Models and the Economy

"I often say that when you can measure something that you are speaking about, express it in numbers, you know something about it; but when you cannot measure it, when you cannot express it in numbers, your knowledge is of the meager and unsatisfactory kind: it may be the beginning of knowledge, but you have scarcely, in your thoughts advanced to the stage of science, whatever the matter might be."

That was from Lord Kelvin. But how does this translate to economics? The uncertainties surrounding human action and society are difficult to model precisely, and it would appear not a problem of "getting more data" because the interactions themselves can change over time. Here is how Nobel laureate Lars Hansen thinks about model uncertainty:

"I like to think about there being three different components to the concept of uncertainty. [.... First,] suppose I write down some model and the model has what economists would call shocks, distributions attached to these shocks and the like. That model, when fully specified, will tell you probabilities of all the future events, what's in the domain of the model. You've got a full--there's uncertainty out there, but it's certainty under which you've got a model that just tells you all the probabilities of everything. And so once you've got the model, it's done. So, I like to think of that as risk. Like, if I fully embrace this model, there's the risk component. 

[....Second,] there are different models out there. Even a given model, I might not know all the details of it--the so-called parameters of the model. There may be multiple models out there, and the like. So now, for me to assign probabilities on the future I have to start say, well how much weight do I want to put on this model versus that model? Each distinct model and the like. So this issue about how I want to weight, how much confidence I put in the different models out there--once I take a specification of that confidence and continuous assign probabilities to things, that process of assigning probabilities across models--there I think of that as a potential source of ambiguity. I'm not really sure how to do that. And how do I confront that component of uncertainty.

There's a third component that I think is probably the hardest part, but I think maybe that in many respects the most important part is all the models are in some sense wrong. How do I use models in sensible ways, in ways that are in some sense robust to different forms of misspecification? I acknowledge that they are wrong, but if I knew exactly how they are wrong, I'd just fix them. So I have to somehow confront that form of uncertainty, as well. So, those are the different pieces that I can think about when I think about uncertainty."

That was from a discussion on EconTalk. So there's uncertainty as to what the model output will be, there are competing models with different inputs, and there is the epistemological problem that all models are wrong in some way. I actually see this quite akin to the uncertainty facing the natural sciences, especially the more detailed or aggregated we look at the universe.