Quality Change and Price Indexes

1
Quality Change and Price Indexes

• Comments by Ellen Dulberger
• Brookings Workshop on Economic Measurement,
  February 1, 2001
• Second Session
• Hedonic Price Indexes Too fast? Too slow? Or
  Just Right?

2
Agenda

• Insights from IBM/BEA price indexes for computing
  equipment
• Comments on NRC panel's recommendations (Chapter
  4)

3
IBM/BEA Price Indexes

• 4 types of computing equipment (See tables below)
• Processors, DASD, Printers, Displays
• Alternative price indexes
• Matched model, Composite, Characteristics,
  Regression
• Comparisons
• Matched model indexes showed much smaller price
  declines for all types of equipment
• Other 3 showed price declines which were much
  more like each other for all types of equipment

4
Key Insights From IBM/BEA Research

• Difference between matched model and other
  indexes indicated multiple price regimes
• Price changes associated with models for which
  prices are estimated in one period are NOT the
  same as price changes for models existing in both
  periods
• If the price changes were the same, the two
  indexes would not move differently
• Multiple price regimes raise important questions
• Sample representativeness
• Quantities AND Price changes?
• Implications for aggregation?
• Use hedonics to test for multiple price regimes?
• If one price regime in each period, don't need
  hedonics to estimate missing prices. It's okay to
  assume same changes in quality adjusted prices
  for old and new models
• If multiple price regimes, hedonics are needed
  and require special care in using

5
Key Insights From IBM/BEA Research

• Characteristics indexes have a lot of intuitive
  appeal, however
• Inappropriate selection of characteristics
  results in biased characteristics' coefficients
• Depending on the specification, coefficients may
  or may not be characteristics' prices. However,
  they are used to calculate the estimates of the
  prices
• How do we handle multiple price regimes? Keep "on
  the line." Maintain distance from "the line"?
• In the case of computer processors, coefficients
  were biased when technology classes were not
  accounted for
• BTW--characteristics' coefficients can remain
  constant, yet characteristics prices change
• Regression indexes are problematic (I agree)
• Problem of weights
• Specification errors
• Choices in how to handle
• It's very important to have product and
  technology expertise from the beginning!
• I couldn't disagree more with "computers are
  easy"

6
Comments on the NRC Panel's Recommendations

• 4-1 Conduct random experiments
• Why? Hedonics are so hard, focus on problems or
  suspects first
• Better to give high priority to hedonics in areas
  undergoing biggest change
• 4-2 Expand testing of Hedonics beyond price
  adjustment
• OK, but this seems minor in importance relative
  to new products problem
• 4-3 Cautious integration of hedonics
• This recommendation relates to the importance of
  a CPI program which allows for historical
  revisions
• The stability issue is important. Single year
  regressions may be problematic
• (Single year vs. multi-year regressions tested
  for computer processors. Could not reject
  hypothesis that single years were from the same
  population as multi-year)
• 4-4 Avoid dummies
• Can't argue with that!
• Again, main problem is specification error

7
Comments on the NRC Panel's Recommendations

• 4-5 Pursue characteristics price indexes
• Proper specification is hard for many reasons
  including...
• Selection of characteristics
• Possibility of multiple price regimes
• Coefficients must be positive (value to buyers
  and costly to produce)
• Characteristics prices can't go to zero
• 4-6 Study indirect method
• OK, but the really problem is understanding the
  product and the market
• 4-7 Congress should provide incremental resources
• There's a bigger issue here potential to
  restructure program.
• National vs. Local prices
• Scanners free up field agent resources
• Skills transition
• 4-8 Independent advisory panel
• Not likely to work. Need more than outside
  review. Need outside input early

8
Recommendations the NRC panel didn't make....

• Explore the use of hedonics for direct comparison
  of quality of items in different strata
• important to focus on value in use
• related to questions of classification hierarchy
• Explore the use of hedonics to develop user cost
  for durable goods
• Investigate differences associated with frequency
  (monthly vs. quarterly vs. annual)

9
Summary

• Hedonics help us deal with multiple price regimes
  and other problems beyond those discussed in the
  NRC report
• Product and technology expertise is required
  early in the research process