3000 word report based on two related papers

Available online at www.sciencedirect.com

www.elsevier.com/locate/jbf

Journal of Banking & Finance 32 (2008) 207–216

The impact of capital market imperfections on investment–cash flow sensitivity

S�enay Ağca a,*, Abon Mozumdar b,1

a School of Business, George Washington University, 2201 G Street, Funger Hall 505, Washington, DC 20052, United States

b Pamplin College of Business, Virginia Tech, 7054 Haycock Road, Room 352, Falls Church, VA 22043, United States

Received 5 June 2006; accepted 20 February 2007 Available online 6 July 2007

Abstract

We examine the investment–cash flow sensitivity of US manufacturing firms in relation to five factors associated with capital market imperfections – fund flows, institutional ownership, analyst following, bond ratings, and an index of antitakeover amendments. We find a steady decline in the estimated sensitivity over time. Furthermore, we find that investment–cash flow sensitivity decreases with increasing fund flows, institutional ownership, analyst following, antitakeover amendments and with the existence of a bond rating. The overall evidence suggests that investment–cash flow sensitivity decreases with factors that reduce capital market imperfections. � 2007 Elsevier B.V. All rights reserved.

JEL classification: G14; G31; G32

Keywords: Investment–cash flow sensitivity; Capital market imperfections

2 See Hubbard (1998) for a detailed review of this literature. 3 Allayannis and Mozumdar (2004) show that including firms with

negative cash flows can lead to these findings, since these firms are financially distressed and therefore their investments are not sensitive to cash flow. Rauh (2006) presents evidence that capital expenditures

1. Introduction

As argued by Modigliani and Miller (1958), the invest- ment decisions of firms are not affected by their financing decisions in perfect capital markets. Capital markets, how- ever, are not perfect, and existing imperfections introduce a wedge between the costs of external and internal funds. Firms facing higher informational imperfections experience a wider wedge, and therefore are more financially constrained.

A measure that has been used in the literature to assess the degree of financial constraints experienced by firms is the sensitivity of investments to the availability of internal funds, controlling for investment opportunities as mea- sured by Tobin’s Q. A number of studies, starting with Fazzari et al. (1988), show that investment is more sensitive

0378-4266/$ - see front matter � 2007 Elsevier B.V. All rights reserved. doi:10.1016/j.jbankfin.2007.02.013

* Corresponding author. Tel.: +1 202 994 9209; fax: +1 202 994 5014. E-mail addresses: [email protected] (S�. Ağca), [email protected] (A.

Mozumdar). 1 Tel.: +1 703 538 8414; fax: +1 703 538 8415.

to cash flow for firms that have a high degree of financial constraints.2 On the other hand, Kaplan and Zingales (1997) and Cleary (1999) show that investment–cash flow sensitivity can be higher for unconstrained firms.3 Addi- tionally, Gilchrist and Himmelberg (1995), Erickson and Whited (2000) and Alti (2003) argue that measurement problems associated with Tobin’s Q affect the estimated sensitivity of investments to the availability of internal funds.

decrease as internal funds reduce due to mandatory pension contributions. Moyen (2004) shows that different criteria used to differentiate between financially constrained and unconstrained firms can lead to results consistent either with Fazzari et al. (1988) or with Kaplan and Zingales (1997). According to Gomes (2001) and Alti (2003), investment–cash flow sensitivity can be positive even without any financial frictions.

208 S� . Ağca, A. Mozumdar / Journal of Banking & Finance 32 (2008) 207–216

There has been little investigation of the evolution of investment–cash flow sensitivities over time. The evidence that exists suggests that it has decreased over time in the US: while earlier papers in the area (Fazzari et al., 1988; Kaplan and Zingales, 1997), using data from the seventies and early eighties, have reported sensitivities in the (0.4, 0.7) range, studies employing data from the late eight- ies and nineties (Cleary, 1999; Erickson and Whited, 2000) have found sensitivities in the (0.1, 0.2) range. Therefore, in this paper, we first examine whether there is a decline in investment–cash flow sensitivity through time.

Applying the Erickson–Whited estimators to US manu- facturing firm data, we find that, consistent with Erickson and Whited (2000), cash flow is not significant in explaining investment for the period from 1992 to 1995. Extending the analysis to a much longer sample period (1970–2001), how- ever, we find that while the role of Q increases and that of cash flow declines (relative to OLS estimates) in explaining investment, cash flow continues to be a significant factor in most of the sub-samples examined.4 Of particular interest is the finding that the declining pattern of estimated cash flow effects is robust to the application of the Erickson–Whited estimators.

If investment–cash flow sensitivity is linked with capital market imperfections, then it should decrease with factors that reduce these imperfections. There is some interna- tional cross-sectional evidence to support this hypothesis. Wurgler (2000) examines cross-sectional data from 65 countries, and shows that capital allocation is more effi- cient in financially developed markets. Using cross-sec- tional data for several countries, Love (2003) and Islam and Mozumdar (2007) show that the sensitivity of invest- ment to cash decreases with financial market development. We pursue this analysis further by focusing on US firms and examining the relation between their investment–cash flow sensitivities and five factors related to capital market imperfections – aggregate fund flows, institutional owner- ship, analyst following, bond ratings, and corporate gover- nance.5 Again, to control for the measurement problems related to Tobin’s Q, we apply the GMM estimators of Erickson and Whited (2000) in addition to OLS.

4 Polk and Sapienza (2004) too find that cash flow effects remain significant when Erickson and Whited (2000, 2002) estimators are used. Also, using Erickson and Whited (2000) sample and estimators, Hennessy (2004) finds cash flow to be significant for firms with junk-rated debt, and insignificant for those with investment-grade debt.

5 Mitigation of capital market imperfections may have two effects: first, it may cause a reduction in the cost of internal funds, and second, it may cause a reduction in the wedge between the costs of external and internal funds. It is the second effect that is of interest from an investment–cash flow sensitivity viewpoint. The mechanism by which this effect may occur is also twofold. First, reduced capital market imperfections could induce a liquidity effect of greater supply of funds to primary capital markets that increases the total pool of external funds available to firms. Second, there could be a narrowing of the information gap between firm insiders and outsiders due to superior information production. The capital market factors that we consider seek to capture one or both of these mechanisms.

Our evidence suggests that the sensitivity of investment to internal funds decreases with factors that reduce capital market imperfections.6 Specifically, we find that invest- ment–cash flow sensitivity reduces with increasing fund flows, institutional ownership, analyst following, antitake- over amendments and with the existence of a bond rating. Therefore, the sensitivity of investments to the availability of internal funds cannot be explained solely as an artifact of measurement error.

The rest of the paper is organized as follows. Section 2 provides a brief summary of the basic q model of invest- ments, the measurement error problem in estimating it, as well as the major hypotheses. Section 3 describes the data. Section 4 analyzes the time series characteristics of investment–cash flow sensitivity. Section 5 examines the relation between investment–cash flow sensitivity and the factors associated with capital markets. Section 6 con- cludes the paper.

2. The q model of investments and hypotheses on investment– cash flow sensitivity in relation to capital market factors

2.1. The q model of investments and measurement error

As a remedy for the measurement error problem high- lighted in their critique, Erickson and Whited (2000, 2002) propose a class of GMM estimators that exploit the infor- mation in the higher order moments of the regression vari- ables. Using these estimators and a sample of US manufacturing firms over the period 1992–1995, they find that the explanatory power of Q improves dramatically rel- ative to traditional OLS estimates, while cash flow loses sig- nificance as a determinant of investment. Naturally then, the first question that needs to be settled is whether the investment–cash flow sensitivity estimates represent any- thing meaningful, or whether they are purely artifactual.

Consider the simplified version of the standard q invest- ment model presented by Erickson and Whited (2000, 2002). The firm chooses an investment policy to maximize the expected discounted value of cash flow subject to the law of motion for capital. Let It denote gross investment; Kt�1 beginning of period capital stock, and qt the marginal cost of capital In perfect markets, investments are deter- mined solely by the shadow price of capital, or marginal q. In particular, considerations of the availability of internal funds should play no role in the process. On the other hand, significant deviations from the perfect market paradigm would result in such considerations playing an important role. In empirical work, marginal q (qt) is typically approx- imated by Tobin’s average Q (Qt), while the most commonly used measure of internal funds is cash flow. Erickson and

6 Our findings, however, do not allow the inference that investment–cash flow sensitivities in the US have declined because capital market imperfections have decreased over time. While such an inference may appear plausible, it may only be reliably made on the basis of an explicit analysis of the temporal evolution of the relation between the two.

S� . Ağca, A. Mozumdar / Journal of Banking & Finance 32 (2008) 207–216 209

Whited (2000) list the several layers of approximation that lie between the marginal q of theory and the various ver- sions of Tobin’s q used in practice. Let Qt be the mismea- sured empirical proxy of the true marginal qt, i.e., Qt = a0 + qt + vt, with vt � i:i:d:ð0; r2vÞ. Then, the invest- ment equation in a regression framework is as follows:

I t K t�1

¼ða � b1a0Þþ b1Qt þ zt B þðut � b1vtÞ

¼ a þ b1Qt þ zt B þ et ð1Þ where z represents some measure(s) of internal funds.

Erickson and Whited (2000, 2002) show that the biases induced by measurement error in q in the above equation may be substantial and may be responsible for the estimated coefficients on Q being low and those on cash flow being high, as reported in earlier papers. They propose a class of measurement error-consistent GMM estimators that utilize the information in the higher order moments of the data. We use their GMM estimators in addition to OLS estima- tors to address problems related to measurement error. In addition to the usual assumption of independence between the errors (u, v) and the (true) regressors (z, q), Erickson and Whited (2000, 2002) approach also requires two techni- cal conditions to be satisfied for the model to be identified.7

2.2. Hypotheses on the relation of investment–cash flow

sensitivity to certain capital market factors

If investment–cash flow sensitivity is not simply an arti- fact of measurement error but rather linked to capital mar- ket imperfections, then it should decrease with factors that reduce these imperfections. We explore this hypothesis by examining the relation between investment–cash flow sensi- tivity and five factors related to capital market imperfec- tions – aggregate fund flows, institutional ownership, analyst following, bond ratings, and corporate governance.

Over the last 50 years, there has been a steady increase in investments through institutions such as mutual funds and pension funds. This increase in fund flows can be seen as a proxy for the increase in overall market liquidity. As doc- umented by Chordia et al. (2000), Hasbrouck and Seppi (2001) and Huberman and Halka (2001), there is common- ality in liquidity across assets, and the liquidity of an asset is positively related to market liquidity. In this respect, increasing fund flows (a proxy for increasing market liquid- ity) should increase liquidity across all assets. Additionally, institutions have better information processing skills, and

7 The two identification conditions require that E(b1) and E(g 3) be

significantly different from 0, where g is the residual from a linear projection of q on z. These two identification conditions are notoriously difficult to satisfy in the data (Erickson and Whited, 2000; Hennessy, 2004; Polk and Sapienza, 2004; Almeida and Murillo, forthcoming). We encounter the problem in some of our tests as well; we report GMM results only for those sub-samples for which the conditions are met. However, the GMM overidentifying restrictions (Hansen, 1982) do not hold for a number of identified samples. We report these results also with the corresponding J statistics.

therefore, reduce informational asymmetries. Thus, increased fund flows should reduce the external financing costs for all firms. Butler et al. (2005) present evidence sup- porting this hypothesis by examining seasoned equity offer- ings. They show that securities issuance costs are lower for firms with stocks that are more liquid, which reduces their cost wedge between external and internal funds. If this reduction is reflected in the sensitivity of investments to the availability of internal funds, then increasing fund flows should reduce this sensitivity. Therefore our first hypothe- sis is

H1: Investment cash flow sensitivity decreases with increasing fund flows.

An overall increase in fund flows reduces the liquidity- related imperfections for all firms by increasing overall market liquidity. However, the impact should be greater for firms with larger institutional shareholdings. Moreover, as documented by Bennet et al. (2003), institutions act on information. Therefore, firms with institutional ownership should have more information reflected into prices, which reduces the information asymmetry between outsiders and insiders and therefore the cost wedge between external and internal funds. If this reduction in the cost wedge is reflected in the sensitivity of investments to the availability of internal funds, one should observe a decrease in these sensitivities. Thus, our second hypothesis is

H2: Investment–cash flow sensitivity decreases with insti- tutional holdings.

Analyst coverage is another important factor related to capital market imperfections. There is an extensive body of literature linking analyst coverage with the flow of infor- mation to prices (see Chang et al., 2006 for a review). The incomplete information theory of Merton (1987) sug- gests that firms that are covered by a large number of ana- lysts should have lower costs of external capital. Consistent with this hypothesis, Leary et al. (2006) find that firms with more analyst coverage rely more on external funds due to reduced information asymmetries. If this lower premium for external funds translates into investment–cash flow sen- sitivities, then we should observe a decrease in these sensi- tivities with increased analyst following. Hence our third hypothesis states that

H3: Investment–cash flow sensitivity decreases as analyst following increases.

Bond ratings also have an impact on the cost wedge between external and internal funds. When a firm obtains a debt rating, it undergoes an independent evaluation that produces additional public information and thus mitigates its asymmetric information problem. Moreover, access to public bond markets extends the set of external financing choices available to the firm. Additionally, the junk bond market, which started in the early 1980s, has created an opportunity for firms without investment-grade bond rat- ings to raise capital through bond markets. The existence of a bond rating reduces the cost of external financing and thus eases the firm’s reliance on internal cash for mak- ing investments. Consequently, our fourth hypothesis is

210 S� . Ağca, A. Mozumdar / Journal of Banking & Finance 32 (2008) 207–216

H4: Investment–cash flow sensitivity decreases with the existence of bond ratings.

The final factor we examine is corporate governance. An important external control mechanism is the takeover mar- ket.8 The corporate governance index developed by Gom- pers et al. (2003) (GIM, henceforth), provides us a composite measure of corporate governance. The GIM index is formed by adding one point for each takeover defense provision9 that increases managerial power. There- fore, a high value for this index corresponds to strong man- agerial rights (low takeover vulnerability). While GIM find that operating performance is better for firms with strong shareholder rights, they fail to find a significant relation between the governance index and returns on equity. Klock et al. (2005) analyze the impact of corporate governance on the cost of debt financing using the same index. They find that high managerial rights reduce the cost of debt financ- ing.10 The firms with available GIM index data are generally large firms. Debt is the major source of external funds for these firms. Therefore, we would expect to find changes in investment–cash flow sensitivity to be in the same direction as the cost of debt financing. Thus, our final hypothesis is

H5: Investment–cash flow sensitivity decreases as take- over vulnerability decreases for firms that depend more on

debt financing.

3. Data and sample description

Our data come from six sources: COMPUSTAT, CRSP, the Federal Reserve’s Flow of Funds Account of the Uni- ted States, I/B/E/S, CDA/Spectrum, and the Investors Research Responsibility Center.

We use annual COMPUSTAT industrial files covering the years 1970 through 2001 as our primary source for US manufacturing firm data. We measure investment, cash flow, Tobin’s Q, and capital in the same manner as Kaplan and Zingales (1997). As in Kaplan and Zingales (1997), we deflate cash flow and investment by capital stock at the beginning of each year.

Aggregate fund flow data are taken from the Flow of Funds Accounts of the United States, which is published by the Federal Reserve Board. The publication contains quarterly data on sector holdings for each major asset class, starting with 1951. For the years 1970 through 2001, annual net fund flows are obtained by adding up the corresponding quarterly data on the net equity pur-

8 See Shleifer and Vishny (1997) and Denis and McConnell (2003) for surveys on corporate governance.

9 The provisions used in the index of Gompers et al. (2003) are from Corporate Takeover Defenses (Rosenbaum, 1990, 1993, 1995, 1998). 10 Several potential explanations for this finding are available in the

literature: takeover related wealth changes of bond holders and wealth transfer from bondholders to shareholders (Warga and Welch, 1993; Billet et al., 2004), managers having the opportunity to engage in longer term projects when shielded from takeover risks (DeAngelo and Edward, 1983; Stein, 1988), reduction in firm risk due to reduced job protection activities of managers when there are takeover defenses (Jensen and Ruback, 1983).

chased by insurance companies, pension funds, mutual funds and closed-end funds. Net fund flow values for each year are converted to 2001 values using the CPI index of the Bureau of Labor Statistics.

The data on analyst following are obtained from I/B/E/ S for the period 1976–2001. To get annual figures for the number of analysts following a firm, the last available quarterly data at or before that firm’s fiscal year end are used. If no data exist for a firm, the number of analysts fol- lowing that firm is taken to be zero.

Institutional ownership data for 1980 through 2001 are from the CDA/Spectrum database. This database provides quarterly reports on institutional holdings derived from the SEC’s 13(f) filings. Spectrum classifies each institution as one of five types: bank, insurance company, investment company, independent investment advisor or ‘other’. The institutional holdings of these institutions are added up in order to determine the institutional ownership of each stock. If a stock has no reported institutional ownership, we assume the institutional holdings of the stock to be zero.

Bond ratings data from 1985 to 2001 are taken from the annual COMPUSTAT industrial files. On the basis of bond ratings data, firms are divided into two categories – firms with rating and with no rating. Dummies are assigned accordingly.11

Annual data on takeover defenses for the period 1990– 2001 are taken from the Investors Research Responsibility Center. GIM created an annual corporate governance index for US firms based on five governance rules that consist of a total of 24 takeover defense provisions We consider these 24 provisions to form the index as in GIM. The value of the index varies between 2 and 18. A high governance index value corresponds to strong managerial rights (or low take- over vulnerability) and weak shareholder rights. The annual index values are used as a measure of corporate governance. If the governance index of a firm is missing for a year, the index value of the previous year is used.

Since all the data are not available for the overall sample period 1970–2001, we report the results for the 1985–2001 and 1990–2001 periods.12 We also examine the time series pattern of investment–cash flow sensitivity by running ten year rolling regressions of investment on cash flow and Tobin’s Q from 1970 to 2001. Additionally, we look at the period from 1992 to 1995, since this is the sample ana- lyzed in Erickson and Whited (2000). We omit firm-years with negative cash flows on account of their potential dis- tortionary impact on investment–cash flow sensitivity esti- mates.13 Additionally, in order to reduce the effects of

11 We carry out the analyses using all junk bonds as well as by removing those that have a below-CCC rating. The latter is carried out to remove the impact of financial distress. The results are comparable to those reported. 12 We also examine the periods 1976–2001 and 1980–2001 with available

variables for these periods. The results are comparable to those reported for the available variables. 13 See Allayannis and Mozumdar (2004) and Bhagat et al. (2005).

Table 1 Descriptive statistics

Obs Total assets I CF Q Fund flows Analysts IO Bond Governance index

1985–2001 21,944 2245.77 0.27 0.62 1.61 5.44 5.62 0.031 0.24 169.29 0.21 0.40 1.32 5.29 2.00 0.002

1990–2001 14,935 2651.88 0.27 0.66 1.68 7.12 5.57 0.030 0.26 203.21 0.20 0.41 1.36 8.50 2.00 0.002

1990–2001a 5848 3606.33 0.23 0.49 1.79 7.34 10.48 0.06 0.50 9.47 993.51 0.20 0.38 1.49 8.88 8.00 0.014 10.00

This table reports the mean and median values of the variables for the 1985–2001 and 1990–2001 sample periods. The Obs column shows the sample size for each period. For the period 1990–2001, there are two samples. When we require a firm to have a corporate governance index, the sample size reduces to the one reported at the bottom of the table. I, CF, and Q represent investment, cash flow, and Tobin’s Q, respectively, calculated as in Kaplan and Zingales (1997). Fund flows are calculated using the 1970 values as the base. Fund flows are converted to 2001 values using the CPI index of the Bureau of Labor Statistics. The analysts column shows the analyst following for the sample firms. The number of analysts following a firm is divided by the mean number of analysts following for each year. The IO column reports the proportion of the institutional ownership of the sample firms relative to the institutional ownership of all firms with fiscal year ends corresponding to the same quarter. The Bond column reports the proportion of firms that have bond ratings. The Govn column reports the corporate governance index, and is taken from GIM. The corporate governance index value of each firm is divided by the mean corporate governance index for that year. For each variable except bond ratings, the first row reports the mean values and the second row the median values.

a Sample size is reduced due to the exclusion of firms that do not have corporate governance index data available.

S� . Ağca, A. Mozumdar / Journal of Banking & Finance 32 (2008) 207–216 211

extreme observations, for each sub-period, the top and bot- tom one percentile of the data with respect to investment, cash flow, and Tobin’s Q are removed. We require a firm to have at least three years of data to be in the sample.14

Since investment and cash flow are deflated by capital stock, and Tobin’s Q is a ratio, we convert all variables to ratios. With respect to fund flows, the net fund flow value for 1970 serves as a base and is assigned a value of one. The fund flow values for subsequent years are calcu- lated relative to this base. We divide the number of analysts following a firm each year by the mean number of analysts following stocks for that year. This ratio gives us a relative figure for analyst following of a firm compared to other firms in that year. This allows us to distinguish reduced information asymmetries due to analyst following across firms in a given year. For similar reasons, for each year, the institutional ownership of a firm is divided by the total institutional holdings for that year, depending on the fiscal year end of the firm. The first quarterly report on institu- tional holdings at or after the firm’s fiscal year end is used to calculate the ratio. Again, to get a relative figure for cor- porate governance index for each firm in a year, each firm’s annual governance index value is divided by the mean gov- ernance index for that year.

Table 1 reports descriptive statistics for these variables. As can be observed in Table 1, investment, cash flow, ana- lyst following, institutional ownership and percentage of firms with bond ratings are fairly stable throughout the 1985–2001 and 1990–2001 periods. Firm size and fund flows, on the other hand, are higher for the 1990–2001 per- iod compared to 1985–2001 period. The governance index values are similar to those reported in GIM. Also, as

14 We require three years of data to calculate beginning-of-year Q and capital stock as well as to examine first difference regressions. First difference regressions are not reported since the results are comparable to those reported.

expected, when we require firms to have governance index values, the sample size is reduced. The proportion of large firms with bond ratings, large institutional ownership and number of analysts following increases for this sample.

4. The time series pattern of investment–cash flow sensitivity

Using data from the late eighties and early nineties, Cleary (1999) and Erickson and Whited (2000) report investment–cash flow sensitivities that are much lower than those estimated by Fazzari et al. (1988) and Kaplan and Zingales (1997), who used data from the 1970–1984 period. To examine if a decline is observed in our sample, we run rolling regressions from 1970 to 2001 for overlapping peri- ods of ten years. This allows us to analyze the time series pattern of investment–cash flow sensitivity. Our first regression is for the period 1970–1979, the second for the period 1971–1980, and so forth. We estimate the following regression:

I t=K t�1 ¼ a þ b1Qt�1 þ b2CF t=K t�1 þ et; ð2Þ where It, and CFt represent investment and cash flow dur- ing period t, respectively; Kt�1 is the amount of fixed cap- ital at the beginning of period t; and Qt is Tobin’s Q, calculated at the beginning of period t. b2 measures invest- ment–cash flow sensitivity for the period t. The data are transformed into their mean deviation form by taking the differences between the raw data and the firm-level means. This allows us to eliminate fixed firm effects. We include year dummies to control for year effects. Heteroscedasticity correction is used for estimating standard errors.

We report the results based on OLS estimation as well as the measurement error-consistent GMM estimation of Erickson and Whited (2000). GMM estimates based on product moments up to third and fourth orders (GMM3 and GMM4, respectively) are reported for the identified models.

Table 2 Time series pattern of investment–cash flow and Tobin’s Q sensitivities

Obs OLS GMM3 GMM4

R 2 CF Q R2 CF Q R2 CF Q J statistics

Panel A: 1979–2001 Rolling regressions

1979 12,347 0.187 0.28 0.053 0.251 0.157 0.352 0.233 0.177 0.271 1.356 (0.013) (0.005) (0.03) (0.113) (0.015) (0.032) (0.508)

1980 13,376 0.168 0.259 0.057 0.231 0.149 0.314 0.223 0.159 0.278 0.469 (0.011) (0.004) (0.024) (0.083) (0.015) (0.036) (0.791)

1981 13,803 0.158 0.243 0.06 0.22 0.152 0.294 0.22 0.153 0.291 0.054 (0.011) (0.004) (0.02) (0.072) (0.015) (0.04) (0.973)

1982 14,072 0.151 0.234 0.057 0.214 0.150 0.316 0.221 0.147 0.331 0.84 (0.01) (0.005) (0.024) (0.107) (0.015) (0.05) (0.657)

1983 14,237 0.16 0.238 0.055 0.193 0.155 0.204 0.23 0.147 0.354 3.23 (0.01) (0.006) (0.021) (0.098) (0.014) (0.049) (0.198)

1984 14,398 0.168 0.241 0.055 0.208 0.156 0.297 0.26 0.119 0.484 8.61 (0.01) (0.006) (0.047) (0.13) (0.018) (0.068) (0.014)

1985 14,474 0.17 0.236 0.067 0.228 0.135 0.364 0.259 0.117 0.447 7.00 (0.01) (0.006) (0.056) (0.145) (0.017) (0.054) (0.03)

1986 14,476 0.163 0.228 0.073 0.211 0.145 0.284 0.27 0.106 0.448 13.49 (0.01) (0.007) (0.054) (0.128) (0.017) (0.052) (0.001)

1987 14,495 0.165 0.216 0.072 0.207 0.138 0.245 0.271 0.093 0.414 14.67 (0.011) (0.007) (0.058) (0.102) (0.019) (0.054) (0.001)

1988 14,497 0.168 0.2 0.074 0.202 0.133 0.207 0.281 0.079 0.398 18.14 (0.01) (0.006) (0.049) (0.098) (0.017) (0.044) (0.00)

1989 14,399 0.168 0.185 0.081 0.217 0.112 0.230 0.29 0.067 0.378 15.59 (0.01) (0.006) (0.038) (0.101) (0.015) (0.036) (0.00)

1990 14,248 0.159 0.167 0.077 0.232 0.083 0.266 0.285 0.055 0.352 11.38 (0.009) (0.006) (0.034) (0.102) (0.014) (0.033) (0.003)

1991 14,099 0.162 0.155 0.080 0.250 0.067 0.294 0.291 0.050 0.350 9.193 (0.009) (0.005) (0.032) (0.097) (0.014) (0.034) (0.01)

1992 14,017 0.165 0.144 0.071 0.263 0.074 0.283 0.308 0.056 0.342 9.770 (0.008) (0.005) (0.028) (0.087) (0.014) (0.035) (0.008)

1993 13,939 0.17 0.138 0.068 0.287 0.055 0.341 0.320 0.043 0.381 6.350 (0.008) (0.005) (0.031) (0.101) (0.016) (0.042) (0.042)

1994 13,876 0.166 0.128 0.067 0.313 0.037 0.429 0.306 0.040 0.376 8.220 (0.007) (0.005) (0.022) (0.122) (0.015) (0.041) (0.016)

1995 13,834 0.159 0.121 0.065 0.315 0.038 0.450 0.287 0.043 0.357 6.670 (0.006) (0.005) (0.028) (0.129) (0.013) (0.037) (0.036)

1996 13,825 0.152 0.110 0.065 0.294 0.037 0.383 0.283 0.041 0.333 6.260 (0.006) (0.004) (0.026) (0.124) (0.013) (0.037) (0.044)

1997 13,702 0.149 0.102 0.064 0.262 0.045 0.308 0.272 0.043 0.315 0.263 (0.006) (0.004) (0.024) (0.083) (0.012) (0.038) (0.278)

1998 13,460 0.14 0.098 0.060 0.256 0.031 0.324 0.277 0.026 0.341 5.973 (0.005) (0.004) (0.033) (0.12) (0.013) (0.039) (0.05)

1999 13,178 0.146 0.097 0.057 0.250 0.038 0.295 0.278 0.029 0.325 6.490 (0.005) (0.004) (0.022) (0.11) (0.013) (0.038) (0.04)

2000 12,846 0.148 0.091 0.054 0.223 0.053 0.223 0.267 0.032 0.293 3.610 (0.005) (0.004) (0.026) (0.084) (0.014) (0.038) (0.165)

2001 12,310 0.146 0.085 0.052 0.215 0.062 0.195 0.284 0.028 0.309 3.810 (0.005) (0.003) (0.022) (0.07) (0.014) (0.041) (0.149)

Panel B: 1992–1995 Regression

5135 0.139 0.117 0.061 0.162 0.100 0.119 0.286 0.023 0.374 5.75 (0.01) (0.007) (0.066) (0.022) (0.022) (0.055) (0.056)

This table presents the sensitivity of investment to internal funds and Tobin’s Q. We run rolling regressions from 1970 to 2001 for overlapping periods of ten years according to the following regression model:

I t=K t�1 ¼ a þ b1Qt þ b2CF t=K t�1 þ et; where It, and CFt represent investment and cash flow during period t, respectively; Kt�1 is the amount of capital at the beginning of period t; and Qt is Tobin’s Q, calculated at the beginning period t as in Kaplan and Zingales (1997). Panel A reports the rolling ten year regressions from 1970 to 2001. The first regression is from 1970 to 1979, and is reported below as 1979. The second is from 1971 to 1980, and is reported as 1980, and so forth. Panel B reports the regression for the period from 1992 to 1995. In the regressions, the data are transformed into their mean deviation form by taking the differences between the raw data and the firm-level means so as to eliminate the fixed-firm effects. We keep the year dummies to control for the year effects. OLS and GMM results are reported. In the columns below, Obs represents the number of observations. Heteroscedasticity adjusted standard errors are in parentheses.

212 S� . Ağca, A. Mozumdar / Journal of Banking & Finance 32 (2008) 207–216

16 Business cycle expansions and recessions are as reported by the NBER

0

0.05

0.1

0.15

0.2

0.25

0.3

1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001

CF_OLS CF_GMM

Regression Coefficients on Tobin's Q (Q)

0 0.05 0.1

0.15 0.2

0.25 0.3

0.35 0.4

0.45 0.5

1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001

Q_OLS Q_GMM

Regression Coefficients on Cash Flow (CF)

Fig. 1. Regression Coefficients on Cash Flow (CF) and Tobin’s Q. Investment is regressed on CF and Q over ten year periods from 1970 to 1992, i.e. 1970– 1979, 1971–1980, and so forth. In the above figure, years correspond to the last years in the rolling regression. For example, the period 1970–1979 is shown as 1979. Both OLS (CF_OLS and Q_OLS) and GMM3 (CF_GMM and Q_GMM) estimates are shown.

S� . Ağca, A. Mozumdar / Journal of Banking & Finance 32 (2008) 207–216 213

Table 2 and Fig. 1 present the results of the above regressions for the overall sample. In Table 2, Panel A reports the results of ten year rolling regressions from 1970 to 2001, and Panel B reports the results of the regres- sion for the sample period from 1992 to 1995 for compar- ison with Erickson and Whited (2000). In line with the results of Erickson and Whited (2000), the application of GMM estimators results in the coefficient on Tobin’s Q being more than three times higher and the coefficient on cash flow being lower than those obtained with OLS esti- mation. For the period from 1992 to 1995, as in Erickson and Whited (2000), cash flow is not significant when GMM estimators are used. Over most of the other years, however, the estimated investment–cash flow sensitivity, b2, is positive and significant, indicating that the Erickson and Whited (2000) sample period was somewhat special in this respect. Only those results for which both the iden- tification and the over-identification criteria are satisfied, as well as the coefficient is significantly different from 0, are reported in bold.15 Further, there is a clear and steady decrease in the estimated sensitivity, b2, over time for most of the sub-samples examined.

Interestingly, the decrease in investment–cash flow sensi- tivity b2 is not associated with any corresponding pattern in

15 Note that this means that the identification test rejects, the over- identification test fails to reject, and the coefficient t-test rejects.

b1, the sensitivity to Tobin’s Q. A possible alternative explanation for the decline in b2 is that Tobin’s q became a less noisy proxy for marginal q over time, such that the incorrectly inflated role of cash flow also reduced. How- ever, since the measurement-error-consistent estimators also yield a declining pattern for b2, and the pattern for b1 remains stable, the decline in b2 cannot be explained by a reduction in measurement error alone. Furthermore, since the sample period spans several business cycles, and almost every ten year sub-sample covers at least one reces- sion and one expansion, these results cannot be explained by business cycle fluctuations.16 Thus, the evidence lends support to the hypothesis that the estimated sensitivity b2 is decreasing through time and is significant most of the time even after controlling for measurement errors.

5. Investment–cash flow sensitivity in relation to capital

market factors

Since the sensitivity of investment to the availability of internal funds is significant (although decreasing) through

at its web site, http://www.nber.org/cycles.html. The only ten year period considered in our rolling regressions that does not cover both an expansion and a recession is March 1991 to March 2001, which is a 120 month expansionary period.

Table 3 Capital market imperfections and investment–cash flow sensitivity

Obs J statistics R2 CF Q CF * Size CF * Fund CF * Analyst CF * IO CF * Bond CF * Govn

1985–2001 21,944 OLS 0.154 0.107 0.057

(0.005) (0.003)

0.167 0.099 0.056 0.008 �0.003 0.007 �0.115 �0.023 (0.012) (0.003) (0.003) (0.001) (0.005) (0.037) (0.016)

GMM3 0.282 0.034 0.295 (0.015) (0.048)

0.299 0.061 0.310 0.007 �0.004 �0.041 �0.569 �0.028 (0.016) (0.049) (0.003) (0.001) (0.012) (0.119) (0.016)

GMM4 4.974 0.287 0.036 0.289 (0.083) (0.008) (0.023) 2.63 0.288 0.064 0.284 0.007 �0.004 �0.036 �0.552 �0.029

(0.26) (0.015) (0.024) (0.004) (0.001) (0.008) (0.086) (0.017)

1990–2001 14,935 OLS 0.142 0.088 0.051

(0.005) (0.003)

0.153 0.084 0.052 0.008 �0.003 0.009 �0.11 �0.037 (0.016) (0.004) (0.004) (0.001) (0.005) (0.038) (0.016)

GMM3 0.247 0.030 0.270 (0.018) (0.08)

0.295 0.074 0.322 0.006 �0.006 �0.041 �0.602 �0.022 (0.016) (0.05) (0.004) (0.001) (0.012) (0.121) (0.011)

GMM4 3.811 0.274 0.026 0.300 (0.149) (0.011) (0.032) 2.043 0.283 0.078 0.291 0.006 �0.006 �0.035 �0.546 �0.022

(0.36) (0.015) (0.029) (0.004) (0.001) (0.008) (0.087) (0.011) 1990–2001a 5848 OLS 0.177 0.041 0.038 0.018 �0.004 0.006 �0.063 �0.021 �0.053

(0.01) (0.004) (0.009) (0.002) (0.007) (0.032) (0.011) (0.024)

This table shows the impact of capital market imperfections on investment–cash flow sensitivity, controlling for firm size, over two sample periods. The periods are 1985–2001, and 1990–2001. Data are available for fund flows for all subperiods, analyst following only after 1976, institutional ownership only after 1980, bond ratings only after 1985, and the corporate governance index only after 1990. We run the following regression:

I t=K t�1 ¼ a þ b1Qt þ b2CF t=K t�1 þ ciðCF t � FactoriÞþ et; where It, and CFt represent investment and cash flow during period t, respectively; Kt�1 is the amount of capital at the beginning of period t; and Qt is Tobin’s Q, calculated at the beginning period t in the same manner as in Kaplan and Zingales (1997). We control for size by interacting cash flow with the natural logarithm of size (CF * Size). CFt * Factort are the interactions of cash flow with the factors related to capital market imperfections. The factors considered are: fund flows (CF * Fund), analyst following (CF * Analyst), institutional ownership (CF * IO), the proportion of firms with bond ratings (CF * Bond), and the corporate governance index (CF * Govn). In the regressions, the data are transformed into their mean deviation form by taking the differences between the raw data and the firm-level means so as to eliminate the fixed-firm effects. We keep the year dummies to control for the year effects. OLS and GMM results are reported. Heteroscedasticity adjusted standard errors are in parentheses.

a Sample size is reduced from 14,935 to 5,848 due to firms that do not have GIM index data available.

17 We also examine 1970–2001, 1976–2001 and 1980–2001 periods. The results are comparable to those reported for the 1985–2001 and 1990–2001 periods.

214 S� . Ağca, A. Mozumdar / Journal of Banking & Finance 32 (2008) 207–216

different time periods, it is not merely an artifact of mea- surement errors, but is linked to capital market imperfec- tions. Therefore it should decrease with factors that reduce these imperfections. We consider factors that prior research has identified as being related to capital market imperfections, and examine the impact of these factors on investment–cash flow sensitivity. Our regression model is as follows:

I t=K t�1 ¼ a þ b1Qt�1 þ b2CF t=K t�1 þ ciðCF t=K t�1 � FactoriÞþ et: ð3Þ

The factors considered in the regression are fund flows, analyst following, institutional ownership, bond ratings, and the corporate governance index. The data are trans-

formed into their mean deviation form by taking the differ- ences between the raw data and the firm-level means to eliminate fixed firm effects. We include year dummies to control for year effects. Heteroscedasticity correction is used for estimating standard errors.

We analyze the interaction of cash flow with these fac- tors, and report the results for the 1985–2001 and 1990– 2001 periods.17 When the interaction of cash flow with the corporate governance index is considered in 1990– 2001 period, the sample size is reduced due to unavailabil-

S� . Ağca, A. Mozumdar / Journal of Banking & Finance 32 (2008) 207–216 215

ity of the index value for many small firms. We include the interaction of cash flow with the natural logarithm of size (book value of total assets) as an additional control vari- able.18 The results are given in Table 3. OLS estimates as well as the GMM estimates based on Erickson and Whited (2000) are reported.

In Table 3, we observe a negative and significant coeffi- cient on the interaction of cash flow with fund flows for all periods. Consistent with hypothesis stated in Section 2, these results suggest that the sensitivity of investments to the availability of internal funds decreases with factors that reduce capital market imperfections. The coefficient on the interaction of cash flow with firm size is positive whenever it is significant, although it is largely insignificant with GMM estimates. The finding that the coefficient is largely insignificant with the GMM estimates but not with the OLS estimates may be indicative of size capturing some unobserved factors related to Tobin’s Q that are not cap- tured by OLS estimates due to measurement error.

Although fund flows have an impact on the overall mar- ket, we would expect the largest effect to be on firms with higher levels of institutional ownership. These firms should be able to raise external funds more easily, since increased liquidity as well as reduced informational asymmetries through information flow to prices should reduce the cost wedge between external and internal funding. In Table 3, we find that the coefficient on the interaction of cash flow with institutional ownership is negative and significant.

Stock analysts produce research that adds to the pub- licly available information about the firm and thus reduces the information asymmetry between insiders and outsiders. Thus analyst following should reduce the wedge between internal and external financing costs. Consistent with this hypothesis, in Table 3, GMM estimates of the coefficient on the interaction of cash flow with the number of analysts following are negative and significant.

Debt markets play a major role in the external financing of corporations. The credit rating process generates addi- tional public information about the firm as well as making public bond markets accessible to it. In Table 3, we observe that the interaction of cash flow and the bond rating dummy is always negative and significant. These results are supportive of an association between a decreasing investment–cash flow sensitivity and a decreasing cost wedge between internal and external funds due to access to bond markets.

Klock et al. (2005) find that the cost of debt financing decreases with an increase in the GIM index. Using institu- tional ownership as a proxy for shareholder control, Cre- mers et al. (forthcoming) show that antitakeover amendments lead to a decrease in the cost of debt financing when there is strong shareholder control. We examine the

18 We control for firm size since Gilchrist and Himmelberg (1995), Kadapakkam et al. (1998), Erickson and Whited (2000), and Allayannis and Mozumdar (2004) find a significant relation between firm size and financing constraints.

relation between investment–cash flow sensitivity and the GIM index for the sample period 1990–2001. Most of the firms in this sample are large firms (Table 1). Debt financ- ing is the major source of external financing for large firms. Thus the cost of debt financing should have a larger impact on the cost wedge between internal and external funds for the firms in our sample. We report OLS estimates in Table 3. GMM estimates are not reported since the correspond- ing GMM models are not identified. The coefficient on the corporate governance index is negative and significant. The evidence thus indicates a negative relation between investment–cash flow sensitivity and the number of anti- takeover amendments. However, since OLS estimates may not be reliable due to measurement error problems related to Tobin’s Q, this evidence can at best be seen as weak support for the hypothesized negative relation between investment–cash flow sensitivity and the GIM index.

6. Conclusion

This paper examines the sensitivity of investments to the availability of internal funds using US manufacturing firm data. The paper first shows that there has been a steady decrease in investment–cash flow sensitivity over time, and that this decline cannot be explained on the basis of measurement error alone. Next, investment–cash flow sen- sitivity is examined in relation to five capital-market- related factors: fund flows, institutional ownership, analyst following, bond ratings, and antitakeover amendments. By analyzing the relation of investment–cash flow sensitivity to these five capital-market-related factors, we explore whether this sensitivity reduces with factors that decrease capital market imperfections.

Our findings indicate that investment–cash flow sensitiv- ity decreases when there is a reduction in capital market imperfections through increased fund flows, institutional ownership, analyst following, antitakeover amendments and with the existence of a bond rating. Thus capital mar- ket factors that decrease market imperfections and as a result reduce the cost wedge between external and internal funds, lead to lower investment–cash flow sensitivities. It is possible that the decline in investment–cash flow sensitivity has been due to reduction in capital market imperfections over time. However, such an inference cannot be made cat- egorically without a direct time series analysis of the rela- tion between the two. We leave that exercise for future research.

Acknowledgements

We are grateful to Andrew Metrick for kindly providing us with the corporate governance data, and to Toni Whited for kindly making available on her web page her GAUSS programs for implementing certain tests and estimators. We thank Gonul Colak, David Dobofsky, Gayane Hova- kimian, Florian Heider, Saiyid Islam, Christopher Jones,

216 S� . Ağca, A. Mozumdar / Journal of Banking & Finance 32 (2008) 207–216

Mark Klock, Amnon Levy, Erlend Nier, Oguzhan Ozbas, Bernell Stone, and the seminar participants at the 2004 European Finance Association meeting, the 2004 Washing- ton Area Finance Conference, and 2005 European Finan- cial Management Association meeting for helpful comments. Ağca acknowledges a summer research grant from the School of Business, George Washington Univer- sity. Mozumdar likewise acknowledges a summer research grant from the Pamplin College of Business, Virginia Tech.

References

Allayannis, George, Mozumdar, Abon, 2004. The investment–cash flow sensitivity puzzle: Can negative cash flow observations explain it? Journal of Banking and Finance 28, 901–930.

Almeida, Heitor, Murillo, Campello, forthcoming. Financial constraints, asset tangibility, and corporate investment. Review of Financial Studies.

Alti, Aydogan, 2003. How sensitive is investment to cash flow when financing is frictionless? Journal of Finance 58, 707–722.

Bennet, James A., Sias, Richard W., Starks, Laura T., 2003. Greener pastures and the impact of dynamic institutional preferences. Review of Financial Studies 16, 1203–1238.

Bhagat, Sanjai, Moyen, Nathalie, Suh, Inchul, 2005. Investment and internal funds of distressed firms. Journal of Corporate Finance 11, 449–472.

Billet, Matthew T., Dolly, King, Mauer, David C., 2004. Bondholders wealth effects in mergers and acquisitions: New evidence from the 1980s and 1990s. Journal of Finance 59, 107–135.

Butler, Alex, Gustavo, Grullon, Weston, James, 2005. Stock market liquidity and the cost of issuing equity. Journal of Financial and Quantitative Analysis 40, 331–348.

Chang, Xin, Dasgupta, Sudipto, Hilary, Gilles, 2006. Analyst coverage and financing decisions. Journal of Finance 61, 3009–3048.

Chordia, Tarun, Roll, Richard, Subrahmanyam, Avanidhar, 2000. Com- monality in liquidity. Journal of Financial Economics 56, 3–28.

Cleary, Sean, 1999. the relationship between firm investment and financial status. Journal of Finance 54, 673–692.

Cremers, Martijn, Nair, Vinay, Wei, Chenyang, forthcoming. Governance mechanisms and bond prices. Review of Financial Studies.

DeAngelo, Harry, Edward, Rice, 1983. Antitakeover charter amendments and stockholder wealth. Journal of Financial Economics 11, 329–359.

Denis, Diane K., McConnell, John J., 2003. International corporate governance. Journal of Financial and Quantitative Analysis 38, 1–36.

Erickson, Timothy, Whited, Toni M., 2000. Measurement error and the relationship between investment and q. Journal of Political Economy 108, 1027–1057.

Erickson, Timothy, Whited, Toni M., 2002. Two-step GMM estimation of the errors-in-variables model using high-order moments. Econometric Theory 18, 776–799.

Fazzari, Steven M., Glenn Hubbard, R., Petersen, Bruce P., 1988. Financing constraints and corporate investment. Brookings Papers on Economic Activity, 141–195.

Gilchrist, Simon, Himmelberg, Charles P., 1995. Evidence on the role of cash flow for investment. Journal of Monetary Economics 36, 541–572.

Gomes, Joao, 2001. Financing investment. American Economic Review 91, 1263–1285.

Gompers, Paul A., Ishii, Joy L., Metrick, Andrew, 2003. Corporate governance and equity prices. Quarterly Journal of Economics 118, 107–155.

Hansen, Lars, 1982. Large sample properties of generalized method of moments estimators. Econometrica 50, 1029–1054.

Hasbrouck, Joel, Seppi, Duane, 2001. Common factors in prices, order flows, and liquidity. Journal of Financial Economics 59, 383–411.

Hennessy, Christopher A., 2004. Tobin’s Q, debt overhang, and invest- ment. Journal of Finance 59, 1717–1742.

Hubbard, Glenn R., 1998. Capital market imperfections and investment. Journal of Economic Literature 36, 193–225.

Huberman, Gur, Halka, Dominic, 2001. Systemic liquidity. Journal of Financial Research 24, 161–178.

Islam, Saiyid S., Mozumdar, Abon, 2007. Financial development and the importance of internal cash: Evidence from international data. Journal of Banking and Finance 31, 641–658.

Jensen, Michael C., Ruback, Richard S., 1983. The market for corpo- rate control: The scientific evidence. Journal of Financial Economics 11, 5–50.

Kaplan, Steven N., Zingales, Luigi, 1997. Do investment–cash flow sensitivities provide useful measures of financing constraints? Quar- terly Journal of Economics, 169–215.

Leary, Mark T., Roberts, Michael R., 2006. The pecking order, debt capacity, and information asymmetry. Working Paper, University of Pennsylvania.

Love, Inessa, 2003. Financial development and financing constraints: International evidence from the structural investment model. Review of Financial Studies 16, 765–791.

Klock, Mark, Mansi, Sattar A., Maxwell, William F., 2005. Does corporate governance matter to bondholders? Journal of Financial and Quantitative Analysis 40, 693–720.

Merton, Robert, 1987. A simple model of capital market equilibrium with incomplete information. Journal of Finance 42, 483–510.

Modigliani, Franco, Miller, Merton Howard, 1958. The cost of capital corporation finance, and the theory of investment. American Eco- nomic Review 48, 261–297.

Moyen, Nathalie, 2004. Investment–cash flow sensitivities: Constrained versus unconstrained firms. Journal of Finance 59, 2061–2092.

Kadapakkam, Palani-Rajan, Kumar, P.C., Riddick, Leigh A., 1998. The impact of cash flows and firm size on investment: The international evidence. Journal of Banking and Finance 22, 293–320.

Polk, Christopher K., Sapienza, Paola, 2004. Real Effects of Investor Sentiment, Working Paper, Northwestern University.

Rauh, Joshua, 2006. Investment and financing constraints: Evidence from the funding of corporate pension plans. Journal of Finance 61, 33–71.

Rosenbaum, Virginia, 1990, 1993, 1995, 1998. Corporate Takeover Defenses, Investor Responsibility Research Center, Washington, DC.

Shleifer, Andrei, Vishny, Robert W., 1997. A survey of corporate governance. Journal of Finance 52, 737–784.

Stein, Jeremy C., 1988. Takeover threats and managerial myopia. Journal of Political Economy 96, 61–80.

Warga, Arthur, Welch, Ivo, 1993. Bondholder losses in leveraged buyouts. Review of Financial Studies 6, 959–982.

Wurgler, Jeffrey, 2000. Financial markets and the allocation of capital. Journal of Financial Economics 58, 187–214.