Agricultural commodities have formed the foundation of economic activity, their production and trade is influential at the local, national and international level. Farmers, traders, manufacturers and consumers each operate under decision processes which drive both the production and trade of agricultural commodities. The critical factor that underlies these decision processes is the volatility in the price of the commodity. This represents the uncertainty in the price, or price risk, which is used to make decisions for future operations. Coffee is an agricultural commodity that is both popular among consumers in developed nations and is important to many developing nations. As such, there is interest in deriving estimates of the volatility in the coffee price to assist the decision process.
The Autoregressive Conditional Heteroskedasticty (ARCH) model and the Generalised Heteroskedasticity (GARCH) model both estimate temporal volatility, they have been used extensively in both economics and finance. Ordinarily, ARCH and GARCH models are estimated with the classical maximum likelihood (ML) framework. Problems arise when estimating through this framework, but these are overcome with the use the Bayesian framework. Estimating ARCH and GARCH models through Bayesian techniques is preferable for three reasons: (1) finite-sample results can be obtain; (2) the underlying inequality constraints can be imposed; and (3) uncertainty between models can be explicitly taken into account. This dissertation examined monthly coffee prices since 1989, estimating both in the maximum likelihood and Bayesian framework. Focussing upon the estimation of volatility in the coffee price within and forecasting outside the sample.