الفهرس 
Model selection methods in autoregressive distributed lag models with missing data
AbstractOnly 14 pages are availabe for public view
 |  | from | 141 |  |  |
| | | from | 141 | | |
Abstract
In the autoregressive distributed lag (ARDL) model, selection criteria are
considering as an important issue. Based on these criteria, one determines the
optimal order of the ARDL model. Moreover, if the dataset contains missing
values, this will of course affect the optimal order of the ARDL model and also
the estimation efficiency. Therefore, in this study, we propose to use three
imputation methods (K. Nearest Neighbors (KNN), Predictive Mean Matching
(PMM), Expectation Maximization (EM)) for handling the missing values and
then get more efficient estimation of the model with the optimal order of lags
depend on two model selection criteria which Akaike Information Criterion
(AIC) and Bayesian Information Criterion (BIC).
In addition, we using combination methods that includes: Minimum,
Maximum, Simple Average, Weighted Mean and Median methods to combine
results of imputation methods and we demonstrate that the combination is
preferable to improve the imputation data instead of using them individually
through fitting ARDL model of all possible combinations. Finally, we compare
all output of methods by MAE, MSE and RMSE criteria.
Practically, we study and compare the performance of these methods based
on a real dataset application and Mote Carlo simulation study to making
comparison between the behavior of estimation different methods at different
sample size and different percentage of missing was held based on two model
selection AIC and BIC to compare between estimation methods. Also, we
evaluate this model selection to choose the correct order of all ARDL models
for all imputation and combination methods after handling missing data.