A longtime friend of this blog wrote in this morning with some useful commentary on the latest GDP figures coming out of the ELSTAT. Due to a professional interest in statistics that are not completely effing bonkers, said reader is understandably upset:
So what prompted this exchange, I hear you ask?
Well, as veteran readers will know, ELSTAT has given up trying to provide seasonal adjustments on the GDP series for now because of, oh, only a major series break round about the most critical time in the whole of Greek economic history since World War II. As a result, no one has any reliable figures on Greek GDP when we kind of need them, and all we have to go on is seasonally unadjusted figures on what is easily the most seasonal economy in the Eurozone. Seriously, just check out the figures. They look like a long march of boobs, which, from a statistical point of view is what this whole sorry affair is.
Prompted by my dear reader's outburst, I set to work on a little experiment to simulate a seasonally adjusted series for Greek GDP. I took all of the available data from 2006 onwards and estimated seasonal adjustments for each quarter, controlling for the series break. I did this for the nominal GDP figures plus the GDP deflator series. In order to avoid having my regressions contaminated by the fact that the post-break series is generally full of quarters with negative growth, I also added one dummy for the post-2009-election period and another dummy for the post-Memorandum period. The effects are small but since recessions tend to put deflationary pressure on the economy it's probably a good precaution; without it I would end up with a series that says that essentially the Greek economy has been stagnant, as opposed to sliding off a cliff helplessly like the Jamaican national bobsleigh team. Trust me, I checked.
The result is the following graph, pegged to Q1 of each year. Because I've used Q1 as the reference, the seasonally adjusted figures tend to be much lower than the unadjusted ones. The choice of reference quarter is irrelevant: one is simply choosing which quarter dummy to eliminate from the regression analysis as they can't all be included. The results (such as adjusted quarterly growth rates) should be the same regardless of which quarter is chosen as a reference.
Anyway it's best to caveat this sort of experimental McGuiver stuff so that people won't accuse me of being a complete amateur.
These figures suggest that the slight growth spurt we supposedly experienced in Q1 2011 was actually zero. That I suspected. But the scary part is when we get to Q2, because my series suggests that GDP shrank by 2.2%. Not annualised, mind you - 2.2% q on q. Only Q2 and Q3 2010 have been worse so far, when the economy shrank by 3.4% and 3.1% respectively. Again, according to my calculations. That's twice the fall reported by ELSTAT for those quarters, when the statistics agency still saw fit to publish the same adjusted real GDP figures they now say they cannot publish. Aaaah, savour the sound of banjos and running water everyone.
Alternatively, as is appropriate on the day we Greeks celebrate the Feast of the Assumption,
HOLY MOTHER OF GOD!!!!!
I will stress once again that these figures are neither official nor reliable; just the best I can come up with for now. But they seem very plausible. They even capture, in a way that ELSTAT's figures did not, the drop in GDP ahead of the elections as (reason dictates) all investment decisions were put on hold until the new Government's fiscal policies were announced in Q4 2009. They also make some sense of the exploding unemployment rate reported for May, which one would not expect of an economy performing a dead cat bounce. That and the out-of-control primary deficit figures for Q2.
UPDATE: Three days after this post was first published, the Greek Government revised its range of gdp growth estimates to include up to -4.5%. Then it revised it again to include -5.3 as the minimum growth rate. I had previously gone on the record suggesting 4.2% was plausible.
PS.
Some readers have asked me to share my workings so that others can decide whether they find the 'seasonal series' convincing. Well here you go. I have no secrets.
"It's got to be the worst data set from any european stats agency: they can't even give you a q/q gdp figure!"
So what prompted this exchange, I hear you ask?
Well, as veteran readers will know, ELSTAT has given up trying to provide seasonal adjustments on the GDP series for now because of, oh, only a major series break round about the most critical time in the whole of Greek economic history since World War II. As a result, no one has any reliable figures on Greek GDP when we kind of need them, and all we have to go on is seasonally unadjusted figures on what is easily the most seasonal economy in the Eurozone. Seriously, just check out the figures. They look like a long march of boobs, which, from a statistical point of view is what this whole sorry affair is.
Prompted by my dear reader's outburst, I set to work on a little experiment to simulate a seasonally adjusted series for Greek GDP. I took all of the available data from 2006 onwards and estimated seasonal adjustments for each quarter, controlling for the series break. I did this for the nominal GDP figures plus the GDP deflator series. In order to avoid having my regressions contaminated by the fact that the post-break series is generally full of quarters with negative growth, I also added one dummy for the post-2009-election period and another dummy for the post-Memorandum period. The effects are small but since recessions tend to put deflationary pressure on the economy it's probably a good precaution; without it I would end up with a series that says that essentially the Greek economy has been stagnant, as opposed to sliding off a cliff helplessly like the Jamaican national bobsleigh team. Trust me, I checked.
The result is the following graph, pegged to Q1 of each year. Because I've used Q1 as the reference, the seasonally adjusted figures tend to be much lower than the unadjusted ones. The choice of reference quarter is irrelevant: one is simply choosing which quarter dummy to eliminate from the regression analysis as they can't all be included. The results (such as adjusted quarterly growth rates) should be the same regardless of which quarter is chosen as a reference.
Anyway it's best to caveat this sort of experimental McGuiver stuff so that people won't accuse me of being a complete amateur.
These figures suggest that the slight growth spurt we supposedly experienced in Q1 2011 was actually zero. That I suspected. But the scary part is when we get to Q2, because my series suggests that GDP shrank by 2.2%. Not annualised, mind you - 2.2% q on q. Only Q2 and Q3 2010 have been worse so far, when the economy shrank by 3.4% and 3.1% respectively. Again, according to my calculations. That's twice the fall reported by ELSTAT for those quarters, when the statistics agency still saw fit to publish the same adjusted real GDP figures they now say they cannot publish. Aaaah, savour the sound of banjos and running water everyone.
Alternatively, as is appropriate on the day we Greeks celebrate the Feast of the Assumption,
HOLY MOTHER OF GOD!!!!!
I will stress once again that these figures are neither official nor reliable; just the best I can come up with for now. But they seem very plausible. They even capture, in a way that ELSTAT's figures did not, the drop in GDP ahead of the elections as (reason dictates) all investment decisions were put on hold until the new Government's fiscal policies were announced in Q4 2009. They also make some sense of the exploding unemployment rate reported for May, which one would not expect of an economy performing a dead cat bounce. That and the out-of-control primary deficit figures for Q2.
UPDATE: Three days after this post was first published, the Greek Government revised its range of gdp growth estimates to include up to -4.5%. Then it revised it again to include -5.3 as the minimum growth rate. I had previously gone on the record suggesting 4.2% was plausible.
PS.
Some readers have asked me to share my workings so that others can decide whether they find the 'seasonal series' convincing. Well here you go. I have no secrets.
Hi there. Intersting stuff, however some more information would help. For example, you could give us the data and the model you fitted.
ReplyDeleteThanks in advance,
Photis Stavropoulos (photis.stavropoulos@gmail.com, @photis_stav)