In search of the bullwhip effect
Gérard P. Cachon Taylor Randall Glen M. Schmidt
The Wharton School David Eccles School of Business McDonough School of Business
University of Pennsylvania University of Utah Georgetown University
Presented
September 9, 2006
Northwestern University
In search of the bullwhip effect: Cachon, Slide 1
Randall, Schmidt 9/06
The bullwhip effect
Demand variability increases as you move up the supply chain from
the customer towards supply
Equipment Tier 1 Supplier Factory Distributor Retailer Customer
In search of the bullwhip effect: Cachon, Slide 2
Randall, Schmidt 9/06
Campbell’s Chicken Noodle Soup
Shipments
Cases
Consumption
One retailer’s buy
7000
6000
Time (weeks) 5000
4000
Cases
3000
2000
1000
Jul
Jan
Feb
May
Jun
Aug
Sep
Oct
Nov
Mar
Apr
Dec
In search of the bullwhip effect: Cachon, Slide 3
Randall, Schmidt 9/06
The bullwhip at Barilla pasta
Upstream variability at CDC is
much higher
Downstream variability at
DC: mean demand is about
300, the standard deviation
is about 75
In search of the bullwhip effect: Cachon, Slide 4
Randall, Schmidt 9/06
In search of the bullwhip effect: Cachon, Slide 5
Randall, Schmidt 9/06
In search of the bullwhip effect: Cachon, Slide 6
Randall, Schmidt 9/06
Autos to machine tools
Autos Machine tools
80%
60%
40%
20%
% change in demand
0%
-20%
-40%
-60% GDP = solid line
-80%
Source:Anderson, Fine and Parker (1996)
In search of the bullwhip effect: Cachon, Slide 7
Randall, Schmidt 9/06
U.S. PC industry
Changes in
demand
80%
60%
Semiconductor
40% Equipment
20% PC
0%
-20% Semiconductor
-40%
1995 1996 1997 1998 1999 2000 2001
Annual percentage changes in demand (in $s) at three levels of the semiconductor
supply chain: personal computers, semiconductors and semiconductor manufacturing
equipment.
In search of the bullwhip effect: Cachon, Slide 8
Randall, Schmidt 9/06
Explanations for the bullwhip effect …
Fixed cost to produce/order, (s,S) models, order synchronization:
Blinder (1981), Caplin (1985), Caballero and Engel (1999), Mosser (1991), Lee, Padmanabhan
and Whang (LPW) (1997), Cachon (1999)
Positive serial correlation of demand shocks: Kahn (1987), LPW (1997),
Graves (1999), Chen, Drezner, Ryan, Simchi-Levi (2000).
Price fluctuations/cost shocks: LPW (1997)
Non-convex production: Ramey (1991)
Demand can be backlogged: Kahn (1987)
Shortage gaming: LPW (1997), Cachon and Lariviere (1997)
Misperception of feedback/irrational behavior: Sterman (1989)
In search of the bullwhip effect: Cachon, Slide 9
Randall, Schmidt 9/06
Empirical evidence of production smoothing
Blinder and Maccini (91,92):
⎯ Data: 1959-1986, monthly, seasonally adjusted, constant 1982 dollars
⎯ Production is more variable than sales in 17 of 20 two-digit manufacturing industries
⎯ “… the basic facts to be explained are … 1) production is more variable than sales in most
industries”.
Blanchard (1983):
⎯ “… in the automobile industry, inventory behavior is destabilizing: the variance of production
is larger than the variance of sales.”
Miron and Zeldes (1988):
⎯ “…The overall assessment of this model … is quite negative: there is little evidence that
manufacturers hold inventories of finished goods … to smooth production.”
Eichenbaum (1989):
⎯ “We find overwhelming evidence against the production-level smoothing model … we
conclude that the variance of production exceeds the variance of sales in most
manufacturing industries.”
Other negative results:
⎯ West (1986), Ramey (1991), Mosser (1991), Kahn (1992)
In search of the bullwhip effect: Cachon, Slide 10
Randall, Schmidt 9/06
A measure of the bullwhip effect: the amplification ratio
Demand, D
Orders Firm
Production Inventory
Sales
(inflow), Y
(outflow), S
Amplification ratio = V[Y] / V[D].
If demand is not available, use sales as a proxy for demand.
We say the bullwhip effect is present in an industry if its amplification
ratio is greater than 1.
In search of the bullwhip effect: Cachon, Slide 11
Randall, Schmidt 9/06
Our data
Sources:
⎯ Census Department, Bureau of Economic Analysis.
Data:
⎯ U.S., 1992-2006, monthly.
⎯ 50 manufacturing industries: Sales, inventory.
• In a subset of 23 manufacturing industries: Demand.
⎯ 16 wholesale industries: Sales, inventory.
⎯ 6 retail industries: Sales, inventory.
Data manipulations:
1) Adjust Demand and Sales series for margins and price.
2) Adjust Inventory series for price.
3) For each industry evaluate a Production series: Yt = St + ∆ It
4) Log and first difference the Production, Demand and Sales series.
In search of the bullwhip effect: Cachon, Slide 12
Randall, Schmidt 9/06
General merchandise stores – margin and price adjusted
55,000
50,000
Production Sales
45,000
40,000
35,000
30,000
25,000
20,000
15,000
10,000
Jan-92 Jan-94 Jan-96 Jan-98 Jan-00 Jan-02 Jan-04 Jan-06
In search of the bullwhip effect: Cachon, Slide 13
Randall, Schmidt 9/06
General merchandise stores – margin and price adjusted
plus logged and first differenced
0.6
Production Sales
0.4
0.2
-0.2
-0.4
-0.6
-0.8
-1
Jan-92 Jan-94 Jan-96 Jan-98 Jan-00 Jan-02 Jan-04 Jan-06
In search of the bullwhip effect: Cachon, Slide 14
Randall, Schmidt 9/06
Telecom – margin and price adjusted
9,000
8,000
Production Demand
7,000
6,000
5,000
4,000
3,000
2,000
1,000
Jan-92 Jan-94 Jan-96 Jan-98 Jan-00 Jan-02 Jan-04 Jan-06
In search of the bullwhip effect: Cachon, Slide 15
Randall, Schmidt 9/06
Telecom – margin and price adjusted plus logged and first
differenced
0.8
Production Demand
0.6
0.4
0.2
0
c
-0.2
-0.4
-0.6
-0.8
Jan-92 Jan-94 Jan-96 Jan-98 Jan-00 Jan-02 Jan-04 Jan-06
In search of the bullwhip effect: Cachon, Slide 16
Randall, Schmidt 9/06
Research questions
To what extent does the bullwhip effect exist in U.S. industry level
data?
⎯ Are amplification ratios greater than 1?
⎯ Do manufacturers experience the highest demand variability and
retailers the lowest?
Understand variation in the amplification ratios:
⎯ What explains variation in the amplification ratio across industries?
⎯ Have amplification ratios been decreasing over time?
In search of the bullwhip effect: Cachon, Slide 17
Randall, Schmidt 9/06
Prevalence of the bullwhip effect
Aggregate series Amplification Ratio
Retail 0.50
Wholesale 1.14
Manufacturing 0.55
Percentage of industries that exhibit
the bullwhip effect
Seasonally Seasonally
unadjusted adjusted
Retail 16% (1 of 6) 100% (6 of 6)
Wholesale 88% (14 of 16) 100% (16 of 16)
Manufacturing 40% (20 of 50) 74% (37 of 50)
In search of the bullwhip effect: Cachon, Slide 18
Randall, Schmidt 9/06
Demand variability at different levels of the supply chain
0.050
0.045
0.040
Variance of demand
0.035
0.030
0.025
0.020
0.015
0.010
0.005
0.000
Retail Wholesale Manufacturing
In search of the bullwhip effect: Cachon, Slide 19
Randall, Schmidt 9/06
Trends in amplification ratios
1.9
1.7
Amplification Ratio
1.5
1.3 All Industries
Retailers
1.1 Wholesalers
0.9 Manufacturers
0.7
0.5
1 2 3 4
Time Period
1=1992-95, 2=1996-98, 3=1999-2001, 4=2002-05
In search of the bullwhip effect: Cachon, Slide 20
Randall, Schmidt 9/06
Future research
Investigate the bullwhip effect at different levels of aggregation –
firm, category, sku.
Investigate the bullwhip effect at different levels of time aggregation
– daily, weekly, quarterly.
Obtain better order and demand data.
Do firms/supply chains that better manage the bullwhip effect
perform better financially?
In search of the bullwhip effect: Cachon, Slide 21
Randall, Schmidt 9/06
