MODELING DYNAMIC MICRO AND MACRO PANEL DATA WITH AUTOCORRELATED ERROR TERMS
No Thumbnail Available
Date
2018-05
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
UNIVERSITY OF ILORIN
Abstract
Estimation and Inference in dynamic panel data models is limited by the presence of autocorrelation of the error terms. This causes the traditional panel estimators to be biased and inconsistent. This study was aimed at investigating the sensitivity of some dynamic panel data estimators in the presence of autocorrelation. The objectives of the study were to:
(i) evaluate the performances of some dynamic panel data estimators in the presence of autocorrelation; (ii) evaluate the performances of the estimators with change in sample sizes and time periods; (iii) propose an estimator by modifying the existing estimators; and (iv) compare the performance of new estimator with some of the existing estimators at different sample sizes and levels of autocorrelation.
Simulation study was carried out using Monte-Carlo experiment in the environment of R statistical package to generate data with panel structure for sample sizes 10, 20, 50, 100 and 200 and time periods 5, 10, 15 and 20, autoregressive coefficients ( and =0.3, 0.5 and 0.7) and autocorrelation coefficients ( and = 0.2, 0.5 and 0.9). Five dynamic panel data estimators; Ordinary Least Square (OLS), Arellano-Bond Generalized Method of Moments one-step (ABGMM1), Blundell-Bond System Generalized Method of Moments one-step (SYS1), Anderson-Hsiao Instrumental Variable in difference form (AH (d)), Proposed modified estimator (P-est.) were compared. Two robust estimators (M and MM) were also used. Real life data from the Organization of Petroleum Exporting Countries was used to confirm the results from simulation study. Absolute bias and root mean square error were used to evaluate finite properties of the estimators
The following were the major findings from the study:
(i) AH (d) and ABGMM1 estimators performed well in the presence of autocorrelation.
(ii) AH (d) estimator performed relatively well when the time period is small while ABGMM1 estimator outperformed all other estimators when sample size (n) is large for all the time periods considered;
(iii) The results showed that all estimators (with the exception Blundell-Bond System GMM and OLS) generally performed better with small and large time periods (T). However, ABGMM1 seemed to show the largest improvement as sample size (n) and time periods (T) increases;
(iv) The proposed modified estimator was obtained from the existing estimators;
(v) The proposed modified estimator outperformed all other estimators in small and large samplesize irrespective of time period; and
(vi) Results obtained from the analysis of the Real life data validated the findings from
Monte-Carlo studies.
In conclusion, the proposed modified estimator is more appropriate in estimating the parameter of lagged dependent and exogenous variables when dealing with dynamic panel data models in the presence of autocorrelation. It is recommended to use the proposed modified estimator when dealing with dynamic panel data models in the presence of autocorrelation
Description
Keywords
MODELING, DYNAMIC, MICRO, MACRO, PANEL DATA, AUTOCORRELATED ERROR TERMS