Title:
BILINEAR REGRESSION METHODS FOR MACROECONOMIC IMPACT ON CORRUPTION IN GHANA

Authors:
Solomon Crusader Abo, Paul Hammond and Edmond Amissah, Ghana

Abstract:

This paper aimed to compare Ordinary Least Square (OLS), Principal Component Regression (PCR), and Partial Least Square Regression (PLSR) methods in handling multicollinearity among macroeconomic variables using real and simulated data. The real data on macroeconomic variables from The Global Economy was employed spinning 2001 to 2020. To assess the effect of a sample size of Root Mean Square Errors (RMSE), multivariate random normal data was simulated where there was the presence of multicollinearity among all explanatory variables with various samples, n = 20, 50, 100, 200, and 500. The simulations, as well as the analysis of the data, were implemented in R software. Comparing the performances of the three methods through Root Mean Square Error (RMSE), R2, and optimal value, the results of the real data indicated that the PLSR model performs better than the OLS and PCR models when there is the presence of multicollinearity in macroeconomic data. It was evident that only the unemployment rate and fiscal freedom influence the corruption perception index in Ghana. Again, based on the simulation results at n = 20, 50, 100, 200, and 500, PLSR still performs better than the PCR in both small and high samples. Finally, the results indicated that sample size affects the value of RMSE, as the higher the sample size used the lower the RMSE value from both PCR and PLSR. It was concluded that the PLSR method is more capable of overcoming multicollinearity problems in both real and simulated macroeconomic data. Again, we concluded that sample size affects the value of RMSE in macroeconomic data. It was recommended that PLSR is the best tool for determining macroeconomic variables that affect Corruption Perception Index (CPI). Also, the government of Ghana should formulate policies to drastically curb unemployment and again, to ensure that people pay various forms of taxes in Ghana. We also recommended that a large sample size must be employed in working with macroeconomic data, considering the PLSR method as efficient and effective in handling multicollinearity in such data.