Original Article

Detecting price artificiality and manipulation in

futures markets: An application to Amaranth
Received (in revised form): 7th May 2012

Atanu Saha
is Senior Vice President and Head of NY Office of Compass Lexecon. Dr Saha has served as an expert in
many prior matters, and he specializes in the application of economics and finance to complex business
litigation. He was a tenure-track professor at Texas A&M University where he taught PhD-level courses in
econometrics and applied economics. He is the recipient of the Graham Dodd Award for financial research.
Dr Saha holds a PhD from the University of California, Davis, and an MA from the University of Alberta, Canada.

Hans-Jürgen Petersen
is a Vice President at Compass Lexecon. He has experience in economic litigation matters across a wide
spectrum of industries and has testified on issues related to statistical sampling. Prior to Compass Lexecon, he
taught mathematics and statistics at Loyola University of Chicago. Dr Petersen holds a PhD in mathematics
from the University of Wisconsin – Milwaukee.

Correspondence: Atanu Saha and Hans-Jürgen Petersen, Compass Lexecon, 156 West 56th Street, 19th Floor,
New York, NY 10019, USA
E-mails: asaha@compasslexecon.com; hpetersen@compasslexecon.com

The first author served as an expert witness in the Amaranth litigation.

ABSTRACT In this article we propose a general method to test whether economic data
support the claim of futures market manipulation. We examine the question of whether or
not Amaranth manipulated the market for natural gas futures using three alternative
methods. The first is our contribution to the existing body of literature on the analysis of
manipulation claims. The subsequent two have previously been discussed in the literature. All
three methods yield the same result: economic data on futures prices and Amaranth’s trades
do not support the claim that Amaranth manipulated the natural gas futures market in 2006.
Journal of Derivatives & Hedge Funds (2012) 18, 254–271. doi:10.1057/jdhf.2012.7;
published online 21 June 2012
Keywords: manipulation; futures; natural gas
The online version of this article is available Open Access

INTRODUCTION market has expanded greatly in depth and scope,

The history of futures trading dates back to the covering a wide range of commodities and
end of the American Civil War, when contracts financial products. It appears that with the
for the delivery of grain were made into proliferation of futures, there has been a
transferable contracts that were frequently used corresponding increase in claims of
to offset each other. Since then, the futures manipulation in these markets, particularly in

& 2012 Macmillan Publishers Ltd. 1753-9641Journal of Derivatives & Hedge Funds Vol. 18, 3, 254–271
www.palgrave-journals.com/jdhf/
 Detecting price artificiality and manipulation in futures markets

recent years. Some examples of litigation matters Chairman Walter Lukken concluded that
involving manipulation claims in futures markets ‘y analysis of Amaranth trading data failed
include copper (In re Sumitomo Copper Litig., to conclude that Amaranth’s trading was
1998, 1999, 2000, 2001), treasuries (Pacific responsible for the spread price level observed
Investment Management Company LLC and during 2006’ (Lukken and Dunn, 2007; US
PIMCO Funds v. Hershey, 2010), natural gas Senate PSI, 2007). Furthermore, a class-action
(Hershey v. Energy Transfer Partners L.P., 2010; lawsuit alleged that Amaranth manipulated
In re Natural Gas Commodity Litig., 2004, 2005, the prices of 14 NYMEX futures contracts
2006; In re Amaranth Natural Gas Commodities (the March 2006 through April 2007 futures
Litig., 2008, 2009, 2010), milk (Anderson v. Dairy contracts), thereby inflating winter-summer
Farmers of America, Inc., 2010), silver (In re price spreads (In re Amaranth Natural Gas
Commodity Exchange, Inc., Silver Futures and Commodities Litig., 2008).1
Options Trading Litig., 2011), platinum (In re Most prior empirical studies on manipulation
Platinum and Palladium Commodities Litig., have examined price movements of the
2010), palladium (In re Platinum and Palladium commodity or commodities at issue and not
Commodities Litig., 2010), and crude oil (CFTC the trades of the alleged manipulator. This is
v. Parnon Energy Inc., 2011). because data on a hedge fund’s positions and
In this article we propose a general method to trades in any asset class are highly proprietary and
test whether economic data support the claim of are virtually impossible to acquire. However, the
futures market manipulation. We illustrate our Amaranth matter was an exception. The Senate
approach by applying it to the case of Amaranth investigation yielded data on Amaranth’s natural
Advisors LLC, which faced allegations of natural gas futures positions and trades. This data set
gas futures manipulation. Amaranth Advisors provides a unique opportunity to empirically test
LLC, founded in 2000, was a multi-strategy the manipulation claims using actual data on the
hedge fund. It engaged in trading of natural gas alleged manipulator’s trades and their potential
futures, among other instruments. Following a impact on prices.
steep decline in natural gas prices in August and In this article, we examine the question of
September of 2006, Amaranth’s futures positions whether or not Amaranth manipulated the
suffered large losses. By mid-September, market for natural gas futures using three
Amaranth was forced to transfer its energy alternative methods. The first is our
portfolio to Citadel LLC and JPMorgan Chase contribution to the existing body of literature
and to close down the fund. on the analysis of manipulation claims. The
The question as to whether or not Amaranth subsequent two have been discussed in the
manipulated natural gas futures prices in 2006 finance literature and serve as a robustness
generated considerable controversy. While check of the results obtained using our
the Senate Permanent Subcommittee on approach. All three methods yield the same
Investigations (PSI) found that ‘Amaranth’s 2006 result: economic data on futures prices and
positions in the natural gas market constituted Amaranth’s trades do not support the claim
excessive speculation’, the US Commodity that Amaranth manipulated the natural gas
Futures Trading Commission’s (CFTC) Acting futures market in 2006.

& 2012 Macmillan Publishers Ltd. 1753-9641 Journal of Derivatives & Hedge Funds Vol. 18, 3, 254–271 255
 Saha and Petersen

THE RELEVANT LITERATURE whether the alleged manipulator’s trades actually

Markham (1991) reviews historical manipulation affected futures prices. In his book, Pirrong
in the commodity futures markets, the role of (1996) proposes an economic model of
the Commodity Exchange Act (CEA), creation manipulation and discusses the conditions that
of the CFTC and manipulation cases brought by facilitate manipulation. In a subsequent article,
CFTC. Markham argues that while Pirrong (2004) sets out a number of econometric
manipulation may take the form of rumors or tests to detect manipulation. These tests examine
false information and rigged trades, ‘capping’ or the empirical patterns of commodity flows and
‘pegging’, the most significant form is market the relationship between spot and futures prices.
power manipulation, including ‘corners’ and Pirrong (2004) applies his methodology to the
‘squeezes’. Fischel and Ross (1991–1992) specific case of Ferruzzi soybean manipulation
provide an excellent overview of the relevant claim and concludes that soybean futures prices
literature. in the Chicago area were artificial as a result of
Several authors have recognized that there is Ferruzzi’s manipulative activities in 1989.
little consensus on the definition of Marthinsen and Gai (2010a) used a Granger
‘manipulation’. For example, Kolb and Overdahl causality model to analyze whether Amaranth’s
(2007, p. 58) write: trades affected the prices of natural gas futures in
2006. In a follow-up article, the authors
One curious feature of the Commodity
examined whether Amaranth’s spread trading
Exchange Act, and the regulations promul-
affected prices of calendar spreads, particularly
gated under the Act, is that ‘manipulation’ is
the winter-summer spreads (Marthinsen and
referred to nearly 100 times without ever
Gai, 2010b). In both papers, Marthinsen and Gai
once being defined. The lack of a definition
do not find evidence that Amaranth engaged in
is not surprising, because the meaning of the
manipulation.
word is so broad. Attempts to precisely
In this article, we contribute to the existing
define the term inevitably dissolve into
literature by proposing a method that allows the
circular logic: a manipulated price is an
investigator to examine both whether prices
artificial price; an artificial price is one that
were artificial and whether the alleged
has been manipulated.
manipulator’s trades caused any price artificiality
In light of this perceived circularity in defining in the futures market. This methodology
manipulation, Perdue (1987–1988) proposes a essentially involves creating a model to explain
definition that focuses on whether the conduct futures prices using market fundamentals and
of the party involved is reasonable, as opposed to then examining the correlation between the
focusing on the outcome. The article defines ‘errors’ (that is, the difference between actual
manipulation as conduct that would be and model-predicted prices) and the alleged
uneconomical or irrational, absent manipulative manipulator’s trades. We test the robustness of
intent. our results by applying Pirrong’s spot-future
By contrast, a number of studies, rather than price relationship test and Marthinsen and
attempting to define price artificiality or Gai’s Granger model to the data on Amaranth’s
interpreting intent, have instead examined trading and natural gas futures prices.

256 & 2012 Macmillan Publishers Ltd. 1753-9641 Journal of Derivatives & Hedge Funds Vol. 18, 3, 254–271
 Detecting price artificiality and manipulation in futures markets

In the next section, we set out our proposed fundamentals. If prices are indeed artificial, one
methodology. In the subsequent section, we will find statistically significant residuals during
discuss Pirrong’s economic and econometric the alleged manipulation period. In other words,
model of manipulation and apply it to the the test for statistically significant residuals from a
natural gas futures prices at issue. In the market fundamental-based regression is the test
penultimate section, we undertake a Granger for price artificiality.
causality analysis, applying it not only to However, a few caveats are in order. First, the
individual contract prices and price spreads, but efficacy of the approach in detecting price
also to price indices. The final section contains artificiality depends critically on correctly
concluding comments. identifying the relevant market fundamental
variables explaining commodity prices. If one or
more key variables are omitted from the
DETECTING PRICE ARTIFICIALITY regression model specification, then the
omission will be reflected in the residuals,
AND MANIPULATION
leading to potentially erroneous conclusions
A general two-step approach regarding price artificiality.
As noted earlier, defining price artificiality based Second, one needs to be cognizant about the
on manipulation alone necessarily creates a data limitations of the variables reflecting
circular logic: ‘A manipulated price is an market fundamentals. While it is true that most
artificial price; an artificial price is one that has demand and supply factors are observable, a key
been manipulated’ (Kolb and Overdahl, 2007, variable that is almost always unobservable is
p. 58). Consequently, in this article, we propose the market participants’ expectations about the
a two-step approach: first, we examine the future values of the demand and supply factors.
presence of price artificiality, and second, we For example, in the case of natural gas futures,
test for manipulative effect. These two steps the supply of natural gas in the United States is
are distinct, although complementary. influenced by the incidence of hurricanes in the
In the first step, we define prices as artificial Gulf of Mexico and the shut-in production
when they are not explained by market caused by hurricanes. While data on
fundamentals. Clearly, the fundamental supply production disrupted by hurricanes are readily
and demand factors that determine prices vary available and can be incorporated in the
by commodity. For example, in the next regression model, it is far more difficult to
subsection, we will discuss the specific supply capture, for example, market participants’
and demand factors that affected natural gas expectations about an approaching tropical
futures prices in 2006. Generally, one should be storm and its likelihood of becoming a supply-
able to examine the prices of the futures disrupting hurricane. In the absence of data on
contracts at issue through a regression model that market participants’ subjective assessments
has the market fundamental factors as regressors, about the future changes in demand and supply
that is, right-hand-side variables. The residuals factors, the explanatory power of the regression
from this regression model reflect the effect on model can be limited, leading to possibly false
prices of factors unrelated to supply and demand inferences about price artificiality.

& 2012 Macmillan Publishers Ltd. 1753-9641 Journal of Derivatives & Hedge Funds Vol. 18, 3, 254–271 257
 Saha and Petersen

Because of these limitations, an inquiry into Market conditions in 2006

price artificiality cannot be viewed in isolation. In this subsection, we illustrate our approach
It has to be examined in conjunction with an through its application to the data on
examination of the trading pattern of the alleged Amaranth’s trades. The first step entails
manipulator, which is the second step in our construction of a regression model for natural
proposed two-step approach. In particular, the gas futures prices. However, before listing the
second step involves examining whether the demand and supply factors that would
regression residuals from the first step are positively constitute the explanatory variables in the
correlated with the alleged manipulator’s trades. model, it might be instructive to discuss the
If the alleged manipulator’s buy-trades pushed up unique conditions that prevailed in the natural
prices – leading to large positive regression gas market in 2006. This discussion provides a
residuals – then those residuals would be positively useful backdrop for identifying the key
correlated with the trades. Conversely, if the demand and supply variables for the regression
alleged manipulator’s large sell-trades pushed down model of natural gas futures prices.
prices, then that would yield large negative Because the plaintiffs’ principal claim in
residuals, which would also be positively correlated the Amaranth litigation was that its trades
with the alleged manipulator’s trades. Standard tests affected the winter-summer price spreads, we
can be performed to examine the statistical focus on the behavior of the winter-summer
significance of the correlations. price spreads of natural gas futures. The year
It is important to recognize that this residual- 2006 was unusual not only for natural gas but
trade correlation analysis is meaningful because also for related energy products, such as crude
the regression model contains variables reflecting oil and heating oil. The levels of winter-
market fundamentals and not the alleged summer price spreads for these three energy
manipulator’s trades.2 Thus, the effect of the products for the years 2000–2009 are shown in
manipulator’s trading activities, if any, would be Table 1. The average price spreads, excluding
necessarily reflected in the prices ‘unexplained’ 2006, are shown at the bottom of this table.
by the model, that is, in the regression residuals. The fact that the winter-summer price spread
Furthermore, while a statistically significant for natural gas futures was unusually high in
positive correlation between the price-model 2006 is clearly apparent from the table: it was
residuals and the trades can suggest US$3.15 in 2006, which was more than four
manipulation, it does not imply that the alleged times higher than the average for the years
manipulator’s trades ‘caused’ the price excluding 2006. However, it is also evident
artificiality. Correlation, although necessary, is from the table that the price spread for heating
not a sufficient condition for causality. However, oil was four times higher in 2006 relative to the
a large and statistically significant positive other years. Crude oil is typically cheaper in
correlation would suggest that, during the winter, with the average winter-summer spread
relevant period, prices deviated from market being $0.21 for the years 2000–2009,
fundamentals, which, in turn, is not inconsistent excluding 2006. Yet in 2006, the price spread
with the proposition that the alleged of crude oil was positive and remarkably high
manipulator’s trades caused prices to be artificial. at þ $2.44.

258 & 2012 Macmillan Publishers Ltd. 1753-9641 Journal of Derivatives & Hedge Funds Vol. 18, 3, 254–271
 Detecting price artificiality and manipulation in futures markets

Table 1: Average historical winter-summer prices. In September, the price spreads collapsed
futures price spread over alleged manipula- as both winter and summer prices fell rapidly,
tion period days with winter prices evidencing a sharper decline.
Several key market fundamental factors
Year NG HO CL
explain the movement of winter-summer natural
2000 0.10 0.02 2.50 gas futures prices in 2006. First, the 2005–2006
2001 0.39 0.02 0.81 winter was unusually warm, leading to lower
2002 0.57 0.03 0.76 demand for natural gas. The Energy Information
2003 0.20 0.01 2.31 Administration (EIA) reported in January 2006
2004 0.59 0.01 1.99 that temperatures were between 22 and 41 per
2005 1.01 0.06 0.99 cent above normal for all Census divisions (EIA,
2006 3.15 0.16 2.44 2006a). The warm weather continued through
2007 1.71 0.10 2.11 February and March 2006 (EIA, 2006b).
2008 0.79 0.06 0.99 Second, the mild 2005–2006 winter created
2009 1.67 0.15 4.36 unusually high levels of natural gas inventory
(Bryden et al, 2006; Dell and Goller, 2006). An
Average ex. 2006 0.78 0.04 0.21
analyst report noted that ‘Naturally, the main
factor fueling lower natural gas prices is the
record surplus storage gas, which is currently y
$5.00
$4.50 48 per cent above the five-year average, and
$4.00
$3.50
21 per cent above last year’ (Covington et al,
Price

$3.00
$2.50
2006, p. 22).
$2.00 Third, analysts were expecting that the 2006
$1.50
$1.00 hurricane season, which was predicted to be
16/02/2006
02/03/2006
16/03/2006
30/03/2006
13/04/2006
27/04/2006
11/05/2006
25/05/2006
08/06/2006
22/06/2006
06/07/2006
20/07/2006
03/08/2006
17/08/2006
31/08/2006
14/09/2006

very active, would strengthen natural gas prices,

particularly the prices of the winter contracts.
Figure 1: Winter-summer price spread index,
According to forecasts by Colorado State
16/2/2006–15/9/2006.
University’s Tropical Meteorology Project
released on 4 April 2006, ‘[t]he 2006 Atlantic
hurricane season [was expected to be] much
The movement of winter-summer natural gas more active than the average 1950–2000 season’
price spreads in the relevant period of 2006 is (Klotzbach and Gray, 2006). The National
depicted in Figure 1. The underlying data for Oceanic and Atmospheric Administration
Figure 1 show that the price spreads increased (NOAA) also forecasted a very active hurricane
through early May 2006 primarily because, season: ‘The outlook calls for a very active 2006
while winter prices rose mildly, summer prices season y The main uncertainty in this outlook is not
fell, causing the winter-summer spread to widen. whether the season will be above normal, but how
During the period between early May and the much above normal it will be. The 2006 season
end of August, the price spreads fell slightly, could become the fourth hyperactive season in
mainly as a result of generally declining winter a row’ (National Oceanic and Atmospheric

& 2012 Macmillan Publishers Ltd. 1753-9641 Journal of Derivatives & Hedge Funds Vol. 18, 3, 254–271 259
 Saha and Petersen

Administration, 2006, emphasis added). hurricanes do similar damage this year or

NOAA’s predictions turned out to be untrue. In supply be disrupted overseas. However,
early July, the EIA observed that natural gas when these expectations did not
futures prices were continuing to drop because materialize, the selloff of contracts began
of the inactive hurricane season (EIA, 2006c). and prices plummeted. (EIA, 2006d)
Late July and early August brought about a
slight increase in natural gas prices due to triple While the report references the dynamics
digit temperatures (leading to a rare summer in the oil market, the same fundamentals also
withdrawal of natural gas) and the threat of affected the market for natural gas. This is
tropical storm Chris making landfall (Heim et al, corroborated by the data in Table 1, which
2006). Yet, soon after, the heat wave broke, and suggest co-movements of spreads in related
the threat of Chris dissipated. In mid-August, an energy products.
analyst noted that the fear factor associated with
the active hurricane season predictions had been The data and regression model
supporting the natural gas futures curve; variables
however, if the storms failed to materialize, gas We first discuss the data on Amaranth’s positions
prices would come under pressure in the fall and trades in natural gas futures. We then discuss
(Adams et al, 2006). the explanatory variables for the regression
Indeed, by the end of August, it became analysis. As part of its Excessive Speculation in the
abundantly clear that the hurricane predictions Natural Gas Market report, the US Senate
for 2006 were grossly overstated. Since its Permanent Subcommittee on Investigations
inception in 1998, 2006 was the first year (PSI) provided detailed information on
wherein NOAA over-forecasted hurricane Amaranth’s daily positions on the NYMEX,
activity both in May and August, and the first Clearport and ICE natural gas derivatives market
year since 2001 that no hurricanes struck the for Henry Hub delivery covering the time
continental United States. This led to a sharp period January 2005-September 2006.
decline in natural gas futures prices in Included are Amaranth’s positions in futures,
September, with winter prices falling more swaps and options. The latter are expressed in
sharply than summer prices. A 4 October 2006 futures equivalents (FEQs). The FEQs are
EIA report summarized the market dynamics available for 2006 only. For each day in 2006
contributing to the collapse of price spreads as and for each delivery month between March
follows: 2006 and October 2007, we aggregate the data
into two measurements of Amaranth’s positions:
The devastating impact last year’s ‘Futures Only’ (this includes swaps) and
hurricanes had on oil infrastructure, along ‘Futures and Option FEQs’. Note that while a
with forecasts of another strong hurricane change in ‘Futures Only’ from one day to the
season this year, led many oil market next does imply a trade in a futures or swap
participants to buy additional contracts contract on that day, the same is not true for a
early in the year, with the expectation that change in FEQs. The FEQ value might change
prices could be much higher should due to changes in the option’s volatility without

260 & 2012 Macmillan Publishers Ltd. 1753-9641 Journal of Derivatives & Hedge Funds Vol. 18, 3, 254–271
 Detecting price artificiality and manipulation in futures markets

any trade. However, in the absence of more choose 10 explanatory variables reflecting
detailed options data, we use change in market supply and demand fundamentals in a
Amaranth’s FEQ positions as a proxy for its regression analysis of natural gas futures prices,
FEQ trades. including futures prices of competing energy
In order to address the co-movement of products, forecasted energy use, inventory,
Amaranth’s trades, and because of difficulties in weather and expected hurricane activity.
allocating Amaranth’s positions in a specific Appendix A contains a detailed list of the
contract to a specific spread position, we 10 variables used in our model.
constructed two winter-summer aggregate We use the same set of explanatory variables
spread trade variables for Amaranth: (a) the sum discussed above for the regression model for all
of Amaranth’s trades in the November 2006 14 contracts. In each regression, we allow for the
through March 2007 contracts minus the sum of coefficient for each explanatory variable to be
Amaranth’s trades in the April 2006 through different for the summer (1 April–31 October)
October 2006 contracts (‘Trade 2006’); and versus winter time period. For each contract, the
(b) the sum of Amaranth’s trades in the regression is run using data for the time period
November 2006 through March 2007 contract 4 January 2006 through each contract’s expiry
minus the sum of Amaranth’s trades in the date. We also ran the same regression to explain
April 2007 through October 2007 contracts the summer and winter volume-weighted price
(‘Trade 2007’). In addition to using the two indices; this index regression has the advantage
aggregate trade variables, we also examined of capturing the seasonal co-movement of
Amaranth’s individual spread trades, such as the contract prices. For each regression equation, we
January 2007–October 2006 trade and the corrected for autocorrelation using the Prais-
March 2007–April 2007 trade. Winsten method (Prais and Winsten, 1954;
Similar to the construction of the aggregate Greene, 1997). In Appendix B, we present the
trade variables, we also constructed a results of the regression for the summer and
corresponding spread price index. We defined winter 2006 price indices. Details of the
the winter-summer spread price index as the regression results for all 14 contracts are available
volume-weighted prices of the November 2006 from the authors upon request.
through March 2007 contracts minus the
volume-weighted prices for the April 2006
through October 2006 contracts, with the Regression results and correlation
volume being the number of contracts traded as analysis
reported by NYMEX. We find that the key demand and supply
Here we introduce an econometric model variables identified above explain a vast majority
to explain natural gas futures prices for the of the movement of the natural gas futures
14 contracts at issue expiring in 2006–2007. prices. Table 2 contains the adjusted R-squares
Our review of industry publications and various for each of the 14 contracts and the two seasonal
empirical studies on energy prices reveals that price indices. Table 2 shows that the regression
most identify a common set of key demand and model explains, on average, nearly 95 per cent of
supply drivers of natural gas futures prices. We the price movements across the 14 contracts, and

& 2012 Macmillan Publishers Ltd. 1753-9641 Journal of Derivatives & Hedge Funds Vol. 18, 3, 254–271 261
 Saha and Petersen

Table 2: Adjusted R-squares for the 14 Amaranth’s trades. The exclusion of the latter
contracts and seasonal price indices and their potential impact on prices would
thus be reflected in the regression errors, that is,
# Contract or price index Adjusted
the residuals.
R-squares (%)
Since we are interested in examining the
1 March 2006 contract 96.1 correlation of Amaranth’s spread trades with
2 April 2006 contract 97.0 regression errors, we need to convert the
3 May 2006 contract 96.7 residuals produced by our regression analyses
4 June 2006 contract 96.9 into spread residuals. In order to do so, we
5 July 2006 contract 94.6 simply take the difference between the residual
6 August 2006 contract 94.0 for the winter contract and that for the summer
7 September 2006 contract 91.2 contract. Accordingly, we define Amaranth’s
8 October 2006 contract 92.5 winter-summer spread trades as Trade 2006 and
9 November 2006 contract 90.9 Trade 2007.
10 December 2006 contract 94.5 In Table 3, we report the correlation between
11 January 2007 contract 94.6 the spread residuals and Amaranth’s spread trades
12 February 2007 contract 94.5 in individual contracts and for the winter-
13 March 2007 contract 94.8 summer price spread index. The first seven are
14 April 2007 contract 95.6 winter-summer spread prices, where the winter
15 Summer 2006 price index 91.7 contract is represented by the January 2007
16 Winter 2006-2007 price index 90.5 contract; in the next six, the winter prices are
those of contracts November 2006 through
Average (1–14) 94.6 March 2007, while the summer price for all six
is represented by the October 2006 contract.
Finally, we examine the winter-summer price
spread index.
more than 90 per cent when prices are Table 3 shows that in all cases the correlation
incorporated into the seasonal indices. between regression errors and Amaranth’s spread
The small portion of the price movements trades is generally extremely low or negative. We
that remains unexplained by market also undertook a formal test to examine whether
fundamentals are the ‘residuals’ in this regression any of these correlations is statistically significant.
analysis. A critical question is whether these Of the 28 correlations listed, not a single one is
residuals are correlated with Amaranth’s trades statistically significant.3
during the alleged manipulation period. This is Thus, our empirical findings show no
the second step of our proposed two-step evidence of impact of Amaranth’s trades on the
approach. As discussed earlier, if indeed relevant natural gas futures contract prices.
Amaranth’s trades affected prices, then that Rather, based on the results of our regression
impact would have been evidenced in the and correlation analysis, we conclude that
regression residuals because the model does not market fundamentals were the drivers of the
include any explanatory variable that reflects price of the 14 natural gas futures contracts at

262 & 2012 Macmillan Publishers Ltd. 1753-9641 Journal of Derivatives & Hedge Funds Vol. 18, 3, 254–271
 Detecting price artificiality and manipulation in futures markets

Table 3: Correlation between regression spread errors and Amaranth’s spread trades,
4/1/2006–15/9/2006

# Spread Number of Correlation between errors and Amaranth’s trades

data points
With FEQs Without FEQs

1 January 2007–April 2006 28 0.06 0.14

2 January 2007–May 2006 46 0.01 0.01
3 January 2007–June 2006 68 0.15 0.16
4 January 2007–July 2006 89 0.21 0.15
5 January 2007–August 2006 107 0.23 0.20
6 January 2007–September 2006 130 0.12 0.09
7 January 2007–October 2006 141 0.04 0.02
8 November 2006–October 2006 141 0.04 0.04
9 December 2006–October 2006 141 0.01 0.01
10 January 2007–October 2006 141 0.04 0.02
11 February 2007–October 2006 141 0.06 0.05
12 March 2007–October 2006 141 0.08 0.06
13 March 2007–April 2007 141 0.03 0.08
14 Winter-summer price spread index 141 0.01 0.01

issue and that there is no causal impact of (equivalently, by liquidating too few contracts),
Amaranth’s spread trades on those price spreads. the large long induces shorts to repurchase their
In the next two sections, we examine whether remaining positions at artificially high pricesy
our findings are corroborated using two [A] cornerer artificially increases demand for the
alternative approaches that exist in the literature. commodity at the delivery point during the
delivery period in order to earn a
supercompetitive profit’ (Pirrong, 2004, p. 34).
PIRRONG’S ANALYSIS OF A In his 2004 paper, Pirrong tests whether
MANIPULATION CLAIM Ferruzzi manipulated soybean futures in the May
Pirrong argues that a manipulation of a contract. To test whether the price of the May
commodity futures market is profitable when a 1989 futures contract was manipulated, Pirrong
trader owns large long futures positions, and the uses historical price data to model the normal
marginal cost of transporting the commodity relationship between May soybean contract
increases as short traders are forced to make prices and non-deliverable spot prices (central
more deliveries. He posits that ‘y by calling for Illinois) using regression analysis. Pirrong then
too many deliveries of the commodity cumulates the regression errors for the relevant

& 2012 Macmillan Publishers Ltd. 1753-9641 Journal of Derivatives & Hedge Funds Vol. 18, 3, 254–271 263
 Saha and Petersen

period in May 1989 and tests whether the daily log-returns of spot prices, for the period
magnitude of these cumulated residuals is 16 February–31 August5 for each of the years
statistically significant. He then concludes that 2000–2009, excluding 2006. These regressions
the statistical significance of the cumulated yield the predicted futures price returns and the
residuals is consistent with the claim of resulting out-of-sample regression residuals for
manipulation by Ferruzzi. 2006. We then test for the statistical significance
In applying Pirrong’s framework, we have of the cumulative residuals for the February
confined the test of artificiality to the through August period in 2006.6
relationship between futures and spot prices. If Table 4 shows the cumulated residuals and
the manipulation claim were true, then one corresponding t-statistics for the 12 regressions.
would have observed that the historical None of the t-statistics for the cumulated
relationship between spot and futures prices for residuals (cumulated through 31 August 2006)
natural gas futures would have ceased to exist is statistically significant (since they are all less
during the class period in 2006. In other words, than 1.96 in absolute value). These results are
the cumulative residuals (from the regression of inconsistent with the claim of manipulated
futures on spot prices) would have been natural gas futures prices.
statistically significant during the period in Since the cumulated residuals come from out-
which Amaranth’s trades allegedly affected the of-sample forecasts for 2006 using a regression
futures prices. model that excludes data for 2006, a possible
In undertaking the analysis of spot and future source of statistical significance of the residuals
price, we have compared the prices of Henry could be a structural break in 2006.7 In fact,
Hub rolling front month, first-deferred month the data in Table 1 seem to indicate this. Two
and second-deferred month natural gas futures observations are in order here. First, while
contracts to the spot prices at various locations: Table 1 data seem to suggest that all energy
the delivery location at Henry Hub and the prices (including natural gas) experienced a
non-delivery locations at Alberta, El Paso and structural break in 2006, the relationship between
an average of West Coast locations.4 This spot and futures prices for natural gas need not
comparison is undertaken by regressing the have changed structurally in 2006. Second, even
relevant daily log-return of futures prices on the if one admits the possibility of structural break in

Table 4: Cumulative residuals as of 31/8/2006

Spot location Front month Second month Third month

CR t-stat CR t-stat CR t-stat

Henry Hub 0.18 0.51 0.21 0.62 0.17 0.53

Alberta 0.18 0.52 0.20 0.61 0.16 0.52
El Paso 0.22 0.60 0.23 0.68 0.19 0.60
West Coast average 0.20 0.54 0.22 0.64 0.18 0.56

264 & 2012 Macmillan Publishers Ltd. 1753-9641 Journal of Derivatives & Hedge Funds Vol. 18, 3, 254–271
 Detecting price artificiality and manipulation in futures markets

the spot-futures price relationship, that break would expect to find this association to be
would have been reflected in the statistical positive and statistically significant.
significance of the cumulated residuals for 2006. The Granger causality test has its limitations.
The results in Table 4 show that this is not the Spurious Granger causality can be inferred when
case. However, if the cumulated residuals were both variables X and Y have a common
statistically significant, the spot-future price (exogenous) cause Z that is absent from the
relationship test would have been incapable of regression equation. In this case, the test could
discerning the source of the statistical erroneously conclude Granger causality between
significance: it could have been either structural X and Y (Hood III et al, 2008; Hartwig, 2009).
break in 2006 and/or the effect of Amaranth’s Classic examples highlighting this issue can be
trades. found in several earlier studies (for example,
We have undertaken two variations of the sunspots causing GNP and/or prices [ Jevons,
analysis reported in Table 4 to verify the 1884], or GNP and/or prices causing sunspots
robustness of our results. The cumulative [Sheehan and Grieves, 1982; Chowdury, 1987]).
residuals are insignificant for every spot price if We follow the Granger causality specification
the period is extended to 15 September 2006. set forth in Marthinsen and Gai (2010a), who
The cumulative residuals are also insignificant for base their approach on the work of Hartzmark
every spot price when the dependent variable is (1991), Wang (2001), and Sanders et al (2004)
a volume-weighted index of summer and winter X
m X
n
futures prices. Rtc ¼ a þ c
gi Rti þ c
bi DAPtj þ et
i¼1 j¼1

where Rct ¼ ln(Ptc )ln(Pt1 c

) is the log return
AN APPLICATION OF GRANGER for contract c at date t, and Pct is the settlement
price for contract c on trade date t. For example,
ANALYSIS
P3OctJan2006
2006 represents the settlement price on
Overview of Granger analysis 3 January 2006 for the October 2006 natural
c
The Granger causality test is concerned with gas futures contract. Finally, DAPtj is the first
testing whether past changes in a time series difference of Amaranth’s position for contract
variable X are useful in predicting the current c over time period tj.
change in another time series variable Y. If the
answer is affirmative, then the experimenter
concludes that ‘X Granger-causes Y’. In the Results of Granger causality analysis
context of allegations that Amaranth We perform Granger causality tests on three
manipulated prices, an implementation of the spreads: January 2007-October 2006, March
Granger causality test is a regression analysis to 2007-April 2007 and winter-summer index
examine whether the past changes in Amaranth’s price spreads. In each case, the Granger
winter-summer spread trades are associated with regression is run using data for the period
the current changes in winter-summer price 16 February through 15 September 2006. For
spreads. If the allegations were true and the first two, the key regressor is the lagged value
Amaranth had manipulated the market, one of Amaranth’s corresponding spread trades; for

& 2012 Macmillan Publishers Ltd. 1753-9641 Journal of Derivatives & Hedge Funds Vol. 18, 3, 254–271 265
 Saha and Petersen

Table 5: Granger regression of price spreads and Amaranth’s spread trades

Winter-summer price spread variable With FEQs Without FEQs

Coefficient t-stat Coefficient t-stat

January 2007–October 2006 0.35 2.14 0.29 1.69

March 2007–April 2007 0.09 0.78 0.05 0.40
Winter-summer price spread index 0.04 0.13 0.04 0.15

the price index, the key regressor is Amaranth’s natural gas futures using three alternative
lagged aggregate trade variable, Trade 2006. In methods. After reviewing the existing body
Table 5, we report the estimated coefficient of of literature, we have proposed a method
the lagged Amaranth trade variable. None of the that allows the investigator to examine both
Granger causality tests shows an estimated whether prices were artificial and whether
coefficient that is both positive8 and statistically the alleged manipulator’s trades caused any
significant. price artificiality in the futures market. To test
To verify the robustness of our analysis, we the robustness of our results, we have applied
performed Granger causality tests for all possible two additional approaches, the spot-future
winter-summer price spread combinations. Each price relationship test and a Granger model,
of the five winter contracts at issue (November to Amaranth’s trading data and natural gas
2006-March 2007) was paired with each of futures prices. Application of all three
the eight summer contracts at issue (April approaches yielded the same conclusion: the
2006-October 2006 plus April 2007), resulting evidence is not consistent with the proposition
in 40 combinations. For the right-hand-side that Amaranth manipulated the market for
variables, we used Trade 2006 if the summer leg natural gas futures in 2006.
of the spread belonged to summer 2006, and we In the class-action lawsuit, the plaintiffs had
used Trade 2007 if the summer leg was April also alleged that Amaranth depressed the
2007. None of the 40 Granger regressions settlement prices of the March 2006, April
yielded positive and statistically significant 2006 and May 2006 contract prices by
coefficient estimates. ‘slamming the close’ of these three expiring
In summary, we find no empirical evidence contracts. We have not addressed this issue
suggesting that Amaranth’s trading in winter- here, although such an analysis would be a
summer spreads Granger-caused changes in useful extension to this article. Analysis of the
winter-summer spread prices in 2006. ‘slamming the close’ claims requires
examination of the intra-day price and trade
data. Furthermore, the methodology proposed
CONCLUDING COMMENTS in this article needs to be suitably modified to
We have examined the question of whether address intra-day trades and their potential
or not Amaranth manipulated the market for impact on prices during the 30-min settlement

266 & 2012 Macmillan Publishers Ltd. 1753-9641 Journal of Derivatives & Hedge Funds Vol. 18, 3, 254–271
 Detecting price artificiality and manipulation in futures markets

period of expiring contracts. We plan to issues at hand. In particular, a natural

extend the methodological framework candidate for an instrumental variable is
that allows us to examine intra-day price Amaranth’s lagged trades. However, the
manipulation claims in a subsequent lagged trade variable would not be an
study. appropriate choice in the context of this
application because the manipulation claim
is that Amaranth’s trades on any given day
ACKNOWLEDGEMENTS
affected prices on that day, and not
The authors are immensely grateful to Cecilia
necessarily the subsequent day.
Seiden for her excellent research support. We
3. To check for robustness, we have also tested
wish to also acknowledge our gratitude to
the correlation between regression spread
Carina Chambarry, Elizabeth Wall, Nicolas Shea
errors and lagged Amaranth’s spread trades
and Hillel Kipnis for their invaluable assistance
(t1), and the correlation between regression
with research and data analytics and their
errors and Amaranth’s trades for individual
thoughtful comments. We are extremely grateful
contracts, and find our results to be robust.
to Mary Shen for her editorial assistance. The
4. Average of West Coast locations is
authors are also indebted to the reviewers for
constructed by Bloomberg, L.P.
their helpful comments.
5. We have started our analysis on 16 February,
as that is the day the class action complaint
NOTES alleges the manipulation began. We have
1. In this litigation, the plaintiffs claimed the chosen 31 August as the end date because
manipulation period to extend from after that date, natural gas futures prices fell
16 February through 15 September 2006 sharply across almost all traded contracts.
(the class period). Thus, the cumulative residuals would likely
2. One might be inclined to combine the two be highest at the end of August 2006.
P
steps in our approach by including the 6. The CRT ¼ tT¼ 1 êt, where êt denotes the
alleged manipulator’s trades as a right-hand- residual from the regression of the futures on
side variable in the regression equation the spot price on the t-th day. Let the root
and testing for the statistical significance of mean square of this regression be denoted by
the trade variable’s coefficient. However, ŝ. Then the t-statistic for testing the statistical
this approach introduces an endogeneity significance of CRT is given by
pffiffiffiffi
bias in the estimation since trades affect CRT =ðs  T Þ.
prices, and prices affect trades. The standard 7. We are indebted to an anonymous reviewer
approach of using instrumental variables for suggesting this point.
that are correlated with the trade variable 8. The importance of the sign of the coefficients
but uncorrelated with the residuals is within a Granger causality framework is
problematic here because it is virtually documented in several earlier studies; see
impossible to find a suitable instrumental Marvell and Moody (1996), Sanders et al
variable that satisfies the required (2004, pp. 435–437), Marthinsen and Gai
conditions and is also responsive to the (2010a, p. 271).

& 2012 Macmillan Publishers Ltd. 1753-9641 Journal of Derivatives & Hedge Funds Vol. 18, 3, 254–271 267
 Saha and Petersen

REFERENCES Kolb, R.W. and Overdahl, J.A. (2007) Futures, Options, and
Adams, D., Mariani, L. and Nuschler, G. (2006) Energy: Swaps, 5th edn. Malden, MA: Blackwell Publishing.
Jefferies North American Natural Gas Weekly. Jefferies & Lukken, W. and Dunn, M. (2007) Written Testimony of
Company, Inc., 14 August. Acting Chairman Walter Lukken and Commissioner
Bryden, P., Shaw, G. and Kelly, F. (2006) Oil & Gas Trusts: Michael Dunn, Commodity Futures Trading
An Overview of Recent Natural Gas Price Weakness and Commission, Before the Permanent Subcommittee on
Potential Implications for Distribution Levels. RBC Capital Investigations, Committee on Homeland Security and
Markets, 24 March. Governmental Affairs, United States Senate. 9 July.
Chowdury, A.R. (1987) Are causal relationships sensitive to Markham, J.W. (1991) Manipulation of commodity futures
causality tests? Applied Economics 19(4): 459–465. prices – The unprosecutable crime. Yale Journal on
Covington, T., Scialla, M. and Pikul, C. (2006) US Natural Regulation 8(2): 281.
Gas Market Update & Outlook: Lowering 2006 Natural Gas Marthinsen, J.E. and Gai, Y. (2010a) Did Amaranth
Price Forecast to $7.50. Edwards, 7 March. advisors LLC engage in interday price manipulation in
Dell, B. and Goller, A. (2006) Bernstein E&Ps: Lower the natural gas futures market? Journal of Derivatives &
Natural Gas Prices and Escalating Costs. Who is Best Hedge Funds 15(4): 261–273.
Positioned for Margin Compression? Bernstein Research, Marthinsen, J.E. and Gai, Y. (2010b) Did Amaranth’s
19 April. absolute, relative and extreme positions affect natural gas
Energy Information Administration. (2006a) Natural Gas futures prices, spreads and volatilities? Journal of
Weekly. EIA report, 19 January. Derivatives and Hedge Funds 16(1): 9–21.
Energy Information Administration. (2006b) Natural Gas Marvell, T.B. and Moody, C.E. (1996) Specification
Weekly. EIA report, 27 April. problems, police levels, and crime rates. Criminology
Energy Information Administration. (2006c) Natural Gas 34(4): 609–646.
Weekly. EIA report, 7 July. National Oceanic and Atmospheric Administration. (2006)
Energy Information Administration. (2006d) This Week in NOAA: 2006 Hurricane Outlook. NOAA report, 22
Petroleum. EIA report, 4 October. May.
Fischel, D.R. and Ross, D.J. (1991–1992) Should the law Perdue, W.C. (1987–1988) Manipulation of futures
prohibit manipulation in financial markets? Harvard Law markets: Redefining the offense. Fordham Law Review
Review 105(2): 503–553. 56(3): 345–402.
Greene, W.H. (1997) Econometric Analysis, 3rd edn. Upper Pirrong, C. (1996) The Economics, Law, and Public Policy of
Saddle River, NJ: Prentice-Hall. Market Power Manipulation. Norwell, MA: Kluwer
Hartwig, J. (2009) A Panel Granger-Causality Test of Academic Publishers.
Endogenous vs. Exogenous Growth. Zurich, CH: Pirrong, C. (2004) Detecting manipulation in futures
KOF Swiss Economic Institute at ETH Zurich. KOF markets: The Ferruzzi soybean episode. American Law
Working Papers, No. 231. and Economics Review 6(1): 28–71.
Hartzmark, M.L. (1991) Luck versus forecast ability: Prais, S.J. and Winsten, C.B. (1954) Trend Estimators and
Determinants of trader performance in futures markets. Serial Correlation. Cowles Commission Discussion
Journal of Business 64(1): 49–74. Paper no. 383.
Heim, M., Mikhareva, J., Donohue, M. and Argiris, P. Sanders, D.R., Boris, K. and Manfredo, M. (2004)
(2006) Weekly Gas Statistics: Natural Gas Futures Hedgers, funds, and small speculators in the energy
Hit Three-Month High. Orion Securities, Inc., futures markets: An analysis of the CFTC’s
4 August. Commitments of Traders reports. Energy Economics
Hood III, M.V., Kidd, Q. and Morris, I.L. (2008) Two sides 26(3): 425–445.
of the same coin? Employing Granger causality tests in Sheehan, R.G. and Grieves, R. (1982) Sunspots and cycles: A
a time series cross-section framework. Political Analysis test of causation. Southern Economic Journal 48(3): 775–777.
16(3): 324–344. United States Senate Permanent Subcommittee on
Jevons, W.S. (1884) Commercial crises and sun-spots. In: Investigations, Committee on Homeland Security and
W.S. Jevons and H.S. Foxwell (eds.) Investigations in Governmental Affairs. (2007) Excessive Speculation in
Currency and Finance. London, UK: Macmillan. the Natural Gas Market. Staff Report with Additional
Klotzbach, P.J. and Gray, W.M. (2006) Extended range Minority Views. http://hsgac.senate.gov/public/_files/
forecast of Atlantic seasonal hurricane activity and US REPORTExcessiveSpeculationintheNaturalGasMarket
landfall strike probability for 2006. Colorado State .pdf.
University, Department of Atmospheric Science, Wang, C. (2001) Investor sentiment and return
4 April. http://hurricane.atmos.colostate.edu/Forecasts/ predictability in agricultural futures markets. The Journal
2006/april2006/. of Futures Markets 21(10): 929–952.

268 & 2012 Macmillan Publishers Ltd. 1753-9641 Journal of Derivatives & Hedge Funds Vol. 18, 3, 254–271
 Detecting price artificiality and manipulation in futures markets

REFERENCED CASES: 2. Forecasted energy use: We use forecasts of

In re Sumitomo Copper Litigation, 182 F.R.D. 85 (S.D.N.Y. daily energy demand through the Dominion
1998).
In re Sumitomo Copper Litigation, 189 F.R.D. 274, 284 Energy Index, contained in Platts Gas Daily;
(S.D.N.Y.1999). 3. Demand: We use a daily measure of US
In re Sumitomo Copper Litigation, 104 F. Supp. 2d 314 demand derived from data provided by
(S.D.N.Y. 2000).
In re Sumitomo Copper Litigation, 194 F.R.D. 480 (S.D.N.Y. Bentek;
2000). 4. Actual impact of hurricane activity: We
In re Sumitomo Copper Litigation, 120 F. Supp. 2d 328 use actual production loss figures due to
(S.D.N.Y. 2000).
In re Sumitomo Copper Litigation, 262 F. 3d 134 (2d Cir. 2001). hurricanes in the Gulf of Mexico produced
Pacific Investment Management Company LLC & PIMCO by the Bureau of Ocean Energy
Funds v. Hershey, 130 S. Ct. 1504, (2010).
Management, Regulation, and
Hershey v. Energy Transfer Partners, L.P., 610 F. 3d 239 (5th
Cir. 2010). Enforcement;
In re Natural Gas Commodities Litigation, 337 F. Supp. 2d 5. Inventory: We measure the differential in
498, 502 (S.D.N.Y. 2004).
In re Natural Gas Commodities Litigation, 231 F.R.D. 171, existing storage level relative to the prior
181-182 (S.D.N.Y. 2005). five-year average;
In re Natural Gas Commodities Litigation, 235 F.R.D. 241, 6. Inventory: We complement the measure
244 (S.D.N.Y. 2006).
In re Amaranth Natural Gas Commodities Litigation, 587 F. of the differential in existing storage level
Supp. 2d 513 (S.D.N.Y. 2008). introduced previously with a measure of
In re Amaranth Natural Gas Commodities Litigation, 612 F. storage injections and withdrawals derived
Supp. 2d 376 (S.D.N.Y. 2009).
In re Amaranth Natural Gas Commodities Litigation, 711 F. from data provided by Bentek and the
Supp. 2d 301 (S.D.N.Y. 2010). EIA;
In re Amaranth Natural Gas Commodities Litigation, 269
7. Weather: We use a population-weighted
F.R.D. 366 (S.D.N.Y. 2010).
Anderson v. Dairy Farmers of America, Inc., No. 08-4726, average of heating and cooling degree days
2010 WL 3893601 (D. Minn. 30 September 2010). (HDD and CDD) for 16 major US cities,
In re Commodity Exchange, Inc., Silver Futures and Options
Trading Litigation 775 F. Supp. 2d 1382 (S.D.N.Y. Filed using HDD and CDD data from NOAA
on 8 February 2011). and population data from the US Census
In re Platinum and Palladium Commodities Litigation, No. Bureau. We also use an HDD and CDD
10-Civ.3617(WHP) (S.D.N.Y. filed 30 April 2010).
CFTC v. Parnon Energy Inc., No. 1:11-cv-03543 (S.D.N.Y. differential from the 15-year historical
24 May 2011). average. HDD and CDD are measures of
temperature typically in winter and
summer; they are widely used determinants
APPENDIX A of energy demand;
8. Weather forecast: We complement the
Regression Model Variables Used measures of heating and cooling degree days
in Analysis of Natural Gas Futures introduced previously with a population-
Prices weighted average of the temperature
forecasts for 16 major US cities, using
1. Futures prices of competing energy Weather Services International (WSI)
products: We use the prices of crude oil and weather forecasts and population data from
heating oil futures; the US Census Bureau.

& 2012 Macmillan Publishers Ltd. 1753-9641 Journal of Derivatives & Hedge Funds Vol. 18, 3, 254–271 269
 Saha and Petersen

9. Expected hurricane activity: We use 10. Other macro economic factors: We use the
forecasts of the ACE Index from the daily data on the level of the S&P 500 index
National Oceanographic and Atmospheric and also an index of the strength of the US
Association (NOAA) on the intensity of the dollar relative to a basket of other
upcoming hurricane season; currencies.

APPENDIX B

Table B1: Regression coefficients of natural gas futures contracts price indices, 4 January 2006
– index expiry

Variables Summer index Winter index

Coefficient t-Stat Coefficient t-stat

Intercept 2.897 0.18 2.643 0.24

(1) Summer dummy 1.901 0.11 8.965 0.64

(2) Competing products
Heating oil summer 1.172 0.77 0.181 0.10
Heating oil winter 1.637 0.69 3.612 1.70
Lag of heating oil summer 1.818 1.18 4.738 2.62
Lag of heating oil winter 1.763 0.82 0.995 0.47
Crude oil summer 0.084 1.74 0.094 1.65
Crude oil winter 0.063 0.89 0.009 0.15
Lag crude oil summer 0.089 1.80 0.115 2.02
Lag crude oil winter 0.076 1.13 0.046 0.75
(3) Forecasted energy use
Dominion energy index summer 0.750 2.17 0.727 1.89
Dominion energy index winter 0.170 0.36 0.429 1.19
Demand summer 0.000 1.64 0.000 0.39
Demand winter 0.000 0.41 0.000 1.70
(4) Production
Shut-in summer 0.007 6.77 0.004 5.39
Shut-in winter 0.005 1.63 0.002 3.60
Lag storage differential summer 0.012 5.69 0.006 2.99
Lag storage differential winter 0.001 0.44 0.002 2.57
Storage injections/withdrawals summer 0.001 0.60 0.003 1.39
Storage injections/withdrawals winter 0.002 1.20 0.000 0.04

270 & 2012 Macmillan Publishers Ltd. 1753-9641 Journal of Derivatives & Hedge Funds Vol. 18, 3, 254–271
 Detecting price artificiality and manipulation in futures markets

Table B1 continued

Variables Summer index Winter index

Coefficient t-Stat Coefficient t-stat

(5) Weather
HDD summer 0.159 2.69 0.028 0.49
HDD winter 0.018 0.51 0.078 2.47
CDD summer 0.101 1.88 0.042 0.64
CDD winter 1.222 1.67 0.192 0.31
HDD differential summer 0.172 2.88 0.005 0.09
HDD differential winter 0.019 0.58 0.052 1.83
CDD differential summer 0.084 1.78 0.009 0.16
CDD differential winter 1.223 1.74 0.231 0.39
WSI weather forecast summer 0.058 1.82 0.032 0.82
WSI weather forecast winter 0.056 1.27 0.049 1.58
ACE index summer 0.001 0.99 0.001 0.41
ACE index winter N/A N/A 0.003 0.15
(6) Economic variables
S&P 500 index summer 0.004 1.48 0.008 2.35
S&P 500 index winter 0.000 0.04 0.004 1.06
(7) Currency
US Trade weighted index summer 0.047 0.72 0.041 0.60
US trade weighted index winter 0.048 0.36 0.120 1.08

Adjusted R2 91.74% — 90.50% —

Observations 178 — 273 —

Note: Regression results are available for individual spreads and contracts upon request.

This work is licensed under a Creative Commons Attribution-NonCommercial-

NoDerivative Works 3.0 Unported License. To view a copy of this license, visit http://
creativecommons.org/licenses/by-nc-nd/3.0/

& 2012 Macmillan Publishers Ltd. 1753-9641 Journal of Derivatives & Hedge Funds Vol. 18, 3, 254–271 271