Download Rede ning Financial Constraints: a Text-Based

Redefining Financial Constraints: a Text-Based Analysis
Christopher Ball, Gerard Hoberg and Vojislav Maksimovic∗
August 19, 2012
ABSTRACT
We score text in firms’ 10-Ks to obtain for each firm an annual measure of
investment delays due to financial constraints and separate measures of each
firm’s marginal constraints due to concerns with equity and debt financing.
Contrary to the focus in the literature, we find that the key constraints are
the financing of R&D expenditures rather than capital expenditures, and that
the main friction firms face is raising equity capital to fund growth opportunities. We also find debt-market constrained firms more closely resemble
distressed firms funding CAPX. Our measures predict investment cuts following the financial crisis better than other indices of financial constraints used
in the literature. Since constraints are most binding within the populations of
small and young firms these results point to the challenges that the financial
system faces in funding innovation.
∗
meta Heuristica LLC, University of Maryland and the Securities and Exchange Commission, and
University of Maryland, respectively. Ball can be reached at [email protected].
Hoberg can be reached at [email protected] and Maksimovic can be reached at [email protected]. We thank Ran Duchin, Alex Edmans, Laurent Fresard, Steve Kaplan,
Tim Loughran, Toni Whited, Luigi Zingales, and participants at seminars at Nanyang Technological University, The National University of Singapore, the Securities and Exchange Commission,
the SFS Cavalcade at Darden, Singapore Management University, Stockholm School of Economics,
UCLA, the University of Michigan, USC and the 2012 UBC Summer Finance Conference for their
comments. We thank Danmo Lin and Austin Starkweather for excellent research assistance. All
c
errors are the authors alone. Copyright 2011
by Christopher Ball, Gerard Hoberg and Vojislav
Maksimovic. All rights reserved. The Securities and Exchange Commission disclaims responsibility
for any private publication or statement by any of its employees. This study expresses the authors views and does not necessarily reflect those of the Commission, the Commissioners or other
members of the staff.
Researchers have identified several theories explaining why financial constraints
might exist (asymmetric information, moral hazard, cost of contract enforcement,
and transaction costs). A limitation of existing measures of financial constraints is
that existing studies consider a single measure of constraints, even as theory recognizes that the binding constraint might separately relate to the ability to raise equity
or to raise debt. Moreover, existing constraint measures rely on predictive models
estimated using small samples of potentially unstable accounting ratios, which are
then applied out of sample for materially different populations of firms. We develop
a novel methodology that overcomes these limitations.
We find that the most constrained firms are those that require external equity
financing for R&D projects. Remarkably, these firms also have high Tobin’s Q.
These findings affirm the predictions of Krasker (1986)’s theoretical extension of
Myers and Majluf (1984). Krasker shows that asymmetric information can lead
to “equity rationing”, where firms with excellent investment opportunities face a
bounded issuance proceeds function and are forced to underinvest. In contrast, we
find that firms funding capital expenditures (those receiving most attention in the
literature) are less affected by financial constraints.
Our measures of the firm’s financial constraints are based on the analysis of
Management’s Discussion and Analysis (MD&A) section in 48,512 10-Ks. We focus
on mandated disclosures regarding each firm’s liquidity, as well as the discussion of
the sources of capital each firm intends to use to address its financing needs. We
calculate four direct measures of financial constraints for each firm in each year: those
due to broad liquidity challenges leading to potential under investment, and those
due to specific liquidity challenges pertaining to equity, debt and private placement
financing. We examine how the constraints for different forms of external finance
are related to firm characteristics, and show how the constraints are differentially
related to actual investment and issuance decisions.
Following KZ, we focus on the discussion of liquidity in the MD&A section of
the 10-K. This section is context sensitive, and when managers indicate the potential need to curtail or delay investment, the intended conclusion for the reader is
that the firm is investing less than what might be optimal due to the existence of
1
challenges to its liquidity.1 Our constraint indices differ substantially from those
generated following studies by KZ and Whited and Wu (WW) (2006). For example,
our new constraint measures are generally 15% or less correlated with the KZ and
WW indices.
We find that firms that are marginally constrained in the equity and debt markets
are highly distinct. Constrained firms focusing on equity are from industries with
high Tobin’s Q, low dividends, high cash holdings, low leverage and lower operating
income to sales. Relative to their industry peers, these constrained firms also invest
less in R&D and capital expenditures. In contrast, constrained firms focusing on debt
have diametrically opposite features on all of these dimensions. They tend to come
from industries with low Tobin’s Q, high dividends, low cash holdings, high leverage
and higher operating income to sales. Relative to their industry peers, these firms
invest more in R&D and capital expenditures. Our findings also go beyond those of
Hadlock and Pierce (2010) (HP), who find that firm size and age proxy for financial
constraints. Supportive of HP, the top-most chart in Figure 1 shows that smaller
and younger firms are more likely to be constrained according to our measures.
Going beyond HP, we also identify a large fraction of small and young firms that are
unlikely constrained, and a non-trivial number of medium sized and aged firms are
likely constrained. We are also the first to show, as depicted in the lower three charts,
that constrained firms focusing on equity, debt, and private placements are distinct.
Equity focused constrained firms are likely to be small, debt focused constrained
firms are likely to be old and large, and private placement focused constrained firms
are likely to be young.
We take our four constraint measures to the data and examine the responsiveness of firms’ investment and issuance policies to shocks. In an Opler and Titman
(1994) style test, we examine whether constrained firms curtail these polices more
than unconstrained firms during the 2008 financial crisis. In a second test aimed at
identifying a unique exogenous channel for liquidity, we consider the Edmans, Goldstein and Jiang forced mutual fund selling shock. As the forced mutual fund shock
1
In the next section, we extend the theoretical overview in KZ and outline why this text query
directly measure the existence of binding financial constraints.
2
is not industry-specific and only affects equities, we view this shock as an exogenous
shock to a firm’s equity market liquidity. This supply side shock should uniquely
affect equity focus constrained firms. We find this to be the case, with no significant
effect for the marginally debt constrained firms. These results are consistent with a
causal link between financial constraints and policy curtailment through the channel
of equity market friction.
We find that firms having an equity, debt, or private placement focus curtail
investment and issuance policies manage their investment and issuance policies differently. Equity focused constrained firms (and, more so, private placement constrained firms) most severely curtail their R&D, capital expenditures, and equity
issues following negative shocks. In contrast, debt focus constrained firms reduce
their debt issuances, but do not curtail R&D or CAPX investment. These firms appear to maintain their investments by substituting toward equity capital and away
from debt capital during the financial crisis. This substitution is uni-directional, as
equity focused constrained firms do not substitute toward debt following shocks.
We split our sample along several dimensions and find that our results are stronger
for small firms than for large firms. Moreover, small firms especially curtail R&D investment following shocks, whereas large firms especially curtail CAPX. Constraints
also have a larger overall impact on firms that are younger and those that focus on
R&D as their primary investment. These differential responses to shocks are economically large. For example, the highest tercile of delay constrained firms reduced
their R&D by 11.9% of sales in the financial crisis of 2008, whereas the lowest tercile reduced R&D by only 1.4% of sales. On the issuance side, the high constraint
tercile firms curtailed equity issuance by 2.6% of assets, compared to just 1.0% for
low constraint tercile firms. These differential reactions are roughly 10% to 20% of
one cross sectional standard deviation of each policy.
Our paper is most strongly related to existing articles that imply measures of
constraintedness including Fazzari, Hubbard and Petersen (FHP)(1988), and the
aforementioned articles by Kaplan and Zingales, Whited and Wu, and Handlock and
Pierce. Also related, a second strand of research that uses surveys to measure financial constraints and predict investment. Graham and Harvey (2001) and Campello,
3
Graham and Harvey (2011) analyze the role of constraints for large firms and Beck,
Demirguc-Kunt and Maksimovic (2006, 2008) for small firms. Our paper also draws
upon a growing literature that considers text-based analysis to test theoretical hypotheses in Finance.2
There is a large literature on the roots of financial constraints. Myers and Majluf
(1984), Krasker(1986), Greenwald, Stiglitz, and Weiss (1984), Jensen and Meckling
(1976), Hart and Moore (1998) have all identified reasons for a wedge between internal and external financing. Jensen (1986) argues that there are conditions under
which it is value enhancing for a firm to commit to reduce its ability to raise investment capital. Due to length constraints, we leave the matter of identifying why
constraints matter to future work. The goal of the current article is to provide improved measures of financial constraints, and to examine empirically whether (and
under what conditions) they have a real effect on firm policies.
Our empirical framework has several advantages. First, we obtain information
on constraints for a large fraction of the Compustat universe directly from firm disclosures. Our variables have the advantage of low ambiguity due to direct textual
context, and we do not rely on potentially unstable specifications based on aggregations of accounting variables. Second, we avoid the difficulty of having to predict
out of sample using unstable or potentially endogenous accounting variables. Third,
we can query the text further regarding related questions akin to a survey (but not
having to deal with issues of low response rates). Through this channel, the approach
has the potential to identify more specific channels through which constraints matter.3 Fourth, the methodology we are using is transparent, consistent, and can be
applied out of sample with the same interpretation. We find a high degree of consistency throughout our sample and believe that our methods can continue to be used
in future years.
The rest of the paper is organized as follows. Section I discusses the prior liter2
Early financial studies using text include Antweiler and Frank (2004) and Tetlock (2007). In
corporate finance, earlier work includes Hanley and Hoberg (2010), Hoberg and Phillips (2010),
and Loughran and McDonald (2011). See Sebastiani (2002) for an excellent review of text analytic
methods.
3
We leave deeper consideration of these channels to future work due to space constraints.
4
ature, Section II describes our data and methods. Our results regarding constraint
variable attributes are in Section III and our results regarding corporate finance
policies and constraints are in Section IV. Section V concludes.
I
Literature and Hypotheses
In early work, FHP argue that, holding investment opportunities constant, the capital expenditures of a financially unconstrained firm will depend on the Net Present
Value of its investment opportunities. Hence, if financial constraints are a first order issue, investment will become positively correlated with cash flow realizations
once the researcher controls for the value of investment opportunities. This simple
framework allows the authors to measure the degree of constrainedness by estimating
the coefficient on cashflow when CAPEX is the dependent variable and Tobin’s Q is
included as a control variable.
KZ criticize FHP on the grounds that the relation between cashflow sensitivity
and the degree of constraints is not necessarily monotonic and depends on the firm’s
production function and its cost of external funds. KZ use expert evaluations of
10-K MD&A statements for 49 firms to measure financial constraints. The analysis
reveals that some of the firms classified as constrained by FHP, in fact, appear to
be unconstrained. Many later studies use the “KZ Index”, which is a linear sum of
the accounting variables (firm cash flow, long-term debt, dividend-to-asset ratio and
Tobin’s Q) shown by KZ to predict constraints.4
The KZ index is estimated using a very small sample of firms. HP consider a larger
random sample of 356 firms over 1995-2004. They read the management’s letter to
shareholders and MD&A and group firms into categories of constrainedness based
on the number and quality of statements indicating financial constraints. HP then
re-estimate the link between accounting variables and constraintedness. They find
that the KZ index is unstable and that only size and age robustly predict financial
constraints. These results cast serious doubt on any attempt to measure constraints
4
Lamont, Polk, and Saa-Requejo (2001) were the first to apply the KZ index to measure constraints out of sample.
5
using essentially endogenous accounting variables.
A second approach, exemplified by Whited (1991) and WW, is to explicitly model
the relation between investment and the cost of external funds. A key advantage
is the possibility of directly estimating the shadow price of external capital. A
key limitation is theoretical assumptions can be restrictive, especially regarding the
functional form of the shadow price of external finance. WW generate a WW index
of constraints by relating this shadow price to firm and industry long-term debt to
total assets, a dividend payer dummy, firm and industry sales growth, firm size, liquid
assets to total assets, and cash flow to total assets. Later studies use the estimated
coefficients out of sample to measure financial constraints.5
Our approach is to use automated textual analysis of the MD&A section of the
10-K to construct direct indices of financial constraints. Our methodology allows full
coverage of all firms, and does not rely on any mapping to accounting variables as
an intermediate step. We focus on direct statements regarding the need to delay or
curtail investments as a result of issues relating to liquidity challenges. Hence, we
only consider statements of curtailed investment appearing alongside a discussion of
financial liquidity (the organization of 10-Ks makes this straight forward), and we
do not consider any other indications of investment delay in other parts of the 10-K.
Separately, ours is the first study to assess whether a firm views either debt or equity
capital as being more central, and related impact on constraints.
Our approach has several advantages. By reading MD&A text directly, we avoid
the use of out of sample coefficients from a potentially unstable predictive model.
The text is also scored in a way that does not rely on human judgment, and thus it
is easily applied in extended samples. Finally, by using direct statements of financial
constraints, we do not rely on data processed through a structural model, which
depend on assumptions regarding firm and market characteristics.
SEC Regulation S-K obligates firms to discuss (A) challenges to their liquidity,
(B) their demand for liquidity (investment plans), and (C) the sources of available
liquidity.6 We find that firms widely comply, and we find the relevant discussion in a
5
6
It should be noted that WW do not use estimates out of sample in this way.
See, for example, http://taft.law.uc.edu/CCL/regS-K/SK303.html for the full text of Item 303.
6
machine readable subsection of MD&A for roughly 80% of all firms. This subsection
is often titled “Liquidity and Capital Resource”.7
(1) Liquidity. Identify any known trends or any known demands, commitments,
events, or uncertainties that will result in . . . the registrant’s liquidity increasing
or decreasing in any material way. If a material deficiency is identified, indicate
the course of action that the registrant has taken or proposes to take to remedy
the deficiency. Also identify and separately describe internal and external sources of
liquidity, and briefly discuss any material unused sources of liquid assets.
(2) Capital Resources. (i) Describe the registrant’s material commitments for
capital expenditures as of the end of the latest fiscal period, and indicate the general
purpose of such commitments and the anticipated source of funds needed to fulfill
such commitments . . .
(ii) Describe any known material trends, favorable or unfavorable in the registrant’s capital resources. Indicate any expected material changes in the mix and the
relative cost of such resources.
”Liquidity” is clarified in Instruction 5: “refers to the ability of an enterprise to
generate adequate amounts of cash to meet the enterprise’s needs for cash...Liquidity
generally shall be discussed on both a long-term and short-term basis.”
We posit a simple extended version of the model in KZ regarding MD&A disclosure and firm liquidity. First consider a firm with adequate cash to fully fund its
investments. This firm will maximize its objective function F (I) − I, where F (I) is
the firm’s net revenue function, which depends on the the firm’s investment level I.
Let I ∗ denote the optimal investment that solves this problem.
Now suppose the firm’s cash is less than I ∗ and hence the firm must use external
financing. Further suppose that the cost of external capital might differ from the
opportunity cost of internal cash. We refer to the difference between the cost of
external and internal capital as a financing wedge. We extend the KZ framework
and further assume that the firm has a choice of financial instruments: equity and
7
The subsection can also appear with a number of related titles, which we discuss in our online
Appendix 1.
7
debt. Let wE (I, θ) and wD (I, θ) be the financing wedges associated with debt and
equity financing respectively, where θ describes the firm’s growth options, and all
characteristics that affect its ability to raise capital.
The relative size of the equity and debt financing wedges depends on firm characteristics and market conditions. The firm prefers the form of financing with the
smallest wedge (ie, it takes the larger of the maximized quantities F (I) − wE (I, θ)
and F (I)−wD (I, θ)). In general the existence of a financing wedge reduces the firm’s
desired investment below the optimal level I ∗ , to a constrained level I C .
We hypothesize that if I ∗ − I C is immaterial, then the firm does not disclose
this shortfall, and does not extensively discuss the financing wedges that is faces.
However, since firms face legal liability with respect to regulation S-K, if I ∗ − I C
is material relative to the firm’s size and operations, the firm discloses a potential
delay or curtailment of investment statement. Pursuant to Regulation S-K, the firm
also comments on the nature of the wedge. If the wedge is smaller for equity (debt),
then the firm will also disclose its intention to try to issue equity (debt).
We thus score each Liquidity and Capital Resource subsection based on indications of broad financial constraints I ∗ − I C , and separately for discussion of equity
and debt financing wedges. Our analysis focuses on several questions:
First, following shocks to θ, are changes in the levels of R&D and CAPX greater
for for firms that report high expected investment delays I ∗ − I C ? Such a relation is
implied by a well-behaved wedge function w(I, θ).8
Second, are changes in R&D and CAPX more likely to be associated with financing wedges involving equity or debt? The reporting requirements of our data set
are particularly well-suited to address this question, which is absent in the existing
literature. R&D expenditures are more likely to be financed using equity (Myers
(1977)), and debt is more likely to fund tangible investments such as CAPX. Thus,
we would expect a higher sensitivity of R&D for equity-focused firms, and greater
sensitivity of CAPX for debt-focused firms.
Third, we estimate the relationship between security issuance and financial con8
See, for example, KZ.
8
straints. We expect that following a shock, constrained firms will experience a greater
reduction in security issuance than unconstrained firms.
Fourth, we explore differential security issuance responses to shocks for equity
or debt-focused firms. First, we expect that an equity (debt) focused firm will experience a greater reduction in equity (debt) issuance following a shock. Second,
there may be a substitution effect, and a shock may induce a firm to switch to a less
desirable form of financing. For example, a debt focused firm may substitute equity
for debt in order to avoid curtailing its investments.
Fifth, we examine whether firms focused on private equity respond to shocks differently. As shown by Gomes and Phillips (2006), private equity issuers are opaque
and might have greater informational asymmetries. We expect these firms’ investment and issuance policies to be particularly sensitive to shocks.
II
Data and Methods
The variables we create derive purely from 10-K text extracted from the “Management’s Discussion and Analysis” (MD&A) section. We utilize text processing
software provided by meta Heuristica LLC to web-crawl and parse the MD&A section from each 10-K for all electronically filed 10-Ks from 1997 to 2009. From each
MD&A, the software then extracts the Liquidity and Capital Resources subsection.
This subsection’s title is not fully uniform, and we explain in Appendix 1 of our
Online Appendix how we unify subsections of similar content. This key subsection,
which is also used by KZ contains the firm’s remarks concerning its financial liquidity
and its intentions regarding future capital market interactions. The software uses
natural language processing to parse and organize textual data, and its pipeline employs “Chained Context Discovery”.9 This approach leverages an empirical ontology
to normalize the structure of the content and map it to a standardized form. The
software then stores processed data in a high speed database.
From a research perspective, the technology enables fast and thorough querying,
while permitting analysis that is also easy to interpret. For example, many of the
9
See Cimiano (2010) for details.
9
variables used in this study are constructed by simply identifying which firm-year
filings (within a set of 40,000+ filings) specifically contain a statement indicating
that the firm may have to delay its investments due to financial liquidity issues.
The database supports advanced querying including contextual searches, proximity
searching, multi-variant phrase queries, clustering, and other capabilities. These
tools are designed to accelerate discovery of linguistic patterns that can further enrich
content.
A
The CRSP and COMPUSTAT Sample
We start with 56,496 firm-years in the COMPUSTAT sample from 1997 to 2009
with sales of at least $1 million and positive assets excluding financials and regulated utilities (SIC 6000 to 6999 and SIC 4900 to 4949, respectively) with adequate
COMPUSTAT data. The years are chosen based on availability of SEC Edgar data.
B
The Sample of 10-Ks
Our sample of 10-K MD&A’s is extracted by web-crawling the Edgar database for
all filings that appear as “10-K,” “10-K405,” “10-KSB,” or “10-KSB40.” MD&A
generally appears as Item 7 in most 10-Ks. The document is processed for text
information, fiscal year, and the central index key (CIK).
We link each document to the CRSP/COMPUSTAT database using the central
index key (CIK), and the mapping table provided in the WRDS SEC Analytics
package. Of the 56,496 observations available in CRSP and COMPUSTAT database
noted above, we are left with 48,512 after requiring that same-year machine-readable
text data is available. This loss of roughly 4,000 observations is due to two reasons:
(1) a mapping from CIK to gvkey is not available or (2) some MD&As are not
filed in the 10-K itself, but are incorporated by reference. Of the 48,512 machine
readable MD&A sections, 38,980 of these also have a machine readable Liquidity and
Capitalization Resource Subsection (henceforth CAP+LIQ). We later show that the
firms that have a machine readable MD&A, but do not have a separable CAP+LIQ,
are generally healthy firms that have few if any liquidity issues to disclose.
10
C
Using Text to Identify Financial Constraints
Our first query identifies firms that discuss the possibility of delaying investment
in CAP+LIQ. Because firms focus on liquidity issues in CAP+LIQ, it follows by
context that these firms are curtailing investment due to challenges to the firm’s
liquidity. Indeed some firms often state explicitly that poor financing options are to
blame. Importantly, we do not identify investment curtailment anywhere else in the
10-K to insure we focus solely on issues relating to financial liquidity.
Because there are many ways to verbally indicate the notion of curtailment, and
even more forms of investment, identifying a reliable set of firms that are curtailing
investment using text searches offers several challenges. Using the two-sided sentence
view features in the metaHeuristica software, we were able to identify key synonyms
to the word curtail, and a long list of words identifying types of investment. To
identify firms that are directly reporting potential delays in investment, we query
CAP+LIQ for instances in which one word from each of the following two lists exists
side by side (separated by no more than one stop word such as “the”, “of”, “and”,
etc). Note that “*” denotes a wildcard.
Delay List 1: delay* OR abandon OR eliminate OR curtail OR (scale back)
OR postpone*
Delay List 2: construction OR expansion OR acquisition* OR restructuring
OR project* OR research OR development OR exploration OR product* OR expenditure* OR manufactur* OR entry OR rennovat* OR growth OR activities OR
(capital improvement*) OR (capital spend*) OR (capital proj*) OR (commercial
release) OR (business plan) OR (transmitter deployment) OR (opening restaurants)
We refer to firms that trigger this query as the “precise training set”, as these firms
are very directly indicating that issues relating to financial liquidity are resulting in
potential under investment. We find that 1.7% of all firm-year observations load
on this query. However, in reading many other CAP+LIQ examples, we also note
that a large number of firms exist that have words from both lists present, but they
11
are not necessarily side by side. This is a natural consequence of the need to delay
investment being a rather intricate issue to express verbally. Some writers add more
detail, whereas others are more Spartan.
As our objective is to identify as many firms as possible who indicate delay of
investment, we thus consider a query that relaxes the requirement that both words
from both lists must exist side by side in the text. Instead, we require that both
words appear within a twelve word window of one another. Indeed this identifies a
larger set of firms, as 5.5% of our sample firms load on this enhanced query. We refer
to this set of firms as the “12-word enhanced training set”.10
We also identify which firms are focused on issuing equity or debt. In using the
two sided sentence views available in metaHeuristica software, we identify a number
of commonly used phrases that identify the intent to issue equity as follows:
Equity Focused List: issuing equity securities OR expects equity securities OR
through equity financing OR sources equity financing OR seek equity investments
OR seek equity financings OR access equity markets OR raised equity arrangements
OR undertake equity offerings OR sell common stock OR issuing common stock
OR selling common stock OR use equity offerings OR offering equity securities OR
planned equity offering OR seek equity offering OR raise equity offering OR equity
offering would add OR additional equity offering OR considering equity offering OR
seek equity financing OR pursue equity offering OR consummates equity offering OR
raises equity capital OR raise equity offering OR sources equity offering.
Regarding debt focused firms, we consider the following commonly used phrases.
Debt Focused List: increased borrowings OR use line of credit OR expanded
borrowings OR funded by borrowings OR additional credit lines OR incur additional
indebtedness OR pursue lines of credit OR anticipates lines of credit OR through
loan financing OR borrowings bond issue OR increase line of credit OR provided by
credit facilities OR seek borrowing transaction OR raise borrowings OR additional
bank financing OR raises debt capital OR secure line of credit OR borrowing of
capital
10
We have experimented with 6- and 8-word windows and they yield similar qualitative conclusions.
12
For both equity and debt focus, we also consider a precise query identifying firms
that use these phrases directly, and an expanded query allowing these phrases to be
expressed in a twelve word window. For the precise query, we find that 1.0% and
3.4% of sample firms are equity and debt focused, respectively. When we use the
twelve word window, we find that 12.8% and 14.6% of sample firms are equity and
debt focused, respectively.
Finally, we also identify which firms are focused on issuing private placements of
equity. We require that one word from each of the following three lists be present
within a 12 word window in CAP+LIQ:
Private Placement List 1: private
Private Placement List 2: placement OR placements OR sale OR sales OR
offering OR offerings OR infusion OR infusions OR issued OR issuance OR financing
OR financings OR funding
Private Placement List 3: equity OR stock
We find that 9.4% of the firms in our sample load on this query.
D
Degrees of Financial Constraints and Scores
The methodology in the above section identifies a set of firms that directly indicate
constrainedness, or intent to issue debt or equity. However, we are concerned about
two potential sources of error in measuring the degree of constrainedness. First, some
firms may indicate constrainedness in their overall discussions, but they may not
conclude their discussion with a direct statement regarding investment curtailment.
Second, some firms may be partially constrained, and this would be overlooked using
all-or-nothing direct queries. We thus consider the methodology used in Hanley and
Hoberg (2010) to tap the CAP+LIQ vocabulary more fully, and to generate a more
informative degree of constrainedness metric.
13
We thus score each CAP+LIQ section based on how similar its vocabulary is to
those firms in a given training set of interest, while controlling for the presence of
standard (boilerplate) text. This method generates a continuous score for each firm,
and firms with higher scores are more constrained. Firms with lower scores are either
unconstrained or only partially constrained.
Where N denotes the number of unique words in the entire corpus, we define an
N -vector for each firm i in each year t as wordi,t . This vector is populated with the
number of times each word corresponding to each of the N elements is used in firm
i’s CAP+LIQ subsection in year t. We next define the normalized vector for each
firm year normi,t as being equal to wordi,t divided by the sum of its components.
This is the vector we seek to “explain” using various text factors related to boiler
plate content, delay investment constraint content, equity focus content, debt focus
content, and private placement focus content.
We define standard (boilerplate) content in year t stant as the average normi,t
over all firms in the corpus in the given year. We define the delay investment text
vector in year t using a two step procedure. First we compute the average word usage
vector using all words in all paragraphs for paragraphs that are members of the “12word delay investment training set” discussed in the previous section. Second, we
regress this vector on the standard content vector stant and compute the residual.
This residual vector, which we label delayt approximates the non-standard (or nonboilerplate) content used by firms that indicate intent to possibly delay investment.
The main idea is that a firm that has content that is similar to this average training
set vocabulary is more likely to be more constrained.
We compute equity focus content using a similar procedure. However, we first
note that our objective is not just to identify firms that are focused on potentially
issuing equity, but more precisely those that are both constrained and focused on
issuing equity. This allows us to explore whether there is a difference between firms
that might be equity constrained or debt constrained. Our first step is to compute
the average word usage vector using all words in all paragraphs for firms that are
members of both the “12-word delay investment” training set, and the “12-word
equity focus” training set, but not members of the “12-word debt focus” training
14
set (all training sets are discussed in the previous section). Second, we regress this
average word usage vector on both the standard content vector stant and the delay
investment vector delayt and take the residual. This residual vector, which we label
equityf ocust approximates the marginal non-boilerplate content in these paragraphs
that is unique to firms that both indicate the need to delay investment, and the
potential desire to issue equity (but not the desire to issue debt). Any firm that
has content that is highly similar to this content is likely constrained, and it is also
focused on issuing equity rather than debt to meet its liquidity demands. We define
our debt focus constraint variable debtf ocust in a fully symmetric fashion.
With this set of word usage vectors, all of length N , we are able to decompose a
given firm’s CAP+LIQ section into components that load on each vocabulary factor,
and we run the following decomposition regression for each firm in each year:
normi,t = cstan,i,t stant + cdelay,i,t delayt + cequity,i,t equityf ocust +
(1)
cdebt,i,t debtf ocust + i,t
The coefficients from this regression are the “degree of constraint” scores for firm i in
year t. We therefore define the “Delay Investment Score” as the coefficient cdelay,i,t ,
“Equity Focused Score” as the coefficient cequity,i,t , and “Debt Focused Score” as
the coefficient cdebt,i,t . These variables respectively measure the degree to which a
firm faces financial constraints, faces constraints but has an equity focus, and faces
constraints but has a debt focus, respectively.
Figure 2 displays the empirical distribution of our text-based constraint scores.
The figure shows that these variables generally have a bell shaped distribution without extreme outliers. The distributions also have two additional features. First, all
three variables have a spike above the bell curve at zero. This occurs because roughly
4% of all firms that have a CAP+LIQ section that is very small. In these cases, the
loading on the constrained training set can be zero if none of the words in the small
CAP+LIQ section coincide with those in the constrained training sets. Due to the
existence of this small spike, we examine robustness to including a dummy variable
for firms with scores of exactly zero, and our inferences are unchanged. We also
consider a test in which we reclassify firms that have a very small CAP+LIQ section
15
as not having a CAP+LIQ section for the purposes of our study (inferences are also
unchanged). Henceforth we use the constraint score variables described above for
parsimony. The second feature of the distributions is that all three are right skew.
This is primarily due to members of each training set receiving higher scores than
firms that are not in the training set. We view this behavior as appropriate and
consistent with scoring direct statements higher.
We also compute a “Private Placement Focused Score”, where the objective is to
identify the focus on selling equity to private investors, holding fixed the degree of
unconditional delay constraints, and holding fixed the focus on selling equity or debt
unconditionally. Our first step is to compute the average word usage vector using all
words from all firms that are members of both the “12-word delay investment” training set, and the “12-word private placement focus” training set. Second, we regress
this word usage vector on the standard content vector stant , the delay investment
vector delayt , the equity focus vector equityf ocust , the debt focus vector debtf ocust ,
and take the residual. This residual vector, which we label privatef ocust approximates the non-boilerplate content that is unique to firms that are both indicating the
need to delay investment, while also mentioning the potential desire to issue a private
placement, holding fixed the desire to issue equity or debt. This private placement
variable is then added to the model in equation 1, and the resulting coefficient on
privatef ocust is the given firm’s degree of focus on private placements.
We also compute a “Covenants Violation” score variable using a similar two step
procedure. First we identify a training set based on firms that mention the words
covenant and violation within a 12-word window. We then compute the average
word usage of paragraphs that match on this query in each year and thus define
the N -vector covenantt . Next we run the following decomposition regression (which
controls for standard content as before):
normi,t = cstan,i,t stant + ccovenant,i,t covenantt + i,t
(2)
We define “Covenant Violation Score” as the coefficient ccovenant,i,t , and this variable
measures how close firms are to satisfying the conditions generally necessary to trigger
covenant violations.
16
E
Investment Style Measures
A key foundation of our framework is that a firm must have viable investments before
it can possibly report potentially having to delay them. As a result, it is natural
to explore the link between financial constraints and investment activity. We thus
construct three measures of investment style based on vocabularies associated with
three classes of investment. In later analysis, these variables will enable us to examine
financial constraints holding fixed each firm’s investment opportunities.
We separately consider the following text queries:
R&D Focused List: research OR development OR new product OR new products.
CAPX Focused List: construction OR expenditure OR rennovation OR capital improvement OR capital spending OR capital project OR opening restaurants.
Unlike our earlier analyses, which were limited to CAP+LIQ, discussion of investment opportunities is generally in the MD&A but generally not in the CAP+LIQ
subsection. Hence we use the entire MD&A to identify which firms are most focused
on various investments, and we do not rely on the CAP+LIQ section alone as we did
in earlier variable constructions.
We compute the fraction of paragraphs in each firm’s MD&A that score on each
query above. This generates two measures of investment focus that are a function of
the entire MD&A: MD&A R&D Focused Score, and MD&A CAPX Focused Score.
Each variable is bounded in the interval [0,1].
F
Measuring Shocks
We also use the text in the Management’s Discussion and Analysis (MD&A) to
identify firms that might be facing shocks. For this analysis, we query the entire
MD&A, as we find that firms mention potential shocks in various locations of this
section. We consider competitive, demand, and cost shocks. In each case, we identify
shocks by requiring that two words from two separate lists exist side by side in the
text of a given firms’ MD&A in a given year. These variables are relevant to exploring
17
which firms move from an unconstrained to a constrained status in time series.
Our first variable is the “Higher Competition Dummy”. This variable is set to one
when a firm-year MD&A has the word increasing or one of its synonyms (increasing
OR increased OR higher OR greater OR intensified OR intensification) appearing
beside the word competition in its MD&A in the given year.
We next define the “Increasing Cost Dummy”. This variable is set to one when
a firm-year MD&A has the word increasing or one of its synonyms (increasing OR
increased OR higher OR greater OR intensified OR intensification) appearing beside
the word cost in its MD&A in the given year.
We next define the “Lower Demand Dummy”. This variable is set to one when
a firm-year MD&A has the word lower or one of its synonyms (decreasing OR decreased OR lower OR reduced OR reduction) appearing beside the word demand in
its MD&A in the given year.
Beyond using text data, we also explore the impact of the 2008 the financial crisis
on constrained versus unconstrained firms. We also explore the role of forced mutual
fund selling as in Edmans, Goldstein and Jiang (2011).11 This measure is based on
mutual fund redemptions, and can be seen as an exogenous shock to equity market
liquidity that is not attributable to firm policies.
G
Compustat Variables
The four key policy variables we examine are R&D/sales, CAPX/sales, equity issuance/assets, and debt issuance/assets. All four are constructed directly from
Compustat data. Regarding equity and debt issuance, we only consider newly issued
equity (SSTK) and newly issued long term debt (Compustat DLTIS), both scaled
by assets. All four policy variables are winsorized at the 1/99% level to control for
outliers. In all cases, we winsorize financial variables that are defined as ratios at the
1/99% level in each year to reduce the impact of outliers. We also include controls
for firm age and size (log assets) in many specifications.
11
We thank Alex Edmans for providing this data on his website. Phillips and Zhadanov (2011)
also use this variable as an instrument for a firm’s equity valuation that is unrelated to fundamentals.
18
III
Financial Constraints
In this section, we first examine the properties of our constraint variables in time
series and cross section. We then compare them to measures used in the literature.
We also note that the Online Appendix to this paper contains three detailed
appendices outlining (1) the heterogeneity in subsection names regarding subsections
that are analogous to Liquidity and Capital Resource, (2) ten sample firms from
each of our sample years that scored highly on our financial constraint measures
along with their Fama-French-48 industries, and (3) sample paragraphs from the
CAP+LIQ section of the firms with high constraint scores. Based on these online
appendices, we note that the firms that are most constrained are often, but not always
in the drug and technology industries. An interesting exception is that home builder
Toll Brothers was financially constrained in 2005 and 2006 exactly when valuations
and uncertainty were high, but not in later years when, arguably, this industry
was more in distress than it was financially constrained. Regarding the sample
paragraphs, the examples show that the text queries are highly reliable in terms of
identifying firms that acknowledge the possibility of delaying their investments in
CAP+LIQ. However, the results also show that the queries are not fully perfect, as
occasional “difficult to parse” 10-Ks such as Gliatech’s 1997 10-K can add some noise
to our calculations. However, because we focus on average word distributions across
thousands of firms to form the average vocabulary of the training set as a whole, we
do not believe this low level of noise is prolematic.
A
Summary Statistics and Correlations
Table I displays summary statistics for our 1997 to 2009 panel of 48,512 firm-year
observations having machine readable MD&A sections in their 10-K. Panel A shows
that 80.7% of these firms also have a machine readable CAP+LIQ subsection. We
also find that 8.3% of MD&A paragraphs mention R&D, but only 1.8% mention
capital expenditures. Because we require a CAP+LIQ section to compute our key
variables, we conduct our empirical analysis in two stages: (1) we examine which
firms have a distinct CAP+LIQ section, and (2) among those that do, we explore
19
our key hypotheses relating to financial constraints. In later sections, our results will
suggest that the firms that do not include a CAP+LIQ section are primarily unconstrained firms. As a result, the fact that we focus on firms that do have CAP+LIQ
is unlikely to be problematic because the primary challenge in the literature is to
identify constrained firms rather than unconstrained firms.
[Insert Table I Here]
Panel B reports summary statistics regarding the membership of our precise
training sets, and Panel C reports summary statistics regarding the members of
our 12-word expanded training sets. Not surprisingly, expanding the text queries to
12 words greatly improves coverage. For example, coverage increase from 1.7% to
5.5% for delay investment constraints. The larger training sets offer improved power.
Panel D reports summary statistics for our score-based constraint variables, and
Panel E reports summary statistics related to covenant violation score and the size
of the CAP+LIQ section. Because the construction of these variables controls for
boiler plate content, intuitively, these scores are relative rankings and have means
near zero. These variables measure the degree to which the given firm’s CAP+LIQ
section is similar to that of firms in each respective training set. This approach
provides a measure of the degree to which a firm is constrained.
Panel F reports the summary statistics for our text-based shocks, which have
averages of less than 15%. Panel G reports summary statistics for the four corporate
finance policies we examine. These results suggest that firms in our sample spend an
average of 18.1% of their sales on R&D and 10.6% of their sales on CAPX. Equity
and debt issuance average 5.2% and 10.1% of assets, respectively.
[Insert Table II Here]
Table II reports Pearson correlation coefficients between our financial constraint
variables and other constraint variables including the KZ index and the WW index.
The three text constraint variables are only slightly correlated. This is by construction using orthogonalized text vectors, as our intent is to (1) identify the degree of
investment delay constraints, and then (2) to identify the marginal effect of debt and
20
equity focus among constrained firms. Our delay investment variable is just -1.9%
(4.3%) correlated with our equity (debt) market constraint variable. The debt and
equity constraint variables are modestly negatively correlated at -8.9%. This correlation is likely due to the fact that a purely equity focused firm is not a debt focused
firm, although many firms might consider both forms of capital.
In comparing our constraint variables to existing measures, we find that our delay
investment constraint is positively correlated with both the KZ index (3.9%) and the
WW index (8.2%). However, readers may find it surprising that these correlations
are rather low. The finding indicates that these variables contains much distinct
information. Our later analysis suggests that the KZ index and the WW index
primarily load on investment opportunities, firm size, and firm age.
Because firms must have viable investment opportunities before they might be
forced to delay them, our three text-based constraint variables also correlate positively with investment opportunities. Firms that are delay investment constrained
tend to be more focused on R&D than CAPX, likely because the former investment
type is more prone to informational asymmetry. Equity focused constrained firms
load even more on R&D relative to CAPX, whereas debt focused firms load more on
CAPX. The results for debt focused constrained firms are consistent with a role for
asset tangibility and lower uncertainty.
The table also shows that the KZ index and the WW index correlate in an opposite way with our equity and debt market constraint variables. The WW index
correlates more with equity-focused constrained firms, and the KZ index correlates
more with debt-focused constrained firms. Also relevant, our debt market constraint
variable correlates positively with firms reporting covenant violations (60.5%). This
suggests that these firms (not surprisingly) experience some distress. However, our
other two constraint variables (delay score and equity focused delay) are somewhat
negatively correlated with covenant violations text, indicating that they contain information potentially linked to market frictions and not distress.
21
B
Time Series and the Financial Crisis
Although not reported to conserve space, we note that our financial constraint variables are somewhat persistent, but not overwhelmingly so. Our three key textual constraint variables have autocorrelation coefficients that range from 0.4 to 0.6. Hence,
a firm that is constrained in a given year is quite likely to be constrained next year,
but unlikely to still be constrained two to three years later.
Table III reports the time series averages of our constraint and shock variables.
Although our results suggest there is a link between financial constraints and business
cycles, we interpret these results with caution as our sample covers just thirteen
years. We display results for all firms (Panel A), large firms (Panel B), and small
firms (Panel C).
[Insert Table III Here]
Panel A of Table III shows that our constraint variables experience meaningful
time series variation, and that the financial crisis of 2008-2009 had a significant
influence on constraints. Both the precise delay dummy and the 12-word window
delay dummy jump sharply from 2007 and reach an all-time high in 2008. This
increase is statistically significant at the 5% or 1% level despite our short sample.
Although both debt and equity focused delay constraints also increase in columns
three to six in Panel A, the increase is only significant for equity focused delay. This
provides evidence that many constrained firms sought equity financing around the
time of the crisis, but they did not have confidence they would obtain it.
Comparing large firms (Panel B) to small firms (Panel C) reveals that the financial
crisis did not affect all firms equally. Although both groups have significant increases
in delay investment constraints during the crisis, small firms have significant increases
in equity focused delays but not debt-focused delays. Larger firms have the opposite
result. This contrast is stark in statistical and economic terms. For example, small
firms actually experienced a slight decline in debt focused delays (12-word window)
during the same time that the equity focused delays increased substantially to more
than double the 2006 levels. These results suggest that although both large and
22
small firms became more constrained, smaller firms became more focused on raising
equity and larger firms focused more on raising debt. In both cases, firms did not
have confidence in their ability to successfully raise capital.
The table also shows that the investment focus variables changed little during
our sample period, and that the shock variables also had little or only moderate
aggregate variation. This is not surprising if shocks are mainly localized within
product markets (as one might expect to be the case for competition, cost, and to
some extent demand shocks). Although significance levels are just below standard
levels, we find suggestive evidence that our low demand shock variable increases
during times of relative economic distress. For example, this variable was relatively
low during the 1990s, but increased in 2001 and 2002 following the events of 9/11
and the technology bust. This variable jumps again in 2008. Hence, low demand
might be linked to some of the problems faced by firms in the financial crisis.
C
Firm Characteristics and Constraints
We begin our analysis by assessing which firms have a machine readable CAP+LIQ
subsection in their MD&A in a given year. Our summary statistics indicate that
80.7% of firms have this subsection, and we conjecture that only unconstrained firms
would omit this content. This conjecture is based on the fact that disclosure of
liquidity is required by law, and only unconstrained firms might have sufficiently
little to disclose to warrant not doing so in a separate subsection.
Table IV formally tests this conjecture using a logistic regression in which the
dependent variable is a dummy equal to one when the firm has a CAP+LIQ subsection. We include industry and year fixed effects (Panel A), or firm and year fixed
effects (Panel B), and all standard errors are adjusted for clustering at the firm level.
[Insert Table IV Here]
The table confirms that firms disclosing a CAP+LIQ subsection indeed are more
likely to be constrained. Consistent with HP, these firms have fewer assets in place
and are younger. Firms having a CAP+LIQ section are also less likely to pay dividends, tend to be more R&D oriented, and tend to have higher leverage. The focus
23
on R&D rather than CAPX by these firms, supports a possible link between liquidity challenges and asymmetric information. Because we are primarily interested in
examining constrained firms, the remainder of this paper focuses only on the sample
of firms that has a machine readable CAP+LIQ subsection.
In Table V, we consider panel data regressions in which we explore the association between various firm and industry level characteristics, and our four constraint variables (delay investment constrained, equity-market constrained, debtmarket constrained, and private placement constrained) plus controls for the size
of the CAP+LIQ section, firm size, and firm age.
[Insert Table V Here]
The results of Table V are consistent with our delay investment constraint variable
indeed being a valid measure of financial constraints. We highlight three major pieces
of evidence, but also note other nuances. First, the table shows that firms with
high delay investment scores operate in industries with high Q. This is consistent
with these firms having good investment opportunities, a necessary requirement for
binding financial constraints (a firm must have a promising investment before it
can delay it). This result also confirms that financial constraints and distress are
distinct concepts, as distressed firms intuitively have low Q. Second, we find that
the delay investment score is negatively correlated with profitability, consistent with
their having little in the way of free cashflow.
The third major finding regarding the delay investment variable is that it is
positively related to both capital expenditures and research and development at the
industry level. This is likely a necessary condition for a financial constraint to be
binding, as the firm must have promising investment opportunities before it can delay
them. More striking, the delay investment variable is negatively related to own-firm
R&D and CAPX. This supports the direct prediction of financial constraints: the firm
has low investment relative to what would be optimal if it had perfect liquidity (so
the constrained firm in an industry invests less than its industry peers on average).
Finally, we note the strong statistical significance of these findings. In all cases,
standard errors are adjusted for clustering at the firm level.
24
Table V also shows that equity market and debt market constraints are different.
For example, the majority of coefficients have opposite signs for these two variables.
This is quite remarkable given that these variables are mutually little correlated as
discussed earlier. Hence, their mirror image results are likely due to economic forces.
The overall coefficient patterns also show that equity market constrained firms are
similar to delay investment constrained firms. These firms have high Q and little
in the way of current profits, and hence they are more growth oriented. Regarding
investment policy, they also reside in product markets with high levels of R&D,
although their own-firm R&D is lower than the industry average. These firms also
reside in product markets where firms tend to hold high levels of cash, consistent
with precautionary savings relating to financial constraints.
The results for the private placement score are similar overall to those for the
equity focused score. Intuitively, private placement focused firms are essentially
more extreme versions of equity focused firms. Firms that score highly on both
variables are likely to be among the most constrained firms, a result we find further
support for in the next section.
In contrast, firms that are debt market constrained are in product markets with
low Q, and these firms have even lower Q relative to their industry peers. These
firms also hold less cash. In all, these firms, unlike equity market constrained firms,
have some resemblance to distressed firms and fallen angels. However, one striking characteristic they have is that their investment levels are above their industry
averages. Like distressed firms, these debt constrained firms are in industries with
few investment opportunities. However, the fact that their own R&D levels exceed
industry levels suggests that these firms might have better than average investment
opportunities relative to their distressed peers.
Finally, Table V also suggests that the size of the CAP+LIQ subsection is informative, although its impact is modest. We include this variable to address concerns
that our text variables are just loading on the size of the CAP+LIQ section. We
attribute the findings for the size of CAP+LIQ variable, as being consistent with the
this variable also potentially being a weak proxy for financial constraints.
25
In all, our results support the conclusion that our delay investment variable measures the degree of constraints, and our equity and debt constraint variables affirm
that both forms of capital play unique roles in financial constraints.
IV
Investment Policy and Financial Constraints
In this section, we examine whether firms with higher levels of financial constraints
react to negative shocks differently than do unconstrained firms. We consider two
types of investment expenditures – CAPX and R&D expenditures – and two forms of
external financing — equity and debt financing. While historically the literature on
constraints has focused on examining financing of CAPX expenditures, the financing
of R&D is more likely to be subject to the constraints, both because R&D is less
tangible and less certain, and because R&D focused firms are often small and young
and lack a well established track record.12
We consider two shock variables as independent variables in the policy regressions.
The first is the aggregate sum of the three text-based shocks (high competition, low
demand, and high cost). Our second is the 2008-2009 financial crisis dummy. Our
shock framework parallels Opler and Titman (1994), who examine the effect of a
macro-economic shock on distressed firms. Later in this section, we also consider an
exogenous equity market liquidity shock, forced mutual fund selling in response to
redemptions, as in Edmans, Goldstein and Jiang (2011). For ease of comparison, we
scale all shock variables to be positive when the shock arrives (mutual fund selling
is scaled to be negative in Edmans, Goldstein and Jiang (2011)).
A central hypothesis in the literature is that constrained firms should more aggressively scale back issuance and investment activities following negative shocks, as
their constraints likely become more binding in times of hardship. Table VI we show
the economic magnitudes regarding the degree of policy curtailment experienced by
firms in various groups when they experience each of the three shocks discussed
above. For the financial crisis, the value after the shock is the firm’s value of the
12
Brown, Fazzari and Petersen (2009) discuss the factors that make financing of R&D expenditures more susceptible to constraints.
26
policy in 2009, and the pre-shock value is the firm’s value of the given policy in 2007.
For the text shock, the value for shocked firms is the average policy for firms experiencing at least one of the three text shocks (low demand, high competition, or high
cost) and the value for non-shocked firms is the average policy for firms experiencing
none of the three text shocks. For forced mutual fund selling, the value shocked
firms is the average policy for firms with above median forced selling and the value
for non-shocked firms is the average policy for firms that have below median mutual
fund selling. Because each of the policy variables is scaled by sales or assets, the
reported mean changes are in fractional units of sales or assets, respectively.
[Insert Table VI Here]
The table shows that constrained and unconstrained firms react differentially to
shocks, and that these differences are economically large. For example, the highest
tercile delay constrained firms curtailed R&D by 11.9% of sales in the financial
crisis of 2008-2009, whereas the lowest tercile firms reduced R&D by only 1.4%
of sales. This inter-tercile range of 10.5% is an economically large 16.1% of one
standard deviation (the cross sectional standard deviation of R&D/sales is 65%).13 .
On the issuance side, high delay constrained firms curtailed equity issuance by 2.6% of
assets during the financial crisis, and low delay constrained firms by just 1.0%. This
inter-tercile difference of 1.6% is an economically meaningful 11.0% of one standard
deviation. For equity focus delay, this inter-tercile range is a larger 2.4% of assets,
and is 1.8% for private placement focused firms. The difference in equity issuance
response to forced mutual fund selling are equally material.
A
Panel Regressions
We now examine the link between constraints and financial policies using a panel
data approach. We first consider firm fixed effect regressions. We then consider the
Arrellano-Bond Differences GMM estimator that allows for consistent estimation of
models with lagged dependent variables. We consider several tests from the existing
13
This estimate is conservative because economic magnitudes are even larger if we consider the
subsample of firms with strictly positive R&D. This result is also not influenced by outliers as we
winsorize R&D/sales at the 99% level.
27
literature to assess the validity of the Arrellano-Bond method in our setting.
In Table VII, we examine four firm policies: R&D/sales, CAPX/sales, equity
issuance/assets and debt issuance/assets in year t as a function of year t − 1 characteristics, year t − 1 constraint measures, and year t − 1 shocks. For each policy we
estimate a simple model that predicts the firm’s policy as a function of the the shock
that the firm experiences and several control variables. Control variables include log
firm age, log Assets, Tobin’s Q and log document size. The independent variables of
interest are our text-based shock variables (high competition, low demand, technological change, and high cost), and the 2008-2009 financial crisis. Our tests focus on
the cross terms between our constraint variables and the shock variables. For both
shocks, we expect to observe a negative and significant cross term coefficient indicating that a firm that scores high on our constraint indices curtails a given investment
or issuance policy more than a less constrained firm.14 The equations are estimated
using OLS firm fixed effect regressions with year fixed effects. Standard errors are
adjusted for clustering at the firm level.
[Insert Table VII Here]
Panel A of Table VII shows that firms with high delay investment scores significantly scale back both R&D and CAPX investments following negative shocks.
Many of the coefficients are significant at the 5% or 1% level. Regarding issuance
policy, we also find that delay constrained firms also curtail equity issuance in Panel
A. Overall, our results support the conclusion that our delay constraint variable is
associated with treated firms reacting more negatively to negative shocks, consistent
with a constraint interpretation. We view this evidence as suggestive, and we defer
a discussion of causality until the next section.
Additional results in Panel A are also in line with expectations. Firms with higher
Tobin’s Q invest more in both R&D and CAPX. They also issue more stock. Finally,
the delay constraint variable level coefficient is positively related to investment policy
and issuance policy. This finding is consistent with the conclusion that firms seeking
14
We do not have direct predictions regarding the coefficients for the observed levels of the shock
or constraint variables since deriving such a relation requires further assumptions about equilibrium
outcomes.
28
very aggressive investment strategies might not be able to issue securities at a rate
that would facilitate full funding of all their opportunities.
Panels B to E of Table VII repeat this test for four additional constraint variables:
marginal equity focused delay and marginal debt focused delay, the KZ index, and
the WW index. The panel only shows the shock cross terms (the other variables are
included in the regressions but not displayed to conserve space).
Regarding the marginal equity and debt focused delay constraints in Panels B
and C, we remind the reader that these are marginal variables. Their construction
using stepwise regressions ensures that each variable describes how an equity or
debt focused constrained firm is different from a regular constrained firm. To infer
the total effect of being equity focus constrained, the reader should combine the
effects of the delay constraint variable in Panel A and the marginal equity focused
constraint variable in Panel B. Panel B thus shows that firms with high marginal
equity focused delay scores curtail investment and equity issuance more than baseline
delay constrained firms in Panel A. Because the delay investment variable and the
marginal equity focus delay variable correlate little (see Table II), there exist many
firms that score high on both and are thus especially sensitive to shocks.
Panel C shows that debt focused delay firms are different. These firms curtail
R&D investment and equity issuance less than standard delay constrained firms.
Because this variable is marginal, it is important to note that the positive coefficients
do not imply that these firms actually increase investment during shocks, but rather,
we conclude that the effects of constraints are mitigated for these firms relative to
other constrained firms. Thus, debt focused firms have more flexibility than equity
focused firms. For example, we find that equity issuance is positively related to the
financial crisis variable, indicating that debt focused firms tend to substitute away
from debt and toward equity when negative shocks arrive. This substitution allows
these firms to avoid curtailing investment.
The results in Panel D do not support a constraint based interpretation of the
KZ Index. Firms with a high KZ index increase R&D and CAPX investment in
the financial crisis. For the WW Index, Panel E provides mixed results. There is
29
no support for a constraint based interpretation for the text shock, but the results
for the financial crisis are supportive. Firms with a high WW index curtail R&D,
CAPX, and equity issuance during the crisis. However, these results are not as strong
as those are for the text-based constraint measures, and results presented in the next
section suggest that the WW index is less robust to controls for endogeneity.
We also perform unreported placebo tests as in Duchin, Ozbas, and Sensoy (2010).
In particular, we run analogous specifications to those in this section, except that
we replace the financial crisis dummy (indicating the years 2008 and 2009) with a
placebo crisis (2006 and 2007) from a nearby time interval. We find in this test
that all of our negative and significant cross term results for all policy regressions
entirely vanish. Moreover, all of the financial crisis cross terms are insignificant,
with the exception of equity issuance, which is 10% level positive and significant in
some tests (note that in the real financial crisis that this cross term is negative as
predicted. This may be consistent with constrained firms taking advantage of the
strong economy in 2006 and 2007, and viewing this as a unique opportunity to issue
abnormally high levels of equity.
B
Alternative Econometric Specification
The use of fixed effect panel regressions is standard in many studies. However, it
is likely that the process driving corporate investment policies has both a firm fixed
effect and an autoregressive term. In this case, such panel regressions will be biased.
Disentangling the impact of firm fixed effects and the lagged dependent variable is
especially critical in our setting because our panel is short in time duration at 13
years even though we have a large number of firms in cross section (See Roodman
(2009)). To address this issue we re-estimate our model using the Arrellano-Bond
estimator, which eliminates this bias, while also addressing potential endogeneity
concerns by using lagged variables as instruments.
The Arrellano-Bond estimator allows for added flexibility to specify which variables are endogenous, pre-determined, and truly exogenous. The quality of these
designations can then be examined using standard tests. These tests allow us to
30
evaluate whether the relationships we find can be linked to a causal interpretation
in the sense that the variables of interest are independent of the error term.
To take full advantage of the Arrellano-Bond estimators, we follow Roodman’s
(2009) recipe. We first designate three variables as exogenous: firm age, firm size,
and the 2008-2009 crisis dummy. Second, we designate financial constraint variables
as potentially endogenous, and we use an additional lag for these variables. We
classify all other variables such as Tobin’s Q as being predetermined by past data,
but not fully exogenous. Our specification is conservative because we use differences
GMM instead of system GMM, and we avoid the orthogonal deviations transform.
Both choices avoid the need for additional assumptions. We also include year fixed
effects and we adjust standard errors for clustering by firm.
We assess the stability of our regressions in five ways. First, we test if the
idiosyncratic disturbance is autocorrelated at the second lag following Arrellano and
Bond (1991). If second order autocorrelation is significant, the instruments may
not be valid. Second, we examine whether the coefficient for the lagged dependent
variable falls within the two reasonable OLS bounds indicated in Bond (2002) and
illustrated by example in Bazzi and Clemens (2009). Third, we examine the Hansen
J-statistic of overidentifying restrictions. A significant J-statistic indicates improper
instrumentation for endogeneity. Fourth, we consider the Hansen Differences statistic
based on our two shock cross terms we draw inferences from. A significant differences
statistic indicates improper instrumentation for endogeneity directly relating to these
key variables. We report the results of the four tests discussed here in a row titled
“regression diagnostics”, where we indicate each test is passed using “a”, “b”, “c”,
and “d”, respectively. Finally, we note that the number of instruments used by the
Arrellano-Bond model outnumber groups by a factor of roughly five to ten across
our specifications, far exceeding the minimum 1:1 ratio below which Roodman (2009)
indicates that problems are most likely. Hence, it is unlikely that our results suffer
from problems relating to too many instruments.
Table VIII displays the results using the Arrellano and Bond (1991) model. The
lagged dependent variable in Panel A is significant in each case indicating that investment policies do contain a relevant autoregressive component, although this com31
ponent is smaller for equity issues which are known to be more lumpy. With one
exception, the signs of the cross-term coefficients are negative and thus consistent
with those in in Table VII. As expected, when instrumental variables are used, significance levels are somewhat lower, although our results are highly robust.
[Insert Table VIII Here]
For investment policies, delay constrained firms are significantly more likely to
curtail R&D and CAPX policies following shocks. These results are significant at
the 1% or 5% level, and the critical R&D specification passes all four regression
diagnostic tests and thus is consistent with a causal conclusion. Panels B and C
show that firms that are marginally equity focused constrained curtail both forms
of investment more in the financial crisis, and firms that are debt focus constrained
tend to curtail policies less. We find notably weaker results for either the KZ index
or the WW index for investment policies in Panels D and E respectively, suggesting
that a similar conclusion is not warranted for these variables.
The results are less conclusive for issuance policies. The Hansen J test (denoted by
the letter “c” in the regression diagnostics) uniformly suggests that the instruments
are not sufficiently strong to draw a causal conclusion for the model as a whole.
However, the models do uniformly pass the critical Hansen differences test (denoted
by the letter “d”), which examines the issue of exogeneity for the two shock cross
terms alone. Overall, the results are least conclusive for the debt issuance model,
which passes only two of the four tests. However, this is somewhat immaterial given
that the key variables are not significant in this model.
C
Subsample Results
We next explore key subsamples based on above or below median size, age, R&D
focus, and CAPX focus. Firms are sorted into these subsamples based on their value
of the given characteristic relative to the median in the year when the given firm
enters the sample. This ensures time-series continuity for each firm, an important
consideration given the nature of the Arrellano-Bond model and its use of lags.
HP find that financial constraints are most relevant for smaller and younger firms.
32
Panels A and B of Table IX display our results for these subsamples, and Panels C
and D display results for young and old firms. We find that results for both small
and young firms are stronger than those for larger and older firms. These results are
quite strong as significance is maintained despite the reduced sample size.
Our results extend those of HP along three dimensions. First, we find that
only some small and young firms are constrained. We conclude that (A) financial
constraints are best measured using textual measures and (B) the sample of small
and young firms is where the most relevant variation is between constrained and
unconstrained firms (this differs quite materially from HP’s conclusion that size and
age are the best proxies for constrained status). Second, we do find that constraints
have some impact on larger and older firms as they significantly curtail CAPX policy
when shocks arrive. Third, we show that marginal equity and debt focus constraints
also matter in a unique way.
[Insert Table IX Here]
Panels E and F of Table IX display analogous results for low and high R&D
Focused firms. Results for these subsamples are especially striking. We find no
significant results in Panel E for low R&D focused firms, and we find strong results
in Panel F for high R&D firms. Many coefficient estimates in Panel F are significant
at the 5% level or better. Panels G and H of Table IX display analogous results for
low and high CAPX Focused firms, and results not surprisingly are strongest for low
CAPX firms. Overall we conclude that investment type is important in determining
the extent to which financial constraints are binding. Investors also appear to be less
willing to provide liquidity when investments are informationally sensitive (R&D),
and more willing to provide funding when investments are tangible (CAPX).
D
Equity and Debt Focused Constraints
In this section, we further consider the hypothesis discussed earlier in this section
that debt focused constrained firms have more flexibility, and respond to shocks by
substituting away from debt issuance and toward equity issuance. To sharpen our
earlier tests, we further propose that equity focused constrained firms facing the most
33
difficulties are most likely to be those that invest in R&D projects, whereas debtfocused constrained firms are most likely to be those investing in CAPX projects.
[Insert Table X Here]
Table X considers the equity and debt focused constraint variables for the high
R&D focused subsample (Panels A and B) and the high CAPX focused subsample
(Panels C and D). The results in Panel A show that firms with high equity focused
delay constraints curtail both equity issuance and R&D spending significantly more
than other firms during the crisis years. To examine debt focused constrained firms,
we first consider Panel D given the earlier discussed links between CAPX and debt
issuance. Panel D shows that debt focused constrained firms curtail debt issuance
and increase equity issuance (and hence do not have to curtail investment) during the
financial crisis. Regarding the intermediate panels, Panel C (like Panel A) shows that
equity focused constrained firms also curtail equity issuance and investment when
they are CAPX focused. Panel B (like Panel D) shows that debt focused constrained
firms have more flexibility and do not curtail investment materially. Overall, our
results indicate that equity focused constrained firms have little flexibility, and thus
curtail both equity issuance and investment. In contrast, debt focused constrained
firms do have flexibility, and substitute toward equity issuance following shocks. As
a result, these firms face little pressure to curtail investment.
E
Private Placements
In this section, we explore the role of marginal private placement focus. As discussed
earlier, this variable is marginal, and the stepwise textual regression ensures that we
are examining the role of private placement mentions in the text, holding fixed the
degree to which firms are delay investment constrained, and also the degree to which
firms are equity or debt focused constrained. This test is motivated by the finding in
Gomes and Phillips (2011) that private placements are more likely to be used when
asymmetric information is particularly high.
[Insert Table XI Here]
34
Table XI displays the results for our private placement focused variable for the
overall sample (Panel A) and for various subsamples in later panels. The findings are
consistent with a role for asymmetric information. Firms that are private placement
focused are significantly more likely to curtail R&D and equity issuance following
negative shocks (R&D and equity are often viewed as more informationally sensitive
than CAPX and debt). As was the case for our earlier analysis, these findings are
also especially strong for the high R&D focused subsample, for younger firms, and for
smaller firms. Many of these results also perform well on the regression diagnostics,
and thus are consistent with a causal relationship between these variables.
F
Shocks to the market for the firm’s equity
In this section, we add forced mutual fund selling, and corresponding cross terms
with constraint indices, to the specifications in Table VIII. We hypothesize, consistent
with Edmans, Goldstein and Jiang (2011), that this variable represents an exogenous
shock to a firm’s equity-market liquidity. Unlike the 2008 crisis and the text shock,
which likely have both supply and demand side effects, as in Opler and Titman
(1994), the forced mutual fund shock variable is uniquely a supply side shock. Our
hypothesis is that shocks to this variable should result in greater curtailment of
equity issuance for constrained firms relative to unconstrained firms.
[Insert Table XII Here]
Panel A of Table XII shows that the mutual fund selling shock does induce firms
to curtail equity issuance policy overall. However, the lack of significance for the
cross term indicates that firms with higher loadings on the delay focus constraint
do not curtail their issuance differentially. In contrast, Panels B and D show that
equity focus constrained firms, and private placement focus constrained firms do
curtail equity issuance more when forced mutual fund selling is high. We do not
find significant results for the marginal debt focus constraint, nor the KZ Index or
the WW Index. These results are particularly supportive of a causal link between
financial constraints and policy curtailment through the channel of equity market
friction. In particular, we would expect this supply side shock to matter most for
35
firms that are indeed equity focus constrained, and for firms that are considering
private placements of equity on the margin. The fact that only these cross terms are
significant, and only for the equity issuance policy and not debt issuance policy, is
especially stark in supporting our key hypotheses regarding financial constraints.
These results clarify that our constraint measures relate to real issues of financial
liquidity. Going further, three key findings help to rule out a possible concern that
our constraint measures are picking up disclosures by managers with low expected
value projects, who may have incentives to blame the markets for their failure to
invest. First, as shown in Table V, firms that report high delay investment and
equity delay constraints have high Tobin’s Q, and thus are likely to have good investment opportunities. Second, in unreported tests, we also find no evidence that
high constraints predict low equity returns, indicating that the high Tobin’s Q of
these firms indeed reflects their fundamentals. Third, the results in this section for
the exogenous mutual fund selling shock uniquely imply a liquidity channel for our
results. All three of these results run counter to the predictions of the alternative
hypothesis. In contrast, all of our paper’s findings are consistent with the theory
presented in Krasker (1986)’s extension to Myers and Majluf (1984). In particular,
informational asymmetries can create equity issuance cascades in which firms with
the best investment opportunities cannot issue sufficient equity to finance all of their
high quality investments.
V
Conclusion
We develop a methodology for identifying financial constraints using mandated disclosures in firm 10-K statements. We focus on discussions in which management
summarizes their firm’s ability or inability to obtain financing for planned investments, and the marginal form of external financing (debt, equity, or private placements of equity) their firm is actively considering. We use this text to construct
indices of delay investment financial constraints, and financial constraints with a
debt-, equity-, or private placement- financing focus. These disclosures are required
by law, and contain a great deal of information concerning these issues. Our indices
36
allow us to directly investigate financial constraints without having to rely on the
assumptions of a specific model. Our text-based measures also have advantages over
surveys: they are available for the entire COMPUSTAT panel from 1997 forward,
are in response to time-invariant regulation, and we do not have to extend our measures from a small survey to the whole population using statistical models based on
potentially endogenous variables.
We find that text-based measures of constraints and financing focus correlate
with firm characteristics in an intuitive way. For example, equity focused constrained
firms are in industries with high investments, higher Tobin’s Q, lower profits, lower
leverage, and higher cash. Hence, they are in industries where planned investment
levels and opportunities are high, and many rivals are cash rich. However, the relative
position of constrained firms within these industries is such that they have lower than
average investment levels (R&D and CAPX) despite having an even higher average
Tobin’s Q than their peers. These firms are focused more on R&D than CAPX, and
these results are consistent with constraints being most severe for firms seeking the
most aggressive and informationally-opaque investments. These findings also clearly
illustrate that financial constraints and financial distress are quite distinct.
Firms that are debt-focused constrained are diametrically opposite, and are from
industries with lower Tobin’s Q and lower investment. In terms of position, constrained firms in these industries have even lower Tobins’ Q than the already-low
levels of their industry peers, and higher leverage than the already high levels of
their peers. Despite this, their investment levels are higher than their peers. Hence,
these firms likely face constraints because lenders may see them as over-leveraged and
overly risky due to their aggressive investment strategies. Interestingly, constraints
are less binding for these firms due to their ability to substitute to equity following
negative shocks, allowing less interruption to their investment levels.
Firms that score high on our constraint measures curtail R&D, capital expenditures, equity issuance and debt issuance more than less constrained firms when they
receive negative shocks. We also find that these curtailments are larger for firms
that are equity-focus constrained and private placement-focus constrained relative
to those that are debt-focus constrained. Because the market perceives these equity
37
focused constrained firms as having good investment opportunities (high Tobin’s
Q), these results strongly affirm the predictions of Krasker (1986) regarding equity
rationing (a form of financial constraints due to information asymmetry).
Our constraint measures outperform measures used in the literature, and our
results are economically large. Firms in the most delay constrained tercile curtail
R&D/Sales by 11 percentage points more than the least constrained firms during
the financial crisis. Consistent with illiquidity and informational asymmetry driving
financial constraints, our results are strongest for small firms, young firms, firms
focused on R&D. Moreover, when we consider an exogenous shock to equity market
liquidity, we find strong results for equity-focused and private placement focused
constrained firms. These results point to a causal link between financial constraints
and policy curtailment through the channel of equity market friction.
Taken together, these findings suggest that researchers need to consider equity
and debt market financing constraints separately when estimating the shadow price
of external capital. One-dimensional measures of financial constraints, as used in
the existing literature, miss this important level of detail. The degree to which constraints bind also depends on the firm’s situation, its preferred type of capital, and
the type of investment it undertakes. Constraints are most binding when the constraint is based on the inability to issue equity. Among investment classes, research
and development is far more prone to constraints relative to capital expenditures,
which are more tangible. Our new constraint indices should have other applications
and we plan to make them available on the web.
38
Antweiler, Werner; Frank, Murray; 2004, Is all that talk just noise? The information content of internet stock message boards, Journal of Finance 52, pp. 1259-1294
Arellano, M.; Bond, S.; 1991, Some tests of specification for panel data: Monte
Carlo evidence and an application to employment equations. Review of Economic
Studies 58, pp. 277-297
Bazzi, Samuel; Clemens, Michael; 2010, Blunt Instruments: A Cautionary Note
on Establishing the Causes of Economic Growth. Center for Global Development
Working Paper
Beck, Thorsten; Demirguc-Kunt, Asli; Maksimovic, Vojislav; 2005, Financial and
Legal Constraints to Growth: Does Firm Size Matter? Journal of Finance 60, iss. 1,
pp. 137-77
Bond, Stephen; 2002, Dynamic panel data models: a guide to micro data methods
and practice. Portuguese Economic Journal 1, iss. 2, pp. 141-162.
Brown, James R.; Fazzari Steven M.; Petersen Bruce C.: 2009, Financing Innovation and Growth: Cash Flow, External Equity, and the 1990s R&D Boom, Journal
of Finance 64, 151-185.
Campello, Murillo; Graham, John R.; Harvey, Campbell R.; 2010, The Real Effects of Financial Constraints: Evidence from a Financial Crisis, Journal of Financial
Economics 97, iss. 3, pp. 470-87.
Cimiano, Phillip; 2010, Ontology Learning and Population from Text: Algorithms, Evaluation and Applications, Springer, New York.
Duchin, Ran; Ozbas, Oguzhan; Sensoy, Berk; 2010, Costly External Finance, Corporate Investment, and the Subprime Mortgage Credit Crisis, Journal of Financial
Economics 97, 418-435.
Edmans, Alex; Goldstein, Itay; and Jiang, Wei, 2011, The Real Effects of Financial Markets: The Impact of Prices on Takeovers, Journal of Finance, Forthcoming.
Fazzari, Steven M., R. Glenn Hubbard, and Bruce P. Petersen, 1988, Financing
Constraints and Investment, Brookings Papers on Economic Activity 1, 141-195.
39
Gomes, Armando; Phillips, Gordon, 2011, Why Do Public Firms Issue Private
and Public Securities? University of Maryland Working Paper
Graham, John R.; Harvey, Campbell R.; 2001, The Theory and Practice of Corporate Finance: Evidence from the Field, Journal of Financial Economics 60, iss.
2-3, pp. 187-243
Greenwald, Bruce; Stiglitz, Joseph; Weiss, Andrew; 1984, Informational imperfections in the capital carket and macroeconomic fluctuations, American Economic
Review Papers and Proceedings 74, pp. 194-199
Hadlock, Charles J.; Pierce, Joshua R.; 2010, New Evidence on Measuring Financial Constraints: Moving beyond the KZ Index, Review of Financial Studies 23,
pp. 1909-40.
Hanley, Kathleen Weiss; Hoberg, Gerard; 2010, The Information Content of IPO
Prospectuses, Review of Financial Studies 23, iss. 7, pp. 2821-641.
Hart, Oliver; Moore, John; 1998, Default And Renegotiation: A Dynamic Model
Of Debt, Quarterly Journal of Economics 113, pp. 1-41
Hoberg, Gerard; Phillips, Gordon; 2010, Product Market Synergies and Competition in Mergers and Acquisitions: A Text-Based Analysis, Review of Financial
Studies 10, pp. 3773-3811
Hoberg, Gerard; Phillips, Gordon; 2010, Dynamic Text-Based Industry Classifications and Endogenous Product Differentiation, National Bureau of Economic
Research, Inc, NBER Working Papers: 15991
Jensen, Michael; 1986, Agency Costs of Free Cash Flow, Corporate Finance and
Takeovers, American Economic Review 76, pp. 323-329
Jensen, Michael; Meckling, William; 1976, Theory of the Firm: Managerial Behaviour, Agency Costs and Ownership Structure, Journal of Financial Economics 3,
pp. 305-360
Kaplan, Steven; Zingales, Luigi; 1997, Do Investment-Cash Flow Sensitivities
Provide Useful Measures of Financial Constraints, Quarterly Journal of Economics
112, 159-216.
40
Krasker, William; 1986, Stock Price Movements in Response to Stock Issues
Under Asymmetric Information, Journal of Finance 41, pp. 93-105.
Lamont, Owen; Polk, Christopher; Saa-Requejo, Jesus; 2001, Financial Constraints and Stock Returns, Review of Financial Studies 14, iss. 2, pp. 529-54
Loughran, Tim; McDonald, Bill; 2011, When is a liability not a liability? Textual
analysis, dictionaries, and 10-Ks, Journal of Finance 66, pp. 35-65.
Myers, Stewart; 1977, The determinants of corporate borrowing, Journal of Financial Economics 5, pp. 147-175
Myers, Stewart C.; Majluf, Nicholas S.; 1984, Corporate Financing and Investment Decisions When Firms Have Information That Investors Do Not Have Journal
of Financial Economics 13, iss. 2, pp. 187-221
Opler, Tim C.; Titman Sheridan: 1994, Financial Distress and Corporate Performance, Journal of Finance 49, 1015-1040.
Phillips, Gordon M.; Zhadanov, Alexei; 2011, R&D and the Incentives from
Merger and Acquisition Activity, University of Maryland Working Paper
Roodman, David; 2009, How to do xtabond2: An introduction to difference and
system GMM in Stata, Stata Journal 9, iss. 1, pp. 86-136
Sebastiani, Fabrizio; 2002, Machine Learning in Automated Text Categorization,
ACM Computing Surveys, iss. 34, pp. 1-47
Tetlock, Paul; 2007, Giving content to investor sentiment: The role of media in
the stock market, Journal of Finance 62, pp. 1139-1168
Whited, Toni M.; 1992, Debt, Liquidity Constraints, and Corporate Investment:
Evidence from Panel Data, Journal of Finance 47, iss. 4, pp. 1425-60
Whited, Toni and Guojun Wu, 2006, ”Financial Constraints Risk,” Review of
Financial Studies 19-2, 531-559.
41
Table I: Summary Statistics
Summary statistics for firms with machine readable 10-Ks, and having machine readable Management’s Discussion
and Analysis. Observations are from 1997 to 2009. We discard all financials, and require COMPUSTAT data
coverage in the current year and the past year. We also discard firms with zero assets or zero sales, and require
valid text data to be available. Please see Section II for a description of variables. All non-binary variables are
winsorized at the 1/99% level.
Variable
Mean
Std.
Dev.
Min.
Median
Max.
# Obs.
1.000
1.000
0.915
48,512
48,512
48,512
Panel A: MD&A Content (Firms with MD&A Section)
Has Cap+Liq Subsection Dummy
MD&A R&D Focused Score
MD&A CAPX Focused Score
0.807
0.083
0.018
0.395
0.100
0.043
0.000
0.000
0.000
1.000
0.051
0.000
Panel B: Precise Text Constraint Declarations (Firms with Cap+Liq Section)
Delay Investment Text
Equity Focused Text
Debt Focused Text
0.017
0.010
0.034
0.128
0.100
0.182
0.000
0.000
0.000
0.000
0.000
0.000
1.000
1.000
1.000
38,980
38,980
38,980
Panel C: 12-Word Text Constraint Declarations (Firms with Cap+Liq Section)
Delay Investment Text
Equity Focused Text
Debt Focused Text
Private Placement Foc. Text
0.055
0.128
0.146
0.094
0.229
0.335
0.353
0.291
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
1.000
1.000
1.000
1.000
38,980
38,980
38,980
38,980
Panel D: Text Constraint Scores (Firms with Cap+Liq Section)
Delay Investment Score
Equity Focused Score
Debt Focused Score
Private Placement Foc. Score
-0.007
-0.012
-0.005
-0.027
0.111
0.067
0.089
0.069
-0.384
-0.194
-0.279
-0.231
-0.008
-0.018
-0.011
-0.025
0.348
0.262
0.269
0.191
38,980
38,980
38,980
38,980
Panel E: Other Text Variables (Firms with Cap+Liq Section)
Covenant Violation Score
Log # Words in Cap+Liq Section
0.037
8.306
0.105
0.848
-0.701
3.951
0.030
8.409
0.541
11.315
38,980
38,980
Panel F: Text-Based Shock Variables (Firms having valid MD&A)
High Competition Dummy
Low Demand Dummy
High Cost Dummy
0.136
0.118
0.134
0.343
0.322
0.341
0.000
0.000
0.000
0.000
0.000
0.000
1.000
1.000
1.000
48,512
48,512
48,512
0.002
0.037
0.006
0.000
10.359
6.983
1.207
1.500
48,512
48,512
48,512
48,512
Panel G: Corporate Policy Variables
R&D/Sales
CAPX/Sales
Equity Issuance/Assets
Debt Issuance/Assets
0.181
0.106
0.052
0.101
0.763
0.276
0.138
0.211
0.000
-0.000
-0.000
0.000
42
43
Equity Focused Delay Score
Debt Focused Delay Score
Private Place. Foc. Delay Score
Covenant Violation Score
Log Assets
KZ Index
WW Index
MD&A R&D Focused Score
MD&A CAPX Focused Score
Row Variable
-0.019
0.043
0.088
-0.153
-0.046
0.039
0.082
0.150
0.019
Delay
Invest.
Score
-0.089
0.038
-0.130
-0.106
-0.119
0.126
0.153
-0.058
Equity
Focused
Score
Private
Placement
Focused
Score
-0.244
0.605
0.137
0.153
-0.136
-0.206
0.057
-0.484
0.015
-0.116
0.014
0.127
-0.018
Correlation Coefficients
Debt
Focused
Score
0.052
0.171
-0.077
-0.225
0.048
Covenant
Viol.
Score
0.076
-0.913
-0.213
0.157
Size
(Log
Assets)
-0.012
-0.150
0.068
KZ
Index
0.262
-0.154
WW
Index
-0.019
MD&A
R&D
Focused
Score
The table displays Pearson Correlation Coefficients between various measures of financial constraints and financial distress. The first six variables are text-based measures of
investment opportunities (the first three being derived from the entire MD&A, and the latter three derived from the Capitalization and Liquidity subsection (“CAP+LIQ”). We also
include our three constraint variables based on investment delay, equity focused delay, and debt focused delay. Finally, we include the covenant violation score. In addition to the text
variables, we include correlations with firm size, the KZ index, the WW index, and the Altman Z-score. Please see Section II for a description of variables. Finally, we consider firm
size and three existing measures of financial constraints and distress used in the existing literature: the KZ index, the WW index, and the Altman Z score.
Table II: Pearson Correlation Coefficients (Investment and Constraint Measures)
44
Precise
Delay
Dummy
1.502
1.658
1.558
1.543
1.768
1.752
1.954
1.636
1.303
1.068
1.574
2.513**
2.273
1.246
1.461
1.019
1.391
1.138
1.274
1.752
1.183
0.986
0.736
1.349
2.313**
1.768
1.757
1.855
2.098
1.694
2.397*
2.229
2.156
2.089
1.620
1.399
1.799
2.713**
2.778
Year
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
6.122
6.689
6.415
6.715
7.789
7.389
5.930
6.198
6.620
6.186
6.597
9.258**
9.091
4.646
5.337
4.194
5.203
4.194
3.949
3.571
3.062
3.380
2.946
3.298
5.263**
4.377
5.384
6.013
5.304
5.959
5.992
5.669
4.751
4.629
5.000
4.566
4.948
7.260***
6.734
12-Win.
Delay
Dummy
0.057
0.056
0.000
0.181**
0.240
0.191
0.135
0.139
0.070
0.000
0.075
0.160
0.000**
0.057
0.112
0.000**
0.000
0.060
0.064
0.067
0.139
0.000***
0.000
0.000
0.000
0.000
0.057
0.084
0.000
0.091*
0.150
0.127
0.101
0.139
0.035**
0.000
0.037
0.080
0.000
Precise
Delay x
Equity
Dummy
2.098
2.361
2.098
2.117
2.876
2.739
2.022
2.019
1.972
1.841
2.324
3.751**
3.367
1.700
1.404
1.138
1.754
0.779**
0.955
0.539
0.696
0.493
0.515
0.675
1.037
0.926
1.899
1.883
1.618
1.936
1.827
1.847
1.280
1.357
1.232
1.178
1.499
2.393**
2.146
12-Win.
Delay x
Equity
Dummy
0.170
0.112
0.180
0.060*
0.060
0.000
0.000
0.000
0.000
0.000
0.000
0.080
0.084
% MDA
R&D
Focus
1.162
1.124
1.259
1.089
0.929
0.828
0.640
0.627
0.951
1.068
1.012
1.237
1.010
0.108
0.102
0.103
0.103
0.099
0.093
0.094
0.093
0.089
0.083
0.083
0.077
0.074
Panel A: All Firms
12-Win.
Delay x
Debt
Dummy
0.085
0.078
0.080
0.088
0.084
0.079
0.080
0.077
0.070
0.065
0.064
0.058
0.057
1.020
1.012
1.259
1.028
1.078
0.764*
0.741
0.696
1.197***
1.252
1.124
1.038
1.094
0.131
0.126
0.127
0.119
0.113
0.107
0.107
0.109
0.108
0.101
0.102
0.096
0.091
Panel C: Small Firms
1.303
1.236
1.258
1.149
0.779
0.892
0.539
0.557
0.704
0.884
0.900
1.435*
0.926*
Panel B: Large Firms
0.227
0.112
0.120
0.121
0.000
0.064
0.202*
0.000***
0.000
0.000
0.150*
0.159
0.084
0.198
0.112
0.150
0.091
0.030
0.032
0.101
0.000
0.000
0.000
0.075
0.120
0.084
Precise
Delay x
Debt
Dummy
0.013
0.012
0.011
0.013
0.013
0.015
0.015
0.014
0.015
0.014
0.015
0.016
0.017
0.024
0.021
0.021
0.022
0.023
0.023
0.026
0.026
0.028
0.029
0.027
0.028
0.030
0.019
0.016
0.016
0.017
0.018
0.019
0.021
0.020
0.021
0.021
0.021
0.022
0.023
% MDA
CAPX
Focus
0.131
0.124
0.119
0.143
0.161
0.173
0.203
0.201
0.125**
0.107
0.102
0.087
0.088
0.144
0.161
0.164
0.160
0.183
0.200
0.223
0.267
0.194*
0.147
0.145
0.130
0.123
0.137
0.142
0.141
0.152
0.172
0.187
0.213
0.234
0.159**
0.127
0.123
0.108
0.105
High
Compet.
Dummy
0.053
0.069
0.074
0.075
0.120
0.147
0.152
0.146
0.099
0.082
0.083
0.123
0.166
0.080
0.094
0.107
0.120
0.187
0.222
0.191
0.174
0.151
0.148
0.142
0.237
0.324
0.067
0.082
0.090
0.098
0.153
0.184
0.171
0.160
0.125
0.115
0.113
0.180
0.245
Low
Demand
Dummy
0.104
0.107
0.094
0.099
0.135
0.148
0.138
0.159
0.169
0.153
0.142
0.152
0.131
0.126
0.117
0.129
0.123
0.146
0.176
0.188
0.201
0.221
0.224
0.215
0.234
0.229
0.115
0.112
0.111
0.111
0.141
0.162
0.163
0.180
0.195
0.189
0.179
0.193
0.180
High
Cost
Dummy
1764
1779
1668
1653
1669
1570
1484
1436
1420
1358
1334
1253
1188
1765
1780
1669
1653
1669
1570
1484
1437
1420
1358
1334
1254
1188
3529
3559
3337
3306
3338
3140
2968
2873
2840
2716
2668
2507
2376
#
Obs.
The table displays annual means of various measures of financial constraints, investment focus, and financial shocks. ***, **, and * denote significant changes relative to the previous
year at the 1%, 5%, and 1% level, respectively. All variables are text-based. Please see Section II for a full description of variables.
Table III: Time Series Statistics (All Firms)
45
–0.1091
(–2.270)
–0.1087
(–2.240)
–0.0912
(–1.850)
–0.1099
(–2.190)
(5)
(8)
(7)
(6)
(4)
(3)
–0.5429
(–5.280)
–0.5429
(–5.280)
–0.5307
(–5.160)
–0.4969
(–4.850)
–0.3800
(–11.550)
–0.3717
(–11.300)
–0.2790
(–8.410)
–0.2635
(–8.370)
–0.2618
(–14.790)
–0.2522
(–14.090)
–0.2130
(–11.490)
–0.2077
(–11.280)
(1)
(2)
Log
Age
Log
Row Assets
0.0025
(0.060)
–0.0008
(–0.020)
0.0216
(0.530)
–0.0582
(–1.860)
–0.0408
(–1.370)
0.0675
(2.810)
oi/
Sales
Tobin’s
Q
Book
Leverage
Dividend
Payer
0.5688
(4.000)
0.7584
(5.440)
–0.6739
(–9.960)
–0.5832
(–8.750)
0.0271
(1.510)
0.0106
(0.560)
–0.0375
(–0.200)
0.0956
(0.490)
–0.1449
(–1.420)
–0.1472
(–1.420)
Panel B: Firm and year fixed effects
–0.0110
(–0.840)
–0.0411
(–3.160)
Panel A: SIC-3 industry and year fixed effects
0.0267
(0.110)
–0.0280
(–0.110)
0.0977
(0.370)
0.4279
(2.110)
0.7091
(3.300)
0.3447
(1.610)
Cash
Assets
8.7929
(20.940)
8.5657
(11.080)
MD&A
R&D
Focus
2.2283
(3.160)
4.0806
(3.340)
MD&A
CAPX
Focus
44,750
44,750
44,750
44,750
44,750
44,750
44,750
44,750
#obs
Panel data Logistic regressions with firm and year fixed effects. The sample includes all firm years from 1997 to 2009. The dependent variable is a dummy equal to one when the given
firm’s MD&A has a distinct Capitalization and Liquidity subsection. Standard errors are adjusted for clustering at the firm level. The independent variables include various firm
characteristics. Please see Section II for a full discussion of variable definitions.
Table IV: Capitalization and Liquidity Subsection and Firm Characteristics
46
Dividend Payer (Industry Specific)
Dividend Payer (Firm Specific)
oi/sales (Industry Specific)
oi/sales (Firm Specific)
oi/assets (Industry Specific)
oi/assets (Firm Specific)
(3)
(4)
(5)
(6)
(7)
(8)
(18) R+D/sales (Firm Specific)
(17) R+D/sales (Industry Specific)
(16) CAPX/sales (Firm Specific)
(15) CAPX/sales (Industry Specific)
(14) Market Leverage (Firm Specific)
(13) Market Leverage (Industry Specific)
(12) Book Leverage (Firm Specific)
(11) Book Leverage (Industry Specific)
(10) cash/assets (Firm Specific)
cash/assets (Industry Specific)
Tobin’s Q (Firm Specific)
(2)
(9)
Tobin’s Q (Industry Specific)
(1)
Dependent
Row Variable
0.711
(10.33)
0.099
(0.75)
-0.027
(-1.88)
0.019
(0.68)
-0.191
(-11.65)
-1.076
(-9.80)
-0.110
(-10.86)
-0.173
(-10.29)
0.046
(6.98)
-0.021
(-1.76)
0.005
(0.67)
0.061
(3.54)
-0.011
(-1.77)
0.023
(1.92)
0.224
(13.13)
-0.049
(-2.45)
1.317
(10.46)
-0.499
(-4.43)
Delay
Invest
Score
1.845
(17.80)
1.039
(5.30)
-0.197
(-8.83)
0.153
(3.46)
-0.398
(-16.97)
-1.991
(-12.30)
-0.258
(-17.55)
-0.350
(-12.91)
0.140
(14.96)
0.024
(1.42)
-0.068
(-6.67)
0.074
(2.94)
-0.093
(-10.20)
-0.010
(-0.60)
0.251
(9.80)
-0.057
(-1.82)
2.214
(12.37)
-0.797
(-4.96)
Equity
Focused
Score
-1.106
(-14.08)
-0.789
(-5.88)
0.152
(8.90)
-0.295
(-8.33)
0.177
(11.64)
0.529
(6.93)
0.125
(12.91)
0.007
(0.43)
-0.121
(-17.55)
-0.147
(-12.05)
0.109
(12.67)
0.301
(15.93)
0.110
(14.63)
0.262
(18.06)
-0.086
(-4.83)
0.085
(4.04)
-1.011
(-9.76)
0.348
(3.86)
Debt
Focused
Score
1.551
(15.32)
0.842
(4.56)
-0.200
(-9.10)
0.001
(0.02)
-0.334
(-16.48)
-0.947
(-7.95)
-0.212
(-16.49)
-0.204
(-8.74)
0.137
(15.62)
0.082
(5.29)
-0.097
(-9.37)
-0.069
(-2.96)
-0.112
(-11.84)
-0.100
(-5.57)
0.273
(11.97)
-0.043
(-1.57)
1.035
(7.46)
-0.418
(-3.37)
Private
Placement
Score
0.016
(2.02)
-0.008
(-0.61)
-0.011
(-5.51)
-0.013
(-3.11)
-0.012
(-6.85)
-0.050
(-5.51)
-0.007
(-6.81)
-0.016
(-9.00)
0.004
(5.01)
-0.005
(-3.79)
0.000
(0.12)
0.024
(11.40)
-0.001
(-0.83)
0.017
(10.99)
0.013
(6.75)
-0.003
(-1.35)
0.049
(4.15)
-0.014
(-1.36)
Log #
Words in
CAP+LIQ
-0.094
(-10.57)
-0.028
(-1.92)
0.030
(11.78)
0.067
(13.92)
0.020
(10.17)
0.077
(8.49)
0.015
(12.36)
0.004
(2.21)
-0.010
(-11.80)
-0.001
(-0.82)
0.004
(3.48)
-0.009
(-3.82)
0.005
(5.27)
-0.009
(-4.94)
-0.020
(-9.48)
-0.007
(-2.60)
-0.085
(-6.38)
0.029
(2.83)
Log
Firm
Age
-0.020
(-3.95)
-0.038
(-3.84)
0.013
(10.14)
0.045
(17.26)
0.010
(9.34)
0.085
(16.37)
0.007
(10.39)
0.040
(28.73)
-0.004
(-8.03)
-0.011
(-14.47)
0.005
(8.75)
0.013
(10.22)
0.004
(7.60)
0.009
(10.25)
-0.002
(-1.89)
0.010
(7.37)
-0.037
(-4.95)
0.016
(2.48)
Log
Assets
35,312
0.404
35,312
0.027
35,312
0.397
35,312
0.109
35,312
0.507
35,312
0.124
35,312
0.488
35,312
0.155
35,312
0.516
35,312
0.049
35,312
0.480
35,312
0.061
35,312
0.497
35,312
0.070
35,312
0.476
35,312
0.101
35,312
0.480
35,312
0.28
# Obs. /
RSQ
OLS regressions. Dependent variables specified by row. This panel only includes firms (80.7% of all firms) that have a CAP+LIQ Section. Industry specific variables are computed
using TNIC industries as in Hoberg and Phillips (2010, 2011). Firm specific components are computed as a firm’s raw value for the specific characteristic minus the industry value. All
regressions include time and three digit SIC fixed effects. All standard errors are adjusted for clustering by firm.
Table V: Firm Characteristics and Constraint Variables
Table VI: Investment Policies versus Constraint Cross Terms (Economic Magnitudes)
Economic Magnitudes of Interactions between policy variables and constraint variables during each of the two
shocks considered in this paper. We consider the two shocks as noted in Panels A and B. For each shock, we
consider the difference in each policy for treated firms minus the policy level for non-treated firms. For the financial
crisis, the treated value is the firm’s value in 2009, and the untreated value is the firm’s value of the given policy in
2007. For the text shock, the treated value is the average policy for firms experiencing at least one of the three text
shocks (low demand, high competition, or high cost) and the untreated value is the average policy for firms
experiencing none of the three text shocks. Note that because each of the policy variables is scaled by sales or
assets, that the reported mean changes are in fractional units of sales or assets, respectively.
Constraint Variable
and Tercile (1 to 3)
Change in
R&D/Sales
Change in
CAPX/Sales
Change in
Equity
Issuance
/Assets
Change in
Debt
Issuance
/Assets
Panel A: Financial Crisis Change from 2007 to 2009
delay constraint 1
delay constraint 2
delay constraint 3
equity focus delay 1
equity focus delay 2
equity focus delay 3
debt focus delay 1
debt focus delay 2
debt focus delay 3
private placement focus delay 1
private placement focus delay 2
private placement focus delay 3
-0.014
-0.033
-0.119
-0.008
-0.069
-0.089
-0.100
-0.029
-0.036
-0.045
-0.046
-0.074
-0.020
-0.024
-0.061
-0.040
-0.021
-0.043
-0.029
-0.026
-0.049
-0.024
-0.032
-0.049
-0.010
-0.015
-0.026
-0.007
-0.013
-0.031
-0.017
-0.018
-0.016
-0.006
-0.021
-0.024
-0.014
-0.039
-0.027
-0.041
-0.020
-0.019
-0.003
-0.037
-0.039
-0.030
-0.012
-0.037
-0.028
-0.028
-0.061
-0.013
-0.031
-0.066
-0.059
-0.044
-0.016
-0.016
-0.042
-0.062
-0.009
-0.003
-0.003
-0.009
-0.001
-0.009
-0.005
-0.001
-0.014
-0.017
-0.007
0.007
Panel B: Text Shock (high minus low)
delay constraint 1
delay constraint 2
delay constraint 3
equity focus delay 1
equity focus delay 2
equity focus delay 3
debt focus delay 1
debt focus delay 2
debt focus delay 3
private placement focus delay 1
private placement focus delay 2
private placement focus delay 3
-0.042
-0.055
-0.137
-0.020
-0.059
-0.124
-0.123
-0.085
-0.036
-0.002
-0.073
-0.175
-0.001
0.000
0.001
0.011
-0.001
-0.013
-0.017
-0.001
0.012
0.004
-0.000
-0.008
Panel C: Mutual Fund Selling (high minus low)
delay constraint 1
delay constraint 2
delay constraint 3
equity focus delay 1
equity focus delay 2
equity focus delay 3
debt focus delay 1
debt focus delay 2
debt focus delay 3
private placement focus delay 1
private placement focus delay 2
private placement focus delay 3
-0.019
-0.031
-0.278
-0.005
-0.052
-0.245
-0.201
-0.078
-0.026
-0.060
-0.098
-0.160
-0.014
-0.010
-0.028
-0.008
-0.008
-0.031
-0.034
-0.002
-0.012
-0.006
-0.018
-0.024
47
-0.010
-0.013
-0.038
-0.003
-0.010
-0.042
-0.036
-0.013
-0.008
-0.012
-0.018
-0.030
-0.012
-0.014
-0.000
-0.011
-0.010
-0.006
-0.015
-0.009
-0.006
-0.007
-0.007
-0.012
Table VII: Investment Policies versus Constraint Cross Terms (Firm Fixed Effects)
Panel data OLS regressions with industry and year fixed effects. The sample includes all firm years from 1997 to
2009 that have a machine readable Capitalization and Liquidity subsection. The dependent variable is firm level
equity issuance/assets, debt issuance/assets, R&D/sales, or CAPX/ sales as noted by column. All independent
variables are lagged one year. We examine a different constraint index and its interactions as noted in each Panel
header. All panels are based on regressions that include all of the variables shown in Panel A. However, Panels B
through E only display the cross terms we draw inferences from in order to conserve space. Standard errors are
adjusted for clustering at the firm level. The independent variables include our key text-based investment and
constraint variables, and our text-based shock variables (high competition, low demand, technological change, and
high cost). We also consider cross terms including our constraint variables and the shock variables, in addition to a
2008-9 dummy. We do not directly include a 2008-9 dummy because we control for time fixed effects. Please see
Section II for a full discussion of variable definitions. ***, **, * denote significance at the 1%, 5%, 10% levels,
respectively.
Dependent
Variable=
R&D/Sales
Variable
Name
Dependent
Variable=
CAPX/Sales
Dependent
Variable=
Equity
Issuance
Dependent
Variable=
Debt
Issuance
Panel A: Constraint Index = Delay Investment Score
Text Shock
0.0034
–0.0001
–0.0009
Constraint Index
0.2431∗∗∗
0.0584∗∗∗
0.0715∗∗∗
Log Firm Age
–0.0395∗∗∗
–0.0157∗∗
–0.0130∗∗∗
Log Size of MD&A
0.0000
0.0000∗∗∗
0.0000∗∗∗
Log Assets
0.0193∗∗
0.0247∗∗∗
–0.0479∗∗∗
Tobins Q
0.0095∗∗∗
0.0101∗∗∗
0.0135∗∗∗
∗∗
∗∗∗
Text Shock x Index
–0.1062
–0.0385
–0.0211∗
2008–9 Dummy x Index
–0.2141∗∗
–0.0643∗∗
–0.0535∗∗∗
Panel B: Constraint Index = Marginal Equity Focus Delay
Text Shock x Index
2008–9 Dummy x Index
–0.0127
–0.7109∗∗∗
0.0400∗∗
–0.1251∗∗∗
–0.0360∗∗
–0.1775∗∗∗
–0.0009
–0.0008
0.0028
0.0000∗∗∗
–0.0134∗∗∗
0.0029∗∗∗
–0.0068
–0.0410
–0.0060
–0.1429∗∗∗
Panel C: Constraint Index = Marginal Debt Focus Delay
Text Shock x Index
2008–9 Dummy x Index
–0.0218
0.1766∗
0.0122
0.0271
0.0119
0.0760∗∗∗
–0.0009
–0.0167
–0.0001
0.0000
0.0000
–0.0001
0.0031
–0.1713∗∗∗
–0.0059
–0.0412
Panel D: Constraint Index = KZ Index
Text Shock x Index
2008–9 Dummy x Index
0.0002
0.0011
0.0002∗∗
0.0001
Panel E: Constraint Index = WW Index
Text Shock x Index
2008–9 Dummy x Index
0.0330
–0.1876∗
0.0209
–0.0454∗
48
Table VIII: Investment Policies versus Constraint Cross Terms (Arrellano-Bond)
The table displays the results of Arrellano-Bond regressions using differences GMM including year fixed effects.
The dependent variable is firm level equity issuance/assets, debt issuance/assets, R&D/sales, or CAPX/ sales as
noted by column. All independent variables are lagged one year. We examine a different constraint index and its
interactions as noted in each Panel header. All panels are based on regressions that include all of the variables
shown in Panel A. However, Panels B through E only display the cross terms we draw inferences from in order to
conserve space. The independent variables include our key text-based investment and constraint variables, and our
text-based shock variables (high competition, low demand, technological change, and high cost). We also consider
cross terms including our constraint variables and the shock variables, in addition to a 2008-9 dummy. Please see
Section II for a full discussion of variable definitions. ***, **, * denote significance at the 1%, 5%, 10% levels,
respectively. We also report the results of four regression diagnostic tests. A specification includes the letter “a” if
the idiosyncratic disturbance is not autocorrelated at the second lag at the 5% level, following Arrellano and Bond
(1991). A specification includes the letter “b” if the lagged dependent variable falls within the two reasonable OLS
bounds indicated in Roodman (2009). A specification includes the letter “c” if the Hansen J statistic of
overidentifying restrictions is not significant at the 5% level. A specification includes the letter “d” if the two shock
cross terms have a Hansen Differences statistic that is not significant at the 5% level, consistent with a failure to
reject their being exogenous. A specification including all four letters passes all four tests. We also note that our
instruments outnumber groups by a factor of roughly ten across specifications, far exceeding the minimum 1:1 ratio
below which Roodman (2009) indicates problems are most likely.
Dependent
Variable=
R&D/Sales
Variable
Name
Dependent
Variable=
CAPX/Sales
Dependent
Variable=
Equity
Issuance
Dependent
Variable=
Debt
Issuance
Panel A: Constraint Index = Delay Investment Score
Lagged Dep. Var.
Text Shock
Constraint Index
Log Firm Age
Log Size of MD&A
Log Assets
Tobins Q
Text Shock x Index
2008–9 Dummy x Index
Regression Diagnostics
0.3262∗∗∗
0.0053
0.0676
–0.0157
0.0016
0.0554∗∗∗
0.0070∗∗
–0.1458
–0.4008∗∗
a, b, c, d
0.3034∗∗∗
–0.0035∗
0.1355∗∗
–0.0106
–0.0006
–0.0113∗
0.0057∗∗∗
–0.1046∗∗∗
–0.1322∗∗∗
a, b, d
0.0589∗∗∗
0.0046∗∗
–0.0246
–0.0125
0.0005
–0.1032∗∗∗
0.0085∗∗∗
0.0008
–0.0735∗∗
a, b, d
0.0416∗∗∗
0.0023
0.0149
–0.0106
0.0004
–0.0639∗∗∗
–0.0051∗∗∗
–0.0210
–0.0404
a, d
Panel B: Constraint Index = Marginal Equity Focus Delay
Text Shock x Index
–0.1013
2008–9 Dummy x Index
–0.6175∗∗
Regression Diagnostics
a, b, c, d
Panel C: Constraint Index =
Text Shock x Index
2008–9 Dummy x Index
Regression Diagnostics
Text Shock x Index
2008–9 Dummy x Index
Regression Diagnostics
Text Shock x Index
2008–9 Dummy x Index
Regression Diagnostics
–0.1932∗
0.2236
a, b, c, d
Panel D: Constraint
–0.0976∗∗
–0.1538∗∗∗
–0.1568∗
–0.3792∗∗∗
a, b, c, d
a, b, d
Marginal Debt Focus Delay
0.0239
–0.0792
a, d
0.0798∗∗∗
0.0612∗∗
–0.0383
0.1122∗∗∗
a, b, d
a, b, d
Index = KZ Index
–0.0118
–0.0523∗
a, d
0.0369
–0.0208
0.0283∗
0.0229
–0.0420∗∗
0.0477∗∗∗
b, c, d
b, d
a, b, d
Panel E: Constraint Index = WW Index
0.0880
–0.2349∗∗
b, c, d
0.0026
–0.0150
b, c, d
49
0.0434
–0.1249∗∗∗
a, b, d
0.0016
0.0138
a, d
0.0126
0.0052
a, d
Table IX: Investment Policies versus Constraint Cross Terms (Arrellano-Bond) (Various Subsamples)
The table displays the results of Arrellano-Bond regressions using differences GMM including year fixed effects for
subsamples based on firm size, firm age, R&D Focus, and CAPX Focus. The dependent variable is firm level equity
issuance/assets, debt issuance/assets, R&D/sales, or CAPX/ sales as noted by column. All independent variables
are lagged one year. Each Panel displays the cross terms for a regression model that is the same as in Panel A of
Table VIII (see header for details), except that we examine a different constraint variable in each Panel as noted in
the Panel header. ***, **, * denote significance at the 1%, 5%, 10% levels, respectively. We also report the results
of four regression diagnostic tests, as also summarized in Table VIII.
Dependent
Variable=
R&D/Sales
Variable
Name
Dependent
Variable=
CAPX/Sales
Dependent
Variable=
Equity
Issuance
Dependent
Variable=
Debt
Issuance
Panel A: Constraint Index = Delay Investment Score [SMALL FIRMS]
Text Shock x Index
2008–9 Dummy x Index
Regression Diagnostics
–0.3077∗∗
–0.2866
a, b, d
–0.1034∗∗
–0.1435∗∗
a, b, c, d
0.0283
–0.1643∗∗∗
a, b, d
–0.0352
–0.0543
a, d
Panel B: Constraint Index = Delay Investment Score [LARGE FIRMS]
Text Shock x Index
2008–9 Dummy x Index
Regression Diagnostics
–0.0600
–0.2063
d
–0.0891∗∗∗
–0.1323∗∗
a, c, d
0.0063
0.0029
a, b, c, d
–0.0167
–0.0132
a, d
Panel C: Constraint Index = Delay Investment Score [YOUNG FIRMS]
Text Shock x Index
2008–9 Dummy x Index
Regression Diagnostics
–0.1079
–0.5745∗∗
a, b, c, d
–0.1357∗∗∗
–0.1196∗
b, c, d
–0.0082
–0.0863∗
a, b, d
–0.0263
–0.0643∗
a, c, d
Panel D: Constraint Index = Delay Investment Score [OLD FIRMS]
Text Shock x Index
2008–9 Dummy x Index
Regression Diagnostics
–0.1211
–0.1127
a, d
–0.0525
–0.1106∗∗
a, c, d
–0.0108
–0.0097
a, b, d
0.0069
0.0010
a, c, d
Panel E: Constraint Index = Delay Investment Score [LOW R&D FOCUS FIRMS]
Text Shock x Index
2008–9 Dummy x Index
Regression Diagnostics
0.0110
0.0340
a, d
0.0087
–0.0171
a, b, c, d
0.0327
–0.0186
a, b, c, d
0.0082
0.0022
a, c, d
Panel F: Constraint Index = Delay Investment Score [HIGH R&D FOCUS FIRMS]
Text Shock x Index
2008–9 Dummy x Index
Regression Diagnostics
–0.2171
–0.6992∗∗
a, b, c, d
–0.1910∗∗∗
–0.1976∗∗
b, c, d
–0.0250
–0.1399∗∗
a, b, d
–0.0321
–0.0841∗∗
a, c, d
Panel G: Constraint Index = Delay Investment Score [LOW CAPX FOCUS FIRMS]
Text Shock x Index
2008–9 Dummy x Index
Regression Diagnostics
–0.1956
–0.5633∗∗
b, c, d
–0.0840∗∗
–0.1959∗∗∗
a, b, d
0.0043
–0.0982∗∗
a, b, d
–0.0416∗
–0.0733∗∗
a, d
Panel H: Constraint Index = Delay Investment Score [HIGH CAPX FOCUS FIRMS]
Text Shock x Index
2008–9 Dummy x Index
Regression Diagnostics
–0.1322
–0.0518
a, d
–0.1379∗∗
0.0268
c, d
50
–0.0292
–0.0039
a, b, c, d
0.0278
0.0065
a, c, d
Table X: Investment Policies versus Constraint Cross Terms (Arrellano-Bond) (Equity and Debt Focus Constraints)
The table displays the results of Arrellano-Bond regressions using differences GMM including year fixed effects for
subsamples based on R&D Focus, and CAPX Focus subsamples. The dependent variable is firm level equity
issuance/assets, debt issuance/assets, R&D/sales, or CAPX/ sales as noted by column. All independent variables
are lagged one year. Each Panel displays the cross terms for a regression model that is the same as in Panel A of
Table VIII (see header for details), except that we examine a different constraint variable in each Panel as noted in
the Panel header. ***, **, * denote significance at the 1%, 5%, 10% levels, respectively. We also report the results
of four regression diagnostic tests, as also summarized in Table VIII.
Variable
Name
Dependent
Variable=
R&D/Sales
Dependent
Variable=
CAPX/Sales
Dependent
Variable=
Equity
Issuance
Dependent
Variable=
Debt
Issuance
Panel A: Constraint Index = Marginal Equity Focus Delay Investment Score [High R&D Firms]
Text Shock x Index
2008–9 Dummy x Index
Regression Diagnostics
–0.2777
–0.8965∗
b, c, d
–0.1189∗
–0.0924
b, c, d
–0.1373∗∗
–0.4924∗∗∗
a, b, d
0.0372
–0.1061
a, d
Panel B: Constraint Index = Marginal Debt Focus Delay Investment Score [High R&D Firms]
Text Shock x Index
2008–9 Dummy x Index
Regression Diagnostics
–0.3808∗
0.2704
b, c, d
0.0562
–0.1540∗∗
b, c, d
0.0630
0.1923∗∗∗
a, b, d
–0.0537∗
–0.0634
a, d
Panel C: Constraint Index = Marginal Equity Focus Delay Investment Score [High CAPX Firms]
Text Shock x Index
2008–9 Dummy x Index
Regression Diagnostics
–0.1440
–0.5863
a, d
–0.2605∗∗
–0.4500∗∗
c, d
–0.1109∗∗
–0.4183∗∗∗
a, b, c, d
0.0083
0.0300
a, c, d
Panel D: Constraint Index = Marginal Debt Focus Delay Investment Score [High CAPX Firms]
Text Shock x Index
2008–9 Dummy x Index
Regression Diagnostics
0.0754
0.7711∗∗∗
a, d
0.1015
0.0304
d
51
0.0492
0.1052∗∗
a, b, c, d
–0.0463
–0.0946∗
a, c, d
Table XI: The Marginal Role of Equity Private Placements (Arrellano-Bond)
The table displays the results of Arrellano-Bond regressions using differences GMM including year fixed effects.
The dependent variable is firm level equity issuance/assets, debt issuance/assets, R&D/sales, or CAPX/ sales as
noted by column. All independent variables are lagged one year. The key independent variables of interest in all
panels are the cross terms between the two economic shock variables and the marginal private placement focused
delay constraint variable. Each Panel displays the results for a regression model based on the whole sample (Panel
A) or based on various subsamples (all other panels). All regressions include the full set of variables used in Panel
A of Table VIII (see header for details), except that we only display the key cross terms to conserve space. ***, **,
* denote significance at the 1%, 5%, 10% levels, respectively. We also report the results of four regression
diagnostic tests, as also summarized in Table VIII.
Variable
Name
Dependent
Variable=
R&D/Sales
Text Shock x Index
2008–9 Dummy x Index
Regression Diagnostics
0.0400
–0.4558∗∗
a, b, c, d
Text Shock x Index
2008–9 Dummy x Index
Regression Diagnostics
0.1601
–0.3734
a, b, c, d
Dependent
Variable=
CAPX/Sales
Dependent
Variable=
Equity
Issuance
Dependent
Variable=
Debt
Issuance
–0.1296∗∗∗
–0.1530∗∗∗
a, b, d
–0.0698∗∗
–0.0261
a, d
–0.2131∗∗∗
–0.3747∗∗∗
a, b, d
–0.0144
0.0650
a, d
0.0056
–0.0529
a, b, c, d
–0.0736∗∗
–0.0400
a, d
–0.1808∗∗∗
–0.1530∗∗
a, b, c, d
–0.0949∗∗
0.0208
a, c, d
–0.0475
–0.1508∗∗∗
a, b, c, d
–0.0495
–0.0141
a, d
–0.0515
–0.0362
a, b, d
–0.0615∗
–0.0776
a, d
–0.1758∗∗∗
–0.2628∗∗∗
a, b, d
–0.0434
0.0957
a, c, d
–0.1249∗∗∗
–0.1318∗∗
a, b, d
–0.0893∗∗∗
–0.0153
a, d
–0.0903∗
–0.1134
a, b, c, d
–0.0288
–0.0055
a, c, d
Panel A: Whole Sample
–0.0542
–0.0862
a, b, d
Panel B: Small Firms
–0.1678∗∗∗
–0.2112
a, b, c, d
Panel C: Large Firms
Text Shock x Index
2008–9 Dummy x Index
Regression Diagnostics
–0.0171
–0.5132∗∗
c, d
0.0852
0.0370
a, c, d
Panel D: Young Firms
Text Shock x Index
2008–9 Dummy x Index
Regression Diagnostics
0.2286
–0.4461
a, b, c, d
–0.0566
0.0457
b, c, d
Panel E: Old Firms
Text Shock x Index
2008–9 Dummy x Index
Regression Diagnostics
–0.2704∗∗∗
–0.1519
a, d
–0.0105
–0.0312
a, d
Panel F: Low R&D Focus Firms
Text Shock x Index
2008–9 Dummy x Index
Regression Diagnostics
0.0157
0.0936∗∗∗
a, c, d
–0.0141
–0.0695
a, b, c, d
Panel F: High R&D Focus Firms
Text Shock x Index
2008–9 Dummy x Index
Regression Diagnostics
0.1337
–0.6979∗
a, b, c, d
–0.0602
–0.0024
b, c, d
Panel F: Low CAPX Focus Firms
Text Shock x Index
2008–9 Dummy x Index
Regression Diagnostics
0.0667
–0.3967
b, c, d
–0.0747
–0.1022
a, b, d
Panel F: High CAPX Focus Firms
Text Shock x Index
2008–9 Dummy x Index
Regression Diagnostics
–0.0141
–0.2401
a, d
–0.0944
0.0545
d
52
Table XII: Investment Policies versus Mutual Fund Selling Cross Terms (ArrellanoBond)
The table displays the results of Arrellano-Bond regressions using differences GMM including year fixed effects.
The dependent variable is firm level equity issuance/assets, or debt issuance/assets as noted by column. All
independent variables are lagged one year. We examine a different constraint index and its interactions as noted in
each Panel header. Each Panel displays the cross terms for a regression model that includes all variables displayed
in Panel A of Table VIII (see header for details), but with two exceptions. First, we examine a different constraint
variable in each Panel as noted in the Panel header. Second, we add forced mutual fund selling from Edmans,
Goldstein and Jiang (2011) along with a cross term based on this variable and the given constraint index. Please
see Section II for a full discussion of variable definitions. ***, **, * denote significance at the 1%, 5%, 10% levels,
respectively. We also report the results of four regression diagnostic tests, as also summarized in Table VIII.
Dependent
Variable=
Equity
Issuance
Variable
Name
Dependent
Variable=
Debt
Issuance
Panel A: Constraint Index = Delay Investment Score
Mutual Fund Selling
Mutual Selling x Index
Regression Diagnostics
–0.0005∗∗
0.0025
a, d
–0.0005∗∗
–0.0036∗∗
a, d
Panel B: Constraint Index = Marginal Equity Focus Delay
Mutual Fund Selling
Mutual Selling x Index
Regression Diagnostics
–0.0007∗∗∗
–0.0128∗∗∗
a, b, d
–0.0004∗∗
0.0007
a, d
Panel C: Constraint Index = Marginal Debt Focus Delay
Mutual Fund Selling
Mutual Selling x Index
Regression Diagnostics
–0.0005∗∗
0.0082∗∗∗
a, d
–0.0004∗∗
0.0025
a, d
Panel D: Constraint Index = Marginal Private Placement Focus Delay
Mutual Fund Selling
Mutual Selling x Index
Regression Diagnostics
–0.0009∗∗∗
–0.0133∗∗∗
a, b, d
–0.0008∗∗∗
–0.0092∗∗∗
a, d
Panel D: Constraint Index = KZ Index
Mutual Fund Selling
Mutual Selling x Index
Regression Diagnostics
–0.0004∗
0.0016
a, d
–0.0006∗∗∗
0.0006
a, d
Panel E: Constraint Index = WW Index
Mutual Fund Selling
Mutual Selling x Index
Regression Diagnostics
–0.0010∗
–0.0018
a, d
0.0009
0.0052∗
a, b, d
53
Figure 1: 3-D Plots of the likelihood of being highly constrained versus size and age.
One figure is displayed for each constraint variable as noted on the left-hand side of
each figure. The variable being plotted is the fraction of firms with a level of the
constraint variable that is more than two standard deviations above the mean based
on annual standard deviations.
0.1
Delay Investment
Constraint
Large Firms
0.05
Medium Firms
0
Young
Equity
Focused
Delay Investment
Constraint
Debt
Focused
Delay Investment
Constraint
Private Placement
Focused
Delay Investment
Small Firms
Mid‐Age
Old
0.1
Large Firms
0.05
Medium Firms
0
Young
Small Firms
Mid‐Age
Old
0.1
Large Firms
0.05
Medium Firms
0
Young
Small Firms
Mid‐Age
Old
0.04
Large Firms
0.02
Medium Firms
0
Young
Small Firms
Mid‐Age
Old
54
55
0.27
0.29
0.31
0.29
0.31
0.29
0.31
0.19
0.17
0.15
0.13
0.11
0.09
0.07
0.05
0.03
0.01
‐0.01
‐0.03
‐0.05
‐0.07
‐0.09
‐0.11
‐0.13
‐0.15
‐0.17
‐0.19
‐0.21
‐0.23
‐0.25
‐0.27
‐0.29
‐0.31
‐0.33
‐0.35
‐0.37
0.25
0.1
0.09
0.08
0.07
0.06
0.05
0.04
0.03
0.02
0.01
0
0.27
marginal debt‐focused delay constraint variable
0.27
0
0.23
0.02
0.25
0.04
0.25
0.06
0.21
0.08
0.23
0.1
0.23
marginal equity‐focused delay constraint variable
0.21
0.19
0.17
0.15
0.13
0.11
0.09
0.07
0.05
0.03
0.01
‐0.01
‐0.03
‐0.05
‐0.07
‐0.09
‐0.11
‐0.13
‐0.15
‐0.17
‐0.19
‐0.21
‐0.23
‐0.25
‐0.27
‐0.29
‐0.31
‐0.33
‐0.35
‐0.37
0.12
0.21
0.19
0.17
0.15
0.13
0.11
0.09
0.07
0.05
0.03
0.01
‐0.01
‐0.03
‐0.05
‐0.07
‐0.09
‐0.11
‐0.13
‐0.15
‐0.17
‐0.19
‐0.21
‐0.23
‐0.25
‐0.27
‐0.29
‐0.31
‐0.33
‐0.35
‐0.37
Figure 2: Frequency distribution of text-based constraint measures.
delay constraint variable
0.1
0.09
0.08
0.07
0.06
0.05
0.04
0.03
0.02
0.01
0