Page 19 - Contributed Paper Session (CPS) - Volume 7

CPS2014 Ma. S.B.P. et al.
in GVA growth rate at − 1; a 35.04 % increase in GVA growth rate at results
from a proportionate increase in GDP growth rate at − 1; and a 35.14 %
increase in GVA growth rate at results from a proportionate increase in GVA
growth rate at − 1. (1) produced negative estimates that contradict the
expected relationship between the output series.
The proposed estimation procedure’s predictive performance is also
superior than (1 ) as shown in Table 3. The proposed procedure has the
advantage of including the contemporaneous effects of the input series.

Table 1. The proportion of variance explained by the first K sparse principal components (SPCs)
of the reduced covariate matrix with =10 and =4.

10 83.32 %, 87.20 %, 89.52 %, 90.34 %, 91.64 %, 93.64 %, 94.72 %, 95.74 %, 95.89 %, 96.52 %

Table 2. Estimates of the output autocorrelation coefficient with their corresponding standard
errors by the proposed procedure.
Estimates

Procedure 11 s.e. 12 s.e. 21 s.e. 22 s.e.
Proposed 0.3649 0.4586 0.3629 0.4595 0.3504 0.4513 0.3514 0.4519
VAR(1) -0.0292 0.0292 -0.0330 0.0313 -0.0291 0.0462 -0.0365 0.0451

Table 3 . Estimated MAPE of the proposed procedure and VAR(1)
Estimated MAPE (Proposed procedure) Estimated MAPE (VAR 1)
MAPE 1 MAPE 2 MAPE 1 MAPE 2
4.13 % 4.21 % 62.19 % 81.35 %

4. Discussion and Conclusion
An estimation procedure is developed for the postulated model of
short bivariate time series with high dimensional inputs. The additive
bivariate model is postulated for a pair of correlated series explained by
its immediate past and by the contemporaneous effects of some input
series. This is to characterize short series that are simultaneously
influenced by the contemporaneous effects of some input series over
time.
Simulation scenarios affirm that the proposed estimation procedure
produces more accurate predictions and better estimates than VAR(1).
The proposed estimation procedure has relatively better predictive
performance than VAR(1) as reflected in the MAPE (less than 15%) given
the varying length of series, number of input series, and lags of input
series. Simulation results also show that the predictive performance of the
proposed procedure is robust to misspecification error (when variance is
three or six times larger). Minimal changes (between 1 % to 7 %) in the
estimates as reflected in the absolute percentage bias is observed across
the scenarios as the misspecification error is induced in the series.

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