Experimentation with heteroskedastic noise

Abstract

The direction of information with respect to the choice variable can easily change under a general class of distributions. The implication is that, when information is a function of the choice variable, the result of a model may be an outcome of the structure of the uncertainty and not of the existence of uncertainty. Thus, incomplete information does not always have such a clear effect on an agent's decision as previous models have suggested. If such results are key to the conclusions, then the structure of the model should be examined. The model presented here can be used to determine the relationship between information and the choice variable and to derive basic insights into the particular model being examined. It can also be used to eliminate the informational aspect of decisions so as to examine other aspects of a model, e.g., the incentive for information transmission (jamming) and other dynamic aspects (e.g., capital accumulation). Finally, the functional form of the unknown function as well as the error function are inportant in determining the direction of increased information, as the direction of increased noise can be the direction of experimentation.