A principal component analysis of multivariate data on inflation for Nigeria

David E. Omoregie ; Virtue U Ekhosuehi

Mathematica Applicanda, Tome 48 (2020) no. 2, pp. 115-132.

Résumé

For quite some time now, the Central Bank of Nigeria (CBN) analyses multivariate data on inflation so as to account for diverse sources of inflationary pressures at the current period and to monitor the inflation pattern in the economy. When the data are subjected to the classical multiple regression analysis using the Ordinary Least Squares (OLS) method, some of the variables may be highly correlated causing statistical insignificance. This may lead to exclusion of some variables from the fitted model. When this happens, the cost of data collection for such variables is a waste. Regardless of the outcome from a variable selection technique, this study is designed to familiarise monetary policy makers with the possibility of integrating Principal Component Analysis (PCA) with regression analysis so that a few variable components are utilised without excluding any explanatory variable. The paper models the multivariate data at the CBN database on inflation, and extracts important artificial orthogonal variables from the linear combinations of the observed explanatory variables, although with a penalty cost of excluding components with observations that are minimally separated. The PC-based model explains 95.91% of variations in the headline inflation with a mean difference (between the actual and the predicted inflation) that is statistically not different from zero. The study is a significant addition to the existing methodologies for inflation forecasting in the literature as it is one of a few works that apply PCA-based technique to predict headline inflation.

Détail

DOI : 10.14708/ma.v48i2.7043

Classification : 62P20, 62J05
Mots-clés : inflation, orthogonal variables, multivariate data, principal component analysis, regression