Young Firms, Old Capital∗

Song Ma† Justin Murfin‡ Ryan Pratt§

(Preliminary. Please Do Not Circulate Without Permission.)

This paper explores the interaction of capital reallocation and entrepreneurship activities.

We document a robust fact about young firm investment using a sample of machine

transactions covering more than 70,000 models of equipment across a wide range of

industries—that young firms are the predominant buyers of vintage physical capital,

originally owned and seasoned by older firms that are geographically proximate. This

pattern is more pronounced when financial constraints are more likely to bind. Because

young firms depend on older capital, which in turn tends to trade more locally, we

demonstrate there are returns to local agglomeration among young and old firms.

Specifically, the local availability of vintage capital provided by older firms shapes the

investment choices, growth and entry of new businesses in the same county and industry

just as the existence of local new businesses facilitates machine turnover and capital

adjustment in older incumbents.

JEL Classification: L26, E22, G31

Keywords: Capital Reallocation, Entrepreneurship, Investment, Agglomeration

∗This version: June 5, 2020. We thank seminar participants at Chicago Booth, Colorado Finance Summit,Duke/UNC I&E Research Conference, FDIC, Federal Reserve Bank of Philadelphia, Federal Reserve Board,GSU Conference on Financing Tangible and Intangible Capital, HEC Workshop on Entrepreneurship, JuniorEntrepreneurial Finance and Innovation Workshop, Kellogg, Pittsburgh, Princeton, Syracuse, Tilburg, Utah,Washington St. Louis Corporate Finance Conference, and Yale (Applied Micro). We also thank TonyCookson, Francois Derrien, Jason Donaldson, William Mann, Sahil Raina, Adriano Rampini, Xinxin Wang,and Liu Yang for detailed comments. Huijun Sun and Ran You provided excellent research assistance. YaleInternational Center for Finance provided research support. We thank Erik Gilje for generously sharing dataon county-level shale oil shocks. All errors are our own.

†Yale University, song.ma@yale.edu, +1 (203) 436-4687;‡Cornell University, justin.murfin@cornell.edu, +1 (607) 255-8023;§Brigham Young University, ryan.pratt@byu.edu, +1 (801) 422-1222.

1. Introduction

Recent work on business cycle dynamics and growth has emphasized the importance of

young firms and, separately, the process of capital reallocation. In this paper, we link these

two important drivers of growth via a core feature of capital reallocation in practice: that

young firms are the predominant buyers of old capital seasoned by older, established firms in

the same county. We argue this pattern of capital reallocation influences both young firm

investment and capital replacement rates of old firms. Combined, these patterns suggest that,

because of trade in vintage capital, the co-location of young and old firms may shape the

returns to agglomeration.

Our paper proceeds in two parts. First we document the pattern of capital reallocation

across young and old firms and suggest some explanations. Next, taking the pattern as given,

we ask how the potential for trade in old capital shapes the behavior of both young and old

firms.

To document the interaction between firm and machine age, we lean on 1.5 million

transactions covering more than 70,000 models of machines used across a broad range of

industries. Across a wide range of industries and equipment types, young firms deploy older

capital, whereas old firms are the predominant investors in new capital. This correlation is not

obviously explained by selection related to omitted machine or firm characteristics and holds

within firm, within make-model, and even within a uniquely identified machine. On average,

a given reallocated machine is purchased by a firm that is six years younger than its prior

owner. Similar patterns emerge in separate samples covering machine tools, woodworking

tools, printers, copiers, lift trucks, and machines used in logging and construction and are

both statistically and economically significant in more than 80 percent of the industries.

Given the observed reallocation dynamic, what features of older machines make them

relatively more attractive to younger firms? We find evidence consistent with a finance

motive described in Eisfeldt and Rampini (2007) and Rampini (2019). Rampini (2019) points

out that more durable goods have higher prices and that this effect dominates their higher

collateral value in a world with imperfect capital pledgeability. Consequently, more durable

goods (that is, younger equipment with a longer remaining productive life) require larger

1

down payments per unit of capital. Financially constrained (young) firms optimally choose

older capital to lower upfront costs at the expense of a higher user cost, as manifest by

increased down time and/or higher ongoing maintenance costs. Exploiting cross-sectional

measures of financial constraints with an instrument for bank branch liquidity developed in

Gilje, Loutskina, and Strahan (2016), we find that the association between firm and capital

age is most evident in periods/places experiencing tighter credit. Related, Eisfeldt and

Rampini (2009) argues that lease contracts enjoy a repossession advantage relative to loan

contracts, increasing pledgeability of capital and mitigating financial constraints. As a result,

machine choice by young firms may be less distorted when machines are available for lease.

Again, consistent with a financial motive, we find weaker links between capital age and firm

age in rental and leasing markets. In contrast, our evidence does not seem to support theories

based on young firms’ different technological demand relative to older firms (Bond, 1983).

Given the potential importance of used capital to young firms, it makes sense to ask

whether that supply needs to be locally provided. If so, this might suggest important

geographic links between young and old firms, as purchasers and providers of used capital

respectively. Using our ability to track machines across users via their serial numbers, we

confirm that trade in used machines is essentially geographically local, and increasingly so

with capital age. The suggestion of gains from trade across young and old firms in the same

locality motivates the second section of the paper, which asks how the opportunity for used

capital purchases by young firms from local older firms (and conversely the opportunity for

old firms to sell used capital to local younger firms) affects their investment choices, and

perhaps supports returns to co-location.

To explore these ideas, we create a measure of the availability of local vintage capital,

using records of debt-financed equipment purchases from UCC filings to track the transaction

history of new machine purchases within a county. We then track the aggregate latent supply

of machines of a given age that would be potentially available to a given buyer in each county

by year. For example, 100 new machines purchases at time t, in county i, provides a measure

of the five-year-old machine supply in county i at time t+ 5.

Using this variation in old capital supply, we find that a more abundant supply of used

machines at start-up investment is immediately reflected in equipment purchase decisions at

2

the micro-level. We estimate a choice model of machine purchase for young and old firms

based on the latent supply of local capital. Conditional on a purchase being made in a

given quarter, the youngest firms are significantly more sensitive to vintage capital supply

as a determinant of the specific type of equipment that they purchase, and that sensitivity

monotonically declines with firm age. This fact would seem to lend credence to anecdotes

common among entrepreneurs about the sensitivity of early decisions to the local availability

of inputs. As a prominent example, the iconic start-up Ben and Jerry’s chose to make

ice cream after finding a used ice cream truck and freezer for sale locally, but only after

abandoning their initial plan to make bagels due to their inability to find affordable used

bagel machines.1

The Ben and Jerry’s example suggests even more interesting questions. Given the success

of the ultimate enterprise, we might wonder if firms that enjoy access to local supply in used

capital show long-term benefits. For example, if young firms invest based on used capital

availability, how does this predict the volume and variety of their subsequent acquisitions?

Aggregating our supply variable up to an industry level and conditioning on the entry of a

young firm into a given industry, we find that firms entering industries with a large local

supply of used capital invest more in their first three years, and they invest in a greater

variety of equipment types than firms with less access to used capital. Surprisingly, firms

with access to used capital even appear to graduate to the purchase of new equipment more

quickly than firms without access.

Again, a plausible explanation for the relationship between capital supply and firms

behavior might be that lagged investment booms in capital that drive today’s used capital

supply have a long-tailed effect on local industry investment. However, when we benchmark

young firms’ sensitivity to that of older firms, we find effects concentrated among start-ups

making their first purchases. We also find effects only for heavy equipment constrained to

trade locally. Although by no means definitive, it is not obvious why business activity that

drives new capital investment should impact firm growth 5-years later differently for young

firms and among firms using only physically heavy equipment, except because of young firms’

1https://www.washingtonpost.com/business/on-small-business/when-we-were-small-ben-and-jerrys/

2014/05/14/069b6cae-dac4-11e3-8009-71de85b9c527_story.html?utm_term=.f2b01c7b9a77.

3

dependence on vintage capital. We also find that used capital supply also appears to attract

entrepreneurial entry. After controlling for time trends at the industry and county level,

start-up entry is particularly strong in county-industries for which the old capital supply is

greater. Again, this only applies to industries that rely on heavy physical capital.

If the local supply of capital seasoned by older local firms benefits start-ups, do older firms

also enjoy gains from trading with their younger counterparts? In our final tests, we use the

fraction of young businesses, measured by their share of industry-county-quarter employment,

as a measure of local demand for used capital. We then compare firms’ propensity to replace

specific equipment based on the presence of local young firm users of the same equipment

type. Because equipment within an industry will experience differential local young firm

demand (for example, Ben and Jerry’s ice cream freezer and cash register will experience

differential demand based on the local industry mix), our tests can absorb fixed-effects at the

level of county-industry-time. By tracking the reallocation of equipment via serial number,

we find that firms sell and replace equipment faster when the mix of local users of the capital

skew towards young firms. Joint with our findings on young firm investment, these results

indicate that young and old firms in a given geography enjoy a symbiotic relationship through

the supply and demand of used equipment.

Our results connect us to a several distinct bodies of work. Most directly, we propose

a key input into entrepreneurship and the investment demand of young firms. Start-ups

often have limited resources (Petersen and Rajan, 1994). As a result, young firms show

distinct features when making decisions on financing (Puri and Zarutskie, 2012), labor (Brown

and Medoff, 2003; Ouimet and Zarutskie, 2014), and business focus (Ewens, Nanda, and

Rhodes-Kropf, 2018). Our paper brings capital investment, another key component in firms’

production function, to the discussion. The documented investment patterns also links

our paper to studies on capital investment and replacement. Our evidence suggests that

differential demand of capital facilitates capital reallocation through decreasing the matching

frictions (Gavazza, 2011a,b).

The fact that local capital reallocation links firms in the same industry may also be

relevant to our understanding of agglomerative forces in the urban economy. Agglomeration is

both a feature of any modern economy and a robust growth driver (Davis, Fisher, and Whited,

4

2014). The Marshallian view (Marshall, 1890) of agglomeration highlights rationales like

labor market pooling, customer-supplier interactions, and knowledge spillovers (Duranton and

Puga, 2004; Kerr and Kominers, 2015). The linkage through capital reallocation suggests a

new motivation for agglomeration, and one that has specific predictions about the importance

for co-location among young and old firms.

2. Background and Data

Our primary data source covers sales and leases of new and used physical capital. It is

collected and sold by Equipment Data Associates (EDA) and is recovered from financing

statements filed by secured lenders. The financing statements are designated as the means

of documenting liens under the uniform commercial code (UCC) and 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

provides cleaned and formatted versions going back to 1990. Similar data are used in Murfin

and Pratt (2019) as well as Edgerton (2012). An introduction to financing statements and

the institutional background of the data can be found in Edgerton (2012) and Gopal (2019).

[Insert Figure 1 About Here.]

Figure 1 shows an example UCC financing statement from Edgerton (2012). The

statements typically contain identifying equipment characteristics, including make, model,

year of manufacture, and serial number, allowing EDA to calculate the age of equipment

and also track equipment across sequential transactions. In addition, the statements provide

information on the location of the purchaser/lessor, which EDA supplements with data on

the firm industry, age, and when available, estimated number of employees and annual sales.

The complete dataset includes more than seven million observations covering more than

150,000 models of equipment, including construction equipment, copiers, lift trucks, logging

equipment, and machine tools. These equipment are coded into 124 broad categories based

5

on the functionality (i.e., cranes), which are then coded into 376 detailed equipment types

based on their more specific characteristics (i.e., crawler crane, truck crane, etc.). For each of

these equipment types, we hand collect the specification information for the top five most

popular models, through which we obtain a measure of average equipment type weight from

dealer and manufacturer marketing materials.

Given our goal to link equipment reallocation with geographic agglomeration, we also

collect regional level economic variables. These include a measure of local banking liquidity

using shale oil shocks, following Gilje, Loutskina, and Strahan (2016). The variable captures

variation to local credit condition by tracking, for each county-quarter, the exposure to shale

discovery-driven windfalls through the banking network.2 For local employment data, we

turn to the U.S. Census Quarterly Workforce Indicators (QWI) to compute total employment

by firm age and by county, similar to Adelino, Ma, and Robinson (2017) and Barrot and

Nanda (2019). The QWI is derived from the Longitudinal Employer-Household Dynamics

(LEHD) program at the Census Bureau and provides total employment in the private sector

tabulated by industries and age groups, allowing us to observe local start-up activities.

[Insert Table 1 About Here.]

Table 1 reports summary statistics for equipment transactions included in the UCC data.

We focus on the sample for which we observe firm-level characteristics, most importantly the

age of the buyer. The mean (median) age of a machine in the UCC transaction data is four

years (one year). The mean (median) firm age in our sample is by 22 (16) years, with 25

percent of all transactions involving firms less than seven years old. The average machine in

our sample had an estimated value at the time of acquisition of roughly $70,000, while the

25th and 75th percentiles of equipment value are $15,577 and $91,324.

3. The Relation Between Firm Age and Capital Age

The first contribution of the paper is to document the relation between capital age and

firm age, and in particular, its robustness and consistency across asset markets.

2See Appendix A.1 for more details.

6

3.1. Univariate Analysis

Our analysis begins with a univariate illustration of how firms use different aged machines

over their life cycle, and conversely, how machines are reallocated across the firm age

distribution over time. In Figure 2, we plot the average age of machines purchased by

firms across different age groups, as well as the 95 percent confidence interval. Newly born

firms purchase machines that are on average 5.5 years old. Older firms purchase younger

machines—1-year old firms’ purchase 5-year old capital on average, and this number drops

to 4.1 years old for firms that are 10 years old. The pattern captures more than just the

distinction between purchasing new vs. used capital. Similar patterns obtain within the

subsample of used machine transactions, suggesting a continuous reallocation of capital to

different vintages of firms as capital ages.

[Insert Figure 2 and 3 About Here.]

The economic magnitude of the effect size appears sizable. Consider that in the sample,

the average effect of age on estimated machine value (per EDA’s valuation estimates) is

–$4,310 per year. Meanwhile, the median acquisition of 0-1 year old firm in the sample is

$43,110. Comparing the roughly 3-year change in average machine age going from a start-up

to a fully mature firm, implies a $12,930 reduction in acquired machine value, or 25 percent

of total price.

Of course, the relationship depicted in Figure 2 may be confounded by biases arising

from the changing composition of firms. Mechanically, the observations within the later age

groups condition on the survival and investment decisions of continuing firms. Moreover,

large differences in the distribution of age across industry, geography, and potentially a host

of unobservable firm characteristics leave the plot open to interpretation.3 To visually isolate

the pattern net of any selection effects driven by unobserved firm heterogeneity correlated

with age, we focus on a balanced sample of the same set of firms as they acquire capital over

time. To compare the same set of firms at each point in the plot, we limit the sample to firms

that have at least 10 transaction records and track the age of machines each firm purchases

3For example, firms that purchase copiers are on average 24 years older than the mean firm buyingconstruction equipment. Yet the average construction equipment transaction involves a machine that is 7.5times older than the average copier transaction.

7

with each successive capital acquisition. In addition to holding fixed firm characteristics

constant, this also allows us to sidestep measurement error in firm age and focus on exogenous

variation due to the passage of time.

Figure 3 presents this within-firm result. On average a firm’s first capital purchase involves

a machine that is 5.25 years old, closely mapping to the average age of equipment purchased

by newly born firms in Figure 2 (5.5 years). By the time firms are purchasing their sixth

piece of equipment, the average equipment age has fallen to 3.6 years old. The decrease

in machine age over the first few transactions is particularly notable, suggesting important

effects near the time of firm establishment.

[Insert Figure 4 About Here.]

Finally, the documented relationship is not driven by patterns particular to some small

set of industries and corresponding capital. Rather, we find that it is ubiquitous across many

different industries and different types of equipment. In Figure 4, we plot the histogram of the

industry-by-industry (4-digit SIC level) and equipment type-by-equipment type coefficients

from a regression of machine age on firm age. The coefficients that are statistically significant

at the 1 percent level are reported in white, those that are statistically significant at the 10

percent level are reported in light gray, and those that are insignificant at the 10 percent level

are reported in dark gray. Across 174 industries, we find that this relationship of young firms,

old capital holds at the 1 percent level for 131 industries and at the 10 percent level for 146

industries. The relation is positive and significant at 10 percent level for only 9 industries.

Of the 116 equipment types in our data, the young firms, old capital relationship holds at

the 1 percent level for 92 types and at the 10 percent level for 98 types.

3.2. Regression Results: Firm Age and Capital Age

While the figures suggest a robust pattern that goes beyond industry or equipment type

effects, Table 2 allows us to flexibly explore the relationship and its sensitivity to conditioning

out confounding machine or firm characteristics. We perform the analysis on a complete set

of equipment transactions i using the following model:

log(1 +Machine Age)i = β × log(1 + Firm Age)i + δFE + εi. (1)

8

In these regressions, each transaction is indexed by the purchasing firm, its industry (3-digit

SIC code), transaction year, machine type (i.e., compact utility tractor), and the machine

itself. Both machine age and firm age are measure as the logarithm of one plus age.4 Since

we are limited by the observable characteristics of machines and the buyer, we sequentially

incorporate a system of fixed effects, which we discuss below.

[Insert Table 2 About Here.]

In column (1), we control for county-by-year and industry fixed effects. In this way, we

remove the effect of county-level time trends and industry-level variations that could drive

the correlation between firm age and machine age. We find that the coefficient is negative

and significant. The economic magnitude, –0.100, suggests that an approximate doubling of

firm age results in a 10 percent decrease in the age of equipment acquired. Moving from a

start-up to the mean age firm (25 years), the average age of machines decrease by 33 percent,

or roughly, 1.8 years relative to the average start-up machine age of 5.5 years old. This

corresponds to an approximate reduction in machine value (using EDA estimates) of $7,758,

or 18 percent of average acquisition value for start-ups in the sample.

Column (2) adds machine type fixed effects. Machine types are correlated with but

different from industry categorizations and broadly describe the machine’s function, but not

necessarily its size or power. For example, all black and white copiers comprise one machine

type and all color copiers comprise another. If young firms and old firms are matched to

different types of assets that have different depreciation dynamics, our results could simply be

picking up this firm-asset matching outcome. The result in column (2), with an economically

large and statistically significant coefficient, suggests that endogenous matching doesn’t

explain the observed relationship between firm and machine age.

Column (3) introduces firm-level fixed effects, mirroring the within-firm plot in Figure 4.

Column (4) pushes the analysis to its natural limit by incorporating machine-level fixed

effects, focusing the variation within exactly the same underlying asset, where the machine is

identified using make-model-serial number. This evidence most clearly depicts the pattern of

reallocation of primary interest: machines are originally purchased and seasoned by mature

4In Appendix Table A2, we show that the analysis shown below is qualitatively and quantitatively similarif we transform the age variables using inverse hyperbolic sine operator.

9

firms and are then serially reallocated to younger and younger entrants over time. Among

the machines that we can follow, the average difference in age between the seller and buyer is

2.4 years—the owner age shrinks by 13 percent with each reallocation.5

In Appendix Table A3, we extend the regressions by including firm size measures in the

regressions. Firm size is measured using both the amount of sales and the total employment,

sourced by EDA from Dun and Bradstreet. To separate the effects of age and size is

particularly important given the recent research on the independent influences of age and size

in firm behavior (Haltiwanger, Jarmin, and Miranda, 2013; Adelino, Ma, and Robinson, 2017).

Not surprisingly, and consistent with Eisfeldt and Rampini (2007), larger firms purchase

younger machines, and the economic magnitudes are large. However, even after controlling for

firm size, firm age still plays an important role in capital investment decisions. Regrettably,

coverage of firm size is only captured once in the database (i.e. we do not observe changes in

firm size over time) and not for all firms, limiting our ability analyze this variation.

Plausible candidate explanations for the “young firms, old capital” pattern include

financing constraints faced by young firms. Technological preferences of young vs old firms

may also play a role. We find evidence of the former, but not the latter.

Regarding financing constraints, Eisfeldt and Rampini (2007) and Rampini (2019) argue

that used capital should be more attractive to financially constrained firms since they are

more willing to exchange higher maintenance costs and less durability for a lower down

payment requirement today. We test this idea by interacting a cross-sectional instrument for

bank liquidity with firm age—if the relaxation of financing constraints for young firms flattens

the relationship with machine age, we might conclude that financing constraints play an

important role. Following Gilje, Loutskina, and Strahan (2016), we capture county-by-quarter

variation in bank liquidity driven by deposit windfalls in distant branches of local banks

related to shale oil discoveries. Bank liquidity shocks are aggregated at the county-year level

to come up with geographic variation in access to credit. We characterize better credit as

a dummy for above median values of the size of the measured shale-driven liquidity shock.

5These numbers vary by industry. Woodworking tools are reallocated to a buyer that is six years younger onaverage than the previous owner. The difference is 5.1 years for logging tools and three years for constructionequipment. Printers, copiers, and machine tools show the smallest differences in user age as the machinetrades at just 2.5, 1.4, and 0.5 years, respectively, although the sign is the same in every major equipmentcategory.

10

We then augment the model in Eq. (1) with an interaction term of firm age and the shale

shock indicator. The results are reported in Table 3 columns (1) and (2), showing that young

firms’ preference for old capital decreases by about 20 percent with better credit access. This

suggests a prominent role for financial constraints in the allocation of older capital.

[Insert Table 3 About Here.]

Explanations rooted in young firms’ technological preferences are harder to detect. In

columns (3) and (4), we find limited evidence that young firms use older capital when

they lease it (Eisfeldt and Rampini, 2009). Finally, when we substitute machine age for a

measure of technological age based on the year a model was introduced, we find no obvious

correspondence with firm age. Column (5) restricts the sample to the subset of transactions

in which the buyer firm purchases a new machine. This allows us to fix variation in the

new vs. used decision and instead focus on variation in technological age, which we measure

as “model age”, defined as the difference between the year of transaction and the year of

model introduction. If young firms purchase old capital because of non-financial reasons

such as preferring older but proven technologies, one would expect that younger firms will

choose longer-established models even when buying new machines. We find that the economic

relation between firm age and model age in this regression is both economically negligible

and statistically weak.

While these facts help shed light on the economic motivations for start-ups’ use of vintage

capital, for the remainder of the paper, we largely take the observed correlation as given and

turn to the bigger issue at hand: given the apparent demand for old capital by young firms,

how does local availability of seasoned capital shape entrepreneurial investment and growth?

3.3. Nonincidental Sample Selection

So far, we’ve avoided any discussion of sample selection—i.e., the data generating process

that allows us to observe equipment acquisitions together with buyer characteristics. Because

our data are generated as a result of secured debt financing with liens perfected under the

UCC, exclusion from the data is not random, but instead will correlate with firms’ decisions

to acquire capital via cash, unsecured debt, or secured debt. For example, older firms may

11

be more likely to purchase with cash, and/or new machines may be collateralized more often,

taking advantage of commonplace manufacturer financing (Murfin and Pratt, 2019). Given

the right (or wrong) selection equation, the relation between firm age and machine age could

easily confound the true relation between firm an machine age.

To measure the severity of the problem, we first investigate the potential selection on

firm age. While we can’t test this at the micro-level, it is possible to check whether indeed

young firms are more likely to use secured debt when acquiring the types of physical capital

in our data. To do so, we compare the firm age distribution in the UCC data with that of

the selection-free Census data. Specifically, we compare the proportion of UCC machine

purchases made by firms in different firm age groups with the proportion of employment by

firms in different firm age groups. In Table A4, we find almost identical firm age distributions.

While by no means definitive, the proximity of the two distributions is inconsistent with

massive selection effects biasing our data to a specifc end of the firm age distribution.

Equally important is how machine age enters the selection equation. In particular, if

both old firms and old equipment were systematically excluded from the data, the unlikely

observation of an old piece of equipment would need be, on average, more likely to be matched

to a young firm (to offset the effects of machine age in the selection equation). While it seems

likely that new machines relative to used machines will tend to benefit from commonplace

manufacturer financing relative, supporting their more frequent sample inclusion (Murfin

and Pratt, 2019), variation in financing status across machine age among used machines is

not obvious. Meanwhile, 65% of equipment transactions are for machines at least one year

old, so these machines will be of most interest to us in terms of understanding the typical

relationship between machine and firm age.

To estimate the sign of machine age in the selection equation, we need a sample of machines

for which we can observe the complete information on whether each was ever selected into

UCC. We exploit a subset of UCC filings that are flagged by the data provider as wholesale

acquisitions, primarily floor-plan financing for dealer inventory. These transactions are useful

if we assume that when dealers borrow against machines in inventory, those machines will

subsequently be sold to retail buyers. For these machines, some are followed by a subsequent

UCC sales (i.e., selected), and some are not (i.e., selected out). We can then estimate the

12

selection equation based on equipment age to understand if machine age predicts selection

into the data.

[Insert Table 4 About Here.]

In Table 4, we model the selection equation, examining whether a wholesale transaction

was followed by a subsequent sale over the next year—that is, selected into sample. We

include a dummy for new vs. used machines as well as a non-binary age variable, allowing

the new vs. used margin to impact selection differently from variation in age within used

machines. The table suggests that brand new machines are more likely to be selected into

our sample. However, among used machines, the relationship between age and selection is

economically tiny and not distinct from zero.

Given that machine age appears to influence financing (and hence selection) only at the

new vs. used margin, how should we interpret our main results? To see what the results

would look like absent these selection effects, we can re-estimate the regressions from Table 2

among used machines—the subset for which age appears unimportant for selection. Because

the regression of interest requires that we truncate machine age, Table 4 Panel B estimates a

truncated regression model to deal with the bias associated with truncated left-hand-side

variables. Consistent with selection biases being small, a comparison of the main effects in

columns (1)–(2) and (3)–(4) yields little difference in effect size.

4. Old Capital, Local Economic Activities, and Agglomeration

In the prior sections, we documented young firms’ apparent demand for used capital

and provided evidence of potential explanations for that fact. Going forward, we take these

facts as given and proceed by asking what are the consequences of this relationship on

entrepreneurship, capital investment, and growth? In particular, if young firms require used

capital, they may benefit from being located near older firms, which serve as producers of

used capital. This may jointly shape how firms invest conditional on entry within an industry,

or even the industries that entrepreneurs choose to enter.

Several questions emerge immediately. For example, do we observe large sample variation

consistent with the Ben and Jerry’s anecdote shared at the paper’s start, in which an

13

entrepreneurial start-up’s investment choice was shaped by vintage capital supply? We

also explore longer term outcomes for firms shaped by used capital supply. How quickly

do start-ups surrounded by used capital grow and expand? Does used capital motivate

entrepreneurial entry in the first place?

Older firms that prefer newer capital may also benefit from having natural buyers of their

used capital in close proximity. If true, this would make a strong case for the importance of

used capital markets, not just to support entrepreneurship, but also to shape the investment

dynamics of incumbent firms. Perhaps more importantly, it would suggest gains to co-

location among young and old firms and provide a novel channel for the returns to geographic

agglomeration.

To evaluate the conjectures above, we propose and test two hypotheses relating young

and old firm trade in vintage capital. First, we hypothesize that local old capital supply will

influence the nature and volume of young firm investment and start-up entry (i.e., old firms’

used capital benefits young firms). Second, we test if the presence of old capital-dependent

young firms allows incumbent capital owners to upgrade capital more frequently (i.e., young

firms’ reliance on used capital benefits old firms).

In the next section, we describe our measurement of the supply of old capital and propose

how we intend to identify the hypothesized links between the trade in vintage capital and

firm outcomes as distinct from other hypotheses.

4.1. Measuring Old Capital Availability

To begin our examination of the role of vintage capital in young firm investment, we

first need a measure of the volume of old capital available to young firms. We approach

this problem by making use of predetermined variation in the local purchase history of new

physical capital in a county reported in EDA. As each machine “enters” a county as a new

purchase, we count it as part of the local supply for that equipment type in the same county

for the remainder of its useful life. From this, we are able to aggregate a measure of the

number of k -year old equipment locally-available based on the new purchases made locally k

years ago.

As an example, a brush cutter purchased by a logging company in Durham, NC in 2010

14

will appear in the local supply of one year old machines in 2011, the local supply of two year

old machines in 2012, and so on. In our analysis, we focus on the old machine availability

measure for equipment aged five to ten years. The lower bound captures our interest in used

vintage capital. The upper bound is limited by the time span of the EDA data. We cannot

capture the supply of machines that are ten years old until the eleventh year of our sample,

since we need to allow ten years to pass from the observation of a new machine purchase.

Also note, we focus on new machine purchases to avoid treating same-machine turnover

as an increase in supply. By using variation in new machine purchases as variation in the

latent supply of old machines, we also ensure a gap in timing between supply shocks and

current economic activity that becomes blurry with used capital transactions (for example,

the purchases of a five-year old machine at time t may reflect variation in used capital supply,

but it is also mechanically linked to investment, a key outcome measure, during the same

period).

We apply the procedure described above to every machine type in every county and

aggregate annually. This provides us with a measure of the total number of a given machine

type at any age category available to local businesses in a given county, at a given time.

Our primary measures of old capital supply focuses on available machines in the age range

between five and ten years old. Moreover, as will be discussed below, this measure can be

flexibly aggregated to reflect the total supply available to an industry by aggregating the

equipment type level variable up to the industry level for all the machines purchased by a

given industry, weighting based on the relative frequency of purchase.

4.2. Geographic Constraints to Used Capital Trade

Calculating the local equipment supply to test its effects on local businesses presupposes

that trade in vintage capital is predominantly local. This is both critical to our identification,

which relies on geographic variation in used capital, but also suggests an empirical question:

how local is trade in used capital?

Given the ability to track machines by way of their serial number over subsequent trades,

our data provide a unique setting to answer this question. Under the assumption that the

closest two observed acquisitions of a machine in time represent a trade from the former

15

owner to the new owner, we can calculate the average distance that a machine travels with

each subsequent realloction. A few novel facts emerge.

[Insert Figure 5 About Here.]

First, the reallocation of the used capital in our sample (a broad sample) is a very local

activity—nearly half of the capital reallocation we observe is within 50 miles, and 80 percent

occurs within 200 miles. Figure 5 provides a histogram documenting the full distribution

of trade distance. Moreover, we find that trade in physical capital becomes increasingly

local as machines age. In Table 5, we regress the logarithm of moving distance on our

measure of machine age, as well as the number of time a machine has been traded. The

regressions include fixed effects for industry, machine type, and the locations of buyers and

sellers (separately), allowing us to control for variation in trade activity that may relate to

a locations remoteness. In each specification, we find that as machines age, and with each

subsequent trade, the trading distance shortens. For example, in column (4), we show that it

translates into a 1.7 percentage points higher probability of being traded within the same

county each time the machine is turned over to a new buyer, which is an 8.5 percent increase

from the base rate.

[Insert Table 5 About Here.]

These new facts both support our identification of the effects of used capital supply using

geographic variation. But, more importantly, they also suggest that the physical location

of used capital, the firms need it and those that supply it, may matter. While it is by no

means surprising that geography matters for trade, the degree to which it matters for our

expensive and potentially moveable machines suggests the potential for new and important

geographic components to how industry concentrations emerge, and which types of businesses

entrepreneurs start.

[Insert Figure 6 About Here.]

Given the importance of geography to machine trade, our tests can then exploit variation

in the locality of capital trade as a means of exploring competing interpretations. If local

16

used capital supply causally affects young firm investment, it should do so more strongly

for machines constrained to trade locally because of physical characteristics like weight. In

contrast, if firm investment and local used supply are co-determined by confounding economic

activity, contrasting predictions for high and low weight-to-value capital supply are less

obvious. We conjecture that measures of logged weight-to-value ratios, taken from machine

specification sheets and EDA estimates of value, should provide a strong proxy for constraints

to the locality of trade. This concept is widely used in supply chain and logistic studies

and recently adopted in economics research (Hummels, 2007; Barrot, Loualiche, Plosser, and

Sauvagnat, 2018; Koch, Panayides, and Thomas, 2020). Of course, a necessary condition for

these tests to make sense is that our measure of weight-to-value provides a strong first-stage

in predicting the distance a machine travels after a typical sale. Figure 6 confirms this is

the case, plotting the observed relationship between five weight to value quintiles against

the average logged trade distance. Moving from the first to the fifth quintile reduces trade

distance by roughly 40 percent.

4.3. Old Capital Supply and Young Firm Investment

With variation in the local supply of old machines across time, place, and machine type,

we can now test how the supply of old machines shapes firm investment choices, conditional

on a firm investment occurring. To this end, in Table 6, we estimate a choice model of

machine purchase based on variation in used capital supply. Conditional on a purchase being

made in a given year-county, we assume a potential choice set for that transaction consisting

of all of the equipment types within the same broader equipment category. We then test

whether the supply of old equipment influences the specific choice of equipment from among

alternatives in the same equipment category.

As an example, a Durham lumberjack considering buying a brush cutter in 2015 might also

consider the available supply of de-limbers, fellers, and tree shears. By focusing on tools that

may be reasonable substitutes for the actual tool ex-post purchased, we are able to explore the

intensive margin of investment. Equipment type and sub-category classifications are provided

by EDA. On average, each equipment category (containing more than one equipment type)

contains 6.7 equipment types, suggesting an unconditional purchase probability within the

17

choice sample of 15 percent.

With the unit of observation being a potential machine purchase and the outcome of

interest an indicator for whether or not a given machine type was chosen, we then estimate

the differential effect of used capital supply on old and young firms via the model

I(Bought) =β × LocalSupply × Y oung+

β′ × LocalSupply + β′′ × Y oung + δFE + ε,(2)

I(Bought) is a dummy variable that takes the value one for the specific equipment type

actually purchased. LocalSupply is at the equipment type-county-year level, capturing

variation in the latent supply of each type of equipment. Y oung is a dummy variable

indicating whether the buyer is younger than three years old.

Of course it won’t be surprising that supply affects the eventual equipment choice. Instead,

our identification depends on the differential effects of the latent supply of local capital across

young vs. old firms, captured by the β coefficient in model Eq. (2). The corresponding

thought experiment compares two lumberjacks—one young, one old—in Durham county in

the market for forestry equipment in 2015, presented with the same opportunity to choose

from various equipment types. Our hypothesis that young firms’ decisions are shaped by old

capital supply benchmarks their sensitivity to locally available equipment against that of

established firms. Later, we add another layer of differencing that compares the sensitivity of

young firms for more/less mobile equipment.

In order for our model to approach this thought experiment, our specifications include

detailed fixed effects at the industry-equipment category-county-time level (i.e., lumberjacks-

forestry equipments-Durham-2015). This fine fixed effect structure exactly captures the

thought experiment described above, by ensuring we are not capturing slow moving industry

booms or local economic trends correlated with supply. Standard errors are clustered at the

buyer-equipment category-year level, taking into account that each buyer’s decisions over a

set of equipment types within the same category are correlated.

[Insert Table 6 About Here.]

Table 6 presents the results. We find the youngest firms are significantly more sensitive

18

to vintage capital supply as a determinant of their investment decision and that sensitivity

monotonically declines with firm age. To interpret the economic magnitude, it is perhaps

easiest to break the regression from column (1) down across firm age groups. In Figure 3, we

plot the average response of equipment purchase choices to a one standard deviation change

in local used capital supply across firm age categories ranging from start-ups to firms older

than fifty years. The reported coefficients across age groups are estimated from the following

model:

I(Bought) = β × LocalSupply + δFE + ε, (3)

using the same set of fixed effects from column (1) of Table 6. If we take the oldest firms as

the control, allowing us to capture the baseline elasticity of demand with respect to used

capital supply shocks, the response by start-ups is approximately 18 percent higher than

the response by 50-year-old firms for the same supply shock. In short, at the margin, the

direction in which young firms grow appears to be sensitive to what used capital is available

locally.

[Insert Figure 7 About Here.]

One hurdle to this interpretation is the alternative hypothesis, also consistent with the

regression results, that our supply variable captures a five year lag in local industry dynamism

to which young firms are simply more sensitive. To parse these competing interpretations,

we lean on the fact that the influence of local capital supply on entrepreneurial investment

hinges on the assumption that physical capital is difficult or costly to relocate. In contrast, it

is less clear why a lagged response by young firms to a five-year-old boom should only show

up in their sensitivity to higher weight-to-value equipment supply.

To exploit these competing predictions, in column (3) we interact the effect of local old

capital supply and firm age with Heavy—a dummy variable capturing machines that fall

within the upper quintile of weight-to-value. The results confirm that it is the supply of heavy

capital, the trade of which is more locally-constrained, that most significantly influences

young firm investment.

19

4.4. Old Capital Supply and Young Firm Growth and Entry

Table 6 suggests that, conditional on investing, firms’ choices of equipment depend on

local capital supply. In this section, we ask whether the local availability of old capital has

longer run effects, allowing start-up companies to grow and expand via additional capital

investments in the early stages of their life cycle. To do this, we examine a cross-section of

firms that made at least one equipment investment during their first three years. We then

estimate how the investment dynamics following the initial investment are influenced by local

capital supply. Specifically, we estimate the following regression:

Investmenti = β × LocalSupplyi + δFE + εi. (4)

We construct LocalSupply at the county-industry-year level to capture the total amount of

old equipment available to the start-ups in the county-industry at the time of the initial

investment. To do this, we map the equipment-type supply variable described in Section 4.1

to the 3-digit SIC industry level based on the distribution of industries that acquire each

machine type over the entire sample. For example, if half of all excavators appear in the

data as mining-industry purchases and half as construction purchases, then we allocate a

local supply of 20 excavators as ten excavators each to mining and construction industries.

In this way, the supply of used capital available to each industry reflects the distribution of

machines purchased by firms in that industry throughout the sample. Because each industry’s

supply comprises various equipment types, we measure supply based on log equipment values

(as new) rather than equipment counts. Finally, we standardize LocalSupply to ease the

interpretation of our estimates. We include combinations of county-year and industry-year

fixed effects to absorb local and industry growth trends and cluster our standard errors at

the industry and county levels.

[Insert Table 7 About Here.]

Table 7 Panel A presents the results using a variety of start-up investment outcome

variables. Column (1) shows that young firms with better access to old capital invest more

in additional equipment between one and three years after their initial investment. A one-

20

standard-deviation increase in the amount of used capital available leads to a 9.4 percent

increase in subsequent capital investment.

While the result in column (1) is consistent with old capital supply facilitating survival

and growth of start-ups, it is also possible that the subsequent capital investments arise as

young firms need to replace their initial used-capital investment. To address this possibility,

in column (2) we examine the total machine types that a young firm invests in between one

and three years after its initial investment. We find that the supply of used equipment at the

time of initial investment results in a broader array of equipment purchases in the future. A

one-standard-deviation increase in the local supply of used capital at the time of a young

firm’s initial investment leads to a 7.3 percent increase in the number of equipment types a

firm invests in over the ensuing years.

In column (3) we examine young firm investment in new (unused) equipment between

one and three years after their initial equipment investment. We find that young firms

subsequently invest more in new equipment when their initial investment was facilitated by a

large supply of used equipment, with a one-standard-deviation increase in used capital supply

resulting in a 5.0 percent increase in future investment in new equipment. That is, young

firms do not simply subsist off of a supply of local used capital. Instead, early investment

opportunities enhanced by available used capital help young firms graduate into investment

in new capital.

As discussed above, the influence of local capital supply on entrepreneurial investment

hinges on the assumption that physical capital is difficult or costly to relocate. Machines

with a high weight-to-value ratio are more constrained to trade locally (as discussed in

Section 4.2), so we would expect the local supply of heavy equipment to impact new firm

investment more than the local supply of light equipment. To test this prediction, we replace

the local supply measure from Equation 4 with the HeavySupply and LightSupply, where

HeavySupply captures the log value of old equipment in the top quintile of weight-to-value

and LightSupply captures equipment in the bottom quintile.

The results are shown in Panel B. Across all measures of start-up growth and expansion,

only the supply of heavy equipment impacts subsequent investment. In each case, the impact

of light equipment is close to zero, while the difference between the effect of heavy and light

21

equipment supply has a p-value less than 1 percent.

Finally, we can benchmark the impact of old capital supply on the future investment and

growth of young firms against the impact on old firms. Given young firms’ preference for old

equipment, we would expect that young firm success is more sensitive to old capital supply.

In measuring young firm success, we capture follow-on investments after an initial investment.

When benchmarking against old firms, however, there is no natural first investment event.

Instead, for old firms, we choose a random investment and measure the amount of additional

investment one to three years after that randomly chosen investment.

In Panel C, we include old firms (ten years and older) and modify the regression from

Panel A, interacting an indicator for young firms with the LocalSupply. This allows us to

include even finer fixed effects at the county-industry-year level to account for any local

industry trends. These fixed effects absorb the main effect on the LocalSupply variable but

still allow us to capture the interaction of interest. Effectively, our estimates compare two

firms in the same county-industry who make an investment at the same time, one of which

is an old firm while the other is young. In columns (1) and (2), we find that young firms’

subsequent investments, in terms of both quantities and breadth of investment, is significantly

more sensitive to local capital supply than that old firms. Column (3) reveals no difference

between young and old firms in the impact of old capital supply on subsequent investment in

brand new machines. In conjunction with Panel A, the results indicate that local capital

supply facilitates future investment in new equipment for young and old firms alike. One

potential reason for this is that a healthy used capital market (made possible by healthy

young firms) allows older firms to refresh their capital stock more frequently, an implication

which we investigate further in the next section.

Table 7 indicates that local supply of old capital facilitates survival and growth of young

firms, conditional on a first investment for young firms. We now ask whether local capital

supply is important in creating start-ups in the first place. Returning to the specification

from Panel A of Table 7, we replace the dependent variable with Start-up Entry, as captured

by the EDA data. Specifically, for all firms that ever make a purchase in our dataset, we

capture the year of firm birth as well as the county-industry. Our estimates capture whether

more firms are born in a county-industry-year when the availability of old capital is high.

22

We report the results in Table 8. The sample includes only those county-industry-years

with at least one new start-up. The coefficient in column (1) suggests that a one-standard-

deviation increase in local used capital results in 0.26 more start-ups (relative to a mean of

1.3 and a standard deviation of 0.97). In column (2), we use the Heavy capital availability

and Light capital availability specification from Panel B of Table 7 to test whether start-up

activity is differentially dependent on the supply of immobile capital. We find this is the case,

with a p-value of 0.042 on the difference between heavy and light supply.

[Insert Table 8 About Here.]

Table 8 suggests that, in the presence of abundant used capital, more firms are born. By

lowering barriers to entry, we would expect that used capital also lowers the quality threshold

for the marginal new entrant. In light of this, the results in Table 7 are more surprising. Not

only are more start-ups born when there is a lot of used capital, but they are more likely to

survive and grow.

4.5. Young Firms and Incumbents’ Investment Decisions

Above we have shown that young firms benefit from the availability of used capital

originally bought and seasoned by older firms in the same location. Does young firms’ reliance

on old capital benefit the firms who already own the capital? In particular, we hypothesize

that existence of an active entrepreneurial sector will increase the base of potential buyers and

thus the overall thickness of the used equipment market. One implication of this increased

market thickness would be more frequent asset turnover (Gavazza, 2011b), implying more

frequent opportunities for all firms, including older firms, to upgrade physical capital more

quickly.

To study this, we examine the frequency of machine replacement, captured by the joint

observation of an acquired machine being sold (showing up as acquired by another firm)

and the same machine type being purchased. We then examine the probability of capital

replacement for a given machine conditional on a measure of the relative employment size

of young firms in the local economy that are natural users of the same equipment type. To

be clear, the unit of observation in this analysis is a purchase of physical equipment. Each

23

of these purchases can be characterized by the machine type, the location at the county

level, the industry of the firm, and year. The primary key dependent variables of interest is

whether the equipment is replaced over three, four, and five year time horizons.

Percentage of Young Firm Employment is the key independent variable and is designed

to capture variation in young firm demand by equipment type by measuring the relative

size (employment weighted) of local young firms divided by total firms in industries that

are active users of a given equipment type. This variable is calculated for each machine

type in each county-year. For each machine type, we measure the frequency with which it

is purchased in all possible industries (2-digit NAICS) to calculate industry weights (e.g.

tractors may be used 40 percent in construction and 60 percent in agriculture). For each

equipment type-county-year, we then average the number of young firm employees divided by

total employees in each county-industry, weighted by the propensity for industry to be a user

of the equipment type in question. If Durham, NC agriculture employment is 25 percent in

young firms and Durham. NC construction employment is 50 percent young firms, then we

would measure % of Young Firm Employment as 0.25 × 0.6 + 0.5 × 0.4 = 0.35—the weighted

average of % young firms by industry, weighted by the importance of each industry for that

equipment type. The raw young firm percentage by industry county employment is obtained

using Census LEHD QWI data as used in Adelino, Ma, and Robinson (2017) and Barrot and

Nanda (2019).

One advantage of this measure is that it varies even within the firm at a given point in

time, and certainly within an industry-county at a point in time. For example, a Durham

farmer will benefit from young firm demand for his tractors via local construction employment–

demand he might not enjoy for a combine-harvester which can only be used by agricultural

firms, which happen to be much older. This allows us to identify the effects of young firm

employment even with industry-county-time fixed effects, taking out local industry trends

in investment and turnover. As in prior tables, we will also exploit variation in equipment

weight-to-value to ensure the effects of local young firm employment matter more when

equipment is constrained to trade locally.

[Insert Table 9 About Here.]

24

Table 9 shows the results. All regressions include fixed effects at the level of year-county-

industry and machine type-county. Column (1)-(3) sequentially measure the impact on the

probability of replacing a machine in three, four, and five years. To deal with truncation, we

require three, four, and five years of remaining data for a given machine type to measure

replacement within three, four and five years, respectively. In all those specifications, we find

that the larger the presence of young firms is, the higher the probability of replacement over

a short window. Again, the effects are concentrated among heavy equipment, as measured by

the highest quintile of equipment type weight-to-value.

5. Conclusion

Economic growth is characterized and driven by its dynamism. At the extensive margin,

start-ups drive the business dynamism through their entry and entrepreneurial investment in

capital. At the intensive margin, the capital are often reallocated between firms with the goal

of satisfying heterogeneous capital needs in the production process. Given their importance,

recent work on business cycles dynamics and growth has emphasized the importance of young

firms and, separately, the process of capital reallocation. The observed relationships between

young firms and old firms in the same industry and geography highlight the deep connection

between these seemingly separate activities.

Our findings suggest that entrepreneurship activities are not isolated or independent

activities and, in fact, they are deeply connected to incumbent firms. While other work

has shown this through human capital connections (entrepreneurial spawning, acqui-hires,

etc.) and innovation relations (innovation flow, collaboration), this paper emphasizes the

reallocation of physical capital.

25

References

Adelino, Manuel, Song Ma, and David Robinson, 2017, Firm age, investment opportunities,and job creation, Journal of Finance 72, 999–1038.

Barrot, Jean-Noel, Erik Loualiche, Matthew C Plosser, and Julien Sauvagnat, 2018, Importcompetition and household debt .

Barrot, Jean-Noel, and Ramana Nanda, 2019, Can paying firms quicker affect aggregateemployment?, Journal of Finance Forthcoming.

Bond, Eric W, 1983, Trade in used equipment with heterogeneous firms, Journal of PoliticalEconomy 91, 688–705.

Brown, Charles, and James L Medoff, 2003, Firm age and wages, Journal of Labor Economics21, 677–697.

Davis, Morris A, Jonas DM Fisher, and Toni M Whited, 2014, Macroeconomic implicationsof agglomeration, Econometrica 82, 731–764.

Duranton, Gilles, and Diego Puga, 2004, Micro-foundations of urban agglomeration economies,in Handbook of Regional and Urban Economics , volume 4, 2063–2117 (Elsevier).

Edgerton, Jesse, 2012, Credit supply and business investment during the great recession:Evidence from public records of equipment financing, Federal Reserve Bank Working Paper.

Eisfeldt, Andrea L, and Adriano A Rampini, 2007, New or used? investment with creditconstraints, Journal of Monetary Economics 54, 2656–2681.

Eisfeldt, Andrea L, and Adriano A Rampini, 2009, Leasing, ability to repossess, and debtcapacity, Review of Financial Studies 22, 1621–1657.

Ewens, Michael, Ramana Nanda, and Matthew Rhodes-Kropf, 2018, Cost of experimentationand the evolution of venture capital, Journal of Financial Economics 128, 422–442.

Gavazza, Alessandro, 2011a, Leasing and secondary markets: Theory and evidence fromcommercial aircraft, Journal of Political Economy 119, 325–377.

Gavazza, Alessandro, 2011b, The role of trading frictions in real asset markets, AmericanEconomic Review 101, 1106–43.

Gilje, Erik P, Elena Loutskina, and Philip E Strahan, 2016, Exporting liquidity: Branchbanking and financial integration, Journal of Finance 71, 1159–1184.

Gopal, Manasa, 2019, How collateral affects small business lending: The role of lenderspecialization .

Haltiwanger, John, Ron S Jarmin, and Javier Miranda, 2013, Who creates jobs? small versuslarge versus young, Review of Economics and Statistics 95, 347–361.

26

Hummels, David, 2007, Transportation costs and international trade in the second era ofglobalization, Journal of Economic perspectives 21, 131–154.

Kerr, William R, and Scott Duke Kominers, 2015, Agglomerative forces and cluster shapes,Review of Economics and Statistics 97, 877–899.

Koch, Andrew, Marios A Panayides, and Shawn Thomas, 2020, Common ownership andcompetition in product markets, Journal of FInancial Economics Forthcoming.

Marshall, Alfred, 1890, Principles of Economics (London, Mcmillan).

Murfin, Justin, and Ryan Pratt, 2019, Who finances durable goods and why it matters:Captive finance and the coase conjecture, Journal of Finance 74, 755–793.

Ouimet, Paige, and Rebecca Zarutskie, 2014, Who works for startups? the relation betweenfirm age, employee age, and growth, Journal of financial Economics 112, 386–407.

Petersen, Mitchell A, and Raghuram G Rajan, 1994, The benefits of lending relationships:Evidence from small business data, Journal of Finance 49, 3–37.

Puri, Manju, and Rebecca Zarutskie, 2012, On the lifecycle dynamics of venture-capital- andnon-venture-capital-financed firms, Journal of Finance 67, 2247–2293.

Rampini, Adriano A, 2019, Financing durable assets, American Economic Review 109,664–701.

27

Figure 1. Sample UCC Filings

Notes. This is an example UCC filing from the state of North Carolina, provided in Edgerton (2012).

28

Figure 2. Firm Age and Machine Age

Notes. This graph plots the average age of machines purchased by firms across different age groups (1 to 50years old), as well as the 95% confidence interval.

Figure 3. Trading Order and Machine Age

Notes. This graph plots the average age of machines purchased by firms across their own life cycles and the95% confidence interval, by focusing on a group of firms that have at least 10 transaction record, and followthrough the age of machines they purchase as the time evolves.

29

Fig

ure

4.

Fir

mA

ge

and

Mach

ine

Age

Est

imati

on

Coeffi

cients

:B

y-I

ndust

ryan

dB

y-E

qu

ipm

ent

(a)By-Industry

Estim

ations

(b)By-E

quipmentTypeEstim

ations

Notes.

This

gra

ph

plo

tsth

ehis

togra

mof

the

indust

ry-b

y-i

ndust

ry(P

anel

(a))

and

equip

men

tty

pe-

by-e

quip

men

tty

pe

(Panel

(b))β

corr

elati

ons

bet

wee

nfi

rmag

ean

dm

ach

ine

age,

esti

mat

edu

sin

g

log(1

+MachineAge)

i=β×

log(1

+FirmAge)

i+ε i.

The

coeffi

cien

tsth

atar

est

atis

tica

lly

sign

ifica

nt

atth

e1%

leve

lar

ere

por

ted

inw

hit

e,th

atar

est

atis

tica

lly

sign

ifica

nt

only

atth

e10

%ar

ere

por

ted

inli

ght

gray

,an

dth

ose

that

are

insi

gnifi

cant

atth

e10%

are

rep

ort

edin

dark

gra

y.

30

Figure 5. Histogram of Reallocation Distance

Notes. This figure presents the histogram of reallocation distances (in miles) for equipment transactions. Thedistances are calculated based on the addresses provided in the UCC filings.

Figure 6. Weight-to-Value Ratio and Average Reallocation Distance

Notes. This figure presents binned scattered plot of the weight-to-value ratio and average travel distancefor different equipment types. Weight-to-value ratio is at the equipment type level, and is calculated as thelogarithm of the average equipment weight (in pounds) divided by the average price (in USD) when theequipment is sold new.

31

Figure 7. Effect of Local Old Capital Supply on Equipment Choice—By FirmAge

Notes. This figure plots the average response of equipment purchase choices on local old capital supply acrossdifferent firm age groups. The reported coefficients are estimated from the following model

I(Bought) = β × LocalOldCapital + δFE + ε,

using the fixed effects from column 1 of Table 6. We cut firm age groups into six groups that are reported inthe horizontal axis. 95% confidence intervals are reported for each estimated point.

32

Table 1Basic Summary Statistics

N Mean Std.Dev p25 Median p75

Firm age (years) 1,494,387 22 24 7 16 29Machine age (years) 1,494,387 4 7 0 1 5Value of equipment ($USD) 1,492,935 69,408 94,667 15,577 40,866 91,324Total number of employees 1,494,387 109 2,040 4 15 50Annual sales ($000 USD) 1,494,387 128,345 1,749,756 440 2,200 12,000

Notes. This table provides descriptive statistics on the equipment transactions in the sample. MachineAge is defined as the difference between the transaction date and the machine date of production. Value ofequipment is defined as the price of the equipment at transaction. Firm age is the difference between thetransaction date and the firm founding date. Total number of employees and annual sales are from Dun andBradstreet.

Table 2Firm Age and Machine Age in Equipment Transactions

(1) (2) (3) (4)Log(1+Machine Age)

Log(1+Firm Age) -0.104*** -0.105*** -0.083*** -0.089***[0.009] [0.010] [0.027] [0.008]

Observations 1,494,387 1,494,387 1,494,387 1,494,387R2 0.128 0.259 0.537 0.588Year x County FE Y YIndustry (SIC-3) FE Y YMachine Type FE YFirm FE YMachine FE Y

Notes. This table documents the relationship between buyer firm age and machine age in equipmenttransactions, based on the following model,

log(1 +Machine Age)i = β × log(1 + Firm Age)i + δFE + εi.

In these regressions, each transaction is indexed by the purchasing firm, its industry (3-digit SIC code),transaction year, machine type and the machine itself. Machine Age is the number of years between theoriginal manufacture date and the date of transaction, Firm Age is the number of years between the firmfounding date and the date of transactions. Both age variables are measured as the logarithm of one plusage. Standard errors clustered at the machine type level are displayed in brackets. ***, **, and * indicatesignificance at the 1%, 5%, and 10% levels, respectively.

33

Table

3F

irm

Age

and

Mach

ine

Age—

Cre

dit

Condit

ions

and

Model

Age

(1)

(2)

(3)

(4)

(5)

Log

(1+

Mac

hin

eA

ge)

Log(1

+M

od

elA

ge)

Log

(1+

Fir

mA

ge)

-0.1

03**

*-0

.111

***

-0.1

01**

*-0

.108

***

-0.0

03

[0.0

10]

[0.0

08]

[0.0

10]

[0.0

11]

[0.0

02]

Log

(1+

Fir

mA

ge)

xI(

Sh

ale

Sh

ock

)0.

019*

*0.

021*

**[0

.008

][0

.008

]L

og(1

+F

irm

Age

)x

Lea

sin

g0.

061*

**0.

074*

**[0

.011

][0

.007

]

Ob

serv

atio

ns

683,

025

683,

025

2,63

3,81

02,

633,

810

845,8

48

R2

0.22

10.

339

0.16

60.

289

0.2

64

Yea

rx

Cou

nty

FE

YY

YY

YIn

du

stry

(SIC

-3)

FE

YY

YY

YM

ach

ine

Typ

eF

EY

YY

Lea

sin

gF

EY

Y

Notes.

Th

ista

ble

stu

die

sth

ere

lati

onsh

ipb

etw

een

bu

yer

firm

age

and

mac

hin

eag

ein

equ

ipm

ent

tran

sact

ion

s,b

ased

onth

efo

llow

ing

mod

el,

log(1

+MachineAge)

i=β×

log(1

+FirmAge)

i×ShaleShock/Leasing i

+δ F

E+ε i.

Th

em

od

else

tup

issi

mil

ar

toth

at

inT

ab

le2.

Inco

lum

ns

(1)

an

d(2

),w

ein

tera

ctth

efi

rmage

vari

ab

lew

ith

ad

um

my

vari

ab

lein

dic

ati

ng

abov

e-m

edia

nav

aila

bilit

yof

cred

itth

atar

etr

ansm

itte

dto

loca

lbra

nch

esfr

omsh

ale

oil

shock

thro

ugh

the

ban

kin

tern

alnet

wor

k,

follow

ing

the

defi

nit

ion

ofG

ilje

,L

outs

kin

a,an

dStr

ahan

(201

6).

The

sam

ple

size

dro

ps

inth

isse

tof

regr

essi

ons

bec

ause

the

shal

eoi

lsh

ock

isre

leva

nt

only

afte

r20

00.

Inco

lum

ns

(3)

and

(4),

we

use

bot

hth

em

ain

sam

ple

ofm

ach

ine

pu

rch

ases

and

mac

hin

ele

ases

.W

ein

tera

ctth

efi

rmag

eva

riab

lew

ith

ava

riab

lein

dic

atin

gle

asin

gd

eals

.In

colu

mn

(5),

the

dep

end

ent

vari

able

ism

od

elag

eof

the

mac

hin

e,w

hic

his

calc

ula

ted

asth

enu

mb

erof

yea

rsb

etw

een

the

tran

sact

ion

an

dth

ein

trod

uct

ion

of

the

spec

ific

make

an

dm

od

el.

Th

ean

aly

sis

isp

erfo

rmed

on

the

tran

sact

ion

sof

new

mach

ines

.S

tan

dard

erro

rscl

ust

ered

at

the

mach

ine

typ

ele

vel

are

dis

pla

yed

inb

rack

ets.

***,

**,

an

d*

ind

icate

sign

ifica

nce

at

the

1%

,5%

,an

d10

%le

vels

,re

spec

tive

ly.

34

Table 4Nonincidental Sample Selection

Panel A: Machine Age and Sample Selection.

(1) (2) (3)Selection into Sample

New Machine 0.016*** 0.006 0.016***[0.006] [0.005] [0.005]

Log(1+Machine Age) 0.002 -0.003 0.009[0.003] [0.004] [0.006]

Observations 67,764 67,764 67,764R2 0.005 0.061 0.242Year FE Y YMachine Type FE YMachine Make-Model FE Y

Panel B: Selection-free Analysis on Firm Age and Machine Age.

(1) (2) (3) (4)Log(Machine Age)

Full Sample (Table 2) Machine Age ≥ 1 Only

Log(1+Firm Age) -0.104*** -0.105*** -0.087*** -0.085***[0.009] [0.010] [0.005] [0.004]

Observations 1,494,387 1,494,387 958,572 958,572R2 0.128 0.259 - -Estimation Methods OLS Truncated RegressionYear X County FE Y Y Y YIndustry (SIC-3) FE Y Y Y YMachine Type FE Y Y

Notes. Panel A documents the relationship between sample inclusion/exclusion and machine age. A missingsale is defined as a machine which did not reappear as a retail sale or lease within one year of being reportedas part of an equipment dealer’s wholesale floorplan financing. Columns (2) and (3) include fixed effects atthe machine type and machine make and model level. Standard errors clustered at the machine type level aredisplayed in brackets. ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively.

Panel B compares the full sample relationship between firm age and machine age (columns 1 and 2)with the same for the subset of used machines which are selection-free (columns 3 and 4). The usedmachine-analysis is estimated using truncated regressions. Standard errors clustered at the machinetype level are displayed in brackets. ***, **, and * indicate significance at the 1%, 5%, and 10%levels, respectively.

35

Table 5Machine Age and Geographical Patterns of Equipment Transactions

(1) (2) (3) (4)Log(Moving Distance) I(Same-County)

Log(1+Machine Age) -0.090*** 0.016***[0.016] [0.003]

Number (order of transactions) -0.081*** 0.017***[0.021] [0.003]

Observations 221,279 221,279 221,279 221,279R2 0.141 0.141 0.114 0.114Year FE Y Y Y YIndustry (SIC-3) FE Y Y Y YMachine Type FE Y Y Y YBuyer County FE Y Y Y YSeller County FE Y Y Y Y

Notes. This table documents the relationship between the moving distance of a machine in a transact andmachine age, based on the following model for each transaction i in year-quarter t,

Distancei = β ×Machine Agei + δFE + εi.

Machine Age is first captured using the number of years between the original manufacture date and the dateof transaction (columns 1 and 3). As an alternative measure to capture machine age, we construct the orderof transactions as the number of transactions the machine went through prior to i (columns 2 and 4). Wecapture moving Distance in two ways—the logarithm of the distance between the buyer and the seller of amachine (in miles) in columns (1) and (2), and a dummy variable indicating whether the transaction wasbetween two parties in the same county in columns (3) and (4). In all analyses we control for fixed effects atthe level of year, industry, machine type, buyer county, and seller county. Standard errors clustered at themachine type level are displayed in brackets. ***, **, and * indicate significance at the 1%, 5%, and 10%levels, respectively.

36

Table 6Equipment Purchase Choice and Local Old Capital Supply

(1) (2) (3)I (Bought) = 1

Local Old Capital 0.204*** 0.203*** 0.205***[0.000] [0.000] [0.000]

Young x Local Old Capital 0.006*** 0.005***[0.001] [0.001]

Young x Local Old Capital x Heavy 0.009***[0.003]

Observations 1,796,465 1,796,465 1,796,465R2 0.260 0.260 0.261Fixed Effects Year×County×Industry×Machine CategoryFully Saturated Model Y Y Y

Notes. This table studies the equipment purchase choice of firms in response to local old capital supply. Foreach realized transaction, we construct a set of “pseudo transactions” by pairing the buyer and all the otherpossible equipment types under the same equipment category. With the unit of observation being a potentialmachine purchase, and the outcome of interest whether or not a given machine was chosen, we then estimatethe differential effect of used capital supply on old and young firms under the model below:

I(Bought) =β × LocalOldCapital × Y oung+

β′ × LocalOldCapital + β′′ × Y oung + δFE + ε,

I(Bought) indicates whether the pair is a realized transaction (= 1) or a pseudo pair. Local Old Capital isthe old capital supply calculated from the natural aging of capital. Heavy is indicated as that the machinetype is at the top quintile of the weight-to-value ratio. Standard errors clustered at the transaction level(grouped by a real transaction and its paired pseudo transactions) are displayed in brackets. ***, **, and *indicate significance at the 1%, 5%, and 10% levels, respectively.

37

Table 7Local Old Capital Availability and Start-up Investment

Panel A: Old Capital Supply and Young Firm Growth

(1) (2) (3)Log(Total Investment) Log(New Machine Types) Log(New Machines)

Local Old Capital 0.094*** 0.073*** 0.050***[0.020] [0.016] [0.013]

Observations 73,584 73,584 73,584R2 0.216 0.241 0.195Year-Industry FE Y Y YCounty-Year FE Y Y Y

Panel B: Heavy vs. Light Machines

(1) (2) (3)Log(Total Investment) Log(New Machine Types) Log(New Machines)

Local Heavy Capital 0.054*** 0.039*** 0.032***[0.012] [0.009] [0.007]

Local Light Capital 0.002 0.004 -0.004[0.007] [0.005] [0.006]

Observations 73,584 73,584 73,584R2 0.216 0.241 0.195Year-Industry FE Y Y YCounty-Year FE Y Y Yp-value of Heavy vs. Light 0.000 0.000 0.000

38

Panel C: Young vs. Old

(1) (2) (3)Log(Total Investment) Log(New Machine Types) Log(New Machines)

Local Old Capital x Young 0.020*** 0.019*** 0.006[0.007] [0.004] [0.005]

Young 0.019** 0.016*** -0.001[0.008] [0.006] [0.006]

Observations 382,310 382,310 382,310R2 0.377 0.392 0.367County-Industry-Year FE Y Y Y

Notes. This table examines the relationship between start-up investment behaviors and local old capitalavailability, based on the following model

Investmenti = β × LocalOldCapitali + δFE + εi.

Local Old Capital is the old capital availability calculated from the natural aging of capital in the county-industry-year. Investment is measured in three ways—total investment measured as the number of machines,total number of new machine types, and the number of machines that are purchased as new. Heavy andLight machines are defined as the top and bottom quitile of the weight-to-value ratio, respectively. Standarderrors are double clustered at industry and county level and are displayed in brackets. ***, **, and * indicatesignificance at the 1%, 5%, and 10% levels, respectively.

39

Table 8Local Old Capital Availability and Start-up Entry

(1) (2)Log(Start-up Entry)

Local Old Capital 0.259***[0.092]

Local Heavy Capital 0.125***[0.033]

Local Light Capital 0.065**[0.030]

Observations 86,974 86,974R2 0.262 0.261Year-Industry FE Y YCounty-Year FE Y Yp-value of Heavy vs. Light 0.042

Notes. This table examines the relationship between start-up entry and local old capital availability, basedon the following model

StartupEntryi = β × LocalOldCapitali + δFE + εi.

The analysis uses a sample at the county-industry-year level. Local Old Capital is the old capital availabilitycalculated from the natural aging of capital in the county-industry-year. Start-up entry the number of newmachine-purchasing firms captured by the UCC data. Heavy and Light machines are defined as the top andbottom quitile of the weight-to-value ratio, respectively. Standard errors are double clustered at industry andcounty level and are displayed in brackets. ***, **, and * indicate significance at the 1%, 5%, and 10% levels,respectively.

40

Table 9Holding Period of Machines

(1) (2) (3) (4)Replace

≤ 3 Years ≤ 4 Years ≤ 5 Years

% of Young Firms 0.021** 0.032** 0.029** 0.038*[0.010] [0.013] [0.013] [0.022]

% of Young Firms x Heavy 0.121**[0.046]

Observations 644,529 617,656 556,213 550,881R2 0.159 0.168 0.173 0.245Year-County-Industry FE Y Y Y YIndustry-Machine Type FE Y Y Y Y

Notes. This table documents the relationship between the holding period of a machine acquired in transactioni and the % of young firms in the same county and industries that may demand the equipment, based on thefollowing model

Holding Periodi = β × % of Y oung Firms+ εi.

Holding Period is captured using whether the machine is held for less than five years. % of Young Firmsis the percentage of firms in the 0-3 year-old category in the same county, averaged across industries thathave historical purchase history of the specific machine type. We also perform the same analysis with thedependent variable focusing replacement, which is jointly defined by selling and buying a newer machinesimultaneously. Standard errors are clustered at the machine type level and are displayed in brackets. ***,**, and * indicate significance at the 1%, 5%, and 10% levels, respectively.

41

Online Appendix (Not For Publication)

A. Additional Details on Data and Measurements

A.1. Constructing Shale Shocks

We also utilize a cross-sectional measure for local liquidity based on Gilje, Loutskina,

and Strahan (2016) that traces deposit shocks due to shale oil discoveries across the branch

network of banks receiving large dollar inflows related to shale discoveries. We use data from

2000 to 2010 provided by Gilje, Loutskina, and Strahan (2016) capturing the timing and

magnitude of major shale discoveries in seven states: Arkansas, Louisiana, North Dakota,

Oklahoma, Pennsylvania, Texas, and West Virginia. For each bank with a branch in the

county receiving a windfall, the liquidity measure allocates a proportional fraction of the

shock, captured using the number of wells discovered, to active banks based on their fraction

of total deposits held in a windfall county. This generates a bank-quarter level variable which

we average at the county-quarter in non-windfall counties using the weight of those banks in

the county. Formally, the variable Shale Shock is defined as

ShaleShockc,t =∑

b∈B(c)

BankWeightb,c,t×∑c∈C

BankWeightb,c,t×BankShareb,c,t×Wc,t. (5)

Wc,t is the number of oil wells that have been discovered in county c by year t; BankShareb,c,t

is the fraction of deposits that bank b holds in county c in year t as a fraction of total deposit

in that county-year; and BankWeightb,c,t is the fraction of deposits that bank b holds in

county c in year t as a fraction of total deposits in that bank-year. Ideally, the measure

will allow us to capture cross-sectional variation in local lending conditions generated by

predetermined geography of bank branch networks, while avoiding the demand effects of local

economic conditions associated directly with a shale discovery.

A.2. Shipping Cost and Weight

Importantly, we logarize the weight in the numerator. This is because shipping cost of

equipment is with a logarithm relation, rather than a linear relation, with the equipment

weight. Conceptually, this non-linearity is due to that equipment shipping cost being a

A1

combination of a fixed cost and a variable cost increasing with weight. Empirically, we

document this nonlinear log-like relation in Figure A1.

Figure A1. Relation between Equipment Weight and Shipping Cost

Notes. This figure plots the average response of equipment purchase choices on local old capital supply acrossdifferent firm age groups. The reported coefficients are estimated from the following model

In specific, we construct a data set of real world shipping cost of equipment using heavy

equipment shipping broker website uShip.6 with the goal of characterizing the weight-cost

relation. For each equipment weight (created at the interval of every 200 lbs), we quote 40

different shipping routes, identical across all weights, ranging from same-county to roughly

3,000 miles. The average shipping cost across those routes can be viewed as a good proxy of

transportation expenses of each equipment after averaging out the influence of any particular

route or distance. In Figure A1, we plot this average shipping cost against weight, and see a

clearly non-linear relation that fits a log-function well.

6Data can be accessed at: www.uship.com.

A2

B. Additional Results

Figure A2. Firm Age and Machine Age—Used Capital Only

Notes. This graph plots the average age of machines purchased by firms across different age groups (1 to 20years old), as well as the 95% confidence interval. The graph uses only transactions of used machines.

A3

Table A1Machine Transactions Tabulated by Buyer Industries

No of Obs Percentage

Consumer NonDurables 168,011 11.11Consumer Durables 21,447 1.42Manufacturing 130,794 8.65Oil, Gas, and Coal Extraction and Products 25,494 1.69Chemicals and Allied Products 5,515 0.36Business Equipment 5,521 0.37Telephone and Television Transmission 1,987 0.13Utilities 4,059 0.27Wholesale, Retail 115,548 7.64Healthcare, Medical Equipment, and Drugs 43,834 2.9Finance 24,421 1.62Mining 20,628 1.36Construction 745,785 49.34Transportation 55,129 3.65Hotel 2,024 0.13Business Service 31,946 2.11Others 109,504 7.24

Total 1,511,647 100

Notes. This table provides descriptive statistics on machine transactions covered in the paper’s main analysissample. Machine transaction records are from the universe of UCC financial statement filings from 1990. Thetable report transactions based on buyer industries across the Fama-French 12 industries, where the “other”industry in Fama-French 12 is further decomposed to mining, construction, transportation, hotel, businessservice, and others.

A4

Table A2Main Results with Inverse Hyperbolic Sine Transformation

(1) (2) (3) (4)IHS(Machine Age)

IHS(Firm Age) -0.099*** -0.100*** -0.082*** -0.085***[0.009] [0.010] [0.027] [0.008]

Observations 1,494,387 1,494,387 1,494,387 1,494,387R-squared 0.128 0.262 0.541 0.599Year X County FE Y YIndustry (SIC-3) FE Y YMachine Type FE YFirm FE YMachine FE Y

Notes. This table reproduces Table 2 with firm age and machine age variables being transformed using inversehypobolic sine transformation instead of logarithm.

A5

Table A3Main Results with Size Control

Panel A: Add Total Employment As Controls

(1) (2) (3) (4)Log(1+Machine Age)

Log(1+Firm Age) -0.048*** -0.053*** -0.047*** -0.020***[0.005] [0.005] [0.011] [0.005]

Log(Total Number of Employees) -0.119*** -0.118*** 0.007 -0.015***[0.012] [0.013] [0.008] [0.004]

Observations 1,535,050 1,535,050 1,535,050 1,535,050R-squared 0.155 0.285 0.559 0.958Year-County FE Y Y Y YIndustry (SIC-3) FE Y Y YMachine Type FE YFirm FE YMachine FE Y

Panel B: Add Total Sales As Controls

(1) (2) (3) (4)Log(1+Machine Age)

Log(1+Firm Age) -0.036*** -0.041*** -0.048*** -0.019***[0.005] [0.004] [0.011] [0.005]

Log(Annual Sales) -0.098*** -0.098*** -0.019 -0.012***[0.010] [0.012] [0.019] [0.004]

Observations 1,556,704 1,556,704 1,556,704 1,556,704R-squared 0.158 0.286 0.556 0.958Year-County FE Y Y Y YIndustry (SIC-3) FE Y Y YMachine Type FE YFirm FE YMachine FE Y

Notes. This table reproduces Table 2 with adding size controls, where size is measured using number ofemployees and annual sales from Dun and Bradstreet.

A6

Table A4New table—later should be moved to appendix

Firm Age Census Employment (%) UCC Equipment Purchase (%)

0-1 YO 7.50 8.732-3 YO 7.61 8.064-5 YO 6.88 7.006+ YO 78.01 76.21

Notes. This table compares the firm age distribution of the UCC sample and the Census LEHD QWI sample.

A7