Forecasting: Bayesian Inference Using Markov Chain Monte Carlo Simulation

Abstract

The evolution of modern computing techniques has targeted researchers and technical professionals to delve into an era where forecasting charts future business plans and events. Time series plays a vital role in forecasting processes. Diverse applications including physics portray time in multiple dimensions. In an attempt at recording and analyzing information, the early methods of forecasting used charts, indicators, and numbers. Many business cycles are not regular because they tend to vary due to multiple factors including Weather, Holiday, Season and Flood. Sustaining the commercial balance under such a scenario is done smoothly with the help of capturing weekly, monthly and daily behaviours. Selection and implementation of right forecasting technique require insights into historical data and mapping it with the market expectations which many a time is not aptly forcing for organizations. This paper represents various forecasting models and an approach to predict the operational/sales data on a daily basis using combined estimators. Markov Chain Monte Carlo (MCMC) has played an extraordinary role in modern engineering applications including economics, physics, statistics and beyond for the past decade. The heart of Monte Carlo simulation lies in the art of drawing random statistical samples. Empirical evaluation of various forecasting models resulted in the understanding of the stochastic nature of the processes. It is observed that out of all the time series models, MCMC yields a satisfactory outcome. Moreover, this research provides a comparison of the available forecasts and formulates the procedure for the best-suited technique. The results show that the outcome is a combined estimate of all the established prediction models.