Comment on Armknecht, Lane and Stewart
“New Products and the US Consumer Price Index”
Frank C. Wykoff^
The consumer price index (CPI), perhaps the most important single price statistic produced by the government, is used to calculate changes in many indexed public benefits and private contracts. It is a major indi-cator of economic and political performance; some analysts even argue that the CPI alone should guide Federal Reserve Board policy. For these reasons, design and construction of the CPI is important.
Nonetheless, many academic and business economists have shown remarkable neglect of the actual construction of the CPI by the Bureau of Labor Statistics (BLS). An occasional call by a Simon Kuznets or by a Zvi Griliches for “better data” tends to fall on deaf ears among those in the profession. This neglect may have been abetted by some institutional in-ertia, a (perfectly understandable) bureaucratic tendency to avoid clear explanations of actual practice and a possible reluctance to change. But the Bureau does produce a detailed BLS Handbook of Methods that in-cludes a chapter on the CPI, and much effort does go into correcting the various biases.
In fact, collection of raw price data, compilation of various sub-aggregates, and the final design of the CPI are quite complex involving hundreds of individuals and thousands of observations drawn from a large and highly dynamic economic system over time. This measurement proc-ess requires theoretical and practical decisions at various levels by many different people. The CPI does not, and indeed cannot, reflect the design of a simple, fixed rule. The world presents the Bureau with too many va-garies to allow this. Instead, field agents must make pragmatic decisions in implementing their instructions, and the instructions themselves must be simple enough to follow and yet complex enough to allow for the changing nature of the market place.
Paul Armknecht, Walter Lane, and Kenneth Stewart (ALS) have made a major contribution in this paper by explaining, as carefully as they can, the architecture and construction of the index. They explain the con-cepts that drive the equation structure and the pragmatic decisions that drive the sampling procedures. Finally, they point out some practical problems with implementation that result from changes in the nature of goods and in the mode of marketing. Some of these decisions are bound to be controversial; some are not exactly clear, and some are unresolved, but now, thanks to ALS, we know enough to start working on how best to allow, in designing price series, for the types of rapid changes that are taking place in today’s markets, one of the most difficult of which is the introduction of entirely new goods, like video recorders and magnetic resonant imagers.
One can think of the CPI as an economic statistic, intended to summarize changes over time in prices of goods purchased by consum-ers, that is a compromise between two competing sets of influences. On the one hand, economic theory, reflecting which questions economists think the CPI is supposed to answer, drives the architecture. On the other hand, the actual object under study, the economy confronts field agents, who are collecting data, with practical problems that are often difficult to anticipate and resolve - design changes, production changes, model changes, marketing changes, shopping pattern changes, and so forth. The index itself is a product of attempts to reconcile these two influences, one theoretical and the other practical. The CPI also is complex because it is used to give a simple one number description of so much that is going on.
In the BLS view the CPI is designed as a “cost of living index”; that is, an index that measures changes in the cost of living over time relative to a base period for which the index value is set to one. In principle, this means that if the period-t CPI equals 1.33, then the same standard of liv-ing costs a representative consumer 33% more than it did in the base pe-riod. Even within the confining framework of classical, representative-consumer economic theory, there are many different ways to measure changes in the cost of living, so more must be said about the theoretical underpinnings of the CPI than that it is a cost of living index.
The CPI is based on a Laspeyres price index model, which is to say that it is intended to be a measure of the period-t cost of a fixed bundle of goods relative to the cost of the exact same bundle in the base period. By construction the CPI index value for period-t is supposed to be a weighted average of the period-t prices of the original bundle in which the weight assigned to each good in the bundle is the period-t quantity of the good consumed divided by the value spent on the good in the base period, which is the product of the base period price and quantity. This formula yields one for the base period. If the original bundle costs 33% more in period-t then the CPI will be 1.33.
Many other cost of living indexes, for instance Paasche, Fisher ideal, Tornqvist could be produced from the similar data, though some collection aspects could be harder, but based on slightly different eco-nomic theories-different perspectives, specific utility functions, different mathematical bases, and so on. Why has BLS chosen the Laspeyres? Two obvious virtues of the Laspeyres formula are its simplicity and its famili-arity. It is easy to explain a measure to compare the price of a fixed mar-ket basket of goods over time, and anyone who has studied a bit of eco-nomics has learned about a Laspeyres index, though perhaps without the title.
The Laspeyres also has the convenient and practical feature that once the weights are determined from the base year(s), only prices need to be collected in order to update the index monthly. Thus its ease of con-struction and its intuitive simplicity, that you simply multiply each good’s price by the relative importance of that good in the basket and that the basket does not change, surely explain its popularity. Were BLS to aban-don the Laspeyres formula, someone else would construct one, and eve-ryone would probably use it instead of whatever newfangled index BLS produced in its place.
Nonetheless, two points need to be made. First, the Laspeyres price index comes from a very static, rigid and limited economic model, and one certainly must question its descriptor qualities of the very complex, dynamic economy it is intended to reflect. Second, the CPI is not, in fact, a Laspeyres index. It only purports to be this for ease of explanation. The actual numbers reflect a more subtle compromise with reality than Mr. Laspeyres formula would suggest. The exact nature of these compromises is what the ALS paper is all about, and it is a very constructive discussion of how BLS solves practical problems not anticipated by those of us who describe Laspeyers indexes to our students.
Briefly, the CPI is a Laspeyres index over 207 strata where a strata is not an actual individual good but a category of goods such as ap-parel, gasoline, cereal. The weight associated with each strata is a con-stant determined by the average quantity of total expenditures devoted to this strata during a three year interval centering at the base year. How-ever, each strata price is itself an aggregate index of the prices of the many goods in that strata. It is not exactly clear to me from the paper how the price of a strata is compiled, however. Field agents do alter their sampling procedures from year to year in compiling the data that lead into these strata prices, thus, sample variations and therefore sampling error do take place within the price of each strata every year. Only in the crudest sense, then, i.e., at the most aggregated level, is the CPI a Laspeyres index. These sampling variations include changes in the geo-graphical regions from which data is actually drawn each year. Of a total of 85 regions, 20% are rotated in or out each year. Again it is not exactly clear what goes on in these rotations.
To illustrate the nature of the difficulties faced by field agents who collect the actual raw data that lead into the price of each strata consider the problems caused by the two facts that the vintage characteristics of products frequently change and the nature of the retail outlets frequently changes. Boutiques, mail order, department stores, discount houses, spe-cialty shops are all different retail modes of transactions. Thus, if we let v index the vintage style or model, o the outlet type and t the date the price of a good is observed, then the difference between what price is reported from the field between period-t and period-t+1 for a given good in any one strata, is:
p(v+1,o+1,t+1) - p(v,o,t)
In period-t+1, a possibly new, in a characteristics sense, product, v+1, is sold by a possibly new outlet type o+1. Just re-sampling this new price can overstate a pure price increase, in the sense of a pricing fixed bundle of goods within a strata, because either v+1, o+1 or both may differ rep-resenting quality improvements. Here BLS has two options. One, they can either observe or estimate, with hedonics or cost estimates, p(v,o,t+1) or p(v+1,o+1,t), which can then be used to strip out the percent of price change attributable to quality change. Estimation of these new-old vin-tages or old-new vintages with hedonic techniques is well known. See for an extensive application to a variety of assets, Gordon (1990) and see Triplett (1990) for an analysis of Hedonics in statistical agencies. For a subtle analysis of difficulties encountered trying to observe new-old vin-tage or old-new vintage assets, see Berndt and Griliches (1993).
If one does not have these prices, either due to time or cost con-straints, then the second option is to compute some average of the prices of stuff in the strata that did not change. This latter procedure in effect amounts to excluding quality improvements embodied in new goods that may hve taken place. This could bias the price index upward by excluding improvements in variety and quality that may have accompanied price in-creases of all products. It could also bias the index downward by exclud-ing the very goods whose prices rose because they were better. To some extent the second option, of leaving out new or imporved goods, is driven by resource limitations at BLS. They simply do not have the resources to allow for every change that occurs. All they can do is try to correct for really big changes once they have clearly been a factor, but not until such evidence is obvious. Unfortunately, as Diewert (1987) has pointed out this will lead to substantial upward bias in price indexes; bias that Triplett (1993) refers to as new introductions bias.
Of course, in some sense, if they are really trying to produce
a Laspeyres index, then perhaps BLS should simply ignore such changes anyway.
If one is trying to price the cost of a fixed bundle of goods pur-chased
by a given consumer with the same utility function in each period, then
perhaps new goods and new outlets should be left out altogether. On the
other hand, though, perhaps economists, in and out of BLS, ought to study
ways we coud abandon this static framework altogether and build an index
that reflects the very dynamic world in which we are trying to price the
cost of what ever it is contemporary consumers are trying to buy compared
to what they would have had to pay for this life style years before. In
other words, perhaps BLS should abandon the fiction that the world is simple
enough to be captured by a Laspeyres formula at all.
It does seem evident to me that the CPI, as it is actually compiled
by BLS, is not in fact a Laspeyres index. As a practical matter, BLS cannot
construct a Laspeyres index because too many changes are occurring over
time for such an index to reflect the economy after just a few years. Thus,
BLS staff are trying to sensibly allow for some of these changes in demo-graphics,
shopping habits, quality change, new goods, and so forth, but all this
work is still imbeded in a Laspeyres framework. Perhaps the only solution
to this conundrum is the ever fashionable 1990’s warning label solution
to bring truth in advertising:
“WARNING: The CPI is not a Laspeyres Index*”
*But we thought you would like to think it is. Have a nice day.
^: I wish to thank Chuck Hulten and Jack Triplett for comments on an
earlier draft. I am responsible, of course, for all errors of commission
or omission.
References:
1. Berndt, Ernst and Zvi Griliches, “Price Indexes for Microcomputers: An Exploratory Study,” Price Measurements and Their Uses, Murray F. Foss, Marilyn E. Manser, and Allan H. Young, NBER Studies in Income and Wealth, Vol. 57, pp. 63-93.
2. Diewert, Erwin, “Index Numbers,” New Palgrave: A Dictionary of Eco-nomics, Vol. 2, (Ed) J. Eatwell, M. Milgate and P. Newman, London, Mac-millan Press, pp. 767-780, 1987.
3. Gordon, Robert J., Durable Goods Prices, 1990.
4. Triplett, Jack E., “Hedonic Methods in Statistical Agency Environ-ments: An Intellectual Biopsy,” Paper 7 in Fifty Years of Economic Meas-urement: The Jubilee of the Conference on Research in Income and Wealth, NBER Studies in Income and Wealth, Vol. 54, pp. 207-233, 1990.
5. ____________, “Comment,” Price Measurements and Their Uses, Murray
F. Foss, Marilyn E. Manser, and Allan H. Young, NBER Studies in Income
and Wealth, Vol. 57, pp.197-206, 1993.