The U-Shaped Investment Curve

Transcription

1 MSc in Finance and International Business Aarhus School of Business University of Aarhus Master thesis The U-Shaped Investment Curve Empirical evidence from a panel of US manufacturing and mining firms Author: Elena Spiridonova Yildiz, MSc student Contact: spirik@gmail.com Supervisor: Stefan Hirth, Assistant Professor, PhD Contact: stefanh@asb.dk January 010

2 Abstract: The study empirically investigates the effect of the availability of internal funds on a firm s investment. The analysis is performed on data for US manufacturing and mining firms within the time period It indicates that investment increases monotonically in internal funds when they are large. But when they are sufficiently low, investment decreases in internal funds. These findings are consistent with the U- shaped investment curve introduced by Cleary, Povel and Raith (007). Moreover, I detect an inverted U-shaped relationship between investment-cash flow sensitivities and internal funds. Investment of firms with low liquidity around zero is the most sensitive to changes in cash flows. Finally, firms with the lowest level of internal liquidity are found to be the smallest and the most leveraged within the sample. They appear to be the most financially constrained. At the same time, these firms have the highest growth opportunities and exhibit high investment rates. On the one hand, these results support the notion that investment choice is based on a trade-off between the cost effect and the revenue effect [Cleary, Povel and Raith (007)]. On the other hand, such a high investment activity of these firms can be attributed to the fact that external investors provide additional finance as a result of a high future profit anticipation. Key words: capital market imperfections, investment, internal funds, financing constraints, investment-cash flow sensitivity Author: Elena Spiridonova Yildiz Student no. ey7783 Exam no Study: M.Sc. in Finance and International Business School: Aarhus School of Business, University of Aarhus i

3 Acknowledgements I would like to express my gratitude and appreciation to my supervisor for his helpful and constructive advices. I would also like to thank my dear husband for his support and care; my sons for being patient; and my friends for their help and support. ii

4 Table of Contents 1. Introduction Problem Statement Delimitation The Structure of the Thesis Literature Review Background of Investment Theories Investment and Financing Constraints: Recent Theoretical and Empirical Studies Market Imperfections and Financing Constraints Monotonicity of Investment Behavior in Financing Constraints Non-monotonicity of Investment Behavior in Financing Constraints Methodology The U-shaped Investment Curve: Theoretical Model The U-shaped Investment Curve: Empirical Approach Sample Characteristics and Variables Description The Manufacturing Industry Sector The Mining Industry Sector Empirical Analysis Summary Statistics Correlation Analysis Mean and Median Investment Levels Standard Regression Analysis Split-Sample Regressions Financing Constraints and Investment-Cash Flow Sensitivities Relation with Previous Investment Studies Split-Sample Regression for Payout Groups Split-Sample Regression for KZindex Groups Analysis of Financial Characteristics of Manufacturing and Mining Firms with Respect to the Level of Internal Funds Conclusion Bibliography Appendices... I

5 1. Introduction 1. Introduction The objective of any company is to maximize its market value by taking investment projects whose costs are less than the cash flows generated from them in the future. Companies invest in resources and competences in order not only to get and sustain competitive advantage, but also to exist and compete in the market. It is essential to make correct investment decisions at the right time. In the real world one can observe constant swings in investment activity with a dissimilar behavior among companies. This puzzling investment pattern creates a theoretical challenge for economists to come up with a model that explains firm investment behavior. Which microeconomic and macroeconomic factors determine the time and the magnitude of investment? Does the method of financing of investment projects matter? How do financing constraints affect investment? Answers to these questions have been sought for many decades. Investment theories went through an evolution starting from the idea by Modigliani and Miller (1958) that in a world of perfect capital markets a firm s financial structure does not affect its financial policy and market value. Later, it is recognized that in conditions of imperfect capital markets financial structure influences the level of investment. Market imperfections cause financing constraints, when a firm faces a gap between costs of internal and external finance. These facts lead to an interdependency of the firm s financing and investment decisions. Particularly, the availability of internal funds or net worth provides a cost advantage over external finance and affects investment, as it reduces agency costs [Jensen and Meckling (1976)] and problems with information asymmetry between managers and outside investors [Myers and Majluf (1984)]. 1 A vast strand of literature shows that the lower the borrower s net worth is, the higher the agency costs are, and consequently, the higher the premium external investors will demand. As a result, it leads to a reduction in investment and production. Researchers are analyzing either investment volume or its sensitivity to internal funds with respect to the level of financing constraints. Most of the investment studies until the late 90 s are based on the traditional idea that the 1 Here it is worth mentioning that the high value of net worth or liquidity may at the same time lead to an increase in agency costs in a form of overinvestment [Jensen (1986), Hirth and Uhrig-Homburg (007)]. See a review of investment literature in Hubbard (1998) and Schiantarelli (1996). 1

6 1. Introduction relationship between internal funds and investment activity is monotonically increasing. The main conflict is on the behavior of investment-cash flow sensitivities in financing constraints. One stream of researches follows the findings of Fazzari, Hubbard and Petersen (1988) (hereinafter FHP) stating that investment of firms identified as more financially constrained is more sensitive to cash flow. Other studies support the contradictory findings of Kaplan and Zingales (1997) (hereinafter KZ) where firms that are least likely to be financially constrained are more investmentcash flow sensitive. A recent paper by Cleary, Povel and Raith (007) (hereinafter CPR) introduces a new idea that the relation between investment and internal funds is U- shaped. Investment increases in internal funds when they are large. But when they are low, investment starts to increase as internal funds decrease further. The authors argue that this non-monotonic behavior of investment is caused by a trade-off between two effects: 1) the risk of default and liquidation; ) the need to generate revenue to repay debt. They find strong empirical support to their model on a large sample of US companies. Moreover, the authors come up with an explanation of the conflicting findings of FHP (1988) and KZ (1997), which is based on their different sample construction criteria and classification schemes Problem Statement I think that the paradox detected by CPR demonstrating relatively high investment activity of non-profitable companies with sufficiently negative levels of internal liquidity requires further investigation. That is why I decide to replicate and extend the study by CPR in order not only to test the predicted U-shaped relationship between investment and internal funds empirically, but also to provide some explanation of the detected increasing investment volume in financing constraints. The aim of this study is to examine the following issues: 1) Is it possible to detect a U-shaped relationship between investment and internal funds on a sample over a more recent time period? Is there any difference in investment behavior between industry sectors? Is the pattern still U-shaped if alternative proxies of internal funds are used? ) What is the behavior of the investment sensitivity to internal funds in financing constraints?

7 1. Introduction 3) Is the explanation of the conflicting findings of FHP and KZ provided by CPR convincing? 4) Is there a significant difference in the financial characteristics between firms with positive and negative liquidity? Are there any explanations why companies with negative internal liquidity are able to invest at such a high level? I construct two unbalanced samples from the Compustat database: for the US manufacturing and mining industry sectors for the period The methodology is based on the model and empirical tests from CPR. My analysis is based on a number of statistical tests: summary statistics, correlation analysis, standard regression analysis, split-sample regressions and plots of mean and median values. 1.. Delimitation The research is delimitated by focusing solely on two US industry sectors within the period I choose not to cover the EU market because of the lack of data availability and much shorter time horizon caused by the replacement of local European currencies by the Euro in 00. The decision to analyze the US market is also driven by a desire to obtain consistency and comparability with CPR. A firm s investment behavior is measured through its investment in net fixed assets (capital expenditures). Other measurements, like changes in inventories, accounts receivable, employment, research and development are not considered. Only the effect of internal funds and investment opportunities on investment activity is analyzed. Thus, I do not take into account other possible influences, like corporate governance, monitoring mechanisms, changes in monetary policy, country tax systems, legal environment or financial development The Structure of the Thesis The thesis is organized as follows. In the next section the literature review is presented. In section three the methodology for the analysis is described. Section four contains the data selection criteria and variable description. An empirical analysis with an interpretation of the results is provided in section five. Section six relates the obtained empirical findings to previous investment studies. The analysis of firm financial characteristics with respect to internal funds availability is performed in section seven. Finally, in the last part of the thesis conclusions and suggestions for future work are presented. 3

8 . Literature Review. Literature Review.1. Background of Investment Theories The foundation for the theory of investment is built on the Fisher Separation Theorem [Fisher (1930)]. It states that investment decision is made according to the preferences of the capital market, e.g. the cost of capital, while an individual investor s subjective preferences are irrelevant. The value of an investment project is equal to the present value of its future cash flows. Moreover, it is independent of the method of financing: debt, equity or cash. The Modigliani-Miller Irrelevance Theorem [Modigliani and Miller (1958, 1963)] further extends this notion. It states that the total market value of a firm is not dependent on its financing decision. This theory is built on the following strict assumptions: 1) a firm s financing choice does not affect the total cash flows to its equity and debt holders; ) the absence of transaction costs; 3) there are no arbitrage opportunities in the market. 1 In a world of perfect capital market, where these assumptions are satisfied, a firm s choice of the level of internal funds, payout policy or debt leverage will neither change its investment policy nor its market value. The Modigliani-Miller Theorem gave birth to the Neoclassical theory of investment which is based on conditions of perfect capital markets. It states that a firm s cost of capital is not affected by its financial structure. This theory is described in detail by Hall and Jorgenson (1967). The authors find empirical evidence that tax policy influences the level, timing and composition of investment expenditures. In the real world, the prediction drawn by the Neoclassical theory fail to hold, as it is observed that investment expenditures respond to current income, sales and output. At the same time, the dependency on the cost of capital is much weaker than expected. This empirical puzzle created a challenge for academics to come up with a better model describing investment behavior. There are a number of alternative models besides the neoclassical theory of investment, where the main are: Accelerator Principal model, based on the effect of the optimum capacity utilization and economies of scale on the level of investment [J.M. Clark (1917), Chenery (195)]; 1 Hillier, Grinblatt and Titman (008, p. 511). See FHP (1988, p.196). 4

9 . Literature Review The Modified Neoclassical (Putty-Clay) model, built on a number of underlying ex-ante factors proportions [Bischoff (1971)]; The Profit Motive theory, based on the market value of a firm. Current and past profits/cash flows influence future profitability which consecutively determines investment level; Tobin s Q or Securities-Value model, where investment depends on the market valuations of a firm s asset [P.K. Clark (1979)]. Liquidity Funds model, which emphasizes the effect of internal liquidity on investment ratio. The empirical evidence on this subject is quite contradictory and created a lot of discussions. One of the earliest studies on investment behavior is performed by Meyer and Kuh (1957) through the analysis of a firm s investment activity and various variables, i.e. sales, annual depreciation, net profit, liquidity stock, dividendprofit ratio, age of assets, etc. The conclusion is that the investment theories are not mutually exclusive but rather based on different arguments. This research got a number of critics, but all in all, it highlights the significance of financial considerations in a firm s investment. Jorgenson and Siebert (1968) test different investment theories by comparing possible investment patterns for fifteen large manufacturing firms. The authors determine the lag distribution function for each theory. The Neoclassical theory of investment is found to fit the data best. Another study by Elliott (1973) criticizes the ability of the Neoclassical theory to explain investment behavior. The author determines that the incorporation of several variables in this model does not show better results compared to the singlevariable models, such as the accelerator, expected profits and liquidity ones. A posterior paper by Bernanke, Bohn and Reiss (1988) argues that none of the models for investment demand is correctly specified. The study claims that the previous empirical studies, which have ranked investment models by their goodness of fit or statistical prediction error criteria, have a number of serious drawbacks. For example, the selected samples are too small, and inappropriate comparative criteria and ranking procedures are applied. Finally, most of the studies at that time did not recognize the fact that the sensitivity of investment expenditures to various variables may depend on the firm-specific characteristics, such as size, age, payout or other 5

10 . Literature Review criteria. In other words, the classification of companies may be essential in order to capture the investment pattern observed in practice... Investment and Financing Constraints: Recent Theoretical and Empirical Studies..1. Market Imperfections and Financing Constraints In the s academics started to set more focus on the idea that market imperfections had to be taken into account. It is recognized that market imperfections may constraint investment and cause interdependency between a firm s financing and investment decisions. 1 At this point, it is essential to clarify that a firm is considered to be financially constrained if it lacks enough of its own funds and is not able to obtain them externally in order to invest in a positive NPV project. There are different types of market imperfections, such as costs of financial distress, transaction costs, costs of under- and overinvestment, agency problems and asymmetric information, which may constrain a firm s investment activity. Asymmetry in information and agency problems are considered to be the primary disturbing factors in the investment literature. The problems with asymmetric information are first described by Akerlof (1970) in The market for Lemons. Applied to the case of investment decision, the problem is described as follows: a manager knows more about the quality of a new investment project but can not reveal it completely to other market participants. It creates a problem for external investors to make a distinction between the quality of the companies. As a result, they value a firm as an average value of all firms and require the premium to cover the loss arising from the investing in low-quality companies. Moreover, the more severe the effect of asymmetric information between managers and outside investors is, the more expensive the issue of new debt or equity becomes. Imperfect information and risk aversion cause additional costs associated with incentive or agency problem. This concept of agency costs is introduced by Jensen and Meckling (1976). The relationship between a manager and an external investor is contractual and described as being principal-agent. Their conflicting interests lead to agency problems. Taking into consideration the manager s incentives to act in his/her 1 See a review of investment literature in the study by Hubbard (1998). See, for example, FHP (1988). 6

11 . Literature Review own interest, the outside provider of funds requires a higher return to cover the costs arising from monitoring and possible moral hazard behavior. Moreover, the lower the borrower s net worth or internal funds are, the higher the agency costs are, and consequently, the higher premium the external investors will demand. 1 As a result, it leads to a reduction in liquidity, and to a subsequent decrease in investment and production. At the same time, the high value of a borrower s internal funds does not suggest an elimination of agency costs. In particular, it may cause the assets substitution problem, when equity holders have a tendency to invest risk-free liquid funds in risky projects. As managers pursue their own incentives, they may invest in perks, projects that pay off earlier or in empire-building instead of distributing funds to equity holders in dividends. Therefore, large internal funds may lead to a higher probability of their wasteful use and to the overinvestment problem which are not optimal for the value maximization of a firm [Jensen (1986)]. 3 Problems with asymmetric information and agency conflicts are further applied in the Pecking Order Hypothesis described by Myers and Majluf (1984). According to this theory, the sources of financing experience different cost of capital and have an impact on the firm s investment policy. In more detail, because of the effect of adverse selection, firms prefer internal funds to external, where the latter are more expensive. In case when outside financing is essential in order to invest in a desirable project, new debt is preferred to new equity. This choice of financing resembles a hierarchy structure.... Monotonicity of Investment Behavior in Financing Constraints Most of the subsequent investment studies become a departure from the simple investment models based on the perfect market conditions toward models built on the asymmetric information argument. It creates a new direction in investment theories and empirical studies, where the recognition of the impact of market imperfections and financing constraints becomes an important issue. The main challenge is to answer the question: What is the effect of financing constraints on investment? 1 See, for example, Bernanke and Getler (1989, 1990). Hillier, Grinblatt, and Titman (008, p. 65). 3 Hirth and Uhrig-Homburg (007) provide a solution to this problem through deriving the optimal level of liquid funds that eliminates agency costs. 7

12 . Literature Review The paper by FHP (1988) is one of the first studies which empirically test the impact of financing constraints and market imperfections on investment demand. They argue that an increase in asymmetric information might cause an increase in the cost of external financing, such as new debt or equity. It creates a difference between opportunity cost of external and internal finance (retained earnings and cash flow). When internal funds are insufficient, a company has to use external funds in order to invest in a desirable project. If their cost disadvantage is essential, it will mostly rely on the level of internal finance as it is predicted by the Pecking Order Hypothesis. As a consequence, investment spending of this company will be dependent on the fluctuations in cash flows and earnings. FHP use data for 4 US manufacturing companies covering the time period The sample is spit into three groups based on a firm s dividend-income ratio. The assumption is that high-dividend firms represent mature, large and financially healthy companies, while the low-dividend group contains younger and smaller ones. The idea is to make a distinction between more constrained, less constrained and non-constrained companies. The former group is expected to face more severe capital market imperfections and subsequently higher cost of external finance. This technique of splitting the sample makes this paper distinctive from the previous studies described in Section.1, which tested investment models on a sample containing all firms. The authors provide empirical support to the notion that financial constraints have influence on investment activity. This study indicates that investment policy of firms classified as more constrained is more sensitive to changes in cash flow compared to unconstrained firms. As a result, investment will decrease and investment-cash flow sensitivity will increase monotonically in financial constraints. Therefore, the authors suggest that the magnitude of investment-cash flow sensitivity provides a measure of financing constraints severity. Hoshi, Kashyap and Scharfstein (1991) follow the same approach but apply a different classification procedure. The authors analyze the effect of information asymmetry on investment activity on two sets of Japanese firms. The first set includes companies that belong to an industrial group (keiretsu). Therefore these firms exploit close financial relationship to a bank, which results in more information sharing and a possibility of monitoring. The second group, containing independent firms with weaker bank connections, faces more severe effect of information asymmetry. The 8

13 . Literature Review argument is that the former group has an advantage in raising external capital. The authors empirically find that the group of independent firms exhibits a greater sensitivity of investment to changes in cash flows. Neslihan (00) investigates how sensitive R&D investment is to the availability of cash flow. Firms are classified into two groups, constrained and unconstrained, on a basis of their interest expense to income ratio and dividend payout ratio. The data on US manufacturing sector shows that R&D investment is more sensitive to internal funds for financially constrained firms than for non-constrained ones. Schaller (1993) argues that firm classification schemes by maturity status, concentration of ownership, availability of collateral and membership of an industrial group detect the severity of asymmetric information in a better way than payout behavior applied by FHP. The analysis is performed on a sample of 1 Canadian firms within the time period of The following conclusion is outlined: more constrained companies classified as young, independent, with dispersed ownership structure and operating in manufacturing sector experience a greater sensitivity of investment to cash flow compared to less constrained companies. Some studies are analyzing investment and financing constraints in a different way. 1 This stream of literature points out the drawbacks of the Q investment model. For example, this model is outperformed by other models empirically; Q may not be a good proxy for investment opportunity; and there are problems with the measurement of Q (average vs. marginal). In order to overcome these problems, the Euler Equation of the standard neoclassical model for inter-temporal capital accumulation is employed. This equation describes optimal investment through the adjustment cost model. These studies distinguish between constrained and unconstrained companies by applying different criteria, for instance, sorting by age, the level of debt, dividend ratio, interest coverage ratio, etc. Most of the researchers find that the standard Neoclassical model holds only for the group of companies classified as nonconstrained. Rejection of the model for constrained firms provides evidence on the impact of capital-market imperfections on investment decision. These results support the basis of FHP (1988) and other studies described earlier, suggesting that companies 1 See, for example, Whited (199); Bond and Meghir (1994); Hubbard, Kashyap and Whited (1995). 9

14 . Literature Review classified as more financially constrained are more sensitive to the level of liquidity than those that are less constrained. Conflicting view of KZ (1997) creates a debate on the reliability of the results of the studies summarized earlier. The authors examine and criticize the findings obtained by FHP (1988). The authors use the data from FHP on 49 low-dividend paying firms, classified as the most financially constrained. KZ extract more detailed financial information about these companies funds, i.e. the current level, future demand and availability of access. The data are obtained from several sources, e.g. letters to shareholders, 10-K, financial statement and notes, annual reports, management operation s discussion, etc. The companies are then further classified into five groups: 1) definitely not financially constrained; ) likely not financially constrained; 3) possibly financially constrained; 4) likely financially constrained; 5) definitely financially constrained. The analysis demonstrates that the level of investment and most of the firm characteristics on financial status, such as cash, sales growth, interest coverage ratio, and cash flow decline monotonically in financing constraints. They discover that firms classified as unconstrained, show a greater investment-cash flow sensitivity compared to more constrained ones. Thus, the investment-cash-flow sensitivity does not necessary increase monotonically in constraints. High sensitivity of investment to cash flow cannot be taken as evidence of the financially constrained status. These findings completely contradict with FHZ (1988) and other previous studies. As a conclusion, the authors mention the possibility of considering a non-monotonic relationship between investment-cash flow sensitivities and the degree of financing constraints. The study by KZ (1997) raises a critique paper by FHP (000). The main argument is that the KZ theoretical model and classification approach, determining the degree of financing constraints across companies are non-informative, subjective and non-effective. For example, FHP claim that the level of cash stock, leverage and unused line of credit fails to capture the degree of financing constraints. A subsequent response paper by KZ (000) deepens this debate on the usefulness of investment and cash-flow sensitivities. Nevertheless, the research by KZ (1997) is followed by a number of studies supporting their findings. For example, Cleary (1999) extends their classification scheme based on a firm s ability to raise funds externally. The authors point out that 10

15 . Literature Review the sample applied by KZ (1997) is too small for further subdivision and could be biased as it includes only fairly high quality firms. That is why Cleary (1999) obtain a much larger sample consisting of 1,317 U.S. companies operating in different industries. Then the author divides the sample into groups of firms according to a special multivariate classification index. This index is similar to the Altman s Z score factor for the prediction of bankruptcy. 1 In addition, the author argues that a firm s financial status is altering every year as a response to the changes in economic conditions. Therefore, the firm s financing constraints state should be assigned not for the entire period but for each firm-year observation. The empirical findings support KZ. The investment behavior of the least constrained companies, which enjoy high creditworthiness, is found to be significantly more sensitive to the level of internal funds compared to the most constrained companies with less creditworthiness. Another paper by Allayannis and Mozumdar (004) supports the findings of KZ (1997) and Cleary (1999) by showing that investment activity of financially weaker firms is less sensitive to changes in cash flow. Moreover, this study detects a trend of decreasing investment-cash flow sensitivities over time, especially for the most constrained companies. Moyen (004) presents an explanation of the conflicting results of FHP and KZ. The authors come up with two models, i.e. the unconstrained model and the constrained model based on a firm s ability to raise external funds. In the case when the constrained model criterion based on the Cleary (1999) index is applied the obtained results are similar to those obtained by KZ (1997). The identification of constrained firms according to the amount of paid dividends leads to the findings supporting FHP (1988). To draw a short conclusion, all the presented studies recognize the impact of financing constraints on investment. The common feature of the investment models considering asymmetric information distortion is that a firm s investment expenditures depend on its net worth or liquidity level. These studies are based on the idea that investment volume is monotonically decreasing in financing constraints. The main conflict is on the behavior of investment-cash flow sensitivities in constraints. Challenges in testing investment and financing constraints empirically are attributed to the choice of: 1) a classification scheme based on various a-priori measures 1 For more detail see Altman (1968). Hubbard (1998). 11

16 . Literature Review (constrained vs. non-constrained); ) a proxy for investment opportunities; 3) a proxy for internal funds; 4) a proxy for investment volume...3. Non-monotonicity of Investment Behavior in Financing Constraints The conflicting positions of FHP and KZ on the behavior of investment-cash flow sensitivities in financing constraints create a challenge for academics to come up with a better explanation of the relationship between investment and financing constraints. Recall, that in spite of their different empirical findings on the behavior of investment-cash flow sensitivities, both approaches suggest that investment is decreasing in financing constraints. A recent paper by CPR (007) introduces a new idea that the relationship between investment and internal funds is U-shaped. This study is based on the extension of the papers by Cleary (1999) and Povel and Raith (001, 003). The authors suggest that investment is not monotonically decreasing in financing constraints, as it was detected by previous studies, but experiences a non-monotonic behavior. The study agrees that investment decreases as internal funds become lower. Contrary to earlier studies, the authors believe that at some intermediate point, where internal funds are sufficiently low, investment may start to increase as internal funds decrease further. The authors consider two sources of financing constraints: the level of internal funds and asymmetric information. That is to say, that they look at the financing constraints effect on investment from two different dimensions. CPR provide empirical evidence on these predictions. Their model and empirical findings are described in more detail in Section 3. The study by CPR is not the only one detecting a non-monotonic relationship between investment and internal liquidity. For instance, Bhagat, Moyen and Suh (005) empirically find evidence that for financially distressed firms the relation between investment and internal funds is diverse. In the case when a firm is distressed but generates positive profit the cash flow sensitivity is positive and its investment is similar to that of financially healthy firms. The sensitivity is lower for firms experiencing operating loss. Surprisingly, 40% of distressed firms with operating losses invest more than in the previous year in spite of the low cash flow level. Moreover, they invest even more as cash flow decreases further. These firms experience a negative investment-cash flow sensitivity. The authors find that the 1

17 . Literature Review additional funds for investment are provided by equity holders, what is consistent with a gamble for resurrection. Flor and Hirth (008) extend the CPR model in order to theoretically test whether it is still possible to obtain the U-shaped investment curve if some of the assumptions are changed. Particularly, the authors consider a static one-period model with symmetric information between a firm and its external investor, and where financing is obtained through a standard debt contract. They argue that investment is driven by a trade-off between two types of financing costs: the expected liquidation costs and the second-best investment costs. The authors derive the model which shows that under these assumptions the investment curve is still U-shaped in liquid funds. In more detail, because of the risk of liquidation unconstrained firms decrease their investment volume in financing constraints until some point where internal funds are sufficiently low. At that point it becomes more important for financially constrained firms to reduce underinvestment costs. As a result, they try to get closer to the first-best investment and increase their investment rate. Thus, in that region investment is increasing in financing constraints. Another paper by Guariglia (008) provides an empirical research on the relationship between investment and financing constraints on a large sample for UK firms over the period of The author distinguishes between internal (the level of internal funds available) and external (access to external finance) financial constraints. The conclusion is that the separation of firms according to the level of internal funds (cash flow) leads to the U-shaped relationship between investment and cash flow. These results are consistent with the CPR model. Moreover, the author finds that the investment-cash flow sensitivity responses differently dependent on the type of financial constraints considered. Particularly, the highest sensitivity is detected for the firms facing strong external financing constraints and weak internal constraints. Drakos and Kallandranis (007) find that the relationship between investment and cash flow is non-linear. Its changing effect is dependent on two types of asymmetry: ex ante caused by the state of expectation and ex post due to the state of the business cycle. One of the most recent papers by Hovakimian (009) investigates the determinants of the investment-cash flow sensitivity. In order to eliminate the problem with the ex-ante classification scheme, the author first estimates investment- 13

18 . Literature Review cash flow sensitivity on a firm-level. Then she classifies firms into three groups: firms with high, low and negative sensitivity. Finally, the comparison of each group s financial characteristics is performed. The obtained data for US manufacturing firms show that a large proportion of the observations experiences a strong positive investment-cash flow sensitivity, while a small number of firms demonstrate a strong negative sensitivity. These results provide support to the U-shaped investment curve in liquid funds proposed by CPR. Moreover, the author identifies that the investmentcash flow sensitivity is non-monotonic with respect to financial constraints, liquidity and growth opportunities. Moreover, Hovakimian (009) finds that firms with strong positive sensitivity are more constrained than cash flow insensitive firms. They represent smaller and younger firms with a lower payout ratio and higher levels of financial slack. In spite of the fact that these firms have lower cash flow than cash flow insensitive firms, they issue more new equity and debt, and show higher investment and growth rates. The group of companies with negative sensitivity represents the most constrained companies. They are the smallest and the youngest, pay the lowest dividends and have the highest amount of financial slack. These firms have the lowest and even negative levels of cash flow. But despite all these facts, they demonstrate a very high investment and growth rate financed via a large proportion of issue of new equity and debt. This study provides an explanation of the negative relationship between investment and cash flow through the corporate life cycle hypothesis. In more detail, over their lifetime firms experience changes in growth opportunities which influence investment and cash flow. Particularly, a firm starts operating in the market with good investment perspectives but low earnings. The market seems to perceive the firm s future investment opportunities as very valuable. As a result, the firm is able to raise a sufficient amount of debt and equity. As the firm becomes more mature it starts to experience higher earnings generated from the past investments. The firm s investment opportunities become less attractive, and its investment and growth rate slow down. Another stream in investment literature uses investment-timing models, which suggest that in a dynamic framework investment can be increasing in financing 14

19 . Literature Review constraints. 1 The main distinction of this type of models is that a firm has an option to invest in a project at any time, i.e. it may decide either to invest now or to postpone investment. Recall, that the static models consider now-or-never investment. For instance, Boyle and Guthrie (003) analyze a dynamic relationship between investment and liquidity. The authors suggest that financially constrained firms are more concerned with the risk of facing a funding shortfall in the future. This fact may results either in forgoing investment or even in investing early in a suboptimal project. Therefore, a more severe effect of financing constraints may encourage firms to increase investment today in order to avoid future constraints. Unconstrained firms, at the same time, have more incentives to delay investment in order to choose the best project. Moreover, the authors argue that the investment-cash flow sensitivity can be greater for firms with higher liquidity. Another good example is a recent study by Lyandres (007), where the author finds that the relationship between the cost of external financing and investment is non-monotonic. In particular, investment is increasing in the cost of external finance when the cost is not too high and decreasing as the cost of external financing become high enough. The relation between the sensitivity of investment to cash flow and the cost of external financing is U-shaped. This effect is explained through two main effects. The first one supposes that the cost of external financing makes the investment more costly and reduces its attractiveness. The second assumes that this cost reduces the profitability of future investment project and the value of the option to delay. The author finds empirical support to these predictions. Dasgupta and Sengupta (00) obtain similar results showing a nonmonotonic relationship between investment and internal funds. In addition, dependent on the level of liquidity the investment cash-flow sensitivity can be non-monotonic as well. The authors agree that the sensitivity is higher for firms with liquid funds below some critical level than for those with liquidity above this level. But at the same time, some firms with high liquidity experience a greater sensitivity than firms with lower liquidity. 1 See, for example, Boyle and Guthrie (003), Dasgupta and Sengupta (00), Hirth and Uhrig- Homburg (007), Moyen (004), Lyandres (007). 15

20 3. Methodology 3. Methodology I believe that the paradox detected by CPR demonstrating a relatively high investment activity of companies with sufficiently negative internal liquidity requires further investigation. That is why I aim to replicate and extend their study in order not only to test the predicted U-shaped relationship between investment and internal funds empirically, but also to provide some explanation of the increasing investment volume in financing constraints. Therefore, I choose the paper by CPR as the main framework for this thesis. This section describes in detail the theoretical model, assumptions and empirical analysis The U-shaped Investment Curve: Theoretical Model The model 1 is static. It describes a risk-neutral entrepreneur owned firm considering debt-financed investment. Initially the firm has a choice to invest in a project. Investment is now-ornever and is equal to amount I 0. This investment generates a stochastic revenue F (I, ) after one period. is a random variable, which is distributed on [, ] with density and c.d.f. The partial derivatives F and. The main assumptions are: F I have both positive values. In other words, a higher value of leads to a greater revenue and marginal revenue on I. For that reason, may be considered as the unknown state of demand for the firm s product. For F < 0 expected value E F I I II, has a unique maximum value I at some positive level of I. I represents the first-best investment level. Moreover, the investment project is scalable. That is to say that the firm may choose between smaller and larger investment projects, which are less or more expensive. If the firm does not invest, the revenue is zero or 0, 0 F. If the firm borrows externally, at state there is a positive probability that it will default on the agreed repayments or F I, =0. The time line of the model is depicted in Figure 1 (Appendix A.1). It can be described as follows: 1 Source: CPR (007, pp. 7-17). 16

21 3. Methodology 1) At time period t=0, the firm has some level of internal funds W, which may be negative or positive. The firm is called financially constrained if its internal funds are not sufficient enough to invest in the best investment project or W I. The firm may obtain the missing amount of I W by borrowing from an outside risk-neutral investor. Therefore, there is an opportunity to design a financial contract with the external investor. At this period there is symmetric information between the firm and its investor, as the investor has information on the firm s investment opportunities. The investor, taking into consideration the firm s financial situation and its future investment projects, decides to either accept or reject the contract. Moreover, the investor s requirement to earn a satisfactory return determines the firm s costs of external funds endogenously. ) At the next period (t=1) the firm earns a revenue F I, dependent on the resolution of the uncertainty. The investor can not observe the realized revenue. Thus, there are asymmetries in information between the firm and the outside investor. Note, that the firm has a limited liability, so that its payments to the investor cannot exceed its funds. Asymmetric information and different incentives between the firm and the investor lead to agency problems. As a result of these capital market imperfections, the external investor demands a premium covering his losses. It makes external funds more expensive than internal funds. 3) As the firm receives its revenue, it makes a debt repayment R to the investor. It is specified in the contract, at what conditions the firm is allowed to continue to operate or to be liquidated. This decision is dependent mainly on the firm s debt repayments. The liquidation decision is stochastic. In other words, a probability of liquidation is stated in the contract as a function of the firm s payments. 4) At time period t= there are two possible outcomes. The first possibility is that the investor allows the firm to continue operating. Then the firm earns an extra nontransferable payoff. Otherwise, the investor decides that the firm should be liquidated. Then all its assets are sold for a liquidation value L, which is less than the benefit. This liquidation value may be verified. The liquidation value L presents the market value of the firm s assets. If the firm fails to make its payments, the investor has the right to overtake its assets. For the owner of the firm the value of the assets is equal to. The resulted difference L represents an inefficiency of liquidation for the firm owner. 17

22 3. Methodology Moreover, the authors assume that investment does not entail any fixed costs and it is not possible for the firm to issue risk-free claims to finance investment. In addition, at the stage of financing the firm has no existing debt that is due in time period one or after the revenue is generated from the investment project. The existing debt may be due immediately, right before the firm starts to invest. It may result in a negative value of W. This assumption allows examining of the underinvestment problem which is not due to the debt overhang effect. CPR come up with four propositions derived from the model. The first one is on the optimal debt contract. This is a type of contract which maximizes the firm s payoff and ensures that the firm makes sufficient payments to the investor. Recall, that the firm s revenue is unobservable and the risk of liquidation is required in order to enforce the firm to make its debt repayments. Suppose that the firm s internal funds W are at the lowest level: L W : E L F I, F I, I. (1) Now consider the case when internal funds are insufficient to cover investment in a desirable project. In order to invest I the firm needs to borrow I-W externally, so it will write a financial contract with an investor. It will be agreed to pay back an amount D. If the firm makes a repayment D, it is allowed to continue operating. Otherwise, if it can not repay fully and pays only R<D back, it will default. In this R case there is a probability 1 D R that the firm is allowed to continue, and a probability 1 R of liquidation. Whether the firm will be able to make the promised debt repayment D and avoid liquidation is dependent on the outcome of the state. Thus, the bankruptcy threshold may be described by: D F I, ˆ. () The investor s participation constraint is defined by: 1 ˆ D I W ˆ D F I, F I, L d. (3) The repayment D can not be greater than. The contract is called optimal if it induces the firm to repay either the full amount of D or the whole revenue F. The threat of ineffective liquidation (L< ) enforces the firm to pay its debt back if there are enough funds. Thus, the optimal 18

23 3. Methodology contract causes minimization of the likelihood of executing a liquidation procedure. This leads to a probabilistic liquidation rule 1. The firm will choose the level of investment I and debt repayment D in order to maximize: ˆ F I d F I d, D. (4) ˆ, Taking into consideration the investor s participation constraint (3), equation (4) can be rewritten: F I, D I, W E, (5) where D(I,W) provides a solution for equation (3). The authors argue that (5), (), (3) can be solved and have a unique solution for I and its behavior is U-shaped as a function of W. The second proposition predicts a dependency between internal funds and investment choice. It states: At W I and at W=W, the firm invests in the first-best level I. On the interval W, I, the optimal investment function I(W) is strictly lower than I and has a unique minimum at a negative level of internal funds W ~. CPR plot investment as a function of the firm s internal funds, which is presented in Figure (Appendix A.). The authors believe that the detected relationship is not attributed to the debt overhang effect or credit rationing but is rather caused by the risk of liquidation, which is a part of the debt contract. When the firm chooses its investment, it faces a trade-off between its future earnings and the expected costs of liquidation. The threat of liquidation forces the firm to invest below the first-best level I in order to make lower debt repayments and escape bankruptcy. Thus, the investment is decreasing as internal funds become lower until it reaches some intermediate level of internal fundsw ~, where it can start to increase. The authors suggest that this non-monotonic relationship is caused by the firm s investment scale effect. The larger the investment project is, the more debt and the larger repayments it will require. But on the other hand, the higher the investment scale is, the higher the expected revenue it will generate. This higher revenue will improve the firm s financial situation and provide funds to repay the debt, which is in the firm s and its investor s interest. 1 Povel and Raith (003). Source: CPR (007, p. 10). 19

24 3. Methodology Thus, there are two main effects which determine the firm s optimal scale of investment: 1) the risk of liquidation; ) the expected revenue. The trade-off between these two effects is made continuously as they vary in internal funds W. In particular, if the firm has a relatively high level of W, which is not sufficient to cover the firstbest investment I, it will have to invest in a less expensive project and will borrow less, as it will reduce the risk of liquidation. The gain from avoiding bankruptcy exceeds the loss associated with the decrease in revenue. As internal funds decrease further until some point W ~, the probability of default increases. It becomes not optimal for the firm to decrease investment, as it leads to a greater loss in revenue. At this point the investor can be stimulated to provide additional funds and reduce the required payment D, as he/she receives the whole revenue if the firm cannot meet its obligations. The gain from the higher revenue generation outweighs the costs of liquidation and causes an increase in investment. Thus, the authors suggest that the relationship between investment and internal funds is U-shaped. They describe the cost and revenue effect as: ˆ I d F I, ˆ d 1 0 FI I. (6), ˆ Recall, that the difference between costs of internal and external funds is associated with the risk premium, which the investor demands because of the financial constraints effect. The authors define the risk premium as: I, W I W D i I, W. (7) I W Equation (7) shows that the notion that lower levels of internal funds lead to a higher risk premium is true for their model. The authors come up with the third proposition describing the risk premium. It states, that when W decreases, keeping either investment I or the required fund I-W fixed, the risk premium increases. The authors point out the importance of the distinction between the marginal and the average costs of debt-financed investment. Changes in W make it inappropriate to determine investment by looking at the average cost, as its behavior is monotonic. The marginal cost explains fluctuations in investment better. Finally, the authors try to extend their model by detecting the effect of changes in asymmetric information between the firm and its investor. They presume that the 0

25 3. Methodology future payoff and the liquidation value L are equal to zero with probability α, and to 1 and L/(1-α) with probability 1-α. At time period t=1, when the revenue is earned, the firm gets information about the future payoff. If its value is zero, the firm will be less incentive to repay its debt. Therefore, a higher value of α causes more severe asymmetric information problem. This fact allows considering two identical firms which face different degrees of information asymmetry. Note, that the original model considers the case when α=0. Incorporation of a different degree of asymmetric information into the model does not change the optimal debt contract described earlier. The investor s participation constraint is now as follows: 1 F I, 1 1 ˆ D I 0 ˆ, D F I L ( ) d W The firm s objective to maximize becomes: E F I (1 ) D.,. (8) The authors argue that this extension of the model still leads to the U-shaped investment curve. For high values of W (W I ) the firm continues to invest at the first-best investment I. The only difference is that, as W starts to decrease, there is more uncertainty about the future revenue and it becomes more expensive to borrow externally in comparison to the case of α=0. As a consequence, the left end of the U- curve, for sufficiently negative W, lies to the right compared to the original curve. CPR come up with the fourth proposition on asymmetric information and investment choice. It states: For infinitesimal increases in α, (a) If I 0, then I 0 ; that is, whenever investment is increasing in internal w funds, it is decreasing in the degree of informational asymmetry. (b) For W sufficiently close to I, we have I 0 ; i.e., the sensitivity of investment with respect to the level of internal funds is increasing in α. (c) The risk premium increases for any given I. 1 w The investment curves identified by the original and the extended models are presented in Figure 3 (Appendix A.). It illustrates, that an increase in α, i.e. from 1 Source: CPR (007, p. 16). 1

26 3. Methodology α=0 to α=0.1, changes the amplitude of the U-shaped curve. The new curve is bent downward and inward, while the right end at (W, I) = ( I, I) end remains unchanged. The authors get the same results when they consider the liquidation value L as a proxy for the degree of agency problem. The lower value of L means less capital may be transferred to the investor, what causes greater agency problem. Furthermore, they suggest that the model could be alternatively extended by considering a probability of verifiability of the firm s revenue. The model introduced by CPR predicts that for firms with positive levels of internal funds, a higher degree of asymmetric information leads to a higher sensitivity of investment to fluctuations in internal funds. At the same time, firms with sufficiently negative internal funds experience a greater sensitivity as well, but the correlation between investment and internal funds is negative. 3.. The U-shaped Investment Curve: Empirical Approach CPR empirically test the relationship between investment and internal funds on a large unbalanced data panel consisting of 88,599 observations within the time period The data includes all US firms from the Compustat financial statement database excluding regulated or financial industries. 1 Two proxies for internal funds are considered: cash flow from operations (CF/K) and beginning-ofperiod net liquid assets (NLA/K). Other main variables for the analysis are: a firm s gross investment (I/K), sales growth and market-to-book ratio. In order to detect the relationship between internal funds and investment four different tests are conducted: 1) First, all observation are split into ventiles of internal funds. Then mean and median of investment (I/K) are calculated for each ventile and the obtained results plotted. The detected plots are presented in Figure 4 and Figure 5 (Appendix A.3). Both figures demonstrate the U-shaped relationship between investment and internal funds. Investment is at the lowest level when internal funds are around zero. Investment increases in both cases when internal funds become positively larger and negatively lower. ) In the second test, investment is regressed on the M/B ratio, sales growth internal funds and its square value. All regressions detect small and positive values of coefficients for internal funds that confirm the predicted nonlinearities in the data. 1 SIC codes 43XX, 48XX, 49XX, 6XXX and 9XXX.

27 3. Methodology 3) In the next test, the spline regressions of investment on internal funds are conducted. The sample is split into different quantiles. Then for each quantile investment is predicted as a piecewise linear, continuous function of cash flow or net liquid assets. The obtained results validate the nonlinear relationship between investment and internal funds as well. 4) In the final test, the regression is conducted for two sub-samples separating observations with positive and negative internal funds. Investment is regressed on the M/B ratio, internal funds, and sales growth. For the group with positive observations the obtained coefficients for internal funds are positive, while they are negative for the group with negative observations. These results are consistent with the previous tests and the theoretical predictions. All four tests provide strong empirical support to the theory on the U-shaped relation between investment and internal funds. In addition, CPR try to explain the conflicting findings of FHP and KZ on the behavior of investment-cash flow sensitivities in financing constraints. First, CPR point out that these academics use different sample selection criteria. In particular, the construction of a balanced data panel results in the elimination of financially weaker firms with negative internal funds, which represent a significant part of the economy. Second, CPR argue that the earlier theoretical models are based on either too restrictive or ad hoc assumptions about a firm s investment and its way of financing. In contrary, the U-shaped investment model is founded on more realistic assumptions. These conditions lead to a more complex model, which is able to capture different dimensions of financing constraints and fits the data well. Third, the authors argue, that these conflicting findings are caused by the differences in classification schemes. In more detail, FHP use proxies for the effect of capital market imperfections, while KZ compare traditional financial ratios. CPR suggest that if firms are classified by the degree of capital market imperfections they face and if the financially weakest firms are not included, then the investment-cash flow sensitivity is expected to be higher for more constrained firms. In the second case, when firms are classified according to the level of internal funds, more financially constrained firms experience lower investment-cash flow sensitivity compared to less constrained firms. 3

28 4. Sample Characteristics and Variables Description 4. Sample Characteristics and Variables Description In order to construct a sample the data are extracted from the Compustat North America Xpressfeed database ( It is a data delivery system for Standards & Poor s which provides independent financial information. The Fundamentals Annual files (FUNDA) are used in order to extract 19 data items of interest. Their description is presented in Table 1 (Appendix A.4). The sample is obtained by delimiting the entire database through specifying the initial criteria which are described in Appendix A.5. The extracted data items are then used for calculations of the main variables for the analysis: Investment (I/K) A company s capital expenditures (data item 18) divided by beginning-of-period net fixed assets (lagged data item 8). Cash Flow (CF/K) The sum of income before extraordinary items (data item 18) and depreciation and amortization (data item 14) divided by beginning-of-period net fixed assets (lagged data item 8). Net Liquid Assets (NLA/K) Beginning-of-period total current assets (lagged data item 4) minus beginning-of-period total current liabilities (lagged data item 5) minus beginning-of-period inventories (lagged data item 3). The obtained number is divided by beginning-of-period net fixed assets (lagged data item 8). Beginning-of-period Cash Stock (Cash/K) - Beginning-of-period cash and marketable securities (lagged data item 1) divided by beginning-of-period net fixed assets (lagged data item 8). Market/Book Ratio (M/B) - Beginning-of-period total assets (lagged data item 6) minus beginning-of-period total common equity (lagged data item 60) minus beginning-of-period deferred taxes (lagged data item 74) plus the product of beginning-of-period share price (lagged data 199) and beginning-of-period number of common shares outstanding (lagged data 5). The obtained figure is divided by beginning-of-period total assets (lagged data item 6). Sales Growth Current sales (data item 1) minus sales form the previous period (lagged data 1). The obtained number is then divided by sales from the previous period and multiplied by 100 in order to show the value in percentages. Following the standard approach the first four variables, namely investment, cash flow, net liquid assets and cash stock, are divided by beginning-of-period net 4

29 4. Sample Characteristics and Variables Description fixed assets (K). This procedure controls for a possible effect of heteroskedasticity caused by differences in firm size. Besides these six key variables additional ten variables for the analysis are calculated. The description and interpretation of all the selected variables are presented in Table (Appendix A.6). 1 Due to the limited scope of this project it is impossible to cover all the companies from the Compustat database as it is done by CPR. That is why I decide to decreased the number of companies by focusing only on two major sectors of economy: manufacturing and mining industry sectors The Manufacturing Industry Sector The sample for the manufacturing sector is constructed by including only firms operating within Panel D of the Standard Industry Classification (SIC) system (SIC codes ). This industry sector consists of companies producing new products through activities of mechanical or chemical transformation of materials or substances. The output of this sector is either consumed as end products or as input to other productions. The detailed classification of the manufacturing sector is presented in Panel A of Table 3 (Appendix A.7). After satisfying these criteria the sample consists of 40,356 observations. This number is decreased further by including only companies that are legally registered in the USA (Current ISO Country Code=USA). As the further analysis requires the lagged data terms availability, companies which contain only one year observations are eliminated. Firm-year observations are deleted if one of the following criteria is met: 1) Total assets value (Compustat data item 6) is equal zero. 3) Sales/Turnover volume (data 1) is equal zero ) Missing values (#VALUE! and #DIV/0!) for either of the main variables: I/K, CF/K, NLA/K, Cash/K, M/B ratio and sales growth. 4) Sales growth exceeding 100%. 3 5) Observations lying above the 99th percentile and below the 1st percentile for some of the key variables: I/K, CF/K, NLA/K, Cash/K. This procedure eliminates possible outliers caused by data entry mistakes. 1 These variables are originally selected by CPR and this study continues to work with them. Note that the obtained sample does not include regulated or financial industries (SIC codes 43XX, 48XX, 49XX, 6XXX, and 9XXX). These industries were excluded by CPR. 3 According to Almeida, Campello, and Weisbach (004), this procedure helps to eliminate distorting observations caused by reorganization, mergers and acquisitions and other large corporate events. 5

30 4. Sample Characteristics and Variables Description 6) Observations with missing values for other variables: market-to-book (M/B) ratio, payout ratio, leverage, current ratio, return on equity (ROE) and interest coverage ratio (TIE). 1 After fulfilling these steps the final number of firm-year observations is equal to 19,077 for,084 companies. The obtained sample consists of manufacturing companies of different size, age and financial situation. In measuring financing constraints I analyze firm year-observations, not some average number for a firm within the whole time horizon. This procedure is consistent with Schiantarelli (1996) and Cleary (1999), who argue that a firm s financial status may be changing over time as a result of changing economic conditions. Panel A of Table 3 (Appendix A.7) presents the number of firms and observations within the manufacturing sector by industry. It shows that the largest number of the observations belongs to the production of electronic and other electrical equipment (19.6%), and to chemicals and allied products (16.95%). I do not require firms to have each year-observation within the entire time horizon. According to CPR, the analysis of an unbalanced data panel leads to less biased results, as the elimination of firms with an incomplete number of yearobservations results in excluding financially weak companies. Their presence is essential in order to detect the non-monotonic relationship between investment and internal funds. Nevertheless, in order to relate this study to some previous empirical findings I construct a balanced subsample by deleting firms with missing year-observations within the whole time period. It results in a significantly smaller sample containing 8,596 observations for 66 companies. At this point it is essential to explain the choice of this industry sector and shortly identify its specific characteristics that have to be taken into account. The manufacturing sector is selected because it is the most used in empirical studies on investment behavior. 3 The attractiveness of this sector is attributed to its large size and significant role in the national economy. Moreover, manufacturing industries enjoy 1 This step is not executed by CPR (007) at this stage of analysis. The authors vary the number of observation within different tests. I decide to have the same number of observations for all tests as this procedure should lead to more consistent and compatible outcomes. Many previous studies work with the balanced panel of data, e.g. FHP (1988), KZ (1998), Cleary (1999), Allayannis and Mozumdar (004), etc. 3 See, for example, FHP (1988); KZ (1997); Hubbard, Kashyap and Whited (1995); Almeida, Campello and Weisbach (004), Hovakimian (009). 6

31 4. Sample Characteristics and Variables Description greater data availability compared to other industry sectors. Moreover, the choice of this industry group provides me with an opportunity to compare the obtained results to other existing studies. The US manufacturing sector is the largest in the world, although the manufacturing activity and employment in the USA have declined over the past two decades. At the same time, the productivity of this sector is showing an increasing trend. The output of this sector experiences fluctuations as a result of technological shocks, free trade and other changes in economic and political activities. Figure 6 Appendix A.8 illustrates the sharp decline in the US manufacturing output from 1998 to 004. The output of the manufacturing sector is more unstable compared to the whole economy. The observed pattern of production where it peaks in some years and drops in others suggests that the manufacturing sector was affected by recessions, the terrorist attacks of September 11, and the wars in Afghanistan and Iraq. In the last decades there is a trend in the manufacturing sector where companies are becoming more service-oriented. In particular, firms are getting more involved in such operations as marketing, accounting, maintenance of machinery, logistics, R&D, etc. The manufacturing industry sector is fairly competitive. For example, Abdel- Raouf (009) finds that in % of the US manufacturing industries operate in competitive market, where the rest operate in loose and tight oligopoly, and none in monopoly. The authors detected the trend of increasing competitiveness in this sector over time. One of the main strength of the US manufacturing sector is the fact that it is the largest market for innovative products and services. During the last decades one of the most growing investment activities is in equipment and software. The rapid changes in technologies and global competitive pressure require large capital investments and substantial investments in research and development in this sector. R&D results in innovation which improves the firm s competitive position on the market. At the same time, R&D leads to higher risks and uncertainty. Moreover, because of the competitiveness in the market, manufacturing firms can not reveal the information on new products or services completely to other market participants. It may create asymmetry in information between a firm and its outside investors. 7

32 4. Sample Characteristics and Variables Description 4. The Mining Industry Sector The sample for the mining sector is constructed by setting the SIC code equal to (Panel B). Mining companies are primarily involved in the extraction of natural minerals, exploration and development of mineral properties, and other various mining activities. The division structure of the mining sector is presented in Panel B of Table 3 (Appendix A.7). The obtained 4,758 observations are decreased by deleting non-us companies and companies with only one year-observation. After fulfilling these conditions the sample contains 3,717 observations. These observations are then further truncated by applying exactly the same six steps described in Section 4.1. The final number of the year observations is equal to 1,684 for 3 companies. The obtained sample for the mining sector is much smaller compared to that for the manufacturing industries. Panel B of Table 3 (Appendix A.7) reports the proportion of the firms and observations for each mining industry. A significant fraction of the observations is accounted for oil and gas extraction (80.34%). In order to relate the obtained results to some previous findings, a balanced subsample is constructed. It consists of 853 observations for 66 firms. The US mining firms represent a significant part of the national industry. According to the United States Geological Survey (USGS), in 008 the total output of the US mining sector is USD 71 bn compared to USD 70 bn in Anyway, this sector is not so widely used in investment studies compared to the manufacturing sector because of its smaller size and lack of data availability. The data on the mining sector are usually applied in studies analyzing risk management and hedging strategies. Fluctuations in activity of this sector are attributed to technological changes, globalization pressure and emergence of new markets. The mining sector is highly exposed to instability in commodity prices. Lamont (1997) detects that the 1986 sharp oil price decrease led to a significant drop in investment of oil companies in non-oil subsidiaries compared to other companies that are less dependent on oil. In addition, the mining sector is exposed to economic downturns which subsequently lead to costcutting procedures in the form of reducing production, decreasing the size and scale of investment, postponing some investment projects and even closing some operations. 1 United States Mining Report (009, p.5). United States Mining Report (009). 8

33 4. Sample Characteristics and Variables Description There is a common opinion that this industry sector is highly affected by political instability, bureaucracy, corruption and bribery. Government regulations have a great impact on the demand for output of the mining sector. For example, the Bush administration s desire to limit imported energy led to some changes in the legislation in 005. As a result, more favorable conditions were created for the US producers of oil, coal and gas. There are, of course, other examples, like the government demand for cleaner coal technologies, which led to a decrease in the competitiveness of the US mining industry on the world market. 1 According to the United States Mining Report (009), the US mining sector is competitive. It consists of hundreds of companies of different size. In the last decade, the mining sector is characterized by an increasing consolidation in the market with an active merger and acquisition activity. Growing environmental concerns, especially the climate change, are becoming a significantly more important issue as they cause more drastic standards, higher costs, loss of jobs, and a necessity to outsource some activities abroad. Recently, the mining firms have set more focus on the relocation of exploration, investment into less developed regions and on the greater diversification of their exploration portfolio. At this point it is hard to identify which industry sector is more affected by asymmetric information. On the one hand, the assets of mining companies are expected to be more specialized compared to those of manufacturing firms. Therefore, they cannot serve as collateral for a bank loan in the same way as standard assets, like computers and cars. Moreover, there may be more uncertainty about the liquidation value of assets L. These facts may affect hurdles on external finance and make it more expensive. On the other hand, the manufacturing sector produces heterogeneous products. It is more competitive and more R&D intensive. The profitability of new products and services can be unpredictable, so there is a higher uncertainty about the firm s future payoff. Thus, manufacturing firms may face a more severe effect of asymmetric information between a firm and its outside investors, which can make external funds more costly. This issue will be investigated in the following sections.. 1 United States Mining Report (009). There is also an uncertainty in the future payoff of mining firms. For example, firms involved in the exploration of oil and gas do not always know how big the potential reserves are. I consider that this uncertainty is not under control of mining firms. That is why this uncertainty does not create any significant effect of information asymmetry between a firm and its external investors. 9

34 5. Empirical Analysis 5. Empirical Analysis 5.1. Summary Statistics The empirical analysis is aimed to examine a dependency between a firm s investment activity and internal funds. The level of capital expenditures (I/K) is considered as a proxy for investment I. There are various measures of internal funds (W). According to CPR, none of them represents a perfect solution. That is why a number of proxies for W are included in this study. The first two measures are the same as those used by CPR, namely operating cash flow (CF/K) and beginning-of-period net liquid assets (NLA/K). The additional proxies are: beginning-of-period cash stock (Cash/K); the sum of cash flow and beginning-of-period cash stock (CF/K+Cash/K); and the sum of cash flow and net liquid assets (CF/K+NLA/K). 1 Cash flow (CF/K) is the most used proxy for W in the existing investment literature. It correctly reflects changes in internal funds, but it does not take into account funds available in the beginning of the period. Cash stock (Cash/K) and net liquid assets (NLA/K) are lagged variables. They include funds carried over from the previous fiscal year. Their drawback is that they do not account for current changes in cash flow caused by immediate investment. Besides, these two variables may include funds borrowed externally in the prior period. These recourses may be wrongly considered as internal funds W in the current period. CPR claim that cash stock is not an appropriate measure of W, as it never experiences a negative value and does not fully capture the effect of financial obligations. I include cash stock in order to examine this argument and to investigate the reasons for its inability to detect the U-shaped relationship. In order to overcome the problems with using cash flow, cash stock and net liquid assets, I apply other alternative measures, namely CF/K+Cash/K, and CF/K+NLA/K. As a result, I consider five different measures of internal funds. The intention is to investigate, whether the way of measurement of W affects the predicted 1 CPR also consider the possibility of applying different measures of internal funds, namely CF/K+Cash/K, CF/K+NLA/K, Cash/K. But the authors do not report their findings. It is stated that all of them, except for cash stock, lead to qualitatively similar results. See a review of investment literature in Hubbard (1998). 30

35 5. Empirical Analysis U-shaped pattern of the investment curve and if yes, to identify the factors causing the problem. According to CPR, a separation of observations with positive and negative internal funds reflects a firm s liquidity situation and helps to identify financially weaker firms. Following their approach, I apply the classification scheme according to the level of cash flow (CF/K). This decision is also based on the fact that this proxy is the most used one in the investment literature. 1 Table 4 presents the summary statistics for the unbalanced sample, and for the subsamples with positive and negative levels of cash flow. Moreover, in the last two columns the analysis of the balanced subsample is reported. Recall, that this subsample includes only firms with data available for the entire time period. Table 4: Summary statistics The table illustrates the mean and median values for the selected variables. The unbalanced sample consists of year-observations within the time period , and firms are not required to have data available for each year. The analysis is performed for the unbalanced sample containing all observations, for the subsamples with positive CF/K and negative CF/K. In addition, the balanced subsample is analyzed where only the firms with data available for the entire period of are included. Panel A provides the results for the manufacturing sector, Panel B for the mining sector. The description of the variables is presented in Table, Appendix A.6. Panel A: Manufacturing industry sector Variables description Unbalanced data panel Balanced data panel All observations Negative CF/K Positive CF/K All observations Mean Median Mean Median Mean Median Mean Median Net Fixed assets Total Assets Sales I/K (investment) CF/K (cash flow) NLA/K (net liquid assets) Cash/K (cash stock) CF/K+NLA/K CF/K+Cash/K M/B (market/book ratio) Payout ratio Leverage ROE (return on equity) TIE (interest coverage ratio) Current ratio Sales growth No. of observations 19,077 5,063 14,014 8,596 No. of firms,084 1,443 1, See Hubbard (1998). 31

36 5. Empirical Analysis Panel B: Mining industry sector Variables description Unbalanced data panel Balanced data panel All observations Negative CF/K Positive CF/K All observations Mean Median Mean Median Mean Median Mean Median Net Fixed assets Total Assets Sales I/K (investment) CF/K (cash flow) NLA/K (net liquid assets) Cash/K (cash stock) CF/K+NLA/K CF/K+Cash/K M/B (market/book ratio) Payout ratio Leverage ROE (return on equity) TIE (interest coverage ratio) Current ratio Sales growth No. of observations 1, , No. of firms The analysis is performed on the data for the manufacturing (Panel A) and the mining (Panel B) sectors. Mean and median are calculated for the selected variables described in Section 4. Some of the variables have a large difference between the mean and median values, for instance, net fixed and total assets, sales, and ROE. It may be attributed to the fact that the data on these variables include outliers which have not been eliminated. I do not conduct this procedure in order to obtain consistent and compatible outcomes to CPR. The results from Table 4 are quite similar to those reported by CPR (007, p. 19). It is obvious that on average firms with negative cash flows are less profitable and smaller in size compared to those with positive CF/K. Moreover, firms with negative CF/K experience not only lower sales levels, but also negative sales growth and interest coverage ratio. A higher M/B ratio indicates that these companies may be younger and have great growth opportunities. They are more leveraged and have a greater buffer of cash. At the same time, these companies invest at about the same level as companies with positive CF/K. These findings are consistent with the theoretical predictions on the U-shaped relation between investment and internal 3

37 5. Empirical Analysis funds, where firms with sufficiently low internal funds can experience an increase in investment activity as internal funds decrease further. There are some differences in the results for the manufacturing and the mining industry sectors. First, firms operating in the mining sector are larger in size and have lower sales than manufacturing ones. Second, mining firms always show a negative mean value for return on equity. Third, the level of net liquid assets is higher for manufacturing firms with negative CF/K compared to those with positive CF/K. Within the mining sector firms with negative CF/K experience a lower and even negative level of NLA/K compared to those with positive CF/K. Furthermore, I decide to calculate additional descriptive statistics for sales, I/K and CF/K in order to identify any difference in the degree of asymmetric information for manufacturing and mining firms, which was discussed in Section 4.. The results are presented in Table 5 (Appendix A.9). Sales and cash flows of manufacturing firms are significantly more volatile compared to those of mining firms. At the same time, investment (I/K) is spread out over a similar range of values for both industry sectors. These results are consistent with Figure 6 (Appendix A.8), where the output of the manufacturing sector is detected to be more unstable compared to the whole economy. Thus, I predict that manufacturing companies may face a more severe effect of asymmetric information as there is a higher uncertainty about their expected future payoff. The analysis of the balanced data panel illustrates that an elimination of firms with missing year-observations leads to a much smaller subsample which consists of larger and more financially healthy firms (measured by total assets and sales). Firms in this subsample appear to have higher cash flows and sales growth, and lower M/B ratios. These findings are consistent with the argument by CPR that the sample construction plays a significant role in obtaining reliable empirical results. Setting too strong criteria, as balanced data panel composition, results in the exclusion of financially weaker companies. This may be one of the reasons why the previous empirical studies were not able to detect the non-monotonic relationship between investment and financing constraints. Table 6 (Appendix A.9) provides support to this notion. The obtained results show that the balanced subsample consists of fewer observations with negative internal funds, except for observations with NLA/K. For example, for CF/K the 33

38 5. Empirical Analysis percentage of negative observations drops from 6.54% to 13.54% for the manufacturing sector; and from 19.00% to 11.61% for the mining sector. Note, that Cash/K is never negative. As mentioned earlier, this proxy does not take into account a firm s financial obligations. 1 Although I only focus on two industry sectors compared to a much larger sample used by CPR, it is still possible to obtain a large part of observations with negative levels of internal funds. Its fraction varies between 10% and 40% dependent on the measure of internal funds. Thus, I conclude that the negative value of internal funds is neither attributed to a specific industry nor time period. Furthermore, the indicated large proportion of observations with negative internal funds demonstrates that these firms represent an essential part of the economy and should not be excluded from the analysis. Their elimination may result in a partial view of the actual situation in the market. To draw a conclusion, the obtained data show a large proportion of firms with negative internal funds. Second, the separation of observations with positive and negative cash flows helps to detect a pattern of the financial characteristics of firms, where firms with negative CF/K are smaller in size and have higher M/B ratios. Third, the analysis provides some evidence of the non-monotonic relationship between investment and internal funds for both industry sectors. Companies with negative CF/K appear to invest almost at the same level as those with positive CF/K. 5.. Correlation Analysis In order to further investigate the relationship between investment and internal funds, a correlation analysis is performed. The theoretical prediction according to the U-shaped investment model is that there is a difference in characteristics between the firms with negative and positive CF/K. The correlation coefficients among investment and cash flow for the subsamples with negative and positive CF/K are expected to be higher and more significant compared to those obtained for the entire sample. Moreover, the sign of these coefficients is predicted to be negative/positive for the subsamples with negative/positive CF/K. The former subsample is expected to show negative relationship between I/K and CF/K, while the latter should demonstrate positive. 1 CPR (007, p. 0). The correlation analysis is not considered in the study by CPR. As it is a popular tool in empirical studies, I decide to extend their approach and to include it in the presented analysis. 34

39 5. Empirical Analysis Correlation matrices are computed for the manufacturing and the mining sectors and presented in Table 7 and Table 8 (Appendix A.10), respectively. Each table consists of three panels. Panel A shows the results for the entire sample containing all observations. Panel B and Panel C report results for the subsamples consisting of observations with negative and positive CF/K, respectively. Recall, that a correlation coefficient measures the strength and direction of the linear relationship between two random variables. 1 As the main interest of presented research is to detect the dependency between investment and internal funds, the focus is set on correlations among I/K and other variables. For the entire sample the correlation coefficient among investment and cash flow is and significant for the manufacturing sector, while equal to 0.04 and insignificant for the mining sector. For the subsample with positive CF/K this correlation coefficient equal to 0.43 for the manufacturing sector and 0.9 for the mining one. Both coefficients are significant at 1% level. It indicates a positively increasing linear relationship between I/K and CF/K. In other words, investment is increasing as cash flow increases. It means that a separation of observation with positive levels of cash flow leads to a clearer pattern of the dependency between investment and cash flow. For the subsample with negative CF/K this coefficient is equal to and significant for the manufacturing sector. It demonstrates that there is a decreasing linear relationship between I/K and CF/K. Thus, investment declines in cash flows, as predicted by the model. Notice, that in the case of the mining sector, the coefficient is and insignificant. It indicates that the data for this sector may show weak or no increase in investment when internal funds become sufficiently low. These findings will be considered in the following tests. Moreover, the examination of the correlation coefficients among investment and other variables for the total sample shows a pattern of higher and significant correlations between I/K and other proxies of internal funds (NLA/K, Cash/K, CF/K+Cash/K), M/B ratio, leverage and current ratio. These results suggest that firms tend to change their investment as a response to the changes in liquidity, growth opportunities and debt. 1 Aczel, A.D. (1990, p.435). 35

40 5. Empirical Analysis There are very low and almost always insignificant correlations between I/K and payout ratio, return on equity and interest coverage ratio. Therefore, these financial ratios are expected not to have any relationship with investment volume. Furthermore, payout ratio, ROE and TIE show very low and almost always insignificant correlations with the rest of the variables. These results are observed for both industry sectors. The separation of observations with positive and negative CF/K leads to small deviations from this pattern, i.e. some coefficients increase or decrease in value, and become less or more significant. There is one interesting difference between these two industry groups. The correlation between I/K and sales growth is relatively high (0.) and significant for the manufacturing sector, while it is very low (0.0) and insignificant in the case of the mining sector. It means that the relationship between sales growth and investment is weaker for mining firms. This fact will be considered in the regression analysis in the following sections. The correlation among CF/K and the M/B ratio is positive/negative for the subsample with positive/negative CF/K. Thus, for firms with positive internal funds the relationship between the M/B ratio and CF/K is increasing, while it is decreasing for those with negative CF/K. To conclude, the separation of negative and positive observation of CF/K leads to higher correlation coefficients among I/K and CF/K compared to those obtained for the entire sample. The different sign of the coefficients provide support to the predicted nonlinearities in the data. It is consistent with the U-shaped relation between investment and internal funds. Moreover, it seems that the model fits the data of the manufacturing sector better compared to the mining one. This difference will be under further investigation in the next tests Mean and Median Investment Levels The main drawback of the summary statistics and correlation analysis is their inability to demonstrate the U-shaped relationship between investment and internal funds in a more illustrative way. That is why it is a good idea to plot investment on internal funds. This procedure helps to make the picture more comprehensive. I follow the approach by CPR described in Section 3.. All observations are split into ventiles of internal funds. More precisely, every observation for internal funds is ranked from 1 to n, i.e. each observation value is compared to other values 36

41 5. Empirical Analysis within the sample. Then the sample is divided into 0 equal groups based on the rank value. As a result, each group reflects the level of internal funds available. In particular, the first group has the lowest level of internal funds, while the twentieth group has the highest one. The calculations are performed for all five proxies for internal funds: CF/K, NLA/K, Cash/K, CF/K+Cash/K, CF/K+NLA/K. Afterwards, the mean and median values of I/K are computed for each ventile. Finally, the obtained results are plotted. Figure 7 and Figure 8 illustrate the plots for the main proxies of internal funds, namely CF/K (part (a)) and NLA/K (part (b)), for the manufacturing and mining sectors, respectively. Figure 7: Mean and Median I/K for Ventiles of Internal Funds (CF/K and NLA/K). The Manufacturing Industry Sector This figure illustrates the plots of mean (blue line) and medium (pink line) of I/K on ventiles of internal funds. The data for the entire sample consisting of 19,077 observations for the manufacturing industry sector are used. The computations are made for two main proxies of internal funds, namely CF/K and NLA/K; and the obtained results are plotted in (a) and (b), respectively. The description of the variables is presented in Table, Appendix A.6. (a) 0,7 0,6 0,5 I/K 0,4 0,3 MEAN I/K MEDIAN I/K 0, 0,1 (b) 0, CF/K 0,8 0,7 0,6 I/K 0,5 0,4 0,3 MEAN I/K MEDIAN I/K 0, 0,1 0, NLA/K The results for the manufacturing sector (Figure 7) demonstrate that the relationship between investment and either CF/K or NLA/K is U-shaped. Investment 37

42 5. Empirical Analysis is monotonically increasing in internal funds when they are large. Then at some point when internal funds are around zero, investment starts to increase as internal funds become negatively lower. The depicted plots are similar to those reported by CPR (Figure 4 and Figure 5, Appendix A.3). Recall, that their sample is much larger and covers the earlier time period. Thus, I can conclude that it is still possible to detect the U-shaped investment curve when a smaller sample and a more recent time period are applied. These finding indicate that financing constraints caused by the level of internal funds do have an impact on investment and their relationship is U-shaped. The results are consistent with CPR who detect that financially weaker firms with negative levels of internal funds invest more than financially stronger firms with low (around zero) levels of internal funds. The similarity in patterns obtained for the mean and median values of I/K suggests that these results are not influenced by outliers and that there is consistency in the data. Moreover, the left part of the curves, where the internal funds are negative, consists of five ventiles of CF/K and six ventiles of NLA/K. It indicates that a large part of the observations has negative levels of internal funds. The fact that both plots for CF/K and NLA/K show a similar U-shaped pattern illustrates the ability of both proxies to serve as useful measures of internal funds. The analysis of the results for the mining industry sector (Figure 8) shows some deviations from the predicted pattern. Part (a) of the figure, where the level of CF/K is used as a proxy of internal funds, illustrates a similar U-shaped investment curve. The only difference is that the left branch of the graph, where CF/K is negative, does not show the predicted constantly increasing investment volume as CF/K decreases. There is a distortion from the pattern in one of the ventiles. There are two possible explanations. First, this effect may be caused by outliers present in this ventile. Second, there may be some factors specific to the mining sector limiting investment of financially weak firms. Recall, that the correlation analysis in Section 5. detected a very low and insignificant correlation between investment and cash flow for the subsample with negative CF/K. Part (b) of the figure demonstrates the plot for NLA/K. The right branch of the investment curve is increasing in NLA/K. In the left branch of the curve, when NLA/K are negative, investment remains at the same level instead of showing the predicted decreasing pattern. Thus, the plot obtained for the mining sector for NLA/K 38

43 5. Empirical Analysis does not illustrate a strong U-shaped relationship between investment and net liquid assets. Figure 8: Mean and Median I/K for Ventiles of Internal Funds (CF/K and NLA/K). The Mining Industry Sector This figure illustrates the plots of mean (blue line) and medium (pink line) of I/K on ventiles of internal funds. The data for the entire sample consisting of 1,684 observations for the mining industry sector are used. The computations are made for two main proxies of internal funds, namely CF/K and NLA/K; and the obtained results are plotted in (a) and (b), respectively. The description of the variables is presented in Table, Appendix A.6. (a) 0,6 0,5 0,4 I/K 0,3 0, MEAN I/K MEDIAN I/K 0,1 (b) 0, CF/K 0,6 I/K 0,5 0,4 0,3 MEAN I/K MEDIAN I/K 0, 0,1 0, NLA/K In addition, I decide to plot together the investment curves for both industry sectors. The intention is to investigate if there is any significant difference in the patterns. In order to get a more illustrative picture I cut-off the left end of the curve for the manufacturing sector as its data on CF/K are spread out over a larger value range. 1 Recall, that manufacturing firms experience greater deviations in cash flows which can be attributed to a greater instability of their output compared to the whole 1 In more detail, I assign the maximum value of CF/K in the first ventile to -0 instead of calculated -77,39. This procedure allows me to compare the plots for both sectors in a better way. 39

44 5. Empirical Analysis economy (Figure 6, Appendix A.8). The plots for ventiles of CF/K are presented in Figure 9 (Appendix A.11). It is obvious from the results that manufacturing firms invest less when internal funds are low and around zero compared to mining firms. But when internal funds are positively large and negatively low, investment is higher for the manufacturing sector. The detected difference in amplitudes of the investment curves for two industry sectors indicates that manufacturing and mining firms may face a dissimilar degree of asymmetric information. Taking into consideration Proposition 4 predicting the effect of asymmetric information on investment (Figure 3, Appendix A.) and the results from Section 5.1, it is possible to conclude that manufacturing firms experience a greater effect of asymmetric information compared to mining ones. This effect can be attributed to more risk and uncertainties in operations of this sector caused by, for example, high R&D activity and great fluctuations in demand. Moreover, the fact that manufacturing firms experience much more negative values of CF/K can be related to the presence of the large fixed costs in this sector, as its operations require a great amount of upfront investment without the certainty of outcome. As I decided to consider alternative measures of internal funds, namely Cash/K, CF/K+NLA/K and CF/K+Cash/K, additional calculations and plots are conducted. The results for the manufacturing and mining sectors are presented in Figure 10 and Figure 11 (Appendix A.1), respectively. Part (a) of the figures illustrates the plot of investment on ventiles of Cash/K. The picture looks quite similar for both industry sectors. But there are some differences. In particular, the data for the manufacturing sector show a clearer pattern, while there are some deviations in the pattern for the mining sector. The plot for the mining sample demonstrates a significant difference between the median and mean curves. It may indicate a presence of outliers in the data. Nevertheless, it is obvious that the relationship between investment and Cash/K is monotonically increasing. An increase in cash stock causes an increase in investment activity. It is consistent with an argument by CPR that cash stock is an inappropriate measure of internal funds, as it does not account for a firm s obligations and never experiences negative values. Using cash stock as a proxy of internal funds leads to an inability to detect the U-shaped investment curve. This fact helps to 40

45 5. Empirical Analysis explain findings in the existing empirical literature, where this proxy was applied. 1 Thus, cash stock can be an ambiguous criterion for a classification of constrained firms from non-constrained ones, as its high value may be present in both groups of firms with positively large and negatively low CF/K. These findings are consistent with KZ (1997, p. 11). Parts (b) and (c) of the figures illustrate plots of investment on ventiles of CF/K+NLA/K and CF/K+Cash/K, respectively. Once again the data for the manufacturing industry sector (Figure 10, Appendix A.1) fit the model better and illustrate a strong U-shaped relation between investment and these two proxies of internal funds. The results are consistent with CPR where the authors applying these proxies get qualitatively similar results to those obtained for CF/K and NLA/K. The main problems with the plot for the mining sample are again in deviations in the pattern, and stagnant investment in the left branch of the curve, where a monotonically decreasing relationship was expected. The findings in this subsection demonstrate that the plots for the manufacturing sample are similar to those obtained by CPR and provide strong empirical support to the predicted U-shaped investment curve. The results for the mining sector are less strong. All proxies of internal funds lead to qualitatively similar results, except for cash stock. That is why I decide to work form now on only with two main proxies of internal funds, namely CF/K and NLA/K Standard Regression Analysis In this step of analysis I investigate the relationship between investment and internal funds in more detail by applying a simple regression analysis. As mentioned earlier, investment is affected by the level of internal funds. The expectation is that larger internal finance helps to reduce the cost of external finance when a firm has to borrow in order to invest in a desirable project. CPR argue that not only internal funds have an impact on investment. Other factors have to be considered, like a firm s investment opportunities. For instance, the detected high investment activity for firms with negative levels of cash flow may be due to the fact that these firms are young and have good investment opportunities. Therefore, they invest aggressively and consequently exhaust their internal funds into 1 For example, cash stock is considered by FHP (1988); KZ (1977); Kashyap, Lamont and Stein (1994). See Hubbard (1998). 41

46 5. Empirical Analysis a negative level. To control for the possible effect of investment opportunities, I follow the approach by CPR. Two additional explanatory variables, namely the market-to-book ratio and sales growth, are included in regressions as proxies for investment opportunities. I consider the M/B ratio as a proxy for Tobin s Q. This ratio is a forwardlooking indicator and is expected to reflect a firm s growth options together with the future profitability on the demand for investment. Tobin s Q is one of the most used explanatory variables in this type of regressions in investment literature. 1 Its impact on investment has been proven to be significant in a lot of studies. But, surprisingly, the Q ratio cannot explain investment volume completely as internal funds still matter. The detected coefficients for cash flow are more significant. In recent years there has been a debate in literature on the problems with the measurements of Q. For example, Erickson and Whited (000) believe that the lower than expected explanatory power of Q is attributed to errors in its measurement. The authors suggest that the proper measurement of marginal Q leads to a higher explanatory power of Q in investment-q regressions. Due to the limited scope of the project I do not consider these measurement adjustments of Q. I decide to follow the approach by CPR suggesting that an incorporation of sales growth as second explanatory variable helps to overcome these problems with Q. Table 9 presents the estimated coefficients for regressions. The estimated regression model considers investment (I/K) as a dependent variable. There are four regression equations. In the first two equations, the independent variables are the proxy for internal funds (CF/K or NLA/K), market-to-book ratio and lagged sales growth. The results are reported in columns (1) for CF/K and (3) for NLA/K. Moreover, in order to test Proposition predicting a quasi-convex shape of function W, CPR additionally include the square value of internal funds in the regression model. I follow the same approach and conduct two supplementary equations where investment is regressed on the M/B ratio, sales growth and on either CF/K and ( K CF / ) or NLA/K and ( NLA / K) columns () and (4), respectively.. The estimated coefficients are presented in 1 See, for example, FHP (1988); KZ (1997); Almeida et al. (004) and others. See a discussion on this issue in Erickson and Whited (000). 4

47 5. Empirical Analysis Table 9: Regression Estimates for Sample Containing All Observations This table presents the results for regressions where investment (I/K) is a dependent variable. It is regressed on the market-to-book (M/B) ratio, sales growth, and on either cash-flow (CF/K) (column (1) and ()) and its square ( CF / K ) (column ()), or net liquid assets (NLA/K) (column (3) and (4)) and its square ( NLA / K ) (column (4)). The description of the variables is presented in Table, Appendix A.6. The sample contains all observation over the period Panel A reports the estimated coefficients for the manufacturing industry sector, Panel B - for the mining industry sector. T-statistics are in brackets. *, ** and *** indicate significance at the 1%, 5% and 10% levels, respectively. Panel A: Manufacturing industry sector. (1) Only CF/K CF/K ( CF / K NLA/K ) ( NLA / K M/B ratio ) Sales Growth Constant [-19.16]* [16.74] * [34.70] * 0.33 [93.04] * () CF/K and ( CF / K ) [-7.35] * [4.08] * [16.8] * [33.66] * 0.33 [93.11] * (3) Only NLA/K [47.78] * [4.00] * 0.00 [3.36] * 0.06 [84.4] * (4) NLA/K and ( NLA / K ) 0.00 [3.65] * [-.13] ** 0.01 [4.10] * 0.00 [3.17] * 0.06 [83.76] * Number of observations 19,077 19,077 19,077 19,077 Number of firms,084,084,084,084 R 9.35% 9.43% 17.48% 17.50% Adjusted R 9.34% 9.41% 17.47% 17.49% S.E. of regression Sum squared residuals Panel B: Mining industry sector. CF/K ( CF / K NLA/K ) ( NLA / K M/B ratio ) Sales Growth Constant 0.04 [4.36] * 0.04 [8.01] * [0.75] 0.7 [9.83] * 0.09 [5.43] * [3.59] * 0.05 [8.40] * [0.73] 0.6 [7.6] * [6.57] * 0.04 [8.41] * [0.75] 0.75 [31.35] * [8.18] * [5.3] * 0.0 [7.9] * [0.73] 0.7 [31.0] * Number of observations 1,684 1,684 1,684 1,684 Number of firms R 3.84% 4.58% 5.19% 6.71% Adjusted R 3.67% 4.35% 5.0% 6.49% S.E. of regression Sum squared residuals

48 5. Empirical Analysis The results reported in this table are quite similar to those of CPR. Almost all obtained coefficients are highly significant at the 1% level. Moreover, the coefficients for the manufacturing sample (Panel A) are more significant than those estimated for the mining sample (Panel B). This effect can be attributed to a much larger size of the manufacturing sample. The low values of the estimated coefficients for the proxies for internal funds varying from to 0.9 indicate a weak impact of changes in W on investment within the entire sample. Note, as the predicted relationship between investment and internal funds is non-monotonic, one part of the observations is expected to exhibit a negative dependency, while another part - positive. On average, this relationship is then anticipated to be small around zero. Thus, the obtained results are consistent with the predictions on the U-shaped relationship between investment and internal funds suggested by CPR. Besides, it is worth mentioning the effect of the M/B ratio on investment. Its coefficients vary between and 0.05 and they are always significant across the regressions. The coefficients for sales growth are around zero and significant at the 1% level for the manufacturing sector and insignificant for the mining one. This difference has already been noticed in the correlation analysis in Section 5., where for the entire mining sample the correlations coefficients between investment and sales growth are around zero and insignificant. The findings in CPR demonstrate that the incorporation of the square value of CF/K as an additional explanatory variable leads to positive and significant coefficients for both CF/K and its square. Moreover, the explanatory power of the regression increases. This outcome is consistent with the predicted quasi-convex shape of function W (Proposition ). The authors find that the same procedure for NLA/K yields similar but less robust results. I conduct this procedure and obtain similar conclusions in the case of the mining sector (Panel B). It is clear from the columns () and (4) that an inclusion of the square value of CF/K and NLA/K leads to positive and significant coefficients for both proxies for internal funds and their square. The explanatory power (adjusted R ) has increased from 3.67% to 4.35% and from 5.0% to 6.49% in the case of CF/K and NLA/K, respectively. The results for the manufacturing sector are less strong. The 44

49 5. Empirical Analysis estimated coefficients for ( K CF / ) and ( K NLA / ) are zero and the explanatory power remains at the same level. To draw a short conclusion, the conducted standard regression analysis on the sample for all observations detects low and significant coefficients for both CF/K and NLA/K. In addition, the explanatory power is low. These findings provide support to the predicted nonlinearities in the data. These results are consistent with the U-shaped investment curve proposed by CPR Split-Sample Regressions The standard regression analysis for the entire sample provides insufficient evidence of a U-shaped dependency of investment on the level of internal funds. That is why it is a good idea to group firms according to the level of asymmetric information and agency costs they are facing, then run separate regressions for subsamples and compare the obtained results. This is a traditional procedure in the empirical studies on investment. 1 It was first proposed by FHP (1988). The main idea is to separate financially constrained companies from non-constrained ones, as these two groups experience different financial characteristics and investment behavior. I follow the approach by CPR and divide the sample into observations with negative and positive levels of internal funds. According to CPR, the level of internal funds reflects a firm s financing constraint state. In order to obtain comparability with the previous tests the same explanatory variables are applied in this regression analysis, namely the market-to-book ratio, sales growth, CF/K or NLA/K. The dependent variable is investment (I/K). Table 10 presents the split-sample regressions estimates. Panel A and Panel B report results for the manufacturing and the mining sectors, respectively. The table includes several regressions. Two main proxies for internal funds are considered: CF/K (columns (1)-(3)) and NLA/K (columns (4)-(6)). The first regression is run on the entire sample, where one equation includes CF/K (column (1)) and another NLA/K (column (4)). The results are the same as reported in Table 9 (columns (1) and (3)). Then the sample is split into subsamples. Columns () and (5) show the estimates for the subsamples with positive observations of CF/K and NLA/K, respectively. The results for the subsamples with negative levels of CF/K and NLA/K are displayed in columns (3) and (6), respectively. 1 See a review of investment studies in Schiantarelli (1996). 45

50 5. Empirical Analysis Table 10: Split-Sample Regression Estimates The table reports the results for regressions of investment (I/K) on the market-to-book (M/B) ratio and sales growth and either cash-flow (CF/K) or net liquid assets (NLA/K),. The regression is run on the entire sample in column (1) for CF/K and in column (4) for NLA/K. The sample contains all observation over the period of Additionally, the sample is split into subsamples. The regression analysis is then performed on the subsamples with positive CF/K (column ()) and positive NLA/K (column (5)); negative CF/K (column (3)) and negative NLA/K (column (6)). Description of the variables is presented in Table, Appendix A.6. Panel A and Panel B report the results for the manufacturing and the mining sectors, respectively. T-statistics are in brackets. *, ** and *** indicate significance at the 1%, 5% and 10% levels, respectively. Panel A: Manufacturing industry sector CF/K NLA/K M/B ratio Sales Growth Constant (1) All [-19.16]* [16.74]* [34.70]* 0.33 [93.04]* () Pos. CF/K [43.34]* 0.07 [0.04]* 0.00 [5.46]* 0.13 [37.93]* (3) Neg. CF/K [-0.65]* [4.6]* 0.00 [13.45]* 0.41 [38.89]* (4) All [47.78]* [4.00]* 0.00 [3.36]* 0.06 [84.4]* (5) Pos. NLA/K 0.01 [43.7]* 0.0 [3.83]* [7.81]* [5.64]* (6) Neg. NLA/K [-14.19]* [6.8]* 0.00 [17.86]* [50.06]* No. of observations 19,077 14,014 5,063 19,077 13,148 5,99 No. of firms,084 1,605 1,443,084 1,884 1,49 R 9.35% 5.06% 1.38% 17.48%.11% 8.75% Adjusted R 9.34% 5.05% 1.33% 17.47%.09% 8.71% S.E. of regression Sum squared residuals Panel B: Mining industry sector CF/K NLA/K M/B ratio Sales Growth (1) All 0.04 [4.36]* 0.04 [8.01]* [0.75] () Pos. CF/K [9.64]* [6.80]* [0.6] (3) Neg. CF/K [-0.53] [3.39]* [.10]** (4) All [6.57]* 0.04 [8.41]* [0.75] (5) Pos. NLA/K [7.19]* [4.83]* 0.00 [6.76]* (6) Neg. NLA/K 0.00 [3.16]* 0.03 [5.7]* [0.37] Constant [9.83]* [9.76]* [10.81]* [31.35]* [18.81]* [19.45]* No. of observations 1,684 1, , No. of firms R 3.84% 11.56% 5.87% 5.19% 10.99% 4.55% Adjusted R 3.67% 11.37% 4.98% 5.0% 10.7% 4.15% S.E. of regression Sum squared residuals

51 5. Empirical Analysis The results for the manufacturing industry sector (Panel A) are qualitatively similar to those obtained by CPR. All estimated coefficients are significant at the 1% level. The columns (1)-(3) demonstrate that the CF/K coefficient is positive/negative for the subsample with positive/negative observations of CF/K. Meaning that when a company experiences a positive level of CF/K, an increase in cash flow by one dollar leads to an increase in investment by 9.1 cents, holding all other variables constant. Thus, the dependency of investment on the level of cash flow is positively increasing. When a company has a negative level of CF/K, an increase in cash flow by one dollar causes a decline in investment volume by 1.4 cents. In other words, for these firms investment is decreasing in cash flow. These findings provide strong support to the predicted U-shaped investment function in internal funds. Besides, the explanatory power of the regressions has improved. The adjusted R has increased in the splitsample regressions compared to that for the entire sample. It indicates that the classification firms according to the level of cash flow leads to a better regression fit. The results are almost the same for the mining sector (Panel B, columns (1)- (3)). The only difference is that the estimated coefficients are less significant, what may be caused by a smaller sample size. When NLA/K is included as an explanatory variable (columns (4)-(6)), the obtained results are quite similar but less strong. 1 For both industry sectors there is a problem with the decreasing adjusted R calculated for the subsamples with negative NLA/K. That is to say, that the explanatory power of the regression has decreased while the opposite effect was expected. It means that the separation of observations with negative value of NLA/K does not lead to an improvement of the regression fit. Moreover, the NLA/K coefficient obtained for the subsample with negative NLA/K for the mining sector (Panel B, column (6)) is positive instead of predicted negative. It supports the results from Figure 8 (b) where in the left branch of the curve investment remains stagnant instead of showing a decreasing pattern in internal funds. One possible explanation could be that it is not sufficient to separate observations into two groups. Most of the studies consider more than two subsamples, for example, financially constrained, partially financially constrained and not financially constrained. But in this case with the sample classification according to the level of internal funds it is difficult to objectively identify these groups. Another 1 CPR report the same tendency but do not discuss the results. See, Cleary (1999). 47

52 5. Empirical Analysis explanation is in the drawbacks with the measurement of W through NLA/K discussed in Section 5.1. All in all, it seems that CF/K provides more robust results on the U- shaped investment curve compared to NLA/K. This fact may clarify the extensive use of CF/K as a proxy for the change in net worth in empirical studies on investment behavior. 1 In addition, the figures in Table 10 indicate that the data on the mining industry sector provide less robust results on the U-shaped relationship between investment and internal funds compared to those for the manufacturing sector. Recall, that there are problems with measurement errors of the M/B ratio as a proxy for Tobin s Q discussed in the previous subsection. Erickson and Whited (000) criticize previous empirical studies reporting poor explanatory power of Tobin s Q, and suggest the method of adjusting the proxy of internal funds in order to overcome this problem. Since in this study the obtained coefficients for the M/B ratio are statistically significant at the 1 % level in all regression equations, I decide to follow the argument by CPR and assume that the measurement errors have minimal distortions on the estimates Financing Constraints and Investment-Cash Flow Sensitivities. In addition, I decide to split the samples into five quantiles of internal funds and run the same regressions as in Section 5.5 for each quantile. The intention is to detect the changes of the estimated coefficients of CF/K and NLA/K in internal funds. This procedure separates firms into five equal groups. Firms in the first quantile have the lowest level of CF/K and subsequently are the most constrained, while those in the fifth quantile exhibit the highest CF/K and are the least constrained. The results are reported in Table 11, Appendix A.13. The regression equations include investment as a dependent variable. The market-to-book ratio, sales growth, and either CF/K or NLA/K are independent variables. The analysis is performed for the manufacturing (Panel A) and the mining (Panel B) industry sectors. For both industry sectors the estimated coefficients for CF/K lead to a conclusion of a nonlinear behavior of investment-cash flow sensitivity in financing constraints To get a clearer picture, the CF/K coefficients are plotted on internal funds in Figure 1 (Appendix A.14). When the level of CF/K is sufficiently negative, firms 1 See a review of investment literature in Hubbard (1998). CPR perform split-sample regression analysis for quantiles as well, but the results are neither reported nor discussed in their study. 48

53 5. Empirical Analysis experience the lowest and even negative investment-cash flow sensitivity. Thus, for these firms investment is the least sensitive to fluctuations in internal liquidity. In the case when CF/K is positively low and around zero, the investment-cash flow sensitivity is the highest. Finally, when cash flows are positively large, the sensitivity is low and positive. This figure illustrates an inverted U-shaped relationship between investmentcash flow sensitivities and internal funds. These results are consistent with CPR and other studies detecting the non-monotonic relationship between investment-cash flow sensitivities and financing constraints. 1 In particular, Bhagat, Moyen, and Suh (005) detect a negative and significant investment-cash flow sensitivity for the companies with operating losses. Hovakimian (009) finds that companies classified as negative investment-cash flow sensitive have the lowest and negative CF/K. Moreover, this figure illustrates that investment of mining companies is more sensitive compared to manufacturing companies when CF/K is large and positive. But when CF/K is around zero and decreases further into a negative region, manufacturing firms become more investment-cash flow sensitive. These results support the findings in Section 5.3 detecting a more severe effect of asymmetric information for firms operating within the manufacturing sector compared to mining ones. The same plots for NLA/K lead to the similar conclusion for the manufacturing sample. The results for the mining sector are less strong. Moreover, the CF/K and NLA/K coefficients are insignificant in almost all regressions for this sector. This fact may be caused by a significantly smaller sample size for the mining sector compared to the manufacturing one. It provides additional evidence that the data on the manufacturing sector tend to show stronger results on the U-shaped relationship between investment and internal funds. 1 See, for example, KZ (1997) and Huang (00). The obtained figures of NLA/K coefficients plotted on internal funds (NLA/K) are qualitatively similar to those obtained for CF/K and not reported in this study. 49

54 6. Relation with Previous Investment Studies 6. Relation with Previous Investment Studies The aim of this section is to relate and compare the obtained results to other empirical studies on investment. As it was described in Section. there is a debate in the literature between FHP and KZ on the relationship between investment-cash flow sensitivities and financing constraints. The former study argues that investment of financially constrained firms is more sensitive to changes in cash flow compared to unconstrained firms. The latter research suggests that the most constrained companies experience the lowest investment-cash flow sensitivity compared to less constrained ones. CPR suggest an explanation for these conflicting findings which is based on their different sample construction criteria and classification schemes. I replicate the approach by CPR in order to investigate whether this explanation is convincing Split-Sample Regression for Payout Groups First, I decide to apply methodology, sample characteristics and grouping method which are similar to FHP (1988). FHP classify firms according to the severity of financial constraints they are facing. The financial constraints status is measured by the proxies for asymmetric information. The authors consider a firm s payout ratio and a number of years when it pays dividends, as a measure reflecting the degree of asymmetric information between a firm and its external investor. CPR state that it is difficult to replicate the FHP grouping method exactly, as CPR analyze an unbalanced data panel on a firm-year basis. CPR assume that a firm s financial status may vary over a time period. That is why the authors modify the FHP classification scheme and exclude a number of payout years as a criterion. I follow the same approach. The sample is divided into three groups based on the payout ratio. The first group FHPVary1 is comprised of firm year-observations with payout ratios below 10%. The second group FHPVary includes year-observations with payout ratios between 10% and 0%. Finally, the last group FHPVary3 consists of yearobservations with payout ratios above 0%. As a consequence, FHPVary1 represents the most constrained firms, FHPVary - less constrained, and FHPVary3 - unconstrained. Moreover, CPR consider two additional groups: the FHPVary1strict group, which contains firm-years with strictly positive payout ratios below 10%; and the FHPVary1& group, which includes firm-years with payout ratios below 0%, or in other words, contains both groups FHPVary1 and FHPVary. The FHPVary1Strict group excludes firms with negative or zero payout ratios. Finally, the regression 50

55 6. Relation with Previous Investment Studies analysis is performed on resulted five subsamples. Following the CPR approach, I consider only CF/K as a measure of internal funds. Other explanatory variables are the market-to-book ratio and sales growth. Table 1: Regression Estimates for Payout Groups for the Manufacturing Industry Sector. The table presents the results for regressions of investment (I/K) on cash-flow (CF/K), market-to-book (M/B) ratio and sales growth. The analysis is performed on the data for the manufacturing sector. Panel A displays estimates for the unbalanced sample. Panel B reports results for the balanced data panel requiring firms to have observation within the whole time horizon Panel C shows the regression estimates for the balanced data panel with excluded observations with negative CF/K. The sample is divided into three main payout groups: FHPVary1 (column (1)), FHPVary (column ()) and FHPVary3 (column (3)) with firm-years payout ratios below 10%, between 10% and 0%, above 0%, respectively. Column (4) presents results for the FHPVary1& group, which is a union of the groups FHPVary1 and FHPVary. Column (5) shows estimates for the group FHPVary1strict, which includes only firm-years with strictly positive payout ratios below 10%. T-statistics are in brackets. *, ** and *** indicate significance at the 1%, 5% and 10% levels, respectively. Panel A. Payout Groups. Unbalanced Data Panel (1) () FHPVary1 FHPVary (3) FHPVary3 (4) FHPVary1& (5) FHPVary1 strict [6.50]* 0.06 [6.6]* 0.00 CF/K [-16.3]* [1.57]* [1.65]* [-17.60]* M/B ratio [13.55]* [4.0]* [9.53]* [14.8]* Sales Growth [30.34]* [6.54]* [6.48]* [3.3]* [7.61]* Constant [81.37]* [.07]* [3.85]* [86.98]* [16.19]* No. of observations 14,4 1,913,74 16,335 1,406 No. of firms 1, , R 9.01% 17.41% 14.75% 9.17% 13.83% Adjusted R 8.99% 17.8% 14.66% 9.16% 13.64% S.E. of regression Sum squared residuals Panel B. Payout Groups. Balanced Data Panel (1) () FHPVary1 FHPVary (3) FHPVary3 (4) FHPVary1& (5) FHPVary1 strict [6.35]* [7.64]* 0.00 CF/K [4.80]* [9.61]* 0.07 [14.1]* [5.8]* M/B ratio [0.70]* [3.5]* [4.7]* [3.64]* Sales Growth [15.75]* [5.7]* [3.39]* [17.88]* [6.30]* Constant [39.81]* [18.7]* [31.76]* [44.06]* [11.05]* No. of observations 5,058 1,504,034 6, No. of firms R 14.1% 13.96% 15.5% 14.04% 0.56% Adjusted R 14.07% 13.79% 15.40% 14.01% 0.3% S.E. of regression Sum squared residuals

56 6. Relation with Previous Investment Studies Panel C. Payout groups. Balanced Data Panel. Positive CF/K Firms Only (1) FHPVary1 () FHPVary (3) FHPVary3 (4) FHPVary1& (5) FHPVary1 strict 0.07 [6.96]* CF/K [.70]* [11.31]* [18.01]* [6.66]* M/B ratio [14.3]* [1.49] [.06]** [15.]* [7.31]* Sales Growth [9.68]* [5.09]* [4.9]* [11.81]* [6.41]* Constant [3.04]* [17.61]* [30.10]* [6.86]* [10.00]* No. of observations 3,984 1,466 1,958 5, No. of firms R 5.57% 15.55% 0.6% 5.40% 1.3% Adjusted R 5.51% 15.38% 0.49% 5.36% 1.06% S.E. of regression Sum squared residuals The regression estimates for the manufacturing industry sector are presented in Table 1. Note, that almost all coefficients are significant at the 1% level. Panel A presents the results for the unbalanced data panel. The conclusions are that the investment-cash flow sensitivity is higher for less constrained companies compared to the most constrained ones. The coefficients for FHPVary and FHPVary3 are higher than those for FHPVary1. Thus, the lowest sensitivity is attributed to the lowest payout ratio. These conclusions are quite similar to those obtained by CPR and conflict with FHP (1988). For the balanced subsample (Panel B) the results remain the same: the FHPVary1 group with the lowest payout ratio experiences lower investment-cash flow sensitivities. These findings are consistent with the study by Huang (00). The author obtains a greater CF/K coefficient for firms with high payout ratios compared to those with low payout ratios. In addition, the author attributes the conflicting results of FHP (1988) to the sample-selection biases. Panel C provides support to this argument. It is detected that the exclusion of observations with negative levels of CF/K from the balanced data leads to a significant increase in the CF/K coefficient for the FHPVary1group. The investment-cash flow sensitivity for this group is almost the same as for the FHPVary and FHPVary3 groups. The elimination of observations with zero or negative payout ratios in FHPVary1strict leads to a higher CF/K coefficient in the case of the balanced and unbalanced panel (Panel A and Panel B). Surprisingly, this procedure does not result 5

57 6. Relation with Previous Investment Studies in the predicted increase of the CF/K coefficient for the balanced sample with positive CF/F. Note that applying stronger criteria for sample composition leads to a drop in the significance level of the M/B ratio coefficient. This fact may shed light into the previously mentioned discussions on poor performance of the Q ratio as an explanatory variable in investment regressions. 1 The results for the mining industry sector are reported in Table 13 (Appendix A.15). The table is divided into three panels in the same manner as it is done in Table 1. Surprisingly, in spite of the poorer performance of the mining sector compared to the manufacturing one within the previous analysis, in this test the data set for the mining sample provides stronger results. The estimated CF/K coefficients for the unbalanced data panel (Panel A) lead to the same conclusion that firms with the lowest dividends are the least investmentcash flow sensitive. Stricter sample composition criteria, namely the elimination of firms with missing year-observations or year-observations with negative CF/K, cause a higher/lower CF/K coefficient for the group of firms with the lowest/highest payout ratio. Panel B and Panel C show that the most financially constrained firms with the lowest payout ratio (FHPVary1) are more investment-cash flow sensitive compared to unconstrained firms (FHPVary3). These results look similar to those obtained by FHP (1988). Note that both Table 1 and Table 13 (Appendix A.15) demonstrate that within the unbalanced data panel the highest investment-cash flow sensitivity is detected for the group FHPVary which includes less financially constrained firms. It provides some support to the detected non-linear relationship between investmentcash flow sensitivities and financial constraints presented in Section 5.6. The results for the manufacturing unbalanced sample (Table 1, Panel A) provide some evidence of the U-shaped relationship between investment and cash flow. The investment is decreasing is cash flows for firms with the lowest payout ratio (FHPVary1). In more detail, for this group, representing the most constrained firms, an increase in CF/K by one dollar leads to a drop in investment by dollar. This tendency is not detected when stricter sample construction criteria are applied (Panel B and Panel C). It is neither obtained for the mining industry sector. 1 See, for example, Erickson and Whited (000) and Gomes (001). These studies analyze the problem with low explanatory power of Q and attribute it to errors in its measurement. 53

58 6. Relation with Previous Investment Studies All in all, it is evident that the investment-cash flow sensitivity increases significantly for the group of the most financially constrained firms as the classification according to the payout ratio and stricter sample composition criteria, i.e. the elimination of financially weak firms, are applied. These findings may explain the findings of FHP where a higher sensitivity of investment to changes in cash flow was detected for more constrained firms. 6.. Split-Sample Regression for KZindex Groups CPR replicate the research by Cleary (1999) in order to relate their findings to KZ. The study by Cleary (1999) is close to the KZ approach and strongly supports their results. It is based on separation of observations according to a firm s financial status. Financial status is allowed to change every year and is measured by a Z-score, which is an index predicting if a firm is going to increase or decrease its dividends, and consequently, reflects its financial strength. Due to the data constraints I find it impossible to perform this test in the same way as in CPR. Instead, I employ the KZindex proposed by Lamont, Polk and Saa- Requero (001). The KZindex is based on the regression coefficients obtained by KZ (1997). This index consists of a linear combination of five accounting ratios. The KZindex provides information on the likelihood that a firm faces financing constraints. A higher value of the index indicates that a firm is more financially constrained. In particular, firms with a great KZindex have high debt, low cash, and pay low dividends. Thus, the interpretation of the KZindex is close to the Z-score. This index is applied in several subsequent empirical studies. 1 Detailed information on the KZindex is presented in Table 14 (Appendix A.16). All year-observations are ranked according to the value of the KZindex and formed into three equal groups. Following Almeida et al. (004), the top 33% of all firms ranked on the KZindex are considered as financially constrained (FC) and the bottom 33% as financially not constrained (FNC). I group the rest of firms into the possibly financially constrained group (PFC). Table 15 reports the regression estimates for the resulted three KZindex groups for the manufacturing (Panel A) and the mining (Panel B) industry sectors. As in the previous regressions investment is regressed on the market-to-book ratio, sales growth and either cash flow or net liquid assets. 1 See: Almeida et al. (004), Baker, Stein and Wurgler (003), Hovakimian (009). 54

59 6. Relation with Previous Investment Studies Table 15: Regression Estimates for KZindex Groups The table presents the results for split-sample regressions. All observations are first divided into three equal groups according to the value of the KZindex: financially constrained (FC), possibly financially constrained (PFC), and not financially constrained (NFC). Then investment is regressed on the marketto-book ratio, sales growth and either cash flow or net liquid assets. The analysis is performed for the manufacturing (Panel A) and the mining (Panel B) industry groups. T-statistics are in brackets. *, ** and *** indicate significance at the 1%, 5% and 10% levels, respectively. The description of the KZindex is presented in Table 14 (Appendix A.15). Panel A: Manufacturing industry sector (1) NFC () PFC (3) FC CF/K [-7.0]* [0.13] [-5.51]* NLA/K [35.13]* [8.99]* [9.94]* M/B ratio [4.51]* [17.0]* [.3]* [19.38]* [7.89]* [17.38]* Sales Growth [16.33]* [18.03]* 0.00 [17.73]* [19.7]* 0.00 [16.36]* [1.86]* Constant 0.15 [1.78]* [1.31]* [30.77]* [30.07]* [50.44]* [53.38]* Number of observations 6,359 6,359 6,359 6,359 6,359 6,359 Number of firms 1,30 1,30 1,390 1,390 1,367 1,367 R 15.48% 8.68% 13.59% 14.67% 15.05% 7.78% Adjusted R 15.44% 8.64% 13.55% 14.63% 15.01% 7.74% S.E. of regression Sum squared residuals Panel B: Mining industry sector CF/K [-0.1] [4.78]* [0.01] NLA/K 0.09 [.57]** [1.3] [.97]* M/B ratio [6.7]* 0.08 [6.98]* [4.50]* 0.08 [5.83]* [4.85]* [6.5]* Sales Growth [.58]* [.70]* [0.48] [0.3] 0.00 [4.91]* [4.63]* Constant 0.35 [7.77]* 0.19 [7.85]* 0.17 [4.78]* [7.74]* 0.08 [15.94]* 0.10 [16.8]* Number of observations Number of firms R 8.69% 9.75% 8.5% 5.99% 9.07% 10.40% Adjusted R 8.0% 9.7% 7.76% 5.48% 8.58% 9.9% S.E. of regression Sum squared residuals The obtained results are different from those in CPR, where the authors detect that the investment-cash flow sensitivity is higher for not constrained companies compared to financially constrained ones. The estimated coefficients for CF/K show that investment decreases as cash flow increases for both constrained (FC) and not 55

60 6. Relation with Previous Investment Studies constrained (NFC) firms. 1 Remark, that the investment-cash flow sensitivity is at about the same level for both groups. The results are quite similar between the industry sectors. I alternatively consider NLA/K as a measure of internal funds. The obtained results lead to a conclusion that firms identified as financially constrained (FC) are less investment-cash flow sensitive compared to not constrained firms (NFC). These findings are consistent with KZ. The only difference is that NLA/K is considered as a proxy of W instead of CF/K. Note, that the NLA/K coefficients always show a positive response of investment to changes in net liquid assets. In general, the obtained results are not as clear as expected. Notice, that the presented tests are conducted on exactly the same samples as in the previous tests. It means that different classification schemes and different proxies for internal funds lead to completely dissimilar conclusions. These dissimilarities can be caused by the fact that I sort companies according to another index than CPR and apply other sample construction criteria. 3 In particular, these conflicting results may be attributed to inability of the KZindex to reflect the degree of financial constraints in the same way as the Z-score index. These facts are under further investigation in the following section. Nevertheless, this section provides support for the argument by CPR that the sample composition, different measures of financing constraints and classification scheme do have a considerable influence on empirical findings on investment behavior and have to be taken into account in the analysis. 1 Grouping observations into five quantiles of the KZindex leads to similar results. An exclusion of sales growth as an explanatory variable from the regression equation results in a decrease in the explanatory power of regression and still leads to similar conclusions. That is why I do not report these findings here. KZ (1997) do not consider net liquid assets as a proxy for internal funds. 3 For example, Cleary (1999) delete observations with negative values for sales and the market-to-book ratio, and winsorized observations. I do not apply these criteria as they will eliminate financially weak firms and may bias the sample. 56

61 7. Analysis of Financial Characteristics of Manufacturing and Mining Firms with Respect to the Level of Internal Funds 7. Analysis of Financial Characteristics of Manufacturing and Mining Firms with Respect to the Level of Internal Funds The empirical findings in Section 5 show that investment increases monotonically when internal funds are large. But it can decrease when internal funds are sufficiently low and negative. According to CPR this U-shaped investment pattern is a result of a trade-off between the cost effect and the revenue effect. I postulate that there is a limitation in the study by CPR. In particular, the authors do not explain why financially weak firms with sufficiently negative internal liquidity are able to invest almost at the same high level as financially healthy firms with large liquid funds. It is also puzzling why external investors provide additional funds for investment even if they know that the firm is not profitable at the moment. That is why I decide to extend the approach by CPR and analyze the financial characteristics of firms with respect to the level of internal funds available in more detail. According to CPR, my major theoretical prediction is to identify any difference in characteristics of firms with negatively low, around zero and positively large internal liquidity. In particular, companies with negative internal funds are expected to be highly leveraged. They are supposed to be concerned with the future revenue generation and invest at high rates in order to be able to repay their existing debt. Firms with internal funds around zero are presumed to be influenced by some specific factors forcing them to slow down investment as a response to the risk of inefficient liquidation. Firms with large internal funds are supposed to show characteristics of being financially strong and stable. Many authors consider three or more groups reflecting a firm s financial constraint status. Since it is impossible to distinguish between the most, less and least constrained firms precisely through measuring the firm s internal liquidity, I decide to work from now on with the ventiles of internal funds. Thereby I can better capture the changing effect of financial characteristics with respect to the level of internal funds. I choose cash flow (CF/K) as a proxy for internal funds. First, this measure of internal funds is the most popular in the investment literature. 1 Second, other alternative measures of internal funds in Section 5.3 have led to qualitatively similar results. That is why I assume that the drawbacks with the measurement of internal funds through CF/K are minimal. 1 See a review of investment literature in the study by Hubbard (1998). 57

62 7. Analysis of Financial Characteristics of Manufacturing and Mining Firms with Respect to the Level of Internal Funds All observations for the entire sample are divided into ventiles of CF/K as it is done in Section 5.3. Then mean values for the selected financial characteristics are calculated for each ventile and plotted on internal funds (CF/K). The results for the manufacturing and mining industry sectors are presented in Figure 13, Appendix A.17 and Figure 14, Appendix A.18, respectively. These figures consist of seven parts depicting the mean values of the selected financial characteristics: (a) investment (I/K) and leverage; (b) Cash/K and NLA/K; (c) the market-to-book ratio and current ratio; (d) payout ratio; (e) sales growth and KZindex; (f) return on equity (ROE) and interest coverage ratio (TIE); (g) total assets and sales. 1 It is obvious that leverage (part (a)) is monotonically decreasing in CF/K for both industry sectors. Companies with the highest level of CF/K borrow least. Thus, internal funds are most likely to be the source of financing of their investments. Companies with negatively low CF/K are the most leveraged. Note that this ratio is calculated as a lagged variable, so it does not explain these firms ability to raise additional funds externally and invest at such a high level. The observed pattern of high debt and investment for firms with negative CF/K is in contradiction with the capital structure theories suggesting that debt has a negative effect on growth and investment. This negative effect is a result of agency costs caused by information asymmetries between a firm and its external investors [Myers and Majluf (1984)] and problems with contract enforcement [Jensen and Meckling (1976)]. Larger debt is supposed to lead to a higher debt payment and a subsequent reduction of cash flow. External investors may interpret a large value of existing debt as a signal of potential insolvency and will be reluctant to lend more. As a result, external funds are expected to become more expensive and insufficient, which forces the firm to cut back or delay its investment projects. 3 The pattern for cash holdings (Cash/K) (part (b)) is U-shaped for both industry sectors. For firms with positive CF/K cash stock is increasing in internal funds, while it is decreasing when CF/K is negative. The relatively high level of cash for firms with the highest level of CF/K may be attributed to their high profitability. Companies with the lowest CF/K accumulate the most cash. These findings are consistent with study by Almeida et al. (004), where the authors find that financially constrained 1 Note that these financial characteristics have originally been selected by CPR, and I continue to work with them. Their short description and interpretation are presented in Table (Appendix A.6). See, for example, Lang, Ofek and Stulz (1996). 3 See, for example, Ndikumana (1999). 58

63 7. Analysis of Financial Characteristics of Manufacturing and Mining Firms with Respect to the Level of Internal Funds firms hold far more cash on their balance sheet compared to unconstrained firms. On the one hand, it can be explained by the fact that these firms have to retain some minimum cash balance as collateral for a loan. On the other hand, according to Calomiris, Himmelberg and Watchel (1995) financially weak firms built up cash reserves in order to avoid the risk of being distressed in the future. Moreover, CPR suggest that firms may currently have some part of an unused loan from the previous periods on its bank account. In general, large cash holdings may be the source of financing of new projects and explain the observed high investment rate. These findings are consistent with the argument by CPR that cash stock is an inappropriate measure of financing constraints. Furthermore, it sheds some light on the conflicting findings in the investment literature discussed by KZ (1997, p.11). They argue that classification criteria in most of the studies are theoretically ambiguous. Considering the level of cash stock, some authors, like Kashyap, Lamont and Stein (1994), classify firms with very high cash holdings as relatively unconstrained. At the same time in other papers, for example in Calomiris et al. (1995), it is assumed that firms with very a high cash stock are relatively constrained. For the manufacturing industry sector the pattern obtained for net liquid assets (NLA/K) (part (b)) is U-shaped. Unexpectedly, the same figure for the mining sector resembles a monotonically increasing relationship between cash flow and net liquid assets. Recall, that the plot of I/K on NLA/K for the mining sector (Figure 8 (b)) does not provide strong evidence for the U-shaped investment curve. The pattern is U-shaped in the case when the market-to-book and current ratios (part (c)) are under investigation. The picture looks similar for both industry sectors. The lowest M/B ratio is detected for firms with cash flows around zero. Firms with positive CF/K experience an increase in the M/B ratio as internal funds become larger. At the same time, firms with the lowest CF/K have the most attractive investment and growth opportunities. The fact that these firms are highly leveraged may indicate that external investors are willing to provide them additional funds as a result of a high future profit anticipation. These findings are consistent with Hovakimian (009), who finds that firms classified as negatively investment-cash flow sensitive exhibit the highest M/B ratio. The analysis of the current ratio leads to the conclusion that firms with negative CF/K experience an increase in this ratio as internal funds become lower. These findings indicate that firms with the lowest CF/K are financially healthy in the 59

64 7. Analysis of Financial Characteristics of Manufacturing and Mining Firms with Respect to the Level of Internal Funds short-term. They are able to repay their short-term debt in spite of the fact of being currently non profitable. At the same time, firms with CF/K around zero have problems with their short-term liquidity. Notice that the market-to-book ratio, current ratio and cash stock are following a U-shaped pattern, which is similar to the investment curve. It means that these variables have influence on a firm s investment activity and this relationship is expected to be monotonically increasing, i.e. an increase in these variables leads to greater investment. These findings are supported by the correlation analysis in Section 5., where positive and significant correlations between these variables and investment are identified. The picture obtained for the payout ratio (part (d)) is interesting. For the manufacturing sector the payout ratio is low and around zero within the whole sample. Notice some deviations in the pattern with a sharp decline and increase in two ventiles. These distortions are caused by outliers in the data, which have not been eliminated in order to obtain consistent and compatible outcomes to CPR. The same plot for the mining sector illustrates a non-monotonic relationship between cash flow and dividends. Firms with positive CF/K around zero pay the highest dividends. A higher level of cash flow leads to a decrease in a firm s payout policy. Finally, firms with negative CF/K pay the lowest dividends. It can explain the fact that the splitsample regressions in Section 6.1 show stronger results for the mining sector compared to those obtained for the manufacturing sector. All in all, it seems that the payout ratio does not have any significant influence on investment activity. Recall from Section 5., that correlations between these two variables are zero and insignificant. These findings support Schaller (1993) criticizing the ability of classification based on the payout ratio to capture a firm s financing constraint status. 1 Sales growth (part (e)) is monotonically increasing in internal funds for both industry sectors. The figure for the mining sector shows a sharp decline in sales growth in one of the ventiles. Companies with negative CF/K experience the lowest and negative sales growth, while those with positive CF/K show higher sales growth as internal funds increase. 1 In addition, Schiantarelli (1996) argues that classification based on a firm s dividend payout behavior is less accurate for the later years compared to the earlier ones. There is an outlier in the data. In order to obtain consistency with CPR extreme observations for sales growth are not eliminated in this study. This procedure is conducted only for the key variables. 60

65 7. Analysis of Financial Characteristics of Manufacturing and Mining Firms with Respect to the Level of Internal Funds The value of the KZindex (part (e)) is monotonically decreasing in internal funds. In the figure for the manufacturing sector there is a deviation from the pattern in one of the ventiles, which is caused by an outlier. 1 In general, the pattern leads to the conclusion that firms with the highest level of CF/K are the strongest financially. Firms with negatively low CF/K are the most financially distressed. The rest of the companies with intermediate levels of CF/K around zero are less constrained. These facts provide support to the argument by CPR that grouping of firms based the level of cash flow reflects a firm s financial situation and the severity of financial constraints it is facing. The pattern detecting relationship between return on equity (part (f)) and CF/K is different between two industry sectors. Within the manufacturing sample ROE is the highest for companies with the lowest CF/K. Surprisingly, within the mining sector ROE is the highest for firms with low internal funds around zero. Note that there is an outlier in one of the ventiles, which causes a deviation from the pattern. All in all, ROE seems to have no significant influence on a firm s investment policy. These results are confirmed by the correlation analysis in Section 5., where correlations between ROE and either investment or cash flow are around zero and insignificant. The results for interest coverage ratio (TIE) (part (f)) are dissimilar between the two industry sectors. Nevertheless, they lead to the conclusion, that firms with negative CF/K have the lowest and even negative TIE. It means that in the previous period these firms earned less than their interest payments. The fact that these firms are operating in the current period indicates that there were either other sources of funds for debt repayments, like new equity or debt, or renegotiations in the existing debt contract. These findings are consistent with the CPR model where an external investor can be stimulated to provide additional funds to a firm and reduce the required debt payments, as he/she receives the whole revenue in the case when the firm cannot meet its obligations Part (g) illustrates the plots of total assets and sales on internal funds. The patterns look similar for both industry sectors. The value of total assets or its natural logarithm is usually considered as a proxy for firm size. Thus, companies with 1 Note that in order to obtain consistency with CPR, extreme observations for the KZindex are not eliminated in this study. This procedure is conducted only for the key variables. Note that in order to obtain consistency with CPR, extreme observations for return on equity (ROE) are not eliminated in this study. 61

66 7. Analysis of Financial Characteristics of Manufacturing and Mining Firms with Respect to the Level of Internal Funds negative CF/K are the smallest within the sample and have a sales level around zero. Firms with low and positive CF/K around zero are the largest and have the highest sales level. 1 Firms with large internal funds which are classified as unconstrained are medium in size and with average sales within the sample. These results support the findings of Hovakimian (009), where firms with negative internal funds are found to be the smallest in size. To summarize the obtained results, I would like to distinguish three groups of firms: with negatively low, low around zero, and positively high internal funds (CF/K). For both industry sectors firms with negatively low internal funds appear to be the smallest in size and the most constrained (in terms of the KZindex). They are highly leveraged, with a large amount of cash stock, and pay the lowest dividends. Their sales are around zero and have decreased compared to the previous period. At the same time, these firms have great investment opportunities and invest at high rates. It is worth mentioning, that their high M/B ratio indicates that these companies may be overvalued in the market. Recall from Figure 1, Appendix A.14 that these firms experience a very low and negative investment-cash flow sensitivity. Another group of companies with low internal funds around zero are less constrained. These firms are the largest, with the highest sales and positive sales growth. Their leverage is sharply decreasing in internal funds. They invest least and their cash holdings are the lowest within the sample. These firms have the lowest M/B ratio, which means that their investment opportunities are either poor or not recognized by the capital markets. According to Figure 1, Appendix A. 14 these firms are the most investment-cash flow sensitive. Finally, as internal funds become larger firms start to accumulate more cash and increase investment. These firms with positively high CF/K are financially healthy and the least constrained. They are medium-sized; the least leveraged, and have good investment opportunities. They exhibit positive sales and the highest sales growth. According to Figure 1, Appendix A.14 the sensitivity of investment of these firms to the level of cash flow is positive and low. The obtained results are consistent 1 Here I would like to note that my results on the U-shaped investment curve can be driven by the fact that the detected low values of CF/K and I/K of these firms are caused by the high values of net fixed asses (K). Recall, that investment and cash flow are always standardized by the firm s net fixed capital in order to control for possible heteroskedasticity due to differences in firm size. I recommend considering other types of normalization of these variables in future research. 6

67 7. Analysis of Financial Characteristics of Manufacturing and Mining Firms with Respect to the Level of Internal Funds with Hovakimian (009), where similar financial characteristics for the groups of firms with negative, low and high investment-cash flow sensitivity were detected. Coming back to the study by CPR, these findings provide support to the theoretical predictions drawn earlier. On the one hand, firms with negatively low internal funds invest at high rate as they are highly leveraged, and earnings from future large-scale investment projects may ensure the required repayments of debt. On the other hand, firms with positively low internal funds slow down their investment activity as a response to the risk of liquidation. These firms have the lowest cash stock and they are experiencing short-term liquidity problems. As their assets are very valuable and sales are high, it is not efficient for them to become liquidated. These facts may enforce them to scale down or postpone their investment projects until more favourable investment periods. The main drawback of this explanation is that the effects of a firm s investment opportunities, size and age are not considered in the model. An alternative explanation of these results may be based on the corporate life cycle hypothesis discussed by Hovakimian (009). According to this hypothesis firms become listed as they are small and young. They have great investment and growth opportunities but very low earnings. The market perceives their investment opportunities as very high and is willing to provide sufficient additional financing through debt and equity. These firms are then exploiting their investment opportunities and invest actively. It takes time before the past investment projects become profitable and generate sufficient revenue. As these firms mature, their cash flows increase. Their investment and growth rate slow down as they face fewer investment opportunities. Thus, a firm s investment activity and cash flow follow a shifting trend as a response to changes in investment and growth opportunities within its lifetime. These findings explain the fact that firms with negative liquidity which are the most financially constrained are able to raise funds for investment externally. It seems that the market has great expectations for their future earnings. Hovakimian (009) confirms this notion as she finds that firms with negative internal funds issue large amount of new debt and equity to finance their projects. The author believes these results imply that the efficiency of capital markets has increased. The trend detected by Allayannis and Mozumdar (004) that investment-cash flow sensitivities decrease over time for the most constrained firms supports this notion. 63

68 8. Conclusion 8. Conclusion The purpose of this thesis is to empirically test the model introduced by CPR (007) which predicts a U-shaped behavior of investment in internal funds. The obtained results provide strong support of the drawn theoretical predictions. It is identified that investment decreases as internal fund become lower until some intermediate point, where internal funds are around zero. Thereafter, investment starts to increase as internal funds decrease further. The detected pattern is qualitatively similar for different measures of internal funds 1, except for cash stock. The findings are stronger for the sample for the manufacturing sector compared to those obtained for the mining one. One of the possible explanations is that the former sample is significantly larger. Another reason is that the mining industries are less competitive. The outcome of this sector is highly influenced by the government and fluctuations in commodity prices. These facts may force mining companies with negative internal liquidity to be less aggressive in their investment policy so that in non-profitable years they most likely scale-down or delay projects until the next period. Moreover, it is identified that manufacturing companies face a more severe effect of asymmetric information caused by greater uncertainty about their future payoff. Manufacturing firms invest less when internal funds are low and around zero compared to mining firms. But when internal funds are positively large and negatively low, investment is higher for the manufacturing sector. The detected investment-cash flow sensitivities are non-linear in internal funds. Investment of firms with positively low internal liquidity around zero is found to be the most sensitive to fluctuations in cash flow. Firms with positively large CF/K are less sensitive. Finally, when the level of CF/K is negatively low, firms experience the lowest and even negative investment-cash flow sensitivity. It is further identified that sample composition, different measures of financing constraints and classification scheme may have a significant impact on the empirical results. These findings shed some light on the contradictory findings of FHP and KZ. 1 CF/K, NLA/K, CF/K+NLA/K and CF/K+Cash/K. For example, Lamont (1997) detects that the 1986 sharp oil price decrease led to a significant drop in investment of oil companies in non-oil subsidiaries compared to other companies that are less dependent on oil. 64

69 8. Conclusion Finally, the analysis the financial characteristics of firms with respect to the level of internal funds is performed. The intention is to explain the paradox of the detected high investment rate of companies with negative internal liquidity. The results show that these firms are the smallest and the financially weakest within the sample. They are the most leveraged, have a large amount of cash and the most attractive investment opportunities. On the one hand, these findings support the argument by CPR that these companies enlarge the scale of their investment as a response to the need for higher earnings generation in order to cover the required debt repayments in the next period. On the other hand, taking into consideration their small size and high M/B ratio, the results are consistent with the corporate life cycle hypothesis [Hovakimian (009)], where small and young companies with great investment opportunities but low earnings invest and grow at high rates. External investors value their investment opportunities high and are willing to provide additional funds for investment. The fact that these firms are able to raise sufficient funds externally may imply that the efficiency of capital markets has increased. 1 There are a number of limitations in the presented research. One of them is its inability to explain exactly where firms with negatively low levels of internal funds raise additional funds for investment. Moreover, why are external investors willing to provide financing even though they know that these firms are currently nonprofitable? For this reason further research is desirable. There are a number of opportunities to improve the presented research. For example, one can consider other industry sectors. In particular, I recommend extracting more data on sources and degree of asymmetric information that firms are facing. Alternatively, an analysis of different countries or regions could be made, as there is an expectation that markets vary in the severity of external finance constraints due to a dissimilar stage of economic development and the type of legal system. Future research may also consider the effect of monetary policy (changes in interest rate) on a firm s investment. The analysis of firm financial characteristics could be extended by including other variables like, a firm s bond and investment grade rating, ownership structure, costs of external finance, financial slack and age. Besides, other proxies for 1 See Allayannis and Mozumdar (004). 65

70 8. Conclusion investment activity could be incorporated into the analysis, for example, investment in inventories or research and development. In addition, I recommend a case study analysis. It may provide more detailed information on the detected high investment rate of firms with negative internal liquidity. In order to investigate the sources of additional external financing for investment I recommend the incorporation into analysis not only the current level of leverage, but also issues of new debt and equity. This information will provide an opportunity to check the gamble for resurrection hypothesis which suggests that equity holders of financially distressed firms choose to continue operations when liquidation would have been optimal. 1 I also find it interesting to replicate the presented analysis on a more recent time period in order to investigate the current ability of companies with negative internal funds to borrow additional funds externally. In particular, has the Global Financial Crisis of made a significant impact on a firm s investment activity? Recall, that this crisis has dramatically influenced credit markets and the banking system. As a result, banks have tightened their lending standards. They have become more reluctant to take high risk and to lend funds. Moreover, the distinction between a long-term and short-term leverage could be made. Alternatively, future research should look at the performance of firms with negative internal funds within some time horizon. The question is whether these firms decrease or sustain their investment rate in the following periods. Is it an overinvestment problem or a successful strategy? In particular, is this high level of investment activity followed by positive or negative sales growth? Does it cause any changes in the firm s financial situation in the next periods? These questions still remain unanswered and could provide a basis for future research. 1 De camps and Faure-Grimaud (000), Bhagat, Moyen, and Suh (005). See Taylor (010). 66

71 9. Bibliography 9. Bibliography Abdel-Raouf F. 009, How Competitive is the US Manufacturing Sector? Eastern Economic Journal, vol. 35, issue: 1, pp: Akerlof, G. A. 1970, The Market for "Lemons": Quality Uncertainty and the Market Mechanism, Quarterly Journal of Economics, vol. 84, issue: 3, pp: Aczel, A.D. 1990, Complete Business Statistics, 3rd edition, Irwin. Alexander, D; Britton, A. & Jorissen. 005, International Financial Reporting and Analysis, nd edition, Thomson Learning. Allayannis, G. & Mozumdar, A. 004, The impact of negative cash flow and influential observations on investment-cash flow sensitivity estimates, Journal of Banking & Finance, vol. 8, issue: 5, pp: Almeida, H.; Campello, M. & Weisbach, M. S. 004, The Cash Flow Sensitivity of Cash, Journal of Finance Year, vol. 59, issue: 4, pp: Altman, E. I. 1968, Financial Ratios, Discriminant Analysis and the Prediction of Corporate Bankruptcy, Journal of Finance, vol. 3, issue: 4, pp: Baker, M.; Stein, J. C. & Wurgler, J. 003, When Does the Market Matter? Stock Prices and the Investment of Equity-Dependent Firms, The Quarterly Journal of Economics, vol. 118, issue: 3, pp: Berk, J & DeMarzo, P. 007, Corporate Finance, International Edition, Pearson Education. Bernanke, B.; Bohn, H. & Reiss, P. C. 1988, Alternative Non-Nested Specification Tests of Time-Series Investment Models, Journal of Econometrics, vol. 37, issue: 3, pp: Bernanke, B. & Gertler, M. 1989, Agency Costs, Net Worth, and Business Fluctuations, The American Economic Review, vol. 79, issue: 1, pp: Bernanke, B. & Gertler, M. 1990, Financial Fragility and Economic Performance, The Quarterly Journal of Economics, vol. 105, issue: 1, pp: Bhagat, S.; Moyen, N. & Suh, I. 005, Investment and internal funds of distressed firms, Journal of Corporate Finance, vol. 11, issue: 3, pp: Bischoff, C. W. 1971, Business Investment in the 1970s: A Comparison of Models, Brookings Papers on Economic Activity, issue: 1, pp: Bond, S. & Meghir, C. 1994, Dynamic Investment Models and the Firm's Financial Policy, The Review of Economic Studies, vol. 61, issue:, pp: Boyle, G. W. & Guthrie, G. A. 003, Investment, Uncertainty, and Liquidity, Journal of Finance, vol. 58, issue: 5, pp: Calomiris, C. W; Himmelberg, C. P & Wachtel, P Commercial paper, corporate finance, and the business cycle: A microeconomic perspective, Carnegie - Rochester Conference Series on Public Policy, vol. 4, pp: Chenery, H. B. 195, Overcapacity and the Acceleration Principle, Econometrica, vol. 0, issue: 1, pp: 1-8. Clark J. M. 1917, Business Acceleration and the Law of Demand: A Technical Factor in Economic Cycles, The Journal of Political Economy, vol. 5, issue: 3, pp: Clark, P. K. 1979, Investment in the 1970s: Theory, Performance, and Prediction, Brookings Papers on Economic Activity, issue: 1, pp: Cleary, S. 1999, The Relationship between Firm Investment and Financial Status, Journal of Finance, vol. 54, issue:, pp:

72 9. Bibliography Cleary, S.; Povel, P. & Raith, M. 007, The U-Shaped Investment Curve: Theory and Evidence, Journal of Financial and Quantitative Analysis, vol. 4, No. 1, pp: Dasgupta, S. & Sengupta, K. 00, Financial Constraints, Investment and Capital Structure: Implications from a Multi-period Model, Working Paper, Hong Kong University of Science and Technology, De camps, J.-P., Faure-Grimaud, A., 000, Bankruptcy costs, ex post renegotiation and gambling for resurrection, Finance, vol. 1, issue:, pp: Drakos, K. & Kallandranis, C. 007, Investment and cash flow: evidence for asymmetries in European manufacturing, Applied Financial Economics, vol. 17, issue: 14, pp: Elliott, J. W. 1973, Theories of Corporate Investment Behavior Revisited, American Economic Review, vol. 63, issue: 1, pp: Erickson, T. & Whited, T. M. 000, Measurement Error and the Relationship between Investment and "q", Journal of Political Economy, vol. 108, issue: 5, pp: Fazzari, S. M.; Hubbard, R. G. & Petersen, B. 1988, Financing Constraints and Corporate Investment, Brookings Papers on Economic Activity, issue: 1, pp: Fazzari, S. M.; Hubbard R. G. & Petersen, B. C. 000, Investment-Cash Flow Sensitivities are Useful: A Comment on Kaplan and Zingales, Quarterly Journal of Economics, vol. 115, issue:, pp: Fisher, I The Theory of Interest. Reprint, New York: Augustus Kelly, Flor, C. R. & Hirth S. 008, Financing Investment: The Cost Trade-Off. Working Paper. August 14. Forbes, K. 004, U.S. Manufacturing: Challenges and Recommendations, Business Economics, vol. 39, issue: 3, pp: Gomes, J. F. 001, Financing Investment, American Economic Review, vol. 91, issue: 5, pp: Guariglia, A. 008, Internal financial constraints, external financial constraints, and investment choice: Evidence from a panel of UK firms, Journal of Banking & Finance, vol. 3, issue: 9, pp: Hall, R. E. & Jorgenson, D. W. 1967, Tax Policy and Investment Behavior, American Economic Review, vol. 57, issue: 3, pp: Hillier, D.; Grinblatt, M. & Titman, S. 008, Financial Markets and Corporate Strategy. European Edition. MvGraw-Hill Education. Hirth, S. & Uhrig-Homburg, M. 007, Investment Timing, Liquidity, and Agency Costs of Debt, 0th Australasian Finance & Banking Conference Paper, Aarhus School of Business and University of Karlsruhe (TH), November 18, Hoshi, T; Kashyap, A. & Scharfstein, D. 1991, Corporate Structure, Liquidity, and Investment: Evidence from Japanese Industrial Groups, The Quarterly Journal of Economics, vol. 106, issue: 1, pp: Hovakimian, G. 009, Determinants of Investment Cash Flow Sensitivity, Financial Management, vol. 38, issue: 1, pp: Huang, Z. 00, Financial Constraints and Investment-Cash Flow Sensitivity, Working Paper. EFA 00 Berlin Meetings Presented Paper. pp Hubbard, R. G. 1998, Capital-Market Imperfections and Investment, Journal of Economic Literature, vol. 36, issue: 1, pp:

73 9. Bibliography Hubbard, R. G.; Kashyap, A. K. & Whited, T. M. 1995, Internal Finance and Firm Investment, Journal of Money, Credit and Banking, vol. 7, issue: 3, pp: Jensen, M. C. & Meckling, W. H. 1976, Theory of the firm: Managerial behavior, agency costs and ownership structure, Journal of Financial Economics, vol. 3, issue: 4, pp: Jensen, M. C. 1986, Agency Costs of Free Cash Flow, Corporate Finance, and Takeovers, The American Economic Review, vol. 76, issue:, pp: Jorgenson, D. W. & Siebert, C. D A Comparison of Alternative Theories of Corporate Investment Behavior, American Economic Review, vol. 58, issue: 4 pp: Kaplan, S. N. & Zingales, L. 1997, Do investment-cash flow sensitivities provide useful measures of financing constraints?, The Quarterly Journal of Economics, vol. 11, issue: 1, pp: Kaplan, S. N. & Zingales, L. 000, Investment-Cash Flow Sensitivities are not Valid Measures of Financing Constraints, Quarterly Journal of Economics, vol. 115, issue:, pp: Kashyap, A. K.; Lamont, O. A. & Stein, J. C. 1994, Credit Conditions and the Cyclical Behavior of Inventories, The Quarterly Journal of Economics, vol. 109, issue: 3, pp: Lamont, O. 1997, Cash Flow and Investment: Evidence from Internal Capital Markets, Journal of Finance, vol. 5, issue: 1, pp: Lamont, O; Polk, C. & Saá-Requejo, J. 001, Financial constraints and stock returns. Review of Financial Studies, vol. 14, issue:, pp Lang, L.; Ofek, E. & Stulz, R. M. 1996, Leverage, investment, and firm growth, Journal of Financial Economics, vol. 40, issue: 1, pp: 3-9. Lyandres, E. 007, Costly external financing, investment timing, and investmentcash flow sensitivity, Journal of Corporate Finance, vol. 13, issue: 5, pp: Meyer, J.R. & Kuh, E. 1957, The investment decision: An Empirical study. Harvard University Press. Modigliani, F. & Miller, M. H. 1958, The Cost of Capital, Corporation Finance and the Theory of Investment, The American Economic Review, vol. 48, issue: 3, pp: Modigliani, F. & Miller, M. H. 1963, Corporate Income Taxes and the Cost of Capital: A Correction, American Economic Review, vol. 53, issue: 3, pp: Moyen, N. 004, Investment-Cash Flow Sensitivities: Constrained versus Unconstrained Firms, Journal of Finance, vol. 59, issue: 5, pp: Myers, S. C. & Majluf, N. S. 1984, Corporate Financing and Investment Decisions When Firms Have Information That Investors Do Not Have, Journal of Financial Economics, vol. 13, issue:, pp: Ndikumana, L Debt service, financing constraints, and fixed investment: evidence from panel data. Journal of Post Keynesian Economics, vol. 1, issue: 3, pp: Neslihan, O. 00, Effects of Financial Constraints on Research and Development Investment: An Empirical Investigation, Applied Financial Economics, vol: 1, issue: 11, pp: Povel, P. & Raith, M. 001, Optimal Investment under Financial Constraints: The Roles of Internal Funds and Asymmetric Information, 69

74 9. Bibliography AFA 00 Atlanta Meetings; Institute of Financial Studies, Working Paper, No. 0103, October, Povel, P. & Raith, M. 003, Optimal Debt with Unobservable Investments. Working Paper, January, Schaller, H. 1993, Asymmetric Information, Liquidity Constraints, and Canadian Investment, Canadian Journal of Economics, vol. 6, issue: 3, pp: Schiantarelli, F. 1996, Financial constraints and investment: methodological issues and international evidence, Oxford Review of Economic Policy, vol. 1, issue:, pp: Taylor, M. P. 010, 'The Global Financial Crisis: Introduction and Overview', Applied Financial Economics, vol. 0, issue: 1-, pp: Whited, T. M. 199, Debt, Liquidity Constraints, and Corporate Investment: Evidence from Panel Data, Journal of Finance, vol. 47, issue: 4, pp: United States Mining Report 009. August 009, Business Monitor International, pp Web Pages: United States Department of Labor. Occupational Safety & Health Administration Compustat North America 70

75 10. Appendices 10. Appendices Appendix A.1... III Figure 1: Time Line of the Model... III Appendix A... IV Figure : Investment as a Function of Internal Funds... IV Figure 3: Investment I(W) with a Different Degree of Asymmetric Information. IV Appendix A.3... V Figure 4: Mean and Median I/K for Ventiles of CF/K...V Figure 5: Mean and Median I/K for Ventiles of NLA/K...V Appendix A.4... VI Table 1: Data Items Description... VI Appendix A.5... VII Initial Criteria for the Sample Construction... VII Appendix A.6... VIII Table : Variables Description...VIII Appendix A.7... X Table 3: Standard Industrial Classification (SIC) Division Structure...X Appendix A.8... XII Figure 6: The US Manufacturing Industry Sector Output ( )... XII Appendix A.9... XIII Table 5: Descriptive Statistics for Sales, Investment and Cash Flows...XIII Table 6: Data Composition...XIV Appendix A XV Table 7: Correlations among Variables. The Manufacturing Industry Sector...XV Table 8: Correlations among Variables. The Mining Industry Sector...XVIII Appendix A.11...XXI Figure 9: Mean I/K for Ventiles of CF/K for the Manufacturing and Mining Industry Sectors...XXI Appendix A.1... XXII Figure 10: Mean and Median I/K for Ventiles of Internal Funds (measured by Cash/K. CF/K+NLA/K and CF/K+Cash/K). The Manufacturing Industry Sector... XXII Figure 11: Mean and Median I/K for Ventiles of Internal Funds (measured by Cash/K. CF/K+NLA/K and CF/K+Cash/K). The Mining Industry Sector...XXIII Appendix A.13...XXIV Table 11: Split-Sample Regression Estimates for Quantiles of Internal Fund (CF/K and NLA/K)...XXIV Appendix A.14...XXVI Figure 1: Investment-Cash Flow Sensitivity in Quantiles of Internal Funds. XXVI Appendix A XXVII Table 13: Regression Estimates for Payout Groups for Mining Industry Sector... XXVII Appendix A.16...XXIX Table 14: Ordered Logit from Kaplan and Zingales...XXIX Appendix A XXX

76 10. Appendices Figure 13: Firm Financial Characteristics in Ventiles of Internal Funds. The Manufacturing Industry Sector... XXX Appendix A XXXII Figure 14: Firm Financial Characteristics in Ventiles of Internal Funds. The Mining Industry Sector... XXXII

77 10. Appendices Appendix A.1 Figure 1: Time Line of the Model This figure illustrates the timing of the investment model introduced by CPR (007). It is a static model which describes a risk-neutral entrepreneur owned firm considering debt-financed investment. There are three time periods. Time period: 0 A firm has: - internal funds W; - new investment I (W< I ); - new debt raised W-I. makes investment I=W+D 1 - stochastic revenue F, I ; - required debt repayment D; - realised debt repayment R; - demand for the firm s product Realized state of uncertainty: ˆ F<D R<D partial recovery of debt F D no liquidation full recovery of debt the firm is liquidated with probability 1 R for a liquidation value of assets L the firm is allowed to continue with probability R 1 D R and earns an extra payoff, where >L the firm is allowed to continue and earns a pay-off III

78 10. Appendices Appendix A. Figure : Investment as a Function of Internal Funds This solid line displays the relationship between internal funds W and scalable investment I. Negative value of W represents financially weak firms facing severe effect of financing constraints, or if there are large fixed costs. Very high value of W gives the firm the opportunity to invest in the first-best investment I. When W starts to decrease, the costs of external funds become more expensive and there will be a higher risk of default. It leads to a decrease in investment volume. For sufficiently negative W, revenue generation effect becomes more important in order to repay the debt and escape liquidation. It causes increase in investment as W decreases further. Figure 3: Investment I(W) with a Different Degree of Asymmetric Information. This figure illustrates the effect of asymmetric information on scalable investment I. The curve marked α=0 is the original curve from Figure 1. The curve labeled α=0.1 represents investment of the firm facing more degree of information asymmetry. For low and positive levels of W, more asymmetric information leads to lower investment. While for negative W, investment starts to increase as W decrease as a result of the revenue generation concern. Source: CPR (007, p.11, p.17). IV

79 10. Appendices Appendix A.3 Figure 4: Mean and Median I/K for Ventiles of CF/K The Figure illustrates the mean (thick line) and median (thin line) values of investment for each ventile of CF/K. The horizontal axis represents CF/K, as a proxy for internal funds. The vertical axis corresponds to the level of investment (I/K). As it was predicted by the theory, both plots detect the U- shaped dependency between CF/K and I/K. Figure 5: Mean and Median I/K for Ventiles of NLA/K In this figure the mean (thick line) and median (thin line) values of I/K are plotted for each ventile of NLA/K. The horizontal axis represents NLA/K, as a proxy for internal funds. The vertical axis corresponds to the level of investment (I/K). Both plots show the U-shaped relation between NLA/K and I/K, which is consistent with previous figure and theoretical predictions. Source: CPR (007, p.1, p.). V

80 10. Appendices Appendix A.4 Table 1: Data Items Description This table illustrates the list of the data items extracted from the Compustat North America database. The label and short name for each data item are presented. 1 Data number Label Short name Data1 Cash and Short Term investment (MM US$) CHE Data3 Inventories Total (MM US$) INVT Data4 Current Assets Total (MM US$) ACT Data5 Current Liabilities Total (MM US$) LCT Data6 Assets Total (MM US$) AT Data8 Property, Plant & Equipment Net (MM US$) PPENT Data9 Long-Term Debt - Total (MM US$) DLTT Data1 Sales Net (MM US$) SALE Data14 Depreciation and Amortization (MM US$) DP Data15 Interest Expense (MM US$) XINT Data18 Income Before Extraordinary Items (MM US$) IB Data1 Dividends - Common (MM US$) DVC Data5 Common Shares Outstanding (MM) CSHO Data34 Debt in Current Liabilities (MM US$) DLC Data60 Common Equity - Total (MM US$) CEQ Data74 Deferred Taxes (Balance Sheet) (MM US$) TXDB Data18 Capital Expenditures (SCF) (MM US$) CAPX Data178 Operating Income After Depreciation (MM US$) OIADP Data199 Price -- Fiscal Year -- Close (US$&c) PRCC_F Source: Compustat North America, as of August 13, Remark, that in summer 007 the Compustat adopted a new Xpressfeed format. The new version of the Compustat resulted in changes of the old FTP version. Data items are not longer described by the number but by their short name. VI

81 10. Appendices Appendix A.5 Initial Criteria for the Sample Construction 1. The time horizon covers the period of 19 years: from January 1990 till April 008. The observations from 1990 are only used to construct lagged variables.. The Level of Consolidation (CONSOL) is set on the Consolidated Level (CONSOL=C), so that the parent and subsidiary accounts are reported combined. 3. Industry Format (INDFMT) which indicates a general industry presentation is chosen to truncate the sample to only the Industrial sector (INDFMT=INDL). 4. Data Format (DATAFMT) which designates how the data are collected and presented is selected to report in the Standardized format (DATAFMT=STD). This format represents data with standardized restated interim (quarterly /annual/semiannual). 5. Population source (POPSRC) indicating the source of the data is set as Domestic (POPSRC=D). 6. ISO Currency Code (CURCD) which describes the reported currency of a company s financial statement is chosen as USD dollar (CURCD=USD). Consequently, all figures are expressed in the same currency. It provides consistency and comparability of the results within the obtained sample. 7. Active/Inactive Status Marker (COSTAT) is set to include only active companies (COSTAT=A) with the purpose of eliminating inactive companies on Compustat database from the sample. VII

82 10. Appendices Appendix A.6 Table : Variables Description This table describes sixteen variables selected for the analysis. It includes the name of the variable, its formula for calculating, short description and interpretation. Name of the variable Net Fixed assets (K), (MM$) Total Assets, (MM$) Formula data8 data6 Description Property, plant and equipment (Net). Tangible items that are held for use in the production or supply of goods/services, for rental or other administrative purposes. Expected to be used during more than one period (taxable year). Assets (Total). Economic resources owned by a firm. The value of total assets or its natural logarithm is usually considered as a proxy for the firm s size. Sales, (MM$) data 1 Sales/Turnover (Net). Money received from the sale of goods/services. Investment (I/K), (MM$) Cash flow (CF/K), (MM$) Net liquid assets (NLA/K), (MM$) Cash stock (Cash/K), (MM$) CF/K+NLA/K, (MM$) Data18/L.data8 (data14+data18)/l.data8 (L.data4-L.data5-L.data3)/ L.data8 L.data1/L.data8 CF/K+NLA/K Capital expenditures or money spent on acquisition or upgrade of net fixed assets scaled by beginning-of-period net fixed assets The sum of income before extraordinary items and depreciation and amortization divided by beginning-of-period net fixed assets. Reflects cash inflows and outflows from operating activities. A measure of liquidity and solvency of a company. Measures how much of a firm s liquid assets would be left if all current liabilities were paid off. Measures the ability of a company to invest in a new project without external financing. The sum of cash at bank and in hand, and short term investment scaled by beginning-of period net fixed assets. Measures a firm s ability to meet its short-term obligations. The sum of cash flow and net fixed assets scaled by beginning-of period net fixed assets. An alternative measure of a firm s liquidity. VIII

83 10. Appendices CF/K+Cash/K, (MM$) Market-to-book ratio (M/B), (MM$) CF/K+Cash/K (L.data6-L.data60- L.data74+L.data199xL.data5)/ L.data6 The sum of cash flow and cash stock scaled by beginning-of period net fixed assets. An alternative measure of a firm s liquidity. A proxy for Tobin s Q ratio. Market value of a firm s assets divided by current cost of replacing the firm s assets or the total value of assets. Provides a measure of stock valuation. High/low value of the ratio suggests that a firm is over- /undervalued. Moreover, it provides an idea about the firm s growth opportunities. Considered to be a driving factor for investment decision. Payout ratio L.data1/L.data178 Measures the proportion of operating income after depreciation paid in dividends. More mature and financially stable companies tend to have high payout ratio. Young and growth firms tend to have low payout ratio around zero. Leverage (L.data9+L.data34)/L.data6 Debt-to-Assets ratio. Compares the value of debt to the value of a firm s total assets. High value indicates that a firm is highly leveraged. It implies that the firm is more risky. Return on equity (ROE), (%) Interest coverage ratio (TIE) L.data18/L.data60x100 L.data178/L.data15 Income before extraordinary items divided by equity. Provides information on a firm s profitability by measuring how much profit a firm generates compared to the money shareholders invested. A high ROE indicates a firm s ability to find profitable investment opportunities. Operating income after depreciation divided by a firm s interest expenses. High ratio indicates that a firm is earning more than it is necessary to meet its interest payments. Current ratio L.data4/L.data5 The relation between current assets and current liabilities. Measures short-term financial status or liquidity situation of a firm. The higher the ratio, the more resources a firm has to repay its short-term debt obligations. Sales growth, (%) (data1-l.data1)/ L.data1*100 Sources: CPR (007, p.19); Alexander, Britton and Jorissen (005); Berk and DeMarzo (007). Percentage changes in sales in current period compared to the previous period. Indicated how fast a firm is growing. IX

84 10. Appendices Appendix A.7 Table 3: Standard Industrial Classification (SIC) Division Structure This table presents the Standard Industrial Classification (SIC) Division Structure description. This classification system is applied by the United States government in order to distinguish every industry by denoting it a four-digit code. Panel A describes the composition of the manufacturing industry sector. Moreover, the number of companies and observations, and percentage of the total observations for each industry within the obtained sample is calculated. Panel B presents the same information for the mining industry sector. Panel A: Manufacturing industry sector Industry group number Industry group name Number of companies Number of observations Percentage of total observations Major Group 0XX Food and Kindred 103 1, % Products Major Group 1XX Tobacco products % Major Group XX Textile Mill Products % Major Group 3XX Apparel and Other % Finished Products Made From Fabrics And Similar Materials Major Group 4XX Lumber and Wood % Products, Except Furniture Major Group 5XX Furniture and Fixtures % Major Group 6XX Paper and Allied Products % Major Group 7XX Printing, Publishing, and Allied Industries % Major Group 8XX Chemicals and Allied 445 3, % Products Major Group 9XX Petroleum Refining and % Related Industries Major Group 30XX Rubber and Miscellaneous Plastic Products % Major Group 31XX Leather and Leather % Products Major Group 3XX Stone, Clay, Glass, and % Concrete Products Major Group 33XX Primary Metal Industries % Major Group 34XX Fabricated Metal Products, % Except Machinery and Transportation Equipment Major Group 35XX Industrial and Commercial Machinery and Computer Equipment 64,46 1.7% Major Group 36XX Electronic and Other 395 3, % Electrical Equipment and Components, Except Computer Equipment Major Group 37XX Transportation Equipment % X

85 10. Appendices Major Group 38XX Measuring, Analyzing, and Controlling Instruments; Photographic, Medical and Optical Goods; Watches and Clocks Major Group 39XX Miscellaneous Manufacturing Industries Panel B: Mining industry sector 98, % % Major Group 10XX Metal Mining % Major Group 1XX Coal Mining % Major Group 13XX Oil and Gas Extraction 188 1, % Major Group 14XX Mining and Quarrying of Nonmetallic Minerals, Except Fuels % Source: United States Department of Labor. Occupational Safety & Health Administration, as of August 13, 009. XI

86 10. Appendices Appendix A.8 Figure 6: The US Manufacturing Industry Sector Output ( ) This figure illustrates the decreasing trend of the US manufacturing industry sector output over the time period from 1998 to 004. In order to relate it to the whole national economy, the US real GDP is plotted as well. It is obvious that the activity in the manufacturing sector has declined. There are more fluctuations in output of the manufacturing sector compared to the whole economy. Source: Forbes (004, p. 31). XII