Bayesian Sequential Updation and Prediction of Currency in Circulation Using a Weighted Prior
Shivangee Misra *
Department of Statistics, University of Lucknow, India.
Rajeev Pandey
Department of Statistics, University of Lucknow, India.
*Author to whom correspondence should be addressed.
Abstract
Aims/ Objectives: The objective of the study is to analyse and predict the dependent variable using bayesian sequential updation of priors. An information weighted criterion is constructed from the previous information. A bayesian multiple regression model under student's t distribution and log normal error distribution is performed. Priors are updated on a sequential basis. Consequently, model comparison evaluates the best model and predictions are made for the future period . The illustration is performed on a real dataset of currency in circulation and its related macro economic variables.
Methodology: A weighted prior of the regression estimates is constructed from two mutually exclusive parts of the same data using the Deviance Information Criterion (DIC) obtained after performing the bayesian regression on the two parts under two dierent likelihoods- student's t and log normal distributions. The prior so constructed is further used and gets updated for prediction purposes.
Conclusion: It was found that the weighted prior thus constructed improved on sequentially after updating the priors and incorporating the previous information into the likelihood. Consequently, since DIC was the lowest in log normal error likelihood, it was concluded to be of the best fit to the dataset.
Keywords: Bayesian sequential updation of prior, log normal error disturbances, t distribution, DIC, CIC, macroeconomic indicators, weighted prior