Optimal Policies and Approximations for A Serial Multi-echelon Inventory System...

58
Optimal Policies and Approximations for A Serial Multi-echelon Inventory System with Time-correlated Demand Lingxiu Dong John M. Olin School of Business Washington University St. Louis, MO 63130 [email protected] Hau L. Lee Graduate School of Business Stanford University Stanford, CA 94305 [email protected] August 2000 Revision: November 2001 Subject classication: Inventory/production: approximations, multi-echelon, stochastic.

Transcript of Optimal Policies and Approximations for A Serial Multi-echelon Inventory System...

Optimal Policies and Approximations for A Serial

Multi-echelon Inventory System with

Time-correlated Demand

Lingxiu Dong

John M. Olin School of Business

Washington University

St. Louis, MO 63130

[email protected]

Hau L. Lee

Graduate School of Business

Stanford University

Stanford, CA 94305

[email protected]

August 2000

Revision: November 2001

Subject classification: Inventory/production: approximations,

multi-echelon, stochastic.

Abstract

Since Clark and Scarf’s pioneering work, most advances in multi-echelon inventory

systems have been based on demand processes that are time-independent. This paper

revisits the serial multi-echelon inventory system of Clark and Scarf, and develops

three key results. First, we provide a simple lower bound approximation to the

optimal echelon inventory levels and an upper bound to the total system cost for the

basic Clark and Scarf’s model. Second, we show that the structure of the optimal

stocking policy of Clark and Scarf holds under time-correlated demand processes

using Martingale model of forecast evolution. Third, we extend the approximation

to the time-correlated demand process, and study in particular for an auto-regressive

demand model the impact of leadtimes and auto-correlation on the performance of

the serial inventory system.

1 Introduction

In a seminal paper, Clark and Scarf (1960) study a serial multi-echelon inventory sys-

tem where demand occurs at the lowest echelon, and show that, when there is no fixed

ordering cost in the serial system, the optimal inventory control of the overall system

follows a base-stock policy for every echelon. The base-stock levels at each echelon

can be solved from a series of single location inventory problems with appropriately

defined penalty functions for not having enough inventory to bring the downstream

site to its target base-stock level.

Extensions of the Clark and Scarf approach have been developed for other

production/inventory systems. Schmidt and Nahmias (1985) use the Clark and Scarf

decomposition technique to characterize the optimal policy for an assembly system

with two components. Rosling (1989) shows, under appropriate assumptions, that a

general assembly system is equivalent to a serial system and thus Clark and Scarf’s

results apply. Axäter and Rosling (1993) establish the condition under which the

echelon and installation based policies are equivalent. Chen (2000) generalizes the

Clark and Scarf result to the serial systems and assembly systems where the material

flow from one level to another has to be in batches (e.g., truck loads). Iida(2000)

studies error bounds of near myopic policies in a multi-echelon system. Lee and

Whang (1999) show that a performance measurement scheme based on Clark and

Scarf’s penalty function exists that enables a decentralized serial inventory system to

achieve the optimal performance of a centralized system.

Although the optimal inventory policy follows a simple form of a base-stock

inventory policy, the computation of the optimal base-stock levels can be quite com-

plicated due to the complexity of the induced penalty cost functions. Federgruen and

Zipkin (1984) extend Clark and Scarf’s results to the infinite horizon case, and show

that the computation can be much simplified if one minimizes discounted cost or

long-term average cost. For a two-echelon system with Normal demand distribution,

they provide a closed-form equation for solving the optimal inventory levels and a

closed-form expression for the corresponding optimal cost. Similar calculations can

be carried over to a system with more than two echelons, but the complexity of the

computation increases considerably. Moreover, even with a closed-form expression,

the analysis of the system inventory performance is still quite tedious and often has

to rely on numerical methods. Gallego and Zipkin (1999) develop several heuristic

methods and conduct numerical studies on system performance sensitivity to the

stock positioning under different assumptions of inventory holding costs. Shang and

Song (2001) extend the restriction-decomposition heuristic in Gallego and Zipkin to

the echelon-stock setting under linear holding cost and give newsvendor type upper

and lower bounds to the optimal echelon policies.

Most of the research work on multi-echelon inventory systems has been based

on stationary demand processes that are independent over time. Although empiri-

cally, time-correlated demands are commonly observed (e.g. see Erkip et al., 1990

and Lee et al., 1999), there has been only limited work with time-correlated demand

processes. Chen and Zheng (1994) construct a simple proof of Clark and Scarf’s

2

results in an infinite-horizon case by finding a lower bound on the system-wide cost

and constructing a feasible policy to achieve such a lower bound. Using this lower-

bounding approach, Chen and Song (2001) relax the independent demand assumption

in the classic model and consider a demand process whose demand distribution is de-

termined by the state of an underlying finite-state Markov chain. They show that

echelon base-stock policies are still optimal when order-up-to levels are adjusted based

on the state of the underlying Markov chain, and provide an algorithm to compute

the optimal base-stock levels.

For time-series based demand processes, the martingale model of forecast evo-

lution (MMFE) developed by Graves et al. (1986, 1998) and Heath and Jackson

(1994) offers a powerful descriptive framework that incorporates both past demands

and other influential factors to characterize the forecast processes. Readers are re-

ferred to Heath and Jackson for an excellent discussion of motivation of MMFE,

to Toktay and Wein (2001) for a review on researches (Güllü, 1996) built upon it.

A common demand process used in the literature is the order-one auto-regressive, or

AR(1), process with positive auto-correlation (see, for example, Johnson and Thomp-

son, 1975; Lovejoy, 1990&1992; Fotopoulos and Wang, 1988; Reyman, 1989; Scarf,

1959&1960; Miller, 1986; Kahn, 1987; Erkip et al., 1990; Lee et al., 1997; and Lee

et al., 1999). In fact AR(1) process with minimum mean-square error forecast is a

special case of MMFE. More complex time series demand processes have also been

modeled more recently, such as the random walk model of Graves (1999) and Lee and

Whang (1998).

3

This paper develops three key results for the Clark and Scarf model. First,

we develop a simple approximation of the optimal echelon inventory stocking levels

that is easy to compute for serial multi-echelon systems with more than two echelons.

This approximation provides a good lower bound to the optimal stocking levels for the

system. This approximation applies to any convex inventory holding and penalty cost

functions, while in the linear function case, the lower bound is same as that of Shang

and Song (2001). Second, we show that Clark and Scarf’s results can be carried over

to the time-correlated demand process using MMFE, but with greater computation

complexity compared to its counterpart in the independent demand environment.

Finally, we extend the approximation in the independent demand case to the time-

correlated demand process and show that the approximation again provides a lower

bound to the optimal stocking levels. The approximation is used in particular for the

AR(1) case to explore the role of the cross-time demand correlation in affecting the

performance of the system.

The remainder of this paper is organized as follows: in section 2 we introduce

the serial inventory system; we then present the approximation of optimal inventory

levels under an i.i.d. demand process in section 3; in section 4 we establish Clark and

Scarf’s results under time-correlated demand processes, present the approximation,

and illustrate the impacts of the echelon leadtimes and demand auto-correlation on

the system performance; in section 5 we show through numerical examples that the

approximations in sections 3 and 4 are very accurate and are useful analytical tools of

unveiling the fundamental relationship between system parameters and the inventory

4

cost; we conclude our findings in section 6.

2 The Serial Multi-echelon System

Consider an M-echelon serial periodic-review inventory system where customer de-

mand arises from echelon 1 and orders are placed by echelon 1 to echelon 2, by echelon

2 to echelon 3, ..., and echelon M is the highest echelon that orders from an external

supplier with infinite supply. The event sequence is as follows: (a) at the beginning

of a period, shipments due in this period arrive, followed by demand occurrence; (b)

demand is satisfied from on-hand inventory with complete backlogging; (c) at the end

of the period, inventory holding and shortage costs are charged against stocks and/or

shortages followed by ordering and shipment decisions. The shipment leadtime from

echelon m + 1 to echelon m is Tm, i.e., an order placed by echelon m at the end of

period t will arrive at echelon m at the beginning of period t+Tm+1. Such an event

sequence, different from that of Clark and Scarf but the same as that of Federgruen

and Zipkin (1984) and Lee and Whang (1999), makes the exposition easier. The

results of this paper hold for other sequences of events with appropriate adjustments.

Assume that there are no ordering setup costs at all sites. The following

notation will be used:

ξ : demand in a period, assumed to be i.i.d. in section 3;

um : on-hand inventory level at site m at the end of a period, just before ordering

decisions are made;

5

wm : vector of amounts due to arrive from site m+1 to site m in future periods (the

dimension of the vector equals Tm, and the ith element, wim, is the amount due

to arrive in i periods);

w0m : vector wm without the first element;

em : column vector of ones with dimension of Tm;

xm : echelon inventory position at levelm at the end of a period, just before ordering

decisions are made; xm = vm +wmem;

vm : echelon inventory at levelm at the end of a period, just before ordering decisions

are made, i.e., vm = xm−1 + um;

ym : echelon inventory position at level m at the end of a period, after ordering

decisions are made;

cm : unit shipment or processing cost from site m+ 1 to site m;

Lm (vm) : holding and shortage cost in a period at level m, given that the echelon

inventory at the end of a period is vm;

α : discount factor per period;

v : column vector (vm)Mm=1;

x : column vector (xm)Mm=1;

y : column vector (ym)Mm=1;

6

W : a collection of column vectors given by (wm)Mm=1;

W0 : a collection of column vectors given by (w0m)

Mm=1;

(W0, z) : a collection of column vectors given by³¡

w0mzm

¢´Mm=1

, where z =(zm)Mm=1, is

any vector of dimension M .

Let C (v,W) be the system minimum expected discounted cost, given initial

echelon inventory levels v and shipmentsW. The overall inventory control problem

can be stated as:

C (v,W) (1)

= miny:xm≤ym≤vm+1

nXM

m=1[cm (ym − xm) + Lm (vm)] + αE

£C¡v +w1 − ξ, (W0,y− x)¢¤o ,

where vM+1 =∞.

Clark and Scarf show that the optimal inventory policy for the overall system

follows an echelon base-stock policy at each echelon. Let S∗m be the optimal order-

up-to level at echelon m. Hence, at the end of each period, echelon m places an

order to its upstream echelon to bring its echelon inventory position up to S∗m. The

actual amount that gets shipped depends on the on-hand inventory at the upstream

site. Clark and Scarf’s results state that we can decompose (1) into a series of single-

echelon inventory control problems with the proper penalty cost function Γm added

to the original holding and shortage cost expression, with S∗m being the respective

solution for echelon m. We state this result explicitly as follows:

C (v,W) =XM

m=1fm (vm,wm) , (2)

7

where

fm (vm,wm)

= miny:xm≤y

©cm (y − xm) + Lm (vm) + Γm (vm) + αE

£fm¡vm +w

1m − ξ, (w0

m, y − xm)¢¤ª

.

Define γ (j), a random variable, as the sum of demands in j periods, where γ (0) = 0.

Hence γ (1) = ξ. Since the decision made at the end of period t will not affect the

system cost until period t + Tm + 1, we consolidate the state space and rewrite the

function fm (vm,wm) :

fm (vm,wm) (3)

= Lm (vm) + Γm (vm) + αE£Lm¡vm +w

1m − ξ

¢+ Γm

¡vm +w

1m − ξ

¢¤+α2E

£Lm¡vm +w

1m +w

2m − γ (2)

¢+ Γm

¡vm +w

1m +w

2m − γ (2)

¢¤+...+ αTmE [Lm (vm +wmem − γ (Tm)) + Γm (vm +wmem − γ (Tm))]

+gm (xm) ,

where

gm (xm) (4)

= miny:xm≤y

{cm (y − xm) + αTm+1E [Lm (y − γ (Tm + 1)) + Γm (y − γ (Tm + 1))]

+αE [gm (y − ξ)]}.

We interpret the function fm (vm,wm) as the adjusted (through the penalty function

Γm (vm)) discounted expected cost for echelon m, given the current state of vm and

wm. For m = 1, the penalty function is defined as Γ1 (v1) = 0 for all values of v1. For

8

m ≥ 1, the penalty function is defined as:

Γm+1 (vm+1) (5)

=

cm (vm+1 − S∗m) + αTm+1E [Lm (vm+1 − γ (Tm + 1)) + Γm (vm+1 − γ (Tm + 1))]

−αTm+1E [Lm (S∗m − γ (Tm + 1)) + Γm (S∗m − γ (Tm + 1))]

+αE [gm (vm+1 − ξ)− gm (S∗m − ξ)] if vm+1 ≤ S∗m;

0 otherwise.

For computation purpose, the cost function Lm (·) has to be explicitly specified.

We follow the linear cost specification used by Federgruen and Zipkin (1984) and Lee

and Whang (1999). Let

Hm = per-period unit holding cost for inventory at sitem, or in transit to sitem−1.

π = per-period unit backorder cost at site 1.

hm = Hm −Hm+1, m = 1, ...,M , where HM+1 = 0. hm is referred to as the echelon

inventory holding cost at level m.

We will specify the holding and shortage costs associated with echelons as:

Lm (vm) = hmvm, m > 1; (6)

L1 (v1) = h1v1 + (π +H1) v−1 ; (7)

where

x− = (−x)+ and x+ = max (0, x) .

9

Lee and Whang (1999) show that such an echelon cost definition fully allocates the

total system cost to all sites, i.e.,

H1v+1 + πv−1 +

XM

m=2Hm (um +wm−1em−1) =

XM

m=1Lm (vm) .

3 An Approximation of the IID Demand Model

Federgruen and Zipkin develop a closed-form solution for a two-echelon system when

inventory holding and shortage cost is linear and the end demand follows a Normal

distribution. However, generalizing their results to serial multi-echelon systems with

more than two echelons involves calculating multi-variate Normal distribution func-

tions and is computationally challenging. Since the difficulty of calculating optimal

stocking level S∗m is caused by the complexity of the penalty function Γm (·), we pro-

ceed to develop an approximation for Γm (·), denoted by bΓm (·). Define: bΓ1 (·) = 0;and, for m ≥ 1,

bΓm+1 (vm+1) (8)

= cm³vm+1 − bSm´+ αTm+1E

hLm (vm+1 − γ (Tm + 1)) + bΓm (vm+1 − γ (Tm + 1))

i−αTm+1E

hLm³bSm − γ (Tm + 1)

´+ bΓm ³bSm − γ (Tm + 1)

´i+αE

hbgm (vm+1 − ξ)− bgm ³bSm − ξ´i,

10

where bSm is the solution ofbgm (xm) (9)

= miny:xm≤y

{cm (y − xm) + αTm+1EhLm (y − γ (Tm + 1)) + bΓm (y − γ (Tm + 1))

i+αE [bgm (y − ξ)]}.

Hence, we have the following approximation for fm (vm,wm):

bfm (vm,wm) (10)

= Lm (vm) + bΓm (vm) + αEhLm¡vm +w

1m − ξ

¢+ bΓm ¡vm +w1

m − ξ¢i

+α2EhLm¡vm +w

1m +w

2m − γ (2)

¢+ bΓm ¡vm +w1

m +w2m − γ (2)

¢i+...+ αTmE

hLm (vm +wmem − γ (Tm)) + bΓm (vm +wmem − γ (Tm))

i+bgm (xm) .

Functions bΓm, bgm, and bfm together constitute an approximate formulation of the

inventory control of the multi-echelon system. Although the approximation is still

inductive, i.e., for echelonm+1, the definition of bΓm+1 requires the approximate lowerechelon target stocking level, bSm, whose calculation involves bΓm, the computation isgreatly simplified. Using this approximate penalty function, echelonm+1 is penalized

even when it can fulfill the order from echelon m, i.e., echelon m+ 1 is penalized for

overstocking! Theorem 1 states an intuitive result: the stocking level bSm derived fromthe approximate inventory control system is a lower bound to the optimal stocking

level S∗m for each echelon m, m = 1, ...,M . The proofs of Theorem 1 and others are

given in the Appendix.

11

Theorem 1 Let bSm be the solution of (9), then bSm ≤ S∗m, m = 1, ...,M .It turns out that bSm can be computed easily for the specific cost functions

Lm (·) of (6) and (7), and an approximate cost function can be derived in closed-

form.

Define:

τ (i, j) = (Ti + 1) + (Ti+1 + 1) + ...+ (Tj + 1) for i ≤ j, and 0 otherwise.

Φ (·; t), Φ (·; t) : the cumulative distribution function (CDF) and the complementary

CDF of demand in t periods, respectively.

Θ (x) = xΦ (x)+φ (x), where Φ (·) and φ (·) are the CDF and the probability density

function (PDF) of the standard Normal distribution, respectively.

µ (t) ,σ (t) : mean and standard deviation of demand in t periods.

Theorem 2 For m = 1, 2, ...,M , bSm satisfies:Φ̄³bSm; τ (1,m)´ = Km,

where

Km =³Xm

i=1α−τ(1,i)

£(1− α) ci + αTi+1hi

¤´/ (π +H1) .

Theorem 2 thus gives us a very simple, non-inductive way to calculate bSm.Let bAα

m be the cost per period at echelon m by following the order-up-to bSm policy,calculated with the approximate induced penalty cost function bΓm and zero initialinventory. Let bAα be the corresponding system cost per period, i.e., bAα =

PMm=1

bAαm.

12

Lemma 1

bAα1 = (1− α) c1 bS1 + αc1µ+ αT1+1h1

hbS1 − µ (T1 + 1)i+αT1+1 (π +H1)

Z ∞

t=bS1³t− bS1´ dΦ (t;T1 + 1) ;

for m > 1,

bAαm = (1− α) cm bSm + αcmµ+ αTm+1hm

hbSm − µ (Tm + 1)i+αTm+1E

hbΓm ³bSm − γ (Tm + 1)´i.

Lemma 2

bΓm (y) =³y − bSm−1´Xm−1

i=1ατ(i+1,m−1) £(1− α) ci + αTi+1hi

¤−ατ(1,m−1) (π +H1)

Z y

t=bSm−1 Φ (t; τ (1,m− 1)) dt.

Furthermore, if the end demand per period is Normal with mean µ and variance σ2,

then

bΓm (y) =³y − bSm−1´Xm−1

i=1ατ(i+1,m−1) £(1− α) ci + αTi+1hi

¤+ατ(1,m−1)σ (τ (1,m− 1)) (π +H1)

hΘ (ι (y))−Θ

³ι³bSm−1´´i ,

where

ι (y) = − [y − µ (τ (1,m− 1))] /σ (τ (1,m− 1)) .

Suppose the end demand per period is Normal with mean µ and variance σ2.

Let bAm be the approximate average cost per period, i.e., α→ 1, for echelon m when

the cost function are given by (8), (9), and (10) and when the stocking levels bS0ms are13

followed. Let bA be the corresponding average cost per period for the whole system,i.e., bA =PM

m=1bAm.

Lemma 3 With Normal demands,

bA1 = c1µ+ h1Φ−1 (1−K1)σ (T1 + 1) + (π +H1)σ (T1 + 1)Θ¡−Φ−1 (1−K1)

¢;

for m > 1,

bAm = cmµ+ hmµ (τ (1,m− 1))

+σ (τ (1,m))£(H1 −Hm+1)Φ−1 (1−Km) + (π +H1)Θ

¡−Φ−1 (1−Km)¢¤

−σ (τ (1,m− 1)) £(H1 −Hm)Φ−1 (1−Km−1) + (π +H1)Θ¡−Φ−1 (1−Km−1)

¢¤.

The telescoping structure of seriesn bAmoM

m=1leads to:

Theorem 3 With Normal demands,

bA = µXM

m=1cm +

XM

m=1hmµ (τ (1,m− 1)) + σ (τ (1,M)) (π +H1)φ

¡Φ−1 (1−KM)

¢= µ

XM

m=1cm +

XM−1m=1

Hm+1µ (Tm + 1) + σ (τ (1,M)) (π +H1)φ¡Φ−1 (1−KM)

¢.

Corollary 1 ∂ bA/∂Tm > 0 and ∂ bA/∂Tm ≥ ∂ bA/∂Tm+1.Theorem 3 provides a closed-form expression of the approximate system cost.

The system cost can be viewed as consisting of three parts: (1) the average ship-

ping/processing cost; (2) the average inventory holding cost; and (3) the safety

stock cost caused by the demand randomness within the total system-wide leadtime

τ (1,M) . Corollary 1 illustrates the role that leadtimes play in affecting the system

14

cost. In general, longer leadtime leads to higher cost. Since holding inventory is more

expensive at the downstream sites, reducing lower-echelon leadtimes would have a

bigger impact to the system cost than reducing the higher-echelon leadtimes.

We are now ready to examine the true cost performance of following the

stocking levels of bSm in the multi-echelon system. Let Aαm denote the echelon-m

cost per period of following the order-up-to bSm policy, calculated with the true

penalty cost Γm. Note that Aαm = (1− α) cm bSm+αcmµ+αTm+1hm

hbSm − µ (Tm + 1)i+ αTm+1E

hΓm³bSm − γ (Tm + 1)

´i. Denote Aα as the corresponding true system-

wide per-period cost for policy bSm. For a two-echelon system, we were able to derivethe cost dominance relationship between bAα and Aα:

Corollary 2 For a two-echelon system:(1) bAα2 ≥ Aα

2 ; (2) bAα ≥ Aα.

For the two-echelon case with α→ 1, i.e., the average cost case, we see that bAprovides an upper bound for the true cost of implementing the approximate policy bSm.This is helpful, since the average cost bA is easily computable, as shown in Theorem3. In section 5, we will show through numerical examples that both bSm and bAα are

very accurate approximations of the optimal stocking levels and the optimal system

cost.

4 The Time-correlated Demand Model

When demands are correlated across time periods, the optimal control of the inventory

system will be based on the demand forecasts which are revised from period to period

15

as information accumulated over time. Let Dt be the end-customer demand in period

t. Let Dt,t+i be the forecast made in period t for demand in period t + i. Hence

we can define a vector of demand forecasts made in period t for future periods,

Dt = (Dt,t,Dt,t+1, ...), with Dt,t = Dt as the demand realized in period t. Then

²t,t+i = Dt−1,t+i −Dt,t+i,

represents the forecast update made for period t + i from period t − 1 to t, i =

0, 1, ..., and ²t = (²t,t+i)∞i=0 represents the forecast update vector in period t. We

adopt the MMFE developed by Heath and Jackson (1994), which treats forecasts

as the conditional expectation of the future demands on the current information set

and assumes that the demand random variables form a martingale relative to the

corresponding information set. We make the following assumptions explicitly:

(1) ²t vectors are i.i.d. multi-variate Normal random vectors with mean 0;

(2) there are only finite number of random variables in ²t that are linearly inde-

pendent, i.e., there exists a set of finite number of random variables in ²t such

that each ²t,t+i can be written as a linear combination of the random variables

in the set.

Assumption (1), essential to MMFE, requires the demand process to be sta-

tionary and forecasts to be unbiased. Assumption (2) implies that only finite number

of uncertainty factors of the information set affect the forecasts.

MMFE is indeed a powerful model of forecast process in that it captures both

historical demands and potential information available through other sources in one

16

framework (see Aviv, 2001a, for an exploration of MMFE). In fact, an ARMA (auto-

regressive and moving average) process with minimum mean-square error forecast is

a special case of MMFE. To illustrate, we take a look at an AR(1) demand process

Dt = d+ ρDt−1 + εt,

with d > 0, −1 < ρ < 1, and εt being i.i.d. Normal with mean zero and variance

σ2 (σ ¿ d). Here ρ is the auto-correlation coefficient of demand in two consecutive

periods. In this case, Dt,t+i = dPi−1

j=0 ρj + ρiDt, ²t,t+i = ρiεt, and clearly ²t,t+i is

linearly dependent on εt.

Theorem 4 For a time-correlated demand process that satisfies assumptions (1) and

(2), a state-dependent base-stock policy {S∗ (D)} is optimal for a single-echelon in-

ventory system, where the state D is the demand forecast made before making the

ordering decision.

Let C (v,W,D0) be the minimum expected discounted cost for the serial in-

ventory system, given the forecastD0 made in the current period, the initial inventory

levels v and shipmentsW. The overall inventory control problem can be stated as

C (v,W,D0) (11)

= miny:xm≤ym≤vm+1

{XM

m=1[cm (ym − xm) + Lm (vm)]

+αED0

£C¡v+w1 −D1, (W0,y− x) ,D1

¢¤},where vM+1 = ∞. The expectation in (11) is taken over the next period’s forecast

D1, but conditional on the demand forecast D0 in the current period. In the rest

17

of this section, we will drop D0 in the conditional expectation for ease of exposition

whenever it is unambiguous.

Define D (i, j) as a random variable representing demand in j periods that

starts from period i, i.e., the cumulative demand in time interval [i, i+ j − 1]. Simi-

lar to the i.i.d. demand case, we define the adjusted discounted expected cost func-

tions, fm (vm,wm,D0) and gm (xm,D0), and the corresponding penalty cost function

Γm (vm,D0) for echelon m, given that the current period forecast is D0.

fm (vm,wm,D0) (12)

= miny:xm≤y

{cm (y − xm) + Lm (vm) + Γm (vm,D0)

+αE£fm¡vm +w

1m −D1, (w0

m, y − xm) ,D1

¢¤}= Lm (vm) + Γm (vm, D0)

+αE£Lm¡vm +w

1m −D1

¢+ Γm

¡vm +w

1m −D1, D1

¢¤+ ...

+αTmE [Lm (xm −D (1, Tm)) + Γm (xm −D (1, Tm) , DTm)] + gm (xm,D0) ,

gm (xm,D0) (13)

= miny:xm≤y

{cm (y − xm)

+αTm+1E [Lm (y −D (1, Tm + 1)) + Γm (y −D (1, Tm + 1) ,DTm+1)]

+αE [gm (y −D1,D1)]}.

18

For m = 1, Γ1 (·, ·) = 0; for m ≥ 1,

Γm+1 (vm+1, D0) (14)

=

cm [vm+1 − S∗m (D0)]

+αTm+1E [Lm (vm+1 −D (1, Tm + 1)) + Γm (vm+1 −D (1, Tm + 1) ,DTm+1)]

−αTm+1E [Lm (S∗m (D0)−D (1, Tm + 1)) + Γm (S∗m (D0)−D (1, Tm + 1) ,DTm+1)]

+αE [gm (vm+1 −D1,D1)− gm (S∗m (D0)−D1,D1)] if vm+1 ≤ S∗m (D0) ;

0 otherwise;

where S∗m (D0) is the solution of (13).

Under the assumptions (1) and (2), the dynamic programming formulation of

the inventory problem differs from the i.i.d. demand case in that more dimensions

are added to the state space, namely the forecasts for future periods, and that the

conditional expectation is used. The following theorem shows that the key result of

Clark and Scarf for i.i.d. demand system, i.e., the property of echelon decomposition

for the optimal policy using the induced echelon penalty cost, remains true in the

time-correlated demand setting.

Theorem 5 C (v,W,D0) =PM

m=1 fm (vm,wm,D0) .

19

We again define the approximate penalty cost function bΓm, similar to theprevious section. Define: bΓ1 (·, D0) = 0; and, for m ≥ 1,

bΓm+1 (vm+1, D0) (15)

= cmhvm+1 − bSm (D0)

i+αTm+1E

hLm (vm+1 −D (1, Tm + 1)) + bΓm (vm+1 −D (1, Tm + 1) ,DTm+1)i

−αTm+1EhLm³bSm (D0)−D (1, Tm + 1)

´+ bΓm ³bSm (D0)−D (1, Tm + 1) , DTm+1

´i+αE

hbgm (vm+1 −D1,D1)− bgm ³bSm (D0)−D1,D1

´i,

where bSm (D0) is the solution to:

bgm (xm,D0) (16)

= minxm≤y

{cm (y − xm) + αTm+1EhLm (y −D (1, Tm + 1)) + bΓm (y −D (1, Tm + 1) ,DTm+1)i

+αE [bgm (y −D1,D1)]}.

Also,

bfm (vm,wm,D0)

= Lm (vm) + bΓm (vm,D0) + αEhLm¡vm +w

1m −D1

¢+ bΓm ¡vm +w1

m −D1, D1¢i

+α2EhLm¡vm +w

1m +w

2m −D (1, 2)

¢+ bΓm ¡vm +w1

m +w2m −D (1, 2) , D2

¢i+...+ αTmE

hLm (vm +wmem −D (1, Tm)) + bΓm (vm +wmem −D (1, Tm) ,DTm)i

+bgm (xm,D0) .

When demand is time-correlated, it is possible that the physical inventory level

at a site is higher than the target stocking level, such that the target inventory level

20

may not always be realizable. Indeed, it arises in inventory problems in other settings

as well. For example, when demands are Normal, stock levels can exceed the target

even when demands are i.i.d., since it is theoretically possible that a negative demand

can occur. Another example is the classic paper by Eppen and Schrage (1981) where

the optimal allocation of stocks among multiple sites may not be feasible due to stock

imbalance. The traditional approach to this problem is to assume that the probability

of such occurrence is negligible. In fact, when demand follows an AR(1) process with

ρ > 0, the probability of excess inventory (overshoot) is smaller than the standard

i.i.d. demand case (see Lemma 1 in Lee et al.,1999, Aviv, 2001b) and the myopic

policy is quite an accurate approximation to the optimal policy. Another approach

is to make the “costless return” assumption (see Lee et al., 1997). Specifically, at

echelon m < M , if the physical echelon inventory level is higher than the target

echelon inventory level, then the excessive inventory will not be charged as echelon m

inventory, and echelon m gets a refund at the original purchasing price which is paid

by an outside source (for example, a bank); however, the excessive inventory does not

leave the system physically and is still available as inventory for the upper echelon,

m + 1, for future replenishment to echelon m; at echelon M , excessive inventory is

returned to the outside supplier with a refund at the original purchasing cost. Under

this assumption the myopic policy is indeed optimal. We will use such an assumption

in the remainder of the analysis.

Again the approximate inventory levels, which can be computed easily, are

lower bounds of the optimal stocking levels.

21

Theorem 6 bSm (D) ≤ S∗m (D), m = 1, ...,M .We now study the AR(1) demand process. Since in any given period t, the

forecasts of future periods are functions only of demand realization Dt, we need only

Dt to represent forecasts made in period t. Given the current period’s realized demand

D0, let Φ0 (·; t) and Φ0 (·; t) be the CDF and the complementary CDF of demand in

the subsequent t periods whose mean and standard deviation are given by µ0 (t) and

σ (t) respectively.

Theorem 7 For m = 1, 2, ...,M , bSm (D0) satisfies:Φ̄0³bSm (D0) ; τ (1,m)´ = Km,

where

Km =³Xm

i=1α−τ(1,i)

£(1− α) ci + αTi+1hi

¤´/ (π +H1) .

Hence bSm (D0) = µ0 (τ (1,m)) + Φ−1 (1−Km)σ (τ (1,m)) .

Let bAαm (D0) be the cost per period at echelon m by following the order-up-

to bSm (D0) policy and using bΓm as the penalty cost function. Let bAα (D0) be the

corresponding system-wide cost per period, i.e., bAα (D0) =PM

m=1bAαm (D0).

Lemma 4

bAα1 (D0) = (1− α) c1 bS1 (D0) + αc1µ0 (1) + αT1+1h1

hbS1 (D0)− µ0 (T1 + 1)i+αT1+1 (π +H1)

Z ∞

t=bS1(D0)ht− bS1 (D0)i dΦ0 (t;T1 + 1) ;

22

for m > 1,

bAαm (D0) = (1− α) cm bSm (D0) + αcmµ0 (1) + αTm+1hm

hbSm (D0)− µ0 (Tm + 1)i+αTm+1E

hbΓm ³bSm (D0)−D (1, Tm + 1) ,DTm+1´i .Lemma 5

bΓm (y,D0) =³y − bSm−1 (D0)´Xm−1

i=1ατ(i+1,m−1) £(1− α) ci + αTi+1hi

¤+ατ(1,m−1) (π +H1)σ (τ (1,m− 1))

hΘ (ι (y,D0))−Θ

³ι³bSm−1 (D0) , D0´´i ,

where

ι (x,D0) = − [x− µ0 (τ (1,m− 1))] /σ (τ (1,m− 1)) .

As the discount factor α→ 1, the average system-wide cost per period bA (D0)has the following closed-form expression:

Theorem 8

bA (D0) = µ0 (1)XM

m=1cm +

XM

m=1hm [µ0 (τ (1,m))− µ0 (Tm + 1)]

+σ (τ (1,M))φ¡Φ−1 (1−KM)

¢= µ0 (1)

XM

m=1cm +

XM

m=1hmE

£µTm+1 (τ (1,m− 1))

¤+σ (τ (1,M)) (π +H1)φ

¡Φ−1 (1−KM)

¢.

where µTm+1 (τ (1,m− 1)) represents the mean demand in the subsequent τ (1,m− 1)

periods starting from period Tm + 1.

Corollary 3 bA (D0) is nondecreasing in T1, T2, ..., TM and ∂ bA (D0) /∂Tm ≥ ∂ bA (D0) /∂Tm+1.23

The above lemmas and theorems share similar structures as those in the i.i.d.

and Normal demand case. Indeed, when ρ is set to zero, Theorem 8 is the same as

Theorem 3.

We now further assume that demands across periods are positively correlated,

i.e., ρ > 0. Erkip et al. (1990) and Lee et al. (1999) both report that positively

correlated demands were commonly observed in industry.

Theorem 9 If ρ > 0, then:

(1) bSm (D0) is nondecreasing in ρ and T1, T2, ..., Tm;

(2) bA (D0) is nondecreasing in ρ and T1, T2, ..., TM ;

(3) ∂ bSm (D0) /∂ρ is nondecreasing in T1, T2, ..., Tm and ∂ bA (D0) /∂ρ is nondecreasingin T1, T2, ..., TM .

Hence, as expected, both the approximate stocking levels and the approximate

average cost are higher with higher auto-correlated demands, or longer leadtimes,

ceteris paribus. Moreover, the sensitivity of stocking levels and cost to the auto-

correlation of demand is greater when leadtimes are long. This seems to imply that

the interaction effect of demand auto-correlation and leadtime is significant.

Similarly, let Aαm (D0) be the actual echelon-m cost using bSm (D0) as the target

stocking level, i.e., Γm is used to calculate the penalty cost. Then:

Aαm (D0) = (1− α) cm bSm (D0) + αcmµ0 (1) + αTm+1hm

hbSm (D0)− µ0 (Tm + 1)i+αTm+1E

hΓm³bSm (D0)−D (1, Tm + 1) ,DTm+1´i .

24

The corresponding system cost is Aα (D0) =PM

m=1Aαm (D0). For a two-echelon sys-

tem, we have the following cost dominance relationship.

Corollary 4 For a two-echelon system: (1) bAα2 (D0) ≥ Aα

2 (D0); (2) bAα (D0) ≥

Aα (D0).

5 Numerical Examples

In this section, we first present examples to show that the approximations developed

in sections 3 and 4 are accurate for both the i.i.d. and AR(1) demand processes. We

then illustrate how demand auto-correlation of the AR(1) process affects the decisions

and performance of the system.

In the i.i.d. demand scenario, the base case has the following parameters: the

demand per period is Normal with mean µ = 100 and standard deviation σ = 20;

replenishment leadtimes, T1 = 2, T2 = 1; transportation or processing costs, c1 = 10,

c2 = 5; inventory holding costs, H1 = 4, H2 = 2; site 1 shortage cost, π = 10; per-

period discount factor, α = 0.9. The true optimal stocking levels and system cost are

calculated by the method given by Federgruen and Zipkin (1984) for a two-echelon

system.

Figure 1 shows the approximate and optimal inventory stocking levels under

varying σ/µ. We observe, as stated in Theorem 1, that the approximate stocking

level, bS2, is a lower bound of the optimal stocking level, S∗2 . Moreover, the differencebetween the two levels is very small (within 3% of the optimal stocking level S∗2).

25

Figure 2 shows the comparison of three system costs: (1) bAα is the system cost using

the approximate optimal stocking levels bS, calculated with the approximate penaltycost function bΓ; (2) Aα is the system cost using bS, but calculated with the penaltycost function Γ; (3) A

αis the system cost using optimal stocking levels S∗, calculated

with the true penalty cost Γ. As shown in Corollary 2, we observe that bAα ≥ Aα.

Moreover, the three cost curves are very close (within 1.01% of the optimal system

cost Aα). Hence, Figure 2 shows that both bAα and Aα are very good approximations

of Aα. Comparisons based on other parameter ranges, such as c2/c1, H1/c1, H2/c2,

π/c1, and T2/T1 (a total of 61 examples, see Table 1), all show the similar accuracy

of the approximation.

Figures 1 and 2 about here.

To verify the accuracy of this approximation for systems with more than two

echelons, we also run comparison for a three-echelon system. The three-echelon sys-

tem parameters are set as: Normal distribution with mean µ = 100 and standard

deviation σ = 20; replenishment leadtimes, T1 = 2, T2 = 1, T3 = 1; transportation or

processing costs, c1 = 10, c2 = 5, c3 = 2; inventory holding costs, H1 = 4, H2 = 2,

H3 = 1; site 1 shortage cost π = 10, per-period discount factor, α = 1. Figures 3 and 4

show that the gap between our approximation and the optimal result is very small. In

fact, comparisons based on other parameter ranges (a total of 59 examples, see Table

2) also shows similar accuracy (gaps are within 2.55% of the optimal stocking level

S∗3 and 0.36% of the optimal system cost A, respectively). Hence, this approximation

26

is reasonably accurate for a three-echelon inventory system as well.

Figures 3 and 4 about here.

For the AR(1) demand process, the base case is the same as that of the i.i.d.

case, with α = 1 and the AR(1) process specified as d = 100, ρ = 0.1, D0 = 10,

and σ = 20. We extended the method of Federgruen and Zipkin (1984) for the i.i.d.

case to the AR(1) case to calculate the optimal stocking levels and system costs

(see Dong, 1999, for details). Figure 5 shows the approximate and optimal stocking

levels under varying σ/d. Again, the approximate stocking level bS2 (D0) is a lowerbound of optimal S∗2 (D0), and their difference is very small (within 2.20% of A (D0)).

Similarly, Figure 6 confirms that bA (D0) ≥ A (D0). The cost curves are very close toone another, with gaps that are within 1.12%. A total of 71 examples (see Table 3)

based on different parameter changes also show similar degree of accuracy. Hence,

for both the i.i.d. and AR(1) demand processes, the approximation not only provides

an effective and simple computation for the target stocking levels, also generates

an accurate representation of the actual system cost by following the approximate

stocking levels that is close to optimal.

Figures 5 and 6 about here.

We further explore the effectiveness of the simple closed-form expression of

the approximate model. Figures 7 and 8 show the optimal inventory levels and

system costs under varying auto-correlation coefficient ρ and replenishment leadtime

T1. As expected, both the optimal inventory levels and system cost increase as ρ or T1

27

increases. In addition, the cost increments due to leadtime increase, is greater when

ρ is higher. Leadtime and demand auto-correlation are thus shown to have significant

interaction effects. Leadtime reduction has greater impact on system cost when the

demand auto-correlation is high. Such observations match the results of Theorem

9. Hence, the analytical results and the associated qualitative insights that we draw

from the approximate model are shown to be valid under the true cost model, based

on this set of examples.

Figures 7 and 8 about here.

6 Conclusion

In this paper, we revisit the Clark and Scarf’s model and develop a simple approxima-

tion of the induced penalty cost function. This approximation leads to a lower bound

on the optimal stocking levels and an upper bound on the average system cost. The

approximation is then extended to the time-correlated demand process with MMFE,

where Clark and Scarf’s decomposition result is shown to hold. In particular, under

the AR(1) process the approximation provides a simple, easy to compute closed-form

expression for the stocking levels and the average system cost. The closed-form ex-

pression from the approximation allows us to investigate how the underlying demand

process affects the performance of the inventory system. As noted earlier, an AR(1)

model may be a better representation of the underlying demand process in many

high-tech and consumer goods industries, and it is indeed used in recent supply chain

28

management research studies. Our study of the AR(1) demand process shows that

both the system cost and target inventory stocking levels increase as the demand

auto-correlation coefficient increases. Hence, in a highly time-correlated demand en-

vironment, simplifying the demand assumption to i.i.d. would understate the actual

system cost and result in target inventory levels that are too low. We have also shown

that the impact of leadtime reduction is greater when the auto-correlation coefficient

ρ, is higher. Hence it is more worthwhile to invest in leadtime reduction in a highly

time-correlated demand environment.

The approximation developed in this paper offers a benchmark for further

study of the forecast/information sharing in supply chains. An important avenue

of future research is on fine tuned MMFE, which offers great flexibility in modeling

information and collaboration in multi-echelon systems.

Acknowledgments: The authors gratefully acknowledge the helpful comments

of the editor and two anonymous referees. The authors also thank Paul Zipkin, Jing-

Sheng Song, Yossi Aviv for insightful discussions. Useful feedback was provided by

participants at the 2001 Multi-echelon Inventory Systems Conference at Berkeley.

Appendix

Proof of Theorem 1. First observe that, in (4), dE [gm (y − ξ)] /dy = −cm.

Hence, the first derivative of the minimand on the R.H.S. of (4) w.r.t. y is given by

(1− α)cm + αTm+1 (d/dy)E [Lm (y − γ (Tm + 1)) + Γm (y − γ (Tm + 1))] . (A.1)

29

Similarly, the first derivative of the minimand on the R.H.S. of (9) w.r.t. y is given by

(1− α)cm + αTm+1 (d/dy)EhLm (y − γ (Tm + 1)) + bΓm (y − γ (Tm + 1))

i. (A.2)

To show bSm ≤ S∗m, it suffices to show that:(d/dy)E

hbΓm (y − Z)i ≥ (d/dy)E [Γm (y − Z)]for any random variable Z. We show this inequality by induction. When m = 1, bΓ1 = Γ1,

and so equality holds. Assume that the inequality holds up to m and bSm ≤ S∗m. Let F (·)be the CDF of Z. First, note that, when bSm ≤ S∗m < y−z, bΓm+1 (y − z)−Γm+1 (y − z) =bΓm+1 (y − z) is nondecreasing in y. Second, for S∗m ≥ y − z, from (5), we have:

(d/dy)Γm+1 (y) = (1− α) cm+αTm+1 (d/dy)E [Lm (y − γ (Tm + 1)) + Γm (y − γ (Tm + 1))]

and from (8), we have

(d/dy) bΓm+1 (y) = (1− α) cm+αTm+1 (d/dy)E

hLm (y − γ (Tm + 1)) + bΓm (y − γ (Tm + 1))

i.

Hence, we have:

(d/dy)EhbΓm+1 (y − Z)− Γm+1 (y − Z)

i=

Zz≤y−S∗m

(d/dy)hbΓm+1 (y − z)− Γm+1 (y − z)

idF (z)

+

Zz>y−S∗m

(d/dy)hbΓm+1 (y − z)− Γm+1 (y − z)

idF (z)

=

Zz≤y−S∗m

hdbΓm+1 (y − z) /dyi dF (z)

+

Zz>y−S∗m

αTm+1 (d/dy)EhbΓm (y − z − γ (Tm + 1))− Γm (y − z − γ (Tm + 1))

idF (z)

≥ 0.

30

Assume that Eh(d/dy)

³bΓm+1 (y − Z)− Γm+1 (y − Z)´i< ∞. The interchange of dif-

ferentiation and integration is justified by the Lebeague Dominance Convergence Theorem.

Proof of Lemma 1. Let bAαm (xm) be the cost per period at echelon m by

following the order-up-to bSm policy with initial inventory of xm. Clearly bAαm (xm) =

(1− α) bgm (xm). Define egm (·) as:egm (xm) = bgm (xm) + cmxm

= (1− α) cm bSm + αcmµ+ αTm+1EhLm³bSm − γ (Tm + 1)

´+ bΓm ³bSm − γ (Tm + 1)

´i+αE

hegm ³bSm − ξ´i.

And

bAαm (xm) = (1− α) [egm (xm)− cmxm] .

Note that egm (xm) is actually independent of xm, hencebAαm = bAα

m (0) = (1− α)egm (0)= (1− α) cmbSm + αcmµ+ αTm+1E

hLm³bSm − γ (Tm + 1)

´+ bΓm ³bSm − γ (Tm + 1)

´i.

By definition of Lm (·), the desired result follows.

Lemma A.1 For all m > 1 and for a random variable Z, we have:

(d/dy)EhbΓm (y − Z)i

= (1− α)Xm−1

i=1ατ(i+1,m−1)ci +

Xm−1i=1

ατ(i,m−1) ddyE [Li (y − Z − γ (τ (i,m− 1)))] .

31

Proof of Lemma A.1. The proof is by induction. We refer the reader to details

in Dong (1999).

Proof of Theorem 2. For m = 1, the desired result is obtained by observing

that, setting (A.2) to zero yields

(1− α) c1 + αT1+1 (d/dy)E [L1 (y − γ (T1 + 1))]

= (1− α) c1 + αT1+1£h1 − (π +H1) Φ̄ (y;T1 + 1)

¤= 0.

For m ≥ 2, using Lemma A.1, (A.2) becomes

(1− α) cm + αTm+1hm + αTm+1 (d/dy)EhbΓm (y − γ (Tm + 1))

i= (1− α) cm + αTm+1hm + αTm+1{(1− α)

m−1Xi=1

ατ(i+1,m−1)ci

+m−1Xi=1

ατ(i,m−1) ddyE [Li (y − γ (Tm + 1)− γ (τ (i,m− 1)))]}

= (1− α) cm + αTm+1hm + (1− α)Xm−1

i=1ατ(i+1,m)ci

+Xm−1

i=1ατ(i,m) d

dyE [Li (y − γ (τ (i,m)))]

= (1− α)Xm

i=1ατ(i+1,m)ci + αTm+1hm +

Xm−1i=2

ατ(i,m)hi

+ατ(1,m)£h1 − (π +H1) Φ̄ (y; τ (1,m))

¤= (1− α)

Xm

i=1ατ(i+1,m)ci +

Xm

i=1ατ(i,m)hi − ατ(1,m) (π +H1) Φ̄ (y; τ (1,m))

Setting (A.2) to zero immediately yields the desired result.

Proof of Lemma 2. Using Lemma A.1 by setting Z to be zero and integrating

32

both sides from bSm−1 to y gives:bΓm (y)− bΓm ³bSm−1´

=

Z y

t=bSm−1{(1− α)m−1Xi=1

ατ(i+1,m−1)ci +m−1Xi=1

ατ(i,m−1) ddtE [Li (t− γ (τ (i,m− 1)))]}dt

=³y − bSm−1´ (1− α)

Xm−1i=1

ατ(i+1,m−1)ci

+Xm−1

i=1ατ(i,m−1)

Z y

t=bSm−1 hidt− ατ(1,m−1) (π +H1)Z y

t=bSm−1 Φ̄ (t; τ (1,m− 1)) dt=

³y − bSm−1´Xm−1

i=1ατ(i+1,m−1) £(1− α) ci + αTi+1hi

¤−ατ(1,m−1) (π +H1)

Z y

t=bSm−1 Φ̄ (t; τ (1,m− 1)) dt.

The result follows from noting that bΓm ³bSm−1´ = 0.For Normal demand, note that Θ0 (x) = Φ (x), so that

Z y

t=bSm−1 Φ̄ (t; τ (1,m− 1)) dt=

Z y

t=bSm−1 Φµ− (t− µ (τ (1,m− 1)))

σ (τ (1,m− 1))¶dt

= −σ (τ (1,m− 1))hΘ (ι (y))−Θ

³ι³bSm−1´´i .

Here we state two lemmas that are useful for calculations involving Normal demands.

The proofs of these two lemmas are quite tedious and are given in Dong (1999).

Lemma A.2

(1/Σ)

Z ∞

t=−∞(− (a− t) /b)Φ (− (a− t) /b)φ ((t− µ) /Σ) dt

=Σ2

b√Σ2 + b2

φ

µµ− a√Σ2 + b2

¶+µ− ab

Φ

µµ− a√Σ2 + b2

¶.

33

Lemma A.3

(1/Σ)

Z ∞

t=−∞φ (− (a− t) /b)φ ((t− µ) /Σ) dt

=³b/√b2 + Σ2

´³1/√2π´exp

Ã− (a− µ)22 (b2 + Σ2)

!.

Lemma A.4

EhbΓm ³bSm − γ (Tm + 1)

´i=

³bSm − µ (Tm + 1)− bSm−1´Xm−1i=1

ατ(i+1,m−1) £(1− α) ci + αTi+1hi¤

−ατ(1,m−1) (π +H1)E

"Z bSm−γ(Tm+1)t=bSm−1 Φ (t; τ (1,m− 1)) dt

#.

If the end demand per period is Normal with mean µ and variance σ2, then

EhbΓm ³bSm − γ (Tm + 1)

´i=

(m−1Xi=1

ατ(i+1,m−1) £(1− α) ci + αTi+1hi¤)ש

Φ−1 (1−Km)σ (τ (1,m))− Φ−1 (1−Km−1)σ (τ (1,m− 1))ª

+ατ(1,m−1) (π +H1)×©Θ¡−Φ−1 (1−Km)

¢σ (τ (1,m))−Θ

¡−Φ−1 (1−Km−1)¢σ (τ (1,m− 1))ª .

Proof of Lemma A.4. The first part is a consequence of Lemma 2. For Normal

demand, we can write

bSm = µ (τ (1,m)) + Φ−1 (1−Km)σ (τ (1,m)) ,

34

for m = 1, ...,M . Hence

E

"Z bSm−γ(Tm+1)t=bSm−1 Φ (t; τ (1,m− 1)) dt

#= −σ (τ (1,m− 1))×

{Z ∞

t=−∞Θ³−hbSm − t− µ (τ (1,m− 1))i /σ (τ (1,m− 1))´ dΦ (t;Tm + 1)

−Θ ¡−Φ−1 (1−Km−1)¢}.

Let a = bSm − µ (τ (1,m− 1)), b = σ (τ (1,m− 1)), eµ = µ (Tm + 1), Σ = σ (Tm + 1),

then Z ∞

t=−∞Θ³−³bSm − t− µ (τ (1,m− 1))´ /σ (τ (1,m− 1))´ dΦ (t;Tm + 1)

= (1/Σ)

Z ∞

t=−∞Θ (− (a− t) /b)φ ((t− eµ) /Σ) dt

= (1/Σ)

Z ∞

t=−∞(− (a− t) /b)Φ (− (a− t) /b)φ ((t− eµ) /Σ) dt

+(1/Σ)

Z ∞

t=−∞φ (− (a− t) /b)φ ((t− eµ) /Σ) dt

=Σ2

b√Σ2 + b2

φ

µ eµ− a√Σ2 + b2

¶+eµ− ab

Φ

µ eµ− a√Σ2 + b2

¶+³b/√b2 + Σ2

´³1/√2π´exp

(−1/2(a− eµ)2

b2 + Σ2

)(by Lemmas A.2 & A.3)

= σ2 (Tm + 1) / (σ (τ (1,m− 1))σ (τ (1,m)))φ¡−Φ−1 (1−Km)

¢− (σ (τ (1,m)) /σ (τ (1,m− 1)))Φ−1 (1−Km)Φ

¡−Φ−1 (1−Km)¢

+σ (τ (1,m− 1)) /σ (τ (1,m))φ ¡Φ−1 (1−Km)¢

= (σ (τ (1,m)) /σ (τ (1,m− 1)))Θ ¡−Φ−1 (1−Km)¢.

The desired result follows.

Proof of Lemma 3. The result follows from taking the limit of bAαm as α→ 1.

See Dong (1999) for the algebraic details.

35

Proof of Theorem 3.n bAmoM

m=2is a telescope series. Using Lemma 3, we get

bA = µXM

m=1cm +

XM

m=1µ (τ (1,m− 1))

+σ (τ (1,M))£H1Φ

−1 (1−KM) + (π +H1)Θ¡−Φ−1 (1−KM)

¢¤.

Note that

H1Φ−1 (1−KM) + (π +H1)Θ

¡−Φ−1 (1−KM)¢

= H1Φ−1 (1−KM)

+ (π +H1) {φ¡−Φ−1 (1−KM)

¢− Φ−1 (1−KM)Φ¡−Φ−1 (1−KM)

¢}= (π +H1)φ

¡−Φ−1 (1−KM)¢,

since Φ (−Φ−1 (1−KM)) = KM = H1/ (π +H1) as α → 1, and the desired result

follows. The alternative expression can be obtained by using hm = Hm−Hm+1,HM+1 = 0.

Proof of Corollary 1. ∂ bA/∂Tm = Hm+1+ 12σ(τ(1,M))

(π +H1)φ (Φ−1 (1−KM)) ≥

0 form = 1, ...,M−1; ∂ bA/∂TM = 12σ(τ(1,M))

(π +H1)φ (Φ−1 (1−KM)) ≥ 0. ∂ bA/∂Tm ≥

∂ bA/∂Tm+1, since H1 ≥ H2 ≥ ... ≥ HM .Proof of Corollary 2. (1)

bAα2 −Aα

2 = αT2+1EhbΓ2 ³bS2 − γ (T2 + 1)

´− Γ2

³bS2 − γ (T2 + 1)´i≥ 0,

since bΓ2 (·) ≥ Γ2 (·) on <. (2) follows for the fact that bAα1 = A

α1 .

For the AR(1) case, we will make use of the property that

Dn = dXn−1

i=0ρi + ρnD0 +

Xn

i=1ρn−iεi,

36

µn = dXn−1

i=0ρi + ρnD0, (A.3)

σ2n = σ2¡1− ρ2n

¢/¡1− ρ2

¢,

D (1, n) = d/ (1− ρ)Xn

i=1

¡1− ρi

¢+D0

Xn

i=1ρi +

Xn

i=1

Xi

j=1ρi−jεj

= dXn

i=1

Xi−1j=0

ρj +D0Xn

i=1ρi + 1/ (1− ρ)

Xn

i=1

¡1− ρi

¢εi,

µ0 (n) = dXn

i=1

Xi−1j=0

ρj +D0Xn

i=1ρi,

σ20 (n) = σ2Xn

i=1

³Xi−1j=0

ρj´2.

Notice that the variances of Dn and D (1, n) are independent of historical demand D0,

hence we can drop subscript 0 from the variance notation.

Proof of Theorem 4. We first study the finite horizon case. With zero leadtime,

the dynamic programming formulation of the problem is

fn (xn,Dn) = minxn≤y

{c (y − xn) + Ln (xn) + αE [fn+1 (y −Dn+1,Dn+1)]} ,

where Ln (x) is convex in x for all n and fN+1 (·, ·) = 0. Note that the expectation is taken

over a finite number of multi-variate random variables, no anomaly should be encountered.

Let V ∗ be the set of functions v (·, ·) that are convex w.r.t. the first variable;

π∗ = ∆∗×∆∗× ...×∆∗ (∆∗ appears N times), where ∆∗ is the set of base-stock policies,

and for δ ∈ ∆∗, δ (x,D) = S∗ (D) if x < S∗ (D), δ (x,D) = x otherwise.

The following three steps are sufficient to show that a base-stock policy is optimal

for period n and for every n, and the optimal base-stock level is a function of demand in

period n.

37

• It can be easily shown that, if fn+1 ∈ V ∗, then E [fn+1 (y −Dn+1,Dn+1)] is convex in

y.

• If fn+1 ∈ V ∗, then there exists δ ∈ ∆∗ that is optimal for period n. To see this, suppose

fn+1 ∈ V ∗. Now, we have cyn + Ln (xn) + αE [fn+1 (yn −Dn+1,Dn+1)] being convex

in yn. Let S∗n be a minimizer of cyn + Ln (xn) + αE [fn+1 (yn −Dn+1,Dn+1)], which is

clearly a function of Dn. Let δ (xn,Dn) = S∗n (Dn) if xn < S∗n (Dn), δ (xn,Dn) = xn

otherwise. Hence, following δ (xn,Dn) is optimal for period n.

• If fn+1 ∈ V ∗, then fn ∈ V ∗. This follows from:

fn (xn,Dn)

=

c (S∗n (Dn)− xn) + Ln (xn) + αE [fn+1 (S

∗n (Dn)−Dn+1,Dn+1)] if xn < S∗n (Dn) ;

Ln (xn) + αE [fn+1 (xn −Dn+1,Dn+1)] otherwise,

and that fn (xn,Dn) is convex and has continuous derivative with respect to xn.

The nonzero leadtime case can be extended as in standard inventory literature.

Finally, we can extend the result to the infinite horizon case using Proposition 1.6 and

Proposition 1.7 in Section 3.1 of Bertsekas (1995), such that the limit of S∗n(D) converges

to an optimal stationary policy S∗(D) as n→∞.

Proof of Theorem 6. Similar to that of Theorem 1. See Dong (1999) for

details.

Lemma A.5 For all m > 1 and for a random variable D (1, T ), we have:

(d/dy)EhbΓm (y −D (1, T ) ,DT )i

= (1− α)Xm−1

i=1ατ(i+1,m−1)ci +

Xm−1i=1

ατ(i,m−1) ddyE [Li (y −D (1, T + τ (i,m− 1)))] .

38

Proof of Lemma A.5. The proof is similar to that of Lemma A.1. See Dong

(1999) for details.

Proof of Lemmas 4 and 5 .

Similar to that of Lemmas 1 and 2. See Dong (1999) for details.

Lemma A.6

EhbΓm ³bSm (D0)−D (1, Tm + 1) , DTm+1´i

=nXm−1

i=1ατ(i+1,m−1) £(1− α) ci + αTi+1hi

¤oשΦ−1 (1−Km)σ (τ (1,m))− Φ−1 (1−Km−1)σ (τ (1,m− 1))

ª+ατ(1,m−1) (π +H1)שσ (τ (1,m))Θ

¡−Φ−1 (1−Km)¢− σ (τ (1,m− 1))Θ ¡−Φ−1 (1−Km−1)

¢ª.

Proof of Lemma A.6. The proof is similar to that of Lemma A.4. See Dong

(1999) for details.

Proof of Theorem 8. The first equality is similar to the proof of Theorem 2.

For the second equality, we note that

µ0 (τ (1,m))− µ0 (Tm + 1) (A.4)

= dXτ(1,m)

i=1

Xi−1j=0

ρj +D0Xτ(1,m)

i=1ρi − d

XTm+1

i=1

Xi−1j=0

ρj −D0XTm+1

i=1ρi

= dXτ(1,m)

i=Tm+1+1

Xi−1j=0

ρj +D0Xτ(1,m)

i=Tm+1+1ρi,

39

and

E£µTm+1 (τ (1,m− 1))

¤(A.5)

= E

·dXτ(1,m−1)

i=1

Xi−1j=0

ρj +DTm+1Xτ(1,m−1)

i=1ρi¸

= dXτ(1,m−1)

i=1

Xi−1j=0

ρj +³dXTm

j=0ρj + ρTm+1D0

´Xτ(1,m−1)i=1

ρi

= dXτ(1,m−1)

i=1

³Xi−1j=0

ρj +XTm

j=0ρj+i

´+D0

Xτ(1,m−1)i=1

ρi+Tm+1

= dXτ(1,m−1)

i=1

Xi+Tm

j=0ρj +D0

Xτ(1,m)

i=Tm+1+1ρi

= dXτ(1,m)

i=Tm+1+1

Xi−1j=0

ρj +D0Xτ(1,m)

i=Tm+1+1ρi.

Proof of Corollary 3. Since σ (τ (1,M)) is nondecreasing in T1, ..., TM and

∂σ (τ (1,M)) /∂Tm = ∂σ (τ (1,M)) /∂Tm+1, we only need to check the termPM

m=1 hm×

E[µTm+1 (τ (1,m− 1))]. Note that

XM

m=1hmE

£µTm+1 (τ (1,m− 1))

¤=

XM

m=1hmE

·dXτ(1,m−1)

i=1

Xi−1j=0

ρj +DTm+1Xτ(1,m−1)

i=1ρi¸

which is nondecreasing in T1, ..., TM . To show ∂ bA (D0) /∂Tm ≥ ∂ bA (D0) /∂Tm+1, weexamine:

∂³XM

m=1hmE

£µTm+1 (τ (1,m− 1))

¤´/∂Ti

= ∂hiE£µTi+1 (τ (1, i− 1))

¤/∂Ti + ∂

³XM

m=i+1hmE

£µTm+1 (τ (1,m− 1))

¤´/∂Ti

40

It follows that

∂³XM

m=1hmE

£µTm+1 (τ (1,m− 1))

¤´/∂Ti

−∂³XM

m=1hmE

£µTm+1 (τ (1,m− 1))

¤´/∂Ti+1

= ∂¡hiE

£µTi+1 (τ (1, i− 1))

¤¢/∂Ti + ∂

³XM

m=i+1hmE

£µTm+1 (τ (1,m− 1))

¤´/∂Ti

−∂ ¡hi+1E £µTi+1+1 (τ (1, i))¤¢ /∂Ti+1−∂

³XM

m=i+2hmE

£µTm+1 (τ (1,m− 1))

¤´/∂Ti+1

= ∂¡hiE

£µTi+1 (τ (1, i− 1))

¤¢/∂Ti + ∂

¡hi+1E

£µTi+1+1 (τ (1, i))

¤¢/∂Ti

−∂ ¡hi+1E £µTi+1+1 (τ (1, i))¤¢ /∂Ti+1.Since

∂E£µTi+1+1 (τ (1, i))

¤/∂Ti − ∂E

£µTi+1+1 (τ (1, i))

¤/∂Ti+1

= ∂ [µ0 (τ (1, i+ 1))− µ0 (Ti+1 + 1)] /∂Ti − ∂ [µ0 (τ (1, i+ 1))− µ0 (Ti+1 + 1)] /∂Ti+1

= ∂µ0 (τ (1, i+ 1)) /∂Ti − ∂µ0 (τ (1, i+ 1)) /∂Ti+1 − µ0 (Ti+1 + 1) /∂Ti

+∂µ0 (Ti+1 + 1) /∂Ti+1

and

∂µ0 (τ (1, i+ 1)) /∂Ti = ∂µ0 (τ (1, i+ 1)) /∂Ti+1,

µ0 (Ti+1 + 1) /∂Ti = 0,

∂µ0 (Ti+1 + 1) /∂Ti+1 ≥ 0,

41

it suffices to show that ∂E£µTi+1 (τ (1, i− 1))

¤/∂Ti ≥ 0, which is true because

E£µTi+1 (τ (1, i− 1))

¤= d

Xτ(1,i−1)k=1

Xk−1j=0

ρj +E [DTi+1]Xτ(1,i−1)

j=1ρj.

Lemma A.7 (1) E£µTm+1 (τ (1,m− 1))

¤is nondecreasing in ρ;

(2) ∂E£µTm+1 (τ (1,m− 1))

¤/∂ρ is nondecreasing in T1, T2, ..., Tm.

Proof of Lemma A.7. (1) Taking first derivative w.r.t. ρ:

∂E£µTm+1 (τ (1,m− 1))

¤/∂ρ

= dXτ(1,m)

i=1

Xi−1j=0jρj−1 + ∂E [DTm+1] /∂ρ

Xτ(1,m−1)i=1

ρi +E [DTm+1]Xτ(1,m−1)

i=1iρi−1

≥ 0.

where ∂E [DTm+1] /∂ρ ≥ 0 follows from (A.3). (2) It is now obvious that

∂E£µTm+1 (τ (1,m− 1))

¤/∂ρ is nondecreasing in T1, T2, ..., Tm.

Lemma A.8 (1) σ (τ (1,M)) is nondecreasing in ρ;

(2) ∂σ (τ (1,M)) /∂ρ is nondecreasing in T1, T2, ..., TM .

Proof of Lemma A.8. (1) In general, for any n,

∂σ (n) /∂ρ

= ∂

rσ2Xn

i=1

³Xi−1j=0

ρj´2/∂ρ

= σnXn

i=1

³Xi−1j=0

ρj´³Xi−1

j=0jρj−1

´o/

rXn

i=1

³Xi−1j=0

ρj´2≥ 0.

42

(2) We need to show

nXn+1

i=1

³Xi−1j=0

ρj´³Xi−1

j=0jρj−1

´o/

rXn+1

i=1

³Xi−1j=0

ρj´2

≥nXn

i=1

³Xi−1j=0

ρj´³Xi−1

j=0jρj−1

´o/

rXn

i=1

³Xi−1j=0

ρj´2.

Let ai =³Pi−1

j=0 ρj´2, bi =

³Pi−1j=0 ρ

j´³Pi−1

j=0 jρj−1´, we will show that

nXn+1

i=1bio/

rXn+1

i=1ai ≥

nXn

i=1bio/

rXn

i=1ai.

It is equivalent to show that

³Xn

i=1bi + bn+1

´2Xn

i=1ai ≥

³Xn

i=1bi´2 ³Xn

i=1ai + an+1

´or

2bn+1Xn

i=1biXn

i=1ai + b

2n+1

Xn

i=1ai ≥

³Xn

i=1bi´2an+1

or

2bn+1Xn

i=1ai ≥

³Xn

i=1bi´an+1

or Xn

i=1(2bn+1ai − bian+1) ≥ 0.

To show that 2bn+1ai − bian+1 ≥ 0 for i = 1, ..n, note that

2bn+1ai − bian+1

=³Xi−1

j=0ρj´³Xn

j=0ρj´×n

2³Xn

j=0jρj−1

´³Xi−1k=0

ρk´−³Xi−1

k=0kρk−1

´³Xn

j=0ρj´o

≥ 0,

43

since

2³Xn

j=0jρj−1

´³Xi−1k=0

ρk´−³Xi−1

k=0kρk−1

´³Xn

j=0ρj´

=Xn

j=0

Xi−1k=0(2j − k) ρj+k−1

=Xi−1

j=0

Xi−1k=0(2j − k) ρj+k−1 +

Xn

j=i

Xi−1k=0(2j − k) ρj+k−1

≥Xi−1

j=0

Xi−1k=0(j − k) ρj+k−1

=Xi−1

j=0

Xj−1k=0(j − k) ρj+k−1 +

Xi−1j=0

Xk=j(j − k) ρj+k−1

+Xi−1

j=0

Xi−1k=j+1

(j − k) ρj+k−1

=Xi−1

j=0

Xj−1k=0(j − k) ρj+k−1 +

Xi−1j=0

Xi−1k=j+1

(j − k) ρj+k−1

=Xi−1

j=0

Xj−1k=0(j − k) ρj+k−1 +

Xi−1k=0

Xk−1j=0(j − k) ρj+k−1

= 0.

Proof of Theorem 9. Straightforward by Lemmas A.7 and A.8.

References

Aviv, Y. 2001a. The Effect of Collaborative Forecasting on Supply Chain Perfor-

mance. Management Science. Vol. 47. No. 10. 1-18.

Aviv, Y. 2001b. Gaining Benefits from Joint Forecasting and Replenishment processes:

The Case of Auto-correlated Demand. Working Paper. Washington University.

Axäter, S. and K. Rosling. 1993. Notes: Installation vs. Echelon Stock Policies

for Multilevel Inventory Control. Management Science. Vol. 39. No. 10.

1274-1280.

44

Bertsekas, D. P. 1995. Dynamic Programming and Optimal Control. Vol. 2. Athena

Scientific. Belmont. Massachusetts. 134-149.

Chen, F. 2000. Optimal Policies for Multi-echelon Inventory Problems with Batch

Ordering. Operations Research. Vol. 48. No. 3. 376-389.

Chen, F. and J.-S. Song. 2001. Optimal Policies for Multi-Echelon Inventory Prob-

lems with Markov-Modulated Demand. Operations Research. Vol. 49. No.2.

226-234.

Chen, F. and Y.-S. Zheng. 1994. Lower Bounds for Multi-echelon Stochastic Inven-

tory Systems. Management Science. Vol. 40. No. 11. 1426-1443.

Clark, A. and H. Scarf. 1960. Optimal Policies for a Multi-echelon Inventory Prob-

lem. Management Science. 6. 475-490.

Dong, L.X. 1999. Contributions to Inventory Management in Distribution Channels,

Doctoral Dissertation, Department of Industrial Engineering and Engineering

Management, Stanford University.

Eppen, G. and L. Schrage. 1981. Centralized Ordering Policies in a Multi-warehouse

SystemWith Lead Times and Random Demand. in L. B. Schwarz (Ed.). Multi-

Level Production/Inventory Control Systems: Theory and Practice. TIMS Stud-

ies in the Management Sciences. Vol. 16. North-Holland. Amsterdam.

Erkip, N., W. H. Hausman, and S. Nahmias. 1990. Optimal Centralized Ordering

45

Policies in Multi-Echelon Inventory Systems with Correlated Demands. Man-

agement Science. Vol. 36. No. 3. 381-392.

Federgruen A. and P. Zipkin. 1984. Computational Issues in an Infinite-Horizon,

Multi-echelon Inventory Model. Operations Research. Vol. 32. No. 4. 818-836.

Fotopoulos, S. and M. C. Wang. 1988. Safety Stock Determination with Corre-

lated Demands and Arbitrary Lead Times. European Journal of Operational

Research. No. 35. 172-181.

Gallego, G. and P. Zipkin. 1999. Stock Positioning and Performance Estimation

in Serial Production-Transportation Systems. Manufacturing & Service Opera-

tions Management. Vol. 1. No. 1. 77-88.

Graves, S.C., H.C. Meal, S. Dasu. Y. Qiu. 1986. Two-stage production Planning in

a Dynamic Environment. S. Axäter, C. Schneeweiss, E. Silver, eds. Multi-Stage

production Planning and Control. Lecture Notes in Economics and Mathemat-

ical Systems, Springer-Verlag, Berlin. 266. 9-43.

Graves, S.C., D.B. Kletter, W.B. Hetzel. 1998. A Dynamic Model for Requirements

Planing with Application to Supply Chain Optimization. Operations Research.

Vol. 46. S35-S49.

Graves, S. C. 1999. A Single-Item Inventory Model for a Nonstationary Demand

Process. Manufacturing & Service Operations Management. Vol. 1. No. 1.

50-61.

46

Güllü, R. 1996. On the Value of Information in Dynamic Production/Inventory

Problems under Forecast Evolution. Naval Research Logistics, Vol. 43. 289-

303.

Heath, D.C. and P.L. Jackson. 1994. Modeling the Evolution of Demand Forecasts

with Application to Safety Stock Analysis in Production/Distribution Systems.

IIE Transaction. Vol. 26. No. 3. 17-30.

Iida, T. 2000. The Infinite Horizon Non-stationary Stochastic Multi-echelon Inven-

tory Problem and Near Myopic Policies. Working Paper. Tokyo Institute of

Technology.

Johnson, G. and H. Thompson. 1975. Optimality of Myopic Inventory Policies for

Certain Dependent Demand Processes. Management Science. Vol. 21. No. 11.

1303-1307.

Kahn, J. “Inventory and the Volatility of Production. American Economic Review.

77. 667-679.

Lee, H.L., V. Padmanabhan, and S. Whang. 1997. Information Distortion in a

Supply Chain: The Bullwhip Effect. Management Science. Vol. 43. No. 4.

546-558.

Lee, H.L., K. C. So, and C. S. Tang. 1999. The Value of Information Sharing in a

Two-Level Supply Chain. Management Science. Vol. 46. No. 5. 626-643.

47

Lee, H.L., and S. Whang. 1998. Value of Postponement. in Product Variety Man-

agement, edited by T. Ho and C. Tang, Kluwer Academic Publishers, 65-84.

Lee, H.L. and S. Whang. 1999. Decentralized Multi-Echelon Inventory Control

Systems: Incentives and Information. Management Science. Vol. 45. No. 5.

633-640.

Lovejoy, W. 1990. Myopic Policies for Some Inventory Models with Uncertain De-

mand Distribution. Management Science. Vol. 36. No. 6. 724-738.

Lovejoy, W. 1992. Stopped Myopic Policies for Some Inventory Models with Uncer-

tain Demand Distributions. Management Science. Vol. 38. No. 5. 688-707.

Miller, B.L. 1986. Scarf’s State Reduction Method, Flexibility, and a Dependent

Demand Inventory Model. Operations Research. Vol. 34. No. 1. 83-90.

Reyman, G. 1989. State Reduction in a Dependent Demand Inventory Model Given

by a Time Series. European Journal of Operational Research. No. 41. 174-180.

Rosling, K. 1989. Optimal Inventory Policies for Assembly Systems Under Random

Demands. Operations Research. Vol. 37. No.11. 565-579.

Scarf, H. 1959. Bayes Solutions of the Statistical Inventory Problem. Annals of

Mathematical Statistics. No. 30. 490-508.

Scarf, H. 1960. Some Remarks on Baye’s Solution to the Inventory Problem. Naval

Research Logistic Quarterly. No .7. 591-596.

48

Schmidt, C. and S. Nahmias. 1985. Optimal Policy for a Two-Stage Assembly

System Under Random Demand. Operations Research. Vol. 33. No. 5. 1130-

1145.

Shang, K.H. and J.-S. Song. 2001. Newsvendor Bounds and Heuristic for Optimal

Policies in Serial Supply Chains. Working Paper. University of California,

Irvine.

Toktay, L.B. and L.M. Wein. 2001. Analysis of a Forecasting-Production-Inventory

System with Stationary Demand. Management Science. Vol. 47. No. 9.

1268-1281.

49

498

499

500

501

502

503

504

505

σ /µ

Inve

ntor

y le

vel

S2^ 500.15 500.18 500.21 500.24 500.27 500.3 500.33 500.37 500.4 500.43 500.46

S2* 501.47 501.76 502.06 502.35 502.65 502.94 503.23 503.53 503.82 504.12 504.41

0.1 0.12 0.14 0.16 0.18 0.2 0.22 0.24 0.26 0.28 0.3

Figure 1: Inventory Level Comparison: Optimal vs. Approximation (Normal Distri-

bution)

2350

2400

2450

2500

2550

2600

2650

2700

σ /µ

Syst

em C

ost

A^ 2470.2 2487.05 2503.89 2520.73 2537.57 2554.41 2571.25 2588.09 2604.93 2621.77 2638.61

A 2469.04 2485.65 2502.26 2518.87 2535.48 2552.09 2568.7 2585.31 2601.92 2618.53 2635.13

A - 2468.94 2485.52 2502.11 2518.7 2535.28 2551.87 2568.46 2585.05 2601.63 2618.22 2634.81

0.1 0.12 0.14 0.16 0.18 0.2 0.22 0.24 0.26 0.28 0.3

Figure 2: System Cost Comparison: Optimal vs. Approximation (Normal Distribu-

tion)

50

710

720

730

740

750

760

770

σ/ µ

Inve

ntor

y le

vel

S3^ 714.97 717.97 720.96 723.96 726.95 729.95 732.94 735.94 738.93 741.93 744.92

S3* 721.47 725.77 730.06 734.36 738.65 742.95 747.24 751.54 755.83 760.13 764.42

0.1 0.12 0.14 0.16 0.18 0.2 0.22 0.24 0.26 0.28 0.3

Figure 3: Inventory Level Comparison for A Three-echelon System: Optimal vs.

Approximation (Normal Distribution)

2600

2650

2700

2750

2800

2850

2900

σ/ µ

Syst

em C

ost

A^ 2625.9 2651.1 2676.3 2701.4 2726.6 2751.8 2777 2802.2 2827.3 2852.5 2877.7

A 2618.4 2642.1 2665.8 2689.5 2713.1 2736.8 2760.5 2784.2 2807.9 2831.6 2855.2

A_ 2615 2638 2661 2683.9 2706.9 2729.9 2752.9 2775.9 2798.9 2821.9 2844.9

0.1 0.12 0.14 0.16 0.18 0.2 0.22 0.24 0.26 0.28 0.3

Figure 4: System cost Comparison for A Three-echelon System: Optimal vs. Approx-

imation (Normal Distribution)

51

541

542

543

544

545

546

547

548

549

550

σ/d

Inve

ntor

y le

vel

S2^ 544.487 544.52 544.553 544.586 544.619 544.652 544.685 544.718 544.751 544.785 544.818

S2* 546.042 546.386 546.731 547.075 547.419 547.763 548.108 548.452 548.796 549.14 549.484

0.1 0.12 0.14 0.16 0.18 0.2 0.22 0.24 0.26 0.28 0.3

Figure 5: Inventory Level Comparison: Optimal vs. Approximation (AR(1) Process)

2450

2500

2550

2600

2650

2700

2750

2800

σ/ d

Syst

em C

ost

A^ 2589 2607 2626 2662 2680 2699 2717 2753 2772A 2588 2606 2624 2660 2678 2696 2714 2750 2768A- 2587 2605 2623 2659 2677 2695 2713 2749 2767

0.1 0.12 0.14 0.18 0.2 0.22 0.24 0.28 0.3

Figure 6: System Cost Comparison: Optimal vs. Approximation (AR(1) Process)

52

0

500

1000

1500

2000

2500

3000

ρ

S2*

T1=2 529.4971 576.6876 632.1269 697.6646 775.4779 868.0883 978.3771 1109.6 1265.402 1449.824

T1=3 631.2226 689.7189 759.2915 842.9569 944.6014 1069.154 1222.77 1413.024 1649.114 1942.058

T1=4 732.9189 802.7237 886.4328 988.2616 1113.977 1271.417 1471.125 1727.104 2057.692 2486.562

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

Figure 7: Inventory Level vs. Auto-correlation for Various Leadtimes (AR(1) Process)

2000

3000

4000

5000

6000

7000

ρ

T1=2 2306.189 2404.918 2521.571 2660.098 2825.148 3022.093 3257.068 3537.001 3869.644 4263.598

T1=3 2528.266 2652.329 2800.806 2980.314 3199.336 3468.593 3801.426 4214.202 4726.742 5362.754

T1=4 2748.097 2897.046 3076.753 3296.564 3569.218 3911.968 4347.936 4907.721 5631.267 6569.989

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

A

Figure 8: System Cost vs. Auto-correlation for Various Leadtimes (AR(1) Process)

53

Param eter R a t io bS2 S∗2O p t im a lG ap

bAα Aα AαO p tim a lG ap

σ/µ

0.10.120.140.160.180.200.220.240.260.280.30

500.15500.18500.21500.24500.27500.30500.33500.37500.40500.43500.46

501.47501.76502.06502.35502.65502.94503.23503.53503.82504.12504.41

0.26%0.32%0.37%0.42%0.47%0.52%0.58%0.63%0.68%0.73%0.78%

2470.202487.052503.892520.732537.572554.412571.252588.092604.932621.772638.61

2469.042485.652502.262518.872535.482552.092568.702585.312601.922618.532635.13

2468.942485.522502.112518.702535.282551.872568.462585.052601.632618.222634.81

0.05%0.06%0.07%0.08%0.09%0.10%0.11%0.12%0.13%0.14%0.14%

c2/c1

0.10.20.30.40.50.60.70.80.91

510.28507.38504.51501.66498.82495.97493.10490.21487.27484.28

527.52516.55509.68504.55500.37496.73493.43490.32487.30484.28

3.27%1.77%1.01%0.57%0.31%0.15%0.07%0.02%0.01%0.00%

1629.971892.272153.722414.312674.032932.853190.733447.613703.373957.78

1618.691885.002149.272411.762672.682932.213190.473447.533703.353957.78

1613.741883.062148.532411.502672.602932.193190.473447.533703.353957.78

1.01%0.49%0.24%0.12%0.05%0.02%0.01%0.00%0.00%0.00%

H1/c1

0.10.20.30.40.50.60.70.80.91

522.24513.37506.26500.30495.15490.59486.48482.73479.27476.05

527.31517.26509.42502.94497.39492.52488.16484.21480.57477.20

0.96%0.75%0.62%0.52%0.45%0.39%0.34%0.30%0.27%0.24%

2137.272279.902418.632554.412687.862819.402949.363077.973205.403331.81

2133.942276.962416.022552.092685.802817.582947.743076.543204.143330.69

2133.442276.592415.742551.872685.632817.442947.643076.453204.073330.64

0.18%0.15%0.12%0.10%0.08%0.07%0.06%0.05%0.04%0.04%

H2/c2

0.10.20.30.40.50.60.70.80.91

504.78503.28501.79500.30498.82497.33495.83494.34492.83491.32

515.60510.09506.08502.94500.37498.17496.25494.50492.88491.32

2.10%1.33%0.85%0.52%0.31%0.17%0.08%0.03%0.01%0.00%

2192.982313.872434.352554.412674.032793.192911.843029.933147.353263.87

2184.772308.232430.632552.092672.682792.472911.513029.803147.313263.87

2182.212307.032430.102551.872672.602792.442911.503029.803147.313263.87

0.49%0.30%0.17%0.10%0.05%0.03%0.01%0.00%0.00%0.00%

π/c1

0.51.01.52.02.53.03.54.04.55.0

466.43500.30515.27524.74531.58536.88541.19544.80547.90550.60

467.46502.94518.67528.60535.75541.28545.76549.50552.71555.51

0.22%0.52%0.66%0.73%0.78%0.81%0.84%0.86%0.87%0.88%

2473.912554.412598.552628.672651.332669.392684.332697.032708.062717.78

2473.532552.092594.462623.122644.562661.572675.612687.512697.822706.89

2473.522551.872594.012622.472643.742660.612674.522686.322696.532705.51

0.02%0.10%0.18%0.24%0.29%0.33%0.37%0.40%0.43%0.45%

T2/T1

0.51.01.52.02.53.03.54.04.55.0

500.30599.51698.48797.20895.64993.771091.561188.971285.941382.41

502.94603.70704.02803.90903.321002.271100.721198.621295.951392.65

0.52%0.69%0.79%0.83%0.85%0.85%0.83%0.81%0.77%0.74%

2554.412562.652572.832585.232599.952617.002636.322657.822681.382706.86

2552.092558.842567.902579.552593.852610.752630.142651.872675.782701.71

2551.872558.402567.272578.782593.002609.862629.252651.022674.992700.99

0.10%0.17%0.22%0.25%0.27%0.27%0.27%0.26%0.24%0.22%

Table 1: Stocking Levels and System Costs Comparisons: IID Normal Demand

54

Param eter R a t io bS3 S∗3O p t im a lG ap

bA A AO p t im a lG ap

σ/µ

0.10.120.140.160.180.200.220.240.260.280.30

714.97717.97720.96723.96726.95729.95732.94735.94738.93741.94744.92

721.47725.77730.06734.36738.65742.95747.24751.54755.83760.13764.42

0.90%1.07%1.25%1.42%1.58%1.75%1.91%2.08%2.24%2.39%2.55%

2625.902651.082676.262701.442726.622751.802776.992802.172827.352852.532877.71

2618.412642.102665.782689.462713.142736.832760.512784.192807.872831.562855.24

2614.972637.962660.962683.952706.942729.942752.932775.922798.922821.912844.90

0.13%0.16%0.18%0.21%0.23%0.25%0.28%0.30%0.32%0.34%0.36%

c2/c1

0.30.40.50.60.7

714.97714.97714.97714.97714.97

722.67721.47720.77720.27720.07

1.07%0.90%0.80%0.74%0.71%

2275.902525.902775.903025.903275.90

2267.812518.412769.363020.203270.74

2261.592514.972767.193018.583269.34

0.28%0.14%0.08%0.05%0.04%

H1/c1

0.30.40.50.60.70.80.91.0

719.48714.97711.40708.43705.90703.70701.75700

724.88721.47718.80716.63714.70713.00711.55710.3

0.74%0.90%1.03%1.14%1.23%1.30%1.38%1.45%

2454.632625.902794.302960.523125.033288.153450.113611.10

2448.992618.412785.352950.353113.813276.023437.183597.46

2445.992614.972781.252945.563108.323269.853430.353590.00

0.12%0.13%0.15%0.16%0.18%0.19%0.20%0.21%

H2/c2

0.30.40.50.60.7

714.97714.97714.97714.97714.97

722.67721.47720.77720.27720.07

1.07%0.90%0.80%0.74%0.71%

2475.902625.902775.902925.903075.90

2467.812618.412769.362920.203070.74

2461.592614.972767.192918.583069.34

0.25%0.13%0.08%0.06%0.05%

π/c1

0.51.01.52.02.53.03.54.04.55.0

703.70714.97721.29725.60728.83731.40733.52735.33736.89738.26

709.20721.47728.19732.80736.13738.90741.02742.93744.49745.96

0.78%0.90%0.95%0.98%0.99%1.02%1.01%1.02%1.02%1.03%

2594.072625.902645.092658.652669.052677.452684.462690.462695.702700.33

2589.782618.412635.332647.152656.142663.352669.342674.452678.892682.81

2587.712614.972631.042642.262650.812657.672663.372668.242672.482676.22

0.08%0.13%0.16%0.18%0.20%0.21%0.22%0.23%0.24%0.25%

T2/T1

0.20.40.60.81.01.21.41.61.82.0

654.32694.76735.19775.60816.01856.40896.79937.16977.541017.91

660.52701.16741.89782.40823.01863.70904.29944.86985.441026.00

0.94%0.91%0.90%0.87%0.85%0.85%0.83%0.81%0.80%0.79%

2560.392604.092647.692691.192734.602777.922821.172864.342907.442950.48

2553.332596.762640.032683.162726.182769.092811.902854.632897.292939.87

2550.492593.522636.372679.082721.672764.152806.542848.842891.082933.25

0.11%0.12%0.14%0.15%0.17%0.18%0.19%0.20%0.21%0.23%

T3/T1

0.20.40.60.81.01.21.41.61.82.0

654.32694.76735.19775.60816.01856.40896.79937.17977.541017.90

659.52700.86742.09783.30824.51865.60906.69947.67988.741029.70

0.79%0.87%0.93%0.98%1.03%1.06%1.09%1.11%1.13%1.15%

2620.392624.092627.692631.192634.602637.922641.172644.342647.442650.48

2614.802617.242619.562621.772623.902625.942627.922629.842631.712633.52

2612.032614.022615.882617.642619.312620.902622.432623.902625.322626.69

0.11%0.12%0.14%0.16%0.18%0.19%0.21%0.23%0.24%0.26%

Table 2: Three-echelon Stocking Levels and System Costs Comparisons: IID Normal Demand

55

Param eter R a t io bS2 S∗2O p tim a lG ap

bA A AO p tim a lG ap

σ/µ

0.10.120.140.160.180.200.220.240.260.280.30

544.49544.52544.55544.59544.62544.65544.69544.72544.75544.78544.82

546.04546.39546.73547.07547.42547.76548.11548.45548.80549.14549.48

0.28%0.34%0.40%0.45%0.51%0.57%0.62%0.68%0.74%0.79%0.85%

2588.972607.252625.532643.812662.082680.362698.642716.922735.192753.472771.75

2587.572605.562623.562641.562659.552677.552695.542713.542731.542749.532767.53

2587.432605.402623.372641.32659.302677.272695.242713.212731.172749.142767.11

0.06%0.07%0.08%0.09%0.10%0.12%0.13%0.14%0.15%0.16%0.17%

ρ

00.10.20.30.40.50.60.70.80.9

500.30544.65596.62657.89730.40816.44918.561039.671183.001352.13

502.94547.76600.37662.51736.23823.89928.211052.231199.361373.40

0.52%0.57%0.62%0.70%0.79%0.90%1.04%1.19%1.36%1.55%

1244.291331.301435.791561.851714.321898.842121.842390.672713.553099.65

1241.971328.491432.291557.371708.451890.962111.162376.092693.663072.63

1241.751328.211431.921556.871707.741889.962109.692373.942690.523068.11

0.20%0.23%0.27%0.32%0.39%0.47%0.58%0.70%0.86%1.03%

c2/c1

0.10.20.30.40.50.60.70.80.91

555.50552.35549.23546.13543.03539.94536.82533.67530.48527.23

575.02562.84555.22549.53544.89540.88537.24533.82530.52527.23

3.40%1.86%1.08%0.62%0.34%0.17%0.08%0.03%0.01%0.00%

1694.171975.342255.602534.922813.303090.713367.103642.413916.514189.15

1681.131966.832250.312531.832811.633089.903366.763642.293916.494189.15

1675.341964.512249.402531.512811.533089.873366.763642.293916.494189.15

1.12%0.55%0.28%0.13%0.06%0.03%0.01%0.00%0.00%0.00%

H1/c1

0.10.20.30.40.50.60.70.80.91

568.50558.85551.13544.65539.05534.09529.62525.55521.78518.28

574.36563.38554.83547.76541.71536.40531.64527.33523.36519.68

1.02%0.80%0.67%0.57%0.49%0.43%0.38%0.34%0.30%0.27%

2218.392376.282529.922680.362828.272974.113118.223260.873402.243542.49

2214.442372.752526.772677.552825.762971.873116.233259.093400.663541.09

2213.832372.302526.422677.272825.532971.693116.083258.983400.573541.02

0.21%0.17%0.14%0.12%0.10%0.08%0.07%0.06%0.05%0.04%

H2/c2

0.10.20.30.40.50.60.70.80.91

549.52547.89546.27544.65543.03541.41539.79538.16536.53534.88

561.88555.74551.26547.76544.89542.45540.31538.38536.59534.89

2.20%1.41%0.91%0.57%0.34%0.19%0.10%0.04%0.01%0.00%

2278.772413.082546.952680.362813.302945.743077.643208.933339.503469.11

2269.192406.422542.512677.552811.632944.843077.213208.763339.463469.11

2266.152404.972541.852677.272811.532944.803077.203208.763339.463469.11

0.56%0.34%0.20%0.12%0.06%0.03%0.01%0.01%0.00%0.00%

π/c1

0.51.01.52.02.53.03.54.04.55.0

507.82544.65560.92571.22578.65584.42589.11593.03596.40599.34

509.09547.76564.89575.70583.47589.49594.36598.43601.92604.96

0.25%0.57%0.70%0.78%0.83%0.86%0.88%0.90%0.92%0.93%

2592.912680.362728.312761.022785.642805.252821.482835.282847.262857.81

2592.432677.552723.402754.402777.582795.982811.142824.012835.142844.94

2592.412677.272722.842753.602776.582794.812809.832822.572833.592843.29

0.02%0.12%0.20%0.27%0.33%0.37%0.41%0.45%0.48%0.51%

T2/T1

0.51.01.52.02.53.03.54.04.55.0

544.65654.90764.88874.58983.971093.031201.701309.951417.711524.93

547.76659.78771.31882.32992.821102.791212.201321.021429.181536.64

0.57%0.74%0.83%0.88%0.89%0.89%0.87%0.84%0.80%0.76%

2680.362689.952701.442715.322731.722750.692772.162796.032822.182850.47

2677.552685.412695.622708.642724.592743.402764.962789.122815.692844.50

2677.272684.872694.852707.712723.562742.342763.912788.112814.762843.66

0.12%0.19%0.24%0.28%0.30%0.30%0.30%0.28%0.26%0.24%

Table 3: Stocking Levels and System Costs Comparisons: AR(1) Demand

56