Brent R. Moulton
Associate Director for National Income, Expenditure and Wealth Accounts
Bureau of Economic Analysis, U.S. Department of Commerce

July 21, 1999

I recognize that our topic this afternoon is the next twenty years, but first I would like to reflect briefly on the remarkable advances that have been made in price measurement during the last ten years.

Ten years ago, I think it's fair to say that the drop-off in inflation during the 1980s had led to an environment in which relatively little attention was being given to price measurement issues, either by the policy makers or by the academic community. Some improvements were being made - note the adoption of the hedonic-adjusted computer price indexes by BEA in 1986, BLS's multifactor productivity program, and research on superlative indexes. But I also recall an environment in which Michael Boskin's initiative to upgrade our economic statistics failed to garner the necessary political support.

Let's review some highlights of the last decade:

Robert Gordon's 1990 book, The Measurement of Durable Goods Prices, systematically critiqued official price indexes and estimated alternative indexes.

The Conference on Research in Income and Wealth (CRIW) held several important conferences on price-related topics, later published as Output Measurement in the Service Sector, Price Measurements and Their Uses, and The Economics of New Goods.

Marshall Reinsdorf's seminal paper on outlet substitution pointed to measurement problems that previously had gone unnoticed and ultimately led to an entire research agenda on the intricacies of price index estimation.

BLS published a series of papers in the December 1993 Monthly Labor Review, thereby beginning to develop estimates of the size of various CPI biases.

Zvi Griliches, in his AEA presidential address, issued a call to the profession to give more attention to economic data. With several co-authors, he studied pharmaceutical prices and price indexes.

The BLS producer price index program began aggressively expanding its coverage of services.

BEA switched to chain-type quantity and price measures using a Fisher formula for measuring changes in real GDP and other aggregates.

The Senate Finance Committee held a series of hearings on CPI bias, then appointed an advisory commission chaired by Michael Boskin. The final report of the commission educated the economics profession as well as the public and resulted in unprecedented attention being given to economic measurement.

The BLS took a number of steps to address CPI biases, including formula bias, adoption of the geometric mean for lower level estimation, more frequent weight updates, expanded use of hedonics, and development of new historically consistent and superlative-type indexes.

At the end of this decade, I see price and quantity measurement as having moved back to center stage in the profession's research agenda. Rather than being seen as an arcane area for specialists and government agencies, the importance of measurement has now been imprinted on the consciousness of the economics profession.

Turning to the next twenty years (or perhaps more modestly, the next five to ten years), what do I see as the research agenda? My answer to that question is probably somewhat different now, after spending the last year and a half working on the national accounts, than it would have been before.

The biases addressed in the Boskin report obviously remain central to our research agenda. While considerable progress has been made toward dealing with upper- and lower-level substitution bias, the problems of quality, new-goods, and outlet-substitution bias remain as difficult, unsolved problems.

I'd like to focus on some specific areas for which improvement is needed:

With the release of the comprehensive revision of the NIPA's (National Income and Product Accounts) scheduled for this October, software will be treated as fixed investment. Along with this change, BEA plans to introduce software price indexes that are, in part, quality adjusted using hedonic methods. We find that the hedonic quality adjustment is important-the price indexes based on hedonic quality adjustment tend to show significant price declines, whereas matched-model indexes (such as the software PPI) tend to show little price change. Software investment will be large as a share of total fixed investment, and it will be important to take account of quality improvements.

The service sector is a perennial problem. Better measures of price and output of bank services, insurance, and other hard-to-measure services are very high on our agenda. Improving our measures of these sectors will require cooperation among the statistical agencies as well as between the statistical agencies and the academic community.

There is a lack of good price data for construction, especially non-residential construction. The recent Gullickson-Harper study of "problem" industries, as identified by long-run negative trends in multifactor productivity, drew attention to construction, as well as banking and insurance.

Measurement of medical care services has received a lot more attention recently, and was the topic of a recent CRIW conference. BLS is to be congratulated for developing PPI's that track a course of treatment instead of inputs. Nevertheless, much work remains in the area of quality adjustment.

Technology change in capital goods other than computers has not received sufficient attention. Communications equipment, instruments, photocopy and related equipment, and machine tools are examples of capital equipment that have recently had large improvements in quality that may not fully be reflected in the PPI's. I would particularly highlight telecommunications equipment, which is rapidly growing and if correctly quality adjusted, might rival computers in importance.

Besides pointing to industries and commodities that need to receive more attention, I'd also like to make a couple of modest suggestions on specific research topics that may have a large payoff:

National accountants make heavy use of the idea of benchmarking-that is, finding data and methodologies that ensure the internal consistency of the data and guard against data series permanently drifting off course. Price indexes, however, are only rarely benchmarked. One way of interpreting Marshall Reinsdorf's influential 1993 outlet-substitution paper is that he came up with a way of benchmarking certain CPI series by comparing trends in price indexes to trends in average prices for goods for which quality change was relatively unimportant. My own opinion is that this approach could, and should, be applied more widely. BLS could produce a number of additional average price series for items like rent and certain apparel items. Even in cases for which quality change is going on, it would be useful to know more about the relationship between average prices and price indexes.

In constructing price indexes, I hypothesize that there may be an important interaction between quality adjustment and sample replacement. The sample replacement strategy of linking together the old and new samples is implicitly based on the assumption that the two samples are both representative at the time of replacement-an assumption that, if true, would appear to imply that sample replacement is unnecessary. If the average quality-adjusted price of an out-of-date sample is systematically different from the current population quality-adjusted price, then it would be appropriate to make a quality adjustment as part of the sample replacement procedure. A paper presented at last year's meeting of the Ottawa group by Tim LaFleur, Karin Moses, and myself included some empirical evidence in the case of televisions that important quality improvements may not be fully accounted for because quality changes are "linked out" during periodic sample replacements.

I recommend additional research on disappearing varieties and the treatment of their replacements in price index construction. Although I think the paper that Karin Moses and I wrote for the 1997 Brookings Papers probably was probably misinterpreted in the context of the debate that was then going on about the Boskin report, it does present compelling evidence that the replacement of old versions of an item with new versions is critical to measuring changes in both price and quality.

Finally, I'll mention computer prices. While I think this is one of the success stories for the U.S. statistical system, it is also important to recognize that real growth in computers is now one of the most important contributors to GDP growth. In view of its importance, I think it would be very worthwhile to carefully critique and monitor the hedonics-based indexes to ensure that the quality adjustments result in indexes that neither understate nor overstate the dramatic price declines that have occurred.


National Bureau of Economic Research, 1050 Massachusetts Ave., Cambridge, MA 02138; 617-868-3900; email:

Contact Us