It was suggested that five is the minimum number of data points for the least likely category when constructing discrete distributions. In many cases, however, we must estimate probabilities that are extremely small or for which relatively few data are available. For example, the probability of bankruptcy for a firm is (hopefully) quite small, but it is important for both borrowers and lenders to know this small probability value. Typically, either expert opinion or proxy measures are used to determine these small probability values instead of using data, What makes us reluctant to use data to estimate small probability values?