on the model with a drift only. The unit root null hypothesis is rejected at conventional levels
for all series in first differences.
Table 2. DF-GLS unit root tests on the first differences of series
|
Variables |
Specification |
Lags |
Test statistic |
|
∆lnIPI |
constant |
4 |
- 1.986 |
|
∆lnPPI |
constant |
2 |
- 3.680 |
|
∆lnM1A |
constant |
1 |
- 6.633 |
|
∆lnXREURO |
constant |
1 |
- 6.753 |
The critical values for 0.95 (0.99) confidence levels with constant are -1.944 (-2.592), with constant and trend
are -3.074 (-3.633). The Akaike Information Criteria was used to determine the lag length.
Because the graphical analysis suggests the possibility of structural breaks, that are not
accounted for by the DF-GLS tests, the Zivot-Andrews unit root test in the presence of
structural breaks is used to double-check the series (table 3).
Table 3. Andrews - Zivot unit root tests
|
Variable |
Structural break |
Lags |
Possible break |
Test statistic |
|
lnPPI |
intercept only |
0 |
2001:06 |
- 3.783 |
|
trend only |
0 |
2000:09 |
- 2.506 | |
|
both |
0 |
2001:06 |
- 3.825 | |
|
lnIPI |
intercept only |
1 |
2003:06 |
- 2.557 |
|
trend only |
1 |
1998:03 |
- 2.413 | |
|
both |
1 |
2001:05 |
- 2.865 | |
|
lnXREUR O |
intercept only |
2 |
2001:05 |
- 3.717 |
|
trend only |
2 |
1998:09 |
- 3.278 | |
|
both |
2 |
1998:08 |
- 3.263 | |
|
lnM1A |
intercept only |
0 |
2003:07 |
- 2.208 |
|
trend only |
0 |
1998:03 |
- 3.655 | |
|
both |
0 |
1998:08 |
- 2.876 |
The critical values for 0.95 (0.99) confidence levels in case of a structural break in the intercept only, are -4.80
(-5.34), trend only -4.42 (-4.93), and both -5.08 (-5.57). The Schwarz Bayesian criterion was used to determine
the lag length.
10