Barrons - Gene Epstein rips into the hedonic pricing cynics
(Actually I'm interested to hear others' views on whether he's correct that "chain-weight" factors fix the problem).
for subscribers:
interactive.wsj.com
SEPTEMBER 11, 2000
Why the ''Hedonics'' Debate Is a Lot of
Baloney
By Gene Epstein
My award for the most misunderstood piece of U.S. output goes to
computer hardware and software. Those presumably in the know keep
informing us that a bit of statistical chicanery bizarrely dubbed "hedonics"
has been artificially blowing up the contribution that computers have been
making to the growth of gross domestic product, and that without this
factor, U.S. economic performance suddenly would look tame.
Last Monday, this popular fallacy began to assume the dimensions of a
global us-against-them data war, with the publication of an article in the
Financial Times that came close to blaming hedonics for the euro's
clobbering.
"By a process called 'hedonic' price indexing," wrote FT columnist James
Grant, chairman of grantsinvestor.com (and in a former life, a Barron's
writer), "the fruits of the information technology revolution are made to
appear even plumper, riper and juicier than they actually are ... [and] the
consequences of the real-life application of these adjustments are
significant and far-reaching." For example, the article asserted, "Crediting
published U.S. productivity data, currency traders have bought the dollar
and sold the euro."
But, Grant added, help may be on the way. "The eerie official silence on
hedonic indexing has at last been broken," he wrote. Analysts at the
Bundesbank have taken a "stab" at calculating how their numbers would
look if they goosed them with the same method the U.S. hedonists are so
fond of employing. The Germans' findings: inconclusive and -- as nearly
as I can determine (I haven't read the Bundesbank report itself) --
probably wrong. But, Grant asserts ominously, "Without hedonic
assistance, growth in the second quarter surged 3.6% from a year
earlier."
Well, sorry, but U.S. growth in the second quarter vaulted 6.0% -- and,
without hedonic assistance, at least 5.7%. Similarly, second-quarter
year-on-year productivity growth jumped 5.1% -- and without hedonics,
about 4.7%. Hardly enough to ruffle the feathers of the average currency
trader.
Grant's error places him in relatively good company, because not only is
hedonics almost universally misunderstood, the role it plays in boosting
GDP is also commonly exaggerated. And it all gets back to the hubristic
assumption that the bean-counters in Washington just don't know what
they're doing, poor, underpaid bureaucrats that they are.
Talk to them, however, and you'll find that the contribution that any
particular sector makes to real -- i.e., inflation adjusted -- output is
essentially calculated in two stages. In stage one, you count how much it
grew in real terms -- and in the case of computers, that's where hedonics
come in. But in stage two, before that sector gets included in the total, you
give it a weight that reflects how much aggregate spending is actually
allocated to it.
That technique is called "chain-weighting." It means, in effect, that where
prices are rising rapidly, the weight gets bigger, because more money is
being spent. But when it comes to computers, prices have been declining,
so the weights have been falling in tandem.
The critics have been making the fatal error of performing stage one on
the computer sector, but not stage two. (The Bundesbank economists
probably made the same mistake.) So, for example, they will calculate that
hedonically driven computer output soared 59%, year-over-year, in the
second quarter, and then they'll assume that all of that increase went into
GDP growth. Hence, they'll say, the role played by those gadgets has
been looming way too large.
But they forget stage two. Owing to falling price indexes in this sector
(which, to compound the irony, have been falling all the more because
hedonics are being used), that 59% gets a relatively small weight. In fact,
computer hardware and software together contributed just 0.9% of the six
percentage points of second-quarter GDP growth. At most, hedonics
counted for about a third of that, or 0.3%.
Also, hedonic modeling is not quite the joke the critics make it out to be.
My award for most misunderstood government statistician goes to a
Bureau of Labor Statistics numbers nerd, Michael Holdway, that agency's
key hedonic model-builder. True, the 49-year-old Holdway is not entirely
without blame for his predicament, since he often talks like a government
statistician. But once you let him speak a little bit more, you soon find he
has a much firmer grip on reality than the gaggle of naysayers who
misconstrue the arcane technique he practices.
Hedonics -- defined by Webster's Third International as "a theory of ethics
dealing with or based on the relation of duty to pleasure" (first definition),
or, more helpfully, "a branch of psychology that deals with pleasant and
unpleasant states of consciousness and their relation to organic life" --
springs from the belief that quality improvements make a good cheaper to
buy.
Or as Holdway puts it in a paper he's done on the subject, "A mainstream
desktop computer that sold for $2,200 in 1993 may have included a
33Mhz cpu, 8 MB of DRAM, a 210MB hard drive, a 15 inch monitor and
many other defining technological characteristics." But by 1998, "a
desktop computer that also sold for $2,200 could easily be configured with
a 450Mhz cpu, 128MB of SDRAM, an 8,000MB hard drive and a 17-inch
monitor, and include other advanced features that were unavailable in
1993, such as a DVD player and 3D graphics capabilities."
Now, anyone who was willing to pay $2,200 for that Model T of '93 would
gladly fork out far more for the vastly improved '98. And this must mean
that on a quality adjusted basis, the '98's price had declined substantially.
By tracking the change in prices paid and features offered, the hedonic
models are set up to estimate the rate of that price decline. For example,
every three months, Holdway takes an updated statistical snapshot of
desktop computers sold to businesses.
Now, let's say that, this time around, another 32 megabytes of memory
have been added to many of the new models. Since his data will include
the prices of models sold with anywhere from 16 to 256 megabytes, his
statistics will be able to tease out the value of that improvement to the
user. If, say, the estimate comes to $1 per megabyte, his model tells him
that all future 32-megabyte improvements mean that $32 must be
subtracted from the actual price to get the hedonically adjusted price.
Also, because these models are updated so frequently (quarterly for
desktops, and at least twice a year for all the others), the values assigned
and the features recorded also keep changing. In that regard, Harvard
economist James Medoff, who is quoted in James Grant's article, speaks
for a thousand ill-informed critics by observing that the technique "relies
on the premises that this year's goods and last year's goods can be
described in terms of the same characteristics, and that the characteristics
mean the same thing this year as they did last year."
But that's not at all so. The updates dictate that the observed
characteristics are changing all the time. For example, some of the new
models feature high resolution monitors, DVDs ("digital video discs") and
CDRWs (disc drives that can both read and write compact discs). And as
for features meaning the same thing well, in the early 'Nineties, an
additional megabyte of hard drive capacity received an estimate of more
than $1, but since that was many megabytes ago, the model now assigns
it less than two cents.
Ironically, hedonics has the fault of its virtues. It should have been
performed years ago on other products showing significant quality
improvements, like planes, trains, automobiles and TVs. And since it
wasn't, one wonders whether the technique is making the past decade's
output growth look somewhat better than that of decades gone by. But as
noted, and as the world should remember, the chain weights take care of
it, aided by the hedonic method itself. Government statisticians aren't that
dumb, after all. |
