Smoothing non-stationary noise of the Nigerian Stock Exchange All-Share Index data using variable coefficient functions

Abstract

We employed variable coefficient models or generalized additive models (GAMs) with scatterplot smoothers such as smoothing splines to establish non-linear relationships that exist between non-stationary Nigeria Stock Exchange All-Share Index and three major monetary policy macroeconomic indicators using data collected from Central Bank of Nigeria and Nigeria Statistical Bulletin between 2004 and 2014. Two flexible smoothing splines were postulated using backfitting algorithm with weights and stringent control on convergence precision and maximum number of iterations for convergence. This was to ensure convergence of the iterative estimation procedure and to provide unbiased estimate of the regression coefficients and their standard errors. Generalized Cross-validation (GCV) re-sampling technique was employed in determining the effective degrees of freedom EDF and smoothing parameters for the smoothing spline functions. We compared models accuracy using the residual deviance for the GAMs. Smoothing splines without interaction terms was selected. The analysis shows that volatile exchange rates, rising inflation rates and Treasury bill rates remain the major monetary policy macroeconomic variables causing instability in the growth of the country’s major stock market index, which is in line with conclusions drawn by [1]. R programming language packages were employed throughout the analysis.

Keywords: Nigerian Stock Exchange All-Share Index, variable coefficient models, generalized additive models, generalized cross-validation, smoothing splines and R programming language.