08fl3
TRANSCRIPT
-
8/13/2019 08fl3
1/63
Inventory Management and Risk
Pooling
Designing & Managing the Supply Chain
Chapter 3Byung-Hyun Ha
mailto:[email protected]:[email protected] -
8/13/2019 08fl3
2/63
Outline
Introduction to Inventory Management
Effect of Demand Uncertainty
(s,S) Policy
Supply Contracts
Continuous/Periodic Review Policy
Risk Pooling
Centralized vs. Decentralized Systems
Managing Inventory in Supply Chain Practical Issues in Inventory Management
Forecasting
-
8/13/2019 08fl3
3/63
Case: JAM USA, Service Level Crisis
Background Subsidiary of JAM Electronics (Korean manufacturer)
Established in 1978
Five Far Eastern manufacturing facilities, each in different countries
2,500 different products, a central warehouse in Korea for FGs
A central warehouse in Chicago with items transported by ship Customers: distributors & original equipment manufacturers
(OEMs)
Problems Significant increase in competition
Huge pressure to improve service levels and reduce costs
Al Jones, inventory manage, points out: Only 70% percent of all orders are delivered on time
Inventory, primarily that of low-demand products, keeps pile up
-
8/13/2019 08fl3
4/63
Case: JAM USA, Service Level Crisis
Reasons for the low service level: Difficulty forecasting customer demand
Long lead time in the supply chain
About 6-7 weeks
Large number of SKUs handled by JAM USA
Low priority given the U.S. subsidiary by headquarters in Seoul
Monthly demand for item xxx-1534
-
8/13/2019 08fl3
5/63
Inventory
Where do we hold inventory? Suppliers and manufacturers / warehouses and distribution
centers / retailers
Types of Inventory Raw materials / WIP (work in process) / finished goods
Reasons of holding inventory Unexpected changes in customer demand
The short life cycle of an increasing number of products.
The presence of many competing products in the marketplace.
Uncertainty in the quantity and quality of the supply, suppliercosts and delivery times.
Delivery lead time and capacity limitations even with nouncertainty
Economies of scale (transportation cost)
-
8/13/2019 08fl3
6/63
-
8/13/2019 08fl3
7/63
Single Warehouse Inventory Example
Key factors affecting inventory policy Customer demand characteristics
Replenishment lead time
Number of products
Length of planning horizon
Cost structure
Order cost
Fixed, variable
Holding cost
Taxes, insurance, maintenance, handling, obsolescence, andopportunity costs
Service level requirements
-
8/13/2019 08fl3
8/63
Economy Lot Size Model
Assumptions Constant demand rate of D items per day
Fixed order quantities at Q items per order
Fixed setup cost K when places an order
Inventory holding cost h per unit per day Zero lead time
Zero initial inventory & infinite planning horizon
Time
Inventory
OrderSize
Avg. Inven
-
8/13/2019 08fl3
9/63
Economy Lot Size Model
Inventory level
Total inventory cost in a cycle of length T
Average total cost per unit of time
Time
Inventory
OrderSize
Avg. Inven
2
hQ
Q
KD
Cycle time (T)
(Q)
D
QT
TDQ
2
hTQK
-
8/13/2019 08fl3
10/63
Economy Lot Size Model
Trade-off between order cost and holding cost
0
20
40
60
80
100
120
140
160
0 500 1000 1500
Order Quantity
Cost
Total Cost
Order Cost
Holding Cost
-
8/13/2019 08fl3
11/63
Economy Lot Size Model
Optimal order quantity Economic order quantity (EOQ)
Important insights
Tradeoff between setup costs and holding costs whendetermining order quantity. In fact, we order so that these costsare equal per unit time
Total cost is not particularly sensitive to the optimal orderquantity
h
KDQ
2*
Order Quantity 50% 80% 90% 100% 110% 120% 150% 200%
Cost Increase 25.0% 2.5% 0.5% - 0.4% 1.6% 8.0% 25.0%
-
8/13/2019 08fl3
12/63
Effect of Demand Uncertainty
Most companies treat the world as if it werepredictable
Production and inventory planning are based on forecasts of
demand made far in advance of the selling season
Companies are aware of demand uncertainty when they create aforecast, but they design their planning process as if the forecast
truly represents reality
Recent technological advances have increased the level of
demand uncertainty
Short product life cycles
Increasing product variety
-
8/13/2019 08fl3
13/63
Effect of Demand Uncertainty
Three principles of all forecasting techniques: Forecasting is always wrong
The longer the forecast horizon the worst is the forecast
Aggregate forecasts are more accurate
News vendor model
Incorporating demand uncertainty and forecast demand into
analysis
Characterizing impact of demand uncertainty on inventory policy
-
8/13/2019 08fl3
14/63
Case: Swimsuit Production
Fashion items have short life cycles, high variety ofcompetitors
Swimsuit production
New designs are completed
One production opportunity Based on past sales, knowledge of the industry, and economic
conditions, the marketing department has a probabilisticforecast
The forecast averages about 13,000, but there is a chance that
demand will be greater or less than this
-
8/13/2019 08fl3
15/63
Case: Swimsuit Production
Information Production cost per unit (C) = $80
Selling price per unit (S) = $125
Salvage value per unit (V) = $20
Fixed production cost (F) = $100,000 Production quantity (Q)
Demand Scenarios
0%5%
10%15%
20%25%30%
8000
1000
0
1200
0
1400
0
1600
0
1800
0
Sales
Probability
-
8/13/2019 08fl3
16/63
Case: Swimsuit Production
Possible scenarios Make 10,000 swimsuits, demand ends up being 12,000
Profit = 125(10,000) - 80(10,000) - 100,000 = $350,000
Make 10,000 swimsuits, demand ends up being 8,000
Profit = 125(8,000) - 80(10,000) - 100,000 + 20(2,000) = $140,000
-
8/13/2019 08fl3
17/63
Swimsuit Production Solution
Problem Find optimal order quantity Q*that maximizes expected profit
Notation
diith possible demand where di< di+1
(d1, d2, d3, d4, d5, d6) = (8K, 10K, 12K, 14K, 16K, 18K) piprobability that ith possible demand occurs
(p1,p2,p3,p4,p5,p6) = (0.11, 0.11, 0.28, 0.22, 0.18, 0.10)
ES(Q)expected sales when producing Q
EI(Q)expected left over inventory when producing Q
EP(Q)expected profit when producing Q
Expected profit
EP(Q) = SES(Q)CQF+ VEI(Q)
-
8/13/2019 08fl3
18/63
Swimsuit Production Solution
Expected sales Case 1: Qd1
Case 2: dk< Qdk+1, k=1,,5
Case 3: d6Q
Expected left over inventory (why?)
EI(Q) = QES(Q)
QQES )(
6
11
)(ki
i
k
i
ii pQdpQES
6
1
)(i
iidpQES
6
1
},min{)(i
ii QdpQES
-
8/13/2019 08fl3
19/63
Swimsuit Production Solution
Quantity that maximizes average profit
Expected Profit
$0$100,000
$200,000
$300,000
$400,000
8000 12000 16000 20000
Order Quantity
Profit
-
8/13/2019 08fl3
20/63
Swimsuit Production Solution
Question Average demand is 13,000
Will optimal quantity be less than, equal to, or greater thanaverage demand?
Note that fixed production cost is sunk cost
We have to pay in all cases
-
8/13/2019 08fl3
21/63
Swimsuit Production Solution
Look at marginal cost vs. marginal profit If extra swimsuit sold, profit is 125-80 = 45
If not sold, cost is 80-20 = 60
In case of Scenario Two (make 10,000, demand 8,000)
Profit = 125(8,000) - 80(10,000) - 100,000 + 20(2,000)= 125(8,000) - 80(8,000 + 2,000) - 100,000 + 20(2,000)= 45(8,000) - 60(2,000) - 100,000 = $140,000
That is,
EP(Q) = SES(Q)CQF+ VEI(Q)= SES(Q)C{ES(Q) + EI(Q)}F+ VEI(Q)
= (SC)ES(Q)(CV)EI(Q)F
So we will make less than average
-
8/13/2019 08fl3
22/63
Swimsuit Production Solution
Tradeoff between ordering enough to meet demandand ordering too much
Several quantities have the same average profit
Average profit does not tell the whole story
Question: 9000 and 16000 units lead to about thesame average profit, so which do we prefer?
Expected Profit
$0
$100,000
$200,000
$300,000
$400,000
8000 12000 16000 20000
Order Quantity
Profit
-
8/13/2019 08fl3
23/63
Swimsuit Production Solution
Risk and reward
0%
20%
40%60%
80%
100%
-30000
0
-10000
0
10000
0
30000
0
50000
0
Revenue
Probab
ility
Q=9000Q=16000
Consult Ch13 of Winston, Decision making under uncertainty
-
8/13/2019 08fl3
24/63
Case: Swimsuit Production
Key insights The optimal order quantity is not necessarily equal to average
forecast demand
The optimal quantity depends on the relationship betweenmarginal profit and marginal cost
As order quantity increases, average profit first increases andthen decreases
As production quantity increases, risk increases (the probabilityof large gains and of large losses increases)
-
8/13/2019 08fl3
25/63
-
8/13/2019 08fl3
26/63
Case: Swimsuit Production
Analysis for initial inventory and profit Solid line: average profit excluding fixed cost Dotted line: same as expected profit including fixed cost
Nothing produced 225,000 (from the figure) + 80(5,000) = 625,000
Producing 371,000 (from the figure) + 80(5,000) = 771,000
If initial inventory was 10,000?
0
100000
200000
300000
400000
500000
5000
6000
7000
8000
9000
10000
11000
12000
13000
14000
15000
16000
Production Quantity
Profit
-
8/13/2019 08fl3
27/63
Case: Swimsuit Production
Initial inventory and profit
0
100000
200000
300000
400000
500000
5000
6000
7000
8000
9000
10000
11000
12000
13000
14000
15000
16000
Production Quantity
Profit
-
8/13/2019 08fl3
28/63
Case: Swimsuit Production
(s, S) policies For some starting inventory levels, it is better to not start
production
If we start, we always produce to the same level
Thus, we use an (s, S) policy
If the inventory level is below s, we produce up to S
sis the reorder point, and Sis the order-up-to level
The difference between the two levels is driven by the fixedcosts associated with ordering, transportation, or manufacturing
-
8/13/2019 08fl3
29/63
Supply Contracts
Assumptions for Swimsuit production In-house manufacturing
Usually, manufactures and retailers
Supply contracts
Pricing and volume discounts Minimum and maximum purchase quantities
Delivery lead times
Product or material quality
Product return policies
-
8/13/2019 08fl3
30/63
Supply Contracts
Manufacturer Manufacturer DC Retail DC
Stores
Selling Price=$125
Salvage Value=$20
Wholesale Price =$80
Fixed Production Cost =$100,000
Variable Production Cost=$35
Condition
-
8/13/2019 08fl3
31/63
Demand Scenario and Retailer Profit
Demand Scenarios
0%
5%10%15%
20%25%30%
8000
1000
0
1200
0
1400
0
1600
0
1800
0
Sales
Probability
Expected Profit
0
100000
200000
300000
400000
500000
6000 8000 10000 12000 14000 16000 18000 20000
Order Quantity
-
8/13/2019 08fl3
32/63
Demand Scenario and Retailer Profit
Sequential supply chain Retailer optimal order quantity is 12,000 units
Retailer expectedprofit is $470,700
Manufacturer profit is $440,000
Supply Chain Profit is $910,700
Is there anything that the distributor andmanufacturer can do to increase the profit of both?
Global optimization?
-
8/13/2019 08fl3
33/63
Buy-Back Contracts
Buy back=$55
0
100,000
200,000
300,000
400,000
500,000
600,000
6000
7000
8000
9000
1000
0
1100
0
1200
0
1300
0
1400
0
1500
0
1600
0
1700
0
1800
0
Order Quantity
RetailerProfit
$513,800
retailer
manufacturer
0
100,000
200,000
300,000
400,000
500,000
600,000
6000
7000
8000
9000
1000
0
1100
0
1200
0
1300
0
1400
0
1500
0
1600
0
1700
0
1800
0
Production Quantity
ManufacturerProfit $471,900
-
8/13/2019 08fl3
34/63
Buy-Back Contracts
Sequential supply chain Retailer optimal order quantity is 12,000 units
Retailer expectedprofit is $470,700
Manufacturer profit is $440,000
Supply chain profit is $910,700
With buy-back contracts
Retailer optimal order quantity is 14,000 units
Retailer expectedprofit is $513,800
Manufacturer expectedprofit is $471,900 Supply chain profit is $985,700
Manufacture sharing some of risk!
-
8/13/2019 08fl3
35/63
-
8/13/2019 08fl3
36/63
-
8/13/2019 08fl3
37/63
Global Optimization
What is the most profit both the supplier and thebuyer can hope to achieve?
Assume an unbiased decision maker
Transfer of money between the parties is ignored
Allowing the parties to share the risk!
Marginal profit=$90, marginal loss=$15
Optimal production quantity=16,000
Drawbacks
Decision-making power Allocating profit
0
200,000
400,000
600,000
800,000
1,000,000
1,200,000
6000
7000
8000
9000
1000
0
1100
0
1200
0
1300
0
1400
0
1500
0
1600
0
1700
0
1800
0
Production Quantity
SupplyChainP
rofit $1,014,500
0
200,000
400,000
600,000
800,000
1,000,000
1,200,000
6000
7000
8000
9000
1000
0
1100
0
1200
0
1300
0
1400
0
1500
0
1600
0
1700
0
1800
0
Production Quantity
SupplyChainP
rofit $1,014,500
-
8/13/2019 08fl3
38/63
Global Optimization
Revised buy-back contracts Wholesale price=$75, buy-back price=$65
Global optimum
Equilibrium point!
No partner can improve his profit by deciding to deviate from theoptimal decision
Consult Ch14 of Winston, Game theory
Key Insights
Effective supply contracts allow supply chain partners to replacesequential opt im izat ionby glob al op timizat ion
Buy Back and Revenue Sharing contracts achieve this objectivethrough r isk shar ing
-
8/13/2019 08fl3
39/63
Supply Contracts: Case Study
Example: Demand for a movie newly released videocassette typically starts high and decreases rapidly
Peak demand last about 10 weeks
Blockbuster purchases a copy from a studio for $65
and rent for $3 Hence, retailer must rent the tape at least 22 times before
earning profit
Retailers cannot justify purchasing enough to cover
the peak demand In 1998, 20% of surveyed customers reported that they could not
rent the movie they wanted
-
8/13/2019 08fl3
40/63
Supply Contracts: Case Study
Starting in 1998 Blockbuster entered a revenue-sharing agreement with the major studios
Studio charges $8 per copy
Blockbuster pays 30-45% of its rental income
Even if Blockbuster keeps only half of the rentalincome, the breakeven point is 6 rental per copy
The impact of revenue sharing on Blockbuster wasdramatic
Rentals increased by 75% in test markets Market share increased from 25% to 31% (The 2nd largest
retailer, Hollywood Entertainment Corp has 5% market share)
-
8/13/2019 08fl3
41/63
Multiple Order Opportunities
Situation Often, there are multiple reorder opportunities
A central distribution facility which orders from a manufacturerand delivers to retailers
The distributor periodically places orders to replenish its inventory
Reasons why DC holds inventory
Satisfy demand during lead time
Protect against demand uncertainty
Balance fixed costs and holding costs
-
8/13/2019 08fl3
42/63
Continuous Review Inventory Model
Assumptions Normally distributed random demand
Fixed order cost plus a cost proportional to amount ordered
Inventory cost is charged per item per unit time
If an order arrives and there is no inventory, the order is lost
The distributor has a required service level
expressed as the likelihood that the distributor will not stock outduring lead time.
(s, S) Policy
Whenever the inventory position drops below a certain level (s)we order to raise the inventory position to level S
-
8/13/2019 08fl3
43/63
Reminder: The Normal Distribution
0 10 20 30 40 50 60
Average = 30
Standard Deviation = 5
Standard Deviation = 10
-
8/13/2019 08fl3
44/63
A View of (s, S) Policy
Time
InventoryL
evel
S
s
0
Lead
Time
Lead
Time
Inventory Position
-
8/13/2019 08fl3
45/63
(s, S) Policy
Notations AVG = average daily demand
STD = standard deviation of daily demand
LT = replenishment lead time in days
h = holding cost of one unit for one day
K = fixed cost
SL = service level (for example, 95%)
The probability of stocking out is 100% - SL (for example, 5%)
Policy
s= reorder point, S = order-up-to level
Inventory Position
Actual inventory + (items already ordered, but not yet delivered)
-
8/13/2019 08fl3
46/63
(s, S) Policy - Analysis
The reorder point (s) has two components: To account for average demand during lead time:
LTAVG
To account for deviations from average (we call this safetystock)
zSTDLTwhere zis chosen from statistical tables to ensure that theprobability of stock-outs during lead-time is100% - SL.
Since there is a fixed cost, we order more than up to
the reorder point:Q=(2KAVG)/h
The total order-up-to level is:S= Q+ s
-
8/13/2019 08fl3
47/63
(s, S) Policy - Example
The distributor has historically observed weekly demand of:
AVG= 44.6 STD= 32.1
Replenishment lead time is 2 weeks, and desired service levelSL = 97%
Average demand during lead time is: 44.6 2 = 89.2
Safety Stock is: 1.88 32.1 2 = 85.3
Reorder point is thus 175, or about 3.9 weeks of supply atwarehouse and in the pipeline
Weekly inventory holding cost: .87 Therefore, Q=679
Order-up-to level thus equals:
Reorder Point + Q = 176+679 = 855
-
8/13/2019 08fl3
48/63
Periodic Review
Periodic review model Suppose the distributor places orders every month
What policy should the distributor use?
What about the fixed cost?
Base-Stock Policy
Inventory
Level
Time
Base-stock
Level
0
Inventory
Position
r r
L L L
Inventory
Level
Time
Base-stock
Level
0
Inventory
Position
r r
L L L
Time
Base-stock
Level
0
Inventory
Position
r r
L L L
-
8/13/2019 08fl3
49/63
Periodic Review
Base-Stock Policy Each review echelon, inventory position is raised to the base-
stock level.
The base-stock level includes two components:
Average demand during r+Ldays (the time until the next order
arrives):(r+L)*AVG
Safety stock during that time:z*STD* r+L
-
8/13/2019 08fl3
50/63
Risk Pooling
Consider these two systems: For the same service level, which system will require more
inventory? Why?
For the same total inventory level, which system will have betterservice? Why?
What are the factors that affect these answers?
Market Two
Supplier
Warehouse One
Warehouse Two
Market One
Market Two
Supplier Warehouse
Market One
-
8/13/2019 08fl3
51/63
Risk Pooling Example
Compare the two systems: two products
maintain 97% service level
$60 order cost
$.27 weekly holding cost
$1.05 transportation cost per unit in decentralized system, $1.10in centralized system
1 week lead timeWeek 1 2 3 4 5 6 7 8
Prod A,
Market 1
33 45 37 38 55 30 18 58
Prod A,
Market 2
46 35 41 40 26 48 18 55
Prod B,
Market 1
0 2 3 0 0 1 3 0
Product B,
Market 2
2 4 0 0 3 1 0 0
-
8/13/2019 08fl3
52/63
Risk Pooling Example
Risk Pooling Performance
Warehouse Product AVG STD CV s S Avg.
Inven.
%
Dec.
Market 1 A 39.3 13.2 .34 65 197 91
Market 2 A 38.6 12.0 .31 62 193 88
Market 1 B 1.125 1.36 1.21 4 29 14
Market 2 B 1.25 1.58 1.26 5 29 15
Cent. A 77.9 20.7 .27 118 304 132 36%
Cent B 2.375 1.9 .81 6 39 20 43%
-
8/13/2019 08fl3
53/63
Risk Pooling: Important Observations
Centralizing inventory control reduces both safetystock and average inventory level for the sameservice level.
This works best for
High coefficient of variation, which increases required safetystock.
Negatively correlated demand. Why?
What other kinds of risk pooling will we see?
-
8/13/2019 08fl3
54/63
Centralized vs. Decentralized Systems
Trade-offs that need to be considered in comparingcentralized and decentralized systems
Safety stock
Service level
Overhead costs
Customer lead time
Transportation costs
-
8/13/2019 08fl3
55/63
Inventory in Supply Chain
Centralized distribution systems How much inventory should
management keep at each location?
A good strategy:
The retailer raises inventory to level Sreach period
The supplier raises the sum ofinventory in the retailer and supplierwarehouses and in transit to Ss
If there is not enough inventory in thewarehouse to meet all demands fromretailers, it is allocated so that theservice level at each of the retailers willbe equal.
Supplier
Warehouse
Retailers
-
8/13/2019 08fl3
56/63
-
8/13/2019 08fl3
57/63
Forecasting
Recall the three rules
Nevertheless, forecast is critical
General overview:
Judgment methods
Market research methods
Time series methods
Causal methods
-
8/13/2019 08fl3
58/63
Forecasting
Judgment methods Assemble the opinion of experts
Sales-force composite combines salespeoples estimates
Panels of expertsinternal, external, both
Delphi method
Each member surveyed
Opinions are compiled
Each member is given the opportunity to change his opinion
-
8/13/2019 08fl3
59/63
Forecasting
Market research methods Particularly valuable for developing forecasts of newly introduced
products
Market testing
Focus groups assembled
Responses tested Extrapolations to rest of market made
Market surveys
Data gathered from potential customers
Interviews, phone-surveys, written surveys, etc.
-
8/13/2019 08fl3
60/63
Forecasting
Time series methods Past data is used to estimate future data
Examples include
Moving averagesaverage of some previous demand points.
Exponential Smoothingmore recent points receive more weight
Methods for data with trends: Regression analysisfits line to data
Holts method combines exponential smoothing concepts withthe ability to follow a trend
Methods for data with seasonality
Seasonal decomposition methods (seasonal patterns removed) Winters method: advanced approach based on exponential
smoothing
Complex methods (not clear that these work better)
-
8/13/2019 08fl3
61/63
Forecasting
Causal methods Forecasts are generated based on data other than the data
being predicted
Examples include:
Inflation rates
GNP Unemployment rates
Weather
Sales of other products
-
8/13/2019 08fl3
62/63
Forecasting
Selecting appropriate forecasting technique What is purpose of forecast? How is it to be used?
What are dynamics of system for which forecast will be made?
How important is past in estimating future?
-
8/13/2019 08fl3
63/63
Summary
Matching supply and demand
Inventory management
Reducing cost and providing required service level
Taking into account holding and setup cost, lead time, forecast
demand, and information about demand variability Demand aggregation
Globally optimal inventory policies
Global optimization