Chapter 4: Individual Items with

Probabilistic Demand
¨ Characteristics

¨ The base-stock policy

¨ The (r, Q) or (s,Q) model

¨ Leadtime Variability

¨ The (r, R) or (s,S) model

¨ Periodic Review System – (R,S) model

 1. Characteristics
¨ Demand per unit time is a random variable X with
mean E(X) and standard deviation σ
¨ Possibility of overstocking (excess inventory) or
understocking (shortages)
¨ There are overage costs for overstocking and
shortage costs for understocking
¨ Net stock = (on hand) – (back orders)
¨ Inventory Position = (on hand) + (on order) – (back
orders) – (demand)
 Types of Stochastic Models
¨ Single period models
¨ Fashion goods, perishable goods, goods with short lifecycles,
seasonal goods
¨ One time decision (how much to order).

¨ Multiple period models

 Base Stock: Q = 1
 Continuous time models:

 (s, Q)
 Up-to-level: (s, S)
 Periodic models: Up-to-Level: (R, S)
 Stockout Probability
¨ If L is the order replenishment lead time, D is demand per unit
time, and r is the reorder point (in a continuous review system),
then:
Probability of stockout = 1- P(demand during lead time ≤ r)
¨If demand during lead time is normally distributed with mean
E(D)L, then choosing r = E(D)L =  leads to
Probability of stockout =0.5
 2. The Base Stock Model
The Base-Stock Policy
¨ In the base stock model, inventory is refilled one unit at a time and demand
is random (arrivals).
¨ Start with an initial amount of inventory R. Each time a new demand arrives, place a
replenishment order with the supplier.
¨ An order placed with the supplier is delivered L units of time after it is placed.
¨ Because demand is stochastic, we can have multiple orders (inventory on-order) that
have been placed but not delivered yet.
¨ The amount of demand that arrives during the replenishment leadtime L is called the
Q=1
leadtime demand (inventory on order). R=r+1
¨ Exponential
Amount distribution
of time customer for time
arrive: t
exponential
Number of customers arrive in a period of time: poisson r
t
t1
t2 Assumptions
1. Demands occur one at a time.
2. Any demand not filled from stock is backordered.
3. Replenishment lead times are fixed and known;
4. Times between consecutive orders are stochastic but
independent and identically distributed (i.i.d.)
5. Inventory is reviewed continuously
6. There is no fixed cost associated with placing an order; and
7. There is no constraint on the number of orders that can be
placed per year.

¨ Last two assumptions imply that there is no incentive to

replenish stock in anything other than one-at-a-time fashion.
 Inventory Profile
t
0 1 2 3 4 5 6 7 8 9 10

t
0 1 2 3 4 5 6 7 8 9 10
 Notation
¨ L = replenishment lead time
¨ X = random demand during leadtime (pcs).
¨ G(x) = Prob{X ≤ x}, c.d.f of demand during leadtime
¨ p(x) = Prob{X = x}, p.m.f of demand during leadtime
¨  = E(X) mean demand during leadtime.
¨  = std dev of demand during leadtime.
¨ h = unit holding cost
¨ b = unit backorder cost
¨ r = reorder point (pcs)
¨ R = r +1, base stock level (pcs)
¨ SS = r - , safety stock level (pcs) (r>).
¨ Q = 1, order quantity (fixed at one)
¨ S(R) = service level (average fill rate), fraction of orders filled from stock.
¨ B(R) = average backorder level
¨ I(R) = average on-hand inventory level
¨ True: cumulative (probability of an area), False: density (probability at one point)
 Base Stock Equations
¨ Under Base Stock Policy:
R=Inventory position = on-hand - backorders + orders
¨ At any point in time, number of orders equals number demands during
leadtime(X)  on-hand inventory - backorders = R - X
¨ Expected Service Level:

¨ On-hand and backorders are never positive at the same time, so if X=x,
then:
¨ Expected backorder level:
¨ Expected inventory level: I(R) = R -  + B(R)
Where:  = E(D)L
¨ G(R) (z) R r  B(R)  I(R)
Base Stock Example: Discrete Case
L = one month

 = 10 units (per month)

Assume Poisson demand, so

Note: Poisson
demand is a good
choice when no
variability data
is available.
 Base Stock Example Calculations

p(r)= POISSON.DIST(r, , false)r= 13, =10 p(r)=0.072908

G(r)= POISSON.DIST(r, , true)r= 13, =10 G(r)=0.864464
B(r)= p(r)+(-r)[1-G(r)]=10*0.072908+(10-13)*(1-0.864464)=0.322
 Base Stock Example Results

¨ Service Level: For fill rate of 90%, we must set r = 14, so

R=15 and safety stock ss = r -  = 4. Resulting service is
91.7%.

¨ Backorder Level:
B(r) = 0.187
¨ Inventory Level:

I(R) = R -  + B(R) = 15 - 10 + 0.187 = 5.187

 “Optimal” Base Stock Levels:
Continuous Case
¨ Objective Function:
TC(R) = hI(R) + bB(R) = h(R -  +B(R)) + bB(R)
 TC(R) = h(R - ) + (h+b)B(R)
¨ Solution: if we assume G is continuous, we can use calculus to get
Implication: set base stock
level so fill rate is b/(h+b).
Note: R* increases in b and
decreases in h.
¨ If G is normal(,  ), then

where (z)=b/(h+b)  z = NORMSINV(b/(h+b)})

¨ =z  R* =  + z 
 “Optimal” Base Stock Example
¨ Data: Approximate by normal with mean 10 units/month and
standard deviation  = 3.16 units/month. Set h=$15, b=$25.

¨ Calculations:

since , z =0.32 and hence

R* =  + z = 10 + 0.32(3.16) = 11.01 ~ 11  r = 10
¨ Observation: from previous table fill rate is G(10) = 0.583, so
maybe backorder cost is too low.
 Base Stock Insights
1. Reorder points control the probability of stockouts by
establishing safety stock = r- 
2. The “optimal” fill rate is an increasing function of the backorder
cost (b) and a decreasing function of the holding cost (h). We
can use either a service constraint or a backorder cost to
determine the appropriate base stock level.
bG zR
h keep more stock G zR
3. Base stock levels in multi-stage production systems are very
similar to kanban systems and therefore the above insights apply
to those systems as well. (production card= order, withdrawal
card = demand)
3. The (r, Q) or (s,Q) Model
 Assumptions
¨ Demand occurs continuously over time
¨ Times between consecutive orders are stochastic but independent and identically
distributed (i.i.d.)
¨ Inventory is reviewed continuously
¨ Supply leadtime is a fixed constant L
¨ Orders that cannot be fulfilled immediately from on-hand inventory are
backordered
¨ Fixed cost associated with replenishment orders and cost per backorder.
¨ Constraint on number of replenishment orders per year and service constraint

17
 The (r, Q) Policy
¨ The base-stock policy is the special case of the (r, Q) policy where Q = 1.
¨ Objective:

or lost sale cost

As in EOQ, this makes

batch production attractive.

18
 Notation
¨ D = expected demand per year
¨ L = (replenishment) lead time (assumed constant)
¨ X = (random) demand during lead time
¨  = E[X]= expected demand during lead time
¨  = standard deviation of demand during leadtime
¨ P(x) = Pr(X = x) = p.m.f of demand during leadtime
¨ G(x) = Pr(X < x) = c.d.f of demand during leadtime
¨ A = fixed order cost
¨ v = unit cost of an item
¨ h = annual unit holding cost ( r =h/v: text book)
¨ B1 = cost per stockout (lost sale)
¨ b = annual unit backorder cost= B2v
 Notation (cont.)
¨ Decision Variables:
Q = order quantity
r = reorder point
ss = r -  = safety stock implied by r
¨ Performance Measures:
F(Q) = average order frequency
S(s, Q) = average service level (fill rate)
B(s, Q) = average backorder level
I(s, Q) = average inventory level
Inventory versus Time

N=D/Q = 1/T
 Demand during Leadtime

2
 Demand during Leadtime

r r)

r
 Costs in (r, Q) Model

¨ Fixed Setup Cost: AF(Q)

¨ Stockout Cost: B1D(1-S(r,Q)), where B1 is cost per

stockout (lost sales)

¨ Backorder Cost: bB(r, Q); b = B2v

¨ Inventory Carrying Costs: hI(r, Q)

 Fixed Setup Cost in (r, Q) Model

¨ Observation: since the number of orders per year

is D/Q,
Stockout Cost (lost sale) in (r, Q) Model

¨ Key Observation: inventory position is uniformly distributed

between r+1 and r+Q. So, service in (r,Q) model is
weighted sum of service in base stock model.

¨ Result:
r Q

Note: this form is easier to use

in spreadsheets because it does
not involve a sum.
 Service Level Approximations

¨ Type I (base stock):

Note: computes number
of stockouts per cycle,
underestimates S(Q,r)

¨ Type II:
Note: neglects B(r,Q)
term, underestimates S(Q,r)
 Backorder Costs in (Q,r) Model

¨ Key Observation: B(Q,r) can also be computed by

averaging base stock backorder level function over the
range [r+1,r+Q].

¨ Result:

Notes:
1. B(Q,r)» B(r) is a base stock approximation for backorder level.

2. If we can compute B(x) (base stock backorder level function),

then we can compute stockout and backorder costs in (Q,r) model.
 Inventory Costs in (r, Q) Model
¨ Approximate Analysis: on average inventory declines from Q+ss to
ss+1 so

¨ Exact Analysis: this neglects backorders, which add to average

inventory since on-hand inventory can never go below zero. The
corrected version turns out to be
 3.1. (r,Q) Model with Stockout Cost

¨ Objective Function:

¨ Approximation: Assume we can still use EOQ to compute Q* but

replace S(r, Q) by Type II approximation and B(r, Q) by base stock
approximation:
Results of Approximate Optimization

¨ Assumptions:
¨ Q, r can be treated as continuous variables
¨ G(x) is a continuous c.d.f

¨ Results:

Note: this is just the EOQ formula

Note: another version

of base stock formula
(only z is different)
if G is normal(q,s),
where F(z)=B1D/(B1D+hQ)
 3.2. (r,Q) Model with Backorder Cost

¨ Objective Function:

¨ Approximation: B(r, Q) makes optimization complicated because it

depends on both Q and r. To simplify, approximate with base stock
backorder formula, B(r):
Results of Approximate Optimization
¨ Assumptions:
¨ Q,r can be treated as continuous variables
¨ G(x) is a continuous cdf

¨ Results:

Note: this is just the EOQ formula

Note: this is just the
base stock formula

if G is normal(q,s),
where F(z)=b/(h+b)
 (r, Q) Example
D = 14 units per year
c = $150 per unit
h = 0.1 × 150 = $15 per unit
L = 45 days
 = (14 × 45)/365 = 1.726 units during replenishment lead
time
A = $10
b = $40
=0.635
Demand during lead time is Poisson
 Values for Poisson(θ)
Distribution

p(r)= POISSON.DIST(r, , false)r= 2, =1.726 p(r)=0.26513

G(r)= POISSON.DIST(r, , true)r= 2, =1.726 G(r)=0.750345
B(r)= p(r)+(-r)[1-G(r)]=0.389
Solution
Performance Measures for Example
 Observations on Example

¨ Orders placed at rate of 3.5 per year

¨ Fill rate fairly high (90.4%)
¨ Very few outstanding backorders (0.049 on
average)
¨ Average on-hand inventory just below 3 (2.823)
 Varying the Example
¨ Change: suppose we order twice as often so F=7 per year, then
Q=2 and:

which may be too low, so increase r from 2 to 3:

¨ This is better. For this policy (Q=2, r=3) we can compute

B(2,3)=0.026, I(Q,r)=2.80.

¨ Conclusion: this has higher service and lower inventory than the
original policy (Q=4, r=2). But the cost of achieving this is an
extra 3.5 replenishment orders per year.
3.3. Uncertain Demand with Safety Stock
¨ Notations:
¨ k =safety factor
¨ = standard normal cdf function of the expected shortages per
replenishment cycle (ESPRC)
¨ probability that a unit normal variable takes on a value of k or
larger = 1 - (k).
¨ Safety stock from simple-minded approach:
 Safety Stock Based on Stock out
Cost B1
¨ Total cost

 k?
¨ Step 1: check  if yes go to step 2. Otherwise, continue step
3 with
¨ Step 2: set k at its lowest alloable value (by management)
¨ Step 3:
 Example

¨ B1 = $300
¨ D= 200pcs/year
¨ Q= 129pcs
¨ v= $2/pcs
¨ =50pcs
¨ =21
¨ h/v = 0.24$/$/yr
 Safety Stock Based on Back
Order Cost B2
¨ Total cost:
 k?
¨ Step 1: check  if yes go to step 2. Otherwise, continue
step 3 select k satisfy

¨ Step 2: set k at its lowest alloable value (by management)

¨ Step 3:
 Example
¨ B2 = 0.25
¨ D= 200pcs/year
¨ Q= 85pcs
¨ A=$21.5
¨ v= $6/pcs
¨ h = 0.2v =$1.2
¨ =50pcs
¨ =10 pcs
 Single Product (r, Q) Insights
¨ Basic Insights:
¨ Safety stock provides a buffer against stock-outs.
¨ Cycle stock is an alternative to setups/orders.

¨ Other Insights:
1. Increasing D tends to increase optimal order quantity Q.
2. Increasing  tends to increase the optimal reorder point. (Note: either
increasing D or l increases .)
3. Increasing the variability of the demand process tends to increase the optimal
reorder point (provided z > 0).
4. Increasing the holding cost tends to decrease the optimal order quantity and
reorder point.
 4. Lead Time Variability
¨ Problem: replenishment lead times may be variable, which
increases variability of lead time demand.
¨ Notation:
L = replenishment lead time (days), a random variable
l = E[L] = expected replenishment lead time (days)
L = Var(L)= std dev of replenishment lead time (days)
D = demand per period, a random variable, assumed i.i.d.
d = E(D) = expected demand per period
D = Var(D) =std dev of daily demand (units)
D,d,sigma(D): demand
X, theta, sigma: demand during lead time
L,l, sigma(l)
Including Lead Time Variability in Formulas
¨ Expected Lead Time Demand: E(X)=E(L)E(D)

¨ Standard Deviation of Lead Time Demand: 

Inflation term due to

lead time variability

¨ Modified Base Stock Formula:

¨ In general:
E(X)=E(L)E(D)
Var(X)=E(L)Var(D) + E(D)2Var(L)
REF: Expected Leadtime
Demand Proof

50
REF: Variance of Leadtime
Demand Proof

51
 5. The (s, S) Model

¨ This is usually called the (s, S) or (r, R) model

¨ Each demand order can be for multiple units
¨ Demand orders are stochastic
¨ A replenishment order is placed to bring inventory position back to R
¨ Decision variables are R (instead of Q) and r

52
 Order-Up-To-Level (s, S) vs (r,Q) System

SS • If all demand
transactions are
unit-sized, two
systems are the
ss same S= r+Q
• (s,S)~ “min-max”
(s,S) (s,
system
S) system rR+Q
+Q

rR

(Q,,R)
(Q,r) system
system
53
 (s, S) System (cont.)
¨ Define two levels, s < S, and let u be the starting inventory at the
beginning of a period.
Then, if u is less or equal to s, order (S – u),
if u is more than s, do not place an order.

¨ In general, computing the optimal values of s and S is extremely

difficult, so few systems operate using optimal (s,S) values.
Approximation: Compute optimal values for (Q,r) model and set
s = r and S = r + Q
6. Periodic Review Systems: (R, S)

55
 Periodic Review Systems
 Continuous Review Systems
 Always knew level of on-hand inventory
 Could place an order at any time
 Often, we are constrained by WHEN we can order --
and it may be periodically
 Train dispatched once a week
 Delivery truck arrives each morning
 Routine products
 Thus, we do not need to continuously review inventory,
just check periodically
 Periodic Review Systems
 If demand were known and constant, we would just resort
to our EOQ solution, possibly modifying it to meet
shipping date
 Now: demand is random variable
 Setting:
 Place an order every T periods
 Policy: Order up to S
 The value of Q (order quantity) will now change periodically
 Previous concern: demand exceeding supply during the
lead time
 Now: demand exceeding supply during the period and lead
time, or T + l

57
 Periodic Review Systems
Order up to S
every T periods of
time.
I(t) Order arrives.
S Cycle continues.
Demand
Q

Lead Time passes…

l l
Time
T
58
 Expected Cost Function
Include expected holding, setup, penalty
and ordering (per unit) costs
Average Inventory Level:
S
At level r*,on average,
order Q = S-r* units.
r* l periods later, units arrive.
Inventory level?
l

R
59
 Expected Cost Function
Include expected: holding, setup,
penalty and ordering (per unit) costs
Average Inventory Level:
S
S-Dl units present when
Q arrive (expected) as
Dl units consumed over
leading time.

60
 Expected Cost Function
Include expected: holding, setup,
penalty and ordering (per unit) costs
Average Inventory Level:
S S - Dl
D*R units
removed
(expected) from
inventory over
time R.
R

61
 Expected Cost Function
Include expected: holding, setup,
penalty and ordering (per unit) costs
Average Inventory Level:
S S-Dl

D*T/2 D*R

S-Dl-D*R
R

𝐷𝑅 𝐷𝑅
( 𝑆=− 𝐷𝑙− 𝐷𝑅 ) +
Average Inventory Level = 𝑆− 𝐷𝑙−
2 2
 Expected Cost Function
 Include expected:
holding, setup, penalty and ordering (per unit) costs

 Average Holding Cost: h ( 𝑆 − 𝐷𝑙−

𝐷𝑅
2 )
 Average Set-up Cost:

63
 Expected Cost Function
 Expected Shortage per Cycle:
 Pr(demand in R + l is between x and x + dx)

 Expected Penalty Cost :

 Expected Cost Function:

 Derivative:
 Recall that R and l are given:


 Example 1
¨ A special control board is used in a version of a product on
the production line
¨ The board cost is $122.50
¨ The holding cost rate is 30% per year
¨ Reorders are placed at the start of each week, and the
supplier delivers these parts in one week
¨ The shortage cost is $100 per board due to worker
downtime
¨ Weekly demand is N (μ=125, σ2=104.17)
¨ Set up cost (A) is $120
¨ Find S

65
 Solution
Holding cost is:
h = ic = .30 (122.50) = 36.75 / 52 = $.7067 per week
Compute:

Demand Distribution is Normal

 mean = 125 variance = 104.17

ZIf =penalty costNormal
2.457 from dropstable
to $10 per unit:
 Compute:  S = 125+(2.457)(104.17)1/2 ~ 150
S   z

 S = 125+(1.47)(104.17)1/2 = 140
If p = $1/pcs  S = 125+(-.54)(104.17)1/2 = 119.49
66
 Solution

If penalty cost drops to $10 per unit:

Compute:

 S = 125+(1.47)(104.17)1/2 = 140
If p = $1/pcs  S = 125+(-.54)(104.17)1/2 = 119.49
 Example 2
¨ A special control board is used in a version of a product on
the production line
¨ The board cost is $122.50
¨ The holding cost rate is 30% per year
¨ If S = 160
¨ The shortage cost is $100 per board due to worker
downtime
¨ Weekly demand is N (μ=125, σ2=104.17)
¨ Set up cost (A) is $120
¨ Find R.

68