Transcript Discussion of Nordhaus on Alchemy and the New Economy

Discussion of CNV and MP
on the CPI for Shelter
Robert J. Gordon
Northwestern University and NBER
Brookings Workshop on
Economic Measurement,
May 23, 2003
Tributes


Congrats to the organizers for finding the right
papers on the most important component in the
CPI
Congrats to the authors for adding enormously
to the body of knowledge



About pitfalls in the CPI as they change over time
Using new data sets, esp. the AHS, to provide new
perspective on potential sources of bias
This topic is the BIG BANANA of CPI bias
Apostle of Upward CPI Bias?

1990 The Measurement of Durable Goods Prices

I still love my monthly Consumer Reports and sure miss the
Sears Catalogue

1996 Boskin Commission Report

BUT




Relative prices change all the time
Now: many goods prices falling, but college tuition and
plumbers’ fees keep rising
Just wait for the decline of the dollar to start boosting a wide
variety of imported goods prices
IN SAME WAY

Bias can differ greatly across products, sometimes up,
sometimes down, relative extent of bias changing over time
Bias may change over time




Hulten-Bruegel paradox
Nordhaus’ suggestion
Hulten’s response
Extend back to Bruegel


Resolution?



¼ lb of potatoes vs. those prosperous-looking Dutchmen
Bias in durable goods (and Nordhaus for light) can’t be applied
to necessities (food, apparel, shelter)
OR, bias in necessities changed over time
Boskin Report: only 95-96, bias could have been higher
or lower (or opposite sign?) at any point in the past
Prima Facie Case that the CPI for
Shelter is Biased Downwards

Change over long historical periods



Median Price of Existing Houses




CPI for Shelter 1999/1925 = 5.1
To get that number, see CNV Appendix 2, where on
their base year, CPI (1925) = 34.5 (1940/1925!!)
Historical Statistics, Washington DC, 1925, $7,809
Statistical Abstract, Washington DC, 1999, $176,000
Ratio = 22.5, not 5.1
OK, rent vs. ownership price including land


But CPI for shelter is used to proxy for home
ownership price
Tenants pay for the land, not just the structure
Another Example

Historical Statistics, 1922

Residential Wealth $71.3 billion








Number of housing units 19.5 million (why so few?)
Value per Unit = $3,656
Statistical Abstract for 1999


Structures $51.1 billion
Land $20.2 billion
Net Residential Capital Stock $9,405 billion
Number of units 115 million
Value per Unit = $81,800
Ratio 1999/1925 = 22.1 not 5.1
Could it ALL be quality change?
If the CPI-Shelter is Biased
Downwards, Over What Interval?


CNV makes a convincing case that bias less in
90s than before (MP limited to 89-01 interval)
Ingredients in Change over Time: CNV Key
Point that Rent Increases Occur with Tenant
Turnover






1942: obtain rent from tenants rather than landlords
(WWII rent controls)
Mail survey 1942-63: nonresponse rate
Telephone survey still misses tenant turnover
1978: interview landlord or tenant every 6 months
1988: aging bias correction
1994: recall bias correction
Shocking Sources of Downward
CPI Bias in the Past

The “Case of the Disappearing Tenant”




quality change
Parallel to Apparel Shocker: “The Case of the
Dress that Goes on Sale”



Why Emphasis on 6 month changes?
Why Not Structure Data Collection Around 12 month
change?
I am a landlord: I know the history of both price and
Ride the Sale Price Down
Link Out the New Dress at the Regular Price
Apparel + Housing Resolve the Hulten-Bruegel
Paradox (But don’t overturn Boskin)
Implications of CPI Bias



CNV (and Weston) create presumption that bias was
greater 1942-78
Growth in real consumption and real GDP overstated
Golden age of productivity growth 1940-70 overstated



Less of a productivity growth slowdown after 1970
Low Bias in 1990s implies bigger productivity revival
2% annual bias for 30% of the CPI implies 0.6% annual
upward bias for real consumption, 0.4% for real GDP

These are big numbers in the “battle of the basis points”
CNV Paper




Real Addition to Our Knowledge
Terrific History of CPI Methodology since
1942
Long-period Time Horizon
Micro Data


BLS Micro Data Set
AHS Hedonic Regressions
First, Let’s Correct those Little
Errors



Abstract says “CPI for rent over 1940-77 may
have been overstated by 1.5 percent annually”
Appendix 2 says 1.8 percent not 1.5 percent
Overall Bias Result from Appendix 2




1940-78 -1.80
1978-88 -1.16
1988-2002 -0.03
Absence of bias after 1988 a stunning result,
just as is -1.80 before 1978
What Did I Think Beforehand?



1940-70 Weston’s Thesis
1970-99 Regressions from AHS Summary Data
Slightly Different Periods, but same general
pattern:




1940-75 -3.49
1975-91 -1.34
1991-99 +1.79
Open Questions:


Interpretation of Weston 1940-70
CNV More Reliable 1975-99
My Little Corner of the World

Historical rent index for Evanston IL



Many versions, hedonic, nonhedonic, identical
addresses etc.
Source: newspaper advertisements going back to
1925
Implications: CPI minus Evanston




1925-40
1940-80
1980-90
1990-99
-0.94%
-1.43%
-0.02%
+0.05%
Issue We Shouldn’t Take for
Granted: Aging Bias



CNV Take This at Face Value
Yes, Clearly a Source of Downward CPI Bias in the Past
But What does it Mean?


Location: Inner-city vs. Suburbs
Omitted quality variables from my perspective as an Evanston
landlord:



New high-rises have views
Central air conditioning, new kitchens
But do the older units actually decline in quality?

My coach house says “no”! Renovations not only by owner but by
tenants!

Distinction between maintenance (roof, gutters) vs. improvements
(e.g., circuit board, kitchen appliances, new and better floors)
Where Do the CNV Estimates
Come From?


Model, Apply corrections for new renters,
recall bias, vacancy imputation while using
CPI estimates of aging bias
Credibility test for 1940-77




Their model bias -1.8%
Median gross rent 1940-77 -2.6%
Implies quality change ~+0.8%
This seems very plausible
Closer Look at 1975-87

Growth Rates from their Table 12




CPI minus Median Gross Rent -1.70%
CPI minus their adjusted index -1.78%
CPI minus their hedonic AHS index -2.00%
Median Gross Rent vs. Hedonic




Implies quality decline of -0.3% per year
They don’t comment on this
Is it credible?
Could it be a location phenomenon – south, suburbs?
Puzzle of Repeat Rent Measure


Blank in Cell 1983-85 (Table 12)
Average for 1975-87 omitting 83-85





CPI
CPI
CPI
CPI
minus
minus
minus
minus
AHS Median Gross Rent -1.74
their model -1.92
hedonic AHS -1.86
AHS Repeated Rent -0.20
Why? They don’t know (p. 28)

This is where the MP paper should help, but it doesn’t


Implied quality improvement missed by hedonic index is not
credible
Is the CNV Repeated Rent Index Identical in Principle
to the MP index? (Homework Assignment)
So MP Step In Regarding AHS



Interesting Hypothesis: Bias by
Insufficient Sampling of Top Part of Market
Recall Paul Pieper on Residential Housing:
Omission of 2000+ sq. ft. houses biases
down housing inflation by ~1% per year
Is it something special about top of the
market or unmeasured quality attributes?
MP Paper Disappointing

Sample Period only 1989-2001



They Miss the Big Period of CPI Bias pre-1989
Why Don’t They Exploit the AHS 1975-89?
Why Don’t They Cumulate Their Own Results
and Convert to Annual Rates?

Charts 4 and 5.




Tenant Rent CPI minus their AHS index +0.37
OER CPI minus their AHS index +0.77
Significant Conflict with CNV paper
Their short-run orientation is unconvincing and, in
this crowd, uninteresting. We care about decades,
not two-year intervals
Nitpicks

P. 4: “[the pre-1983 CPI treatment of
housing was] analogous to the CPI’s
current treatment of automobiles and
other consumer durables.


NO! The current CPI is MUCH better than
that!
Poor explanation of what was wrong with CPI
pre-1983.

Wrong weights, applying the current mortgage
rate to all outstanding mortgages
Big Issue of Principle: OER Costs

MP (p. 10) includes




“Cost of Capital” equals value of the property
net of 8 percent transaction costs
Times “The yearly average of the 30-year
fixed mortgage rate minus ½ percentag
point”
Add “real estate taxes, utilities, insurance
costs”
Add depreciation = 1% of property value
What if Mortgage Interest Declines
from 7.0 to 5.2 over Two Years?





Property Value net of 8% = $200,000
Mortgage Component falls from $14,000
to $10,400
Say Taxes, utilities, etc. = $7,000,
increasing at 4% per year
Depreciation = $2,000
Total OER cost declines from $23,000 to
$19,970

Annualized inflation = -7.06%
Why We Don’t Want Mortgage
Rates in the CPI


Didn’t We Settle This in 1983?
These Fed Economists are Monetary Policy
Imperialists



The Fed Controls the CPI!
This Nonsense Plagued the British CPI for
years!
Low Interest Rates Create Capital Gains,
and we don’t want Capital Gains in CPI or
GDP
Dramatic Example of Why We
Don’t Capture OER Inflation with
CPI for Rents

My friend:



Apply this to a $1 million house, of which there
are a LOT


Condo worth $225K
Annual Rental $19,800
Implies monthly rent = $7,333
Where is the rental market for $7,333 per month
houses? It doesn’t exist for good reasons that
home ownership makes big houses very cheap
Two Alternatives for OER

Option #1



Option #2



Take seriously the role of capital gains in making
homeownership cost virtually zero over time
Let housing costs in CPI ride the roller-coaster of the
current mortgage interest rate
Pretend that capital gains don’t exist
Base CPI on rate of inflation in property taxes,
insurance, utilities, and home maintenance
I want our CPI and GDP deflator to be based on
#2 not #1. How about You?