PolySciFi Blog

Thursday, May 26, 2005


Three notes on GDP and forecasts

It's all just statistics about the past, so perhaps it's not that important, but GDP growth for the first quarter was revised today up to 3.5% from 3.1% (to clarify: that's US GDP for our international readers) .

Note 1:
Now one of the things I find interesting is what was being said when the initial estimate was released in April.
"The newest snapshot of the economy is likely to disappoint economists. Before the report's release, they were forecasting a 3.5 percent growth rate for the first quarter."
which means the economists did a pretty good job forecasting.

Note 2:
The next thing I find interesting is that there while there was some reporting that economists had been expecting a revision up to 3.6% (which means this revised forecast is still close enough for me to call "pretty good"), I'm having a hard time finding a note in current articles about the initial forecast of 3.5%. I'm not alleging bias or anything like that; I'm just wondering if I'm the only one with a long term memory.

Note 3:
Then consider this discussion in the corner (which also missed the fact that the new estimate matches the original forecasts):
"Hey Jonah, doesn't it seem like the "revised" figures for the nation's gross domestic product are routinely 10 percent higher than the originally reported figures? If the initial figures are almost always too low, isn't it time for the feds to start changing the way the initial calculations are done? This is getting ridiculous, no?"
This prompted the following email from me: [Update: this email is now posted in the Corner]
Here's an assertion which I don't have the time to check today (I'll see if I can find some stats to either prove or disprove this assertion tonight):
"For an accelerating economy, revisions are generally upwards. For a decelerating economy, revisions are generally downwards."
This makes some abstract sense if you assume a) that GDP estimations are measuring "things" that are coming into and out of existence and b) your best first estimate of the rate of change in the current number of those "things" is the measure from the previous quarter. For instance, consider the creation of companies during an economic cycle.

Now estimating the number of companies in a large economy can be a pretty difficult task. Assuming the change in number of companies is the same as the previous quarter is not a bad first estimate. However, for an accelerating economy, that leads to an undercount in the first estimate as the acceleration has led to a faster growth rate. Similarly during decelerations, that first estimate leads to an overcount.

Now, I'm an engineer, not an economimst, (Dammit Jim!) so the assertion might well be wrong, but it does gives the GDP economists an "out" for being off by 10% for some scenarios.
As I said in the email, I'll try to find a good dataset to analyze my assertion tonight.

Update 2
This report (pdf) has some good information on the assertion. Repeatedly, the report claims that any bias is insignificant, but the data presented alongside belie that point.

For instance, see the following two charts that I've extracted:

(from page 11)

(from page 20)

Takeaway points:
1. Regardless of what the conclusions the report draws, empirically, there appears to be a methodological flaw that initially underestimates the rate of growth in a growing economy by about 10% (I'm assuming about 3-4% annual growth for the page 11 chart. For an error of 0.4 percentage points, that's about 10% ). This 10% error appears to swamp most other effects.

2. The revisions only tend to overestimate growth (i.e., revise down) in a contracting economy (and then only by a little bit).

3. Perhaps I gave the fed economists too much credit in my acceleration explanation - a second order effect. It appears that they can't even account for first order effects (is the economy growing or shrinking?).


This page is powered by Blogger. Isn't yours?