Can a Decision Tree Forecast Real Economic Growth from Relative Depth of Financial Sector In Nigeria?

Abstract

We employed a decision tree statistical learning method which is lately gaining wide usage in the field of econometrics to establish the relationships between real gross domestic products growth rate and financial depth indicators such as stock market turnover ratio, credit to private sector (CPS) and broad money supply (M2) relative to gross domestic product (GDP) in Nigeria between 1981 to 2016. The data was divided into training and test datasets. The former was used to train the decision tree while the later was used to test the performance of the fitted decision tree model. Recursive binary splitting produced a fitted tree with nine nodes (leaves). This tree was pruned using cost complexity pruning procedure which uses a tuning parameter α to control the tradeoff between the tree complexity and overfitting the data. Pruning produced a tree with four terminal nodes and improved predictability in terms of lower model MSE on test dataset and interpretability. Bagging and Random Forest procedure were employed to further improve the performance of the model by aggregating bootstrapped training samples in order to reduce the variance.