Kitov's warning

This is the most interesting econ topic I have seen in some time.

Ivan Kitov:

In this blog, we have discussed already [here and here] the incompatibility of real GDP data caused by the change in definition of the GDP deflator... in the USA - in 1977... One can see that the CPI inflation rate is approximately equal to the rate of the GDP deflator change multiplied by a factor of 1.22 since 1978...

Kitov adds:

This observation naturally leads to the assumption that real GDP in the United States is biased by the change in definition of the GDP deflator.

Yes. The data called "nominal" is based on actual prices. If you take "nominal GDP" and divide by the GDP Deflator, you get the so-called "real" values. And if you change the Deflator, you also change "real GDP".

The deflator is the price component of nominal GDP, and "real GDP" is the output component. If you make the price component bigger, the output component gets smaller. Oh, but that's not what we want! Making the price component bigger tells everyone that prices went up more than we thought, and output grew less. That's not what we want.

If you make the price component smaller, you're saying prices went up less than we thought, and output grew more. Yeah, that's better. That's what we want.

Guess which way the Deflator changed.


Okay. In Long-term economic growth is linear (10/24/14) Kitov refers to "the change in definition of the GDP deflator... in 1977".

In Is real GDP correct? (12/24/10) he says "All in all, the notion of real GDP is a virtual one and is highly biased by the change in its definition in 1979."

And in Real GDP is NOT correct (10/22/11) he refers to "the change in real GDP definition in 1978."

He's a little iffy on the date.

I don't know Kitov. But I do know that the methods for calculating inflation and unemployment have changed, and I know it is hard to find information on such changes, even on the internet. I remember a change in the calculation of the CPI back around Reagan's time, or maybe Clinton's. But my memory is fuzzy: It could have been the Deflator that changed, and maybe it was before Reagan. So I cannot dismiss Kitov's warning.

I see him saying

There is no direct statement about the reasons of the change in definitions in [Concepts and Methods of the U.S. NIPA], but we might guess that this is likely related to the introduction of a new methodology to evaluate the overall price inflation.

This bothers me. I need something definite. I cannot proceed based on a guess.

Kitov also says

there is no such macroeconomic measurable parameter as real GDP (see Concepts and Methods of the U.S. NIPA for details). There are two actually measured variables: nominal GDP and GDP deflator (price index). Real GDP is estimated using nominal GDP less the change in prices.

Yeah, that's what I thought; and I think it is how things once were. But Nick Rowe put doubt in my brain about how things are now. And there may have been changes in methodology such that Kitov is no longer correct on this point. I wish I could be sure.


In Is real GDP correct?, Kitov links to Concepts and Methods of the U.S. NIPA. The link brings up a list of Methodology Papers. The "Concepts and Methods" item is on the list. Also listed (with no link) is "A Guide to the National Income and Product Accounts of the United States", which, we are told, was replaced by the "concepts and methods" paper.

I remember the Guide. I remember making fun of the filename -- nipaguid.pdf -- for being DOS-compatible.

Under the heading Real Output and Related Measures on page 15 of the nipaguid (or page 16 of 28 in Adobe Reader) we read:

In addition to estimating the current-dollar market value of GDP, BEA estimates “real,” or inflation-adjusted, GDP and its components.

Yeh. But this gets v.e.r.y interesting:

The annual changes in quantities and prices in the NIPAs are calculated using a Fisher formula that incorporates weights from 2 adjacent years. For example, the 2003–04 change in real GDP uses prices for 2003 and 2004 as weights, and the 2003–04 change in prices uses quantities for 2003 and 2004 as weights.

There is a footnote attached to that last sentence, which says:

Because the source data available for most components of GDP are measured in dollars rather than in units, the quantities of most of the detailed components used to calculate percent changes are obtained by deflation. For deflation, quantities are approximated by real values (expressed, at present, with 2000 as the reference year) that are calculated by dividing the current-dollar value of the component by its price index, where the price index uses 2000 as the reference year.

They put it in the sentence twice, but the year they were using as the "reference year" is not the important thing. The important thing is that, for most components of GDP, the quantity numbers are figured by dividing the nominal value by the price index. RGDP is figured from the Deflator, not the other way around. Kitov is right.


Granted, the 28-page nipaguid has been replaced by the 447-page "Concepts and Methods" PDF. Yeah, yeah, nice.

On page 4-17 and 4-18 of the new PDF, they say that the "chain-type" index that they started using in 1996 is better than the older method, except it is "not additive". This seems to mean that if you add up the numbers, you get the wrong answer. Nice!

Get back on topic, Art. From the last paragraph on page 4-18:

For real GDP and its major components, BEA provides tables that present accurate estimates of contributions to growth rates that are based on chain-type quantity indexes rather than on the chained-dollar estimates (see the appendix).

Yeah, they said something like that in the nipaguid, too. I wonder if they still admit that the chain-type quantity indexes are "obtained by deflation".

They do! Page 4-19:

For most NIPA components, estimates of physical quantities are not available. Instead, “real” estimates—that is, estimates that exclude the effects of price change—are derived by “deflating” (dividing) the current-dollar value by appropriate price indexes.

They use chaining, rather than the pre-1996 method, to figure the price indexes. So, okay. But they're still doing all the work in prices, and converting to quantities. So remember: When they say their estimates of Real GDP are "based on chain-type quantity indexes": Yeah sure, but the quantity indexes are derived by “deflating” the current-dollar value by using price indexes.

They tell you again and again that "Real GDP" is based on quantities. But the fact remains that they figure the quantities by working backwards from actual prices and price indexes.

In follow-up comments to those linked above, Nick Rowe focuses intently on the calculations that are used:

I had thought that StatsCan uses the "Chain Laspeyres" index for GDP (see page 32). That's what I described above. But it sounds like they are maybe using Fisher, which is a geometric average of Laspeyres and Paasche.
...
In Laspeyres you first multiply today's vector of quantities by yesterday's vector of prices, to get real GDP today. Then you divide NGDP by RGDP to get the price index. (That's what I had thought they all did, so that RGDP comes before P.)

In Paasche you first multiply today's vector of prices by yesterday's vector of quantities to get a price index. Then you divide NGDP by P to get RGDP. (Which is the opposite).

And Fisher takes a geometric average of those two methods.

Okay. And I can see I'm going to have to put some numbers in Excel and see how those calculations work. (Not today.)

But notice that Nick takes the "vector of quantities" as a given. It is rarely a given: For "most" of the components of GDP, they figure the quantities by working backwards from actual prices and price indexes. And that's according to both the nipaguid and the "Concepts and Methods" PDF.

That's how I read it, but I'll leave a door open: If I have it wrong, let me know.

But don't focus on the smoke and mirrors of Laspeyres and Fisher and Paasche. Go back to the start. Go back before the part where they use "quantity indexes rather than chained-dollar estimates". Go back to where the quantity indexes are "derived by “deflating” the current-dollar value by appropriate price indexes". In most cases, the price indexes come before the quantity indexes.

Nick says:

I'm now starting to think that this is a non-question. Like debating whether the chicken or the egg comes first. You can do the math either way around. You can calculate P first, or you can calculate RGDP first. It doesn't matter.

It matters, Nick, because you can't use numbers you don't have.

I'm not saying we should avoid using quantity information, when we have it. I'm saying we don't usually have it.

And when the quantity information is calculated by working backwards from price information, let's not pretend that it really is quantity information.

So let's don't say the price index is figured by dividing nominal GDP by real GDP when, in most cases, if you trace things back far enough, it is real GDP that is figured by dividing nominal GDP by the price index.

Almost every case, I would bet.


I opened this discussion by observing that this concern of Ivan Kitov's is the most interesting econ thing I've come across in a long time.

But I had reservations about Kitov. I don't know him or his work.

I was concerned that he was "iffy" about his dates. I now understand that the change occurred in the 1990s, and data was changed for what were even then years past, back into the 1970s. So I'm no longer troubled to see that Kitov does not pin down the date precisely. That's the least important issue raised today.

I was concerned because Kitov was guessing about "the introduction of a new methodology", and I could not proceed based on a guess. I resolved this concern by checking the nipaguid and the "Concepts and Methods" PDF. Yes: The introduction of the "chain-type" calculation is the new methodology. I'll have to test that myself to prove to myself that the old methodology slash new methodology difference is the cause of the Deflator discrepancy that Kitov finds. But I'm good for now.

I was concerned that Kitov might be wrong when he said "Real GDP is estimated using nominal GDP less the change in prices." Because Nick Rowe challenged me on exactly that issue some than five years ago, and doubt lingered in my mind ever since. But I have now resolved this concern, barring a new challenge, by looking into the nipaguid and the other PDF.

I was concerned that my weak memory, which supported Kitov's argument, was weak and possibly flawed. I can resolve that concern right now:

Graph #1: The CPI and Five Vintages of GDPDEF, the GDP Deflator

The thin black line that extends all the way to the right is the Consumer Price Index (CPI). This is the one that the GDP Deflator formerly matched, according to Ivan Kitov. The other five lines on the graph all show the GDP Deflator, as it was at different dates in the past.  You can click the graph to see it bigger. Or you can click the text "Graph #1" in the caption below the graph to see the source page at ALFRED.

The blue line that extends all the way to the right is the most recent version of the Deflator. You can get some idea which line is which not only from the color, but also from the date where the line ends. Other than the CPI and this Deflator line, there are four other Deflator lines on the graph. These end in 1991, 1995, 1996, and 2000.

The fat red line that starts around 1960 and ends in 1991 (vintage 1991-12-04). This is the earliest "vintage" of the GDPDEF Deflator data available at the ALFRED site. It runs close to the CPI all the way to the end in 1991.

The bright blue line (vintage 1995-01-27) closely follows the fat red one and the CPI. The line ends in 1995.

The brown line (vintage 1996-01-19) was the first issue of 1996 data. You can see it reaches just a little past the end of the bright blue line. You can also see that it is lower than the 1995-01 vintage data -- but only back to about 1977 or so, apparently. As Kitov pointed out.

Note that the data issued in 1996 is different from the data issued in 1995, all the way back to the 1970s.

Not on the graph, the first-vintage data for the years 1997, 1998, and 1999 all follow the same path as the 1996 data shown.

The green line (vintage 2000-01-28) is the first year where the first vintage of the year has moved downward again, near to the path of the most recent vintage of the Deflator (blue).

In sum, the GDPDEF data follows the CPI until January 1995 (or possibly later that year). From 1996 to 1999 the data follows a "mid way" path, lower than the CPI but higher than the recent path of the Deflator. And since 2000, the data path is low, like the recent Deflator.

So Ivan Kitov is correct in saying the Deflator seems to depart the path of the CPI  in the latter 1970s. However, the change which caused that departure did not occur until 1996 or thereabouts. And then there was another change, in the year 2000.

So here's a question: If the change in the Deflator was caused by the change in methodology that required using "chain-type" calculations, then why do we see two changes, one around 1996 and one around 2000?

Dunno, buddy. Maybe later.


So much for the most interesting econ thing I've seen in a long time.

The most interesting non-econ thing I've seen recently? Stephanie Martini's eyes in Prime Suspect: Tennison.