Abstract:
This study was intended to forecast the inflation of Rwanda using macroeconomics variables, the variables that used were money supply, interbank rate, exchange rate, international oil price and lending rate. The change in consumer price index was treated as inflation. The objectives of the study were to forecast inflation of Rwanda and compare the performance of machine learning algorithms with the existing methods used. The study used quarterly time series data from National Bank of Rwanda from 2006 Q1 to 2019 Q4. The data were analyzed using various methods such as ARMA model used to model dependent variable when the predictors are the past values of that variable with specified number of lags. VAR model which uses its previous values and the previous
values of other variables was also used with specified number of lags. Machine learning techniques Random Forest, Ridge Regression, LASSO Regression and K Nearest Neighbor (KNN) were used to forecast inflation and evaluation were made on RMSE. The results of different techniques gave different RMSE for the ARMA model the resulted RMSE was 1.176938, for VAR model the RMSE was 1.17868, for K Nearest Neighbor the resulted RMSE was 1.2382, for Random Forest the resulted RMSE was 0.5111, for Ridge regression RMSE was 1.2148 and for LASSO Regression, the resulted RMSE was 1.2868. so, Random Forest model was the first model to have small root mean squared error and is the best model that work well in forecasting inflation followed by existing models ARMA and VAR model due to the small RMSE compared to the others. Based
on results, implementing machine learning techniques for forecasting Rwandan inflation is a promising endeavor. So, first line of work could be the improvement in the application of methods that underperformed in the study, as well as the potential to extend the work to include other machine learning techniques such as Neural networks methods.
