An Artificial Neural Network Model for Predicting Retail Maize Prices in Kenya

Abstract

The intention of food price forecasting is to achieve reliability and usefulness for agricultural, non-policymakers, policy makers and agricultural related business. In the current globalization era, food security management in developing countries like Kenya that consider agriculture as a dominant economic activity require efficient and reliable food price forecasting models more than ever. Due to rare data availability and data time lag in developing agricultural dominated economies, normally needs reliance on time series forecasting models. Artificial Neural Network (ANN) modelling methodology gives a possible potential price forecasting method in developing countries based on available data. This study demonstrated the superiority of ANN over linear model methodology based on Root Mean Squared Error (RMSE) and Mean Absolute Deviation (MAD) performance metrics using monthly retail price (real prices) series of maize in three major counties. I.e. Kisumu, Nairobi and Eldoret. This study also portrayed the superiority of the ANN model in its univariate form over its multivariate form based on MAD and RMSE performance metrics. Lower comparative RMSE value would imply a better prediction while results with lower MAD were more close to actual values. Based on empirical study, showed that an ANN model is able to capture adequate number of directions of monthly price fluctuations as compared to models that employ the linear approach. It has also been observed that feeding the model with lagged observation of the same variable (univariate form) leads to more accurate forecasts than its performance in its multivariate form (Feeding it with different variables). Most models reviewed during this study, showed little effort in development of research tools, therefore this study has purposed to develop a user- friendly ANN prototype based on the proposed model.