Although microeconometric models have many advantages over macroeconometric models, they have some disadvantages that stem from the characteristics of the available data. Of particular relevance here is the dichotomy between the information which is observed by the individual decision maker and the information available to the investigator, often referred to as the observational rule. For example, many models of individual choice are predicated on an underlying continuous random variable representing utility. However, in many instances the revealed preference of the agent provides only discrete indicators signaling, as an example, the preferred brand, occupation, or model choice.

If we consider a single equation then we may view the progression from the least squares model with continuous regressors, to the censored, truncated, and discrete choice models, as one of progressive information loss. This information loss is manifest in the need for more complex econometric models. In a multivariate setting the consequences of non-observability for estimation are considerably greater, and have provided much of the impetus for the developments in simulation-based inference.

It is instructive to consider the recent developments in simulation-based inference in a historical context. At any given point analysts may face the trade-off between a preferred modeling strategy and what is tractable given computer technology. If we examine the development of microeconometric models this trade-off is very much apparent. For the sake of exposition it is convenient to consider three stages.