NuclearPhynance Archive

http://www.bea.gov/newsreleases/national/gdp/2008/pdf/gdp407a.pdf



In the latest US GDP release Q4 GDP growth estimate on a seasonally adjusted at annual rates basis was 0.6% (Table 2).

On a year-over-year basis (table 8) it is 2.5%. I am assuming table 8 numbers are calculated on the raw data. That is a pretty big difference imo.

From what i understand the seasonally adjusted numbers are generated using some kind of ARIMA model. I am not that well up in econometrics so i was wondering if there are any well known problems in using the seasonally adjusted method as an indicator of the underlying trend.

I would say that ARIMA uses seasonal adjustment algorithms and not that ARIMA is the tool.  You only care about the "I" part so AR and MA are orthogonal to this.  I did not look at the link but common ways to deseasonalize things would include the Census X-11 procedure, probably explained in Wikipedia or somewhere.  As for common problems, I believe one example is the monthly retail sales figure that uses seasonal coeffiicients (based on history) that miss the new trend of gift cards that effectively defer December sales into January and February and beyond.  The result would be December sales figures that appear weaker than they really were.

Mr. Ball Lightning,



Ignoring any statistical models or seasonal adjustments at all, why is it a surprise to have a difference there? Those two numbers are two different beasts:



Beast 1: GDP Growth measured from start of Q4 2007 to end of Q4 2007

Beast 2: Percent change between Q4 2007 GDP and Q4 2006



each number in Table 8 deals with two different years while the table 2 just refers to change within a single year. The growth estimate for Q4 2007 could be zero, but if there was large growth in Q3 2007, then the difference between Q4 2007 GDP and 2006 GDP is still going to be large.



I don't see why seasonal adjustments have to do with it -- just the raw numbers themselves will be fundamentally different.

functor: I do realise that these are different measures. My question was related to the definition of GDP growth (given recessionary fears) as calculated by a X11Arima model, which is basically a annualised quarter-over-quarter growth of the seasonally adjusted series.

I was trying to figure out if there were any known issues with this approach. The difference with the simple year-over-year numbers prompted the question.