24
α 0 = 0;
α1 = α 2 = 1
However, correlation tests of price levels face the problem of common economy-wide phenomena such as general inflation and agricultural seasonality which affect all prices and therefore raise correlation coefficients between prices. Correlation between price changes is one means of addressing this problem. However, in addition to the problem of spurious correlation, correlation tests fail to address the problem of heteroskedasticity common in high frequency price data. In addition, correlation tests may overestimate market segmentation if lags in information, delivery or contract expirations result in natural lags in the price response between markets (Barrett, 1996, p. 826). In view of the statistical problems in the use of correlation coefficients, later analyses of market integration have shifted to cointegration analysis and more sophisticated approaches.
2.
Ravallion’s model and cointegration analysis. A major methodological innovation came from Ravallion (1986) whose model avoids the inferential dangers of bivariate correlation or regression coefficients. The error correction form of the Ravallion model allows for autocorrelation, distinct short run and long run dynamics, and common inflationary and seasonal components; the model has become the standard for market integration testing (Barrett, 1996, p. 826).
The basic Ravallion model is as follows; for N regions: Pit = ∑ j =1 aij P1t − j + ∑ j =0 bij P1t − j + X it ci + eit N
N
(i = 2,...N )
P1t = ∑ j =1 aij P1t − j + ∑k = 2 ∑ j =0 b Pkt − j + X 1t ct + e1t N
N
N
k
1