Content uploaded by Ana Aizcorbe
All content in this area was uploaded by Ana Aizcorbe on Dec 14, 2015
Content may be subject to copyright.
Price Deflators for High Technology Goods
and the New Buyer Problem
A B S T R A C T
Some items in a household’s market basket, notably durable goods, are purchased
only occasionally. In contrast, standard price measures implicitly assume that
consumers purchase some amount of every available good in every period. The
occasional purchase of an existing good by a new buyer generates a “new buyer”
problem that is similar to the traditional “new goods” problem generated by the entry of
This paper uses an idea introduced by Fisher and Griliches (1995) and Griliches
and Cockburn (1995) to develop price indexes for goods that are not purchased in each
period. A comparison of the resulting price indexes with those calculated under the
standard assumption suggests that the sharp declines typically exhibited by price indexes
for many high technology goods may be overstated. However, it is impossible to make
any definitive statements about the numerical magnitude of this potential problem. These
preliminary findings simply underscore the importance of further research to study this
problem for high tech goods and to explore the possibility that similar problems may
arise for other durable goods.
* Bureau of Economic Analysis, 1441 L Street, NW, Washington, DC 20005. I have
benefited from comments from Carol Corrado, Jason Cummins, Joe Gagnon, David
Lebow, Marshall Reinsdorf, Jack Triplett, and workshop participants at the Brookings
Institution, the Federal Reserve Board, and the NBER/CRIW Summer Institute Meetings.
This paper was essentially completed while I was employed at the Federal Reserve Board
and visiting the Brookings Institution. The views expressed here are mine alone and do
not represent views of staff members either of these institutions or the Bureau of
Some items in a household’s market basket are purchased only occasionally.
Durable goods—like new vehicles, computers and household appliances—are typically
purchased only every few years. Nondurable goods may also be purchased infrequently
when the purchase involves indivisibilities—toothpaste, laundry detergent—adjustment
costs—goods purchased at far-away outlets—or other types of market imperfections.1
Finally, many services are also purchased only on occasion—legal advice, visits to
physicians, vacations, visits to movies or restaurants.
In contrast, standard price measures implicitly assume that consumers purchase
some amount of every available good in every period. For example, traditional index
number theory assumes that a representative consumer purchases some amount of all the
existing goods in every period.2 Similarly, the model typically used to justify hedonic
analysis (Rosen (1974) ) allows heterogeneous consumers to purchase different quantities
of goods each period, but nonetheless assumes that consumers purchase at least some
amount of all available goods every period.
The effect that entry and exit of goods has on standard price measures–the
traditional “new goods” problem—has been studied extensively.3 In contrast, the effect
of entry and exit of buyers has received relatively little attention. In both cases, the
market prices used in standard measures no longer necessarily reflect consumers’ internal
valuations of goods. For the new goods problem, one solution is to use Hicksian
reservation prices to measure price change experienced by consumers. This approach
was applied by Fisher and Griliches (1995) and Griliches and Cockburn (1995)
(hereafter, FGC) for what one might call the “new buyer” problem in prescription
1 Index number theory typically relies on static optimization problems to describe the consumer problem
while these cases inherently point to a dynamic model for the consumer. For durable goods, the seminal
article is Hall and Jorgenson(1967). For nondurable goods, see, for example, Betancourt and Gauchi
(1992), Feenstra and Shapiro (2000), and Hendel and Nevo(2003).
2 A notable exception is Feenstra(1995) who constructs price indexes that allow the entry and exit of
consumers making discrete choices over products. His approach, however, is to consider specific
functional forms and other cases where the utility functions for individual consumers can be aggregated up
to an aggregate utility function that serves the same role as that of a representative consumer.
3 See “New Goods and New Outlets,” Chapter V of the National Academy’s report titled “At What
Building on their work, this paper considers price indexes for goods that are
purchased intermittently and constructs FGC-type indexes for several high-technology
goods, a class of goods with important macroeconomic implications.
A preliminary look at this problem for these goods suggests that numerical
differences in the FGC and standard indexes can be quite large. Moreover, in most of the
cases considered here, the FGC price indexes show slower price declines than standard
indexes, suggesting that standard indexes may overstate the rate of price declines for
some high tech goods. However, theoretically, the bias can work in either direction and,
as shown below, calculations done using the Prud’homme and Yu(2003) data on
prepackaged software provide one example where the bias goes the other way.
Hobijn(2001), Harper(2003) and Feenstra and Knittel (2004) have explored other
potential sources of bias that might lead constant-quality price measures constructed
using standard methods to overstate the observed price declines for high technology
goods. More generally, Bils and Klenow (2001a, 2001b) empirically assessed the ability
of standard methods to adequately capture quality change for a wide array of consumer
durable goods. As such, this paper contributes to a strand of the literature that is
concerned with the ability of standard measures to provide accurate measures of
technological change and product innovation.4
The paper is organized as follows. After a brief look at the reservation-price
solution to the new-goods problem, the FCG approach is reviewed and extended slightly
to define an FGC-type Fisher index. The key requirement in applying their approach is
that one be able to place bounds on consumers’ reservation prices. Section 3 provides a
formal definition of “reservation prices” in the context of a simple discrete-dynamic
choice model and uses that framework to explore conditions under which reservation
prices are well defined for durable goods like IT products. The fourth section shows that,
under fairly weak assumptions, the downward-sloping price contours typically seen for
high technology goods may be used to bound reservation prices and construct FGC-type
4 Some recent studies in this literature have explored the sensitivity of price indexes for high technology
goods to the particular choice of measurement technique (Aizcorbe, Corrado and Doms(2000)), the
assumption of perfect competition (Aizcorbe(2002) and Hobijn (2002)); others have used growth
accounting frameworks to assess how much of the movements in standard indexes may be directly
attributed to technological change (Aizcorbe, Oliner and Sichel(2003) and Flamm(2003).
indexes. A comparison of the FGC price measures to standard ones suggests the
differences in the two indexes may be nontrivial. Given the potentially important
implications for economic growth and productivity, these preliminary findings call for
further work to pin down the potential importance of this issue.
2. The New Goods Problem and the Role of Reservation Prices
The importance of entry and exit for price measurement has been studied in the
context of the “new goods” and the “new outlets” problems.5 The reservation price
solution to that problem forms the basis for what follows and is reviewed here.
Consider a Fisher index that measures aggregate price change from time t-1 to t
that is stated as the ratio of two weighted averages of price relatives:
Σ i ωi,t-1 (P i,t / P i,t-1 )
(1) FSTDt,t-1 = _________________
Σ i ωi,t (P i,t-1 / P i,t )
where the ωi,t-1‘s are nominal expenditure weights and the (Pi,t/Pi,t-1) are price relatives.
The “new goods” problem is that if a new good is introduced at time t, a price for
that good does not exist for the previous period (t-1) and the price relatives necessary to
form price measures like the Fisher index cannot be formed.6
Chart 1 illustrates this “new goods” problem and the “reservation price” solution
in the context of the downward-sloping price contours that characterize prices for many
high technology goods and perhaps other goods as well.7 Assume that the models in
chart 1 are defined such that quality is constant over the life of the good so that quality
5 See, for example, Fisher and Shell (1972) and Reinsdorf(1993).
6 This is different from the “new goods” problem faced by statistical agencies. There, goods that are being
tracked by the agency are obsoleted by a new good and the disappearance of the old good and appearance
of the new good must be dealt with. See Pakes (2003) for a full discussion of this problem. Here, the FGC
approach provides a “fix’ for a particular type of matched-model indexes. Unlike procedures used by
statistical agencies, these matched-model indexes are typically constructed using near-universe data and
exploit such data to split out quality change from pure price change—see, for example, Grimm (1998),
Aizcorbe, Corrado and Doms (2000), Silver and Heravi (2001), and Abel, Berndt and White (2003))
7 The shape of these profiles for DRAM memory chips are discussed in Flamm(1989) and Irwin and
Klenow(1994) and for microprocessor chips in Aizcorbe (2002).
change occurs only when new goods appear or old goods exit. Here, model 2 replaces
model 1 at time t and the "new goods" problem is that P2,t-1 does not exist.
The goal of price indexes is to separate out changes in prices that result from
changes in quality from those that result from other influences (i.e., "pure price change").
The reservation price solution measures "pure price change" explicitly and quality
change implicitly. Thus, given an estimate for the consumer's reservation price at t-1
(i.e., PR2,t-1) one can measure "pure price change" as the ratio of the price paid for the new
good at time t to the reservation price for the good at t-1 (i.e., P 2,t/PR2,t-1). To see how
this explicit measurement of "pure price change" implicitly measures quality change,
note that had the consumer purchased the old model at t-1, then the difference between
the price he paid (for the old good) and his reservation price (for the new good) at time t-
1 would be a measure of his valuation of the quality improvement in the new good at t-1.8
In practice, estimates for reservation prices are either estimated econometrically
using hedonic techniques—as in the imputation method—or by making the assumption
that the implicit price change for the representative consumer is the same as the actual
price change measured using observed prices—as in matched-model methods. In terms
of chart 1, the hedonic imputation implicitly attributes some part of the gap between the
two market prices at time t to quality change and the rest to pure price change while the
matched-model method assumes that the entire vertical distance is the “market’s
valuation” of the quality improvement in the new good.
Finally, note that, despite the literature's emphasis on the entry of new goods,
exiting goods also create a problem. In measuring prices from time t to t+1, one would,
in principle, need a reservation price for the exiting model.
8It can be shown that, in general, the use of a reservation price to fix the new goods problem implies a
valuation of the quality improvement for the new good that is some function of the gap between the prices
in both periods. Define quality change as the difference between changes in average prices and changes in
a constant-quality price index. In the simple example in chart 1, the change in average prices from t-1 to
t+1 is P2,t+1/P1,t-1. Using a simple geometric mean of price change as our index, a chained price index that
uses a reservation price for the new good would be the product of the (multiplicative) price changes from t
to t+1: ( P2,t+1/ P2,t ) and the average price changes from t-1 to t (i.e, (P1,t/ P1,t-1)½ ( P2,t/ PR
2,t-1)½, where PR
is the reservation price: I GEO
t-1,t+1 = ( P2,t+1/ P2,t )(P1,t/ P1,t-1)( P2,t/ PR
2,t-1). The difference between this index
and the average price change gives a geometric average of quality differences in the two goods at time t-1
and time t, where the quality differences are measured as the gaps between the two price contours:
(P2,t+1/P1,t-1) / [ ( P2,t+1/ P2,t )(P1,t/ P1,t-1)( P2,t/ PR
2,t-1) ] = (PR
2,t-1/P1,t-1)½ (P2,t/P1,t) ½
3. Indexes for Intermittent Purchases
When consumers do not purchase every good in every period, price indexes must
take into account the presence of new and exiting buyers. The FGC papers provide a
framework based on the reservation price approach that may be used to define a Paasche
index that includes new buyers. As is usually the case, the Paasche index is not affected
by the presence of exiting buyers because it uses the current period as its point of
reference: that is, the index values price change using current period expenditures and
those purchases are zero for buyers that exited the market at t-1. To account for exiting
buyers, a straightforward extension of the FGC idea is applied here to the Laspeyres
index and that resulting index is then combined with the FGC Paasche to form a Fisher
3.1 The FGC Paasche Index
FGC expanded our understanding of the new goods problem by considering the
case where a new good is introduced but heterogeneous consumers do not necesarily
purchase the new good at the period of introduction. In their particular application, the
patent for a branded drug expires and a generic drug enters the market. But, the slow
diffusion of knowledge about the generic drug generates staggered switching to the new
good: not everyone switches to the lower-cost generic drug in the period of introduction
even if the two drugs are chemically identical because it takes time for consumers to
realize that the generic is a safe alternative to the higher-priced branded drug.
In this context, FGC make three important contributions that have wider
applicability than the particular case of prescription goods.9 First, they define a Paasche
index that explicitly accounts for the fact that some buyers are new buyers:
(2) ΠΝCt,t-1 = [Σi Σb P i,t X i,t b)] / Σi [(Σb∈C P i,t-1 X i,t b) + (Σb∈N PR,b i,t-1 X i,t b)]
where “NC” denotes that the Paasche index measures price change for new and
9 This paper ignores another important contribution of the FGC work. They also provide a COLI
interpretation to their Paasche Index. However, because there are well-known problems with applying the
COLI interpretation to durable goods (or, more broadly, to goods where the decision to make a purchase is
dynamic in nature), the interpretation is not exploited here.
continuing buyers, summations are over goods, i, and buyers, b; and b∈C and b∈Ν
denote continuing and new buyers, respectively. As with any Paasche index, the
numerator measures actual expenditures in the current period. The denominator
evaluates the current period market basket for continuing buyers using the market prices
they paid for the good (P i,t-1 ) and the market basket for new buyers using their
(individual) reservation prices (PR,b i,t-1). Note that although market prices do not vary
over buyers, reservation prices are buyer-specific.
FGC then propose a way to deal with the unobserved reservation prices. In (2),
consumer heterogeneity complicates things because, in principle, one needs a reservation
price for each consumer. However, FGC show that all one needs is an average
reservation price. In particular, one can adapt (2) to allow for new buyers by simply
replacing individual reservation prices in the base period with an average reservation
(3) ΠΝCt,t-1 = [Σi Σb P i,t X i,t b] / [ Σi (Σb∈C P i,t-1 X i,t b) + (PR i,t-1 Σb∈N X i,t b) ]
where PR i,t-1= [Σb∈N PR,bi,t-1X i,t b] / [Σb∈N X i,t b ].
Finally, to measure this average reservation price, FGC show that if one can place
bounds on consumers’ reservation prices, and if one assumes that the reservation prices
are uniformly distributed within those bounds, then FGC show that one can calculate the
average reservation price as the average of the bounds. Although the uniform
distribution is, in some sense, unsatisfying, as pointed out by FGC, there does not seem
to be any obvious reason for choosing any distribution in particular.10
10 In terms of chart 1, measuring price change over the life of the first chip—from t-6 to t—involves
measuring price change over six time periods, and if new buyers exist in each of those time periods, then
one would need to make an assumption about the distribution of reservation prices at each of those six
points in time. Although there are probably weaker distributional assumptions under which the average of
the bounds will give one the average of the distribution—if the bounds are symmetric about the mean, for
example—that would generate clumps of consumers with reservation prices exactly between market prices
at each period over the life of the good and there’s no reason to believe that reservation prices will fit this
One can, in principle, apply the FGC idea to the case of exiting buyers—those
buyers that purchased the good in the base period but not the current period. In
particular, the following Laspeyres index measures price change for exiting and
(4) ΛΧCt,t-1 = [ Pi,t (Σb∈C Xi,t-1b) + Σi PRi,t (Σb∈XXi,t-1b) ] / [ Σi P i,t-1 (ΣbXi,t-1b)]
Here, the denominator is unchanged—it measures actual expenditures in the base
period—but the numerator now values the base period market basket using market prices
for continuing buyers and reservation prices for exiting buyers.
This Laspeyres index may then be combined with the Paasche index in (3) to
form a Fisher Ideal index that allows for the presence of new and exiting buyers:
(5) FFGCt,t-1 = [ ΛΧCt,t-1 ΠΝCt,t-1 ] 1/2
As shown in the Appendix, sufficient conditions for the numerical equivalence of
the FGC and standard Fisher indexes are that (1) the gap between the average reservation
price and the market price be the same for new and exiting buyers—i.e., (Pi,t-1 / PR i,t-1 ) =
(PRi,t / Pi,t)—and (2) that markets be stable in the sense that the number of entering buyers
just equals that of exiting buyers. These conditions suggest that price growth will be
understated in growing markets—where entry exceeds exit—and overstated in shrinking
markets—where the reverse is true.
3.3 Applicability to Durable Goods
The remainder of the paper focuses on this problem as it relates to durable goods
purchases. In this context, two issues with potential implications for price measurement
must be considered: what is the relevant price and when are these prices well-defined.
For non-durables that are not storable, the purchase and consumption of the good
occur in the same period, so that the observed price may be used to measure changes in
consumer welfare as well as the value of the industry’s output. In contrast, durable goods
11 Just as the Paasche index is the relevant index for the price change experienced by new buyers—the
Paasche uses the current period quantities to compare prices—the Laspeyres is the relevant index for
exiting buyers—it uses base period weights to compare prices.
are purchased in one period but the stream of services is consumed over time so that the
price at the time of acquisition cannot be used to measure consumer welfare. For the
latter, the relevant ”good” is the per-period flow of services and the attendant price is the
per-period cost—either a user cost or a rental price.12 This is the preferred measure for
constructing cost-of-living indexes (COLIs) where one is interested in the use of the
durable.13 Hence, an index based on acquisition prices does not have a clear COLI
interpretation except under the (strong) assumption that changes in acquisition prices
track those of user costs.14,15
But, what if the focus is the construction of price deflators to value industry
output? What is the relevant price for valuing production (rather than the use) of the
good? For production taking place today, it makes sense to value the good at today’s
acquisition price because the acquisition price gives the price that consumers are willing
to pay for the durable good and, thus, their valuation of the good at the time of purchase.
In what follows, acquisition prices are used to split out quality differences from pure
price change in the construction of a deflator to measure the value of real output. 16
The second issue that arises in the application of the FGC method to the case of
durable goods has to do with reservation prices. Consumer purchases of durable goods—
particularly the large-ticket items like computers—are likely to be one-time purchases.
The implication of one-time purchases for the FGC indexes is that there are no
continuing buyers—all buyers are either new or exiting buyers.
The Paasche index then measures price change for new buyers only and the
Laspeyres measures price change for exiting buyers. But, notice that this assumes that
12 See, for example, Hall and Jorgenson (1967).
13 Feenstra and Knittel (2004) approach this issue from a COLI perspective, in part, because they are
interested in constructing an input price rather than a price deflator. Moreover, their data allow them to
base their estimates of any potential bias on user cost estimates rather than acquisition prices.
14 Note that if all durable goods were actually leased or rented, the problem disappears. In that case, the
“good” is defined as the flow of services each period and the “price” is the rental price. In effect, the
problem has been redefined to fit the static problem that forms the basis for standard price measures.
15 The unfortunate fact is that, in practice, the paucity of data with which to construct user costs or rental
prices has led many to construct simple indexes based on acquisition prices. In fact, the majority of
available indexes for durable goods—including some that we examine below—are based on acquisition
16 This ignores the case where production is also dynamic—for example, the building of Boeing jets
requires many months of production—it’s not at all clear when output should be recorded and which prices
should be used to value it.
reservation prices exist for buyers that have already made a purchase. As argued below,
it’s not clear that exiting buyers have a well-defined reservation price in the period after
making a purchase.17
In what follows, it is assumed that consumers making a purchase at time t leave
the market once they make a purchase and that therefore, do not have a well-defined
reservation price at time t+1. Moreover, the focus is narrowed to the case of one-time
purchases—a plausible assumption for many durables.
The implication for measurement when all consumers are one-time buyers is that
only the Paasche is well defined. It is the only index that focuses on buyers that were in
the market in both periods and that measures their assessment of changes in quality over
the period. Given reservation prices for new buyers, the FGC Paasche index provides a
best guess at pure price change:18
(3') ΠΝCt,t-1 = [Σi Σb P i,t X i,t b] / [ Σ (PR i,t-1 Σb∈N X i,t b) ]
4. FGC Indexes for High Tech Goods
This section applies the FGC idea to the case of high technology goods. The
notion of a “reservation price” is pinned down in order to explore when a reservation
price exists and how reservation prices are likely to change over time.
4.1 Reservation Prices for Durable Goods
In general, one can think of a consumer’s “reservation price” as the maximum
price he is willing to pay for a good. Discrete, dynamic choice models provide a
framework that allows one to explore the properties of these reservation prices under
fairly general conditions. Suppose there are Jt goods on the market at time t=1,…T and
17 This problem arises in the traditional new goods problem in that the use of reservation prices requires, in
principle, that the good exist in the period prior to introduction (See the discussion in Fisher and Griliches
(1995). Similarly, in the present paper, the construction of a Laspeyres index requires that the consumer
"exist" or be in the market after making a purchase.
18 The traditional interpretation of a Paasche index is as a lower bound to a true cost of living index
(COLI). As discussed earlier, however, the COLI interpretation is not relevant for durable-goods price
indexes that are based on acquisition prices.
that each of the I households already owns an (older) version of the durable that was
bought at time τ < t. Households face Jt + 1 choices each period; they either buy one of
the Jt durables on the market or they keep the durable they already own.
The I households in the market face per-period utility functions of the form
U(z,s(qjt,), pjt(qjt)), where z is a vector of the household’s characteristics, and both the
service flow s( ) and user cost p( ) depend on the characteristics of the good (qjt) a vector
that includes the age of the durable they already own. An intuitive way to think about the
service price is to define it as the erosion of the value of the asset that occurs over the
period. Let λt be the fraction of the value at acquisition that remains at time t, where the
fraction is, in turn, a function of the good’s characteristics, including age. Then, the
erosion of the asset from t-1 to t is defined as the rental cost:
p(qjτ) = λt-1(qjτ) / λt(qjτ) Pjτ
This rental cost for the asset held by the household is a function of the age of the asset—
an attribute included in the vector of characteristics—and the acquisition price. Although
the rental cost is a function of the acquisition price, in this specification, the relevant
price change for the consumer’s COLI is not: pt / pt-1 = λt / λt-1.
Following the notation in Keane and Wolpin (1994), define dj(t)=1 if the
household chooses option j at time t, and dj(t)=0 otherwise. And assume that the
decisions are mutually exclusive so that Σj=0,J+1 dj(t)=1. Then, the per-period utility
function may be written as:
j=0,J+1 dj(t) Ujt.
Conditional on the household’s characteristics and all the options available at
time t—including the durable they already own—(S(t))—each household chooses a path
so as to maximize future utility. Given Markov assumptions, there are J+1 Bellman
equations that give the time t expected utility associated with each option:
(7) Vjt= U(z,s(qjt,), pjt(qjt)) + δE[V(S(t), dj(t)] .
The first term is the utility for option j at time t and the second term is the expected future
utility conditional on having chosen option j at time t (that is, of all the future decisions
that might be made, the second term gives the expected utility across all possible paths).
The optimal choice, V*t, is the option that gives the highest expected utility, V*t > Vjt,
for all j ≠ *.
Given this framework, one may define a reservation price for each good on the
market as the highest (acquisition) price for the good such that the consumer would
choose that good over all other options, conditional on actual price of the other options:
the reservation price for option j is the highest price for option j that satisfies Vjt(PRjt ) >
Vkt for all k, and depends on all current and future service flows and rental prices for all
goods. Thus, although utility depends on service flows and user costs, the reservation
price is defined as a hypothetical acquisition price. One may define such a reservation
price for each option.
In what follows, it is assumed that consumers leave the market in the period
following a purchase and that, therefore, a reservation price does not exist in that period.
A consumer will choose to be “in the market” when it is likely that he can find an option
that could sufficiently raise his utility over his current option (Vot ). Immediately after
making a purchase, the good becomes a “second-hand” good and its value typically falls
sharply. In the framework above, this is represented as a high service price in the first
period of acquisition (high λτ+1(qjτ) / λτ(qjτ)) that makes it relatively expensive to hold a
durable in the first period and inexpensive to hold it in subsequent periods. If that service
price is sufficiently high, the consumer will not find an (unanticipated) alternative in the
next period that will bring him sufficiently high utility to offset the costs involved in
switching and will, therefore, exit the market. A framework that allows for search costs
would only make exit more likely.
It is also assumed that reservation prices fall over time. As pointed out above,
reservation prices depend on the prices and quality levels of all current and future goods
so that changes in the prices of substitutes and complements will affect reservation
prices. For example, a decline in the price of a substitute lowers the reservation price for
a good. Moreover, in markets for high-technology goods—and perhaps other markets
with rapid rates of product innovation—substitution exists along an intertemporal
dimension as well as contemporaneously. So, the reservation price for a good available
today depends not only on the prices and qualities of currently-available goods, but also
on those of goods that are expected to enter the market in the future. In this intertemporal
dimension, an important determinant of the reservation prices for today’s goods is the
amount of time the consumer must wait before the good appears. In particular, as the
arrival of future, higher-quality goods approaches, the reservation prices for goods
available today will fall. This is exactly the explanation often given for rapid declines in
market prices over the life of high tech goods. If this intertemporal substitution effect
dominates that of contemporaneous substitutes and complements in generating
movements of reservation prices, then reservation prices will fall over time. As seen
below, this turns out to be an assumption that is important for bounding reservation
4.2 Bounding Reservation Prices when Price Contours are Downward-Sloping
The challenge in constructing the FGC-type indexes is in finding a way to
measuring reservation prices. In what follows, the patterns of price change typically
exhibited by high technology goods are exploited to obtain bounds for reservation prices
that may ultimately be used to form a price index for one-time buyers.19
High tech goods like semiconductors, personal computers, and other consumer
electronics typically have price contours that are downward sloping. The shape of these
contours can be generated by cost- or demand-side factors (See Flamm(1987) and Irwin
and Klenow(1994)) for a discussion of the role of learning-by-doing in generating these
price contours for memory chips and Aizcorbe and Kortum(2004) for a vintage-capital
model that generates these contours without appealing to learning economies).
In either case, these downward-sloping price contours reflect what consumers are
willing to pay for the good and may be used to obtain bounds for their reservation prices.
Chart 2 traces out a typical price contour, where a new good enters the market at time t-3.
19 This framework will not accommodate cases where the durable good held by the household “dies” and is
replaced by another good in the same period. In that case, a reservation price does not exist at t-1 and the
FGC indexes cannot be calculated.
Consider a Paasche measure of price change over t,t-1 (i.e., Πt,t-1 ). For one-time buyers,
the index contains only new buyers whose reservation prices are such that they do not
purchase the new good until time t. The reservation-price approach to measure the price
change experienced by a buyer is the ratio of the price he actually paid at time t to his
reservation price at time t-1 (i.e., Pi,t//PRi,t-1 ).
To see how downward sloping contours allow one to bound his reservation price
at time t-1, note that, for the current period, the fact that the buyer did purchase the new
good implies that his reservation price was at least as high as the market price in that
period (time t). Similarly, for the earlier period, the fact that the buyer did not purchase
the new good implies that his reservation price was at least as low as the market price in
that period. As long as his reservation price did not rise over the period—the assumption
discussed in the previous section—then the observed prices in the two periods provide
bounds for his reservation price. That is,
(8) Pi,t < PRi,t < PRi,t-1 < Pi,t-1 ,
Making the assumption that these bounds hold for all buyers, and that the
reservation prices are uniformly distributed—the FGC assumption—one can form the
average of the bounds gives the average reservation price: PR
i,t = [Pi,t + Pi,t-1] / 2.
Substituting this average price into the Paasche in (3’) yields a simple expression
that relates that FGC Paasche for new buyers (ΠΝΒt,t-1) to one constructed using standard
(9) ΠΝΒt,t-1 = [ 1/2 ( 1 + Πt,t-1
-1 ) ] -1
For goods with downward-sloping price contours, the standard Paasche will show price
declines (0 < Πt,t-1
-1< 1 ) and will fall faster than the FGC Paasche: Πt,t-1 < ΠΝΒt,t-1
example, if the standard Paasche shows a 10% decline (Πt,t-1
-1= .9), the Paasche for new
buyers will fall less than that, say 5% ( ΠΝΒt,t-1 = .95).
Table 1 illustrates how various degrees of price declines as measured in the
standard Paasche (in column 1) translate into the constant-quality price decline
experienced by new buyers (column 2). Comparing these two measures shows that the
FGC-Paasche for new buyers shows substantially slower price declines than the standard
Paasche: these differences range from about 50 to 60 percent of the price declines
measured in the standard Paasche. This makes sense because, given the bounds defined
above, the average reservation price is about half the distance between the prices used in
the standard index.
4.3 Comparison to Conventional Price Measures for Selected High-tech Goods
Using the FGC Paasche as the relevant price measure for one-time purchases,
one can compare movements in this index with those of standard indexes to assess the
potential numerical importance of this problem. Here, the FGC Paasche is compared to
indexes typically reported in the literature as estimates of true price change: superlative
indexes like the Fisher and Tornquist indexes or measures obtained from hedonic
Table 2 uses results reported in the literature to do this comparison. The standard
Paasche indexes reported in each study (column 1) were used in (9) to obtain the FGC
Paasche indexes (column 2). Comparing these two columns illustrates the theoretical
argument made above: declines measured in the FGC Paasche indexes are substantially
less than those in the standard Paasche indexes.
Column 3 provides information—either a point estimate or a range—for the
preferred indexes reported in the studies. Comparing those estimates with the FGC
indexes in column 2 provides some perspective on the potential direction and magnitude
of differences in the two indexes of true price change Among the three studies that used
Fisher indexes as preferred measures, two of them report Fisher indexes that fall faster
than the FGC Paasche: Abel et. al.(2004) and Kokoski et. al.(2000). In these cases, the
differences are not trivial: for the Abel et. al. study, price declines in the FGC index were
about ½ those in the standard Fisher—declines of 5.2 percent vs. 10.6 percent; for the
Kokoski et. al. study, the differences are smaller but still nontrivial—declines of 42
percent vs. 51 percent.
The third study that used a Fisher index—Prudhomme and Yu (2003)—reports a
Fisher index that lies above the FGC index. This case illustrates that if the gap between
the Paasche and Laspeyres is sufficiently wide, the resulting Fisher can lie above the
FGC index. In this case, the gap between the Laspeyres and Paasche was unusually
wide: the calculated Laspeyres in this study actually showed price increases and pulled
up the Fisher index to a substantially slower price decline than that in the FGC Paasche.
The remaining two studies in table 4 used hedonic techniques as their preferred
measure. As may be seen, the hedonic measures fall faster than the FGC index and,
again, the differences can be large This comparison is problematic, however, in that the
FGC and hedonic measures differ not only in their treatment of new buyers but also in
their treatment of new goods. A comparison that would control for this would adopt the
strategy proposed by Pakes (2003) for standard indexes and would use reduced-form
hedonic regressions to fill in the missing market prices created by new goods before
constructing the FGC Paasche index.
On balance, the differences between the FGC and preferred price measures are
sufficiently large to suggest that allowing for consumer heterogeneity might be
numerically important in the construction of price indexes.
5. Discussion and Future Research
This paper builds on the work in Fisher and Griliches (1995) and Griliches and
Cockburn (1994) to define indexes for goods that are purchased only intermittently. For
high-tech goods that display downward-sloping price contours, one may bound
reservation prices and apply the FGC approach to obtain constant-quality price indexes
that allow for the presence of heterogeneous consumers that make one-time purchases.
The theoretical discussion above does not provide guidance on when the standard Fisher
index will overstate or understate price growth relative to the FGC Paasche or the
potential magnitude of those differences. However, calculations based on results
reported in the empirical literature found several cases where the reported Fisher indexes
fall substantially faster than the FGC Paasche; one can also point to other cases where the
two indexes are close or where the Fisher falls less fast than the FGC Paasche.
With regard to the potential applicability of these issues for other durables, as
highly disaggregate data become available, the evidence is mounting that durable goods
other than high technology goods also have downward-sloping price contours. For
example, Corrado, Dunne and Otoo(2003) show that prices for new motor vehicles
typically decline over the model year. Similar patterns of price change were found for
apparel (Pashigian and Bowen (1991)), television sets (Silver (1999)), consumer audio
products (Kokoski et. al. (2001)) and college textbooks (Chevalier and Goolsbee(2004)).
It would be interesting to study the price contours for other goods and investigate the
possibility that intertemporal price discrimination in the presence of consumer
heterogeneity generates those contours. If so, the methods developed here would be
relevant for those goods as well.
The FGC approach provides a “fix” for matched-model indexes. Two alternative
approaches seem worth pursuing. First, in the context of standard hedonic techniques,
one could use household characteristics to control for consumer heterogeneity in a
standard hedonic regression. The idea would be that, in addition to controlling for
differences in quality across goods, the presence of heterogeneous consumers requires
that one also control for differences across households.
The second possibility is to use empirical models of the demand for durable goods
that have been recently extended to allow for the presence of heterogeneous consumers
who make purchases only intermittently (See, for example, Melnikov(2001), Nair(2002),
and Gowrisankaran and Rysman(2005)). Those models implicitly estimate entry and exit
probabilities for different groups of consumers and values for their utilities (reservation
prices). Those values could be folded into the FGC framework to explore the potential
magnitude of the new buyer problem across different durable goods.
Abel, J., E. Berndt and A. White (2004) “Price Indexes for Microsoft’s Personal
Computer Software Products,” NBER Working Paper No. 9966, March.
Aizcorbe, Ana (2002) "Why Are Semiconductor Prices Falling So Fast?" FEDS
Working Paper 2002-20,
Aizcorbe, A., C. Corrado and M. Doms (2000) "Constructing Price Indexes for High
Technology Goods," Available at www.nber.org.
Aizcorbe, A. and S. Kortum (2004) “Moore’s Law and the Semiconductor Industry: A
Vintage Model,” Paper presented at a conference in honor of Tor Jacob Klette,
“Technology and Change,” Oslo, August.
Aizcorbe, A., S. Oliner and D. Sichel (2001) “Semiconductor Price Puzzles,” Paper
presented at NBER/CRIW Summer Institute Workshop, July 2003.
Anderson, S.P., dePalma, A. and Thisse, J. (1992) The discrete choice theory of product
differentiation. Cambridge, MA: MIT Press.
Berndt, E. R. and Zvi Griliches, “Price Indexes for Microcomputers: An Exploratory
Study,” in Price Measurements and Their Uses, ed. Murray Foss, Marilyn Manser, and
Allan Young. (Chicago: The University of Chicago Press, 1993), pp. 63-93.
Berndt, E.R., Z. Griliches and N. Rappaport (1995) “Econometric Estimates of Price
Indexes for Personal Computers in the 1990s,” Journal of Econometrics, 68:243-268.
Berry, Steven, James Levinsohn, and Ariel Pakes (1995). "Automobile Prices in Market
Equilibrium," Econometrica, vol. 63(4), pages 841-90.
Betancourt, Roger R., and David Gautschi (1992) “The Demand for Retail Products and
the Household Production Model: New Views on Complementarity and Substitution,”
Journal of Economic Behavior and Organization, 17, 257-275.
Bils, Mark, and Klenow, Peter J. 2001a. The acceleration in variety growth. American
Economic Review 91 (May): 274–80.
___________. 2001b. Quantifying quality growth. American Economic Review 91
Chevalier, Judith and Austan Goolsbee, (2004) “How do Consumers make Durable
Goods Purchase Decisions? Evidence from the College Textbook Market," mimeo, Yale
Coase, Ronald (1972) "Durability and Monopoly," Journal of Law and Economics 15,
Cole, Rosanne, Y.C. Chen, Joan A. Barquin-Stolleman, Ellen Dulberger, Nurhan
Helvacian, and James H. Hodge (1986) “Quality-Adjusted Price Indexes for Computer
Processors and Selected Peripheral Equipment,” Survey of Current Business 66
(January), pp. 41-50.
Corrado, Otoo and Dunn (2003) An Initial Look at Incentives and Prices for Motor
Vehicles: What Has Been Happening in Recent Years? Paper presented at the
CRIW/NBER Summer Institute, July 28.
Dean, J. (1969) “Pricing Pioneering Products,” Journal of Industrial Economics,
Diewert, W.E. (1995) "Axiomatic and Economic Approaches to Elementary Price
Indexes" Discussion Paper Nol. 95-01, Department of Economics, University of British
Dulberger, E.R. (1989) “The Application of a Hedonic Model to a Quality-Adjusted
Price Index for Computer Processors,” in Technology and Capital Formation, ed. Dale
Jorgenson and Ralph Landau. Cambridge, Mass.: MIT Press.
Dulberger, E.R. (1993) “Sources Of Price Decline In Computer Processors: Selected
Electronic Components,” in Price Measurements and Their Uses , ed. Foss, Murray F.,
Marilyn E. Manser, and Allan H. Young. Cambridge, Mass.: MIT Press.
Feenstra, R. C. (1995) "Exact Hedonic Price Indexes," Review of Economics and
Statistics, 77(4):634-53, November.
Feenstra, R.C. and Knittel, C. (2004) "Re-Assessing the U.S. Quality Adjustment to
Computer Prices: The Role of Durability and Changing Software," Paper presented at the
CRIW Conference on Price Index Concepts and Measurement, Vancouver, BC, Canada,
Feenstra, R.C. and M. Shapiro (2000) High-Frequency Substitution and
the Measurement of Price Indexes, Paper presented at the CRIW conference “Scanner
Data and Price Indexes,” Sept. 15-16, 2000, Washington, DC.
Fisher, F. and Z. Griliches (1995) "Aggregate Price Indices, New Goods, and Generics,"
Quarterly Journal of Economics, 229-44, February.
Fisher, F.M. and K. Shell. (1972) “Taste and Quality Change in the Pure Theory of the
True-Cost-of-Living Index,” in Z. Griliches, ed. Price Indexes and Quality Change.
Cambridge, Mass.: Harvard University Press, Pp. 16-54.
Flamm, K. (1989) Mismanaged Trade? Strategic Policy and the Semiconductor
Industry,Washington, D.C.: Brookings Institution.
Flamm, K. (2003) "Moore's Law and the Economics of Semiconductor Price Trends"
(paper presented to the National Research Council Symposium on Productivity and
Cyclicality in Semiconductors: Trends, Implications, and Questions, Kennedy School of
Government, Harvard University, Cambridge, Mass., September 24, 2001; updated draft,
Gordon, R. (2000) “Does the New Economy Measure Up to the Great Inventions of the
Past?” NBER Working Paper No. 7833, August. Available at:
Gowrisankaran, Gautam and Marc Rysman (2005) “Determinants of Price Declines
for New Durable Goods” Paper presented at the ASSA meetings, Philadelphia, January.
Griliches, Z. and I. Cockburn (1994) "Generics and New Goods in Pharmaceutical Price
Indexes," American Economic Review 84(5):1213-1232.
Grimm, B. T. (1998) “Price Indexes for Selected Semiconductors, 1974-96,” Survey of
Current Business, February.
Hall, Robert E. and Dale W. Jorgenson. 1967. “Tax Policy and Investment Behavior.”
American Economic Review, 57, No. 3 (June): 391-414.
Harper, M. (2003) “Technology and the Theory of Vintage Aggregation,” Paper
presented at the CRIW conference titled “Hard-to-Measure Goods and Services: Essays
in Memory of Zvi Griliches” Bethesda, Md. September.
Hendel, I. and A. Nevo (2002) “Measuring the Implications of Sales and Consumer
Stockpiling Behavior,” mimeo.
Hicks, J.R. (1940) “The Valuation of Social Income,” Economica, VII, Pp. 105-24.
Hobijn, B. (2002) "On both sides of the quality bias in price indexes," Federal Reserve
Bank of New York Staff Report No. 157.
Irwin, D. A. and P. Klenow (1994) “Learning by Doing Spillovers in the Semiconductor
Industry,” Journal of Political Economy, 102(6):1200-1227, December.
Jorgenson, D.W. (2000) “Information Technology and the U.S. Economy,” Presidential
Address to the American Economic Association, New Orleans, Louisiana, January 6.
Keane, M., K. Wolpin(1994) “The Solution and Estimation of Discrete Choice Dynamic
Programming Models by Simulation and Interpolation,” The Review of Economics and
Kokoski, M., K. Waehrer and P. Rozaklis (2001) "Using Hedonic Methods for Quality
Adjustment in the CPI: The Consumer Audio Products Component" BLS Working
Paper No. 344, Office of Prices and Living Conditions, March.
McAfee, R. P. and T. Wiseman “Capacity Choice Counters the Coase Conjecture,”
mimeo available at
National Research Council (2002) At What Price? Conceptualizing and Measuring Cost-
of-Living and Price Indexes. Panel on Conceptual, Measurement, and Other Statistical
Issues in Developing Cost-of-Living Indexes, Charles l. Schultze and Christopher
Mackie, Editors. Committee on National Statistics, Division of Behavioral and Social
Sciences and Education. Washington, DC: National Academy Press.
Melnikov, O. (2001) “Demand for Differentiated Products: The Case of the
U.S. Computer Printer Market,” mimeo, Cornell University.
Nair, H. (2004), Dynamics of Pricing in Durable Good Markets: Application
to 32-bit Console Video Games, mimeo, Chicago GSB.
Oliner, S. and D. Sichel (2000) “The Resurgence of Growth in the Late 1990s: Is
Information Technology the Story?” Journal of Economic Perspectives, 14(4):3-22.
Pakes, A. (2003) “A Reconsideration of Hedonic Price Indices with an Application to
PC’s,” American Economic Review, Vol: 93 (5) pp. 1578 – 1596.
Pashigian, B. P and B. Bowen (1991) “Why are products sold on sale? Explanations of
Pricing Regularities.” Quarterly Journal of Economics, 106(4):1015-1038.
Prud’homme, M. and K. Yu (2003) “A Price Index for Computer Software Using
Scanner Data” Paper presented at the Brookings Institution workshop on Economic
Measurement, May 23.
Reinsdorf, M. (1993) "The Effect of Outlet Price Differentials in the US Consumer Price
Index," in M.F. Foss, M.E. Manser, and A.H.Young, eds. Price Measurements and Their
Use, Chicago, Ill: University of Chicago Press, Pp. 227-254.
Rosen, S. (1974) “Hedonic Prices and Implicit Markets: Product Differentiation in Pure
Competition, Journal of Political Economy, Pp. 34-55.
Silver, M.S. (1999) “Disparities in the prices of new and old models when they coexist
in the market: some econometric estimates,” International Journal of Management
Science 27(1999): 13-24.
Triplett, Jack (1990) "Hedonic methods in statistical agency environments: an intellectual
biopsy," in Berndt, E.R., and J.E. Triplett (eds.) Fifty years of economic measurement:
the Jubilee Conference on Research in Income and Wealth, NBER Studies in Income and
Wealth, Chicago: University of Chicago Press.
Triplett, J. (1998) “The Solow Productivity Paradox: What Do Computers Do to
Productivity?” Canadian Journal of Economics, Volume 32, No. 2, April 1999. p. 319.
Available at http://www.csls.ca/jrn/v32n2_04.pdf.
Table 1. Comparison of Standard vs. FGC Paasche Indexes
for One-Time Buyers
Table 2. Comparison of FGC Paasche Lower Bound and Preferred Price Measures.
____ Paasche ______ Preferred Measure___
Study Market Standard FGC Ave % chg Method
(1) (2) (3) (4)
Abel et. al. (2004)
Software - 9.9 - 5.2 -10.6 Fisher
Cole et. al. (1986)
Printers - 3.5 - 1.8 -10 to -14 Hedonic
Processors - 8.5 - 4.4 -18 to -19 Hedonic
Kokoski et. al. (2000)
CD Players -58.5 -41.7 -51 Fisher
Prudhomme and Yu (2003)
Software -24.9 -14.2 -6 Fisher
Source: Author’s calculations
Appendix: Comparison of FGC Indexes With Standard Indexes
Comparing the FGC indexes with those constructed under standard methods, the
FGC Paasche and the FGC Laspeyres understate and overstate price growth, respectively,
relative to the standard indexes. To see this, consider the magnitude of reservation prices
relative to market prices. In periods where consumers did not make a purchase, the
reservation price is less than the market price. For the Paasche index, replacing the
market price in the denominator of a standard index with a (lower) reservation price
increases the overall index. So, for example, with falling prices, a standard index that
equals .8 (showing a 20 percent price decline) would translate into an FGC index that is
greater than that: say, .9 (a 10 percent price decline). A similar argument shows that the
standard Laspeyres overstates price growth relative to the FGC Laspeyres.
A simple framework is used here to develop some sense of the potential
numerical differences in the standard and FGC indexes. Moreover, because the effect on
the Paasche and Laspeyres work in opposite directions, this framework is also used to
explore conditions under which the net effect on a Fisher index—an average of the two—
is just offsetting.
Beginning with the Paasche, define the units of each good that were sold to
continuing buyers as a fixed percentage of the total number of units sold, and simplify the
analysis by assuming that this percentage is equal across all goods: Σb∈C X i b = γC t-1
ΣbX i b for all i. Imposing this assumption on (3) yields:
(3') ΠΝCt,t-1 = [ΣiPi,t (ΣbX i,tb)] / [ Σi (ΣbX i,tb ) ( γCt-1 P i,t-1 + (1- γCt-1) PRi,t-1) ]
Second, express the average reservation price for new buyers as some fraction of the
previous period’s market price and, again, assume that this fraction is equal across goods:
PRi,t-1=λΝ P i,t-1 for all i, with λΝ <1 because market prices exceed reservation prices.
Substituting this into (3') yields a Paasche index that is a scalar blowup of the unadjusted
(3'') Π NC t,t-1 = [ Σb Σi P i X i,t b] 1
[ ΣbΣi Pi,t-1 X i,t b] (γCt-1 + λN (1 − γCt-1))
Because both parameters are positive fractions, the denominator is also a positive fraction
(0 < (γC
t-1 + λN (1 − γC
t-1)) < 1 ) and the FGC index will take on a higher value--show
slower price declines--than the standard index (where the denominator equals 1).
As an example, table 1 uses this expression to illustrate the potential numerical
importance of new buyers for a Paasche index that falls about an 84 percent average
quarterly percent change at an annual rate--the magnitude of price declines typically seen
for microprocessors.20 Two of the possible values for the parameters yield a scalar of 1
and, thus, return the usual Paasche: one is when the reservation price just equals the
20 The average quarterly percent changes at an annual rate are calculated as 4 x [ ( Σ Π-1, s / S ) – 1] x 100).
previous period’s market price (the top row of each panel) and the other is when there are
no new buyers (the first column).
As one would expect, the FGC Paasche indexes diverge from the standard indexes
as one increases the share of new buyers (i.e., reading left to right) or lowers the
reservation price (moving down the rows). So, for example, as shown in the top panel, if
reservation prices are, on average, about 90 percent of last period’s market price, and if
about ½ of expenditures on the new good in the current period went to new buyers, then
the adjusted Paasche falls about 67 percent per quarter (italics) versus an 84 percent fall
for the standard Paasche.
Scanning the table, the range of adjusted indexes is very wide. Because buyers’
reservation prices can range from the market price in the previous period to that in the
current period, the price change experienced by consumers can range from the value of
the standard index to zero (no price change).
A similar exercise yields a Laspeyres index that is a simple multiple of the
(4') ΛX t,t-1 = [ Σb Σi P i,t X i,t-1 b]
------------------------ (γCt + λ X (1 − (γCt))
[ ΣbΣi Pi,t-1 X i,t-1 b]
where γCt is the share of unit sales in the current period that were purchased by
continuing buyers and λ X is the ratio of the exiting buyers' reservation prices--averaged
over buyers--to the market price, assumed constant across goods and buyers. Again, both
parameters are positive fractions so the scalar is also a positive fraction. Thus, the FGC
Laspeyres index will take on a lower value--show faster price declines--than the standard
The FGC Fisher index is the geometric mean of the Paasche and Laspeyres and
may be expressed as a multiple of the standard Fisher:
(γCt + λ X (1 − (γCt))
(6) FFGCt,t-1 = FSTDt,t-1 [ ----------------------------- ] 1/2
(γCt-1 + λN (1 − γCt-1))
The FGC Fisher index equals the standard Fisher when the second term--call it the
adjustment term--equals one and that occurs when the numerator and denominator are
equal. A sufficient condition for this to occur is when the gaps between reservation and
market prices are the same for new and exiting buyers (λ X = λ N ) and when the new
buyers' share of unit sales in the current period just equals the exiting buyers' share of
unit sales in the base period ( γCt = γCt-1). Little is known about the magnitude of
reservation prices, let alone the magnitude of the gap between reservation and market
prices. Suppose for the moment that the gaps are, on average, equal. Then, for growing
industries (where entry exceeds exits), one would expect the effect of the Paasche to
dominate, in which case the FGC Fisher would take on a larger value--show slower price
declines--than the standard Fisher. One would expect the opposite in industries where
sales are falling (exits exceed entry).
Table A-1 illustrates this point by providing two examples of calculations of the
adjustment term in (6) using different assumptions about continuing buyers' share of unit
sales at t-1 (in each row) and time t (in each column). The two panels give calculations
when average reservation prices are 95 percent (top panel) and 90 percent (bottom panel)
of market prices. Notice that in each table, the numbers on the diagonal are 1, indicating
that when markets are neither growing nor shrinking, the FGC Fisher equals the standard
Fisher. Numbers for growing markets are given above the diagonal where, as may be
seen, the FGC Fisher is a larger number (shows slower price declines) than the standard
Fisher. So, for example, if the gap between reservation and market prices is 90 percent,
and if the share of unit sales that went to continuing buyers grew from 25 percent at time
t-1 to 50 percent at time t, the adjustment factor is 1.04 and a, say, 20 percent decline in
the standard index would translate into a 17 percent decline in the FGC index.
In sum, differences between a standard and FGC Fisher index depend on the
relative magnitude of new and exiting buyers and on the relative gap between reservation
and market prices for new and exiting buyers. So, for example, the analysis suggests that
price growth is understated in growing markets and overstated in shrinking markets.
Table A-1. Estimated Ratio of FGC Fisher to Standard Fisher Price
PRs / Ps = .95
Percent of unit sales purchased by continuing buyers
At time t-1 At time t
0% 25% 50% 75% 100%
0% 0 0.25 0.5 0.75 1
25% 1.00 1.01 1.01 1.02 1.03
50% 0.99 1.00 1.01 1.01 1.02
75% 0.99 0.99 1.00 1.01 1.01
100 0.98 0.99 0.99 1.00 1.01
0.97 0.98 0.99 0.99 1.00
PRs / Ps = .90
Percent of unit sales purchased by continuing buyers
At time t-1 At time t
0% 25% 50% 75% 100%
0% 0 0.25 0.5 0.75 1
25% 1.00 1.01 1.03 1.04 1.05
50% 0.99 1.00 1.01 1.03 1.04
75% 0.97 0.99 1.00 1.01 1.03
100 0.96 0.97 0.99 1.00 1.01
0.95 0.96 0.97 0.99 1.00
Chart 1. The "New Goods" Proble
Chart 2. The "New Buyer" Problem
t-1 t t+1