Price Differentials Across Outlets in CPI Data, 20022007

1
Price Differentials Across Outlets in CPI Data,
2002-2007

• John Greenlees
• Robert McClelland
• May 15, 2008

2
New Outlets Bias

• New Outlets Bias can arise from the failure of
  the CPI to adequately reflect the gains to
  consumers from the appearance of new types of
  product outlets
• These welfare gains can arise from
• Greater convenience (e.g., Internet shopping)
• Greater product variety (e.g., Tuscan
  restaurants)
• Lower prices (e.g., Wal-Mart, Costco)
• This paper focuses only on the lower-price effect

3
Price Effects of New Outlets

• The CPI does not reflect differences in prices
  between products at different sample outlets
• Differences across outlets are implicitly treated
  as entirely reflecting quality differentials
• Currently, the most interest concerns the low
  prices offered at discount department stores like
  Wal?Mart and warehouses and club stores like
  Costco

4
Empirical Studies

• New Outlets Bias is a long-recognized issue
• Hoover and Stotz (1964)
• Reinsdorf (1993)
• White (2000)
• Hausman and Leibtag (2004, 2005)

5
Empirical Approach

• Use CPI Research Database for 2002-2007
• Select relatively, but not completely,
  homogeneous CPI food item categories
• Regress price on item characteristics, with
  dummies for time and for outlet fixed effects
• Estimate changes in average outlet premium or
  discount, and average item quality, over time
• Decompose outlet effects within and across outlet
  categories

6
Key Advantages of Our Approach

• Uses actual CPI microdata
• Uses regression estimation to incorporate
  variations in item characteristics
• Examines outlet differentials in general, not
  just across pre-specified outlet categories
• Analyzes item quality change, not just outlet
  effects
• Compares hedonic to matched-model approach

7
CPI Sample

• 14 Food ELIs corresponding to categories used
  earlier by Reinsdorf and Hausman/Leibtag
• Some more homogeneous than others
• 69 months from January 2002 through September
  2007
• About 8,000 outlets
• About 16,000 quote strings
• About 360,000 price quotes

8
Sample Shares by Outlet Type, 2002-2007
9
Hedonic Regression Models

• We estimate 14 regressions, one for each of our
  item categories
• Dependent variable is ln Pijt, the log-price of
  item i in outlet j in time t.
• RHS variables include item characteristics,
  outlet fixed effects, and dummies for month
• Regressions yield a monthly price index for each
  item category with January 2002100.

10
Price Trends with Outlet Fixed Effects
11
Index log-changes by Item Category, 2002-07
12
Overall Results

• Outlet effect -0.26 percent/year
• Item characteristics effect
• 0.20 percent/year
• Difference between hedonic and matched-model
  index
• -0.25 percent/year

13
Weighted Sample AverageOutlet Effects by Outlet
Category
14
Conclusions (1)

• We find significant new outlet effects
• Averaging -0.26 percent per year
• Lowering prices for 10 of 14 items
• This result is after adjusting for differences in
  item characteristics across outlets
• Remember, some of these effects may be due to
  differences in outlet quality

15
Conclusions (2)

• Most of the outlet effects do not arise from
  growth in the discount department store category
• Warehouse category growth is also important
• About 1/3 of total outlet effect comes from
  changes in average outlet premiums within
  categories

16
Conclusions (3)

• We also identify large effects of changes in item
  characteristics
• Even within our relatively homogeneous item
  categories
• Hedonic model estimates average quality increase
  at 0.20 percent per year
• Differences between hedonic and matched model
  estimates warrant further study of how CPI
  adjusts for quality