An empirical Bayes procedure for adaptive forecasting of shrimp yield

In order to make decisions regarding when to harvest cultured shrimp, short-term forecasts of future growth must be compared with expected future costs and changes in the value of the shrimp harvested. As with most agricultural products, researchers and business professionals collect and maintain data on the various factors that influence growth. Though many general fixed relationships between, say salinity, temperature, time and turbidity are known, exactly how these affect a particular crop changes from year to year. Vitality of the stock, unmeasured characteristics of the pond, and other factors which vary over time have a significant affect on the yield of a harvest. Consequently, to make reasonable short-term forecasts, forecasting models must include both the general factors gathered from past experience and crop specific variations from these overall effects. This paper presents a method of obtaining such a forecasting equation using a general mixed model setup currently available in many computer packages. These general formulas are illustrated with a real life example. SAS code necessary to implement the analysis is also included. The paper concludes by presenting a general equation for short-term forecasts using the estimates obtained from a SAS analysis. q 2000 Elsevier Science B.V. All rights reserved.