The model of spatial integration predicts that, under competitive conditions, price differences between two regions in the same economic market for a homogeneous commodity will approximately equal the inter-regional transportation costs. Market integration thus involves a test of price efficiency by examining how food markets in different regions respond jointly to supply and demand forces. If price movements in different parts of the country tend to behave similarly, reflecting the cost of transferring the product between two regions, then markets are said to be integrated.
Several studies on market liberalization have tested for food market integration Early empirical studies of market integration used static price correlations to test for spatial market integration in agricultural markets. This involves the estimation of bivariate correlation or regression coefficients between the time series of spot prices for an identical good at different marketplaces. In these analyses, a statistically significant coefficient implies that the two markets are integrated. This kind of modeling of spatial market integration has been criticized for masking other effects like inflation and seasonality. Price correlation assumes instantaneous price adjustment and cannot capture the dynamic nature of a marketing system. It is possible that price correlation might suggest being the result of spurious market integration, like common trends, common seasonality, monopoly price-fixing, etc. (Harriss, 1979; Delgado, 1986; and Heytens, 1986). Price correlation tests may also overestimate a lack of market integration if a lag in market information produces a lag in the price response between markets (Barrett, 1996). Finally, price correlation tests only a pair of markets at a time and cannot be used to evaluate the marketing system as a whole (Delgado, 1986). In order to overcome the weaknesses of price correlation tests, various alternative methods have been developed.
Time-series methods have been introduced in the study of market integration to overcome the problems of common trends and nonstationarity of food prices inherent in bivariate price correlation models. Studies employing time series methods also formally modeled issues pertaining to short-run and long-run integration, seasonality, and the degree of market integration.
The Granger causality method employs an error correction mechanism to determine the extent to which current and past price changes in one market explain price changes in another. The error correction model (ECM) (Engle and Granger, 1987) holds that if the price of a local market and the price of the central market are co-integrated, then the error term from the co-integrating equation should be included, otherwise a first differencing regression between the two prices will be mis-specified and cannot be used to test for market integration (Palaskas and Harriss, 1993; Dercon, 1995). The advantage of the ECM is that not only can short-run and long-run information be conveyed between markets, but the relevant direction of the flow of price information can also be determined. Another advantage of the ECM is that it helps to alleviate the problems of auto-correlation and multicollinearity of most food price series (Baulch, 1997). The shortcoming of Granger cointegration analysis is that it does not allow for the investigation of all possible co-integrating vectors in a multivariate system.
Johansen (1988) developed a multivariate method of cointegration analysis which is a more recent development in this field. The method uses a maximum likelihood methodology to test the hypothesis of co-integrating relationships among several economic time series. The use of multivariate analysis is the most suitable approach to use if prices are endogenously determined, which is usually the case for food markets, and in which prices are simultaneously determined. The multivariate approach was used by Silvapulle and Jayasuriya (1994) in their study of Philippine rice markets to study the co-integration of markets; they found that rice markets were co-integrated. Chang and Griffith (1998) applied the multivariate approach to Australian monthly beef prices at the farm, wholesale, and retail levels and found all three prices to be co-integrated.
For the present study, there is some validity in the above-mentioned criticisms especially as far as non-stationary transfer costs are concerned. Nonetheless, time series analysis can provide useful insights into the issue of market integration if an appropriate testing framework is employed and the results are interpreted correctly. Co-integration tests and ECMs provide an analytical tool that can focus beyond the case of market integration in testing notions such as completeness, speed, and asymmetry of the relationship between prices.
In analyzing spatial integration, data on daily prices or average weekly prices are preferable. Like other developing countries, in Bangladesh, daily prices are available for only a few central markets and only for a short period of time. For the purpose of this study, data pertaining to weekly wholesale rice prices were collected from different marketing intelligence centers (MICs) as assigned by the Department of Agricultural Marketing, Government of the People’s Republic of Bangladesh, for the period January 2004 to November 2006. Data on prices pertain to Friday of each week for twelve months. The prices were reported in Tk/quintal. The selected six MICs for this study were Dhaka, Chittagong, Rajshahi, Khulna, Barisal, and Sylhet.
The individual price series are tested for the order of integration to determine whether or not they are stationary. A number of tests for stationarity are available in the literature; these include the Dickey-Fuller (DF) test (Dickey and Fuller, 1979), the Augmented Dickey-Fuller (ADF) test (Dickey and Fuller, 1981), and the Philips-Perron (PP) test (Perron, 1988). Having established that the variables are nonstationary a maximum likelihood approach based on a finite vector autoregression (VAR) model as developed by Johansen (1991) can be specified to determine whether the system of equations is co-integrated.