Dealer Pricing

INTERNATIONAL ECONOMIC REVIEW
Vol. 46, No. 4, November 2005
DEALER PRICING OF CONSUMER CREDIT∗
BY GIUSEPPE BERTOLA, STEFAN HOCHGUERTEL,

AND WINFRIED KOENIGER1
Università di Torino, Italy; Free University Amsterdam,

The Netherlands; IZA, Bonn, Germany
Price discrimination incentives may induce dealers to bear the financial cost
of their customers’ credit purchases. We focus on how financial market imperfec-
tions make it possible to segment the customer population. When borrowing and
lending rates differ from each other and from the rate of interest on a durable
good purchase, the structure of those rates influences customers’ choices to pur-
chase on credit or cash terms, and the scope for dealers’ price discrimination.
Empirical analysis of a set of installment-credit, personal-loan, and regional in-
terest rate data offers considerable support to the assumptions and implications
of our theoretical framework.
1. INTRODUCTION
Why are installment payment terms for durable good purchases so affordable
as familiar ads suggest? Compared to personal loans and other forms of borrow-
ing, installment payment plans may be safer from the point of view of lenders:
They are backed by collateral if lenders can repossess valuable second-hand cars,
for example, and purchase of items such as household appliances may be cor-
related with repayment-relevant features of the borrower’s lifestyle. Very often,
however, favorable credit terms are not granted to consumers by the lending insti-
tutions (banks) that bear repayment risks, but by sellers of durable goods. When
a purchase takes place on an installment payment basis, the bank pays the dealer
immediately and is entitled to receive future installment payments from the cus-
tomer. In cases where a dealer advertises a zero rate—or any rate lower than what
would be required by the bank’s cost of funds, processing costs, and assessment of
repayment probabilities—the amount paid by the bank to the seller’s account is
lower than the amount that the seller would receive had the customer paid with
cash.
∗ Manuscript received January 2003; revised December 2003.

1 Helpful comments were offered by Fumio Hayashi, Frank Vella, Guglielmo Weber, two anony-
mous referees, and participants at the first Economics of Consumer Credit workshop, the EEA 2001
Congress in Lausanne, the 2002 Royal Economic Society Annual Congress, and a Bonn University
seminar. We are also grateful to Findomestic Banca for making available the data set analyzed in
this article and for sponsoring, with CETELEM, the Finance and Consumption in the EU Chair to
which the authors were all affiliated while working on this article. Please address correspondence to:
Giuseppe Bertola, Università di Torino, Dipartimento di Economica, Via Po 53, 10124 Torino, Italy.
Tel: 39-0116702730. Fax: 39-0116702762. E-mail: giuseppe.bertola@unito.it.
1103
1104 BERTOLA, HOCHGUERTEL, AND KOENIGER
We argue that dealer subsidization of consumer credit can be explained by in-

centives for durable good sellers to engage in monopolistic price discrimination.
Since borrowing rates are higher than lending rates on the financial market, dis-
tinct groups of consumers are inclined to purchase on cash and on credit terms.
Hence, sellers can set those terms so as to offer different prices to cash-rich and
liquidity-constrained customers, in much the same way as lower prices are some-
times charged to consumers who own particularly old trade-ins or take the time
to clip coupons. We model this intuition formally and show that it can explain not
only the existence of dealer-subsidized consumer credit, but also its heterogeneous
incidence as observed in real-life data.
In our model, both installment payment terms and cash prices for a given item
are posted by dealers. They are not tailored to a specific consumer’s characteris-
tics. Hence, customers face a complicated discrete choice problem when deciding
whether to purchase a given good and, if so, whether to pay cash. We propose and
analyze a simple explicit model of how such choices depend on the amount and
timing of consumers’ purchasing power, and on the structure of interest rates of
consumers’ assets, liabilities, and installment purchases.
At the market level, the elasticities of demand by consumers who find credit
purchases attractive and by consumers who choose to pay cash depend on the
relationship between inclination to borrow and willingness to pay across the pop-
ulation of potential customers. If the rate of return implicit in the comparison of
cash and credit prices is larger than the interest rate on consumers’ assets, but
lower than that charged on their liabilities, then customers’ inclination to borrow
and willingness to pay are related to each other in ways that depend on finan-
cial market features. That relationship in turn determines the profitability of the
seller’s discrete credit-subsidization strategy that is observable in our data.
On the empirical front, we analyze a novel data set of installment-credit con-
tracts. To the best of our knowledge, ours is the first analysis of data on the ex-
tent and incidence of dealer-subsidized credit in the academic literature. Most
of the sample refers to small durable good purchases, for which the assumption
of posted financing terms is completely realistic. Regressions explaining the in-
cidence of dealer-subsidized credit, across the widely heterogeneous regions of
Italy and over time, offer remarkable support to our theoretical model’s sign and
curvature predictions regarding the role of interest rate levels and spreads.
Our theoretical and empirical results contribute to a small, but growing liter-
ature focused on microeconomic details of credit relationships. Seller-financed
credit has been studied from a monopolistic price discrimination perspective in
the context of business credit. Suppliers rather than banks may provide credit
when they are in a better position to screen, select, and discipline the lender, or
to repossess and use the loan’s collateral, as well as for price discrimination pur-
poses. Brennan et al. (1988) study incentives for sellers of investment goods to
finance their customers’ purchases in the presence of ad hoc liquidity constraints,
and an extensive literature (surveyed by Petersen and Rajan 1997) studies more
generally forms of trade credit. As regards consumer credit, some related market-
structure issues have been studied in specific instances. The profitability of addi-
tional sales generated by credit availability plays a role in models of credit card
usage (Murphy and Ott, 1977; Chakravorti and To, 1999), as well as in models of
DEALER PRICING OF CONSUMER CREDIT 1105
voluntary or legal provisions that make the lender jointly liable for the seller’s
failure to deliver suitable goods (Iossa and Palumbo, 2004). Credit may also be
packaged with sales for a variety of other purposes, including facilitating impulse
purchases and reducing transaction costs. The consumers’ side of credit relation-
ships, however, has been the subject of relatively little formal work. Juster and
Shay (1964) noted that interest rates are different on consumers’ assets, liabilities,
and durable purchases, characterized qualitatively the implications of this state of
affairs for consumer choices, and explored survey data empirically, focusing in par-
ticular on the sensitivity of aggregate consumption to changes in macroeconomic
monetary conditions. Although the extensive literature analyzing consumers’ con-
strained borrowing did not follow up on these early efforts, focusing on simple
quantity constraints instead, there are a few notable exceptions: Attanasio (1995)
stresses the importance of cash outlays for liquidity-constrained consumers, who
are prepared to pay higher interest rates in exchange for longer loan duration, and
Brugiavini and Weber (1994) and Alessie et al. (1997) also analyze empirical re-
lationships between borrowing opportunities and durable good purchases. These
and other contributions, however, propose and study models where borrowing
opportunities depend on the existing stock of durable goods, rather than on new
purchases as would be implied by the mechanisms outlined above.
After reviewing the data in Section 2, in Section 3 we set up and solve the
consumer’s discrete choice problem. In Section 4 we study how interest rates
influence consumers’ choices and sellers’ credit subsidy incentives, and in Section
5 we bring empirical analysis of a credit contract database to bear on the real-life
relevance of the model’s intuitive theoretical insights.
2. DEALER CREDIT SUBSIDIES IN ITALIAN DATA
The data set available for our analysis includes a random sample of more than
200,000 credit applications by more than 120,000 individuals, over the period 1995–
1999, to the leading supplier of consumer credit in Italy. Focusing on installment
credit applications filed in the period 1995/III to 1999/II, we retain about 123,000
observations. We keep all records in the data, irrespective of the outcome of the
credit application.2
In the Italian markets from which our data are drawn, dealers can choose to
absorb interest payments and offer attractive financial terms to those among their
customers who are inclined to purchase on credit. Since most items are small, there
is no negotiation of financing terms. The financing conditions are posted along
with the item’s price in the store, and customers are faced by a take-it-or-leave-
it choice when deciding whether or not to buy and use the credit opportunity.
The data set contains information on the characteristics of both the individual
customer (conditional on applying for credit) and of the item whose purchase the
credit application would finance. There is no information about customers who
pay cash and neither characteristics of nor identifiers for individual dealers.
2 Alessie et al. (2005) offer a detailed description of the data set and empirical analysis of the
time-series and cross-sectional impact of introduction of a law on usury rates in 1997.

TABLE 1
DEALER SUBSIDIZATION BY REGION
Region 1996/I 1999/I

North
Piemonte 25.3 23.2
Valle d’Aosta 14.3∗ 21.4∗
Lombardia 26.2 31.6
Trentino/Alto Adige 41.7∗ 32.1
Veneto 38.0 33.3
Friuli-Venezia-Giulia 51.4 44.8
Liguria 21.2 23.9
Emilia Romagna 41.1 29.7
Center
Toscana 38.0 26.9
Umbria 26.5 25.0
Marche 51.0 35.0
Lazio 27.1 23.4
South & Islands
Abruzzo 22.1 22.0
Molise 5.3∗ 15.0∗
Campania 19.3 29.2
Puglia 21.8 18.7
Basilicata 31.0∗ 3.8
Calabria 11.1 39.8
Sicilia 42.7 39.4
Sardegna 21.3 45.2
All (unweighted) 28.8 28.2
All (weighted) 30.4 30.5
NOTE: Percentage of installment applications with dealer subsidization by region.

∗ Cell size <30 observations.
SOURCE: Findomestic Banca; own calculations.
The crucial information from the present article’s perspective is the application-
level indicator of whether the dealer pays interest fully, in part, or not at all: Only
in the last case can financial charges be computed, as the internal rate of return of
the consumers’ repayment obligations. In practice, sellers typically pay all interest
if they do subsidize their customers’ credit.3 The resulting “zero” percentage rate,
familiar from advertisements, does not make credit completely costless, as con-
sumers bear transaction costs on installment payments and may also be required
to pay processing and insurance fees.
The incidence of subsidized credit purchases is high and variable across regions
of Italy, as well as across goods and over time. Tables 1 and 2 show percentages of
complete dealer subsidy in region-quarter-item cells, averaged over time for the
3 “Zero interest” deals are presumably preferable to low interest rates because their simple structure
is easily communicated to potential customers. Less than 1.5% of the contracts in the sample see
dealers paying only a portion of the bank’s interest charges; 74% of such contracts relate to motorcycle
purchases, perhaps as a reflection of the high rates charged on these relatively expensive items. We
decided to drop all partially subsidized installment plans from the sample.
TABLE 2
DEALER SUBSIDIZATION BY ITEM BOUGHT
Item bought 1996/I 1999/I
White goods 42.8 50.4

Household appliances 37.0 39.8
Brown goods 32.4 41.5
Computers 21.2 15.7
Furniture 26.4 24.6
Telephony 20.7 24.1
(Used) cars 6.5 7.8
New cars 15.1 4.7
Motorcycles 52.1 41.0
Sport and leisure 12.5 31.0
Home maintenance 14.3∗ 47.4∗
Air-conditioning na 82.9
Other 26.0 21.6
All (unweighted) 25.6 33.3
All (weighted) 30.4 30.5
NOTE: Percentage of installment applications with dealer subsidization by class of

good bought.
∗ Cell size <30 observations.
quarters 1996/I and 1999/I.4 Table 1 takes averages within regions across items,
and Table 2 takes averages within items across regions. The overall incidence of
subsidized credit is rather stable over time at about 30%. Consistently low rates
of complete dealer subsidization are observed, for instance, in Liguria (North)
and Abruzzo (South). High rates are found in, for instance, Friuli-Venezia-Giulia
(North) or Sicily (South). In 1996, the regional discrepancies were larger than
three years later.5
The incidence of dealer subsidies is also highly heterogenous across items
bought. Relatively low subsidization rates are observed for cars, whereas white
goods (kitchen appliances), brown goods (consumer electronics), and other house-
hold appliances are relatively frequently subsidized, as shown in Table 2. Variation
over time of item-specific subsidization is also present, and quite pronounced in
some cases: The incidence of subsidized credit for white good purchases, for in-
stance, increased from 42.8% to 50.4%, whereas for new cars it fell from 15.1%
to 4.7%. We discuss below possible sources of such heterogeneity in light of theo-
retical insights and on the basis of controlled-regression estimates of item-specific
effects.
Table 1 reported stark regional differences in the incidence of dealer subsi-
dization. We document in Table 3 similarly pronounced regional heterogeneity
in terms of financial market indicators, such as borrowing and lending rates. The
4 Our data set includes observations for earlier and later dates, but is most representative for the
1996–1999 period; we focus on the first quarter to mitigate seasonal variation in the raw data.
5 Some of the variation is, however, due to sampling error, as some cells have fewer than 30
observations.
TABLE 3
DISTRIBUTIONS OF QUARTERLY REGIONAL INTEREST RATES
Variable Quarter Mean SD Coeff. Var. Min Max
Bank borrowing rate 1996/IV 5.92 0.23 0.04 5.40 6.27

1998/IV 2.50 0.20 0.08 2.14 2.93
Bank lending rate 1996/IV 12.24 1.40 0.11 10.33 15.22
1998/IV 7.97 1.15 0.14 6.17 9.77
Level 1996/IV 9.08 0.66 0.07 8.10 10.44
1998/IV 5.24 0.58 0.11 4.38 6.18
Spread 1996/IV 6.33 1.51 0.24 4.47 9.56
1998/IV 5.47 1.16 0.21 3.58 7.26
SOURCE: Bank of Italy, Sintesi delle note sull’andamento dell’economia delle regioni italiane,
various years. The level is the midpoint of borrowing and lending rate, the spread is the
difference, both calculated at the regional level.
table also reports the level (average of borrowing and lending rates) and spread
(difference between lending and borrowing rates) of these rates. These statistics
are based on quarterly regional measures of financial rates applicable to rela-
tionships between the banking sector and the private sector. The available data,
collected and published by the Bank of Italy, refer to loans exceeding Italian
Lire (LIT) 150 m (approximately €75,000) and deposits exceeding LIT 20 m (ap-
proximately €10,000). Clearly, these rates are not directly relevant to consumers’
financial environment, but their level and the spread between active and passive
rates are related, over time and across regions, to those applicable to smaller loans
and deposits. The table shows that interest rates declined over time from 9.1%
to 5.2%. The rate spread also declined over time, from 63 basis points to 55 on
average. The interregional differences of the spread measured by the coefficient
of variation declined. In particular, lending rates declined sharply. In terms of re-
gional segmentation we find higher levels and larger spreads in the relatively less
well-developed Southern regions.
Our modeling approach indicates that such financial market variables should be
relevant to dealers’ incentives to offer subsidized credit. To see whether and how
the incidence of dealer subsidization covaries with financial market structure, we
next show patterns of temporal correlation of dealer subsidization with interest
rate spreads controlling for the rate level. We partition available data in three
macro regions (North, South, and Center, as defined in Table 1). To control for
interest rate level variability, we further partition each subsample in five quintiles
of the mean bid–ask interest rate distribution. Within each of the 15 subsamples, we
analyze the role of interest rate bid–ask spreads (a measure of financial market
imperfection) by regressing the region- and quarter-specific incidence of seller-
subsidized credit on a third-degree polynomial in the applicable bid–ask spread.
The resulting descriptive evidence is displayed graphically in Figures 1, 2, and 3.
A nonlinear hump shape is apparent in most of the curves displayed in those
figures. We show in what follows that such a nonmonotonic relationship is indeed
implied by a theoretical perspective based on the desirability and cost of interest
FIGURE 1
RELATIONSHIP BETWEEN DEALER SUBSIDIZATION AND INTEREST RATE SPREADS, AVERAGE WITHIN FIVE
QUINTILES OF THE RATE LEVEL DISTRIBUTION: NORTHERN ITALY
FIGURE 2
SAME AS FIGURE 1: CENTRAL ITALY

FIGURE 3
SAME AS FIGURE 1: SOUTHERN ITALY
rate subsidies from the dealers’ point of view, and we provide additional empirical
evidence in a controlled-regression framework. The raw data also indicate that
the incidence of dealer subsidies varies widely across goods (see Table 2). Our
model makes it possible to interpret this phenomenon in terms of underlying
characteristics of the relevant market demand functions.
3. A CONSUMER’S PROBLEM
We illustrate more general theoretical insights with an admittedly simplified,

but far from trivial formal model. In general, a consumer’s decision to purchase a
durable good depends on tastes, on prices, and on the funds available for that and
other purposes. When financial markets are imperfect, current and future funds
are not perfectly substitutable: Hence, optimal choices do not only depend on the
total amount but also on the timing of the consumer’s resources, which interacts
in interesting and complex ways with the relationship between the cash and credit
prices of the good and with the intertemporal rates of transformation applicable
to borrowing and lending contracts.
To model the relevant choice in the simplest possible way, we consider a two-
period representation of the consumer’s preferences. We assume utility to be
increasing and concave in current nondurable consumption, C, and in the amount
of funds available for future consumption, denoted A, and to be higher when the
durable good is purchased and available for use in the current and future periods.
A general time-separable representation of such preferences has the form U(C,

δ) + V(A, δ) for δ, an indicator equal to 1 if the durable good is available and 0
otherwise, with
U(C, 1) + V(A, 1) > U(C, 0) + V(A, 0),

∂U(·) ∂ 2 U(·) ∂ V(·) ∂ 2 V(·)
> 0, < 0, > 0, <0
∂C ∂C 2 ∂A ∂ A2
for all C and all A.6 We choose to illustrate the character of the solution in a simple
special case where closed-form solutions are available, namely:
(1) max log(C) + log(A) + δk

C,δ
where the additive constant k represents the positive impact of the durable’s
services on utility. This parameter summarizes a host of possible reasons why the
durable is attractive. In reality, k is heterogeneous across customers in ways that
also depend on the customers’ ownership of other durable goods, resulting from
previous purchases in a dynamic environment. In the model we specify and solve,
k is treated as a given parameter. We briefly discuss below its implications for
the dealer’s pricing decision, but data limitations make it impossible to account
explicitly for it in our empirical work.
Maximization of (1) is subject to budget constraints. We denote with W the
consumer’s current cash on hand; with Y the amount of purchasing power that
will become available (with certainty) in the future; with r a the rate of return on
the consumer’s assets; and with rb ≥ ra the interest rate charged on the borrowed
amount (negative assets). Thus, when the durable is not purchased the budget
constraint reads

(W − C)(1 + ra ) if C < W
(2) A= Y +
(W − C)(1 + rb) otherwise
When the durable good is purchased, the budget-constraint relationship between

current and future resources and utility-function arguments is similar, but needs
to account for the durable’s price by subtracting it from the consumer’s resources.
To simplify notation, suppose the durable good may be either paid in cash, at
price P0 , or fully financed, with no money down and a single installment payment
P1 in the future period. Payment plans with positive down payments would have
substantially similar implications as long as the financial structure of the payment
plan (like the price of the durable) is taken as given by the consumer. Thus, if the
consumer purchases the good on a cash basis, then P0 needs to be subtracted from
6 Preferences in this form are assumed in, for example, Spence’s (1977) study of durable-good
pricing when product quality is not directly observable. With little loss of generality, Spence assumes
that U(C, 1) = U(C, 0), and notes that specifications such as V(A, 1) = V(A, 0) + k or V(A, 1) = V(A+
k, 0) for some constant k may be interesting special cases for characterization purposes.
W in the relevant version of budget constraint (2). If the good is purchased on credit
instead, then P1 is subtracted from Y. We assume that the internal rate of return
of the durable good’s financing terms is neither lower than the rate of return on
consumers assets nor higher than the interest rate on general consumption loans,
P1
(3) 1 + ra ≤ ≤ 1 + rb
P0
so that credit purchase cannot be preferable to cash purchase, or vice versa, for
both borrowers and lenders in the consumer population. Since no customer would
choose one of the two payment terms if condition (3) failed, it must be satisfied
when cash and credit terms are chosen by a durable good dealer who finds it
optimal to sell the durable at different prices to different segments of the hetero-
geneous population of potential customers.
Conditional on the durable purchase decision, the impact on utility of current
consumption and future purchasing power is easily characterized by familiar Euler
and slackness conditions over the two periods. Since utility is logarithmic, optimal
allocation of available funds across current consumption and future uses implies
that
(4) C (1 + ra ) ≤ A ≤ C (1 + rb)
the first inequality holds as an equality if the consumer has positive assets (is
lending), the second if the consumer borrows (has negative assets). If the lending
and borrowing rates are not the same, it can be optimal for the Euler condition to
be slack, in which case the consumer is liquidity constrained.
The discrete choices of whether to purchase the durable good and whether to
use current or future funds for that purpose, conversely, needs to be based on
pairwise comparisons of utility levels achievable under each strategy, rather than
on first-order conditions. Through the budget constraint, current and future funds
available for allocation to the continuous variables C and A depend on whether
the durable is purchased and, if so, on whether current or future funds are used
to pay for it. Figure 4 illustrates the problem. The horizontal axis of the figure
measures amounts of current cash on hand W, and the vertical axis measures
amounts of funds Y that will be available in the future. If the durable is not
purchased, then the consumer’s logarithmic objective makes it optimal to lend
some of the current funds at rate r a if Y < (1 + ra )W and to borrow against some
of the future funds at rate r b if Y > (1 + rb)W. When the intertemporal pattern of
the consumer’s resources falls in the cone from the origin between these two lines,
the constrained consumption pattern simply coincides with available resources
and, since the durable is not purchased, utility only accrues from C = W and
A = Y.
The other lines plotted in the figure, whose slopes also coincide with the in-
tertemporal rates of transformation associated with positive and negative assets,
identify cones originating from P1 on the vertical axis and from P0 on the hori-
zontal axis. If the consumer purchases the durable good on credit, then the levels
FIGURE 4
A GRAPHICAL REPRESENTATION OF THE EULER CONDITIONS FOR CHOICE OF C AND A, CONDITIONAL ON
WHETHER THE GOOD IS PURCHASED WITH CASH, ON AN INSTALLMENT LOAN BASIS, OR NOT AT ALL
of C and A coincide with the amount of current and residual future funds if these
lie within the cone originating from P1 and reflect optimal borrowing or lending
choices if they lie outside of it. It is similarly easy to characterize the implications
for C and A of a decision to purchase the durable on a cash basis, referring to the
cone originating from P0 on the horizontal axis of the figure.
3.1. Character of the Solution. In order to establish optimality of cash, credit,

or no purchase, the utility levels achieved in those cases need to be compared
with each other. The three utility levels depend univocally on the parameters of
the problem (resources and prices) through the intertemporal allocation choices
determined by slack Euler conditions like (4). Such comparisons are conceptually
easy and could be implemented numerically for much more general preference
specifications than that proposed in (1).
The character of the solution, however, does not depend on such details. Since
the marginal utility of C and A is decreasing, the consumer chooses to divert
some of his or her purchasing power to the durable purchase (which affords a
given utility level k) only if current and/or future funds are sufficiently plentiful.
Hence, the endowment structures that make consumers indifferent to purchasing
FIGURE 5
THE SOLUTION FOR THE PURCHASE INDIFFERENCE AND CASH–CREDIT INDIFFERENCE FRONTIER
the good identify a (weakly) downward-sloping locus in (W, Y) space.7 Also in-
tuitively, indifference between cash and credit payments in cases where the con-
sumer does purchase the durable is depicted by an upward-sloping locus in that
space: A steeper expected income profile, such as that of young households, leads
consumers to prefer credit purchases if they are sufficiently rich.
Figure 5 illustrates the qualitative character of the solution, and the derivations
that follow characterize how the shape of the relevant indifference loci depends
on the relationship between the borrowing and lending rates on the one hand and
on the cash and credit price of the good on the other. To organize those derivations
and to offer some intuition for their shape, it will be helpful to refer to regions of
the three points M, N, and Q in Figure 4. In what follows, we present the solution
for these three regions; the Appendix discusses the complete solution of all the
10 regions defined by the Euler conditions in the (W, Y) plane, within which the
possible lending, borrowing, and purchase choices are restricted in different ways.
7 Suppose instead the purchase-indifference locus were positively sloped. Consider a point
(W ∗ , Y ∗ ) on that locus and a point (W ∗ + x,Y ∗ ) to its right, where x > 0. The consumer would
purchase the durable at (W∗ , Y∗ ), but not at (W ∗ + x,Y ∗ ) despite the fact that overall resources have
increased. This cannot be optimal.
3.2. Indifference to Purchase. If the consumer’s endowment is such that op-

timal assets are positive regardless of whether the durable good is purchased on
a cash basis or on credit or not at all (as is the case at point Q in Figure 4), it is
straightforward to characterize the optimal purchase decision. If the durable good
is not purchased, then current consumption is

1 Y
arg max[log(C) + log(Y + (W − C)(1 + ra ))] = W+ ≡ CN,a
C 2 1 + ra
whereas a cash purchase reduces current consumption to
arg max[log(C) + log(Y + (W − C − P0 )(1 + ra ))]

C

1 Y
= W − P0 + ≡ CD,a
2 1 + ra
and affords additional utility k. The choice is a matter of indifference for the
consumer when
(5) log(CD,a ) + log(Y + (W − CD,a − P0 )(1 + ra )) + k

= log(CN,a ) + log(Y + (W − CN,a )(1 + ra ))
This is a quadratic equation in P0 and W + Y/(1 + ra ), the present value of funds

discounted at the lending rate. The left-hand side of (5) is larger than the right-
hand side (to imply that cash purchase is preferable to no purchase) if 8
1
Y e2k
W+ > 1 P0
1 + ra e2k − 1
namely, if the consumer’s endowment of current and future resources, at a point

such as Q in Figure 4, lies to the north east of a downward sloping line. It remains
to be checked whether credit purchase might in turn be preferable to cash or no
purchase. Quite intuitively, however, credit purchase cannot be strictly optimal
for the consumer if the endowment, as is the case at point Q, is such that assets
are positive regardless of whether and how the durable is purchased. In fact, when
assets are positive, use of future rather than current funds can never increase the
amount of future purchasing power A for any choice of C, and hence utility, since
(6) Y + (W − C)(1 + ra ) − P1 ≤ Y + (W − C − P0 )(1 + ra )
as long as P1 ≥ (1 + ra )P0 as assumed in (3).

1k
8 The left-hand side is also larger if W + Y
1 + ra < e2
1k P0 , but such low wealth levels would imply
e 2 +1
negative current consumption in the relevant
region:
1k
CD,a = 12 W + 1 +Yra − P0 < 12 1
e2
P0 − P0 = − 1
2
1
1k P0 <0
k
e 2 +1 e 2 +1
Symmetric reasoning is applicable when the endowment is such that assets are
negative regardless of whether and how the durable is purchased (see point M in
Figure 4). In that case, credit purchase is always at least weakly preferable to cash
purchase, and preferable to no purchase if
1
Y e2k P1
W+ > 1
1 + rb e2 − 1
k 1 + rb
again a downward-sloping line in the plane depicted by the figures. The two in-
difference lines would coincide if the borrowing and lending rates were equal to
each other and, by (3), to the internal rate of return of the durable’s installment-
credit plan. In such a perfect-capital-markets case, the solution would be easy and
uninteresting: All the cones would collapse to lines in the figure, consumers would
always be indifferent between cash and credit purchase, and only the present value
of intertemporal resources would affect their choice of whether to purchase or not.
When r a < r b, conversely, the indifference locus is a more steeply declining line
in the region of point M than in the region of point Q, and becomes interestingly
nonlinear when the consumer’s endowment lies outside of those regions.
Characterization of the purchase decision is also quite straightforward when
the (W, Y) endowment of purchasing power lies in the region of point N, that
is, within all three of the cones plotted in Figure 4. When the consumer has zero
assets regardless of whether and how the good is purchased, a cash purchase is
preferable to no purchase if log(W) + log(Y) < log(W − P0 ) + log(Y) + k, that is,
if
ek
(7) W > P0
ek − 1
and credit purchase is preferable to no-purchase if log(W) + log(Y) < log(W) +

log(Y − P1 ) + k, that is, if
ek
(8) Y > P1
ek − 1
The parameters may be such that these conditions are automatically satisfied
for all endowment patterns in the region of point N. In this case, the horizontal
and vertical coordinates of all points in that region are larger than those of
the intersection point of the flatter boundary of the cone originating from P1
on the vertical axis with the steeper boundary of the cone originating from P0
on the horizontal axis. Formally, the intersection point can be calculated as the
solution of P1 + (1 + ra )W = (W − P0 )(1 + rb)
P1 + P0 (1 + rb)
(9) W=
rb − ra
1 + rb
(10) Y = (P1 + P0 (1 + ra ))
rb − ra
If the right-hand sides of (7) and (8) are larger than those of (9) and (10), then
purchase of the durable is always optimal in the region of point N. Otherwise,
the purchase-indifference locus goes through that region, where it consists of a
vertical and horizontal segment.
To complete the characterization of optimal choices in the region of point N,
note that for a consumer who always has zero assets, credit purchase is preferable
to cash purchase if
log(W − P0 ) + log(Y) + k < log(W) + log(Y − P1 ) + k
that is, if Y/W > P1 /P0 . Thus, any portion of the upward-sloping cash–credit
purchase–indifference locus that lies in the region of point N is a straight line
segment, with slope given by the installment plan’s internal interest factor. We
have already shown that all points in the region of point M (Q) lie above (below)
that locus; we derive the shape in other regions of the (W, Y) plane in the Appendix.
Whenever the purchase-indifference locus goes through the region of point
N, it has a horizontal and a vertical segment. In general, the nonlinearity of the
purchase-indifference locus becomes more pronounced as the consumer’s imper-
fect access to borrowing and lending opportunities becomes more relevant to the
purchase decision. Intuitively, the indifference locus tends to become (and is, in
the region of point N) horizontal when the consumer is liquidity constrained, be-
cause when future resources are too low to make a credit purchase appealing (and
financial markets are not accessed), then an increase of current resources does not
make such a purchase any more desirable. For a given degree of financial market
imperfection, as represented by the difference between r b and r a in the model,
the relevance of financial market access for durable purchase decisions depends
on k.
Indifference conditions in the other regions, where purchasing the durable good
is always associated with a discrete change in the consumer’s asset position, can be
derived by much the same reasoning as that applied to the regions—considered
above—where assets are positive, negative, or zero regardless of purchase deci-
sions. In all cases, indifference to purchase is characterized by comparisons of
utility reflecting the optimal intertemporal pattern of consumption if the con-
sumer smooths it by either borrowing or lending, or just the intertemporal pattern
of resources if the consumer is liquidity constrained. We discuss solution methods
in the Appendix, where we characterize in closed form the subset of the (W, Y)
space where cash purchase is preferable to no purchase and that where credit
purchase is preferable to no purchase. The intersection of these sets includes all
intertemporal endowment patterns that make purchase optimal. The Appendix
shows that the locus of purchase indifference is continuous.
3.3. Indifference between Cash and Credit Purchase. To complete character-

ization of the solution, the set of (W, Y) points for which purchase is preferable
to no purchase can be further partitioned according to whether cash or credit
payment is optimal. The boundary of these subsets is an (upward-sloping) locus
of points such that the choice of cash or credit terms is a matter of indifference.
We have already characterized this locus in the region of point N, where the
consumer has no assets in all cases and, quite intuitively, prefers credit to cash
purchase when Y/W > P1 /P0 , that is, when the financing terms of the durable
purchase offer an attractive intertemporal rate of transformation for the assumed
logarithmic specification of preferences. As mentioned above, in other regions
(where the purchase is associated with a change in the consumer’s asset position)
complex considerations are relevant to the choice of whether to purchase on a cash
or credit basis. We report in the Appendix the exact solution for the cash–credit
indifference locus in all regions, which is continuous with slope lower than 1 + r b
and larger than 1 + r a . Intuitively, in all regions where the purchase implies a qual-
itative change in the consumer’s asset position (which can be positive, negative,
or zero) the slope of the cash–credit indifference locus is a weighted average of
the applicable intertemporal marginal rates of substitution, which all lie between
1 + r a and 1 + r b.
3.4. Discussion. In our model, derivation and description of the solution are
considerably complicated by the need to consider all the possible interactions
between the discrete decision to purchase the durable good and the (also discrete)
changes in the budget set. Such changes imply possible switches from positive to
negative asset positions, or vice versa, when a credit or cash purchase alters the
intertemporal pattern of residual funds. This feature captures important elements
of reality, since not only the rate of interest but also the structure of installments
is severely limited in most durable good purchases, and has complex implications
for optimal behavior.
Some of the complications arising from discontinuities of the consumer’s budget
set may be of practical as well as theoretical interest. For example, the discreteness
of purchase and financing choices implies that it may be optimal for a consumer
to buy the durable on credit and hold positive liquid assets (though less than in
the case of no-durable purchase). This phenomenon is reminiscent of the puzzling
coexistence of credit card debt and liquid assets in many consumers’ portfolios
(identified and studied by Bertaut and Haliassos, 2002, and their references). In
our model of discrete purchase and financing choices, holding positive assets and
purchasing a durable good on credit is optimal for some intertemporal endow-
ments: The first-period consumption compression implied by a cash purchase can
yield lower utility than a smaller consumption decline in the first period, smoothed
intertemporally by a larger positive asset position, and payment of the good on a
credit basis. In terms of Figure 4, installment liabilities and liquid assets can co-
exist in regions of the (W, Y) plane where the purchase of the durable affects the
consumer’s decision whether to lend or to borrow whereas Equation (6), which
rules out such coexistence, only holds in the region of point Q, where assets are
positive independently of the durable purchase.
Our functional form assumptions conveniently imply that all conditions of in-
difference to purchase define quadratic equations in the (W, Y) plane. Allowing
for discounting of future utility or for nonhomothetic utility would imply more
complex but qualitatively similar relationships that could be studied by numerical
methods. Applying a discount factor β = 1 to future utility would require solution
of equations of order 1 + β, and the borrowing, lending, and liquidity-constrained

ranges would be delimited by nonlinear loci in Figure 4 for nonlogarithmic utility
functions.
4. CONSUMER CHOICES AND DEALER PRICING
In Figure 5, the purchase-indifference locus is the lower envelope of the loci,

which define when the consumer is indifferent between cash purchase and no pur-
chase or credit purchase and no purchase. The shape of the purchase-indifference
locus is intuitive. For example, the locus defining indifference between credit pur-
chase and no purchase becomes flatter in the region of the (W, Y) plane in which
the consumers’ endowment is more tilted toward the present. Starting from an
endowment at which the consumer does not purchase the durable, the consumer
needs relatively more additional current resources W to buy the durable on credit
if his initial endowment is relatively more tilted toward current resources. In the
figure, and below, we denote with ω the level of current wealth such that the
purchase-indifference and the cash–credit indifference locus intersect.
Crucially, for our purposes, the nonlinearity of the purchase-indifference fron-
tier depends on the relationship between r b and r a . This is illustrated in Figure 6,
where we plot purchase-indifference loci for increasingly large values of r b,
FIGURE 6
CONSUMER PURCHASE DECISIONS: IMPLICATIONS OF DIFFERENT BORROWING RATES; INTERNAL RATE OF

RETURN SET EQUAL TO THE AVERAGE OF LENDING AND BORROWING RATE
keeping k, r a , and P0 constant, and varying P1 so that the internal rate of return of
the installment payment plan is always equal to the average of the borrowing and
lending rates in the financial market (P1 /P0 = 1 + (r a + r b)/2). The straight down-
ward sloping line in the figure represents the case of perfect financial markets, that
is, r a = r b, which is equal to the internal rate of return of the installment plan. We
do not plot the cones defining liquidity-constrained regions in the figure. It is not
difficult to see, however, that regions where assets are zero (contingent on one
or more of the possible choices open to the consumer) become larger as financial
market access becomes more difficult. In the figure, financing terms for durable
good purchase become relatively more favorable compared to unconditional loan
rates, and the nonlinear character of the consumer’s choices becomes more and
more pronounced. Thus, as borrowing becomes more expensive, customers whose
total resources are tilted toward the future need to be richer in order to consider
purchasing the good.
Additional qualitative features of the solution are illustrated in Figure 7, where
for a given k, we keep P0 , r a , and r b constant but vary P1 . Not surprisingly, as
P1 increases the purchase-indifference locus shifts outward, that is, the consumer
needs to be richer in order to find it optimal to purchase the durable good. This is
more pronounced for configurations of the endowment that are relatively tilted
toward the future: Any change of the delayed payment terms for the good (as long
as the internal rate is larger than ra ) is irrelevant for consumers who have positive
FIGURE 7
CONSUMER PURCHASE DECISIONS: IMPLICATIONS OF DIFFERENT INSTALLMENT-CREDIT INTEREST RATES
assets and buy the durable with cash. Hence, if installment payment terms become
more favorable, customers with steep lifetime-resource profiles can be induced to
buy the durable on credit.
4.1. A Dealer’s Optimal Pricing of Financial Terms. The simple qualitative in-
sights outlined in the Introduction, and the more precise quantitative perspective
offered by our modeling approach, can be brought to bear on various aspects of
reality and of the data available to us. In this section, we show that the incidence
of dealer subsidies can be explained by the relationship between customers’ will-
ingness to pay and inclination to borrow—which are likely to be heterogeneous
across goods, but are not easily measured—and also by the structure of interest
rates, for which data on heterogeneous financial market conditions are available
in our empirical implementation.
We consider the case of a monopolist dealer who has a specific durable good
available for sale. The basic insights generalize to other models with imper-
fect competition as long as there is scope for price discrimination. Normalizing
marginal cost to zero with no loss of generality, the objective of the dealer’s pricing
decisions is defined as follows:
P1
(11) R(P0 , P1 ) = P0 D0 (P0 , P1 ) + D1 (P0 , P1 )
1 + rf
where D0 (P0 , P1 ) is the quantity purchased on a cash basis at price P0 , and D1 (P0 ,
P1 ) is the quantity purchased on a credit basis. For each customer who purchases
the good on credit, the dealer receives from the bank the customer’s installment
payment P1 , discounted at rate rf . In the model, as in reality, the bank handles the
financial side of the durable purchase transaction, and the dealer faces no default
risk.
Suppose the dealer is faced by a heterogeneous population of potential cus-
tomers that behave like the one characterized in the previous section. The param-
eter k, of course, is in general different across consumers as well as across goods.
No additional insight would be afforded by making that heterogeneity explicit,
however, over and above the implications of consumer heterogeneity in terms
of current and future financial resources. Formally, we suppose that purchasing
power is distributed across consumers according to the bivariate density function
f (W, Y).
Let Y = χ (W; P0 , P1 , k, ra , rb) denote the upward-sloping cash–credit indiffer-
ence locus of Figure 5, and let Y = π(W; P0 , P1 , k, ra , rb) denote the downward-
sloping line of indifference between purchasing and not purchasing the good in
that and other figures of the previous section. As shown above, the future-resource
level Y identified by each of these loci depends on current resources W, as well
as on the cash and credit prices, on the specification of tastes (parameterized by
k in the model), and on the financial market environment (parameterized by r a
and r b).
If ω denotes the current-resource coordinate of the two schedules’ intersec-
tion point, implicitly defined by χ (ω; P0 , P1 , k, ra , rb) = π (ω; P0 , P1 , k, ra , rb), the
quantity sold on a cash basis can be expressed as an integral over the appropriate
region of the previous section’s figures:

∞ χ (W;...)
(12) D0 = f (W, Y) dY dW
ω π (W;...)
Similarly, the quantity sold on credit is given by:

∞ ∞ ω ∞
(13) D1 = f (W, Y) dY dW + f (W, Y) dY dW
ω χ(W;...) 0 π (W;...)
The model features three distinct interest rates, r a < rf < r b. Since “the bank”
(or financial sector) is the counterpart of the consumers’ deposit and borrowing
relationships, the wedges rf − r a > 0 and r b − r a > 0 reflect intermediation costs
and nonrepayment risk, left unmodeled for this article’s purposes. The wedge
r b − rf > 0 between the interest rates charged by the bank on unrestrained con-
sumer borrowing and on durable good installment payments reflects differential
transaction costs and repayment behavior, through the selection effects outlined
in the Introduction. It would be interesting in future research to account for these
mechanisms explicitly, in particular in the consumers’ optimization problems.
If the cash and credit prices are related according to P1 = (1 + r f )P0 , then the
dealer’s revenues are not affected by the proportion of cash and credit sales. Be-
sides choosing the overall level of the item’s price, however, the dealer can also
choose to set the installment price as a different ratio of the cash price. Price
discrimination is optimal whenever feasible, and it is feasible in the situation we
consider because the utility derived from the good being purchased is different
for groups of consumers with different intertemporal rates of return. Dealers do
not observe individual income profiles and willingness to pay (in reality, the terms
of the contract are posted irrespective of individual customer characteristics), but
maximize profits by setting prices that sort consumers according to whether the
cash or credit payment is more attractive to them. The information needed to es-
tablish the optimal discriminating pricing structure in our model is not qualitatively
different from that needed to implement optimal decisions in other monopolistic
pricing models (see Dana, 2001, and references therein). In our application, it
includes information regarding the customer pool’s need of and access to credit:
It is not unreasonable to suppose that dealers are able to base their decision on
relevant aggregate statistics, as well as on trial-and-error methods.
Indeed, it will be generally optimal for the dealer to exploit the opportunity to
price discriminate among customers. As long as
P1
1 + ra < < 1 + rb
P0
different sets of consumers are attracted by cash and credit purchases. Hence, a
dealer faced with a population of potential customers that is heterogeneous across
the two dimensions of the figures above (current and future purchasing power)
should in general find it optimal to charge different prices to the subsets of that
population that are attracted by cash and credit purchase terms. Of course, if it
were the case that ra = rb = r f = r̄ , then the terms of durable good financing deals
would be unable to discriminate among customers. In the face of this perfect-
financial-markets configuration, if the dealer offered favorable financing terms
(P1 /P0 < 1 + r̄ ), then all customers would take advantage of them, and the dealer’s
interest rate subsidy would be equivalent to a lower price. Even less interestingly,
if the dealer tried to charge his credit customers more than the market interest
rate (P1 /P0 > 1 + r̄ ), then no consumer would accept such unfavorable terms, and
those who prefer to devote future resources to the durable purchase would simply
borrow on the market and pay cash. In reality, consumers’ borrowing and lending
rates are different: Qualitatively, it will not be surprising to find that the wider
the spread between them, the more pronounced are incentives for sellers to price
discriminate.
We illustrate the model’s implications with numerical computations based on a
simple parameterized model. We suppose that the population’s distribution of cur-
rent and future resources is well approximated by a bivariate normal distribution
over the relevant region (this distribution attaches positive probability to nega-
tive levels of W and Y, which however are irrelevant because they can never be
associated with purchase decisions). Although the qualitative point could be illus-
trated under simpler distributional assumptions, the normal distribution provides
a suitable theoretical basis for empirical work. Its variance, mean, and correlation
parameters can be varied very flexibly, and extensive numerical experimentation
confirmed the robustness of the characterization results discussed below.
On the basis of the previous section’s analysis of consumers’ choices, any price
structure such that 1 + ra ≤ P1 /P0 ≤ 1 + rb has nontrivial implications for the
quantities demanded from dealers on a cash and credit basis. In the data we
observe, however, the dealer either fully subsidizes or does not subsidize at all.9
Hence, we focus on the case
P1 = P0 (1 + ra )
when the dealer subsidizes. We compare profits in this case with profits when
dealers do not subsidize, in which case
P1 = P0 (1 + r f )
The optimal choice of whether to subsidize depends on the discrete comparison

between maximum profits in the two cases. Within each case, maximum profits can
9 As mentioned in footnote 2, only very few installment purchases are reported to be partially
financed by dealers. Unconstrained choices could make a negative subsidy optimal for dealers, whereby
P1 /P0 > 1 + r f . Dealers can mark up the cost of their customer’s credit when banks are not in direct
contact with customers. In some segments of the Italian data set we analyze, banks may pay fees to
dealers, but information on incidence of that phenomenon is not available to us.
be computed evaluating numerical counterparts to the demand expressions (12)

and (13) for every pricing choice by the dealer. Our program then determines the
optimal one by numerically integrating the dealer’s objective function (11) and
performing a grid search over the relevant region of P0 and P1 for each set of
parameter values, updating the trial solution values according to the steepness
and direction of the objective function’s gradient.
Figure 8 reports the difference between the profits when dealers “subsidize
completely” and “pass on all finance charges” as a function of the spread between
customers’ borrowing and lending rates. We increase the spread symmetrically
around rf = ra = rb = 0.16 (with 21 grid points each for rf and the spread), for
a few values of rf . The difference in profits is a hump-shaped function. When
the spread is small, there is little scope for price discrimination. As the spread
increases, subsidization makes more of a difference in terms of possibly desirable
price discrimination. However, as ra falls relative to rf , subsidizing becomes more
expensive. Thus, the effect of an increase in the spread on dealer subsidies becomes
less pronounced, and eventually decreasing. The smaller effect also reflects the
fact that at high levels of the spread already most of the customers are liquidity
constrained, and an additional increase of the spread makes little difference.
FIGURE 8
THE DIFFERENCE IN PROFITS WHEN SUBSIDIZING AND NONSUBSIDIZING AS A FUNCTION OF THE SPREAD (IN
BASIS POINTS). PARAMETER k IS SET AT 0.35; CURRENT RESOURCES W HAVE MEAN 3.5 AND FUTURE
RESOURCES Y HAVE MEAN 4; THE STANDARD DEVIATION IS 2 FOR BOTH, AND THEIR CORRELATION IS 0.5
It is also possible to characterize the implications of increasing rf starting from

r a , going toward r b. This is also hump shaped: in the figure, a higher rf shifts
the difference in profits upward and to the left, that is, dealer subsidies become
relatively more profitable for a given spread, for relatively small interest rate
levels and spreads. In that region, the cost of subsidizing credit is relatively small,
and a higher rf increases the scope for price discrimination. Initially, subsidizing
means charging almost exactly the same price pattern and it is not surprising
that it makes little difference to profits (in fact, the dealer might like to charge a
higher borrowing rate than r f ≈ ra since r b is much higher). As rf increases, the
large spread gives scope for price discrimination, and once subsidizing becomes
attractive, a larger rf actually makes it more attractive (because it implies a more
pronounced change in the intertemporal price pattern). However, subsidizing also
becomes more costly, because the dealer has to absorb all charges. At some point,
the latter effect dominates, and the profit difference becomes a declining function
of rf .
In reality, the character of the distribution of current and future resources across
the customer population also affects the scope for price discrimination (in the limit
case of no variation across the population in the relevant respects, the dealer could
not possibly discriminate among them). However our data, which we analyze
empirically next, offer no information regarding this aspect. Instead, we focus
on the model’s intuitive insights into the role of financial market conditions in
determining dealer subsidies’ desirability and cost, and proceed to seek empirical
support for them in a controlled-regression framework.
5. EMPIRICAL EVIDENCE
As illustrated in Figure 8 and discussed above, our theory predicts that the dif-
ference between profits in the two cases where the dealer does and does not fully
subsidize should be hump shaped in relation to both the financing rate rf and
the financial market spread ≡ rb − ra . The descriptive evidence of Section 2
indicated that such a nonlinear relationship is broadly consistent with patterns of
covariation over time and across Italian regions. In what follows, we seek further
support for our model by also controlling for (broad) durable goods market and
financial contract characteristics. We construct a measure for the bank’s financing
rate rf and use it along with a measure for the spread in formal models of the
probability of dealer subsidies at the credit application level. We discuss econo-
metric issues resulting from measurement error, endogeneity, omitted variables,
and clustered errors.
A possible empirical representation of the theory’s implications is in the form
(14) s∗ = Xγ + α1r f + α2r 2f + α3r 3f + β1 + β2 2 + β3 3 + µ(r f ) + u
where s∗ is the dealer’s inclination to subsidize as a function of variables, on the

right-hand side, pertaining to a specific microeconomic observation. We observe
the choice s of no (s = 0) or complete (s = 1) subsidization, and the parameters
of equations in the form (14) can be estimated by standard binary-choice mod-

els relating Pr(s = 1) to the determinants of the latent variable s∗. In order to
detect evidence in favor or against a hump shape, the specification includes a
third-degree polynomial in both rf and . We will focus in particular on estimates
of the parameters αi , βi , and µ. The vector X contains observable characteristics
that we should control for, in particular item indicators and characteristics of the
financial contract as determinants of the relationship between customers’ willing-
ness to pay and their inclination to borrow. Additional controls for the tastes and
lifetime income patterns of the potential customer population could be included
as well, but it would be difficult to offer a structural interpretation of their coeffi-
cients. The parsimonious regressions we report do not include demographic and
other customer characteristics. However, we have checked and discuss below the
robustness to their inclusion of the estimates we are interested in.
Estimation of Equation (14) needs to take into account potential endogeneity
of regressors. We simply assume away general equilibrium feedback effects from
the dealer subsidization decision on the financial market rates r a and r b, and
view as an exogenous variable. The durable good financing rate, however, is
determined by the interaction between dealers and the banking sector, which
we leave unmodeled but which occurs at the same microeconomic level as the
decisions we model. Hence, we may want to instrument rf . Since that rate is not
directly observed in the data when the customer does not pay interest, before
estimating the empirical model (14) of dealer subsidization, we need to construct
a measure for the bank’s financing rate on the basis of the predictions of interest
rate regressions.
5.1. Interest Rate Regressions. When the installment purchase is subsidized,

the financial rate of interest we observe in the data is zero. We do observe the
interest rate charged by banks on installment credit when the dealer does not
subsidize durable good purchases and in the case of personal loans.10 We estimate
a model of interest rate determination on that subsample: Selection effects are
not an issue if the bank’s financing rate is (to a first approximation) not directly
dependent on dealers’ price discrimination incentives.11 We run and report in
Table 4 separate regressions for each item category, in the form
(15) r f = Zη + ε
where the vector Z includes a personal-loan dummy and its interaction with some
loan characteristics; time, region, and quarter dummies and their interactions
(the reference case is an installment payment contract issued in Lazio in the third
10 The intended purchase is recorded in the personal-loan observations. In 7 of the 13 item categories
we consider, personal loan observations are numerous enough to allow estimation of the interest rate
surcharge entailed by customers dealing with the bank directly rather than through a dealer.
11 Estimation of a fully simultaneous equation system by Maximum Likelihood would be compu-
tationally demanding and would only be possible if we were willing to rely on specific functional form
assumptions on the errors’ distribution. Nonlinearities and unobservabilities could then be accounted
for by integrating out over that distribution, but the results would be sensitive to the latter’s likely
misspecification.
TABLE 4
INTERNAL RATES OF RETURN, INSTALLMENT CREDIT AND PERSONAL LOANS
Household Home
Item White Goods Appliances Brown Goods Computers Furniture Telephony (Used) Cars New Cars Motorcycles Sport & Leisure Maintenance Air-conditioning Other
No. of observations 4419 8460 16156 3722 8238 17982 5059 6883 5553 1201 1234 216 2941
Adj. R2 0.4554 0.3803 0.4415 0.4130 0.5583 0.5020 0.7435 0.6140 0.6754 0.6958 0.8895 0.0918 0.6475
Coeff. t-value Coeff. t-value Coeff. t-value Coeff. t-value Coeff. t-value Coeff. t-value Coeff. t-value Coeff. t-value Coeff. t-value Coeff. t-value Coeff. t-value Coeff. t-value Coeff. t-value
Contract Characteristics
Log(loan size in 0.0062 2.50 0.0007 0.54 −0.0068 −7.25 −0.0151 −7.79 −0.0209 −24.12 0.0024 2.08 −0.0166 −17.64 −0.0038 −4.92 −0.0072 −6.11 −0.0159 −6.94 −0.0189 −7.28 −0.0155 −1.03 −0.0138 −8.67
LIT)
Duration in months −0.0014 −4.19 −0.0001 −0.95 0.0002 1.32 −0.0010 −5.78 −0.0003 −4.77 −0.0008 −5.24 0.0000 0.43 −0.0001 −2.45 −0.0001 −0.73 −0.0001 −0.37 −0.0001 −2.16 0.0011 0.82 −0.0006 −4.00
Insurance∗ 0.0990 13.95 0.0780 15.66 0.1105 36.00 0.1160 10.24 0.0187 4.29 0.1050 26.84 – – 0.0222 4.03 0.0413 10.39 0.0973 10.55 0.0768 9.11 0.0209 0.62 0.0886 12.61
Pay by bank∗ −0.0086 −3.52 −0.0090 −4.61 −0.0057 −4.58 −0.0073 −3.61 −0.0084 −6.94 −0.0056 −4.74 −0.0087 −9.61 −0.0030 −5.69 −0.0029 −2.18 −0.0112 −3.38 −0.0099 −7.38 −0.0035 −0.25 −0.0126 −5.69
Dealer–bank communication
Fax 0.0190 3.86 0.0055 1.61 0.0155 5.81 0.0029 0.97 −0.0140 −8.79 0.0093 5.25 −0.0106 −3.16 0.0013 1.34 0.0000 0.01 −0.0216 −4.03 −0.0152 −2.35 −0.0488 −2.03 0.0107 3.73
Phone 0.0127 3.03 0.0023 0.83 0.0105 5.44 0.0096 2.72 −0.0054 −3.28 0.0068 3.91 −0.0116 −3.45 – – 0.0071 5.02 −0.0251 −6.52 −0.0079 −1.24 0.0290 0.57 0.0244 8.92
Other 0.0271 2.43 0.0146 1.65 0.0121 2.05 0.0181 2.84 −0.0016 −0.56 0.0042 0.76 −0.0127 −3.36 0.0027 2.17 0.0040 1.21 −0.0223 −3.34 −0.0083 −1.28 0.0340 0.38 0.0197 4.67
Personal loan
Personal loan (PL) – – – – – – – – 0.0958 1.98 – – 0.3668 10.31 0.2968 8.98 0.2598 2.64 0.1500 1.85 0.1459 3.32 – – 0.1975 3.65
PL × log loan size – – – – – – – – −0.0061 −2.00 – – −0.0200 −9.17 −0.0164 −8.25 −0.0166 −2.65 −0.0093 −1.77 −0.0085 −3.05 – – −0.0122 −3.41
PL × insurance – – – – – – – – −0.0235 −4.43 – – −0.0180 −3.54 −0.0334 −6.80 0.0177 1.10 −0.0591 −6.38 −0.0307 −6.10 – – −0.0603 −9.78
PL × quarter – – – – – – – – −0.0014 −2.80 – – −0.0022 −3.83 −0.0027 −6.54 −0.0021 −1.96 −0.0020 −1.99 −0.0036 −6.20 – – −0.0050 −7.76
dummy
Regions
Piemonte 0.0257 2.79 0.0320 4.65 0.0418 8.91 −0.0074 −0.51 −0.0072 −1.25 0.0394 6.80 −0.0197 −3.61 0.0001 0.02 −0.0224 −3.73 −0.0128 −0.92 −0.0150 −1.19 0.0308 0.42 −0.0214 −2.36
Val d’Aosta 0.0598 1.72 0.0055 0.31 0.0638 4.72 −0.0816 −2.25 −0.0232 −1.75 0.0235 1.87 −0.0221 −3.16 0.0019 0.34 −0.0400 −1.48 – – −0.0207 −0.82 – – 0.0000 0.00
Lombardia −0.0067 −0.76 −0.0021 −0.32 0.0161 3.49 −0.0142 −1.00 −0.0073 −1.33 0.0104 1.84 −0.0156 −2.90 0.0036 0.84 −0.0081 −1.48 −0.0094 −0.74 −0.0208 −1.66 0.0725 1.13 −0.0267 −3.02
Trentino/Alto −0.0001 −0.01 −0.0025 −0.22 0.0168 2.52 −0.0121 −0.72 −0.0401 −4.86 0.0156 2.00 −0.0171 −2.11 −0.0001 −0.01 −0.0181 −1.69 −0.0279 −0.97 – – – – −0.0374 −1.75
Adige
Veneto −0.0208 −2.29 −0.0209 −3.04 −0.0025 −0.52 −0.0232 −1.58 −0.0102 −1.73 −0.0013 −0.22 −0.0178 −3.06 0.0011 0.25 −0.0242 −4.15 −0.0243 −1.95 −0.0070 −0.53 0.0404 0.63 −0.0286 −2.97
Friuli-Venezia- 0.0218 2.09 0.0278 3.48 0.0325 5.82 −0.0088 −0.57 −0.0232 −2.53 0.0034 0.54 −0.0215 −3.13 −0.0027 −0.50 −0.0343 −5.19 0.0153 0.69 −0.0181 −1.33 0.1527 1.86 −0.0285 −2.29
Giulia
Liguria −0.0234 −2.36 −0.0120 −1.59 0.0021 0.38 −0.0142 −0.93 −0.0182 −2.89 0.0089 1.34 −0.0141 −2.28 −0.0061 −1.29 −0.0079 −1.16 0.0063 0.36 −0.0185 −1.39 – – −0.0280 −2.49
Emilia Romagna −0.0269 −2.91 −0.0177 −2.58 −0.0070 −1.47 −0.0217 −1.50 −0.0244 −4.31 0.0078 1.36 −0.0208 −3.66 0.0039 0.88 −0.0245 −4.12 −0.0127 −0.92 −0.0143 −1.10 0.0171 0.26 −0.0355 −3.77
Toscana −0.0178 −4.05 −0.0086 −2.83 −0.0171 −7.44 −0.0262 −6.36 −0.0038 −1.62 −0.0082 −3.85 −0.0134 −6.69 −0.0052 −3.98 −0.0344 −14.93 −0.0164 −2.77 −0.0049 −1.55 −0.0157 −0.48 −0.0165 −3.60
Umbria −0.0070 −0.61 −0.0060 −0.81 −0.0072 −1.46 −0.0078 −0.94 0.0038 0.59 0.0010 0.24 −0.0063 −1.49 0.0004 0.11 −0.0203 −3.80 0.0116 1.06 0.0051 0.68 0.2453 2.61 −0.0044 −0.51
Marche −0.0204 −3.89 −0.0221 −5.14 −0.0145 −5.05 −0.0067 −1.13 −0.0237 −5.73 −0.0158 −5.25 −0.0099 −2.85 −0.0052 −1.70 −0.0317 −7.34 −0.0223 −2.27 0.0027 0.46 −0.0256 −0.59 −0.0034 −0.50
Abruzzo −0.0070 −0.77 −0.0077 −1.03 −0.0041 −0.81 −0.0170 −1.09 −0.0084 −1.35 −0.0034 −0.63 −0.0267 −4.39 −0.0138 −3.60 −0.0076 −1.39 −0.0271 −1.67 −0.0040 −0.34 – – −0.0208 −1.95
Molise −0.0222 −1.47 −0.0267 −2.25 0.0008 0.09 −0.0305 −1.67 −0.0129 −0.98 −0.0047 −0.43 −0.0166 −1.22 −0.0187 −2.33 −0.0281 −1.89 −0.0531 −2.90 −0.0014 −0.08 – – −0.0531 −2.09
Campania −0.0056 −0.67 −0.0123 −1.96 0.0004 0.08 −0.0028 −0.19 0.0034 0.62 −0.0002 −0.03 −0.0159 −3.26 −0.0269 −8.20 −0.0283 −6.47 −0.0170 −1.58 −0.0022 −0.20 0.0112 0.17 −0.0091 −1.07
Puglia 0.0037 0.45 −0.0006 −0.09 0.0040 0.89 −0.0097 −0.63 0.0072 1.29 0.0089 1.84 −0.0284 −5.57 −0.0289 −8.46 −0.0303 −6.25 −0.0224 −1.92 −0.0089 −0.78 0.0192 0.30 −0.0154 −1.68
Basilicata −0.0201 −1.44 −0.0110 −0.92 0.0122 1.87 −0.0109 −0.63 −0.0012 −0.12 −0.0022 −0.36 −0.0192 −3.29 −0.0215 −4.88 −0.0160 −2.36 −0.0172 −0.80 −0.0001 −0.01 – – −0.0262 −1.87
Calabria 0.0012 0.14 0.0033 0.49 −0.0035 −0.73 −0.0044 −0.28 0.0077 1.33 −0.0026 −0.52 −0.0186 −3.78 −0.0244 −7.37 −0.0195 −4.10 −0.0199 −1.73 −0.0017 −0.15 0.0322 0.46 −0.0137 −1.50
Sicilia −0.0098 −1.20 −0.0159 −2.60 −0.0097 −2.18 −0.0162 −1.06 −0.0051 −0.96 −0.0136 −2.84 −0.0258 −5.38 −0.0260 −8.05 −0.0316 −7.32 −0.0195 −1.79 −0.0030 −0.27 0.0158 0.25 −0.0103 −1.20
Sardegna 0.0090 1.00 0.0221 3.28 0.0127 2.58 −0.0161 −1.01 −0.0171 −3.04 0.0148 2.82 −0.0277 −5.42 −0.0243 −7.16 −0.0317 −5.94 −0.0365 −2.91 −0.0037 −0.33 −0.0187 −0.25 −0.0177 −1.95
Intercept 0.2629 7.98 0.3116 16.71 0.4135 31.60 0.5548 17.60 0.6118 47.50 0.2906 18.40 0.5442 37.80 0.2721 21.90 0.4079 23.71 0.5446 17.00 0.5910 15.56 0.4053 1.85 0.5037 22.15
SOURCE: Findomestic Banca, own calculations.
NOTE: Time dummies, their interaction with insurance, and the interaction of the region dummies with time and insurance, respectively, are included in the regression, but not reported. Goods are classified as follows: furniture includes all sorts of home furnishing, among
which are living and bedrooms and (modular) kitchens; motorcycles includes motorcycles and scooters; used cars include cars, motorhomes, and caravans and they may or may not be second-hand; new cars include cars, motorhomes, and caravans that are classified as
being new; white goods include fridges, freezers, washing machines, dishwashers; household appliances are those not classified as white or brown goods; brown goods are consumer electronics like TV sets, VCRs, radios, cameras, etc., excluding home computers; computers
and telephony cannot further be subdivided. ∗ : Dummy variable; –: not enough variation to identify the coeffient.
quarter of 1995); the size and duration of the loan; whether or not the customer
has purchased insurance toward repayment in case of severe sickness and its
interaction with time and space; and whether or not payments are made via a
bank account. We also control for the technology the dealer uses to communicate
the application data to the bank, with online technology as the reference case.12
The latter set of dummies is an indicator of the nature and quality of the business
relation between dealer and bank. Importantly, we will exclude it (and various
interaction terms) from the vector of variables X in Equation (14). Such exclusion
restrictions are needed to identify a component of financing-rate variation that is
independent of observed subsidization. The dealer–bank relationship’s role as an
instrumental variable can indeed be given a natural economic interpretation in our
model. It seems plausible to assume that this relationship, which is at least partly
predetermined, affects the subsidization choice only through the interest rate, but
not directly. In our model, the dealer’s decision to post subsidized rates is based
on characteristics of the customer pool, as captured by the variables included in
X. These are of course time varying and related to the item and region variables
that also explain financing-rate variation. It is much harder, however, to see how
they could be related to variation of the dealer–bank relationship after controlling
for those variables. Dealer–bank communication technologies may affect interest
rates through transaction costs or through more complex and history-dependent
effects; so it is difficult to state theoretical implications regarding their coefficients’
signs. For our purposes, it suffices to find in Table 4 that these dummies are a
statistically significant source of variation in interest rates across observations
with similar characteristics as regards the relationship between dealers and their
customers.
Although the data strongly reject pooling of observations across item categories,
the pattern of coefficient signs is broadly similar across many columns of Table 4. In
order to interpret the magnitude of the coefficients, note that the included interac-
tions of the personal loan dummy with contract characteristics and time dummies
are quite significant, indicating that (e.g.) fixed costs of credit provision play a dif-
ferent role across personal and installment loans as well as across different item
categories. The error term ε does not have a structural interpretation, but for de-
scriptive purposes it is comforting to find that the model yields large R2 values for
the largely cross-sectional data we analyze. Consistently, with our modeling frame-
work, where r b > rf , the personal-loan dummy is associated with a higher borrow-
ing rate for all item groups where the test is possible, and typically the effect is
significant.
12 In preliminary regressions, not reported, we included all available contract and customer infor-
mation as explanatory variables, finding that interest rate variability is mostly accounted for by time,
region, and type of item purchased, plus a limited number of contract characteristics (such as loan
size). Given that in reality the terms of the financing deals are posted and thus do not differ across
customers, it is not surprising that variables that relate to the individual loan applicant are much less
relevant: Creditworthiness does not affect the interest rate charged by the bank and individual char-
acteristics become known (and may only affect whether credit is granted) after the terms of financing
are decided.
5.2. Determinants of Dealer Subsidies. Estimation of the interest-rate regres-

sion (15) allows us to compute predicted values r̂ f = Zη̂ for both subsidized and
nonsubsidized installment payment observations. Since these predicted values
are noisy measures of the true latent financing rate, measurement error will
tend to bias the estimated coefficients, typically toward zero (the “attenuation
bias” case, as discussed in, e.g., Aigner et al., 1984). To estimate the relationship
in (14) we impute to all subsidized and unsubsidized observations in the same
item/region/quarter cell the average within that cell of the predicted values, r̂ f ,
from regressions similar to those reported in Table 4 but estimated on the subsam-
ple obtained by excluding personal loan applications, which are never subsidized.13
As to the borrowing/lending spreads in the financial market, denoted in
(14), we rely on the substantial variation across regions and over time of the
interest rates reported by the Bank of Italy (and discussed in our Section 2 above).
Those data, which are only available from 1996/IV, are likely to approximate
very imperfectly the rate spread applying to households. Lacking better data, or
suitable instruments, we include them as an indicator of variation in the structure
of financial markets across both time and regions, after controlling for pure time
and region effects. Interpretation of the results, however, will need to take into
account measurement-error bias.
In Table 5, we report not only the coefficients of a linear probability model in
the form of Equation (14), but also the marginal effects from a probit model, that
is, the change in the probability of subsidization when the corresponding regressor
increases infinitesimally (or, in case of a dummy variable, changes from zero to
one) keeping all other regressors at their mean values.14 The two specifications
indicate qualitatively and quantitatively similar effects of the regressors on subsi-
dization, and both fit the data well in terms of R2 . A logit specification also gives
essentially identical results.
The regressors denoted X in (14) are a vector of contract characteristics, item
dummies, and region and quarter dummies (not reported). Inclusion of item
dummies and contract characteristics is important to control for retail market
and financial contract heterogeneity. Although their coefficients do not have a
structural interpretation, some findings deserve to be mentioned. The relative
sizes of the item effects are similar to those of the raw frequencies reported in
Table 2, but contract characteristics are empirically important controls in these
regressions. The sign and size of these coefficients is theoretically ambiguous, as
the amount of credit extended affects both the cost of subsidization and the ben-
efits of successful segmentation of the customer population. Empirically, larger
loans and shorter-maturity loans are more likely to be subsidized. Item dummies
13 Averaging mitigates the impact of measurement error on the resulting rf proxy, while also elim-
inating the variation of the true latent variable within the cell. The latter may or may not be a source
of additional bias, depending on whether variation of financing rates within the cell is driven more
strongly by exogenous factors or by joint endogeneity of rf and of the dealer’s subsidy choice.
14 The predicted values of the linear probability model have similar meaning, but a probability
interpretation would be valid only if the specification was fully saturated and all regressors were
dummy variables. The probit regression is more appropriate if the functional form of the error term’s
distribution is approximately normal, which may but need not be the case.
TABLE 5
DEALER SUBSIDIZATION: ESTIMATES
(a) Probit (b) Linear Model

Cluster- Cluster-
Marginal Adjusted Coefficient Adjusted
Effect t-Value t-Value Estimate t-Value t-Value
Item financed (ref.: new cars)
White good 0.4897 16.97 5.80 0.2799 13.88 4.58
Household appliance 0.4300 15.17 4.97 0.2279 11.53 3.57
Brown good 0.4315 15.83 5.21 0.2414 12.39 3.90
Computer 0.0571 2.42 0.84 −0.0419 −2.37 −0.76
Furniture 0.5036 22.81 7.50 0.2774 19.13 5.89
Telephony 0.3008 10.71 3.45 0.1350 6.63 2.04
(Used) cars 0.0920 5.27 2.10 −0.0357 −3.61 −1.94
Motorcycles 0.5914 32.74 11.75 0.3742 32.54 11.08
Sport and leisure 0.4614 14.66 5.35 0.2418 11.07 3.86
Home maintenance 0.4548 8.92 4.23 0.2169 5.74 2.83
Air-conditioning 0.6134 19.29 8.27 0.4126 17.18 7.42
Other 0.4555 15.83 5.22 0.2463 12.38 3.84
Contract characteristics
Log amount 0.1364 43.03 25.29 0.1031 38.06 23.99
Duration (months) −0.0307 −76.36 −19.44 −0.0167 −72.54 −36.10
Insurance −0.3268 −79.36 −40.61 −0.3474 −93.92 −35.30
Pay via bank 0.0953 24.83 20.63 0.0850 26.34 21.52
Interest rate spread polynomial
Linear −0.1717 −2.17 −1.00 −0.1883 −2.89 −1.19
Quadratic 0.0262 1.88 0.90 0.0282 2.49 1.06
Cubic −0.0017 −2.09 −1.01 −0.0017 −2.65 −1.14
Interest rate level polynomial
rf, linear −4.5913 −4.31 −1.11 −2.1120 −2.61 −0.89
Quadratic 23.3460 5.25 1.32 13.0395 3.76 1.28
Cubic −38.8616 −6.69 −1.60 −23.8782 −5.14 −1.69
Level/spread interaction
Linear 0.0433 1.76 0.56 0.0535 2.95 0.96
NOTE: 93,899 observations. Probit pseudo-R2 = 0.2782. Linear model R2 = 0.2709. Probit marginal
effects of dummy variables refer to discrete change of regressor from 0 to 1. Intercept, region, and
quarter dummies (“fixed effects”) are included, but not displayed; “cluster-adjusted” t-values also
allow for heteroskedasticity within region-quarter cells.
are also significant, indicating that the interaction of willingness to pay and incli-
nation to borrow (within the population of potential buyers) is different across
durable good markets.
The main variables of interest are the proxies for the spread of the financial
market rates and for the financing rate. The spread varies only across region-
quarter cells and the level rf , imputed from the previous section’s regressions,
varies only across region-quarter-item cells. Since we control for item, region, and
quarter effects, identification of interest rate spread and level effects is achieved
by the interaction of time and space. The functional form we estimate is flexible
enough to let the data determine the shape of the relationship between these
variables and the occurrence of subsidization.
The point estimates are negative for the linear and cubic terms both for the
spread and the level rf , and positive for both quadratic terms. The estimated in-
teraction coefficient is positive. Since available proxies for rf and vary only across
clusters of observations rather than at the individual level, it may be important to
account for a nonstandard structure of error covariances when assessing the sta-
tistical fit of the results. Table 5 reports two sets of t statistics. The first column next
to the coefficient estimates reports statistics computed under the assumption that
after controlling for “fixed” region and time (quarter) effects, the model features
a diagonal homoskedastic variance–covariance matrix. The statistics in the next
column allow for arbitrary covariance structures within region-quarter “clusters.”
It is not surprising to see in the table that the coefficients appear much less signifi-
cantly different from zero on the basis of cluster-adjusted t statistics, which provide
a conservative basis for inference (see Wooldridge, 2003, for a recent discussion).
These statistics, however, are likely to be very conservative in our application.
Our financial market rate data are measured for regions/quarter (and, in the case
of the level, also item) aggregates. Imputing them to all observations within such
cells may induce correlation across them, but the presence of “fixed” effects in the
regression should eliminate that correlation. Within-cell heteroskedasticity may
still be a problem but there is no clear evidence of any heteroskedasticity in our
application (the Huber/Eicker/White corrected standard errors are very similar
to the unadjusted ones).
5.3. Testing the Theory’s Implications. We proceed to offer formal tests of

our theoretical restrictions. The solution of the model features a negative interac-
tion between the spread and the level of interest rates. In Table 5, the estimated
coefficient is positive instead, but less significantly different from zero than the co-
efficients of the polynomials in the two variables of interest. As regards the latter
estimates, our theoretical derivations indicate that the incidence of subsidization
should be a concave function of interest rate spreads and levels: Larger spreads
increase the scope for segmentation of the customer population, but also imply
a larger cost of offering subsidized credit, and the former effect tends to become
less important than the latter as spreads and levels of interest rates increase.
On the basis of the regression results of Table 5, we compute and report in Table 6
one-sided tests of concavity of the polynomial at the three quartiles of the rate level
or spread distribution.15 We find that dealer subsidization is a concave function of
both the rate level and the spread, at various points of their distributions. All point
estimates of measures of concavity are negative, as predicted by theory, except the
one computed at the first quartile of the spread distribution. To assess statistical
significance, we report both unadjusted and cluster-adjusted standard errors, and
the limit of a one-sided 95% confidence band. When this is negative, the data
deliver a statistically significant rejection of the (linear, or convex) relationships
ruled out by our theoretical model’s hump-shaped predictions. They always do so,
15 In terms of Equation (14), we test the null hypothesis H : 2α + 6α r ≥ 0 and a similar expres-
0 2 3 f
sion for . A suitable test statistic can be computed from the estimated variance–covariance matrix
and a matrix of linear restrictions at any value of the two variables.
TABLE 6
TEST FOR CONCAVITY OF SUBSIDIZATION
No Clustering Clustering
Concavity
Point Estim. SE Conf. Band SE Conf. Band
(a) Probit spread
1st quartile 0.0236 0.0222 0.0601 0.0477 0.1021
Median −0.0119 0.0107 0.0057 0.0235 0.0268
3rd quartile −0.0512 0.0192 −0.0196 0.0391 0.0130
rf, level
1st quartile −22.2328 4.4057 −14.9860 12.7340 −1.2872
Median −55.5081 4.4353 −48.2126 19.2321 −23.8742
3rd quartile −72.6370 6.2516 −62.3541 28.4963 −25.7648
(b) Linear model spread
1st quartile 0.0106 0.0059 0.0204 0.0146 0.0347
Median −0.0009 0.0029 0.0039 0.0074 0.0112
3rd quartile −0.0137 0.0048 −0.0058 0.0107 0.0040
rf, level
1st quartile −6.9548 1.0269 −5.2656 2.9141 −2.1412
Median −13.4565 1.2308 −11.4319 4.2612 −6.4177
3rd quartile −16.8033 1.7322 −13.9540 5.8894 −7.0750
NOTE: The table displays tests of concavity in both the spread and the level, based on results in Table 5.
Column “point estim.” gives the numerical value of the second derivative, evaluated at three different
quartiles of the respective distribution. See text, in particular note 14, for explanations. Columns “no
clustering” do not, and columns “clustering” do allow for heteroskedasticity within region-quarter
cells in regressions that already include “fixed” effects for those cells. “Conf. band” indicates the 95th
percentile of the distribution of the test statistic (one-sided test).
regardless of estimation and standard error computation methods, in the case of

the interest rate level.
There is less support for theoretical predictions as regards the spread, since
almost all tests fail to reject the linearity null at the 5% level. Even though test
statistics are negative at both the median and the upper quartile, statistically sig-
nificant evidence of concavity is found only in the case of the top quartile on
the basis of unadjusted standard error, whereas the conservative clustered stan-
dard errors deliver very wide confidence bands. Hence, the data do not exclude
a nonconcave relationship between the incidence of subsidization and the spread
between financial borrowing and lending rates. Of course, only very imprecise
measures of interest rate spreads are available to us. Since attenuation bias is a
likely implication of the resulting measurement error, it may not be surprising to
find that confirmation of theoretical effects is empirically more elusive as regards
spreads than in the case of interest rate levels, where we are able to compute and
employ more directly relevant (if imperfectly measured) data.
As to robustness, estimating the subsidization equation separately for each item
category delivers similar patterns for the coefficients of interest, offering more
support for the theory’s curvature prediction in some cases and less in others.
Experimentation with the specifications different from that reported in Tables 5
and 6 indicates that the results of interest are not driven by omitted-variable bias.
The benchmark specification of Table 5 is relatively parsimonious: It excludes cus-

tomer characteristics, their interaction with item dummies, as well as interactions
of loan size and item dummies. These variables are statistically significant when
included in the regression, but leave the results of interest unchanged. Omitting
quarter dummies also preserves the curvature results of interest. Omitting region
dummies leads to somewhat different results. The coefficients of interest are not
surprisingly more significant in this case, and the estimated interaction between
the interest rate spread and level turns out to be negative, as predicted by the-
ory. However, the test statistic for concavity is positive for spreads at the median,
and there is evidence of nonconcavity for rate levels at the first quartile only for
clustered standard errors.
6. CONCLUDING COMMENTS
We have motivated, proposed, and solved a model of interrelated purchase

and borrowing decisions when, on the one hand, wedges between borrowing and
lending rates in the financial market make consumers’ willingness to pay depend
on the relative as well as the absolute size of current and future funds and, on
the other hand, installment payment plans toward purchase of a specific item are
available at relatively favorable rates. Furthermore, we have rationalized in terms
of price-discrimination incentives the behavior of dealers who choose to bear the
financial cost of their customers’ credit purchases. Wedges between borrowing and
lending rates segment the population of potential customers into groups that are
more or less inclined to borrow. Hence, any relationship between such inclinations
and the overall willingness to pay induces dealers endowed with monopoly power
to charge different present-value prices to the two groups.
Although the model proposed is very much simplified, its closed-form solu-
tion and numerical experiments offer intriguing insights into subtle aspects of
real-life market interactions. Empirical analysis of a rich set of installment-credit
and personal-loan data offers considerable support for the assumptions and im-
plications of the proposed theoretical perspective. In particular, we have offered
evidence that heterogeneity across geographic, market-segment, and time dimen-
sions of the structure of borrowing, lending, and installment-plan interest rates has
nontrivial implications for the incidence of dealer-subsidized credit in our data.
From a measurement perspective, our model implies that aggregate financial
market development indicators might hide important interactions between more
detailed imperfections in unconditional and conditional credit markets. Moreover,
the modeling perspective we propose may also have interesting implications for
the pattern of durable and nondurable expenditure by individual consumers. For
example, consumers with fast-increasing income patterns may not find it optimal
to borrow in order to finance current nondurable consumption (especially when
such borrowing is expensive and/or severely constrained). They should, however,
be inclined to tilt their consumption bundles toward the kinds of durable goods
that feature favorable financing arrangements. The resulting interaction between
durable stocks and expenditure flows on the one hand, and financial market im-
perfections on the other, is interestingly different from that featured in Alessie
et al. (1997) and other related contributions. It might be possible to study such
phenomena combining detailed information on the characteristics of customers
who do apply for credit (and may or may not be offered a subsidy by dealers)
with the information on financial market access and durable expenditure patterns
available in representative data sets, such as the Bank of Italy survey studied by
Alessie et al. (2005) and Bertola et al. (forthcoming).
Further research could extend both our theoretical and empirical analysis in
fruitful ways. More comprehensive data would make it possible to obtain clearer
insights on the relationship between the incidence of dealer subsidies and financial
market imperfections. Our theoretical model provides a structural link between
the two and thus may provide guidance for further empirical analysis of such
data. In particular, more detailed data on the distribution of current and future
income in markets with different financial development might make it easier to
detect the mechanisms at work. Finally, in our model, heterogeneity of financial
market development is taken as exogenous, but they may well be endogenous to
higher-level economic interactions. Since market imperfections for unconditional
credit increase the scope for consumer-credit dealer subsidies, competition among
dealers and among credit institutions are likely to play a very important role in
determining the scope of financial market imperfections and their relevance for
different consumer groups.
APPENDIX
We first derive algebraic formulae defining the combinations of current and

future resources that
(a) make a consumer indifferent between purchasing or not purchasing the
durable good and
(b) make a consumer indifferent between purchasing the durable good with
cash or credit.
To organize the derivation of such indifference loci, it will be helpful to refer to
Figure A.1, which partitions the (W, Y) plane according to whether the durable
good is purchased with cash, credit, or not at all, and whether the resulting in-
tertemporal choices entail positive assets, negative assets, or a binding liquidity
constraint.
Within each of the 10 resulting regions, labeled with Roman numerals, the
possible lending, borrowing, and purchase choices are restricted in different ways.
Note that the regions of the points M, N, and Q mentioned in the main body of
the article correspond to regions I, IV, and X, respectively.
A.1. Exact Solution: Purchase Indifference. We proceed to list algebraic for-

mulae defining the combinations of current and future resources that make a
consumer indifferent between purchasing or not purchasing the durable good.
We omit details of the derivation, which are in general quite similar to those dis-
cussed in the article for regions I, IV, and X and always lead to quadratic equations.
The sign of the root in the solution is always uniquely determined by considering
FIGURE A.1
LABELING OF REGIONS IN THE MODEL’S RESOURCES SPACE
restrictions on the possible values of W and Y in each of the regions. Again taking
such restrictions into account, the slope of the indifference locus can be shown to
be weakly negative in all regions considered here. The indifference locus is contin-
uous at the boundaries of the regions: Values at all boundary points are reported
in the tables in Section A.2. It is also continuously differentiable, unless indiffer-
ence between cash and credit purchase also happens to obtain when a boundary
is crossed by the purchase-indifference locus.
Region I. From inspection of Figure A.1 it follows that in this region the con-
sumer has negative assets independent of the purchase decision. Credit purchase
is always at least weakly preferable to cash purchase because
Y + (W − C)(1 + rb) − P1 ≥ Y + (W − C − P0 )(1 + rb)
given that
P1
≤ (1 + rb)
P0
We compare the utility afforded by a credit purchase of the durable with the utility
the consumer obtains if he does not buy the durable. The endowments that make
the consumer indifferent between a credit purchase and no purchase lie on the
locus:
1

e2k P1
Y= − W (1 + rb)
e 2 k − 1 1 + rb
1
Region II. As is clear from inspection of Figure A.1, in this region the consumer
has no assets (is liquidity constrained) if the durable good is purchased on credit,
and has negative assets (borrows) otherwise.
The endowments that make the consumer indifferent between a cash purchase
and no purchase lie on the locus:
k

e2
Y = P0 k − W (1 + rb)
e2 − 1
The endowments that make the consumer indifferent between a credit purchase

Y = W(1 + rb)(2ek − 1) − 2 W2 (1 + rb)2 (e2k − ek) − ek(1 + rb)W P1
Region III. The asset position of the consumer depending on whether the
durable is purchased and, if so, on a cash or credit basis can again be inferred
easily from Figure A.1 in this and all other regions. The endowments that make
the consumer indifferent between a cash purchase and no purchase lie on the
locus:

2 2
Y = (1 + rb) P0 + W k − 1 + k W ek P0 + (1 − ek)W
e e
and those that generate indifference toward a credit purchase are defined by
ek
Y = P1
ek − 1
Region IV. From inspection of Figure A.1, it follows that in this region the
consumer does not hold assets independent of the purchase decision. The en-
dowments that make the consumer indifferent between a cash purchase and no
purchase lie on the locus:
ek
W = P0
ek −1
and those that generate indifference toward a credit purchase are defined as in
Region III.
Region V. The endowments in this region that make the consumer indifferent
to purchasing the good with cash are described by the same relationship as in
Regions I or II. Indifference to credit purchase occurs when
k
(1 + ra )W(1 − ek) + ek P1 (1 + rb) + e 2 (1 + ra )(1 + rb)((rb − ra )W + P1 )
Y=
ek(1 + rb) − (1 + ra )
Region VI. The endowments in this region that make the consumer indifferent
to purchasing the good with cash are described by the same relationship as in
Region III.

2 2
Y= P1 + W(1 + ra ) k − 1 + k W(1 + ra ) P1 e + (1 + ra )W(1 − e )
k k
e e
Region VII. The endowments in this region that make the consumer indiffer-
ent to purchasing the good with cash are described by the same relationship as
in Region IV. The endowments in this region that make the consumer indifferent
to purchasing the good with credit are described by the same relationship as in
Region VI.
Region VIII. The endowments that make the consumer indifferent between a
cash purchase and no purchase lie on the locus:
k ra −rb
(1 + rb) (1 + ra )(W − (W − P0 )ek + e 2 (1 + ra )(1 + rb) 1+rb
W + P0
Y=
(1 + ra )ek − (1 + rb)
k
e2
Y = P1 k − W(1 + ra )
e2 − 1
Region IX. The endowments that make the consumer indifferent between a
cash purchase and no purchase lie on the locus:

Y = (1 + ra ) 2e (W − P0 ) − W − 2 (W − P0
k
)ek (W − P0 )ek −W
The endowments in this region that make the consumer indifferent to purchasing
the good with credit are described by the same relationship as in Region VIII.
Region X. From inspection of Figure A.1, it follows that in this region

the consumer holds positive assets independent of the purchase decision. The
endowments that make the consumer indifferent between a cash purchase and no
purchase lie on the locus:
k

e2
Y= P0 k − W (1 + ra )
e2 − 1
Cash purchase is always at least weakly preferable to credit purchase because
Y + (W − C)(1 + ra ) − P1 ≤ Y + (W − C − P0 )(1 + ra )
given that
P1
≥ (1 + ra )
P0
A.2. Continuity. Here we report the values of the indifference locus at all
points where it crosses the boundaries of two regions. In all cases, the value is the
same whether it is computed with the above analytic expressions for either one of
the regions.
At the intersection points for the cash purchase–no-purchase–indifference lo-
cus, current resources W are given by the following expressions:
k1 k 1
1 P0 e 2 (1 + rb ) − P1 (e 2 − 1)
Region I–II: 2 1k
(e 2 − 1)(1 + rb )
1
1 e2k
Region II–III: P
2 0 1k
e2 −1
1 1
P0 e 2 k (1 + rb ) − P1 (e 2 k − 1)
Region II–V: 1
(2 + rb + ra )(e 2 k − 1)
Region II–VI: P1
rb − ra
only if k = 2 log 2 2P1 − PP0 (r
1
b − ra )
k
Region III–IV: P0 eke− 1
ek (2 + rb + ra )(P0 (1 + rb ) − P1 ) + 2P1 (1 + rb )
Region III–VI: +
√ 4(e − 1)(1 + ra )(1 + rb )+e (rb − ra )
k k 2
2 ((1 + ra )(1 + rb )ek P02 + P1 P0 ek (rb − ra ) + P12 (1 − ek ))(1 + rb )2

4(ek − 1)(1 + ra )(1 + rb ) + ek (rb − ra )2
k P1 + P0 (1 + rb )
Region III–VII: P0 eke− 1 only if k = log P0 (1 + ra ) + P1
k
Region IV–VII P0 eke− 1
1
1 e2k
Region V–VI: P
2 0 1k
e −1
2
√
ek (2 + ra + rb ) + 2 ek (1 + ra )(1 + rb )
Region VI–VIII: 4(ek − 1)(1 + ra )(1 + rb ) + ek (rb − ra )2
(1 + rb)P0
P0 (1 + rb ) 1 + rb
Region VI–IX: rb − ra
only if k = log 1 + ra
ek
Region VI–VII: P0 ek − 1
k
Region VII–IX: P0 eke− 1
√
− 4ek (1 + ra )(1 + rb ) + (1 + ra )(1 + rb ) + (1 + rb )2 − 2 ek (ra + 1)3 (1 + rb )
Region VIII–IX: 4(1 − ek )(1 + ra )(1 + rb ) + (rb − ra )2
P0
k 1
Region IX–X: 1
P 2e 2 − 1
2 0 12 k
e −1
Instead, at the intersection points for the credit purchase–no-purchase–

indifference locus, current resources W are given by
1 P1
Region I–II: 2 1
(1 + rb )(e 2 k − 1)
Region II–III: ek (1 + rbP)(e1 k − 1)
√
− 2 − (ra + rb ) − 2 ek (1 + ra )(1 + rb ) P1
Region II–V: (ra − rb )2 − 4(ek − 1)(1 + ra )(1 + rb )
+ rb
Region II–VI: P1
rb − ra
only if k = log 11 + ra
k
Region III–IV: Y = P1 eke− 1
P1
Region III–VI: (ek − 1)(1 + ra )
P1 (1 + rb ) + P0 (1 + rb )(1 + ra )
Region III–VII: P1
(ek − 1)(1 + ra )
only if k = log P1 (1 + ra ) + P0 (1 + rb )(1 + ra )
P1
Region IV–VII: (ek − 1)(1 + ra )
√
ek (2 + ra + rb ) + 2 ek (1 + ra )(1 + rb ) P1
Region V–VI: 4(ek − 1)(1 + rb )(1 + ra ) + ek (rb − ra )2
1
1 e2k
Region VI–VIII: P
2 1 1k
(e 2 − 1)(1 + ra )
P0 (1 + rb ) + ra )(1 + rb )
Region VI–IX: rb − ra
only if k = 2 log 2P0 2P0 (1 + r(1a )(1 + rb ) − P1 (rb − ra )
ek (2 + rb + ra )(P1 + P0 (1 + rb )) − 2(1 + ra )(1 + rb )P0
Region VI–VII: +
√ 4(e − 1)(1 + ra )(1 + rb ) + e (ra − rb )
k k 2
2 − (1 + ra )(1 + rb )((ek − 1)P02 (1 + ra )(1 + rb ) − ek P0 P1 (rb − ra ) − P12 ek )

4(ek − 1)(1 + ra )(1 + rb ) + ek (ra − rb )2
1
1 e2k
Region VII–IX: P
2 1 1k
(e 2 − 1)(1 + ra )
1 1
P0 (e 2 k − 1)(1 + rb ) + P1 e 2 k
Region VIII–IX: 1k 1k
2(e 2 − 1) + (rb − ra )(e 2 − 1)
1
e2k
Region IX–X: 1
2
P +
0 1k P 1
(e − 1)(1 + ra )
2
A.3. Exact Solution: Cash–Credit Indifference. We list below algebraic for-

mulae defining the combinations of current and future resources that make a
consumer indifferent between purchasing the durable good with cash or with in-
stallment credit. Recall from the discussion in the text that cash purchase is never
optimal in region I whereas credit purchase cannot be optimal in region X. We
omit details of the derivation, which are in general quite similar to those dis-
cussed in the article for region IV. The sign of the root in the solution is always
uniquely determined by considering that the cash–credit indifference locus has to
be upward sloping.
This can be shown by an argument similar to the one used to prove the weakly
negative slope of the purchase-indifference locus. Suppose the cash–credit indif-
ference locus were negatively sloped. Then it would separate the W − Y space in
two regions, one of which would be further away from the origin than the other.
(i) Imagine a pair (W0 , Y0 ) for which it is optimal to purchase the durable
with cash. Then the pair must lie above the negatively sloped frontier.
Otherwise, a sufficiently large increase of present resources, W, induces a
change from cash to credit purchase. This, however can never be optimal.
Hence, the region where cash purchase is optimal must be the region
further away from the origin.
(ii) Imagine a pair (W0 , Y0 ) for which it is optimal to purchase the durable
on credit. Then the pair must lie above the negatively sloped frontier.
Otherwise, it would be possible that an increase of future resources, Y,
induces a change of credit to cash purchase. This, however, can never be
optimal.
Now, (i) implies that in the region above the cash–credit indifference locus, cash
purchase is optimal, and (ii) implies that in this region credit purchase is optimal.
By contradiction, the cash–credit indifference locus cannot be downward sloping.
The indifference locus is continuous and continuously differentiable at the
boundaries of the regions. Its values at all boundary points are reported in section
A.4.
Regions II, III. In Region III the frontier is characterized by indifference be-
tween buying the durable cash and borrowing versus buying the durable on credit
and depleting assets, that is,
1
(Y + (W − P0 )(1 + rb))2 ek = W(Y − P1 )ek
4(1 + rb)
which yields

Y = (W + P0 )(1 + rb) − 2 W(1 + rb)(P0 (1 + rb) − P1 )
Region IV. In Region IV, the frontier is characterized by indifference between

buying the durable in cash or on credit where the consumer holds zero assets in
both cases, that is,
log(W − P0 ) + log(Y) + k = log(W) + log(Y − P1 ) + k
so that
P1
Y= W
P0
Regions V, VI, VIII. In Regions V, VI, and VIII, the frontier implies indiffer-
ence between buying the durable cash and borrowing versus buying the durable
credit and lending. This yields

(P1 − P0 (1 + ra ))(1 + rb) + (1 + rb)(1 + ra )(W(rb − ra ) − P0 (1 + rb) + P1 )
Y=
rb − ra
Regions VII, IX. In Regions VII and IX, the frontier is characterized by indif-
ference between buying the durable cash and borrowing versus buying the durable
credit and lending, which results in
Y = (1 + ra )(W − P0 ) + (P1 − (1 + ra )P0 )

+ 2 (P1 − (1 + ra )P0 )(W − P0 )(1 + ra )
A.4. Continuity. At the intersection points for the credit purchase–cash

purchase–indifference locus current resources W are given by
1 (1 + rb )P0
2
Region II–III: 4 (1 + rb )P0 − P1
Region III–IV: P02 P0 (1 1++rbr)b − P1
− (P1 −(1 + rb )P0 )
Region V–II: √
1 + ra − 2 (1 + ra )(1 + rb ) + (1 + rb )

Region V–III: P1
rb − ra
only if P1
P0
= 12 (1 + rb) + (1 + rb)(1 + ra )
√
(1 + rb )[P0 (1 + ra ) − P1 ] + ((1 + rb )(1 + ra ))[P0 (1 + rb ) − P1 ]
Region V–VI: √
((1 + rb )(1 + ra )) − (1 + rb ) (rb − ra )
√
1+ra + ((1 + rb )(1 + ra )) (P0 (1 + rb ) − P1 )
Region VI–III: √
((1 + rb )(1 + ra )) − (1 + ra ) (rb − ra )
P1 + P0 (1 + rb )

Region VI–IV: rb − ra
only if PP10 = ((1 + rb)(1 + ra ))
√
(1 + rb )(2P0 (1 + ra ) − P0 (1 + rb ) − P1 ) + (P0 (1 + rb ) − P1 ) ((1 + rb )(1 + ra ))
Region VI–VII: √
((1 + rb )(1 + ra )) − (1 + rb ) (rb − ra )
P1
Region VII–IV: P
P1 − P0 (1 + ra ) 0
√
(1 + rb )[P0 (1 + ra ) − P1 ] + ((1 + rb )(1 + ra ))[P0 (1 + rb ) − P1 ]
Region VIII–VI: √
((1 + rb )(1 + ra )) − (1 + ra ) (rb − ra )
√
((1 + rb ) − 2(1 + ra ))P0 + 2
(rb − ra )
(P1 − (1 + ra )P0 )((1 + ra ) + (1 + ra )(1 + rb )) + P1
Region VIII–IX: (rb − ra )
(1 + ra )(1 + rb )
Region VIII–VII: P0 r1b +−rrba only if P1
P0
=2 √
1 + rb + ((1 + rb )(1 + ra ))
1 P12
Region IX–VII: 4 (1 + ra )(P1 − P0 (1 + ra ))
REFERENCES
AIGNER, D. J., C. HSIAO, A. KAPTEYN, AND T. WANSBEEK, “Latent Variable Models in Econo-
metrics,” in Z. Griliches and M. D. Intriligator, eds., Handbook of Econometrics, vol.
II (Amsterdam: North-Holland, 1984), 1321–93.
ALESSIE, R., M. P. DEVEREUX, AND G. WEBER, “Intertemporal Consumption, Durables and
Liquidity Constraints: A Cohort Analysis,” European Economic Review 41 (1997),
37–59.
——, S. HOCHGUERTEL, AND G. WEBER, “Consumer Credit: Evidence from Italian Micro
Data,” Journal of the European Economic Association 3 (2005), 144–78.
ATTANASIO, O. P., “The Intertemporal Allocation of Consumption: Theory and Evidence,”
Carnegie-Rochester Conference Series on Public Policy 42 (1995), 39–89.
BERTAUT, C. C., AND M. HALIASSOS, “Debt Revolvers for Self Control,” Working paper,
University of Cyprus and Board of Governors of the Federal Reserve System, 2002.
BERTOLA, G., L. GUISO, AND L. PISTAFERRI, “Uncertainty and Consumer Durables Adjust-
ment,” Review of Economic Studies (forthcoming).
BRENNAN, M. J., V. MAKSIMOVIC, AND J. ZECHNER, “Vendor Financing,” Journal of Finance
43 (1988), 1127–41.
BRUGIAVINI, A., AND G. WEBER, “Durables and Non-Durables Consumption: Evidence
from Italian Household Data,” in A. Ando, L. Guiso, and I. Visco, eds., Saving and
the Accumulation of Wealth: Essays on Italian Household and Government Saving
Behavior (Cambridge, UK: Cambridge University Press, 1994), 305–29.
CHAKRAVORTI, S., AND T. TO, “A Theory of Merchant Credit Card Acceptance,” Federal
Reserve Bank of Chicago Research Department, Working Paper Series WP-99-16,
1999.
DANA, J. D. JR, “Monopoly Price Dispersion under Demand Uncertainty,” International
Economic Review 42 (2001), 649–70.
IOSSA, E., AND G. PALUMBO, “Product Quality, Lender Liability, and Consumer Credit,”
Oxford Economic Papers 56 (2004), 331–43.
JUSTER, F. T., AND R. P. SHAY, “Consumer Sensitivity to Finance Rates: An Empirical and
Analytical Investigation,” NBER Occasional Papers No. 88, 1964.
MURPHY, M. M., AND M. OTT, “Retail Credit, Credit Cards and Price Discrimination,”
Southern Economic Journal 43 (1977), 1303–12.
PETERSEN, M. A., AND R. G. RAJAN, “Trade Credit: Theories and Evidence,” Review of
Financial Studies 10 (1997), 661–91.
SPENCE, M., “Consumer Misperception, Product Failure and Producer Liability,” Review
of Economic Studies 44 (1977), 561–72.
WOOLDRIDGE, J. M., “Cluster-Sample Methods in Applied Econometrics,” American Eco-
nomic Review, Papers and Proceedings 93 (2003), 133–8.

Dealer Pricing

Uploaded by

Copyright:

Available Formats

You might also like

Dealer Pricing

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Dealer Pricing

Uploaded by

Copyright:

Available Formats

INTERNATIONAL ECONOMIC REVIEW

Vol. 46, No. 4, November 2005

DEALER PRICING OF CONSUMER CREDIT∗

BY GIUSEPPE BERTOLA, STEFAN HOCHGUERTEL,

Università di Torino, Italy; Free University Amsterdam,

∗ Manuscript received January 2003; revised December 2003.

We argue that dealer subsidization of consumer credit can be explained by in-

2. DEALER CREDIT SUBSIDIES IN ITALIAN DATA

time-series and cross-sectional impact of introduction of a law on usury rates in 1997.

Region 1996/I 1999/I

NOTE: Percentage of installment applications with dealer subsidization by region.

Item bought 1996/I 1999/I

White goods 42.8 50.4

NOTE: Percentage of installment applications with dealer subsidization by class of

Variable Quarter Mean SD Coeff. Var. Min Max

Bank borrowing rate 1996/IV 5.92 0.23 0.04 5.40 6.27

SAME AS FIGURE 1: CENTRAL ITALY

SAME AS FIGURE 1: SOUTHERN ITALY

We illustrate more general theoretical insights with an admittedly simplified,

A general time-separable representation of such preferences has the form U(C,

U(C, 1) + V(A, 1) > U(C, 0) + V(A, 0),

(1) max log(C) + log(A) + δk

When the durable good is purchased, the budget-constraint relationship between

3.1. Character of the Solution. In order to establish optimality of cash, credit,

3.2. Indifference to Purchase. If the consumer’s endowment is such that op-

whereas a cash purchase reduces current consumption to

arg max[log(C) + log(Y + (W − C − P0 )(1 + ra ))]

(5) log(CD,a ) + log(Y + (W − CD,a − P0 )(1 + ra )) + k

This is a quadratic equation in P0 and W + Y/(1 + ra ), the present value of funds

namely, if the consumer’s endowment of current and future resources, at a point

(6) Y + (W − C)(1 + ra ) − P1 ≤ Y + (W − C − P0 )(1 + ra )

as long as P1 ≥ (1 + ra )P0 as assumed in (3).

and credit purchase is preferable to no-purchase if log(W) + log(Y) < log(W) +

log(W − P0 ) + log(Y) + k < log(W) + log(Y − P1 ) + k

3.3. Indifference between Cash and Credit Purchase. To complete character-

of equations of order 1 + β, and the borrowing, lending, and liquidity-constrained

4. CONSUMER CHOICES AND DEALER PRICING

In Figure 5, the purchase-indifference locus is the lower envelope of the loci,

CONSUMER PURCHASE DECISIONS: IMPLICATIONS OF DIFFERENT BORROWING RATES; INTERNAL RATE OF

Similarly, the quantity sold on credit is given by:

The optimal choice of whether to subsidize depends on the discrete comparison

be computed evaluating numerical counterparts to the demand expressions (12)

It is also possible to characterize the implications of increasing rf starting from

(14) s∗ = Xγ + α1r f + α2r 2f + α3r 3f + β1  + β2 2 + β3 3 + µ(r f ) + u

where s∗ is the dealer’s inclination to subsidize as a function of variables, on the

of equations in the form (14) can be estimated by standard binary-choice mod-

5.1. Interest Rate Regressions. When the installment purchase is subsidized,

5.2. Determinants of Dealer Subsidies. Estimation of the interest-rate regres-

(a) Probit (b) Linear Model

5.3. Testing the Theory’s Implications. We proceed to offer formal tests of

regardless of estimation and standard error computation methods, in the case of

The benchmark specification of Table 5 is relatively parsimonious: It excludes cus-

We have motivated, proposed, and solved a model of interrelated purchase

We first derive algebraic formulae defining the combinations of current and

A.1. Exact Solution: Purchase Indifference. We proceed to list algebraic for-

LABELING OF REGIONS IN THE MODEL’S RESOURCES SPACE

Y + (W − C)(1 + rb) − P1 ≥ Y + (W − C − P0 )(1 + rb)

Region X. From inspection of Figure A.1, it follows that in this region

(14) s∗ = Xγ + α1r f + α2r 2f + α3r 3f + β1 + β2 2 + β3 3 + µ(r f ) + u