The Effects of the Source of Policy Deviation in a Decentralized

SC

Joong Y. Son*Chwen Sheu**

*MacEwan School of BusinessGrant MacEwan College

**Department of ManagementKansas State University

Oct. 31, 2008

Research Forum

Contents

Policy Deviation – Examples & Issues

Research Questions

Literature Review

Model Descriptions

Policy Deviations in a Decentralized SC

Numerical Results & Managerial Implications

Future Research

Examples

Campbell’s winter sales promotion of Chicken noodle soup

Volvo’s special deal on green cars Cisco’s over-reliance on its forecasting

technology and misaligned incentives with partners

Mar. 2000 most valuable company, MV of $555 bn

May 2001 inventory write-off ($2.2 billion)

Supply Chain Underperformance

Who/What is responsible? Misaligned incentives

(Lee & Whang 1999)

Information asymmetry (Corbett & de Groote 2000)

Behavioral causes(Sterman 1987, 1989)

Decentralized and myopic

supply chain policies(Croson & Donohue 2002)

Remedies? SC coordination

(Lee & Whang 1999; Klastorin et. al 2002)

Information sharing/ exchange (Cachon & Lariviere 2001; Moinzadeh 2002; Huang et. al 2003)

Team approach in ordering policy (Chen 1999; Chen & Samroengraja 2000)

Focus of this Research

Coordination (benchmark) vs. Decentralized repl. policies

Benchmark policy: base stock policy at each installation Cost of policy deviation: decentralized replenishment

policies Order for order policy MA based ES based

The impact of deviation based on the source (relative position with SC)

Managerial implications

Research Questions

What are the penalties (local & system-wide) for deviating from the benchmark policy?

Relationships between structural parameters (costs, demand variations) vs. policy parameters (replenishment policies and base stock levels) in a steady state.

What is the relationship between the source (relative position) of deviations and supply chain performance?

Incorporating incentive compatible design (stock outs in SC)

Related Works

Microeconomics perspective: Radner (1987),

Marschak & Radner (1972)

SC coordination – incentive compatibility:

Jeuland and Shugan (1983), Lee and Rosenblatt (1986),Balakrishnan et al (2004)

Information sharing: Lee & Whang (1999),

Gilbert & Ballou (1999), Huang et al (2003), Chen et al (2000)

Setting: Clark & Scarf (1960),

Sterman (1989, 1992), Steckel et al (2004), Chatfield (2004)

Replenishment Policies (smoothing algorithms)

Dejonckheere et al (2002) Warburton (2004), Balakrishnan (2004)

Multi-agent based modeling: Kimbrough et al (2002)

Sikora & Shaw (1998) Swaminathan et al (1998)

Supply Chain Structure

Information delays

no delays 2 weeks 2 weeks 2 weeks 2 week

no delays 2 weeks 2 weeks 2 weeks 2 weeks

Transportation Delays

Factory Distributor Retailer Wholesaler F Supply F Demand

A four-stage serial supply chain

Model Settings and Assumptions

Factor Setting

Demand distribution Normal random variable. Discrete and truncated at zero

Demand and variation Avg weekly demand = 50 units std weekly demand = 5, 10, 20 units

Shipping & Info delays Two weeks between any two adjacent positions

Initial on-hand inventory 120 units at all positions

Final demand information availability

Benchmark case: Information is known to the central coordinator. Policy deviation case: Information is either asymmetric or underutilized.

Replenishment Decisions Benchmark case: Base stock, imposed by the central coordinator. Policy deviation case: Base stock, LFL, MA, ES Policy deviations could occur at one or more positions.

Unit backorder cost/week $1, $2, $5, and $10 for all positions.

Unit holding cost/week Retailer: $1; Wholesaler: $0.75; Distributor: $0.50; Factory: $0.25

Model Settings and Assumptions

A four-stage decentralized serial supply chain Beer distribution game Incentive design on stockouts at each station

Based on Clark & Scarf (1960) Different from Chen’s (1999) “team approach” or “cost

centre” Stockout penalty is incurred at each position Upstream position is responsible for portion of stockouts

at its immediate buyer

Downstream positions have better access to demand information. In case of policy deviations at multiple stations, deviations are

more likely to occur from upstream

Benchmark Replenishment Policy

Benchmark policy for a four-stage serial supply chain Base stock policy at each installation with incentive to

stock Minimize long run average supply chain costs Central coordinator/ planner Information availability

Base stock level at position i (si) satisfies standard newsvendor results:

ii

ii hb

bsF

)(

Sequence of Simulation Run

Steps Actions at Each Position

0 Customer demand is generated based on normal distribution with an average of 50

units and a st. dev. of 20 units per week.

1 Advance shipping delays by one period between any two successive stations in the

supply chain (i.e., move shipping delay 1 to the current inventory and shipping delay 2

to shipping delay 1).

2 Take the incoming orders (demand) from the downstream position.

3 Fill the incoming demand plus any backorders carried over from the previous weeks.

4 Advance the information delays by one week between any two successive stations.

5 Determine the number of backorders at each position.

6 Identify the party responsible for the backorders incurred at each position.

7 Appropriately allocate backorder costs to the responsible party within the chain.

8 Determine order amount to be placed and place orders with the upstream position.

(based on the replenishment policy adopted)

Costs for the Benchmark Case

0.0

20.0

40.0

60.0

80.0

100.0

120.0

140.0

160.0

180.0

200.0

b=1 b=2 b=5 b=10

unit stockout penalty

cost

RetWhlslrDistFacSC

Results: Benchmark case

On Hand Inventory

0.0

50.0

100.0

150.0

200.0

250.0

300.0

b=1 b=2 b=5 b=10

unit stockout penalty

inv

leve

l

Ret

Whlslr

Dist

Fac

SC

Results: Benchmark case

Results: Benchmark case

The base stock level and SL at each station increases monotonically in unit backorder cost The base stock / SL: lowest at the retailer level

highest at the factory

Service level remains relatively constant in demand variations Service level determined directly by the cost

structure (bi/hi)

The base stock level increases in demand variations

Policy Deviations

Policy Deviations from the Base-stock Policy (BS)*

Order-for-Order (OFO) Moving Average (MA)

N=2, 4, 10

Exponential Smoothing (ES)

= 0.1, 0.5, 0.9

Ret Whole Dist Factory Ret W h o le Dist F a c to ry Ret W h o le Dist F a c to ry

OFO BS BS BS BS BS OFO

BS OFO BS BS BS OFO OFO

BS BS OFO BS OFO OFO OFO

BS BS OFO OFO OFO OFO OFO

MA BS BS BS BS BS MA

BS MA BS BS BS MA MA

BS BS MA BS MA MA MA

BS BS BS MA MA MA MA

ES BS BS BS BS BS ES

BS ES BS BS BS ES ES

BS BS ES BS ES ES ES

BS BS BS ES ES ES ES

* Benchmark case: Base stock policy at all positions

Replenishment Scenarios

Policy Deviations

Order for order policy (OFO/ LFL) Similar to the base stock policy (order =

demand)

Moving average based policy N= 2, 4, 10

Exponential smoothing based policy = 0.1, 0.5, 0.9

Total # of decentralized policies tested = 588 2,000 replications over 1,000 weeks

Results: Policy Deviations OFO

Parameters: St. dev. of Weekly Demand=20, Backorder Cost/unit=$5.00

Order-for-order deviation at: Average weekly

costs at

Base-Stock

Policy Retailer Wholesaler Distributor Factory

Retailer

Wholesaler

Distributor

Factory

Supply Chain

$57.8

$47.7

$35.5

$20.8

$161.8

$65.6

47.5

35.5

20.8

$169.4

$54.2

59.7

35.6

20.8

$170.3

$56.5

44.2

53.1

20.8

$174.6

$57.5

46.9

32.9

46.7

$184.0

Base stock vs. OFO

Results: Policy Deviations OFO

0.0

20.0

40.0

60.0

80.0

100.0

120.0

140.0

160.0

180.0

200.0

Base Stock Policy OFO at Retailer OFO at Wholesaler OFO at Distributor OFO at Factory

Co

sts

Retailer

Wholesaler

Distributor

Factory

Supply Chain

OFO policy:

σ=20, bi=$5.00

Results: Policy Deviations OFO

Deviating station (s): incurred the highest costs

Non-deviating stations: not significantly different from the benchmark case (in fact, slightly lower)

OFO: results in under-stocking at the deviating station (e.g., the distributor)

Unfilled orders downstream lower OH at the wholesaler The distributor accountable for much of backorders at

downstream stations

With high bi, upstream deviation more costly to the entire SC

Results: Policy Deviations-MA based

MA2 based

σ=20, bi=$5.00

0

20

40

60

80

100

120

140

160

180

200

MA2 at Retailer

MA2 atWholesaler

MA2 atDistributor

MA2 at Factory

Base StockPolicy

Co

sts

Retailer

Wholesaler

Distributor

Factory

Supply Chain

Results: Policy Deviations-MA based

MA 10 based:

σ=20, bi=$5.00

0.0

50.0

100.0

150.0

200.0

250.0

Base StockPolicy

MA10 at Retailer

MA10 atWholesaler

MA10 atDistributor

MA10 at Factory

Co

sts

Retailer

Wholesaler

Distributor

Factory

Supply Chain

Results: Policy Deviations-MA based

Shorter MA period, N=2 or 4 Bigger order sizes and overstocking Little impact on other stations Downstream deviations more costly

Longer MA period N=10 Smoothing effects Frequent demand-supply misalignment Deviating party responsible for stock outs at

downstream positions Upstream deviations more costly

Results: Policy Deviations-ES based

ES based

σ=20, bi=$5.00

0.0

50.0

100.0

150.0

200.0

250.0

300.0

Base stock ES 0.5 atRetailer

ES 0.5 atWholesaler

ES 0.5 atDistributor

ES 0.5 atFactory

ES 0.5 atDistributor/

Factory

ES 0.5 atWholesaler/Distributor/

Factory

ES 0.5 at All

positions

Co

sts

Retailer

Wholesaler

Distributor

Factory

Supply Chain

Results: Policy Deviations-ES based

Consistent with MA based results Small α=0.1

Significant under-stocking at the deviating station demand – supply timing mismatch high stock-out penalties

Benefit from low OH outweighed by huge stock-out penalty borne by the deviating station

Large α=0.5, 0.9 Overstocking at the deviating station Downstream deviations more costly

Results: % increase SC costwith Policy Deviations

Potential cost savings for the SC by implementing the benchmark case

)(benchmark costs SC

)(benchmark costs SC- deviation)(policy costs SC costs SCin increase %

Results: % increase SC costPolicy Deviations - OFO

OFO based

σ=20

0.00%

10.00%

20.00%

30.00%

40.00%

50.00%

60.00%

70.00%

80.00%

90.00%

100.00%

b=$1.00 b=$2.00 b=$5.00 b=$10.00

Unit Backorder Cost

% I

ncr

ease

in

Su

pp

ly C

hai

n C

ost

s

Retailer

Wholesaler

Distributor

Factory

Distributor/ Factory

Wholesaler/ Distributor/ Factory

All Positions

Policy deviations Order-for-order at

Results: % increase SC costPolicy Deviations – ES based

ES based

σ=20, α=0.9

0.00%

20.00%

40.00%

60.00%

80.00%

100.00%

120.00%

140.00%

160.00%

b=$1.00 b=$2.00 b=$5.00 b=$10.00

Unit Backorder Cost

% I

ncr

ease

in

Su

pp

ly C

hai

n C

ost

s Retailer

Wholesaler

Distributor

Factory

Distributor/ Factory

Wholesaler/ Distributor/ Factory

All Positions

Policy deviations ES 0.9 at

Results: % increase in SC cost

Order-for-order based policy Tendency to understock With high bi higher potential savings staying

with BS

MA with small N/ ES with high α Overstock Downstream deviation costly With high bi potential savings lower with BS

Managerial Implications

In a steady state, OFO policy deviations (from the benchmark base stock policy) at a given station result in Under-stocking at the deviating station Higher costs at the deviating station Non-deviating stations may benefit

Especially at an immediate downstream positionSome cases result in local costs lower than the

benchmark

Managerial Implications

Decentralized policies with greater smoothing effects (larger N for MA, and smaller for ES) tend to Display significant under-stocking/ misaligned D-S Result in the worst SC performance both locally

and globally (could be parameter-specific)

Decentralized policies with higher responsiveness (smaller N for MA and greater for ES) exhibit Overstocking Relatively strong SC performance

Managerial Implications

Downstream positions (e.g. the retailer or the wholesaler) have incentive NOT to share demand information with others

When others deviate, the retailer exhibits the lowest local cost (even lower than the benchmark case)

Need incentive compatible mechanism in SC to have the retailer share demand information with the rest of the stations

Regardless of policy deviations at other stations Staying with base stock policy warrants local protection

against deviations at other positions

Managerial Implications

Regardless of whether companies have seemingly perfect operations or not, it is crucial for parties to coordinate and share information (Booker, 2001)

Think globally, act locally!

Future Research

Supply chain design Multiple stations at each stage Determining the number of buyers and

suppliers Degree of heterogeneity of buyers and

suppliers

Incorporating risk pooling and other coordination mechanisms (e.g. revenue sharing)