Abstract
We consider a multi-component assembly system that produces a single end product in order to satisfy the one-time demand at a (known) future time with an exogenous selling price. Components can be outsourced from outside suppliers with positive leadtimes under either time-inflexible or time-flexible contracts. One of these components, say component 1, faces an uncertainty in its procurement price, depending on the spot market price that is governed by a geometric Brownian motion, while the prices of other components are constant. Given supply contracts, the assembler needs to determine the procurement strategy to maximize the total expected profit that equals the expected revenue minus procurement cost, holding cost, and tardiness penalty cost. Under time-inflexible contracts, the problem is static and a variant of the classic newsvendor problem. We show that leadtime uncertainty will cause the assembler to be more conservative in procurement quantity, but more aggressive in procurement timing than if the leadtime is deterministic. For time-flexible contracts, we show that under certain conditions, the original problem is equivalent to an optimal single stopping problem whose optimal strategies follow either upward or downward base-price procurement policies. Under the general condition, we propose an efficient Monte Carlo simulation method to calculate the optimal solutions. Numerical studies also provide several interesting insights: first, both procurement quantity and profit are non-monotone in the leadtime length; second, the value of time-flexible contracts compared to time-inflexible contracts is close to zero if the price of component 1 has a decreasing trend, but otherwise significant.
Similar content being viewed by others
Change history
11 September 2019
This erratum is published because vendor overlooked corrections and section headings were given in numbered and run-on format.
References
Akella, R., Araman, V. F., & Kleinknecht, J. (2001). B2B markets: Procurement and supplier risk management in e-Business. In H. E. R. J. Geunes & P. M. Pardalos (Eds.), Supply chain management: Models, applications, and research directions. Dordrecht: Kluwer.
Andersen, L., & Broadie, M. (2004). Primal-dual simulation algorithm for pricing multidimensional American options. Management Science, 50(9), 1222–1234.
Belomestny, D. (2011). Pricing Bermudan options using nonparametric regression: Optimal rates of convergence for lower estimates. Finance and Stochastics, 15, 655–683.
Bender, C., Schoenmarkers, J., & Zhang, J. (2015). Dual representations for general multiple stopping problems. Mathematical Finance, 25(2), 339–370.
Benjaafar, S., & ElHafsi, M. (2006). Production and inventory control of a single product assemble-to-order system with multiple customer classes. Management Science, 52, 1896–1912.
Benjaafar, S., ElHafsi, M., Lee, C. Y., & Zhou, W. (2011). Optimal control of assembly systems with multiple stages and multiple demand classes. Operations Research, 59, 522–529.
Berling, P., & Martínez de Albéniz, V. (2011). Optimal inventory policies when purchase price and demand are stochastic. Operations Research, 59(1), 109–124.
Chen, N., & Liu, Y. (2014). American option sensitivities estimation via a generalized infinitesimal perturbation analysis approach. Operations Research, 62(3), 616–632.
Choi, T. M., Cheng, T. C. E., & Zhao, X. D. (2016). Multi-methodological research in operations management. Production and operations management. Production and Operations Management, 25(3), 379–389.
Dixit, A., & Pindyck, R. (1994). Investment under uncertainty. Princeton, NJ: Princeton University Press.
Doğru, M. K., Reiman, M. I., & Wang, Q. (2010). A stochastic programming based inventory policy for assemble-to-order systems with application to the W model. Operations Research, 58(4), 849–864.
Gavirneni, S. (2004). Periodic review inventory control with fluctuating purchasing costs. Operations Research Letters, 32(4), 374–379.
Haksöz, Ç., & Seshadri, S. (2007). Supply chain operations in the presence of a spot market: A review with discussion. Journal of the Operational Research Society, 58, 1412–1429.
Haugh, M., & Kogan, L. (2004). Pricing American options: A duality approach. Operations Research, 52, 258–270.
Hruska, J. (2018). Why RAM prices are through the roof. Extreme Tech.
Hull, J. C. (1997). Option, futures and other derivatives (3rd ed.). London: Prentice Hall International.
Jacka, S. D., & Lynn, J. R. (1992). Finite-horizon optimal stopping. Obstacle Problems and the Shape of the Continuation Region, Stochastics and Stochastic Reports, 39(1), 25–42.
Kobylanski, M., Quenez, M., & Rouy-Mironescu, E. (2011). Optimal multiple stopping time problem. The Annals of Apllied Probability, 21(4), 1365–1399.
Kushner, H. J. (1999). Control and optimal control of assemble-to-order manufacturing systems under heavy traffic. Stochastics and Stochastics Reports, 66, 233–272.
Li, C. L., & Kouvelis, P. (1999). Flexible and risk-sharing supply contracts under price uncertainty. Management Science, 45(10), 1378–1398.
Longstaff, F. A., & Schwartz, E. (2001). Valuing American options by simulation: A simple least-squares approach. Review of Financial Studies, 14(1), 113–147.
Lu, Y., & Song, J. S. (2005). Order-based cost optimization in assemble-to-order systems. Operations Research, 53(1), 151–169.
Martínez de Albéniz, V., & Simchi-Levi, D. (2005). A portfolio approach to procurement contracts. Production and Operations Management, 14(1), 90–114.
Plambeck, E. (2008). Asymptotically optimal control for an assemble-toorder system with capacitated component production and fixed transport costs. Operations Research, 56(5), 1158–1171.
Plambeck, E., & Ward, A. R. (2006). Optimal control of a high-volume assemble-to-order system. Mathematics of Operations Research, 31, 453–477.
Plambeck, E., & Ward, A. R. (2015). Optimal control of a high-volume assemble-to-order system with guaranteed maximum delays and expediting. Queueing Systems, 60(1), 1–69.
Rogers, L. C. (2002). Monte Carlo valuation of American options. Mathematical Finance, 14(3), 113–147.
Stentoft, L. (2004). Convergence of the least square Monte Carlo approach to American option valuation. Management Science, 50(9), 1193–1203.
Song, J. S., Yano, C. A., & Lerssrisuriya, P. (2000). Contract assembly: Dealing with combined supply lead time and demand quantity uncertainty. Manufacturing & Service Operations Management, 50(5), 889–903.
Song, J. S., & Zipkin, P. (2003). Supply chain operations: Assemble-to-order and configure-to-order systems. Handbooks in Operations Research and Management Science, 11, 561–596.
Tsitsiklis, J., & Van Roy, B. (1999). Optimal stopping of Markov processes: Hilbert space theory, approximation algorithms, and an application to pricing high-dimensional financial derivatives. IEEE Transanction on Automatical Control, 44(10), 1840–1851.
Tsitsiklis, J., & Van Roy, B. (2001). Regression methods for pricing complex American-style options. IEEE Transanction on Neural Networks, 12(4), 694–703.
Xiao, G., Yang, N., & Zhang, R. (2015). Dynamic pricing and inventory management under fluctuating procurement costs. Manufacturing & Service Operations Management, 17(3), 321–334.
Xiao, Y., Chen, J., & Lee, C. Y. (2010). Single-period two-product assemble-to-order systems with a common component and uncertain demand patterns. Production and Operations Management, 19(2), 216–232.
Xu, S. H. (2001). Dependence analysis of assemble-to-order systems. In Supply chain structures: Coordination, information and optimization, Kluwer, Norwell, MA.
Yi, J. & A. Scheller-Wolf. (2003). Dual sourcing from a regular supplier and a spot market. GSIA-Carnegie Mellon University. Working Paper.
Zhao, Y., & Simchi-Levi, D. (2006). Performance analysis and evaluation of assemble-to-order systems with stochastic sequential leadtimes. Operations Research, 54(4), 706–724.
Acknowledgements
This work is partially supported by National Natural Science Foundation of China under Grant Numbers 71501196, 71522003, 71721001, 71821002, 71871166, and 71871198, Hong Kong General Research Fund under Project Number 9042129, and the Natural Science Foundation of Guangdong Province of China under Grant Number 2014A030312003.
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
Time-inflexible contract with uncertain leadtime
We now consider the setting with the uncertain leadtime of component 2 which follows distributions c.d.f G(x) and p.d.f g(x). Without confusion, we retain the notations used in the main text. The formulation is a little bit different in that the terms involving \(L_2\) should be taken expectation. Then, the expected tardiness of the end product is
And, the expected holding time of component 2 from the arrival of the order to the assemble time (due date plus tardiness, if any) is as follows:
Consequently, the total expected profit function is
where \(C(q,t_{1},t_{2})=s\theta T_{d}+q(h_{1}T_{h,1}+h_{2}T_{h,2})+q({\mathbb {E}}[c_{1,t_{1}}\mid c_{1,0}]+c_{2}).\)
The objective is then to solve the optimization problem stated in (6). Due to uncertain leadtime, the assembler cannot perfectly time the procurement of component 2 to avoid its holding cost. That is, Lemma 1 is no longer available in this setting. This indeed complicates the analysis of the problem. Define \(t_1^{u,*}\), \(t_2^{u,*}\), and \(q^{u,*}\) as the optimal procurement timings and quantity under uncertain leadtime, respectively. The following theorem partially characterizes the optimal solutions for the case with \(\mu >0\) an \(\mu <0\), respectively.
Theorem 4
- (a) :
-
When \(\mu >0\), the following results hold:
- (i):
-
the optimal procurement time of component 1 is
$$\begin{aligned} t_{1}^{u*}=\left\{ \begin{array}{ll}0 &{}\quad \text{ if } \quad \frac{h_{1}}{\mu c_{1,0}}\le 1;\\ \frac{1}{\mu }\ln (\frac{h_{1}}{\mu c_{1,0}}), &{}\quad \text{ if } \quad 1<\frac{h_{1}}{\mu c_{1,0}}\le e^{\mu (T-L_{1})};\\ T-L_{1}, &{}\quad \text{ if } \quad e^{\mu (T-L_{1})}<\frac{h_{1}}{\mu c_{1,0}}. \end{array}\right. \end{aligned}$$(A1) - (ii):
-
the optimal procurement quantity and procurement time of component 2 satisfy
$$\begin{aligned} q^{u*}&=F^{-1}\left( \frac{p-h_{1}{\hat{T}}_{h,1}-h_{2}{\hat{T}}_{h,2}-c_{1,0}e^{\mu t^{u*}_{1}}-c_{2}}{p}\right) ;\\ t_2^{u*}&=T-G^{-1}\left( \frac{sd+q^{u*}h_{1}}{sd+q^{u*}(h_{1}+h_{2})}\right) , \end{aligned}$$where \({\hat{T}}_{h,1}={\mathbb {E}}[(t^{u*}_{2}+L_{2}-T)^{+}]+T-t^{u*}_1-L_1\) and \({\hat{T}}_{h,2}={\mathbb {E}}[(t^{u*}_{2}+L_{2}-T)^{+}]+T-t^{u*}_2-{\mathbb {E}}[L_2]\).
- (b) :
-
When \(\mu <0\), the following results hold:
- (i):
-
the optimal procurement time of component 1 is
$$\begin{aligned} t_{1}^{u*}=\left\{ \begin{array}{ll}T-L_1 &{}\quad \text{ if } \quad \frac{sd}{-\mu c_{1,0}q^{u*}}\ge e^{\mu (T-L_1)};\\ \frac{1}{\mu }\ln (\frac{sd}{-\mu c_{1,0}q^{u*}}), &{}\quad \text{ if } \quad e^{\mu T}<\frac{sd}{-\mu c_{1,0}q^{u*}}< e^{\mu (T-L_1)};\\ T, &{}\quad \text{ if } \quad \frac{sd}{-\mu c_{1,0}q^{u*}}\le e^{\mu T}. \end{array}\right. \end{aligned}$$(A2) - (ii):
-
the optimal procurement time of component 2 is
$$\begin{aligned} t_2^{u*}=t_1^{u*}+L_1-G^{-1}\left( \frac{sd+q^{u*}h_{1}}{sd+q^{u*}(h_{1}+h_{2})}\right) . \end{aligned}$$ - (iii):
-
the optimal order quantity satisfies
$$\begin{aligned} q^{u*}&=F^{-1}\left( \frac{p-h_{1}{\hat{T}}_{h,1}-h_{2}{\hat{T}}_{h,2}-c_{1,0}e^{\mu t^{u*}_{1}}-c_{2}}{p}\right) , \end{aligned}$$where \({\hat{T}}_{h,i}={\mathbb {E}}[\max \{(t^{u*}_{1}+L_{1}-T)^{+}, (t^{u*}_{2}+L_{2}-T)^{+}\}]+T-t^{u*}_i-{\mathbb {E}}[L_i]\).
Proof of Theorem 4
First, taking the first partial derivative of \(T_{d}(t_1,t_2)\) respect to \(t_{1}\) and \(t_{2}\), respectively, we can obtain
Accordingly, the first partial derivatives of \(S(q,t_{1},t_{2})\) respect to q, \(t_{1}\) and \(t_{2}\) are respectively given as:
(a) For the case \(\mu >0\), we first prove that it is optimal to order component 1 before time \(T-L_1\), i.e., \(t_1^{u*}\le T-L_1\). Suppose \(t_1^{u*}> T-L_1\), then \(t_1^{u*}\) and \(t_2^{u*}\) should satisfy
which, together with (A5), implies that \(-\,\mu c_{1,0}qe^{\mu t_{1}}-sd=0\). This contradicts with the fact that \(\mu >0\). Therefore, \(t_1^{u*}\le T-L_1\). Define \(\hat{t}:=\frac{1}{\mu }\ln (\frac{h_{1}}{\mu c_{1,0}})\). Refer to (A5). If \(\hat{t}\le 0\), i.e., \(\frac{h_{1}}{\mu c_{1,0}}\le 1\), then \(S(q,t_{1},t_2)\) is decreasing in \(t_{1}\in [0, T-L_1]\) for any q and \(t_2\), and thus, \(t_{1}^{u*}=0\). If \(0<\hat{t}\le T-L_1\), i.e., \(1<\frac{h_{1}}{\mu c_{1,0}}\le e^{\mu (T-L_{1})}\), then \(t_{1}^{u*}=\hat{t}\). If \({\hat{t}} > T-L_1\), i.e., \(\frac{h_{1}}{\mu c_{1,0}}> e^{\mu (T-L_{1})}\), then \(t_1^{u*}=T-L_1\). Moreover by (A6), the optimal ordering time of component 2 satisfies
By the first derivative of \(S(q_1,t_1,t_2)\) respect to \(q_1\), we have
where \({\hat{T}}_{h1}={\mathbb {E}}[(t^{u*}_{2}+L_{2}-T)^{+}]+T-t^{u*}_1-L_1\) and \({\hat{T}}_{h2}={\mathbb {E}}[(t^{u*}_{2}+L_{2}-T)^{+}]+T-t^{u*}_2-{\mathbb {E}}[L_2]\).
(b) For the case \(\mu <0\), it is clear that \(S(t_1,t_2,q)\) is increasing in \(t_1\le T-L_1\) for any given q. Therefore, it is optimal to order component 1 after time \(T-L_1\), i.e., \(t_1^{u*}>T-L_1\). Moreover, in this case, by the first optimality condition, \(t_1^{u*}\) and \(t_2^{u*}\) should satisfy (A7). That is,
Putting it into (A5), we can obtain
Let \({\hat{t}}_1=\frac{1}{\mu }\ln (\frac{sd}{-\mu c_{1,0}q}).\) If \({\hat{t}}_1\le T-L_1\), i.e., \( \frac{sd}{-\mu c_{1,0}q}\ge e^{\mu (T-L_1)}\), then \(t_1^{u*}=T-L_1\); if \(T-L_1<{\hat{t}}_1<T\), i.e., \(e^{\mu T}<\frac{sd}{-\mu c_{1,0}q}< e^{\mu (T-L_1)}\) then \(t_1^{u*}={\hat{t}}_1\); if \({\hat{t}}_1 \ge T\), i.e., \(\frac{sd}{-\mu c_{1,0}q}\le e^{\mu T},\) then \(t_1^{u*}=T\).
By the first derivative of \(S(q_1,t_1,t_2)\) respect to \(q_1\), we have
where \({\hat{T}}_{hi}={\mathbb {E}}[\max \{(t^{u*}_{1}+L_{1}-T)^{+}, (t^{u*}_{2}+L_{2}-T)^{+}\}]+T-t^{u*}_i-{\mathbb {E}}[L_i]\). \(\square \)
Auxiliary proofs
Proof of Lemma 1
We first consider the case with \(t_1^*\le T-L_1\). Referring to (2) and (3), we have \(T_{d}=\max _{i=2,\ldots ,N}\left\{ (t_{i}+L_{i}-T)^{+}\right\} \) and \(T_{hi}=T_d+T-(t_i+L_i)\), \(i=2,\ldots ,N\). Clearly, both holding and tardiness costs are minimized at \(t_i^*=T-L_i\), \(i=2,\ldots ,N\), with the minimal values \(T_d=0\) and \(T_{hi}=0\). Referring to (5), this result implies that given \(t_1^*\le T-L_1\), \(S(q,{\mathbf {t}})\) is minimized at \(t_i^*=T-L_i\), \(i=2,\ldots ,N,\) for any q.
Now turn to the case with \(t_1^*\ge T-L_1\). Define \({\mathbf {t}}_{-1}=(t_2,\ldots ,t_N)\). We write \(T_d\) and \(T_{h2}\) as a function of \({\mathbf {t}}_{-1}\). By the definition, we have for any \({\mathbf {t}}_{-1}\),
Note that \(T_d({\mathbf {t}}_{-1})|_{t_i=t_1^*+L_1-L_i}=(t_{1}^*+L_{1}-T)\) and \(T_{h2}({\mathbf {t}}_{-1})|_{t_i=t_1^*+L_1-L_i}=0\) which means that both lower bounds can be achieved at \(t_i^*=t_1^*+L_1-L_i\). Therefore, the results hold. \(\square \)
Proof of Theorem 1
(a) Recall that \(c_{1,t}\) follows a Brownian motion (1) and thus, \({\mathbb {E}}[c_{1,t_1}|c_{1,0}]=c_{1,0}e^{\mu t_{1}}\). Then, \({\widehat{C}}(q,t_1)\) can be rewritten as
The first partial derivatives of \({\widehat{S}}(q,t_{1})\) respect to q and \(t_{1}\) are given as follows.
The second partial derivatives of \({\widehat{S}}(q,t_{1})\) respect to q and \(t_{1}\) are given as follows.
Therefore, for any fixed \(t_{1}\), \({\widehat{S}}(q,t_{1})\) is concave in q and for any fixed q, \({\widehat{S}}(q,t_{1})\) is concave in \(t_{1}\).
(b) Referring to (B9), we have that if \(\mu > 0\), \(\frac{\partial {\widehat{S}}(q,t_{1})}{\partial t_{1}} < 0\) for any \(t_{1}> T-L_{1}\), which implies that it is more profitable to procure component 1 before or at time \(T-L_{1}\), i.e., \(0 \le t_{1}^{*}\le T-L_{1}\). Define \(\hat{t}\doteq \frac{1}{\mu }\ln (\frac{h_{1}}{\mu c_{1,0}})\). If \(\hat{t}\le 0\), i.e., \(\frac{h_{1}}{\mu c_{1,0}}\le 1\), then \({\widehat{S}}(q,t_{1})\) is decreasing in \(t_{1}\in [0, T-L_1]\) for any q, and thus, \(t_{1}^{*}=0\). If \(0<\hat{t}\le T-L_1\), i.e., \(1<\frac{h_{1}}{\mu c_{1,0}}\le e^{\mu (T-L_{1})}\), then \(t_{1}^{*}=\hat{t}\). If \({\hat{t}} > T-L_1\), i.e., \(\frac{h_{1}}{\mu c_{1,0}}> e^{\mu (T-L_{1})}\), \({\widehat{S}}(q,t_{1})\) is increasing in \(t_{1}\in [0, T-L_1]\) for any q, and thus, \(t_1^*=T-L_1\).
Observe that for \(0\le t_{1}\le T-L_{1}\), the first partial derivative of \({\widehat{S}}(q,t_1)\) respect to q can be reexpressed as
Thus, the optimal order quantity should be
(c) Again from (B9), if \(\mu < 0\), \(\frac{\partial {\widehat{S}}(q, t_{1})}{\partial t_{1}} > 0\) for any \(t_{1} \le T-L_{1}\). That is, it is optimal to procure component 1 after time \(T-L_{1}\), i.e., \(t_1^*\ge T-L_1\). From the first derivative respect to q, we have
Clearly, \(q^{*}\) is increasing in \(t_1^*\). Recall that \(T-L_1 \le t_1^*\le T\). As a result, \({\underline{q}}\le q^*\le {\overline{q}}\).
By (B9), if \( q^*\le \frac{sde^{-\mu (T-L_1)}}{-\mu c_{1,0}}\), then \(\frac{\partial {\widehat{S}}(q,t_{1})}{\partial t_{1}}<0\) for any \(t_1>T-L_1\) and thus, \(t_1^*=T-L_1\). Following a similar logic, one can readily prove the results for other cases. \(\square \)
Proof of Theorem 2
The results for the scenario with an uncertain leadtime of component 2 are presented in Theorem 4. Consider the case with \(\mu >0\). It is clear that \(t_1^*=t_1^{u*}\). From the proof of Theorem 1, we know that
From (A8), in order to prove \(q^*\ge q^{u*}\), it suffices to prove \(h_{1}{\hat{T}}_{h1}+h_{2}{\hat{T}}_{h2}\ge h_1 (T-L_1-t_1^*).\) This result holds because \({\hat{T}}_{h1}\ge (T-L_1-t_1^*)\).
Now, turn to the case with \(\mu <0\). We prove the result by contradiction. Suppose that \(q^* < q^{u*}\). Note that from the proof of Theorem 1, \(t_1^*\) and \(q^*\) also satisfy (A2). Therefore, one can readily prove that \(t_1^{u*}\ge t_1^*\). Hence, we can obtain
This result contradicts our hypothesis and \(q^{u*}\le q^*\). By (A2), \(t_1^{u*}\ge t_1^*\). \(\square \)
Proof of Lemma 2
(i) We prove by contradiction that for any \(m_1<m_2\), \(t_{m_1}^*\ge t_{m_2}^*\). Suppose \(t_{m_1}^*<t_{m_2}^*\). That is, at time \(t_{m_1}^*\), it is optimal to procure component \(m_1\) and delay the procurement of component \(m_2\). Then, there always exists a small enough time interval dt such that component \(m_2\) will be procured after time \(t_{m_1}^*+dt\), i.e., \(t_{m_2}^*\ge t_{m_1}^*+dt\). Next, we show that the assembler will be better off by procuring component \(m_1\) at time \(t_{m_1}^*+dt\) than at time \(t_{m_1}^*\) with other decisions being identical. It suffices to show that procuring component \(m_1\) at time \(t_{m_1}^*+dt\) leads to smaller holding and tardiness costs.
Denote by \(T_d\) and \({\widehat{T}}_d\) the tardiness time of procuring component \(m_1\) at time \(t_{m_1}^*\) and \(t_{m_1}^*+dt\), respectively. By the definition, we can obtain
where the first equality follows from \(t_{m_2}^*\ge t_{m_1}^*+dt\) and \(L_{m_2}\ge L_{m_1}\).
As procurement decisions of other components are identical, so are their holding costs. Now, we check the holding cost of component \(m_1\). By the definition, \(T_{h,m_1}=T_d+T-(t_{m_1}^*+L_{m_1})\) and \({\widehat{T}}_{h,m_1}={\widehat{T}}_d+T-(t_{m_1}^*+dt+L_{m_1})\), where \(T_{h,m_1}\) and \({\widehat{T}}_{h,m_1}\) represents the holding costs of procuring component \(m_1\) at time \(t_{m_1}^*\) and \(t_{m_1}^*+dt\), respectively. Clearly, \(T_{h,m_1}>{\widehat{T}}_{h,m_1}\). As a result, it is better off by procuring component \(m_1\) at \(t_{m_1}^*+dt\), which contradicts the optimality of \(t_{m_1}^*\).
(ii) Consider component i in \(\Omega ^L\). To prove \(t_i^*\ge T-L_i\), we prove that for any time \(t_i\in [0, T-L_i)\), the assembler will be better off by procuring component i at time \(T-L_i\) than at time \(t_i\), with other decisions being identical. By the definition of \(T_d\), the tardiness costs under these two strategies are the same and thus, so are the holding costs of other components. However, procuring component i at time \(T-L_i\) will lead to a lower holding cost of component i. Therefore, it is better off to procure component i after time \(T-L_i\). Recall that component 1 must be procured before the due date and thus arrive before time \(T+L_1\). As a result, other components should arrive before time \(T+L_1\), which implies \(t_i^*\le T-(L_i-L_1)\). \(\square \)
Proof of Lemma 3
We now prove that the Problems (13) and (14) are equivalent. By Observation 2, we only need to prove \({\widehat{U}}^*(q,c_{1,0})\ge U^*(q,c_{1,0})\). To achieve this, we propose for Problem (13) a feasible strategy as follows: \(t_i^*=T-L_i\), for \(i=2,\ldots ,m\), and \(t_1^*={\widehat{t}}_1^*\), under which the total expected profit of Problem (13) is \(U^*(q,c_{1,0})\).
We use a result of Problem (14) from Lemma 4(i) that if \(\mu \ge 0\), \({\widehat{t}}_1^*\le T-L_1\). Therefore, under the constructed strategies, all components arrive before due date and thus, \(T_d={\widehat{T}}_d=0\). Moreover, as component \(i\in \Omega ^L\) arrive at its due date, no holding cost is incurred, i.e., \(T_{h,i}=0\), \(i\in \Omega ^L\). Finally, \(T_{h,1}={\widehat{T}}_{h,1}=T-({\widehat{t}}_1^*+L_1)\). In summary, all the costs in two systems, as well as the expected revenue, are equal. Therefore, under the proposed feasible strategy, Problem (13)’s expected profit is \(U^*(q,c_{1,0})\). Consequently, the optimal expected profit of problem (13) satisfies \({\widehat{U}}^*(q,c_{1,0})\ge U^*(q,c_{1,0})\) and the proposed strategy is optimal. \(\square \)
Proof of Lemma 4
Let \({\widehat{U}}(q, t_1,c_{1,t_{1}})\) be the maximal expected total profit at time \(t_1\) with the current price \(c_{1,t_{1}}\). Then, the function \({\widehat{U}}(q,t_1,c_{1,t_{1}})\) satisfies the Wald-Bellman equations (Shiryaev 1978):
with the boundary condition \({\widehat{U}}(q, T, c_{1,T})={\widehat{S}}(q,T,c_{1,T})\).
The first term in (B12) represents the total profit if the assembler makes the purchase immediately, while the second term represents the expected profit if the assembler waits for another dt units of time.
By Wald-Bellman equations (B12), it is easy to verify that for any time \(t_1\),
By Ito’s Lemma, we have
and hence,
Taking partial derivatives of \({\widehat{S}}(q,t_1,c_{1,t_{1}})\) respect to \(t_1\) and \(c_{1,t_{1}}\), we can obtain
Substituting (B15) into (B14), we can obtain
(i) If \(\mu < 0\), then for any \(t_{1}\le T-L_{1}\), \({\mathbb {E}}[d{\widehat{S}}(q,t_1,c_{1,t_{1}})]>0\). This, together with (B13), implies that
In other words, it is more profitable for the assembler to make the purchase at least after time \(T-L_{1}\). That is, \(T-L_{1}\le t_{1}^{*}\le T \).
(ii) For the case with \(\mu > 0\), we prove an equivalent result by induction that it is optimal to purchase immediately for any time \(t_1\in (T-L_1, T]\) and observed price. The result clearly holds for time T. Suppose it also holds for time \(t_1+dt\), i.e.,
Then, for time \(t_1\), we have that
By (B16) and \(\mu >0\), one can easily verify that \({\widehat{S}}(q, t_{1},c_{1,t_{1}})>{\mathbb {E}}[{\widehat{S}}(q,t_1+dt,c_{1,t_{1}+dt})\mid c_{1,t_{1}}]\). That is, it is optimal to purchase immediately and the induction completes. \(\square \)
Proof of Lemma 5
The main idea in the following is similar to that in the proof of Theorem 4.3 in Jacka and Lynn (1992). For any \(t \in [0,T]\), define the continuation value function as follows:
which means the optimal value the decision maker can expect if she decides not to stop at time t with \(c_t=x\). It is clear that the optimal policy is to stop at t if \(g(t,x)\ge V(t,x)\) and not to stop if \(V(t,x)>g(t,x)\). For any stopping time \(\tau _t\in (t,T]\), we have
Because \(\tau _t > t\), we can directly get from (B17) that if \(a>0\) and \(b\mu >0\),
and if \(a<0\) and \(b\mu <0\),
Note that (B18) implies that
for any \(\tau _t\in [0,T]\) and \(c_t\). Then the decision maker should always wait until the time horizon T and part (i) is proved. In contrast, (B19) results in \(V(t,c_t)<g(t,c_t)\) for any \(t\in [0,T]\) and \(c_t\), which means the decision maker should immediately stop at \(t=0\). Then, part (ii) holds.
Next, we prove parts (iii) and (iv). Note that for any stopping time \(\tau _t\in (t,T]\) and \(c_t\),
Now suppose a stopping time \(\tau _t^x \in (t,T]\) is optimal for \(c_t=x\), we have
On the other hand, \(\tau _t^x\) may not necessarily be optimal for \(c_t=y\) and \(y \ne x\), which implies
where the last equality is because \(\tau _t^x\) has nothing to do with \(y \ne x\).
(iii) Assume that the decision maker should wait at \(c_t=x\), which means \(V(t,x)-g(t,x)>0\). If \(b \mu <0\), then for any \(y<x\), (B20) and (B21) implies \(V(t,y)-g(t,y)>V(t,x)-g(t,x)>0\) and then the decision maker should wait at \(c_t=y\), too. Hence, part (iii) holds, where \({C}_{Lt}\) is the lowest price at which it is optimal to stop.
(iv) Assume that the decision maker should wait at \(c_t=x\), which means \(V(t,x)-g(t,x)> 0\). If \(b \mu >0\), then for any \(y>x\), (B20) and (B21) implies \(V(t,y)-g(t,y)>V(t,x)-g(t,x)>0\) which in turn implies that the decision maker should wait at \(c_t=y\), too. Then, part (iv) holds, where \({C}_{Ht}\) is the highest price at which it is optimal to stop. \(\square \)
Proof of Theorem 3
Lemma 3 has already proved that the original system is essentially equivalent to a standard stopping time problem involving only component 1 for the case with \(\mu \ge 0\). Moreover, \(t_i^*=T-L_i\), for \(i=2,\ldots ,m\). Clearly, when \(\mu <0\) and \(\Omega ^L\) is an empty set, the original system is also equivalent to an OSS problem. Therefore, we only need to study Problem (14).
(i) For the case with \(\mu > 0\), it is clear that Problem (16) is equivalent to the following problem:
Problem (B22) corresponds to the case of Lemma 5 with \(a=qh_1\) and \(b=-\,q\). Then, part (iii) of Lemma 5 implies that there exists a threshold price \({\tilde{C}}_{1,t}\) at which the assembler should procure component 1 at t if and only if \(c_{1,t} \ge {\tilde{C}}_{1,t}\). That is, the upward base-price procurement policy is optimal for component 1.
(ii) If \(\mu <0\) and \(\Omega ^L\) is an empty, Problem (17) is equivalent to the following problem:
Clearly, Problem (B23) corresponds to the case of Lemma 5 with \(a=-\,sd\) and \(b=-\,q\). Then, part (iv) of Lemma 5 implies there exists a threshold price \({\widehat{C}}_{1,t}\) such that the assembler should procure component 1 at t if and only if \(c_{1,t} \le {\widehat{C}}_{1,t}\). That is, the downward base-price procurement policy is optimal for component 1. \(\square \)
Proof of Proposition 1
(i) In the case with \(\mu > 0\), the assembler should order component 1 before time \(T-L_1\), i.e., \(0 \le t_{1}^{*}\le T-L_{1}\). The original problem is thus reduced to the one stated in (16). In this case, we have
Define \({\widehat{U}}^1(q, t_1,c_{1,t_{1}})\) as the maximal expected total profit at time \(t_1\) with the current unit price \(c_{1,t_{1}}\) for Problem (16). To prove the result, it suffices to prove that when \(c_{1,t_1}<\frac{h_{1}}{\mu }\), it is optimal to wait, which means that for any \(c_{1,t_1}<\frac{h_1}{\mu }\),
Recall that for any time \(t_1\), \({\widehat{S}}^1(q, t_{1},c_{1,t_{1}})\le {\widehat{U}}^1(q,t_1,c_{1,t_{1}})\). Therefore, we can obtain
which by (B12) implies that when \(c_{1,t_1}<\frac{h_{1}}{\mu }\), it is optimal to wait which in turn implies that \({\tilde{C}}_{1,t}\ge \frac{h_{1}}{\mu }\).
(ii) In the case with \(\mu < 0\), the assembler should order component 1 after time \(T-L_1\), i.e., \(T-L_1\le t_{1}^{*}\le T\). The original problem is thus reduced to the one stated in (17). In this case, we have
Define \({\widehat{U}}^2(q, t_1,c_{1,t_{1}})\) as the maximal expected total profit at time \(t_1\) with the current unit price \(c_{1,t_{1}}\) for Problem (17). To prove the result, it suffices to prove that when \(c_{1,t_1}>\frac{-sd}{\mu q}\), it is optimal to wait, which means that for any \(c_{1,t_1}>\frac{-sd}{\mu q}\),
Recall that for any time \(t_1\), \({\widehat{S}}^2(q, t_{1},c_{1,t_{1}})\le {\widehat{U}}^2(q,t_1,c_{1,t_{1}})\). Therefore, we can obtain
which by (B12) implies that when \(c_{1,t_1}>\frac{-sd}{\mu q}\), it is optimal to wait at time \(t_1\) which in turn implies that \({\widehat{C}}_{1,t}\le \frac{-sd}{\mu q}\). \(\square \)
Proof of Proposition 2
Recall that given the procurement quantity q, \(t_1^{*}\) can be determined by solving the following optimal stopping problem (see (14)):
Moreover, Lemma 4 implies (i) when \(\mu >0\), \(0 \le t_1^{*} \le T-L_1\) and \(T_d=0\); (ii) when \(\mu <0\), \(T-L_1 \le t_1^{*} \le T\) and \(T_d=t_1^*+L_1-T\); (iii) when \(\mu =0\), \(t_1^{*}=T-L_1\).
By Theorem 3.1 of Chen and Liu (2014), we have
Hence the optimal \(q^{*}\) can be characterized by the following equations: (a) when \(\mu >0\),
(b) when \(\mu <0\),
(c) when \(\mu =0\),
To obtain the bounds for \(q^{*}\), we first observe the following property based on the dynamics of \(c_{1,t}\):
Then, for \(\mu >0\), because \(0 \le t_1^{*} \le T-L_1\), we know
and hence
which implies the bounds for \(q^{*}\). The bounds for \(q^{*}\) when \(\mu <0\) can be obtained in a similar way.
Proof of Lemma 6
For notation convenience, we take two components for example, i.e., \(\Omega ^L=\{2\}\).
(i) We first prove \(T-L_1\le t_1^*\le T\). Given q and \(t_i^*\) where \(i\in \Omega ^L\), it is easy to prove that \(T_d\) and \(T_{h,i}\), \(i\in \Omega ^L\), is independent of \(t_1\) for any \(t_1\le T-L_1\), and that the holding costs of component 1 is strictly decreasing in \(t_1\le T-L_1\). Note that the price of component 1 has a decreasing trend. Therefore, the expected total costs are decreasing in \(t_1\le T-L_1\). That is, the assembler will be better off procuring component 1 at time \(T-L_1\) than before time \(T-L_1\) which in turn implies \(T-L_1\le t_1^*\le T\).
(ii) Now we prove by contradiction that \(t_2^*+L_2-L_1\le t_1^*\le T\). If \(t_1 < t_2+L_2-L_1\), then \(T_d=t_2+L_2-T\), \(T_{h1}=t_2+L_2-t_1-L_1\), \(T_{h2}=0\), and hence
It is obvious that \({\mathbb {E}}[dS]>0\) which implies that the assembler will never make procurement before \(t_2+L_2-L_1\), which verifies (ii).
(iii) Applying Theorem 3.1 of Chen and Liu (2014) to the Problem (11) with \(\mu <0\) and \(\Omega ^L=\{2\}\), we can get:
which implies
Because \(h_1T_{h,1} \ge 0\), \( h_2T_{h,2} \ge 0\), and \({\mathbb {E}}[c_{1,t_1^{*}}|c_{1,0}]={\mathbb {E}}[c_{1,0}e^{\mu t_1^{*}}|c_{1,0}] \ge e^{\mu T}c_{1,0}\), the bounds for \(q^{*}\) are,
\(\square \)
Rights and permissions
About this article
Cite this article
Yang, Y., Wang, J., Chen, Y. et al. Optimal procurement strategies for contractual assembly systems with fluctuating procurement price. Ann Oper Res 291, 1027–1059 (2020). https://doi.org/10.1007/s10479-019-03314-y
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10479-019-03314-y