Prior to testing for the presence of cointegration, iShare price series data are checked for
non-stationarity using conventional unit root tests, namely Augmented Dickey-Fuller (ADF; Said
and Dickey (1984)) and Phillips-Perron (PP; Phillips and Perron (1988)). All series were found to
be non-stationary in levels and stationary in differences.3
5. Methodology
5.1 Gregory-Hansen Residual Based Cointegration Test
Engle and Granger (1987) suggested that two non-stationary variables might converge to a
common equilibrium in the long run. Then a stationary combination of the two non-stationary
variables should exist. Such variables are then called cointegrated and the vector that transforms
the two non-stationary variables into a stationary one is called cointegration vector. Test for
cointegration, suggested by Engle and Granger (1987) was extended by Johansen (1988) to a
multivariate case. Both tests rely on the assumption that stability of cointegration vector is stable
over time. However, it is highly likely that during longer periods a fundamental or non-
fundamental shock may disrupt the equilibrium, which would result in a change in the parameters
of cointegration vector.
Results of Monte Carlo experiments (Campos, Ericcson, and Hendry, 1996; Gregory and
Hansen, 1996) demonstrate that standard tests for cointegration may lose power when a shift in
parameters takes place and in fact, falsely signal the absence of equilibrium in the system.
Gregory and Hansen (1996) test assumes the null hypothesis of no cointegration against the
alternative hypothesis of cointegration with a single structural break of unknown timing. The
3 The critical values are from MacKinnon (1991). For the purposes of brevity tables are not reported here, but
they are available upon request.
13