following instruction finish essay with A+++ work

Today

I Measuring consumer well-being I How bad is inflation? I Are we as well off as past generations? I Will the next generation be as well off as we are?

Cost of Living

I A way to gauge how well off someone is relative to another time or place is by calculating the cost of living

I How much money does a person need to spend today to have as much utility as they had yesterday?

I Abbreviate cost of living as COL I If they have to:

I spend less then they are better off I spend more then they are worse off

Cost of Living

I Economists often define an expenditure function: I

e(p, ū) I The minimum amount to spend to get ū utility

I For two goods: I

minq1,q2 p1q1 + p2q2 I s.t. u(q1, q2) = ū

Cost of Living

I Economists often define an expenditure function: I

e(p, ū) I The minimum amount to spend to get ū utility I

p is a vector (p = (p1, ..., pn) if there are n goods) I For two goods:

I minq1,q2 p1q1 + p2q2

I s.t. u(q1, q2) = ū

Ideal Index

I I will index time periods with super scripts I

p

1 and q1 prices and quantities in time period 1 I

p

2 and q2 prices and quantities in time period 2 I One idea is to look at the ratio of the expenditure functions in

each period I e(p

2, ū) e(p1, ū)

= Index I If Index > 1 then the consumer is worse off (COL went up) I If Index < 1 then the consumer is better off

Ideal Index

I This would be ideal: but we can’t observe people’s utility I Can’t assume their chosen quantities in period 1 and 2 lead to

the same utility I Two simple approaches to try to solve this problem:

I Laspeyres index: use the first periods chosen quantities in both periods

I Paasche index: use the second periods chosen quantities in both periods

Ideal Index

I Since utility is a function of the quantities chosen we guarantee ū equal in both periods

I Ignore consumer’s ability to substitute across goods I Can achieve same ū with different goods when prices change

I Therefore these indexes can never truly measure the cost of living only approximate it

Laspeyres Index

I This index compares the cost of the consumption bundle in the first period with its cost in the second period

I For two goods: L = p 2 1q

1 1+p

2 2q

1 2

p11q 1 1+p

1 2q

1 2

I This index will be an over-estimate of the ideal index I Sometimes the index will appear to rise when the cost of the

old choice of consumption q1 goes up I But consumers will have substituted away from q1

Paasche Index

I This index compares the cost of the consumption bundle in the first period with its cost in the second period

I For two goods: P = p 2 1q

2 1+p

2 2q

2 2

p11q 2 1+p

1 2q

2 2

I This index will be an under-estimate of the ideal index I Buying the new bundle at the old prices may not have been

the lowest cost way to achieve the same utility I Therefore the index does not rise as much as it should

The CPI

I The BLS publishes the Consumer price Index (CPI) I The CPI is based on a Laspeyres index (L-index) for practical

reasons I Main practical reason is that price quotes have been faster to

obtain the information on changing consumption patterns I If you don’t know the current consumption pattern Paasche is

not possible I The CPI reference period consumption bundle (basket) is

derived from the CEX I Consumer expenditure survey I The 2016 CPI uses the CEX from 2013-2014

I Since the basket is updated every few years it is not a pure L-index

Superlative Indexes

I In 1999 the BLS recognized that the CPI was usually interpreted as a COL index

I But it was systematically overestimating the COL I Boskin Commission published estimates of bias in 1996 I A period of reform and discussion continued until 2003

I Therefore it has moved toward a superlative index I Superlative indexes can be thought of as weighted averages of

L and P indexes I The BLS uses the Fisher Index which is the geometric mean of

the two indexes I F-index is the ideal index for specific consumer preferences

Superlative Indexes

Types of Bias

I Substitution I This is what we have discussed previously

I Outlet I Quality and new goods I Excluded goods and services

I Environmental quality, crime levels

Types of Bias

I All three of the types of indexes discussed will have these biases

I Superlative, Laspreyes, and Paasche I The consensus is not to try to fix bias from excluded goods

I Deaton (1998) suggests a holistic approach to well being: CPI, crime, mortality, ...

I Estimating the value that individuals receive from these goods is difficult

I They are not purchased at a price and quantity

Substitution Bias

I Substitution bias I Consumers shift their purchases to reflect new prices I This source of bias is the most obvious problem with the

existing indexes I It can be alleviated using Superlative indexes

I Just guesses between I upper bound (L-index) I lower bound (P-index)

I Only a complete fix for certain consumer preferences

Substitution Bias

I Could be fixed: I If q⇤(p1,p2) the demand function can be estimated I Then e(p, ū) = p1q⇤(p1,p2) + p2q⇤(p1,p2)

I Utility maximization implies expenditure minimization I (Just like profit maximization implies cost minimization)

I Diewert 1998 estimates this bias to be about 0.5%

Substitution Bias

Outlet Bias

I Outlet bias is related to substitution bias I Instead of substituting to new goods consumers may change

where they buy goods I Think of the outlet-good pair as a single new good with a

specific price I The entry of discount retailers like Wal-Mart means that

consumers can shift purchases to lower prices I Estimates show that this bias was small prior to the 1960s and

grew during the 1980s I Its likely to have gotten smaller again as the market share of

discount retailers has risen dramatically I The BLS does not currently weight the prices by the outlet

type I Can be fixed by estimating demand for outlet-good pairs I Diewert 1998 estimates this bias to be about 0.4%

Quality and New Goods Bias

I The introduction of a new good brings about significant increases in consumer welfare

I Usually 10s to 100s of millions of dollars I Cell phones provided $50 billion in 1994 and $111 billion in

1999 I New goods rotate slowly into the CPS

I It takes time to adjust the sampling procedures I Goods with small market share are difficult to get good

consumer survey data on

Quality and New Goods Bias

I Even after inclusion the gains to consumers from the new goods are overlooked

I The CPI will only count decreases in prices after the good enters the sample

I This creates dramatic issues in healthcare technologies I Statins are a powerful drug developed in 1994 I They reduce cholesterol (# 35%) and prevent heart attacks

(# 45%) I The decreases in price after patents ran out were counted by

the BLS I But no accounting for the extended life expectancy and reduce

hospital costs I Estimated benefits would have been 3% of GDP in 2008

Quality and New Goods Bias

I Quality improvements generate many of the same issues I Prices may rise, but the price to quality ratio may improve I This makes it easier to achieve the same or better utility I But using past consumption levels, expenditures will appear to

rise with no offset

Quality and New Goods Bias

I Again solving these issues requires estimating the demand curve

I Can calculate the sizes of the consumer surplus gains I Solve estimated demand curve for the price where quantity is 0

I This area is referred to as the compensating variation I Amount needed to transfer to consumers before the product

introduction I ...To make them as well off before as after I In other words to make utility constant across the two periods

Quality and New Goods Bias

I Diewert I estimates new goods bias to be about 0.51% I estimates quality bias to be about 0.40%

I Hausman provides some suggestive evidence that the bias may be larger

I For consumer durables that make up 80% of consumer durables spending bias was 2.2%

I (likely one of the worst offending type of goods) I demand estimation relies on assumptions that make result an

overestimate

Quality and New Goods Bias

Types of Bias

I The only way to estimate the ideal index is to estimate demand

I essentially requires making assumptions about or estimating the utility function

I allows you to solve directly for expenditures that maintain the same utility level

I i.e. know substitution patterns and value of quality I Estimating demand functions is common in empirical work

I Not been used on the scale of official statistics I BLS would need better data and more high skilled workers I Becoming more feasible as aggregate and individual level

scanner/purchase data becomes available

Measurement Issues

I Purchase data only solves part of the problem I Services are increasingly important in the economy

I But demand for services may depend on too many unobserved qualities

I This makes it difficult to measure how much demand has shifted out in response to quality

Measurement Issues

I Triplett and Bosworth (2004) find that productivity in healthcare fell from 1987 - 2001

I Ratio of total output to total hours worked I Output is measured by the number of treatments applied I Cataract surgery switched from a week long in-patient

procedure to an out-patient procedure I Output appears to fall: hospitals are no longer using the same

number of treatments per cataract I many treatments like pain meds, drugs, and services from

nurses I to fewer medications and any nurse services are now home

production I But actually the healthcare sector probably can treat more

cataracts per unit time

Measurement Issues

I It is important to clearly define services I Is healthcare service the provision of medical goods or curing

diseases? I In this case the problem with the official measurement is clear I In general the relationship between the treatments provided

and the alleviation of disease/symptoms is complex I Understanding these relationships completely requires a lot of

domain knowledge and scientific testing I Hard to account for it in government statistics

I There also may be unobservable components to the quality of a service or good

I E.g. physicians with good bedside manner may provide better care

Heterogeneity

I The biases discussed above I Compare some COL index to an ideal index defined by 1

expenditure function I i.e. we looked at CS under the market or aggregate demand

curve I But demand curves vary from individual to individual

I This means that the bias in the CPI may be higher or lower for different individuals

Heterogeneity

I Individuals face their own rates of inflation and COL because I Geography

I Types of stores available I Competitiveness of local markets I Cost of transportation

I Taste and habit I Some people like expensive things I inelastic demand for goods that have gotten relatively

expensive

Heterogeneity

I Kaplan and Schulhofer-Wohl (2016) attempt to address this issue

I They estimate individual level COL indexes I Ultimately interested in the distribution of COL indexes

I What is the range of COL indexes?

Scanner data

I Kaplan and Schulhofer-Wohl (2016) have the following data: I Kilts-Nielsen Consumer Panel I Prices, quantities at UPC level for 50,000 U.S. households

from 2004-2013 I 500 million transactions

I Based only on goods sold at retail outlets I Participants scan items that they purchase using a barcode

scanner to generate the data set

Scanner data

I Kaplan and Schulhofer-Wohl (2016) have the following data: I Kilts-Nielsen Consumer Panel I Prices, quantities at UPC level for 50,000 U.S. households

from 2004-2013 I 500 million transactions

I Based only on goods sold at retail outlets I Participants scan items that they purchase using a barcode

scanner to generate the data set

Four Indexes

I Construct annual indexes (Laspreyes) comparing Q1 with Q4 I Does variation reflect differences in the goods purchased or

differences in prices across households? I Use the price the household paid for the same good in each

quarter I Use the average price for that barcode

I Use the CPI prices instead, how much variation is due to what people buy compared to the standard baskets

I Use aggregate CPI for retail goods as the price I Use underlying CPIs for categories of goods (i.e. match all

canned fruit to the canned fruit CPI)

Distribution of COL

I The distribution of COL indexes has a high variance I The 75 percentile household had 6.7% points more inflation

than the 25th I The 90th had 14.8% points more inflation than the 10th I i.e. if the rate of inflation for the 25th percentile was 1% then

the 75th had 7.7%

Distribution of COL

I The variance decreases dramatically when prices are set to the strata level prices

I Variation in household indexes seems to come from paying different prices within strata

I Variance does not change as much when going from strata level to agg. CPI

I Household variance not from differences in the share of expenditures in each strata

I A significant amount of variance goes away when prices are set to the barcode average

I Differences in prices for the same barcode are important

Distribution of COL

Distribution of COL

Distribution of COL

Distribution of COL

Distribution of COL

I The breakdown: I 66.9% of variance in household level COL comes from

differences in prices at barcode level I 30.6% from differences in barcode choices within item strata I 2.5% from differences in shares of consumption across strata

Distribution of COL

Distribution of COL

I Paasche v. Laspeyres I Recall that the L-Index > P-Index I The L-index understates peoples ability to substitute away and

the P-index overstates (relative to fisher) I If the P-Index > L-Index that indicates that people are unable

or do not substitute toward lower priced goods I This happens for 40% of households in the period from

2004-2005

Distribution of COL