Global Journal of Human Social Science, E: Economics, Volume 23 Issue 2

Volume XXIII Issue II Version I 59 Global Journal of Human Social Science - Year 2023 ( ) E Impact of COVID-19 on Stock Market Volatility and Forecast using ARIMA and EGARCH R-squared 0.996476 Mean dependent var 7.779222 Adjusted R-squared 0.996472 S.D. dependent var 0.180569 Log likelihood 3162.471 Akaike info criterion -7.530956 Note: *, **, and *** denote statistical significance at 10%, 5%, and 1% level The exponential terms (exp -0.03326 =1.2156646) indicate that for the Ghana Stock exchange composite index, the bad news of COVID-19 has a rather large symmetric effect on the volatility of the stock. The exponential term was however not significant even at 10% level. But for the insignificance of the asymmetric term, negative shocks invoke greater volatility than a positive shock. The bad news of the COVID-19 did not influence the volatility of the stock exchange. f) EGARCH Diagnostics Table 5: Diagnostic test of Appropriateness Logged GSECI Normal Gaussian Student t’s GED Student’s twith fixed df Significant Coefficient s All* 2 2 3 ARCH Significance Yes Yes Yes Yes GARCH Significance Yes Yes Yes Yes Log-likelihood 2758.957 3074.970 3162.471* 3020.470 Adj R 2 0.996443 0.996469 0.996472* 0.996472 Schwarz IC -6.536429 -7.282604 -7.491437* -7.160566 Heteroscedasticity No No No No Autocorrelation Yes Yes Yes Yes Note: * represents the best modelSource: Authors computation In choosing the preferred model, we depend on the four different error constructs in Table 5 above. The model must be parsimonious. Thus, the ARCH and GARCH coefficients must be statistically significant. The generalized error model has the highest adjusted R-square and the log-likelihood ratio. The Generalized Error Distribution (GED) model also the lowest SIC information criterion which gives the heaviest penalties for loss of degrees of freedom. All the models have the same results for test of heteroscedasticity and serial correlation. The reasonable tradeoff is to choose the generalized error distribution model. From the GARCH (1,1) model in Table 4, both the GARCH and ARCH models are positive and significant at one 1% level. The residual test reveals that the model passes the residual test since the F-statistic is not significant at 1% level. From Table 5, there is no evidence of heteroscedasticityin the residuals. Using the correlogram Q-statistics, there existed no serial correlation in the residuals. The ACF and the PACF lie within the confidence intervals as shown in Figure 2. There exist no probability values of the Q-statistics below the alpha level of 1% indicating that there is no serial correlation. Evidence of serial correlation here is when the p-values of the Q-statistics is are statistically significant. 7.4 7.5 7.6 7.7 7.8 7.9 I II III IV I II III IV I 2019 2020 2021 LGSE_CIF3 ± 2 S.E. © 2023 Global Journals