Bayesian Model Averaging for Model Uncertainty in Inflation Rates Modeling in Nigeria

Abstract

Applied researchers are frequently faced with the issue of model uncertainty in situations where many possible models exist as a result of several regressors or predictors variables motivated by different theories. For instance, having over 40 regressors (k) to formulate model can cummulate into trillions of possible models (2^k). Thus, data analysts are unsure of which regressors are useful.  An alternative approach to model selection is to compute a weighted average of the estimates of the all competing models. Bayesian model averaging (BMA) has a coherent mechanism for dealing with model uncertainty This is evident in the properties and uses of posterior model probabilities. Also of concern is the issues of which of the predictor variables included the models are relevant or significant in the data generating process. This study investigates the key drivers of inflation rates using thirteen likely predictors resulting in 8192 plausible models comprising of all possible combinations of the predictors. Each model was weighted accordingly with a model uniform prior and parameter prior. Using Markov Chain Monte Carlo (MCMC) algorithm, which generate draws from a Markov chain on the model space with the posterior model distribution as its stationary distribution. Model posterior probability and posterior inclusion probability were determined in order to obtain the most appropriate model for inflation rates. Hence the Bayesian model averaged for inflation rates consists of an average of four predictors showing the real interest rates with posterior inclusion probability equals to 1 and the mean number of regressors is 3.425 for the best 1527 models.