Thread by @vanishingcorp

It's fun that certain flavors of (financial) economists are absolute zealots when it comes to DiD, identification strategies, instrumental variables...and then use "total revenues" from Compustat to calculate industry concentration. A short 🧵

Wharton did the world a service in creating WRDS to stitch together the unholy mess that was Compustat with CRSP and other data sources. Mostly it gets things right. (I'm pretty sure it correctly considers the original predecessor to JP Morgan Chase to be Manufacturers Hanover.)

But industry classification is...not trivial, esp. when things change a lot over time. Ask a room full of academics what "Westinghouse" is called today, or "Woolworth," or "American Can," and you get blank stares. Why does this matter?

Here are Apple's annual entries for SIC and NAICS, both current and "historical":

Not too bad: Apple went from being a computer manufacturer to a "Radio and television broadcasting and communication equipment" manufacturer in the historical series in 2012. But: observe SIC historical is absent before 1987. Oops.

That means if you carelessly downloaded the data, sorted by industry (SICH) and used "revenues" to calculate industry size and firm market shares, you'd lose Apple. But if you used current SIC (SIC), you'd be classifying Apple as a phone maker back to 1980.

If you were wondering "Yeah, what DID happen to Woolworth?", here's your answer. 1994 onward, it was called "Foot Locker." (Same issue with time series data on industry.)

What about a big messy company like GE? Er, "Not classifiable." What are we to make of the HHI in "Business services, not elsewhere classified"?

You could also ask companies to classify themselves. Here were some recent self-classifications:

And if you enjoyed this rant, you may like this piece about how we are misled by using Depression-era tools and categories to map the economy today.

journals.sagepub.com/doi/10.1177/14…