WP5 I q 4- POLICY RESEARCH WORKING PAPER 1944 Detecting Price Links in the Links between international cotton prices have improved World Cotton Market in the past decade - in the short run largely because prices are now transmitted MohBaffes L Awadmore quickly. To a lesser Moha7ned I. Ajwad extent, and for different reasons, prices should also converge somewhat more in the long run. The World Bank Development Research Group July 1998 POLICY RESEARCH WORKING PAPER 1944 Summary findings Baffes and Ajwad examine the degree to which Central Asia to other parts of the FSU were considered international cotton prices are linked and test whether part of domestic trade. Now cotton exports from such links have improved over the past decade. Uzbekistan are the most important component of that They conclude that the degree of linkage has improved country's foreign trade. over the past decade, in the short run largely as the result With such changes, one should expect cotton prices to of short-run price transmission - and to a lesser extent converge somewhat more in the long run. because of long-run comovement. Price links between West Africa and Central Asia are Improvements in information technology have made it much greater than between the United States and other much easier for information about demand to be markets - in part because most West African and disseminated across markets, so changes in cotton prices Central Asian cotton is exported, compared with only 40 attributable to a price shock in one place are soon percent of U.S. cotton (and 60 percent of Greek cotton). transmitted to prices in other places. Prices in countries that export most of their cotton are Moreover, many countries have liberalized their cotton more likely to converge than prices in countries where subsectors, and in some countries the government's role prices are subject to both domestic and international has changed substantially. demand conditions. In East Africa, for example, cotton marketing and To improve price risk management, there should be trade was handled entirely by government parastatals. futures contracts other than those traded on the New Now Tanzania, Uganda, and Zimbabwe have liberalized York Cotton Exchange, which mostly serves domestic their marketing and trade regimes, to varying degrees. U.S. needs and is not used extensively by non-U.S. In the former Soviet Union (FSU) cotton shipped from hedgers and speculators. This paper - a product of the Development Research Group - is part of a larger effort in the group to investigate the behavior of world prices. Copies of the paper are available free from the World Bank, 1818 H Street NW, Washington, DC 20433. Please contact John Baffes, room MC3-545, telephone 202-458-1880, fax 202-522-1151, Internet address jbaffesCy.worldbank.org. July 1998. (36 pages) The Policy Research Working Paper Sedbes disseminates the findigs of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, int'erpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the viewv of the World Bank, its Executive Directors, or the countries they represent. Produced by the Policy Research Dissemination Center DETECTING PRICE LINKS IN THE WORLD COTTON MARKET JOHN BAEFES W7orld Bank MOHAM:ED I. AJWAD University of Illinois CORRESPONDENCE: John Baffes Development Research Group World Bank 1818 H Street, NW, room MC3-545 Washington, DC 20433 tel: (202) 458-1880 fax: (202) 522-1151 em: jbaffes@worldbank.org The authors would like to thank Alain d'E[oore and Uma Lele for valuable comments and suggestions on an earlier draft. I. INTRODUCTION In the absence of impediments, the comparative advantage argument of trade theory dictates that resources will be allocated in an efficient manner, in turn implying that factor and product prices in different locations will be equalized - subject to transfer costs. Under certain conditions, the existence of strong price linkages, therefore, may be viewed as a necessary requirement for efficient allocation of resources and hence maximum welfare [Samuelson (1952); Takayama and Judge (1964)]. This paper focuses on the degree of price linkages of the world market of cotton. The issue of price linkages in product markets both at national and international levels has been studied in the literature rather extensively either under the notion of the law of one price [e.g. Protopapadakis ancd Stoll (1983, 1986), Ardeni (1989), Baffes (1991)] or under the notion of market integration [e.g. Ravallion (1986), Sexton, Kling, and Carman (1991), Gardner and Brooks (1994), Fafchamps and Gavian (1996), Baulch (1997a)]. Moreover, reflecting on the market liberalization and structural adjustment efforts undertaken by a number of developing countries in recent years, the degree to which markets are integrated has been used quite often as a yardstick in assessing the success of policy reforms [e.g. Goletti and Babu (1994), Alexander and Wyeth (1994), Gordon (1994), Dercon (1995)]. As many authors have cautioned, however, price convergence does not necessarily imply efficient allocation of resources unless the setting in which trade takes place is competitive [e.g. Faminow and BFenson (1990), Baulch (1997b)]. For example, consider the extreme case of two duopolists who agree to charge the same price in two segmented markets. While convergence in prices (whenever price changes occur) would take place instantaneously, the oligopolistic setting of the market may not necessarily allocate resources in the most efficient manner. The same argument may be advanced for a number of developing countries where parastatals assign panterritoriat and panseasonal prices on certain commodities. In such cases the law of one price holds by definition without necessarily implying that resources are allocated efficiently. The present paper examines the strength of price linkages in the world market of cotton. In pursuing this objective, the paper contributes to the literature of price 1 linkages in two respects. On the theoretical side, it introduces a measure of price linkage and also identifies its source (i.e. short-run price transmission versus long-run comovement.) On the empirical side, it applies this measure to the world market of cotton for two different time periods, thereby examining whether improvements in price linkages have taken place over the last decade. There are at least two reasons as to why one would expect that price linkages may have improved over the last decade. First, improvements in information technology have made it much easier for information on demand conditions to be disseminated across markets; therefore one would expect that cotton price changes from one origin due to a demand shock would be transmitted immediately to the price of the other origins. Second, many countries have undertaken steps to liberalize their cotton subsectors while in other countries the role of the government has been substantially altered. For example, under the Former Soviet Union (FSU) structure, cotton from Central Asia shipped to other parts of the FSU was domestic trade. Currently, cotton exports from Uzbekistan constitute the single most important component of its foreign trade. Changes have also taken place in Africa. For example, until the early 1990s, cotton marketing and trade in East African countries was handled in its entirety by government parastatals. Now, Uganda, Zimbabwe, and Tanzania, operate - to different degrees - within liberalized marketing and trade regimes. One, therefore, would expect faster long-run convergence of cotton prices (if convergence existed) or at least some convergence (if convergence did not exist in the first place). The remainder of the paper proceeds as follows. In the next section the model along with the explicit measure of the degree of market linkage is outlined. In discussing the model, we undertake a rather extensive literature review on the subject of price linkages. The review indicates that a number of commonly used models are, in fact, restricted versions of the same dynamic specification. The penultimate section discusses the data and presents the empirical findings. The data cover the periods August 1985 through December 1987 and August 1995 through January 1997 and refer to CIF prices in North European ports for US, Greek, West African, and Central Asian types of cotton. The findings indicate that price linkages between W. Africa and C. Asia have been much higher than between the US and the other markets. In fact, in the 2 first period, no comovement between the US and the other markets was detected. The last section concludes by addressing sorne policy implications and subjects for further research. II. DETECTING PRICE LINKAGES Earlier studies examining the relationship between prices either have looked at correlation coefficients [e.g., Lele (1967); Southworth, Jones, and Pearson (1979); Timmer, Falcon, and Pearson (1983); Stigler and Sherwin (1985)] or have used the following type of regression [e.g. Isard (1977), Mundlak and Larson (1992), Gardner and Brooks (1994)]:' (= S + p2 + El, where p7' and p12 denote prices from two origins of the commodity under consideration, g and BI are parameters to be estimated while Łt denotes an lD(O, c02) term. The hypothesis that the slope coefficient equals unity and (possibly) the intercept term equals zero can be tested; formally, Ho: LL + 1 = = 1. Under H,, the deterministic part of (1) becomes p,i = p,2, in turn implying that the price differential, p,' - p7, is an IID(O, C2) term. Estimating (1) and testing Ho, while intuitively appealing and computationally implementable, presents two fundamental shortcomings. First, some statistical properties of the series involved in (1), namely nonstationarity, may invalidate standard econometric tests and thus give misleading results regarding the degree to which price signals are being transmitted from one market to another. Second, in primary commodity markets with characteristics such as (small or even perceived) differences in quality, high transfer cosl:s relative to the price, etc., it is rather unlikely that the two prices will only differ by an IID(O, 62) term as H,1 of (1) dictates. Therefore, H, is expected to be rejected without necessarily ruling out a relatively high degree of price linkage. Consequently, it is deemed necessary to employ a general enough model that imposes no a priori requirements on the stationarity properties of the series in question and at the same time allows for some degree of flexibility. 3 With respect to the nonstationari,ty problem one can examine the order of integration of the error term in (1) and make inferences regarding the validity of the model (Ardeni, 1989). If prices are indeed nonstationary, the existence of a stationary error term implies comovement between the two prices. However, if the slope coefficient is different from unity, the uniqueness of the cointegration parameter in the bivariate case implies that the corresponding price differential would be growing and such growth would not be accounted for, although prices may move in a seemingly synchronous manner. Hence, stationarity of the error term of (1) (given non-stationary prices) while establishing proportional price movement, should not be considered as a testable form equivalent to that of the Ho of (1). Note that a number of authors have warned against interpreting non-unity slope coefficient as a sign of market integration (e.g. Barrett (1996)). To account for the non-unity slope coefficient one can restrict the parameters of (1) according to Ho, in which case the problem is equivalent to testing for a unit root in the following univariate process (Engle and Yoo, 1987): (2) (pl p2) I(0). If the price differential as defined in (2) is stationary, then one can conclude that price signals are transmitted from one market to another, in the long run. The assumption (or finding) that the cointegration parameter is unity is very crucial, as it ensures that there is no other nonstationary component entering the system. As Meese (1986) and West (1987) observe, the absence of cointegration (with unity slope coefficient in the present setting) can be attributed to omitted nonstationary variables, in turn implying that an additional component would have to be included in (2) in order to fully account for the variability of the price differential. As a sidelight, it should be emphasized that if the cointegration parameter is unity, it is immaterial for all relevant aspects of the analysis whether (1) or (2) is employed. This is the case because as the sample size increases, regression (1) should yield ,X equal to unity. However, in finite samples this may not be necessarily the case. For example, Ardeni (1989), using (1) in logarithms for a number of internationally 4 traded primary commodities, found that the corresponding error term was not stationary, thus rejecting the law of one price. Baffes (1991), on the other hand, by using the same data set found that in the majority of cases the price differential was stationary, hence providing supportive evidence for the law of one price as a long run relationship.' From the preceding discussion, it is rather evident that cointegration tests are not very powerful as they only make inferences about the existence of the moments of the distribution of (Pt7 _ p72) and not about certain restrictions that may be required by economic theory [e.g. Ho of (1)]. Therefore, (2) cannot serve as a substitute for the H, of (1); it can only serve as an intermediate step in establishing its validity. With respect to the restrictive nature of (1), one can circumvent it by introducing a more general autoregressive structure. Appending one lag to (1), gives:' Pt = +t + + I2Pt-1 + I3Pt-1 + Ut, where u1 is IID(O, a2) and | 3 I <1. Despite its simplicity, (3) encompasses a wide variety of commonly used dynamic models with different economic interpretations. For example, Hendry, Pagan, and Sargan (1983) discuss a number of testable hypothesis, results of corresponding restrictions on the parameter space of (3). The most important one is the long-run proportionality or homogeneity hypothesis, the validity of which ensures that price movements in one market (p72) will eventually be transmitted to the prices of the other market (p7). Such a hypothesis can be tested by restricting all slope parameters of (3) to sum to unity (i.e., :3,B = 1). Under long-run proportionality, (3) can be re-parameterized as follows: (4) 0- P) = . + (1-I3)(Pt' - P' ,) + f31(Pt - ) + Ut. Relationship (4) belongs to the family of error-correction models (ECM). Because of the equivalence of the existence of cointegration and ECM, stationarity of the price differential (2) implies the existence of (4) (in the sense that (1 - ,3,) is significantly different from zero) and vice-versa [see Appendix A for a formal proof of this argument as well as the steps involved in going from (3) to (4)]. Note that the restriction | |