Do the Selection of Volatility Models Depend on the Choice of Error Distributions: A Study on DS30 Index of the Dhaka Stock Exchange Limited, Bangladesh

Abstract

Research purpose: This study aims to explore whether the choice of the error distributions matters in the case of selecting the best-fitted models for the 30 blue-chip stocks from the Dhaka Stock Exchange (DSE), Bangladesh. Design/methodology/approach: The study encompasses the DS30 index of the DSE, from 1 January 2015 to 31 March 2025, and the data have been converted into logarithmic return form. To fulfill the objectives, the researchers compared the Generalized Autoregressive Conditional Heteroscedastic (GARCH) along with the Exponential GARCH (EGARCH) under three alternative error distributions, namely normal, Student's t, and generalized error distributions. Then, the researchers employed ARMA with GARCH and EGARCH techniques. For the model selections, two popular criteria — AIC and BIC —have been considered. Findings: The models MA(1)-GARCH(1,2), MA(1)-EGARCH(1,1); MA(1)-GARCH(1,1), MA(1)- EGARCH(1,1); and MA(1)-GARCH(1,1), MA(1)-EGARCH(1,1) have been determined as the preferred models under Normal, Student's t, and Generalized error distributions, respectively. However, the models selected under Student’s t error distributions have been proven to be superior. Hence, thechoice of error distributions is crucial when modeling the volatility of the DS30 index. Practical implications and originality: This is a novel research work that might guide the investors, who generally target the blue-chip stocks having good fundamentals, for their investments in DSE, Bangladesh. Research limitations: The comprehensive nature of this investigation necessarily imposed certain methodological constraints, precluding the inclusion of several additional ARCH/GARCH variants and
alternative distributional assumptions within the analytical framework.