Commodity Price Dynamics in
Do speculators destabilize the market?
Getaw Tadessea and Atle Guttormsenb
Department of Economics and Resource Management,
Norwegian
Fax:+476496
aTel: +4764965695
E-mail: getaw.gebreyohanes@umb.no
bTel: +4764965691
E-mail: atle.guttormsen@umb.no
ABSTRACT
The impact of speculative storage on price dynamics in Ethiopian grain markets is tested using monthly data for the period 1996-2006. The analysis relies on the classical storage model, modified to account for periodic correlation of shocks and the effects of interest rate change. The predictions of the model are tested using a reduced form threshold autoregression. Several non-linearity tests are adapted. Regime switching normalized maximum likelihood estimation method was formulated. The results show that speculative storage is detrimental in the price formation process of commodity markets, causing threshold non-linearity and impacts price levels to serially correlate over time. Storage fails to adequately stabilize price volatility. Interest rate change in formal financial markets appears to have no effect on storage and price dynamics. Policy implications are discussed.
JEL classification: Q11; Q13;
Keywords: speculative storage, price dynamics,
periodic thresholds, grain markets,
1. Introduction
The rationale for state intervention in stabilizing agricultural
prices has been long questioned (Newbery and Stiglitz
1981).
Nevertheless, the ever-widening of price volatility and its adverse effect on
producers and consumers present both an opportunity and a challenge for policy
making concerning how to define an appropriate state role in terms of price
risk management. The challenge becomes worse when the sources of enhanced price
volatility are not clearly known. Untimely and excessive government intervention
and speculative storage[1] are two
often mentioned sources of price volatility in
At the first indication of a price spike, the hoarding behavior of grain trader is instantly raised as culprit. This claim is not without reason. Grain storage has long been recognized as having substantial impact on the price dynamics of storable commodities (Williams and Wright 1991). The explanation is straightforward. If the market offers temporal arbitrage due to supply fluctuations, profit-maximizing traders will engage in intertemporal trading. These speculators buy when price is low and sell when the price rises so that the price difference across time will depend only on storage costs and the opportunity cost of capital. Thus, speculative storage would cause prices to stabilize over seasons. However, it would also cause strong autocorrelation that explodes price level in the long run. Since Gustafson (1958), many economists (Samuelson 1971; Wright and Williams 1982; Scheinkman and Schechtman 1983; Deaton and Laroque 1992) have attempted to systematically investigate the role of commodity storage on price dynamics using classical commodity storage theory. However, due to the variability of the nature of the problems across economies and methodological shortcomings, the topic still remains live and new insights and empirics are required (Wright 2001).
Two major methods are used in empirical commodity price analysis. These are simulations based on dynamic programming and threshold analysis based on observed time series. The methods assume a shock at harvest time that translates into a constant threshold in all periods. This assumption is quite restrictive, particularly for developing countries where production is constant over seasons in a year. A new approach that considers the periodic correlation of harvests seems more relevant to such settings.
In response to these perceived shortcomings in previous approaches and the on-going policy debate, this paper aims to evaluate the effectiveness of speculative storage and structural change[2] in characterizing the intertemporal price dynamics of commodity markets using a periodic threshold analytical method. To these ends, the presence of threshold non-linearity is tested, where price behavior is understood to depend on the price level. Price levels and price volatility in models with and without storage are compared using alternative threshold models. In addition, the effect of interest rate changes on speculative storage and price dynamics is examined.
The paper contributes to the ongoing debate regarding the sources of serial correlation in commodity prices by providing empirical evidence in an environment where markets are thin, production is stochastic and interregional trade is limited. Moreover, it demonstrates how periodic threshold models can be used to explain price dynamics of a commodity whose harvests are correlated over time. Studies on commodity storage and speculative behavior help to shed light on price stabilization interventions that have long been controversial features of many commodity markets (Wright 2001). Thus, the study aims to identify possible price stabilization strategies and provide market performance information that could be used for analyzing the efficiency and distributional implication of stabilization strategies.
The rest of the paper is organized as follows. Section 2 reviews pervious studies on Ethiopian
grain prices. The storage model and its extension are presented in section 3.
Section 4 outlines the econometric methods that are employed to estimate the parameters
of different threshold and structural break models. Starting with tests of non-linearity, section
5, presents and discusses the estimated results. The final section points the
concluding remarks and policy implications. It also indicates the shortcomings
of the paper and their implications for future research.
2. Review of Price dynamics in
The Ethiopian grain market has undergone a series of reforms in recent years. Following the 1990 market liberalization program, many structural changes have been carried out. These include the reformation of the state-owned marketing boards, lifting of trade-control check points and the opening up of markets to private traders. Many farmers’ cooperative organizations have been established. They are actively involving on grain marketing, primarily to safeguard producers from price risk. The government-owned grain trade enterprise has engaged in accumulation of buffer stocks and engaged in grain import and export. Rationing of low-price supplies, restrictions on grain exports and increases in bank interest rates are the most recent polices aimed at curbing inflation observed in 2005 to 2007.
Despite these reforms, price instability remains a major problem of the economy. Post-reform grain prices are subject to significant and continuing inter-annual price volatility that ranks among the highest in the developing world. Monthly price instability, an indication of a market’s relative performance in realizing arbitrage opportunities, has also worsened in the most recent period. As indicated below (see Figure 2) between 1996 and 2006, monthly grain prices exhibit an average variability of 23%. A substantial price collapse in 2001/2002 resulted in a situation where many farmers were unable to cover even their fertilizer costs. In contrast, in 2005 and 2006 the price of all food grains in all markets increased as much as 80% (MoARD 2006). Price instability, both at the farm gate and in retail markets can reduce the welfare of consumers and producers welfare (Sahn and Delgado 1989; Sadoulet and Janvry 1995). Price instability is detrimental to the flow of agricultural households’ income received for food crops and it undermines consumers’ ability to purchase food.
Previous studies on grain price dynamics concentrated
on spatial price analysis. The studies evaluated the integration and efficiency
of spatially separated markets using monthly price series. (Negassa, Myers et al.
2004),
find that Ethiopian grain markets are characterized by high probability of
spatial inefficiency. The study confirms the existence of periodic gluts and
shortages. The nature of inefficiency, however, varies across commodities.
Maize traders are found to be making losses most of the time while wheat
traders are found to be making profits most of the time. (Negassa
1998)
assesses the vertical and horizontal integration of grain markets at individual
market levels and for groups of spatially-linked markets. Results indicate that
grain markets in
These studies help to illustrate the spatial price
formation process and provide valuable information on where to intervene in the
market and how to integrate markets. However, an appropriate price
stabilization policy requires not only the understanding of spatial price
relationships but also temporal dynamics of the market and its prices. In areas
like
3. Theory and Prediction
3.1. The storage model
with i.i.d. harvest
In order to
comprehend the empirical tests, the storage model presented by (Williams and Wright
1991)
is reformulated as follows. To begin, the
basic theoretical price formation process is presented with supply shocks that
are independently and identically distributed (i.i.d.) and a constant opportunity
cost of tied capital. Consider a grain
market in which risk-neutral speculators engage in storage trade. Further, assume that the traders utilize all
available consumption and production information to set future expected prices.
Let the grain price at time t be
, annual harvest, 
, which is identically and independently distributed due to
weather variability. In the absence of storage, the total consumption is
exactly equal to the total harvest so that price entirely depends on the
stochastic harvest:
Equation represents the inverse demand
function where demand is equal to supply at equilibrium (
). In this framework,
price follows an i.i.d. process. No
serial correlation is expected as long as the harvest is i.i.d.
In the presence of profit-maximizing and risk-neutral speculators in the market, both the demand and supply will change to account for storage in such a way that
Where 
is the state variable
that consists of the stochastic harvest and the carryover level of grain from the
pervious period (
). This is the available grain at any time either for current
consumption or storage. Using equation and , the price formation equation
becomes:
Storage in this case
is a decision variable that must be determined from the speculator’s problem.
The speculator’s problem is to determine the discounted optimal level of
storage that maximizes net profit subject to storage cost and the equation of motion
. (Deaton and Laroque
1992)
showed that a formal derivation of the problem provides us unique stationary
rational expectation equilibrium (SREE) at optimal market price
Where 
represents the
marginal cost of storage that includes all variable costs related to managing
inventory and inventory decay while 
represents the market interest rate. The first term in the parenthesis of
equation indicates the gain of
consuming a unit of grain currently while the second term shows the expected
and discounted gain of storing grain for one more period. The expectation is
made based on the realized current period stock and the expected future
harvest. Here storage is assumed as windfall strategy whereby speculators do
not plan to store a head of time. This means that expected future storage has no
effect on current period price expectation.
The SREE of equation (4) implies the following temporal arbitrage conditions
- If
, then 
and the market
price reduces to (1). Substituting 
from equation (3)
where 
, one obtains:
Since consuming today provides higher value than storing, storage becomes zero and price depends only on the level of harvest. The intertemporal connection is broken and as a result the price series is expected to be less serial correlated than under an assumption of storage. Price volatility mainly arises from supply shocks. This could happen if storage is costly, say due to price risk, high costs of storage or a high opportunity cost of capital. If there exists any speculative trade under such conditions, as commonly observed in developed economies, it must be attributed to either a desire for convenience or ease of transaction (Chavas, Despins et al. 2000).
- If
>, then 
and the market
price is less than the expected gain from storage, that is.
.Equation states that, along the optimal storage path, the current price must be lower than the discounted expected future price and the marginal storage costs. Since storage is positive, prices are expected to be serially correlated. Furthermore, price volatility should depend not only on the stochastic harvest but also on the predetermined level of stocks. Thus, lower volatility is an equilibrium outcome of such a condition.
3.2. Storage with periodic harvest
The i.i.d
assumption is restrictive, and fails to adequately explain price dynamics at
higher prices. Prices could correlate in
the absence of effective private storage due to the correlations of harvests
across seasons. Modeling commodity price dynamics in the presence of serially
correlated stochastic harvest has been attempted earlier (Chambers and Bailey
1996; Deaton and Laroque 1996). Here we present the special form where the
harvest is seasonally correlated but varies over years. If a season is represented by a single month,
say July, the harvest shocks of every July in each year are i.i.d but the
harvest shock of July and other months, say August, are correlated. Thus, the harvest shock behaves as a periodic
disturbance. Periodic disturbances offer a plausible and realistic assumption
for empirical work using monthly price data. It is particularly important for
developing countries where supply shocks are seasonally correlated. As a result of periodic disturbance, the
price formation process becomes different for different seasons. Suppose 
represent seasons or
periods, such that the mean and disturbances of the harvest are 
and 
respectively. The SREE can be modified to accommodate
periodic harvest as
where 
є
.
Equation implies that each period will have different rational expectation equilibrium conditional on observed and expected harvests. Harvests are constant across periods. An important implication of the periodic SREE is that prices correlate not only over time but also over periods. The latter corresponds to harvest while the former corresponds to storage. Extending the i.i.d assumption to a periodic disturbance thus helps to account for a seasonal harvest effect.
- Econometric
Methods
4.1.Threshold Autoregression (TAR)
The analytical methods of section 3 imply empirically testable hypotheses. We turn now to a series of threshold autoregression models, which vary depending on the underlying assumption regarding the probability distribution of shocks.
The major
testable prediction derived from the commodity storage model is that storage
causes prices to behave differently above and below a certain decision
price. The two arbitrage conditions described
by equations and imply that there exists a price level 
below which
speculators maintain stock and prices are functionally related over time, and
above which speculators do not maintain any stock and prices are disconnected.
The reduced form of the SREE is, thus, specified as:
where 
is a “threshold” price.
Assuming rational expectations on the part of agents, the stochastic
form of is
where 
is the disturbance
associated with deviations from the actual price due to demand and supply
shocks.
The probability
distribution of the random disturbance term () has an important implication for the formulation and estimation
of . The disturbance term could behave as i.i.d.,
time-dependent and/or periodic. Furthermore,
the opportunity cost of capital () may change over time as a result of structural policy
changes. Such distinctions give rise to several different possible forms for the
TAR.
The i.i.d
assumption of supply shocks implies that the threshold remains constant over
the entire sample period (Deaton and Laroque
1992).
The model [3] is specified as:
where 
are parameters to be
estimated from observed prices. As shown
by , there is no intercept term
due to the assumption that price is formed based on variable storage costs instead
of fixed storage costs. 
and the total per unit
storage cost is given by 
.
Under the assumption of periodic harvest correlation, the disturbances are i.i.d within a period but correlated across periods, i.e,
The major
implication of periodic disturbance assumption is that the threshold price,
is no longer constant, but instead varies across periods. In
particular, each period will have its own threshold price that speculators use for
decision making (Chambers and Bailey
1996).
The periodic threshold form of is thus,
Periods can be defined in different ways based on the objectives and the practical conditions of the area under investigation. In this paper we define two periods: a harvest period and a non-harvest period. These periods are group of months observed over several years. Each period is hypothesized to have its own threshold to account for seasonal effects of price correlation.
4.2. Regime Switching MLE
The TAR models are estimated using regime-switching maximum likelihood estimation. For T independent observations, the log-likelihood function is given by
Where 
denotes the standard
normal distribution. Equation is modified for regime
switching regression using probability function that identifies the probability
of a given observation falling in a particular regime,
. The probability function, 
, denoted as 
takes a value of 1 if 
and 0 otherwise for 
and the reverse for 
. Furthermore it has been normalized using the proportion of
observation within the regime (Hamilton 1994). The normalized log-likelihood function is
where 
is the proportion of
observations under regime , 
and 
. 
is the variance for each
regime;