Who Finances Durable Goods and Why it Matters:

Captive Finance and the Coase Conjecture∗

Justin Murfin

Yale University

Ryan Pratt

Brigham Young University

First Draft: November 2013

This Draft: February 2017

Abstract

We propose that, by financing their own product sales through captive finance subsidiaries, durable goods manufac-

turers commit to higher resale values for their products in future periods. Using data on captive financing by the

manufacturers of heavy equipment, we find that captive backed models have lower price depreciation. The evidence

is consistent with captive finance helping manufacturers commit to ex-post actions that support used machine prices.

This, in turn, conveys higher pledgeability for captive backed products, even for individual machines financed by

banks. Although motivated as a rent seeking device, captive financing generates positive spillovers by relaxing credit

constraints.

∗Correspondence: Murfin: justin.murfin@yale.edu, (203)436-0666. Pratt: ryan.pratt@byu.edu, (801)422-1222.This paper was formerly titled “Captive Finance and the Coase Conjecture.” We thank Jean-Noel Barrot, TanakornMakaew, Rich Matthews, Ralf Meisenzahl, Rodney Ramcharan, Adriano Rampini, seminar participants at Yale Schoolof Management, BYU economics and finance departments, Southern Methodist University, University of Rochester,Ohio State, Drexel, Tulane, Colorado University, University of Pittsburgh, University of Illinois, the Federal Re-serve Bank of Philadelphia, Dartmouth Tuck, MIT Sloan, Temple University, the University of South Carolina, andconference participants at Red Rock Finance Conference, Olin Corporate Finance Conference, Financial ResearchAssociation Annual Meeting, Northern Finance Association Conference, and American Finance Association AnnualMeeting.

1 Introduction

A substantial share of durable goods financed with credit in the US is not financed by banks,

but rather by the manufacturer of the good itself. For firms making new investments in durable

equipment used in agriculture, construction, logging, manufacturing, and printing, the share of

financing done by the wholly-owned subsidiary lenders of equipment manufacturers was 58% as of

2013 and ranged from dominant (76% in agricultural equipment) to non-trivial (18% in printing)

(see Figure 1). And while manufacturers are important lenders within these industries, they are also

sizable enough to be important in aggregate. Manufacturing firms such as Toyota, John Deere, and

Caterpillar originate loan portfolios in a given year which would rank them among the top banks in

terms of business and non-credit card installment lending (as of 2012, #9, 15, and 17, respectively).1

Recent work by Stroebel (2016) and Benmelech, Meisenzahl, and Ramcharan (2016), meanwhile,

demonstrates the importance of these vertically integrated lenders in the markets for new housing

and cars, respectively.

As the line between traditional banking and manufacturing firms’ bank-like activities—so-called

“captive finance”—is blurred, a natural question arises. What are the economic motives behind

manufacturers financing their own sales? If banks are specialists in credit evaluation, monitoring,

and fundraising and thus would be natural candidates to finance durable goods investment, what

is the comparative advantage of captive finance?

This paper explores a set of candidate answers to the question above informed by the expansive

literature on durable goods. In particular, we begin with the insight that, because consumers of

durables care about the future value of their goods, durable goods producers require a means of

signaling that machines bought today will retain their value tomorrow. Captive finance, by linking

future periods’ profits to the future value of products sold today (whether through collateral value

on loans or residual values on leases), perhaps can serve as a commitment device for manufacturers

to avoid excessive depreciation of their product line, supporting rents today.

Following this intuition, we explore the link between who finances capital—whether traditional

1This is based on bank holding company call report data, where bank loan portfolios excluded interbank, charge-card, or mortgage lending for comparability.

1

2040

6080

2040

6080

2000

2001

2002

2003

2004

2005

2006

2007

2008

2009

2010

2011

2012

2013

2000

2001

2002

2003

2004

2005

2006

2007

2008

2009

2010

2011

2012

2013

2000

2001

2002

2003

2004

2005

2006

2007

2008

2009

2010

2011

2012

2013

AGRICULTURE CONSTRUCTION LIFT TRUCKS

LOGGING MACHINE TOOLS PRINTING

% N

ew M

achi

nes

Cap

tive

Fina

nced

Figure 1: Captive Finance and Durable Investment The figure plots the percentage of newdurable goods investment financed by captive finance subsidiaries of manufacturers in a variety ofindustries. Industry definitions were chosen by Equipment Data Associates, which provided thesesummary statistics.

banks or the manufacturer itself—and realized resale values of the same capital. Combining auction

data to measure depreciation rates for varying makes and models of heavy construction equipment

with new data on the financing provided by captives in support of equipment sales, we document

a robust link between products’ realized depreciation paths and the financing support offered by

captive lenders. On average, moving from zero captive financing for a given make and model to

machine sales being 100% financed by the manufacturer is associated with a 14-19 percentage point

decrease in observed depreciation. The results are driven by the source of financing and are not

dependent on the nature of the financing contract; we find captive lending and leasing portfolios

are both associated with lower depreciation, but that in equilibrium, installment loans are more

prevalent in the data.

Given the strong observed correlation between captive financing and the realized price deprecia-

tion associated with financed products, we proceed by exploring the mechanisms at play, differenti-

2

ating in particular between two closely related ideas. On one hand, we emphasize a new idea to the

vendor financing literature: that captive finance may be operating as a solution to the famous Coase

conjecture, that even a monopolist producer of a durable good faces competition from its own future

production, as customers may opt to delay their purchases if they expect prices to fall (Coase 1972).

Absent the ability to commit to future production, this competition can erode the producer’s mar-

ket power, as customers won’t be willing to pay monopolist prices today if they expect prices to fall

tomorrow. Whether through leasing or lending against its own product, manufacturers internalize

the price impact of their own future production choices to resolve time-inconsistencies in their op-

timal production paths (Bulow 1982). We refer to this hereafter as the limited commitment view

of captive financing.

Note, this idea is distinct from an intuitive and closely related interpretation: that lower depre-

ciation rates on captive financed models may indicate financing serves as a signal of quality in a

market with adverse selection. For durable goods, lower machine quality may manifest in a shorter

productive life and, as a result, faster price depreciation. Hence, the commitment to future resale

values offered by captive finance may arise through variation in ex-ante unobservable machines

characteristics as opposed to (or in addition to) ex-post producer actions. This idea, which we

hereafter label the information asymmetry view of captive financing, has support in prior work on

vendor financing.2

While the limited commitment and asymmetric information views both predict a relationship

between price depreciation and captive finance, a decomposition of the drivers behind depreciation

can help distinguish between the two. In particular, variation in ex-ante quality might suggest

depreciation driven by faster physical degradation. Immediately, however, we find that aspects

of depreciation related to changes in machines’ physical condition have little bearing on the link

between captive finance and resale values. Moreover, when we isolate the component of depreciation

driven by the price differential for old vs. new vintages at a given point in time—a measure that

would presumably capture the average physical or technological durability of a given model—we find

no evidence of a link to captive financing. Instead, depreciation appears to arise from supply effects

2For example, Stroebel (2016), Emery and Nayar (1998), Long, Malitz, and Ravid (1993), Lee and Stowe (1993)suggest financing may resolve issues related to unknown product quality.

3

that jointly impact new and old prices alike. Using auction sales of new or close to new machines,

we show that the effects of captive financing on depreciation are matched by a corresponding effect

on the time series of new machine prices. These findings lend support to the limited commitment

view whereby manufacturers lacking commitment discount newly produced machines over time.

Of course, the limited commitment mechanism depends critically on firms’ ability to affect prices;

firms without market power have no incentives to use captive financing to commit to prices they can’t

control. We reaffirm earlier findings (Mian and Smith 1992, Bodnaruk, O’Brien, and Simonov 2016)

that captive finance is concentrated among producers with plausible market power and extend this

result within firm to show that manufacturers concentrate their captive financing in machines for

which they have high market share. More importantly, the relationship between financing and

depreciation is limited to markets where producers are relatively large and concentrated.

The foregoing analysis establishes the predicted correlations we would expect to find in the data

under a limited commitment view of captive financing. We next move closer to causal inference

in a narrower sub-setting by exploiting a natural experiment that abruptly shifted the share of

captive financing for a specific set of equipment models. In 2007, Volvo, which had a large financing

arm, acquired a division of Ingersoll Rand, which had a history of very little captive financing. We

observe a sharp uptick in resale performance following the acquisition and corresponding increase

in captive financing share by Volvo. Importantly, even used machines produced by Ingersoll Rand

prior to the division sale began to retain their value more after the sale, indicating that our results

are not being driven by ex-ante machine characteristics such as unobservable machine quality.

Finally, although limited commitment can affect prices through a variety of ex-post firm actions

(we discuss potential mechanisms at length in Section 2.2, the original papers in the literature on

durable goods monopolists focused on production quantities. Using available measures of new ma-

chine sales, we find strongly suggestive evidence of a link between financing and ex-post production.

Within a given manufacturer, future sales are significantly lower for equipment types that have a

history of captive finance backing.

The evidence above in favor of the limited commitment hypothesis would point to captive

finance as a rent seeking device for oligopolistic producers seeking to maintain higher prices/lower

4

production over time. While healthy banks might arguably provide a public good, the initial

analysis would suggest welfare losses from a healthy captive sector. But our final results explore

a positive externality associated with a healthy captive finance sector. Products receiving strong

captive finance support enjoy increased pledgeability due to their lower depreciation rates, and

this is true even for individual machines which are financed by bank lenders. If credit constraints

are a significant drag on profitable investment, captive finance provides a social good by relaxing

these constraints. This becomes most valuable in periods of tight credit. In particular, there

is a shift towards models receiving captive support (and thus higher resale values) when lenders

report tightening their conditions for collateral. This is evident even among purchasers using bank

financing and therefore cannot be attributed to a pull-back in bank credit. An evaluation of the

welfare benefits of captive finance would therefore trade off the costs of enhancing the market power

of large manufacturers within the product market against the benefits of reduced frictions in credit

markets through enhanced pledgeability.

Our findings are related to both the literature on vendor financing as well as the literature on

durable goods. Within the vendor financing literature, we add to a long list of motivations for

firms to finance their own product . Our contribution is not to rule out all other motivations for

captive finance, but rather to present an idea that is perhaps new to this literature, and importantly,

one with distinct testable predictions. Most relevant to our work are the papers discussed above

under the information asymmetry hypothesis that document a signaling role for trade credit in a

market with adverse selection. Our findings suggest a nuanced retelling of this story whereby the

relevant signal is about hidden actions as opposed to hidden information. Our paper also relates

to earlier work by Biais and Gollier (1997), who first suggested vendor financing could resolve

information asymmetry about borrower risk and thereby generate positive spillovers by facilitating

bank financing. We observe similar spillover effects driven by signaling about the quality of borrower

collateral.

Within the literature on durable goods, the idea that manufacturer leasing can solve a limited

commitment problem is as old as the limited commitment problem itself. In contrast, ours is the

first paper to show that risky debt contracts can serve a similar role, both in theory and in practice.

5

2 Background and Hypothesis Development

2.1 Institutional Background

A captive finance company is typically a wholly-owned subsidiary of a manufacturer that provides

retail and wholesale financing for the parent’s products. Familiar examples include the finance

subsidiaries of major automobile manufacturers (e.g., Toyota Financial Services, GM Financial,

Ford Credit). Among heavy construction equipment manufacturers, the largest captive finance

companies include Caterpillar Financial, John Deere Capital, CNH Capital, Komatsu Financial,

Kubota Credit, and Volvo Financial Services, which together held more than $100 billion in total

finance receivables as of 2013.

While these companies actively provide short-term wholesale financing for dealers, the bulk

of their assets consist of longer-term secured loans made to retail customers. Although the prior

literature on durable goods manufacturers and the time-inconsistency problem has suggested leasing

might serve as a commitment to support resale values, empirically, lease financing is small relative to

secured lending, constituting 18%, 8%, and 10% of the retail finance portfolios reported in 2013 10-

K’s for Caterpillar, Deere, and Case-New Holland, respectively. This matches the characterization

of contracts in our data, for which less than 10% of total contracts are flagged as leases. The

construction finance industry skews more heavily toward secured lending than does the broader

equipment finance industry, where leasing made up approximately 30% of retail finance portfolios

from 2000-2009.3

The captive finance companies in our sample are largely funded by issuing long-term debt,

mostly consisting of medium term notes structured to match the maturity of the financing assets.

Such notes make up 77% of Caterpillar Financial’s liabilities and 63% of John Deere Capital’s. It

is also typical for captive finance companies to access commercial paper markets and to borrow

directly from their parent companies.

Of particular interest for the hypotheses tested in this paper are the securitization practices of

the captives. Loan sales or risk transfer by lenders could easily undo the effect of captive finance

3“Captive Finance Firms in a Challenging Economy,” Equipment Leasing & Finance Foundation

6

on manufacturer incentives to either restrict production or to signal product quality. Sutherland

(2016), however, reports that securitization plays a minor role in this market, which is consistent

with our reading of regulatory filings by large captives. As of 2013, the six largest captives had

securitized 14% of their finance portfolios. Most of this is driven by Case-New Holland; excluding

CNH, only 5% of financing assets were securitized. In all cases, securitized assets remain on balance-

sheet with significant retained interest such that even Case-New Holland discloses its exposure to

loan performance and collateral values in its 2013 10-K as follows: “An increase in customer credit

risk may result in higher delinquencies and defaults, and a deterioration in collateral valuation

may reduce our collateral recoveries, which could increase losses on our receivables and leases and

adversely affect our financial position and results of operations.”4 Hence, even with securitization,

captive lenders appear to remain significantly exposed to the resale price risk of their collateral,

consistent with security design preserving incentive compatibility.

While historically some captives have grown into more diversified finance companies (e.g.,

GMAC, now Ally), the companies in our sample primarily provide financing for their parents’

products, competing with commercial banks and other finance and leasing companies. Anecdotal

evidence from industry insiders indicates that, relative to these competitors, captives have histor-

ically lagged in terms of operating efficiencies, indicating that the prominence of captive finance

is unlikely to be explained by the superior industry expertise of the manufacturer.5 Additionally,

captives tend to provide financing for riskier borrowers. During 2004-2008, captives experienced

delinquencies beyond 30 days at nearly three times the rate of banks in their equipment financing

portfolios. Similarly, in 2008 captives approved 91% of credit applications, compared to 72% for

the entire industry. This focus on funding risky debt may be consistent with theories about captive

finance that depend on the threat of default to affect manufacturer behavior.

2.2 Hypothesis Development

In this section we explore potential mechanisms by which a manufacturer of durable goods can

enhance its profits by offering financing for its products. In particular, we focus on the idea that,

4Similar disclosures are found in the annual reports of the other major captives as well.5“Captive Finance Firms in a Challenging Economy,” Equipment Leasing & Finance Foundation.

7

by providing financing, manufacturers provide credible signals of future product value, raising the

willingness to pay of current period customers. This idea is grounded in two closely related theories,

each with distinct testable predictions.

On one hand, captive financing’s potential role in generating commitment to value might be

naturally interpreted as a form of product quality guarantee. This idea plays heavily in the literature

on trade credit. By way of example, Long, Malitz, and Ravid (1993) proposes slow payment by

buyers as a way to allow time for inspection of the underlying good, facilitating the separation of

high quality and low quality producers. Lee and Stowe (1993) and Emery and Nayar (1998) also

discuss mechanisms whereby trade credit provision serves to signal the underlying good’s quality. In

the context of longer-term secured captive lending against durable equipment, we might naturally

reinterpret quality as shorthand for long-term machine productivity, whereby low quality machines

cease to be productive, either due to physical degradation or technological obsolescence, over a

shorter time horizon. A sufficient condition for captive finance to function as a commitment to

produce goods which retain their productivity is the threat that unproductive machines will be

returned to the manufacturer. This might be achieved through leases or through secured lending

contracts, provided that there is a threat of default when realized equipment values are low enough.

While this idea is compelling, it is also somewhat limiting. If captive financing serves producers

by allowing them to signal higher future values for their product, this doesn’t have to occur just

through ex-ante production choices (e.g. quality choice), but alternatively, may be driven by a

commitment to ex-post production choices that have an equal hand in determining future machine

value.

In particular, we have in mind Coase’s classic 1972 analysis of the time-inconsistency problem

faced by a durable goods monopolist; that absent commitment to take actions that support product

value in future periods, consumers’ rational anticipation of excessive depreciation of the product

erodes the manufacturer’s rents today.6 While Coase (1972) and subsequent authors pointed out

6Coase (1972)’s analysis focused on the manufacturer’s production volume. However, the time-inconsistency prob-lem is much more far-reaching than that, encompassing any future action that the firm might take which couldaffect the value of past customers’ used goods. Previous theoretical work has shown that the inability to commitleads durable goods producers to overinvest in R&D and to introduce new products too frequently (Waldman 1996),to allow excess availability of used units (Waldman 1997), and to monopolize maintenance markets (Borenstein,

8

that leasing products as opposed to selling them can serve as such a commitment device, this

observation has not played a prominent role in our understanding of vendor financing to date.

Meanwhile, we show below that to commit manufacturers to low ex-post depreciation, leasing

contracts aren’t strictly required. Both leases and loans funded by the manufacturer can achieve

the same ends.

To fix ideas, below we sketch out the simple intuition of how captive lending and leasing can

serve to solve a limited commitment problem of a durable goods manufacturer. The more formal

presentation of the model, an extension of Bulow (1982), is kept to the appendix.

Consider a two-period setting in which a durable goods producer faces a downward sloping

demand schedule for its product. The firm produces qt machines in each period t = 1, 2 using a

production technology with constant marginal cost. Machines do not physically depreciate between

periods 1 and 2, except in the sense that they only have one period of usefulness remaining in period

2. The stock of machines at given point in time, then, includes current production plus any past

periods’ production.

In this setting, Bulow (1982) showed that a manufacturer that could commit to future production

would choose only to produce in the first period. This maximizes total profits because restrictions

on future production raise the purchase price of the good today sufficiently to more than offset any

potential profits which could have been earned through second period sales.

A manufacturer that lacks an effective commitment mechanism to avoid a second round of

production, however, arrives at the second period with no interest in the value of machines sold to old

customers and responds to any residual demand by selling additional units. This time inconsistency

in the producer’s incentives prevents first-best monopolist profits from being achieved. Meanwhile,

the ex-post inefficient second period production that occurs not only lowers first period prices, but

also leads to faster depreciation of the goods.7

A large literature follows this basic point to show the various production or contracting solutions

Mackie-Mason, and Netz 1995), all of which undermines the producers’ market power.7That lack of commitment leads to higher depreciation rates is not a consequence of the two-period environment.

Stokey (1981) studied the time-inconsistency problem in a discrete-time, infinite-horizon model, and Kahn (1986) an-alyzed a continuous-time setting with convex production costs. In both cases, without commitment, the manufacturerproduces too much over time, driving down prices.

9

that can return the producer to first-best rents. Bulow (1982) and Desai and Purohit (1998) studied

leasing; Butz (1990) analyzed best-price provisions, in which the difference in price is refunded to

any customers who paid a higher price for the same product, an alternative similar to the repurchase

agreements proposed by Coase; Bulow (1986) showed that the durable goods monopolist can reduce

its time-inconsistency problem by making goods with uneconomically short useful lives (planned

obsolescence); Kahn (1986) showed that increasing marginal costs in the production function make

the problem less severe; and Karp and Perloff (1996) and Kutsoati and Zabojnik (2001) studied

the deliberate adoption of inferior production technology. Probably the most applicable of these

proposed solutions to our empirical setting is that the manufacturer lease rather than sell its output.

By leasing, the manufacturer retains legal ownership of its products and thus internalizes the future

resale value of equipment, providing the necessary incentives to commit to the optimal production

path.

Yet in the model presented in the appendix, we show it is not necessary for the manufacturer

to retain ownership in order to align its incentives. The manufacturer can sell its output so long

as it retains exposure to future prices on the down side. This can be efficiently achieved when the

manufacturer provides secured debt financing for its customers. Returning to the simple two-period

problem, suppose a manufacturer finances the sale of its own equipment in the first period. By

choosing low down payment financing, the manufacturer ensures first period customers retain high

loan balances in the second period. Then, if the manufacturer overproduces in the second period, or

takes any ex-post action to depress second period prices, first period customers become underwater

on their loans. In the simplest version of the model where borrowers have no outside wealth or

loans are without effective recourse, the secured loan contract acts exactly like a lease in forcing

the manufacturer to internalize the full price effects of second period production. However, strict

assumptions on recourse or outside borrower wealth are not necessary; as long as lender profits

are sensitive to the threat of borrower defaults, lending contracts can satisfy the same role Coase

initially proposed for leases.8

8While we are not able to fully characterize the extent to which loans are subject to recourse, there do appearto be some practical limitations. For example, Case-New Holland makes the following disclosures regarding theirloan portfolio: “some states may impose prohibitions or limitations on deficiency judgments,” and, in cases where adeficiency judgment is obtained, “it may be uncollectible or settled at a significant discount.” A deficiency judgment

10

There are two important differences between the manufacturer’s solution with outside financing

and that with captive financing. First, since captive finance creates incentives for restricted produc-

tion in both periods, machines depreciate more slowly when they are financed by the manufacturer.

Second, the manufacturer is able to offer a higher loan-to-value when it finances its own machines.

This is tied to the depreciation rate. Because the machines will be worth more in the future, the

manufacturer can lend more without pressing the buyers into default. In fact, since the threat

of default is what provides commitment to the manufacturer, it is more accurate to say that the

manufacturer must offer a higher loan-to-value. If it were to offer the same financing contract as we

find under bank financing in the model, it would lack the commitment to keep secondary market

prices high.

Hence, under models where buyers face asymmetric information about the productivity or

longevity of their durable purchases, or where sellers have limited commitment to take ex-post

actions in support of their older machines, we can generate predictions whereby producers who

finance their own sales can also support lower depreciation rates on those machines. In section 4.1

we proceed by showing that, in practice, there is a strong relationship between realized depreciation

on physical machinery and the identity of the financier, consistent with the hypotheses put forward

above. We then attempt to assess the relative importance of the two potential mechanisms to this

empirical fact.

A few caveats are in order. First, note that while we have emphasized the role of secured

lending above, given that leases and loans can achieve similar ends under both models of information

asymmetry and limited commitment, when we examine the data we will focus more on who provides

the financing and less on the form of the contract. Meanwhile, while leasing might appear to have an

advantage over lending in the sense that, with leases, the manufacturer is exposed to residual values

even without default, as described in the institutional details section above, secured lending turns

out to be the dominant contractual form for captive finance companies in the construction industry.

is required to collect on any amounts owed by a borrower after the liquidation of collateral. It states further thatcourts have applied “general equitable principles [that] may have the effect of relieving an obligor from some or all ofthe legal consequences of default.” Caterpillar suggests the same in its discussion of how it calculates allowances forcredit losses, claiming that this primarily depends on “the most probable source of repayment, which is normally theliquidation of collateral.”

11

This might be consistent with the well-known problem associated with lease contracts when there

is scope for abuse by the user of the underlying good. Bulow (1982) and Eisfeldt and Rampini

(2009) both point out that leases suffer from the costs of separating ownership from control. Thus,

benefits of leasing relative to lending may be offset by the higher monitoring costs when machine

value is sensitive to the manner in which it is used. While the literature may benefit from further

investigation of contract determination, this is perhaps beyond the scope of this paper.9 Having

said that, in Table 1 we separate captive leasing from lending and find that it does appear to have

the predicted relationship with depreciation, consistent with it being an effective mechanism when

feasible.

Second, with any model where financing serves as a signal or commitment device, we might

worry that securitization of loans or leases would undermine the seller’s incentives. Of course,

with informed loan/lease buyers, this shouldn’t be the case—loan buyers require sellers to maintain

enough skin in the game to retain their incentives to protect collateral values. Although we doc-

umented the limited use of securitization in practice in the prior section, to the extent that loan

sales do occur, manufacturers’ retained interests are sufficient to make them sensitive to collateral

values, as we might expect.

3 Data

To test the hypotheses derived above, we focus our attention on the market for heavy equipment used

in construction and agriculture. We focus on heavy equipment for a number of related reasons. First,

to match the interesting features of the model, we need a less-than-perfectly competitive industry

such that production and financing choices can interact meaningfully. While not monopolistic, the

market for heavy machinery is controlled by a handful of large firms. By way of example, the most

purchased piece of equipment in our sample is a skid-steer loader—a small, four-wheeled machine

with lift-arms capable of pushing or lifting heavy material. In 2012, 5 manufacturers produced

9Other proposed solutions to the time-inconsistency problem, such as repurchase agreements, also suffer from theseparation of ownership and control. Insofar as the terms offered by the manufacturer will lead to the customerchoosing to sell back the machine, the customer will not internalize its care. The same argument applies to best-priceprovisions/money-back guarantees.

12

93% of debt-financed skid-steers in the US. Thus, it seems plausible to presume that individual

manufacturers have pricing power.

The durable quality of goods is also central to our argument. Returning again to skid-steers, the

median used skid-steer financed with secured credit in 2012 was 8 years old. Durability, meanwhile,

begets a healthy secondary market, which is both a feature of the model and allows us to track

prices over time.

Finally, we focus on heavy equipment because of its nature as capital investment (as opposed

to a consumption good). Although, in theory, our hypotheses apply to consumer durables equally

well, we think that some of the more interesting implications of our theoretical findings pertain to

the potential link between captive finance and pledgeability and how this may impact borrowers

facing credit constraints. To the extent that the relaxation of credit constraints is an important

outcome of captive financing, any resulting impact on firm investment could have large spillover

effects for the aggregate economy.

For our primary tests, we rely on two distinct sources of data. One, produced and sold by

Equipment Data Associates (“EDA”), tracks financing statements filed by secured lenders for sales

– new or used – of heavy equipment financed by secured debt (hereafter, the “UCC data”, because

statements of financing are designated as the means of documenting liens under the uniform com-

mercial code). The data represent the universe of complete filings with the exception of filings from

the state of Nevada, which does not allow for bulk downloads. We use these financing statements

to infer the extent to which financing is done by manufacturers or by competing banks, our central

predictions having to do with how much financing support a manufacturer lends to a given model.

Although it is debatable as to whether or not financing statements are required for leases, the data

and conversations with the data provider suggest that it is a common practice to file financing

statements for both loans and financing leases. As a result, we include both leases and loans in

our analysis. Ultimately, the focus of our analysis will be who finances the equipment, not the

legal form of the contract. Nevertheless, we can show that our main results will hold for loans or

leases. We also use a limited sample of the data which allows us to observe the actual loan amount

extended by the lender and an EDA-formulated estimate of the equipment value, thereby allowing

13

us to infer loan-to-value, or percent down payment.

The data on financing statements are self-reported by lenders motivated by the need to “stake

a claim” to specific pieces of collateral. In the event of a default on a secured loan in which

multiple lenders report liens against the same piece of equipment, the first lender to have filed a

UCC financing statement on that specific piece of equipment is given priority. Thus lenders have

strong incentives to promptly report the collateral they have lent against. Financing statements

are publicly available, but EDA sells cleaned and formatted versions going back to 1990. An

introduction to financing statements and the claim-staking process is available in Edgerton (2012),

which is the first and only other work we are aware of to use these data.

The second dataset we exploit is produced by EquipmentWatch, a data provider which reports

results from heavy equipment auctions going back to 1993. While not comprehensive, we have

data on the sales/purchases of over one million pieces of equipment from the largest auctioneers of

heavy equipment.10 The machine-level observations include sales price, as well as both auction and

equipment information such as age, condition, and make/model. The machines in our data represent

substantial purchases for the average buyer. The average estimate of equipment value from EDA

and actual sales prices at auction are $89,455 and $30,750, respectively. The difference partially

reflects the fact that, whereas auction sales include both cash and debt financed sales, the EDA

estimates only reflect sales financed by secured debt. Additionally, auctions are composed almost

exclusively of used sales, while financing statements include new and used sales. For reference, the

appendix lists the most common machine types and manufacturers in the UCC data by observation

count.

4 Results

4.1 Captive Finance and Resale Prices

Our central hypothesis, and thus the first testable implication we investigate, is that firms that

provide their own financing to buyers are able to commit to higher future resale values. As we’ve

10EquipmentWatch claims it has 90% coverage of equipment auctions going back to 1990, notably excluding auctionsrun by IronPlanet.

14

noted above, any observed correlation between machine financing and subsequent price path could

suggest financing serves as a signal to resolve information asymmetry about machine characteristics

known to the manufacturer but not to markets. After documenting the relationship between resale

performance and financing, we’ll work to separate the importance of signaling machine quality from

financing’s role in providing ex-post commitment to maintaining price levels.

Table 1 begins the analysis by comparing, on a model-by-model basis, estimated depreciation

rates in the used equipment market with the degree of financing support offered for each model.

Specifically, the variable ModelCaptiveSupport is the percentage of new machines of a given model

financed by a captive finance lender over the entire sample. (Note, we only observe transactions

conditional on financing, so our measure of captive support does not take into account cash sales,

for example). Our focus on captive support only over new machines allows us to isolate the effect

of captive financing availability, rather than variation in new vs. used sales, which is closely related

to lender type. As a point of reference, for the average make and model, 28% of new machine

sales/leases were financed by the manufacturer.

Meanwhile, we estimate depreciation rates using the auction sample, by regressing the log sales

price on log machine age. Regressions are estimated on a model-by-model level. Before running

model-level regressions, year fixed effects, estimated over the entire sample, are removed from both

equipment age and price. Thus, time fixed effects are treated as constant across model-specific

regressions. Formally, our measure of depreciation comes from the regression

˜ln(AuctionPricei,t) = αmodel + δmodel˜ln(1 + EquipmentAgei,t) + εi,t (1)

where δmodel is estimated separately for each model based on the sales of individual machines, i,

at auctions in year t, and ˜ln(AuctionPricei,t) and ˜ln(1 + EquipmentAgei,t) are demeaned by the

average yearly log price and log (1+) machine age at auction. Age is calculated as the number of

years between the original manufacture date and the date of resale for each machine being sold.

δmodel captures the model-specific depreciation rate. It should be interpreted as the percentage loss

in value for a proportional increase in age.

15

With a measure of depreciation in hand, we can now project it on the level of captive financing

support available for a given model, ModelCaptiveSupport, again, estimated as the percentage of

new model sales reported in the UCC data that were financed by a captive. To reduce noise, we

limit ourselves to models for which we have 30 or more transactions in both the auction sales and

the UCC data. After dropping equipment models with too few observations, we are left with 1,750

models with which to estimate the conditional mean depreciation rate based on captive support.

To isolate the effect of financing, we control for fine level equipment-type fixed effects, as well as a

set of 26 size dummy variables, where size is characterized by EDA based on important machine

characteristics, often horsepower or weight.

Column 1 of Table 1 reports the estimates from the model

− δmodel = α+ β1ModelCaptiveSupport+ β2...kControlsmodel + ε. (2)

We estimate a coefficient on ModelCaptiveSupport of -0.14, suggesting that moving from a fully

bank financed model to one which is completely financed by the manufacturer would predict a

reduction in the machine’s depreciation rate by 14 percentage points. Meanwhile, the mean depre-

ciation rate is 60%. Figure 2 shows this graphically, plotting mean depreciation rates across levels

of captive finance. While the relationship is flat for small amounts of captive financing, we can see

that above some threshold amount (roughly 50% of machines), depreciation rates appear strongly

inversely related to captive support.

A sensible response to the observed correlation between captive financing and machine character-

istics is to look for alternative machine characteristics that might drive both who finances machines

and how the machine holds its value. Note, the inclusion of machine-type and size fixed effects

appears to rule out omitted variables related to physical features of machines or intended uses, say,

of backhoes vs. excavators. Instead, it seems more likely that plausible confounding variables are

in fact driven by manufacturer characteristics. One such argument might be that bigger, better

companies make better machines and are capable of financing their own products. To address this,

column 2 of Table 1 includes manufacturer fixed effects, testing for differentials in rates of product

16

Figure 1. Resale depreciation and captive finance.

�

Figure 2. Captive Finance and Market Power

-.65

-.6-.5

5-.5

-.45

d ln

(Pric

e)/d

ln(A

ge)

0-10

%

20-3

0%

40-5

0%

60-7

0%

80-9

0%

10-2

0%

30-4

0%

50-6

0%

70-8

0%

90-1

00%

% Model Sales Captive Financed

20%

25%

30%

35%

40%

45%

Mea

n %

Mod

el S

ales

Cap

tive

Fina

nced

1 2 3 4 5 6 7 8 9 10Equipment Type Herfindahl Decile

Figure 2: Resale Depreciation and Captive Finance. The figure plots mean depreciationrates across different levels of captive finance. On the horizontal axis is the percentage of new salesof a given model that are financed by the manufacturer. On the vertical axis is the proportionaldecrease in resale price for a proportional increase in age.

depreciation across models within a manufacturer. The coefficient on ModelCaptiveSupport is not

statistically distinct from the results in column 1 without manufacturer fixed effects. That is, un-

observed manufacturer characteristics related to the level of captive financing do not explain the

observed correlation.

Notice that our measure of captive financing varies based on who the financier is, and not

the form of contract. Whereas our presentation of the limited commitment problem emphasizes

the potential role of debt in enforcing commitment to match the fact that the machines in our

sample are overwhelmingly debt financed, Coase and subsequent authors focused on leasing by the

manufacturer as a solution. In either case, as long as the manufacturer is sensitive to residual

value on old machines, the mechanism works similarly, although one might naturally think that

sensitivity would be higher for leases, which are returned to the manufacturer even without the

threat of default.

17

The final columns of Table 1 exploit our limited ability to break down captive fi-

nance by contract type, separating ModelCaptiveSupport into ModelCaptiveSupportLease and

ModelCaptiveSupportLoan based on a UCC data flag indicating whether a given transac-

tion was a lease or a loan. Among captives, the percentage of lease financing is rare—

ModelCaptiveSupportLease averages less than 3% compared to 26% for ModelCaptiveSupportLoan,

consistent with separation of ownership and control being particularly problematic for heavy ma-

chinery. Moreover, we should be wary that it is impossible to distinguish between operating and

financing leases.

Having said that, re-estimating column 2 on the separate measures would seem to suggest

that, when leasing is feasible, it has significant power in generating commitment. Doubling lease

support reduces depreciation by 0.45, compared to 0.16 using loan captive support. Thus, the

data are consistent with Coase’s earliest prediction regarding lease financing. Our perhaps small

contribution regarding the theory—the suggestion that debt works as well—is also strongly evident,

albeit weaker in magnitude than the effect of leasing support.

However, the inclusion of loan and lease measures side-by-side in column 5 leaves only captive

lending significant, perhaps due to the limited amount of variation in leasing support (for 50% of

models, captive leasing is 0). Given that both loans and leases generate commitment in the model

and appear to be effective in the data, and more to the point, given our emphasis on who provides

financing as opposed to contract type, going forward we focus on the total measure of captive

support covering both leases and loans.

4.2 Limited Commitment vs. Asymmetric Information Hypotheses

Absent further information, the result is open to interpretation. As discussed in the hypothesis

development section, our interest is in differentiating the role of captive financing in solving a

limited commitment problem from that of a more traditional adverse selection model, whereby

vendor financed models are of higher quality, and thus have better ex-post price performance. For

example, Stroebel (2016) finds that when new homes are financed by the home builder, they are

less likely to suffer capital losses, and he links this result to better construction techniques which

18

are unobservable to the home buyer. Note the key differences in hypotheses: whereas an adverse

selection model of captive finance suggests captive financing commits the firm to ex-ante higher

quality, the limited commitment hypothesis suggests manufacturers will have incentives to maintain

prices in secondary markets through ex-post production choices, independent of quality and in the

absence of information asymmetry.

Fortunately, the data allow us to refine our definitions of depreciation in a way that can help

differentiate among competing interpretations. In particular, the most natural application of signal-

ing models predicts a link between financing and depreciation caused by physical degradation as a

machine ages (in Stroebel (2016) for example, home foundations crack over time). In contrast, our

mechanism emphasizes predictions that would occur absent any physical degradation. An idealized

experiment might, then, track the market price of “new, old-stock” machines over time—unused

machines for which the passage of time isn’t related to changes in condition—to separate the effects

of physical condition from our predicted price effects.

Table 2 attempts to approximate this experiment by exploiting information on machine condition

provided by auctioneers at the time of sale. For a subsample of machines and auctions, Equipment

Watch codes individual machines as excellent, very good, good, fair, and poor based on a physical

evaluation of the machines coming up for sale. For machines which are coded based on condition, we

can estimate a separate depreciation path which holds condition constant—that is, we can estimate

the change in price for a model over time which is unrelated to its physical deterioration.

We add this information to our tests in stages, beginning with a straightforward replication of the

specifications from Table 1, this time controlling for machine condition in the first stage depreciation

regressions. Following the methodology in Table 1, before running model-level regressions, year fixed

effects, estimated over the entire sample, are removed from equipment age, price, and now condition

indicators. This time, however, for each make × model, we estimate the elasticity of price with

respect to age, holding condition constant. That is, for each equipment model for which we have

at least 30 observations in the auction data with condition information, we estimate δ in the model

19

below:

˜ln(AuctionPricei,t) = αmodel+δmodel˜ln(1 + EquipmentAgei,t)+

∑cond

γmodel,cond˜I(Conditioni,t)+εi,t.

(3)

Columns 1 and 2 from Table 2 reproduce the findings from Table 1, with and without manufac-

turer fixed effects. However, this time, by holding condition fixed as time passes, we have eliminated

one likely explanation for the correspondence between financing and resale performance related to

the speed with which machines physically deteriorate.

Columns 3 and 4 further refine our measure of depreciation by redefining the unit of observation

at the make × model × vintage level. Whereas prior tests have treated machines within a given

model name as homogenous, if manufacturers modify design or quality over time, any resulting

variation in auction prices could confound our estimates of depreciation. By focusing attention at

the make × model × vintage level, we eliminate any potential for variation in quality or design.

Meanwhile, we continue to control for condition at auction. Hence, the depreciation estimates in

columns 3 and 4 can be interpreted as the expected change in price over time for a given vintage

of model, holding condition fixed (e.g. what percent decline in price did a 2002 vintage John Deere

210LE Tractor Loader experience as it aged, controlling for any changes in condition over time?).

We think this perhaps best approximates the idealized test of “new, old-stock” prices.

Moreover, by drilling down to the vintage year level, we can include year fixed effects based on

the year in which a model was produced. This removes any times series covariation in financing and

depreciation that might be behind Table 1 (and eliminates the need to partial out year fixed effects

before the depreciation regression). Finally, whereas earlier tests estimate the captive support for a

given model over the entire sample, once operating at the model × year level, we can focus on the

backward-looking history of captive finance for a make-model from the years prior to the current

vintage year.

For the sample of make × model × vintage year machines for which we observe 30 or more sales

from the auction data with complete information on age and condition, we now estimate δmodel×year

by running make × model × vintage year level regressions of log price on log age alongside dummy

20

controls for condition.

Although executed on a substantially smaller sample (the growth in observations due to each

model observation multiplying by the number of years it was produced is swamped by the loss of

model years for which the auction data has 30 or more observations), columns 3 and 4 of Table 2

replicate the baseline regression from Table 1 with unchanged effect sizes. The stability of the

coefficients suggests that the baseline regressions capture the variation of interest to test models

of commitment. Even after removing price effects due to changes in condition over time, machine

vintages for which the model has a history of captive backing depreciate significantly more slowly.

Column 4 adds manufacturer fixed effects—the coefficient of -0.20 is nearly identical to the -0.19

from Table 1.11

Finally, column 5 pushes the data to its limit to approximate the path of newly produced

machines (as opposed to the approximation of new, old-stock machines in columns 1-4). Whereas

models of information asymmetry make no prediction about the time series path of new machine

prices, a commitment role for captive finance will affect the path of new and used prices alike.

Although the auction data provide very little coverage of new machine sales, we can limit the

sample to machines less than two years old and in very good or excellent condition. For this subset

of close-to-new machines, we examine the price trend based on years since model introduction. That

is, for each model for which we have at least 30 auction observations, we estimate

˜ln(AuctionPricei,t) = αmodel + δmodel˜ln(1 +ModelAgei,t) + εi,t, (4)

where model age is the machine vintage year less the minimum vintage year for that model in the

sample, and price and age have been demeaned by auction year as in (1) and (3). We refer to

−δ as “new price depreciation” in our specifications, as it refers directly to the price path of newly

11Appendix Table A3 provides a slightly different take on the tables above, estimating a measure of physicaldurability at the model level using auction data on condition and adding it to the regressions in columns 1 and 2 ofTable 1. Specifically, we define failure as a machine transitioning to fair or poor condition at auction and use survivalanalysis to estimate a survival function for each model. We then define durability as the estimated log median survivaltime for a given model. We describe the analysis in detail in the Appendix. To summarize, however, we find thatadding durability as a control strongly predicts dollar depreciation rates, but does not attenuate the coefficient onmodel captive support. Finally, consistent with these two results, model captive support is not a strong predictor ofestimated durability.

21

produced machines for a given model, as best we can approximate it. Under the limited commitment

view, new price depreciation should be linked to captive finance through producers’ ex-post supply

(or other) decisions—for example, if manufacturers lacking commitment discount model pricing over

time. This measure, however, should not be affected by the speed with which machines lose their

productivity as they age, the variation predicted under the information asymmetry view. For the

160 models for which we can estimate new price depreciation, column 5 replicates column 2 and

finds a coefficient of -0.16, consistent with the magnitude of earlier estimates. The fact that results

are unchanged even when focusing on our best approximation of new machine prices would seem to

add additional support for the price commitment role for financing.

Given the results in columns 1-5 of Table 2, it seems unlikely that physical quality drives the

relationship between prices and the prevalence of captive financing in our data. However, a closely

related idea spans the gap between both limited commitment and quality. Waldman (1996) argues

that one manner in which the time-inconsistency problem might operate is through the introduction

of new technologies by manufacturers. If these newer technologies drive down old machine prices,

then manufacturers would like commitment to delay technological introduction. The prediction for

captive financing is that captive backed machines will suffer less technological depreciation at the

hands of the manufacturer. Yet the same story can easily be retold as relating to ex-ante quality. If

manufacturers can forecast future changes in technological standards better than consumers, they

might signal machines’ technological longevity through financing.

Column 6, however, would appear to rule out both such hypotheses. In the final test of Table 2,

we modify our measure of model level depreciation to capture how prices vary with age at a given

point in time. That is to say, what would be the price discount between an old and a new version

of a given make and model observed by a customer walking into a dealer today? Note, this is

distinct from the measure of depreciation used in columns 3 and 4, for example, which captured the

expected change in price for a unit increase in age based on the passage of time for a given model

× vintage.12

12In contrast, the latter measure of depreciation (used in columns 3 and 4) is identified entirely off time seriesvariation in age which affects a given vintage uniformly. Obviously, there is no within-auction-year variation in agewhen regressions are estimated at the vintage level, given no two machines of the same vintage can have different agesin the same auction year. Meanwhile, Table 1 and columns 1 and 2 of Table 2 pool the two sources of variation to

22

This distinction is important in sorting through the mechanism. If manufacturers commit to

keep prices high through, for example, restricted production, this has no clear predictions about

how old machines should be valued relative to new machines at a given point in time; in the

case of old and new machines, prices are higher because of the manufacturer’s commitment. In

contrast, if the manufacturer has used its financing arm to signal physical quality or technological

longevity, then relative to models with no financing, we would unambiguously expect to observe a

shallower discount for older vintages relative to newer vintages at a given point in time. Hence, we

can use a variant of depreciation rates estimated based on cross-sectional variation in equipment

age at a fixed point in time to differentiate between the competing hypotheses. We call this new

measure “productivity-driven depreciation” to capture the idea that it reflects changes in value with

age that correspond with changes in realized productivity, either through physical or technological

characteristics. We isolate within-auction-year variation in price with respect to age for a given

model by simply including auction year fixed effects in each depreciation regression. That is, the

left-hand-side variable in column 6 is −δ from the following model-level regression:

ln(AuctionPricei,t) = αmodel + γauctionyear + δmodelln(1 + EquipmentAgei,t) + εi,t. (5)

ModelCaptiveSupport is again measured over the life of the model.

Note, when we re-estimate the basic fixed effects regression from Table 1, but with the modified

depreciation measure, the effect of captive support falls away. Moreover, the coefficient in column 6

is statistically different from that in column 2 of Table 1 with a p-value of 0.055. Here, the absence

of any evidence linking captive finance to price differences between old and new machines at a

given point in time appears to rule out ex-ante quality or technological depreciation as the relevant

mechanism.

estimate depreciation (within time and across time).

23

4.3 Market power

The case for examining captive finance as an outcome of the problem facing durable goods producers

with market power was partly driven by the observation that captives tend to concentrate around

durable goods, but also partly based on the observation that captive financing is associated with

large firms (see Mian and Smith (1992) and Bodnaruk, O’Brien, and Simonov (2016)). The use of

fixed effects and the persistence of the observed relationship between captive financing and resale

values within manufacturer in Tables 1 and 2 meanwhile raises an interesting question: why would a

manufacturer provide more financing to some models than others? Understanding captive finance as

a commitment device for firms with market power suggests one potential answer: manufacturers do

not need the commitment afforded by captive finance in markets where they have limited ability to

set prices. That is, if a manufacturer’s production level has no impact on price, then commitment to

restrict production over time is not valuable. The limited commitment hypothesis can only explain

the use of captive finance by firms with market power and predicts the within-firm variation in

ModelCaptiveSupport underlying Tables 1 and 2 should follow within-firm variation in product

market power.

It turns out this is consistent with the data. Figure 3 plots the use of captive finance across

different models against the market concentration for the equipment type. The x-axis is the decile

of either the Herfindahl index (“HHI”) or a measure of concentration defined by the inverse number

of active producers of a given machine type (e.g. skid-steer loader, mini-excavator, etc.) in a

given year, while the y-axis presents the average model-year captive support within each decile.

Consistent with the findings in Bodnaruk, O’Brien, and Simonov (2016), an active role for captive

finance appears to manifest more prevalently in high HHI industries and industries with fewer

producers.

Table 3 brings us to a similar, but more pronounced, conclusion. Column 1 regresses

ModelCaptiveSupport on the natural log of a model’s market share, along with controls for equip-

ment type, size, and manufacturer fixed effects. Market share is estimated for new equipment sales

within a given equipment type. It is calculated annually and averaged over the life of the model.

We find that, even within a given manufacturer, models receiving captive support are more likely

24

.2.2

5.3

.35

.4M

ean

% M

odel

Sal

es C

aptiv

e F

inan

ced

1 2 3 4 5 6 7 8 9 10

Equipment Type HHI Decile Equipment Type 1/N Concentration Decile

Figure 3: Captive Finance Intensity and Market Power. The figure plots the level of captivefinance support received by an equipment model as a function of the concentration of the model’smarket. The vertical axis is the average percentage of new sales financed by the manufacturer,averaged over machine type × year observations. The horizontal axis is the decile of either theHerfindahl index or a measure of concentration defined by the inverse number of active producersof a given machine type in a given year.

to be models with substantial market share. Although earlier work has suggested a relationship

between firm size or market share and the use of captive finance, we are able to exploit machine-

level data to show that this relationship holds within manufacturer. This allows us to rule out

explanations that would link market share and captive support indirectly through observed or

unobserved manufacturer characteristics.

In interpreting the table, a natural question is whether manufacturers direct financing at models

with market share, or if financing directed at specific models causes high market share for the same

models. Barrot (2016), for example, shows how larger firms may use trade credit provision as a

competitive device to crowd out financially constrained competitors. To resolve this, in column 2,

we estimate market share based only on bank financed sales. This helps us avoid capturing market

share which is driven by captive financing, as opposed to captive financing which is plausibly driven

25

by market share. The effect is slightly smaller, but still significant at 0.02.

Columns 3-5 combine this result with the results from Table 1 to show that captive finance can

only predict resale values when the manufacturer has market power. We replicate the regression of

depreciation rates on ModelCaptiveSupport and controls for equipment type, size, and manufac-

turer from Table 1, but add an interaction with a dummy variable for models with above-median

market share. In column 3, market share is estimated broadly. In column 4 it is limited to bank

financed sales. In each case, the interaction is negative and significant. For models with above-

median market share within a given machine type, going from 0 to 100% captive finance of new sales

reduces depreciation rates by 14-18 percentage points, versus a statistically insignificant reduction

of 2-3 percentage points for manufacturers with below-median market share.

Because we might still worry that causal effects of captive financing on market share muddy

the interpretation, column 5 zooms out and replaces market share with a market-wide estimate of

concentration, specifically the inverse of the number of firms producing a given equipment type.

Similar to market share, it is calculated annually and averaged over the life of the model. Although

market share is what matters—even in a concentrated industry, captive finance will be of limited

commitment value for small players—we can think of this regression as using the number of firms in

an industry almost as an instrument for market share. Findings are consistent with the relationship

between depreciation and financing attenuating in more competitive markets. Going from the 10th

to the 90th percentile in one over the number of firms (moving from roughly 35 to 4 producers)

shifts the sensitivity to ModelCaptiveSupport from -0.04 to -0.19.13

13Although in Figure 3 above we also demonstrate the correspondence between HHI and use of captive finance,in the make-model level regressions, focusing on the number of producers has more natural predictions. As statedabove, our sharpest prediction are with respect to market share. Hence, to the extent we substitute market sharefor an equipment-type level measure of concentration, we want a measure that unambiguously affects the averagefirm’s market share. The inverse number of active producers satisfies this role—if four firms produce and one leavesthe market, the average firm’s market share must increase. In contrast, note that HHI can be decomposed into twocomponents—first, HHI is driven by 1

Nand second, it is driven by a component capturing the variance of market

shares (formally, HHI = 1N

+∑

(si − s̄)2, where si is firm i’s market share and s̄ is the mean market share). Unlike1N

, a shock to the variance of market shares has perfectly offsetting predictions on the average firm’s market share(one firm’s gain must be another firm’s loss). Hence, any relevant variation in HHI must be driven by 1

N. Meanwhile,

as a secondary issue, because subsequent tests will examine quantity share as an outcome of interest, measures ofconcentration which don’t directly depend on market share may be preferable. Therefore, in this test and those goingforward, we focus on 1

Nas measure of concentration.

26

4.4 Evidence from Volvo’s Acquisition of Ingersoll Rand Unit

Absent a large scale natural experiment to generate random variation in captive financing across

our entire sample, we’ve relied on mostly broad sample correlation-based evidence to interpret the

observed link between how firms finance their sales and the depreciation rates associated with their

products. So far, the evidence strongly suggests a role for what we refer to as a limited commitment

role for captive financing in the spirit of Coase (1972). The inclusion of manufacturer fixed effects in

Table 1 rules out alternative mechanisms having to do with unobserved manufacturer characteristics;

controlling for measures of the physical condition of machines in Table 2 and, to the extent possible,

tracing out the effect on new machine prices separates our findings from the predictions of a model

of product quality signaling; and the results based on market power in Table 3 provide evidence of

a prediction unique to our hypothesis.

Now we add to the evidence a quasi-experiment affecting a set of machines which went from

receiving no financing support from the manufacturer to substantial captive backing over a short

period of time, and then examine their subsequent resale performance. While we won’t argue that

change in financing was randomly assigned, the motivations for the change in financing are at least

readily understandable. Moreover, while financing patterns changed, the machines being produced

and the means of servicing them seem to have been largely unchanged. This fact helps narrow the

set of plausible interpretations of prior results.

Specifically, we focus on the acquisition of Ingersoll Rand’s road construction division by Volvo

in 2007. The road construction unit was part of Ingersoll Rand’s construction technology division.

Though a small part of the total firm, generating less than 10% of operating income in 2006 (Ingersoll

Rand annual report 2006), the division was a large part of the road construction market, with 25%

market share as of 2004, measured using new units financed in the UCC data. The sale to Volvo

was driven by a strategic realignment that redirected resources to other divisions, including the

larger climate control and security divisions. The proceeds from the $1.3BN sale were used to fund

acquisitions and a continued share buyback program. The sale does not appear to have been driven

by distress, but rather strategic fit for the two companies. Reuters’ announcement provided the

following commentary on the sale from Volvo’s perspective: “In terms of products this is right.

27

These are things they don’t have in their product line-up. In terms of the price, you always pay

quite a lot for these types of assets since they have rather high margins.” (Volvo to buy Ingersoll

Rand Road unit for $1.3 BN, Reuters, 2/27/2007).

Two aspects of the sale are of interest. First, the acquisition gives sharp variation in captive

financing support. Whereas Ingersoll Rand was not historically active in financing, Volvo had a

large consumer finance division that immediately began aggressively financing the formerly Ingersoll

Rand machines. Second, until recently, the machines acquired and sold by Volvo were under the

same designs and manufactured in the same plants as they had been when owned by Ingersoll

Rand. As an example, the specifications for the most popular Ingersoll Rand roller (the DD-24)

are listed in the appendix alongside the same key specifications for the Volvo DD-24 roller. A

side-by-side comparison suggests these are very nearly identical machines. Also included in the sale

were Ingersoll Rand’s service and maintenance facilities. Thus, when we study resale performance,

neither physical characteristics of the machine nor service quality are likely to be material drivers of

any changes in resale performance. That is, the unit sale helps us isolate the hypothesized role for

captive financing from potentially confounding characteristics of bank financed vs. manufacturer

financed machines.

Figure 4 documents the patterns of interest. Using the UCC data, we plot the market share

of Volvo and Ingersoll Rand among affected machines within the road construction segment. In

the two plots below, we observe the market share and share of captive financing for Ingersoll Rand

and later Volvo as active producers of road compaction equipment and road paving equipment

(the two broad segments of road construction equipment included in the sale). Road compaction

equipment includes a broad range of machine types, from manually operated rammers to larger

single drum and tandem drum vibratory compactors, all broadly designed to accomplish the goal

of flattening surfaces on soil, gravel, or pavement. The spin-off, however, also included the sale of

road paving equipment, including tracked pavers, wheeled pavers and road wideners. All told, 23

different equipment types were included in the sale, including 61 different affected models (those

models which were produced by Ingersoll Rand/Volvo and which traded both before and after the

spin-off). Meanwhile, these represent diverse classes of equipment, ranging in weight from 500

28

kilograms to 11.9 metric tons and in price from $3,500 (the first percentile of transaction price

within the auction data) to $127,000 (the 99th percentile).

0.2

.4.6

.8

2004 2005 2006 2007 2008 2009 2010

Compaction Equipment

0.2

.4.6

2004 2005 2006 2007 2008 2009 2010

Share of new compactors/pavers captive financed (Ingersoll Rand or Volvo)

Market Share Ingersoll Rand

Market Share Volvo

Paving Equipment

Figure 4: Ingersoll Rand sale of road construction equipment unit to Volvo. The figureplots the market share of new road construction equipment of Ingersoll Rand and Volvo (based onunits financed in the UCC data) as well as the combined share of new machines which were financedby a captive. The left panel represents compaction equipment and the right panel shows pavingequipment.

In the plots, with compaction equipment represented on the left and paving equipment on the

right, notice the acquisition by Volvo shows up in the data beginning in 2007 and 2008, as sales of

Ingersoll Rand equipment gradually decline and are slowly replaced by Volvo-branded equipment.

As of 2008, combined sales of Volvo/Ingersoll Rand compaction and road paving equipment captured

in the UCC data represented 14% of the number of total machines sold, second only in market

share to Caterpillar. More important, however, is that the transaction increased captive financing

of affected Ingersoll Rand products from 0.91% in 2006 to 66% in 2008. Column 1 of Table A4

in the appendix presents this chart in a make × model fixed effect regression, where whether or

not a new compaction or paving machine was captive financed is regressed on a post-2007 dummy,

29

interacted with a dummy for formerly Ingersoll Rand machines (post-2007 Volvo machines), along

with model-level and year fixed effects. Holding model-specific attributes fixed, the spin-off raised

captive financing by 44%, an effect which is both economically and statistically significant at the

1% level.

How do we explain the large change in machine financing? As mentioned above, Volvo had an

active financing arm useful in serving Volvo’s largest business line (trucks and buses), which are

naturally amenable to captive financing. Meanwhile, while we can’t say with certainty why Ingersoll

Rand was less active in captive finance, we suspect this is also related to the company’s broader

business strategy. At the time of the divestiture, Ingersoll Rand’s largest divisions were producing

climate control equipment and refrigeration units for buses and refrigerated trailers, along with

grocery store refrigerated displays. They later used cash from the division sale to expand this brand

with the acquisition of Trane heating and cooling systems and services. Whereas our explanation for

captive financing is intimately related to the ability to originate secured debt against the product

being sold, for practical reasons, the primary products being sold by Ingersoll Rand would not

appear to easily lend themselves to security interest. Meanwhile, there are limited opportunities

for secondary markets in air conditioning systems. Thus, Ingersoll Rand’s broader disinterest in

captive financing is consistent with the hypotheses we have suggested. Combined, we think of the

variation in captive financing as coming from fixed costs of establishing a lending unit interacted

with the usefulness of having a lending arm across the producer’s portfolio of products.

Using the divestiture and resulting change in machine financing as a source of variation, Table 4

again tests for the relationship between machine depreciation and source of financing. The estima-

tion strategy approximates a triple difference-in-difference whereby we estimate the impact of age

on machine value for Ingersoll Rand machines before and after the acquisition of the business by

Volvo. The difference in depreciation is then benchmarked against changes in depreciation observed

for similar machines made by rival producers over the same time period.

Take the simplest presentation of the test described above. For the four subgroups (treatment,

control, pre and post), we run separate regressions of log price on log age and make × model ×

vintage fixed effects. Here, we are approximating earlier depreciation regressions, but pooling our

30

estimates of depreciation rates across the four subgroups (instead of estimating the make-model-

vintage specific depreciation regressions) to economize on parameters.

The results of this simple exercise, shown in Panel A of Table 4, are telling. For the control

group—manufacturers of road construction equipment who were not Ingersoll Rand or Volvo—the

average depreciation rate captured by the slope on log age is 0.33 in the years before 2007 and 0.37

in 2007 and after. That is, for the control group, depreciation rates are stable and not statistically

distinct. In contrast, prior to 2007, Ingersoll-Rand machines exhibited depreciation of 0.44. The

difference relative to the control in the pre-period of 0.10 is significant and perhaps consistent with

the lack of captive financing capabilities. Yet in the post-2007 period, once Volvo’s captive financing

arm is active, that number improves to 0.33 when applied to both old Ingersoll Rand and new Volvo

machines, an improvement relative to the control of 0.15 (significant at the 5% level).

Panel B of Table 4 extends these results to a more fully specified model, combining the four

regressions into one and adding various controls and combinations of fixed effects. The economic

message, however, is largely unchanged. Formally, we estimate the regression

ln(AuctionPrice) = β1ln(1 +MachineAge) + β2Post× ln(1 +MachineAge) + β3IR× Post

+β4IR× ln(1 +MachineAge) + β5IR× Post× ln(1 +MachineAge) + λControls+ ε

where post is a dummy equal to one for machines sold at auctions in 2007 or after, and IR is a

dummy for machines produced by Ingersoll Rand prior to 2007 and Volvo during and after 2007.

The coefficient on IR×Post× ln(1+MachineAge) tells us whether or not Ingersoll Rand machines

depreciated differently post-divestiture relative to broader trends in machine depreciation over the

pre and post periods (captured by Post × ln(1 + MachineAge)). Notice, Post is defined by the

year of the auction and not the year the machine is produced. This is important because, unlike a

depreciation effect driven by differences in how the machines were produced, which could only take

effect for machines produced post-2007, the commitment to higher resale values induced by Volvo’s

increased role in financing should also benefit machines produced prior to 2007. In fact, we can

(and do) test this specific prediction by limiting the sample to machines produced prior to 2007, for

31

which depreciation rates can only be dictated by ex-post producer behavior and not ex-ante design

or build choices by Volvo that differed from Ingersoll Rand.

Given that our auction data only go to mid-2013, we limit the analysis to post-2000 auction

data, roughly balancing the sample around the 2007 handover. We also limit the age of the ma-

chines at auction to those less than 7 years old, given we only observe secondary sales of Volvo

compactors/pavers up to this age. Finally, we control for auctioneer fixed effects and for the re-

ported condition of the machine at auction when available (condition is classified as excellent, very

good, good, fair, poor, new, or unknown), although for the majority of machines, the condition is

unknown.

The results are reported in Panel B of Table 4. Column 1 includes make-model fixed effects.

Columns 2 and 3 include make-model-vintage fixed effects. Finally, column 3 excludes machines

actually manufactured under Volvo ownership. That is, it reports how machines made by Ingersoll

Rand depreciated before and after Volvo ownership relative to the control group.

In each case, the evidence is consistent with the prediction that the increase in captive financing

post-2007 changed how machines maintained their value over time. The coefficient on IR×Post×

ln(1+EquipmentAge) ranges from 0.15 to 0.17 and suggests that following the acquisition by Volvo,

machine depreciation slowed by 15-17 percentage points relative to the pre-acquisition period. The

coefficient on ln(1 + EquipmentAge) in column 1 suggests a baseline 41% elasticity of machine

price with respect to age for non-Ingersoll Rand/non-Volvo machines before 2007.14 Again, the

improvement in depreciation is relative to the change in depreciation of non-Ingersoll Rand/Volvo

machines over the same time period (although this benchmarking is not critical, as Panel A suggests

there is no evidence that depreciation rates of road construction equipment changed for the control

group from 2000-2006 vs. 2007-2013).

Because the experiment leans on a single acquisition that simultaneously affects an entire class

of machines, there is scope for concern that we’ve missed a confounding aspect of the transaction

that drives the observed variation in machine prices, but has nothing to do with the change in

financing. Ideally, we would have an additional control group within the set of affected machines—

14The coefficient on age is harder to interpret in columns 2 and 3, which include both model-vintage fixed effectsand auction year dummies, which combined, are collinear with age and nearly collinear with ln(1+Machine Age).

32

models which were part of the spin-off but which were not impacted by the change in financing.

Of course, if we were to simply compare the newly financed machines to models for which Volvo,

ex-post, chose to withhold financing, we would introduce the risk that Volvo chose not to support

some models based on private information.

In the absence of an instrument for treatment intensity within the set of acquired makes and

models, we lean instead on pre-determined machine characteristics which, out of sample, are as-

sociated with a propensity for captive finance. Specifically, for any machine type involved in the

acquisition, we estimate the percentage of sales or leases which were captive financed in the pre-2000

era, excluding any Ingersoll Rand or Volvo manufactured machines. Our conjecture is that, if some

machine types lend themselves more naturally to captive vs. bank financing, then given the newly-

found capacity to offer captive financing for old product lines, Volvo will increase its financing share

relatively more for these machines (that is, relative to the more typically bank financed machines).

This in turn, generates pre-determined variation in the expected intensity of treatment and allows

us to compare outcomes based on their exposure to captive finance. The test is similar in spirit to

Rajan and Zingales (1998), who estimated financial dependence by industry in the US and applied

the resulting industry classifications to non-US firms to learn about the causal effects of financial

development.

The first stage regression plays out as we might expect. As described above,

MachineCaptivePropensity is defined as the captive finance share for each equipment type

in the pre-2000 sample for non-Ingersoll Rand/non-Volvo producers. We then add to

our first stage regression of captive financing on the treatment and post variable an in-

teraction with MachineCaptivePropensity. The positive interaction of Post2007 × IR ×

MachineCaptivePropensity in column 2 of Table A4 in the appendix shows that the increase in

captive financing for formerly-Ingersoll Rand machines was more pronounced for machines which,

out of sample and for unrelated producers, also tended towards captive financing. Said otherwise,

machine types which rarely benefit from captive financing out of sample also tended to receive less

financing from Volvo post-acquisition.

Columns 4 and 5 proceed by using MachineCaptivePropensity as a source of variation in

33

the realized growth in financing around the spin-off. We re-run the specification from column 2,

this time splitting the sample around the median MachineCaptivePropensity. A comparison of

the relative depreciation rates captured by the interaction on IR× Post× ln(1 + EquipmentAge)

suggests that the financing aspect of the acquisition is indeed critical to the observed price effects.

While we might have predicted some impact on depreciation among the less affected machines,

there is no obvious change in depreciation rates evident in column 4. Although these machines did

experience growth in captive financing (Volvo increased financing by 29 percentage points for low

captive propensity machines), the magnitude of the financing effect of the spin-off was considerably

smaller than for the machines captured in column 5 (for which Volvo increased financing by 50

percentage points). Meanwhile, the depreciation effect for the more affected machines in column 5

is both significantly larger than that of column 4 (with a p-value of .01) and economically larger

than the population-averaged result in column 2. So, while we can’t rule out unknown, non-financial

effects of the merger driving the changing depreciation rates, those effects would also need to co-vary

with the same machine characteristics that drive our ex-ante predictions regarding the size of the

financing effect.

While our predictions and tests primarily focus on the role of financing in preserving machine

value over time, we also expect that the level effect on machine value should be positive, as com-

mitment to support prices in the future leads to higher prices today. Column 6 shows this directly,

bypassing the interaction between age and post-divestiture and simply focusing on Post2007× IR.

The coefficient of 0.06 suggests that, controlling for age and model, the price of the average affected

machine increased by 6% following the divestiture. Meanwhile, Table A5 in the appendix estimates

the year fixed effects for affected machines in the years leading up to and following the treatment

and demonstrates an immediate increase in average machine prices suggested in column 6, with no

evidence of a pre-existing trend.

4.5 Evidence on the Quantity Mechanism

The evidence to this point has focused on distinguishing among different hypotheses about the

role of captive finance companies by examining patterns in equipment prices. We now turn our

34

attention directly to potential mechanisms by examining the production choices of the manu-

facturers in our sample. While we’re sympathetic to wide-ranging behaviors that could affect

vintage machine prices under the umbrella of the limited commitment problem—again, previ-

ous authors have suggested aftermarket support, new product development, and scrap rates are

all ex-post decisions that firms might like to commit to ex-ante (Borenstein, Mackie-Mason, and

Netz 1995, Waldman 1996, Waldman 1997)—given the emphasis on production quantities in earlier

work and the clarity of predictions associated with this mechanism, we would be remiss in not

examining this hypothesis. Our findings suggest a strong link between manufacturers’ history of

captive financing for a given equipment type and attenuated production growth.

Before proceeding, it is worth noting that our measurement of quantities is imperfect. Infor-

mation on sales volume by model and/or equipment type is closely guarded by producers. Instead,

we can only measure quantities from the UCC data. However, because this measure of quantities

depends on machines being financed and reported in the UCC data, we can’t distinguish between

hypotheses for which captive finance drives future quantities produced from alternative hypotheses

for which captive finance drives future quantities financed. Of course, if captive financing ensures

manufacturer commitment, then we might expect a history of captive financing would embolden

banks to do more financing going forward—we show evidence shortly that, indeed, banks lend higher

loan to values for models which benefit from captive support. If our measure of quantity growth is

partially capturing growth in lending, then this would seem to work against the predicted pattern

wherein growth is slower for captive backed machines.

With this caveat in mind, we test the hypothesis that captive financing inhibits production by

examining the future sales of new equipment (as observed in the UCC data) as a function of the

history of captive finance support. The analysis is done at the make × equipment type × size

× vintage year level. Defining the unit of observation this way allows us to avoid problems in

measuring quantities that would arise from the introduction of new models to replace old models.

Our measure of captive finance is backward-looking as in columns 3 and 4 of Table 2, so that

ModelCaptiveSupport measures the proportion of new machines of a given type and size that were

captive financed prior to the current vintage year. As in prior tables, we require more than 30

35

sales to estimate the measure of captive support. The outcome of interest is average sales of new

machines over the next one, three, and five year periods, relative to current period quantities. An

observation in year t is conditional on positive quantities in the current and lagged year, but does

not require future quantities. Finally, we include fixed effects for the manufacturer, the equipment

type, size, and year.

Column 1 of Table 5 shows a negative but insignificant relationship between one year production

growth and a history of financing. But over the longer three year and five year periods, an effect

is evident. Three year average sales scaled by lagged sales drops by 0.21 moving from full bank to

full captive financing, relative to a mean of 1.07. Similarly, five year average sales scaled by lagged

sales drops by 0.34, relative to a mean of 1.05. Interpreted through the lens of limited commitment,

captive finance would appear to lessen the temptation to overproduce in future periods at the

expense of current and past period buyers.

4.6 Captive Finance and Pledgeability

Taken together, the evidence above suggests a rent-seeking motivation for captive finance, helping

firms with market power commit to high prices and limited production quantities. While this

casts captive finance in a decidedly negative light, there may also be benefits associated with

manufacturers lending against their own product. In particular, captive financing allows for—

or to be more precise, requires—lower down payments from the buyer, relaxing credit constraints

for borrowers with limited internal cash. High loan-to-values provide commitment to the producer

not to lower prices in the future, but if the captive’s financing activities are observable to bank

lenders, they may free-ride and extend the same loan-to-value without incurring excess risk of

default. Thus, sufficient captive finance backing will generate lower down payments, even when

machines are financed by banks. Whereas simply observing more aggressive lending standards by

captives (discussed below) would perhaps not be surprising, the prediction that banks will lend

more on machines that receive large amounts of captive finance suggests a broader importance for

how machines are financed.

We test this prediction in Table 6 by using a subset of financing statements for which the bank

36

or captive reports the mortgage amount on their UCC filing. Combining this amount with an EDA-

provided estimate of purchase price based on machine value at that point in time, we calculate a

variable, Downpayment, as the difference between the log price and the log loan amount. Of inter-

est is the effect of captive financing on predicted down payment amounts, controlling for machine

characteristics as well as borrower characteristics. While borrower characteristics are limited, we

observe the borrower’s state and industry and include dummy variables for each (industry dummies

are at the level of a 2-digit SIC code). For the machine, we can control for the equipment manufac-

turer, type, size, and the age of the machine, all of which may plausibly impact both the required

down payment and the lender of choice. Among the transactions for which we can calculate down

payment, a small minority (6.5%) are characterized as leases in the UCC filing. Given that our

predictions would hold across loans or leases, we leave these in the data and add a control for the

contract type.

Finally, and most importantly, we can limit the analysis to bank financed transactions. Whereas

Benmelech, Meisenzahl, and Ramcharan (2016) suggest an important role for captives in providing

credit directly, we are interested in the indirect effects of captive financing through machine collat-

eral characteristics. To isolate these effects, it may be important to set aside transactions where

captives had a direct role and instead assess how captive lending interacts with non-captive lenders’

assessment of collateral.

Table 6 reports the results. In column 1, we look at bank and captive financed transactions and

find the relationship between captive financing support and down payments is large and significant,

controlling for year, machine size, age, type, manufacturer, and borrower state and industry fixed

effects. A model without captive support requires a down payment (as a fraction of value) which

is 25% larger than a machine which is otherwise always financed by the captive. With 20% as a

typical rule-of-thumb required down payment and average machine values of $89,455, this implies an

additional $4,547 down payment amount required of machines receiving no captive support (relative

to $17,891 for a fully captive financed model). Column 2 isolates just bank financed transactions

and reports a near identical coefficient. In other words, captive financing raises loan-to-values even

on transactions where the buyer financed at a bank, indicating that the coefficient in column 1 is

37

not simply capturing differences in terms offered by captive lenders. Consistent with earlier tests,

we show in column 3 that the relationship between model captive support and down payment is

significantly stronger in concentrated industries (as measured by the inverse number of producers),

where manufacturers are more likely to be able to affect prices. Finally, column 4 repeats the

specification in column 1 but replaces the pooled ModelCaptiveSupport with the backward-looking

version used in columns 1-4 of Table 2. Although we lose observations based on the requirement that

captive support be estimated with more than 30 observations, we finds similar signs and significance.

These results suggest a positive spillover effect of captive finance. While the reduced depreci-

ation and higher resale values may serve as rent-seeking devices, the resulting low required down

payments/higher pledgeability of machines that have received captive financing support may be

of value when credit constraints bind. Extending this logic, can the outcomes from rent seeking

generate positive spillover effects in periods when financing frictions, and thus the shadow value of

pledgeable equipment, are high?15

To test this possibility, and to provide additional support for the argument that captive finance

may help commit companies to production paths fostering greater pledgeability, we look to the

revealed preference for makes and models receiving strong captive support during periods of credit

tightening by banks. Our prediction is that borrowers facing rising shadow costs of internal capital

will choose machines with greater pledgeability—machines that have received captive financing

support from their makers. Because we would like to avoid documenting substitution effects of

bank financing being replaced by captive financing during periods of tight credit, we can again limit

our attention to the machine choice of borrowers who are financing their purchases with banks.

Figure 5 gives a taste of our findings. Using two measures of bank credit tightness, we observe

a strong variable demand for machines with captive finance support during periods of tight credit.

The top panel plots the probability a given new machine purchase, when financed by a bank,

will be of a make and model that received strong captive finance support over the entire sample

period (specifically, the fraction of total machines financed by the manufacturer is more than the

15Benmelech and Bergman (2009) show that pledgeability in the airline industry is most valuable during industrydownturns when rationing is likely to be greatest.

38

 

 

0.0

5.1

.15

.2.2

5

-20

020

4060

1991

q1

1993

q1

1995

q1

1997

q1

1999

q1

2001

q1

2003

q1

2005

q1

2007

q1

2009

q1

2011

q1

2013

q1

Net % Loan Officers Tightening Collateral Req. (left axis)Pr(Purchase Captive Backed Machine|X) (bank financed purchases only)

0.0

5.1

.15

.2.2

5

200

300

400

500

1991

q1

1993

q1

1995

q1

1997

q1

1999

q1

2001

q1

2003

q1

2005

q1

2007

q1

2009

q1

2011

q1

2013

q1

Mean spread on sepculative grade bank loans (left axis in bps)Pr(Purchase Captive Backed Machine|X)(bank financed purchases only)

Figure 5: Machine Choice and Credit Tightness. The figure plots the time series of theproportion of new, bank-financed purchases of models with high captive support (more than themedian value captive financing percentage) on the right-hand axis of both panels. In the firstpanel, we compare this to the net percentage of loan officers that reported tightening collateralrequirements on the left-hand axis. In the second panel, we plot the mean spread on speculativegrade bank loans on the left-hand axis.

39

median across all models), controlling for equipment type and size.16 This series (the solid line) is

plotted against a dashed line representing a survey-based measure of banks’ demand for pledgeable

assets taken from the senior loan officer survey performed quarterly by the Federal Reserve Board

of Governors. The survey asks loan officers whether they have tightened or loosened collateral

requirements for small businesses. The Fed then reports the net percentage reporting tightening

(percentage tightening minus percentage loosening). There is an apparent correlation between the

two time series driven by large changes during boom and busts periods, but also in the quarterly

changes. The bottom panel, meanwhile, shows a similar relationship between machine choice and

tightening as seen via increases in loan spread. The dashed line represents the mean spread over

Libor of new issue speculative grade loans in the DealScan database. Again, in periods of tight

credit/high spreads, the demand for machines with captive finance support increases, even among

bank financed purchases. Note, this is not driven by time series variation in the availability of

captive finance relative to bank lending—if anything, the share of captive finance for new machine

purchases actually decreases in periods of tight bank credit.

Table 7 reduces the picture in Figure 5 into a regression of an individual purchaser’s machine

choice on financing conditions, this time allowing for new and used transactions, as well as controls

for machine age, type, and size. Out of concern that the recent credit crunches may have dispro-

portionately affected industries that use machines receiving more captive support, we include both

2-digit SIC industry and state fixed effects for machine buyers. Meanwhile, the left-hand-side vari-

able captures the consumer’s choice, within an equipment type, of a captive backed vs. traditionally

bank financed model. Captive-backed models are those for which the model’s captive finance sup-

port is greater than the median for that equipment type. Model captive support is estimated over

the entire sample for models with more than 30 observations.

Column 1 reports the basic finding from Figure 5 that buyer preference for machines receiving

above median captive support increases during periods of credit tightness across the entire sample.

The measure of credit tightness has been given zero mean and unit standard deviation, such that

a coefficient of 0.02 suggests that for a one standard deviation increase in tightness is associated

16The figure plots the time fixed effect from a regression of machine choice (captive supported machine or not) onquarter, equipment size, and type.

40

with a 2% higher probability a consumer chooses a captive backed model. The immediate apparent

concern, of course, is that this effect may be driven by greater availability of captive financing during

periods of tight credit. To rule this out, we can focus on the subset of bank-financed transactions

as we did in column 2 of Table 6. If the preference for captive backed machines is driven by

availability of captive financing, it should not affect buyers who choose to finance at the bank (or, if

anything, we might expect the sample of bank transactions to be skewed away from captive backed

models). The effect, however, is unchanged when we focus on bank-financed transactions, showing

a positive association between machine choice and aggregate conditions even among bank financed

transactions.

Columns 3 and 4 explore the robustness of the finding. Column 3 demonstrates that, consistent

with prior results, the effects are stronger in more concentrated industries. Finally, column 4

identifies captive backed models using only the backward-looking measure of financing used in

Table 2. That is, for each equipment model, we measure the percentage of all lagged annual

transactions which were captive financed; we then define a captive backed model as being above the

median with respect to that measure. As before, we require 30 or more past transactions to estimate

the percent of captive financing, causing some shrinkage in sample size. The results, however, are

largely unchanged.

Of course periods of credit tightness covary closely with other business cycle measures. As a

result, it may be reasonable to interpret this pattern loosely as a preference for captive financed

machines in downturns generally and not necessarily as evidence of a direct link to credit frictions.

Moreover, we are sensitive to the fact that, while the time-series results appear statistically signif-

icant under our most conservative standard error estimates, we only have data over a few credit

cycles.

5 Discussion

In this paper, we motivate and document evidence of a strong correlation between use of captive

financing and resale price performance for capital goods. The evidence appears consistent with

41

a commitment motive, more so than alternative explanations. Meanwhile, we find evidence that

captive financed machines require lower down payments and are favored by borrowers when ac-

cess to funding is constrained, perhaps as a result of the increased pledgeability associated with

manufacturer commitment to supporting prices.

The welfare implications of captive finance, therefore, are blurred. On one hand, the evidence

points to captive finance as a rent-seeking device. Manufacturers that offer it would appear to do

so, at least in part, to raise prices today and restrict production tomorrow. This has a clearly

negative impact on consumer surplus. On the other hand, in a world with credit constraints where

pledgeability is valuable, welfare implications become less clear. As we saw in the crisis (Figure 5),

the demand for machines offering greater pledgeability appears to jump exactly when the need for

lending was greatest. This suggests a value to captive finance which is perhaps largest in the worst

states. Thus, it may be reasonable from a welfare context to think of captive finance as a costly

hedge, where the cost of restricted supply and the associated manufacturer rents affords the benefit

of relaxed credit constraints on capital investment during periods of tight credit.

42

Appendix

Appendix A Theoretical Motivation

In this section we formalize the argument that secured debt financing from the manufacturer can help

solve the time-inconsistency problem highlighted by Coase (1972). We begin with the environment

studied by Bulow (1982), which we will shortly extend to allow for secured debt financing from the

manufacturer. Consider a two period setting in which a durable goods producer faces a demand

schedule for rental services of its output given by

pr = a− bQ, (6)

where Q denotes the total stock of machines in the economy and pr is the one period rental price.

The firm produces qt machines in each period t = 1, 2 using a production technology with constant

marginal cost which we normalize to 0. Machines do not physically depreciate between periods 1

and 2, except in the sense that they only have one period of usefulness remaining in period 2. We

then have Q1 = q1 and Q2 = q1 + q2. Finally, we normalize the discount rate to 0.

There is a frictionless resale market for used equipment. Since there are only two periods, the

market price to purchase a machine in period 2 is equal to the rental price, p2 = a − bQ2. In

contrast, a first-period buyer receives the equipment’s services in both periods, the present value of

which is given by p1 = a− bQ1 + a− bQ2 = a− bQ1 + p2. This equation reflects the fact that the

sales price of equipment in the first period is affected by the anticipated future value of equipment,

which in turn is affected by second-period production. That is, buyers care about the resale value

of their equipment so that the price they pay for equipment is the sum of the first-period rental

price and the resale value.17

As shown by Bulow (1982), a manufacturer that could commit to future production in this

17Equivalently, we could think of the first-period sales demand as arising from the buyers’ option to wait to purchase.Given an anticipated price path {p1, p2}, a buyer with productivity x will purchase in the first period if 2x − p1 ≥x − p2 ⇐⇒ x ≥ p1 − p2. From the demand for rental services, the quantity of buyers with such productivity isQ1 = 1

b(a− (p1 − p2)). This is equivalent to the inverse sales demand above.

43

setting would choose a production path characterized by:

q∗1 =a

2b, p∗1 = a, π∗ =

a2

2b

q∗2 = 0, p∗2 =a

2.

(7)

Here the manufacturer produces the amount that would be chosen by a monopolist producer of

consumption goods in the first period. In the second period the manufacturer produces nothing,

leaving the total stock of machines in the economy at the monopolist level. Total profits, given by

π∗, are first-best.

A.1 Equilibrium with Bank Financing

A manufacturer that lacks an effective commitment mechanism, however, will not be able to capture

π∗. Consider such a manufacturer that sells its output to buyers who finance through a bank (hence

the b superscripts below), which are equivalent to cash sales from the manufacturer’s perspective.

We solve the manufacturer’s optimization problem by backward induction. Conditional on first

period production, the manufacturer will choose second period production to maximize second

period profits. So at time 2 the firm solves:

maxq2

(a− b(q1 + q2))q2 (8)

It follows immediately from the first-order condition that the optimal production and corresponding

price and profit (conditional on q1) are qb2 = 12b(a− bq1), pb2 = 1

2(a− bq1), and πb2 = 14b(a− bq1)2.

Given this solution to the second period problem, the first period sales price is p1 = 32(a− bq1),

and the firm’s first period problem can be written as:

maxq1

3

2(a− bq1)q1 +

1

4b(a− bq1)2 (9)

The first term in the objective function is the monopolist’s profit from first period production, and

the second is his profits from second period production. The solution to this problem yields the

44

following production and price paths:

qb1 =2a

5b, pb1 =

9a

10, πb =

9a2

20b

qb2 =3a

10b, pb2 =

3a

10

(10)

Two differences are immediately apparent when comparing this solution to the full-commitment

solution. First, the manufacturer’s profits are lower, illustrating the value that the firm can derive

from any mechanism that enables it to commit to keep future prices high. This is because the

firm’s production in the second period drives down the market-clearing price in both the first and

second periods. Second, because of second period production, first-period machines have a higher

depreciation rate. This has important implications for the financing contract that banks can offer,

which we now consider.

The financing contract consists of a down payment, db1, and a promised second period payment,

db2. Competitive banking implies that db1 + db2 = pb1 (given our assumption of no discounting). For

simplicity we assume that borrowers default when they owe more than their machines are worth

(though we relax this assumption later), which implies the collateral constraint db2 ≤ pb2. The down

payment must then satisfy db1 ≥ pb1 − pb2 = 3a5 . Indeed, under the assumption that funds are scarce

for the buyer, this constraint would bind, and the unique optimal debt contract would be given by

db1 =3a

5, db2 =

3a

10. (11)

Intuitively, borrowers’ down payments must be sufficient to cover depreciation in order to ensure

repayment.18

A.2 Equilibrium with Vendor Financing

We now extend the environment analyzed by Bulow (1982) to allow for secured debt financing

from the manufacturer and ask: can the manufacturer use the terms of financing to solve its time-

18Technically, we need an additional assumption to rule out the optimality of contracts with the same down paymentbut higher second period payments. Such contracts would result in default for sure and could be ruled out by theexistence of any deadweight costs of default.

45

inconsistency problem?

As with bank financing, the vendor financing contract consists of a required down payment, d1,

and a promise to pay, d2. We again start by analyzing the firm’s problem in period 2. In the second

period the firm inherits the promised debt payment, d2, and the first period production, q1, as

state variables. For convenience, let φd be an indicator variable for default, which occurs whenever

first period buyers find that they owe more than their machines are worth (d2 > p2). Notice that

conditional on the state variables d2 and q1, φd is a function of second period production. In

particular, d2 > p2 = a − b(q1 + q2) ⇐⇒ q2 >1b (a − bq1 − d2). Thus, borrowers will default if

the manufacturer produces any more than q̄2 ≡ 1b (a − bq1 − d2). We will refer to q̄2 as the default

boundary. Using this notation, we can write the firm’s problem at time 2 as:

maxq2

(a− b(q1 + q2))q2 + (1− φd) · d2 · q1 + φd · (a− b(q1 + q2))q1 (12)

The first term in (12) represents the profit from selling new machines in the second period; the

second term represents the cash flow from the remaining debt payment on first period machine

sales; and the third term represents the firm’s proceeds from repossessing machines in default.

The first-order condition for (12) is given by

q2 =1

2b(a− bq1(1 + φd)). (13)

When φd = 0 (no default), (13) reduces to the same production choice made by the manufacturer

with bank financing above. Intuitively, if the threat of default is not in play, it cannot provide any

commitment to the manufacturer. In contrast, when φd = 1, (13) becomes q1 + q2 = a2b , the same

production choice made by the manufacturer with full commitment.

As indicated by (13), the optimal second period production drops discontinuously at the default

boundary. Thus, for any q̄2 ∈[a2b − q1,

a2b −

q12

], the optimal production will occur at the default

boundary. Of course, since the default boundary is determined by the manufacturer’s choice of

production and financing terms in the first period, the manufacturer could choose the default

boundary so as to achieve the full-commitment equilibrium.

46

To see this, we begin by assuming that second period production occurs at the default boundary,

solve for the optimal first period production and financing terms, and then check that this solution

is consistent with our assumption. If second period production occurs at the default boundary, then

q2 = 1b (a− bq1 − d2), p2 = d2, and π2 = 1

b (a− bq1 − d2)d2. Substituting these relationships into the

manufacturer’s first period objective function gives:

maxq1,d1,d2

(d1 + d2)q1 +1

b(a− bq1 − d2)d2

subject to d1 + d2 ≤ a− bq1 + d2

(14)

The first term in (14) is the proceeds from the sale of equipment produced in the first period, and the

second term is the second period profits assuming production occurs at the default boundary. The

constraint says that the sum of the debt payments cannot exceed the price of machines produced in

the first period. The right-hand side of the constraint is the purchase price of machines produced in

the first period, where we have substituted d2 for p2 since these are equal at the default boundary.

It is clear that the constraint must bind, since otherwise the firm could simply increase d1 to

make itself better off. Substituting the constraint into the objective function and taking first-order

conditions gives q∗1 = a2b , d

∗2 = a

2 . Combining this solution with the solution to (12) gives the firm’s

optimal production and financing policies under vendor financing:

q∗1 =a

2b, p∗1 = a, d∗1 =

a

2, π∗ =

a2

2b

q∗2 = 0, p∗2 =a

2, d∗2 =

a

2

(15)

By providing secured debt financing, the firm is able to achieve the full-commitment solution to

its optimization problem, with the accompanying features of lower depreciation rates and higher

loan-to-value ratios.19 It is straightforward to verify that our assumption that the firm optimally

produces at the default boundary in the second period holds.20

19Note that this solution is the unique optimum without default. The optimal production path could also be achievedby lending more than d2 = a

2, though this would result in default.

20In this case, q̄2 = 0, and the condition for optimal production to occur at the default boundary is q̄2 ∈ [0, a4b

].

47

A.3 Predictions

There are two important differences between the manufacturer’s solution with bank financing and

vendor financing. First, since vendor finance facilitates commitment to high future prices, machines

depreciate more slowly when they are financed by the manufacturer. Second, the manufacturer is

able to offer a higher loan-to-value when it finances its own machines. This is tied to the depreciation

rate. Because the machines will be worth more in the future, the manufacturer can lend more

without pressing the buyers into default. In fact, since the threat of default is what provides

commitment to the manufacturer, it is more accurate to say that the manufacturer must offer

a higher loan-to-value. If it were to offer the same financing contract as we found under bank

financing, it would lack the commitment to keep secondary market prices high.

A.4 Discussion

We have made a number of simplifying assumptions in order to present the intuition behind secured

vendor financing as a solution to the Coase conjecture as simply as possible. In this section we discuss

the importance (or lack thereof) of these assumptions in arriving at the empirical predictions.

First, we have assumed a two period model. Stokey (1981) studied the Coase conjecture with

an infinite horizon in both discrete and continuous time. In discrete time, the manufacturer (absent

commitment) chooses production to maximize the area under the residual (unmet) demand curve

each period, asymptotically approaching market saturation. In this case, there is always production

and, therefore, quantity-driven depreciation for a manufacturer who cannot commit to restrict

production.

In continuous time, the uncommitted manufacturer instantly saturates the market, confirming

Coase’s intuition. In this case, neither the committed nor the uncommitted manufacturer would

produce after their initial, instantaneous production, so there would be no wedge in the deprecia-

tion rates based on levels of commitment. However, Kahn (1986) showed that, in a continuous-time

setting with increasing marginal production costs, the uncommitted monopolist will smooth his

production over time, thus preventing the stark (and “profoundly unsatisfying,” in Kahn’s words)

outcome of instant market saturation. Production dynamics of committed vs. uncommitted mo-

48

nopolists have properties very similar to those in the two period setting: there is a startup period

of time during which the committed monopolist produces more than the uncommitted monopolist

(as in period 1 in the two period setting), more quickly approaching his lower asymptotic stock.

After this startup period, production (and depreciation) are forever higher for the uncommitted

manufacturer.

The predicted relationship between vendor finance (as a proxy for commitment) and depreciation

is, thus, not dependent on the two period setup. It holds in an infinite-horizon, discrete-time

environment, or in a continuous-time environment with some friction that prevents instantaneous

market saturation.

In deriving the optimal financing contracts, we have assumed that borrowers are constrained.

With bank financing, this assumption ensures that borrowers prefer the largest loan available, as

opposed to being indifferent between any two fairly priced loans of different sizes. Borrower con-

straints have no effect on the optimal debt contract under vendor financing, since the manufacturer

benefits from providing sufficiently large loans to achieve commitment. Thus, borrower constraints

are not important for our results on the relationships between vendor financing and depreciation

nor between vendor financing and loan amounts.

Finally, we have made two assumptions that merit further attention. First, we have assumed

that either banks or captives do all of the financing. In the next section we relax this assumption

and derive the relationship between the level of financing that comes from the manufacturer and the

depreciation of equipment. In the final section, we relax the assumption that all borrowers default

when they owe more than their machines are worth.

A.5 Equilibrium with Captive and Bank Financing

Empirically we observe that most machine types are neither purely captive nor purely bank financed,

and the level of captive support is a key variable in many of our tests. To support the empirical

analysis, we now analyze the relationship between the level of captive financing and depreciation

when (for exogenous reasons) the manufacturer finances a fraction α of the machines it produces.

Financing only a fraction of the machines limits the manufacturer’s ability to commit to high resale

49

prices. A manufacturer which internalizes the second period price of the machines it finances faces

the problem:

maxq2

p2(q2) · q2 + p2(q2) · αq1. (16)

The first term in (16) captures the manufacturer’s proceeds from second period production, while

the second term captures the manufacturer’s exposure to resale prices. The first-order condition

and associated price are:

q∗2 =1

2b(a− bq1(1 + α)), p∗2 =

1

2(a− bq1(1− α)). (17)

Clearly, second period production (price) is decreasing (increasing) in the proportion of machines

financed, α.

Substituting this second period solution into the manufacturer’s first period problem, it must

solve:

maxq1

p1(q1) · q1 + π∗2, (18)

where π∗2 = p∗2 · q∗2. Taking first-order conditions gives the following solution:

q∗1 =2a

b(5− 2α+ α2), p∗1 =

a

2

(9− 4α+ 3α2

5− 2α+ α2

), π∗ =

a2

4b

(9− 2α+ α2

5− 2α+ α2

)q∗2 =

a

2b

(3− 4α+ α2

5− 2α+ α2

), p∗2 =

a

2

(3 + α2

5− 2α+ α2

).

(19)

The financing contract offered by the manufacturer that supports this solution is given by setting

the loan size, d2, equal to the desired second period price, p2. Thus, the financing contract is given

by

d∗1 = a

(3− 2α+ α2

5− 2α+ α2

), d∗2 =

a

2

(3 + α2

5− 2α+ α2

). (20)

It is straightforward to show that the manufacturer’s profit is strictly increasing in the proportion

of machines it finances internally. Similarly, the depreciation rate and the down payment required

are strictly decreasing in the proportion of machines financed internally. Thus we expect to see

50

that when manufacturers finance a larger fraction of their machines, the machines depreciate more

slowly and the lender is able to give larger loans.

In this setting, competitive banks would offer the same financing contract being offered by the

manufacturers. Thus, the lower depreciation made possible through vendor financing will lead to

larger loans being made on strongly captive backed machine models, even for individual machines

which happen to be financed by banks.

A.6 Relaxing Default Assumptions

To this point, we have assumed that any borrower who owes more than his machine is worth will

default. We now relax this assumption, allowing instead for a proportion λ to default. λ may be

interpreted as the proportion of borrowers who are in financial distress. Alternatively, any limits to

recourse that exist in practice would result in a higher λ.

Suppose, as in the previous section, that the manufacturer finances a proportion α of its first

period output. In this case, the manufacturer’s exposure to resale values is capped at a proportion

α · λ of its first period sales. Thus, the manufacturer’s problem is identical to that in the previous

section, with αλ replacing α. The optimal solution, then, would be given by:

q∗1 =2a

b(5− 2αλ+ (αλ)2), p∗1 =

a

2

(9− 4αλ+ 3(αλ)2

5− 2αλ+ (αλ)2

), π∗ =

a2

4b

(9− 2αλ+ (αλ)2

5− 2αλ+ (αλ)2

)q∗2 =

a

2b

(3− 4αλ+ (αλ)2

5− 2αλ+ (αλ)2

), p∗2 =

a

2

(3 + (αλ)2

5− 2αλ+ (αλ)2

).

(21)

The financing contract that supports this optimal solution is given by:

d∗1 = a

(3− 2αλ+ (αλ)2

5− 2αλ+ (αλ)2

), d∗2 =

a

2

(3 + (αλ)2

5− 2αλ+ (αλ)2

). (22)

It is straightforward to show that the relationships of interest hold up. As long as the manu-

facturer is exposed to some threat of default (λ > 0), machines which receive more captive finance

support depreciate more slowly and are associated with larger loan sizes. Moreover, this analysis

highlights an additional mechanism available to the manufacturer apart from the level of captive

finance support. Since the level of commitment depends only on α ·λ, the manufacturer can achieve

51

a given level either by providing lots of financing (increasing α) or by focusing its financing on

riskier borrowers (increasing λ). This is consistent with anecdotal evidence that captives lend to

riskier borrowers than do banks.

52

Appendix B Supplementary Tables and Figures

Table A1: Machine Types and Manufacturers. The table shows the top 10 equipment typesand manufacturers from the UCC filing data. Included are both new and used sales from 1990-2012.Ranking is based on number of observations.

Top 10 Machine Types Top 10 Manufacturers(1) (2)

1 SKID STEER LOADERS CATERPILLAR2 EXCAVATORS DEERE3 CRAWLERS CASE4 INDUSTRIAL TRACTORS BOBCAT5 WHEEL LOADERS KOMATSU6 ROUGH TERRAIN FORKLIFTS NEW-HOLLAND7 COMPACTORS INGERSOLL-RAND8 TRENCHERS VOLVO9 AERIAL LIFTS DITCH-WITCH10 DRILLS FORD

53

B.1 Ingersoll Rand vs. Volvo DD-22/24 Compactors

After Ingersoll Rand’s sale of its road construction unit to Volvo in 2007, machines continued to

be manufactured with the same designs in the same facilities. This helps to rule out the possibility

that the changes we observe in resale performance after the acquisition are driven by changes in

machine characteristics or build. As an example, the figures show the spec sheets for the most

popular Ingersoll Rand roller (DD-24) in 2006 and the Volvo DD-24 roller in 2008. Comparison of

the spec sheets reveals nearly identical machines.

Introducing reliable performance in a smaller package

Ingersoll Rand combined our global engineering resources to bring you the DD-22 andDD-24 vibratory asphalt compactors. Offering high productivity and unsurpassed value,these machines feature industry-leading drum performance, superior propulsioncharacteristics, and unmatched operation intervals in a class-leading 2.5-tonne machine.Designed for optimum compaction of complex materials, the DD-22 and DD-24 providehigh performance in a small package. Move your business and bottom line in the rightdirection with the DD-22 and DD-24.

AVAILABLE OPTIONSn Biodegradable oiln Dual drive leversn Edge compactor and cuttern Foldable ROPS assemblyn Hydraulic traction control n Inside neoprene wipersn Offset articulation joint (50 mm)n Work and road light packages

MODEL DD-22 DD-24MACHINE WEIGHTSOperating Weight – lb (kg) 5,400 (2450) 5,725 (2600)Weight @ Front Drum – lb (kg) 2,700 (1125) 2,863 (1300)Weight @ Rear Drum – lb (kg) 2,701 (1125) 2,863 (1300)Shipping Weight – lb (kg) 5,070 (2300) 5,400 (2450)MACHINE DIMENSIONSLength – in (mm) 95 (2420) 95 (2420)Width – in (mm) 39.4 (1000) 47.2 (1200)Height (top of steering wheel) – in (mm) 75 (1909) 75 (1909)Height (top of ROPS / FOPS) – in (mm) 103 (2621) 103 (2621)Drum Base – in (mm) 68 (1720) 68 (1720)Curb Clearance – in (mm) 9.8 (250) 9.8 (250)Inside Turning Radius (to drum edge) – in (mm) 110 (2800) 104 (2650)DRUMWidth – in (mm) 39 (1000) 47 (1200)Diameter – in (mm) 28 (700) 28 (700)Shell Thickness – in (mm) 0.5 (13) 0.5 (13)Finish MachinedVIBRATIONFrequency – vpm (Hz) 3,300 / 4,020 (55 / 67) 3,300 / 4,020 (55 / 67)Nominal Amplitude – in (mm) 0.02 (0.53) 0.02 (0.53)Centrifugal Force – lb (kN) High 8,318 (37) 8,992 (40)

Low 5,620 (25) 6,295 (28) PROPULSIONType Closed-loop hydrostaticDrum Drive Radial piston Travel Speed – mph (km/h) 0 – 6.2 (0 – 10) 0 – 6.2 (0 – 10)ENGINEMake / Model Cummins A1700 Cummins A1700Engine Type 3-cylinder dieselRated Power @ 2,500 rpm – hp (kW) 32 (24) 32 (24)Electrical 12 V, 45 A alternatorBRAKESService HydrostaticParking Secondary Spring-applied, hydraulic release on each drumWATER SYSTEMType PressurizedPumps Electric diaphragmSpray Bars 1 per drum, PVCNozzles 4 per spray barFiltration Inlet and nozzle filtrationDrum Wipers Spring-loaded, self-adjusting neoprene wipersWater Tank Capacity – gal (L) 68 (260) 68 (260)MISCELLANEOUSArticulation Angle + / - 30° + / - 30°Oscillation Angle + / - 10° + / - 10°Fuel Tank Capacity – gal (L) 10.6 (40) 10.6 (40)Hydraulic Oil Capacity – gal (L) 7.1 (27) 7.1 (27)Gradeability (theoretical) 30% 30%

Product improvement is a continuing goal at Ingersoll Rand. Designs and specifications are subject to changewithout notice or obligation.

(877) IR BRAND • ingersollrand.com

© 2006 Ingersoll-Rand Company 14-0013

PERFORMANCE FEATURESn Dual amplitude drums n Excellent visibility to drum edges at front and rear

of machine n High frequency (over 4,000 vpm) allows for maximum

rolling speeds n Liquid cooled 32-hp, 3-cylinder diesel enginen Neoprene wipers to keep water on drum surfacen Pressurized water system with flow control; 4 nozzles

per drum

SERVICEABILITY

Unobstructed daily checkpoints, easy-to-lift engineenclosure, plus ground-level access to the engine,radiator, battery, and filters. Color-coded service chartsshow detailed checkpoints, service intervals, and properlubricants.

Figure A1: 2006 Ingersoll Rand DD-22/24 Compactor Spec Sheet.

54

Specifications

Model DD22 DD24 Machine Weights (w/ ROPS) Operating Weight kg (lb) 2 450 (5,400) 2 600 (5,725) Weight @ Front Drum kg (lb) 1 225 (2,700) 1 300 (2,863) Weight @ Rear Drum kg (lb) 1 225 (2,700) 1 300 (2,863) Shipping Weight kg (lb) 2 300 (5,070) 2 450 (5,400) Machine Dimensions Length mm (in) 2 420 (95) 2 420 (95) Width mm (in) 1 090 (43) 1 290 (51) Height (top of steering wheel) mm (in) 1 909 (75) 1 909 (75) Height (top of ROPS) mm (in) 2 624 (103.3) 2 624 (103.3) Drum Base mm (in) 1 720 (68) 1 720 (68) Curb Clearance mm (in) 490 (19.3) 490 (19.3) Inside Turning Radius (to drum edge) mm (in) 2 800 (110) 2 700 (106) Drum Width mm (in) 1 000 (39) 1 200 (47) Diameter mm (in) 700 (28) 700 (28) Shell Thickness mm (in) 13 (0.5) 13 (0.5) Finish Machined Vibration Frequency Hz (vpm) 55 / 67 (3,300 / 4,000) 55 / 67 (3,300 / 4,000) Nominal Amplitude mm (in) 0,53 (0.02) 0,53 (0.02) Centrifugal Force kN (lb) High 37 (8,318) 40 (8,992)

Low 25 (5,620) 28 (6,295) Propulsion Type Closed-loop hydrostatic Drum Drive Radial piston Travel Speed km/h (mph) 0 – 10 (0 – 6.2) 0 – 10 (0 – 6.2) Engine Make / Model Cummins A1700 Engine Type 3-cylinder diesel Rated Power @ Installed Speed kW (hp) 24 (32) 24 (32) Electrical 12 volts, 45 A alternator Brakes Service Hydrostatic Parking Secondary Spring-applied, hydraulic release on each drum Water System Type Gravity Pressurized Gravity Pressurized Nozzles – 4 per drum – 4 per drum Filtration Inlet filter Inlet and nozzle filtration Inlet filter Inlet and nozzle filtration Drum Wipers Spring-loaded, self-adjusting urethane wipers Spring-loaded, self-adjusting urethane wipers Water Tank Capacity l (gal) 225 (59) 260 (68) 225 (59) 260 (68) Miscellaneous Articulation Angle + / - 30° + / - 30° Oscillation Angle + / - 10° + / - 10° Fuel Tank Capacity l (gal) 40 (10.6) 40 (10.6) Hydraulic Oil Capacity l (gal) 27 (7.1) 27 (7.1) Gradeability (theoretical) 30% 30%

Product improvement is a continuing goal at Volvo. Designs and specifications are subject to change without notice or obligation.

Figure A2: 2008 Volvo DD-22/24 Compactor Spec Sheet.

55

B.2 Distribution of Estimated Depreciation Rates

0.5

11.

5D

ensi

ty

-.5 0 .5 1 1.5 2Estimated Model Depreciation (winsorized at top and bottom 1%)

Figure A3: Depreciation Rates. The figure plots the depreciation rates used in Table 1. Weestimate depreciation rates by regressing the log sales price on log machine age. Regressions areestimated on a model-by-model level. Before running model-level regressions, year fixed effects,estimated over the entire sample, are removed from both equipment age and price. Thus, time fixedeffects are treated as constant across model-specific regressions. Depreciation should be interpretedas the expected percentage loss in value for a proportional increase in age.

56

B.3 Captive Finance vs. Secured Finance (Table 1 Robustness)

One potential weakness of our measure of captive finance support is that it is conditional on secured

financing so that the denominator misses any sales financed with cash or unsecured credit. It is

possible, then, that there is an an omitted variable capturing total secured financing (banks and

captives) that is correlated with our captive share measure but not controlled for in regressions.

While we are not able to observe model level sales except through the UCC statements, we

are able to come up with some crude measures of manufacturer level sales figures for a handful of

publicly-traded manufacturers. Specifically, for seven of the largest producers of machines (Deere,

Caterpillar, CNH, Hitachi, Komatsu, Kubota, and Terex), we infer the value of North American

equipment sales from annual reports as total sales times the percentage of total sales in North

America times the percentage of sales due to construction equipment. These percentages are often

reported in the text of annual reports—for a few manufacturers, the breakdown of North American

construction sales is directly reported. We then use this as the denominator for a new captive share

with a summation of estimated machine values for new captive financed sales from the UCC data in

the numerator. This gives us the percentage of total sales—not financed sales—which the captive

had a role in. More importantly, we can then make a similar calculation for the total percentage of

sales (captive and non-captive) which were financed in the UCC data. This allows us to control for

the share of purchases which are financed with secured financing. Note that these calculations are all

done as of 2012 but applied retroactively to all past models produced by these seven manufacturers.

Going back further, the quality of data on segment sales breakdowns becomes less consistent.

Table A2 shows the results of regressing model-specific depreciation rates on both the captive

finance share and secured finance share. The large caveat here is that we can only examine man-

ufacturer level covariation between financing and depreciation. Moreover, we can only do so for

seven manufacturers, so we view the results with some caution. However, it is encouraging to note

that, even controlling for the omitted variable of concern (total percent of sales financed via secured

debt), our result is robust and within striking distance of the magnitude in the main results in

Table 1. The key takeaway is that the slope on the captive finance share is -0.321 after controlling

for the total percentage of sales financed, significant at the 5% level. Perhaps more importantly,

57

total financing is not significantly correlated (economically or statistically) with depreciation rates.

Table A2: Captive finance vs. secured finance. The table provides an alternative estimateof the relationship between the resale value of heavy machinery and the extent of captive financingin the primary market. Depreciation rates are estimated as in Table 1 for each make and model(conditional on having 30 or more resale observations). −δmodel, the model-specific depreciationrate, captures the percentage loss in value for proportional increases in age. The table then reportsthe results of the following regression:

−δmodel = α+ β1CaptiveF inancing

Sales + β2TotalSecuredF inancing

Sales + β3...kControls+ ε.The level of sales is inferred from North American equipment sales times the percentage of sales inNorth America times the percentage of sales due to construction equipment from the 2012 annualreports of seven of the largest publicly-traded producers of machines (Deere, Caterpillar, Case-NewHolland, Hitachi, Komatsu, Kubota, and Terex). Captive Financing is measured as the summationof the estimated machine values for new captive financed sales in the UCC data in 2012, andsimilarly for Total Secured Financing. These 2012 calculations are then applied retroactively toall past models produced by the seven manufacturers. Foreign Manufacturer is an indicator forwhether the manufacturer is headquartered outside the US. ln(GlobalSales) is the natural log of thetotal global sales for each manufacturer. Standard errors are clustered at the manufacturer level,are robust to heteroskedasticity, and are reported in parentheses. ***, **, and * signify resultssignificant at the 1, 5, and 10% levels, respectively.

Depreciation Rate (1)

Captive Financed Volume/Sales (2012) -0.32**(0.15)

Total Financed Volume/Sales (2012) -0.03(0.07)

Foreign Manufacturer -0.09**(0.04)

ln(Global Sales) 0.03**(0.02)

Manufacturer Fixed Effects NOEquipment Type Fixed Effects YESEquipment Size Fixed Effects YESObservations 865R-squared 0.43

58

B.4 Alternative Control for Physical Condition (Table 2 Robustness)

In columns 1-2 of Table 2 we measure price depreciation absent the effect of physical deterioration by

controlling for machine condition in the make × model level depreciation regressions. The resulting

measure of depreciation holds condition fixed and thus approximates the changes in prices for “new,

old-stock” machines. Here we present an alternative approach to controlling for machine quality.

Table A3 adds an estimated measure of physical durability at the model level as a control to the

regressions in columns 1 and 2 of Table 1. Specifically, we use survival analysis to estimate a survival

function for each model in the auction data, defining failure as a machine transitioning to fair or

poor condition. Our survival function is estimated using an exponential model with a constant

hazard for each model. As a measure of durability, we use the logged median survival time for a

machine, backed out of the estimated survival function.

Because we lose observations due to missing data on machine condition, columns 1 and 2 first

replicate columns 1 and 2 of Table 1 with limited effect on coefficient magnitudes. Columns 3

and 4 add log median survival time for the model level observations as a measure of durability

as a control. Two results are worth noting. First, log median survival time is very effective in

predicting dollar depreciation rates. In each regression, it is highly significant (t-statistics near 6)

and generates at least a modest increase in the regression r-squared. While this should not be

surprising, it suggests the measure of durability is doing its job. Second, as we add durability as a

control, the effect of model captive support on dollar depreciation is all but unchanged. The fact

that adding a potentially noisy control does nothing to attenuate the coefficient of interest suggests

that even a perfectly measured control would be unlikely to qualitatively change the results. Finally,

consistent with durability predicting depreciation but not affecting the coefficient on model captive

support, columns 5 and 6 confirm there is no obvious significant relationship between the measured

durability and captive financing.

59

Table A3: Durability and captive finance. The table documents the relationship betweenmachine durability and captive financing in the primary market. Durability is estimated for eachmake and model as the log median survival time (in years) from a survival model. Since actualmachine failure is not observable, we instead treat a machine failure as moving from good to fair orpoor condition. The survival function is estimated in the auction data using an exponential modelwith a constant hazard for each machine, and median survival time is the value at which the survivalfunction equals 0.5. Columns 1 and 2 repeat the regressions from Table 1 for the limited sample ofmachines for which we can measure durability, given that condition is only reported for a subset ofthe auction observations. Columns 3 and 4 then add durability as a control. Finally, columns 5 and6 estimate the relationship between durability and the level of captive finance support. Errors areclustered at the manufacturer level, are robust to heteroskedasticity, and are reported in parentheses.***, **, and * signify results significant at the 1, 5, and 10% levels, respectively.

Depreciation Durability

(1) (2) (3) (4) (5) (6)

Model Captive Support -0.14*** -0.23*** -0.14** -0.20** 0.07 -0.24(0.04) (0.03) (0.06) (0.08) (0.10) (0.18)

Model Durability -0.09*** -0.08***(0.01) (0.02)

Manufacturer Fixed Effects NO YES NO YES NO YESEquipment Type Fixed Effects YES YES YES YES YES YESEquipment Size Fixed Effects YES YES YES YES YES YESObservations 882 882 882 882 882 882R-squared 0.58 0.67 0.60 0.68 0.46 0.54

60

B.5 Ingersoll Rand First Stage

Figure 4 in the text documents the jump in captive financing that occurred for road construction

equipment models that were acquired by Volvo from Ingersoll Rand in 2007. This evidence motivates

the regressions in Table 4, in which we test for the hypothesized impact of an increase in captive

finance support on the subsequent resale performance of the formerly Ingersoll Rand-made machines.

In Table A4, we formally present the first-stage regression of the effect of the acquisition on the

captive support received by the acquired equipment models. In column 1, a dummy representing

whether or not a new compaction or paving machine was captive financed is regressed on a post-

2007 dummy, interacted with a dummy for machines made by Ingersoll Rand prior to 2007 or

Volvo during and after 2007, along with model-level and year fixed effects. Holding model-specific

attributes fixed, the spin-off raised captive financing support by 44%. The effect is economically

significant and has a t-statistic of 8.4, indicating that this is a strong first stage.

In columns 3 and 4 of Table 4, we divide the sample into machine types that received either

high or low (relative to the median) captive support from manufacturers other than Ingersoll Rand

or Volvo prior to 2000. This out-of-sample measure of machine captive propensity provides a pre-

determined source of expected treatment intensity. Column 2 of Table A4 runs the associated

first-stage regression. The coefficient on the interaction with Machine Captive Propensity indicates

that the treatment effect is, in fact, significantly larger for those machines that we would expect to

receive more captive finance support from Volvo after the acquisition.

61

Table A4: Ingersoll Rand sale to Volvo first stage. The table documents the change inthe level of captive finance support for machines formerly made by Ingersoll Rand after the saleof its road construction unit to Volvo. The sample consists of all new transactions of paving andcompaction equipment in the UCC data between 2000-2013. IR is a dummy for machines madeby Ingersoll Rand prior to 2007 and made by Volvo in 2007 and after. Post2007 is a dummyfor transactions occurring in 2007 or after. Column 2 adds an interaction with Machine CaptivePropensity, which measures the captive support provided for each machine type out of sample (fornon-Ingersoll Rand/Volvo producers and before 2000), as a pre-determined source of variation inexpected treatment intensity. Standard errors are clustered at the manufacturer level, are robustto heteroskedasticity, and are reported in parentheses. ***, **, and * signify results significant atthe 1, 5, and 10% levels, respectively.

Transaction Captive Financed (1) (2)

Post2007 × IR 0.44*** 0.15***(0.05) (0.04)

Post2007 × IR × Machine Captive Propensity 1.72***(0.36)

Make × Model Fixed Effects YES YESYear Fixed Effects YES YESMachine Captive Propensity × Treat NO YESMachine Captive Propensity × Post NO YESObservations 65,969 65,969R-squared (within Make × Model) 0.03 0.03

62

B.6 Ingersoll Rand Parallel Price Trend

Table 4 documents the impact of Ingersoll Rand’s sale of its road construction division to Volvo and

the related increase in captive financing on the resale performance of machines sold at auction. After

the spin-off, the formerly Ingersoll Rand-made machines show lower depreciation rates, consistent

with captive financing providing the commitment necessary to support machine prices. Column

6 showed the related effect of the spin-off on the level of machine prices, with acquired models

appreciating by 6% on average. Table A5 examines the parallel trends assumption by estimating

the level of machine prices during the years leading up to and following the acquisition. The table

shows an immediate increase in average machine prices upon acquisition with no evidence of a

pre-existing upward trend.

63

Table A5: Ingersoll Rand price trend. The table documents the average annual price level forIngersoll Rand/Volvo made machines in the years leading up to and following the sale of IngersollRand’s road construction unit to Volvo. The sample consists of all transactions of paving andcompaction equipment in the auction data between 2000-2013. IR is a dummy for machines madeby Ingersoll Rand prior to 2007 and made by Volvo in 2007 and after. Each year’s effect is measuredrelative to the pre-2003 price level. Standard errors are clustered at the manufacturer level, arerobust to heteroskedasticity, and are reported in parentheses. ***, **, and * signify results significantat the 1, 5, and 10% levels, respectively.

ln(Machine Price) (1)

ln(1+ MachineAge) -0.43***(0.02)

IR × 2003 0.01(0.02)

IR × 2004 -0.00(0.02)

IR × 2005 0.02(0.04)

IR × 2006 -0.06(0.04)

IR × 2007 0.05(0.03)

IR × 2008 0.07**(0.03)

IR × 2009 0.06***(0.02)

IR × 2010 0.02(0.02)

IR × ≥2011 0.09(0.06)

Make × Model Fixed Effects YESYear Fixed Effects YESCondition Fixed Effects YESAuctioneer Fixed Effects YESObservations 14,368R-squared 0.42

64

References

Barrot, Jean-Noel, 2016, Trade credit and industry dynamics: Evidence from trucking firms, The

Journal of Finance 71, 1975–2016.

Benmelech, Efraim, and Nittai K. Bergman, 2009, Collateral Pricing, Journal of Financial Eco-

nomics 91, 339–360.

Benmelech, Efraim, Ralf Meisenzahl, and Rodney Ramcharan, 2016, The Real Effects of Liquidity

During the Financial Crisis: Evidence from Automobiles, .

Biais, Bruno, and Christian Gollier, 1997, Trade credit and credit rationing, The Review of Financial

Studies 10, 903–937.

Bodnaruk, Andriy, William O’Brien, and Andrei Simonov, 2016, Captive finance and firm’s com-

petitiveness, Journal of Corporate Finance 37, 210 – 228.

Borenstein, Severin, Jeff Mackie-Mason, and Janet Netz, 1995, Antitrust policy in aftermarkets,

Antitrust Law Journal 63, 455–482.

Bulow, Jeremy, 1982, Durable Goods Monopolists, Journal of Political Economy 90, 314–332.

, 1986, An Economic Theory of Planned Obsolescence, Quarterly Journal of Economics 101,

729–749.

Butz, David, 1990, Durable-Good Monopoly and Best-Price Provisions, American Economic Review

80, 1062–1076.

Chevalier, Judith, and Austan Goolsbee, 2009, Are Durable Goods Consumers Forward-Looking?

Evidence from College Textbooks, Quarterly Journal of Economics 124, 1853–1884.

Coase, Ronald H., 1972, Durability and Monopoly, Journal of Law and Economics 15, 143–149.

Desai, Preyas, and Devavrat Purohit, 1998, Leasing and selling: Optimal marketing strategies for

a durable goods firm, Management Science 44, S19–S34.

65

Edgerton, Jesse, 2012, Credit Supply and Business Investment During the Great Recession: Evi-

dence from Public Records of Equipment Financing, Working paper, Federal Reserve Board.

Eisfeldt, Andrea L., and Adriano A. Rampini, 2009, Leasing, Ability to Repossess, and Debt Ca-

pacity, Review of Financial Studies 22, 1621–1657.

Emery, Gary, and Nandkumar Nayar, 1998, Product quality and payment policy, Review of Quan-

titative Finance and Accounting 10, 269–284.

Hortacsu, Ali, Gregor Matvos, Chad Syverson, and Sriram Venkataraman, 2013, Indirect Costs of

Financial Distress in Durable Goods Industries: The Case of Auto Manufacturers, Review of

Financial Studies 26, 1248–1290.

Kahn, Charles, 1986, The durable goods monopolist and consistency with increasing costs, Econo-

metrica 54, 275–294.

Karp, Larry, and Jeffrey Perloff, 1996, The Optimal Suppression of a Low-Cost Technology by a

Durable-Good Monopoly, Rand Journal of Economics 27, 346–364.

Kutsoati, Edward, and Jan Zabojnik, 2001, Durable goods monopoly, learning-by-doing, and ‘sleep-

ing patents’, Working paper, University of Southern California.

Lee, Yul W., and John D. Stowe, 1993, Product risk, asymmetric information, and trade credit,

The Journal of Financial and Quantitative Analysis 28, 285–300.

Long, Michael, Ileen Malitz, and Abraham Ravid, 1993, Trade credit, quality guarantees, and

product marketability, Financial Management 22, 117–127.

Mian, Shehzad, and Clifford Smith, 1992, Accounts receivable management policy: Theory and

evidence, The Journal of Finance 47, 169–200.

Rajan, Raghuram G., and Luigi Zingales, 1998, Financial dependence and growth, The American

Economic Review 88, 559–586.

66

Stokey, Nancy L., 1981, Rational expectations and durable goods pricing, The Bell Journal of

Economics 12, 112–128.

Stroebel, Johannes, 2016, Asymmetric information about collateral values, Journal of Finance 71,

1071–1112.

Sutherland, Andrew, 2016, The economic consequences of borrower information sharing: Relation-

ship dynamics and investment, Working paper.

Waldman, Michael, 1996, Planned obsolescence and the r&d decision, Rand Journal of Economics

27, 583–595.

, 1997, Eliminating the market for secondhand goods, Journal of Law and Economics 40,

61–92.

, 2003, Durable Goods Theory for Real World Markets, Journal of Economic Perspectives

17, 131–154.

67

Table 1: Captive finance and resale prices. The table documents the relationship betweenthe resale value of heavy machinery and the extent of captive financing in the primary market.Depreciation rates are estimated for each make and model (conditional on having 30 or more resaleobservations) based on the following regressions:

˜ln(AuctionPricei,t) = αmodelyear + δmodelyear˜ln(1 + EquipmentAgei,t) + εi,t

where subscript i indexes an individual machine and t is the year of the auction. Year fixed effectsare estimated in the full auction sample and partialled out of age and price in a first stage regression.Equipment age is the number of years between the original manufacture date and the date of resale.−δmodel, the model-specific depreciation rate, captures the percentage loss in value for proportionalincreases in age. Columns 1 through 5 report regressions of depreciation rate on the fraction ofnew machines which are financed by captive finance units for each model, along with controls forequipment type and size,

−δmodel = α+ β1ModelCaptiveSupport+ β2...kControlsmodel + ε.Model Captive Support is calculated by model from the UCC filing data and is only included formodels which had 30 or more machines financed from 1990-2012. In columns 1 and 2 it includes bothcaptive loans and leases; in columns 3 and 4 loan and lease support is considered separately; and incolumn 5 the separate measures are both included in the regression. Standard errors are clusteredat the manufacturer level, are robust to heteroskedasticity, and are reported in parentheses. ***,**, and * signify results significant at the 1, 5, and 10% levels, respectively.

Depreciation Rate (1) (2) (3) (4) (5)

Model Captive Support -0.14*** -0.19***(0.04) (0.04)

Model Captive Support (loans) -0.16*** -0.15**(0.06) (0.06)

Model Captive Support (leases) -0.45* -0.32(0.25) (0.20)

Manufacturer Fixed Effects NO YES YES YES YESEquipment Type Fixed Effects YES YES YES YES YESEquipment Size Fixed Effects YES YES YES YES YESObservations 1,750 1,750 1,750 1,750 1,750R-squared 0.46 0.56 0.56 0.56 0.56

68

Table 2: Captive finance and drivers of depreciation. The table documents the relationship between the depreciation ratesof heavy machinery and the extent of captive financing in the primary market. Depreciation rates are estimated in a variety ofways to isolate different drivers of depreciation. Columns 1 and 2 use model-specific depreciation controlling for physical conditionas estimated in the regression in Equation (3) in the text, while columns 3 and 4 use model-vintage-specific depreciation from thesame regression at the model × vintage year level. The depreciation rates capture the percentage loss in value for proportionalincreases in age, holding condition fixed. This represents estimates of the price paths of “new, old-stock” machines. In column 5,the measure of depreciation captures the time-series change in new machine prices by including only machines less than two yearsold and in very good or excellent condition, where machine age is replaced with model age, defined as the number of years sincethe introduction of a given model. Finally, column 6 isolates cross-sectional variation in model prices at a point in time drivenby differences in machine productivity by adding auction year fixed effects to the depreciation regressions from Table 1. Eachcolumn, then, reports regressions of the particular measure of depreciation on the fraction of new machines which are financedby captive finance units, along with controls for equipment type and size. Model Captive Support is calculated as in Table 1,except for in columns 3 and 4, wherein it is calculated for each model × vintage in a backward-looking fashion. Standard errorsare clustered at the manufacturer level (manufacturer and vintage year for columns 3 and 4) , are robust to heteroskedasticity,and are reported in parentheses. ***, **, and * signify results significant at the 1, 5, and 10% levels, respectively.

Depreciation (holding condition constant) ProductivityNew Price Driven

Make × Model Make × Model × Vintage Depreciation Depreciation

(1) (2) (3) (4) (5) (6)

Model Captive Support -0.13*** -0.14** -0.14** -0.20** -0.16*** -0.04(0.05) (0.06) (0.06) (0.08) (0.04) (.04)

Vintage Year Fixed Effects NO NO YES YES NO NOManufacturer Fixed Effects NO YES NO YES YES YESEquipment Type Fixed Effects YES YES YES YES YES YESEquipment Size Fixed Effects YES YES YES YES YES YESObservations 891 891 596 596 160 1,735R-squared 0.36 0.47 0.49 0.54 0.27 0.33

69

Table 3: Captive finance and market power. Columns 1 and 2 estimate the link between market share for a particularmodel and the extent to which that model receives captive support. Market share is estimated for new equipment sales withina given equipment type. It is calculated annually and averaged over the life of the model. Market share estimates in column 1are based on all UCC filings and in column 2 are based only on bank financed sales to avoid capturing market share which isdriven by aggressive captive financing terms. Columns 3-5 extend the regressions of Table 1 to include an interaction term withmodel-specific market share. HighMarketShare is a dummy variable equal to one if a model is above the median market sharein the sample. Column 3 uses the broad measure of market share, whereas column 4 uses the share within the bank-financeduniverse. Column 5 includes as an interaction an alternative measure of market power based on the inverse number of activeproducers for each type of equipment. Sample size is somewhat larger than in Table 1, where only models with more than30 recorded sales were included in the analysis. Below, we include these models so as not to exclude models with low marketshare (the variable of interest). All regressions include manufacturer, equipment type, and size fixed effects. Standard errorsare clustered at the manufacturer level, are robust to heteroskedasticity, and are reported in parentheses. ***, **, and * signifyresults significant at the 1, 5, and 10% levels, respectively.

Model Captive Support Depreciation Rate

(1) (2) (3) (4) (5)

ln(Market Share) 0.04*** 0.02**(0.01) (0.01)

Model Captive Support -0.03 -0.02 -0.02(0.05) (0.06) (0.06)

Model Captive Support × High Market Share -0.11** -0.16***(0.05) (0.04)

High Market Share 0.03 0.07***(0.03) (0.02)

Model Captive Support × Market Concentration ( 1N ) -0.67**

(0.27)

Manufacturer Fixed Effects YES YES YES YES YESEquipment Type Fixed Effects YES YES YES YES YESEquipment Size Fixed Effects YES YES YES YES YESObservations 2,614 2,446 2,611 2,443 2,604R-squared 0.59 0.61 0.45 0.46 0.42

70

Table 4: Ingersoll Rand unit sale and resale performance. The table estimates the impact of Ingersoll Rand’s sale ofits road construction equipment unit to Volvo and the related increase in captive financing that followed on the depreciation ofmachines sold at auction. The sample consists of all individual machine sales of road paving and compaction equipment in theauction data. IR is a dummy for machines made by Ingersoll Rand prior to 2007 and made by Volvo in 2007 and after. Post2007is a dummy for auction sales occurring in 2007 or after. Panel A reports depreciation regressions with make × model × yearbuilt fixed effects for each of the four groups defined by {Pre2007, Post2007} × {Control, Treat}, where the control group isproducers other than Ingersoll Rand or Volvo. Panel B includes all of these interactions in the same regression. The interactionof Post2007 × IR × ln(1+MachineAge) documents the change in depreciation of formerly Ingersoll Rand-made machines afterthe acquisition, relative to similar machines over the same time period. Columns 1 and 2 use the full sample of road paving andcompaction equipment made between 2000 and 2013. Column 3 excludes machines produced by Volvo, limiting the treatmentgroup to only the machines actually produced by Ingersoll Rand before 2007. Columns 4 and 5 divide the sample into low and highcaptive propensity machine types based on the captive support provided for each machine type out of sample (for non-IngersollRand/Volvo producers and before 2000) to focus on pre-determined differences in expected treatment intensity. The sample isdivided at the median captive propensity. Column 6 excludes the interactions with Machine Age to examine the level effect onprices captured in the coefficient on Post2007 × IR. Supporting regressions for the first-stage effect of the acquisition on captivefinance intensity are in Appendix Table A4, while a regression documenting the parallel trend in machine prices is in AppendixTable A5. Standard errors are in parentheses and are clustered at the level of the manufacturer. ***, **, and * signify resultssignificant at the 1, 5, and 10% level, respectively.

Panel APre2007 × Control Post2007 × Control Pre2007 × IR Post2007 × IR

ln(MachinePrice) (1) (2) (3) (4)

ln(1+ MachineAge) -0.33*** -0.37*** -0.44*** -0.33***(.04) (.04) (.06) (.08)

Make × Model × Year Built Fixed Effects YES YES YES YES

71

Panel B Excluding Low Captive High CaptiveVolvo-Made Propensity Propensity

Machines Machines Machinesln(MachinePrice) (1) (2) (3) (4) (5) (6)

ln(1+ MachineAge) -0.41*** 0.09 0.06 0.11** 0.04 -0.43***(0.03) (0.07) (0.07) (0.05) (0.12) (0.02)

Post2007 × ln(1+Machine Age) -0.02 0.03 0.01 0.13 -0.01(0.08) (0.06) (0.06) (0.09) (0.11)

IR × ln(1+Machine Age) -0.07*** -0.08* -0.08* -0.02 -0.20***(0.03) (0.04) (0.04) (0.03) (0.07)

Post2007 × IR -0.20* -0.21** -0.18* -0.04 -0.50** 0.06***(0.12) (0.09) (0.10) (0.07) (0.20) (0.02)

Post2007 × IR ×ln(1+MachineAge) 0.17** 0.17*** 0.15** 0.04 0.37***(0.08) (0.06) (0.06) (0.05) (0.12)

Make × Model Fixed Effects YES NO NO NO NO YESMake × Model × Year Built Fixed Effects NO YES YES YES YES NOYear Fixed Effects YES YES YES YES YES YESCondition Fixed Effects YES YES YES YES YES YESAuctioneer Fixed Effects YES YES YES YES YES YESObservations 14,368 12,606 11,184 6,153 6,429 14,368R-squared 0.42 0.42 0.46 0.52 0.40 0.42

72

Table 5: Captive finance and quantities. The table estimates the relationship between thehistory of captive financing support for equipment and subsequent production volumes of that equip-ment. The analysis is at the make × equipment type × size × vintage year level to avoid problemsin measuring production quantities at the model level that would arise from the introduction ofnew models to replace old ones. Quantities are measured based on new machines financed in theUCC data. Average future production for one, three and five years is normalized by current periodproduction:

1N

N∑n=1

qt+n/qt

Model Captive Support is backward-looking and measures the proportion of new machines of a giventype and size that were financed by the manufacturer prior to the current vintage year. Standarderrors are clustered at the manufacturer level, are robust to heteroskedasticity, and are reported inparentheses. ***, **, and * signify results significant at the 1, 5, and 10% levels, respectively.

N=1 N=3 N=5N-Year Average Future Sales/Current Sales (1) (2) (3)

Model Captive Support -0.06 -0.21** -0.34***(0.06) (0.10) (0.11)

Year Fixed Effects YES YES YESManufacturer Fixed Effects YES YES YESEquipment Type Fixed Effects YES YES YESEquipment Size Fixed Effects YES YES YESObservations 22,414 22,414 22,414R-squared 0.05 0.12 0.13

73

Table 6: Captive finance and pledgeability. The table estimates the relationship between the down payment required forindividual machine buyers and the captive financing support at the model level. ln(Downpayment) is measured as the log ratio ofthe estimated machine value to loan or lease amount as reported for a sub-sample of the UCC data. Model Captive Support is theproportion of new machines of a given model that are financed by the manufacturer over the entire sample, conditional on havingat least 30 transactions financed. Column 1 includes all leases and loans. Column 2 includes only bank financed transactions toavoid capturing differences in terms offered by captives vs. banks. Column 3 includes an interaction with a measure of marketpower based on the inverse number of active producers for each equipment type. Column 4 replaces the Model Captive Supportmeasured over the entire sample with a backward-looking measure of Model Captive Support. Standard errors are clustered atthe manufacturer and year level, are robust to heteroskedasticity, and are reported in parentheses. ***, **, and * signify resultssignificant at the 1, 5, and 10% levels, respectively.

All Bank All AllTransactions Transactions Transactions Transactions

ln(Downpayment) (1) (2) (3) (4)

Model Captive Support -0.25*** -0.24*** 0.05(0.03) (0.08) (0.06)

Model Captive Support × Market Concentration ( 1N ) -1.98***

(0.64)Market Concentration ( 1

N ) 1.32***(0.34)

Model Captive Support History -0.10***(0.03)

Year Fixed Effects YES YES YES YESManufacturer Fixed Effects YES YES YES YESBorrower Controls (State, Industry FEs) YES YES YES YESMachine Controls (Type FEs, Size FEs, ln(age)) YES YES YES YESSale/Lease Dummy YES YES YES YESObservations 58,893 25,369 56,541 45,307R-squared 0.26 0.35 0.30 0.26

74

Table 7: Machine choice and pledgeability in the time series. The table estimates the relationship between the yearlychanges in collateral requirements reported by senior loan officers and the machine choice of bank-financed buyers. The left-hand-side variable is a dummy for whether the consumer chose a captive backed model, defined as a model that received greaterthan the median level of captive support within its equipment type. As in earlier tables the model captive support is estimatedover the life of the model for those models which have at least 30 transactions, except in column 4 where it is backward-looking.Collateral tightening reflects the net percentage of loan officers responding to the Senior Loan Officer Survey reporting tightercollateral requirements on commercial loans to small and medium sized firms during the year prior to the purchase. It is demeanedand given unit standard deviation for ease of interpretation. Column 1 include all sales and leases. Column 2 limits the sampleto bank-financed sales and leases to rule out the effect being driven by greater availability of captive finance during downturns.Column 3 adds an interaction with a measure of market power based on the inverse number of active producers within eachequipment type. Column 4 replaces the left-hand-side variable with a dummy indicating whether the model chosen was abovethe median within its equipment type with respect to a backward-looking measure of captive finance support. Standard errorsare clustered at the manufacturer and year level, are robust to heteroskedasticity, and are reported in parentheses. ***, **, and* signify results significant at the 1, 5, and 10% levels, respectively.

Capt. HistoryCaptive Support ≥ Median ≥ Median

All Transactions Bank Transactions All Transactions All TransactionsCaptive Backed Model Chosen (1) (2) (3) (4)

Collateral tightening over past year 0.02*** 0.02*** 0.01 0.03**(0.01) (0.01) (0.01) (0.01)

Collateral tightening × Market Concentration ( 1N ) 0.14***

(0.01)Market Concentration ( 1

N ) -0.87*(0.42)

Borrower Controls (State, Industry FEs) YES YES YES YESMachine Controls (Type FEs, Size FEs, ln(age)) YES YES YES YESSale/Lease Dummy YES YES YES YESObservations 2,835,511 1,150,262 2,824,264 1,993,691R-squared 0.08 0.06 0.08 0.06

75