Forecasting errors pose a serious problem of identification, often neglected in empirical applications. Any attempt of estimating choice models under uncertainty may lead to severely biased results in the presence of forecasting errors even when individual expectations on future events are observed together with the standard outcome variables.