Agricultural Commodity Risk

Management
IE 441 Final Report
 Team 10

Yuan Zong, Feng Wang and Shengtao Li

2
Executive Summary

Since 2007, agricultural commodity markets, especially grains, have experienced a

number of price surges. In June 2008, December 2010 and October 2012, food prices

increased sharply and then declined dramatically, only to remain at relatively high levels,

as compared with those five years ago. Thus, there is high concern that we may be

experiencing a period of commodity bubble in these markets.

In this paper, we discuss whether bubbles are present in agricultural commodity

markets in the United States. Then we identify possible bubble episodes and look at

possible reasons behind these price surges. We apply the generalized sup augmented

Dickey-Fuller (GSADF) test (Phillips, Shi and Yu, 2012) to weekly prices from 16

agricultural futures contracts to test whether speculative bubbles are present in these

markets and identify whether patterns of bubble behavior exist over a very long sample

period. The tests indicate at the 5% significance level, bubbles are present in five

markets: CBOT soybean and soybean oil, KCBOT wheat, and ICE cotton and sugar.

In the advanced analysis part, we try to answer the questions related to the drivers of

these explosive behaviors. By applying the Granger Causality Test, we successfully

detect the presence of casual links between the index investment positions and the futures

prices in all five markets. However, in soybean, soybean oil, wheat and sugar markets, we

can conclude at the 5% significance level that futures price changes Granger-cause index

investment, whereas in the cotton market, we can conclude at the 5% significance level

that index investment Granger-causes futures price changes.

3
Contents
Literature Review..............................................................................................................................1
Definition and Mechanism..............................................................................................................1
Debate.............................................................................................................................................2
Preliminary Analysis..........................................................................................................................3
Theory.............................................................................................................................................3
Testing and Date-Stamping Procedures..........................................................................................4
The Standard (Augmented) Dickey-Fuller Test................................................................................4
The sup Augmented Dickey-Fuller Test...........................................................................................5
The Generalized Augmented Dickey-Fuller Test..............................................................................6
Data.................................................................................................................................................9
Results and Analysis........................................................................................................................9
Advanced Analysis..........................................................................................................................14
Granger Causality Test...................................................................................................................15
Data...............................................................................................................................................15
Results and Analysis......................................................................................................................16
Appendix.........................................................................................................................................20
References......................................................................................................................................22

4
List of Tables
Table 1 Test Results.......................................................................................................................10
Table 2 Bubble periods for the five markets..................................................................................13
Table 3 Value of lag order p...........................................................................................................16
Table 4 The Granger Causality Test results....................................................................................17
Table 5 Granger Causality Test results (Soybean Oil market, after 2006)......................................18
Table 6 Granger Causality Test result (Cotton market, non-bubble-existing period).....................19
Table 7 Critical values at 90% level, by lags 0 to 12 and number of observations 100 to 500......20
Table 8 Critical values at 95% level, by lags 0 to 12 and number of observations 100 to 500......20
Table 9 Critical values at 99% level, by lags 0 to 12 and number of observations 100 to 500......21

List of Figures
Figure 1 Recursive ADF Statistics, CBOT Wheat Contract..............................................................11
Figure 2 Recursive ADF Statistics, CBOT Soybean, KCBOT Wheat and ICE Cotton Contracts.........11
Figure 3 Recursive ADF Statistics, CBOT Soybean Oil and ICE Sugar Contracts.............................12

5
Literature Review

Since 2007, agricultural commodity markets, especially grains, have experienced a

number of price surges. In June 2008, December 2010 and October 2012, food prices

increased sharply and then declined dramatically, only to remain at relatively high

levels, as compared with those five years ago. Thus, there is high concern that we may

be experiencing a period of commodity bubble in these markets.

In this paper, we discuss whether bubbles are present in agricultural commodity

markets in the United States. We will identify possible bubble episodes and also

investigate possible reasons behind these price surges.

Definition and Mechanism

An economic bubble, sometimes referred to as a speculative bubble, is described as

“trade in high volumes at prices that are considerably at variance with intrinsic

values”. (King et al, 1993) It is an unsustainable situation where asset prices are based

on implausible or inconsistent views about the future. (Krugman, 2013) At a high

level, when a bubble is present, asset prices rise at an increasingly fast rate until

eventually the bubble bursts and the prices collapse back to their fundamental values.

Stiglitz (1990) presents a straightforward way to define a bubble: if a price is high

today only because investors believe that the price will be high tomorrow – when

fundamental factors do not seem to justify such a price – then a bubble exists.

Examples include the dot-com bubble in the 1998-2000, and the Dutch Tulip mania in

1
the 1636-37.

Both the boom and burst phases of a bubble are driven by a positive feedback

mechanism. Under normal market circumstances, a negative feedback mechanism

adjusts the supply and demand forces and then determines the equilibrium price.

However, in a bubble, price momentum is self-enhanced and it becomes impossible to

predict prices based on the supply and demand relationship alone. Shiller (2012)

describes this phenomenon as follows: “A speculative bubble is a social epidemic

whose contagion is mediated by price movements. News of price increase enriches

the early investors, creating word-of-mouth stories about their successes, which stir

envy and interest. The excitement then lures more and more people into the market,

which causes prices to increase further, attracting yet more people and fueling 'new

era' stories, and so on, in successive feedback loops as the bubble grows.” The same

mechanism applies to the burst phase of a bubble.

Debate

A number of economists deny the existence of bubbles. They argue that the

fundamental forces of supply and demand are the root causes of the food prices surges

over the past six years. Due to growing population and increasing income, emerging

countries see a significant growth of demand for food in recent years, while growth

rate of supply cannot keep up with demand. (Alston et al, 2010; Bioversity et al,

2012) This alone exerts great amount of pressure on commodity prices. The growing

demand for food and feed crops for the production of biofuels is another important

driver. (Serra et al, 2010; Balcombe and Rapsomanikis, 2008) A strong indicator of

the increasing demand is the decline of aggregate grain stocks relative to utilization.

2
Indeed, the global grain market stocks-to-utilization ratio has been fluctuating at a low

point since 2005-06, and even small supply and demand shocks can generate wide

price variations. (Wright, 2010)

Even among the scholars who agree that bubbles do exist, i.e., asset prices often

deviate strongly from their true values, the cause of bubbles remains hotly disputed.

Many explanations have been suggested. For example, some researchers claim it is

excessive monetary liquidity combined with speculative activities that causes bubbles.

According to this point of view, a large amount of new money, which partly comes

from excessive liquidity, chases too few assets and finally finds its way into

commodity markets through index funds, a process that in turn causes speculative

bubbles. Speculators are also to blame in this perspective, because speculation, both

rational and irrational, can create self-enforced price trends by herd behavior and

result in bubbles. Moreover, noise trading increases market risk that deters rational

investors from betting against mispricings. Other researchers attempt to explain

bubbles through social psychology. They have devised several popular theories, such

as greater fool theory, extrapolation, herding and moral hazard. Robles et al (2009)

stress that, along with market fundamentals, rising expectations, speculation,

hoarding, and hysteria played a significant role in the increasing level and volatility of

food prices and attribute the 2008 food price increase partly to bubbles.

Preliminary Analysis

In this paper, we apply the generalized sup augmented Dickey-Fuller (GSADF) test

(Phillips, Shi and Yu, 2012) to weekly prices from 16 agricultural futures contracts to

3
test whether speculative bubbles are present in these markets and identify whether

patterns of bubble behavior exist over a very long sample period.

Theory

Within the rational expectation framework, the present value model is widely used to

define the fundamental price by discounting expected future cash flows that will

accrue to the owner of the asset. Pindyck (1993) draws an analogy with the rational

pricing of assets and states that for a storable commodity, the stream of payoffs is the

convenience yield accruing to the owner of the inventory in terms of benefits related

to the facilitation of processing, sales and the avoidance stock-outs. Then the

equilibrium commodity price has two components: the market fundamental

component, which is defined as the present value of expected future convenience

yields, and the bubble component, which is the expected capital gains or discounted

resale value. Several economists have argued that bubbles may exist. (Stiglitz, 1990;

Wang and Wen, 2012) Thus, under certain conditions, any solution for the commodity

price can be written as: 𝑃t= Ft + 𝐵t, where 𝐹t is the fundamental component and 𝐵t

is the bubble component. If 𝐵t ≠ 0, current prices will show explosive behavior

pattern, as 𝐵t obeys a submartingale process, that is, a stochastic process in which the

expected value of next period's value, based on the current period's information, is

greater than the current period's value.

Testing and Date-Stamping Procedures

The Standard (Augmented) Dickey-Fuller Test

Given the different stochastic properties of the fundamental component and the

4
bubble component, early tests for bubbles were based on the standard Dickey-Fuller

test, estimating the parameters of the following equation:

Δ y t=α + β y t−1 +ε t (1)

where y t is the logged price of the commodity at timet ; and α , β are parameters to be

estimated. The extended form – the Augmented Dickey-Fuller test – is more often

used than the standard Dickey-Fuller test, by estimating the parameters of the

following equation:

(2)
k
Δ y t=α + β y t−1 +∑ φi Δ y t −i+ ε t
i=1

where y t is the logged price of the commodity at timet ; and α , β , φ are parameters to

be estimated.

These tests are carried out under the null hypothesis H 0 : β=0 against the

alternative hypothesis H 1 : β< 0. Once the test statistic value

ADF= ^β /se ( β^ ) (3)

is computed, it can be compared with the relevant critical value. If the test statistic is

less than the (negative) critical value, then the null hypothesis is rejected and no unit

root is present, i.e., the time series data is stationary). Otherwise, there is a unit root,

i.e., the time series is non-stationary).

However, this method was criticized for failing to make a distinction between a

unit root or a stationary auto-regression and a process that exhibits periodically

collapsing bubble-like behavior. (Diba and Grossman, 1988; Evans, 1991)

5
The sup Augmented Dickey-Fuller Test
The recursive tests proposed by Phillips, Wu and Yu (2011, PWY hereinafter) are free

from the criticism mentioned above. This methodology is based on repeated

applications of the augmented Dickey-Fuller (ADF) test, estimating through a

recursive regression the parameters in the following equation:

(4)
k
Δ y t=α r + βr y t−1 +∑ φi , r Δ y t −i+ ε t
w w w
i=1

where y t is the logged price of the commodity at time 𝑡; α , β , φ are parameters to be

estimated; r w is the variable sample window size; and 𝑘 is the lag order.

The recursive regression involves the estimation of (3) by least squares starting

with r w =r 0 fraction of the whole sample, and expanding the sample forward

repeatedly, with the last regression using the full sample [1, T ].

PWY test the null hypothesis of a unit root ( H 0 : β=0) against the right-tailed

alternative hypothesis of explosive behavior ( H 1 : β> 0) with the sup ADF (SADF), by

comparing the test statistic value

(5)
¿
SADF= r w ∈[r 0 ,T ]{ ADF r } w

with the relevant critical value. If the test statistic is greater than the critical value,

then the null hypothesis is rejected and bubble-like behavior is present, i.e., the time

series data is explosive. Otherwise, there is no evidence of bubbles, i.e., the time

series is stationary or a random walk..

Unfortunately, the SADF method has its weaknesses, too. First, when there are

multiple bubbles occurring within the whole sample period, this method may fail to

detect the existence of bubbles. Second, SADF is sensitive to the starting point r 0 .

6
The Generalized Augmented Dickey-Fuller Test
Given the weaknesses of the SADF test, Phillips, Shi and Yu (2012, PSY hereinafter)

improve it by introducing a moving window instead of a fixed one and allowing for

the possibility of periodically collapsing bubbles. This new method is called the

generalized version of the sup augmented Dickey-Fuller (GSADF) test.

To identify and date-stamp multiple bubble periods accurately, the testing

procedure must first distinguish the explosive behavior of a price series at t 1 e from its

non-explosive behavior at t 1e-1. Similarly, at t 1 f the testing procedure must be capable

of identifying the transition from an explosive path to a random walk. To achieve this,

PSY use variable window widths in the recursive regressions.

For the estimation starting at r 1and ending at r 2, the Augmented Dickey-Fuller

regression model is:

k
(6)
Δ y t=α rr + β rr y t −1 + ∑ φri ,r Δ y t −i+ ε t
2

1
2

1
2

1
i=1

where 𝑦t is the logged price of the commodity at time 𝑡; 𝛼, 𝛽, 𝜑 are parameters to be

estimated; and 𝑘 is the lag order.

The ADF statistic obtained is denoted as ADF rr 21. Defining r 0 as the minimum

window size required to estimate Equation (6) and a fixed starting point r 1=1 , the

ending point r 2 can vary from r 0 to T, resulting in T −r 0 +1 ADF statistics for the

fixed starting point r 1=1 and

¿ r2 (7)
SADF r 1= r 2 ∈[r 0 ,T ]{ ADF r 1 }

Then let r 1 vary from 1 to r 2−r 0 +1 and

7
 r2
GSADF(rw0)= ¿ ADF r 1 (8)
r 2 ∈ [ r0 , T ]
r1 ∈[1 ,r 2−r0+1 ]

Following the GSADF approach, PSY test the null hypothesis of non-stationarity
r2 r2
H 0 : β r =0 against the alternative hypothesis H 1 : βr > 0which implies bubble-like
1 1

behavior. Significant GSADF test statistics indicate bubble(s) exist in the sample

period.

PSY also carry out a new method to date the starting and ending dates of the

bubbles. The first explosive episode in the sample is specified by [r 1 e , r 1 f ] , and

ρ
r 1 e = inf {r 2 : SAD F r 2 >c v r 2 } (9)
r2 ∈[r 0 ,T ]

ρ
r1 f = inf {r 2 :SAD F r 2< c v r 2 } (10)
r2 ∈[r1 e +h , T ]

where c v rρ2 is the100 ρ % critical values and h is the minimium defined length of the

bubble episode. In essence, the bubble emerges with a sign that statistic value

SAD F r 2 exceeds the corresponding critical value and terminates as soon as SAD F r 2

falls below the critical value. The estimated starting and collapse dates of the second

explosive episode are:

ρ
r 2 e = inf {r 2 : SAD F r 2 > c v r 2 } (11)
r2 ∈[r 1f ,T ]

ρ
r2 f = inf {r 2 : SAD F r 2 <c v r 2 } (12)
r2 ∈[r 2e +h , T ]

In this way, PSY date-stamp all bubbles in the sample period.

We need to make a few comments before carrying out our replication. First, the

lag order k in Equation (6) must be specified. Phillips et al (2013b) contend that size

distortions actually are smallest when a fixed lag length is used. Consequently, we use

a fixed lag order of one (k =1) to allow for the possible low order autocorrelation that

8
has been observed in weekly agricultural futures returns. (Taylor, 1986; Yang and

Brorsen, 1994; Isengildina et al, 2006) Second, the minimum window size r 0 must be

chosen so that the chance of finding explosive periods is maximized and, at the same

time, there are sufficient observations for reliable estimations. In our test, the initial

startup sample contains 40 observations, or roughly one year (r 0 =40). Third, when

defining the end dates of the explosive periods in equations (9)-(12), price

explosiveness needs to last at least h periods to be considered economically

meaningful. Phillips et al (2011) suggest that the minimum length of the explosive

period be set to log ( T ). In our case, with sample size of 537 weeks, this results in a

minimum length of about 6-7 weeks (h=6). Intuitively, in a competitive futures

market it is reasonable to assume that new information is reflected quickly in futures

prices, even if not instantaneously, and market participants react rapidly. Hence, price

movements away from fair prices based on fundamentals are likely to be short-lived.

Data

As noted, explosive periods identified by previous work that relies on cash prices

(Phillips and Yu, 2011) or rolling nearby futures prices (Gilbert, 2010; Etienne et al,

2012; Gutierrez, 2013) may be driven by fundamental demand and supply factors

rather than an explosive bubble component. In contrast, futures prices should behave

as a random walk under fairly general conditions. Specifically, the futures price at

time t for a contract maturing at T is the expected cash price of a certain commodity at

time T conditional on the information available at time t (e.g., Fama and French,

1987; Tomek, 1997), assuming rational expectations, no risk premium, and no basis

risk. Hence, prices from futures contracts will behave approximately as a sequence of

9
expected cash prices at maturity and follow a random walk. (Peterson and Tomek,

2005) Deviations from a random walk in the series of futures prices may thus provide

more reliable evidence of bubble components in agricultural commodity prices.

We consider 16 agricultural commodity futures contracts, including five grain

futures (Chicago Board of Trade (CBOT) corn, soybeans, soybean oil and wheat;

Kansas City Board of Trade (KCBOT) wheat), three livestock futures (Chicago

Mercantile Exchange (CME) feeder cattle, live cattle, and lean hogs), and eight soft

futures (Intercontinental Commodity Exchange (ICE) cocoa, coffee, cotton, orange

juice and sugar; London International Financial Futures and Options Exchange

(LIFFE) cocoa, coffee and sugar). We use weekly prices from July 18, 2003 to

October 22, 2013. All data come from Bloomberg.

Results and Analysis

Phillips, Shi and Yu (2012) provide critical values of GSADF tests against a bubble-

like behavior using Monte Carlo simulations, with a lag order of zero. We obtained

the asymptotic 90, 95 and 99% critical values for lag order of one (k =1)using the

same method. The critical values obtained are shown in Appendix.

Table 1 shows our test results by applying the GSADF approach to agricultural

commodity markets. GSADF statistics greater than the critical value show evidence of

bubbles. The tests indicate that:

 at the 10% significance level, bubbles are present in six markets: CBOT soybean,

soybean oil and wheat, KCBOT wheat, and ICE cotton and sugar;

 at the 5% significance level, bubbles are present in five markets: CBOT soybean

10
 and soybean oil, KCBOT wheat, and ICE cotton and sugar;

 at the 1% significance level, bubbles are present in two markets: CBOT soybean

oil and ICE sugar.

Table 1 Test Results

Futures Contract GSADF
CBOT corn 1.44
CBOT soybean 2.69 **
CBOT soybean oil 3.09 ***
CBOT wheat 2.42 *
KCBO wheat 2.66 **
T
CME live cattle 0.92
CME lean hog 0.72
CME feeder cattle 1.17
ICE cocoa 1.41
ICE coffee 1.22
ICE cotton 2.75 **
ICE orange juice 0.91
ICE sugar 3.39 ***
LIFFE cocoa 2.07
LIFFE coffee 1.40
LIFFE sugar 1.63

In order to date-stamp the starting and ending dates of the bubbles, we plot the

relevant recursive ADF statistics (Figures 1-3).

11
 3.00

2.00

1.00

0.00
2002/9/1 2005/5/28 2008/2/22 2010/11/18 2013/8/14 W1
-1.00

-2.00

-3.00

-4.00

Figure 1 Recursive ADF Statistics, CBOT Wheat Contract

4.00

3.00

2.00

1.00

0.00 S1
2002/9/1 2005/5/28 2008/2/22 2010/11/18 2013/8/14 KW1
-1.00 CT1

-2.00

-3.00

-4.00

-5.00

Figure 2 Recursive ADF Statistics, CBOT Soybean, KCBOT Wheat and ICE Cotton
Contracts

12
 4.00

3.00

2.00

1.00
BO1
SB1
0.00
2002/9/1 2005/5/28 2008/2/22 2010/11/18 2013/8/14
-1.00

-2.00

-3.00

Figure 3 Recursive ADF Statistics, CBOT Soybean Oil and ICE Sugar Contracts

Our date-stamping results show:

 at the 1% significance level, none of the bubbles lasts longer than 6 weeks.

 at the 5% significance level, eight bubble episodes are CBOT soybean from

2006-3-10 to 2006-6-23; CBOT soybean oil from 2005-11-18 to 2006-8-11, from

2007-1-12 to 2007-2-23, from 2007-3-23 to 2007-4-27, and from 2007-7-20 to

2007-10-12; KCBOT wheat from 2007-2-9 to 2007-4-20, and 2007-7-6 to 2007-

8-17; ICE sugar from 2005-6-24 to 2005-11-04.

 at the 10% significance level, ten bubble episodes are CBOT soybean from 2005-

11-25 to 2006-11-24; CBOT soybean oil from 2005-4-22 to 2005-9-30, from

2005-11-18 to 2006-8-25, from 2006-10-20 to 2006-12-8, from 2007-1-12 to

2007-5-18, from 2007-7-13 to 2007-11-16, and from 2007-11-30 to 2008-2-29;

KCBOT wheat from 2007-2-2 to 2007-9-28; ICE cotton from 2010-8-13 to

2010-10-1; ICE sugar from 2005-6-24 to 2005-12-30.

13
Our analysis provides the following findings:

1. The general bubble periods for the five markets are

Table 2 Bubble periods for the five markets

Market From To
CBOT soybean 2005-10-21 2006-10-20
CBOT soybean oil 2005-3-18 2008-1-25
KCBOT wheat 2006-12-29 2007-8-24
ICE cotton 2010-6-4 2010-10-8
ICE sugar 2005-2-18 2005-12-16

and there are three periods in our sample from 2003 to 2013 that are free from

bubbles: before February 2005, between February 2008 and May 2010, and after

October 2010. This may suggest that since October 2010, agricultural commodity

prices have basically returned to their fundamental values.

2. Considering the five bubble-existing markets, a cotton bubble was present

during 2010, whereas bubbles in the other four markets existed primarily

before 2008. Moreover, the cotton bubble period is the shortest among all five

markets. This may suggest that cotton bubbles are driven by different factors,

or the cotton bubble test results are less reliable and worth investigating

further.

3. Among all 16 futures contracts, two out of five grain futures and two out of

eight soft futures show evidence of bubble-like behavior, while none of the

three livestock futures suggests the existence of a bubble. This indicates that

livestock commodities are less exposed to speculative bubbles, possibly

because they are much more difficult to store and transfer.

4. Interestingly, futures contracts of different exchanges may differ in test results.

We can see, although both ICE coffee/cocoa and LIFFE coffee/cocoa are free

14
 from bubble-like behavior, that the analysis of CBOT and KCBOT data

suggests different answers on the wheat bubble tests1, and ICE and LIFFE

hold different opinions about the sugar bubble issue as well. The reason may

lies in the difference of the underlying commodities and of the exchange

mechanisms.

Advanced Analysis

Our analysis so far gives answers to the questions whether bubbles exist in

agricultural commodity markets and when these bubbles are present. However, the

test results may need further investigation and verification, which require answering

questions related to the drivers of these explosive behaviors.

A popular belief is that long-only index investment drives these prices up

high. This argument is also named as the ‘Masters Hypothesis’ because the investment

manager Michael Masters claimed it in testimony before a Senate committee and on

the TV program 60 minutes. According to the ’Masters Hypothesis’, long-only index

investment was a major driver of the 2007-2008 spike in commodity futures prices

and energy futures prices in particular. The rationale behind this argument is when

commodity futures markets are not liquid enough, large flow of index funds can cause

a temporary price deviation from the fundamental value. Also, index investors are

noise traders and make arbitrage risky, which opens the possibility of index fund

traders’ creating their own space if their positions are large enough. (De Long et al.,

1990) If the Masters Hypothesis is true, a positive correlation will be detected

1
Although the GSADF test indicates there is bubble in the CBOT wheat market during the sample period, the
bubble is too short-lived (only lasts one week) to make sense economically.
15
between the changes of index position and the changes of futures price.

In what follows, we apply the Granger Causality Test to the five bubble-existing

markets and test whether the empirical results offer some support for the Masters

Hypothesis in these markets.

Granger Causality Test

We can carry out the Granger Causality Test by estimating the parameters of:

p p
(13)
Δln f t=α 1 + ∑ β 1 j Δln f t − j + ∑ γ 1 j Δ x t − j +ϵ 1 t , t=1, … , T
j=1 j=1

where is the weekly return on agricultural futures contract on day and

is changes in positions on day . The null hypothesis is H 0 :γ 11=…=γ 1 p=0 and

the alternative hypothesis is H 1 : γ 11 ≠ 0∨…∨γ 1 p ≠ 0. If we reject the null hypothesis,

empirical evidence shows the long-only index investment Granger-causes the futures

price changes. At the same time, we estimate:

p p
(14)
Δ x t =α 2+ ∑ β 2 j Δln f t− j+ ∑ γ 2 j Δ x t− j+ ϵ 2t ,t=1 , … , T
j =1 j=1

The null hypothesis is H 0 :γ 21=…=γ 2 p =0 and the alternative hypothesis is

H 1 : γ 21 ≠ 0∨…∨γ 2 p ≠ 0 . If we reject the null hypothesis, there is empirical evidence

that shows the price changes Granger-causes the long-only index investment.

However, if these two test results are both significant, then the causal links may be

16
fake and another factor may be causing both the price changes and the index

investment position changes.

Data

To carry out the Granger Causality Test, we need index position data of the five

bubble-existing markets. Since 2006, the US Commodity Futures Trading

Commission (CFTC) began releasing index trader information in the Supplemental

Commitments of Traders (SCOT) reports. However, we have no access to the

accurate data of index position before 2006. In this paper, we use CFTC

noncommercial position as a proxy for the index position for year 2003-2005. The

CFTC usually categorizes trader positions as commercial and non-commercial.

Noncommercial positions mean that investors do not have commercial interest in the

underlying physical commodity, thus those positions are largely speculative positions.

Commercial positions refer to investors who have commercial interest in the

underlying physical commodities and are largely hedging positions. We can use

noncommercial data as a proxy for index investment positions because index

investments are mainly driven by noncommercial reasons.

Results and Analysis

We first need to determine the lag order in the Granger Causality Test model before

we carry out the test. Here we apply AIC and SC criteria in choosing the best p. The

table below shows the lag order p that gives the smallest AIC/SC value.

Table 3 Value of lag order p

Market Lags 1 2 3 4 5 6

17
 AIC -6.5351 -6.5853 -6.5939 -6.5638 -6.5396 -6.5274
Soybean
SC -6.3099 -6.2065 -6.0585 -5.8688 -5.6820 -5.5039
Soybean AIC -5.7041 -5.6606 -5.6361 -5.6274 -5.5794 -5.5262
Oil SC -5.5831 -5.4581 -5.3513 -5.2596 -5.1278 -4.9900
AIC -6.1377 -5.9735 -5.8187 -5.7249 -5.7452 -5.6837
Wheat
SC -5.8683 -5.5200 -5.1774 -4.8923 -4.7177 -4.4579
AIC -5.0587 -4.7378 -4.8116 -6.5179 - -
Cotton
SC -4.7619 -4.2477 -4.1356 -5.6683 - -
AIC -6.8104 -6.6481 -6.4938 -6.5216 -6.5041 -6.7530
Sugar
SC -6.5647 -6.2344 -5.9087 -5.7616 -5.5657 -5.6325

For soybean oil, wheat and sugar markets, p=1 . For soybean market, p=3

according to the AIC rule and p=1 according to the SC rule. We will input both in the

Granger Causality Test. For cotton market, the lag order should be no larger than 4 to

guarantee enough observations in the test because the bubble only lasted for 19

weeks. Thus we choose p=4 .

Then, we can apply the Granger Causality Test to the five bubble-existing

markets. The test results are shown below.

Table 4 The Granger Causality Test results

H0: Price change does H0: Index Investment
not Granger-cause does not Granger-cause
Market Critical Value Index Investment price change

Soybean F Statistics 22.0598 0.3160

Lag = 1 P-value 0.0000 0.5766

Soybean F Statistics 6.3468 0.1177

Lag = 3 P-value 0.0012 0.9492

Soybean F Statistics 102.7560 0.0556

Oil P-value 0.0000 0.8139

F Statistics 4.1746 0.5718

Wheat
P-value 0.0496 0.4553

F Statistics 2.0372 5.2247

Cotton
P-value 0.2080 0.0370

F Statistics 5.8909 0.1129

Sugar
P-value 0.0198 0.7387

18
For all five markets, the Granger Causality Tests successfully detect the presence of

casual links between the index investment positions and the futures prices. However,

results in soybean, soybean oil, wheat and sugar markets show evidence against the

Masters Hypothesis. In other words, in soybean, soybean oil, wheat and sugar

markets, we can conclude at 5% significance level that index investment does not

Granger-cause futures price changes, but futures price changes do Granger-cause

index investment. At 1% significance level, we reach the same conclusion for soybean

and soybean oil markets. Cotton market gives us some empirical evidence that

supports the ‘Masters Hypothesis’. At 5% significance level, we can conclude that

futures price changes do not Granger-cause index investment, but index investment

does Granger-cause futures price changes.

As we mentioned above, we use proxy variable for index positions before

2006 and we expect some error may occur. In order to determine whether this error

would compromise our result accuracy, we carry out another Granger Causality Test

for the soybean oil bubble period after year 2006. The results are shown below.

Table 5 Granger Causality Test results (Soybean Oil market, after 2006)
Market Lags 1 2 3 4 5 6
Soybean AIC -6.0390 -6.0624 -6.0298 -5.9906 -5.9165 -5.8452
Oil SC -5.8892 -5.8111 -5.6759 -5.5330 -5.3538 -5.1761
Critical H0: Price change does not H0: Index Investment does not
Lags
Value Granger-cause Index Investment Granger-cause price change
F Statistics 99.2888 0.6451
1
P-value 0.0000 0.4237
F Statistics 46.1429 1.1662
2
P-value 0.0000 0.3157

Using the results with p=1 and p=2, we conclude at 1% significance level that index

investment does not Granger-cause futures price changes, but futures price changes do

19
Granger-cause index investment. This result is in line with our test using the whole

bubble period data and we conclude that the error caused by data availability is

tolerable in soybean oil market.

We now try to explain the unexpected test results in soybean, soybean oil,

sugar and wheat market. We attributed this phenomenon to some behavioral factors. If

investments are driven by price itself, instead of fundamentals, price increase does

cause long-only investment position increase. When investors detect a sign of price

surge, they are inclined to take a long position in the futures market because they

believe in a further price increase. Later if they want to quit their position and cash

out their profits, they can simply sell the financial holdings to other market agents

who hold more optimistic beliefs. In this manner, the price bubble will provide a

motivation for more index fund trading.

We carry out a Granger Causality Test on non-bubble-existing period of cotton

market to justify the causal link found above is only present during bubble periods.

We use data from 2006 to 2009 and the results are shown below.

Table 6 Granger Causality Test result (Cotton market, non-bubble-existing period)

Market Lags 1 2 3 4 5 6
AIC -5.6077 -5.6291 -5.6175 -5.5747 -5.5729 -5.5388
Cotton
SC -5.4461 -5.5325 -5.3905 -5.2819 -5.2139 -5.1130
Critical H0: Price change does not H0: Index Investment does not
Lags
Value Granger-cause Index Investment Granger-cause price change
F Statistics 2.4110 1.0630
2
P-value 0.0923 0.3473

Following the AIC and SC rules, we choose p=2. At the 10% significance level, we

obtain that index investment does not Granger-cause price changes, but price changes

20
do Granger-cause index investment. At the 5% significance level, we fail to reach any

solid conclusion, since both p-values are greater than 0.05. Thus we conclude with

more confidence that index investment does cause the cotton market bubble.

Appendix

Table 7 Critical values at 90% level, by lags 0 to 12 and number of observations 100 to 500
#lag 100 150 200 250 300 350 400 450 500
s
0 1.145 1.449 1.593 1.723 1.814 1.892 1.957 2.009 2.045
1 1.267 1.567 1.753 1.868 1.97 2.028 2.111 2.16 2.208
2 1.415 1.748 1.926 2.056 2.153 2.236 2.289 2.343 2.384
3 1.462 1.775 2.004 2.154 2.239 2.331 2.396 2.451 2.501
4 4.607 1.987 2.159 2.311 2.397 2.473 2.534 2.577 2.631
5 1.628 1.988 2.21 2.335 2.435 2.539 2.62 2.688 2.728
6 1.691 2.089 2.317 2.457 2.571 2.678 2.75 2.785 2.841
7 1.768 2.162 2.382 2.514 2.639 2.733 2.8 1.875 2.93
8 1.834 2.258 2.502 2.639 2.733 2.834 2.891 2.961 3.024
9 1.838 2.263 2.465 2.659 2.763 2.855 2.925 2.988 3.051
10 1.953 2.389 2.613 2.763 2.877 2.962 3.037 3.126 3.182
11 1.985 2.439 2.671 2.806 2.93 3.046 3.092 3.161 3.22
12 2.093 2.477 2.749 2.916 3.039 3.132 3.215 3.273 3.325

Table 8 Critical values at 95% level, by lags 0 to 12 and number of observations 100 to 500
#lag 100 150 200 250 300 350 400 450 500
s
0 1.455 1.709 1.863 1.992 2.061 2.14 2.203 2.266 2.308
1 1.589 1.836 2.039 2.163 2.251 2.31 2.374 2.423 2.458
2 1.78 2.061 2.209 2.347 2.426 2.52 2.567 2.618 2.662
3 1.783 2.131 2.331 2.474 2.566 2.649 2.715 2.759 2.789
4 2.015 2.335 2.496 2.605 2.695 2.765 2.836 2.873 2.923
5 2 2.334 2.517 2.659 2.759 2.852 2.923 2.974 3.003
6 2.083 2.484 2.664 2.806 2.896 2.992 3.037 3.097 3.138
7 2.168 2.492 2.735 2.881 2.985 3.058 3.109 3.157 3.185
8 2.28 2.645 2.874 2.99 3.061 3.144 3.212 3.26 3.317
9 2.235 2.659 2.869 3.007 3.127 3.218 3.305 3.359 3.418

21
10 2.35 2.761 2.97 3.127 3.239 3.311 3.378 3.438 3.49
11 2.418 2.821 3.008 3.122 3.25 3.349 3.424 3.476 3.514
12 2.452 2.87 3.103 3.259 3.358 3.445 3.564 3.601 3.654

22
Table 9 Critical values at 99% level, by lags 0 to 12 and number of observations 100 to 500
#lags 100 150 200 250 300 350 400 450 500
0 2.048 2.229 2.381 2.493 2.639 2.716 2.766 2.791 2.803
1 2.16 2.44 2.641 2.738 2.848 2.917 2.968 2.994 3.033
2 2.484 2.667 2.787 2.916 2.993 3.093 3.173 3.222 3.232
3 2.488 2.849 3.003 3.059 3.159 3.241 3.281 3.33 3.346
4 2.751 3.072 3.216 3.254 3.325 3.352 3.431 3.472 3.517
5 2.778 2.977 3.144 3.281 3.377 3.498 3.539 3.576 3.629
6 2.885 3.179 3.296 3.424 3.529 3.589 3.66 3.688 3.736
7 2.972 3.272 3.411 3.539 3.663 3.712 3.786 3.825 3.835
8 3.061 3.325 3.441 3.559 3.684 3.815 3.829 3.864 3.977
9 3.084 3.508 3.678 3.788 3.842 3.89 3.994 4.065 4.086
10 3.225 3.452 3.621 3.752 3.864 3.937 4.025 4.078 4.105
11 3.088 3.545 3.692 3.811 3.893 3.934 3.957 4.017 4.063
12 3.253 3.575 3.792 3.937 4.054 4.162 4.243 4.243 4.307

23
References

Abolafia, M. Y., & Kilduff, M. (1988). Enacting market crisis: The social

construction of a speculative bubble. Administrative Science Quarterly, 33(2),

177-193.

Alexandratos, N. (2008). Food price surges: Possible causes, past experience, and

longer term relevance. Population and Development Review, 34(4), 663-697.

Baffes, J., & Haniotis, T. (2010). Placing the 2006/08 commodity price boom into

perspective (Policy Research Working Paper No. 5731). Washington D.C.:

World Bank. Retrieved from

http://www-wds.worldbank.org/external/default/WDSContentServer/IW3P/IB/

2010/07/21/000158349_20100721110120/Rendered/PDF/WPS5371.pdf

Blanchard, O. J., & Watson, M. W. (1983). Bubbles, rational expectations and

financial markets (NBER Working Paper No. 945). Cambridge, MA: National

Bureau of Economic Research. Retrieved from

http://www.nber.org/papers/w0945

Camerer, C. (1989). Bubbles and fads in asset prices. Journal of Economic Surveys,

3(1), 3-41.

24
Casassus, J., & Collin-Dufresne, P. (2005). Stochastic convenience yield implied from

commodity futures and interest rates. The Journal of Finance, 60(5), 2283-2331.

Davidson, P. (2008). Crude oil prices: "Market fundamentals" or speculation?

Challenge, 51(4), 110-118.

Erb, C. B., & Harvey, C. R. (2006). The strategic and tactical value of commodity

futures. Financial Analysts Journal, 62(2), 69-97.

Etienne, X. L., Irwin, S. H., & Garcia, P. Bubbles in food commodity markets: Four

decades of evidence. Journal of International Money and Finance, (0)

Fawley, B. W., & Juvenal, L. (2011). Commodity price gains: Speculation VS.

Fundamentals. The Regional Economist, (July), 4-9.

Gilbert, C. L. (2008). Commodity speculation and commodity investment. Retrieved

from Department of Economics, University of Trento, Italy.

Gilbert, C. L. (2010). How to understand high food prices. Journal of Agricultural

Economics, 61(2), 398-425. doi:10.1111/j.1477-9552.2010.00248.x

Gilles, C., & LeRoy, S. F. (1992). Bubbles and charges. International Economic

Review, 33(2), 323-339.

Irwin, S. H., Sanders, D. R., & Merrin, R. P. (2009). Devil or angel? The role of

speculation in the recent commodity price boom (and bust). Journal of

Agricultural and Applied Economics, 41(2), 377-391.

25
Irwin, S. H., & Sanders, D. R. (2012). Testing the Masters Hypothesis in commodity

futures markets. Energy Economics, 34(1), 256-269.

Jiang, Z., Zhou, W., Sornette, D., Woodard, R., Bastiaensen, K., & Cauwels, P.

(2010). Bubble diagnosis and prediction of the 2005-2007 and 2008-2009

Chinese stock market bubbles. Journal of Economic Behavior & Organization,

74(3), 149-162.

Johansen, A., & Sornette, D. (1999). Financial" anti-bubbles": Log-periodicity in gold

and Nikkei collapses. International Journal of Modern Physics C, 10(4), 563-

575.

Kawai, M. (1983). Price volatility of storable commodities under rational expectations

in spot and futures markets. International Economic Review, 24(2), 435-459.

Phillips, P. C. B., Wu, Y., & Yu, J. (2011). Explosive behavior in the 1990s Nasdaq:

When did exuberance escalate asset values? International Economic Review,

52(1), 201-226.

Phillips, P. C. B., & Yu, J. (2011). Dating the timeline of financial bubbles during the

subprime crisis. Quantitative Economics, 2(3), 455-491.

Piesse, J., & Thirtle, C. (2009). Three bubbles and a panic: An explanatory review of

recent food commodity price events. Food Policy, 34(2), 119-129.

26
Pindyck, R. S. (2004). Volatility and commodity price dynamics. Journal of Futures

Markets, 24(11), 1029-1047.

Sanders, D. R., & Irwin, S. H. (2010). A speculative bubble in commodity futures

prices? Cross-sectional evidence. Agricultural Economics, 41(1), 25-32.

Schwartz, E. S. (1997). The stochastic behavior of commodity prices: Implications for

valuation and hedging. The Journal of Finance, 52(3, Papers and Proceedings

Fifty-Seventh Annual Meeting, American Finance Association, New Orleans,

Louisiana January 4-6, 1997), 923-973.

Schwartz, E., & Smith, J. E. (2000). Short-term variations and long-term dynamics in

commodity prices. Management Science, 46(7), 893-911.

Shiller, R. J. (2000). Measuring bubble expectations and investor confidence. Journal

of Psychology and Financial Markets, 1(1), 49-60.

Sornette, D., & Woodard, R. (2010). Financial bubbles, real estate bubbles, derivative

bubbles, and the financial and economic crisis. In M. Takayasu, T. Watanabe &

H. Takayasu (Eds.), Econophysics approaches to large-scale business data and

financial crisis (pp. 101-148) Springer Japan.

Sornette, D., & Zhou, W. (2002). The US 2000‐2002 market descent: How much

longer and deeper? Quantitative Finance, 2(6), 468-481.

27
Sornette, D., Woodard, R., & Zhou, W. (2009). The 2006–2008 oil bubble: Evidence

of speculation, and prediction. Physica A: Statistical Mechanics and its

Applications, 388(8), 1571-1576.

Zhou, W., & Sornette, D. (2008). Analysis of the real estate market in Las Vegas:

Bubble, seasonal patterns, and prediction of the CSW indices. Physica A:

Statistical Mechanics and its Applications, 387(1), 243-260.

28