Asymmetric cash flow sensitivity of cash holdings

Using a sample of manufacturing firms from 1972 to 2006, we document that the cash flow sensitivity of cash is negative when a JEL classification: G00 firm faces a positive cash flow e

“= |CORPORATE —= |FINANCE

Journal of Corporate Finance

Asymmetric cash flow sensitivity of cash holdings

Dichu Bao 2, Kam C Chan P*, Weining Zhang “

* Department of Accounting, Naveen Jindal School of Management, University of Texas at Dallas, Richardson, TX 75080-1407, United States

© Department of Finance, Gordon Ford College of Business, Western Kentucky University, Bowling Green, KY 42101, United States

© Cheung Kong Graduate School of Business, Beijing, 100738, China

Article history: Almeida, Campello, and Weisbach (2004) and Riddick and Whited (2009) offer contrasting Received 9 August 2010 conclusions regarding the corporate cash flow sensitivity of cash We use an augmented

Received in revised form 2 May 2012

Accepted 4 May 2012

Available online 12 May 2012

empirical model to affirm the conclusion in Riddick and Whited that the cash flow sensitivity

of cash is generally negative In addition, we contend that the cash flow sensitivity of cash is asymmetric to cash flow The asymmetry may be due to several reasons, including binding project contracts, bad news withholding, and agency costs Using a sample of manufacturing firms from 1972 to 2006, we document that the cash flow sensitivity of cash is negative when a

JEL classification:

G00 firm faces a positive cash flow environment, supporting Riddick and Whited (2009), but the

cash flow sensitivity of cash is positive when a firm faces negative cash flows We further divide firms into financially constrained and unconstrained ones and find that the cash flow

Keywords: sensitivity of cash asymmetry continues to hold in both groups When we use institutional

cash holdings holding as a control for the agency problem, we find that firms with better outside monitoring Asymmetry dissave to capture good investment opportunities All the results support our hypotheses that

firms have different levels of responses to their cash holdings when facing positive and negative cash flows

© 2012 Elsevier B.V All rights reserved

1 Introduction

Prior literature has widely studied why firms hold cash and has arrived at several explanations The precautionary motive suggests that firms use cash holdings to finance new investments or incoming short-term liabilities when there is an anticipated adverse cash flow shock Cash holding, in essence, helps avoid the high cost of external financing in case of cash shortfall For example, Bates et al (2009) attribute the increase in cash to the growing liquidity demand to buffer against cash flow shock Almeida et al (2004) have developed a model to show that only financially constrained firms, because of their restricted

access to external capital markets, save cash out of cash flow, while unconstrained firms do not The authors examine the effect of

changes in cash holdings on cash flow (called the cash flow sensitivity of cash) for a sample of manufacturing firms and their findings are consistent with their argument The positive relation between changes in cash holding and cash flow indicates that firms decrease (increase) cash holdings when they have negative (positive) cash flows Riddick and Whited (2009) reexamine the corporate cash flow sensitivity of cash! with a different theoretical and empirical model specification and find the relation between changes in cash holding and cash flow to be negative

* We acknowledge the helpful comments from an anonymous reviewer and the seminar participants at National University of Kaohsiung, Sun Yat-Sen University, Ling Tung University, and National Changhua University of Education An earlier version of the paper was presented at the Asian Financial Association conference in Macau Weining Zhang acknowledges the financial supports from National Natural Science Foundation of China on the projects #71102077 and

#71002058 The usual caveats apply

* Corresponding author at: Gordon Ford College of Business, Western Kentucky University, Bowling Green, KY 42101, United States Tel.: +1 270 745 2977 E-mail address: Johnny.Chan@wku.edu (K.C Chan)

1 Riddick and Whited (2009) call the cash flow sensitivity of cash the corporate propensity to save To avoid confusion, we follow the terminology of Almeida et

al (2004)

0929-1199/$ - see front matter © 2012 Elsevier B.V All rights reserved

doi:10.1016/j.jcorpfin.2012.05.003

D Bao ef al / Journal of Corporate Finance 18 (2012) 690-700 691 Riddick and Whited reconcile their contradictory conclusion of the cash flow sensitivity of cash relative to that of Almeida et al in two aspects First, unlike the model of Almeida et al, the Riddick-Whited model is more realistic since it allows for a variety of capital depreciation rates and cash flow shocks that may or may not be tied to productivity Second, Riddick and Whited correctly point out that when an explanatory variable has a measuring error, the signs of the other explanatory variables may change They show that the cash flow sensitivity of cash is positive if there is no correction for a measuring error in Tobin's Q (a control variable in the empirical model) After correcting the bias induced by Tobin's Q measuring error using a general method of moment (GMM} estimation, their findings suggest a negative cash flow sensitivity of cash Nonetheless, the empirical model of Riddick and Whited Jacks several control variables such as firm size, capital expenditure, non-cash working capital, and short-term debt level,

The linear relation between changes in cash holding and cash flow, whether positive or negative, implicitly assumes that the magnitude of cash holding change is the same regardless of the direction of cash flow However, the possibility of cash shortfall varies with the level of cash reserve (Faulkender and Wang, 2006) When a firm faces positive cash flow, the cash flow sensitivity

of cash follows normal intuition; in contrast, when a firm faces negative cash flow, the cash flow sensitivity of cash differs from

that in a positive cash flow environment We argue that the cash flow sensitivity of cash is asymmetric to cash flow due to several reasons, inchiding binding project contracts, bad news withholding, and agency costs, While Riddick and Whited (2009, p 1793} briefly show that medium and large firms exhibit a negative nonlinearity in the cash flow sensitivity of cash, the focus of their study is not the nonlinear nature of the relation

We have two objectives in this study First, we reexamine whether the cash flow sensitivity of cash is positive or negative, using a comprehensive specification of the empirical model While Riddick and Whited (2005) document a negative cash flow sensitivity of cash, their empirical model lacks several of the control variables used in Almeida et al (2004) Hence a reexamination using an augmented empirical model is warranted Second, after affirming the negative cash flow sensitivity of cash, we investigate whether firms exhibit asymmetric cash sensitivity when facing positive or negative cash flows The nonlinearity suggests that a firm's cash flaw sensitivity of cash, while negative in a positive cash flow environment, is not necessarily negative in a negative cash flow environment

Cash holding enables firms to fund investments and other liabilities to avoid the high cost of raising funds (Acharya et al., 2007;

Almeida et al, 2004; Bates et al, 2009) Riddick and Whited (2085) contend that firms with an increase in cash flow tend to turn cash

holdings into investments because the positive cash flow shock indicates higher productivity of physical assets Therefore the firm draws down its cash holdings to fund its productive projects Intuitively this explains Riddick and Whited's results, that when a firm's cash flow is positive, the change in cash holdings is negative In contrast, when a firm exhibits a negative cash flow, this indicates the low productivity of existing physical assets or the firm has negative net present value (NPV) projects Thus, according to the argument

of Riddick and Whited, a firm accumulates cash holdings given they terminate less profitable projects

We argue that a firm cannot immediately terminate all negative NPV projects when it faces negative cash flows for three passible reasons First, some bad projects have binding contracts such that the firm cannot terminate them immediately This is particularly common when the projects are the result of bidding When a firm underbids, ic has negative cash flow from the project but it cannot back out of the contract Second, as suggested by Kothari ef ai, (2009), managers have incentives to withhold bad news Ifa firm terminates bad projects, then the market is more likely to know that problems are occurring in the firm Hence some managers may choose to keep bad projects going to minimize bad news to the market By “burying” the bad news (Le., not terminating bad projects), managers hope to buy time until some good news arrives to neutralize the impact of bad news Therefore, when a firm faces negative cash flow, it may not immediately terminate all bad projects Third, the agency cost perspective of jensen and Meckling (1976) and Jensen (1986) suggests that managers have incentives to overinvest and thus they are likely to hald onto some negative NPV projects to maximize their personal benefits If managers seek private benefits by overinvesting in unprofitable projects and are unwilling to terminate them, the increase in cash holdings due to savings from terminating unprofitable projects would be less than the decrease in cash holdings due to dissaving from pursuing profitable projects, even incurring negative cash holding

Because of binding project contracts, bad news withholding, and agency costs, a firm facing negative cash flows may not immediately terminate all bad projects As a result, the negative cash flow sensitivity of cash holds only when a firm has positive cash flows If a firm has negative cash flows, we contend that the negative cash flow sensitivity of cash may or may not hold Using a sample of manufacturing firms from 1972 to 2006, we examine the cash flow sensitivity of cash by incorporating the passibility of nonlinearity mì the response of changes in cash holdings between positive and negative cash flaws The empirical evidence is consistent with our prediction: The cash flow sensitivity of cash is negative when a firm faces a positive cash flow environment, supporting Ruddick and Whited (2009), but it is positive when a firm faces negative cash flows We further categorize firms as financially constrained and unconstrained based on four criteria: the Whited-Wu (WW) index (in Whited and

Wu, 2066), payout ratio, firm size, and bond rating The cash flow sensitivity of cash asymmetry continues to hold in both financially constrained and unconstrained firms Financially constrained firms, compared to unconstrained firms, are less likely to invest in pew projects or fund existing unprofitable projects because they find it hard to obtain external financing In addition, when we use institutional holdings as a contro! for agency problems, we find that firms with better outside monitoring (ie, a larger proportion of institutional ownership) dissave to capture good investment opportunities AH the results support our hypothesis that firms have different levels of responses to their cash holdings when facing positive and negative cash flows

We make several contributions to the literature First, consistent with Riddick and Whited (2609), we document a generally negative cash flow sensitivity of cash The findings are robust to different empirical model specifications Second, we show that the relation between changes in cash holdings and cash flow is nonlinear The negative cash flow sensitivity of cash holds in a positive cash flaw environment In a negative cash flaw environment, the cash flow sensitivity of cash becomes positive That is,

the response to a change in cash holding to a positive cash flow is different from that of a negative cash flow.” Finally, we offer several arguments as possible explanations to the asymmetry of the cash flow sensitivity of cash

2 Hypothesis development

We extend the literature (Almeida et al., 2004; Riddick and Whited, 2009) to incorporate the asymmetry of the cash flow sensitivity of cash in positive and negative cash flow environments We argue that because of possible binding project contracts, bad news withholding, and agency costs, the magnitude and direction of cash holding responses to negative cash flows differ from those to positive cash flows Firms experiencing positive (negative) cash flow have substantially more (fewer) investment opportunities When a firm has positive cash flows, it dissaves to pursue investment opportunities, which leads to a negative cash flow sensitivity of cash When facing negative cash flows, a firm would not save the same magnitude of cash holdings by terminating all bad projects or even by drawing down cash holdings to support ongoing bad projects due to binding project contracts, incentives to withhold bad news, and/or agency problems Hence, the cash flow sensitivity of cash exhibits asymmetry

in positive and negative cash flow environments Our first hypothesis is as follows

H1 The cash flow sensitivity of cash when a firm faces positive cash flows is different from that when the firm faces negative cash flows

As in the literature, we also consider the cash flow sensitivity of cash asymmetry when a firm faces financial constraints When cash flows are negative, we expect a financially constrained firm to continue to fund some negative NPV projects due to binding project contracts, bad news withholding, and agency problems However, we contend that the extent of the funding of the negative NPV projects is less due to the financial constraints Thus our second hypothesis is as follows

H2 The magnitude of the asymmetry of the cash flow sensitivity of cash is less for a financially constrained firm than for a financially unconstrained firm

To explicitly consider the impact of the agency problem, we also incorporate the impact of institutional holdings for each firm

We contend that a firm with a high percentage of institutional holdings has less of an agency problem due to strong outside monitoring Accordingly, the magnitude of the cash flow sensitivity of cash is larger for firms with stronger outside monitoring Our third hypothesis is as follows

H3 The magnitude of the cash flow sensitivity of cash is larger for firms with stronger outside monitoring

3 Methods

3.1 Base model

We contend that the cash flow sensitivity of cash is different when a firm faces positive cash flows from when it faces negative cash flows We use the following model to test our hypothesis:

ACashHoldings;, = lg + a, CashFlow,, + a,Neg;, + a3Cashflow,, * Neg;, + œAQ¡, + A; Size;

+oa,Expenditure;, + a7Acquisition;, + Ag ANCWC,, + @gShortDebt;,1 + €it; 0) where

CashHoldings _ the ratio of holdings of cash to total asset,

ACashHoldings cash in year t minus cash in year t — 1 divided by total assets,

CashFlow earnings before extraordinary items and depreciation divided by total assets,

Neg an indicator variable that equals one if the firm has negative cash flow in that year and zero otherwise,

Q the sum of the market value of equity and total book assets minus the book value of equity divided by book assets,

Size the natural log of total assets,

Expenditure capital expenditures divided by total assets,

Acquisition an indicator variable that equals one if the firm makes an acquisition in that year and zero otherwise,

NCWC net non-cash working capital (working capital minus cash) divided by total assets,

ANCWC NCWC in year t minus NCWC in year t—1,

ShortDebt short-term debt divided by total assets, and

? Though Riddick and Whited (2009) also discuss the nonlinear relation, they focus on the different cash flow sensitivities of cash holding for firms with and without financial constraints However, we focus on the different cash flow sensitivities of cash for positive and negative cash flows.

D Bao ef al / Journal of Corporate Finance 18 (2012) 690-700 693 The main variables in Eq (41) closely follow the definition of the variables in Riddick and Whirted (2005, 5 1750) and other contro! variables follow the model in Almeida et al (2004, p 1788), with the addition of an indicator variable (Neg) that reveals negative cash flows and an interaction term (CashFlow*Neg) to determine how cash flow sensitivity varies with the sign of cash flow We also include industry and year dummy variables in Eq (1) Following the suggestion of Riddick and Whited, we use the fourth-order GMM (GMMA) of Erickson and Whited (2060) to estimate Eq (1) Consistent with prior literature, o, is expected to

be positive in the ordinary least squares (OLS) estimation and negative in the GMM4 estimation According to Riddick and Whited, the negative a, indicates a firm's negative cash flow sensitivity of cash

in testing the asymmetric cash flow sensitivity of cash, we include the indicator and interaction terms We expect to find a positive a3 because firms with negative cash flow use prior cash reserves to fund existing projects A significantly positive a3 in Eq (1) would be consistent with H14, that is, there is an asymmetric cash flow sensitivity of cash We have no a priori reasons to argue that the magnitude of the sensitivity is significantly larger ar smaller for positive cash flow firms than for negative cash flow firms

For other control variables, we include a size variable (Size) to mitigate economies of scale in cash savings The Q (market-to- book) ratio accounts for the future investment opportunity because such an opportunity could affect a firm's incentive to hold cash Capital expenditure (Expenditure) and acquisition activity (Acquisition) variables are included because investment and acquisition reduce a firm's cash holdings Net non-cash working capital (NCWC)} works as a substitute for cash Thus we use ANCWC to control for the effect of net working capital Short-term debt at the beginning of the year indicates possible cash outflow during the year, which either draws out cash or increases managers’ incentives to save more cash Therefore we include the variable ShortDebt in Eq (1) The inchision of more contro! variables represents an improvement of the empirical model of Riddick and Whited (2009)

3.2 Financial constraints and the asymmetric cash flow sensitivity of cash

After considering the asymmetric cash flow sensitivity of cash, we examine how cash flow sensitivity varies between financially constrained and unconstrained firms We use four measures to partition our sample, which are as follows

1 We construct an index of a firm's external financial constraints based on the results in Whited and Wu (2006) Compared with the common Kaplan and Zingales (1997) index, Whited and Wu argue that their index is more consistent, with firm characteristics related to financing constraints We first calculate the WW index for sample firms each year according to the following equation:

WW index, = —0.091Cashflow;,—0.062DIVPOS, + 0.02 1TLTD,,—0.044Size, + 0.1021SG,,—-0.0355G,,

where

DIVPOS a dummy variable that equals one if firm i pays out a cash dividend in year t and zero otherwise,

TLTD the ratio of long-term debt to the book value of total assets,

ISG the firm's three-digit industry sales (data12) growth,

SG the firm's sales growth,

Ọ the sum ofthe market value of equity at the end of the fiscal year and the value of debt divided by the book value of total

assets,

Size the natural low of total assets,

Expenditure capital expenditures divided by total assets, and

CashFlow, Size, and ¢ are defined as before In each fiscal year we rank firms according to the WW index Firms in the top quartile in the annual distribution are considered financially constrained firms

2 if a firm does not pay out a cash dividend in year ¢, the firm is assigned to the financially constrained group

3 We rank firms according to the book value of total assets in each fiscal year; firms in the bottorn quartile of the annual size distribution are considered financially constrained

4 We collect data on Standard & Poor's domestic long-term firm credit ratings each year; firms that never had their public debt rated and those with credit ratings below B— are considered financially constrained,

After we partition the sample into financially constrained and unconstrained firms, we modify the base model in Eq (1) to study the possible asymmetric cash flow sensitivity of cash, which is as follows:

ACashHoldings,, = By + 6, CashFlow, + B.Neg; + ,CashFlow,, « Neg, + B4Constraint,,

+6,CashFlow,, * Constraint, + BgConstraint, « Neg; + B,CashFlow,, « Constraint, « Neg, x

-+ÐgsQ¡ + ÐaS1Z6u + BạoEXpendifUure, + Bị+Acquisiiornti + By ANCWC, + BịsShorfDebti,ì + Eụ,

Eq (2} includes the constraint dummy variable (financially constrained firms have a value of one), various interaction variables, cash flow, and negative cash flow dummy variables We estimate Eq (2) by the GMM4 procedure of Erickson and

Whited (2000) The GMM4 estimation enables us to determine whether asymmetric cash flow sensitivity still exists after categorizing firms as financially constrained and unconstrained

We expect B, to be negative and B3 to be positive in Eq (2) since the base model should continue to exhibit a significant level

of cash flow sensitivity and asymmetric cash flow sensitivity after controlling for financial constraints Given the different incentives to hold cash between the two types of firms, we expect Bs to be positive because financially constrained firms will give

up a good investment opportunity to save cash out of positive cash flow According to our hypothesis of asymmetric sensitivity of cash flow to cash, we expect to see a negative and significant estimate for B7 because it is harder for constrained firms to sustain existing projects with prior cash holdings

3.3 Controlling for agency costs

To consider the impact of the agency problem on the asymmetry of the cash flow sensitivity of cash, we estimate the following equation by the GMM4:

ACashHoldings,, = yg + y¡ CashFlow + Y2/nst;, + y3Cashflow,, x Insti, + YsQit + Y>SizZeit (3)

+y,Expenditure;, + y„Acquisiioni + ysANCWC), + ygShortDebtj 1 + 8ị, where Inst is one if the total institutional holdings of the firm's stock rank in the top decile in the annual outstanding share distribution To isolate the impact of positive and negative cash flows, we estimate Eq (3) for subsamples of positive and negative cash flows For firms with positive cash flows, we expect ‘y; and ‘y3 to be negative because firms with positive cash flow would show a negative cash flow sensitivity of cash and firms with less agency costs (greater institutional monitoring) would draw down more cash holdings to fund good projects For firms with negative cash flows, we expect ‘y; to be positive or negative and 3 to be negative This

is because firms with negative cash flow may have a positive or negative cash flow sensitivity of cash Nonetheless, with a greater institutional monitoring, the cash flow sensitivity of cash for the negative cash flow firms would be negative

4 Data

Following Riddick and Whited (2009), we select our sample of U.S nonfinancial firms with available financial data from Compustat for the period from 1972 to 2006 We exclude firms with Standard Industrial Classification codes ranging from 4900 to

4999, from 6000 to 6999, or greater than 9000 Observations with the main variables (ACashholding, Cashflow, Size, and Q) ranking

in the top or bottom 1% are deleted All other control variables are winsorized at the 1% level Our final sample has 105,492 firm- year observations

Table 1

Descriptive statistics This table reports the descriptive statistics and mean comparisons of the variables in testing the difference in asymmetric cash sensitivity between financially constrained and unconstrained firms The variable CashHoldings is cash divided by total assets,ACashHoldings is the difference in cash between year t and year t— 1 divided by total assets, CashFlow is earnings before extraordinary items and depreciation (data18 + data14) divided by total assets,

Q is the sum of the market value of equity and total book assets minus the book value of equity divided by the book value of total assets, Size is the natural log of assets, Expenditure is capital expenditures divided by total assets, Acquisition is an indicator variable that equals one if the firm makes an acquisition that year and zero otherwise, NCWC is net non-cash working capital (working capital minus cash) divided by total assets, ANCWC is NCWC in year t minus NCWC in year t—1, and ShortDebt is short-term debt divided by total assets, *** indicates significance level of 0.01

Panel A: descriptive statistics

Panel B: mean comparison of financially constrained and unconstrained firm based on the WW index

CashFlow 0.036 0.092 — 0,056*** — 80.92

Size 2.802 5.639 — 2,838*** — 342,97

ANCWC —0.003 —0.004 0.000 0.63 ShortDebf,_ ạ 0.079 0.045 0.0347 52.13

D Bao et al / Journal of Corporate Finance 18 (2012) 690-700 695

Table 2

Pearson and Spearman correlation This table reports the descriptive statistics and mean comparisons of the variables in testing the difference in asymmetric cash sensitivity between financially constrained and unconstrained firms The variable CashHoldings is cash divided by total assets, ACashHoldings is CashHoldings in year t minus CashHoldings in year t— 1, CashFlow is earnings before extraordinary items and depreciation minus dividends divided by total assets, Expenditure is capital expenditures divided by total assets, Acquisition is an indicator variable that equals one if the firm makes an acquisition that year and zero otherwise, NCWC is net non-cash working capital (working capital minus cash) divided by total assets, ANCWC is NCWC in year t minus NCWC in year t — 1,and ShortDebt is short-term debt divided by total assets Below the diagonal shows the Pearson coefficients and above the diagonal shows the Spearman coefficients

ACashHoldings CashFlow Q Size Expenditure Acquisition ANCWC ShortDebt

ACashHoldings 0.20 0.11 0.05 — 0,08 — 0,05 — 0,28 —0.10 CashFlow 0.24 0.33 0.13 0.34 —0.01 0.13 —0.21

Q 0.16 0.19 0.12 0.08 0.13 0.04 —0.22 Size 0.05 0.16 0.03 0.12 0.28 —0.03 —0.10 Expenditure —0.09 0.23 0.06 0.04 —0.07 —0.03 —0.01 Acquisition — 0,05 0.01 0.06 0.28 — 0.10 — 0,03 0.01 ANCWC — 0,27 0.17 0.04 — 0,02 — 0.04 — 0.04 —0.15 ShortDebt —0.06 —0.25 —0.12 —0.17 —0.07 —0.04 —0.27

Panel A of Table 1 reports the descriptive statistics of the variables in Eq (1) The mean (median) change in cash holdings (ACashHoldings) is 0.005 (0.002), showing that there is only a small change in the firms’ cash holdings for the full sample On average, cash holdings account for 12% of firms’ total assets The mean CashFlow for our sample is 0.078, compared to the median

of 0.088, showing that CashFlow is left skewed The mean (median) Q for our sample is 1.524 (1.201) The mean (median) firm size (Size) for our sample is 4.930 (4.764) Up to now, all these main variables have been comparable to the descriptive statistics of

US firms in Riddick and Whited (2009), except that CashFlow in our sample is lower When we turn to other control variables, Expenditure is right skewed, with a mean (median) of 0.071 (0.052) The mean (median) short-term debt (ShortDebt) is 0.054(0.024) Similar to the situation with ACashHoldings, there is not much change in the firms’ net non-cash working capital (NCWC), with a mean (median) of — 0.003 (— 0.002) About 28% of our sample has conducted an acquisition

Table 3

Asymmetric cash sensitivity The models are estimated using OLS and the GMM4 estimator of Erickson and Whited (2000) The sample firms include nonfinancial firms during the period from 1972 to 2006 The variable ACashHoldings is the difference in cash between year t and year t — 1 divided by total assets, CashFlow is earnings before extraordinary items and depreciation (data18 + data14) divided by total assets, Neg is an indicator variable that equals one if CashFlow is negative and zero otherwise, Q is the sum of the market value of equity and total book assets minus the book value of equity divided by the book value of total assets, Size is the natural log of assets, Expenditure is capital expenditures divided by total assets, Acquisition is an indicator variable that equals one if the firm makes an acquisition that year and zero otherwise, NCWC is net non-cash working capital (working capital minus cash) divided by total assets, ANCWC is NCWC in year t minus NCWC in year t — 1, ShortDebt is short-term debt divided by total assets, and T square is an index of measurement quality for Q that varies between zero and one The superscripts ***, **, and * indicate significance levels of 0.01, 0.05, and 0.10, respectively

Dependent = ACashHoldings

Panel A

Panel B

Intercept —0.020*** — 12.58 —0.140*** —5.77 —0.018*** — 10.90 —0.066*** — 10.24

Expenditure —0.202*** — 29.25 —0.218*** — 10.23 — 0.206*** —29.71 —0.170*** —17.12 Acquisition —0.013*** — 14.38 —0.017*** —11.10 —0.014*** — 14.29 —0.017*** — 15.44 ANCWC —0.304*** — 34.29 —0.316*** — 28.90 — 0.306*** — 34.94 —0.327*** — 38.39

Based on the WW index, Panel B of Table 1 compares the variables between financially constrained and unconstrained firm-years With the exception of the change in net working capital, all other variables are significantly different between the two groups Consistent with the definition of the WW index, financially constrained firms are smaller firms with lower ACashHoldings and Cashflow Constrained firms also exhibit a higher Q, indicating more growth and investment opportunities As expected, firms with liquidity constraints have lower capital expenditures and higher short-term debt Compared to unconstrained ones, a smaller proportion of the constrained firms conduct acquisition activities

Table 2 reports the Pearson and Spearman correlation coefficients between all the variables While many correlation coefficients are small, Q and CashFlow have a correlation of 0.33 The high correlation shows that the measurement error in Q will greatly bias the estimated coefficient of CashFlow in the OLS regression The partial correlation between ACashHoldings and CashFlow is significantly positive The variables CashFlow and Expenditure are positively correlated, indicating that firms with higher cash flow are more likely

to invest in new projects

5 Results and discussion

The first four columns in Panel A of Table 3 report the results of a reduced base model in Eq (1) with fewer control variables, for comparison with the results of Almeida et al (2004) and Riddick and Whited (2009) While the OLS regression exhibits a positive coefficient for CashFlow, the GMM4 results show that the coefficient of CashFlow is negative and significant at the 5%

level Consistent with Riddick and Whited, the results indicate that the measurement error in Q biases the estimate of CashFlow

upward; we therefore use GMM4 estimation in remaining regressions In Panel B, we add other control variables adopted from Almeida et al (2004) to exclude cash flow from other sources; the findings are qualitatively the same as in Panel A The estimates

of CashFlow in both the OLS and GMM4 regressions are significantly higher than those in the first two columns, consistent with Almeida et al (2004) Therefore including all the control variables will provide a stronger test of our hypothesis

The last four columns in both Panels A and B of Table 3 report findings of the asymmetric cash flow sensitivity of cash We incorporate a dummy variable (Neg equals one for firms with negative cash flow), as well as an interaction term with the cash flow variable For the GMM4 results, the dummy variable (Neg) is negative and the interaction (CashFlow« Neg) is positive Both

Table 4

Financial constraints Panel A reports the firm-year cross-classifications for the four criteria used to categorize firms as either financially constrained or unconstrained Panel B displays the summary statistics for cash holdings across financially constrained and unconstrained firms Letter A denotes constrained firms and letter B denotes unconstrained firms in each row/column The sample firms include nonfinancial firms during the period from 1972 to 2006 for the first three measures and from 1985

to 2006 for bond ratings

Financial constraint criteria WW index Payout ratio Firm size Bond rating

A B A B A B A B

Panel A: cross-classification of constraint types

1 WW index

Constrained firms (A) 26,368

Unconstrained firms (B) 79,124

2 Payout ratio

Constrained firms (A) 23,484 28,524 52,008

3 Firm size

Constrained firms (A) 23,757 7,882 23,300 8,339 31,639

4, Bond rating

Constrained firms (A) 17,989 38,725 37,316 19,398 21,617 35,097 56,714

Panel B: summary statistics of cash holdings

1 WW index

2 Payout ratio

3 Firm size

4, Bond rating

D Bao et al / Journal of Corporate Finance 18 (2012) 690-700 697 are significant at the 1% level, suggesting that negative cash flow firms experience a downward drift in cash holdings The sum of coefficients (Cashflow + CashFlowNeg) is significantly positive, indicating that firms with negative cash flow experience a decrease in their cash holdings In other words, the cash flow sensitivity of cash is asymmetric: The negative cash flow sensitivity

of cash of Riddick and Whited (2009) applies only to firms with positive cash flows; in a negative cash flow environment firms use prior cash holdings to fund existing projects Other control variables show expected signs and are significant at the 1% level The results support H1

Table 4 presents the summary statistics of the four financial constraints The diagonal in Panel A of Table 4 shows the cross- classification of the constraint types As expected, the number of constrained and unconstrained firms varies with the different

constraints If we use WW index, payout ratio, firm size, and bond rating as the criteria, there are 126,368, 52,008, 31,639, and

56,714 firm-years from financially constrained firms, respectively Panel B of Table 4 shows the cash holdings of financially constrained versus unconstrained firms The mean of the cash holdings (cash divided by total assets) for financially constrained firms is around 0.152, approximately the same for all four measures For the unconstrained firms, the mean of the cash holdings ranges from 0.079 to 0.112

Table 5 reports the empirical model of Eq (2) We incorporate additional dummy and interaction variables using different measures of financial constraints (financially constrained firms have a value of one) in addition to the negative cash flow dummy variable (negative cash flow has a value of one) To compare the cash flow sensitivity of cash between a financially constrained and an unconstrained environment, we pay attention to the coefficients associated with the interaction variables of CashFlow« Constraint and CashFlow« Constraint « Neg (correspondingly showing the cash holding response of constrained firms with positive and negative cash flow) The asymmetric cash flow sensitivity of cash documented in Table 3 still holds for both constrained and unconstrained firms The variable CashFlow is negative and CashFlow « Neg is positive, both significant at the 1%

level, consistent with the results in Table 3

For all constraint measures, the coefficient of CashFlow+Constraint is significantly positive, indicating that, compared to unconstrained firms, constrained firms are less likely to invest in new projects because they find it hard to obtain external financing and thus need to hold more cash Facing profit shocks, they save money and must give up good investment opportunities Here CashFlow « Constraint « Neg are negative and significant at the 1% level in all measures of financial constraints, with the exception of the bond rating, suggesting that financially constrained firms are likely to spend less cash reserves to fund existing projects when they have negative cash flows Our constraint measures can proxy for sustainability so it is harder for constrained firms to keep funding projects using prior cash holdings As a result, they are more likely to terminate projects with negative cash flows Other control variables have the expected signs and are significant in most of the cases The findings support H2

We conjecture that the asymmetric sensitivity of cash flow can be attributed to the fact that managers seek private benefits by overinvesting in unprofitable projects and are unwilling to terminate them We use the percentage of institutional holdings in year t — 1

as a proxy for agency costs (Inst equals one when the percentage of institutional holdings ranks in the top decile in its annual

Table 5

Asymmetric cash sensitivity considering financial constraints The models are estimated using the GMM4 estimator of Erickson and Whited (2000) The sample firms include nonfinancial firms during 1972 to 2006 for the first three constraint measures, and from 1985 to 2006 for bond ratings The variable ACashHoldings

is the difference in cash between the years t and year t—1 divided by total assets, CashFlow is earnings before extraordinary items and depreciation (data18 + data14) divided by total assets, Neg is an indicator variable that equals one if CashFlow is negative and zero otherwise, Constraint is an indicator variable that equals one if the firm in year t is financially constrained and zero otherwise, Q is the sum of the market value of equity and total book assets minus the book value

of equity divided by the book value of total assets, Size is the natural log of assets, Expenditure is capital expenditures divided by total assets, Acquisition is an indicator variable that equals one if the firm makes an acquisition that year and zero otherwise, NCWC is net non-cash working capital (working capital minus cash) divided by total assets, ANCWC is NCWC in year t minus NCWC in year t— 1, ShortDebt is short-term debt divided by total assets, and T square is an index of

measurement quality for Q that varies between zero and one The superscripts ***, **, and * indicate significance levels of 0.01, 0.05, and 0.10, respectively Dependent = ACashHoldings

1 WW index 2 Payout constraint 3 Firm size 4, Bond rating

Intercept —0,052*** — 9,57 —0,049*** — 8.75 — 0.046*** — 8.16 —0,057*** — 7,50 CashFlow —0,173*** — 3.65 —0,181*** — 3.20 —0,171*** — 3.29 — 0,384*** —7,93 Neg —0,030*** — 9,27 —0,025*** — 7.34 — 0,028*** — 8.70 —0.030”” —9.32 CashFlow «Neg 0.434*** 7.81 0.491*** 6.77 0.447*** 6.90 0.635*** 9.12 Constraint — 0,032*** — 10.22 —0,018*** — 8.78 — 0.030*** — 10.70 — 0,022*** —7.79

CashFlow «Constraint 0.312*** 9.90 0.184*** 10.54 0.261*** 10.26 0.203*** 7.22

Constraint+ Neg 0.0157” 5,09 0.002 1.02 0.008*** 3.06 — 0,006* —1.81

CashFlow «Constraint « Neg —0,224*** —5.83 —0,151*** —4,52 —0.192*** —4,52 — 0.068 —0.98

Q 0.0517” 10.61 0.048*** 8.12 0.049*** 8.48 0.063*** 11.56 Size 0.002*** 4,32 0.002*** 5.90 0.001*** 3.10 0.002*** 3.76 Expenditure — 0,168*** — 16.83 —0,170*** — 18.49 — 0.166*** — 17.33 —0,157*** — 12.85 Acquisition —0.018** — 15.55 —0,017*** — 15.18 — 0.017*** — 15.40 — 0,020*** — 22.61 ANCWC — 0,329*** — 39.66 —0,327*** — 41.05 — 0,327*** — 40.41 —0,321*** —44.41

Table 6

Asymmetric cash sensitivity considering institutional holdings The models are estimated using the GMM4 estimator of Erickson and Whited (2000) The sample firms include nonfinancial firms during the period from 1985 to 2006 The variable ACashHoldings is the difference in cash between year t and year t— 1 divided by total assets, CashFlow is earnings before extraordinary items and depreciation (data18 + data14) divided by total assets, Inst is an indicator variable that equals one if the firm's percentage of institutional holdings in year t— 1 ranks in the top decile of its annual distribution and zero otherwise, Q is the sum of the market value of equity and total book assets minus the book value of equity divided by the book value of total assets, Size is the natural log of assets, Expenditure is capital expenditures divided by total assets, Acquisition is an indicator variable that equals one if the firm makes an acquisition that year and zero otherwise, NCWC is noncash working capital (working capital minus cash) divided by total assets, ANCWC is NCWC in year t minus NCWC in year t— 1, ShortDebt is short-term debt divided by total assets, and T square is an index of measurement quality for Q that varies between zero and one The superscripts ***, **, and * indicate significance levels of 0.01, 0.05, and 0.10, respectively

Dependent = ACashHoldings

Inst 0.016*** 5.34 —0.010”7 —2.51

Q 0.002** 2.62 0.012*** 20.50 Size — 0,148*** — 9,08 —0,185*** — 8.95

ANCWC 0.069*** 12.41 0.047 21.49 ShortDebt,_ ¡ 0.124*** 11.79 0.251*** 13.69

N 57,653 14,460

distribution and zero otherwise) We include the dummy Inst and its interaction term with the CashFlow variable Table 6 reports the regression results for firms with positive and negative cash flows separately, as specified in Eq (3) The first two columns show that, on average, firms with positive cash flow exhibit a negative cash flow sensitivity of cash The interaction term CashFlow « Inst is significantly negative, suggesting that institutional investors motivate firms to take good advantage of profit shocks In comparison, the coefficient of CashFlow is significantly positive for firms with negative cash flow (last two columns); that is, firms with greater agency costs keep funding bad projects using cash reserves Here CashFlow«Inst is significantly negative, indicating that outside monitoring makes managers to terminate bad projects and prevents them from seeking private benefits The findings support H3

Our overall finding of asymmetry in the cash flow sensitivity of cash echoes the results of Riddick and Whited (2009) Their graph (Riddick and Whited, 2009, p 1793) shows that the substitution effect (negative cash flow sensitivity of cash) is stronger for medium and large firms because these firms do not have financing constraints In addition, large firms sell capital when productivity is low Our results for financially constrained versus unconstrained firms in Table 5 are consistent with theirs However, the positive coefficients in Table 5 for negative cash flows do not provide direct evidence that firms sell capital or disinvest Table IV of Riddick and Whited (2009, p 1758) presents results on the impact of uncertainty since they argue that large firms and firms with bond ratings face less uncertainty in income We add something new by focusing on the difference in the asymmetry of the cash flow sensitivity of cash between positive and negative cash flow environments, with explanations of binding project contracts, bad news withholding, and agency costs

Another interesting question regarding the asymmetry of the cash flow sensitivity of cash is whether it is driven by macroeconomic patterns or by firm-specific issues To shed some light on the question, we conduct a simple count of negative cash flow firms by year across different macroeconomic conditions We use the economic recession data from National Bureau of Economic Research business cycle contractions and expansions.” The year t is a recession year if at least one month falls within a contraction (peak to trough) We have two findings in Panel A of Table 7: First, the percentage of firms with negative cash flows varies over time It ranges from a low of 2.8% of the sample in 1973 to a high of 23.4% in 2001 Specifically, the percentage of negative cash flow firms in the sample is at a low level during the early 1970s to the mid-1980s Roughly speaking, the negative cash flow sensitivity of cash is more dominant in these years, given that fewer firms having negative cash flows Second, the result from a chi-square test of the stable percentage of negative cash flow firms in the sample over time is statistically significant, suggesting that the percentage is not stable over time Panel B of Table 7 presents the correlation coefficients between the mean cash flows of all firms, the percentage of negative cash flow firms, and recession years The mean cash flow of all firms captures the net impact from positive and negative cash flows in a year The correlation between the mean cash flow and recession, while positive, is not statistically significant Similarly, the percentage of firms with negative cash flow is negatively correlated with recession, but not statistically significant Hence the findings of the asymmetry of the cash flow sensitivity of cash are not likely due to macroeconomic patterns When we divide our sample into recession and expansion years, the asymmetry continues to hold.* Thus the asymmetry is likely due to firm-specific issues rather than macroeconomic factors.°

3 See http://www.nber.org/cycles.html

4 The results are available upon request

° We find that our sample firms with high sales growth and high market-to-book value are more likely to have positive cash flow We conjecture that these firm-specific factors contribute to the asymmetry.

D Bao et al / Journal of Corporate Finance 18 (2012) 690-700 699 Table 7

Annual distribution of cashflow This table reports the annual distribution of the number of firms and firms with negative cashflow Here N is the number of observations in each fiscal year, CashFlow is earnings before extraordinary items and depreciation (data18 + data14) divided by total assets, N_negative is the number of observations with negative cashflow in each fiscal year, Percent_negative is the percentage of observations with negative cashflow in each fiscal year, and Mean_CashFlow is the mean of CashFlow for each fiscal year We obtain recession data from National Bureau of Economic Research U.S business cycle contractions and expansions (http://www.nber.org/cycles.html) The variable Recession equals one if at least one month in year t is within the contraction (peak

to trough) and zero otherwise It is noted that the N and N_negative values are different from those in Table 6 due to missing information in some control variables

in Table 6, Above diagonal are the Pearson correlation coefficients and below the diagonal are Spearman correlation coefficients Here *** denotes 1% significance

Panel A: annual distribution of firms with negative cashflow

1973 2,478 69 2.8% 1

1976 2,790 109 3.9% 0

1978 2,642 79 3.0% 0

1982 2,959 343 11.6% 0

1983 2,934 351 12.0% 0

1984 3,095 375 12.1% 0

1985 3,057 424 13.9% 0

1987 3,040 449 14.8% 0

1988 3,080 443 14.4% 0

1989 2,934 438 14.9% 0

1990 2,896 441 15.2% 1

1991 2,929 464 15.8% 1

1992 3,082 436 14.1% 0

1993 3,286 468 14.2% 0

1994 3,591 459 12.8% 0

1996 3,874 575 14.8% 0

1997 4,073 631 15.5% 0

1998 3,991 681 17.1% 0

1999 3,691 592 16.0% 0

2000 3,359 587 17.5% 0

2002 3,239 685 21.1% 0

2003 3,230 587 18.2% 0

2004 3,165 465 14.7% 0

2005 3,050 453 14.9% 0

2006 2,509 337 13.4% 0

Chi-square for year and percent_negative 2,475.69

Panel B: correlation between cash flow and macroeconomic condition

6 Conclusion

Prior literature has widely studied why firms hold cash and has agreed on several explanations Almeida et al (2004) and Riddick and Whited (2009) offer contrasting conclusions regarding the corporate cash flow sensitivity of cash We use an augmented empirical model to affirm the conclusion of Riddick and Whited that the cash flow sensitivity of cash is generally negative In addition, we contend that the cash flow sensitivity of cash is asymmetric to cash flow due to several reasons, including binding project contracts, bad news withholding, and agency costs That is, a firm's cash flow sensitivity of cash, while negative in a positive cash flow environment, is not necessarily negative in a negative cash flow environment

Using a sample of manufacturing firms from 1972 to 2006, we document that the cash flow sensitivity of cash is negative when

a firm faces a positive cash flow environment, a result supporting Riddick and Whited (2009), but the cash flow sensitivity of cash

is positive when a firm faces negative cash flows We further categorize firms as financially constrained and unconstrained and find that the cash flow sensitivity of cash asymmetry continues to hold in both financially constrained and unconstrained firms Financially constrained firms, when compared to unconstrained ones, are less likely to invest in new projects and fund existing unprofitable projects because they find it hard to obtain external financing In addition, when we use institutional holdings as a