SEARCH

SEARCH BY CITATION

Keywords:

  • Biofuels;
  • commodity prices;
  • exchange rates;
  • food prices;
  • futures markets;
  • money
  • Q11

Abstract

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Commodity Price Booms
  5. 3. The Oil Price and Biofuels
  6. 4. Exchange Rates
  7. 5. Monetary Factors and Futures Markets
  8. 6. Granger Causality Analysis, 1970–2008
  9. 7. 2006–2009
  10. 8. Conclusions
  11. References
  12. Appendices

Agricultural price booms are better explained by common factors than by market-specific factors such as supply shocks. A capital asset pricing model-type model shows why one should expect this and Granger causality analysis establishes the role of demand growth, monetary expansion and exchange rate movements in explaining price movements over the period since 1971. The demand for grains and oilseeds as biofuel feedstocks has been cited as the main cause of the price rise, but there is little direct evidence for this contention. Instead, index-based investment in agricultural futures markets is seen as the major channel through which macroeconomic and monetary factors generated the 2007–2008 food price rises.


1. Introduction

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Commodity Price Booms
  5. 3. The Oil Price and Biofuels
  6. 4. Exchange Rates
  7. 5. Monetary Factors and Futures Markets
  8. 6. Granger Causality Analysis, 1970–2008
  9. 7. 2006–2009
  10. 8. Conclusions
  11. References
  12. Appendices

The prices of food commodities2 more than doubled from 2005 to mid-2008. These price rises were general across the range of agricultural products with only a small number of exceptions, the most prominent being sugar. Table 1 lists rises in prices from the 2000 average to the peak month and also the price rises over the 30 months from January 2006 to July 2008. The entire range of grains and oil seeds saw large price increases.

Table 1.    Percentage Price Rises, 2006–2008
 Peak month2000 average to peak monthJanuary 2006 to July 2008
  1. Notes: Prices are in nominal US dollars. All values are in percentage. Source: International Monetary Fund: International Financial Statistics.

Food commodities
 MaizeJune 2008225160
 Palm OilMarch 2008339172
 RiceApril 2008398181
 SoybeansJuly 2008203151
 SugarFebruary 2006123−18
 WheatMarch 200828696
IMF Indices
 Agricultural foodsJune 200811976
 BeveragesJune 200812155
 Agricultural raw materialsJuly 20082420
 MetalsMay 200722851
 EnergyJuly 2008363115

The causation of these price rises remains controversial. Drought resulted in poor 2006 and 2007 wheat harvests in Australia. The World Bank estimates that these may have reduced wheat supplies by 4% (Mitchell, 2008). The 2007 European grains harvest was also poor. However, these outcomes were largely offset by good harvests in Argentina, Kazakhstan and Russia with the result that the aggregate wheat shortfall was small and manageable. Headley and Fan (2008) concur that weather shocks were not a major factor behind the price rises. Input prices rose sharply in 2008, particularly fertiliser prices, but they lagged food price rises. Stock levels had become low for many grains, but this was not a new situation. By reducing potential supply responsiveness, low stocks will have tended to amplify the impact of other factors but, absent a shock to stock demand, will not themselves have directly impacted prices.

The rises in food prices took place in the context of a general rise in commodity prices led by energy and metals. The rises in the agricultural food price indices were low relative to those of energy and metals products (see Table 1). The prices of agricultural raw materials rose little or even fell over the same period, as did the price of sugar. The entire range of commodity prices saw sharp falls over the second half of 2008, and in particular from September in the ‘post-Lehman’ months. Most notably, the price of crude oil fell from inline image to inline image per barrel at the year end. Food prices fell in line with these developments, although they have subsequently partially recovered.

Many commentators have noted the parallels between the 1973–1974 and 2005–2008 commodity price booms (see, for example, Radetzki, 2006 who focuses on metals and energy commodities). As in 2006–2008, the 1973–1974 boom in agricultural prices occurred against the backdrop of a general rise in commodity prices. Both took place in the context of enormous world liquidity resulting, in part, from large US trade deficits and loose monetary policies. In both cases, oil prices jumped sharply upwards. Both booms ended sharply with the onset of recession, in the second quarter of 1974 and third quarter of 2008, respectively. Metals prices rose strongly in both booms. Coffee and cocoa were largely sidelined in both cases. There were also important differences. Firstly, the 1973–1974 boom was shorter than the 2003–2008 commodity price boom. Secondly, although grains participated in the booms, they led in 1973–1974, ahead of the rise in oil prices, but lagged in 2003–2008 when the rise came only at the end of the boom.

Cooper and Lawrence (1975), who discussed the causes of the 1973–1974 boom, emphasised (p. 672) that the markets were inter-related and that the boom required a ‘general explanation’ over and above ‘intriguing and sometimes significant’ stories relating to particular markets. This paper argues that the same is true of the more recent boom. Section 2 sets out a framework which attempts to provide this type of general explanation. This framework indicates that common (macroeconomic and monetary) demand-side factors should be seen as the main candidates for explaining major changes in agricultural food prices in aggregate.

The diversion of food crops as biofuels stands out as an important new single factor which many have seen as explaining the 2006–2008 food price boom. Rosegrant et al. (2008) argue that there is a clear ‘food-vs.-fuel’ trade-off. According to this view, the diversion of food crops and, more generally, agricultural land into the production of biofuel feedstocks puts upward pressure on food prices and links changes in food prices to changes in the crude oil price. This is the subject of section 3 of this paper.

Abbott et al. (2008) have emphasised the importance of the depreciation of the US dollar as an explanation of commodity price rises in general and food prices specifically over 2006–2008. Exchange rate impacts on food prices are discussed in section 4. Section 5 moves to non-fundamental factors looking particularly at monetary expansion and the role of futures markets. There have been suggestions that excess speculation (‘over trading’ in an older parlance) inflated commodity prices over the period 2006–2008, and a recent US Senate subcommittee report has made this charge with respect to wheat (US Senate Permanent Subcommittee on Investigations, 2009).

The quantitative analysis is in sections 6 and 7, in the former case looking at long-term determinants of food prices and in the latter case specifically at developments in 2006–2008. Section 8 concludes.

2. Commodity Price Booms

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Commodity Price Booms
  5. 3. The Oil Price and Biofuels
  6. 4. Exchange Rates
  7. 5. Monetary Factors and Futures Markets
  8. 6. Granger Causality Analysis, 1970–2008
  9. 7. 2006–2009
  10. 8. Conclusions
  11. References
  12. Appendices

Explanations of the changes in the general level of agricultural food prices differ from explanations of changes in the price of specific food crops. This aggregation argument has two components. The first arises out of averaging – factors which are of minor importance in the explanation of individual prices may, if common, be of much greater importance in explaining aggregate price movements. The second component arises out of substitution between crops which is important at the level of the individual crop price but less so at the aggregate level.

First consider the averaging argument. Explanations of food price movements typically adopt a partial equilibrium, market-by-market, approach and focus on shifts in demand and supply curves. Supply shocks are generally seen as dominant with demand movements playing a less important role. However, supply shocks will be at most weakly correlated across crops, whereas demand movements may have a large common component. The result can be that demand factors predominate at the level of aggregate food prices even if movements in supply are more important in each of the markets considered separately.

This argument is similar to that employed in deriving the standard capital asset pricing model (CAPM) for financial asset prices in which asset returns are decomposed into a common market component and an idiosyncratic (asset-specific) component (Sharpe, 1964; Lintner, 1965). According to the CAPM, idiosyncratic risk becomes negligible in a diversified portfolio, but the market risk remains. A simple CAPM-like agricultural model views the change in the jth food commodity price as

  • image(1)

where x is the common demand shift variable (the ‘market return’ in the CAPM) and ηj is the commodity-specific supply shock (time subscripts are omitted). For simplicity, consider an equally weighted geometric price index P defined over n prices

  • image

Then

  • image(2)

where

  • image

In line with CAPM, assume that the supply shocks are uncorrelated with the demand shock, i.e. inline image (j = 1,…, n) and write inline image and inline image where inline image. Denote the jth diagonal element of Σ as inline image. The (population) correlation of the changes in the jth food price and the demand shock is

  • image(3)

The corresponding correlation for changes in the aggregate price index is

  • image(4)

where inline image, the vector of units.

The relative size of the aggregate and disaggregate correlations depends on the covariance structure of the supply shocks. I consider two polar cases and an intermediate case.

  • 1
     The supply shocks are idiosyncratic with a common variance inline image. Σ is therefore diagonal with inline image and the correlation on the aggregate equation becomes
    • image( (5a))
    In this case inline image and as the number of commodities in the index becomes large, supply-side factors disappear, irrespective of their importance in the commodity-specific equations. This is the basis for the CAPM portfolio diversification proposition.
  • 2
     The supply shocks are common with variance inline image. In this case inline image and
    • image( (5b))
    and the correlation on the aggregate equation is comparable with that on the commodity-specific equations. Diversification fails to reduce risk.
  • 3
     More generally, denote the ijth element of Σ as σij. Suppose an equicorrelation structure such that inline image (i ≠ j = 1, …, n) with 0 ≤ ρ ≤ 1. Hence, inline image
    • image( (5c))
    In this case,
    • image
    and supply-side factors remain important even when the number of component prices in the index is large.3

If agricultural food commodities are sufficiently diverse in their modes and locations of production that supply shocks are generally weakly correlated, we will be closer to the first polar case than to the second and common demand factors will dominate changes in aggregate food price indices.

The discussion so far has been in terms of the contrast between common and crop-specific shocks where the former have been interpreted as demand and the latter as supply shocks. Demand shocks may also be specific. The second argument demonstrates that, even on a commodity-by-commodity basis, demand shocks are likely to have a larger price impact than specific demand shocks of the same magnitude.

Consider the idiosyncratic (commodity-specific) demand shock DD′ in Figure 1. Factors are drawn from other markets and supply is elastic, illustrated as S, with the result that the demand shock leads to the small price rise p1 − p0. If, instead, the demand shock is common across a range of agricultural markets, the position becomes more complicated. First, there may be cost increases as outputs from one sector are used in others, e.g. grain inputs into feed in meat production. This is reflected in the upward shift of the supply curve to S′. Second, it will be less attractive to reallocate factors across markets with the result that additional factors must be drawn from outside the sector. The supply curve becomes less elastic, rotating to S″. The result is that the same demand shock in terms of the market in question will lead to the much larger price rise p2 − p0. This is demonstrated formally in the Model Appendix.

image

Figure 1.  Price responses to idiosyncratic and common demand shocks

Download figure to PowerPoint

In summary, common shocks will dominate movements in aggregate price indices at the expense of idiosyncratic shocks. In agricultural markets this implies that demand shocks are likely to be more important at the aggregate level than supply shocks even if the reverse is the case for individual food commodities. Secondly, common demand shocks will generate a larger price response than idiosyncratic demand shocks. The overall effect is that aggregate equations for food price indices will give greater weight to common demand factors than do crop-specific equations and show greater price responsiveness to common demand shocks than to shocks to specific factors. Typically, these common demand-side shocks will be macroeconomic. These macroeconomic shocks are likely to provide the dominant explanation of large common food price movements.

3. The Oil Price and Biofuels

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Commodity Price Booms
  5. 3. The Oil Price and Biofuels
  6. 4. Exchange Rates
  7. 5. Monetary Factors and Futures Markets
  8. 6. Granger Causality Analysis, 1970–2008
  9. 7. 2006–2009
  10. 8. Conclusions
  11. References
  12. Appendices

The rise, and subsequent fall, in agricultural food prices over 2006–2008 coincided with the sharp upward movement and even sharper fall of crude oil prices. It is therefore natural to suspect that the (larger) oil market may have driven movements in agricultural markets. Furthermore, the jump in oil prices was indeed a common shock that will have affected all agricultural markets.

Oil prices may affect food commodity prices in two ways. First, increases in oil prices will result in higher food production costs – the shift from S to S′ in Figure 1. This pass-through will partly be through the costs of nitrogen-based fertilisers and partly through transport costs. However, agriculture is not highly energy intensive. Baffes (2007) estimated the pass-through of oil prices into agricultural commodity prices as 17%. Mitchell (2008) put that the combined effects of higher energy and transport costs were to raise production costs in US agriculture by 15–20%. Moreover, these effects may be dependent on the state of the market and this vary over time – cost increases will have the greatest price impact in periods of low demand, whereas in periods of high demand they will primarily erode the rents accruing to landowners.4

The second route is through demand – the shift from D to D′ in Figure 1– arising out of the new demand for food commodities as biofuel feedstocks. Mitchell (2008) suggested that biofuels demand was responsible for the largest part of the rise in food prices but resisted the temptation to quantify this share. Abbott et al. (2008) divided the blame for high food prices between biofuels demand and dollar depreciation. Mitchell (2008) concluded that, although the 2006–2008 increase in food prices was caused by ‘a confluence of factors’, the most important of these was the large increase in US and EU biofuels production.5

Schmidhuber (2006) provides a framework for analysing the transmission from the oil market to agricultural food markets. He argues that the prices of crude oil and fertilisers define a break-even price for maize and palm oil at which of biodiesel production yields zero profit (and similarly for sugar and ethanol). Figure 2 illustrates Schmidhuber’s model in relation to maize.

image

Figure 2.  Biofuels demand and grains prices

Download figure to PowerPoint

In the case that the oil price is low, the threshold conversion price is T. The price P at which the demand curve D for maize in food uses intersects the maize supply curve exceeds T making it uneconomic to use maize as a biofuel feedstock. In these circumstances, the maize price will be determined without reference to the oil price except in so far as it affects production and transportation costs. If instead, the oil price is higher so that the threshold price T′ is above P, biofuels production is economic. In the long run, the total (food plus biofuels) demand curve becomes horizontal at this level with the effect that the maize price is pulled up to the threshold T′. So long as the threshold price remains at or above the food demand and supply price P, the maize price should move in step with oil prices. In effect, maize becomes a ‘petro-commodity’. Subsidies and tariffs, such as the US tariff on imported ethanol, complicate these relationships, but the principles remain the same.

In relation to the common factor discussion in section 2, the biofuels story locates the shift in demand in a small number of commodities, specifically maize and palm oil, and relies on inter-crop substitution by farmers to generalise the price effects across the spectrum of food commodities. As biofuels are a new phenomenon, there are insufficient data points on biofuels production to enable direct confrontation of this theory with data. Instead, we need to look for indirect evidence. The theory also predicts that we should see a closer correlation between movements in oil and food prices over the recent past than historically. The correlation between monthly logarithmic changes in the IMF’s nominal agricultural price index and an average of the WTI and Brent crude oil prices was 0.287 over the 36 months 2006–2008 as against 0.199 over the 36 months 2005–2005 and 0.043 over 2000–2002. Correlations over the previous 36-month period were generally close to the 2000–2002 level. Although this provides weak support for the claim of an increased comovement of oil and food prices, the comovement remains small and the 2006–2008 correlation of 0.287 is not statistically significant at the 5% level. This question is further examined below in sections 6 and 7.

4. Exchange Rates

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Commodity Price Booms
  5. 3. The Oil Price and Biofuels
  6. 4. Exchange Rates
  7. 5. Monetary Factors and Futures Markets
  8. 6. Granger Causality Analysis, 1970–2008
  9. 7. 2006–2009
  10. 8. Conclusions
  11. References
  12. Appendices

Almost all international commodities are traded in terms of the US dollar. Dollar deprecation effectively shortens the measuring rod against which we measure their prices. If this were the only change, dollar prices should rise. If the index uses appropriate weights, defined below, the response must have less than unit elasticity such that the resulting rise in dollar prices is offset by falls in prices measured in terms of at least some other currencies. From peak to trough, the dollar depreciated by around 25% against the euro over the period 2005–2008 but much less against other currencies. The likely dollar depreciation effect was therefore small but should be evident.

The impact of exchange rates on commodity prices was analysed by Ridler and Yandle (1972). To understand the Ridler–Yandle theory, suppose there are n countries which both produce and consume a commodity. Write the consumption demand in country j as inline image where p is the international dollar price of the commodity and xj is country j’s dollar exchange rate (local units per dollar). Similarly, write country j’s production as inline image. The USA is country 1 so x1 = 1. Ignoring stocks, market equilibrium requires

  • image(6)

Total differentiation of equation (6) gives

  • image(7)

where

  • image

are, respectively, the elasticities of consumption and production in country j (both measured as positive) where a prime denotes a derivative. Write j’s shares in world consumption and production, respectively, as

  • image

to get

  • image(8)

where the numerator summation omits country 1 (the USA) as inline image by definition. Equation (8) informs us that, in this simple model, exchange rate changes should be weighted by the product of market shares and elasticities. Write country j’s weight as

  • image

so that equation (8) becomes

  • image(9)

If we consider a (small) uniform dollar depreciation of θ% we obtain inline image so dollar prices rise in proportion to the depreciation by a factor of one minus the elasticity-weighted share of the USA in the world market. Gilbert (1989) shows that this ‘less than unit elasticity’ result is robust in more general specifications in which prices for a range of commodities are jointly determined.

In practice, it is frequently found that commodity prices, whether agricultural or mineral, appear to exhibit excess sensitivity to exchange rate movements. One reason for this may be that both exchange rate changes and commodity price movements have a business cycle component which may not be fully reflected by available demand-side variables. A second reason may be because causation runs in part from commodity prices to exchange rates. This is particularly likely to be true if production weights are used in constructing the exchange rate index on the basis that, in the long run, supply is more elastic than demand. High commodity prices result in appreciation of commodity currencies which, if included in the exchange rate index, register as dollar depreciation. The ubiquity of commodity currencies has been documented by Cashin et al. (2003, 2004).

5. Monetary Factors and Futures Markets

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Commodity Price Booms
  5. 3. The Oil Price and Biofuels
  6. 4. Exchange Rates
  7. 5. Monetary Factors and Futures Markets
  8. 6. Granger Causality Analysis, 1970–2008
  9. 7. 2006–2009
  10. 8. Conclusions
  11. References
  12. Appendices

The discussion in sections 2–4 has focused on the so-called fundamental factors. In particular, demand has been seen as consumption-based demand for the physical commodity. In a monetary economy, demand may also take the form of demand for title to (present or future) commodity flows. This opens the door to non-fundamental-based explanations of changes in food prices.

Monetary explanations of changes in price levels and relative prices attracted wide support in the 1970s and 1980s. Cooper and Lawrence (1975) argued that commodities were seen as a safe real asset in a period of unreliable monetary values. Bordo (1980) and Chambers and Just (1982), who considered the impact of monetary growth on agricultural prices, found that monetary expansion could raise agricultural prices relative to a more general price deflator.6 Over the past 25 years, retail price inflation has apparently been conquered through the nominal anchor provided by implicit or explicit inflation targeting undertaken by independent and increasingly credible central banks. Monetary explanations of asset prices have consequently fallen from fashion.

Despite this, there remains a suspicion that monetary growth may generalise into asset prices, a channel stressed by Taylor (1995). The 2006–2008 boom in agricultural prices took place contemporaneously not only with booms in other commodity prices but also with booms in equity and real estate prices. This suggests that, in an environment in which central banks were controlling goods prices, monetary growth may have spilt over into asset prices. Svensson (1985) set out a cash-in-advance model which has this implication. Agricultural futures markets provide a possible route through which this transmission may have taken place.

There are active futures markets for many of the most important agricultural commodities – wheat, maize, soybeans, where prices rose sharply over 2006–2008, but also cocoa, coffee, cotton and sugar, where there was no boom. These markets facilitate the transfer of risk from the so-called ‘commercial’ traders, generally referred to as hedgers, who are exposed to movements in the commodity price through their regular commercial activities, to ‘non-commercial’ traders, often referred to as speculators. The second important function of futures markets is that of price discovery – markets allow agents who believe that they have information to trade on the basis of that information. Finance theory distinguishes between informed and uninformed speculation (see O’Hara, 1995). Informed speculation is expected to have an impact on the market price as information becomes impounded in the market price. Efficient markets theorists argue that commodity price rises have been driven completely by market supply and demand fundamentals and that futures markets form the mechanism by which information about fundamentals becomes incorporated in market prices.

According to this theory, if uninformed trades move a market price away from its fundamental value, informed traders, who know the fundamental value, will take advantage of the profitable trading opportunity with the result that the price will return to its fundamental value. The informed speculators stabilise prices as set out by Friedman (1953). This argument supposes that all trades are informed. In practice, the information content of trades will only become clear over time. For these reasons, there is a worry that trend-following can result in herd behaviour.

IFPRI (2008) has suggested that commodity speculation may be a contributory factor in recent agricultural price rises and has suggested market-calming regulations. Many so-called commodity funds operate by identifying trends and investing accordingly. There is a concern that a chance upward movement in a price may be taken as indicative of a positive trend resulting in further buying and hence driving the price further upwards, despite an absence of any fundamental justification. De Long et al. (1990) show that informed traders may bet on continuation of the trend rather than a return to fundamentals. The conditions under which the informed traders will act in this way is that they have short time horizons (perhaps as the result of performance targets or reporting requirements) and that there are sufficiently many uninformed trend-spotting speculators. Irwin and Yoshimaru (1999) and Irwin and Holt (2004) provide empirical evidence deriving from confidential CFTC data which indicates that this phenomenon does take place. Phillips and Yu (2009) have suggested that there was an oil price bubble between March and August 2008 and that this was one of a sequence of bubbles which moved across financial markets as investors sought high returns (see also Gilbert, 2010). If that view turns out to be correct, it would be unsurprising to find that agricultural markets had participated in these bubbles.

A possible response is that non-fundamental factors, such as monetary expansion or futures market activity, can only affect agricultural prices in so far as they affect inventory levels. In particular, if the futures price is driven above its market clearing level, inventories should accumulate. This argument is correct, but it may take time to play out: an increase in the demand for futures contracts will tend to raise long-dated futures prices and thereby generate an increased incentive to hold inventory. In the short-to-medium term, inventories are largely predetermined at a level implied by the carryover from previous crop years. With the inventory supply curve near vertical, the increased demand can only be met by an increase in the cash price. Over time, as production and consumption respond to the higher price, inventories will rise and price will fall.

The traditional discussion of commodity futures activity has been in terms of hedgers and speculators. However, a new class of transactors in commodity futures markets has become important over the past two decades. These are investors who regard commodity futures as an ‘asset class’, comparable with equities, bonds, real estate and emerging market assets, and who take positions on commodities as a group based on the risk-return properties of portfolios containing commodity futures relative to those confined to traditional asset classes (for a discussion, see Fabozzi et al., 2008).

Masters (2008) testified before a US Senate committee that the behaviour of this group of transactors is quite different from that of traditional speculators, and it is therefore possible that this will result in different effects on market prices. The behaviour of commodity investors differs from that of speculators in three important respects:

  • 1
     Investors take position across the entire range of commodity futures, not in specific futures as traditional speculators.
  • 2
     Investors are almost always long only, whereas speculators may equally be long or short.
  • 3
     Investors hold their commodity positions for long periods of time – months or years – rolling positions forward from expiring to more distant months, whereas speculators hold positions for short periods of time – days or weeks – and seldom roll.

Commodity investors do not generally invest directly in commodity futures. Instead, banks and other financial institutions facilitate such investments by providing suitable instruments, typically exchange traded funds (ETFs), commodity certificates or swaps. In the case of certificates and swaps, they offset much of their net position by taking opposite positions on the futures markets. The majority of such institutions will aim to replicate a particular commodity futures index in the same way that equity tracking funds aim to replicate the returns on an equities index. Such institutions are referred to as Index Funds.

The most widely followed commodity futures indices are the S&P GSCI and the DJ-UBS Index (previously the DJ-AIG index). The S&P GSCI is weighted in relation to world production of the commodity averaged over the previous 5 years. These are quantity weights and hence imply that the higher the price of the commodity future, the greater is its share in the S&P GSCI. High energy prices in 2006–2008 implied a very large energy weighting – 71% in September 2008. The DJ-UBS Index weights the different commodities primarily in terms of the liquidity of the futures contracts, measured in terms of volume and open interest, but in addition considers production. Averaging is again over 5 years. Importantly, the DJ-UBS Index also aims for diversification and limits the share of any one commodity group to one-third of the total. The September 2008 energy share fell just short of this limit. In 2008, agricultural futures comprised 16% of the S&P GSCI and 37% of the DJ-UBS Index.

According to many commentators, index positions are sufficiently large that index-based inventors have come to dominate the commodity futures markets relegating fundamental factors to a minor, supporting, role. Commodity index providers had invested a total of $43bn in US agricultural futures at the end of 2007, rising to $58bn by the end of June 2008 (CFTC, 2008). This is shown in Table 2 which also gives the shares of the index funds’ net positions in total open interest. These are generally in the 25–35% range, although higher for wheat, live cattle and lean hogs. They average 27%. In June 2008 testimony before a second US Senate committee, Soros (2008) asserted that investment in instruments linked to commodity indices had become the ‘elephant in the room’ and suggested that investment in commodity futures might exaggerate price rises. Cooke and Robles (2009) offer academic support for the view that futures market developments may have contributed to the 2006–2008 food price boom.7 I will argue in section 7 that it is index investors and not traditional speculators who were instrumental in driving up food prices in 2006–2008.

Table 2.    Index fund values and shares: US Agricultural Markets 30 June 2008
 Index fund value ($bn)Share (%)
  1. Source: CFTC (2008) valued at front position closing prices. The wheat figures aggregate positions on the Chicago Board of Trade and the Kansas City Board of Trade. Total shares are price weighted.

Corn13.127.4
Soybeans10.920.8
Soybean oil2.621.7
Wheat9.741.9
Feeder cattle0.630.7
Live cattle6.541.8
Cocoa0.814.1
Coffee3.125.6
Cotton2.921.5
Sugar4.931.1
Lean hogs3.240.6
Total58.327.1

6. Granger Causality Analysis, 1970–2008

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Commodity Price Booms
  5. 3. The Oil Price and Biofuels
  6. 4. Exchange Rates
  7. 5. Monetary Factors and Futures Markets
  8. 6. Granger Causality Analysis, 1970–2008
  9. 7. 2006–2009
  10. 8. Conclusions
  11. References
  12. Appendices

The quantitative analysis proceeds in two complementary stages. In this section, I report an exploratory univariate analysis, based on Granger causality tests and using a long sample of quarterly data. The objective is to isolate those factors which appear to have been responsible for food price movements over recent decades. Then, in section 7, I develop a monthly model of food prices with the objective of attributing the observed price rise over the 3 years 2006–2008 to the various factors (the oil price, biofuels demand, speculation, dollar depreciation) which have been argued as being important.

Granger causality tests (strictly Granger non-causality tests) may be thought of as a non-controversial preliminary to explicit modelling. A finding that a variable x Granger-causes a second variable y indicates that x must enter either directly into a structural (i.e. causal) model for y as an explanatory variable or do so indirectly, as a variable correlated with a third variable z which has a direct causal relationship with y. Specification issues arise solely in relation to the choice of lag length used for the Granger causality tests.

In what follows, three price indices are considered. These are the IMF agricultural foods price index IAF, a grains price index IAG, defined as the simple average of the IMF price indices for corn (maize), rice and wheat and a vegetable oil index IAV, the simple average of the IMF price indices for palm oil, soybean oil and sunflower oil. Data for all variables are quarterly from 1971q1 to 2008q4.

The discussion in section 2 suggests we should expect to find demand-side factors and other variables, such as exchange rate changes, which are common across all agricultural products as likely for causal factors for sharp rises in aggregate food price indices. These arguments apply more directly to the aggregate index IAF than to the sub-indices IAG and IAV where specific sectoral variables may be important. The main demand variable considered is a measure Y of world GDP volume. Importantly, this measure includes GDP for China, India and Russia as well as for the advanced economies. National currencies are converted into US dollars using PPP exchange rates taken from Ahmad (2003).

Section 3 focused attention on the oil price O with the implication that this may have been more influential in the recent past than historically as the result of the move toward biofuels. This argument relates more clearly to grains and vegetable oils than to the overall food price index. Section 4 implied that changes in the value of the dollar should impact dollar commodity prices. The analysis employs a measure X of the real value of the US dollar against a basket of major traded currencies. Section 5 suggested monetary expansion and futures market activity as candidates for non-fundamental explanation of price changes. This motivates examination of the impact of changes in world real money supply M. Futures market activity is measured by open interest FOI summed over the three main Chicago grains futures markets (corn, soybeans and wheat). To the extent that transmission is via futures markets, one should expect to see the greatest effect on grains prices as, with the exception of rice, this is the sector where futures trading is most active (for full details on data construction, see the Data Appendix).

Table 3 reports Augmented Dickey–Fuller (ADF) tests on the variables used in the Granger causality analysis. The table confirms that all variables are I(1) with the exception of the exchange rate index where the hypothesis of non-stationarity is marginally rejected. More importantly, all the differenced variables are I(0). The Granger causality tests are based on ADL(2,2) (autoregressive distributed lag) equations

  • image(10)

These equations are balanced in view of the fact that the differenced variables are all integrated in the same order. The choice of lag length reflects prior tests from a more general specification with a uniform specification of nine lags. In each case, the null hypothesis of Granger non-causation is

  • image(11)

and the alternative is its negation. The null and alternative hypotheses are unchanged. Results are reported in the third column of Table 4.

Table 3.    Augmented Dickey–Fuller tests
  1. Notes: ADF tests for the variables used in the Granger causality analysis and their first differences are reported. Tests are conducted over the sample 1971q1–2008q4 and include a constant but no trend except where indicated. In no case is a trend included in tests on the differenced variables. Lag lengths were selected on the basis of non-significance the t-statistic on subsequent lags. 5% critical values −2.88 (no trend), −3.44 (with trend). Statistically significant values are shown in bold face.

lnIAFADF(1) −1.25ΔlnIAFDF −8.87
lnIAGADF(3) −2.03ΔlnIAGADF(3) −5.49
lnIAVADF(1) −2.59ΔlnIAVADF(3) −6.01
lnYADF(1)Trend−1.58ΔlnYADF(3) −5.69
lnXADF(3) 3.03ΔlnXADF(3) −3.88
lnOADF(3) −2.62ΔlnOADF(3) −6.53
lnMADF(3)Trend−3.36ΔlnMADF(2) −4.95
lnFOIADF(2)Trend−2.56ΔlnFOIADF(1)10.99
Table 4.    Granger causality tests
 1971q1–1989q4 (F2,71)1990q1–2008q4 (F2,71)1971q1–2008q4 (F2,14)Chow (F5,142)
  1. Notes: All variables entered are log differences. Tail probabilities are given as percentage in parentheses. The Chow test tests for homogeneity across the two sub-samples.

Agricultural foods price index, ΔlnIAF
 GDP, Y3.40 (3.9)2.66 (7.7)5.99 (0.3)0.85 (67.5)
 Dollar exchange rate, X2.48 (9.1)1.48 (23.4)4.17 (1.7)0.75 (74.5)
 Oil price, O2.85 (6.5)3.68 (0.3)0.15 (86.2)3.15 (9.8)
 Money supply, M6.74 (0.2)1.33 (27.1)7.79 (0.1)1.39 (39.0)
 Futures open interest, FOI0.48 (62.3)2.59 (8.2)2.34 (10.0)0.68 (80.0)
Grains price index, ΔlnIAG
 GDP, Y2.56 (8.4)0.85 (43.1)2.90 (5.8)2.32 (17.3)
 Dollar exchange rate, X3.43 (3.8)3.89 (2.5)4.86 (0.9)3.17 (9.7)
 Oil price, O0.22 (79.6)3.86 (2.6)0.98 (37.6)3.41 (8.4)
 Money supply, M4.23 (1.8)4.61 (1.3)6.27 (0.2)3.21 (9.4)
 Futures open interest, FOI0.24 (78.6)1.79 (17.5)0.93 (39.9)2.60 (14.2)
Oils and fats price index, ΔlnIAV
 GDP, Y1.77 (17.8)0.23 (79.5)2.30 (10.4)1.07 (54.0)
 Dollar exchange rate, X3.02 (5.5)0.55 (57.9)3.65 (4.4)1.34 (41.2)
 Oil price, O0.12 (88.4)1.11 (31.7)0.01 (99.0)1.36 (40.3)
 Money supply, M4.21 (1.9)0.48 (62.0)3.57 (3.1)1.85 (25.5)
 Futures open interest, FOI1.14 (32.6)0.44 (64.6)0.84 (43.3)5.38 (3.2)

With a 38-year sample, there may be a concern that different factors have been operative at different periods of time. I therefore split the sample in two at its mid-point (end-1989). The table reports Granger causality tests for the two sub-samples and the entire sample. A standard Chow test (Chow, 1960) for sample homogeneity indicates that, in every case, it is acceptable to consider the test statistics for the entire sample.

I first consider the two variables which can be taken as both fundamental and common across the entire group of food commodities, GDP growth Y and the exchange rate X. Granger causality is established for the overall index IAF but the tests for the grains index (narrowly) and the vegetable oils index fail to reject Granger non-causality. This finding is in line with the analysis in section 2 which implied that demand-side factors should appear more important for indices at high levels of aggregation. In the case of exchange rate changes, Granger causality is established with respect to all three indices.

The biofuels discussion in section 3 suggested that we should expect that the link between food prices and the oil price should be expected to have become stronger over recent years. Granger causality is established both for IAF and for IAG in the more recent sub-sample but not in the earlier sub-sample. These results are supportive of the view that there may have been a closer link between the oil price and food prices over recent years relative to earlier periods. There is no evidence of a statistically significant causal link from the crude oil price to vegetable oil prices.

Finally, consider the effects of monetary expansion M and futures open interest FOI. The test results confirm a Granger-causal relationship between monetary expansion and changes in food prices for the earlier sample for all three indices. Granger causality is also established for grains prices in the second sub-sample. Although the second sub-sample does not support Granger causality for the overall index and the vegetable oil sub-index, by virtue of sample homogeneity and the positive results for the entire sample, neither does it provide any evidence against this hypothesis. Taken together, these tests support the conclusion that money supply has affected food commodity prices not only in the earlier part of the period considered but also in more recent years. In the case of futures open interest, the tests fail to reject the null of Granger non-causality. However, high open interest may result either from long or short speculative positions and it is therefore not entirely surprising that no consistently signed effect is found.

Granger causality tests are subject to a well-known list of qualifications. They do not say anything about contemporaneous causation, likely to be particularly important (in either direction) for futures market variables. In relation to causation over time, they are vulnerable to missing variable bias which would amount, in the context of causal analysis, to the case of a third variable z which causes both x and y which are themselves causally independent – the grass is green (y) when the river is high (x) because of recent rain (z). Despite these qualifications, a finding that x Granger-causes y does provide very strong prima facie evidence of a (direct or indirect) causal relationship from x to y.

Although the results reported in Table 4 cannot be claimed as tests of the hypotheses advanced in sections 2–5, they do provide strong corroborative evidence that:

  • 1
     demand-side (macroeconomic) variables, such as GDP growth, are important in explaining aggregate food price indices;
  • 2
     the importance of the oil price as a determinant of food prices may have increased over the recent past, perhaps because of demand for food crops for biofuels purposes; and
  • 3
     monetary expansion does directly impact agricultural food prices, and appears to have done so consistently over time.

7. 2006–2009

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Commodity Price Booms
  5. 3. The Oil Price and Biofuels
  6. 4. Exchange Rates
  7. 5. Monetary Factors and Futures Markets
  8. 6. Granger Causality Analysis, 1970–2008
  9. 7. 2006–2009
  10. 8. Conclusions
  11. References
  12. Appendices

Analysis of price movements in 2006–2009 is more complicated because the time span is shorter. We may take advantage of the availability of data on index positions on agricultural futures markets which are available from the start of that year. Nevertheless, the number of data points remains small. This limitation can be partially counteracted by moving to monthly data, but some of the series (for example, GDP) analysed in section 6 are unavailable on a monthly basis. The analysis reported below is based on monthly data from March 2006 to June 2009 giving 40 observations (the January and February 2006 observations are lost through lag construction). The maximum ambition is therefore determination of the proximate causes of price movements. Background factors, such as low investment over the preceding years, biofuels demand and low levels of stocks, will remain broadly constant over a period of this length.

As previously, food prices are measured using the IMF’s index of agricultural food prices (IAF). Preliminary data analysis narrowed down the choice to three variables:

  • 1
     An exchange rate variable X, constructed in the same way as in the historical analysis.
  • 2
     The oil price O is as previously.
  • 3
     An index FI of futures positions on 12 major US agricultural futures markets constructed from the data in the CFTC’s Supplementary Commitments of Traders Reports available from the start of 2006.8

Variables are period averages and are measured in nominal terms as changes in the aggregate price level have been small. World money supply and industrial production series were also analysed, but these explained little over the sample in question. This was also the case with US ethanol production.

Table 5 lists the ADF statistics for the four variables. The food price variable and the exchange rate index are confirmed as I(1). The oil price appears stationary, but this result evaporates if the sample is extended backwards. There is evidence that the futures investment index may be integrated of order two. Granger causality tests, conducted within an ADL(1,1) framework, rejects the absence of causality from the futures index to the exchange rate food price index but fails to show causal links from the other two variables.9Table 6 lists the correlations of the variables in levels (below the diagonal) and in differences (above the diagonal). Movements in all three candidate explanatory variables show high correlations both with food prices and with each other making it difficult to distinguish their separate effects.

Table 5.    Augmented Dickey–Fuller and Granger causality tests
 ADF testsGranger causality (F1,37)
  1. Notes: ADF tests for the variables used in the 2006–2009 analysis and their first differences and Granger causality tests relative to the variable ΔlnIAF are reported. The ADF tests are conducted over the sample May 2006 (levels) and June 2006 (differences) to June 2009. Lag lengths were selected on the basis of non-significance the AIC criterion. The Granger causality tests, which are within a bivariate ADL(1), are over the sample March 2006 to June 2009. F-tail probabilities are given as percentage in parentheses. ADF 5% critical values −2.94 (no trend). Significant values at the 5% level are indicated in bold face.

lnIAFADF(1)−1.88ΔlnIAFDF−3.44
lnXADF(1)−1.72ΔlnXDF−4.820.51 (48.0)
lnOADF(2)−3.64ΔlnOADF(2)−2.600.79 (38.1)
lnFIADF(1)−2.39ΔlnFIDF−2.224.50 (4.1)
Table 6.    Correlations
 ΔlnIAFΔlnXΔlnOΔlnFI
  1. Notes: Below the diagonal, the correlations between the levels of the variables (January 2006 to June 2009) and, above the diagonal, between their differences (February 2006 to June 2009) are listed.

lnIAF −0.6330.6580.762
lnX−0.967 −0.471−0.579
lnO0.742−0.725 0.703
lnFI0.743−0.7360.884 

I proceed by estimating an equation linking the change in the agricultural food price index ΔlnIAF to changes in the oil price ΔlnO, the exchange rate ΔlnX, and the futures investment index ΔlnFI.10 Efficient market considerations suggest that only contemporaneous variables should be included in the regression. However, averaging may induce first-order serial dependence (Working, 1960). I investigated the use of both the current and single lagged variables but was able to set the coefficients of the lagged exchange rate change and lagged change in the futures investment index to zero. The resulting equation is

  • image(10)

where u1t is a disturbance term.

Possible simultaneity bias resulting from joint determination is clearly a potential problem here. I therefore suppose that the changes in the oil price ΔlnOt and in the futures investment index ΔlnFIt to be endogenous. However, changes in the dollar exchange rate index ΔlnXt are regarded as exogenous to the commodity sector. This requires specification of instruments for the two endogenous regressors or, if a system estimator is used, equations for their determination. Even if using a single equation estimator, it is preferable to adopt the latter approach since this implies the optimal set of instruments for two-stage least squares estimation is the complete set of exogenous regressors in the model excluded from the question in question (Sargan, 1988, p. 48) (here equation 10). The focus remains on the food price equation with the oil price and index investment equations serving to guarantee consistent estimation of, and valid inference from, the food price equation.

I specify the change in the oil price ΔlnO as depending on the change in index futures positions ΔlnFI, the change in Chinese industrial production ΔlnCIP and the a post-Lehman dummy variable LEH defined to be one in October, November and December 2008 and zero otherwise

  • image(11)

where u2t is a disturbance term. There is no evidence of either serial dependence or of a direct exchange rate impact on oil prices. The index investment variable ΔlnFI is specified as depending on the change in the dollar exchange rate ΔlnX, current and lagged changes in Chinese industrial production ΔlnCIP and current and lagged changes in the Standard and Poor 500 and Hang-Seng stock market indices ΔlnSP and ΔlnHS, respectively. This specification is motivated by the results reported in Gilbert (2010) and reflects the view that index-based investment in commodity futures markets has been driven by views about China’s likely appetite for raw materials. The proposed equation is

  • image(12)

where u3t is a disturbance term. There is strong evidence for partial adjustment in index positions, as reflected in the lagged dependent variable but no evidence of any contemporaneous effect from changes in the Hang–Seng index or from lagged changes in the exchange rate index.

Estimation results are reported in Table 7. I estimate using OLS (ignoring potential simultaneity bias), two-stage least squares (2SLS) and three-stage least squares (3SLS). The 2SLS estimates use the complete set of exogenous variables in the model as instruments. (The equation for net index position, ΔlnFI, does not contain endogenous regressors; so, 2SLS and OLS become identical). In general terms, there is little difference between the three sets of estimates and the impact of potential simultaneity appears slight. The discussion which follows focuses on the 3SLS estimates.

Table 7.    Estimates of the monthly model
 ΔlnIAFtΔlnOtΔlnFIt
OLS2SLS3SLSOLS2SLS3SLSOLS3SLS
  1. Notes: Sample: March 2006 to June 2009 (40 observations). t-statistics are given in parentheses and P-values in square brackets. ΔlnIAFt, ΔlnOt and ΔlnFIt are treated as endogenous. Each equation uses as instruments the complete set of excluded variables. The equations all include an intercept. The Sargan instrument validity test is given for each 2SLS equation. The comparable model test of over-identifying restrictions is given at the foot of the 3SLS estimates for the ΔlnIAFt equation.

ΔlnIAFt−10.145 (0.99)0.119 (0.74)0.107 (0.68)     
ΔlnOt0.096 (1.69)0.092 (0.86)0.127 (1.18)     
ΔlnOt−1−0.128 (2.47)−0.140 (2.63)−0.139 (2.65)     
ΔlnFIt0.504 (3.29)0.609 (2.60)0.551 (2.36)0.876 (2.52)1.046 (2.61)1.015 (2.46)  
ΔlnFIt−1      0.512 (5.11)0.520 (5.43)
ΔlnXt−0.458 (1.79)−0.358 (1.31)−0.365 (1.34)   −0.432 (2.51)−0.458 (2.75)
ΔlnCIPt   1.200 (1.17)1.023 (0.98)0.994 (0.92)1.128 (4.23)1.073 (4.20)
ΔlnCIPt−1      0.647 (2.07)0.641 (2.15)
LEHt   −0.205 (3.41)−0.186 (2.90)−0.196 (2.97)  
ΔlnSPt      0.242 (3.42)0.235 (3.48)
ΔlnSPt−1      −0.365 (2.62)−0.372 (2.81)
ΔlnHSt−1      0.178 (2.00)0.180 (2.12)
SE0.0266 0.02680.0269 0.07210.0724 0.07440.0191 0.0186
R20.709  0.664  0.876 
Instrument validity inline image = 3.60 [60.9%]inline image = 18.5 [23.6%] inline image = 7.32 [29.2%]   

The estimates show a statistically significant impact of index investment on both the food price index and on the oil price. In quantitative terms, the latter impact is around twice the size of the former in line with the greater weight of crude oil in the indices which the investments track. The contemporaneous impact of the oil price on the food price index is poorly determined, possibly as the result of weak instrumentation of the oil price, and is offset by an approximately equal, and better determined, lagged impact. The net effect is the food prices appear to have been impacted by the second and not the first difference in the oil price. The oil price equation is less well determined. Finally, the index investment equation is consistent with investors regarding the ‘commodity investment class’ as both a dollar hedge and a means of taking exposure to Chinese economic growth.

We can use this model to examine the causes and channels of the 2006–2008 food price spike. Write the three endogenous variables in equations (10)–(12) as yt and the six exogenous variables as zt so that

  • image

The model may be summarised as inline image where L is the lag operator. Inversion gives

  • image

with static long-run solution inline image.

Table 8 reports the estimated long-run multipliers (omitting those associated with the intercept and the dummy variable LEH). Both food prices and the oil price have an elasticity almost exactly equal to unity despite there being no direct exchange rate impact on the oil price. The oil price is seen as twice as sensitive to Chinese growth as is the food price. Both prices react to the differential between returns on the Hang-Seng and Standard and Poors indices consistent with price rises being generated by a shift in activity from North America to China. This indicates that the fundamental drivers of the 2006–2008 commodity price boom were (current and anticipated) Chinese growth and dollar depreciation.

Table 8.    Estimated long-run impacts
 ΔlnIAFΔlnOΔlnFI
  1. Note: The table reports the estimated static solution of the 3SLS estimates reported in Table 7.

ΔlnX−0.98−0.97−0.96
ΔlnCIP2.144.623.57
ΔlnSP−0.17−0.29−0.29
ΔlnHS0.230.380.37

Index investment in commodity futures disappears from this breakdown as it is seen as endogenous and hence a channel rather than a fundamental cause of the price rises. In order to analyse the importance of this channel we need to return to the estimates of equation (10) and condition on the exchange rate and index investment variables. The agricultural food price index rose by 77.2% from January 2006 to its peak in June 2008. This corresponds to a logarithmic price change of ΔlnIAF = 0.5735. Table 9 reports the breakdown of this price change into the components implied by the 3SLS estimates of equation (10). The estimates in the final column of the table show index futures investment to have been the preponderant channel through which the fundamental causal effects in Table 8 affected food prices over the food price boom. However, the channels through which monetary factors influence agricultural and other prices are likely to vary over time as institutions evolve and this conclusion may turn out to be specific to the particular period under investigation.

Table 9.    Channels of impact
  29-month changeImplied price change
%log change%log change
  1. Notes: The changes in the logarithm of the agricultural food price index lnIAF over the 29-month period January 2006 to June 2008 attributable to each of the three factors entering equation (10) using the 3SLS estimates reported in the third column of Table 7 are presented. Note that the log changes (including the residual) sum to the total, but this property does not hold for percentage changes.

Dollar depreciationlnX15.40.16757.10.0685
Oil pricelnO104.50.7156−1.0−0.0096
Futures index positionslnFI77.40.573542.50.3538
Total   48.60.4217
Residual   17.30.1595
Agricultural food priceslnIAF  77.20.5722

Dollar deprecation, which has been regarded as an exogenous factor, appears in both Tables 8 and 9. Table 9 attributed a 7.1% direct impact of dollar deprecation to the rise in food prices. However, depreciation also, and perhaps more importantly, drives commodity futures investment as a dollar hedge. The near unit elasticity reported in Table 8 implies that the total, direct plus indirect, impact of dollar appreciation on food prices, was 15.1%, somewhat less than the estimate given by Mitchell (2008).

The results reported in Table 8 indicate that the oil price was driven by the same set of common factors as food prices albeit with different weights. This can account for the high correlation of the two series over this period (see Table 6). Nevertheless, there does not appear to have been any direct net impact of oil prices on food prices over this period (see Table 9). This is in line with the Granger causality test reported in Table 5 and the observation, in section 4, that cost pass-through will primarily be observed when markets are weak. It also accords with the results in Table 6 where the oil price effects on food prices were seen as ambiguous with a sign reversal between the period 1970–1989 and 1990–2009.

Neither the level of stocks nor biofuels demand appears explicitly in Table 8 or Table 9. Low stock levels will have been a determinant of the magnitude of the estimated elasticities, whereas any shock to stock demand will be in the residual. Any biofuels effect on agricultural prices should also be contained within the Table 9 residuals. This implies that biofuels demand was responsible for at most one-quarter to one-third of the rise in food prices over 2006–2008, considerably less than implied in a number of discussions. Even this estimate may be too high as other factors may have contributed to the residual. This judgement does not, of course, rule out the possibility that biofuels demand may have been important in relation to the price rises for particular food crops. Neither does it directly contradict the projections advanced by Rosegrant et al. (2008) which show biofuels demand pushing up food prices over the period 2020. However, echoing the Cooper and Lawrence (1975) discussion of the 1973–1974 price spike, biofuels demand in 2006–2008 may only have been an ‘intriguing’ and perhaps ‘significant story’ relating to a number of particular markets. In any event, the available evidence does not support the suggestion that use of food commodities as biofuel feedstocks should be radically curtailed.

8. Conclusions

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Commodity Price Booms
  5. 3. The Oil Price and Biofuels
  6. 4. Exchange Rates
  7. 5. Monetary Factors and Futures Markets
  8. 6. Granger Causality Analysis, 1970–2008
  9. 7. 2006–2009
  10. 8. Conclusions
  11. References
  12. Appendices

Discussions of the causes of commodity prices tend to adopt a commodity-specific framework in which price changes are seen as the result of largely idiosyncratic supply shocks. This approach may not be helpful in analysing major booms, such as those of 1972–1974 and 2006–2008, in which a large number of prices rise together. Market-specific explanations require too much coincidence and the resulting price responses to shocks may seem disproportionate. Idiosyncratic shocks, which may dominate explanations in particular markets, will tend to fade in importance if one is concerned with food prices in general.

The argument of paper has been that common factor relating to demand growth and monetary and financial developments, are likely to be the main determinants of changes in the overall level of agricultural prices. In line with this view, Granger causality tests show world GDP growth, monetary expansion and exchange rate changes as determinants of changes in world agricultural prices over a 38-year period. The oil price and the dollar exchange rate have also been important causal factors, but the impact of the former has varied over time while exchange rate effects, which are consistent over time, are relatively small.

There is a widespread view that the major demand shock experienced by the agricultural sector over the recent boom arose from the demand for grains and oilseeds as biofuel feedstocks. Some commentators, most notably Mitchell (2008), have argued that potential and actual biofuels demand has resulted in a strengthened link between food prices and the oil price and have suggested that diversion of food commodities into biofuels was the main cause of the 2007–2008 price spike. This is a residual argument with the attribution of price rises to biofuels demand reflecting lack of other available explanations. There is little direct evidence for the claimed link which appears less important than has been suggested. The evidence in this paper suggests that the correlation between the oil price and food prices, both in terms of levels and changes, is the result of common causation and not of a direct causal link. These results do not offer support for restrictions on the use of food commodities as biofuel feedstocks.

Monetary factors were an important determinant of food price rises in the 1970s, but monetary transmission channels tend to vary over time. The evidence in this paper suggests that index futures investment was the principal channel through which monetary and financial activity have affected food prices over recent years. The general rise in energy and metals prices stimulated interest in commodity futures as an asset class. This activity is sufficiently large that it has the potential to move prices. Purchases of long futures positions by index investors may be seen as a positive shock to inventory demand. By investing across the entire range of commodity futures, index-based investors appear to have inflated food commodity prices. Paradoxically, this may have generated the comovement predicted by the commodity asset class perception. The direct impact of dollar deprecation on food prices was smaller but dollar depreciation also had an indirect effect by stimulating the build-up of index positions as a dollar hedge.

There has been some discussion of the need for increased futures market transparency and of tighter positions limits on speculative activities. These proposals are premised on viewing index investment as similar to traditional short-term, trend-chasing, speculative activities. However, the fundamental driver of index investment has been the belief that rapid economic growth in China (and perhaps the rest of Asia) will increase demand for the entire commodity asset class. Food commodities, for which China is not always important, joined in the general rise in commodity prices as the result of their share of the indices the investments sought to track. It is difficult to believe that more stringent regulation of futures markets could make the China factor go away (although there may be other merits in tougher regulation).

Footnotes
  • 2

    I confine myself in this account to discussion of food commodities and do not consider the transmission of changes in these prices into retail food prices.

  • 3

    I have simplified this discussion by assuming that the idiosyncratic shocks have a common variance inline image. Relaxation of this assumption introduces an additional term into the denominator of the R2 expressions in equations (5a)–(5c) but does not change the qualitative conclusion.

  • 4

    Consider the polar case in which the medium-term food crop supply curve is a reverse-L with the kink at land availability, taken as variable only in the long run. A cost increase shifts the L-supply curve upwards. So long as the demand curve intersects the L on its horizontal section, there is full pass-through. If the intersection is on the vertical, there will be no pass through. As different crops have different energy intensities, there will be changes in relative food prices but little change in aggregate food price indices.

  • 5

    An early draft of the paper saw dollar depreciation and the oil price as contributing 35% of the 140% rise in grain prices with the remainder attributed to biofuels demand.

  • 6

    Monetary and exchange rate explanations of commodity price movements are related. Taylor (1995) argued that the effects of unilateral monetary expansion in a single country will come largely through currency depreciation. Consistently with this argument, Awokuse (2005) concluded that US agricultural prices were determined primarily by changes in the value of the dollar and that monetary factors had relatively little impact on agricultural prices. In the opposite case of general worldwide monetary expansion, exchange rates may not change markedly and transmission to commodity prices will be through general asset price inflation.

  • 7

    Irwin et al. (2009) have advanced a number of ‘logical inconsistencies’ in the view that futures market activity can move prices. These arguments are repeated in Sanders and Irwin (2010). Their main argument is that the link between futures activity and cash prices is complex and unclear (see also Headley and Fan, 2008). In particular, index investors neither participate in the delivery process where, they claim, price discovery takes place nor hold physical inventory. However, in many markets, the bulk of off-market transactions are at futures and not cash prices so that discovery takes place in relation to nearby contracts as well as at delivery, which is often only the discovery mechanism of last resort. And even though index traders do not hold physical inventory, by bidding up more distant prices they will provide others with an increased incentive to hold inventory. I conclude that there is no logical inconsistency in the claim that index traders may have been at least partially responsible for the food price boom and that the matter needs to be settled empirically.

  • 8

    The markets covered are corn (maize), soybeans, soybean oil and soft wheat (Chicago Board of Trade), hard wheat (Kansas City Board of Trade), cocoa, coffee, cotton and sugar (New York Board of Trade) and feeder cattle, lean hogs and live cattle (Chicago Mercantile Exchange). Positions are measured in contracts. The index uses base period price weights of 3 January 2006. I am grateful to Elena Corazzolla for research assistance in the construction of this index.

  • 9

    A Granger causality test for the effect of US ethanol production on the food price index gave a test statistic of F1,37 = 1.08 with tail probability 30.5%. Data source for ethanol production: US Energy Information Administration.

  • 10

    An earlier version of this paper adopted a cointegration approach. With the extension of the dataset to the first 6 months of 2009, the oil price appears stationary (see Table 5) which precludes this approach. The vector autoregression (VAR) approach is unattractive in the current context because of the importance of contemporaneous interactions. This motivates choice of a simultaneous equations model (SEM).

References

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Commodity Price Booms
  5. 3. The Oil Price and Biofuels
  6. 4. Exchange Rates
  7. 5. Monetary Factors and Futures Markets
  8. 6. Granger Causality Analysis, 1970–2008
  9. 7. 2006–2009
  10. 8. Conclusions
  11. References
  12. Appendices
  • Abbott, P. C., Hurt, C. and Tyner, W. E. What’s Driving Food Prices? (Oak Brook, IL: Farm Foundation, 2008).
  • Ahmad, S. Purchasing Power Parity (PPP) for International Comparisons of Poverty: Sources and Methods (Washington, DC: World Bank, 2003).
  • Awokuse, T. O. Impact of macroeconomic policies on agriculture’, Agricultural and Resource Economics Review, Vol. 34, (2005) pp. 226237.
  • Baffes, J. Oil Spills on Other Commodities, Policy Research Working Paper #4333 (Washington, DC: World Bank, 2007).
  • Bordo, M. D. The effect of monetary change on relative commodity prices and the role of long term contracts’, Journal of Political Economy, Vol. 88, (1980) pp. 10881109.
  • Cashin, P., Céspendes, L. and Sahay, R. Commodity currencies’, Finance and Development, Vol. 40, (2003) pp. 4548.
  • Cashin, P., Céspendes, L. and Sahay, R. Commodity currencies and the real exchange rate’, Journal of Development Economics, Vol. 75, (2004) pp. 239268.
  • CFTC. Staff Report on Commodity Swap Dealers and Index Traders with Commission Recommendations (Washington, DC: CFTC, 2008).
  • Chambers, R. G. and Just, R. E. An investigation of the effects of monetary factors on agriculture’, Journal of Monetary Economics, Vol. 9, (1982) pp. 235247.
  • Chow, G. Tests of equality between sets of coefficients in two linear regressions’, Econometrica, Vol. 28, (1960) pp. 591605.
  • Cooke, B. and Robles, M. Recent Food Price Movements: A Time Series Analysis, IFPRI Discussion Paper, 00942 (Washington, DC: IFPRI, 2009).
  • Cooper, R. N. and Lawrence, R. Z. The 1972–75 commodity boom’, Brookings Papers on Economic Activity, Vol. 1975, (1975) pp. 481490.
  • De Long, J. B., Shleifer, A., Summers, L. H. and Waldman, R. J. Positive feedback investment strategies and destabilizing rational expectations’, Journal of Finance, Vol. 45, (1990) pp. 379395.
  • Fabozzi, F. J., Füss, R. and Kaiser, D. G. A primer on commodity investing’, in F. J.Fabozzi, R.Füss and D. G.Kaiser (eds.), The Handbook of Commodity Investing (Hoboken, NJ: Wiley, 2008, pp. 337).
  • Friedman, M. Essays in Positive Economics (Chicago, IL: University of Chicago Press, 1953, pp. 157203).
  • Gilbert, C. L. The impact of exchange rates and developing country debt on commodity prices’, Economic Journal, Vol. 99, (1989) pp. 773784.
  • Gilbert, C. L. Speculative Influences on Commodity Futures Prices 2006–08, Discussion Paper #197, UNCTAD/OSG/DP/2010/1 (Geneva: UNCTAD, 2010).
  • Headley, D. and Fan, S. Anatomy of a crisis: The causes and consequences of surging food prices’, Agricultural Economics, Vol. 39, (2008) pp. 375391.
  • IFPRI, High Food Prices: The What, Who, and How of Proposed Policy Actions, Policy Brief (Washington, DC: IFPRI, 2008).
  • Irwin, S. H. and Holt, B. R. The effect of large hedge fund and CTA trading on futures market volatility’, in G. N.Gregoriou, V. N.Karavas, F.-S.Lhabitant and F.Rouah (eds.), Commodity Trading Advisors: Risk, Performance Analysis and Selection (Hoboken, NJ: Wiley, 2004, pp. 151182).
  • Irwin, S. H. and Yoshimaru, S. Managed futures, positive feedback trading, and futures price volatility’, Journal of Futures Markets, Vol. 19, (1999) pp. 759776.
  • Irwin, S. H., Sanders, D. R. and Merrin, R. P. Devil or angel? The role of speculation in the recent commodity boom (and bust)’, Journal of Agricultural Applied Economics, Vol. 41, (2009) pp. 393402.
  • Lintner, J. The valuation of risk assets and the selection of risky investments in stock portfolios and capital budgets’, Review of Economics and Statistics, Vol. 47, (1965) pp. 1337.
  • Masters, M. W. Testimony before the U.S. Senate Committee of Homeland Security and Government Affairs (Washington, DC, 20 May 2008).
  • Mitchell, D. A Note on Rising Food Prices, Policy Research Working Paper #4682 (Washington, DC: World Bank, Development Prospects Group, 2008).
  • O’Hara, M. Market Microstructure Theory (Oxford: Blackwell, 1995).
  • Phillips, P. C. B. and Yu, J. Dating the Timeline of Financial Bubbles during the Subprime Crisis (Cowles Foundation for Research in Economics, Yale University, 2009).
  • Radetzki, M. The anatomy of three commodity booms’, Resources Policy, Vol. 31, (2006) pp. 5664.
  • Ridler, D. and Yandle, C. A. A simplified method for analyzing the effects of exchange rate changes on exports of a primary commodity’, IMF Staff Papers, Vol. 19, (1972) pp. 559578.
  • Rosegrant, M. W., Zhu, T., Msangi, S. and Sulser, T. Global scenarios for biofuels: Impacts and implications’, Review of Agricultural Economics, Vol. 30, (2008) pp. 495505.
    Direct Link:
  • Sanders, D. R. and Irwin, S. H. Bubbles, Froth and Facts: The Impact of Index Funds on Commodity Futures Prices (University of Illinois at Urbana-Champaign, processed, 2009).
  • Sanders, D. R. and Irwin, S. H. A speculative bubble in commodity futures prices? Cross-sectional evidence’, Agricultural Economics, Vol. 41, (2010) 2532.
  • Sargan, J. D. Lectures in Advanced Econometric Theory (Oxford: Blackwell, 1988).
  • Schmidhuber, J. Impact of Increased Biomass Use on Agricultural Markets, Prices and Food Security: A Longer Term Perspective (Rome: Global Perspectives Unit, FAO, 2006). Available at: http://www.fao.org/ESD/pastgstudies.html.
  • Sharpe, W. F. Capital asset prices: A theory of market equilibrium under conditions of risk’, Journal of Finance, Vol. 19, (1964) pp. 425442.
  • Soros, G., Testimony Before the U.S. Senate Commerce Committee Oversight Hearing on FTC Advanced Rulemaking on Oil Market Manipulation (Washington, DC, 4 June 2008).
  • Svensson, L. E. O. Money and asset prices in a cash-in-advance economy’, Journal of Political Economy, Vol. 93, (1985) pp. 919944.
  • Taylor, J. B. The monetary transmission mechanism: An empirical framework’, Journal of Economic Perspectives, Vol. 9, (1995) pp. 1126.
  • US Senate Permanent Subcommittee on Investigations. Excessive Speculation in the Wheat Market, Majority and. Minority Report (Washington, DC: US Senate, 24 June 2009).
  • Working, H. Note on the correlation of first differences of averages in a random chain’, Econometrica, Vol. 28, (1960) pp. 916918.

Appendices

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Commodity Price Booms
  5. 3. The Oil Price and Biofuels
  6. 4. Exchange Rates
  7. 5. Monetary Factors and Futures Markets
  8. 6. Granger Causality Analysis, 1970–2008
  9. 7. 2006–2009
  10. 8. Conclusions
  11. References
  12. Appendices

Model Appendix

Consider a single fixed input z, a vector of m > 1 outputs x and a vector of n ≥ 1 variable inputs y. Define the minimal fixed input requirements to produce x with y as inline image. Let the market prices of the m outputs be p and those of the n variable inputs be q. Gross profits are

  • image( (A1))

The product supply and input demand functions are

  • image

respectively. The Slutsky matrix

  • image

where S11 has ijth element sij where

  • image( (A2))

The matrix S11 is positive definite. Write the ijth element of inline image as sij.

Let demand for the m outputs be inline image which for simplicity I take to be unaffected by prices. The initial prices p0 and q0 satisfy inline image

Now consider a set of demand shocks u and input price shocks w. For simplicity, take supplies of variable inputs to be perfectly elastic. Expanding about p0 and q0, the prices move to p1 = p0 + Δp where

  • image

Inverting,

  • image( (A3))

where P is the diagonal matrix with jjth element inline image.

Focus on demand shocks. We may suppose that these have a common component v and an idiosyncratic component ε, i.e. u = vl + ε, where l is the vector of units. These two components are independent and the idiosyncratic shocks are mutually independent. Consider a unit idiosyncratic shock to the demand for product 1, ε1. In this case

  • image

by positive definiteness of S11. By contrast, the effect of a unit common shock v on the price of product 1 is

  • image

again by positive definiteness of S11. The difference in the two effects is

  • image( (A4))

We should expect the off-diagonal terms in the S11 matrix to be negative. Write inline image where Σ is the diagonal matrix

  • image

and hence all the diagonal elements of inline image are zero. Define the vector

  • image

Positive definiteness of S11 requires inline image. The inverse matrix inline image satisfies inline image. Provided the elements of inline image are small, to a first-order approximation inline image. It follows that

  • image( (A5))

A sufficient condition for a unit common shock to have a greater impact on the price of a good is therefore that the weighted sum of the cross-price effects be negative, i.e.

  • image

where the weights are the own price effects.

Data Appendix: Variable Definitions

Data are quarterly over the sample 1969q1–2008q4 or monthly over the sample January 2006 to June 2009. The data file used in this paper is available from the author upon request.

CIP

industrial production, People’s Republic of China (2005 = 100). Source: Haver Analytics

FI

net index provider positions on US agricultural futures markets (number of contracts). Source: Author’s calculations from data in CFTC, Supplementary Commitments of Traders Reports

FOI

futures market open interest (contracts of 5,000 bushels) in the Chicago Board of Trade corn, soybeans and wheat contracts, summed over all contracts and averaged over the quarter. Source: Chicago Board of Trade

HS

Hang–Seng equities index, closing prices, monthly averages. Source: Yahoo Finance

IAF

index of agricultural food commodity prices (2000 = 100), deflated by the US GDP deflator (2000 = 100) in the quarterly dataset; (2005 = 100) undeflated in the monthly dataset. Source: IMF, International Financial Statistics

IAG

index of grain (maize, rice, soybeans and wheat) commodity prices (2000 = 100), deflated by the US GDP deflator (2000 = 100). Source: IMF, International Financial Statistics

IAV

Index of vegetable oil (palm oil, soybean oil and sunflower oil) commodity prices (2000 = 100), deflated by the US GDP deflator (2000 = 100). Source: IMF, International Financial Statistics

LEH

Post-Lehman dummy variable, 1 in 2008.10, 2008.11 and 2008.12, zero otherwise

M

World money supply. Money supply of China (M2), the Euro zone (M2), India (money and quasi-money), Japan (M2), Russia (M2), Switzerland (M2), the UK (M4) and the USA. (M2) converted into US dollars at the average exchange rate of the month and deflated by the US GDP deflator (2000 = 100). Pre-1999 Euro zone money supply figures are calculated from the money supplies of France, Germany and Italy scaled up to the 1999q1 Eurozone aggregate. Source: IMF, International Financial Statistics

O

Oil price: WTI, New York, $ per barrel, deflated by the UD GDP deflator (2000 = 100) in the quarterly dataset, undeflated in the monthly dataset. Source: IMF, International Financial Statistics

SP

Standard and Poors 500 equities index, adjusted closing prices, monthly averages. Source: Yahoo Finance

X

US dollar exchange rate relative to basket comprising the euro (Deutschmark prior to 1999), yen, Australian dollar and Canadian dollar with weights 2 : 2 : 1 : 1. In the quarterly dataset, exchange rates are deflated by the relativity between the country’s GDP deflator and the US GDP deflator. The German GDP deflator is used to deflate the euro exchange rate. Exchange rates are undeflated in the monthly dataset. Source: IMF, International Financial Statistics

Y

World GDP at constant 1993 prices. National currencies are converted to US dollars using the PPP exchange rates in Ahmad (2003). GDP is aggregated over Australia, Canada, China (mainland, from 1978), France, Germany, India (from 1988), Italy (from 1980), Russia (from 1995), Spain (from 1970), Switzerland, the UK and the USA. Annual figures for China and India (to 1996) are interpolated on a quarterly basis. As data are absent for many countries at the start of the sample, GDP levels are extrapolated backwards iteratively using the world GDP estimate such that, in any particular quarter, the estimated GDP growth rate depends only on data for countries for which there are published data. Source: IMF, International Financial Statistics