Forecasting performance of mixed data sampling (MIDAS) regressions, autoregressive distributed lag (ADL) model and hybrid of GARCH-MIDAS model: a comparative study

• Main Authors: Bawa, M. U., Shabri, Anil, Dikko, H. G., Garba, J., Sadiku, S.
• Format: Article
• Language: English
• Published: JMPES 2021
• Subjects: QA Mathematics
• Online Access: http://eprints.utm.my/id/eprint/97762/1/AnilShabri2021_ForecastingPerformance0fMixedDataSampling.pdf; http://eprints.utm.my/id/eprint/97762/; https://www.mathematicaljournal.com/archives/2021.v2.i1.A.19

This paper considers the Comparison of forecasting performance between Mixed Data sampling (MIDAS) Regressions model, Autoregressive distributed lag (ARDL) Model and hybrid of GARCH-MIDAS. The data employed for this study was secondary type in nature for all the variables and it is obtained from the publications of Central Bank of Nigerian bulletin, National Bureau of Statistics and World Bank Statistics Database dated, January, 2005 to Dec, 2019. The result of unit root test shows that all variables are stationary at level and after first differences at 5% level of significant. From the results we found that F-statistics 1.895554 is inside the regions defined as the lower and upper bound (3.62 and 4.16) at 5% level of significant, this implies that there’s no long-run relationship between dependent variable (NSE) and independent Variable (CC). using forecasting evaluations with shows that that GARCH-MIDAS has a least value of RMSE and MAPE than ARDL and MIDAS model (1823.531 and 3.976542) is least than for MIDAS and Ardl models (2372.846, 4.765421 and 2134.732, 5.952348). Finally, we can conclude that GARCH- MIDAS model outperform MIDAS and ARDL models of Nigeria Stock Exchange.