* Corresponding author. Tel.: #886-35-712121ext.57125. E-mail address: [email protected] (C. Chiang).
Order splitting under periodic review inventory systems
Chi Chiang*
Department of Management Science, National Chiao Tung University, Hsinchu, Taiwan, Republic of China
Accepted 12 April 2000
Abstract
Most research on order splitting have focused on the reduction of safety stock in the multiple sourcing setting. Moreover, all works study the use of order splitting for the continuous review inventory systems. In this paper, we investigate the possibility of the multiple-delivery arrangement in the sole sourcing environment. In addition, we concentrate on the reduction of cycle stock for periodic review systems. We show that splitting an order into multiple deliveries can signi"cantly reduce the total cost especially if the cost of despatching an order for an item is not small. Although the use of information technology such as EDI decreases the ordering cost and thus shortens the period length, order splitting remains a cost-e!ective approach as long as the cost of despatching an order is not close to zero. 2001 Elsevier Science B.V. All rights reserved.
Keywords: Inventory; Order-splitting; JIT; Periodic review; Information technology
1. Introduction
The use of order splitting during an order cycle seems to have received much attention recently. For example, Kelle and Silver [1] analyze the safety stock reduction by order splitting assuming the Weibull-distributed lead times, and Sculli and Shum [2] present numerical results on the e!ect of order splitting on the lead-time demand. Ramasesh et al. [3], Lau and Zhao [4], and Chiang and Benton [5] further develop a total cost model, respectively to obtain the optimal reorder point and order quantity jointly. However, these works have focused on the reduction of safety stock in the multiple sourcing setting. In comparision with cycle stock, safety stock is only a small portion of a company's inventory. In a recent paper, Chiang
and Chiang [6] propose the arrangement of multiple deliveries during each order cycle, and consider the reduction of cycle stock in the sole sourcing environment. All of these studies have concentrated on the use of order splitting for con-tinuous review inventory systems. In this paper, we investigate the use of order splitting during each period (also called order cycle thereafter) for peri-odic review (R, S) systems (without a reorder point). In addition, we focus on the reduction of cycle stock in the sole sourcing setting. Our research also provides a rationale for the just-in-time (JIT) frequent-delivery approach.
In a typical periodic review system, an order quantity which brings the inventory level to S is placed with a speci"c vendor whenever inventory is reviewed every period of length R. Such an operat-ing policy, known as a replenishment cycle system, is often found in practice (see, e.g. [7]). Although the use of computer systems nowadays has made
0925-5273/01/$ - see front matter 2001 Elsevier Science B.V. All rights reserved. PII: S 0 9 2 5 - 5 2 7 3 ( 0 0 ) 0 0 0 4 6 - 3
continuous review systems more popular, periodic review systems are still applied in many situations (see, e.g. [8]). Often, periodic systems are found to have the review periods which are longer than the supply lead times. For instance, a retailer may place regular replenishment orders biweekly while the supply lead time is of the order of one week.
It is generally assumed in a periodic system that the whole order quantity from the supplier arrives in a single delivery in each period. It is possible, however, that the buyer could have the supplier agree to order splitting so that portions of an order quantity arrive at the receiving point at di!erent times of a period. For companies who work with their suppliers on a long-term relationship, this multiple-delivery approach is particularly feasible and useful. For example, Hotai Motor Co. Ltd., the distributor of Toyota products (and the largest auto distributor) in Taiwan, has recently adopted this multiple-delivery approach, which will be com-monly used in the future by other auto distributors. Hotai Motor Co. Ltd. orders thousands of service parts for domestically manufactured cars (such as Toyota Corona Exsior and Tercel) monthly from six major suppliers (not including the Toyota Mo-tor Company in Japan) (manufacturing of these cars is in a di!erent company). Each of the six suppliers makes at least 10 deliveries (low-usage items are shipped less frequently than high-usage items) per month to the central warehouse of Hotai Motor Co. Ltd.
Apparently, the bene"t of this multiple-delivery approach is the reduction in the average cycle stock, while the disadvantage of this approach is that ordering costs (which include transportation and inspection costs) may increase. The major goal of this research is to develop a multiple-delivery model and evaluate this tradeo!. In addition, if it bene"ts the buyer to arrange multiple deliveries with the supplier, does there exist an optimal num-ber of deliveries per order cycle? This research also investigates this issue.
We assume that both R and S are decision vari-ables. In addition, we assume that lead time is constant, demand is non-negative and indepen-dently distributed in disjoint time intervals, and that demand during a time interval of length q is normally distributed with mean kq and variance
pq. Note that in practice R is often predetermined by the "rm. For example, a retailer may coordinate a group of items to a distribution center weekly or biweekly. Also, vendors in a department store often make routine visits to customers to take fresh orders [8]. There may exist other practical or or-ganizational considerations (see, e.g. [9]). In this paper, however, we assume that R is a variable inside the model.
In addition, as the supply chain management of materials has been a trend in industry, some "rms have invested in information technology to reduce the communication and transaction time among trading partners. The use of electronic data inter-change (EDI) in inventory control systems has been particularly noteworthy and the bene"ts of reduced logistics and order processing costs are reported [10]. In this paper, we also discuss the impact of the reduction of ordering costs on the shortening of the review periods and on the performance of the multiple-delivery model, after the "rm and its sup-plier(s) have decided to invest in EDI (i.e., the deci-sion of establishing an EDI-based inventory system has been made and is not considered inside the model).
This paper is organized as follows. First, we brie#y review the ordinary single-delivery approach under periodic review systems. Then we present a two-delivery model, which is followed by some computational results. Next, we generalize the analysis to the multiple-delivery model. Finally, this paper ends with the conclusion.
2. Review of the single-delivery approach
We "rst review the ordinary single-delivery approach. Let ¸ be the constant lead time and
h(>) the probability density function (PDF) for
the demand > over a period of length R. Then
h(>) is N(kR, pR). Suppose that we review
inventory at the time point t and > is the demand during the preceding period (t!R, t). Then we will order > and raise the inventory position up to
S which should be large enough to meet the
de-mand for the upcoming time interval of length
R#¸. Let B(¸) be the average backorder that
Fig. 1. A two-delivery model.
t#R#¸. Then,
B(¸)"p(R#¸G(k), (1) where
k"S!p(R#¸k(R#¸) (2) and G( ) ) is the partial expectation function tabled in Brown [11] or Silver and Peterson [7]. Instead of having to estimate the backorder cost, we use a service level (SL) constraint for the objective func-tion (as in [6]). Service level is de"ned as the per-centage of demand to be served directly from stock. For the single-delivery approach, service level is given by
SL"100!100B(¸)
kR . (3)
Let D be the average annual demand, A the "xed ordering cost, J the review cost, and h the annual carrying cost per unit. To simplify the notation, we incorporate the review cost into the ordering cost, i.e., A also includes the review cost. The ordering cost A, as described by Lau and Zhao [4], consists of two components. One is the cost of despatching an order for an inventory item each time, denoted by O, such as administrative and processing costs (note that the review cost is included in O). The other is the cost of receiving an incoming procure-ment, denoted by I, such as costs of transportation, handling and inspection of the procurement after it arrives, etc. O and I togeter constitute the "xed ordering cost A (i.e., A"O#I). The use of in-formation technology such as EDI will decrease
O and thus shorten the period length R. In Section
4, we will investigate this issue. As the average amount ordered per period is kR and the safety stock is S!k¸!kR, the decision problem for the single-delivery model can be expressed by Min C(R, S)"[D(O#I)/kR] #h
S!k¸!kR 2 (4) s.t. 100!100B(¸) kR *t, (5)wheret is a preassigned service level. Note that the expression for average on-hand inventory is only an approximation, i.e., it assumes that the average backorder level is quite small (see, e.g. [12]). Given the nature of the problem, the optimal solution will automatically satisfy constraint (5) at equality. To "nd the optimal R and S, we use (5) to "nd the optimal S for a given value of R. Then we tabulate the total cost as a function of R to determine the optimal R [12].
3. A two-delivery model
We now present a two-delivery model for the periodic review (R, S) system. Let ¸ (which is smaller than R) be the inter-arrival time between the "rst and second shipments. We suppose that we order > (i.e., demand during the preceding period) at the review epoch t and the supplier agrees to deliver part of the order quantity after time ¸ and the remaining part after time ¸#¸ (see Fig. 1). (Notice that ¸#¸ need not be less than R as shown in Fig. 1.) Let (1!w)kR be the size of the second shipment and thus >!(1!w)kR is the size of the "rst shipment (note that the average size of the "rst shipment is wkR). The idea is to raise the inventory up to S!(1!w)kR in the "rst ship-ment. It is assumed that >'(1!w)kR. If
Fig. 2. Average cycle stock of the two-delivery model.
> should be less than (1!w)kR, there is only one shipment of size > delivered after ¸#¸. > is at least (1!w)kR if ¸ is not too short so that w is very small (the fact that w depends on ¸ is shown in the following computation). Intuitively, to make the two-delivery arrangement attractive, the inter-arrival time ¸ should not be too short, since otherwise the reduction in the average cycle stock (to be explained below) would be very small and the two-delivery approach may only re-sult in an increase in the material handling cost (part of the ordering cost). It is assumed for the time being that ¸ is "xed. Later we will explore how ¸ can have an impact upon the total cost of the buyer.
Let h(>) be the PDF for demand > during the upcoming time interval (t, t#¸#¸). Then
h(>) is N(k(¸#¸), p(¸#¸)). In the
single-delivery approach, it is generally assumed that the order placed at the review epoch t would clear all the shortages (if any) at the time of arrival (see, e.g. [12]). In the two-delivery model, shortages may not be all cleared at time t#¸ (since part of the order arrives at time t#¸#¸). More im-portantly, shortages can occur between time t#¸ and t#¸#¸. Thus, we need to compute the shortages that might build up before the receipt of the second shipment. Note that there is the possibil-ity of double-counting the same shortages as long as a shipment could not clear all the shortages at the time of arrival. Although this should rarely happen (since the average backorder level is quite small), we assume that we are willing to accept possible double-counting [6]. Let B(¸) denote the average backorder that might build up before the recipt of the second shipment. For the two-delivery approach, the service level is given by
SL"100!100B(¸)#B(¸) kR (6) and B(¸) is expressed as B(¸)"
1\\UI0>![S!(1!w)kR]h(>) d> "p(¸#¸G(k), (7) where k"[S!(1!w)kR!k(¸#¸)]/p(¸#¸. (8) Also, if we let the arrival of the "rst shipment of an order initiate a cycle, then a cycle consists of two time intervals of length ¸ and R!¸ (see Fig. 1). The average cycle stock is wkR!k¸/2 for the time interval of length ¸ and kR/2!k¸/2 for the time interval of length R!¸ (see Fig. 2). Thus, the overall average cycle stock is kR/2!(1!w)k¸. By splitting an order into two deliveries, we see that the average cycle stock is reduced by (1!w)k¸ (if R remains the same as in the single-delivery model).On the other hand, when two deliveries of an order are arrenged with the supplier, the ordering cost may increase. While the cost of despatching an order is unchanged, the cost of receiving incoming procurements may nearly double. We assume that when two deliveries of an order are arranged with the supplier, the ordering cost becomes O#2I.
It follows that the decision problem for the two-delivery model discussed above is expressed by Min C(R, S, w)"[D(O#2I)/kR]
#h
S!k¸!kR2!(1!w)k¸
(9)Table 1
E!ect of the inter-arrival time ¸ on the performance of the two-delivery model. Data: k"10 units/day, p"2 units, 1 year"250 days,
A"$2(O"I"$1),t"99.90, h"$0.5/unit/year
¸ S w C(R, S) ¸ S w C(R, S, w)
(A) ¸"10 days, R"20 days (B) ¸ "10 days, R"25 days
8 321 0.3774 73.1 10 372 0.3792 72.5 9 321 0.4301 72.4 11.5 372 0.4422 71.4 9.5 321 0.4564 72.2 12 372 0.4632 71.3 10.0 321 0.4827 72.1 12.5 372 0.4841 71.3 10.5 321 0.5090 72.2 13 372 0.5051 71.3 11 321 0.5353 72.4 13.5 372 0.5260 71.5 12 321 0.5878 72.3 15 372 0.5888 72.7
(C) ¸"5 days, R"20 days (D) ¸"5 days, R"25 days
8 269 0.3719 71.9 10 320 0.3766 71.3 9 269 0.4250 71.1 11.5 320 0.4400 70.3 9.5 269 0.4515 70.9 12 320 0.4611 70.2 10.0 269 0.4779 70.9 12.5 320 0.4821 70.1 10.5 269 0.5044 71.0 13 320 0.5032 70.2 11 269 0.5308 71.2 13.5 320 0.5243 70.4 12 269 0.5836 72.0 15 320 0.5874 71.6 s.t. 100!100B(¸)#B(¸) kR *t or B(¸)#B(¸)!(100!100t)kR)0, (10) 0(w(1. (11)
Note that (10) can be easily shown to be convex with respect to S and w (for a given R). Given the nature of the problem, the optimal solution will always have constraint (10) held at equality. By formulating the Lagrangian
[D(O#2I)/kR]#h
S!k¸!kR2 !(1!w)k¸
#jB(¸)#B(¸)!(100!100t)kR
,and setting the derivatives with respect to S, w and j equal to zero, we can obtain
¸
R"P(k)#P(k)P(k) , (12) B(¸)#B(¸)"(100!100t)kR, (13)
where 0(w(1 and P( ) ) is the complement of the cumulative distribution function for the standard normal variable. It is evident from (12) and (13) that the optimal S and w (given a certain R) do not depend on the values of O, I, and h. To "nd the optimal combination of R, S, and w, we use (12) and (13) to "nd the optimal S and w for a given R. Then we tabulate the total cost as a function of R to determine the optimal R.
4. Computational results
We investigate the e!ect of the inter-arrival time ¸on the performance of the two-delivery model. We also examine the e!ect of cost parameters, demand variability, and service level on the perfor-mance of the two-delivery model relative to the single-delivery model.
4.1. Ewect of the inter-arrival time
We "rst investigate the e!ect of ¸ on the perfor-mance of the two-delivery model. It appears from Table 1 that the total cost is at a minimum when ¸ is approximately equal to R/2. Note that this result is also obtained under other levels of ¸ and
Table 2
Single-delivery model versus two-delivery model under di!erent levels of O. Data:k"10 units/day, p"2 units, 1 year"250 days, ¸"10 days, ¸"R/2, I"$1, h"$0.5/unit/year, t"99.90
O Single-delivery model Two-delivery model % savings
R S C(R, S) R S w C(R, S, w) $0.0 10 217 58.5 20 321 0.4818 59.6 !1.88 0.125 11 227 61.6 20 321 0.4818 61.2 0.65 0.25 11 227 64.4 20 321 0.4818 62.7 2.64 0.5 12 237 69.8 22 342 0.4780 65.7 5.87 1.0 14 258 79.7 24 362 0.4822 71.2 10.67 2.0 17 288 95.6 28 403 0.4818 80.9 15.38 Table 3
Single-delivery model versus two-delivery model under di!erent levels of I. Data:k"10 units/day, p"2 units, 1 year"250 days, ¸"10 days, ¸"R/2, O"$1, h"$0.5/unit/year, t"99.90
I Single-delivery model Two-delivery model % savings
R S C(R, S) R S w C(R, S, w) $0.25 11 227 64.4 17 291 0.4744 52.7 18.17 0.5 12 237 69.8 20 321 0.4818 59.6 14.61 1.0 14 258 79.7 24 362 0.4822 71.2 10.67 2.0 17 288 95.6 31 433 0.4862 89.5 6.38 4.0 22 339 121.3 41 535 0.4861 117.2 3.38 8.0 30 420 160.0 51 697 0.4895 157.8 1.37
R, although the computations are not shown here.
We are unable to prove this result due to the partial expectation functions involved. However, we could explain as follows. As mentioned above, we can let the arrival of the "rst shipment of an order initiate a cycle. Then if ¸"R/2, the arrival of second shipment is exactly halfway through a cycle. It is as though we place an order (in a single shipment) every period of length R/2 and the arrival epochs would be the same. For computational purposes, it is much easier to "x ¸ at R/2 and "nd the optimal combination of R, S and w, instead of treating ¸also as a variable. Moreover, it is easy for both the buyer and the supplier to implement the two-delivery contract when ¸ is simply "xed at R/2. Consequently, we set ¸"R/2 in the computa-tions thereafter.
4.2. Ewect of cost parameters
We next consider the e!ect of cost parameters. Note that the cost structure is characterized by the ratio of A/h or (O#I)/h. Thus, we "x the value of
h to investigate the e!ect of the amount of A (relative
to h) on the performance of the two-delivery model. Since A"O#I, we vary the value of O and I, respectively, to examine their e!ect. As we see from Tables 2 and 3, the two-delivery model becomes more attractive as O increases and I decreases, re-spectively (other things being equal). These results came with no surprise. Since the ordering cost equals
O#2I in the two-delivery model, a smaller I results
in only a slight increase in the ordering cost and thus the two-delivery approach becomes more ef-fective by reducing the average cycle stock.
Table 4
Single-delivery model versus two-delivery model under di!erent levels of the ratio O/(O#I). Data:k"10 units/day, p"2 units, 1 year"250 days, ¸"10 days, ¸"R/2, A"O#I"$2, h"$0.5/unit/year, t"99.90
O Single-delivery model Two-delivery model % savings
R S C(R, S) R S w C(R, S, w) $2.0 14 258 79.7 19 311 0.4801 59.6 25.22 1.75 14 258 79.7 21 331 0.4851 62.8 21.20 1.5 14 258 79.7 22 342 0.4780 65.7 17.56 1.25 14 258 79.7 23 352 0.4802 68.5 14.05 1.0 14 258 79.7 24 362 0.4822 71.2 10.67 0.75 14 258 79.7 25 372 0.4841 73.8 7.40 0.5 14 258 79.7 26 382 0.4860 76.2 4.39 0.25 14 258 79.7 27 393 0.4801 78.6 1.38 0 14 258 79.7 28 403 0.4818 80.9 !1.51
On the contrary, a smaller O saves the two-deliv-ery model vtwo-deliv-ery little of the ordering cost and thus makes it less e!ective. This has a very important implication. As we mentioned before, the use of EDI in inventory systems will lower O and thus the optimal period length R is shortened, as can be seen from Table 2. If we assume that the "rm and its supplier(s) have made an investment in EDI, the use of order splitting does not appear particularly at-tractive as O is decreased to a smaller level in the long run. Nevertheless, the "rm bene"ts from the use of the delivery approach, since the two-delivery approach obtains a lower total cost than the traditional single-delivery approach as long as
O is not decreased to a level of near zero. This result
also is apparent from Table 4. If we "x the value of
A but change the proportion of O in A, we see that
the two-delivery approach yields a smaller percent-age cost savings as the ratio of O/A decreases to zero. In an extreme case of O/A+0, splitting an order into two deliveries during each cycle will increase the total cost.
As a note, the two-delivery approach yields a larger R than the single-delivery model, although the inter-arrival time ¸ between the two deliveries is shorter than the optimal R of the single-delivery model. This implies that the buyer will order a larger quantity and thus is more likely to ob-tain quantity discounts under the two-delivery model.
4.3. Ewect of demand variability and service level
Next, we examine the e!ect of demand variability and service level on the performance of the single-delivery model versus the two-single-delivery model. It appears from Tables 5 and 6 that the two-delivery model performs better under lower levels of p or t (other things being equal). This is because there are two stockout possibilities (one more than the single-delivery model) during each order cycle in the two-delivery model. Thus, the two-delivery model is less vulnerable to stockouts if demand variability or service level is low. This agrees with the "nding of Chiang and Chiang [6].
5. A multiple-delivery model
We consider the possibility of the arrengement of
n shipments during each order cycle. Let ¸G"R/n, i"2,2, n, be the inter-arrival time be-tween the (i!1)th and ith shipments. We suppose that the supplier agrees to deliver the ith shipment after GH¸H, i"1,2, n, and the "rst shipment has size >!(1!w)kR (note again that the aver-age size of the "rst shipment is wkR), the second shipment has size wkR,2, the (n!1)th shipment has size wL\kR, and the nth shipment has size (1! L\HwH)kR. Let B(¸G), i"2,2, n be the aver-age backorder that might build up before the
Table 5
Single-delivery model vs. two-delivery model under di!erent levels ofp. Data:k"10 units/day, 1 year"250 days, ¸"10 days, ¸"R/2, A"$2, (O"I"$1),t"99.90, h"$0.5/unit/year
p Single-delivery model Two-delivery model % savings
R S C(R, S) R S w C(R, S, w) 0.5 14 243 72.2 24 344 0.4953 63.0 12.74 1.0 14 247 74.2 24 349 0.4958 65.5 11.73 2.0 14 258 79.7 24 362 0.4822 71.2 10.67 4.0 14 280 90.7 24 390 0.4595 83.8 7.61 Table 6
Single-delivery model vs. two-delivery model under di!erent levels oft. Data: k"10 units/day, p"2 units, 1 year"250 days, ¸"10 days, ¸"R/2, A"$2(O"I"$1), h"$0.5/unit/year
t Single-delivery model Two-delivery model % savings
R S C(R, S) R S w C(R, S, w)
95.00 14 235 68.2 24 336 0.5066 59.6 12.60
99.00 14 247 74.2 24 350 0.4895 65.6 11.59
99.90 14 258 79.7 24 362 0.4822 71.2 10.67
99.99 14 266 83.7 24 371 0.4765 75.3 10.03
receipt of the ith shipment. Then,
B(¸G)"p
G H¸HG(kG), i"2,2, n, (14) where kG"S!1!G\ HwHkR!k G H¸Hp G H¸H . (15) We assume that the ordering cost is O#nI whenn shipments during each cycle are arrenged with the
supplier. Noticing that the average cycle stock is reduced by (1!w)k¸#(1!w!w)k¸#2 #(1! L\
HwH)k¸L, we can express the decision
problem as Min C(R, S, w,2, wL\)"[D(O#nI)/kR] #h
S!k¸!kR 2 !(1!w)k¸ ! (1!w!w)k¸!2!1!L\ HwHk¸L (16) s.t. L GB(¸G)! (100!t)kR 100 "0, (17) 0(wH(1, j"1,2, n!1, (18) L\ HwH(1. (19) To "nd the optimal combination of S and wH,j"1,2, n!1, for a given R, we formulate the
Lagrangian of this multiple-delivery model and set the derivatives with respect to S and wH, j"1,2,
n!1, equal to 0, yielding ¸ R"P(k)#P(k)#2#P(kL)P(k) , ¸ R"P(k)#P(k)#2#P(kL)P(k) ,2, ¸L R"P(k)#P(k)#2#P(kL)P(kL) , (20) L GB(¸G)" (100!t)kR 100 (21)
Table 7
Two-delivery model versus three-delivery model under di!erent levels of the ratio O/(O#I). Data:k"10 units/day, p"2 units, 1 year"250 days, ¸"10 days, O#I"$2, h"$0.5/unit year, t"99.90
O Two-delivery model Three-delivery model % savings
R S w C(R, S, w) R S w w C(R, S, w,w) $2.0 19 311 0.4801 59.6 24 364 0.3050 0.3463 51.1 14.26 1.5 22 342 0.4780 65.7 29 415 0.3076 0.3458 60.6 7.76 1.0 24 362 0.4822 71.2 33 456 0.3086 0.3446 68.7 3.51 0.5 26 382 0.4860 76.2 37 496 0.3120 0.3462 75.8 0.52 0 28 403 0.4818 80.9 41 537 0.3124 0.3448 82.2 !1.61 Table 8
Multiple-delivery models. Data:k"10 units/day, p"2 units, 1 year"250 days, ¸"10 days, O"$1.5, I"$0.5, h"$0.5/unit/year, t"99.90 n R S C(R, S, w,2, wL\) n R S C(R, S, w,2, wL\) 2 22 342 65.7 3 29 415 60.6 4 36 487 58.2 5 42 550 56.9 6 48 612 56.1 7 55 684 55.7 8 61 746 55.5 9 67 808 55.4 10 74 880 55.5 11 80 942 55.7
(see the appendix for details). Since ¸G"R/n,
i"2,2, n, it follows from (20) that k"k"2
"kL. Hence, we can use the following procedure to obtain the optimal S and wH, j"1,2, n!1.
Step 1. Substitute k"k,2, kL"k into (21)
to obtain k and thus S.
Step 2. Use kG"k, i"2,2, n, to obtain wH by
using (15), j"1,2, n!1, respectively.
We then tabulate the total cost as a function of
R to determine the best R. Table 7 gives the
com-putational results for the relative performance of the two-delivery model versus the three-delivery model. As we see, the total cost may be further reduced if we split an order into three deliveries during each cycle.
A question arises at this point: does there exist an
optimal number of deliveries per cycle that results in
minimum total cost (as in [6]). To investigate this, we carry out the computation further. As we see, for example, the optimal number of deliveries per cycle is 9 in Table 8. This also illustrates the frequent-delivery approach that Hotai Motor Co. Ltd. (as mentioned in the introduction) employs to reduce the inventory carrying cost. As Hotai Motor Co.
Ltd. works with its major suppliers on a long-term relationship, the ordering cost of an item is small. Often, delivery of a split procurement for an item is part of a joint shipment which includes hundreds of items, and there is no inspection after the procure-ment arrives. The suppliers also absorb some of the transportation cost.
In summary, if the ordering cost structure agrees with what we assume, the buyer should consider negotiating an optimal number of deliveries for each cycle with the supplier.
6. Conclusion
In this paper, we investigate the possibility of the multiple-delivery arrangement during each order cycle for periodic review systems. We show that splitting an order into multiple deliveries can reduce the average cycle stock and thus the total cost, especially if the cost of despatching an order (which includes the review cost) is not small. Al-though the use of information technology such as EDI decreases the ordering cost and thus shortens
the period length, order splitting remains a cost-e!ective approach as long as the cost of despatch-ing an order is not close to zero. Moreover, we show that there exists an optimal number of delive-ries per cycle such that the lowest total cost is obtained. As very few assumptions are made in this research, "rms can apply the approach of order splitting in practice immediately, as long as mul-tiple shipments of an order can be arranged with suppliers. Finally, we should note that this research also provides a rationale for the JIT frequent-deliv-ery approach.
Appendix A
In this appendix, we derive expression (20) of Section 5. The Lagrangian including (16) and (17) with a multiplierj is [D(O#nI)/kR]#h
S!k¸!kR 2 ! (1!w)k¸!(1!w!w)k¸ !2!1!L\ HwHk¸L #j L HB(¸H)! (100!t)kR 100 .Di!erentiating it with respect to S and setting the derivative equal to zero, we obtain
j"h/(P(k)#P(k)#2#P(kL)). (A.1) Next, we di!erentiate the Lagrangian with respect to wL\, set it to zero, and substitute (A.1) into the expression to give
¸L/R"P(kL)/(P(k)#P(k)#2#P(kL)). (A.2) Then, we di!erentiate the Lagrangian with respect
wL\, set it to zero, and substitute (A.1) into the
expression to give
(¸L\#¸L)/R"(P(kL\)#P(kL))/(P(k) #
P(k)#2#P(kL)),
and substitute (A.2) into the above expression to yield
¸L\/R"P(kL\)/(P(k)#P(k)#2#P(kL)). (A.3) Continue this way and di!erentiate with respect to
wL\,2, and w to obtain respectively
¸L\/R"P(kL\)/(P(k)#P(k)#2#P(kL)), $
¸/R"P(k)/(P(k)#P(k)#2#P(kL)), The above expressions together with (A.2) and (A.3) are expression (20) in Section 5.
References
[1] P. Kelle, E.A. Silver, Safety stock reduction by order split-ting, Naval Research Logistics 37 (1990) 725}743. [2] D. Sculli, Y. Shum, Analysis of a continuous review stock
control model with multiple suppliers, Journal of Opera-tional Research Society 41 (1990) 873}877.
[3] R.V. Ramasesh, J.K. Ord, J.C. Hayya, A.C. Pan, Sole versus dual sourcing in stochastic lead-time (s, Q) inven-tory models, Management Science 37 (1991) 428}443. [4] H. Lau, L. Zhao, Optimal ordering policies with two
suppliers when lead times and demands are all stochastic, European Journal of Operational Research 68 (1993) 120}133.
[5] C. Chiang, W.C. Benton, Sole sourcing versus dual sourc-ing under stochastic demands and lead times, Naval Research Logistics 41 (1994) 609}624.
[6] C. Chiang, W.C. Chiang, Reducing inventory costs by order splitting in the sole souring environment, Journal of the Operational Research Society 47 (1996) 446}456. [7] E.A. Silver, R. Peterson, Decision Systems for Inventory
Management and Production Planning, Wiley, New York, 1985.
[8] R.B. Chase, N.J. Aquilano, Production and Operations Management, Irwin, Homewood, IL, 1997.
[9] M.K. Starr, Operations Management, Boyd & Fraser, MA, 1996.
[10] T. Lester, Squeezing the supply chain, Management Today (1992) 68}70.
[11] R.G. Brown, Decision Rules for Inventory Management, Holt, Rinehart, and Winston, New York, 1967.
[12] G. Hadley, T.M. Whitin, Analysis of Inventory Systems, Prentice-Hall, Englewood Cli!s, NJ, 1963.
