Bubbles in Food Commodity Markets: Four Decades of EvidenceFour Decades of Evidence

Xiaoli L. Etienne, Scott H. Irwin, andPhilip Garcia

“The Masters Hypothesis”

http://www.nytimes.com/2008/09/11/washington/11speculate.html

http://www.loe.org/images/content/080919/Act1.pdf

December 2012“The current state of knowledge indicates only a few, and weak, findings that few, and weak, findings that verify the assumption that the rise in financial speculation in recent years speculation in recent years has increased (1) the level or (2) the volatility of agricultural commodity prices……Seen in this light, the alarmism about financial the alarmism about financial speculation should be classified as a false alarm…”

FAO Index of Real Food Commodity Prices,January 1990-May 2012January 1990 May 2012

Previous Literature

• Four recent studies test for bubbles in agricultural prices– Gilbert, 2010; Phillips and Yu, 2011; Gutierrez, 2012; Etienne, Irwin,

and Garcia, 2012

– Use “right-tail ADF” bubble tests developed recently by Phillips et al.

– Detect and date-stamp bubbles by determining whether prices deviate from a random walk and become mildly explosive

– Mixed evidence on existence of bubbles since 2005

• Limitations– Use cash prices or nearby futures prices

– Both may behave in “bubble-like” fashion in a rational storage model

– Differencing nearby futures prices on roll dates may create large outliers due to contango or backwardation in term structure

Present Study

• Test whether speculative bubbles exist in 12 agricultural Test whether speculative bubbles exist in 12 agricultural futures markets

– Daily prices for individual futures contracts

O i ll i h hi h i– One contract per year, typically one with highest open interest

– Long sample: 1970-2011

– Identify whether patterns of bubble behavior exist over time

• Phillips, Shi, and Yu (2012) test– Forward and backward recursive ADF tests

– More powerful than previous tests since multiple bubbles may be detected

Price Behavior in Non-Explosive Periodsp

What Constitutes an Explosive Period?What Constitutes an Explosive Period?

Mixed Random Walk & Explosive ProcessesProcesses

How to determine when the transitions occurred?

Phillips, Shi, and Yu (2012)’s Generalized R i P d (1)Recursive Procedure (1)

Phillips, Shi, and Yu (2012)’s Generalized Recursive Procedure (2)Recursive Procedure (2)

Initial sample

Backward recursive until minimum window size

Phillips, Shi, and Yu (2012)’s Generalized R i P d (3)Recursive Procedure (3)

Futures Data

• 12 agricultural futures markets– Grains: CBOT corn, soybeans, soybean oil, wheat, and KCBOT wheat

– Softs: ICE cocoa, coffee, cotton, and sugar

– Livestock: CME live cattle, feeder cattle, and live hogs

• Data construction• Data construction– Daily futures prices for one individual futures contract per year over

1970-2011

42 i di id l t t k t– 42 individual contracts per market

– Each price sequence start 13 months before the contract expiration date and end on the last trading day of the month before the contract expiresexpires

– 240-260 daily observations for each contract

Example of Bubble Test(December 1973 corn contract)(December 1973 corn contract)

6.0 3.0 Prices, SADFs and Critical Values, Corn Dec 1973

( i ) l f

5.7 1.5

Log(price), leftSADF, rightCV95, right

5 4Pric

e)

0 0 atist

ic

5.4

Log(

P 0.0

Test

Sta

5.1 -1.5

4.8 Dec72 Jan73 Mar73 May73 Jun73 Aug73 Oct73 Nov73

-3.0

Did Bubble Frequency Increase Across Ti ? SADF T t R lt b YTime? SADF Test Results by Year

16h = 5 days

19732008

12

1414.3

12.8

19732008

8

10

8.6 8.1 7.9 7.9ith

Bub

bles

6

8

5.4 5.5

6.7 6.4

5.4 5.1 5.4 5.8

6.1 6.1 5.7

4.7

6.3 7.0

4 0

6.1

4 0

6.3

4.3

% D

ays

wi

2

4 3.6 3.0

2.2

3.5 2.9 2.8

0.9

3.8

2.7

1.2

3.6

2.2

4.0 4.0

1.6

3.8

2.2 1.9 2.1

1971 1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 20110

Year

How Frequent Were the Bubbles?SADF Test ResultsSADF Test Results

(% bubble days, h=5)

Corn Cotton Cattle Total

1970-1990

6.6 6.9 2.8 5.1

1991 3 0 5 5 3 9 4 01991-2011

3.0 5.5 3.9 4.0

1970- 4.8 6.2 3.4 4.62011

How Long Were the Bubbles?SADF Test Results (1970-1990)SADF Test Results (1970-1990)

Bubbles during 1970-1990 (h = 5 days)70

56.9Avg. Length = 1260

uenc

y 40

50

% F

requ

20

30

16.4

10.7

6.5

2 3 3.1

10

20

2.3 3.1 1.5 1.1 0.8 0.4 0.4

Bubble Length (days)5 10 15 20 25 30 35 40 45 50 55 60

0

How Long Were the Bubbles?SADF Test Results (1991-2011)SADF Test Results (1991-2011)

Bubbles during 1991-2011 (h = 5 days)70

64.3

Avg. Length = 1160

uenc

y 40

50

% F

requ

20

30

13.9

9.76.7

10

20

1.3 1.3 1.3 0.4 0.4 0.4 0.4

Bubble Length (days)5 10 15 20 25 30 35 40 45 50 55 60

0

How Big Were the Bubbles?E t St dEvent Study

4.5 Soybean Oil December 2008 Contract

3

4.0

Pric

e)

1

2

atist

ic

3.5

Log(

P

0

1

Test

Sta

-1

3.0 Nov07 Dec07 Feb08 Apr08 May08 Jul08 Aug08 Oct08

-2

Log(price), left SADF, right CV95, right

How Big Were the Bubbles?Event Study ResultsEvent Study Results

(% returns, h=5)

Positive Bubble Negative Bubble

N % Start to Peak

% Peak to End

N % Start to Peak

% Peak to End

1970- 150 +11.5 -2.8 112 -7.7 +2.019701990

150 11.5 2.8 112 7.7 2.0

1991-2011

131 +8.2 -1.8 107 -6.6 +1.72011

How Big Were the Bubbles?Event Study ResultsEvent Study Results

(% returns, h=5)

Positive Bubble Negative Bubble

N % Start to Peak

% Peak to End

N % Start to Peak

% Peak to End

1971- 58 +15.2 -3.9 24 -8.9 +2.319711976

58 15.2 3.9 24 8.9 2.3

2006-2011

57 +8.3 -2.1 28 -7.6 +1.92011

How Big Were the Positive Bubbles?Event Study Results (1970 1990)Event Study Results (1970-1990)

0 Positive Bubbles during 1970-1990, h=5

-5

-10

eak

to E

nd

Sugar 1974

-15

% R

etur

n Pe

y= 0.04 -0.25x

-20

0 10 20 30 40 50 60 70 80 90 100-25

% Return Start to Peak

How Big Were the Positive Bubbles?Event Study Results (1991 2011)Event Study Results (1991-2011)

0Positive Bubbles during 1991-2011, h=5

-5

-10

eak

to E

nd

y= 0.58 -0.29x

-15

% R

etur

n Pe

-20Coffee 1994

Cotton 2010

0 10 20 30 40 50 60 70 80 90 100-25

% Return Start to Peak

Conclusions

• All 12 agricultural markets experienced multiple bubble periodsbubble periods

• Bubble periods represent a small fraction of samples typically less than 5%samples, typically less than 5%

• Length of bubbles is relatively short, few longer than 20 daysy

• Modest tendency for over-reaction and correction during bubbles, about 3%

• Declining frequency and size of bubbles between 1970s and 2000s, with one big exception—cotton in 2010in 2010

Thank You!!Thank You!!

Were There Bubbles?GSADF Test ResultsGSADF Test Results

(# of commodity-years w/bubbles)

10% Sig. 5% Sig. 1% Sig. Total

1970-1990

13 31 50 94

1991-2011

13 27 33 73

1970 26 54 83 1631970-2011

26 54 83 163

Did Bubbles Cluster Across Time?SADF Test Results (top 10 years)SADF Test Results (top 10 years)

20 Top 10 Years, h = 5 days

16

18

14 3

12

1414.3

12.8

h Bu

bble

s

6

8

10 8.6

8.1 7.9 7.9 7.0 6.7 6.4 6.3%

Day

s with

2

4

1973 2008 1974 1981 2010 1994 1999 1976 1977 20030

Year

Did Bubbles Cluster Across Time?SADF Test Results (bottom 10 years)SADF Test Results (bottom 10 years)

20 Bottom 10 Years, h = 5 days

16

18

10

12

14

th B

ubbl

es

6

8

10

% D

ays w

it

2

42.8 2.7

2.2 2.2 2.2 1.9 2.11.6 1.2 0.9

1990 1993 1984 2007 1997 2009 2011 2005 1995 19910

Year

How Big Were the Negative Bubbles?Event Study Results (1970 1990)Event Study Results (1970-1990)

15 Negative Bubbles during 1970-1990, h=5

y= 0.13 -0.24x

10

ough

to E

nd

Sugar 1982

5

% R

etur

n Tr

o

-50 -45 -40 -35 -30 -25 -20 -15 -10 -5 00

% Return Start to Trough

How Big Were the Negative Bubbles?Event Study Results (1991 2011)Event Study Results (1991-2011)

15 Negative Bubbles during 1991-2011

y= -0.03 -0.27x

10

ough

to E

nd

5

% R

etur

n Tr

o

-50 -45 -40 -35 -30 -25 -20 -15 -10 -5 00

% Return Start to Trough