Risk programming and sparse data: how to get more reliable results

Because relevant historical data for farms are inevitably sparse, most risk programming studies rely on few observations of uncertain crop and livestock returns. We show the instability of model solutions with few observations and discuss how to use available information to derive an appropriate multivariate distribution function that can be sampled for a more complete representation of the possible risks in risk-based models. For the particular example of a Norwegian mixed livestock and crop farm, the solution is shown to be unstable with few states of nature producing a risky solution that may be appreciably suboptimal. However, the risk of picking a sub-optimal plan declines with increases in number of states of nature generated by Latin hypercube sampling.