Inventory Policies

By: Jason Miller, Ph.D.

 Purposes of Inventory Policies
1. Determine the Quantity to
Order
– Three common methods are to
order for the (1) lead time demand,
(2) day of supply setting, and (3) the
economic order quantity
2. Determine when to Order
– Policies are either periodic [order
every 10 days] or continuous [order
when inventory reaches a certain
level
 Push vs. Pull Inventory Strategies
• Pull-driven inventory strategies
– Each stocking location (e.g., warehouse) is treated as independent
– Demand forecasting and replenishment quantities are done at each
location (e.g., each location places its own orders)
– Typically found in retail
• Push-driven inventory strategies
– Replenishment decisions are considered collectively across stocking
locations
– Done in settings where the economies of scale in
purchasing/production are substantial
Replenishment with a Push System
I
1. Begin by assuming we have product ready to be distributed from the
centralized location
2. For item i at each jth downstream location develop a forecast of demand
that covers the length of the next review interval (R) combined with the
lead time (i.e., R + L)
– Note, the lead time refers to the subsequent lead time for the order after next
3. Determine the inventory position for item i at each jth downstream
location
– Inventory position = inventory on-hand + inventory in transit – inventory allocated to
customers (or on backorder)
4. Calculate order quantity as projected needs (i.e., forecasted demand for
(R+L)) less inventory position for item i at each jth downstream location
– May include an additional order component to cover forecasting error (e.g., an extra 5
percent)
Replenishment with a Push System
II
5. Calculate the total amount of item i that will need ordered across the j
downstream locations
– Sum the required orders across all these locations
6. Determine if fulfilling all of the demand for item i across the j
downstream locations will drive the inventory levels at the centralized
facility too low (e.g., there simply isn’t enough of the item or we would
cut too deep into safety stock)
7. If “no” to #6 then fulfill demand
8. If “yes” then utilize some allocation rule to allocate available stock of
item i across the j downstream locations
– Could utilize a heuristic such as a proportion of the quantity demanded
 Pull Models
• Pull models have been the predominant focus of the
extensive literature on inventory management
• Several ways to classify pull models
– Single period vs. multi-period
– Single echelon vs. multi-echelon
– Continuous review process vs. periodic review process
– Fixed quantity systems vs. order-up-to systems
– Single item versus multi-item
– Stationary demand versus nonstationary demand
 Single Period Pull Model
• Often termed the news-vendor problem
– Start by calculating profit for selling a unit as price per unit – cost per unit
– Then calculate loss as cost per unit – salvage value per unit
– Calculate ratio of profit divided by profit + loss

– Make an assumption about the probability distribution that demand follows.

This requires that we know enough moments to define the probability
distribution in question (e.g., if Poisson only need the mean, but for Normal
need the mean and standard deviation)
– The order quantity corresponds to the value from the cumulative distribution
function of the chosen probability distribution
 (s,Q) Inventory Model (I) for Stationary Demand

• (s,Q) model is designed such that when the current inventory level hits a reorder point (s)
that a replenishment order for Q units is placed.
• Q is typically calculated as the economic order quantity (often incorporating transportation
costs). When Q would exceed a full truckload/containerload quantity, Q is set simply to the
full truckload or containerload quantity.
• Reorder point equals safety stock plus expected demand during lead time
• Model works well when an order quantity is much larger than demand during lead time
(s,Q) Inventory Model (II) for Stationary Demand
• An alternative for the (s,Q) model places
orders based on the inventory position,
not the quantity on hand
• Superior to use the inventory position
when (i) we are allowing demand to be
backordered, (ii) our order quantity is
less than demand during lead time
• This is a much more flexible form than
using the inventory on hand to make
orders (hint, this is a superior model)
 (s,S) Inventory Model for Stationary Demand

• A different form of the reorder point model is called an (s,S) model

whereby an order is placed once the inventory position hits “s” to bring
the inventory position up to a specified maximum level “S”
• This policy is often called a “min-max” system because the inventory
position will always be between a minimum and maximum value
• The (s,S) model is preferable to the (s,Q) model when item use quantities
tend to occur in large transactions
– Example, imagine our reorder point is 20 units and our current quantity on hand is 21
units. If the next transaction is for 10 units, an (s,Q) model will be placing an order when
the inventory position is 11 units rather than 20, which can increase the likelihood of a
stockout
• Typically set the difference between S and s to equal the economic order
quantity
 (s,S) Inventory Model for Stationary Demand

S
Q1 Arrives

Q1

Solid black = quantity on hand

Dashed black = inventory position
 (R,S) Inventory Model for Stationary Demand

• (R,S) model is designed such that we review our inventory at a set interval, denoted by “R”,
and then place an order quantity that brings our inventory position up to a preset level, S.
• The review interval R is often set by calculating the days of supply for an EOQ model (ideally
including transportation cost) and then setting R equal to the days of supply for an optimal
order.
• S equals the expected demand during the lead time and review interval plus safety stock
 (R,s,S) Inventory Model for Stationary Demand

• (R,s, S) model is an extension of the (R,S) model whereby we require the inventory position to
below a certain level (s) at the time of a review in order for us to place an order
• Designed to keep us from placing orders that are so small that we greatly increase our costs
– With a pure (R,S) policy we could place an order as small as Q = 1
 Which Policy for Stationary
Demand?
• Most companies managing independent inventory demand
operate on some form of (R,S) policy, usually a (R,s,S) policy
• Allows coordination of replenishment decisions across items
• Can specify different review intervals based on the
importance of items
– Review intervals for “C” items will tend to be rather longer than “A” items
• Can adjust “s” and “S” parameters to changes in demand
– Silver, Pike, and Peterson (1998) suggest that “R” be changed very infrequently
 Inventory Models for Non-Stationary
Demand
• Stationarity demand implies that demand
(i) has a constant mean over time, (ii) has a
constant variance over time, and (iii) as any
two observations of demand become more
separated by time, their correlation
approaches zero (Enders 2015)
• This is often not the case in practice
– Items have short life cycles
– Items have highly seasonal demand
– Items trend up or down
• Tend to abandon the idea of optimal
policies

Neale & Willems (2009, p. 389)

 Seasonal Effects of Daily Sales

• One particularly important form of non-stationary is that the demand for an

item is higher on some days of the week relative to others
– Inventory systems typically assume that demand is identical for each day
– Problem is that we may get far higher stockouts on the peak demand days

Ehrenthal (2014, p. 528)

Managing Non-Stationary Demand
• Even though many products have non-stationary demand, Tunc et
al. (2011) and Ehrenthal et al. (2014) note that most companies
don’t adjust their inventory systems to account for non-stationary
– Reason: non-stationary inventory models can get complex!
• This has spawned some research to understand when using a
stationary policy is most problematic
1. Greater degree of non-stationary increases the penalty from using a
stationary policy
2. Larger order quantities (or case pack quantities) reduce the penalty from
using a stationary policy
 Retail Replenishment Example
• Consider the example setting from Ehrenthal et al. (2014) for a retailer
open from 7:00 AM till 10:00 PM 7 days a week
– Retailer places orders at 5:00 PM for replenishment by 10:00 AM the next day
• Retailer checks inventory position at 5:00 PM (say 10 units)
• Demand for the next day is forecasted to be 50 units
– Must order at least 40 units
• Case pack quantities are 24 units per case
– Order 2 case packs (Q = 48)
• By having the inventory system set such that average demand for each day
of the week is allowed to vary based on historical records, we can have an
effective non-stationary inventory policy
 Stock to Demand

• Simple non-stationary demand model

1. Start by developing a forecast for the expected demand over the
length of time till the next review (R) and the lead time (L)
• If your review window is 4 weeks and the lead time once the order is placed is 1
week, you forecast demand for 5 weeks
2. Add a desired level of safety stock to the quantity from Step 1
• Possibly an additional 10% above the forecasted demand
3. Determine the inventory position for the item
• Inventory position = quantity of inventory on-hand + quantity of inventory
currently in transit – quantity of inventory allocated to customers (e.g., inventory
that is on backorder or is about to be shipped to the customer)
4. Calculate the quantity (Q) to be ordered by taking the number
calculated in Step 2 – Step 3
 Stock to Demand Example
• There is an item which is reviewed every 20 days and has a
lead time of 5 days. Average demand per day over the next 25
days (subsequent of the lead time of this next order) is
estimated to be 7 units. You additionally assume a forecast
error of 15 percent of forecasted demand. Your records
indicate that you have 50 units on hand, no units on order
from suppliers, and 12 or those 50 units on hand are allocated
to a customer. How much should you order
– Q = (7×(20+5))×1.15 – 50 + 12 = 163.25 or 164 units
 Adjusting for Changes in Current
Inventory Position
• We may also want to adjust order quantity to take into anticipated
inventory position for when the next order would arrive
– Example: We have a 2-week review interval and a 1-week lead time from our suppliers.
This means our order quantity must cover 3 weeks of demand. Imagine we develop a
forecast at the start of Week 1 for demand over the next three week forecast interval
(note, if we are at the start of Week 0, the forecast starts for Week 1 [i.e., we need to
exclude next week due to the 1-week lead time] and covers Weeks 1, 2, and 3.
– Imagine our current inventory position is 100 units at the start of Week 0, and we expect
to sell 70 units this week. This means our anticipated inventory position at the start of
Week 1 is 30 units [assuming there are no units backordered or incoming]
– Let’s imagine our forecast for demand for Weeks 1, 2, and 3 is 200 units
– We would then place an order for 170 units
– At the start of Week 3 we would then make a forecast for demands in Weeks 4, 5, and 6
and place the order for arrival in Week 4
– Note how we would order 70 units more with this approach than in the earlier slide
because we adjust the inventory position based on demand during the next lead time
 Adaptive Base Stock Policy
• Developed by Graves (1999) for demand series that follow a
simple exponential moving average structure
– Errors are assumed to be uncorrelated with the magnitude of demand

= order quantity placed in period t for delivery L periods from now

= demand in the current period
= average lead time in periods
= forecast for the next period
= forecast for the current period
• And now we will look at an example
Safety Stock with an Adaptive Base
Stock Policy
• Graves (1999) demonstrates that the standard deviation of
demand over lead time is much more complex than assumed
when demand is stationary

• Graves (1999) demonstrates this has very important

implications when lead time is longer than a period
Inventory Models for Non-Stationary
Demand
• Assuming a constant lead time, the standard
deviation of demand during lead time reduces to
• This implies a concave relationship in that the
pooled standard deviation increases at a less than
linear rate as lead time increases
• However, when we have a demand series
characterized by a positive moving average
parameter that is integrated of order 1 (very
common in practice), we see that starts to
become convex as lead time increases
• Implication: we need to hold a lot more safety
stock when demand is nonstationary than when it
is stationary, with this effect becoming more
pronounced as lead time lengthens

Graves (1999, p. 54)