Throughout this book I have attempted to analyze critically the explicit process of building models of the economy or aspects of the economy. Model building is so widely practiced today that it becomes very difficult to question the soundness of the process. Even when we accept the traditional theories of the textbooks, we still must make decisions regarding the process of applying them to the real world. As such, the process of model building is certainly not automatic. There may be infinitely many potential models that could be built to use any given behavioural hypothesis. However, it does not matter whether we view model building as a process of choosing between existing complete models or a process of building from the ground up, so to speak, assumption by assumption.

As is evident in much of the discussion so far, a model builder must begin by addressing a simple but key question: What is the purpose for building any model? No single model will serve all purposes. The models we build to describe or explain the real world will not necessarily be usable for testing the behavioural hypotheses used in their construction. Similarly, models built for quick application to policy issues may not always be sufficiently realistic for plausible explanations or convincing tests – recall from Chapter 5, this is the distinction Mankiw [2006] was making between engineers and scientists. Sometimes one might suspect that these considerations may not be well understood or appreciated by most economists. Too often it is assumed that a model which has been shown to be successful for one purpose (e.g., description of data) will automatically be sufficient for another (e.g., testing). Of course, in this chapter I will be discussing only models designed for the purpose of testing. However, another distinction is that testing can be on the basis of either singular observations or, as will be discussed in the next chapter, be on the basis of statistical observational data.