I propose the hypothesis:
It is very rare for models to change beliefs about reality.
We (researchers) spend a rather large amount of time on structural models. Especially on getting quantitative answers out of them. Do the answers we get have much of an impact on what people believe about reality. I think the answer is “no”.
I am familiar with research on economic growth, cross-country income differences, and wealth distribution. Consider the basic questions from that literature.
Our current understanding seems alarmingly similar to that of 1996, when I got my degree.
Lots of research has changed beliefs very little. Why? Because we don’t have clear evidence that we can interpret without a model.
As for deep causes, most researchers probably believe that institutions are important. I would argue that this belief has not changed much over time. It has been supported by clever empirical papers (Acemoglu et al.). That work has changed beliefs. Notably, it is purely empirical, supported by a coherent historical narrative.
We don’t really know anything about this.
We like to write down “R&D” models, but that seems to have more to do with prejudice than with evidence. Is there any evidence to support that “R&D” is really what drives growth?
It’s probably true, but I say this because of my intuitive understanding of how the world works, not because of evidence.
Here is a case where a model probably did change beliefs (Huggett 1996). I think this worked because households don’t do very much in that model.
We have a bunch of candidate solutions for generating high wealth holdings (inheritances and entrepreneurship are probably the leading candidates). How important these are is still not settled.
Most researchers probably believe that entrepreneurship is important. But then this is somewhat obvious from looking at the data. I don’t think that models did much to generate this belief.
If models don’t change beliefs, then we are probably wasting a lot of time in macro research.
Perhaps the old-fashioned approach of showing correlations (e.g., OLS regressions) and offering a coherent interpretation is underrated.
Last updated: 2016-Dec