Main menu

Modeling the Nigerian Inflation Process Using the Bayesian Model Averaging (BMA) Approach

Abstract: 

Model fitting or variable selection in standard econometric practice ignores model uncertainty. Bayesian Model Averaging (BMA) is an important Bayesian tool capable of dealing with model uncertainties. It is a weighted averaging method based on posterior distribution and it provides an improved out-of-sample predictive performance. In implementing BMA, a prior distribution is specified in two parts; a prior for the regression parameters and a prior over the model space. In this paper, we apply the BMA to an empirical model on inflation which originated from both demand and supply sides. The parameter prior used is the Unit Information Prior (UIP) since from the literature it has been shown that it outperforms all other parameter priors. And the Uniform prior is considered for the model space because of its equal probability assigns to all possible models in the models space. The inflation rate in this study is simply the percentage annual growth rate of the headline consumer price index (CPI) which represents the CPI in the empirical analysis. Other series that are of interest to the study are expected inflation, real output, real exchange rate, interest rate, the ratio of fiscal deficits to GDP, money supply, oil prices and rainfall. The real gross domestic product (real GDP) is used to represent the real output in the empirical analysis. The data are annual, covering the period (1970-2011), and are all sourced from the CBN Annual Reports and Statements of Account and Statistical Bulletin (various issues) , the International Financial Statistics of the International Monetary Fund (IMF) and World Development Indicators (WDI) & Global Development Finance (GDF). Considering the framework of BMA approach, it reveals that the best 5 models are selected from the available model space of 134,217,728 (227) based on their Posterior Model Probabilities (PMP) as well as their Bayesian Information Criterion (BIC). Model 1 is chosen as the best because of the highest value of its PMP which includes the variables; expected inflations, lending rate and real exchange rate. If interest is the variable (inclusion), real exchange rate should be given the most priority for modeling the Nigerian Inflation process because of the highest value of its Posterior Inclusion Probability (PIP): it appears in all the 5 best models for the Inflation in Nigeria. Moreover, examining the uncertainty imbedded in the model, if we had chosen the best model here, we are 94% uncertainty about the model. Furthermore, almost all the variables have their standard deviations greater than their means except those that are highly significant

Corporate Author: 
Ethipian Economics Association (EEA)
Publisher: 
Ethipian Economics Association (EEA)
ISBN/ISSN: 
978-99944-54-46-4
Primary Descriptors: 

 Inflation

Secondary Descriptor: 

Bayesian Model Averaging; Posterior Model Probability

Geographic Descriptors: 
Nigeria
Cataloge Date: 
10/10/2015
Broad Subject heading: 
Inflation (Finance) - Nigerian- Econometric Models
Call Number: 
330.963 PRO 2015
Serial Key Title: 
Proceedings of the 19th Annual Conference of Africa Region Chapter of the Econometric Society
Publication catagory: 
Content type: 
Volume: 
III
Publication date: 
2015-06-01 00:00:00
Conference Place: 
EEA Conference Center
Place of publication: 
Addis Ababa, Ethiopia
Type of material: 
Book
Current frequency: 
Annually
Conference date: 
July 16 to 19, 2014