An Artificial Neural Network Model For Forecasting Inflation In Kenya

Abstract

The greatest priority of Monetary Authorities in an economy is price stability. Unstable prices result to inflation whose economic effects are undesirable. Price stability, calls for timely prediction of forecasting of inflation. This enables the Central Bank to take quick action by managing the situation causing persistent price changes. In Kenya, Inflation is forecasted using linear forecasting models which assume that economic data is linear in nature. Economic data is complex and nonlinear. Therefore, forecasts made using linear models may be inaccurate. Nonlinear models have been applied with much success in forecasting in inflation. Artificial Neural Networks stand out as demonstrated by reviewed literature. Different models of inflation forecasting were explored on their suitability to forecast inflation in Kenya. The viability of these models and performance was explored through literature review. This project was realised by collecting inflation, GDP growth and oil prices from the Kenya National Bureau of Statistics, Energy Regulatory Commission and Petroleum Institute of East Africa. The data was obtained from publications and the websites. Through experimentation, an ANN model of configuration 3: 12 was developed using 70% data for training, 20% percent data for testing and 10% for validation. This model had a RMSE of 9.52 against the ARIMA model whose RMSE was 19.49. Based on the RMSE values, was concluded that the Neural Network model is a better inflation forecasting model when compared to the ARIMA model. This project describes the development, architecture and implementation of an ANN forecasting model for Kenya and its benchmarking with the ARIMA model.