3000 word report based on two related papers

范菁仪�

F. Moshirian et al. / Journal of Banking and Finance 77 (2017) 1–17 5

t

(

i

d

i

t

e

m

t

H

d

t

a

e

t

a

o

H

i

p

i

i

g

i

f

t

g

s

n

fi

2

t

c

fl

t

t

i

i

o

s

H

t

H

o

i

m

c

m

c

c

t

c

f

t

3

l

I

(

m

i

t

c

v

t

t

b

Z

e

fi

c

t

v

X

c

r

a

c

s

fl

i

i

i

fl

U

fl

t

s

t

c

p

11 Fazzari et al. (1988) scale their regression variables by physical assets, while we

o knowledge-based capital. In addition, U.S. firms hoard liquidity

cash) reserves, well above the historical average, and depress cap-

tal investment ( Bates et al., 2009; Pinkowitz et al., 2012 ). 10 Other

eveloped economies have experienced somewhat similar changes

n production technologies, product markets and capital structures.

We hypothesize that ICFS reflects this long-lasting transi-

ion in the U.S. economic structure and that in other developed

conomies. We identify the effect of asset tangibility on invest-

ent and highlight its importance in predicting ICFS, sidestepping

he wedge between the cost of internal and external funds.

1: Declining asset tangibility and falling investment contribute to the

ecline in ICFS in developed economies .

Developing economies are structurally different. Their produc-

ive capital structures are more tangible because their firms oper-

te with more tangible capital. Capital investment rates are gen-

rally higher. Over time, their levels of tangible capital stock also

end to decline due to productivity enhancement induced by R&D

nd new product markets but at a slower pace. Therefore, our sec-

nd hypothesis is as follows:

2: Higher asset tangibility and more stable investment support ICFS

n developing economies .

The second channel is based on the notion that cash flow ex-

lains investment because it predicts future cash flow and because

nvestment is made in pursuit of future income. High predictabil-

ty of cash flow causes firms to invest more for future income

rowth ( Riddick and Whited, 2009 ). Tangible and intangible cap-

tal are conceived as inputs in production and contribute to the

uture cash flow of the firm. The uncertainty associated with in-

angible capital is higher than that of physical capital; thus, intan-

ible capital is expensed and written off from the firm’s balance

heet. Because intangible capital has become an important compo-

ent of production and the “new economy” environment in which

rms operate is more competitive and complicated ( Hansen, et al.,

005 ), cash flow has become less predictable over time. Because

here is less information in the current cash flow regarding future

ash flow, investment becomes less dependent on the current cash

ow.

We hypothesize that ICFS reflects changes in cash flow persis-

ence. Over time, the declining share of tangible capital attenuates

he persistence of cash flow and contributes to the decline in ICFS

n advanced economies. In contrast, more tangible asset structures

mprove the predictability of cash flow and support ICFS in devel-

ping economies. Therefore, our third and fourth testable hypothe-

es are as follows:

3: Declining asset tangibility contributes to the decline in informa-

ion content of cash flow in developed economies.

4: Higher asset tangibility contributes to stable information content

f cash flow in developing economies.

Under the proposed identification, asset tangibility affects the

nvestment rate and the information content in cash flow. Invest-

ent varies with tangible capital because the share of tangible

apital in total productive capital contains information about its

arginal productivity. Investment varies with cash flow because

ash flow provides information about marginal q with respect to

apital productivity. As we confirm below, only high tangible capi-

al firms have cash flow that explains investment beyond its simple

orrelation with q .

10 According to the World Bank, at the macro level, the ratio of gross fixed capital

ormation to GDP in OECD countries has fallen from nearly 25% in the early 1970s

o 22% in the early 20 0 0s and 19% in 2013.

s

r

U

e

t

t

b

.2. Research design

The implications of our hypotheses are formalized in the fol-

owing model specifications:

I i,t = αc, j,t + β0 + β1 q i,t−1 + β2 �i,t + ε i,t (2)

i,t = αc, j,t + β0 + β1 q i,t−1 + β2 �i,t + β3 �i,t Z i,t−1 + β4 Z i,t−1 + β5 I i,t−1 + β6 X i,t + ε i,t (3)

I i,t = αc, j,t + β0 + β1 q i,t−1 + β2 q i,t−1 Z i,t−1 + β3 �i,t + β4 �i,t Z i,t−1 + β5 I i,t−1 + β6 I i,t−1 Z i,t−1 + β7 X i,t + β8 X i,t Z i,t−1 + β9 Z i,t−1 + ε i,t

(4)

Model (2) corresponds to the standard ICFS in Fazzari et al.,

1988) , whereas Model (3) corresponds to a dynamic investment

odel with the cross-product term of cash flow and tangible cap-

tal and with alternative sources of financing. 11 To control for po-

ential omitted variable bias, Model (4) additionally includes the

ross-product term between tangible capital and all explanatory

ariables. We estimate regressions of these forms, where the ra-

io of a firm’s physical investment to the beginning-of-period to-

al assets is denoted by I i, t . The ratio of a firm’s cash flow to the

eginning-of-period total assets is denoted by �i, t . The variable

i,t−1 is the share of tangible capital assets in total assets at the nd of year t – 1 . This variable measures the extent to which a

rm relies on tangible capital in production. Its interaction with

ash flow serves to test our hypotheses and identifies the effect of

angibility on ICFS. The variable I i,t−1 is the lagged investment. The ariable q i,t−1 is the market-to-book ratio at the end of year t – 1. i, t is a vector of non-cash flow sources of financing, such as the

ontemporaneous change in total debt, net equity issue, and cash

eserves at the end of year t – 1. αc, j, t represents country, industry nd time fixed-effects, respectively.

ICFS is obtained from the expression for ∂ I i,t ∂ �i,t

. The investment–

ash flow–tangible capital sensitivity is obtained from the expres-

ion for ∂ I i,t

∂ �i,t Z i,t−1 . Insignificant (significant) ICFS (investment–cash

ow–tangible capital sensitivity) would indicate that tangible cap-

tal predicts investment. Under H 1 , we expect to document declin-

ng investment–cash flow–tangible capital sensitivity for econom-

cally developed countries. The sum of ICFS and investment–cash

ow–tangible capital sensitivity should also decline over time. 12

nder H 2 , we expect to observe non-declining investment–cash

ow–tangible capital sensitivity for economically developing coun-

ries. The sum of ICFS and investment–cash flow–tangible capital

ensitivity should also remain stable over time.

We estimate the autoregression model of cash flow and the au-

oregression model of the combination of cash flow and tangible

apital. This estimation examines the role of asset tangibility in ex-

laining cash flow persistence.

�i,t = αc, j,t + θ0 + θ1 �i,t−1 + μi,t (5)

�i,t Z i,t−1 = αc, j,t + θ0 + θ1 �i,t−1 Z i,t−2 + ηi,t (6)

cale our variables by total assets. 12 Controlling for the tangibility of asset base, ICFS summarizes the portion of the

elation between investment and cash flow that is not explained by the tangibility.

nder our hypotheses, it is still possible that ICFS is statistically significant but not

conomically. Similarly, the decline in ICFS is not solely due to a decline in asset

angibility. It is also due to a decline in investment that is not explained by the

angibility. ICFS can decline over time, even after controlling for asset tangibility,

ut the rate of decline should be either statistically or economically insignificant.

6 F. Moshirian et al. / Journal of Banking and Finance 77 (2017) 1–17

3

w

f

T

o

(

t

B

t

o

b

m

r

t

a

i

a

b

r

l

a

T

o

m

4

s

i

s

e

e

t

f

d

s

v

d

t

s

1

0

0

A

fi

e

c

t

0

h

14 The order number is an empirical choice. The minimum value is 3, which cor-

responds to a precisely identified Geary (1942) estimator. We use a starting value

of 4. As a general rule, we aim to select the order that returns the highest possible

magnitude of the q coefficient. Thus, we use order values ranging from 4 to 8. 15

As documented in Chen and Chen (2012) , the autoregressive co-

efficient of cash flow and the explanatory power of Model (5) have

declined over time. Under H 3 and H 4 , the autoregressive coefficient

θ 1 and its predictive power are expected to decline for advanced economy firms and to remain stable for developing economy firms.

More importantly, if our hypotheses are correct, the persistence

of the cross-product term ( θ 1 in Model 6) should be significantly stronger than the persistence of standalone cash flow ( θ 1 in Model 5). In other words, the degree of asset tangibility should have a

pervasive impact on cash flow predictability.

3.3. Data and variable definitions

We focus on manufacturing firms (SIC codes from 20 0 0 to

3990) from the Worldscope files over the 1993–2013 period. We

estimate the investment regressions over consecutive seven-year

periods (1993–1999, 20 0 0–20 06, and 20 07–2013) and track how

the estimated coefficients develop over time.

To mitigate the effects of outliers, we require firms to have to-

tal assets, physical assets and market capitalization of at least US$1

million. We also require firms to have at least three non-missing

firm-year observations over the sample period. Moreover, we drop

observations for years in which net income exceeds market capital-

ization. Unlike U.S.-based studies of ICFS, we do not exclude obser-

vations with sales (asset) growth rates greater than 100%, given the

large proportion of young and fast-growing firms in our interna-

tional sample. We consider sample firms with strictly positive cash

flow. The rationale for examining profitable firms is to control for

the distortionary effect of negative cash flow on ICFS ( Allayannis

and Mozumdar, 2004 ). In doing so, we rule out the possibility that

the decline in ICFS is simply driven by an increase in the number

of firm-years with negative income. Finally, all regression variables

are trimmed at the 1% level.

For the purpose of our study, we divide firms into three sub-

samples: (i) U.S. firms, (ii) firms from 21 non-U.S. developed

economies, and (iii) firms from 19 developing economies. The level

of a country’s economic and financial development is defined ac-

cording to the Morgan Stanley Composite Index (MSCI) market

classification. 13 This classification scheme should reflect the cross-

country differences in the productive capital structures.

We also construct a semi-balanced panel sample of firms. The

sample consists of firms that exist through the entire sample pe-

riod. The number of firm-years in regressions may change over

time because several important variables (not firms) are missing

in some parts of the sample. Through this sample, we show how

firm characteristics change over time, even for the same set of

firms. Unfortunately, we cannot construct a strictly balanced panel

because this restriction would dramatically reduce the number of

firm-years, particularly for non-U.S. firms.

We measure (physical) investment as capital expenditures

(WC04601). Cash flow is the sum of the income before extraor-

dinary items (WC01551) and depreciation (WC01151). Both vari-

ables are scaled by the beginning-of-period total assets (WC02999).

Tangible (physical) assets are defined as net property, plant, and

equipment (WC02501). Tobin’s q is the ratio of the market value of

equity (WC08001) minus the book value of equity (WC03501) plus

total assets to total assets. Non-cash flow sources of financing in-

clude the change in total debt ( WC03255), net equity issue (total

equity issued, WC04251 minus common and preferred redeemed,

WC04751), and cash reserves (WC02001).

13 See http://www.msci.com/products/indices/market _ classification.html . The MSCI

classification is similar to the Dow Jones and the Standard and Poor’s classifications.

fi

e

i

c

(

.4. Methodology

To ensure that our results are not driven by a few countries

ith the highest numbers of observations, our regression results

rom Models (2)–(6) are based on weighted least squares (WLS).

he technique weighs each country equally such that firm-year

bservations receive more (less) weight in countries with fewer

more) firm-years. The standard errors are robust and clustered at

he country level.

The problem is that the WLS estimates are likely to be biased.

ecause there is a positive correlation between mismeasured q and

he cash flow regressor, the attenuation bias causes the coefficient

n q to be biased downward and the coefficient on cash flow to

e biased upward. To mitigate the measurement problem, we esti-

ate the investment Models (2)–(4) by linear errors-in-variables

egression with identification from high-order cumulant estima-

ors. In this respect, we follow the recent work by Erickson et

l., (2014) , who developed the q measurement error remedy that

s asymptotically equivalent to the moment estimators in Erickson

nd Whited (20 0 0, 20 02 ). The cumulant estimators are an advance

eyond the moment estimators. Overidentified moment estimators

equire a numerical minimization procedure, but cumulants are

inear and have a closed-form solution. Thus, using moments adds

level of complexity that is absent from high-order cumulants. 14

he goodness-of-fit measure of the regression, R 2 ( Rho ), and the R 2

f the measurement equation ( Tau ), which is an index of measure-

ent quality, are also reported.

. Descriptive statistics

Table 1 reports the descriptive statistics. Panel A reports the

ample means and medians by country for the key variables used

n this paper. The average fraction of tangible assets in total as-

ets is 0.25 for U.S. firms, 0.31 for firms from other developed

conomies, and 0.41 for firms from developing economies. The av-

rage fraction of tangible assets in total productive assets, which is

he sum of tangible and intangible assets, is 0.62 for U.S. firms, 0.76

or firms from other developed economies, and 0.92 for firms from

eveloping economies. 15 Firms from advanced economies have as-

et structures which are less tilted to physical assets. Tobin’s q

aries across countries and is generally higher in the U.S. and other

eveloped economies.

Panel B reports the sample means and medians by period for

he full sample. The average fraction of tangible assets in total as-

ets (in total productive assets) declined from 0.29 (0.69) in 1993–

999 to 0.21 (0.57) in 2007–2013 for U.S. firms, from 0.33 (0.90) to

.28 (0.77) for firms from other developed economies, and from

.42 (0.97) to 0.35 (0.91) for firms from developing economies.

lthough the rates of decline of the first ratio are similar for

rms from non-U.S. developed economies and from developing

conomies, the latter firms still operate with much more tangible

apital. Investment as a fraction of total assets declined from 0.07

o 0.04 for advanced economy firms and slightly declined from

.08 to 0.07 for developing economy firms. The latter firms have

igher rates of investment.

This measure is not perfect. Some firms do not report their R&D costs. Some

rms do not capitalise their costs, and therefore, their intangible assets are under-

stimated by the accounting data. Still, the size of intangible assets provides useful

nformation. If economic benefit is expected to flow to the entity as a result of in-

urring R&D costs, then these costs are treated as an asset rather than an expense

IAS, U.S. GAAP, U.K. GAAP).

F. Moshirian et al. / Journal of Banking and Finance 77 (2017) 1–17 7

Table 1

Descriptive statistics. The table reports the country-level means and medians for the key variables used in Models (2)–(6): physical investment ( I i, t ), cash flow ( �i, t ), Tobin’s

measure ( q i,t−1 ), and beginning-of-period tangible capital ( Z i,t−1 ). The table also reports the ratio of tangible capital to the sum of tangible capital and intangible capital (Y i, t ). Investment, cash flow and tangible capital are deflated by the beginning-of-period total assets. The symbols correspond to the notation used in Models (2)–(6). US

denotes U.S. economy. DME denotes non-U.S. developed economy. EME denotes a developing economy. N is the number of firm-years.

Panel A. Cross-country statistics

I i, t �i, t q i,t−1 Z i,t−1 Y i, t

Country Market N Mean Median Mean Median Mean Median Mean Median Mean Median

UNITED STATES US 20 ,905 0 .05 0 .04 0 .11 0 .10 1 .73 1 .43 0 .25 0 .22 0 .62 0 .65

AUSTRALIA DME 1651 0 .05 0 .04 0 .11 0 .10 1 .45 1 .26 0 .31 0 .31 0 .68 0 .76

AUSTRIA DME 785 0 .07 0 .06 0 .10 0 .10 1 .23 1 .14 0 .34 0 .36 0 .84 0 .92

BELGIUM DME 674 0 .07 0 .05 0 .11 0 .10 1 .41 1 .16 0 .28 0 .26 0 .73 0 .82

CANADA DME 2298 0 .06 0 .04 0 .11 0 .10 1 .47 1 .22 0 .35 0 .33 0 .72 0 .84

DENMARK DME 1049 0 .07 0 .05 0 .11 0 .10 1 .41 1 .18 0 .34 0 .33 0 .83 0 .97

FINLAND DME 1114 0 .07 0 .05 0 .12 0 .10 1 .50 1 .25 0 .32 0 .31 0 .73 0 .82

FRANCE DME 4311 0 .05 0 .04 0 .10 0 .09 1 .38 1 .17 0 .21 0 .19 0 .66 0 .72

GERMANY DME 4752 0 .07 0 .05 0 .11 0 .10 1 .40 1 .23 0 .28 0 .27 0 .77 0 .86

HONG KONG DME 3322 0 .06 0 .04 0 .11 0 .10 1 .19 0 .98 0 .30 0 .28 0 .89 0 .99

IRELAND DME 460 0 .05 0 .04 0 .11 0 .11 1 .68 1 .50 0 .31 0 .32 0 .60 0 .63

ISRAEL DME 1054 0 .04 0 .03 0 .10 0 .08 1 .36 1 .14 0 .24 0 .22 0 .77 0 .88

ITALY DME 1621 0 .05 0 .04 0 .08 0 .07 1 .28 1 .11 0 .25 0 .24 0 .71 0 .79

JAPAN DME 22 ,317 0 .04 0 .03 0 .07 0 .06 1 .10 1 .00 0 .32 0 .31 0 .95 0 .98

NETHERLANDS DME 1223 0 .06 0 .05 0 .11 0 .11 1 .54 1 .29 0 .30 0 .31 0 .74 0 .91

NEW ZEALAND DME 231 0 .06 0 .05 0 .11 0 .10 1 .51 1 .27 0 .38 0 .36 0 .78 0 .93

NORWAY DME 709 0 .07 0 .05 0 .11 0 .10 1 .49 1 .25 0 .30 0 .29 0 .71 0 .78

SINGAPORE DME 2518 0 .06 0 .04 0 .11 0 .09 1 .26 1 .07 0 .32 0 .31 0 .92 0 .99

SPAIN DME 849 0 .06 0 .05 0 .11 0 .10 1 .38 1 .18 0 .38 0 .39 0 .82 0 .94

SWEDEN DME 1755 0 .05 0 .04 0 .12 0 .11 1 .57 1 .32 0 .26 0 .24 0 .63 0 .70

SWITZERLAND DME 1832 0 .05 0 .04 0 .11 0 .10 1 .59 1 .28 0 .31 0 .30 0 .74 0 .82

UNITED KINGDOM DME 6074 0 .06 0 .05 0 .12 0 .11 1 .64 1 .42 0 .30 0 .29 0 .74 0 .92

All 60 ,599 0 .06 0 .04 0 .11 0 .10 1 .42 1 .21 0 .31 0 .30 0 .76 0 .86

ARGENTINA EME 388 0 .07 0 .05 0 .12 0 .11 1 .15 1 .05 0 .45 0 .45 0 .96 0 .99

BRAZIL EME 1506 0 .07 0 .05 0 .11 0 .10 1 .16 0 .99 0 .39 0 .39 0 .93 1 .00

CHILE EME 945 0 .06 0 .05 0 .11 0 .09 1 .32 1 .12 0 .45 0 .46 0 .91 0 .97

CHINA EME 11 ,408 0 .08 0 .06 0 .09 0 .07 1 .99 1 .66 0 .32 0 .30 0 .87 0 .91

GREECE EME 584 0 .05 0 .03 0 .07 0 .06 1 .17 1 .01 0 .42 0 .42 0 .91 0 .98

INDIA EME 8943 0 .08 0 .06 0 .11 0 .09 1 .35 1 .03 0 .38 0 .38 0 .96 1 .00

INDONESIA EME 1521 0 .07 0 .04 0 .11 0 .09 1 .27 1 .02 0 .41 0 .40 0 .97 1 .00

MALAYSIA EME 4328 0 .05 0 .04 0 .10 0 .08 1 .17 0 .94 0 .41 0 .40 0 .94 1 .00

MEXICO EME 905 0 .06 0 .05 0 .11 0 .11 1 .28 1 .09 0 .53 0 .56 0 .86 0 .94

PAKISTAN EME 1385 0 .06 0 .04 0 .12 0 .11 1 .23 1 .03 0 .46 0 .46 0 .99 1 .00

PERU EME 468 0 .06 0 .04 0 .12 0 .11 1 .20 0 .97 0 .45 0 .44 0 .93 0 .99

PHILIPPINES EME 459 0 .06 0 .04 0 .10 0 .09 1 .18 0 .99 0 .40 0 .38 0 .91 0 .98

POLAND EME 971 0 .07 0 .05 0 .11 0 .09 1 .29 1 .06 0 .42 0 .42 0 .88 0 .96

PORTUGAL EME 325 0 .06 0 .04 0 .09 0 .08 1 .14 1 .04 0 .41 0 .39 0 .86 0 .97

SOUTH AFRICA EME 1143 0 .07 0 .06 0 .13 0 .12 1 .41 1 .26 0 .35 0 .34 0 .84 0 .98

SOUTH KOREA EME 8467 0 .07 0 .05 0 .09 0 .08 1 .02 0 .91 0 .37 0 .36 0 .93 0 .98

TAIWAN EME 10 ,241 0 .06 0 .04 0 .10 0 .09 1 .37 1 .17 0 .34 0 .33 0 .96 0 .99

THAILAND EME 2648 0 .07 0 .05 0 .12 0 .11 1 .18 1 .03 0 .42 0 .42 0 .97 1 .00

TURKEY EME 1382 0 .06 0 .04 0 .12 0 .10 1 .39 1 .16 0 .35 0 .36 0 .92 0 .99

A ll 58 ,017 0 .06 0 .05 0 .11 0 .09 1 .28 1 .08 0 .41 0 .40 0 .92 0 .98

Panel B. Time-series statistics (full sample)

I i, t �i, t q i,t−1 Z i,t−1 Y i, t

Period Market N Mean Median Mean Median Mean Median Mean Median Mean Median

1993–1999 US 7295 0 .07 0 .06 0 .12 0 .12 1 .77 1 .47 0 .29 0 .26 0 .69 0 .74

20 0 0–20 06 US 7634 0 .05 0 .03 0 .11 0 .09 1 .73 1 .40 0 .25 0 .21 0 .61 0 .63

2007–2013 US 5976 0 .04 0 .03 0 .11 0 .10 1 .67 1 .42 0 .21 0 .17 0 .57 0 .57

1993–1999 DME 14 ,420 0 .07 0 .05 0 .10 0 .09 1 .44 1 .25 0 .33 0 .31 0 .90 0 .98

20 0 0–20 06 DME 22 ,310 0 .05 0 .04 0 .09 0 .08 1 .31 1 .10 0 .31 0 .31 0 .83 0 .96

2007–2013 DME 23 ,869 0 .04 0 .03 0 .09 0 .08 1 .24 1 .06 0 .28 0 .26 0 .77 0 .92

1993–1999 EME 5207 0 .08 0 .06 0 .11 0 .09 1 .44 1 .17 0 .42 0 .42 0 .97 1 .00

20 0 0–20 06 EME 16 ,862 0 .07 0 .05 0 .11 0 .09 1 .29 1 .07 0 .40 0 .39 0 .95 0 .99

2007–2013 EME 35 ,948 0 .07 0 .05 0 .10 0 .08 1 .43 1 .12 0 .35 0 .34 0 .91 0 .98

Panel C. Time-series statistics (semi-balanced panel)

I i, t �i, t q i,t−1 Z i , t −1 Y i, t

Period Market N Mean Median Mean Median Mean Median Mean Median Mean Median

1993–1999 US 3018 0 .07 0 .06 0 .13 0 .13 1 .87 1 .57 0 .30 0 .27 0 .70 0 .74

20 0 0–20 06 US 3273 0 .05 0 .04 0 .11 0 .10 1 .76 1 .48 0 .26 0 .23 0 .60 0 .62

2007–2013 U S 3100 0 .04 0 .03 0 .11 0 .10 1 .64 1 .44 0 .22 0 .19 0 .56 0 .55

1993–1999 DME 7351 0 .06 0 .05 0 .09 0 .08 1 .42 1 .25 0 .32 0 .31 0 .91 0 .99

20 0 0–20 06 DME 11 ,267 0 .05 0 .04 0 .08 0 .07 1 .22 1 .06 0 .32 0 .32 0 .87 0 .97

2007–2013 DME 11 ,566 0 .04 0 .03 0 .08 0 .07 1 .19 1 .04 0 .29 0 .29 0 .81 0 .94

1993–1999 EME 2731 0 .08 0 .06 0 .11 0 .10 1 .45 1 .16 0 .43 0 .43 0 .97 1 .00

20 0 0–20 06 EME 3498 0 .06 0 .04 0 .11 0 .10 1 .20 0 .99 0 .44 0 .44 0 .95 1 .00

2007–2013 EME 3614 0 .06 0 .04 0 .11 0 .09 1 .35 1 .09 0 .39 0 .39 0 .92 0 .98

8 F. Moshirian et al. / Journal of Banking and Finance 77 (2017) 1–17

Fig. 1. Descriptive statistics (1). The figures plot physical investment, I ( I i, t ) (long dash line), cash flow, CF ( �i, t ) (round dot line), and beginning-of-period tangible capital,

K ( Z i,t−1 ) (square dot line). The variables are deflated by the beginning-of-period total assets. The left figures correspond to the U.S. full sample (upper) and semi-balanced panel (lower), respectively. The middle figures correspond to non-U.S. developed economies. The right figures correspond to developing economies.

Fig. 2. Descriptive statistics (2). The figures plot the ratio of tangible capital to the sum of tangible capital and intangible capital (Y i, t ). The long dash line corresponds to the

U.S. economy; square dot line – to non-U.S. developed economies (DME); round dot line – to developing economies (EME). The left figure corresponds to the full sample;

the right figure – to semi-balanced panel.

t

m

5

5

(

t

a

1

a

s

U

d

(

t

t

fi

t

fl

Panel C contains the sample means and medians for the semi-

balanced panel. Tangible capital and investment fell sharply in the

U.S. The rate of investment also plunged in other economies, al-

though at different paces, and with a more stable share of tangible

assets. Although the firms in the restricted sample are the same,

their characteristics have changed, and in particular, their asset

structures and investment rates have changed. It is these changes,

rather than the composition of the firms in the sample, that matter

for the sensitivity of investment to cash flow.

The reported statistics are built based on seven-year subsample

periods. The problem, however, is that we have only three periods.

Using an alternative sampling approach, we report summary statis-

tics for the key variables for each year in Figs. 1 and 2 . The average

fraction of tangible assets in total assets (in total productive assets)

declined from 0.31 (0.74) in 1993 to 0.22 (0.57) in 2013 for U.S.

firms, from 0.33 (0.91) to 0.27 (0.76) for other developed economy

firms, and from 0.42 (0.96) to 0.35 (0.90) for developing economy

firms. The physical investment as a fraction of total assets fell by

two-fifths of its value for U.S. firms, by over a third for firms from

other developed economies, and between a quarter and a third for

firms from developing economies. The summary statistics indicate

a global change in the productive capital structure, with a sharp

fall in investment and tangible capital in the subsample of U.S.

firms. It is therefore natural to expect a steady decline in ICFS in

I

he U.S and other advanced economies (to a lesser degree) and a

odest decline in ICFS in less developed economies.

. Investment–cash flow sensitivity and asset tangibility

.1. Baseline results

We report the WLS results returned from investment Models

2) and (3) in Panel A and Panel B of Table 2 , respectively. The es-

imated coefficients on Tobin’s q in all regressions are insignificant,

t approximately 0.01, in sharp contrast to the theoretical value of

under the q theory model with constant returns to scale and no

djustment costs. Clearly, the empirical result is driven by the mea-

urement error in q , which we address later in this study.

For investment Model (2), ICFS has dramatically declined in the

.S. In particular, the magnitude and statistical significance have

eclined from 0.25 (t = 14.5) in the first seven-year period to 0.11 t = 7.96) in the last period. Similarly, ICFS has steadily declined in he subsample of non-U.S. advanced economies. Both the magni-

ude and significance have decreased from 0.34 (t = 12.2) in the rst period to 0.14 (t = 7.28) in the final reporting period. Consis-

ent with the literature, a strong response of investment to cash

ow is evident only in the early parts of the sample. In contrast,

CFS is stable in the subsample of developing economies. We docu-

F. Moshirian et al. / Journal of Banking and Finance 77 (2017) 1–17 9

Table 2

Investment regressions The table reports coefficients estimated from the WLS regression of physical investment ( I i, t ) on cash flow ( �i, t ), beginning-of-period tangible

capital ( Z i,t−1 ), lagged investment ( I i,t−1 ), change in total debt ( D i, t ), new equity issue ( E i, t ), beginning-of-period cash reserves ( C i,t−1 ), and Tobin’s measure ( q i,t−1 ). Cash flow has its cross-product term with tangible capital. The model specifications in Panel A and Panel B correspond to Models (2) and (3), respectively. US denotes U.S.

economy. DME denotes non-U.S. developed economy. EME denotes a developing economy. The t-statistic reported in parentheses adjusted for clustered standard errors at

the country level. Country, industry and time fixed effects are included. N is the number of firm-year observations. The R 2 is the adjusted R 2 .

Panel A. Investment Model (2)

Period Market �i, t (t-stat) q i,t−1 (t-stat) R 2 N

1993–99 US 0 .25 (14 .52) 0 .00 (1 .58) 16 .1% 7295

20 0 0–06 US 0 .16 (12 .59) 0 .01 (6 .75) 16 .4% 7634

2007–13 US 0 .11 (7 .96) 0 .00 (2 .55) 10 .7% 5976

1993–99 DME 0 .34 (12 .21) 0 .00 (0 .75) 18 .1% 14 ,420

20 0 0–06 DME 0 .28 (12 .21) 0 .00 (0 .39) 18 .8% 22 ,310

2007–13 DME 0 .14 (7 .28) 0 .00 (2 .53) 14 .9% 23 ,869

1993–99 EME 0 .22 (6 .51) 0 .02 (3 .94) 18 .0% 5207

20 0 0–06 EME 0 .24 (11 .70) 0 .00 (0 .89) 12 .8% 16 ,862

2007–13 EME 0 .21 (7 .89) 0 .01 (3 .22) 13 .5% 35 ,948

Panel B. Investment Model ( 3 )

Period Market �i, t (t-stat) ( �i,t ∗ Z i,t−1 ) (t-stat) I i,t−1 (t-stat) D i, t (t-stat) E i, t (t-stat) C i,t−1 (t-stat) q i,t−1 (t-stat) R 2 �i,t + ( �i,t ∗ Z i,t−1 ) (t-stat) 1993–99 US 0 .02 (1 .23) 0 .55 (8 .14) 0 .41 (25 .69) 0 .09 (13 .22) 0 .05 (6 .40) 0 .02 (3 .58) 0 .00 (1 .62) 48 .8% 0 .57 (10 .78)

20 0 0–06 US 0 .01 (1 .05) 0 .40 (8 .65) 0 .41 (18 .49) 0 .06 (11 .43) 0 .05 (5 .61) 0 .00 (1 .28) 0 .00 (5 .05) 47 .5% 0 .41 (10 .68)

2007–13 US 0 .02 (1 .25) 0 .24 (3 .33) 0 .44 (17 .52) 0 .05 (7 .72) 0 .04 (3 .63) 0 .01 (3 .49) 0 .00 (2 .86) 45 .1% 0 .26 (4 .22)

1993–99 DME 0 .08 (2 .80) 0 .47 (7 .63) 0 .32 (12 .54) 0 .14 (7 .05) 0 .04 (2 .47) 0 .02 (2 .90) 0 .00 (0 .41) 41 .3% 0 .55 (12 .26)

20 0 0–06 DME 0 .05 (2 .59) 0 .46 (7 .52) 0 .35 (23 .23) 0 .11 (10 .56) 0 .03 (2 .82) 0 .01 (1 .84) 0 .00 (−0 .16) 45 .5% 0 .51 (11 .07) 2007–13 DME −0 .01 (−0 .55) 0 .43 (8 .76) 0 .39 (21 .09) 0 .09 (9 .60) 0 .04 (2 .86) 0 .01 (3 .26) 0 .00 (3 .13) 45 .7% 0 .42 (9 .53) 1993–99 EME 0 .05 (0 .96) 0 .32 (2 .62) 0 .30 (12 .83) 0 .20 (13 .83) 0 .10 (3 .82) 0 .06 (4 .02) 0 .01 (3 .20) 43 .5% 0 .36 (4 .30)

20 0 0–06 EME 0 .07 (3 .33) 0 .29 (5 .62) 0 .35 (15 .88) 0 .18 (15 .31) 0 .11 (3 .67) 0 .01 (1 .76) 0 .00 (−0 .80) 40 .3% 0 .37 (10 .12) 2007–13 EME 0 .01 (0 .40) 0 .42 (7 .69) 0 .37 (16 .19) 0 .17 (12 .47) 0 .08 (4 .55) 0 .02 (4 .23) 0 .00 (1 .89) 44 .5% 0 .42 (9 .42)

Note: For brevity, the coefficient estimates on tangible capital ( Z i,t−1 ) are not tabulated.

Fig. 3. Investment regressions by year. The figures plot coefficients estimated from the WLS regression of investment ( I i, t ) on cash flow ( �i, t ) and Tobin’s measure ( q i,t−1 ) from Model (2). The left figure corresponds to the U.S. economy. The middle figure corresponds to non-U.S. developed economies. The right figure corresponds to developing

economies. One data point in 1993 for developing economy firms is missing due to the lack of sufficient observations. The R 2 is the adjusted R 2 .

m

n

i

i

o

o

fi

e

v

t

w

n

s

I

i

r

e

s

c

f

t

j

f

o

s

p

t

(

p

r

e

t

l

fl

p

fl

fl

ent a magnitude of 0.22 (t = 6.51) in the first period and a mag- itude of 0.21 (t = 7.89) in the last period.

Above, we divide the sample into seven-year periods, estimat-

ng the ICFS for each period. The disadvantage of such grouping

s that with only three periods, we lack a more detailed picture

f the time-series variation of ICFS. An alternative approach relies

n cross-sectional estimations for each year. For this purpose, we

rst demean all variables by firm to remove fixed effects for the

ntire period and then estimate a cross-sectional regression of in-

estment on cash flow and q for each year. We also include coun-

ry and industry effects. The results are plotted in Fig. 3 . Again,

e observe a strong decline in ICFS for U.S. firms. In 1993, ICFS is

early 0.2. In 2013, the figure is below 0.1. From 2005 onward, the

ensitivity is never greater than 0.1. We also observe a declining

CFS for firms from non-U.S. advanced economies. ICFS is approx-

mately 0.25 and below 0.15 in the early and later sample years,

espectively. There is no clear pattern for firms from developing

conomies. ICFS varies between 0.1 and 0.25, although it is more

table in the earlier years. The only significant drop in ICFS oc-

urred in 2008. The goodness-of-fit of the investment regressions

ollow the significance of ICFS. The q–investment sensitivity con-

inues to be insignificant. This finding on the global downward tra-

ectory is interesting per se because we document that ICFS has

allen beyond the U.S., although at different paces.

For investment Model (3) with the added cross-product term

f cash flow and tangible capital, we observe two important re-

ults. First, ICFS is indistinguishable from zero, even for the early

eriods, after controlling for the cross-product term. ICFS is statis-

ically significant in some estimations, but it is economically low

below 0.1). However, the cross-product term itself is significantly

ositive and economically meaningful in each of the periods. This

esult implies that ICFS is driven, to a very large extent, by the

ffect of asset tangibility. ICFS is in fact the investment–cash flow–

angible capital sensitivity because the investment of a firm with

ow tangible capital assets is not systematically sensitive to cash

ow. Only firms with high tangible capital have cash flow that ex-

lains investment. They invest more when they have marginal cash

ow, displaying a higher ICFS.

Second, over time, the sum of ICFS and investment–cash

ow–tangible capital sensitivity declines from 0.57 (t = 10.8) to

10 F. Moshirian et al. / Journal of Banking and Finance 77 (2017) 1–17

Table 3

Investment regressions: alternative sources of financing. The table reports coefficients estimated from the WLS regression of physical investment ( I i, t ) on cash flow ( �i, t ),

beginning-of-period tangible capital ( Z i,t−1 ), lagged investment ( I i,t−1 ), change in total debt ( D i, t ), new equity issue ( E i, t ), beginning-of-period cash reserves ( C i,t−1 ), and Tobin’s measure ( q i,t−1 ). Each explanatory variable has its cross-product term with tangible capital. The model specification corresponds to Model (4). US denotes U.S. economy. DME denotes non-U.S. developed economy. EME denotes a developing economy. The t-statistic reported in parentheses adjusted for clustered standard errors at

the country level. Country, industry and time fixed effects are included. N is the number of firm-year observations. The R 2 is the adjusted R 2 .

Period Market �i, t ( �i,t ∗ Z i,t−1 ) I i,t−1 ( D i,t ∗ Z i,t−1 ) ( E i,t ∗ Z i,t−1 ) ( C i,t−1 ∗ Z i,t−1 ) q i,t−1 R 2 N �i,t + ( �i,t ∗ Z i,t−1 ) 1993–99 US 0 .03 0 .52 0 .42 0 .22 0 .16 0 .20 0 .00 49 .8% 7295 0 .55

(1 .40) (7 .10) (12 .57) (4 .53) (2 .44) (5 .50) (1 .55) (9 .56)

20 0 0–06 US 0 .02 0 .36 0 .40 0 .25 0 .15 0 .08 0 .00 48 .8% 7634 0 .38

(1 .82) (7 .40) (10 .24) (7 .18) (2 .45) (2 .64) (2 .62) (9 .20)

2007–13 US 0 .03 0 .19 0 .49 0 .23 0 .09 0 .17 0 .00 47 .0% 5976 0 .22

(2 .01) (2 .58) (10 .61) (4 .94) (1 .16) (5 .13) (−0 .38) (3 .49) 1993–99 DME 0 .09 0 .45 0 .41 0 .23 0 .11 0 .15 0 .00 42 .2% 14 ,420 0 .54

(2 .48) (5 .07) (8 .40) (3 .14) (1 .33) (2 .81) (0 .19) (8 .37)

20 0 0–06 DME 0 .06 0 .43 0 .40 0 .29 0 .09 0 .10 0 .00 46 .7% 22 ,310 0 .49

(3 .31) (7 .00) (8 .66) (4 .57) (1 .34) (3 .18) (−0 .89) (10 .23) 2007–13 DME 0 .01 0 .34 0 .42 0 .34 0 .19 0 .13 0 .00 47 .6% 23 ,869 0 .35

(1 .05) (6 .69) (8 .89) (9 .81) (2 .88) (4 .18) (0 .54) (7 .83)

1993–99 EME 0 .10 0 .18 0 .39 0 .41 0 .29 0 .22 0 .00 45 .8% 5207 0 .27

(1 .99) (1 .79) (8 .13) (5 .36) (2 .99) (3 .73) (−0 .55) (3 .27) 20 0 0–06 EME 0 .09 0 .25 0 .40 0 .32 0 .17 0 .10 0 .00 41 .4% 16 ,862 0 .34

(3 .64) (3 .97) (9 .29) (6 .22) (1 .69) (2 .70) (−1 .33) (7 .83) 2007–13 EME 0 .02 0 .38 0 .45 0 .36 0 .13 0 .16 0 .00 46 .0% 35 ,948 0 .40

(1 .07) (6 .26) (13 .69) (6 .30) (1 .97) (7 .10) (0 .77) (8 .06)

Note: For brevity, the table reports the main variables of interest.

p

b

a

a

a

e

t

c

i

t

i

c

fl

fi

e

5

fi

d

o

s

d

t

c

p

t

s

m

f

t

0.26 (t = 4.22) in the U.S. and from 0.55 (t = 12.3) to 0.42 (t = 9.53) in non-U.S. developed economies. Individually, investment–cash

flow–tangible capital sensitivity also declines in developed

economies. This monotonic pattern supports H 1 . The aggregate

cash flow sensitivity is stable in developing economies. The value

stands at 0.36 (t = 4.30) in the first period and at 0.42 (t = 9.42) in the last period. Investment–cash flow–tangible capital sensitivity is

also non-declining. This pattern supports H 2 . Overall, our results

show that the declining ICFS documented in prior studies simply

reflects the fact that the effect of asset tangibility on ICFS weakens

over time. 16

5.2. Alternative sources of financing

Alternative (non-cash flow) sources of financing may affect both

investment and cash flow sensitivities. Bates et al., (2009) find that

the average cash-to-assets ratio of U.S. industrial firms has doubled

from 1980 to 2006. Thus, it is possible that firms finance their in-

vestment from their cash holdings rather than from cash flow. In

this case, a higher level of cash holdings as internal funds will at-

tenuate ICFS. Omission of the cash holdings in the model specifi-

cation would bias the estimated coefficient of tangible capital. It is

also possible that internal cash flow has become a less important

source of financing compared with equity and debt financing.

Investment Model (4) controls for cash reserves, net equity is-

sues, and the contemporaneous change in debt. The model fur-

ther includes the cross-product term of tangible capital with each

source of financing, lagged investment and Tobin’s q . We now ex-

amine how this addition affects the coefficients on cash flow. The

results are presented in Table 3 .

Before we examine the main results, we briefly discuss our

findings on firm financing characteristics. Debt finance contributes

to new investment, which is consistent with the notion of asset

16 In untabulated results, we document a substantial cross-country variation in

ICFS. Over the sample period, ICFS ranges from 0.05 (Philippines) to 0.44 (Austria).

The presence of the product term of cash flow and tangible capital in the invest-

ment regression reduces the magnitude and significance of ICFS (the average coef-

ficient is only 0.03). In contrast to ICFS, the coefficient on the product term itself

(the investment–cash flow–tangible capital sensitivity) is on average 0.45.

e

(

v

g

h

i

0

ledgeability. In particular, the cross-product term between tangi-

le capital and the change in debt is significant across countries

nd stable over time. Our results further indicate that cash reserves

nd net equity issues are less important in predicting investment.

None of the alternative channels of financing can predict ICFS

nd investment–cash flow–tangible capital sensitivity. Specifically,

ven though the cross-product terms of the firm financing charac-

eristics are added into the regression model, the coefficients of the

ross-product term of cash flow and tangible capital are still pos-

tive and significant. The value continues to steadily decline over

ime for developed economy firms. Conversely, it is strengthen-

ng for less developed firms. ICFS is economically low or insignifi-

ant in most estimations. Therefore, we conclude that internal cash

ow remains an important source of financing for capital-intensive

rms and that changes in alternative sources of financing cannot

xplain why ICFS changes over time.

.3. Investment-intensive, capital-intensive, and semi-balanced panel

rms

We verify our results using subsamples of firms along several

imensions. First, we define investment-intensive firms by the size

f their investment and estimate the cash flow sensitivities from a

ample of firm-years for which actual investment is greater than

epreciation ( Table 4 ). One could argue that the connection be-

ween asset tangibility and investment is driven by a simple me-

hanical relation that dictates that investment ought to replace de-

reciating assets. Of course, the level of investment is set to main-

ain the productive assets, and to test the argument that firms

pend above necessary investments, we need a measure of the

inimum required investment. We use depreciation as a proxy

or the required investment outlay and define net investment as

he difference between actual investment and depreciation. After

liminating those firm-years for which physical capital deteriorates

i.e., with negative net investment), we find that firms from de-

eloped economies continue to experience a decline in the tan-

ibility and in the scale of their investment. Thus, they tend to

ave sensitivities with the significance declining over time. For

nstance, the sum of the sensitivities falls from 0.57 (t = 7.46) to .32 (t = 3.67) in the U.S subsample. Individually, investment–cash

F. Moshirian et al. / Journal of Banking and Finance 77 (2017) 1–17 11

Table 4

Investment regressions: investment-intensive firms. The table reports coefficients estimated from the WLS regression of physical investment ( I i, t ) on cash flow ( �i, t ),

beginning-of-period tangible capital ( Z i,t−1 ), lagged investment ( I i,t−1 ), change in total debt ( D i, t ), new equity issue ( E i, t ), beginning-of-period cash reserves ( C i,t−1 ), and Tobin’s measure ( q i,t−1 ). Each explanatory variable has its cross-product term with tangible capital. The model specification corresponds to Model (4). The sample is restricted to firm-years with positive net investment (the difference between investment and depreciation). US denotes U.S. economy. DME denotes non-U.S. developed economy. EME

denotes a developing economy. The t-statistic reported in parentheses adjusted for clustered standard errors at the country level. Country, industry and time fixed effects

are included. N is the number of firm-year observations. The R 2 is the adjusted R 2 .

Period Market �i, t ( �i,t ∗ Z i,t−1 ) I i,t−1 ( D i,t ∗ Z i,t−1 ) ( E i,t ∗ Z i,t−1 ) ( C i,t−1 ∗ Z i,t−1 ) q i,t−1 R 2 N �i,t + ( �i,t ∗ Z i,t−1 ) 1993–99 US 0 .04 0 .53 0 .36 0 .22 0 .20 0 .32 0 .00 46 .9% 4331 0 .57

(1 .22) (5 .30) (8 .68) (3 .26) (2 .84) (6 .52) (1 .16) (7 .46)

20 0 0–06 US 0 .03 0 .32 0 .33 0 .28 0 .06 0 .17 0 .00 46 .5% 2804 0 .35

(1 .53) (4 .04) (6 .18) (4 .52) (0 .65) (2 .86) (0 .41) (5 .35)

2007–13 US 0 .01 0 .31 0 .31 0 .10 0 .09 0 .28 0 .00 43 .9% 2346 0 .32

(0 .61) (2 .92) (5 .31) (0 .97) (1 .02) (5 .08) (0 .58) (3 .76)

1993–99 DME 0 .08 0 .54 0 .42 0 .23 0 .10 0 .15 0 .00 42 .8% 8944 0 .62

(1 .62) (4 .46) (6 .71) (2 .63) (0 .80) (1 .90) (1 .15) (6 .86)

20 0 0–06 DME 0 .06 0 .47 0 .36 0 .29 0 .07 0 .16 0 .00 47 .5% 10 ,295 0 .53

(1 .51) (3 .57) (6 .24) (2 .91) (0 .68) (3 .36) (−0 .88) (5 .61) 2007–13 DME 0 .03 0 .33 0 .30 0 .24 0 .18 0 .23 0 .00 44 .5% 10 ,582 0 .36

(1.23) (4.37) (4.95) (4.31) (2.19) (4.60) (0.46) (6.11)

1993–99 EME 0 .10 0 .36 0 .33 0 .58 0 .31 0 .18 0 .00 47 .1% 3614 0 .46

(1 .27) (2 .26) (4 .88) (4 .51) (2 .61) (2 .21) (−1 .05) (4 .53) 20 0 0–06 EME 0 .10 0 .37 0 .39 0 .37 0 .04 0 .10 −0 .01 42 .6% 9606 0 .47

(2 .73) (4 .69) (8 .00) (4 .85) (0 .20) (1 .83) (−1 .90) (9 .24) 2007–13 EME −0 .01 0 .50 0 .35 0 .42 0 .12 0 .21 0 .00 44 .9% 22 ,575 0 .49

(−0 .29) (5 .68) (8 .80) (5 .87) (1 .57) (6 .35) (1 .33) (7 .18)

Note: For brevity, the table reports the main variables of interest.

Table 5

Investment regressions: capital-intensive firms. The table reports coefficients estimated from the WLS regression of physical investment ( I i, t ) on cash flow ( �i, t ), lagged

investment ( I i,t−1 ), change in total debt ( D i, t ), new equity issue ( E i, t ), beginning-of-period cash reserves ( C i,t−1 ), and Tobin’s measure ( q i,t−1 ). The sample is restricted to firm-years in the top quarter of the asset tangibility distribution in each country and year. Asset tangibility is defined as the ratio of tangible assets to total assets. US denotes

U.S. economy. DME denotes non-U.S. developed economy. EME denotes a developing economy. The t-statistic reported in parentheses adjusted for clustered standard errors

at the country level. Country, industry and time fixed effects are included. N is the number of firm-year observations. The R 2 is the adjusted R 2 .

Period Market �i, t I i,t−1 D i, t E i, t C i,t−1 q i,t−1 R 2 N

1993–99 US 0 .28 0 .40 0 .14 0 .09 0 .05 0 .00 49 .3% 1827

(10 .29) (16 .92) (8 .71) (3 .15) (1 .91) (0 .00)

20 0 0–06 US 0 .20 0 .39 0 .12 0 .05 0 .02 0 .00 46 .1% 1911

(9 .54) (11 .37) (8 .98) (1 .91) (1 .02) (1 .20)

2007–13 US 0 .11 0 .43 0 .10 0 .07 0 .03 0 .00 41 .2% 1496

(5 .21) (11 .98) (6 .40) (2 .11) (1 .88) (1 .78)

1993–99 DME 0 .36 0 .32 0 .17 0 .03 0 .04 0 .00 47 .1% 3658

(13 .21) (11 .71) (8 .30) (1 .72) (2 .71) (−2 .21) 20 0 0–06 DME 0 .31 0 .33 0 .19 0 .09 0 .01 0 .00 49 .8% 5639

(11 .41) (18 .69) (8 .93) (2 .71) (0 .32) (0 .75)

2007–13 DME 0 .21 0 .38 0 .19 0 .11 0 .03 0 .00 50 .8% 6024

(8 .09) (20 .62) (7 .36) (3 .92) (1 .81) (2 .04)

1993–99 EME 0 .27 0 .24 0 .31 0 .21 0 .09 0 .01 51 .0% 1348

(6 .24) (12 .16) (12 .45) (5 .80) (2 .14) (4 .10)

20 0 0–06 EME 0 .31 0 .30 0 .29 0 .10 0 .05 0 .00 53 .2% 4261

(10 .74) (26 .20) (12 .48) (2 .24) (1 .69) (−1 .14) 2007–13 EME 0 .32 0 .28 0 .30 0 .15 0 .05 0 .00 50 .7% 9040

(11 .51) (10 .94) (8 .32) (4 .92) (2 .80) (1 .04)

fl

(

e

o

h

fl

(

t

t

d

t

B

c

f

I

U

f

i

o

0

t

a

s

p

t

s

e

w

r

a

ow–tangible capital sensitivity drops from 0.53 (t = 5.30) to 0.31 t = 2.92). A similar trend is observed for non-U.S. developed conomies. In contrast, as investment-intensive firms from devel-

ping economies operate with more tangible capital, they tend to

ave non-declining sensitivities. As reported, their aggregate cash

ow sensitivity stands at 0.46 (t = 4.53) in 1993–1999 and 0.49 t = 7.18) in 2007–2013.

Second, we rank firms based on their share of tangible assets in

otal assets in each country and year and assign to the capital in-

ensive group those firms in the top quarter of the asset tangibility

istribution. The purpose of this test is to rule out the possibility

hat our main results are driven by firms with low physical capital.

ecause we select capital-intensive firms for this test, we drop the

ross-product term of tangible capital with explanatory variables

rom our model. The estimation results are presented in Table 5 .

CFS dramatically falls from 0.28 (t = 10.3) to 0.11 (t = 5.21) in the

.S subsample. ICFS weakens from 0.36 (t = 13.2) to 0.21 (t = 8.09) or firms from other advanced economies. No evidence of decline

n the sensitivity of investment to cash flow is found for less devel-

ped economies. Over the sample period, their ICFS improves from

.27 (t = 6.24) to 0.32 (t = 11.5). Third, we construct a semi-balanced panel of firms that exist

hrough the 1993–2013 period. The number of firm-years varies

cross periods because several regression variables are missing in

ome parts of the sample. The purpose of this test is to rule out the

ossibility that ICFS and its relation to asset tangibility is driven by

he composition of the firms in the sample. Specifically, it is pos-

ible that more “new economy” firms enter the sample and “old

conomy” firms play a less essential role. In descriptive statistics,

e report that even for the unchanged firms in the panel, the

ate of investment and the degree of asset tangibility decline on

verage.

12 F. Moshirian et al. / Journal of Banking and Finance 77 (2017) 1–17

Table 6

Investment regressions: semi-balanced panel. The table reports coefficients estimated from the WLS regression of physical investment ( I i, t ) on cash flow ( �i, t ), beginning-of-

period tangible capital ( Z i,t−1 ), lagged investment ( I i,t−1 ), change in total debt ( D i, t ), new equity issue ( E i, t ), beginning-of-period cash reserves ( C i,t−1 ), and Tobin’s measure ( q i,t−1 ). Each explanatory variable has its cross-product term with tangible capital. The model specification corresponds to Model (4). The sample includes firms that exist through the entire sample period. US denotes U.S. economy. DME denotes non-U.S. developed economy. EME denotes a developing economy. The t-statistic reported in

parentheses adjusted for clustered standard errors at the country level. Country, industry and time fixed effects are included. N is the number of firm-year observations.

The R 2 is the adjusted R 2 .

Period Market �i, t ( �i,t ∗ Z i,t−1 ) I i,t−1 ( D i,t ∗ Z i,t−1 ) ( E i,t ∗ Z i,t−1 ) ( C i,t−1 ∗ Z i,t−1 ) q i,t−1 R 2 N �i,t + ( �i,t ∗ Z i,t−1 ) 1993–99 US 0 .00 0 .68 0 .49 0 .24 0 .14 0 .17 0 .00 53 .5% 3018 0 .67

(−0 .15) (5 .96) (8 .64) (3 .18) (1 .31) (2 .93) (1 .20) (7 .48) 20 0 0–06 US 0 .02 0 .38 0 .46 0 .27 0 .20 0 .07 0 .00 57 .7% 3273 0 .40

(1 .29) (5 .07) (6 .76) (5 .04) (2 .85) (1 .58) (0 .93) (6 .48)

2007–13 US −0 .02 0 .48 0 .50 0 .10 0 .15 0 .09 0 .00 60 .3% 3100 0 .45 (−1 .27) (5 .24) (10 .88) (2 .02) (2 .05) (2 .41) (−0 .40) (5 .94)

1993–99 DME 0 .09 0 .44 0 .28 0 .24 −0 .06 0 .18 0 .00 44 .9% 7351 0 .53 (1 .50) (3 .29) (4 .84) (2 .93) (−0 .66) (1 .97) (0 .02) (5 .90)

20 0 0–06 DME 0 .06 0 .39 0 .53 0 .35 0 .00 0 .14 0 .00 49 .2% 11 ,267 0 .45

(2 .39) (3 .41) (8 .08) (2 .40) (−0 .05) (1 .54) (−0 .21) (4 .67) 2007–13 DME 0 .01 0 .30 0 .61 0 .36 0 .16 0 .06 0 .00 59 .7% 11 ,566 0 .31

(0 .24) (2 .40) (11 .72) (3 .25) (1 .54) (1 .19) (−0 .38) (3 .28) 1993–99 EME −0 .06 0 .46 0 .26 0 .36 0 .24 0 .11 0 .00 46 .5% 2731 0 .40

(−0 .69) (2 .65) (3 .12) (3 .13) (1 .32) (1 .11) (0 .64) (3 .57) 20 0 0–06 EME 0 .11 0 .19 0 .37 0 .23 0 .44 0 .17 −0 .01 46 .3% 3498 0 .30

(2 .73) (1 .75) (4 .27) (2 .41) (5 .42) (3 .58) (−1 .51) (4 .44) 2007–13 EME −0 .02 0 .53 0 .54 0 .38 −0 .34 0 .21 0 .00 55 .4% 3614 0 .51

(−0 .64) (5 .67) (11 .39) (4 .15) (−2 .20) (4 .53) (0 .28) (6 .89)

Note: For brevity, the table reports the main variables of interest.

(

t

i

d

e

o

e

t

m

t

i

p

c

(

c

r

s

s

d

t

r

s

t

i

o

c

b

a

r

l

17 Although important for the ICFS literature, the q measurement concern is not

problematic for reaching the conclusion drawn in our study. In particular, if q is

a poor control for investment demand, it should be equally poor for firms with

different asset structures and investment rates. Thus, the measurement error would

not account for why economically motivated factors such as asset tangibility explain

the existence of and variation in ICFS.

Table 6 reports the regression results. In general, for firms that

exist through the entire sample period, the patterns are similar to

those reported in Table 2 , but they also show some differences

across periods. ICFS is still largely non-existent. The investment–

cash flow–tangible capital sensitivity is strong. The value declines

in the subsamples of U.S. firms and firms from other developed

economies, although not monotonically for the U.S. firms. The re-

sult for the subsample of firms from developing economies is less

distinct. The coefficient on the sensitivity stands at 0.46 in 1993–

1999, drops to 0.19 in 20 0 0–20 06 and improves to 0.53 in 2007–

2013. The sudden drop in the second period is accompanied by

relatively strong ICFS, making it difficult to explain. The aggre-

gate cash flow sensitivity steadily declines in the subsamples of

developed economy firms and varies within a certain range for

firms from developing economies. The results returned from the

full sample and semi-balanced panel are broadly consistent. This

consistency indicates that the documented patterns cannot be at-

tributed to the changing firm composition.

5.4. Measurement error in Tobin’s q

We now address the concern that the phenomenon docu-

mented in the previous sections is due to the errors-in-variables

problem. The alternative explanation is that ICFS, investment–

cash flow–tangible capital sensitivity, and their time-series pat-

terns could be explained by the correction for poorly measured q

or by improved quality of q , or both. To explore this possibility and

to address the attenuation bias in q , we use the cumulant estima-

tors from Erickson et al., (2014) . The cumulant-based estimation

results of the investment Model (4) are reported in Table 7 .

We find two important results. First, the measurement error-

consistent ICFS is consistently negative. The treated q is positive. As

discussed, we report the estimation results with the highest possi-

ble magnitude of the q coefficient. The magnitude of the treated q

is several times stronger than the magnitude of the WLS (biased) q .

Significantly, investment–cash flow–tangible capital sensitivity re-

mains highly positive and statistically significant in all regressions

performed.

Second, over the sample period, the sum of ICFS and

investment–cash flow–tangible capital sensitivity falls from 0.35

z-stat = 3.59) to 0.15 (z = 1.41) in the U.S. and from 0.52 (z = 7.09) o 0.33 (z = 5.01) in non-U.S. developed economies. Individually, nvestment–cash flow–tangible capital sensitivity also falls across

eveloped economies. The aggregate cash flow sensitivity strength-

ns from 0.21 (z = 2.49) to 0.37 (z = 9.49) for firms from less devel- ped economies. The patterns returned from the cumulant-based

stimations are qualitatively similar to their WLS counterparts;

herefore, we conclude that the proposed relation between invest-

ent and asset tangibility and its time-series patterns is immune

o the measurement error. 17

The measurement quality of q ( Tau index) is stable or declin-

ng over time. It is never above 0.4. Low proxy quality is ex-

ected because measurement error typically stems from large con-

eptual gap between empirical (average) q and the underlying true

marginal) q . The measurement quality does not improve and thus

annot explain the decline in ICFS.

Additionally, we estimate Model (4) without the q variable. The

esults are presented in Table 8 . The objective is to show how the

ensitivities evolve over time without q . If the time series of the

ensitivities are due to the improvement in the quality of q and not

ue to the changes in firm asset tangibility and investment activity,

hen we should observe a non-declining (or less declining) ICFS in

egressions without q . Further investigation, however, shows that

uch an explanation is unlikely. ICFS continues to decline across

he globe, whereas investment–cash flow–tangible capital sensitiv-

ty and the sum of sensitivities decline only for developed econ-

my firms.

Finally, we examine whether the developments of ICFS and its

onnection to asset structure are due to the changes in correlation

etween cash flow and q . If the developments of the sensitivities

re due to the correlation, then we should observe a declining cor-

elation in the advanced economies and flat (or increasing) corre-

ation in less developed economies. Fig. 4 plots the cross-sectional

F. Moshirian et al. / Journal of Banking and Finance 77 (2017) 1–17 13

Table 7

Investment regressions: cumulant estimators. The table reports coefficients estimated from the errors-in-variables regression of physical investment ( I i, t ) on cash flow ( �i, t ),

beginning-of-period tangible capital ( Z i,t−1 ), lagged investment ( I i,t−1 ), change in total debt ( D i, t ), new equity issue ( E i, t ), beginning-of-period cash reserves ( C i,t−1 ), and Tobin’s measure ( q i,t−1 ). Each explanatory variable has its cross-product term with tangible capital. The model specification corresponds to Model (4). The investment regressions are estimated with high-order cumulant estimators. US denotes U.S. economy. DME denotes non-U.S. developed economy. EME denotes a developing economy.

The z-statistic is reported in parentheses. N is the number of firm-year observations. The R 2 is the R 2 of the regression. Tau measure is the R 2 of the measurement equation

(i.e., index of measurement quality).

Period Market �i, t ( �i,t ∗ Z i,t−1 ) I i,t−1 ( D i,t ∗ Z i,t−1 ) ( E i,t ∗ Z i,t−1 ) ( C i,t−1 ∗ Z i,t−1 ) q i,t−1 R 2 Tau �i,t + ( �i,t ∗ Z i,t−1 ) 1993–99 US −0 .46 0 .81 0 .22 0 .03 0 .94 −0 .29 0 .06 49 .0% 0 .38 0 .35

(−3 .12) (4 .46) (4 .84) (0 .21) (2 .75) (−1 .79) (3 .27) (3 .59) 20 0 0–06 US −0 .30 0 .45 0 .19 0 .00 0 .35 −0 .12 0 .06 53 .1% 0 .36 0 .14

(−4 .33) (3 .84) (3 .70) (0 .03) (2 .25) (−1 .20) (5 .97) (1 .46) 2007–13 US −0 .35 0 .50 0 .27 0 .16 0 .42 −0 .08 0 .06 48 .9% 0 .32 0 .15

(−3 .19) (2 .98) (4 .38) (0 .96) (1 .34) (−0 .76) (3 .76) (1 .41) 1993–99 DME −0 .26 0 .78 0 .34 0 .08 −0 .09 −0 .34 0 .05 42 .7% 0 .29 0 .52

(−5 .30) (7 .27) (11 .58) (0 .55) (−0 .61) (−2 .49) (17 .22) (7 .09) 20 0 0–06 DME −0 .26 0 .69 0 .26 0 .04 0 .62 −0 .17 0 .05 46 .9% 0 .28 0 .43

(−8 .10) (9 .65) (10 .49) (0 .38) (3 .58) (−2 .02) (18 .65) (8 .30) 2007–13 DME −0 .35 0 .68 0 .29 −0 .12 0 .44 −0 .32 0 .06 47 .8% 0 .33 0 .33

(−8 .68) (7 .45) (11 .86) (−1 .13) (2 .48) (−3 .38) (22 .60) (5 .01) 1993–99 EME −0 .24 0 .45 0 .27 0 .36 −0 .06 −0 .26 0 .05 41 .7% 0 .27 0 .21

(−2 .26) (3 .28) (10 .42) (2 .11) (−0 .17) (−1 .19) (4 .72) (2 .49) 20 0 0–06 EME −0 .29 0 .59 0 .21 0 .37 0 .01 −0 .26 0 .09 42 .7% 0 .19 0 .30

(−3 .11) (4 .93) (9 .50) (2 .98) (0 .04) (−1 .77) (5 .50) (5 .37) 2007–13 EME −0 .11 0 .48 0 .25 0 .25 0 .04 −0 .30 0 .05 45 .3% 0 .23 0 .37

(−3 .08) (8 .16) (17 .88) (3 .08) (0 .34) (−4 .53) (7 .07) (9 .49)

Note: For brevity, the table reports the main variables of interest.

Table 8

Investment regressions without q . The table reports coefficients estimated from the WLS regression of physical investment ( I i, t ) on cash flow ( �i, t ), beginning-of-period

tangible capital ( Z i,t−1 ), lagged investment ( I i,t−1 ), change in total debt ( D i, t ), new equity issue ( E i, t ), and beginning-of-period cash reserves ( C i,t−1 ). Each explanatory variable has its cross-product term with tangible capital. The model specification corresponds to Model (4) (without Tobin’s measure). US denotes U.S. economy. DME

denotes non-U.S. developed economy. EME denotes a developing economy. The t-statistic reported in parentheses adjusted for clustered standard errors at the country level.

Country, industry and time fixed effects are included. N is the number of firm-year observations. The R 2 is the adjusted R 2 .

Period Market �i, t ( �i,t ∗ Z i,t−1 ) I i,t−1 ( D i,t ∗ Z i,t−1 ) ( E i,t ∗ Z i,t−1 ) ( C i,t−1 ∗ Z i,t−1 ) R 2 N �i,t + ( �i,t ∗ Z i,t−1 ) 1993–99 US 0 .05 0 .49 0 .42 0 .21 0 .15 0 .19 49 .7% 7295 0 .53

(2 .71) (7 .47) (12 .97) (4 .49) (2 .39) (5 .41) (10 .25)

20 0 0–06 US 0 .03 0 .38 0 .43 0 .25 0 .15 0 .09 48 .4% 7634 0 .42

(3 .04) (8 .13) (10 .89) (7 .23) (2 .44) (2 .94) (10 .56)

2007–13 US 0 .03 0 .23 0 .50 0 .24 0 .12 0 .18 46 .7% 5976 0 .26

(2.53) (3.35) (10.85) (5.01) (1.52) (5.57) (4.49)

1993–99 DME 0 .09 0 .45 0 .41 0 .23 0 .11 0 .15 42 .2% 14 ,420 0 .54

(2 .90) (6 .07) (8 .55) (3 .15) (1 .32) (2 .87) (10 .14)

20 0 0–06 DME 0 .05 0 .45 0 .40 0 .29 0 .10 0 .10 46 .7% 22 ,310 0 .50

(3 .31) (7 .71) (8 .89) (4 .57) (1 .46) (3 .66) (11 .15)

2007–13 DME 0 .02 0 .39 0 .42 0 .34 0 .20 0 .14 47 .3% 23 ,869 0 .41

(1 .64) (8 .43) (8 .89) (10 .12) (2 .85) (4 .95) (10 .54)

1993–99 EME 0 .08 0 .30 0 .40 0 .43 0 .29 0 .26 45 .2% 5207 0 .38

(1 .79) (2 .98) (8 .16) (5 .46) (2 .82) (4 .40) (5 .42)

20 0 0–06 EME 0 .08 0 .28 0 .40 0 .32 0 .17 0 .11 41 .4% 16 ,862 0 .35

(3 .07) (4 .93) (9 .28) (6 .34) (1 .72) (2 .98) (9 .70)

2007–13 EME 0 .02 0 .39 0 .45 0 .36 0 .13 0 .16 46 .0% 35 ,948 0 .41

(1 .42) (6 .73) (13 .82) (6 .42) (1 .99) (6 .88) (8 .63)

Note: For brevity, the table reports the main variables of interest.

Fig. 4. Correlation between cash flow and Tobin’s q . The figure plots the cross-sectional correlation between cash flow and Tobin’s (empirical) q in each year between 1993

and 2013. The left figure corresponds to the U.S. economy. The middle figure corresponds to non-U.S. developed economies. The right figure corresponds to developing

economies. Two data points in 1993 and 1994 years for developing economy firms are missing due to the lack of sufficient observations. The variables are demeaned by firm

to remove fixed effects.

14 F. Moshirian et al. / Journal of Banking and Finance 77 (2017) 1–17

Table 9

Cash flow regressions. Panel A reports coefficients estimated from the WLS regres-

sion of cash flow ( �i, t ) on lagged cash flow ( �i,t−1 ) from Model (5). Panel B re- ports coefficients estimated from the WLS regression of the cross-product term

of cash flow and beginning-of-period tangible capital ( �i,t ∗ Z i,t−1 ) on its lagged cross-product term ( �i,t−1 ∗ Z i,t−2 ) from Model (6). US denotes U.S. economy. DME denotes non-U.S. developed economy. EME denotes a developing economy. The t-

statistic reported in parentheses adjusted for clustered standard errors at the coun-

try level. Country, industry and time fixed effects are included. N is the number of

firm-year observations. The R 2 is the adjusted R 2 .

Panel A. Cash flow autoregressions

Period Market �i,t−1 (t-stat) R 2 N

1993–99 US 0 .35 (21 .86) 26 .4% 7295

20 0 0–06 US 0 .25 (19 .34) 17 .3% 7634

2007–13 US 0 .21 (14 .40) 14 .0% 5976

1993–99 DME 0 .47 (12 .44) 42 .0% 14 ,420

20 0 0–06 DME 0 .36 (14 .16) 30 .9% 22 ,310

2007–13 DME 0 .32 (11 .28) 26 .5% 23 ,869

1993–99 EME 0 .41 (8 .80) 39 .0% 5207

20 0 0–06 EME 0 .42 (10 .11) 39 .0% 16 ,862

2007–13 EME 0 .46 (17 .79) 40 .2% 35 ,948

Panel B. Cash flow-tangible capital autoregressions

Period Market �i,t−1 ∗Z i,t−2 (t-stat) R 2 N 1993–99 US 0 .58 (28 .14) 50 .8% 7292

20 0 0–06 US 0 .46 (20 .99) 40 .9% 7628

2007–13 US 0 .36 (9 .54) 32 .0% 5973

1993–99 DME 0 .62 (24 .89) 57 .0% 14 ,413

20 0 0–06 DME 0 .57 (16 .04) 55 .1% 22 ,302

2007–13 DME 0 .47 (8 .28) 44 .7% 23 ,863

1993–99 EME 0 .49 (9 .64) 43 .4% 5201

20 0 0–06 EME 0 .46 (8 .47) 41 .7% 16 ,857

2007–13 EME 0 .53 (19 .27) 47 .0% 35 ,941

p

e

h

o

v

fl

s

c

i

fi

n

c

s

7

a

i

p

correlation between cash flow and q in each year between 1993

and 2013. 18 The correlation is U-shaped in the U.S. economy: it de-

creases monotonically from 0.5 down to 0.3 in the early period but

increases above 0.4 in the later period. The correlation is some-

what similar, but generally lower, in other developed economies.

In contrast, the correlation shows a sharp decline and subsequent

recovery in developing economies. The correlation stands at 0.35

in the early period. The correlation is lowest in 2009 when it is

0.07. In 2013, it is 0.25. Interestingly, the correlation coefficients are

highest in the U.S and lowest in the developing part of the world. 19

In conclusion, the correlation between cash flow and q cannot be

the main driver of the time-series patterns of ICFS.

6. Cash flow persistence and asset tangibility

6.1. Cash flow persistence

According to q theory in general, investment varies with cash

flow because cash flow provides information about marginal q .

If the correlation between cash flow and q declines, ICFS can be

lower. In the previous section, we show that the simple correla-

tion between the two cannot completely explain the patterns in

ICFS. Still, Chen and Chen (2012) show that the information con-

tent in cash flow regarding investment opportunities has declined

for U.S firms. We now test whether the degree of asset tangibil-

ity is a determinant of the information content in cash flow. More

specifically, we investigate whether the declining share of tangible

capital can explain the declining predictability of cash flow.

Panel A and Panel B of Table 9 report the cash flow autore-

gressions from Models (5) and (6), respectively. The autoregres-

sive coefficient and its explanatory power have substantially de-

clined in the U.S. (from 0.35, t = 21.9 down to 0.21, t = 14.4) and non-U.S. developed economies (from 0.47, t = 12.4 down to 0.32, t = 11.3) but improved in emerging market economies (from 0.41, t = 8.80 up to 0.46, t = 17.8). When we estimate the predictabil- ity of the cross-product term of cash flow and tangible capital, we

find that it exhibits greater magnitude, higher goodness-of-fit and

similar time-series patterns. The patterns in cash flow persistence

are broadly consistent with the patterns in cash flow sensitivities.

As firms become less capital-intensive, tangibility plays a less es-

sential role, and its shrinking importance attenuates the predictive

power of the current cash flow. This loss of information eventually

contributes to the decline in ICFS in advanced economies. In sharp

contrast, due to the higher share of tangible capital in developing

economies, the current cash flow is more predictable. Its informa-

tion stability supports the magnitude of ICFS. These results are the

basis for the argument that asset tangibility predicts investment,

as well as the future value of cash flow. 20

6.2. Cash flow volatility

We now explore how cash flow volatility affects cash flow

sensitivities in the investment regression. Minton and Schrand

(1999) show that volatility is associated with depressed invest-

ment. It is therefore natural to expect that more volatile (and less

18 We omit two data points in 1993 and 1994 years due to the lack of sufficient

observations for firms from developing economies. We also demean the variables

by firm to remove fixed effects. 19 The correlation between cash flow and marginal q exhibits similar patterns . The

correlation of cash flow and marginal q is simply the correlation between cash flow

and empirical q scaled by the square root of the Tau measure. 20 In unreported results, we estimate cash flow persistence for capital-intensive

firms using the cash flow autoregressive Model ( 5 ). The predictability of cash flow

declines from 0.6 to 0.25 for U.S. firms and from 0.6 to 0.4 for other developed

economy firms. Cash flow predictability remains stable and varies between 0.4 and

0.5 for capital-intensive firms from emerging economies.

a

p

o

i

b

T

t

v

a

s

e

redictable) cash flow negatively affects the strength of ICFS. How-

ver, based on the evidence presented in the previous section, a

igher degree of tangibility should induce a greater persistence

f cash flow and increased stability of investment. The cash flow

olatility is defined as the standard deviation of the ratio of cash

ow over total assets in each seven-year period.

We find that cash flow volatility is associated with lower, but

till statistically significant, ICFS ( Table 10 ). However, when we in-

lude the combination of cash flow and tangible capital into the

nvestment regression, both the impact of volatility and the coef-

cients of ICFS become largely insignificant. Moreover, our origi-

al findings on the sensitivity of investment to the combination of

ash flow and tangible capital continue to hold. The results clearly

how that asset tangibility is a strong predictor of ICFS.

. Financial constraints and asset tangibility

Thus far, we have documented that the most conspicuous char-

cteristics of ICFS are driven by tangible capital. However, there

s another potential channel – the credit multiplier originally pro-

osed by Kiyotaki and Moore (1997) and manifested by Almeida

nd Campello (2007) . The latter study examines the effect of asset

ledgeability on corporate investment and documents that the ICFS

f financially constrained firms increases with asset tangibility. The

ntuition is that tangible assets (pledged collateral) support more

orrowing, which allows for further investment in tangible assets.

he investment of unconstrained firms is not sensitive to the sta-

us of cash flow; therefore, the share of tangible capital is irrele-

ant to ICFS. To the extent that physical assets are more pledgeable

nd support more investment, our results are consistent with their

tudy. However, we do not state that ICFS can be used to gauge the

ffect of financing frictions on investment.

F. Moshirian et al. / Journal of Banking and Finance 77 (2017) 1–17 15

Table 10

Investment regressions with cash flow volatility. The upper panel reports coefficients estimated from the WLS regression of physical investment ( I i, t ) on cash flow ( �i, t ),

cash flow volatility ( V i, t ) and its cross-product term with cash flow ( �i, t ∗ V i, t ), lagged investment ( I i,t−1 ), change in total debt ( D i, t ), new equity issue ( E i, t ), beginning-

of-period cash reserves ( C i,t−1 ), and Tobin’s measure ( q i,t−1 ). The lower panel reports coefficients estimated from the WLS regression of physical investment on cash flow, cash flow volatility, beginning-of-period tangible capital ( Z i,t−1 ), lagged investment, change in total debt, new equity issue, beginning-of-period cash reserves, and Tobin’s measure. Each explanatory variable has its cross-product terms with cash flow volatility and with tangible capital. US denotes U.S. economy. DME denotes non-U.S. developed

economy. EME denotes a developing economy. The t-statistic reported in parentheses adjusted for clustered standard errors at the country level. Country, industry and time

fixed effects are included. N is the number of firm-year observations. The R 2 is the adjusted R 2 .

Period Market �i, t (t-stat) ( �i, t ∗ V i, t ) (t-stat) ( �i,t ∗ Z i,t−1 ) (t-stat) �i,t + ( �i,t ∗ Z i,t−1 ) (t-stat) R 2 N

1993–99 US 0 .23 (13 .97) −0 .50 (−3 .73) 46 .0% 7295 20 0 0–06 US 0 .13 (9 .53) −0 .21 (−2 .00) 44 .9% 7546 2007–13 US 0 .12 (9 .60) −0 .40 (−4 .98) 42 .8% 5919 1993–99 DME 0 .32 (12 .39) −0 .93 (−2 .95) 39 .5% 14 ,420 20 0 0–06 DME 0 .25 (15 .51) −0 .51 (−4 .31) 42 .8% 22 ,238 2007–13 DME 0 .15 (7 .29) −0 .50 (−3 .76) 42 .8% 23 ,754 1993–99 EME 0 .15 (4 .43) −0 .43 (−2 .46) 42 .7% 5207 20 0 0–06 EME 0 .24 (13 .32) −0 .49 (−3 .55) 39 .6% 16 ,829 2007–13 EME 0 .20 (6 .01) −0 .52 (−1 .85) 42 .9% 35 ,904 1993–99 US 0 .04 (1 .64) −0 .13 (−0 .94) 0 .51 (6 .77) 0 .55 (9 .57) 49 .9% 7295 20 0 0–06 US 0 .02 (1 .55) −0 .04 (−0 .44) 0 .37 (7 .44) 0 .39 (9 .40) 49 .2% 7546 2007–13 US 0 .07 (3 .27) −0 .34 (−3 .65) 0 .18 (2 .55) 0 .25 (4 .37) 47 .3% 5919 1993–99 DME 0 .11 (2 .88) −0 .28 (−0 .99) 0 .43 (4 .50) 0 .54 (7 .37) 42 .7% 14 ,420 20 0 0–06 DME 0 .08 (3 .52) −0 .26 (−1 .91) 0 .41 (6 .31) 0 .49 (10 .45) 46 .8% 22 ,238 2007–13 DME 0 .03 (1 .31) −0 .18 (−1 .40) 0 .34 (6 .63) 0 .37 (7 .67) 47 .9% 23 ,754 1993–99 EME 0 .09 (1 .74) −0 .14 (−0 .43) 0 .17 (1 .77) 0 .26 (3 .07) 45 .8% 5207 20 0 0–06 EME 0 .13 (4 .04) −0 .58 (−3 .32) 0 .27 (4 .46) 0 .40 (10 .00) 41 .8% 16 ,829 2007–13 EME 0 .04 (1 .53) −0 .26 (−1 .04) 0 .38 (6 .63) 0 .42 (7 .20) 46 .1% 35 ,904

Note: For brevity, the table reports the main variables of interest.

o

a

fi

h

m

t

fl

w

t

t

i

t

e

s

s

s

s

c

o

t

o

(

c

b

t

a

j

t

e

f

q

s

i

a

(

f

fi

t

c

o

s

c

p

t

fi

i

fi

h

m

fi

e

a

d

t

m

t

t

c

c

g

s

f

t

t

n

21 We omit another broadly used scheme — the Kaplan–Zingales index ( Kaplan

and Zingales 1997 ) — because this measure is endogenously determined with the

investment decision. We also do not consider bond or commercial paper ratings

To explore the possibility that the credit multiplier overshad-

ws the effect from two other channels, we examine the impact of

sset tangibility on ICFS for financially more and less constrained

rms. If the first channel (i.e., the intensity of physical investment)

olds, the high tangible capital firms invest more from higher

arginal cash flow, irrespective of their exposure to financing fric-

ions. If the second channel (i.e., the information content in cash

ow) holds, a marginal cash flow is more predictable for firms

ith higher tangible capital. Both channels yield a similar predic-

ion with respect to the sign and significance of ICFS. However,

here is no clear prediction of how a firm with high tangible cap-

tal should react when it faces costly external finance. According

o the credit multiplier, the positive effect of tangibility on ICFS is

vident only for financially constrained firms, and no such effect

hould be observed for unconstrained firms.

To empirically test the credit multiplier, we estimate the regres-

ion Model (4) for financially constrained and unconstrained firms

eparately. If the credit multiplier channel dominates, more con-

trained firms should have a higher investment–cash flow–tangible

apital sensitivity, whereas unconstrained firms should have lower,

r no, sensitivity. We adopt two popular classification schemes

o differentiate our sample firms: firm size and dividend pay-

ut. Gilchrist and Himmelberg (1995) and Hennessy and Whited

2007) demonstrate that firm size is an accurate indicator of the

ost of raising external funds. Size can be considered exogenous

ecause it is not a choice variable for the manager in the short

erm. For the purpose of our study, the sample firms with total

ssets below the 30th percentile of the distribution for country

in year t are considered financially constrained. Firms with to-

al assets above the 70th percentile of the distribution are consid-

red unconstrained. Fazzari et al., (1988) also posit that financing

rictions are more binding on non-dividend-paying firms. Conse-

uently, non-dividend-paying firms are treated as financially con-

trained, and dividend-paying firms are treated as unconstrained

n a particular year. For robustness check (unreported), we also use

nother popular scheme — the Whited–Wu Index. Whited and Wu

2006) developed an index to estimate the likelihood that a firm

aces financial constraints. The index is a linear combination of six

b

rm and industry-specific characteristics. Firms in the top (bottom)

hree deciles of the annual distribution are considered financially

onstrained (unconstrained). Firms are able to change their status

ver the sample period because they are ranked on an annual ba-

is. 21

The regression results are reported in Table 11 . First, for both

onstrained and unconstrained firms, the presence of the cross-

roduct term of cash flow and tangible capital continues to reduce

he significance of the conventional ICFS. In most cases, the coef-

cient of cash flow becomes statistically or economically insignif-

cant (below 0.1) after the product term has been included. This

nding is similar to that observed previously, which means that

ow the investment of a firm is sensitive to the cash flow is deter-

ined, to a large extent, by the size of tangible capital it operates.

Second, irrespective of which criterion is used to differentiate

nancially more and less constrained firms, we find that the co-

fficients (i) on the combination of cash flow and tangible capital

nd (ii) on the aggregate cash flow sensitivity are not consistently

ifferent between the subsamples of firms. This finding suggests

hat tangibility is more likely to affect ICFS through the invest-

ent channel and the information content channel than through

he credit multiplier channel. Our results do not necessarily reject

he notion of the pledgeability of tangible assets but, rather, indi-

ate that financial constraints are not important for reaching the

onclusion drawn in this study.

Finally, and consistent with previous findings, the effect of tan-

ible capital on ICFS declines in general for both financially con-

trained and unconstrained firms in developed economies. The ef-

ect does not diminish for firms in developing economies, irrespec-

ive of their exposure to financing frictions. This evidence indicates

hat the explanatory power of tangibility does not derive from fi-

ancial constraints.

ecause too few firms in the international sample have such ratings.

16 F. Moshirian et al. / Journal of Banking and Finance 77 (2017) 1–17

Table 11

Investment regressions: constrained and unconstrained firms. The table reports coefficients estimated from the WLS regression of physical investment ( I i, t ) on cash flow

( �i, t ), beginning-of-period tangible capital ( Z i,t−1 ), lagged investment ( I i,t−1 ), change in total debt ( D i, t ), new equity issue ( E i, t ), beginning-of-period cash reserves ( C i,t−1 ), and Tobin’s measure ( q i,t−1 ). Each explanatory variable has its cross-product term with tangible capital. Firms are classified as the constrained (FC) and unconstrained (FUC) by dividend payout (Panel A) and firm size (Panel B). US denotes U.S. economy. DME denotes non-U.S. developed economy. EME denotes a developing economy. The

t-statistic reported in parentheses adjusted for clustered standard errors at the country level. Country, industry and time fixed effects are included. N is the number of

firm-year observations. The R 2 is the adjusted R 2 .

Panel A. Dividend scheme

Period Market Status �i, t (t-stat) ( �i,t ∗ Z i,t−1 ) (t-stat) R 2 �i,t + ( �i,t ∗ Z i,t−1 ) (t-stat) 1993–99 US FC 0 .02 (0 .66) 0 .70 (6 .93) 49 .7% 0 .71 (8 .77)

20 0 0–06 US FC 0 .01 (0 .50) 0 .43 (6 .62) 47 .4% 0 .44 (7 .71)

2007–13 US FC 0 .03 (2 .09) 0 .22 (3 .34) 44 .0% 0 .25 (4 .34)

1993–99 US FUC 0 .04 (1 .53) 0 .36 (3 .77) 52 .1% 0 .41 (5 .53)

20 0 0–06 US FUC 0 .08 (3 .67) 0 .20 (2 .57) 54 .2% 0 .28 (4 .67)

2007–13 US FUC 0 .06 (1 .81) 0 .09 (0 .61) 54 .5% 0 .15 (1 .21)

1993–99 DME FC 0 .09 (2 .15) 0 .34 (2 .45) 44 .8% 0 .43 (4 .12)

20 0 0–06 DME FC 0 .10 (2 .80) 0 .33 (2 .83) 37 .9% 0 .43 (4 .69)

2007–13 DME FC 0 .04 (2 .28) 0 .30 (3 .82) 41 .0% 0 .35 (5 .34)

1993–99 DME FUC 0 .10 (4 .67) 0 .50 (6 .56) 47 .7% 0 .60 (8 .96)

20 0 0–06 DME FUC 0 .06 (2 .84) 0 .46 (7 .97) 52 .9% 0 .52 (10 .45)

2007–13 DME FUC 0 .04 (2 .61) 0 .36 (4 .33) 51 .1% 0 .40 (5 .20)

1993–99 EME FC 0 .18 (1 .64) 0 .15 (1 .67) 44 .9% 0 .33 (2 .46)

20 0 0–06 EME FC 0 .09 (2 .44) 0 .33 (3 .97) 43 .3% 0 .43 (6 .76)

2007–13 EME FC 0 .06 (3 .61) 0 .44 (9 .47) 42 .2% 0 .50 (13 .46)

1993–99 EME FUC 0 .15 (2 .44) 0 .19 (1 .84) 43 .2% 0 .33 (3 .13)

20 0 0–06 EME FUC 0 .06 (2 .98) 0 .40 (5 .64) 44 .8% 0 .46 (8 .61)

2007–13 EME FUC 0 .01 (0 .37) 0 .53 (8 .68) 47 .2% 0 .54 (10 .02)

Panel B. Firm size scheme

Status �i, t (t-stat) ( �i,t ∗Z i,t−1 ) (t-stat) R 2 �i,t + ( �i,t ∗Z i,t−1 ) (t-stat) FC 0 .02 (0 .75) 0 .49 (3 .87) 42 .2% 0 .51 (4 .96)

FC 0 .01 (0 .48) 0 .47 (5 .67) 38 .1% 0 .48 (6 .64)

FC 0 .01 (0 .83) 0 .25 (2 .99) 33 .1% 0 .27 (3 .59)

FUC 0 .07 (2 .17) 0 .31 (2 .83) 64 .9% 0 .38 (4 .76)

FUC 0 .04 (1 .50) 0 .30 (3 .95) 68 .6% 0 .34 (5 .74)

FUC 0 .00 (0 .13) 0 .35 (3 .07) 71 .4% 0 .35 (3 .86)

FC 0 .06 (1 .64) 0 .62 (5 .55) 55 .7% 0 .68 (8 .08)

FC 0 .07 (2 .39) 0 .48 (5 .17) 63 .2% 0 .55 (7 .75)

FC 0 .04 (2 .82) 0 .34 (4 .55) 65 .7% 0 .38 (5 .51)

FUC 0 .07 (1 .58) 0 .45 (3 .75) 55 .7% 0 .52 (5 .50)

FUC 0 .03 (1 .07) 0 .50 (5 .05) 63 .2% 0 .53 (6 .29)

FUC 0 .06 (2 .41) 0 .27 (2 .61) 65 .7% 0 .34 (3 .77)

FC 0 .16 (1 .78) 0 .24 (1 .74) 34 .6% 0 .40 (2 .91)

FC 0 .04 (2 .15) 0 .43 (6 .22) 36 .9% 0 .47 (7 .95)

FC 0 .03 (1 .39) 0 .44 (5 .50) 34 .7% 0 .47 (7 .16)

FUC 0 .06 (0 .72) 0 .24 (1 .79) 48 .7% 0 .31 (3 .47)

FUC 0 .09 (2 .69) 0 .33 (4 .20) 52 .7% 0 .42 (7 .69)

FUC 0 .02 (0 .73) 0 .51 (10 .38) 56 .0% 0 .53 (14 .05)

Note: For brevity, the table reports the main variables of interest.

fl

i

g

g

c

p

o

e

s

o

v

T

a

w

s

a

8. Conclusion

The literature on ICFS is rich but inconclusive. In particular, no

published paper provides a plausible explanation for why this sen-

sitivity has declined and virtually disappeared over time in the U.S.

There is also little evidence from non-U.S. economies. In this study,

we examine two channels that could potentially explain the exis-

tence and the time-series development of ICFS. First, we empha-

size the importance of asset tangibility in predicting physical in-

vestment. The share of tangible capital in total productive capital

contains information about its marginal productivity. The rate of

investment varies with the productivity and with the productive

capital structure. Our argument is that firms with high tangible

capital spend more on investment. Second, we extend the notion

that the current cash flow explains investment because it predicts

future cash flow. The firms with more predictable cash flows ex-

ploit new growth opportunities and invest more. Our argument is

that firms with high tangible capital have more predictable cash

ows. Because asset tangibility predicts the intensity of physical

nvestment and the persistence of cash flow, it contributes to ICFS.

Our empirical results confirm our prediction. We show that the

lobal patterns in ICFS are explained well by the variation in tan-

ible capital. Because developed economy firms use less tangible

apital and adopt more intangible capital in production, the com-

osition of their investments has changed, and the persistence

f their cash flows has weakened. Their ICFS is virtually non-

xistent, and investment–cash flow–tangible capital sensitivity has

teadily declined over time. In contrast, because developing econ-

my firms operate with more tangible capital, their physical in-

estments are more stable and cash flows are more predictable.

heir ICFS and investment–cash flow–tangible capital sensitivity

re therefore non-diminishing.

Finally, we show that q measurement error is not the reason

hy ICFS exists. This error does not sufficiently explain the time

eries of this sensitivity or the relation between this sensitivity and

sset tangibility. Furthermore, we provide evidence that the ex-

F. Moshirian et al. / Journal of Banking and Finance 77 (2017) 1–17 17

p

s

I

p

R

A

A

A

A

A

B

B

B

B

B

B

C

C

C

C

E

E

E

F

F

F

G

G

G

G H

H

H

H

I

K

K

K

K

K

L

L

L

M

M

P

R

T

S

W

lanatory power of tangibility does not derive from financial con-

traints. The net results of this study support our explanation of

CFS as a reflection of capital (investment) intensity and income

redictability, rather than an indication of financial constraints.

eferences

gca, S. , Mozumdar, A. , 2008. The impact of capital market imperfections on invest-

ment–cash flow sensitivity. J. Bank. Finance 32, 207–216 . llayannis, G. , Mozumdar, A. , 2004. The impact of negative cash flow and influential

observations on investment–cash flow sensitivity estimates. J. Bank. Finance 28, 901–930 .

lmeida, H. , Campello, M. , 2007. Financial constraints, asset tangibility, and corpo-

rate investment. Rev. Financ. Stud. 20, 1429–1460 . lti, A. , 2003. How sensitive is investment to cash flow when financing is friction-

less? J. Finance 58, 707–722 . scioglu, A. , Hegde, S. , McDermott, J. , 2008. Information asymmetry and investmen-

t–cash flow sensitivity. J. Bank. Finance 32, 1036–1048 . ates, T. , Kahle, K. , Stulz, R. , 2009. Why do U.S. firms hold so much more cash than

they used to? J. Finance 64, 1985–2021 .

eatty, A. , Liao, S. , Weber, J. , 2010. The effect of private information and monitor- ing on the role of accounting quality in investment decisions. Contemporary

Account. Res. 27, 17–47 . iddle, G. , Hilary, G. , 2006. Accounting quality and firm-level capital investment.

Account. Rev. 81, 963–982 . ond, S. , Elston, J. , Mairesse, J. , Mulkay, B. , 2003. Financial factors and investment in

Belgium, France, Germany, and the United Kingdom: a comparison using com-

pany panel data. Rev. Econ. Stat. 85, 153–165 . rown, J. , Fazzari, S. , Petersen, B. , 2009. Financing innovation and growth: cash flow,

external equity, and the 1990s R&D boom. J. Finance 64, 151–185 . rown, J. , Petersen, B. , 2009. Why has the investment–cash flow sensitivity declined

so sharply? rising R&D and equity market developments. J. Bank. Finance 33, 971–984 .

hen, H. , Chen, S. , 2012. Investment–cash flow sensitivity cannot be a good measure

of financial constraints: evidence from time series. J. Financ. Econ. 103, 393–410 . leary, S. , 1999. The relationship between firm investment and financial status. J.

Finance 54, 673–692 . leary, S. , 2006. International corporate investment and the relationship between

financial constraint measures. J. Bank. Finance 30, 1559–1580 . leary, S. , Povel, P. , Raith, M. , 2007. The U-shaped investment curve: theory and

evidence. J. Financ. Quant. Anal. 42, 1–39 .

rickson, T. , Jiang, C. , Whited, T. , 2014. Minimum distance estimation of the er- rors-in-variables model using linear cumulant equations. J. Econometrics 183,

211–221 . rickson, T. , Whited, T. , 20 0 0. Measurement error and the relationship between in-

vestment and q. J. Polit. Econ. 108, 1027–1057 . rickson, T. , Whited, T. , 2002. Two-step GMM estimation of the errors-in-variables

model using higher order moments. Econometric Theory 18, 776–799 . azzari, S. , Hubbard, R. , Petersen, B. , 1988. Financing constraints and corporate in-

vestment. Brookings Papers Econ. Activity 141–195 .

azzari, S. , Petersen, B. , 1993. Working capital and fixed investment: new evidence on financing constraints. Rand. J. Econ. 24, 328–342 .

azzari, S. , Hubbard, R. , Petersen, B. , 20 0 0. Investment–cash flow sensitivities are useful: a comment on Kaplan and Zingales. Q. J. Econ. 115, 695–705 .

atchev, V. , Pulvino, T. , Tarhan, V. , 2010. The interdependent and intertemporal na- ture of financial decisions: an application to cash flow sensitivities. J. Finance

65, 725–763 . eary, R. , 1942. Inherent relations between random variables. Proc. R. Irish Acad. A

47, 63–76 . ilchrist, S. , Himmelberg, D. , 1995. Evidence on the role of cash flow for investment.

J. Monet. Econ. 36, 541–572 .

omes, J. , 2001. Financing investment. Am. Econ. Rev. 91, 1263–1285 . all, R. , 2001. The stock market and capital accumulation. Am. Econ. Rev. 91,

1185–1202 . ansen, L. , Heaton, J. , Li, N. , 2005. Intangible risk?. In: Corrado, C., Haltiwanger, J.,

Sichel, D. (Eds.), Measuring Capital in the New Economy. The University of Chicago Press .

ennessy, C. , Whited, T. , 2007. How costly is external financing? Evidence from a

structural estimation. J. Finance 62, 1705–1745 . oshi, T. , Kashyap, A. , Scharfstein, D. , 1991. Corporate structure liquidity and invest-

ment: evidence from Japanese panel data. Q. J. Econ. 106, 33–60 . slam, S. , Mozumdar, A. , 2007. Financial development and the importance of internal

cash: evidence from international data. J. Bank. Finance 31, 641–658 . adapakkam, P. , Kumar, P. , Riddick, L. , 1998. The impact of cash flows and firm size

on investment: the international evidence. J. Bank. Finance 22, 293–320 .

aplan, S. , Zingales, L. , 1997. Do investment–cash flow sensitivities provide useful measures of financing constraints? Q. J. Econ. 112, 169–215 .

aplan, S. , Zingales, L. , 20 0 0. Investment–cash flow sensitivities are not valid mea- sures of financing constraints. Q. J. Econ. 115, 707–712 .

ashyap, A. , Lamont, O. , Stein, J. , 1994. Credit conditions and the cyclical behavior of inventories. Q. J. Econ. 109, 565–592 .

iyotaki, N. , Moore, J. , 1997. Credit cycles. J. Polit. Econ. 105, 211–248 .

amont, O. , 1997. Cash flow and investment: Evidence from internal capital markets. J. Finance 52, 83–110 .

ewellen, J. , Lewellen, K. , 2016. Investment and cashflow: new evidence. J. Financ. Quant. Anal. 51, 1135–1164 .

ove, I. , 2003. Financial development and financing constraints: international evi- dence from the structural investment model. Rev. Financ. Stud. 16, 765–791 .

inton, B. , Schrand, C. , 1999. The impact of cash flow volatility on discretionary

investment and the costs of debt and equity financing. J. Financ. Econ. 54, 423–460 .

oyen, N. , 2004. Investment–cash flow sensitivities: Constrained versus uncon- strained firms. J. Finance 59, 2061–2092 .

inkowitz, L. , Stulz, R. , Williamson, R. , 2012. Multinationals and the high cash hold- ings puzzle. National Bureau of Economic Research Working Paper, 18120 .

iddick, L. , Whited, T. , 2009. The corporate propensity to save. J. Finance 64,

1729-1765 . obin, J. , 1969. A general equilibrium approach to monetary theory. J. Money Credit

Bank. 1, 15–29 . hin, H. , Stulz, R. , 1998. Are internal capital markets efficient? Q. J. Econ. 113,

531–552 . hited, T. , Wu, G. , 2006. Financial constraint risk. Rev. Financ. Stud. 19, 531–559 .