Application of Bootstrap Data Generating Processes on the Nigerian Capital Market

Abstract

Nigeria stock market data set (1987-2015) was analyzed using parametric and nonparametric bootstrap DGPs. Simple linear regression (SLR) models were employed to fit the data set. Evidence showed that the original data set () without bootstrap resulted in a poor model with high bias. Results also showed that when the sample size was small ≤ 200, in almost all the bootstrap conditions, the nonparametric bootstrap method performed better than all of the parametric bootstrap models by showing the smallest conditional bias. If the parametric bootstrap model () with the high bootstrap level had not been explored in this study, the nonparametric bootstrap model () would have been exclusively ranked first in producing the smallest bias in the three tests length. But for large sample (≥10000), the reverse is the case. The  and models show that real income and saving rate are positively significant linear functions of the stock market capitalization and covaries with it. Across all the bootstrap conditions, it was obvious that all the models that worked well had the lowest values HQIC, SBIC, AIC, including the adjusted R² with standard error term ≤ 0.005 and minimum bias confirming that the models are good models for further studies and predictions in the economic sectors especially in the stock market.