Financial Constraint, Liquidity Management and Investment

RiddioughSchool of BusinessUniversity of Wisconsin-Madison triddiough@bus.wisc.edu Zhonghua WuSchool of BusinessFlorida International University to financial constraint, where more less

Financial Constraint, Liquidity Management and Investment*

Timothy J RiddioughSchool of BusinessUniversity of Wisconsin-Madison

triddiough@bus.wisc.edu

Zhonghua WuSchool of BusinessFlorida International University

to financial constraint, where more (less) financially constrained firms in our sample exhibit high (low) investment and liquidity management sensitivity to variables that measure financial market frictions Empirical results provide support for debt overhang, free cash flow and asset tangibility

as important financial market frictions that influence investment outcomes.

* We thank David T Brown, Jim Clayton, Morris Davis, Piet Eiccholtz, Erasmo Giambona, Don Hausch,

François Ortalo-Magné, Steve Malpezzi, Armin Schwienbacher, James Seward, David Shulman, Ko Wang, Toni Whited and seminar participants at Baruch College, University of Amsterdam, University of Wisconsin-Madison,

Financial Constraint, Liquidity Management and Investment

1 Introduction

Fazzari, Hubbard, and Petersen (1988) convincingly argue that internal versus externalsources of finance are imperfect substitutes in the context of funding investment, and hence thatfinancial constraints impede the efficient allocation of resources Their study has had wide impact,and has come under intense scrutiny

Critics, beginning with panelists that provided comments and discussion published

alongside the original Brookings paper, have generally focused on three instrumental issues: i) endogeneity of financial constraint proxies; ii) measurement error in Tobin’s q; and iii) omitted

variables and channels that provide more complete information about the link between financialmarket frictions and real investment outcomes Chirinko (1993) concisely summarizes theseconcerns by stating, “It is unclear whether significant liquidity and net worth variables arecapturing a structural element heretofore missing in the investment equation or are merelyreflecting a general misspecification.”

Previous studies have addressed one or two of these instrumental issues at a time, but nonehave addressed all three in a systematic and comprehensive manner For example, Whited (1992)and Kaplan and Zingales (1997) primarily address the financial constraint issue, while Erickson and

Whited (2000) focus on measurement error in q and Almeida, Campello and Weisbach (2004)

emphasize the link between cash flow sensitivity of cash holdings and financial constraint Theintent of this study is to address all three issues—endogeneity of financial constraint proxies,

measurement error in q, and omitted variables/channels—simultaneously and comprehensively in

order to provide additional perspective on the effects of financial constraint on investmentdecisions

To address endogeneity in the financial constraint proxy and measurement error in q, we

analyze a specific sector that provides an attractive natural economic laboratory: publicly tradedfirms that own commercial real estate assets in an investment vehicle called a Real EstateInvestment Trust (REIT) These firms are regulated to pay out at least 90 percent of their GAAP netincome as dividends, and most pay out at least 100 percent of GAAP net income to avoid negativetax effects This implies that the entire sector is constrained in its ability to retain cash, andtherefore depends heavily on external finance to fund investment, which mitigates concerns over

confounding effects in identifying constrained firms These firms also have well measured q values,

due to the competitive nature of the industry and characteristics of the underlying commercial realestate assets

The third instrumental issue revolves around omitted variables and resource channels Wemake two contributions in this regard First, we recognize that cash is not a sufficient statistic foravailable liquidity Firms will vary in their capacity and need to hold liquidity, and may decide tohold less internal liquidity when low-cost external sources such as bank lines of credit (L/C) exist.Consequently, firms that might appear to be liquidity constrained may in fact have more thanadequate stores of liquidity when external sources are recognized Second, we specify and estimate

a structural model that accounts for endogeneity in cash flow retention, bank L/C usage, andinvestment decisions Cash flow retention and bank L/C usage together account for a firm’sliquidity management policy as related to investment, where simultaneous consideration allows us

to better disentangle cause and effect as well as to better assess investment-cash flow and othersensitivities that have been a focus in the literature

A unique panel data set covering the years 1990-2003 has been assembled to analyze theseissues Preliminary analysis shows that REITs retain less cash flow, have a lower stock of cash, anduse more bank L/C than a broad cross-section of other publicly traded firms In other words, based

on these measures, REITs appear to be financially constrained We also find that REIT bank L/Cusage increases monotonically with investment This suggests that, in the short run, and given their

significant constraints on cash flow retention, bank L/C substitutes for internal cash in fundinginvestment

Full sample structural 3SLS estimation produces a number of noteworthy results First, in

the investment equation, q and the liquidity flow measures of retained cash flow and bank L/C use

all significantly affect investment, with coefficient estimates that imply high investment

sensitivities Investment sensitivity to q is such that the elasticity of investment with respect to q is just shy of one, which places it near what standard q-theory would predict Given the cash flow

constraints faced by REITs together with the fact that commercial real estate assets are tangiblewith significant debt capacity, high investment sensitivity to both retained cash flow and bank L/Cuse is consistent with effects of asset tangibility (Almeida and Campello (2007)) and incentives toaccelerate current investment in order to create additional external financing capacity in the future(Hennesey, Levy, Whited (2007))

Across the full sample, firms are seen to invest at a rate of approximately 20 percent peryear, which exceeds rates of investment by the broad cross-section of comparison industrial firms.Moreover, most REITs pay well in excess of the minimum dividend payout requirement Thisraises the issue of whether these cash flow constrained and equity dependent firms are reallyfinancially constrained In other words, why is external finance available and affordable to thesefirms?

We conjecture that limited discretion on cash retention mitigates adverse selection costsassociated with raising outside finance This chain of reasoning implies that, contrary toconventional wisdom that emphasizes the primacy of information-based costly external finance as apremier financial market friction, cash flow constraints and equity dependence are not sufficientconditions for financial constraint

To differentiate between the effects of cash flow constraint and financial constraint oninvestment and liquidity management, we split the sample based on Kaplan and Zingales’ (1997)methodology for indexing financial constraint Based on KZ index scores, we find the more

constrained sub-sample invests less, generates lower cash flow and has a lower stock of cash, paysfewer dividends, employs more leverage, and is less likely to maintain relationships with banklenders and security underwriters In other words, the KZ index method appears to accuratelyclassify firms in our sample as more or less financially constrained.

Simultaneous equation estimation reveals substantial differences between firms that aremore versus less financially constrained Consistent with arguments of Gomes (2001), the lessfinancially constrained firms are responsive to investment signals contained in their stock prices,while the more constrained firms are not This outcome refines results of Baker, Stein and Wurgler(2003), who do not differentiate equity dependent firms on the basis of financial constraint

Sensitivity of investment and liquidity management to proxies for financial market frictions

is generally much higher in the sub-sample of more financially constrained firms For example,cash retention policy responds to a number of variables in the financially constrained sub-sample offirms, including investment Establishing a statistically significant link between dividend policyand investment is a new result (see Fama and French (2002) for additional background), in whichfirms decrease their dividend payout when investment increases—presumably to redirect scarcecash flow away from shareholders towards capital acquisition Variables that cause dividend payout

to increase in the more constrained sub-sample include equity issuance, a positive change in bankL/C capacity and a positive change in bank L/C use None of these variables have any effect ondividend payout in the less constrained sub-sample

Stark differences between sub-samples also exist with respect to bank L/C usage An extra

$1 of retained cash flow causes L/C usage to decrease by more than $1 in the more constrained sample, whereas retained cash flow has no effect on L/C use in the sample of less constrainedfirms Thus, more financially constrained firms treat cash as negative short term debt by savingcash out of cash flow, whereas cash constrained but less financially constrained firms do not.Consistent with Sufi (2007), these results suggest that financially constrained firms closely monitortheir bank L/C use due to concerns over covenant violations that would impose significant

sub-additional costs Bank L/C use is also highly responsive to investment, leverage and firm age in themore constrained sub-sample, while there is either less or no responsiveness to these effects in theless constrained sub-sample.

Thus, we show that cash flow constraint is not the same thing as financial constraint sample estimation reveals that cash flow constrained firms that are classified as financiallyconstrained are highly responsive to shocks in variables that proxy for financial market frictions.Higher cash flow results in a simultaneous paydown in bank L/C use and increase in investment,achieved in part through reduction in dividend payout Less financially constrained firms, incontrast, exhibit stability in their dividend policy with no sensitivity to investment or L/C use.These findings point to the importance of agency costs over information-based costs of externalfinance in governing investment and liquidity management policies of financially constrainedfirms Our results are also generally consistent with cash flow focused liquidity management effectsemphasized in Almeida et al (2004) and Almedia and Campello (2007)

Sub-The paper is organized as follows Section 2 provides further background on REITs andbank L/C usage Hypothesis development and empirical model specification are addressed insection 3 Data are described and a preliminary analysis of the data are reported and analyzed insection 4 Simultaneous system equation estimation for the full sample is undertaken in section 5,and sub-sample results are considered in section 6 Section 7 concludes the paper

2 Further Background on REITs and Bank Lines of Credit

The data employed in previous studies of corporate investment generally have limited andnoisy variation One solution to the problem is to apply alternative specifications and moresophisticated econometric analysis (see, e.g., Hoshi, Kashyap and Scharfstein (1991), Erickson andWhited (2000)) A more direct solution is to try to obtain better data We emphasize the latterapproach, and examine the Real Estate Investment Trust (REIT) sector

The REIT sector, for several reasons, provide an attractive natural laboratory to study theeffects of financial market frictions on firm investment First, all REITs are cash flow constrained

by regulation, as they are required to pay at least 90 percent of taxable income to shareholders inthe form of dividends.1 After accounting for the effects of depreciation and the fact that most REITspay in excess of the minimum payout requirement, 65 to 90 percent of current cash flow is typicallypaid out as dividends.2 Cash flow constraints of this magnitude are typically thought to implyfinancially constraint due to the presumed high costs of accessing external finance Consequently,based on this logic, exogenously imposed cash flow constraints substantially reduce endogeneityproblems associated with identifying financially constrained firms

Combined adjustment and purchasing costs of investment should not exceed the shadowvalue of newly installed capital Shadow value follows from investor expectations of the marginal

contributions of new capital gains to future profit In theory, marginal q provides a direct (isomorphic) measure of the shadow value of capital Marginal q is generally unobservable in the data, however, so analysts rely on average q If marginal q is badly measured by average q, an

investment-cash flow relation may be a spurious, as current cash flow may contain informationregarding investment opportunities

Hayashi (1982) has shown that average q is a sufficient statistic for investment when the

following necessary conditions are satisfied: i) there is perfect competition in factor and productmarkets, ii) fixed capital is homogeneous, and iii) product and adjustment costs are linearlyhomogeneous Commercial real estate asset markets and the firms (REITs) that own these assetssatisfy these conditions to a remarkably close approximation The factor market is primarily landand physical capital, with relatively little reliance on human capital, and these markets are generallyquite competitive Competitive market structure is important, since imperfectly competitive

1 Prior to 2000, the dividend payout requirement was 95 percent.

2 REITs that pay out less than 100 percent of net income incur an excise tax on the difference, which causes most to pay

at least 100 percent of net income The average payout in our data and in other studies is approximately 120 percent of net income (see also Chan et al (2003)) The annual flow of depreciation expense (a non-cash item) is generally

industries will generate quasi-rents that can cause a spurious correlation between cash flow and

investment after controlling for average q (Abel and Eberly (2001), Cooper and Ejarque (2003)).

A large proportion of real estate asset operating expenses go to pay utilities, insurance andproperty taxes, which are effectively linear in scale Investment, which in this sector is primarilythe acquisition of built (productive) assets, results in adjustment costs that are linearlyhomogeneous Furthermore, investment in built assets requires very little “time-to-build”, and also

contain little option value that potentially distorts the marginal-average q relation In addition,

regulation requires REITs to be monoline (non-integrated) companies This suggests that imperfectproduct substitution that confounds many multi-product firms is less problematic with REITs,

which strengthens the link between average and marginal q.3 Finally, there are no taxes at the entity

level to distort investment incentives

Compounding the usual marginal q–average q measurement error problem is that average q

is often badly measured in the data due the reliance on asset book values to proxy for thereplacement cost of firm assets As Hartzell, Sun, and Titman (2006) and others have pointed out,however, book asset value is a relatively accurate measure of replacement cost with commercialreal estate assets For example, they report a correlation of 92 between their book asset measure of

q and a net asset value measure of q that is based on market sales data.

To begin to get a sense of the data, Figure 1 shows how average q varies by year for REITs

in our sample, where q is defined as market value of equity plus book value of debt divided by asset

book value as of the beginning of the year Quartile cutoff values are displayed in addition to mean

values Mean and median q values generally exceed 1.0 over the sample period, but not by a large amount It is also apparent that there is significant cross-sectional variation in q values in the early

years of the sample period (particularly in 1992 and 1993), whereas this variation decreases after

1994

Figure 1 Here

3 See Hayashi and Inoue (1991) for more on the issue of imperfect asset substitution and investment

In Figure 2 the time-series of average q and rates of investment by year as a percentage of

year-beginning asset value are displayed There is a clear direct contemporaneous relation between

investment and average q, with cross-correlation measured at 78 Note that investment is in the 10

percent range for most years, but that the years 1995-98 resulted in higher rates of investment thatgenerally exceeded 20 percent of year-beginning book assets

Figure 2 Here

In their analysis of the REIT sector, Ott, et al (2005) document that only seven percent offirm-level investment was funded by retained cash earnings, as compared to 70 percent for otherpublicly traded firms Because of their inability to retain cash, REITs rely on outside financingsources to facilitate investment Seasoned equity, long-term unsecured debt, and secured mortgagedebt are the claims typically issued to permanently finance acquisitions (see Brown and Riddiough(2003) for additional background)

In the short term, REITs rely heavily on bank lines of credit (L/C) to fund investment.4,5 Thetypical funding cycle is as follows A firm identifies an investment opportunity, which oftenrequires partial or full payment at closing Anticipating these investment opportunities, the firmarranges a bank L/C with sufficient capacity to meet its liquidity needs The bank L/C is drawndown to fund the investment, where the firm subsequently begins to work with an investment orcommercial bank to secure permanent sources of finance Once there is sufficient scale, equity or

4 We have explored whether REITs utilize the commercial paper market, and have found no evidence that they do This

is because REITs are generally younger firms without the AAA and AA credit ratings required to access this market The ratings outcomes are in significant part because REITs are unable to retain cash flow Thus, it appears that firms which have access to the commercial paper market are the larger, more mature firms that are able to retain cash— precisely the type of firms that are not likely to be financially constrained In contrast, REITs, which are by definition cash constrained, almost exclusivily rely on bank L/C for external-source liquidity needs.

5 Bank L/C account for a large proportion of total firm-level bank debt in the U.S A recent Federal Reserve Board survey reports that approximately 80 percent of commercial and industrial loans made by banks are arranged as short- term bank loan commitments or lines of credit According to Martin and Santomero (1987) and Avery and Berger (1991), the primary stated reasons why firms use bank L/C are financial flexibility and speed of action In practice,

long-term debt is issued with proceeds used to pay down bank L/C and hence recreate capacity tofund the next round of acquisitions.

Table 1 compares REITs to other publicly traded firms (C-Corporations) that are not subject

to dividend payout requirements We show how five ratios vary and compare by year from 1990 to

2003 The five ratios, all as a percentage of beginning-year total book assets (K), are net investment(INV), dividends paid (DIV), net cash flow (NCF), the stock of cash and liquid securities (CS), andbank L/C capacity (L/C) We also report how investment correlates with the other reportedvariables over the sample period

Table 1 Here

Observe the high rates of investment by REITs in the middle 1990s, and that averageinvestment by REITs exceeded average investment by C-Corporations by almost 70 percent duringthe 1990-2003 sample period.6 As noted earlier, acquisitions were the largest component of netinvestment for REITs over the sample period, whereas capital expenditures and depreciation (anegative adjustment) were major components of net investment for C-Corporations.7

As a result of the dividend payout requirement, paid dividends are significantly higher andretained cash flow is significantly lower for REITs Interestingly, as noted earlier, a significantfraction of REITs pay dividends in excess of the minimum 90 percent of net income required byregulation Specifically, further analysis reveals that 70 percent of REITs pay at least 100 percent

of the net income as dividends in any given year, with a median payout ratio of 120 percent Thisequates to most firms retaining between 10 and 35 percent of cash flow as deployable liquidity or

an addition to cash stock

6 Average rates of investment for REITs in this table do not exactly match those reported in Figure 2 because different data sources were used to generate the respective table and figure.

7 Real estate assets are highly durable with depreciation periods that generally exceed 30 years, whereas assets held by industrial firms typically depreciate at a much faster rate Consequently, capital expenditures are significantly higher for C-Corporations than for REITs.

Cash stock is significantly lower and bank L/C capacity is significantly higher for REITs incomparison to C-Corporations These two variables also display interesting covariation with respect

to investment Prior to 1993, equity REITs were a small sector with a total capitalization ofapproximately $20 billion Rates of growth were relatively slow during this period As a result,cash stocks were relatively high and bank L/C capacity was relatively low Then, in 1993, rates ofinvestment increased rapidly In 1993 and 1994, the data indicate that REITs were able to tap theircash reserves to help fund investment By 1995, however, cash reserves were largely depleted, andbank L/C capacity increased substantially as an apparent substitute for (deficient) internal-sourceliquidity Finally, in 1999-2003, when investment rates drop off from previous high levels, bank L/

C capacity likewise declines Note that cash flow also drops of during this latter period, whichlimits REITs ability to replenish their cash reserves

In comparison, bank L/C capacity and cash stock display a distinct negative correlation with investment for C-Corporations Whereas cash flow constrained REITs are forced

cross-to rely on both internal and external sources of liquidity cross-to help fund investment, the typical (cashunconstrained) C-Corporation appears to use its vastly greater internal store of liquidity to fundinvestment And, because unconstrained firms don’t require external-source liquidity to meet theirinvestment needs, C-Corporations maintain lower L/C capacity relative to the more cashconstrained REITs

REITs thus appear to use bank L/C as a substitute for cash, and do so as a result of beingcash flow constrained At the same time, high levels of investment and paid dividends that exceedthe minimum payout requirement suggest that many of these cash flow constrained firms are notnecessarily financially constrained, in the sense that financial constraints are thought to cause lowrates of investment and create strong incentives to hoard available cash flow The availability ofexternal-source liquidity in the form of bank L/C provides an intriguing link that may help explainthese complex relations

3 Hypothesis Development and Empirical Model Specification

3.a Hypothesis Development

To motivate our empirical model specification we appeal to a recent paper by Hennessy,Levy, and Whited (2007), who develop an internally consistent theory of dynamic investment with

financial market frictions In their model, marginal q depends on average q plus factors that account

for distortions associated with costly external finance and debt overhang problems In a dynamicsetting, financial constraints and costly external finance create incentives for the firm to invest now

in order to create financial capacity in the future The dynamic investment-collateral capacityfeature of Hennessy et al.’s analysis has particular relevance to firms such as REITs, which, besidesexperiencing significant constraints on cash flow retention, own tangible-durable collateral thatoffers significant debt capacity (see also Almeida and Campello (2007))

For hypothesis development, we primarily focus on the unique role of bank L/C as a cost liquidity source in a world with the identified financing frictions. 8,9 As a starting point, it is

low-important to distinguish between the effects of bank L/C capacity and usage Bank L/C capacity is

a negotiated outcome that is strongly influenced by a bank that closely monitors the financialcondition of the firm We interpret changes in bank L/C capacity as a proxy for the magnitude ofthe firm’s debt overhang problem, where greater L/C debt capacity implies greater overall debtcapacity and hence smaller debt overhang problems This in turn implies greater investment

Conditional on bank L/C capacity, bank L/C utilization is chosen by the firm in response tocurrent and expected future investment opportunities Financially constrained firms will valuefuture financial capacity created through investment, and intensely utilize their available L/Ccapacity to install capital when investment opportunities are available Thus, controlling for other

8 As Joe Stiglitz noted in the original general discussion to Fazzari, Hubbard, and Petersen (1988), “liquidity of a firm includes its lines of credit as well as its stock of cash This is an alternative explanation of why the stock of cash has little explanatory power in the cross-sectional investment equations even if finance constraints are important.”

9 Banks are known to play a unique role in resolving financial market frictions, and hence can lessen (or tighten) financial constraints (Fama (1985), James (1987)) Houston and James (2001) document that multiple bank

relationships relax investment-cash flow sensitivities, but they do not consider the direct impact of bank finance capacity or utilization on investment Others, including notably Sufi (2007), have considered the link between firm financial characteristics, bank L/C availability and/or the cost of bank finance, but again a direct channel between bank finance and investment is largely unexplored.

factors including q and bank L/C capacity, we posit that L/C usage proxies for preemptive

investment incentives as they relate to relaxing financial constraints going forward

This leads to our first hypothesis:

Hypotheses 1: An increase in bank L/C capacity is positively related to investment Furthermore,

conditional on L/C capacity, an increase in bank L/C utilization is positively related to investment.

Hypothesis 1 suggests that L/C usage causes investment in a world with financial market frictions Bank L/C usage is also important to financially constrained firms as a liquidity source to

facilitate investment This suggests that causation actually goes both ways, in that L/C usage is animportant source of capital and therefore highly sensitive to actual levels of investment This results

in the following hypothesis:

Hypotheses 2: Bank L/C utilization is particularly sensitive to and depends positively on actual

investment.

Financially constrained firms are thus hypothesized to utilize bank L/C as a substitute forscarce cash In a dynamic setting, however, financially constrained firms must carefully managetheir liquidity to ensure that financial slack is available to fund investment opportunities as theyarrive over time This is a major focus of Hennessey et al (2007), and is also central to the analysis

of Almeida, Campello and Weisbach (2004) in their paper on the cash flow sensitivity of cash Inthis latter paper, cash holdings of financially constrained firms are sensitive to current cash flow asthey (implicitly) relate to creating financial capacity for anticipated future investment

In our context, when cash flow is high, and because the opportunity cost of L/C usage willgenerally exceed that of the cash stock, constrained firms will use their cash flow to reduce their L/

C usage in order to economize on cost and to recreate capacity for the next round of investmentopportunities In contrast, we would expect less financially constrained firms have a lower L/Cusage rate to begin with and a greater ability to secure additional L/C capacity to meet liquidityneeds As a result, they show less propensity to reduce L/C usage

After controlling for investment, high cash flow realizations are thus predicted to be used byREITs to reduce bank L/C outstanding That is, L/C use substitutes for cash flow Moreover, giventhat external debt and equity issuances are infrequent and expensive to undertake relative toutilizing bank L/C, we expect external security issuances to be used (in part) to reduce bank L/Coutstanding in order to reduce cost and replenish the liquidity stock This results in our thirdhypothesis:

Hypotheses 3: Bank L/C utilization depends negatively on cash flow as well as equity and debt

issuances.

In the short term, financially constrained firms can fund investment from two possiblesources: cash or bank L/C One possible way to increase cash available for investment is to reducethe dividend payout Thus, dividend policy is hypothesized to respond to the high relative cost ofexternal finance, and is endogenous as a result We summarize this in the following hypothesis:

Hypotheses 4: Financially constrained firms dynamically adjust their dividend payout downward

to fund higher levels of investment.

At least two factors complicate the intuition embedded in this hypothesis First, it isimportant to recognize that limits to debt capacity and exogenous constraints on retaining cashreduce a firm’s discretion to select against outside equity investors This is in turn lowers the cost

of external finance and hence reduces the need to retain earnings to finance investment Second,implications of financial capacity management as in Almeida et al (2004) suggest that cash flow isvaluable to financially constrained firms in all states of the world as a way to hedge against futureincome or investment shocks This will cause dividend payouts to be smoothed, which will dampenthe anticipated negative investment-dividend payout relation

3.b Empirical Model Specification

To specify an empirical model we focus on three financial market frictions previouslydiscussed: costly external finance, debt overhang, and collateral borrowing constraints

Furthermore, the four hypotheses stated in the previous section collectively imply a structuralrelation between investment and liquidity management as expressed through cash retention policyand bank L/C usage This produces the following simultaneous system of equations to beestimated:

in the RetainedCashFlow equation) and retained cash flow isolates the dividend payout.

Table 2 identifies and summarizes variables that are common across the three equations,

with their interpretation as related to investment Retained cash flow and cash stock are used to proxy for costly external finance and debt overhang problems Total leverage proxies for debt

overhang problems Its effect is complicated, however, by incentives for collateral-constrainedfirms to invest more today in order to create additional financial capacity in the future Thus therelationship between investment and total leverage is unclear

Table 2 Here

In terms of investment, the change in bank L/C capacity is considered as a proxy for debt

overhang costs, and is of particular interest in the context of liquidity management for cash

constrained firms The change in bank L/C usage measures the firm’s actual utilization of its

external-source liquidity stock After controlling for L/C capacity and other relevant effects, intenseutilization of available L/C capacity is interpreted to be consistent with incentives to install tangiblecollateral today (accelerate investment) in order to relax financial constraints in the future.

Additional variables to control for the effects of financial market frictions include Firm age

(years after IPO), which proxies for unspecified financial market frictions such as time-varyingcostly external finance or collateral capacity effects For example, firms begin their life without anysort of managerial track record, and therefore may be more financially constrained than seasoned

firms Equity and debt issuance dummies are included as controls for major financing events that

affect short-run liquidity management decisions

The purpose of equation (2) is to isolate determinants of cash flow retention policy as itinteracts with investment and bank L/C utilization We would expect cash flow retention policy to

be sensitive to proxies for financial market frictions if firms are truly financially constrained In

addition to gross cash flow, lagged gross and retained cash flow are included as instrumental

variables in the specification to account for inertia in the firm’s dividend policy.10

Change in L/C use is specified in equation (3) The relation between L/C usage andinvestment is of particular interest, as is the relation between L/C usage and retained cash flow

Instrumental variables in this equation are year-beginning total L/C capacity and year-beginning total L/C use We would expect a positive relation between the change in L/C use and available

total L/C capacity and a negative relation between the change in L/C use and the outstanding L/Cdebt level

This system of equations that account for interactions between liquidity management andinvestment allows us to collectively assess hypotheses 1 through 4 Critical relations betweeninvestment and L/C use or capacity, as articulated in hypotheses 1 and 2, are contained in equations(1) and (3) If REITs are truly financially constrained, we would expect high sensitivity of L/C use

to investment given the importance of bank L/C in facilitating investment for these

cash-10 The inclusion of lagged retained cash flow as an instrumental variable in the net cash flow equation creates concerns regarding estimation consistency and bias in the system of equations This issue will be addressed in detail in the estimation section of the paper.

constrained firms Predictions associated with Hypothesis 3, which addresses the relation betweenbank L/C use and cash, are captured by equation (3) The relation between investment and dividendpayout as articulated in Hypothesis 4 is contained in equation (2), where the maintained hypothesispredicts that cash flow retention (dividend payout) moves directly (inversely) with investment.

4 Data and Preliminary Results

4.a Data

Our primary data source for model estimation is the SNL REIT database This databaseprovides detailed information on REIT investment, bank L/C availability and usage, and firmfinancial characteristics To be included in the sample, a firm has to meet the following criteria: (1)listed on NYSE, AMEX or NASDAQ; (2) elected REIT tax status at the beginning of each sampleyear; (3) registered with the National Association of Real Estate Investment Trusts (NAREIT), theindustry’s trade association; and (4) classified as an equity REIT by NAREIT.11 Given a sampleperiod of 1990 through 2003, the original sample from the SNL REIT database has 3,667 firm-yearobservations

Capital offering data (equity and public debt issuance) from the NAREIT Capital Offeringdatabase is hand-matched into the SNL data set These data consist of 156 equity IPOs, 1,401seasoned equity offerings, and 950 public debt offerings We also obtain firm-level REIT bank L/Cinformation from Loan Pricing Corporation’s DealScan database

We eliminate observations that do not fit within the following bounds: 0.3<Q t4.0;

−1.0Investment t3.0; 0.0≤CashStockt ≤0.5; 0.0Leverage t≤1.0 where all variables are scaled by

year-beginning total book asset value The wide bounds on investment result because some firms

grew significantly in a given year due to a merger or other significant investments We limit cashstock to 50 percent of book assets due to regulation that constrains REIT holdings in assets other

than real estate interests.12,13 The final data set produces an unbalanced panel consisting of 1,257firm-year observations from 1990 to 2003.

Summary statistics are presented in Table 3 Average investment of just under 20 percentover the sample period indicates significant growth at the firm and sector level.14 The mean Q value

is 1.217, which is not terribly high given the rapid rate of investment by many firms during the

sample period One explanation for the lower Q values is that Q is better measured in our data due

to book asset values that more accurately reflect replacement cost Average cash flow net ofdividends is only 1.64 percent of assets, while the average stock of cash and marketable securities

is only 1.89 percent of assets.15,16 This reaffirms that REITs are unable to retain significant amounts

of cash as a result of dividend payout requirements Note that, even with severe constraints on cashretention, both variables exhibit significant cross-sectional variation Total leverage as measured bylong-term debt is approximately 40 percent of firm assets on average

12 For example, one observation which we eliminated had a cash stock value of 993 This firm was apparently in liquidation mode, as its asset base was declining significantly in the two years leading up to the 993 observation

13 In addition, one observation with an annual change in L/C capacity of 17.35 times year-beginning book assets was eliminated This firm experienced close to a 100-fold increase in book assets in one year, with total L/C capacity increasing from zero to 20 percent of year-end book assets.

14 Note that data used in this section, culled primarily from the SNL REIT data base, is different from the data used in Table 1 to compare REITS and C-Corporations That data were derived primarily from Compustat and DealScan data bases, where we also did not apply the selection-screening criteria stated above.

15 Retained cash flow is defined as GAAP net income plus depreciation and amortization minus paid dividends.

16 Significant increases and decreases in assets in a given year caused by (dis)investment are the reasons for the

relatively high max and min values, particularly in the flow variables.

17 As of 2003, a firm could have been in existence a total of 43 years as a REIT (original REIT legislation was passed in 1960) The SNL REIT database counts age from the firm’s IPO date, so some firms that initially went public as a C- Corporation converted to REIT status at some later date prior to the start of our sample period in 1990.

The average changes in bank L/C capacity and actual bank L/C usage as a percentage ofassets are 7.20 and 1.54 percent per year, respectively.18 Note that the median value for bothvariables is zero There are significant differences in the composition of the two median values,however The change in L/C use is zero 14.2 percent of the time, where a significant proportion ofthose observations are from earlier in the sample period by firms that do not actively utilize theirbank lines In comparison, the change in L/C capacity is zero 42.7 percent of the time This largerpercentage reflects the fact that capacities are not always renegotiated on an annual basis.

4.b Erickson−Whited Test of Q Measurement Quality

Investment models that account for financial market frictions often posit that average q is

informative and well measured, in the sense that the necessary conditions stated in Hayashi (1982)are satisfied and data used for empirical testing accurately describe the true current value of thecapital stock Although these presumptions are typically problematic, we have argued that the REITdata are both informative and well measured This in turn implies that: i) only variables thatmeasure financial market frictions are required to augment the classical investment equationspecification (no additional controls for real market effects are required), and ii) coefficientestimates should more accurately reflect true economic magnitudes associated with the relevant

variables, including Q.

To test our assertions, we apply a generalized threshold test supplied by Erickson andWhited (2005) that allows us to assess coefficient sign robustness of regressors as they depend on

the measurement quality of our proxy for Tobin’s q The basic idea is to posit that Tobin’s q is

unobservable while other variables are observable A proxy is chosen for the unobservable

regressor, thus causing the true value of q to be measured with error For our purposes, we assume that the measurement error in q is uncorrelated with the disturbance term from an OLS regression

18 As noted earlier, one observation with a change in L/C capacity value of 17.35 was eliminated from the sample The next largest observation, which is the maximum in our data set used for model estimation, is 4.68 In this case there was approximately a ten-fold increase in book assets over the year, accompanied with L/C capacity that roughly doubled Thus, extremely rapid firm-level growth that sometimes occurred is the primary cause of the large percentage increases

Threshold estimates are used to assess whether the signs of other regressors might beaffected by the errors-in-variable problem A threshold estimate near zero implies that thehypothesis that the coefficient of interest has the incorrect sign can be rejected, whereas acoefficient estimate near one makes it hard to reject the hypothesis that the coefficient of interest iszero A coefficient in between zero and one is indeterminate Thus, for our purposes, the nullhypothesis is that the coefficient sign of the variable of interest (cash flow, for example) is zero andtherefore not robust to the errors-in-variable problem If the test does not reject the null hypotheses,

then one can infer that Tobin’s q is sufficiently well measured so as to produce a reliable

coefficient sign in an OLS investment equation estimation

Column A of Table 4 reports OLS estimation results for the investment equation Note that

the coefficient for average q is statistically as well as economically significant The size of the coefficient at 139 is particularly noteworthy, and is several magnitudes larger than q coefficient

estimates generated by OLS investment models that utilize industrial data (see, e.g., analysiscontained in Erickson and Whited (2005)) Indeed, the size of the coefficient is such that the

elasticity of investment with respect to Q is near one, which is what theory would predict.19

Table 4 Here

Using estimates supplied to us by Toni Whited, in columns (B) and (C) we report thresholdvalue estimates together with standard errors of the estimates We find that all of the statisticallysignificant variable coefficients reported in Table 4 pass the Erickson and Whited (2005)robustness test based on partial correlations.20 That is, the test allows us to reject the null hypothesis

that the non-q explanatory variable coefficients are zero as a result of errors-in-variables problems associated with our q measure These test results thus provide additional evidence supporting our

19 Using the coefficient estimate of 139, a mean investment rate of 20 and a mean q value of 1.22, an elasticity

claim that Tobin’s q is well measured using data from the REIT sector, and that investment

equation results are more reliable than those encountered in many other studies of the effects ofmarket frictions on investment

5 Full Sample Structural Model Estimation

This paper has two primary objectives The first is to analyze the endogenous effects ofliquidity management on investment, and the second is to assess whether cash flow constraint is thesame thing as financial constraint This section emphasizes the former objective by focusing on fullsample results, while the following section emphasizes the latter objective by analyzing sub-samples

Concern over the effects of endogenous dividend payout policy on investment has existedsince Fazzari, Hubbard, and Petersen’s (1988) initial work Although it is true that, in our data,REITs are constrained to pay out a significant portion of their gross cash flow as dividends, mostREITs actually pay well in excess of the required minimum, with significant variation in actualpaid dividends (see Chan et al (2003)) This suggests that cash retention policy is potentiallyendogenous as it relates to investment Bank L/C usage is also likely to be endogenous, since weknow that REITs use their bank L/C to fund investment (Brown and Riddiough (2003))

We estimate a linear system based on equations (1) through (3) using a 3SLS procedure.The endogenous variables—investment, retained cash flow, and L/C use—are estimated in a firststage Because two of the endogenous variables are used to estimate the third endogenous variable

in each of the three equations, we pool all other variables to be used as instruments in the first stageestimation (including all fixed effects) This approach is conservative, in the sense that the pooling

of all non-endogenous variables as instruments can increase the standard errors of the endogenous

variables to bias against statistical significance To address this latter issue, we also estimated thesystem using an iterated 3SLS procedure (see Hausman (1975)) Three stage least squares and

iterated 3SLS estimation results for the sample are displayed in Table 5 All model specificationsinclude year and firm property type fixed effects in addition to an intercept term.

Table 5 Here

Investment equation results are reported in panel A The coefficient estimate on Q decreases

slightly from the OLS model coefficient estimate reported in Table 4, but remains relatively large

and statistically significant High investment-q sensitivity is consistent with results of Baker, Stein

and Wurgler (2003), who find that equity dependent firms display a higher sensitivity of investment

to q than non-equity dependent firms Our finding makes sense in that equity-dependent firms like

REITs repeatedly access external capital markets, and therefore must be careful to reactappropriately to capital cost-investment signals contained in the firm’s stock price

At the same time that investment shows significant sensitivity to Q, a number of variables

meant to proxy for financial market frictions, including variables measuring the endogenous effects

of liquidity on investment, also exert a strong influence on investment In particular, the coefficient

on retained cash flow implies extreme investment sensitivity to available cash flow This sensitivity

of investment to retained cash flow can be explained by the tangibility of real estate as loancollateral and its effects on financial constraint.21 Almeida and Campello (2007) show thatinvestment sensitivity to cash flow for financially constrained firms is increasing in assettangibility Because REITs hold very high percentages of highly tangible real estate, their credit

21 We have also considered the possibility of data or specification issues in explaining the high sensitivity of investment

to retained cash flow After examining histograms of all variables reported in Table 5 for the potential distorting effects

of outliers, and eliminating 31 observations that might influence the results, we find that our results are robust to the potential effects of outliers (this holds for retained cash flow as well as all other variables) We also examine the model specification, in which we use lagged retained cash flow as an instrument in the retained cash flow equation This variable is included because dividend policy is known to be smoothed, where current dividend payouts typically depend on the previous period’s dividend payout (see Lintner (1956)) The use of lagged endogenous variables as instruments is, however, known to potentially produce inconsistent and biased coefficient estimates in simultaneous equation estimation, since the lagged instrument in question may be correlated with the system disturbances To address this issue, we conduct two diagnostic tests First, we conduct a Sargan-Hansen misspecification test that assesses the effect of the lagged retained cash flow instrumental variable on the investment (as well as L/C use) equation The test does not reject the null hypothesis that the instrument is an over-identifying restriction, thus lending credibility that the lagged instrumental variable does not unduly affect the results The second test assesses serial correlation in the system disturbance term A Durbin-Watson test fails to reject the hypothesis that there is no serial correlation in error terms.