Académique Documents
Professionnel Documents
Culture Documents
3/2012 19
Exact Fill Rates for the (R, S) Inventory Control with Discrete Distributed
Demands for the Backordering Case
The fill rate is usually computed by using the traditional approach, which calculates it as the
complement of the quotient between the expected unfulfilled demand and the expected demand
per replenishment cycle, instead of directly the expected fraction of fulfilled demand. Fur-
thermore the available methods to estimate the fill rate apply only under specific demand
conditions. This paper shows the research gap regarding the estimation procedures to com-
pute the fill rate and suggests: (i) a new exact procedure to compute the traditional approxi-
mation for any discrete demand distribution; and (ii) a new method to compute the fill rate di-
rectly as the fraction of fulfilled demand for any discrete demand distribution. Simulation re-
sults show that the latter methods outperform the traditional approach, which underestimates
the simulated fill rate, over different demand patterns. This paper focuses on the traditional
periodic review, base stock system when backlogged demands are allowed.
Keywords: Inventory, Fill Rate, Periodic Review, Backordering, Discrete Demand
In this sense, [3], [8], [9] and [10] suggest However, expression (1) and expression (2)
methods to estimate it when demand is nor- are not equivalent. Note that if X and Y are
mal distributed whereas [11], [12], [13] when independent random variables it is true that
demand follows any continuous distribution. E X Y E X E Y and, analogously,
When demand is discrete, only [3] suggest a
X 1
method to estimate βTrad for Poisson demands E E X E .
but to the rest of our knowledge no methods Y Y
are available to estimate it when demand fol- 1 1
However E and therefore
lows any discrete distribution function and Y E Y
backlog demands are allowed.
Another approach to compute the fill rate X EX
E (see for example [14]).
consists of directly estimating the fraction of Y E Y
the fulfilled demand per replenishment cycle Thus, applying this reasoning to the defini-
instead of determining the expected shortage, tion of the fill we find that
as follows:
fulfilled demand per replenishment cycle
E (2)
total demand per replenishment cycle
fulfilled demand per replenishment cycle E fulfilled demand per replenishment cycle
E
total demand per replenishment cycle E total demand per replenishment cycle
Since
E fulfilled demand per replenishment cycle E unfulfilled demand per replenishment cycle
1
E total demand per replenishment cycle E total demand per replenishment cycle
then
fulfilled demand per replenishment cycle E unfulfilled demand per replenishment cycle
E 1
total demand per replenishment cycle E total demand per replenishment cycle
[15] proposes methods to estimate both ex- The discussion and summary of this work are
pression (1) and (2) for any discrete demand summarized in Section 5.
pattern when inventory is managed following
the lost sales case principle. However, there 2 Basic Notation and Assumptions
is not available any method to estimate βTrad The traditional periodic review, base stock
and β when demand is modeled by any dis- (R, S) system places replenishment orders
crete distribution and the inventory is man- every R units of time of sufficient magnitude
aged following the backordering case, i.e. to raise the inventory position to the base
when unfulfilled demand is backlogged to stock S. The replenishment order is received
the following cycle. This paper fulfills this L periods after being launched. Figure 1
research gap and suggests two new and exact shows an example of the evolution of the on
methods to estimate both expressions (Sec- hand stock (= stock that is physically on
tion 3). Furthermore, we present and discuss shelf), the net stock (= on hand stock -
some illustrative examples of the perfor- backorders) and the inventory position (= on
mance of both versus a simulated fill rate and hand stock + stock on order – backorders) for
over different demand patterns (Section 4). the backordering case.
Informatica Economică vol. 16, no. 3/2012 21
stock
IPR-L IP2R-L
S
OH0
On hand stock (OH)
OHR Inventory position (IP)
OHR-L
Net stock (NS)
OH2R-L
S
E unfulfilled demand per replenishment cycle
NS0 1
P NS0 D
DR NS0 1
R NS0 P DR (3)
Since the net stock is equivalent to the inven- P NS0 P DL S NS0 f L S NS0 .
tory position minus the on order stock, the
Therefore, βTrad when demand follows any
net stock balance at the beginning of the cy-
discrete distribution function can be estimat-
cle is:
ed with the following expression:
NS0 S DL .
Then,
S
NS0 1
f L S NS 0 D
DR NS0 1
R NS0 f R DR
Trad 1
(4)
D
DR 1
R f R DR
where the denominator represents the ex- vice metric when there is no demand to be
pected total demand per replenishment cycle. served. Therefore, cycles that do not show
Note that expression (4) can be used by any any demand should not be taken into ac-
discrete demand distribution. count. According to [15], in order to derive
an exact method to compute β over different
3.2 Derivation of an Exact Method to demand patterns including intermittent de-
Compute β mand is necessary to include explicitly the
As Section 1 points out, the fill rate is de- condition of having positive demand during
fined as the fraction of demand that is imme- the cycle. Then the fill rate can be expressed
diately fulfilled from shelf. From a practical as
point of view, it is useless to consider a ser-
Hence, positive demand during a cycle can therefore the fill rate will be the fraction of
be: (i) lower or equal than the net stock at the that demand which is satisfied by the on hand
beginning of this cycle, i.e. DR≤NS0, and stock at the beginning of this cycle. There-
therefore the fill rate will be equal to 1; or (ii) fore
greater than the net stock, i.e. DR>NS0, and
NS0
NS0 P DR NS0 DR 0 P DR DR 0 (6)
DR NS0 1 DR
where the first term indicates the case (i) and mass and cumulative distribution functions of
the second term indicates the case (ii). Re- demand, ft(∙) and Ft(∙), respectively, results
writing expression (6) through the probability into
FR NS0 FR 0
NS0 f R DR
NS0 (7)
1 FR 0 DR NS0 1 DR 1 FR 0
Therefore, by applying expression (7) to eve- demand follows any discrete distribution
ry positive net stock level at the beginning of function results as follows:
the cycle, the method to estimate when
Informatica Economică vol. 16, no. 3/2012 23
S
FR NS0 FR 0
NS0 f R DR
f L S NS 0
1 FR 0
DR NS0 1 DR 1 FR 0
(8)
NS0 1
Fig. 2. Demand patterns used in the simulation according to the categorization framework of
demand suggested by [16]
Figure 3 and Figure 4 show the comparison viations that Figure 3 shows arise from esti-
between βTrad and β versus βSim respectively mating the fill rate using the traditional ap-
for the Table 1 cases. In Figure 3, we see that proach (expression (1)) and not from how it
βTrad tends to underestimates the simulated is calculated.
fill rate and therefore the traditional approx-
imation seems to be biased. [9] pointed out
similar results when demand is normally dis-
tributed whereas [15] when demand is Pois-
son distributed for the lost sales case. Note
that expression (4) leads to the exact value of
the traditional approximation. Therefore de-
24 Informatica Economică vol. 16, no. 3/2012
1
0.9
0.8
0.7
Fill Rate
0.6
0.5 Sim
Simulado
0.4 Trad
tradicional
0.3
exacto
0.2
0.1
0
1 2 3 4 5 6 7 8 9 10
Order up to level
Fig. 5. Comparison between βTrad, β and βSim with negative binomial demand with r=4 and
=0.7 (smooth), R=1 and L=1
This paper is part of a wider research project gineering and Management, vol. 4, no.
devoted to identify the most simple and ef- 2, pp. 194-205, 2011.
fective method to find the lowest base stock [3] G. Hadley and T. Whitin, Analysis of In-
that guarantee the achievement of the target ventory Systems. Englewood Cliffs, NJ:
fill rate under any discrete demand context. Prentice-Hall, 1963.
Therefore, further extensions of this work [4] E. A. Silver, "Operations-Research in
should be focused on: (i) assessing the exact Inventory Management - A Review and
method when using other discrete distribu- Critique," Operations Research, vol. 29,
tion functions of demand; (ii) characterizing no. 4, pp. 628-645, 1981.
the cases where the approximation (including [5] S. Chopra and P. Meindl, Supply Chain
some possible new ones) has the most im- Management, 2nd Edition ed Pearson.
portant deviations; (iii) analyzing risks of us- Prentice Hall, 2004.
ing different fill rate approximations to set [6] H. Tempelmeier, "On the stochastic un-
the parameters of the stock policy and finally capacitated dynamic single-item lotsiz-
(iv) exploring the possibility of embedding ing problem with service level con-
results achieved in this work in information straints," European Journal of Opera-
systems with the aim of helping decision tional Research, vol. 181, no. 1, pp. 184-
processes. 194, Aug.2007.
[7] H. L. Lee and C. Billington, "Managing
Acknowledgments Supply Chain Inventory - Pitfalls and
This work is part of a project supported by Opportunities," Sloan Management Re-
the Universitat Politècnica de València, Ref. view, vol. 33, no. 3, pp. 65-73, 1992.
PAID-06-11/2022. [8] E. A. Silver and R. Peterson, Decisions
system for inventory management and
References production planning, 2nd ed. New York:
[1] M. C. van der Heijden, "Near cost- John Wiley & Sons, 1985.
optimal inventory control policies for di- [9] M. E. Johnson, H. L. Lee, T. Davis, and
vergent networks under fill rate con- R. Hall, "Expressions for Item Fill Rates
straints," International Journal of Pro- in Periodic Inventory Systems," Naval
duction Economics, vol. 63, no. 2, pp. Research Logistics, vol. 42, no. 1, pp.
161-179, Jan.2000. 57-80, 1995.
[2] E. Babiloni, M. Cardos, and E. Guijarro, [10] E. A. Silver and D. P. Bischak, "The ex-
"On the exact calculation of the mean act fill rate in a periodic review base
stock level in the base stock periodic re- stock system under normally distributed
view policy," Journal of Industrial En- demand," Omega International Journal
26 Informatica Economică vol. 16, no. 3/2012
of Management Science, vol. 39, no. 3, [14] C. M. Grinstead and J. L. Snell, Intro-
pp. 346-349, June2011. duction to probability, 2nd ed. United
[11] M. J. Sobel, "Fill rates of single-stage States of American: AMS Bookstore,
and multistage supply systems," Manu- 1997.
facturing and Service Operations Man- [15] E. Guijarro, M. Cardós, and E. Babiloni,
agement, vol. 6, no. 1, pp. 41-52, 2004. "On the exact calculation of the fill rate
[12] J. Zhang and J. Zhang, "Fill rate of sin- in a periodic review inventory policy
gle-stage general periodic review inven- under discrete demand patterns," Euro-
tory systems," Operations Research Let- pean Journal of Operational Research,
ters, vol. 35, no. 4, pp. 503-509, Ju- vol. 218, no. 2, pp. 442-447, Apr.2012.
ly2007. [16] A. A. Syntetos, J. E. Boylan, and J. D.
[13] R. H. Teunter, "Note on the fill rate of Croston, "On the categorization of de-
single-stage general periodic review in- mand patterns," Journal of the Opera-
ventory systems," Operations Research tional Research Society, vol. 56, no. 5,
Letters, vol. 37, no. 1, pp. 67-68, pp. 495-503, May2005.
Jan.2009.