The unit root test was used to confirm the GOLD, OIL and NTDUSD
variables were significant at first order differential, so variables
were stationary in the ADF unit root test.
Since the Johansen co-integration test was based on the autoregressive
model (VAR) and the optimal lag length needed to first be
determined to count the total number of integrated vectors. The
backward default was 12, the optimal lag was selected according to
the AIC and SBC criteria, and the results of the optimal lag were 1. The
Johansen co-integrating test was subsequently performed to confirm
the existence of co-integration among the variables.
From Table 3, GOLD had a right-skewed distribution, while
NTD_USD and OIL exhibited left-skewed distributions.
Table 4 shows that Trace Statistics and Max-Eigen Statistics
could not reject the null hypothesis r¼0 (None), rr1 (At most 1)
and rr2 (At most 2) at a 5% significance level, so the cointegrating
relationship did not exist among all the variables.