Rice Genomics and Genetics 2015, Vol.6, No.5, 1-10
3
discriminating test for spurious correlation:
conducting co-integration analysis between apparently
correlated I(1) series and finding co-integration
confirms the regression.
Several methods have been used to measure market
integration. Simple bivariate correlation coefficient,
also called the Law of One Price (LOP) has long been
the most popular model. The concept of constant
market price and arbitrage are consistent with LOP.
This method was strongly criticized most notably by
Harriss (1979) and Ravallion (1986). Ardeni (1989)
and Baffles (1991) found the LOP to be a short-run
phenomenon. Advancements in time-series econo-
metrics led to the development of models that address
some of the perceived weaknesses in the correlation
coefficient approach. In this respect, Ravallion (1986)
proposed a dynamic model of spatial price differen-
tials incorporating time lags.
In spite of this modification by Ravallion, one major
drawback still remained. Both the LOP and Ravallion
models according to Marks (2008) test whether price
changes in one market will be translated on a
one-for-one basis to the other market, either
instantaneously (LOP) or with lags as in the Ravallion
model. It should be understood however that prices in
different markets will only move on a one-for-one
basis if the inter-market price differential is equal to
transfer costs. Thus, price movements inside the
bandwith set by the transfer costs do no harm to the
hypothesis of market integration, whereas these
models will possibly reject this hypothesis.
Palakas and Harris-White (1993) and Alexander and
Wyeth (1994) extended Ravallion’s model using
Co-integration and Granger causality ordinary least
squares (OLS) technique. This allowed testing for
more general notions between markets and measuring
whether prices in two markets wander within a fixed
bandwith (Baulch, 1997). According to Marks (2008),
the limitation of these models however, is that all
models are in fact “static” with the interpretation that
markets are either integrated or not. This requires the
assumption of a constant market structure throughout
the period covered by the time-series data collected
for analysis. It implies that when observations for
different sub-periods are limited, then doing
integration analysis is not feasible.
Presently, the most common approach for testing for
market integration is the Johansen Co-integration
technique and Vector Error Correction model applied
among others by Rufino (2008), Mohammad and Wim
(2010) and Mafimisebi (2012). This paper used this
approach.
3. Methodology
3.1. Sources of Data Used in the Study
Secondary data were used for this study. The
secondary data were sourced from National Bureau of
Statistics (NBS). Data points were monthly retail
prices of imported rice in urban areas of the six
Southwest states of Nigeria which are Ekiti, Lagos,
Ogun, Ondo, Osun and Oyo States. The data covered
the period from January 2001 to December 2010,
giving a total of 120 data points per state.
3.2 Analytical Procedure
The data analytical techniques that were used in this
study comprised of descriptive statistics and
co-integration technique (Johansen co-integration test).
The descriptive statistics that were used included
frequency counts, means and co-efficient of variation
(CV). Augmented Dickey Fuller (ADF) test and Philip
Perron (PP) test were used for the stationarity test.
Johansen Co-integration model was conducted to test
for long run market integration between spatial
markets that were stationary of the same order.
3.2.1 Mean Spatial Prices and Variability Index
Average monthly growth rate of prices for the whole
period considered were computed as well as
coefficient of variation (CV).
3.2.2. Test for Order of Econometric Integration
(unit root test)
A stationary series is one with a mean value which
will not vary with the sampling period. In contrast, a
non-stationary series will exhibit a time varying mean
(Juselius 2006, Mafimisebi, 2012). Before examining
integration relationships between or among variables,
it is essential to test for unit root, and identify the
order of stationarity, denoted as I(0) or I(1). This is
necessary to avoid spurious and misleading regression
estimates (Adams, 1992).
The framework of ADF methods is based on an
analysis of the following model:
