Thoughts on How Economists Can Avoid Reinventing the Wheel¶
In writing research papers, economists solve the same problems over and over again.
- Drawing joint Normal random variables with fixed standard deviations (e.g. agents' endowments).
- Coding common production functions (nested CES).
- Constructing individual histories of earnings / schooling using NLSY / PSID data.
- Consistently constructing measures of schooling / earnings across year in CPS / Census / ACS data.
The list is endless.
Consider what other disciplines do.
People who work on numerical math have access to a set of trusted code libraries (BLAS, NLopt, ...).
Consider the development of the Julia language.
- All the code libraries live on github.
- The major packages are moderated by groups who take responsibility for the code's correctness.
- All code comes with automated tests.
The result: Much of the time when users ask questions on how to solve a problem in
Julia, someone will simply point to an existing library.
Economics needs a similar setup for code and for data.
Instead of creating the same variables over and over again from each of the major datasets, there should be a place where the construction of such variables can be discussed and where code for constructing variables can be posted (and trusted).
The Organizational Problem¶
Why does this not happen?[^QuantEconFn]
Because there is no payoff for individuals to run such an effort.
Can this problem be solved?
As a small first step, I have started to post reusable code I wrote at github.
[^QuantEconFn]: A partial exception is QuantEcon, which is organized by Sargent and Stacchurski. But it is not an open repository.