الفهرس 
1.3.1 Objective of Previous Studies
Causal Autoregressive ModelsOnly 14 pages are availabe for public view
 |  | from | 140 |  |  |
| | | from | 140 | | |
Abstract
In this study, we examine the marginal benefit of allowing the noncausality for forecasting the inflation rate in Egypt. In the empirical section, the study compares the forecasting performance of causal AR models that are dependent only on the past, with corresponding mixed causal-noncausal AR models that allow for explicit dependence on the future. We consider a time series consisting of 117 monthly values of the inflation rate in Egypt over the time period from January 2007 until September 2016. The results investigate from the presence of noncausality components in the inflation series in Egypt. Then, an approximate maximum likelihood estimator assuming non-Gaussian disturbances for the mixed causal-noncausal model is used. This model is found to outperform the causal model in terms of the mean square errors and mean absolute forecast errors. The empirical study concluded that allowing for noncausality to the inflation series in Egypt may lead to estimating suitable model and improving the forecast accuracy. In a simulation study, we compare the performance of the MLE, assuming a t-distribution, with that of the LAD estimator. The simulation shows that both methods capture the mixed causal-noncausal process as well, particularly when the distribution has very fat tails. Also, the study investigates whether the distribution of the error process for a mixed causal-noncausal models has an effect on the performance of the maximum likelihood estimation method. To that end, via simulations, the study compares the performance of the MLE under t-distribution, Lévy distribution and other stable distributions. The results indicated that the heavier tails distributions for the error term may be suitable for identification of the mixed causal-noncausal models and estimation techniques.