The Impact of Index Funds in

Commodity Futures Markets:

A Systems Approach
DWIGHT R. SANDERS AND SCOTT H . IRWIN

I
DWIGHT R . SANDERS nvestment in long-only commodity index and Whaley [2010]; Sanders and Irwin [2010;
is a professor in the funds soared over the last five years. Some 2011]). However, a concern with these studies
Department of Agri-
refer to this surge and its attendant impacts is the power of the time-series statistical tests
business Economics at
Southern Illinois Univer- as thefinancializatiott of commodity futures used, which may lack the statistical power nec-
sity in Carbondale, IL. markets. In view of the scale of this invest- essary to reject the null hypothesis, because
dwight$@siu.edu ment—in excess of $250 billion in 2008—it the dependent variable—the change in futures
is not surprising that a worldwide debate has prices—is extremely volatile. Another issue is
SCOTT H . IRWIN
ensued about the role of index funds in com- that direct tests of the relationship between
is the Lawrence J. Norton
modity futures markets. On one side, champi- index fund positions and price movements in
Chair of Agricultural
Marketing in the Depart- oned by Masters [2008], are those who declare energy futures markets have been hampered
ment of Agricultural and that the "wall" of investment money works by the lack of publically available data on posi-
Consumer Economics at to drive prices away from fundamental value. tions of index funds in these markets.
University of Illinois in Others (e.g., Krugman [2008]; and Irwin, The objective of this research is to provide
Urbana-Champaign, IL.
Sanders, and Merrin [2009]) are dismissive new empirical evidence on the market impact
sirwin@illinois.edu
of the notion that commodities are mispriced of commodity index funds that addresses the
in liquid and efficient markets and point out deficiencies found in previous studies. Our
logical inconsistencies in the bubble argument empirical analysis uses new data from the
and several contradictory facts. Commodity Futures Trading Commission
Several studies have been completed (CFTC) contained in the "Disaggregated
recently in an attempt to sort out which side Commitments of Traders" (DCOT) report.
of the index fund debate is correct. A few This report includes data on the positions
conclude that commodity index funds have held by swap dealers—which are assumed to
impacted commodity futures prices (e.g., reflect index-type investments since much
Gilbert [2009; 2010]; Einloth [2009]; and of the commodity index-based investment
Tang and Xiong [2010]). Irwin and Sanders is delivered through swaps'—in a variety of
[2011] argue that the data and methods used markets including energy futures markets. We
in these studies are subject to a number of focus on a group of 14 grain, livestock, soft,
important criticisms that limit the degree and energy futures markets. Bivariate Granger
of confidence one can place in the results. causality tests are used to investigate lead—lag
Most studies find little or no evidence of a dynamics between index fund positions and
relationship between index fund positions futures returns (price changes) or price vola-
and movements in commodity futures prices tility in each commodity futures market. In
(e.g., Buyuksahin and Harris [2009]; Stoll addition, a new systems approach to testing

40 THE IMPACT OF INDEX FUNDS IN COMMODITY FUTURES MARKETS: A SYSTEMS APPROACH SUMMER 2011
lead-lag dynamics is introduced and applied. The systems Swap dealers (SD) are defined by the CFTC as
approach improves the power of statistical tests by taking those traders who deal primarily in swaps and hedge
into account the contemporaneous correlation of model those transactions in the futures market. Processors and
residuals across markets and allows a test of the overall merchants (PM) include traditional commercial users—
impact of index funds across markets. processors, and producers of the commodity who are
actively engaged in the physical markets and are using the
DISAGGREGATED COMMITMENTS futures to hedge associated price risks. Managed money
OF TRADERS REPORT (MM) represents positions held by commodity trading
advisors, commodity pool operators, and hedge funds,-*
The CFTC has long had a large trader reporting who manage and conduct futures trading on the behalf
system for the purpose of detecting and deterring futures of clients. Other reportable (OR) are noncommercial
and options market manipulation (Fenton and Marti- traders who are large enough to report, but do not fit
naitas [2005]). Positions must be reported to the CFTC into one of the other categories.
on a daily basis if they meet or exceed reporting levels. It is important, especially for the energy futures
A weekly snapshot of the position data is compiled in markets, to understand the manner in which index fund
aggregate form and released to the general public as the positions are categorized in DCOT reports. The first
"Commitment of Traders" report (COT). The COT notable point is that pension and other investment funds
pools traders into two broad categories (commercial and that place index investment directly into the futures mar-
noncommercial), all contract maturities are aggregated kets rather than going through a .swap dealer are classified
into one open-interest exhibit, and the report is released as managed money (MM) or other reportables (OR) in
each Friday with the data as of the end-of-day on the the DCOT. At least in agricultural futures markets, this
preceding Tuesday. leakage into other categories is rather small; Sanders,
In response to requests for more information about Irwin, and Merrin [2010] show that nearly 85% of index
the composition of open interest, the CFTC began pub- positions are held by swap dealers in these markets. Of
lishing the DCOT report in September 2009 and ulti- potentially greater concern, a swap dealer's position in
mately provided historical data back to June 2006 (CFTC the futures market may represent a variety of over-the-
|2009J).= The DCOT reports breaks down combined counter (OTC) positions. That is, how active are the
futures and delta-adjusted options total open interest swap dealers in nonindex-related swaps? Calculations by
(TOI) as follows in Equation 1: the CFTC [2008] suggest that for commodities, such as
wheat, over 90% of the swap dealer long positions rep-
resent index traders; whereas, in crude oil it may be as
[SDL + SDS + 2(SDSP)] + [MML + MMS + 2(MMSP)] +
low as 41%. The difference between the two extremes
[PML+PMS] + [ORL -H ORS + 2(ORSP)] reflects the amount of nonindex-related OTC or swap
Kcponing trading. Wheat, for example, has very little OTC trading
+ [NRL + NRS] = 2(TOI), (1) other than positions reflecting index-related positions.
Non-Reporting Conversely, crude oil has very active OTC markets used
by index funds, hedgers, and other speculators. There-
where reporting traders (those traders with positions fore, a swap dealers' overall long position will reflect a
over a specific size) are disaggregated into processors and mix of index, hedging, and other speculative positions.
merchants (PM), swap dealers (SD), managed money Swap dealers net these various positions internally
(MM), and other reportables (OR). Positions are divided (offset long and short positions on their own books),
into long (L), short (S), and spreading (SP) as indicated and then look to lay off the residual risk in the futures
by the corresponding suffixes. For example, the SDL, market. As a result, a swap dealer's long position in
SDS, and SDSP are the swap dealers' long, short, and the futures market may represent only a portion of the
spreading positions, respectively. Spreading positions are overall index investments to which they are party (see
simply offsetting long and short positions in the same Exhibit 1 for a hypothetical example).
market, but with different contract months and are Certainly, it can be argued that the net position—
largely considered market neutral. representing the residual component, which must be

SUMMER 2011 THEJOURNAL OF ALTERNATIVE INVESTMENTS 41

EXHIBIT 1 where p^ is the futures price of the nearest-to-expiration
A Hypothetical Example of Swap Dealers contract on Tuesday of each week corresponding with
Hedging Activities the DCOT as-of position dates. In order to avoid dis-
tortions associated with contract rollovers, p^, in the log
Long OTC Non-index Swap
Exposure
Short OTC Non-Index Swap
Exposure
relative price return is always calculated using futures
(+70 Futures Contract (-40 Futures Contraci
Equivalents) Equivalents) prices for the same nearest-to-expiration contract as p^.
Rollover dates for the 12 agricultural commodities occur
Long OTC Index Swap Exposure
Short OTC Index Swap on or before the first of the delivery month, such that
Exposure
I
(^lOFuturesContract
Equivalents)
(-50 Futures Contract no return is calculated using prices during the delivery
Hquivak-nts)
month. For the energy markets, the rollover occurs on
or before the 16th of the month preceding the delivery
Net Exposure
(-10 Futures Contract month, also avoiding any issues associated with delivery
Equivalents)
intentions.
Two measures of volatility are computed for use
Hedged in Exchange
Traded Futures
in the empirical tests. First, a forward-looking vola-
(Long 10 Futures
tility measure, or implied volatility, is extracted from
Index Swap Exposure)
the options quotes for each market. Implied volatility is
provided by CRB online quote services and is calculated
in the following manner:
bedged in the commodity futures market—is most
likely to impact prices. Regardless, a large portion of
swap dealer trading represents commodity index inves- This volatility is measured by entering the prices
tors, making their positions a useful measure of their of options premiums into an options-pricing
direct trading activity in commodity futures markets model, then solving for volatility. The implied
(Buyuksahin and Harris [2009]). Indeed, in agricul- volatility value is based on the mean of the two
tural markets, Irwin and Sanders [2010] demonstrate a nearest-the-money calls and the two ncarest-
high correlation between net long positions held by swap the-money puts using the Black options-pricing
dealers and those reported for index investors, suggesting model. This value is the market's estimate of
that swap dealer positions are an acceptable, if imperfect, how volatile the underlying futures will be from
proxy for index trader positions in time-series tests. the present until the option's expiration. (CRB
[2010]).
The DCOT swap dealer position data are avail-
able weekly (reflecting Tuesday's closing positions) from
June 13, 2006 through December 29, 2009, resulting in While this method of deriving the market's implied
186 observations for empirical work. The data are col- volatility assumes that Black's pricing model is correct,
lected for 14 grain, livestock, soft, and energy futures it is a commonly accepted method of calculating the
markets: Chicago Board of Trade (CBOT) corn, CBOT volatility implied in the options markets and is generally
soybeans, CBOT soybean oil, CBOT wheat, Kansas City regarded as providing reasonably efficient volatility esti-
Board of Trade (KCBOT) wheat. New York Board of mates (e.g., Hull [2000] p. 255). The resulting volatility
Trade (NYBOT) cotton, Chicago Mercantile Exchange is expressed as an annualized percentage.
(CME) live cattle, CME feeder cattle, CME lean hogs, In Equation 3, actual or realized market volatility
NYBOT coffee, NYBOT sugar, NYBOT cocoa. New is calculated using the high—low range estimator origi-
York Mercantile Exchange (NYMEX) crude oil, and nally developed by Parkinson [1980] and widely used
NYMEX natural gas. For these markets, weekly returns by financial analysts (e.g., Sitmo |201()])-':
are calculated using nearby futures contracts adjusting
for contract rollovers as shown in Equation 2:
(3)

R = (2)

42 THE IMPACT OF INDEX FUNDS IN COMMODITY FUTURES MARKETS: A SYSTEMS APPROACH SUMMER 2011
where Zis equal to 52 in order to annualize the volatility EXHIBIT 2
estimate, n is equal to 1 week, and H. and L. are the Summary Statistics, Net Long Position Held by
high and low prices for the week. Specifically, H and Swap Dealers (number of contracts), June 2006 to
L are extracted from the daily highs and lows of the December 2009
nearest-to-expiration futures contract recorded from
Market Mean Maximum Minimum St. Dev.
Wednesday through Tuesday (inclusive) corresponding
Com 313,172 430,100 163,606 77,941
with the DCOT as-of position dates. The same contract Soybeans 121,557 193,888 73,898 27,892
rollover rules are used for computing realized volatility Soybean Oil 61,453 89,502 27,442 16,234
and returns. CBOT Wheat 142,550 189,217 91,681 25,373
KCBOT Wheat 22,073 33,863 9,952 6,906
Prior to formal statistical modeling, it is important Cotton 72,092 118,380 42,637 16,797
to determine if the series in question are stationary (fixed Live Cattle 88,844 128,967 65,368 16,351
Feeder Cattle 4,161 6,723 1,730 1,194
mean and variance). If not, then tests of statistical sig- Lean Hogs 69,149 114,377 36,326 16,858
nificance may be invalid (Enders [1995] p. 216). Here, Coffee 37,179 56.959 21,667 8,718
we follow the standard testing procedure popularized by Sugar 132,099 271,255 -32,149 81,371
Cocoa 8,380 16,474 -5,103 4,763
Dickey and Fuller to test the null hypothesis of a unit Crude Oil 40,912 106,176 -10,534 27,504
root. The results suggest that returns and volatility are Natural Gas 49,018 253,500 -67,553 78,063
stationary, while the swap dealer positions need to be
differenced to be stationary. So, in the following time-
To gauge the relative size of the swap dealer posi-
series models, the swap dealer positions will be differ-
tions in each market, the percentage of the long-side of
enced, while returns and volatility will be used in levels
each market held by swap dealers is also tabulated. The
to achieved balanced regression equations (same order of
summary statistics for percentage of long positions are
integration on the right and left side of the equations).
provided in Exhibit 3. The percentage of long positions
held by DCOT swap dealers can vary over a wide range.
EMPIRICAL METHODS AND RESULTS In cotton, the average is 31.1%, but it ranges from a
maximum of 41.9%) to a minimum of 22.1%. Likewise, in
Arguments that index funds impact commodity CBOT wheat, live cattle, and lean hogs the swap dealers
prices focus primarily on the role of long-only index (index positions) can be at times over 40%i of the long
funds that purchase commodity futures as an asset class. side of the market. Even the energy markets show rela-
Indeed, Masters and White [2008] suggest that this is tively high maximum levels, with the percentage of long
essentially a demand shift that pushes up prices. In a positions for swap dealers having a maximum of 45.0%
demand framework, the relevant measure of positions in crude oil and 38.6% in natural gas. It is often these
is that of quantity. Therefore, the analysis that follows
concentrates on swap dealers' net positions (long posi-
tions-short positions) reported in the DCOT report as EXHIBIT 3
a proxy for their overall buying or demand."' Summary Statistics, Percentage of Long Positions
The summary statistics for DCOT swap dealers net Held by Swap Dealers, June 2006 to December 2009
long positions are shown in Exhibit 2. They indicate the Maricet Mean Maximum Minimum St. Dev.
size of index positions had a wide range over the sample Com 22.5 29.0 17.0 2.4
period. For instance, in the corn futures market the swap Soybeans 23.7 28.2 19.8 1.6
Soybean Oil 23.9 30.5 17.4 2.9
dealer position had a maximum of 430,100 contracts CBOT Wheat 37.4 44.5 31.5 2.9
and a minimum of 163,606 for a range of over 100%. In KCBOT Wheat 18.4 24.6 11.4 3.2
the sugar, cocoa, crude oil, and natural gas markets the Cotton 31.1 41.9 22.1 4.7
Live Cattle 33.7 41.0 27.4 4.0
effect of swap dealer netting is seen in the occurrence of Feeder Cattle 15.6 22.2 10.1 2.9
a net short position (the minimum net long position is Lean Hogs 36.0 42.6 26.7 3.4
negative). This confirms that the swap dealer positions Coffee 24.0 36.8 17.6 4.7
Sugar 25.9 34.2 20.5 3.7
in these markets contain positions other than those held Cocoa 14.6 23.8 9.8 3.4
by long-only index funds. Crude Oil 37.5 45.0 30.4 4.2
Natural Gas 30.2 38.6 22.3 5.1

SUMMER 2011 THEJCÍURNAL OF ALTERNATIVE INVESTMENTS 43

large relative position sizes that spawn accusations of price In this research, the power of causality tests based on
influence. But, clearly, both the net positions (Exhibit 2) Equation (4) is increased by modeling the K markets as a
and the percentage of the long positions held (Exhibit 3) system of seemingly unrelated regressions. Since the error
can swing over wide ranges. The important question is term, 8^ ^, in Equation (4) is correlated across markets the
whether these swings in positions affect prices. power of causality tests can be increased by employing
Hamilton [1994] suggests the direct, or bivariate. a GLS estimator within Zellner's seemingly unrelated
Granger test for examining the lead-lag or causal rela- regression (SUR) framework (see Harvey [1991] p. 66).''
tionship between two series. Granger causality is a stan- Previous researchers have used the SUR framework to
dard linear technique for determining whether one time increase the power of market efficiency tests in foreign
series is useful in forecasting another. In our case, the exchange markets (Frankel [1980]; Bilson [1981]). Under
time series of interest are market measures of returns {R), the SUR approach, GLS parameter estimates are the best,
implied volatility (IV), and realized volatility (RV). The linear, unbiased, coefficient estimates. The efficiency
causal variable is the net swap dealer position (NET). gains over OLS estimates increase with the correlation
As an example, consider the causal relationship between between the residuals (e ^) and with the number of equa-
market returns and net positions. Under the null hypoth- tions. Harvey also points out additional efficiency gains
esis that index traders' net position does not Granger cause by imposing the constraint of equal parameters across
market returns, the following linear regression (Equation equations, where appropriate. Moreover, estimating
4) is estimated for each market (k) over time (i): Equation (3) as a system of K regressions also allows for
a more complete set of hypothesis tests.
Traditional bivariate causality in a single market, k,
is tested under the null hypothesis in Equation (4) that
traders' net positions (NET) cannot be used to predict
where the lag structure (m, ti) for each market is deter- (do not lead) market returns: ^ ^= 0 for all /. A rejec-
mined by a search procedure over m = 4 and n — 4 tion of this null hypothesis, using an F-test of the stated
using OLS and choosing the model that minimizes the restriction, provides direct evidence that trader positions
Schwartz criteria to avoid overparmeterization (Enders are indeed useful for forecasting returns in that market.
[1995] p. 88). If the OLS residuals demonstrate serial In order to gauge the aggregate impact of trader posi-
correlation (Breusch-Godfrey Lagrange multiplier test), tions in that market, the null hypothesis that Z"^, ß ^, = 0
additional lags of the dependent variable are added until in each k market will reveal the cumulative directional
the null of no serial correlation cannot be rejected. impact of traders positions on returns, if any. Clearly, in
Previous time-series studies of lead—lag relation- the event that the lag structure is n = 1, then the test of
ships between index fund market participation and com- null hypothesis that 2^"=,ß,t. = 0 is equivalent to a simple
modity futures returns conduct tests market-by-market test on the parameter restriction that ß| ^ = 0.
when data on multiple markets are available (e.g., Stoll When estimated as a system of K equations in a SUR
and Whaley [2010]; Sanders and Irwin [2011]). As noted framework. Equation (4) allows for the testing of system-
earlier, standard time-series approaches can be criticized atic or aggregate impacts across all markets. To further
for a lack of statistical power when the dependent variable increase the efficiency gains, coefficients are restricted
demonstrates extreme volatility such as that observed across market equations where appropriate (see Harvey
for market returns. Summers [1986] demonstrates that [1991] p. 69). Specifically, the strategy for selecting the
traditional time-series tests may also have low statistical restricted SUR model follows the sequential testing pro-
power against a random walk null hypothesis if the true cedure outlined by Harvey (p. 186) where the most gen-
data-generating process is driven by fad or sentiment. eral model is first estimated (no crossmarket parameter
Researchers have proposed alternative, more powerful restrictions). Then, using a Wald test, the hypothesis of
time-series tests (e.g., Jegadeesh [1991]) and they have equal parameter estimates is tested across markets. When
been applied in previous work on the market impact of the null of equal parameter estimates is not rejected, then
commodity index funds; however, they can still suffer the restriction is placed on the model. Specifically, all K
from a lack of robustness under the alternative hypoth- models are first estimated as a SUR system using the lag
esis (see Daniel [2001]). structure chosen with the OLS search procedure. Second,

44 THE IMPACT OF INDEX FUNDS IN COMMODITY FUTURES MARKETS: A SYSTEMS APPROACH SUMMER 2011
for each estimated parameter the null hypothesis that the of parameter restrictions was unable to reject that the
cross-equation parameters are equivalent is tested (e.g., ßj ^ coefficients were equivalent across markets, k, which
y, I = y, , = •••= y, J . If we fail to reject the null hypoth- results in a pooled estimation of ß^., across all Kmarkets.
esis at the 10% level, then that parameter restriction is As a consequence, all the markets for which n = 1 have
imposed resulting in a pooled estimate or single parameter the same causality test result where ß^ = -0.0018 and it
across equations (e.g., y^.). A 10% significance level is is not statistically different from zero. The ß., ^ was only
used (instead of 5%) to be somewhat conservative in the specified and estimated for live cattle. So, this market
imposition of parameter restrictions. By pooling parame- generates an incremental result, but also fails to reject the
ters—when we fail to reject that they are equivalent—the null of no causality (/»-value = 0.2994). In this particular
number of parameter estimates is decreased and statistical system, there is a common ß^ and a single ß., ^ estimated
efficiency is further enhanced. for live cattle. Therefore, the test for systemwide causality
To illustrate, consider again Equation (4) and two (ß|. =ß, |,,= 0) happens to be equivalent to the test for just
extreme examples. First, the most restrictive case, where the live cattle market. So, the system results likewise
the null of equivalency across markets is not rejected for
all parameters. In this case, the parameter estimates are all
pooled. Then, causality testing proceeds using the pooled EXHIBIT 4
parameter under the null, H^^:^, = 0 for all / (where Granger Causality Test Results for DCOT Net Swap
/. represents a parameter common across the K markets). Dealer Net Positions Do Not Lead Returns, June 2006
Rejection of the null hypothesis suggests that positions to December 2009
lead returns across the system (all markets jointly). Addi-
^a. b r e a c h market, fe.
tionally, the aggregate impact that 2"^,ß . = 0 can be tested
and time, /.
for the system as a whole. In this naost restrictive case,
there are no tests for individual markets as the pooled
result applies equally to all markets. /»-value Estimate /»-value
Market, k HI, n ß;=O,Vj
At the other extreme, assume that we reject the null
hypothesis of equal crossmarket parameters in all cases. Com 1,1 0.1694 -0.0018
This is the least restrictive case where none of the param- Soybeans 1,1 0.1694 -0.0018
eter estimates are pooled and individual coefficients are Soybean Oil 1,1 0.1694 -0.0018
CBOT Wheat 1,1 0.1694 -0.0018
estimated for all iC markets. In this case, the null hypoth-
KCBOT Wheat 1, 0.1694 -0.0018
esis is tested for each k under the null, H^^^ ^^ — 0 for all Cotton 0.1694 -0.0018
1,
/, as well as the aggregate impact, 2^"=|ß,j, =0, is tested Live Cattle 0.2994 -0.0017 0.6868
2,2
for each k market. Moreover, the SUR estimation allows Feeder Cattle 2, 0.1694 -0.0018
for the testing of systemwide causality, H,:ß ^, = 0 for all Lean Hogs 1, 0.1694 -0.0018
7 and k, and for the systematic impact across markets, Coffee 1, 0.1694 -0.0018
Z|,.^iZ"^iß^^, = 0. This is an important improvement over Sugar 1, 0.1694 -0.0018
a strictly market-by-market OLS approach to causality Cocoa 1, 0.1694 -0.0018
testing, because it allows for broader statements about Crude Oil 1, 0.1694 -0.0018
systematic impacts. Natural Gas 3, 0.1694 -0.0018
The first set of empirical causality tests focus on /»-value Estimate /»-value
the null hypothesis that net positions do not lead rettirns.
In this case, the independent variable in Equation (4) is System 0.2994 0.0017 0.6868
market returns (R) and the explanatory variable is the
change in the net position (ANET). Exhibit 4 shows the Notes: Y. ß, values are taken to the Í 0 ' power. The models are estimated
SUR test results for the null hypothesis that the swap across the K markets as an SUR system. Wald tests could not reject
the following crossmarket coefficient restrictions: a , = a , = ••• = a ^ ;
dealer positions do not lead returns for each market.
Y2., = Y2.2 = ••• = 7,,^.- ß , , , = ß,,..> = ••• = ?,.Kßr M K 'markets. These
In every market, except live cattle, changes in restrictions are imposed on the system and the common coefficients are
net swap dealer positions enter the model specification estimated as a single pooled parameter across all K markets.
with the minimum lag structure of w = 1. The testing

SUMMER 2011 THEJOURNAL OF ALTERNATIVE INVESTMENTS 45

show a cumulative impact that is not statistically different EXHIBIT 5
from zero. In total, there is no evidence that the net posi- Granger Causality Test Results for DCOT Swap
tions held by DCOT swap dealers lead market returns.^ Dealers Net Positions Do Not Lead Implied
The second set of causality tests focus on the Volatility, June 2006 to December 2009
linkage between net positions and market volatility.
While there is a growing body of empirical w^ork on e,,, for each market, k.
the impact of trader positions on returns, there is less and time, f.
empirical evidence with regard to trader positions and
/»-value Estimate p-value
market volatility. Granger causality tests are run in a
similar fashion to those used to test the linkages between
Market m, n ß, = O,Vj
Com 1, 0.1822 -0.3630
positions and returns. Equation (5) is used to test the null Soybeans 0.0000 2.9400
3,
hypothesis that net trader positions {NET^ do not lead Soybean Oil 4, 0.1688 -1.0500
implied volatility {IV) in an SUR framework: CBOT Wheat 2, 0.2661 0.6510
KCBOT Wheat 2, 0.8405 -0.3430
Cotton 3, 0.1614 2.6000
-¿ß.,AN£T,_., + e , (5) Live Cattle 3, 0.9342 0.0538
i=t Feeder Cattle 1, 0.3466 -3.1100
Lean Hogs 1, 0.5871 2.5700
The results for estimating Equation (5) are shown in Coffee 2, 0.2567 -2.8600
Exhibit 5, where the null of equal coefficients on lagged Sugar 2, 0.3046 -O.6420
Cocoa 2, 0.2979 1.9600
positions is not rejected." Therefore, individual ß ^^ coef-
Crude Oil 4, 0.5051 0.2700
ficients are estimated for each market, k'. In the system, Natural Gas 1. 0.0045 -0.8260
there are two rejections of the no causality null hypothesis
at the 5% level across the 14 markets. The null hypothesis p-value Estimate p-value
of no causality is rejected at the 5% level for soybeans and
also for natural gas. The estimated direction of the impact System 0.0005 -4.0300 0.5821
is negative in both markets, suggesting that larger swap
dealer positions actually lead to lower implied volatility. Notes: X ß, t'alties are taken to the It/ power. The models are estimated
The two individual market rejections result in a rejec- across the K markets as an SUR system. Wald tests could not reject
the foilowing crossmarket coefficient restrictions: Y^, = Y2 2 ~ ' " ~ Yj /c'
tion of the null hypothesis of no causality for the system Y,, = Yj, = • = Y, ^fit all K markets. These restrictions are imposed on
(p-value = 0.0005). However, the mixed signs on the coef- the system and the common coefficients are estimated as a siii(;te pooled
ficients result in a cumulative directional impact (—4.0300) parameter across aii K markets.
that is not statistically different from zero (p-value =
0.5821), suggesting that while there are market specific cocoa at the 5% level. Coffee is the only market with two
impacts (soybeans and natural gas) there is not a pervasive lags of positions and it has a negative cumulative impact
directional impact across the system of markets. (p-value = 0.0581). Causality is also tested for the system
The prior set of tests focused on implied or for- as a whole utilizing the pooled estimation. The null of no
ward-looking volatility. Granger causality tests are also causality across the market system is rejected at the 5%
performed to see if positions lead actual or realized level and the cumulative impact is systematically negative
volatility (RV) as measured by Parkinson's high—low (/)-value = 0.0131).
estimator in Equation (6): The results in Exhibits 4 and 5 are reasonably con-
sistent in suggesting that higher levels of index trader or
.ANET (6)
swap dealer participation in the markets are associated
with lower levels of volatility in subsequent weeks.
/=!
While there are some significant individual market find-
Exhibit 6 presents the estimates and hypothesis ings, the strongest results are manifest in the systemwide
tests associated with Equation (6). Again, individual ß ^^ rejections. Specifically, in Exhibit 6, a negative leading
coefficients are estimated for each market, k. The SUR relationship is suggested between index trader positions
system rejects the null of no causality in soybeans and and realized volatility across all markets.

46 THE IMPACT OF INOEX FUNDS IN COMMODITY FUTURES MARKETS: A SYSTE.MS APPROACH SUMMER 2011
EXHIBIT 6 through 2008. At the same time, commodity prices also
Granger Causality Test Results for DCOT Swap increased rather dratnatically with a number of com-
Dealer Net Positions Do Not Lead Realized modity prices hitting all-time highs in the first half of
Volatility, June 2006 to December 2009 2008. Not surprisingly, this tandem of events brought
commodity index funds under intense scrutiny and a
^,, for each market.
.1 (=1
worldwide debate ensued as to the role of index funds
k, and time, t. in commodity futures markets. Those who believe index
funds were responsible for a bubble in commodity futures
/»-value Estimate p-value
Market m,n ßy = O.Vj
prices (e.g.. Masters |2008]) make what seems like an
Iß; 2ßy = 0
obvious argument—the sheer size of index investment
Com 2,1 0.8258 0.2000
Soybeans 4,1 0.0242 -3.3700 overwhelmed the normal functioning of these markets.
Soybean Oil 2,1 0.5347 -0.9500 Importantly, an empirical linkage must be made between
CBOT Wheat 2,1 0.6975 0.4370 commodity index fund positions and prices or there is
KCBOT Wheat 3,1 0.1308 -5.5000 no obvious mechanism by which a bubble can form.
Cotton 3,1 0.9358 0.2340
Live Cattle 3,1 0.0600 -2.4600
Therefore, continued empirical tests are an important
Feeder Cattle 3,1 0.5317 -5.8200 element of this debate.
Lean Hogs 3,1 0.1531 3.7900 Our empirical analysis relies on data compiled by the
Coffee 1,2 0.1568 -11.8200 0.0581
Sugar
CFTC in the "Disaggregated Commitments of Traders"
3,1 0.8018 -0.3200
Cocoa 4,1 0.0420 -12.0300 report. In particular, the positions held by swap dealers
Crude Oil 3,1 0.0889 1.0500 are assumed to reflect index-type investments. The data
Natural Gas 1,1 0.5975 0.4610 are used to test if index funds impact either returns or
p-value Estimate p-value price volatility across 14 grain, livestock, soft, and energy
1 futures markets. The sample period begins on June 13,
System 0.0408 -36.1000 0.0131 2006, and ends on December 29, 2009, yielding a total of
186 weekly observations for analysis. Bivariate Granger
Notes: S ß, I'alues are taken to the Iff power. The models are estimated causality tests are used to investigate lead-lag dynamics
across the K markets as an SUR system. Wald tests could not reject the between index fund positions and futures returns (price
folloiinnf; crossmarket coefficient restrictions: 7,, = 7, , = •• = 7,, ^. for all
K markets. These restrictions are imposed on the system and the common
changes) or price volatility in each commodity futures
coefficients are estimated as a single pooled parameter across all K markets. market. In addition, a new systems approach to testing
lead-lag dynamics is introduced and applied. The systems
Some care must be taken when interpreting causality approach improves the power of statistical tests by taking
results (see Newbold [1982]). The lead-lag relationship into account the contemporaneous correlation of model
does not necessarily imply that index trader positions residuals across markets and allows a test of the overall
have some dampening impact on implied volatility. It impact of index funds across markets.
may simply indicate that larger index fund or swap dealer Examination of the data characteristics and sub-
positions coincide with other fundamental factors that sequent empirical modeling lead to the following gen-
portend lower volatility levels. Still, the results provide eral conclusions. First, there is no convincing evidence
evidence that is contrary to popular notions that index that positions held by index traders (as represented by
investments exacerbate market volatility. Importantly, swap dealers) impact market returns. The system of
the results for market returns show no evidence linking Granger-style causality tests fails to reject the null hypoth-
futures prices to index investments suggesting that there esis that that trader positions do not lead market returns.
is not a mechanism by which these market participants Hence, there is no evidence of a linkage between index
could create a price bubble. trader positions in commodity futures markets and price
levels. Second, larger net long positions by index traders
SUMMARY AND CONCLUSIONS lead to lower market volatility in a Granger sense. There
is a consistent tendency to reject the null hypothesis that
Money flowing into the commodities markets as index trader positions do not lead market volatility. The
an alternative investment increased rapidly from 2004 direction of the impact is routinely negative. While index

SUMMER 2011 THEJOURNAL OF ALTERNATIVE INVESTMENTS 47

positions lead to lower volatility in a statistical sense, it ^We also ran Granger causality tests usmg index trader
is possible that trader positions coincide with some other positions for 12 agricultural futures markets found in CFTC
fundamental variable that is actually causing the lower "Supplemental" (CIT) reports. Similar hypothesis test results
market volatility. Still, this result is contrary to popular were found using the CIT data. See Irwin and Sanders [2010].
notions about index traders increasing market volatility. "Equations (5) and (6) were also estimated with monthly
dummy variables to test for seasonality in volatility. For this
The policy implications of the results are straightfor-
sample period, the null hypothesis of equal volatility across
ward: current regulatory proposals to limit speculation—
months was not rejected, so the monthly dummy variables
especially on the part of index funds—are not justified and were not included in the final model specifications.
likely will do more harm than good. In particular, lim-
iting the participation of index fund investors would rob
REFERENCES
the commodity futures markets of an important source of
liquidity and risk-absorption capacity at a time when both Bilson, John F.O. "The 'Speculative Efficiency' Hypothesis."
are in high demand. Most likely, increased U.S. regulation Journal of Business, 54 (1981), pp. 435-451.
will force commodity index funds off-shore to rapidly
growing international futures markets that have a lower Buyuksahin, Bahattin, andJ.H. Harris. "The Role of Specu-
regulatory burden. Or, more ominously, tighter position lators in the Crude Oil Futures Markets." Working paper,
limits on speculation in commodity futures markets com- U.S. Commodity Futures Trading Commission, 2009.
bined with the removal of hedge exemptions could force Available at URL: http://papers.ssrn.com/sol3/papers.
commodity index funds into thinly traded cash markets, cfm?abstract_id=1435042.
where truly chaotic results could follow.
Commodity Futures Trading Commission (CFTC). "Staff
Report on Commodity Swap Dealers & Index Traders
ENDNOTES with Commission Recommendations," 2008. Available at
http://www.cftc.gOv/ucm/groups/public/@newsroom/
The authors are indebted to Hongxiajiao for her assis- documents/file/cftcstaffreportonswapdealers09.pdf
tance in collecting the data for this study.
'See CFTC [2008]. . "Disaggregated Commitments of Traders Reports:
^Since 2007, the CFTC has also released the, "Supple- Explanatory Notes," 2009. Available at littp://www.cftc.gov/
mental Commodity Index Traders," (CIT) reports, which ucm/groups/public/@newsroom/documents/file/disaggre-
break out the positions of index traders for 12 agricultural gatedcotexplanatorynot.pdf.
markets. See Irwin and Sanders [2010] for a detailed com-
parison of the CIT and DCOT reports. CRB Trader. "Option Implied Volatility and Futures His-
*Funds that identify themselves as hedge funds may be torical Volatility." Available at http://www.crbtrader.coni/
registered with the CFTC and, thus, technically would be support/options.asp.
more accurately described as CTAs. Registration requirements
can be found at www.NFA.futures.org/NFA-registration/ Daniel, Kent. "The Power and Size of Mean Reversion Tests."
CTA/index.html. Journal ol Empirical Finance, 8 (2001), pp. 493-535.
••See Spurgin [2001] for a discussion of bias and effi-
ciency of Parkinson estimators. Einloth, James T. "Speculation and Recent Volatility
^We tested other position measures including the per- in the Price of Oil." Working paper. Division of Insur-
centage of long positions as shown in Exhibit 3. The empirical ance and Research, Federal Deposit Insurance Corpora-
results were generally not sensitive to the position measure tion, 2009. Available at http://papers.ssrn.coni/sol3/papers.
and the results presented here are representative of those cfm?abstract_id=1488792.
found for all of the position measures. For a more thorough
description of the relevant position measures, see Irwin and Enders, Walter. Applied Econometric Time Series. New York,
Sanders [2010]. NY: John Wiley and Sons, 1995.
''The correlation matrix of residuals for the OLS esti-
mates of (4) showed correlation coefficients ranging from a Fenton, John, and G. Martinaitas. "Large Trader Reporting:
high of 0.96 (CBOT wheat and KCBOT wheat) to a low of The Great Equalizer." Futures //ií/w.sír}', July/August 2005, pp.
—0.12 (corn and feeder cattle) with an average of 0.28. 34-39.

48 THE IMPACT OF INDEX FUNDS IN COMMODITY FUTURES MARKETS: A SYSTEMS APPROACH SUMMER 2011
Frankel, Jeffrey A. "Tests of Rational Fxpectations in the Masters, Michael W., and A.K. White. "The Accidental Hunt
Forward Exchange Market." Southern Economic Journal, 46 Brothers: How Institutional Investors are Driving up Food
(1980), pp. 1083-1101. and Energy Prices," 2008. Available at http://accidentalhunt-
brothers.com.
Gilbert, Christopher L. "Speculative Influences on Com-
modity Futures Prices, 2006-2008." Working paper. Depart- Newbold, Paul. "Causality Testing in Economics." In Time
ment of Economics, University of Trento, 2009. Available Series Analysis: Theory and Practice I, edited by O.D. Anderson,
at http://www.nottingham.ac.uk/economics/documents/ pp. 701-716. Amsterdam, Netherlands: North Holland Pub-
seminars/senior/christopher-gilbert-04-11-09.pdf lishing Company, 1982.

. "How to Understand High Food Prices." Journal of Parkinson, Michael. "The Extreme Value Method for Esti-
Agricultural Economics, 61 (2010), pp. 398-425. mating the Variance of the Rate of Return." Journal of Busi-
ness, 53 (1980), pp. 61-65.
Hamilton, James D. Time Series Analysis. Princeton, NJ:
Princeton University Press, 1994. Sanders, Dwight R., and S.H. Irwin. "A Speculative Bubble
in Commodity Futures Prices? Cross-Sectional Evidence."
Harvey, Andrew C. The Econometric Analysis of Time Series, Agricultural Ecottomics, 41 (2010), pp. 25-32.
2nd ed. Cambridge, MA: MIT Press, 1991.
. "New Evidence on the Impact of Index Funds in U.S.
Hull, John C. Options, Futures, and Other Derivatives, 4th ed. Grain Futures Markets." Canadian Journal of Agricultural Eco-
Upper Saddle River, NJ: Prentice Hall, 2000. nomics, 59 (2011), forthcoming.

Irwin, Scott H., and D.R. Sanders. "The Impact of Index and Sanders, Dwight R., S.H. Irwin, and R.P. Merrin. "The
Swap Funds in Commodity Futures Markets: Preliminary Adequacy of Speculation in Agricultural Futures Markets:
Results (Annex)." Trade and Agriculture Directorate, Com- Too Much of a Good Thing?" Applied Economics Perspectives
mittee for Agriculture, Organization for Economic Coopera- and Policy, 32 (2010). pp. 77-94.
tion and Development, 2010. Available at http://www.oecd.
org/officialdocumentsearch/0,3673,en_2649_201185_l_l_ Sitmo: Resources for Financial Engineers. "Historical
l_l_l,OO.html. High-Low Volatility: Parkinson." 2010. Available at http://
www.sitmo.com/eq/173.
. "Index Funds, Financialization, and Commodity
Futures Markets." Applied Economics Perspectives and Policy, 33 Spurgin, Richard B. "Variance Estimators Using the Par-
(2011), forthcoming. Available at http://aepp.oxfordjournals. kinson Approach: A Note on Bias and Efficiency." The Journal
org/content/early/2011/02/02/aepp.ppq032.full. of Alternative Investments, 4 (2001), pp. 69-71.

Irwin, Scott H., D.R. Sanders, and R.P. Merrin. "Devil or Stoll, Hans R., and R.E. Whaley. "Commodity Index
Angel? The Role of Speculation in the Recent Commodity Investing and Commodity Futures Prices." Journal of Applied
Price Boom (and Bust)." Journal of Agricultural and Applied Finance, 20 (2010), pp. 7-46.
Economics, 41 (2009), pp. 393-402.
Summers, Lawrence H. "Does the Stock Market Rationally
Jegadeesh, Narasimhan. "Seasonality in Stock Price Mean Reflect Fundamental Values." Journal of Finance, 41 (1986),
Reversion: Evidence from the U.S. and the U.K." fouriuil of pp. 591-601.
Finance, 46 (1991), pp. 1427-1444.
Tang, Ke, and W. Xiong. "Index Investing and the Finan-
Krugman, Paul. "More on Oil and Speculation." New York cialization of Commodities." Working paper. Department of
Times, May 13, 2008. Available at http://krugman.blogs. Economics, Princeton University, 2010. Available at http://
nytimes.com/2008/05/13/more-oii-oil-and-speculation. www.princeton.edu/~wxiong/papers/commodity.pclf

Masters, Michael W. "Testimony before the Committee on

Homeland Security and Government Affairs, U.S. Senate." To order reprints of this article, please contact Dewey Palmieri
May 20, 2008. Available at http://hsgac.senate.gov/public/_ at dpalmieri@iijournals.com or 212-224-3675.
files/052008Masters.pdf

SUMMER 2011 THEJOURNAL OF ALTERNATIVE INVESTMENTS 49

©Euromoney Institutional Investor PLC. This material must be used for the customer's internal business use
only and a maximum of ten (10) hard copy print-outs may be made. No further copying or transmission of this
material is allowed without the express permission of Euromoney Institutional Investor PLC. Source: Journal of
Alternative Investments and http://www.iijournals.com/JAI/Default.asp.