Does speculation drive commodity future
prices?
Abstract:
This thesis investigates whether speculation drives commodity future prices. A group of 17 energy commodities, grain commodities, soft commodities, livestock commodities and precious metals is analyzed from 1993 to 2013. The data indicates that speculation, proxied by the Working T Index, is generally not a significant driver of commodity future prices. However, for some less liquid commodity markets, significant results have been found.
JEL classifications: C22, G13, Q02
Author: Aart Hatzmann MSc Finance
University of Groningen, faculty of Economics and Business Date: January 2013
1. Introduction
In recent years, commodity prices have reached all-time highs while exhibiting increased volatility. For example, the price of crude oil showed a level of approximately $88.96 per barrel in February 2008 and an increased level of approximately $145 in November of the same year. In December 2008, the price dropped dramatically to almost $33 per barrel. Currently, the price of crude oil is approximately $107. Not only has the oil price reached all-time highs and exhibited increased volatility, but other commodity markets have also reached all-time highs. The corn market is one example of this. Figure I illustrates the development of corn prices since 1990.
Figure I: Corn price
Figure I represents the average weekly future price of corn traded on the CBOT for the 1993-2013 period.
Public opinion generally holds speculators responsible for surges in commodity derivative prices, which result in high food prices. Although much research has been done on the impact of speculation on commodity prices, the focus is typically on just a few commodities for a limited time span. Therefore, the current research includes all large commodity markets over the course
0 100 200 300 400 500 600 700 800 900 19 90 19 91 19 92 19 93 19 94 19 95 19 96 19 97 19 98 19 99 20 00 20 01 20 02 20 03 20 04 20 05 20 06 20 07 20 08 20 09 20 10 20 11 20 12 20 13
of many years. The key question that this paper therefore seeks to answer is whether speculation drives the prices of large commodity future markets for the period from 1993 to 2013.
From a historical perspective, this is not the first time that commodity prices have experienced increasingly high volatilities. The first major global commodity price boom was actually experienced in 1950-1951. This boom was caused by a massive inventory buildup after the Korean War (Radetzki, 2006). In 1973-1974, there was another boom; this time the boom was accentuated by harvest failure and inadequate market management of the OPEC (Radetzki, 2006).
In addition to increasing prices, the characteristics of commodity markets are also currently changing. This might be why the explanation for the commodity price rise is different than for the two previous booms. Cifarelli and Paladino (2010) state that investment funds raised their holdings in commodity markets from $13 billion in 2003 to $260 billion in mid-2008. They argue that this could be a reason for the increase in prices, which encourages further debate about the role of speculation in the marketplace. Since the commodity price spike occurred in 2008, there has been an ongoing world-wide debate about whether speculation drives commodity prices.
a mechanism that results in commodity prices, especially crude oil prices, exceeding their fundamental values.
While there are several supporters of Masters’ (2008) argument, also called “the Master hypothesis” (Irwin and Sanders, 2012a), the existing academic literature has not provided abundant acceptance of his claim. Next to that, it is important to note that the existing literature concentrates mainly on the spike in prices which occurred in 2008. It does not investigate the recent rise in prices (see Figure I).
Whereas the Master Hypothesis is mainly based on crude oil, Manera et al. (2012) found that spillovers between agriculture and energy commodities are present. Therefore, while analyzing the role of speculators in energy price activity, non-energy commodities should also be considered. Manera et al. (2012) only investigate a group of four energy commodities and a group of five agricultural commodities.
Because other researchers have found some evidence of speculation in certain commodity markets and for certain timeframes, a comprehensive overview will be offered through the inclusion of a variety of commodity groups. A total of five groups will be included: energy commodities (crude oil, heating oil and natural gas), soft commodities (coffee, sugar, cocoa and cotton), grain commodities (corn, wheat, oats and soybeans), metal commodities (gold, silver and copper) and livestock commodities (lean hogs, live cattle and feeder cattle). Furthermore, the analyses will cover a timespan which begins in 2008 and reaches 2013, due to the fact that commodities experienced a sharp rise in prices during this time.
trader positions, made available by the U.S Commodity Futures Trading Commission (CFTC), is analysed. With the use of Working’s T (1960) speculative index and by controlling certain macro-economic variables, this paper investigates whether speculation drove the prices of commodity futures between 1993 and 2013. The methodology of the current study allows for the examination of two periods that show large spikes in commodity prices. The main result of the current study, which will be further discussed in the conclusion, is that speculation does not drive the commodity future price.
The following chapters include a discussion on the current literature written on speculation in future markets and a description of the commodity markets. In the next section, the methodology of the current study will be explained, and a data description will then be provided. After the data description, the results will be presented and discussed. Finally, a summary and conclusion will be provided.
2. Background
2.1. Literature review
In the extant literature, the authors investigated a wide range of commodities. While some focus only on crude oil, others address a small group of commodities, like, for example, a group of grains. Different methods have been used on the causal relationship between speculation and commodity prices. Although most researchers do not find evidence that the activity of speculators drives the prices of commodity futures, some evidence has been found.
index of commodity prices, and this in turn could drive up future prices (Irwin and Sanders, 2012). Gilbert and Pfuderer (2012) support Irwin and Sanders (2012) because they did not find evidence either for the grain markets considered above. In addition, Stoll and Whaley (2012) find no evidence in a study investigating 12 agricultural markets. Irwin and Sanders (2012) used cross-sectional Fama-MacBeth regression tests and find very little evidence in 19 commodity markets.
Irwin et al. (2011) argue that, due to the rolling of positions, commodity index investors could put upward pressure on the spreads between future prices for the same commodity. However, they find no evidence of an increased spread for corn, soybeans or wheat markets during the roll period for index funds. Using this method, Hamilton and Wu (2011) find little evidence in 12 agricultural markets and Aulerich et al. (2012) even find negative relationships. Brunetti and Buyuksahin (2009) look at trader positions and changes in trader positions with the help of commitment of traders (COT) data made available by the CFTC. Their findings indicate no influence on corn, natural gas or crude oil returns. Buyuksahin and Harris (2011) look at lead and lag relations between prices and trader position data at daily and multiple times per day intervals. They find little evidence for a causal relationship, and they even find reverse causality, indicating that price changes lead to trader position changes.
focused on energy and agricultural commodities and also suggested that speculator activity is generally irrelevant in explaining the returns.
Kaufmann (2011) recognizes that it is difficult to measure the activity of speculators. He argues, however, that speculative expectations affected crude oil prices for several reasons. Firstly, he observes a significant increase in private US crude oil inventories since the year 2004. Secondly, he finds repeated and extended break-downs in the co-integrating relationship between spot and far month future prices. According to Kaufmann (2011), those break downs are inconsistent with arbitrage opportunities and the law of one price. Thirdly, Kaufmann (2011) finds statistical and predictive failures when using an econometric model of oil prices that is based only on market fundamentals.
All in all, evidence exploring the question of whether speculation can be held responsible for rising and fluctuating commodity prices is mixed. Despite Masters’ (2008) statement that financial speculation can be held responsible for the price spike in commodity markets, the majority of researchers in the current literature have found opposing results. Nevertheless, some researchers do find evidence in certain markets for particular time-spans. In addition, results indicate that a combination of macroeconomic fundamentals and speculation is the driver of commodity prices. Manera et al. (2012), however, did not find evidence for speculative influence after using Working’s (1960) T index in combination with macroeconomic fundamentals. Buyuksahin and Robe (2013), on the other hand, found evidence of excess speculation after using Working’s (1960) T index in the 2000-2010 period. Furthermore, some speculative influence on specific commodities has been found in other research.
the current study, which has been based on all prior research, is that the activity of speculators will not drive the future prices of many commodities. However, it is expected that, for some commodities, speculation will drive commodity future prices.
2.2. Commodity market descriptions
This section gives a short description of every commodity market considered. The descriptions include the most important characteristics of each individual commodity market. Characteristics might be commodity specific and are therefore important to discuss in order to reflect on the results. For every market description, the average open interest has been reported. The CFTC defines the open interest in their explanatory notes as the total of all futures contracts entered into and not yet offset by a transaction, a delivery or by an exercise.1
Corn: The US is the largest producer of corn, followed by China.2 In the US, corn is the largest crop in terms of acres planted and value (Fontanills, 2007). The usage of corn is primarily based in the livestock market as nutrition (Fontanills, 2007). Corn is also used as an additive in the food processing market. It is used, for example, to produce sweeteners and margarine (Fontanills, 2007). Because corn is a typical summer crop, its prices differ during the year due to seasonality (Fontanills, 2007). The average open interest for corn over the full sample period was 1.029.537 contracts (CFTC data).
Wheat: The main use of wheat is for producing flour. However, wheat is also used for the
manufacture of oil and for brewing and distillation (Fontanills, 2007). In the US, wheat is typically planted in late fall and harvested in early summer of the next year. This type of wheat is called winter wheat, which accounts for approximately 75% of the total market in the US
1
http://www.cftc.gov/MarketReports/CommitmentsofTraders/ExplanatoryNotes/index.htm
2
(Fontanills, 2007). The world-wide production of wheat is widespread (Fontanills, 2007). The average open interest for wheat over the full sample period was 295.221 contracts (CFTC data).
Soybeans: Soybeans are the third largest agricultural crop in the US (Fontanills, 2007).
Soybeans are used in the production processes of many products due to their high protein content. This also makes soybeans a good substitute for meat and dairy (Fontanills, 2007). The main producers of soybeans are the US, Brazil and Argentina (Fontanills, 2007). The average open interest for soybeans over the full sample period was 454.382 contracts (CFTC data).
Oats: Oats are also used primarily as livestock feed, and they are also used for human
consumption in the form of oatmeal and rolled oats (Fontanills, 2007). However, the oats market is not as large as the previously discussed agricultural variables (Fontanills, 2007). The average open interest for oats over the full sample period was 25.052 contracts (CFTC data).
Cocoa: Cocoa is mainly produced in Africa; two thirds of the worldwide production is in the Ivory Coast, Ghana, Nigeria and Camaroon, although Brazil is also a large producer of cocoa (Fontanills, 2007). Cocoa is not a commodity with seasonality (Fontanills, 2007). The average open interest for cocoa over the full sample period was 114.617 contracts (CFTC data).
Coffee: The most widespread form of coffee is Arabica, and the main producers of this kind
Sugar: Sugar is produced in over one hundred countries, in the form of either sugar beets or
sugarcane. Brazil is the world’s largest exporter of sugarcane (Fontanills, 2007). The average open interest for sugar over the full sample period was 372.713 contracts (CFTC data).
Cotton: The five largest cotton producing countries are China, India, the US, Pakistan and
Brazil.5 Dry weather is needed to grow cotton; therefore, when weather conditions are not optimal prices may rise (Fontanills, 2007). The average open interest for cotton over the full sample period was 111.168 contracts (CFTC data).
Gold: Gold has been a valued storage commodity for millennia due to its rarity and beauty
(Fontanills, 2007). In times of crisis and high inflation, gold prices typically rise (Fontanills, 2007). South Africa is the largest producer of gold in the world, followed by the US and Australia (Fontanills, 2007). Gold is used for jewels, electronics and in dentistry practices (Fontanills, 2007). The average open interest for gold over the full sample period was 278.893 contracts (CFTC data).
Silver: Like gold, silver has been used as money for ages. The world’s largest silver suppliers
are Mexico, Peru, the US, Australia and China (Fontanills, 2007). Silver is used mainly in photographic materials; its other minor uses include the production of batteries, jewels, electronics and mirrors (Fontanills, 2007). The average open interest for silver over the full sample period was 105.280 contracts (CFTC data).
Copper: Copper has diverse industrial applications. Its main application is in electronics,
which derives 75% of the copper demand (Fontanills, 2007). Chile is the largest supplier of copper, followed by the US, Indonesia and Australia (Fontanills, 2007). The average open interest for copper over the full sample period was 87.492 contracts (CFTC data).
5
Live cattle: Live cattle are considered steers from the calf stage until they weigh around
600-800kg.6 It takes roughly six to ten months for them to reach maturity.7 Cattle prices can be influenced by the weather, since very hot weather will slow down the growth of cattle.8 This is an American market with the following seven major cattle producing states: Arizona, California, Colorado, Iowa, Kansas, Nebraska and Texas.9 The average open interest for live cattle over the full sample period was 168.703 contracts (CFTC data).
Feeder cattle: Feeder cattle are in fact the same as livestock; however, the steers are
considered feeder cattle when they weigh around 600-800kg. At this point, they will be transferred to feedlots. 10 The average open interest for feeder cattle over the full sample period was 22.649 contracts (CFTC data).
Lean hogs: Lean hogs are pork meat. The largest producer of lean hogs is the US, with states
like Iowa, North Carolina, Minnesota and Illinois leading as the main producers.11 Usually it takes up to six months to raise a pig from birth to slaughter, and prices for hogs tend to be at their highest value between May and July.12 The average open interest for lean hogs over the full sample period was 116. 513 contracts (CFTC data).
Crude oil: Crude oil is the first product that comes from oil drilling. Due to increased road
transportation, crude oil prices are the highest during the summer months, specifically June to September (Fontanills, 2007). The world’s three largest oil producers are: Russia, Saudi Arabia, and the United States (Fontanills, 2007). The average open interest for crude oil over the full sample period was 811.644 contracts (CFTC data).
Heating oil: Around 25% of crude oil is transformed into heating oil; therefore, prices of
heating oil are highly correlated with prices of crude oil (Fontanills, 2007). Heating oil prices tend to be higher during cold winter months, since some parts of the country use heating oil to warm their houses (Fontanills, 2007). The average open interest for heating oil over the full sample period was 192.913 contracts (CFTC data).
Natural gas: Natural gas is used to provide energy. The United States consumes 25% of
world production. It consumes its own production of natural gas and imports the remainder, mainly from Canada.13 Similar to heating oil, the highest demand for natural gas is during cold winter months.14 The average open interest for natural gas over the full sample period was 517.611 contracts (CFTC data).
3. Methodology
3.1. Working T index
As a measure of financial speculation, Working’s (1960) T index, which is considered the standard method for proxying speculative demand, is used. Sanders et al. (2010), for example, support this method, stating that ‘Working’s T still provides an objective measure of speculative
activity’.
Before considering the mathematical definition of the Working (1960) T index, however, it is first useful to discuss the intuition behind the index. Working (1960) argues that future markets are mainly hedging markets and that speculation tends to follow hedging volumes. In commodity markets, natural hedgers purchase or sell future contracts as temporary substitutes for a cash transaction that occurs on a later date. Buyuksahin and Robe (2013) discuss in their paper the working mechanism of the index. The idea is that if long and short hedgers in a futures market
13
http://commodities.about.com/od/profilesofcommodities/p/natural_gas.htm
14
balanced up, speculators would not be needed in the market. In practice, however, long and short hedgers do not always offset each other; hence, speculators have to step in to balance hedging demand.
The speculative T index, in turn, measures the level at which speculation is in excess of the amount that is needed to offset unbalanced hedging between long and short hedgers. In Peck’s (1980) words, “The speculative index…reflects the extent by which the level of speculation
exceeds the minimum necessary to absorb long and short hedging, recognizing that long and short hedging positions could not always be expected to offset each other even in markets where these positions were of comparable magnitudes.” Working (1960) is careful to point out that
what may be “technically an ‘excess’ of speculation is economically necessary for a
well-functioning market.” Buyuksahin and Robe (2013) use a similar definition for Working’s T
index. They state, ‘Working’s “T” measures the extent to which speculation is in “excess” of the
level required to offset any unbalanced hedging at the market-clearing price (i.e., to satisfy hedgers’ net demand for hedging at that price).”
traditionally been used in order to calculate the index. The Working (1960) T index is defined as follows:
(1) (2)
In this equation: SS abbreviates Speculation Short; SL abbreviates Speculation Long; HS abbreviates Hedging Short; and HL abbreviates Hedging Long.
In order to explain the working of the index, a small example from Sanders et al. (2010) can be employed. In a market with speculators and hedgers, SS + HS = SL + HL must hold. When HL = 0 it becomes SS + HS = SL, and dividing by HS yields . From this, it follows that if Speculation Long (SL) = Hedging Short (HS), T has a minimum value, and speculation is exactly sufficient to meet hedging needs. Sanders et al. (2010) also consider another case, where HL= 0, HS = 100, SL = 110, and SS = 10. In this case, T equals 1.10, or there is 10% excessive speculation. Otherwise, there is 10% more speculation than is needed to offset hedging needs. In the case that Hedging Long is not 0, Working (1960) notes that a purely logical basis for writing the formula is missing. However, he shows empirically in his paper that the index can be calculated with long hedging, as stated above in equations (1) and (2).
Other methods have also been used. Till (2009) considers all non-reporting agents as speculators, arguing that, due to the small position size, the traders are not hedgers. Sanders et al. (2010) divide the non-reporting agents in the same proportion as the reportable agents. Buyuksahin and Robe (2013), do not include the group of non-reportable traders in their analysis. In order not to overestimate the impact of speculators in the market, all non-reporting agents are first considered as hedgers. Afterwards, Till’s (2009) study is followed, thus creating an upper bound on the proportion of speculation to hedging. The exact amount of speculative non-reporting traders lies somewhere in between the lower bound and the upper bound.
3.2. Macro-economic variables
It is quite plausible that future prices are affected by macroeconomic variables. However, most literature focuses only on Granger causality tests without macroeconomics. Manera et al. (2012) and Morana (2012) use an approach which includes both macroeconomics and speculation measured by Working’s (1960) T index.
This paper will follow an approach in which Working’s (1960) T index is used as a proxy for financial speculation, in combination with macroeconomic control variables, in order to forecast returns on commodity futures. Therefore, the following section will discuss relevant variables in predicting returns on commodity futures.
interest rate in an empirical study. In addition, Akram (2009) found in his empirical research that shocks to the real interest rate significantly contribute to movements in commodity prices. The 3-month US T-bill will represent the real interest rate.
Secondly, Bailey and Chan (1993) have shown that the yield spread in the equity and bond market explain a large portion in the variation of commodity futures. They suggest that commodity futures are priced to reflect common risk factors that are used to asset markets, such as stock, bond and commodity markets. This implies that there should be a correlation between the basis variations across commodity markets and measures for market risk premiums in the bond and stock market. Thus, when stock and bond markets yield higher returns, it should be reflected in commodity markets. In this scenario, a positive relationship is expected. Chevallier (2009) also states that the price of future commodities reflect systematic risk embedded within the stock market. On the other hand, Chevallier (2009) found opposing results regarding the bond markets. However, to be sure that the results of the current study are authentic, the bond market is incorporated, using the corporate spread calculated, as suggested by Bailey and Chan (1993), by subtracting the return on AAA US Moody’s corporate bonds from BAA Moody’s corporate bonds. In order to incorporate the systematic risk embedded within the stock market, Manera et al. (2012) is followed and a price index of the S&P 500 is used.
rate and commodity future prices. Furthermore, Akram (2009) find evidence that a weaker dollar leads to higher commodity prices. It is expected that not that all commodities returns can be explained by the exchange rate, due to commodities such as live stocks that are mainly produced in the US. A weighted average exchange rate dataset from the Federal reserve is used for these measurements.
The next macroeconomic control variable is the level of inventories. Gorton et al. (2012) state that, at low inventory levels, the basis of the future price increases. They found empirical evidence that there is a negative relationship between inventory levels and commodity futures.
Lastly, a lag value of the commodity return is added to the model since this study concerns a chronological series of events. Brooks (2008) argues that the lag value of the explanatory variable can capture an important dynamic structure in the dependent variable. It could be that markets overreact, and a change in one of the explanatory variables does not necessarily affect the dependent variable immediately. In addition, Manera et al. (2012) found persistence in returns due to positive significant results in their regression analysis for a lagged value of the dependent variable.
3.1 Equations and Econometric specification
In consideration of all previously discussed variables, the following equation is estimated:
- (3)
currencies. - represents the lagged value of . All data used for the variables is discussed in more detail in the data section.
Firstly, I estimate the time series of returns of each commodity separately, using OLS. I then test all specifications for autoregressive conditional heteroscedasticity effects using the White’s test. If ARCH effects are present, I re-estimate a Garch (1,1) model.
Estimations will be made both for the total time period of 1993-2013, as well as for a subsection from 2008, because commodity price fluctuations have been increasing significantly since 2008, compared with the period before 2008. Thus, it is likely that evidence could be found for this subsection.
In addition to the time series regression, I will take advantage of the panel structure of my data set. This could have certain advantages. Firstly, the power of the test will be increased by combining cross-sectional and time series data (Brooks, 2008). Secondly, by structuring the model in an appropriate way, I can remove the impact of certain forms of omitted variables bias in the regression results (Brooks, 2008). It is not unlikely that some evidence of speculative activity driving future prices will be found when making use of the panel structure. The data is suitable for a panel analysis because it entails both time series and cross-sectional elements. Obviously, the time series consists of data from the period between 1993 and 2013, and the cross sectional elements are the different commodities.
of the Hausman test implies that the random effects are correlated with the explanatory variables and that fixed effects are the preferred model.
As a robustness test, the equation will be re-estimated with a different treatment of the Working (1960) T index. In the main equations, a rather conservative approach was taken by assuming all non-reportables as hedgers. Here, the other extreme that all non-reportable are speculators has been assumed, and the regression is estimated again. The true value of the index lies somewhere in the middle; however, by using both treatments, an upper and a lower bound on the Working (1960) T index has been achieved.
4. Data
4.1. Summary statistics commodities
Table I: Commodity Summary Statistics
This table presents summary statistics on all individual commodities for the 1993-2013 sample. Returns are daily returns converted into weekly averages. The Working (1960) T index is calculated as T=1+SS/(HS+HL) if HS ≥ HL and T=1+SL/(HS+HL) if HS< HL. All non-reporting traders are considered hedgers. For lean hogs, Working T data has only been available since 1996; thus, it has fewer observations. For gold and silver, inventory data is available from 2002 until the present. For coffee, inventory data is available from 1997 until the present. Many commodities have only monthly, quarterly or yearly inventory data available. In these cases, data is transformed to weekly values which remain constant for several weeks.
Index Obs Mean Median Minimum Maximum Std. Dev.
Returns
Cocoa US ICE Futures 1073 0.09% 0.04% -12.47% 14.33% 3.52%
Coffee US ICE Futures 1073 0.04% 0.02% -19.75% 34.43% 4.58%
Copper COMEX 1073 0.11% 0.12% -17.13% 9.83% 2.98%
Corn CBOT 1073 0.08% -0.05% -22.39% 13.45% 3.36%
Crude oil NYME 1073 0.16% 0.41% -18.93% 17.20% 3.98%
Cotton US ICE Futures 1073 0.04% 0.10% -26.10% 18.14% 3.61%
Feeder cattle CME 1073 0.05% 0.11% -14.75% 7.42% 1.73%
Gold COMEX 1073 0.13% 0.15% -11.79% 12.81% 1.92%
Heating oil NYME 1073 0.16% 0.41% -18.93% 17.20% 3.98%
Lean hogs CME 1073 0.08% 0.01% -18.67% 27.18% 3.90%
Live cattle CME 1073 0.04% 0.04% -16.24% 7.32% 1.97%
Natural gas NYME 1073 0.07% -0.06% -26.35% 34.81% 6.16%
Oats CBOT 1073 0.08% 0.09% -14.15% 23.89% 3.93%
Silver COMEX 1073 0.16% 0.27% -20.30% 13.36% 3.42%
Soybeans CBOT 1073 0.09% 0.16% -15.60% 9.91% 2.99%
Sugar US ICE Futures 1073 0.06% 0.16% -29.65% 83.50% 4.84%
Wheat CBOT 1073 0.06% -0.13% -17.21% 15.60% 3.48%
Working T
Cocoa US ICE Futures 1073 1.081 1.073 1.006 1.216 0.045
Coffee US ICE Futures 1073 1.102 1.089 1.007 1.305 0.064
Copper COMEX 1074 1.108 1.102 1.004 1.287 0.061
Corn CBOT 1075 1.068 1.061 1.004 1.247 0.038
Crude oil NYME 1073 1.062 1.060 1.002 1.179 0.040
Cotton US ICE Futures 1073 1.088 1.073 1.007 1.334 0.058
Feeder cattle CME 1075 1.132 1.120 1.003 1.455 0.072
Gold COMEX 1074 1.087 1.075 1.009 1.296 0.048
Heating oil NYME 1073 1.042 1.033 1.000 1.173 0.031
Lean hogs CME 901 1.121 1.112 1.000 1.333 0.056
Live cattle CME 1075 1.110 1.098 1.011 1.319 0.057
Natural gas NYME 1073 1.082 1.041 1.001 1.357 0.088
Oats CBOT 1075 1.044 1.037 1.000 1.179 0.031
Silver COMEX 1074 1.089 1.066 1.010 1.443 0.079
Soybeans CBOT 1075 1.091 1.081 1.015 1.252 0.046
Sugar US ICE Futures 1073 1.054 1.046 1.000 1.220 0.040
Wheat CBOT 1075 1.128 1.132 1.016 1.335 0.053
Unit Definition Inventories
Cocoa bags 1074 1840325 1819185 2392 5393598 1803693
Coffee bags 869 2730705 2913822 17295 5092735 1532267
Copper volume 1074 382785 344700 25550 978025 235013
Corn bushels (mill.) 1074 4667 4261 426 10902 2801
Crude oil barrels (thous.) 1074 956709 926624 818781 1100348 79290
Cotton bales 1074 282438 214130 1311 1743226 289731
Feeder cattle pounds (mill.) 1074 395 405 262 525 63
Gold volume 576 7637778 7693048 1794127 11653770 2917961
Heating oil barrels (thous.) 1074 127226 126590 87153 175974 17526
Lean hogs pounds (mill.) 1074 463 460 299 701 85
Live cattle pounds (mill.) 1074 397 406 262 525 63
Natural gas volume 1074 6610979 6653184 5041971 8293610 760560
Oats bushels (thous.) 1074 103 99 36 220 39
Silver volume 576 120000000 117000000 97859630 167000000 15766526
Soybeans bushels (thous.) 1074 1100783 998020 112414 2701366 740657
Sugar volume (tons.) 1074 3219 3370 1241 5390 1095
4.2.Summary statistics macro-economic variables.
Table II: Macroeconomic Summary Statistics
This table presents the summary statistics for all macro-economic variables included in the analyses. The period considered for all macroeconomic variables spans from 1993 to 2013. The three month T Bill is the weekly average of daily middle rate data. The S&P 500 is a composite price index with weekly averages of daily data. The corporate spread is the difference in yield between Moodys BAA and Moodys AAA corporate bonds, transformed from daily data into weekly averages. The WA exchange rate is the weekly weighted average exchange rate of the US dollar with a set of major currencies taken from daily figures. The Unit root test reports the Augmented Dickey Fuller statistic that tests for the null hypothesis, which states that there is a unit root in the variable. ***, ** and * denote significance at 1%, 5% and 10% levels, respectively.
Obs Mean Median Minimum Maximum Std. Dev. Unit Root test
Macroeconomic variables
3 month T Bill 1074 0.97 0.86 0.52 3.47 0.45 -0.30
S&P 500 1074 87.12 86.44 68.11 112.57 11.04 -1.26
Corporate spread 1074 2.89 3.07 0.01 6.22 2.07 -3.13**
WA exchange rate 1074 1067.56 1133.55 432.8 1691.15 322.82 -1.14
Macroeconomic variables returns
3 month T Bill 1073 -0.44% 0.04% -112.60% 203.69% 15.16% -28.73*** S&P 500 1073 0.13% 0.33% -15.28% 8.31% 1.94% -28.73*** Corporate spread 1073 0.01% 0.00% -15.42% 38.15% 3.51% -21.69*** WA exchange rate 1073 -0.02% -0.04% -4.12% 3.38% 0.80% -25.56 *** 5. Results
5.1. Results from time series regressions
With the least squares method, I could not reject the null hypothesis of heteroscedasticity for 16 out of 17 commodities at a 10% level and for 14 out of 17 commodities at a 1% level (see Table III in the appendix). Therefore, I opted for a GARCH (1,1) model as the most suitable model to estimate all individual commodities. Table IV in the appendix displays the time series regression results for all individual commodities for the total sample period.
the oats and feeder cattle equation has the expected positive sign of 1% and a 10% level of significance, respectively. This implies that for oats and feeder cattle, speculation measured by the Working T index drives future prices. However, for cotton the sign is significantly negative at a 5% level. This implies that speculation lowered the commodity future prices of cotton. All other commodities have no significant results for the Working T index, indicating that excess speculation did not cause prices to rise.
For the macroeconomic variables, the S&P500 variable was in line with earlier research, showing the expected positive sign in all commodity equations. Moreover, the S&P 500 variable is significantly positive in fourteen of the commodity equations. This means that, for those commodities, higher returns in the stock market have caused future prices to rise.
Results for the T bill variable differ across commodities. In three commodity equations, the T bill variable is significant with the expected negative sign. In seven commodity equations, the T bill variable is significant with a positive sign. This indicates that a higher interest rate drives some commodity prices up and some commodity prices down. The corporate spread variable is only significant in two commodity equations; for copper, it results in a negative sign, and for wheat it results in a positive sign. This is partly in line with the existing literature.
The level inventories only appear significant in three commodity equations, namely for coffee, sugar and lean hogs. However, the value of the sign is near null. This indicates that the level of inventories is generally not relevant in explaining the commodity future price. This is opposed to the findings of Gorton et al (2012), who find a negative relationship between inventory levels and commodity returns. The difference could be explained by the availability of inventory data. Gorton et al. (2012), who required monthly data while the current study required weekly data. Thus, data needed to be transformed to similar weekly observations, which in turn could have influence the current results. The lagged values are significantly positive in all equations. This means that a positive commodity return in the previous period drives the commodity price in the next period.
In order to find out whether speculation has caused commodity prices to rise in more recent years, the model was re-estimated for a time span including only the years 2008-2013. Table V shows that the Working T index has a negative sign in nearly all commodity equations. The index is significantly negative in four commodity equations. For coffee, sugar and gold, the index is significant at a 10% level; for cotton, it is again significant at a 5% level. The negative sign indicates that excess speculation caused future prices to decrease during the 2008-2013 period. No other commodities equations have significant signs for the Working T index, implying that speculation has no influence on the commodity future price.
5.2. Results from panel regressions
Breusch – Pagan test for random effects. The null hypothesis of no random effects was rejected for both regressions, which implies that panel techniques are more appropriate than a Pooled OLS regression analysis. The panel data was then estimated again, with fixed and random effects, and compared using the Hausman test. The p-value was more than 5% for all four panel regressions, indicating that the more efficient random effects model is the preferred model.
For the 1993-2013 period, the Working T value is not significant at any level and has the opposite sign. As for the other variables in the analysis, the S&P500, the T-Bill, the lagged value and the exchange rate were all significant at a 1% level. The T-bill, however, has a positive sign, which is different than expected. Neither the inventory level nor the corporate spread were significant.
For the 2008-2013 period, the estimation with random effects was also the preferred model. The T value is significant at a 10% level and the sign is negative, indicating that excess speculation caused a significant decrease in commodity prices. This is an intriguing result, which opposes the popular view that speculation increases food prices. Apparently, in the last five years, speculation has led to decreasing commodity prices instead of increasing commodity prices. For the other variables, the results are similar to the full sample period.
5.3. Robustness tests
Table VI shows individual estimation outputs when all non-reporting traders are considered as speculators for the 1993-2013 period. Again, the majority of commodities have a negative sign for the Working T index in the commodity equations. However, more commodities have a significant sign. Feeder cattle, cotton, coffee, sugar and copper are significantly negative, indicating that excess speculation has led to lower commodity returns. Oats are again significantly positive, indicating that, only for oats, speculation drives commodity future returns. For feeder cattle the sign for the Working T index changed to significantly negative from significantly positive when all non-reporting traders were considered as speculators. Thus, results also pointed towards a negative relation when non-reporting traders were considered as speculators.
For the 2008-2013 period, coffee, sugar, cotton and gold were significantly negative when all non-reporting traders were considered as hedgers. Table VII shows the estimation outputs for the adjusted Working T for the 2008-2013 period. Again, the above mentioned commodities were significantly negative. In addition, lean hogs were significantly negative and oats were significantly positive. Also for the 2008-2013 period, the estimation results with the adjusted calculation of the index were to some extent similar to the original calculation.
All in all, it is very important to note that results may be influenced by the application of the Working T index. Despite this, the majority of commodities generally have the same sign for the Working T index with both approaches. Thus, the results for these commodities are robust. However, with the adjusted index, more commodity equations have a significant sign for the Working T index.
the panel analyses. Table IX shows the regression outputs for the 1993-2013 period and for the 2008-2013 period. Whereas in the prior analyses the Working T index was not significant for the full sample and significant at 10% for the 2008-2013 period, the adjusted Working T index is now significant at a 5% level for the full sample and not significant for the 2008-2013 period. The sign is negative for the full sample period. This indicates that speculation had a negative influence on the return of commodities for the total sample. Thus, current results are not robust for the different assumptions on the exact role of the non-reportables. The key result, however, remains unchanged. The sign for the Working T index is negative in every panel equation, indicating that speculation has no positive effect on commodity returns.
6. Discussion
The majority of commodities have a negative sign for the Working T index in the commodity equations. For cotton, the index has a significantly negative sign at a 5% level and for oats a significantly positive sign at a 1% level. Feeder cattle is significantly positive at a 10% level. All other commodities have no significant value, indicating that excess speculation has no influence on those commodity returns. When comparing these results with prior research, it becomes apparent that most researchers do not find evidence for a causal relationship between speculation and commodity future returns either.
Secondly, no significant results have been found for the energy markets. This is in accordance with prior research done by Brunetti and Buyuksahin (2009) and Buyuksahin and Harris (2011), who found no evidence of a causal relationship between trader positions and the returns on crude oil and natural gas.
Thirdly, for the soft commodities, a significantly negative relationship for cotton and no significant results for coffee and cocoa have been found. Irwin and Sanders (2012) also included soft commodities in their analyses, and they found no significant relationship for cotton, coffee, cocoa or sugar. In addition, Hamilton and Wu (2012) did not find evidence that trader positions in coffee, cotton, cocoa, or sugar influence commodity price returns either.
As for the livestock market, a significant result has been found for feeder cattle. This is consistent with Gilbert and Pfuderer’s (2012) results. They found evidence of causality in the livestock market.
Lastly, for precious metals, there was no evidence of speculation since copper, gold and silver were not significant for the total sample period. Most researchers did not include metals in their samples, so it is hard to compare these results with prior research.
Figure II: Open Interest of Commodities
This figure presents the open interest of all commodities, sorted smallest to largest. The CFTC defines the open interest in their explanatory notes as the total of all futures contracts entered into and not yet offset by a transaction, a delivery or by an exercise.15 Feeder cattle, oats and cotton are the markets of interest, since in those markets significant signs are found in the regression equations.
Gilbert and Pfuderer (2012) found similar results in their evidence that index investment affects returns in less liquid markets. They state that, in more liquid markets, speculation is harder to detect due to market efficiencies. Thus, it seems that in less liquid markets, speculators have greater impact on the commodity return.
The 2008-2013 period is of special interest due to large price fluctuations in all commodity markets during this period, and of course, due to the financial crisis. For this period, different commodities show more significant results than for the total period. Coffee, sugar, cotton and gold are significant and have a negative sign. The common factor between these commodities is
that three of them, namely coffee, sugar and cotton, are soft commodities. The open interests or liquidity in those markets is not smaller than other commodities; thus the argument in favour of less liquid markets does not seem plausible for this period. Due to the negative sign of all soft commodities, it can be said that speculation had a negative influence on the return of these three soft commodities.
Looking at other market characteristics, there seems to be no common trend. The commodities are produced in different countries, and the market characteristics seem dissimilar. All in all, besides the fact that all three commodities are soft, there seems no clear explanation for the negative influence on the returns. Also, the return on gold seems negatively affected by speculation. The other precious metals have no significant results. When considering the market characteristics of gold, the most important characteristic in this context seems the value of gold’s storage quality, which spans millennia. Gold prices tend to be higher in times of crisis. Thus, the rationale behind speculators would be to speculate on an increasing price in times of crisis, therefore increasing the price of gold even more. The results show opposite results, however, so there seems to be no economic interpretation for this negative relationship.
7. Summary & Conclusions
the analysis. The research was executed for all commodities individually, using a Garch (1,1) model and for the total sample with a random effects panel data analysis. The main finding was that speculation cannot be held responsible for high food prices since speculative activity does not drive the returns of the commodities.
The Working (1960) T index generally has a negative sign in the individual commodity market equations, indicating that speculation leads to lower commodity returns. In addition, the sign is usually insignificant. The index has a significant sign in three markets, but it is only positive in the oats market equation, with a significantly positive value at a 1% level, and in the feeder cattle market, with a significantly positive value at a 10% level. Cotton has a significantly negative sign at a 5% level. This indicates that speculation drives the future price of oats and feeder cattle.
A common characteristic for these three markets is that they are all less liquid. This indicates that speculators have more impact in less liquid markets. For all other commodities, results indicate a negative or non-significant result. These findings reflect results from other empirical studies and contrast claims made in the literature that speculation positively affects commodity returns. Future research could further elaborate on the role of speculators in less liquid markets, since it seems that speculators have more influence in those markets.
In addition, due to peak periods in commodity prices since 2008, a subsection analysis was executed for the period of 2008-2013. The main result did not change for this subsection; the Working (1960) T index variable was generally negative or insignificant in the commodity market equations.
8. References
Akram, Q.F., 2009. Commodity prices, interest rates and the dollar. Energy Economics 31, 838-851.
Aulerich, N.M., Irwin, S.H., Garcia P., 2012. Bubbles, food prices, and speculation: Evidence from the CFTC’s daily large trader data files. Working paper.
Bahattin, B., Robe, A., 2013. Speculation, Commodities and Cross-Market Linkages. IMF working papers.
Bailey, W., Chan, K.C., 1993. Macroeconomic Influences and the Variability of the Commodity Futures Basis. The Journal of Finance 48, 555-573.
Brooks, C., 2008. Introductory Econometrics for Finance. Cambridge University Press, Cambridge.
Brunetti, C., Reiffen D., 2012. Commodity index trading and hedging costs. Federal Reserve Board.
Brunetti, C., Büyükşahin, B., 2009. Is Speculation Destabilizing? U.S. Commodity Futures Trading Commission Working Paper.
Büyükşahin, B., Harris, J.H,. 2011. Do Speculators Drive Crude Oil Futures Prices? The Energy Journal 32, 167-202.
Büyükşahin, B., Robe, M.A., 2013. Speculation, Commodities and Cross-Market Linkages. In International Monetary Fund (IMF) Conference on "Understanding International Commodity Price Fluctuations". Washington, D.C.
Chevallier, J., 2009, Carbon futures and macroeconomic risk factors: A view from the EU ETS. Energy Economics 31, 614-625.
Ciffarelli, G., Paladino, G., 2010. Oil price dynamics and speculation: a multivariate financial approach. Energy Economics 32, 363–372.
Frankel, J.,A., 2008a. Commodity Prices, Again: Are Speculators to Blame? Unpublished working paper. National bureau of research, Cambridge.
Gilbert, C.L., Pfuderer,S., 2012. Index funds do impact agricultural prices. Workshop “Understanding Oil and Commodity Prices”, London.
Gorton, G. B., F. Hayashi, and K. G. Rouwenhorst. 2013. The fundamentals of commodity futures returns. Review of Finance 17, 35–105.
Hamilton, J.D., Wu, C., 2011. Risk premia in crude oil futures prices. University of California San Diego, mimeo.
Irwin, S.H., Garcia, P., Good, D.L., Kunda, E.L., 2011. Spreads and non-convergence in CBOT corn, soybean, and wheat futures: Are index funds to blame? Applied Economic Perspectives and Policy. 33, 116-142.
Irwin, S.H., Sanders D.R., 2011. Index funds, financialization, and commodity futures markets. Applied Economic Perspectives and Policy 33, 1-31.
Irwin, S.H., Sanders, D.R., 2012. Testing the Masters Hypothesis in commodity futures markets Energy Economics 34, 256-269.
Irwin, S.H., Sanders, D.R., Merrin, R.P., 2009. Devil or Angel? The Role of Speculation in the Recent Commodity Price Boom (and Bust). Journal of Agricultural and Applied Economics 41, 377–391.
Kaufmann, R.K., 2011. The role of market fundamentals and speculation in recent price changes for crude oil. Energy Policy 39, 105-115.
Kaufmann, R.K., Ullman, B., 2009. Oil prices, speculation, and fundamentals: Interpreting causal relations among spot and futures prices. Energy economics 31, 550-558.
Masters, M.W., 2008. Testimony of M.W. Masters before the committee on Homeland Security and Governmental Affairs United States Senate. Downloaded from
http://www.hsgac.senate.gov//imo/media/doc/052008Masters.pdf?attempt=2 on 20-08-2013. Morana, C., 2013. Oil price dynamics, macro-finance interactions and the role of financial speculation. Journal of banking and finance 37, 206-226.
9. Appendix
Table III: White’s Heteroscedasticity Test in OLS Method
This table presents the F-statistic, the probabilities of the F-statistic and the Chi-Square for the White heteroscedasticity test using the OLS method. The null hypothesis denotes no heteroscedasticity. ***, ** and * denote significance at the 1%, 5% and 10% levels, respectively, indicating heteroscedasticity.
Observations F-statistic Prob. F (7,1064) Prob.
Table IV: Estimates of the Time Series Regressions, 1993-2013
This table presents the estimated Garch (1.1) outputs for all commodities individually for the 1993-2013 period, with commodity return as the dependent variable. The impact of speculation is measured by the Working T index and is shown in bold print. The working T is calculated regarding all non-reporting traders as hedgers. The S&P 500 is a composite price index with weekly averages of daily data. The 3 month T Bill is the weekly average of daily middle rate data. The corporate spread is the difference in yield between Moodys BAA and Moodys AAA corporate bonds transformed from daily data into weekly averages. The WA exchange rate is the weekly weighted average exchange rate of the US dollar with a set of major currencies taken from daily figures. The lagged return is the commodity return of the previous week. Inventory data is weekly data or is transformed to weekly data. The standard errors are in parentheses. ***, ** and * denote significance at the 1%, 5% and 10% levels, respectively.
Crude oil Heating oil Natural
gas Corn Wheat Soybeans Oats
Lean hogs
Live cattle
Feeder
Cattle Coffee Sugar Cocoa Cotton Gold Silver Copper
Table V: Estimates of the Time Series Regressions, 2008-2013
This table presents the estimated Garch (1.1) outputs for all commodities individually for the 2008-2013 period, with commodity return as the dependent variable. The impact of speculation is measured by the Working T index and is shown in bold print. The working T is calculated regarding all non-reporting traders as hedgers. The S&P 500 is a composite price index with weekly averages of daily data. The 3 month T Bill is the weekly average of daily middle rate data. The corporate spread is the difference in yield between Moodys BAA and Moodys AAA corporate bonds transformed from daily data into weekly averages. The WA exchange rate is the weekly weighted average exchange rate of the US dollar with a set of major currencies taken from daily figures.The lagged return is the commodity return of the previous week. Inventory data is weekly data or is transformed to weekly data. The standard errors are in parentheses. ***, ** and * denote significance at the 1%, 5% and 10% levels, respectively.
Crude oil Heating oil Natural
gas Corn Wheat Soybeans Oats
Lean hogs
Live cattle
Feeder
Cattle Coffee Sugar Cocoa Cotton Gold Silver Copper
Table VI: Estimates of the Time Series Regressions, 1993-2013
This table presents the estimated Garch (1.1) outputs for all commodities individually for the 1993-2013 period, with commodity return as the dependent variable. The impact of speculation is measured by the Working T index and is shown in bold print. The working T is calculated regarding all non-reporting traders as speculators. The S&P 500 is a composite price index with weekly averages of daily data. The 3 month T Bill is the weekly average of daily middle rate data. The corporate spread is the difference in yield between Moodys BAA and Moodys AAA corporate bonds transformed from daily data into weekly averages. The WA exchange rate is the weekly weighted average exchange rate of the US dollar with a set of major currencies taken from daily figures. The lagged return is the commodity return of the previous week. Inventory data is weekly data or is transformed to weekly data. The standard errors are in parentheses. ***, ** and * denote significance at the 1%, 5% and 10% levels, respectively.
Crude oil Heating
oil
Natural
gas Corn Wheat Soybeans Oats
Lean hogs
Live cattle
Feeder
Cattle Coffee Sugar Cocoa Cotton Gold Silver Copper
Table VII: Estimates of the Time Series Regressions, 2008-2013
This table presents the estimated Garch (1,1) outputs for all commodities individually for a 2008-2013 period, with commodity return as the dependent variable. The impact of speculation is measured by the Working T index and is shown in bold print. The working T is calculated regarding all non-reporting traders as speculators. The S&P 500 is a composite price index with weekly averages of daily data. The 3 month T Bill is the weekly average of daily middle rate data The corporate spread is the difference in yield between Moodys BAA and Moodys AAA corporate bonds transformed from daily data into weekly averages. The WA exchange rate is the weekly weighted average exchange rate of the US dollar with a set of major currencies taken from daily figures. The lagged return is the commodity return of the previous week. Inventory data is weekly data or is transformed to weekly data. The standard errors are in parentheses. ***, ** and * denote significance at 1%, 5% and 10% levels, respectively.
Crude oil
Heating oil
Natural
gas Corn Wheat Soybeans Oats
Lean hogs
Live cattle
Feeder
Cattle Coffee Sugar Cocoa Cotton Gold Silver Copper
Table VIII: Panel Random Effects Regression Output
This table presents the panel random effects output for all commodities individually for the 1993-2013 and the 2008-2013 period, with commodity return as the dependent variable. The impact of speculation is measured by the Working T index and is shown in bold print. The working T is calculated regarding all non-reporting traders as hedgers. The S&P 500 is a composite price index with weekly averages of daily data. The 3 month T Bill is the weekly average of daily middle rate data The corporate spread is the difference in yield between Moodys BAA and Moodys AAA corporate bonds transformed from daily data into weekly averages. The WA exchange rate is the weekly weighted average exchange rate of the US dollar with a set of major currencies taken from daily figures. The lagged return is the commodity return of the previous week. Inventory data is weekly data or is transformed to weekly data. The standard errors are in parentheses. ***, ** and * denote significance at 1%, 5% and 10% levels, respectively. 1993-2013 2008-2013 S&P500 0.193*** 0.382*** (0.015) (0.023) T Bill 0.010*** 0.008*** (0.002) (0.002) Corporate Spread -0.013 0.012 (0.008) (0.015)
W.A. Exchange rate -0.521*** -0.598***
(0.035) (0.055) Lagged return 0.164*** 0.195*** (0.008) (0.013) Inventory 0.000 0.000 (0.000) (0.000) Impact of speculation -0.002** -0.002 (0.001) (0.002) Constant 0.003** 0.003 (0.001) (0.003) R-squared 0.057 0.148 F-statistic 145.810 122.518 Prob(F-statistic) 0.000 0.000
Cross sections included 17 17
Periods included 1072 291
Total panel observations 16755 4936
Hausman test 13.620 0.000
(Probability Hausman test) (0.058) 1.000
Breusch Pagan under Pooled OLS, both sections 1716.507 1009.186
Table IX: Panel Random Effects Regression Output
This table presents the panel random effects output for all commodities individually for the 1993-2013 and the 2008-2013 period, with commodity return as the dependent variable. The impact of speculation is measured by the Working T index and is shown in bold print. The working T is calculated regarding all non-reporting traders as speculators. The S&P 500 is a composite price index with weekly averages of daily data. The 3 month T Bill is the weekly average of daily middle rate data The corporate spread is the difference in yield between Moodys BAA and Moodys AAA corporate bonds transformed from daily data into weekly averages. The WA exchange rate is the weekly weighted average exchange rate of the US dollar with a set of major currencies taken from daily figures. The lagged return is the commodity return of the previous week. Inventory data is weekly data or is transformed to weekly data. The standard errors are in parentheses. ***, ** and * denote significance at 1%, 5% and 10% levels, respectively. 1993-2013 2008-2013 S&P500 0.192*** 0.383*** (0.015) (0.024) T Bill 0.010*** 0.008*** (0.002) (0.002) Corporate Spread -0.013 0.013 (0.008) (0.015)
W.A. Exchange rate -0.521*** -0.594***
(0.035) (0.055) Lagged return 0.165*** 0.194*** (0.008) (0.013) Inventory 0.000 0.000 (0.000) (0.000) Impact of speculation -0.001 -0.014* (0.005) (0.008) Constant 0.002 0.016* (0.005) (0.009) R-squared 0.057 0.149 F-statistic 144.996 122.836 Prob(F-statistic) 0.000 0.000
Cross sections included 17 17
Periods included 1072 291
Total panel observations 16755 4936
Hausman test 0.000 0.000
(Probability Hausman test) 1.000 1.000
Breusch Pagan under Pooled OLS, both sections 1725.697 1003.407
