Can bank dependency explain the puzzle of financial constraints and investment cash flow sensitivity? : evidence from U.S. during the 2008 financial crisis

(1)

University of Amsterdam

Faculty of Economics and Business

Quantitative Finance

Can Bank Dependency Explain the Puzzle of Financial Constraints and

Investment Cash Flow Sensitivity? Evidence from U.S. during the 2008

financial crisis

Master Thesis

Name: Aijia Ma

Student Number: 10683305

Date of Submission: 01/07/2018

Supervisor: Tomislav Ladika

Abstract:

This paper reconciles the puzzling relationship between financial constraints and investment cash flow sensitivity using a large sample of U.S. public firms during the period of 2000-2010. The result shows a negative relationship between financial constraints and sensitivity of cash flow, which is more pronouced during the time of the crisis. Moreover, this paper also proposes explanations for the negative relationship from the aspect of bank dependency, suggesting that bank-dependent firms are relatively more constrained during the crisis and displays lower level of investment cash flow sensitivity. Corporate excess cash holdings can explain the negative relationship to some extent as well since firms which exhibit lower levels of investment cash flow sensitivity are mostly financially constrained and tend to hoard more cash. Additionally, firms that depend on banks that were in the bad shape during the banking crisis are found to have lower investment cash flow sensitivities than those firms that rely on banks which were in good shapes.

(2)

Statement of Originality

This document is written by Student Aijia Ma who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

1. Introduction

The standard dynamic model of Q theory of investment suggests that in the absence of market frictions, the movements in the investment rate should be entirely explained by Tobin’s q and firms optimally choose investments that make the marginal value of capital equals to the marginal value of costs (Hayashi, 1982). However, firms operate with frictions and imperfect financial flexibility, which prevent them from funding all attractive investments. These frictions are described by Lamont, Polk and Saá-Requejo (2001) as financial constraints. Since external financing is more expensive than internal financing, firms rely more on internal finance such as cash flow and cash holdings. Moreover, a firm’s performance would be sensitive to unexpected shocks bring by capital providers if it cannot easily access to the external capital market for funds. Especially for small and medium-sized firms, they face higher probabilities of being denied from loans, and only high-quality borrowers are chosen over them.

Even though there are lots of literature demonstrating a substantial impact of financial constraints on investment cash flow sensitivity, it is interesting to look at the relationship between bank lending and firm’s investments during the crisis in 2008. The banking crisis of the 1930s in the U.S. can be seen as the central event of the Great Depression. Due to crashes of the stock market, bank failures and closures caused a chain reaction among other banks in the U.S. However, it has been a long time for the world to witness another banking crisis until 2008. Concerns about the “liquidity bubble” caused subprime mortgage to default, which is the origin of the substantial shock on the performance of banking sector.

The banking crisis may exert an impact on firm’s investment cash flow sensitivity, which is more pronounced than normal times. Firstly, the lack of liquidity forces banks to either reduce their lending to firms or issue equity to restore their capital positions. Firms that rely most on the bank loans as their main credit providers may suffer from the reduction of the bank loans and therefore have to cut down their investment activities. Secondly, banks require collaterals from firms and collaterals are worth less during the time of recession. Accordingly, external finance would be more expensive for firms which are in bad condition and therefore would limit their investment activities. Furthermore, financial institutions that are saved by governments face stricter regulations and supervision, which causes limitations of providing loans to businesses. For instance, Dodd–Frank Wall Street Reform and

(5)

Consumer Protection Act was signed into law that requires stronger capital and liquidity positions for firms to improve financial stability. Hence, it is reasonable to assume a close relationship between bank lending and firm’s investment cash flow sensitivity. Nevertheless, despite the overall negative impact of the financial crisis, well-established firms with better access to external capital were less affected by the crisis and operated profitably. Those large firms reported that they did not have trouble raising the fund of operations and they operated in an unconstrained situation (Cull, 2012). Also, Shakina & Barajas (2014) emphasize a phenomenon called “Too big to fail”, indicating that human and intellectual capital enabled large firms to survive during the financial crisis. Hence, investments of these large well-established firms were not affected significantly by the crisis and the impact of difficult economic condition on investment activities are still needed to be investigated.

The changes in cash flow can be considered as a proxy for internal financing as well as an important determinant of optimal capital spending. There exist a significant relationship between cash flow and investment, especially for firms which facing a high level of financial constraints. The monotonicity assumption is that the sensitivity of investment to cash flow is higher for firms that encounter a large wedge between the internal and external funds. Nonetheless, there is an ongoing debate about the positive or negative relationship between financial constraints and investment cash flow sensitivity, represented by contradictory view of Fazzari et al. (1988) and Kaplan and Zingales (1997). Although some of the previous literature partly explained the puzzle from fixing the endogeneity issues and the measurement error, the bank-dependency of firms might also have an impact on the negative relationship. If there is a significant and negative influence of bank-dependency on firms’ investments regardless of the increasing level of financial constraints during the time of crisis, it would downwards the investment cash flow sensitivities considerably. Hence, this paper will investigate the puzzling relationship between investment cash flow sensitivity and financial constraints using a more recent sample with larger sample size and try to find an explanation by connecting bank dependency with investment cash flow sensitivity. The research question is: what are the impacts of financial constraints on investment-cash flow sensitivity and can bank dependency explain the puzzling relationship between them? Thus, the contribution to the existing literature of this paper is to propose that bank dependency has a significant impact on firm’s investment cash flow sensitivity and can partially explain the contradictory relationship.

(6)

First, this paper examines the investment cash flow sensitivity of financially constrained and unconstrained U.S. public firms between 2000 to 2010 and finds that there is a negative relationship between financial constraints and investment cash flow sensitivity, which is consistent with the finding of Kaplan and Zingales (1997). Moreover, due to the fact that constrained access to external funds (i.e. bank loans) are more pronounced during time of a banking crisis, I implement a test on the behavior of investment cash flow sensitivities around an event date which is the 2008 financial crisis and divide the sample period into three subperiods: normal times period, pre-crisis period and the crisis period. Firms are also divided into two categories base on their borrowing relationship with banks, which are the bank-dependent group and the non-bank-dependent group. By looking into investment cash flow sensitivities of bank-dependent firms during the crisis and non-crisis period, the result shows that bank-dependent firms, which mostly consists of financially constrained firms, have lower cash flow sensitivity to investment than non-band-dependent firms. This relationship between bank dependency and investment cash flow sensitivity further supports the idea that financially constrained firms generally have lower sensitivity to cash flow. In addition, according to their relationship with banks that were in good and bad shapes during the crisis, I split the sample into good-shape-dependent and bad-shape-dependent respectively and investigate the effect of the banking crisis on firm’s cash flow sensitivities to investment. The result obtained shows that the crisis has a negative impact on investment cash flow sensitivity of firms and this impact is more pronunced on firms that depend on banks with bad shapes.

The remainder of the thesis is organized as follows. There are five sections in total, excluding the references, which is presented at the end of the thesis. Section 2 provides a literature review on: (1) internal and external funds of finance; (2) different measures of financial constraints; (3) financial constraints and investment cash flow sensitivity; (4) bank lending and firms’ investment during the 2008 financial crisis. Section 3 illustrates the theoretical framework, consisting of the hypotheses and methodology. Data and descriptive statistics are elaborated in Section 4. Then, section 5 displays the results of the research and a robustness check test is performed in section 6. At last, Section 7 concludes with the discussion and potential limitations of the research.

(7)

2. Literature review

2.1 Internal and external funds of finance

Erickson and Whited (2000) believe that firms choose to invest when the investment is expected to be profitable and fund all value-enhancing investment opportunities in a frictionless world as well as generating their capital by paying the cost of capital. However, Fazzari et al. (1988) conduct research based on market imperfections and criticize this assumption. When financial structure matters for investment choices, the substitutability for internal and external funds is not perfect anymore. Kadapakkem et al. (1998) propose that the effects of transaction costs, asymmetric information and agency problems are crucial factors that influence the demand for internal funds.

Firstly, asymmetric information is a determinant of internal financing. If insiders have better knowledge of the firm than outsiders do, the wedge between internal and external funds is larger and credit rationing will therefore arise (Degryse and de Jong, 2006). In this case, investors will encounter difficulty evaluating the probability of default. To compensate for their disadvantages regarding asymmetric information, investors would require risk premiums, thus making external financing more expensive and causing a larger wedge between internal and external funds. This wedge is especially harmful to firms in bad financial situations since they do not have sufficient internal capital to support their investments and are perceived as risky borrowers.

Secondly, agency problems arise when management is craving more cash for personal benefits regardless of the interests of other stakeholders. For instance, managers may want to expand the firm size to satisfy their own needs, which can be represented as empire building (Jensen, 1999). Dittmar, Mahrt-Smith, and Servaes (2003) conduct research on over 11,000 firms from 45 different countries to study the determinants of cash holdings. Their evidence suggests that firms in countries with serious agency problem hoard twice as much cash as firms in countries with good protection for shareholder rights. Agency problems also exist between managers and creditors. Managers may repurchase stocks by taking on more debt and thus increase the leverage position, which will harm the benefits of creditors.

Finally, transaction costs incurred when converting noncash assets into cash and using the proceeds to conduct payment for business operations, consisting of administrative costs and

(8)

shelf registration fee. Hence, it is less costly when firms possess adequate cash which has higher liquidity to cope with urgent payments need. Furthermore, when issuing equity to fund investments, the tax advantages also reduce the costs of financing. These can also be seen as transaction motives and taxation motives for holding cash. Denis and Sibilkov (2010) conclude that cash holdings are more valuable for financially constrained firms than for unconstrained firms. Their results indicate that greater cash holdings allow higher levels of investment for constrained firms and the relationship between investment and firm value is stronger for constrained firms.

2.2 Measures of financial constraints

Due to the fact that financial constraints cannot be directly observed, empirical research has to rely on indirect proxies or one of three popular indices based on linear combinations of observable firm characteristics (Farre-Mensa and Ljungqvist, 2015). An overview of different financial constraints is listed in Table 1 below.

Table 1. Proxies for measuring financial constraints

Author(s) Year Variables for financial constraints

Fazzari, Hubbard & Petersen 1988 Payout ratio

Devereux & Schiantarelli 1990 Firm size; Firm age; Sector Hoshi, Kashyap &

Scharfstein

1991 Group membership

Whited 1992 Bond rating; Debt to asset ratio; interest coverage ratio Chirinko & Schaller 1995 Firm age; Concentration of ownership

Kaplans & Zingales 1997 Cash; leverage figures Lamont, Polk &

Saá-Requejo

2001 KZ index

Whited and Wu 2006 WW index

Denis and Sibilkov 2010 Payout ratio; Firm size; Bond rating; Commercial paper rating

Hadlock and Pierce 2010 Size-and-age index

When external financing is costly, firms prefer retaining their internal funds which are at lower costs. Therefore, Fazzari et al. (1988) believe that financially constrained firms have lower payout ratio since firms with a high dividend payout ratio implicating that they do not face the difficulty of raising funds when external financing is more costly.

Following the paper of Fazzari et al. (1988), numerous papers implement different methods to classify financial constraints and used these criteria as variables in the regression models.

(9)

Traditional approaches are sorting financially constrained firms depending on related firm characteristics. Devereux & Schiantarelli (1990) argue that firms with larger sizes have a lower probability to be financially constrained. The reasoning behind this is that larger firms have easier access to external funds because they possess more fixed assets which can be used as collateral when applying for loans. Moreover, smaller-sized firms could be harmed by market information imperfections (i.e. information asymmetry). Chirinko and Schaller (1995) hold the view that firm age is also relevant in determining access to external capital. Their results show that more mature firms tend to have less problem with asymmetric information and lower transaction costs, which give less likelihood of arising financial constraints. A mature firm may have a closer relationship with the lender and enter into multiple cooperation with the lender, thus the lender would have more information about the firm. Credit rating of the firm also matters for the ability to raise external funds, thus bond rating and commercial paper rating can be considered as a criterion of assessing financial constraints (Denis and Sibilkov, 2010). Independent agencies such as Moody’s and S&P ratings give a reliable measure of the quality of firm’s debt and the rating outcomes indicate the abilities of firms to repay their debt. Accordingly, firms with higher credit ratings are less likely to be financially constrained whereas firms with poor ratings may have difficulties repaying the principal and interest and therefore have high probabilities to become financially constrained.

In recent years, researchers constructed several indexes for measuring financial constraints. Lamont, Polk & Saá-Requejo (2001) develop a KZ index based on the regression coefficients of Kaplan and Zingales (1997). However, the main drawback of KZ index is that Kaplan and Zingales (1997) only consider public manufacturing firms with positive sales growth and the dependent variable includes both quantitative and qualitative information, which will cause certain biases. Finally, Hadlock and Pierce (2010) evaluate all previous approaches and build up a size-age-index that has been recognized as the most reliable measure of financial constraints by now. They collect detailed qualitative information and exploit the approach of Kaplan and Zingales (1997). Their findings do not support the validity of the KZ index and suggest that their size-age-index provides a better measure of financial constraints, which considers mainly the firm’s size and age as crucial predictors of financial constraints.

(10)

2.3 Financial constraints and investment cash flow sensitivity

Financial constraints also impose significant influence on firm’s investment cash flow sensitivity. According to the economic theory, internal financing is more favorable for firms than external financing because of the capital market imperfections, and the magnitude of imperfections is measured by the correlation between firm’s investment and internal cash flows. Especially for financially constrained firms, higher external financing cost lead them to rely more on internal financing to funds their investments. Thus, investment spending would vary a lot with internal funds availability due to this cost differential, implying that firms will show a higher investment cash flow sensitivity when rasing external funds is considered to be more costly.

Fazzari et al. (1988) first implement a priori classification scheme of the dividend payout ratio. They use a sample consisting of three groups of US manufacturing firms categorized by the dividend payout ratio to investigate the effect of financial constraints on investment cash flow sensitivity. Their results easily show that investment cash flow sensitivity for financially constrained firms is higher compared to financially unconstrained firms. Following the steps of Fazzari et al. (1988), a large number of studies apply the similar methods and wind up with the same results. Hoshi, Kashyap and Scharfstein (1991) examine two sets of Japanese firms who are members according to the Keiretsu no Kenkyu’s classification scheme and those who are not. They find that the latter group’s (i.e. financially constrained group) investment is more sensitive to liquidity than the first group. Their results also show that the unconstrained group of firms generally have closer ties with banks whereas financially constrained firms have weaker links with banks. This can be interpreted as banks are likely to be well informed by firms with strong bank ties which serve as their primary source of external funds. Besides, Whited (1992) conclude that Euler investment equation fits well for unconstrained firms measured by bond ratings and rejected for others, which also support Fazzari et al. (1988) results by confirming exogenous financial constraints are especially binding for constrained firms.

On the contrary, Kaplan and Zingales (1997) challenge the results of Fazzari et al. (1988) using a combination of qualitative and quantitative criteria to classify financially constrained firms. They question the validity of the measure for financial constraints suggested by Fazzari et al. (1988) and choose a sample of 49 low dividend payout firms that FHP classify as

(11)

financially constrained firms. Surprisingly, they find that less financially constrained firms exhibit a higher sensitivity of cash flow to investment than those defined as more financially constrained and this pattern is robust to different classification schemes. This interesting finding shows that it is not necessarily true that cash flow sensitivity increases as the degree of financial constraints increases. They also point out that FHP ignored testing the fundamental theory of a monotonic relationship and failed to rank the extent to which firms are financially constrained. Several reasons are provided by Kaplan and Zingales (1997) for the negative relationship between financial constraints and investment cash flow sensitivity, even though the intuition is the opposite. First, they argue that the difference in sensitivities might be caused by influential outliers. Moreover, the result obtained from FHP might be applied to a few financially constrained firms that are forced payback their debt by cash flows, which is not representative of other samples. Kaplan and Zingales (1997) argument has been further supported by Cleary (1999) using a larger sample size based on the same objective sorting mechanism. He discovers that firms that are more creditworthy have higher level of investment cash flow sensitivity compared to those that are less creditworthy.

Some earlier studies also propose several explanations for the puzzling relationship between financial constraints and investment cash flow sensitivity. Moyen (2004) obtains the same result as Fazzari et al. (1988) by using the dividend payout ratio to identify financially constrained firms while he obtains the same results as Kaplan and Zingales (1997) by using the constrained model. Therefore, he concludes that heterogeneous measures for financial constraints can explain the conflicting empirical evidence. Similarly, Almeida et al. (2004) use 5 different approaches to divide the sample into constrained and unconstrained groups. They find a significant positive relationship between financial constraints and cash flow sensitivity under 4 classification schemes whereas they discover a very opposite result when using the KZ index. Moreover, Allayannis and Mozumdar (2004) argue that the bias of Kaplan and Zingales (1997) may be caused by the negative cash flow and the problem of small sample size. They also discover a decline of investment cash flow sensitivity for financially constrained category over the period of 1977 to 1996, which could partially explain the puzzle as well. In addition, Erickson and Whited (2000) discover that the disappointing finding may be due to the measurement error in the q theory of investment by using a measurement error-consistent generalized method of moment estimators.

(12)

problem of small sample size, which consisted of only 49 firms that are most constrained, and Cleary (1999) conquered this small sample size problem afterwards. However, the sample Cleary (1999) used are quite outdated, which is between the period from 1987 to 1994. Furthermore, Allayannis and Mozumdar (2004) find a decreasing trend of investment cash flow sensitivity in recent years. Thus, it is necessary to re-visit the analysis using a more recent sample with larger sample size and the findings may differ from the studies conducted in 90s. Accordingly, this paper reconciles the findings of Kaplan and Zingales (1997) and indicates that the level of internal cash flow has a major impact on investment cash flow sensitivity.

2.4 Bank lending and investment during the 2008 financial crisis

Financial crisis can be seen as the implications of market imperfections, which market frictions play an important role. These market frictions increase the wedge between internal and external financing and impose an adverse impact on firm’s investment and profit ratios. As the decrease of firm’s profitability, creditors tend to require higher rates to compensate for the risks accompanied by the falling net worth of the firm. During this unusual time, the falling of supply of external funds leads to higher value of financing flexibility with internal funds, thus firms would suffer when raising capital for operation (Gamba & Triantis, 2008). What is more, firms with poor externally financing ability (i.e. financially constrained firms) are in particular negatively affected by capital market imperfections. Therefore, it is reasonable to believe that the difference between financially healthy and unhealthy firms are more pronounced during the crisis.

The peak period (4th_{quarter of 2008) of the 2008 financial crisis can be seen as a banking}

panic, caused by the meltdown of the mortgage market. After Lehman Brothers filed for bankruptcy in September 2008, several financial institutions such as AIG and Merrill Lynch faced huge liquidity issues and in urgent need to be bailed out by the United States government. Consequently, prices of financial instruments fell dramatically and the cost for firms borrowing capital rose significantly due to the collapse of the mortgage market. Moreover, the run of short-term bank creditors raised difficulties for banks to roll over their short-term debt and caused banks to cut their lending substantially. This has been proved by Ivashina and Scharfstein (2010), they demonstrate that during the peak period of the recession, new loans to borrowers fell by 47% relative to the prior quarter.

(13)

According to the bank lending supply shock theory, the status of bank operation and the behaviour of bank lending can exert certain effects on firm’s external financing and therefore influence the sensitivity of investment cash flow. Capital expenditures and debt issuance of firms should decrease more significantly for bank-dependent firms as a consequence of a bank lending supply shock brought by a crisis (Shleifer and Vishny, 2010). In order to investigate the effect of the banking crisis, Chava and Purnanandam (2011) conduct some empirical tests and indicate that banking collapse has a significant negative impact on the performance of bank-dependent borrowers. Furthermore, Casey and O'Toole (2014) test the impact of bank lending constraints on firm demand for alternative external capital in the time of crisis, they find that constrained firms are more likely to use informal lending from other companies. Besides, it is proven that bank lending also affects firm’s financial constraints as an increase in borrowing from state-owned banks significantly cuts down firm’s financial constraints (Behr, Norden and Noth, 2013). Hence, the above findings prove that bank-dependent borrowers experience an increasing level of financial constraints during the banking panic, which will then lead to a higher level of investment cash flow sensitivity.

On the other hand, the adverse impact of the crisis on firm using internal financing is not as severe as that of firm which rely on bank loans. Santioni, Schiantarelli and Strahan (2017) illustrate that firms within the Italian business groups use their access to internal capital markets as the substitute for external funds. Firms that had greater abilities to borrow externally shared their internal cash across the group and avoided the costs of limited credit stemming from banking crisis. Hence, by moving capital from cash-rich to cash-poor firms, these business groups help a lot of firms to survive the crisis and mitigated the costs of raising funds.

Existing literatures believe that the supply shock leads to decreases in firms’ future investment as well as their demands for external financing to investments opportunities, and this effect is greater for firms with more leverage. Because of higher borrowing costs and fewer lending agreements, financially constrained firms with no other way to obtain funds may suffer a considerable decline in capital expenditure. However, the degree of reducing lending is different among banks that were in different conditions during the crisis. Therefore, this paper will further investigate whether the effects of decreasing lending of banks on firm’s investment differs between the banks that were in good shapes (i.e. did not suffer much) during the crisis and those that were in bad shapes (i.e. suffered huge losses).

(14)

3. Hypothesis and methodology

3.1 Hypothesis development

Firstly, in accordance to Myers and Majluf (1984) and financing hierarchy, when financially constrained firms have difficulty raising external funds than internal funds, they tend to rely more on internal financing since external financing is too costly for them. Therefore, following this fundamental theory, Fazzari et al. (1988) and other previous literature indicate that investment cash flow sensitivity should be higher for financially constrained firms than unconstrained firms. Thus, this paper is going to test whether this inequality of investment cash flow sensitivity actually holds as the theory suggested. Furthermore, as the financial crisis can be seen as an unusual time consisting of market frictions, the difference of investment cash flow sensitivity between constrained and unconstrained firm increases during unusual times because of the growing wedge between the cost of internal and external funds. These lead to the first hypothesis:

Hypothesis 1: The investment cash flow sensitivity is higher for financially constrained firms than unconstrained firms, and the difference between two groups is more pronounced during the time of crisis.

Moreover, previous literature (Almeida et al., 2004) emphasizes the importance of reserved cash holdings on corporate investment activities. According to the precautionary motive theory, financially constrained firms would like to hold more cash in response to an increase in cash flow volatility (Han and Qiu, 2007). Hence, I implement another measure to test the same hypothesis from the aspect of cash holdings following the method of Almeida (2004).

Secondly, to propose factors that xplain the negative relationship between financial constraints and investment cash flow sensitivity, I further consider the influence of bank-dependent characteristics on investment cash flow sensitivity. Chava and Purnanandam (2011) point out that the banking crisis in the late 2000s had an enormous impact on bank-dependent firms. As a lot of banks experienced the liquidity crisis due to the devaluation of financial instruments, it became more difficult for firms to borrow. Banks’ reluctances to lend make it more difficult for bank-dependent firms to get loans and to acquire external funds. Therefore, bank-dependent firms face a higher level of financial constraints and their investment cash flow sensitivities are expected to increase during the time of crisis. Besides, due to the reason that bank-dependent firms have more reliance on big banks which may be affected by the

(15)

crisis to some degree, it is expected that bank-dependent firms may experience a relatively higher increase than non-bank-dependent firms during the time of crisis. These arguments lead to the second hypothesis:

Hypothesis 2: The investment cash flow sensitivity increases significantly for bank-dependent firms during the time of crisis.

Thirdly, as a matter of fact, different banks are affected differently by the 2008 banking crisis. Some of them were in considerably bad shape and eventually failed during the peak time of the crisis while the other banks survived and operated profitably. If lead arranger banks in syndicated loans failed or were in the need of a government bailout, it will be even impossible for firms to keep borrowing from them. Hence, it is reasonable to assume that banks which were in bad shape that bank-dependent firms borrow from may have a greater impact on the cash flow sensitivity of investment. This leads to the third hypothesis:

Hypothesis 3: The investment cash flow sensitivity increases more significantly for firms which rely heavily on banks that were in bad shapes than those rely heavily on banks that were in good shapes during the time of crisis.

3.2 Empirical model

3.2.1 The measure of financial constraints

As the previous discussion of different measures of financial constraints, currently there is no consensus in classifying financially constrained firms and the size-age index which proposed by Hadlock and Pierce (2010) is considered as the most appropriate measurement. The principle behind this is that the possibility of becoming financially constrained is declining when firms grow larger and more mature. By using two most related firm characteristics, which is size and age, SA-index can measure each firm more specifically. Furthermore, it is highly possible that a firm’s level of financial constraints changes during a given time period. SA-index can deal with this time-varying situation since it uses continuous variables. Thus, size-age index is used as the measure of financially constrained firms in this paper and is formulated as below:

SAit = (-0.737×Sit) + (0.043×Sit2) – (0.040×Ait) ... (1)

Where Sit is the firm size (log value of total assets), Ai is the firm age, subscript i stands for

(16)

Hadlock and Pierce (2010) state in their paper that the data used to construct the SA-index need to be winsorized at approximately 95% level. The reason for this cutoff is that the relationship between firm characteristics and financial constraints factors below the cutoff is quadratic and non-flat. Therefore, variables used for constructing SA-index are winsorized at 95% level in this paper. According to Hadlock and Pierce (2010), a higher level of financial constraints is consistent with a higher value of SA-index. Thus, observations from above the 60% percentile of the SA-index are considered as financially constrained whereas observations from below the 40% percentile are considered as financially unconstrained. Observations in the middle are neither classified as financially weak nor financially healthy.

3.2.2 The effect of financial constraints on investment cash flow sensitivity

There are two common ways to test the investment cash flow sensitivity, which are Q model and Euler model. Tobin’s Q which included in the Q model is defined as the ratio of market value of the firm to the replacement cost of the firm’s capital stock (Abel & Eberly, 2011), which measures the firm’s investment opportunities. However, there are some concerns about measurement error in Q model. To cope with the measurement error in Q model, Euler model is introduced as it does not contain marginal Q. Although the Euler model can eliminate the investment opportunity bias, there are many counter arguments with regard to this model. Whited (1998) states that the Euler model is a highly parametric model and its empirical power is relatively weak. Moreover, same as Q model, Euler model also needs to be based on a large number of assumptions so there is no particular advantage over Q model in this aspect.

Hence, I include Q in the regression model to take into account for unobservable information of growth opportunities since the influence of attractiveness of future investment opportunities is difficult to measure directly. Following the method of Fazzari et al. (1988), the main regression to estimate the investment cash flow sensitivity as follow:

I𝑖𝑡 K𝑖𝑡−1 = 𝛼𝑖+ 𝛼𝑡+ 𝛽1× 𝑇𝑜𝑏𝑖𝑛′𝑠 𝑄𝑖𝑡−1+ 𝛽2 𝐶𝐹𝑖𝑡 𝐾𝑖𝑡−1+ 𝛽3× 𝐿𝑒𝑣𝑒𝑟𝑎𝑔𝑒+ 𝛽4× 𝑆𝑖𝑧𝑒 + εit ...(2) Where I𝑖𝑡

K𝑖𝑡−1 is firm’s capital expenditure and

𝐶𝐹𝑖𝑡

𝐾𝑖𝑡−1 is firm’s internal cash flow deflated by

beginning-period-of capital stock Kit-1. Control variables are included to eliminate the omitted

(17)

opportunities, are included in the model to capture unobservable information about long-term growth opportunities since firm’s capital expenditure is affected by the attractivness of future investment opportunities. Firm Size, which measured by the logarithm of the book value of the firm’s total assets, accounts for eliminating the size effect, which is the ability that larger firms may have better access to external financing and thus have more opportunities.

Leverage, which measured by the total book value of debt deflated by the capital stock, can

be considered as an important contribute of capital sturcture along with retained earnings and equity since the capital structure also has a significant impact on firms investment decisions. 𝛼_𝑖 and 𝛼_𝑡 control for time and firm fixed effects and

ε

_it is the error term. The coefficient of interest is 𝛽2, which measures the sensitivity of investment to cash flow.

Firstly, different tests are conducted on normal time period, pre-crisis period and the crisis period. In order to test whether investment is more sensitive to cash flow during the time of the crisis, I further include a difference test which compares the investment cash flow sensitivities between financially constrained group and non-constrained group during three different time periods.

Secondly, cash holding plays an important role in determining firm’s investment activities and hoarding cash can facilitate investments oppotunities. Thus, there exist potential endogeneity problem in the baseline investment regression. To accounts for endogeneity problem, I implement another model including cash holdings as dependent variables introduced by Almeida et al. (2004) to test investment cash flow sensitivity as below:

𝐶𝑎𝑠ℎ𝐻𝑜𝑙𝑑𝑖𝑛𝑔𝑠 = 𝛼𝑖 + 𝛼𝑡 + 𝛽1× 𝑇𝑜𝑏𝑖𝑛′𝑠 𝑄𝑖𝑡−1+ 𝛽2× 𝐶𝐹𝑖𝑡 + 𝛽3× 𝑆𝑖𝑧𝑒𝑖𝑡 + εit ………(3)

Where the dependent variable 𝐶𝑎𝑠ℎ𝐻𝑜𝑙𝑑𝑖𝑛𝑔𝑠is the ratio of cash and marketable securities to total assets and the independent variable 𝐶𝐹𝑖𝑡 is the ratio of earnings before extraordinary

items and depreciation (minus dividends) to total assets. Control variables as 𝑇𝑜𝑏𝑖𝑛′𝑠 𝑄_𝑖𝑡−1, is the market value divided by the book value of assets and 𝑆𝑖𝑧𝑒_𝑖𝑡, which is the natural log of assets. The coefficient of interest is 𝛽2, which measures the sensitivity of investment to cash

flow.

3.2.3 The effect of bank dependency on investment cash flow sensitivity

Due to the presence of 2008 banking crisis, firms that rely heavily on bank loans suffer a higher level of financial constraints, which leads to an increase of investment cash flow sensitivity. As for the empirical test, I divide the sample into two subgroups, one is the

(18)

bank-dependent group and the other one is the non-bank-bank-dependent group, then comparing the investment cash flow sensitivities between two groups during normal and the crisis period. Follow the method of Chava and Purnanandam (2011), firms which do not have public debt (i.e. no public debt ratings) are defined as bank-dependent firms whereas firms do possess public debt are defined as non-bank-dependent. After the classification, regression for testing the investment cash flow sensitivity are being runned again by including the BD dummy and its interaction term with cash flow in equation (2):

I𝑖𝑡 K𝑖𝑡−1= 𝛼𝑖 + 𝛼𝑡 + 𝛽1× 𝑇𝑜𝑏𝑖𝑛′𝑠 𝑄𝑖𝑡−1+ 𝛽2 𝐶𝐹𝑖𝑡 𝐾𝑖𝑡−1+ 𝛽3× 𝐿𝑒𝑣𝑒𝑟𝑎𝑔𝑒+ 𝛽4× 𝑆𝑖𝑧𝑒 + 𝛽5× 𝐵𝐷 + 𝛽6_𝐾𝐶𝐹𝑖𝑡 𝑖𝑡−1 × BD + εit...(4)

Where BD represents bank-dependency dummy and the other variables are same as the previous regression. 𝛽₆ is the coefficient of interest that represent the relationship with bank-dependency with investment cash flow sensitivity. The reason to include the interaction variable is that the bank-dependency influences the relationship between internal cash flow and capital expenditure by interacting with internal cash flow.

Firstly, a time-series comparison of the investment cash flow sensitivity between bank-dependent group and non-bank-bank-dependent group during the crisis and pre-crisis period is necessary since this paper is going to test whether bank-dependent firms experience more significant changes than non-bank-dependent firms during the time period of crisis.

Secondly, to further looking into the impact of the relationship between bank-dependent firms and major banks on the sensitivity of cash flow, I divide the bank-dependent firms into two subgroups. By looking into the lead arranger banks in syndicated loans, I define firms that borrow from the composition of more than 50% banks (i.e. 50% of the number of lead arranger banks in the syndicate) which were in bad shape during the crisis as bad-shape-dependent group and the rest are good-shape-bad-shape-dependent group. The reason for this classification is that apart from the simple bank dependence, looking at the influence of specific banks which experienced a failure during the crisis allows me to directly comment on the impact of banking crisis on borrower’s investment performance.

In terms of the classification of banks that were in good shape and bad shape during the crisis of 2008, I collect information from online media articles and a summary of classification of

(19)

banks in the sample is listed in Table 2. The group of banks which are defined as in the bad shape consisting of the Bank of America which said getting a big government bail out (Rucker and Stemple, 2018); Lehman Brothers that experienced a tragic collapse (Telegraph.co.uk, 2018); HBOS that reported a big loss (Finch and Treanor, 2018); Citibank and Wachovia Bank that crashed down (“Treasury banks bailout”, 2018) and so on. On the other hand, banks which survived and do not suffer big losses are defined as in the good shape, such as JP Morgan that stayed profitable (Money.cnn.com, 2018); Barclays that succeeded during the crisis (Finch and Treanor, 2018); Deutsche Bank that escaped the subprime loss (Landler, 2018) and so on.

Therefore, I will run the investment regression from equation (2) separately for good-shape-dependent and bad-shape-good-shape-dependent firms and compare the difference of coefficients between two group of firms. And the subsample used to run this regression only consists of bank-dependent firms.

Table 2. Banks in good shape and bad shape during the financial crisis

Shape of the banks Banks in syndicated loans

Good shape JP Morgan; Wells Fargo Bank NA; Barclays; Deutsche Bank;

Marshall & Ilsley Bank; Goldman Sachs & Co; SunTrust Banks;

Bank of Tokyo-Mitsubishi UFJ; Morgan Stanley & Co; Royal Bank of Canada; Comerica Bank; BNP Paribas SA

Bad shape Bank of America; Lehman brothers; Citi group; ABN AMRO;

Merrill Lynch; HBOS; Wachovia Bank NA; KeyBank Corp;

BB&T Corp; UBS; Fortis; RBS; Scotiabank; Bear Stearns & Co Inc; Commerzbank AG

4. Data and descriptive statistics

The sample is obtained by selecting public U.S. companies over the period of 2000-2010. The reason for choosing this time period is that this paper mainly focuses on the impact of 2008 financial crisis, thus the time period surrounding the crisis would be relevant and appropriate. Daily data are used for credit ratings and quarterly data are used for all other variables. Financial data regarding testing the impact of financial constraints on investment cash flow sensitivity are available at Compustat and CRSP dataset, while data using for syndicated loans are acquired from screening and analysis database from Thomson ONE.

(20)

cash holdings and capital expenditures. Thus I exclude non-positive observations to eliminate most extreme cases that firms are deeply in financial distress. b) Financial firms (SIC 6000-6999) are excluded due to the reason that they might hold cash to meet capital requirements, thus avoiding bias. Firms in utilities industries (SIC 4910-4939) are also excluded to eliminate possible effects of regulation. c) Firms are required to have at least $10 million of market value of equity to exclude firms with considerable big or small sizes, thus to control for the firm size properly. d) Junk-rated firms are dropped because this paper aim to compare bank-dependent firms which have better access to capital in the debt market (i.e. investment-grade rated firms) during time of the crisis. At the end, I winsorize all variables at the 1st and 99th percentiles to eliminate the potential outliers. This sample selection criterion generates 211,290 firm-quarter observations.

After screening and acquiring all the data needed, financial data from Compustat are merged with CRSP by aligning the gvkey codes. However, Thomson ONE uses 6-digit cusip whereas Compustat uses 8-digit cusip. Thus, merging data from Thomson ONE and Compustat are done manually by matching company names and some discards are made. Firms with insufficient data for both datasets are deleted and it generates 170,512 firm-quarter observations in the end. The detailed definition of construction for variables needed in regressions is displayed in Table 3.

Table 3. Definition of variables

Variable Definition

Iit/Kit-1 Fixed investments/net property, plant and equipment

Tobins Qit-1 (Book value of assets + market value of equity − book value of

common equity – balance sheet deferred taxes) /book value of assets

CFit (Income before extraordinary items + depreciation and

amortization)/total assets

CFit/Kit-1 (Income before extraordinary items + depreciation and

amortization)/net property, plant and equipment

CashHoldings Cash and marketable securities/total assets

Size Age

Leverage TCFit

ΔSales

Log value of total assets

The number of years the firm is contained in Compustat with a non-missing stock price

Book value of total debt/net property, plant and equipment

(Income before extraordinary items + depreciation and amortization + all costs incurred to development of new products & services)/ total assets

(21)

Investment ratio

and equipment

(Investment in year 1 + investment in year 2)/2×investment in year 0

Due to the fact that the impact of financial constraints on investment cash flow sensitivity is more pronounced during the time of crisis and banks are reluctant to lend to firms which are highly constrained. Thus, I implement a test focusing on the impact of the financial crisis by looking at the investment cash flow sensitivity in a window around the 2008 financial crisis. I constructed a pre-crisis (data precede the Lehman debacle) period which consists of 2007Q3, 2007Q4, 2008Q1, 2008Q2 and 2008Q3, and the crisis peak period which is 2008Q4. Since the period after the crisis may have some lasting effects and will therefore cause some biasedness, the control group is the normal times sample period from 2000-2007 before the crisis. Table 4 presents the descriptive statistics of financially unconstrained and financially constrained groups respectively.

Table 4. Descriptive statistics

(1) Full sample (2) Financially unconstrained firms (3) Financially constrained firms

Mean Median SD Mean Median SD Mean Median SD

Iit/Kit-1 0.1847 0.1052 0.1369 0.2203 0.1133 0.1495 0.1478 0.0996 0.1150 Qit-1 1.6010 1.3737 0.7815 1.5800 1.3512 0.7874 1.6311 1.4052 0.7721 CFit/Kit-1 0.1414 0.0765 0.2599 0.1435 0.0699 0.2757 0.1383 0.0861 0.2339 CFit 0.0237 0.0226 0.0192 0.0219 0.0236 0.0206 0.0241 0.0236 0.0169 Cash holdings 0.1971 0.0991 0.2295 0.1228 0.0686 0.1484 0.2680 0.1717 0.2677 Size 5.6278 5.2296 1.3773 7.4457 7.3551 1.2148 3.8935 3.8206 0.9402

Notes: All variables are constructed from Compustat variables. This table reports mean, median and standard deviation of the variables for different group of firms. Column 1 reports the summary statistics for the full sample. Column 2 reports the summary statistics of group of firms that is financially constrained while column 3 reports the summary statistics of group of firms that is

financially unconstrained.

As shown in Table 4, financially constrained firms have a lower level of capital expenditures (Iit/Kit-1) ratio than financially unconstrained firms, which are 0.1478 and 0.1847 respectively.

As financially unconstrained firms may have better access to external funds and have more capital to invest in different activities, thus yielding higher capital expenditure than their counterparts. Moreover, the cash flow ratio (CFit/Kit-1) of financially constrained firms

(22)

(0.1383) is lower than unconstrained firms (0.1435), which is consistent with the expectation that financially constrained firms have difficulty generating enough cash flows for future growth activities. Furthermore, financially constrained firms are generally with smaller firm size (3.8935) whereas unconstrained firms’ sizes (7.4457) are much bigger.

In addition, a cross-correlation test is conducted to further check for multicollinearity in the regression, results are shown in Table 5. As we can see from this table, correlations between dependent and independent variables in each regression such as Iit/Kit-1 and CFit/Kit-1 (0.5100),

Iit/Kit-1 and Tobins Qit-1 (0.1515), CashHoldings and CFit (-0.2917), CashHoldings and Tobins

Qit-1 (0.3869) and CashHoldings and Size (-0.2829) are relatively higher compared to other

correlations. In addition, correlations between control variables and dependent variables such as Size and Iit/Kit-1 (-0.0951) and Leverage and Iit/Kit-1 (0.1270) are not trivial, indicating that

these have some influence on the dependent variable and should be included as control variables. Thus, although it is impossible to conclude any relationship between variables only based on correlations, this gives me the motivation to examine the influence of those independent variables on the capital expenditure (Iit/Kit-1) and cash holdings ratio

(CashHoldings). Moreover, correlations between independent variables in regressions such as

CFit/Kit-1 and Tobins Qit-1 (-0.0192), CFit and Tobins Qit-1 (-0.0168) and CFit and Size (0.0326)

are not high enough to deduce a significant relationship. Hence, this result shows that there is no concern about multicollinearity problem between those independent variables.

Table 5. Cross-correlation matrix table

Iit/Kit-1 CFit/Kit-1 CFit CashHo

ld-ings

Tobins Qit-1

SA Size Age Levera

-ge Iit/Kit-1 1.0000 CFit/Kit-1 0.5100 1.0000 CFit -0.0319 0.5298 1.0000 CashHolding 0.1333 -0.2549 -0.2917 1.0000 Tobins Qit-1 0.1515 -0.0192 -0.0168 0.3869 1.0000 SA -0.1042 0.1760 0.3424 -0.2903 -0.1795 1.0000 Size -0.0951 0.1752 0.0326 -0.2829 -0.1458 0.8207 1.0000 Age -0.1308 0.0668 0.1631 -0.1863 -0.1185 0.7489 0.3254 1.0000 Leverage 0.1270 -0.0104 -0.0811 -0.0317 0.0089 0.0227 0.0267 0.0029 1.0000 Notes: All variables are constructed from Compustat. This table reports the correlation coefficient of relevant variables. The detailed definition of variables are in Table 2.

(23)

5. Results

5.1 The impact of financial constraints on investment cash flow sensitivity

Firstly, a test on the impact of financial constraints on investment cash flow sensitivity is conducted. Panel A and Panel C of Table 6 represents the coefficients of regression for financially constrained and unconstrained group during normal times, pre-crisis period and the crisis period, using both equation (2) and equation (3). Panel B and Panel D compare the differences between financially constrained and unconstrained firms during each of the three sample periods. In terms of the normal times, as displayed in column (1) and (2) of Panel A, the coefficients of cash flow ratio for both financially constrained (0.269) and unconstrained groups (0.270) are only significant at 10% level. The significance of these coefficients implies that firms do not necessarily need to forego investment activities when facing a decrease in internal cash flows. Moreover, as we can see from the difference test in column (1) of Panel B, the difference between the magnitude of financially constrained and unconstrained firms is relatively small and not significant. This can be inferred as the wedge between internal and external financing is not so large in the period of normal times and the impact of financial constraints on investment cash flow sensitivities is relatively insignificant.

When it comes to the pre-crisis period before 2007, as shown in column (3) and (4) of Panel A, financially constrained firms exhibit a significant investment cash flow sensitivity coefficient of 0.221, illustrating that financial constraints have spilled into their investment activities. The coefficient for financially unconstrained firms is 0.269, also significant at 1% level. From the difference test in column (2) of Panel B, the difference between firms in two states is 0.0590, significant at 1% confidence level as well. Finally, column (5) and (6) in Panel A exhibit the coefficient estimates for both groups of firms during the period of crisis. There is a considerable increase of the investment cash flow sensitivity for financially unconstrained firms and the coefficient is 0.255 while it is only 0.0900 for financially constrained firms, both significant at 1% level. It is also important to note that the coefficient for financially constrained firms is 0.165 lower than that of financially unconstrained firms, which is a relatively bigger difference than other periods. This larger difference between the investment cash flow sensitivity for financially constrained and unconstrained firms indicates that they react much more differently to the cash flow shortfalls during the crisis period, and financially constrained firms are affected less by the crisis than unconstrained firms.

(24)

As for the results obtained using the cash holding model Almeida (2004)’s, Panel C and Panel D shows that the general pattern is essentially the same. The result still holds when implementing another model that accounts for endogeneity, further supporting the validity of the results obtained using the main regression from FHP (1988). According to the theory, financially constrained firms should use more internal cash flows than external funding since the cost of external financing is higher and information asymmetry exists, thus resulting in higher investment cash flow sensitivity. Nonetheless, it is interesting to note that the investment cash flow sensitivity for financially unconstrained firms are greater than that of financially constrained firms for all time periods, which is not consistent with the rationale. Thus, it can be concluded from the results that the impact of financial constraints on investment cash flow sensitivity is negative and significant, and is more pronounced during the crisis period. This negative relationship is consistent with previous literatures including Kaplan and Zingales (1997).

Table 6. Investment cash flow sensitivities: Effects of Financial Constraints Panel A: Fazzari’s model

Normal times Pre-crisis period Crisis period

Constrained Unconstrained Constrained Unconstrained Constrained Unconstrained

(1) (2) (3) (4) (5) (6) CF/K 0.269* 0.270* 0.221*** 0.269*** 0.0900*** 0.255*** (0.1502) (0.1634) (0.00239) (0.0299) (0.0268) (0.0323) Tobins Q -0.00624 0.000992 -0.00259 0.00206 -0.0271*** -0.0538*** (0.00496) (0.00370) (0.00549) (0.00462) (0.00709) (0.0109) Size Leverage 0.0414** (0.0101) -0.00498** (0.00232) 0.0288** (0.00528) -0.00326 (0.00256) 0.0561*** (0.0130) -0.00594** (0.00288) 0.0450*** (0.00301) -0.00654** (0.00301) 0.0683*** (0.0168) -0.00752* (0.00386) 0.0944*** (0.0224) -0.00828* (0.00480) N 75,579 63,739 56,653 43,024 20,135 19,420 R2 _0.273 _0.362 _0.446 _0.294 _0.489 _0.438 Firm FE Industry-year FE YES YES YES YES YES YES YES YES YES YES YES YES

Panel B: Difference Tests of investment cash flow sensitivity

(1) (2) (3)

Difference: Difference: Difference:

(1) - (2) (3) - (4) (5) - (6)

CF/K -0.0010 -0.0590*** -0.1650***

(25)

Panel C: Almeida’s Model

Constrained Unconstrained Constrained Unconstrained Constrained Unconstrained

(1) (2) (3) (4) (5) (6) Cash flow 0.0901 0.113** 0.119** 0.283*** 0.0932*** 0.240*** (0.108) (0.0472) (0.0552) (0.0849) (0.0160) (0.0314) Tobins Q 0.0144*** 0.0115*** 0.0119*** 0.0156*** 0.0125*** 0.0117*** (0.00407) (0.00325) (0.00375) (0.00489) (0.00103) (0.00208) Size 0.00728 -0.0111** -0.0122** 0.00801 -0.00113 -0.0102** (0.00478) (0.00437) (0.00535) (0.00519) (0.00440) (0.00450) N 76,351 64,515 57,231 43,640 20,300 19,603 R2 _0.813 _0.832 _0.853 _0.841 _0.897 _0.897 Firm FE Industry-year FE YES YES YES YES YES YES YES YES YES YES YES YES

Panel D: Difference Tests of investment cash flow sensitivity

(1) (2) (3)

Difference: Difference: Difference:

(1) - (2) (3) - (4) (5) - (6)

Cash flow -0.0116* -0.0257** -0.0868***

(0.00613) (0.01507) (0.01553)

Notes: This table reports coefficients of cash flow from estimating regression (2) and (3). All variables are defined in Table 3. In Panel A and B, the dependent variable is capital expenditure

(Iit/Kit-1) and the dependent variable is cash holdings (Cashholdings) in Panel C and D. Panel A shows

the coefficient estimates for financially constrained and unconstrained firms during normal times, pre-crisis period and the pre-crisis period. Panel B reports the difference tests for coefficients from Panel A and only coefficients of CF/K is reported for brevity. Panel C and D display the same as Panel A and B except using a different model. Year and firm fixed effects are included and all observations are clustered at firm level. Robust standard errors in parentheses: *** p<0.01, ** p<0.05, * p<0.1.

Although the empirical results are consistent with the puzzling finding of Kaplan and Zingales (1997), Allayannis and Mozumdar (2004) discover a declining trend of investment cash flow sensitivity. Therefore, it is crucial to confirm that this negative influence of financial constraints on investment cash flow sensitivity still exists in recent years and a test of time series of investment cash flow sensitivity is conducted. The sample is divided into 5 subsamples by sorting the years and separating them evenly, each sub-period consists of 2 years. As we can see from Table 7, there exhibits a decreasing trend of the investment cash flow sensitivity over time. In the first sub-period from 2000 to 2001, the investment cash flow sensitivity of the whole sample is 0.594. However, it drops to 0.153 in the period from 2008 to 2010, which is a significant decline. Specifically, investment cash flow sensitivity has declined for both financially constrained firms and unconstrained firms, but it is still

(26)

significant at 1% level in the last sub-period. As shown by comparing the difference of cash flow sensitivities between financially constrained and unconstrained firms, though the magnitude of the difference between them is declining, they are still statistically significant at 1% confidence level. The investment cash flow sensitivity for financially unconstrained firms is 0.03 higher than that of constrained firms in the period of 2000 to 2001 and is 0.0666 higher in the period of 2008 to 2010. Thus, this result indicates that the puzzling finding of investment cash flow sensitivity and financial constraints still exists in recent years. During the period of 2000 to 2001, dot-com bubble burst so I omit that period to avoid potential bias in Figure 1. Nevertheless, the previous results have not changed, as shown in Figure 1.

Table 7. Time series of investment cash flow sensitivity

(1) 2000-2001 (2) 2002-2003 (3) 2004-2005 (4) 2006-2007 (5) 2008-2010

(A) All sample 0.594*** 0.231*** 0.193*** 0.208*** 0.153***

(0.0168) (0.0280) (0.0380) (0.0337) (0.0193) (B) unconstrained 0.206*** 0.120*** 0.1752*** 0.0919*** 0.0769*** (0.0159) (0.0275) (0.0178) (0.0215) (0.0068) (C) constrained 0.176** 0.0555*** 0.1229*** 0.0553** 0.0103** (0.0842) (0.0148) (0.0295) (0.0317) (0.00417) (B) – (C) 0.030** 0.0645** 0.0523*** 0.0366*** 0.0666*** (0.0142) (0.0285) (0.0174) (0.0118) (0.0202)

Notes: This table reports coefficients of cash flow from estimating equation (2). The dependet variable is capital expenditure (Iit/Kit-1) and independent variables are cash flow ratio (CF/K) and

tobinsq (Tobins Q). Investment cash flow sensitivity is obtained for 5 sub-periods, which is 2000-2001, 2002-2003, 2004-2005, 2006-2007 and 2008-2010 and only investment cash flow sensitivity is reported in the table. Year and firm fixed effects are included and all observations are clustered at firm level. Robust standard errors in parentheses: *** p<0.01, ** p<0.05, * p<0.1.

Figure 1. Time series of investment cash flow sensitivity: Investment cash flow sensitivity is

obtained for 5 sub-periods, which is 2000-2001, 2002-2003, 2004-2005, 2006-2007 and 2008-2010. The blue line represents the full sample, red line represents financially unconstrained

0 0.05 0.1 0.15 0.2 0.25 2002 2004 2006 2008 All unconstrained constrained

(27)

firms and green line represents financially constrained firms.

5.2 The relationship between bank dependency and investment cash flow sensitivity

As suggested by the second hypothesis, banking crisis has a negative influence on bank-dependent firms due to their inability of borrowing from the banks. Hence, it is possible that bank-dependent firms suffer a significant increase of financial constraints, which leads to higher investment cash flow sensitivities during the time of crisis. From the regression results of equation (4) in Table 8, the coefficients estimate of the interaction variable bd×CF/K indicates the impact of bank-dependency dummy on investment cash flow sensitivity. In Table 8, coefficients of bank-dependency dummy are negative in all periods, and the magnitude increases from the normal times period (-0.0161) to the crisis period (-0.0717), which is significant at 1% level. This decline of the coefficient indicates that bank-dependent firms cut their investments significantly after the crisis. Moreover, the coefficient of the interaction term bd×CF/K is -0.117 in column (1) and is only significant at 10% level. And

the coefficient of the interaction term is -0.244 in the crisis period in column (3), which is significant at 1% level. The bigger magnitude of the negative coefficient estimates of bank-dependency dummy and interaction term during the crisis period indicates that investment cash flow sensitivities for bank-dependent firms are significantly lower than non-bank-dependent firms, and the difference between two groups are larger during the time of crisis.

Table 8. Investment-cash flow sensitivity for bank-dependent and non-bank-dependent firms

(1) (2) (3) bd -0.0161 -0.0585*** -0.0717*** (0.0122) (0.00793) (0.0105) bd×CF/K -0.117* -0.220*** -0.244*** (0.0716) (0.0196) (0.0228) CF/K 0.0982*** 0.199*** 0.221*** (0.0206) (0.0188) (0.0219) Tobins Q 0.00786*** 0.0178*** 0.0155*** (0.00278) (0.00124) (0.00148) Size 0.0498*** 0.0428*** 0.0714*** (0.0100) (0.00319) (0.00458) Leverage -0.00391** -0.00241*** -0.00248** (0.00165) (0.000812) (0.00107) N 138,504 99,181 39,257 R2 _0.334 _0.37 _0.425

(28)

Firm FE YES YES YES

Industry-year FE YES YES YES

Notes: This table reports coefficients of cash flow from estimating regression (4). All variables are defined in Table 3. The dependent variable is capital expenditure (Iit/Kit-1). Column (1) shows the

coefficient estimates for bank-dependent and non-bank-dependent firms during normal times, column (2) shows the coefficients estimate for pre-crisis period and column (3) shows the coefficients estimate for the crisis period. Year and firm fixed effects are included and all observations are clustered at firm level. Robust standard errors in parentheses: *** p<0.01, ** p<0.05, * p<0.1.

Moreover, it is more straightforward to see the trend by comparing the bank-dependent and non-bank-dependent firms’ investment cash flow sensitivities annually. Table 9 exhibits the time series analysis by running the baseline investment regression for the sample of each year. Investment cash flow sensitivity for both bank-dependent and non-bank-dependent firms are displayed in the table and only coefficients of CF/K are included in the table for brevity. The reason for looking at annual coefficients is that by showing the trend of investment cash flow sensitivities, it enables us to see what happened exactly during the crisis time in the year of 2007 and 2008. Additionally, the annual investment cash flow sensitivities also help to discover the short-term and long-term effect separately and this might help explaining the negative relationship found between financial constraints and investment cash flow sensitivity. Figure 2 is a time series plot of the investment cash flow sensitivities for bank-dependent and non-bank-bank-dependent firms, which shows the trend more directly.

As we can see from Table 9, bank-dependent firms’ investment cash flow sensitivities are much lower than non-bank-dependent firms in most of the periods. Nonetheless, in the year of 2007, there is no significant difference between two groups of firms when the crisis started. The investment cash flow sensitivity for bank-dependent firms rises slightly to 0.173 in the year of 2007, because of the higher level of financial constraints they encounter. Therefore, it can be confirmed that a supply shock increases the level of financial constraints of bank-dependent firms and boosts up their investment cash flow sensitivities immediately.

There is no upward or downward trend of investment cash flow sensitivity for non-bank-dependent firms during the crisis period and the level of cash flow sensitivity fluctuate around 0.15 during the whole period. However, for bank-dependent firms, the investment cash flow sensitivity decreased during the year of 2008 significantly. To be more specifically, the level of investment cash flow sensitivity declined significantly to 0.0627 in 2008 and stayed low at 0.0795 in the year of 2009, which is the exact opposite of the expectation that