Modeling and Forecasting Primary Commodity Prices

Walter C. Labys

• Publisher: Routledge
• Year: 2017
• eBook ISBN: 9781351917087
• Topic: Business
• Subtopic: Business General
• Index: Business

Recent economic growth in China and other Asian countries has led to increased commodity demand which has caused price rises and accompanying price fluctuations not only for crude oil but also for the many other raw materials. Such trends mean that world commodity markets are once again under intense scrutiny. This book provides new insights into the modeling and forecasting of primary commodity prices by featuring comprehensive applications of the most recent methods of statistical time series analysis. The latter utilize econometric methods concerned with structural breaks, unobserved components, chaotic discovery, long memory, heteroskedasticity, wavelet estimation and fractional integration. Relevant tests employed include neural networks, correlation dimensions, Lyapunov exponents, fractional integration and rescaled range. The price forecasting involves structural time series trend plus cycle and cyclical trend models. Practical applications focus on the price behaviour of more than twenty international commodity markets.

Chapter 1
History of Commodity Price Analysis

Prices are ever changing. They change from hour to hour, from day to day, from season to season, and year to year. Every change affects the relationships of individuals, of groups of people, and of nations … Many factors combine to make prices what they are … Prices are both a cause and an effect. The causes may be analyzed just as substance may be analyzed chemically, and the proportion due each cause ultimately may be determined. The science of price analysis is still new but has progressed far enough to be of help.

Warren and Pearson, Prices (1933, p. 2)

Commodity prices again! The twentieth century has only been the latest spectator to the impacts and importance of commodity price fluctuations. According to Fischer (1996), commodity price records have come down to us from the ancient civilizations of India, Mesopotamia, Egypt, Greece and Rome, some from as early as circa 1800 B.C. Beginning in the 12^th century, price series of high quality can be found. Recall the Granger and Elliott (1967) time series study of 18^th century grain prices. Formal research on the relationships between agricultural demand, supply and prices in a market context began earlier in the 20^th century. This research not only evolved in sophistication but extended to mineral and energy commodities. In substance it analyzed market history, explained commodity prices, evaluated commodity policies and forecast commodity prices. Also at the beginning of that century, some of the earliest work took place on applying statistical methods to price series. In fact the study of prices has been one of the few areas in economics in which we have allowed the data to help us to formulate theory, as has been strongly advised (Granger, 1992; Hendry, 1995; and Roehner, 1997). This work recently reached a peak in the awarding of the Nobel Prize in Economics to Dr. Clive Granger in 2003 (shared) for his advances in time series econometrics, including applications to commodity prices.

The purpose of this chapter is to review how this progress has contributed to analyzing commodity markets and prices and to solving price forecasting problems, concentrating on more recent advances in quantitative modeling and time series analysis. It consists of the following sections: the Challenge, Structural Approach, Nonstructural Approach, Some Remaining Problems, and Conclusions.

The Challenge

The possibility that commodity market quantities and prices are often random introduces a large amount of risk and uncertainty into the process of market analysis and forecasting. Of course, randomness is implied generally and the nature of the price fluctuations varies as we observe them and their likely causes in the long, medium or short term. The economic analysis of long term price movements has had an extensive and interesting history. Some studies of mention include Abel (1935), Barnett and Morse (1963), Dick (1998), Drame et al. (1991), Fischer (1996), Froot (1995), Granger and Hughes (1971), Kondratief (1935), Labys (1993), Lewis (1949), Mills (1927, 1936, 1940, 1946), Persson (1994), Terraza (1981), Usher (1930, 1931) and Warren and Pearson (1933). These prices also relate to the long run inflations that have occurred in different nations and in different periods (Brown, 1985, 1988). More recently long term price trends have been studied concerning their role in declining commodity terms of trade in developing countries (Cuddington, 1992). From a practical viewpoint, the prediction of long run price trends has been important for evaluating investments in commodity industries, particularly in mineral and energy projects in developing countries (Duncan, 1984; Torries, 1998; World Bank, 1994).

In the long term, commodity markets are subject to shocks or changes in trend, which range from natural catastrophes and political/military interventions to structural market changes. The econometric methods of interest have been those dealing with structural breaks, booms and slumps, and secular movements. For example, see Cashin and McDermott (2002), Perron (1989), and Cuddington and Urzua (1989). Commodity shocks tend to be irregular in nature and cause abrupt shifts in prices usually to higher, but, sometimes, to lower levels. Examples include the impacts of the Korean war and Vietnam war, the petroleum price increases of 1973-1978, the Gulf war, and the Iraq invasion. Sometimes the return of a market to normality is quick; at times the shocks persist; and, at other times price changes reoccur, resulting in a series of consecutive turning points. Methods recently developed that help to analyze such trend changes appear in Andrews (1993), Badillo et al. (1999), and Perron (1989).

In the medium term, factors that shock commodity markets can also be of a political or cataclysmic nature, but they tend to be more related to national economic conditions or to market forces themselves. Such market forces tend to be observed in demand and supply conditions and in underlying market equilibrium. The analysis of medium term price movements or price cycles also has had a long history. Kondratief (1935) proposed that commodity and consumer prices reveal cycles that reoccur every 50-60 years. Much greater importance has been placed on the role of commodity price cycles in motivating the great depression (Lewis, 1949). Fluctuations in national economic conditions, commonly observed in the form of business cycles, can cause changes in industrial production and consequently in mineral demands or in interest rates and ultimately in commodity prices. This led the National Bureau of Economic Research in the United States to spawn a series of business cycle studies that dealt with agricultural prices as well as with minerals and raw materials commodity prices, e.g. Mills (1927, 1936). More recent studies have examined the interrelations between commodity prices and business cycles (Bosworth and Lawrence, 1982; Ding, 1998; Kaldor, 1987). Related econometric methods have not only used spectral analysis but also have involved structural time series models, which emphasize cyclical components (Labys and Granger, 1970; Harvey, 1985).

Variations in weather conditions induce changes in agricultural supply and hence in product prices. The formal analyses of the impact of these kinds of shocks has appeared in modeling studies, such as Adams and Behrman (1978), Ghosh et al. (1987), Labys (1973, 1999), Marquez (1984), and Rausser and Hochman (1979).

In the short term, market shocks come primarily from financial factors, particularly those related to speculation and hedging on commodity futures, options and other derivative markets. The resulting price behavior has been termed random, because it reflects the flow of randomly appearing information. In fact the price behavior can be identified more specifically as being stochastic or following a nonlinear dynamic or other form of stochastic process (e.g., autoregressive conditional heteroscedastic), or even a chaotic process. It can also be related more specifically to financial shocks such as in interest rates or exchange rates. A substantial literature exists attempting to explain this short-term behavior. Examples include Adams and Vial (1988), Barkoulas et al. (1997, 1999), Decoster et al. (1992), Holt and Aradhgula (1990), Hudson et al. (1987), Teysseire et al. (1997), Yang and Brorsen (1992), and the NC134 Conferences on Applied Commodity Price Analyses, Forecasting and Risk Management (1994-99).

Short term price analysis has recently experienced the most interest, particularly in relation to the study of futures markets and the discovery of chaos and nonlinear dependence. Much of this work has been related to tests of the efficient market hypothesis in futures prices. Short term commodity price movements had early been discovered to possess random walk behavior or a variant known as a martingale (Working, 1958; Samuelson, 1965). While this behavior implies an independence of price changes, other work has confirmed deviations from random walk in the form of occasional autocorrelations or linear dependence, e.g. Houthakker (1961), Labys and Granger (1970). Hints of prices possessing nonlinear rather than linear dependence have been revealed in studies testing for fractal or chaotic structure, e.g. Mandelbrot (1963), Frank and Stengos (1989). More recently, nonlinear dependence has been confirmed in some price series using fractional integration methods and a shift to examining volatility as well as means (Cheung and Lai, 1993). Such discoveries also have led to improved possibilities for forecasting commodity price movements.

Structural Approach

The most comprehensive commodity market analytical methods stem from structural models which are soundly based in microeconomic and econometric theory, but also include other modeling theories, e.g. optimization, programming, input-output, computable general equilibrium. Because these structural models trace the interaction between endogenous market variables such as prices and demand and exogenous variables such as industrial production, they can explain market behavior and performance. The scientific process involved usually requires model specification, estimation, and simulation. Model simulation can trace the historical behavior of price and quantity variables over time and/or space; it can provide conditional estimates of various commodity policy impacts; or it can forecast the variables into the future. In the case of conditional forecasts, one can predict endogenous variables conditional upon forecasts of macroeconomic variables or upon maintained assumptions concerning the behavior of policy makers. These models come from a distinguished background of theoretic developments in agricultural, mineral and energy economics. A previously prepared bibliography (Labys, 1987) lists a variety of such models. Apart from the Labys (1999) update on mineral and energy models, no recent appraisal exists for agricultural commodity models.

Standard Commodity Model

The beginnings of structural commodity modeling stem from the earliest work on econometric models. Some landmark works include those of the Cowles Commission (Hood and Koopmans, 1953; Christ, 1994 (re-edition); Tinbergen, 1939; and Klein and Goldberger 1955). Some of the first work on supply-demand models concerned agricultural commodities. Examples include the research of Fox (1958), Hildreth and Jarrett (1955), Meinken (1955a, b), Moore (1914), Wallace and Judge (1959), and Waugh (1964). This work expanded to include other commodities such as Adams and Griffin (1972) on petroleum, Behrman (1971) on rubber, Burrows (1971) on cobalt, Desai (1966) on tin, Houck et al. (1972) on soybeans, Labys (1973) on lauric oils, Labys et al. (1979) on coal, Rausser (1971) on oranges, Weymar (1968) on cocoa, Witherell (1968) on wool, and Zusman (1962) on potatoes. Commodity modeling became formalized in the works of Labys (1973, 1975) and Adams and Behrman (1978). Other econometric modeling guides include Hallam (1990), Ghosh et al. (1987) Guvenen et al. (1991), Labys (1999) and Lord (1991).

The most basic type of commodity model from which econometric and modeling methodologies have developed is the competitive market model. Such a model initially neglects market imperfections and assumes that commodity demand and supply interact to produce an equilibrium price reflecting competitive market conditions. Such a model may consist of a number of combined regression equations, each explaining separately a single market or sector variable. Market models or the equivalent industry models are applicable to all agricultural, mineral or energy production and use categories. Their greatest utility is in providing a consistent framework for planning industrial expansion, forecasting market price movements, and studying the effects of regulatory policies. The basic structure of such a model typically explains market equilibrium as an adjustment process between demand, supply, inventory and price variables; typical specifications appear in Labys (1973, 1975, 1999) or Lord (1991).

Most simply it would consist of the following equations:

1. D_t = d (D_t-1, P_t, PC_t, A_t, T_t)
2. Q_t = q(Q_t-1, P_t-Θ, N_t, Z_t)
3. P_t = p(P_t-1, dI_t)
4. I_t = I_t-1, + Q_t – D_t

where:

1. D = Demand
2. Q = Supply
3. P = Prices
4. PC = Prices of substitutes
5. P t-Θ = Prices with lag distribution
6. I = Inventories
7. A = Income or activity level
8. T = Technological factors
9. N = Resource characteristics
10. Z = Policy variables influencing supply

Commodity demand is explained as being dependent on prices, economic activity, prices of one or more substitutes and possible technological influences. Other possible influencing factors and the customary disturbance term are omitted here and below to simplify presentation. Accordingly supply would depend on prices as well as the underlying production factors, such as geology or resource exhaustibility, and a possible policy variable. A lagged price variable is included since the supply process is normally described using some form of the general class of distributed lag functions. Commodity prices can be explained by changes in inventories, although this equation is sometimes inverted to explain inventory demand. The model is closed using an identity that equates inventories with lagged inventories plus supply minus demand. The utilization of this model involves further specification, estimation and simulation (Labys, 1973, 1978, 1999). Some experimentation has also taken place using system dynamics (Meadows, 1970; Ruth and Hannon, 1997).

There is no doubt that commodity modeling evolved from agricultural economics. In the beginning of the twentieth century, Lehfeld (1914) and Moore (1914, 1917) employed regression methods to analyze relationships between agricultural commodity demand and prices. Moore's writings continued to influence econometric research not only to measure true demand and supply curves but also to forecast expected prices (Stigler, 1962). Emphasis also was placed on analyzing price behavior itself, as witnessed in studies by Haas and Ezekiel (1926) on hogs, Killough (1925) on oats, and Working (1922) on potatoes.

This approach spread very rapidly. Warren and Pearson (1928) concentrated on the supply-side of modeling. Works by European economists included Hanau (1928) on hogs, Leontief (1929) on price indexes and Roy (1935) on price methods. Several modeling books began to emerge. Worthy of interest are Ezekiel (1959) on market correlation and Koopmans (1950) on econometric models. Demand analysis included works by Schultz (1938), Stone (1958), Brandow (1961), Working (1927), Fox (1953), Waugh (1964), Wold and Jureen (1953), Meinken (1955a, b) and Nerlove (1958). Price analysis also expanded with studies by Thomsen (1936), Shepherd (1941), Waite and Trelogan (1948), Thomsen and Foote (1952), Foote (1958), Harlow (1962), Labys and Granger (1970) and Labys (1973). Discussion of the structural modeling of agricultural markets appears in Chen (1994), Brookfield (1991), Ferris (1998), Labys (1973, 1975), Heady and Kaldor (1954), Hinson (1991) and Hallam (1990).

Another direction of this work has been to construct models more of a reduced-form nature to explain commodity price behavior. Some studies such as Chu and Morrison (1984) have been longer-term and more macroeconomic in their approach. Studies which concentrate more directly on the use of reduced form models are numerous. More recently Labys and Kouassi (2004) have applied the structural time series approach to discover stochastic cyclical behavior in commodity prices. Special interest in electricity pricing can be found in works such as Munasingha and Meier (1993), Schweppe et al. (1988) and Turvey (1968). Other studies such as those presented in Winters and Sapsford (1990) employ models which are larger scale in nature and thus closer to the above econometric models.

More recently models of this type have been placed in a general equilibrium framework and and various hybrids also developed to assess the impact of the Common Agricultural Policy in the European Union. A summary of these modeling efforts appears in Conforti (2001). Cited as different modeling approaches in this document are the World Agricultural Simulation Model (WATSIM), Static World Policy Simulation Model (SWOPSIM), Food and Agricultural Research Policy Model (FAPRI), Modele International Simplifie de Simulation (MISS), the FAO World Food Model (FAOWFM), and the European Simulation Model (ESIM).

The development of econometric models suitable for analyzing mineral markets possessing competitive behavior has evolved more recently. One of the first was the Desai (1966) tin model that explained tin price fluctuations on a world basis. The copper market has also been subject to several modeling efforts. Most notable, Fisher et al. (1972) built a world copper model that was recognized as one of the first major econometric mineral modeling efforts. Charles River Associates (CRA, 1978) later extended the supply sector of the Fisher copper model to include long run adjustments in exploration and discovery as well as subsequent mining capacity formation. This long run adjustment process was combined with a short-run inventory adjustment process in a distinctly disequilibrium form of copper model by Labys (1980a). Such an approach to modeling the copper market was suggested by Richard (1977) with his continuous time, differential equation approach. More recent developments on the structural modeling and forecasting of mineral markets appear in Labys (1999).

Applications to energy markets have not been as extensive because of the difficulties in dealing with regulatory and non-competitive influences on market behavior. Nonetheless, a variety of such models can be found in Labys (1999) and in Lesourd et al. (1996). Verleger (1982, 1993) has shown how such models can be applied to explain disruptive shortages. His model links econometric equations for oil spot prices, consumer demand, and supply shortage conditions. MacAvoy and Pindyck (1975) have built an econometric model of the natural gas industry which has been used extensively to analyze the effect on the industry of federal regulation of the wellhead price of gas and of permissible rates of return for the pipeline industry. Labys et al. (1979) have modeled the U.S. coal market using this approach to forecast future levels of coal demand, supply, prices and inventories. Most recently, Trieu (1994) et al. have reported their modeling of the world uranium market.

The several econometric approaches taken to model noncompetitive market configurations are essentially similar. For example, the monopoly case involves one dominant (monopolist) producer and many (perfectly competitive) consumers. The single producer thus maximizes his own profits given the aggregate demand function for the commodity of interest and the supply response of the other firms in the industry. Examples of applications to OPEC and the crude oil market include Blitzer et al. (1975), Cremer and Salehi-Isfahani (1991), and Pindyck (1978b). Regarding intermediate mineral market configuration, Pindyck (1978a) developed a model to determine optimal price and quantity paths that would result from cartel behavior on the part of producers’ organizations in the copper and bauxite markets. To this day, OPEC would have been richer had they followed Pindyck's advice.

Finally it should be recognized that the specification of the above model is given in its structural form. This formulation can be conveniently converted to a reduced form in which the endogenous or dependent variable appears only on the left-hand side of an equation and the ex...

Table of contents

• Cover Page
• Title Page
• Copyright Page
• Contents
• List of Figures
• List of Tables
• List of Contributors
• Foreword
• Preface
• Acknowledgments
• List of Abbreviations
• Introduction
• Chapter 1 History of Commodity Price Analysis
• PART 1 Long Term Price Movements
• PART 2 Medium Term Price Movements
• PART 3 Short Term Price Movements
• PART 4 Price Forecasting
• Bibliography
• Index