Applied Econometrics is like that old car that your grandparents gave to your parents that your dad drove you in to your little league baseball game. It didn’t work quite as well as both of you hoped. After a two hour, four strike out game at the plate, that blue sedan might not turn on.

A lot of the applied econometrics that we’ll encounter in this class is concerned with assuring you that one’s approach works. That their estimate is super robust. That it doesn’t matter how they estimate the thing they get the same answer every time (or there about). In my experience, that’s not typically how it works. It’s more like that blue sedan.

The process starts with a research question like — how would the availability of affordable housing change if Fannie Mae securitized construction loans? That’s an interesting question. You then do a google search of (1) applied econometric techniques and (2) of historical events to see how best you could take some real world data and produce an estimate of the causal impact.

Most applied econometric techniques (we’ll cover them shortly) more or less assume that one’s data can be roughly thought of as being formed by a collection of local randomized control trials. And so you realize pretty quickly that this research question while important can’t be credibly tackled with applied econometric techniques - It’s really hard to find a context where the availability of securitized construction loans (or even an offer!) is randomly assigned.

So in practice, we often cannot tackle the most interesting questions with causal inference. But that said, as practitioners, this shouldn’t disuade us from pursuing questions which are clearly important.