Agenda for Applied Economics

My objective in this project is to set out a more detailed strategy for applied research in economics, which takes account of the issues raised in my book, Putting Econometrics in its Place, and the generic and widespread Signal-to-Noise Problem that I believe afflicts most econometric work.

This project is work-in-progress, but one important principle has already emerged.

While the vast majority of modern applied economic research dives immediately into econometric analysis, my work on the Signal-to-Noise Problem indicates that the wise researcher must first do other work before engaging in any applied econometrics.  

The reasons for this stem from four observations:

a) The signal-to-noise ratio is unacceptably low in most applied studies which estimate complex models with very large numbers of parameters (see the paper Doubtful Significance).

b) By contrast, case studies do not suffer so severely from the signal-to-noise problem – though they do have some other shortcomings (see the paper Clear Case)

c) To increase the signal-to-noise ratio, we need to impose valid restrictions on complex models (Putting Econometrics in its Place, 2006, pp. 52-53, 227-228).

d) There are many other types of economic data and knowledge besides econometrics-ready data, and these other data can be used to impose valid restrictions.  This point was discussed at length in Putting Econometrics in its Place.

A preliminary agenda for applied economics - which is radically different from the way most economists pursue their applied research - could look like this:

1) Start with research techniques that have a high signal-to-noise ratio. The obvious place to start is with some case studies, but in some cases it may be appropriate to use economic history, engineering economics and interviews.

2) Collect all the additional material suggested by soft techniques such as, common sense and intuition, metaphor, everyday life as research technique, journalism and the reported views of ‘Captains of Industry’ (and other economic actors).

3) Use what is learnt from the above to simplify any complex models with very large numbers of parameters (by imposing valid restrictions).

4) Examine existing econometrics-ready data to see if it is adequate.  Then consider what might be added by new data collection (including by surveys and questionaires).

5) Then, and only then, start some applied econometrics.

This agenda puts econometrics 'in its place’ as the last step after some essential (and probably more important) earlier steps.  But econometrics definitely has a place on this agenda.

Comments