Demand Information Distortion and

Bullwhip Effect

Chopra: 3rd ed.: Chap. 17

Lessons of the Game

• Such oscillations are common

– Bullwhip effect (demand distortion)

• Everyone blames others - but problem is

with the structure

–

Lessons of the Game

• Such oscillations are common

– Bullwhip effect (demand distortion)

• Everyone blames others - but problem is with the structure

– The root cause: uncertainties in end demands

– Lack of information

– Lack of coordination - centralization

– Long leadtime

Bullwhip EffectBullwhip Effect

Manufacturers Regional Local Local Local

Distributors Wholesalers Retailers Customers

Bullwhip effect: increased demand variability up the SC

A baby diaper (尿片）problem

P&G executives investigated their diaper SC

• number of babies in most regions of US was stable

• the usage of diaper is stable (why?)

• But their factories received orders with huge fluctuated

sizes

• going up to see their orders to 3M (plastics), even more

variable

They coined this as bullwhip effect

Prof Hau Lee of Stanford studied it and popularized it

A bullwhip (牛鞭)

牛鞭子

Bullwhip, Snowball, Ripple Effect

Example

ATO – Sales vs. PO Shipment

0

2000

4000

6000

8000

10000

12000

14000

16000

18000

20000

2003/

Jun 

2003/

Jul 

2003/

Aug 

2003/

Sep 

2003/

Oct 

2003/

Nov 

2003/

Dec

 

2004/

Jan 

2004/

Feb 

2004/

Mar

 

2004/

Apr 

2004/

May

 

2004/

Jun 

2004/

Jul 

2004/

Aug 

2004/

Sep 

2004/

Oct 

2004/

Nov 

2004/

Dec

 

Month of Withdraw

Sales: PDBDT-305 PO Shipment: PDBDT-305Sales: PRF-275-19 PO Shipment: PRF-275-19

Changes in the Value World’s Merchandise

Trade, Production and GDP, 1950-2005 (in %)

Cambell Chicken Soup

Bullwhip effect in the US PC supply chain

Semiconductor

1995 1996 1997 1998 1999 2000 2001

-40%

-20%

0%

20%

40%

60%

80%

PC

Semiconductor

Equipment

Changes in

demand

Semiconductor

1995 1996 1997 1998 1999 2000 2001

-40%

-20%

0%

20%

40%

60%

80%

PC

Semiconductor

Equipment

Changes in

demand

Annual percentage changes in demand (in $s) at three levels of the semiconductor

supply chain: personal computers, semiconductors and semiconductor manufacturing

equipment.

Barilla Case

Orders from Cortese Northeast DC to Pedrignano CDC

0

100

200

300

400

500

600

700

800

900

10001 3 5 7 9

11

13

15

17

19

21

23

25

27

29

31

33

35

37

39

41

43

45

47

49

51

Weeks

Ord

ers

(Q

uin

tals

)

Week

4L 5L 4L 5L

HP- Shipment Wholesaler-see-

thru

HP Laser: L –Series

5L

Elek Tek

Micro Electronic

PC Warehouse

Comp USA

Office

Depot

Best Buy

OfficeMax

Staple

120002000 6000

Sel

l –o

ut

Std

Dev

Order Std Dev

• Causes for Poor SC Performance

– Demand uncertainty ( how to cope with it?)

– Product variety ( -- )

– Information distortion along the SC -- bullwhip ( -- )

Safety stock

Better forecast.

Better plan.

Bullwhip EffectBullwhip Effect

Manufacturers Regional Local Local Local

Distributors Wholesalers Retailers Customers

Bullwhip effect: increased demand variability up the SC

Curses of Bullwhip EffectCurses of Bullwhip Effect

• Curses

–

–

–

X

= 350= 350

Svc Level = .95Svc Level = .95

P(Stockout) = .05P(Stockout) = .05

FrequencyFrequency

xx = ?= ?

= 10= 10

Safety Stock = Safety Stock = xx --

Curses of Bullwhip EffectCurses of Bullwhip Effect

• Curses

– Inventory out of control

– Customer service degradation

– Misguided capacity planning (W/H, staffing, etc)

R

dt

qt

dt

qt

Causes of Bullwhip EffectCauses of Bullwhip Effect

• Key causes

– Demand forecasts update (by different parties)

• “information distortion”

– Leadtimes

– Price promotion - forward buying

– Order synchronization

– Batch ordering practice

– Shortage “Gaming”

Not in the

game

Psychological effect?

Some observations

• Roughly speaking,

average in = average out

• But, the standard

deviations of “in”/”out”

can be diff .

• CV & bullwhip effect

• Order-up-to level (OUT):

– If all retailers use OUT

policy (with a constant S),

then the Stdev of the

retailer’s orders per period

equals the Stdev. Of

consumer demand per

period – no bullwhip effect

If there is no need to update

demand information at the

retailer level

Rdt

qt

Order Synchronization

• Synchronized ordering occurs when

retailers tend to order at the same time:

– end of the week orders

– beginning of the month orders

– end of the quarter orders

Cachon &

Terwiesch,

2009

Order batching

• Retailers may be required to

order in integer multiples of

some batch size, e.g., case

quantities, pallet quantities, full

truck load, etc.

• The graph shows simulated

daily consumer demand (solid

line) and supplier demand

(squares) when retailers order

in batches of 15 units, i.e.,

every 15th demand a retailer

orders one batch from the

supplier that contains 15 units. 0

10

20

30

40

50

60

70

T ime (e a c h p e rio d e q u a ls o n e d a y )

Un

its

Average number of periods between orders = Batch size/Mean demand per period

• Smaller min order quantity (lower Q), so retailers

order more frequently

• Unsynchronize retailer order intervals

– Retailers may order every T periods

– Min batch size Q=1, so no min order Q restriction

– Retailers are placed on balanced schedules s.t. average

demand per period is held constant

• e.g., 100 identical retailers and T=5 implies 20

retailers may order each period

Order batching solutions

Trade Promotion Trade Promotion

• Why trade promotion?

• Consequences of trade promotion?

Forward Buying

Hey, I am offering

a discount x%

if u will buy in

(greater)bulk

OK! Triple the

qty I usually order!

I can try and sell

some at a promotion

and keep the rest and sell

at the regular price next time

Earn & save money!

Why big orders cause problem in this case?

Trade promotions and forward buying

• Supplier gives retailer a temporary discount, called a trade promotion.

• Retailer purchases enough to satisfy demand until the next trade

promotion.

• Example: Campbell’s Chicken Noodle Soup over a one year period:

One retailer’s buy

T im e (w e e ks)

Ca

se

s

Shipm e nts

C onsum ption

0

1000

2000

3000

4000

5000

6000

7000

De

c

Ja

n

Fe

b

Ma

r

Ap

r

Ma

y

Ju

n

Ju

l

Au

g

Se

p

Oc

t

No

v

Ca

se

s

Total shipments and consumption

0

500

1000

1500

2000

2500

3000

Time (measured in weeks)

On

-han

d in

ven

tory

(u

nit

s)

Reactive and Overactive Ordering

• In reality, the retailer may not know the

demand to fluctuate around an “average”

• Each location forecasts demand to determine

shifts in the demand process

• How should a firm respond to a “high” demand

obs. – an outlier?

– Is this a signal of higher future demand or just

random variation in current demand?

• If the firm’s inventory is low, hedge by assuming this signals

higher future demand, i.e., order more than usual

• Each location forecasts demand to determine shifts in

the demand process

• How should a firm respond to a “high” demand obs?

– Is this a signal of higher future demand or just random

variation in current demand?

• If the firm’s inventory is low, hedge by assuming this signals higher

future demand, i.e., order more than usual

• How should a firm respond to a “low” demand obs?

– If the firm’s inventory is high, be more conservative and

wait to see if demand has really shifted, i.e., no order now

• Rational reactions at one level propagate up the SC

Demand forecast updating – by

Intuition

Forecast Updating - 121 SC:

An Example – Order-up-to Level

Forecast Updating - 121 SC:

An Example – Order-up-to Level

Period t t-1 t-2 t-3 t-4 t-5

Demand 64 40 45 35 40

Forecast 64 40 45 35 40

Order Upto 128 80 90 70 80

Order q 112 30 65 25

Assumptions: retailer uses Dt-1 to forecast future demand

Dt, as Ft = Dt-1; order-up-to-level 2 Ft.

No Safety stock?

What if you’re facing a stable demand pattern?

Impact of Forecasting on BEImpact of Forecasting on BE

• The BE is due, in part, to the need to forecast

demand & hold safety stock

• Moving ave and exponential smoothing are “bad”

• The fancier the method, the worse the BE

• Smoother demand forecasts can reduce the

bullwhip effect (MA & ES methods)

• The longer the leadtime, the higher the BE

• Centralised information sig reduces the BE

Forecast Updating - 121 SCForecast Updating - 121 SC

.25.21.5 )(DVar / )(qVar

:5.1 ,0),(,exampleFor

)(qVar

)(DVar

)(qVar

)]D ,(D Cov - [2 z

)D - (DVar z )(qVar

)(

:quantity ordering

z : tperiodfor and

z :1- tperiodfor Thus,

. D

:method gforecastin naive a follows and

level to-upOrder :Retailer ).,(: :DemandMkt

2

tt

21

2

2

t

t

t

22

2-t1-t

22

2-t1-t

2

t

2111

1

211

1-T

2

zDDCov

z

z

DDzDSSq

z DFS

z DF S

F

zFS

D

tt

tttttt

ttt

t-tt-

T

TT

T

RetailerRetailer

DDTT

qqT, T, L=0L=0

If all updating If all updating

their forecasts, their forecasts,

variability variability

amplifies amplifies

exponentiallyexponentially

in [in [-- , +, +]]

Assume: extra inventory

can be returned without

any cost

Gerard Cachon

(The Wharton School)

• I worked with Campbell Soup to evaluate their VMI

system in which they decided what to ship to their client

retailers. They were able to lower their retailers’

inventories by about 2/3rds and raise their fill rates at the

same time. What made that achievement remarkable was

that their system was quite simple, painfully simple –

forecast sales for the next few days based on a rolling

average of sales in the previous weeks, choose an order up-

to level that would achieve a given fill rate assuming a

reasonable level of demand volatility.

Avoiding Demand Forecast Updates

• Channel Alignment

– VMI - vendor managed inventory scheme

– Consumer direct

– Discount for information sharing, including

plan of promotion activities

• Operational Efficiency

– Leadtime reduction

– Echelon-based inventory control

Avoiding Demand Forecast Updates

• BE resulted from the chain effect along the SC

– Repetitive multiple forecast updating

• Share demand information so that every one can

obs demand shifts without distortions:

– Demand forecasts should be based on final sales to

consumers

Bullwhip can occur within a firm

Sales

We need to promote and get

rid of these green cars

Production

All green cars are sold out,

time for replenishement

Volvo Green Cars

A Bigger Colour

Problem

• Retailers submit orders for delivery in a

future period

• Supplier might not be able to fill all orders

– He might not get enough components

– His production yield might not be as high as

expected

• Phantom orders

– Reatilers order more than they think they need

to make sure they get a good allocation if

demand is high or if capacity is tight

Shortage game

• Supplier allows retailers to cancel order or accepts

returns

• High retailer profit margin, i.e., costly to not have

goods

• Retailer demand expectations positively

correlated(i.e., if one retailer has high demand

expectation, the other retailers probably do too.)

• Retailer competition (if retailer A takes more

inventory, retailer B has less to sell)

• Capacity is expensive, so the supplier will not

build unlimited cap

When is shortage game likely?

Classic Bullwhip Effect:

Semiconductor Industry, 1995

• Perception: Demand for semiconductors

would have a tremendous increase

• Result: Customers, worried about a supply

shortage, tripled their orders

• Reality: Semiconductor companies

scrambled to meet demand, realized

information was inflated and suffered huge

losses

Motorola's Inventory Problem

• Feb. 17, 1996. Announced it has excess inventory

of cellular phones

• Stock price tumbled 10%; Profit estimates down

$39 million

• Reason

– In 1994, distributors faced shortages

– In 1995, they over-ordered, early and often

• Now new-orders are down - idle capacity

Cisco’s 2002/03’s inventory problem.

• Bad:

– supplier can’t use initial orders to forecast

demand, so it builds the wrong level of capacity

– allocation among retailers is poor: some

retailers get more than they need, others are

starved

• Good:

– Reduces idle cap., assuming the retailers

actually take and sell the product

Shortage gaming: bad and good

• Don’t let retailers cancel orders

• Don’t offer retailers generous return

policies

• Share cap. And inventory data prevent false

scares

• Prioritize retailers (customers, e.g., by past

sales)

How to stop phantom ordering

Breaking Order Batches

• Making smaller batches or more frequent orders

• Information sharing

– Internet, B2B, & CAO - reducing ordering/billing

cost

• Channel Alignment -- making more frequent

– Encouraging truckload assortment

– Use of 3rd party transportation servers

– Managed/coordinated replenishment/deliveries