A James Alexander post
My last post was on the need to change the 2% inflation target to higher one or, better still, switch to NGDP growth targeting. However, by even talking about inflation I feel it is easy to get sucked into a black hole of nonsense chatter about a concept so hard to practically measure.
Nominal GDP (as measured by the value of output, total income or total expenditure) is the reality. Real GDP is a highly artificial construct. And inflation is also a highly artificial construct, the mere residual between the reality of actual Nominal GDP and the artificial Real GDP. It is necessary to calculate some version of inflation to get from real Nominal GDP to artificial Real GDP.
To create a Real GDP figure the statisticians are meant to collect a huge number of price indices for each product, and then deflate the real or actual nominal value of output (ie sales) to derive a supposedly inflation-free “real” but artificial output figure.
Although they collect all these price indices they don’t create indices for the changing quality of each product. They should do. The price indices are thought of as “inflation”, but that is because of the assumption that the inflationary element of prices changes more quickly than quality or nature of the goods and services change. But how do we know? Has it been tested? Has it even been thought about in any methodical manner. I don’t think so. Occasionally, hedonistic or quality changes are incorporated into the price indices, but in a highly haphazard way. Statisticians do track changes in the basket of goods and services via surveys or by observing actual patterns of expenditure but can’t track changes in the nature of service – like a switch from learning on the job to learning at college, or a switch from spending on alcohol to spending on a gym, vice to virtue.
Of course, any quality or nature indices would create huge debates. But the price indices are largely meaningless without them. How can price inflation be observed without as much monthly effort going in to assessing the quality and nature of the product or service being tested.
Entertainment is the classic example, 15% of the CPI basket. The switches from street entertainers, to theatres, to movie theatres, to black and white television, to colour television, to broadband internet all involved major changes in product quality and very often the fundamental nature of the service. The pure inflation element may be able to be measured from one week to the next, but quality and nature also move ahead rapidly. Transport, another 15%, is the same as walking gave way to horse drawn transport, to railways, to motor vehicles, to aircraft, to not travelling but having people and products brought to you virtually. The cost is not then transport but the cost of the broadband connection. Restaurants and hotels, 10% of the basket, change in quality all the time. Housing provokes similar questions.
I am not denying it is quite hard to compile NGDP as it has one or two theoretical issues itself: the final vs intermediate consumption issue, the issue of how to value self-owned housing or the scale of the informal economy. But RGDP has the exact same issues, plus the massive issue of divining pure inflation from changes in quality and nature.
Paul Krugman likes to throw the “voodoo economics” tag around when non-mainstream economists come up with ideas, but what should be done when mainstream economics has formed a consensus around a very silly idea like inflation targeting.
The 2% target is voodoo upon voodoo
On top of the targeting of inflation, seemingly out of thin air a 2% target was created. It was possibly invented because the long run real growth has often been calculated as around 3%. So a 2% inflation seemed a nice balance. Not more than real growth, but not too close to zero and risking deflation. Not too high as to upset the current bunch of Republicans, the Germans, the famous Japanese housewives, or …? William Dudley of the NY Fed recently gave as a reason that it meant most people in their 30 year working lives would see a doubling of prices. Assuming inflation can be so precisely calculated, so what? Why not no change or quadrupling? What difference can it make?
Why has inflation targeting appeared to have worked?
There is much discussion about the “divine coincidence” that while targeting inflation, central bankers actually targeted the output gap. And during the Great Moderation got monetary policy more or less right. The “output gap” is an even more tricky theoretical concept.
NGDP Targeting is so sensible, so simple. It does not rely on any theoretical concepts to target like inflation, RGDP or another voodoo upon voodoo concept like the gap between artificially-created RGDP and where the artificial RGDP should be, theoretically speaking.
Some have suggested that central bankers were implicitly targeting NGDP growth. Well, maybe. If they were, it came very unstuck in 2007-08 when they seemed blinded by high headline inflation, and were very slow to react to falling actual NGDP growth and crashing NGDP growth expectations.
The process the Bureau of Labor Statistics uses to deflate industry outputs goes some distance in fulfilling your needs. Going back many decades the BLS has attempted to estimate the portion of the annual price increase in autos that can be allocated to an increase in quality. It does not include that amount in the inflation measure. It has carried this process to a number of other industries. If one looks at the tech industries there are a number of deflators less than one reflecting the fact that the quality increases in the products exceed the price increases.
What this means is that the BLS by its deflation process increases the real value of GDP on the output side to a extent nowhere explicitly shown and is no made to the input side. This is obvious when one considers Moore’s Law concerning integrated circuits where the number of transistors doubles every two years thus increasing the real value of the integrated circuit with the necessity of doubling the inputs. To which factors of production should this increase in real value be allocated. The BLS can easily calculate the value of these quality adjustments by industry and publish them with the deflators.
The BLS has an Office of Occupational Statistics and Employment Projections that creates annual Input/Output tables in nominal and real dollars. That are forced to scale the real tables to the appropriate real dollar value of output since there is no explicit means of allocating the quality adjustments.
Ironically the National Accounts faced a similar problem on how to account for the increase in the value of inventories resulting from price changes year over year. They created a dummy industry, Inventory Valuation Adjustment, that explicitly shows the increase and just adds it directly to GDP. The quality change should be handled the same way. A dummy industry, Quality Change In Output, that is explicitly shown and added to the value of each industry.
Art, thanks for the comment. I hoped someone as well informed, and interested, as you would reply. There is hope!
A small point first, it is the BEA within the Department of Commerce that produces the GDP deflators, while the BLS within the Department of Labor that produces the CPI. I also suspect a slight “turf war” between the two, which probably doesn’t help our understanding of the issues.
However, the fact that hardly anyone is interested in the topic, and that it is actually quite complex, somewhat proves my point. People, in general, just see the prices changing and think little or nothing at all about quality or nature. The BLS should in a perfect world publish a monthly “quality and nature index”. It really is as important as that, forcing people in general, and the CPI hawks in particular, to appreciate the size of the error bars in calculating “inflation” if nothing else. The BEA should also produce a separate “quality and nature index” for their series of GDP deflators like the PCEPI.
I am happy to acknowledge that the US is far ahead of the rest of the world in these adjustments, but the BLS still only adjusts less than 3% of the CPI basket, ex-housing, hedonically, and don’t even include automobiles:
This UK study showed very little hedonic adjustment activity, internationally, for CPI:
Click to access review-of-hedonic-quality-adjustment-in-uk-consumer-price-statistics-and-internationally.pdf
This 10 year old discussion also from the UK illustrates well the complexity when it comes the GDP deflator:
Click to access chapter-6—quality-adjustment.pdf
To be fair, this equally old US study of hedonic adjustments to GDP estimated that changes in the quality of computer equipment was less than 0.1% of GDP per annum between 2001 and 2005:
Click to access hedonicGDP.pdf
However, that so little is written about the subject, despite a small flurry of activity a decade ago, makes me think that it is just too difficult, too controversial, rather than that it is too inconsequential. I suspect that the pace of change forced by the internet is just too fast for statisticians to cope with at the moment.
The central bankers, commentators, “financial market types”, and big name macroeconomists chatter endlessly about “inflation” and intensely analyse core vs headline, CPI vs PCEPI, goods versus services, but less than 1% appreciate the challenges.
And then there is not just the change in quality problem, but the change in nature. Things like a switch from travelling to a meeting, to skyping a meeting for “free”, or at least no extra marginal cost to the internet access; medical breakthroughs, in treatment or just in more healthy behaviour; educational change; or, more generally, the fact that services have replaced goods as the majority of expenditure/income generation/output. Hedonics has validity, but is still very hard to apply to goods, so when it comes to service quality hedonics must really struggle.
The thousands of microscopic price changes, mostly to goods prices, monitored by the statisticians get swamped by these broader challenges.
Of course, it makes concepts such as the “output gap” even more ridiculously unmeasurable, but that is another post.
Just target cheap and easy-to-measure stable growth in NGDP and forget about “inflation”.
(Apologies for the length!)
James: Thank you for your prompt and knowledgeable response. My problem is that the BEA uses BLS industry output deflators to create the GDP deflators. BEA deflates the bills of goods of each Final Demand component with these industry deflators.
As I mentioned previously the BLS increases the real value of each industry output by the dollar amount that the BLS considers a quality increase rather than a price increase. Again, this increases only the real output side of the National Accounts but does not have a balancing increase on the real input side. BEA carries incorporates this error in its deflation of GDP.
BEA has created a dummy industry, Inventory Valuation Adjustment, for the minor case of the price increase in inventories held for a year in order to preserve the sanctity of the real side of the Accounts. I am sure the total value of all the imputed quality increases more than swamps that of IVA. In the creation of the BLS industry deflators a nominal value of the estimated quality increase could be shown. It would also give some insight into certain discrepancies that appear in the use of productivity and labor income relationships.
Thanks for that helpful response. Sounds a bit of a mess. Some productivity increases might occur to the value of government output of the BLS were merged with the BEA.
If I understand you correctly, there are some modest (but undisclosed) quality adjustments to final output deflators, but no quality adjustments to the intermediate inputs deducted from total sales of a particular industry. As a result the intermediate input are overstated and then deduction is too high, thus underestimating the final output figure? This error would mean underestimating Real GDP and leading to an artificially higher implied GDP, thus overestimating “inflation”.
If this is the case it would add some power to the argument of using total output. But nominal GDP makes no use of deflators at all so should still be the best thing to target.
Excellent blog and commentary. Add on: what is the value of cleaner air and water, which can show up in the cost of production but there is no mechanism for capturing the added value.
Lesson: inflation is just not that important.
James: My understanding is that BEA applies the BLS industry output deflator to the specific industry purchases , i.e, the bills of goods, of each of the individual Final Demand components to derive real GDP. BEA does not publish deflated I/O tables only the BLS does and the BLS applies the same individual industry deflator to the rows which contain both intermediate demand and Final Demand. This leads to massive scaling to compensate for quality adjustments.
I am not questioning the appropriateness of using NGDP targeting at all . I am saying that the way the BLS and BEA incorporate quality changes impacts only the output or demand side of the National Accounts and not the input or income side. This means that measures of industry productivity base on real output have an estimate of quality increases imbedded in them but the measures of real wages do not.
The present handling of quality changes, first, does not show the total value of these changes on the measure of real GDP. And second, no offset is made to the income side and there is no attempt to assign it to the components of National Income. The present handling of quality changes is a mess, if it was extended to incorporate your suggestions it would become even worse.
Art. thanks for the follow up. I am not sure I entirely follow your arguement.
I was just looking at the BLS Real Wages data set and couldn’t believe they deflate them with CPI-U. What nonsense. They should use the relatively more quality-adjusted PCEPI deflator at the very least.
On the input-output issue, my understanding is those tables are only used for calculating GDP via the Output Method: used to value the output of every industry, and thus the country.
GDP via the Expenditure Method, essentially adding final sales to investment, and is the primary method used in the US, as I understand things.
GDP via the Income Method essentially adds up labour compensation and gross profits, plus a few other things.
The two latter methods, GDP(O) and GDP(I), are “adjusted” to match GDP(E). In the UK GDP(I) and GDP(E) are “adjusted” to match GDP(O).
All three methods use price indexes to deflate the nominal figures, either an actual price index or an implied index (eg when nominal sales are used for a nominal data set and “hours worked” or “total cars produced” or “tonnes of coal mined” for the real data set, and an implied deflator is then created).
Sometimes, in the UK at least, you even have a real data set multiplied by a price index, to “inflate” the real data to get an implied nominal figure.
James: The deflation the nominal input/output tables relies on the BLS industry output deflators. These BLS industry deflators are applied to the nominal rows of the Use Table which contains the industry distribution of each industry’s intermediate demand as well as the industry distribution of the industry purchases of the demand components.
My contention is that the industry deflators contain an estimate of the real value of industry quality increases in them since the nominal price has been lowered to reflect the quality increase. By using these deflators to create real GDP the real GDP value has the quality increase embedded in it. However there is no deflator that is used on the income side that has an offsetting reflection of the quality increase.
Take a Use Table with the column for the integrated circuit industry containing the intermediate purchases as well as the value added components. Default each row of this industry by the BLS industry deflator and the value added rows by the income deflators. Get a derived real value for the integrated circuit industry as a sum of these deflated inputs, By Moore’s Law the quality increase in the integrated circuit output will have doubled every two years. Aside from the deflator that has been applied to the diagonal value in the integrated circuit column what other input would show a quality increase of that magnitude? Thus the derived real value of the deflated output will be much less than will the value obtained from the integrated circuit industry deflator. Since the BLS deflator defines the real value of the industry output the inputs in the column must all be doubled to match the target output.
This is why I keep banging on with the comparison with IVA. The inventory measurement in nominal dollars contains an increase due to a price increase. This increase has no commensurate increase in any factor of production and so must be accounted for. The same problem occurs when the BLS incorporates a value for quality in its output deflators that has no commensurate increase in the factors of production. The quality problem can be handled by the BLS specifically showing for each industry the value of the quality adjustment calculated in nominal terms and creating a dummy industry that absorbs the discrepancy between the derived and calculated real industry output value.
Since BEA has not correctly handled this aspect of the deflation process if they extend the procedure to your areas of concern the Accounts will be even more skewed.
Sorry to drone on so long.
Art. A very helpful new analysis included in this report just out [see new post upcoming too]. Brings out lots of the quality adjustment issues very well. The issues with the service sector and other changes in nature are well discussed, not just the narrow issues around IT goods. To be fair, it shows problems with NGDP too, but at least they are not then multiplied with the problems of the deflator.
Click to access 2904936_Bean_Review_Web_Accessible.pdf
Pingback: Mind your P’s and Q’s | The Corner