Search

The main purpose of this study is to address the question of how price forecasts can be produced in an accurate, reliable, and cost-effective form for planning and policy purposes in Mexican agriculture. The need for this stems from the recent opening of the Mexican economy, which resulted in a decrease government intervention, such that market demand, supply and prices are now influenced more by international events. Thus, this work investigates the application of univariate time series models to forecasting the trade prices for corn, dry beans, and sorghum in an import side, and tomatoes, broccoli, coffee, and sugar as export commodities. After reviewing the advances in econometric time series and analyzing commodity price formation, two models (ARMA and STS) were selected to fit the price series. These models are used to forecast monthly prices for a whole year. The forecasting accuracy, bias and risk of these models are compared, in terms of validation statistics, to the Random Walk and the Holt-Winter methods for forecasting. Considering the characteristics of the price series, the findings in this study suggest that: (1) The random walk model is useful to adjust medium to long-run forecasts. (2) For medium and long-run forecasts, the accuracy and bias displayed by the ARMA and STS models are similar and the differences in forecasting performance are not significant. (3) In terms of risk, both the ARMA and STS models show similar behavior. Since uncertainty for producers is different than for buyers, the optimal model depends on the users point of view. From the market analysis, the study concludes that: (1) Support programs must be specified in real monetary terms to avoid inflation uncertainty in Mexico. (2) Price forecasts should be included as a tool for designing programs, since it can lead to better agricultural production planning. (3) Risk must be included in planning since, for some products, Mexico is a buyer whereas for others, Mexico is the producer. Finally, further research for dealing with the undesirable statistical characteristics displayed by price series is required to improve the forecasting performance of univariate time series models.