A Comparative Study for Detection and Remedy Methods of Heterocedasticity Problem in Linear Model Using Actual and Simulated Data

Abstract

Constructing an econometric model to be suitable for forecasting necessarily requires that it should be free from measurement problems (Heterocedasticity, Auto-correlation and Multicollinearity). The main object of this research focused on the problem of Heterocedasticity by comparing common detection methods (Breuch-Pagan Godfry, Harvey, Glejser, ARCH LM, White, Park, Spearman's Rank Correlation Coefficient and Gold-Field Quandt) and apply all five Remedies in addition to sixth remedy (suggested by the researcher) on the actual data (government expenditures as a dependent variable with GDP, inflation, money supply and exchange rate as independent variables in the period from 1977 to 2018) and other simulated data was corresponding to the actual data in order to testing hypotheses which most important of it are: There are no Significant differences among the results of all common detection methods and remedies when applied to the actual data once and re-applied to the simulation data again, and by analyzing them using statistical packages (SPSS V.20), (E.Views V.9) and (Excel V.10) program. And the most important result in actual data was that the best test led to the detection of Heterocedasticity is White's Test based on the criteria of the coefficient of determination ( ) and the probability value (Prob-Value), which proved its advantage in helping to detect the problem when applied in the original model and the remedies and the best remedy that led to remove the problem was the fifth remedy using the logarithm. It was proven that 7 out of the 8 detection methods led to the remedy, followed by the sixth remedy (suggested by the researcher) based on the general condition, where it is proven 5 out of the 8 methods of detection led to the remedy of the problem. In other side the most important results after applied in simulated data was that the best test led to the detection of Heterocedasticity is also White's Test, based on the determination coefficient ( ) and the probability value (Prob-Value) too, which proved its advantage in helping to detect the problem when applied in the simulated model and the remedies. The best remedy that led to remove of the problem here was the first remedy because it was proven that 6 out of the 8 detection methods led to the remedy, followed by the third and fifth (by using logarithm) Assumptions. It was proven that 5 out of the 8 detection methods led to the remedy, According to all research results for actual data and simulated data, the research recommend using White's Test to detect the problem of Heterocedasticity and remedy by taking logarithms.