Abstract:
Price fluctuations affect livelihoods and social stability; thus it is crucial for farmers, distributors, processors, and policymakers to study price behavior and predict future trends when planning agricultural activities. Over time, various forecasting methodologies have been developed to predict grain prices. However, some of these methodologies like Autoregressive Integrated Moving Average (ARIMA) and neural networks do not produce accurate forecasts due to inherent limitations such as the ARIMA’s linearity assumption and the specific data requirements for neural networks. To address these issues, this study aims to use non-parametric models, like Dynamic Generalised Additive Models (DGAMs), to predict future grain prices in South Africa. The study will also be looking at the Bayesian Dynamic Generalised Additive Models (BDGAMs) since DGAMs can be approached from a Bayesian perspective. BDGAMs is where prior beliefs about function nonlinearity influence model complexity and smoothness penalties. Historical time series data from Republic South Africa’s South African Future Exchange (RSA SAFEX) Domestic future prices spanning the period 1996 to 2024 and factors such as weather variables, exchange rate, and fuel prices from appropriate sources are used in fitting the DGAMs model. The results will then be compared to traditional ARIMA models to assess the performance of the DGAMs model. DGAMs are more flexible thus they can capture
complex patterns and may yield more accurate results. The DGAMs are implemented in the CRAN R mvgam package to estimate the parameters for discrete time series and to generate probabilistic forecasts. The results shows that the best model that can be used for forecasting maize and wheat prices is the DGAMs and the factors that are significantly contributing to the rise in wheat and maize prices in South Africa are exchange rate and inflation. The study recommends that the results can be used to provide farmers and stakeholders with price certainty and the impact of factors affecting the price for planning and policy directions on tackling food insecurity. Other researchers can focus on using more variables or expanding the model to different commodities for broader applicability.
