (t))>0, then there is h>0 with (t,x+h)∈A with u(t,x+h)−v(t,x+h)>0; the same holds for t=t j .
We further assume that the comparison principle is already shown for the intervals [t l ,t l−1[ with l∈{j−1,…,m}, i.e. u(t,x)≤v(t,x) on \([t_{j-1},T]\times\mathbb{R}_{+}\). Note that Lemma 3.1 yields u(x,T)=v(x,T) for all \(x\in \mathbb{R}_{+}\).
Define v k=kv for k>1. It is easy to check that \(k\tilde{v}\) is still a supersolution with lsc envelope kv. Choose k>1 such that \(u(\hat{t},\hat{x})-v^{k}(\hat{t},\hat{x})>0\). Due to Lemma 2.1, we obtain the following inequality:
It is clear that u(t,x)−v k(t,x)≤0 for \(x\ge\frac{\lambda (0)}{k-1}e^{\beta t_{j-1}}=:\eta\). We can choose k in such a way that η≥γ j,i+1(t) for all t∈[t j ,t j−1[ and define
Define further
From above we know that M<λ(0)<∞ and obtain \(0<u(\hat{t},\hat{x})-v^{k}(\hat{t},\hat{x})\le M\). Since u−v k is continuous on A, there is (t ∗,x ∗)∈A with \(u(t^{*},x^{*})-v^{k}(t^{*},x^{*})>\frac{M}{2}>0\). Further, we let H:={(t,x,s,y): (t,x),(s,y)∈A, y−x≥0, t−s≥0} and for ξ>0:
In the following, we mean by u(t,γ j,i+1) and v k(s,γ j,i+1) the corresponding lsc envelopes u ∗(t,γ j,i+1) and kv ∗(s,γ j,i+1), respectively. Then it holds that
Note that \((t,x,s,\gamma_{j,i}(s)), (t,\gamma_{j,i+1}(t),s,y), (t_{j},x,s,y)\in\bar{H}\) only if t=s and x=γ j,i (s), t=s and y=γ j,i+1(t) or s=t j which yields
Let M ξ =sup H f ξ . Since f ξ is continuous on H, there is \((t_{\xi},x_{\xi}, s_{\xi},y_{\xi})\in\bar{H}\) such that M ξ =f ξ (t ξ ,x ξ ,s ξ ,y ξ ). Since (t ∗,x ∗)∈A, it holds that γ j,i (t ∗)<x ∗<γ j,i+1(t ∗), from which it follows that
We obtain directly: M ξ >0 for
In particular, we have \(\liminf_{\xi\to\infty}M_{\xi}\ge \frac{M}{2}>0\). With the usual techniques, see for example [3], one can show that there is ξ 1>0 such that (t ξ ,x ξ ,s ξ ,y ξ )∈H∖∂H for ξ>max{ξ 1,ξ 0}.
Note that by definition of (t ξ ,x ξ ,s ξ ,y ξ ) it holds that
Choose a sequence ξ n →∞ such that \((t_{\xi_{n}},x_{\xi _{n}},s_{\xi_{n}},y_{\xi_{n}})\to(\bar{t},\bar{x},\bar{s},\bar{y})\).
Let \(z_{n}:= (\xi_{n}(t_{\xi_{n}}-s_{\xi_{n}})+1 ) (\xi_{n}(y_{\xi _{n}}-x_{\xi_{n}}+t_{\xi_{n}}-s_{\xi_{n}})+1 )\). Then using (P2) and (18) we obtain
Here \(\lim_{n\to\infty}\xi_{n}(y_{\xi_{n}}-x_{\xi_{n}}+t_{\xi _{n}}-s_{\xi_{n}})=\infty\) implies \(\lim_{n\to\infty}\xi _{n}(y_{\xi_{n}}-x_{\xi_{n}})^{3}<0\), which is a contradiction. Thus, we conclude \(\lim_{n\to\infty} (t_{\xi_{n}}-s_{\xi_{n}})= \lim_{n\to\infty} (y_{\xi_{n}}-x_{\xi_{n}})=0\), and consequently \(\bar{x}=\bar{y}\), \(\bar{t}=\bar{s}\). On the other hand, \(\lim_{n\to\infty}\xi_{n}(y_{\xi_{n}}-x_{\xi_{n}}+t_{\xi_{n}}-s_{\xi_{n}})\le \frac{2k}{1+k}\), because otherwise we would again obtain \(\lim_{n\to\infty}\xi_{n}(y_{\xi_{n}}-x_{\xi_{n}})^{3}<0\). Note that \(\bar{x}\) is bounded away from \(\gamma_{j,i+1}(\bar{t})\) and \(\gamma_{j,i}(\bar{t})\).
Define the functions
These functions are continuously differentiable in t and in x. Furthermore, \(u(t,x)e^{\beta s_{\xi_{n}}}-\psi(t,x)\) attains its maximum at \((t_{\xi_{n}},x_{\xi_{n}})\) and \(v^{k}(s,y)e^{\beta t_{\xi_{n}}}-\phi(s,y)\) attains its minimum at \((s_{\xi_{n}},y_{\xi _{n}})\). Thus, \(\psi(t,x)e^{-\beta s_{\xi_{n}}}\) and \(\phi(s,y)e^{-\beta t_{\xi_{n}}}\) are test functions for u(t,x) and v k(s,y), respectively. Due to the arguments after relation (19), we have
Thus, there is \(N\in\mathbb{N}\) such that for n>N it holds that
Therefore, it holds by Definition 3.2 of viscosity sub- and super-solutions that:
Subtracting the above inequalities, rearranging the terms and letting ξ n →∞ yields the following relation:
On the other hand, we know that
which is a contradiction. □
Rights and permissions
About this article
Cite this article
Eisenberg, J., Grandits, P. & Thonhauser, S. Optimal Consumption Under Deterministic Income. J Optim Theory Appl 160, 255–279 (2014). https://doi.org/10.1007/s10957-013-0320-x
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10957-013-0320-x