Have a look at the following picture that comes from a 2012 paper by Gotz and Hecq on forecasting growth. It tells you what the official US growth rate at 3 different dates was estimated to be over time. This means that the start of the thick black line tells you what the statisticians thought in the 4th quarter of 1986 what economic growth in the 3rd quarter of 1986 was (basically 0.7% a quarter). Following the black line you get updates on what they thought growth in that particular quarter in 1986 was. In 1990 they thus thought that growth in the 3rd quarter of 1986 was just 0.2%. In 1993 they thought it was 0.6% and since about 2000 they think the growth was 1% in that same 1986 quarter. Hence depending on when you ask the question, you would either think that the US economy was close to recession or booming in the 3rd quarter of 1986.
The two other lines give you the estimated growth levels from particular other quarters, with equally astonishing changes over time.
Just think what our inability to measure growth accurately means for policy, media, and academia. For policy, it basically means you have to wait 20 years before you get to know whether you are in a recession today or not and hence whether or not you should stimulate the economy or depress it. For media it means almost every article written on current growth levels has a strong chance of being substantially wrong. For academia, it means the thousands of articles written in the 80s and 90s on growth dynamics in the 80s were using very dodgy data of which you cannot really trust the conclusions. It probably means all the ones written now on wobbles in growth the last 10 years face serious limitations.
Think of this the next time you sit in a seminar or a presentation and someone is trying to convince you that due to some miniscule blip in the data, or the coefficient on the AR(3) component of their Structural VAR (the tech equivalent of ‘some blip in the data’), they have found conclusive evidence of a coming recession, the advent of a new growth era, or the effect of admission to the WTO, to take but 3 ‘conclusions’ I have seen people make out of milking data like this.
What does this mean for the regular economist not looking to make a living out of data blips? Well, unless the economic movements are very strong and backed up by ancillary evidence (employment, deficits, trade), don’t put too much faith in the headline GDP figures.
Core Economics 
Surely GDP estimates are derived from samples and hence have some kind of standard sampling error associated with it. Also the percentage uses data from the previous period and the current period and by my reckoning the percentage errors add when you are estimating the growth rate. So you get a number close to zero with an error double that of the things you’ve sampled from.
I guess what I’m asking is, are the changes outside of what is expected from the standard error of the samples, or can the variation be explained by just the expected noise?
LikeLike