Dynamic linear regression by bayesian and bootstrapping techniques

Abstract

Estimation of a dynamic linear regression is said to be of importance as events are measured on past events. However, estimation of dynamic models using the OLS is inefficient since it cannot produce the least variance and disregarding this problem would potentially lead to severe statistical problems. Therefore, the main objective of the study is to employ the bootstrap technique and Bayesian method in estimating the parameters of a dynamic regression model and compare their performances using standard deviation, absolute bias and mean square error of the estimates. The residual resampling technique is used for the bootstrap approach and the Normal-gamma model for the Bayesian approach. The results showed that the bootstrap technique outperformed the Bayesian method with lower standard error values. The bootstrap method also displayed an asymptotic property.