Abstract
When solving for optimal strategies, a financial engineer needs to take into consideration of preferences heterogeneities, which involve not only present bias, but also future-focused preferences. We provide a reusable tool (i.e. algorithm) for explicitly solving optimal strategy in the presence of preferences variation over time, decision-makers and goods. In this framework, a new discount function identifies many hitherto unknown preference heterogeneities, a non-standard HJB yields sophisticated solution, a behavior equation produces naive and precommitted solutions. An application of the framework shows that a decision-maker can be neither impatient nor patient, and that sophisticated paradigm engenders immediate gratification in earlier phases of life cycle and self-control in later phases of life cycle.
Similar content being viewed by others
References
Arrow KJ et al (2014) Should governments use a declining discount rate in project analysis. Rev Environ Econ Policy 8:145–163
Becker RA (1980) On the long-run steady state in a simple dynamic model of equilibrium with heterogeneous households. Q J Econ 95:375–382
Bernheim D, Laibson D (2019) Intertemporal choice. In: Handbook of behavioral economics foundations and applications. Elsevier, North Holland
Caraballo SC, Fernandez RA (2019) A performance-based design framework for enhancing decision-making at the conceptual phase of a motorcycle rear suspension development. Optim Eng. https://doi.org/10.1007/s11081-019-09475-w
de-Paz A, Marín-Solano J, Navas J, Roch O (2014) Consumption, investment and life insurance strategies with heterogeneous discounting. Insur Math Econ 54:66–75
Drugeon J, Wigniolle B (2016) On time-consistent policy rules for heterogeneous discounting programs. J Math Econ 63:174–187
Ekeland I, Lazrak A (2010) The golden rule when preferences are time inconsistent. Math Financ Econ 4:29–55
Ekeland I, Pirvu TA (2008) Investment and consumption without commitment. Math Financ Econ 2:57–86
Ekeland I, Mbodji O, Pirvu TA (2012) Time-consistent portfolio management. SIAM J Financ Math 3:1–32
Feigenbaum JA (2016) Equivalent representations of non-exponential discounting models. J Math Econ 66:58–71
Harris C, Laibson D (2013) Instantaneous gratification. Q J Econ 128:205–241
Karp LS (2007) Non-constant discounting in continuous time. J Econ Theory 132:557–568
Kılınç MR, Sahinidis NV (2018) Exploiting integrality in the global optimization of mixed-integer nonlinear programming problems with baron. Optim Methods Softw 33:540–562
Kloeden PE, Platen E (1995) Numerical solution of stochastic differential equation. Springer, Berlin
Liu J, Ploskas N, Sahinidis NV (2019) Tuning baron using derivative-free optimization algorithms. J Glob Optim 74:611–637
Marín-Solano J, Navas J (2009) Non-constant discounting in finite horizon: the free terminal time case. J Econ Dyn Control 33:666–675
Marín-Solano J, Navas J (2010) Consumption and portfolio rules for time-inconsistent investors. Eur J Oper Res 201:860–872
Marín-Solano J, Patxot C (2012) Heterogeneous discounting in economic problems. Optimal Control Appl Methods 33:32–50
Marín-Solano J, Navas J, Roch O (2013) Non-constant discounting and consumption, portfolio and life insurance rules. Econ Lett 119:186–190
Miller CW, Yang I (2017) Nonlinear pde approach to time-inconsistent optimal stopping. SIAM J Control Optim 55:557–573
Myerson J, Green L (1995) Discounting of delayed rewards: models of individual choice. J Exp Anal Behav 64:263–276
Ngo H, Taguchi D (2017) Strong convergence for the Euler–Maruyama approximation of stochastic differential equations with discontinuous coefficients. Stat Probab Lett 125:55–63
Ni Y-H, Li X, Zhang J-F, Krstic M (2019) Mixed equilibrium solution of time-inconsistent stochastic linear-quadratic problem. SIAM J Control Optim 57:533–569
O’Donoghue T, Rabin M (2015) Present bias: lessons learned and to be learned. Am Econ Rev 105:273–279
Peng L, Hager WW (2017) Non-constant quasi-hyperbolic discounting. Econ Comput Econ Cybern Stud Res 2:145–164
Peng L, Kloeden P (2019) Impulsivity and heterogeneity. Econ Comput Econ Cybern Stud Res 4:313–324
Pliska SR, Ye J (2007) Optimal life insurance purchase and consumption/investment under uncertain lifetime. J Bank Finance 31:1307–1319
Ramsey FP (1928) A mathematical theory of saving. Econ J 38:543–559
Sahinidis NV (1996) Baron: a general purpose global optimization software package. J Glob Optim 8:201–205
Shen Y, Wei J (2016) Optimal investment–consumption–insurance with random parameters. Scand Actuar J 2016:37–62
Shreve SE (2004) Stochastic calculus for finance II. Springer, Berlin
Yang C, Cheng LL (2013) Optimal guidance law in quantum mechanics. Ann Phys 338:167–185
Acknowledgements
We thank the Editor, Professor Nikolaos Sahinidis, and anonymous referees for their insightful suggestions, which enhance this paper. Funding for this research was provided by China Scholarship Council under CSC Number 201808525077. This work was also supported by Scientists Recruitment Program of Guizhoiu University (resp. School of management, Guizhou University) under Agreement Number 2017-008 (resp. 17GLR007).
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendices
Appendix 1: Proof of Theorem 4.1
-
1.
Time is partitioned into N intervals. The grid size is specified as equidistant \(\Delta {t_j} = t_{j + 1} - t_j = \varepsilon = \dfrac{T}{N} \rightarrow 0^ +\) such that all subintervals shrink to zero in the limit as \(N \rightarrow \infty\) or \(\varepsilon \rightarrow 0^ +\). We can find \(t_j = j \times \Delta t = j\varepsilon\) such that \({\hat{X}}_k = {\hat{X}} ( k\varepsilon ), ~{\hat{u}}_k = {\hat{u}} (k\varepsilon ),~~ {\hat{V}}_j \buildrel \Delta \over = {\hat{V}}({\hat{x}},j\varepsilon ), ~~ {\hat{V}}_N \buildrel \Delta \over = {\hat{V}}({\hat{X}}_N,N\varepsilon ) ={\hat{F}}( {\hat{X}}_N, N\varepsilon )\). For \(j=0,\ldots ,N-1\) and \(k=j,\ldots ,N-1\), the original problem (6−7) becomes
$$\begin{aligned}{\hat{V}}_j &= \inf _{{\hat{u}} } {\mathbb {E}}^{ {\hat{x}}}_{j\varepsilon } \left( L({\hat{X}}_j,{\hat{u}}_j,j\varepsilon ) \varepsilon \nonumber \right. \\&\quad \left. + \sum \limits _{i = 1}^{N - j - 1} {\mathbb {D}} ( {0,i\varepsilon }) L({\hat{X}}_{i + j}, {\hat{u}}_{i + j }, (i + j)\varepsilon )\varepsilon +e^{ - (N - j)\rho \varepsilon } {\hat{V}}_N \right) \end{aligned}$$(15)subject to
$$\begin{aligned} \left\{ \begin{aligned} {\hat{X}}_{k + 1}&= {\hat{X}}_k + {\hat{\eta }} ( {\hat{X}}_k, {\hat{u}}_k, k\varepsilon ) \varepsilon + \sum \limits _b^m {\hat{\sigma }} ^b ( {\hat{X}}_k, {\hat{u}}_k, k\varepsilon ) (z_{k + 1}^b - z_k^b)\\ {\hat{X}}_j&= {\hat{x}}\\ \end{aligned} \right. \end{aligned}$$(16)where \({\mathbb {E}}^{ {\hat{x}}}_{j\varepsilon }= {\mathbb {E}} \left[ \cdot \mid {\hat{X}}_j={\hat{x}} \right]\) is the conditional expectation. Our choice of functions guarantees (cf. Kloeden and Platen 1995; Ngo and Taguchi 2017) that Eq. (15) (resp. Eq. 16) converges to Eq. (6) (resp. Eq. 7).
-
2.
On one hand, multiplication of Eq. (15) by \(e^{ - (N - j-1)\rho \varepsilon }\) implies
$$\begin{aligned} \begin{array}{l} e^{ - (N - j - 1)\rho \varepsilon } e^{ - (N - j )\rho \varepsilon } {\hat{V}}_N = \\ \inf_{\hat u } {\mathbb {E}}^{ {\hat{x}}}_{j\varepsilon } \left( \begin{array}{l} e^{ - (N - j-1)\rho \varepsilon } {\hat{V}}_j - e^{ - (N - j-1)\rho \varepsilon } L({\hat{X}}_j,{\hat{u}}_j,j\varepsilon ) \varepsilon +\\ \sum \limits _{i = 1}^{N - j - 1} e^{ - (N - j-1)\rho \varepsilon } {\mathbb {D}} ( {0,i\varepsilon })L \Big ({\hat{X}}_{i + j}, {\hat{u}}_{i + j }, (i + j) \varepsilon \Big )\varepsilon \\ \end{array} \right) \end{array} \end{aligned}$$(17)On the other hand, Eq. (15) implies
$$\begin{aligned}e^{ - (N - j-1)\rho \varepsilon } {\hat{V}}_N &= \inf _{{\hat{u}} } {\mathbb {E}}^{ {\hat{x}}}_{(j+1)\varepsilon } \left( {\hat{V}}_{j+1} \nonumber \right. \\&\quad \left. - \sum \limits _{i = 1}^{N - j - 1} {\mathbb {D}} ( {0,(i-1)\varepsilon }) L \Big ({\hat{X}}_{i + j}, {\hat{u}}_{i + j }, (i + j)\varepsilon \Big )\varepsilon \right) \end{aligned}$$(18) -
3.
Inserting Eq. (18) into Eq. (17)
$$\begin{aligned} {\hat{V}}_{j+1} -{\hat{V}}_j=\inf _{{\hat{u}} } {\mathbb {E}}^{ {\hat{x}}}_{j\varepsilon } \left[ {{\hat{K}}({\hat{x}},j\varepsilon )+ \rho }{\hat{V}}_j \varepsilon - L ( {\hat{X}}_j, {{\hat{u}}}_j, j\varepsilon )\varepsilon + o(\varepsilon ) \right] \end{aligned}$$(19)where
$$\begin{aligned}{\hat{K}}({\hat{x}},j\varepsilon ) &= \left\{ \sum \limits _{i = 1}^{N - j - 1} L \Big ({\hat{X}}_{i + j}, {\hat{u}}_{i + j }, (i + j)\varepsilon \Big ) \nonumber \right. \\&\left. \quad \Big [ {\mathbb {D}} ( {0,(i-1)\varepsilon }) (1 - \rho \varepsilon ) - {\mathbb {D}} ( {0,i\varepsilon }) \Big ] \right\} \varepsilon \end{aligned}$$(20) -
4.
By the SQH algorithm (e.g. Harris and Laibson 2013) and the discretizing algorithm (e.g. Karp 2007), Eq. (20) can be rewritten as
$$\begin{aligned}{\hat{K}}({\hat{x}},j\varepsilon ) &= \left\{ \sum \limits _{i = 1}^{N - j - 1} \Big [ ( 1 - \beta ) \lambda e^{ - \lambda i\varepsilon } \varepsilon + (r - \rho )\varepsilon \Big ] e^{ - ri\varepsilon } \nonumber \right. \\&\quad \left. L \Big ({\hat{X}}_{i + j}, {\hat{u}}_{i + j }, (i + j)\varepsilon \Big ) \right\} \varepsilon \end{aligned}$$(21)LHS of Eq. (19) can be rewritten as (e.g. Yang and Cheng (2013), equation 2.16),
$$\begin{aligned} \begin{aligned}V ( x (t+\varepsilon ), t+\varepsilon ) - V(x,t) &= \frac{\partial V(x,t) }{\partial t}\varepsilon + \nabla _x V(x,t) \cdot \eta ( x,u,t) \varepsilon \\&\quad + \frac{1}{2} trace \Big ( \sigma ^{\mathrm{T}} (x,u,t) \cdot \sigma (x,u,t) \cdot \nabla _{xx} V(x,t) \Big ) \varepsilon \\&\quad + \nabla _x V(x,t) \cdot \sigma (x,u,t) \cdot \Big [ z(t+\varepsilon ) -z(t) \Big ] + o(\varepsilon ) \end{aligned} \end{aligned}$$(22) -
5.
$$\begin{aligned} \begin{aligned}&\frac{\partial V(x,t) }{\partial t}\varepsilon + \nabla _x V(x,t) \cdot \eta ( x,u,t) \varepsilon \\&\qquad + \frac{1}{2} \Big ( \sigma ^{\mathrm{T}} (x,u,t) \cdot \sigma (x,u,t) \cdot \nabla _{xx} V(x,t) \Big ) \varepsilon \\&\qquad + \nabla _{x} V(x,t) \cdot \sigma (x,u,t) \cdot \Big [z(t+\varepsilon ) -z(t)\Big ] +o(\varepsilon )\\&\quad = \rho {\hat{V}}_j \varepsilon - L ( {\hat{X}}_j, {{\hat{u}}}_j, j\varepsilon )\varepsilon + o(\varepsilon )\\&\qquad +\sum \limits _{i = 1}^{N - j - 1} \Big [ ( 1 - \beta ) \lambda e^{ - \lambda i\varepsilon } \varepsilon \\&\qquad + (r - \rho )\varepsilon \Big ] e^{ - ri\varepsilon } L \Big ({\hat{X}}_{i + j}, {\hat{u}}_{i + j }, (i + j)\varepsilon \Big )\varepsilon \end{aligned} \end{aligned}$$(23)
Dividing by \(\varepsilon\) and taking the limit \(\varepsilon \rightarrow 0\) complete the proof.
\(\square\)
Appendix 2: Proof of Remark (4.2)
This appendix will explain how the SHQH discount function includes other discount functions as its special cases.
1.1 Reduction to exponential discount function and its HJB
The exponential discount function at time t used to evaluate a payoff at time s is
The stochastic optimal control problem under the exponential discounting (24) is
subject to the constraint (7).
The HJB associated with the exponential discounting (24) is
When
in Eq. (6), the SHQH discount function (6) and its HJB (8) reduce to the exponential discount function (24) and its HJB (25).
1.2 Reduction to the HG discount function and its HJB
The heterogeneous (HG) discount function (e.g. Marín-Solano and Patxot 2012; Marín-Solano et al. 2013; de-Paz et al. 2014) at time t used to evaluate a payoff at time s is
The stochastic optimal control problem under the HG discounting is
subject to the constraint (7).
The HJB associated with the HG discount function (26) is
where
When
in Eq. (6), the SHQH discount function (6) and its HJB (8) reduce to the HG discount function (26) and its HJB (27). \(\square\)
1.3 Reduction to the SQH discount function and its HJB
The SQH discount function (e.g. Harris and Laibson 2013) at time t used to evaluate a payoff at time s is
The stochastic optimal control problem under the SQH discounting (28) is
subject to the constraint (7).
The HJB associated with the SQH discount function (SQH-HJB) is
where
\(\square\)
When
in Eq. (6), the SHQH discount function (6) and its HJB (8) reduce to the SQH discount function (28) and its HJB (29).
Appendix 3: Proof of Lemma 5.1
Proof of Lemma 5.1
Time is partitioned into N intervals. The grid size is specified as equidistant \(\Delta {t_j} = t_{j + 1} - t_j = \varepsilon = \dfrac{T}{N} \rightarrow 0^ +\) such that all subintervals shrink to zero in the limit as \(N \rightarrow \infty\) or \(\varepsilon \rightarrow 0^ +\). We can find \(t_j = j \times \Delta t = j\varepsilon\) such that \({\hat{W}}_k = {\hat{W}} ( k\varepsilon ), ~{\hat{c}}_k = {\hat{c}} (k\varepsilon ),~~ {\hat{V}}_j \buildrel \Delta \over = {\hat{V}}({\hat{w}},j\varepsilon ), ~~ {\hat{V}}_N \buildrel \Delta \over = {\hat{V}}({\hat{W}}_N,N\varepsilon ) ={\hat{F}}( {\hat{W}}_N, N\varepsilon )\). For \(j=0,\ldots ,N-1\) and \(k=j,\ldots ,N-1\), the original problem (10−11) becomes
subject to
where \({\mathbb {E}}^{ {\hat{w}}}_{j\varepsilon }= {\mathbb {E}} \left[ \cdot \mid {\hat{W}}_j={\hat{w}} \right]\) is the conditional expectation. On one hand, multiplication of Eq. (30) by \(e^{ - (N - j-1)\rho \varepsilon }\) implies
Following the same procedure as Appendix 1 leads to
Dividing by \(\varepsilon\) and taking the limit \(\varepsilon \rightarrow 0\) complete the proof. \(\square\)
Appendix 4: Proof of Table 1
This section solves the heterogeneity problem (10–11) under the logarithmic utility explicitly.
1.1 Precommitted and naive paradigms
Let a and b be constants. Under logarithmic utility, the functions in the SHQH behavior Eq. (14) are of the form
Using Eq. (14) and Eq. (34) (e.g. Shreve 2004), one arrives at the following plans made by the precommitted and naive decision-makers standing at any arbitrary point in the life cycle \(t_0\in [0,T]\)
where
The precommitted decision-maker of cohort \(t_0\) governs the behavior of other selves. Her optimal policy is obtained by evaluating her planned consumption program (35−36) at \(t=0\)
where
For the naive decision-maker, the pain incurred by a old plan is continuously overwhelmed by the happiness from a new plan. Her optimal strategy is found by evaluating her planned consumption program (35−36) at \(t=t_0\) and at \(t_0=t\)
where
1.2 Sophisticated paradigm
Under logarithmic utility, the functions in the SHQH non-standard HJB (12) are of the form
Differentiating the SHQH non-standard HJB (12) with respect to c and Q and using Eq. (37), one arrives at the following first order condition (FOC):
Applying Ito’s lemma (e.g. Shreve 2004) on the wealth dynamics (11), and substituting the resulting equation into the K term (13) give
where
Inserting the K term (39) into the SHQH non-standard HJB (12) gives,
Appendix 5: Proof of Table 2
This section solves the heterogeneity problem (10−11) under the CRRA utility explicitly.
1.1 Precommitted and naive paradigms
Under the CRRA utility, the functions in the SHQH behavior Eq. (14) are of the form
Combining(e.g. Shreve 2004) the conjecture (40) and the SHQH behavior Eq. (14), one arrives at the plans made by the precommitted and naive decision-makers standing at any arbitrary point in the life cycle \(t_0\in [0,T]\)
where
The precommitted decision-maker of cohort \(t_0\) governs the behavior of other selves. Her optimal policy is obtained by evaluating Eq. (41) at \(t=0\)
where
The naive decision-maker of cohort t re-optimizes at each instant. Her optimal policy is obtained by evaluating her planned consumption program (41) at \(t=t_0\) and \(t_0=t\),
where
1.2 Sophisticated paradigm
Under CRRA utility, the functions in the SHQH non-standard HJB (12) are of the form
Differentiating the SHQH non-standard HJB (12) with respect to c and using equation (42), one arrives at the following FOC,
Applying Ito’s lemma on the wealth dynamics (11) and substituting the resulting equation into the K term (13) yield
where
Or, equivalently,
where
Substituting the K term (45) into the SHQH non-standard HJB (12) gives,
where
Appendix 6: Optimal strategies under CARA utility
This section solves the heterogeneity problem (10−11) under the CARA utility explicitly.
1.1 Precommitted and naive paradigms
Under the CARA utility, the functions in the SHQH behavior Eq. (14) are of the form
Combining (e.g. Shreve (2004), page 453) the conjecture (47) and the SHQH behavior equation (14), we arrive at the plans made by the precommitted and naive decision-makers standing at any arbitrary point in the life cycle \(t_0\in [0,T]\)
where
The precommitted decision-maker of cohort \(t_0\) governs the behavior of other selves. Her optimal policy is obtained by evaluating her planned consumption program (48) at \(t=0\)
where
For the naive decision-maker, her planned consumption program continues to become invalid and suboptimal with the progression of time . Her optimal policy is obtained evaluating her planned strategies (48) at \(t=t_0\) and \(t_0=t\)
where
1.2 Sophisticated paradigm
Under the CARA utility, the functions in the SHQH non-standard HJB (12) are of the form
Plugging equation (49) into the SHQH non-standard HJB (12) and collecting similar terms (e.g. Shreve 2004), one arrives at optimal consumption–wealth ratio,
where
Rights and permissions
About this article
Cite this article
Peng, L., Kloeden, P.E. A solution method for heterogeneity involving present bias. Optim Eng 21, 1167–1194 (2020). https://doi.org/10.1007/s11081-020-09505-y
Received:
Revised:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11081-020-09505-y