Tuesday, April 10, 2018

Why Is Macro So Hard? Artificial Precision

I’m speculating here, and in a somewhat nasty way: I suspect that personality tests would show that governments are full of “control freaks”. Control issues are how therapists describe behavior in which people think things perform better because they specifically are the ones in charge.

A symptom of this is the artificial precision in many government statistics. In the U.S. we announce quarterly real GDP growth rates to an accuracy of one tenth of a percentage point. Due to annualization this is actually something a tad sharper than one fortieth of a point.

Yet, my personal opinion is that most people have trouble feeling a GDP growth rate difference of less than a percentage point. So the announcements are ten times sharper than they need to be.

Why do they do that? Most of us have experience with or as parents taking the temperature of a sick child. Doctors usually tell us not to worry (even a little) if the temperature is not above 100º F, and to not worry seriously unless the temperature exceeds 102º F. Yet many parents agonize over the tenths digit on their thermometers. At least parents have a reasonable excuse to be control freaks.

The government is doing this with GDP figures. And those are probably the most precisely measured macroeconomic statistic: others, like the deficit, are far less accurate.

We’re more mature than this. Announcers of weather forecasts get this:

RFD 18-01-11 Weather an an Approximation

They can make point estimates of forecasts that are very precise, but instead they provide us with interval estimates that are reasonably informative: like the high will be in the mid 60’s today.

Why don’t government officials behave the same way?

I think this encourages us to focus too much on unrealistic details. For example the Obama administration (and its critics) agonized over differences between 2.1 and 2.2%, when the real issue was that the economy was growing at 2% rather than 3%.

I work with U.S. real GDP data all the time. A reasonable autoregression shows that with annual data going back to 1929, the 95% confidence interval for growth is –5% to +11%.

Why worry about tenths when the range of what’s possible is so large?

Of course, you could make the argument that the annual data includes the unusual periods of the Great Depression and World War II. Fair enough. If you run the quarterly data from 1947 onwards you still get –1% to +7%.That’s a huge range of possibilities for controllers within the government to encourage people outside the government to fret over.

No comments:

Post a Comment