Page 427 - Contributed Paper Session (CPS) - Volume 4

CPS2509 D.L Sepato et al.
Stage II: Remove outliers
If any of the outliers turn to be non-significant then they are removed from
the set of potential outliers. An outlier free data was used for further analysis.
Thus the new ARMA (p, q)-GARCH (p, q) type models were modelled with 1312
observations after removal of 18 significant outliers from a set of data.

Stage III: iterate stages I and II for the adjusted series (model fit)
After the outliers were removed the ACF and PACF were plotted. According
to the results, AR (2), MA (2) are seen to be significant in this section. The,
parameters for ARMA (0, 2) were estimated. ARMA (0, 2)-GARCH (1, 1) and
ARMA (0, 2)-EGARCH (1, 1) and ARMA (0, 2)-GJR-GARCH (1,1) are fitted.

Table 7 shows the results that suggest ARMA (0, 2)-EGARCH (1, 1) as an
adequate model based on the information criterion and the model
performance evaluations. Therefore, this model was used to make forecasts
for outlier free series. The returns of ARMA (0, 2)-EGARCH (1, 1) are formulated
as:
= 0.0126 +
2 2 2
= −0.005998 + 0.0942−1 + 0.000901 −2 + 0.0541ℎ−1 + 0.9811ℎ −1
− 2.6465
= 1.0352

416 | I S I W S C 2 0 1 9