The U-shaped Investment Curve: Age Perspective and Lifecycle Analysis

Transcription

1 The U-shaped Investment Curve: Age Perspective and Lifecycle Analysis Authors: Supervisor: Stefan Hirth December 2010 Aarhus

2 Abstract This paper investigates relations between investment and internal funds from age and lifecycle perspectives. The analysis is performed on the sample of U.S. manufacturing companies within the time frame This paper confirms the U-shaped relationship between investment and internal funds captured by Cleary, Povel and Raith (2007). However, splitting the companies into five age groups provides evidence that the U-shaped relationship does not hold for old companies. This paper also confirms that investment is, in fact, dependent on age regardless of specifications of the model. At last, this paper finds some evidence of the existence of lifecycle of a firm. However, this evidence is rather weak. 2

3 Acknowledgements We express sincere acknowledgements to our supervisor, Stefan Hirth, for his help and support during our research. We also are thankful to the WRDS Representatives of Aarhus University, Carsten Tanggaard and Henning Bunzel for providing us with the access to the WRDS database. We would like to thank our families and friends for supporting and encouraging us to pursue this degree. 3

4 Table of Contents 1 Introduction Literature Review Overview of Investment Theories Investment Behaviour and Financial Constraints Inefficient Markets Investment and Financial Constraints Monotonic Evidence Investment and Financial Constraints Non-monotonic Evidence Lifecycle Theories Methodology Theoretical Framework Lifecycle Model Data Empirical Results Descriptive Statistics Correlation Analysis Graphic Based Analysis Regression Analysis Split Sample Regression Analysis Analysis of Key Financial Ratios Discussions Conclusions Works Cited Appendices

5 1 Introduction Relations between investment and cash flow have been widely discussed for more than 20 years already. Fazzari, Hubbard, and Petersen (1988) were pioneers in examining internal and external investment financing. They found out that firms, which do not experience shortages of liquidity, are less sensitive to changes in levels of internal funds, compared to financially constrained companies. The authors concluded that liquidity is an important determinant of investment behaviour of a firm, if a firm experiences asymmetric information problems. Other researchers (Hoshi, Kashyap, & Scharfstein, 1991; Schaller, 1993, among others) found similar results. The paper by Kaplan and Zingales (1997), in turn, criticized Fazzari, Hubbard, and Petersen (1988) conclusions. Kaplan and Zingales (1997) stated that Fazzari, Hubbard and Petersen wrongly interpreted their own results, and that there is no strong theoretical base for their conclusions. Kaplan and Zingales (1997) make revolutionary conclusions about the fact that financially stable firms experience higher sensitivity to changes in internal funds. Their findings provoked extensive debates, which are still discussed by scholars in different studies that analyse relations between investment and internal funds. Generally acknowledged conclusions still showed the monotonic function between internal funds and investment, meaning that the more constrained is a firm, the less it tends to invest. The paper by Cleary, Povel and Raith (2007), however, showed that a firm s investment is a U-shaped function of its internal funds. It means that until certain low levels of internal funds, the relationship between investment and cash flow is positive and monotonic. However, when internal funds reach some particular low-level point, this relationship turns around, and a further decrease in cash flow leads to an increase of investment. These findings were also confirmed by the recent paper by Yildiz (2010). This thesis continues to study the model introduced by Cleary et al. (2007), and further tested by Yildiz (2010). We introduce a variable age into the analysis, in order to investigate the U-shaped investment curve from the age and lifecycle perspective. Several papers studying the influence of age (or its proxies) were conducted by e.g. Fazzari, Hubbard, and Petersen (1988), Schaller (1993), Hovakimian (2009). However, there was no research on age and lifecycle perspective of the U-shaped relations 5

6 between investment and internal funds. This is why we think that it is important and interesting to analyse this topic. We identified the following aims and objectives of our research: 1. To see whether there exists the U-shaped investment curve in different stages (age groups) of the development of a firm. 2. To find out whether in situations of financial constraints older firms experience weaker effect of changes in internal funds on investment, as compared to younger firms. 3. To see whether age has an influence on the level of investment of a firm. 4. To find out whether the age groups differ in financial characteristics. As a result we will see if there is any evidence of existence of lifecycle of a firm in our analysis. The paper is organized as follows. Section 2 provides extensive review of literature on investment theories, relationships between investment and internal funds, as well age and lifecycle theories. Literature review is followed by the methodology section (Section 3). In this section we present the theoretical framework, on which we base our analysis and the model we are using in our research. Section 4 describes the data used in the analysis, the criteria for data extraction and the reasons for using our particular data sample. In Section 5, we describe the variables using descriptive statistics and correlation analysis. In addition, this section provides graphic based analysis of data, empirical results from conducted econometric modelling, as well as analyses financial characteristics of different age groups. The next section includes discussions of the obtained results. In addition, a comparison with previous studies is also provided in this section. The last section concludes. 2 Literature Review In order to analyse the U-shaped investment curve in the aspect of lifecycle approach, it is essential to discuss the foundation of the theories of investment, research about relations between investment and internal funds, and different lifecycle techniques used by scholars in previous studies. Making preliminary overview of investment literature we found out that there is no common consent in conclusions about relationships between investment and internal funds. Therefore, we think that it is very important to show and understand all major findings of research on investment and internal funds. 6

7 2.1 Overview of Investment Theories The foundation of modern investment theories is built essentially upon the Irving Fisher ideas, which he formulated in his paper The Theory of Interest (1930), published 80 years ago, namely Fisher Separation Theorem. Fisher argues that individual participants perceptions of investment opportunities and their opportunities to borrow and lend interact with their initial endowments of income and personal utility functions (time preferences), providing Pareto-optimal equilibrium market prices, i.e. interest rates (Lintner, 1967). Investment decision of a firm, though, is viewed as an intertemporal problem and is based on the preferences of capital markets, and is independent of individual participants preferences and financing decisions. Moreover, Fisher states that the value of capital is the present value of the flow of income that an asset generates. And as we know, this is, in fact, the way how economists and scholars think about capital and income today. In 1936, John Maynard Keynes formulated his General Theory. In his paper, Keynes provides a new view on the idea of investment. He believes that investment decisions are the driving force of aggregate demand and short-run variations in economic activity, rather than its importance is purely the result of long-run effect on capital stock growth. Keynes (1936) also indicates new fundamental determinants of investment, namely uncertainty, finance, and monetary factors, hence not only technological conditions of capital productivity. Virtually at the same time Michal Kalecki (1937) presented his theory of investment behaviour of a firm. Similarly to Keynes (1936), he stresses that financial conditions of a firm and effective demand are the fundamental determinants of investment. Keynes (1936) and Kalecki (1937) argue that investment decisions are a combination of both, demand expectations of a firm (given its existing capacity), and its ability to generate investment funds through internal cash flow and external debt financing. The theory of Keynes (1936) and Kalecki (1937) has been further extended by Meyer and Kuh (1957), Duesenberry (1958), Minsky (1975), Steindl (1976), and others. It was tested in several comparative studies of investment theories by e.g. Jorgenson (1971), Bischoff (1971), Elliott (1973), Clark (1979), among others. The theory, however, has not fared well, and was subsequently eclipsed by neoclassical models or models based on Tobin's Q theory of investment. 7

8 Modigliani and Miller in 1958 (further developed in 1963) presented their capital structure (leverage) theorem, which stated that the market value of corporation is independent of its capital structure, given the assumptions of perfect capital markets (PCM). The assumptions, on which the theory is constructed, are very strict: Equal access to information by all market participants. Completely rational economic actors, hence no arbitrage opportunities. No transaction costs. PCM rarely, if ever, exist in the real world, thus the conclusions of the theorem that a firm s market value and investment decisions are independent from the level of internal cash flow, leverage or dividend payout policy hold only in theory. The Modigliani-Miller theorem is necessary for the validity of the neoclassical theory of investment, which states that marginal rate of investment is equated with interest rate, and that a firm s cost of capital is independent of its capital structure. The neoclassical theory of investment has been tested and discussed massively. Jorgenson and Siebert (1968) tested the theory (actually it was Jorgenson s representation of the neoclassical theory) against different alternatives, and found that the neoclassical theory performed best. However, other scholars, who tried to replicate the results of Jorgenson and Siebert, were unsuccessful (see Eisner & Nadiri, 1970; and Chamberlain & Gordon, 1989). Chamberlain and Gordon (1989) also found that the model by Jorgenson and Siebert (1968) misrepresented several variables (among them liquidity variable), which biased the conclusions. They also argue that investment models, which are based on profitability variables, such as e.g. Tobin s Q, present more accurate results than the neoclassical model. None of the above-mentioned models hold sufficiently in the real world. A variety of empirical tests show different results. Thus, there appears to be a room for a lot of improvements in this field of research. Financial literature incorporates a number of alternative investment models. Several econometric analyses were conducted (see e.g. Kuh 1963; Jorgenson & Siebert 1968; Clark, 1979; Bernanke et al. 1988) in order to find out which model can explain investment behaviour of firms the best. The most successful and widely used models, which can be distinguished from others, are: Accelerator model (Generalized Accelerator Model) the model assumes that the desired amount of capital at any time is proportional to the level of output at that time, correspondingly. This means that the desired amount of investments is 8

9 proportional to the first difference of output (see e.g. Clark, 1917; Chenery, 1952; Koyck, 1954). Cash Flow model the model basically adds profit (cash flow) term to the accelerator model equation. Cash flow model justifies the addition of the term by employing the ideas of Klein (1950), who states that current change in profits should influence future profitability of a firm, changing expected level of output, and, thus, changing the optimal future capital stock. The other reason of including the term is that internal funds could me more beneficial to use than external funds (see e.g. Duesenberry, 1958; Eisner, 1978). Modified Neoclassical model - unlike the standard neoclassical approach, the modified neoclassical model allows for putty-clay capital. The model was developed by Bischoff (1971), who found that new equipment and structures incorporate most of modifications in the capital-output ratio, while existing capital stock embodies less modifications due to changes in the relative prices of input. Consequently, Bischoff (1971) investigated that it might be easier to modify the capital-output ratio ex ante. Securities Value (Q) model the model attempts to explain investment behaviour on the basis of portfolio balance, rather than making investment a function of changes in output (as in all above mentioned models) investment depends on the ratio of a market value of capital to its replacement cost (Tobin s Q). The implication of the model indicates that if a market value of capital is greater than its replacement cost, a firm can increase the market value by investing in more fixed capital. On the contrary, if replacement cost is greater than a market value of a firm s assets, reduction in fixed assets will increase the value of equity (initial idea by Brainard & Tobin, 1968, and Tobin, 1969; further developed by Clark, 1979, and Summers, 1981). The topic of investment behaviour has been researched widely. Edwin Kuh was one of the first researchers, who studied investment patterns of firms, initially with Meyer (1957), and further alone (Kuh, 1963). He analyses the investment behaviour using such variables as depreciation, liquidity, dividends etc., but the main focus is devoted to the question, which factor is more influential, profit or sales. The answer, however, is not that straightforward, despite the fact that sales perform better throughout the analyses. Separate tests are highly informative, while tests for different aspects indicate 9

10 conflicting results, thus leading to controversial conclusions. Meyer and Kuh (1957) also suggest that one of the possible assumptions of investment behaviour could be the fact that firms are more interested in maximizing utility, rather than profits, meaning that the assumptions of the neoclassical theory of optimal capital allocation are too narrow in the real world. Jorgenson and Siebert (1968) study analyses five alternative models of investment behaviour. They conclude that the neoclassical model performed best, while models based on profit expectations or capacity utilization shows weaker results. Internal funds model appears to be the worst at describing investment behaviour. Authors decline the broader view of Meyer and Kuh (1957), as it is not supported by evidence from econometric studies. Jorgenson continues the analysis of different investment models together with Hunter and Nadiri. They compare four models of investment behaviour (see Table 1) using a dataset of US manufacturing firms (1970a). In addition, they test the predictive performance of these models by comparing prediction errors for a period of prediction with errors for a period of fit and testing both periods for structural changes (1970b). Conclusions of both papers are almost identical. Anderson and Meyer-Glauber models are misspecified, as these models could not explain investment expenditures, as Stephenson model produces better results. However, in the analysis of the predictive performance of the models, Eisner model is superior to the one by Jorgenson and Stephenson. Table 1. Models of Investment Behaviour. Table gathers models of investment behaviour studied by Jorgenson et al. (1970a, 1970b). The table also shows variables that determine investment behaviour of firms according to these models. Model (by name) Anderson Eisner well as due to autocorrelation of residuals. Out of two remaining models, Jorgenson- Jorgenson- Stephenson Meyer- Glauber Determinants of investment behaviour Capacity utilization, profits, interest rate, stocks of gov. securities, accrued tax liabilities, debt capacity Change in sales, change in profits, capital stock Change in value of output divided by price of capital services, capital stock Capacity utilization, profits, interest rate, % change in securities price index Source: Jorgenson et al., 1970a, 1970b. Bischoff (1971) tries to study the magnitude and timing of allowance for rising prices on the private spending for fixed capital. He also analyses the predictive performance of investment models over the next 2-3 years given alternative scenarios of economics as a whole. Bischoff (1971) finds out that the modified neoclassical model 10

11 fits the data best, while other models do not prove to be successful. The only alternative to the modified neoclassical model might be the accelerator model. Another research by Clark (1979) examines the investment behaviour in the 1970s. Besides the historical performance he also studies the forecasting performance of different investment models. Clark (1979) finds that accelerator model is superior to other models in terms of forecasting, while the modified neoclassical model proves to be a better model for historical analysis. The paper by Bernanke et al. (1988) opens a question regarding the specification of the most famous investment models. The authors also argue that previous studies of classification and ranking of investment models (goodness-of-fit and prediction analyses) were not fully representative, as these analyses had a lot of limitations and drawbacks like small samples, power properties, ranking techniques etc. The authors goodness-of-fit analysis also could not identify the model that would be statistically superior to other. The non-nested hypothesis test shows that none of four investment models, namely accelerator model, simple and modified neoclassical models, and securities value model, were correctly specified. One of the most recent studies conducted by Rapach and Wohar (2007) made a horse race experiment between 6 investment models to find out their forecasting abilities. Besides such models as accelerator, neoclassical, average Q, and cash flow, the authors also include two more recent models, real stock price and excess stock returns. Real stock price model was developed by Barro (1990), who virtually used real stock prices as a substitute to average Q, as author argues that stock prices better incorporate market evaluation of future performance of investments. Excess stock returns model was discussed by Lettau and Ludvigson (2002), who argue that variables, which showed successful forecasting abilities with respect to excess stock returns, should also be superior in terms of forecasting of investment spending. Rapach and Wohar (2007) find that average Q and real stock price models had more evident predictive abilities than other models. However, they also find structural breaks in all the models used in the analysis; hence, even the forecasting abilities of superior models remain questionable. 11

12 2.2 Investment Behaviour and Financial Constraints Inefficient Markets Unfortunately, or fortunately, markets are not that perfect and efficient. Agency costs, asymmetric information, transactions costs, risks and costs of financial distress, opportunity costs, absence of perfect substitutes etc. are among many market imperfections and financial constraints, which may influence the investment behaviour of a firm. The pioneers in stressing that financial constraints may be connected to investment behaviour were Meyer and Kuh (1957). The works by Modigliani and Miller (1958 and 1963), however, damped down the enthusiasm of researchers for further investigation of the effect of financial constraints on investment, and the topic had not as much attention as it should have had. On the other hand, market imperfections were discussed a lot in the literature. Asymmetric information and agency costs have been widely discussed in the last years from different angles. Asymmetric information was perfectly described in the article by Akerlof (1970). He explains the problem by means of second-hand cars. He argues that only sellers know which car is good (cherry), and which one is bad (lemon), but buyers in the market cannot tell them apart. This is why a buyer offers an average price for a car, which will be accepted by a seller only if the offered price is above the value of this car to the seller. This easily leads to the situation when offered prices are less than the value of good cars to the seller. Subsequently, cherries cannot be sold, which leads to the market failure, as only lemons are sold. In case of investments, only managers of investment projects know the whole information about the projects, while investors cannot distinguish between cherries and lemons because of a lack of information. The risk that a manager (agent) will act in his or her own interest instead of investor s (principal s) interests is quite high, and the costs associated with a principalagent problem can also be substantial. That is why agency costs could have an impact on investment decisions and may influence the development of investment projects. The topic of agency costs had a number of research (the pioneers were - Wilson, 1968; Jensen and Meckling, 1976; among others), which shows that information asymmetry between an agent and a principal causes problems of moral hazard and adverse selection. The principal should monitor and control the actions of the agent. These monitoring and controlling costs should be taken into account by an investor 12

13 when entering a new project, and should hence be incorporated in the required return of the project. It is obvious that when the company managing the investment project is financially constrained (e.g. low on cash; high D/E ratio etc.), agency costs increase as incentives to cheat and make riskier decisions are higher. Does it mean that investing in a firm with a lot of cash and small amount of debt will result in reduction of agency costs? According to Jensen (1986) and Stulz (1990) it is not the case. They found out that firms with extensive internal resources may invest more than they should, thus, causing overinvestment, which is also a part of agency costs. Opler et al. (1999) findings are also consistent with Jensen (1986) and Stulz (1990) that firms with considerable amounts of cash invest more than firms with lower amounts of cash. A research paper by Shin and Kim (2002) argues that agency theory explains the fact that investments by firms with large internal resources are less likely connected with their (firms ) growth opportunities, than investments by firms with low internal resources. Their findings that diversified firms invest more, but less efficiently than stand-alone firms, and that large firms invest more than small ones are support factors of the fact that agency costs affect the investment decisions of firms Investment and Financial Constraints Monotonic Evidence Conventional investment models discussed above are based generally on the assumption that business investments are independent of capital structure of the firm, as external finance is a perfect substitute to internal finance. It means that these models assume that investment behaviour of a firm does not depend on firm s financial conditions, but rather barely on the investment demand. Although this is true in case of perfect capital markets, it is questionable in the real world. Recent literature on capital market imperfections let researchers think about the fact that external and internal finance are not perfect substitutes. Asymmetric information and agency problems make it difficult, if not impossible, for external finance providers to evaluate the quality of a particular firm and its investments. That is why the cost of new debt or equity finance may not be the same, as the cost of internally generated cash. This leads to a view that investment may also depend on the availability of internal funds, access to new debt and equity finance, and the availability of external funds (operational behaviour of credit markets). 13

14 Fazzari, Hubbard, and Petersen (further Fazzari et al., 1988) provide one of the first researches on the topic of internal vs. external finance of the investment. The authors analyse a sample of 422 US manufacturing companies along the time frame of 15 years - from 1970 till They divide the companies into three categories based on their earnings retention practices. The authors assume that firms with high dividend-income ratio are less sensitive to the cash flow variations, as they are mature and well functioning firms, which experience small cost disadvantage of external finance. Alternatively, firms having low dividend-income ratio are young and inexperienced companies, which exhaust all internal funds, and thus are more sensitive to variations in cash flow. Fazzari et al. (1998) test this hypothesis using three conventional models of investment: Q model, neoclassical, and accelerator. The authors conclude that in all three cases investment behaviour of firms that use more internal funds, hence paying less in dividends, are more sensitive to fluctuations in cash flow than investments of high-dividend firms. These conclusions are in line with the evidence from capital market imperfections - that young and immature firms are not yet recognized by credit markets. Hence, the cost disadvantage of external finance is substantially higher for them, compared to mature firms with good reputation. Despite the fact that the work by Fazzari et al. (1998) is a step forward in terms of an endeavour to understand the investment behaviour of a firm, critics have identified several drawbacks of the analysis presented by the authors: Dividend policy of a firm is endogenous. No evidence that asymmetric information is an explanation of the wedge between the costs of internal and external financing. Analysis is limited to manufacturing firms - the assets of manufacturing firms are quite specific, and it may be difficult to use them as a collateral for bank loans, thus, the effect of information problems may be exaggerated. Hoshi, Kashyap, and Scharfstein (further Hoshi et al., 1991) find pretty similar results using a sample of 337 Japanese manufacturing firms. The analysis is based on the initial idea that information problems (capital market imperfections) may influence the financial structure and investment of firms. The authors use similar approach as Fazzari et al. (1988), but different classification of firms, to avoid classification according to dividend policy. They divided the sample into two groups. Group one 14

15 companies are affiliated with the industrial group, keiretsu, which consists of a number of large banks and other institutions, which coordinate the operation of the group and provide financing for a great portion of group s investments. Group two companies are stand-alone independent firms, which are not affiliated with any industrial groups, and have weaker relationships with banks. Such division of companies allows to distinguish between firms, which should not face information problems (group one) and which should (group two). The analysis shows that companies with close relationships with banks are less sensitive to liquidity (availability of internal finance), than independent companies. Therefore, the conclusions are in line with the initial idea that liquidity is an important determinant of investment for firms, which experience high information problems. Another study by Schaller (1993) uses different tests to analyse the influence of capital market imperfections on the corporate investment of 212 Canadian firms over the period To address the problems identified in the analysis by Fazzari et al. (1988), author introduces three original tests, which are based on the maturity of the firm, concentration of ownership, and the availability of collaterizable assets, respectively. Schaller (1993) also uses the approach by Hoshi et al. (1991), who categorized the firms based on a membership in an industrial group. The author finds that young and immature firms pay more for new equity financing and are more sensitive to liquidity. The same result is obtained for firms with dispersed ownership compared to firms with more concentrated ownership. Firms with specific assets that might be more problematic to use as collateral (e.g. independent firms, which does not belong to industrial groups) are more influenced by liquidity. Schaller (1993) also finds out that the effect of liquidity on investment behaviour of manufacturing firms is about three times higher than that of other firms. This conclusion casts some doubt on research by Fazzari et al. (1988) and Hoshi et al. (1991), as their results might be exaggerated. Mills, Morling, and Tease (1994) analyse 66 non-financial Australian firms for the 11-year period ( ). The authors find an evidence of the hypothesis that availability of internal funds and capital structure of a firm can, in fact, influence investment behaviour. Despite the fact that both factors are statistically significant, cash flow influence is found to be more economically important. The effect of both internal funds and capital structure, however, differs among firms. The small firms, firms with high D/E ratio, and firms that exhaust nearly all internally generated funds are more 15

16 liquidity sensitive than their opposites. Another conclusion by the authors concerns monetary policy, which is expected to influence investment of firms by influencing cash flows, as well as by influencing discount rate applied to investment projects. The effect of changes in monetary policy is expected to be more significant for small firms and highly leveraged firms. Hubbard, Kashyap, and Whited (1995) use Euler equation for intertemporal capital accumulation on the sample of the US manufacturing firms to identify the effect of internal finance on the investment decisions of firms. The authors use two sets of firms (with high and low dividend payout ratios) and compare them using neoclassical model of investment. The standard neoclassical model is not rejected for firms with high dividend payout ratio, and it is easily rejected for firms, which paid less dividends. Bernanke, Gertler, and Gilchrist (further Bernanke et al., 1996) analyse the topic in terms of financial accelerator. They argue that firms facing high agency costs in credit markets are the most vulnerable in times of economic downturns (flight to quality), which leads to reductions in production and investments, and hence, even worsen the effects of the downturn. Bernanke et al. (1996) also state that these firms should be the first, who respond to a turnaround in economy. Therefore, their findings that highagency-cost firms are more sensitive to liquidity than low-agency-cost firms are in line with other earlier research conducted on the topic. Hubbard (1998) reviews a number of recent empirical studies of investment behaviour of a firm. He argues that despite the fact that empirical studies are consistent with the idea that investment and liquidity are correlated; the magnitude of this correlation is still questionable and requires further research. Kaplan and Zingales (1997) question the pioneering approach by Fazzari et al. (1988) in their paper Do Investment-Cash Flow Sensitivities Provide Useful Measures of Financing Constraints? The authors argue that Fazzari et al. (1988) (as well as other researchers, who used the same or alike methodology) wrongly interpreted their results. In particular, firms, which were expected to have a wider range between internal and external financing costs, were more sensitive to liquidity, and hence were classified as more financially constrained. They state that there is no strong theoretical base under conclusions by Fazzari et al. (1988) and their successors. Kaplan and Zingales (1997) take 49 low-dividend firms studied by Fazzari et al. (1988), which were found to have a higher sensitivity between investment and cash flow, and make a thorough analysis of these companies. They analyse companies 16

17 annual reports, managers reports, and investment plans. They try to integrate it with quantitative data, as well as public news to capture the full picture about firms investment opportunities and possible financing abilities. Their findings are revolutionary. Less financially constrained firms experience higher sensitivity to liquidity. The authors also discuss several possible weak points in approach by Fazzari et al. (1988). For example, Tobin s Q may not provide accurate results, as internal funds may act as a proxy for investment opportunities, but Q does not capture this issue. The other possible weakness is outliers, which influenced sensitivity results. Kaplan and Zingales (1997) argue that they proved that investment-liquidity sensitivity is not an appropriate measure of companies financing constraints. They capture non-monotonic behaviour of investment-liquidity sensitivity, as more financially successful (less constrained) companies rely more on internal funds, in spite of the possibility of cheap external financing. These conclusions, if true, bring up questions about credibility of many studies that use Fazzari et al. (1988) methodology. Besides that, the paper by Kaplan and Zingales (1997) provokes extensive debates among researchers and scholars, which study this topic. Fazzari et al. (2000) provide a comment on Kaplan and Zingales (1997), where they criticize inability of Kaplan and Zingales (1997) to capture the methods used in their paper, as well as other literature. The authors argue that Kaplan and Zingales (1997) are unable to understand correctly their approach. That is why the arguments and criticism in the paper by Kaplan and Zingales (1997) do not have enough force. Fazzari et al. (2000) also show that the opponents approach to classify firms according to the degree of constraints fails to hold, and thus, is not noteworthy. Therefore Fazzari et al. (2000) show that the attempt by Kaplan and Zingales (1997) to analyse whether investmentcash flow sensitivities is a useful measure of financing constraints is not informative in any aspect. Kaplan and Zingales (2000) in turn provide a comment on the comment by Fazzari et al. (2000). They argue that Fazzari et al. (2000) comments in fact confirm the findings by Kaplan and Zingales (1997), but disagreements mainly concern empirical findings, which are pure math. Non-monotonic relationship of investment-cash flow sensitivity and financing constraints is found empirically, but is not explained theoretically. There is no clear evidence also of the fact what really causes investmentliquidity sensitivities. As Kaplan and Zingales (1997) show, sensitivity is not actually 17

18 caused by financing constraints. Therefore, the question remains open for discussion. Kaplan and Zingales (2000) suggest that the answer may lie in behavioural finance, and not be driven by value maximizing behaviour of firms. They argue that one of the possible causes of these sensitivities may be managers non-optimal behaviour, as well as theirs excessive conservatism. But Kaplan and Zingales (1997) were not the only ones, who obtained opposite to Fazzari et al. (1988) results. Cleary (1999) finds the same patterns in his analysis of 1,317 US firms over the time period He uses similar to Kaplan and Zingales (1997) approach. He classifies companies according to financial variables related to financial constraints, e.g. current ratio, debt ratio etc. Methods used by Cleary (1999) to identify firms financial status are similar to Altman s Z-score (see Altman, 1968), which is used for bankruptcy prediction. Such methods allow sample firms to reclassify over time, if firms financial status becomes better or worse off. The author concludes that firms with better financial situation (more creditworthy firms) experience significantly more sensitivity to liquidity than firms experiencing financial problems (less creditworthy firms). Both Fazzari et al. (1988) and Kaplan and Zingales (1997) obtained different and conflicting results. But nobody could explain why you get different results, when you change an approach of classification of firms in a sample. Despite conflicting conclusions, both papers argue that there is a strong evidence of investment-cash flow sensitivity, and that investment decreases when firms become more financially constraint. All papers presented above recognize the fact that firms investment behaviour depends on liquidity level of these firms. This dependence is found to be monotonic, as all studies argue that investment expenditures drop in financing constraints Investment and Financial Constraints Non-monotonic Evidence As we discussed above, due to capital market imperfections (information asymmetry and agency problems), the cost of internal and external finance is not the same. A lot of research has been done in order to solve the problem of capital allocation in the investment process, and generally acknowledged conclusions show that the more firm is constrained the less it invests (see, e.g. Bernanke et al., 1996; Hubbard, 1998; Stein, 2003 etc.). 18

19 The recent paper by Cleary, Povel, and Raith (further Cleary et al., 2007), however, shows that a firm s investment is a U-shaped function of its liquidity. This means that until certain low levels of internal funds relationship between investment and cash flow is positive and monotonic. However, when liquidity reaches some particular low-level point, this relationship turns around, and a further decrease in cash flow leads to an increase in investment. Cleary et al. (2007) base their model on three main assumptions: External funds are more costly than internal funds due to capital market imperfections. The cost of raising external funds is determined endogenously by an investor s requirement to earn sufficient returns. Investment is scalable, meaning that a firm can choose between more or less costly investments, but not just to decide whether to invest in a given project or not. Cleary et al. (2007) provide an illustrative numerical example of employing these three assumptions in practice and show the intuition behind a U-shaped investment curve (description of the example can be found in Appendix I). This simple example shows the rationale behind Cleary et al. (2007) reasoning. The model used by the authors will be better described further in Methodology part (Section 3.1), as we will base our analysis on their empirical model with several developments, which are important for our investigation. Cleary et al. (2007) clearly show that for U-shaped investment curve to be present, satisfaction of these three assumptions is essential. If one assumes, e.g. that a firm is not able to choose between smaller or larger projects, but can only choose whether to invest in a given project, investment becomes monotonic in liquidity (assumption is made in e.g. Bernanke & Gertler, 1989, Bernanke, Gertler & Gilchrist (1999). The same result can appear if one determines cost of external funds exogenously, as investment preferences are not correctly specified in such a case (assumption made in Kaplan and Zingales, 1997; Stein, 2003). It is also important to capture the effect of negative cash flow, as a lot of previous research excluded negative levels of internal funds from the analysis. Cleary et al. (2007) also argue that their analysis help to recognize the differences between two conflicting results, namely recent debate between Fazzari et al. (1988), and 19

20 Kaplan and Zingales (1997). Fazzari et al. (1988) find that investment-liquidity sensitivity is higher for firms, which were initially identified as experiencing more severe financial constraints, while Kaplan and Zingales (1997) demonstrate that foundation of Fazzari et al. (1988) approach is poorly structured, and present results opposite to Fazzari et al. (1988). Cleary et al. (2007) show that these differences come up when you employ different classification methods in the analysis. Fazzari et al. (1998) use proxies for capital market imperfections for classification of firms, but Kaplan and Zingales (1997) classify firms according to financial ratios, which are correlated with internal funds. As it was mentioned before, researchers could not come up with explanations why different classification techniques lead to opposite results. Cleary et al. (2007) clarify their findings as following. When liquidity level is not significantly low, then more asymmetric information leads to higher costs of external finance, and investment decreases. When liquidity level is sufficiently low investment increases, hence, producing opposite results. Therefore, when firms are classified according to capital market imperfections (e.g. information asymmetry), and financially weak firms are excluded from the analysis (as Fazzari et al. (1988) did), investmentliquidity sensitivity is more striking for firms with more financial constraints. Contrariwise, when firms are classified according to financial ratios, which are proxies for the level of internal funds, the more constrained firms experience lower sensitivity between investment and cash flow. The paper by Cleary et al. (2007) is based basically on three previous studies on the topic: Cleary (1999), Povel and Raith (2001), and Povel and Raith (2004). Povel and Raith (2001) found out that U-shaped investment curve exists using the following intuition. Liquidity level of a firm affects its investment in two ways, in particular, by cost and a revenue effect. The cost effect comes up when a firm s internal funds decrease. Lower liquidity means larger external financing for any given investment, thus, increasing the marginal cost of debt and eventually decreasing investment. In contrast, increasing investment leads to higher future revenues, hence, decreasing the marginal cost of debt. Povel and Raith (2001) argue that for higher liquidity levels the cost effect dominates and we obtain positive monotonic relationship between internal funds and investment. For lower liquidity levels, however, the revenue effect becomes prevailing, but for negative levels of internal funds the revenue effect is dominant. This means that investment is a U-shaped function of the firm s internal 20

21 funds, and it can go both directions (i.e. decrease or increase) when internal funds move only one way (e.g. decrease). Bhagat, Moyen, and Suh (further Bhagat et al., 2005) also find similar results, but focus of their research is devoted to firms operating performance. Firms in financial distress document different results, if their operating performance differs. The authors find positive sensitivity of investment to liquidity if constrained firms have operating profits. The authors argue that financially distressed firms having operating losses have zero or negative investment-liquidity sensitivity depending on whether next year investment decreases or increases. Bhagat et al. (2005) find that 40% of financially distressed firms invest more next year, thus, investment of these firms increases while cash flow goes down (negative sensitivity). The authors discover that these investments are generally sponsored by equity holders. This fact suggests that equity holders gamble for resurrection. Guariglia (2008) investigates relations between investment and cash flow using panels of UK firms, as, according to the author, European data is not widely researched. The study confirms findings by Cleary et al. (2007). Guariglia (2008) explores the effect of two types of financial constraints on investment behaviour of a firm, namely internal and external constraints. The effect of both types of constraints is investigated separately, as well as jointly. This time around 99% out of 7,534 UK manufacturing firms (and out of 24,184 firms operating also in other industries) were non-traded. The author argues that it is interesting to look at firms, which are not quoted on the stock market, as quoted firms are generally large and financially stable, with good credit history, while non-quoted firms are relatively small, experiencing solvency problems, as well as having worse reputation compared to quoted firms. The findings by Guariglia (2008) show that sensitivity of investment to liquidity reacts differently, depending on a type of financial constraints. Sensitivity is strong, when it is problematic for the firm to access external financing, and when the firm has no problems with liquidity. What is more, investment is found to be the U-shaped curve of internal funds, when the sample of firms is divided according to internal financial constraints. While monotonic relationship between investment and cash flow is obtained, if dividing the sample according to external financial constraints. These findings support conclusions by Cleary et al. (2007), as well as show again that different results by Fazzari et al. (1988), and Kaplan and Zingales (1997) are the consequence of different criteria used to classify data samples. 21

22 Flor and Hirth (2008) provide their own view on the relationships between investment and liquidity. The authors get rid of asymmetry, and try to study the simplest debt contract, in order to examine cost trade-off between expected liquidation costs and second-best investment costs, and its influence on the investment-liquidity relations. Flor and Hirth (2008) find that this trade-off is a reason of the U-shaped investment curve. They found that unconstrained firms do not want to be involved in debt financing, because liquidation costs dominate underinvestment costs, while constrained firms tend to invest more, due to the fact that firms benefit more if they get closer to the first-best investment. The results are somewhat similar to the findings by Povel and Raith (2001) with their cost and revenue effects. Another study by Hovakimian (2009) uses different approach to classify firms into groups. The author obtains firm-level estimates of investment-liquidity sensitivity and classifies firms according to these estimates. Namely firms with positive and negative investment-cash flow sensitivities, as well as cash flow insensitive firms. Hovakimian (2009) examines different characteristics of firms. In particular, these are financial constraints, internal funds, growth opportunities, and investment and financing behaviour. She founds that firms experiencing positive sensitivity are small and young; they pay lower dividends compared to insensitive firms. These firms also have lower levels of internal funds, and compensate it by having more debt and equity finance, despite higher costs of external funds. Such characteristics make it available for positive sensitivity firms to invest more, hence, grow faster than their insensitive counterparts. Negative sensitivity firms, however, are the weakest in terms of liquidity. These firms are even smaller and younger than firms with positive cash flow sensitivity. But size, age, liquidity constraints, and high costs of external financing do not prevent these firms from demonstrating high levels of investment and growth rates. Hovakimian (2009) explains these findings using corporate lifecycle hypothesis. Over their lifetime firms with negative sensitivity experience opposite trends in cash flows and investment rates. This can happen due to the fact, that negative sensitivity firms are small and young. They become quoted with low levels of liquidity and valuable set of investment projects, which are highly valued by the market, as these firms are able to raise needed amounts for investments using equity and debt financing. As firms grow up and mature these investments materialize into higher levels of cash flow. At the same time, they invest less, as investment opportunities are not perceived 22

23 as very attractive any more. This leads to negative relationships between investment and liquidity. Yildiz (2010) replicates the model by Cleary et al. (2007) using a sample of US manufacturing and mining firms over the time frame (we are using data sample of manufacturing firms used in the work by Yildiz (2010) in our analysis, hence more information about the sample will be provided further). The study confirms the U- shaped investment curve predicted by Cleary et al. (2007). Yildiz (2010) finds that results of manufacturing companies are more statistically significant than that of mining firms. She proposes that these findings may be caused by the fact that mining firms experience less competition, and are influenced by the government. The author also argues that manufacturing firms are more sensitive to asymmetry, due to higher uncertainty about their future profitability. Manufacturing firms are found to invest more at high and negative levels of internal funds compared to mining firms. In contrast, when cash flow is around zero mining sector companies invest more. Concerning sensitivities, Yildiz (2010) findings are in line with Hovakimian (2009). The author identifies that the most investment-cash flow sensitive firms are the ones with the positive around zero level of internal funds. Firms with high levels of internally generated funds are less sensitive, but firms with negative cash flows experience negative investment-cash flow sensitivity. The models discussed above can be classified as static models, which restrict research to now-or-never investment opportunities. In some cases a firm is assumed to invest or not in a given project (see e.g. Bernanke & Gertler, 1989; Bernanke, Gertler & Gilchrist, 1999). In other cases a firm can choose between larger and smaller projects (see e.g. Cleary et al., 2007). However, in both cases investment behaviour is evaluated exclusively according to projects NPVs. Dynamic models, on the other hand, also embody the possibility to study the trade-off between current and future investments. Boyle and Guthrie (2003) apply a dynamic model to study investment behaviour of a firm. They analyse the dynamic relationship between investment and liquidity by including financial constraints into the investment-timing model of McDonald and Siegel (1986). The timing model assumes that the firm has rights to enter investment whenever it wants, and does it when it expects the highest payoff. Boyle and Guthrie (2003) develop the model by introducing capital market frictions that may affect investment decision. The authors find that possible future decrease of liquidity influence the value of firms timing options and leads to early and not optimal investment 23

24 decisions. The findings are actually in line with the ones found by Whited (2002) that small firms invest more often than large ones. Boyle and Guthrie (2003) also find that availability of internal funds makes it easier to invest currently, but at the same time, more cash flow makes it less risky to wait and see if the investment is become more attractive. The authors identify that more internally constrained firms are greatly influenced by less risky waiting. Hence, they will most probably wait and see, while firms with more cash will invest earlier and more often. This conclusion is in line with findings by Kaplan and Zingales (1997) that less financially constrained firms (firm with more cash flow) rely greatly on internal funds compared to financially constrained firms. In addition, Boyle and Guthrie (2003) analyse relationships between investment and uncertainty. They shed light on the fact, why most studies almost do not find significant relations between investment and uncertainty, as it is hard to measure correctly different kinds of uncertainties. This problem leads to a dilemma. More uncertainty about investment leads to higher opportunity costs of current investment, hence to wait and see situation. More uncertainty about firm s financial conditions leads to more risky waiting, hence increasing investment. Dasgupta and Sengupta (2002) use a dynamic multi-period model in their analysis. They describe the same example as Boyle and Guthrie (2003) that more cash flow makes it easier to invest more, and alternatively, makes a firm to be more conservative now, as it is less risky that the firm will experience future shortfall in cash. Authors find non-monotonic relationships between internal funds and investment. They argue that increase in internal funds (but only to a particular point) influences firm s investment decision in such a way that it leads to wait and see situation, hence decrease and investment. But when internal funds are low or high, relationships are monotonic. Providing a real life example, Dasgupta and Sengupta (2002) show that small and financially constrained Japanese firms, which were increasing the proportion of internal funds to assets since mid-nineties, have not experienced increase in investments yet. Moyen (2004) analyse two dynamic models: unconstrained (where firms have easy access to external finance) and constrained (where firms have not access to external finance). When using dividends as classification of firms with financial constraints (firms that pay lower dividends are more financially constrained, and vice versa), Moyen (2004) gets the results similar to Fazzari et al. (1988). She also finds that constrained firms are less investment cash flow sensitive, if classify firms according to 24

25 constrained model (findings similar to Kaplan and Zingales, 1997). The author explains this conclusion by the fact that unconstrained firms issue more debt, which is not taken into account by econometric specification. Hence, it increases the effect of cash flow on investment. 2.3 Lifecycle Theories After analysing investment literature, we devote our attention to examination of research about lifecycle and age of a firm. Miller and Friesen (1984) conduct a study on how corporations develop over time. The authors state that theoretical literature suggests five stages of development of a firm. These stages, although named differently from one article to another, can be classified as birth, growth, maturity, revival and decline. Further, we will shortly describe each phase: The birth phase usually involves substantial risk taking and niche strategy with innovative products. The growth phase, not surprisingly, means a period, when young, but already established company quickly increases its sales and assets to take advantage from the economies of scale. Companies in maturity phase experience stabilisation of sales, decreased level of innovation, and increased bureaucracy. In the revival phase company diversifies and gains from the economies of scope. This stage is also characterized by divisionalised structure of a company and more complex control and planning systems. The decline stage means stagnation, lack of innovation and external challenges. The authors try to observe such aspects as strategy, structure, decision-making methods and organizational situation over the lifecycle of a firm. Their study is based on longitudinal analysis. The authors analyse 36 corporations over the time frame of at least 20 years. The database is rather unconventional and includes book records, annual reports and series of magazine articles. The authors verify information with questionnaires sent to previous and current top executives. The findings of the paper point out that organizations situations has a tendency to become more complex, more dependent on customers and with less concentrated ownership. The observed companies surprisingly remain rather centralized in terms of 25

26 decision-making, even after becoming more divisionalised. However, routine tasks are delegated to divisions. The hypothesis regarding the strategy is also supported empirically. For example, firms have more significant innovation-related variables during the growth phase compared to the maturity phase. The authors describe all five phases using the mean-difference analysis and publicly available information from annual accounts. Further, Miller and Friesen (1984) find out that for many companies growth, maturity and revival phases lasts for ten years or more, because of a tendency to remain within a current phase. However, the tendency for firms to progress through lifecycle periods has also been observed. The interesting fact is that some firms from the revival phase go back to maturity phase. Other firms show a movement from decline to revival or even maturity, which is movement opposite to the lifecycle proposed by theoretical literature. Therefore, the authors claim that conventional literature oversimplifies lifecycle patterns of a firm. Another work by Miller and Friesen (1983) analyse successful and unsuccessful phases of corporate lifecycle. The objective of the research is to determine whether there are significant differences between successful and unsuccessful phases of the lifecycle. The authors define successful phases as those in which a company performs well financially. Correspondingly, unsuccessful phases are those in which a company performs poorly. The first hypothesis is that successful phases will show a linear tendency of increase in information processing and decision making variables. Unsuccessful phases should not show such a tendency. As an extension of this hypothesis Miller and Friesen (1983) also argue that successful companies should have increased use of intelligence systems and controls, like employee performance appraisals, quality controls, profit and cost centres etc. The second hypothesis suggests that firms in successful phases will tend to have more risk taking, pro-activeness and innovation, as opposed to firms in unsuccessful phases. Firms in unsuccessful phase should have more excesses of innovation and stagnation, which lead to diminishing resources and archaistic product line correspondingly. The authors find out that there is no significant difference between successful and unsuccessful subsamples regarding information processing and decision-making style in a birth stage. The difference becomes significant only in later stages, starting from the 26

27 growth phase. The interesting finding here is that in successful phases, information processing is constantly increasing with only slight levelling-off in a decline phase. Unsuccessful phases show significant reduction in information processing. Decision making style has a similar pattern. The authors also find empirical support for their second hypothesis. The unsuccessful phases do in fact show higher variability of innovation in different phases. This means that firms in unsuccessful stages tend to go to extremes in increasing or decreasing the level of innovation. Firms in successful stages have much less variability in this factor. Loderer and Waelchli (2009) study how age of a firm affects its performance. The recent financial crisis and economic downturn showed that even major companies in different industries might be unprepared to face the drastic changes in business environment. The authors argue that often older and established companies struggle in situations, where younger and more flexible companies excel. Old companies may become more efficient, develop know-how and hire human capital. Alternatively, old companies may find their knowledge and abilities obsolete, processes overburdened with bureaucracy. To analyse the situation the authors test organizational rigidity hypothesis, because it is unclear, which effects of aging are stronger. Another hypothesis is that the firms might become increasingly more interested in rent-seeking. In some cases, as the authors state, rent seeking can take the form of aiming to quieter life, meaning that firms become lazier and less aggressive. To find empirical support for their hypothesis authors study about 11,000 listed firms between 1984 and It turns out that profit margins, return on assets and Tobin s Q ratios fall with age. In addition, the authors find out that older companies tend to have higher costs, slower growth and reduced R&D. Corporate governance quality falls with age as well. What is essential, these results are valid for different industries in different time periods. The only exception is very young companies, which just have been listed. These companies observe increasing performance indicators for some time. However, after certain time performance starts to decrease. The authors draw much attention to the fact that the results might be spurious. They discuss, among other, possibilities for the age of a firm to be a proxy for the age of industry, ageing human capital and declining uncertainty. To sum up, both hypotheses are not rejected and could not be explained by other factors. 27

28 Another paper by Loderer, Neusser and Waelchi (2009) (further Loderer et al., 2009) studies the dependence of survivability of a firm, depending on its age. The authors define the death of a firm as an actual bankruptcy or a takeover by another firm. Therefore, comparing to other studies, they focus more on loss of independence. Loderer et al. (2009) use the same database as mentioned in the previous paper. The authors find out that comparing to previous studies on bankruptcy, their work gives more insight on actual resource redistribution. This is due to the fact that bankruptcy is not as common for listed companies as loss of independence. The median listing age for U.S. firms, as the authors mention is only 10 years. A typical death for a firm is to be recycled into another organisation. The authors find the U-shaped function of takeover hazard to age, which, in their opinion, is a strong empirical evidence of the existence of the corporate lifecycle. This effect is even more intensive in high-tech and R&D firms. Balasubramanian and Lee (2008) study the relationship between a firm s age and innovation. The authors use age as a proxy of experience of a firm. On the one hand, experience helps companies to choose paths for innovation, use resources more effectively and learn-by-doing. On the other hand, age is associated with corporate inertia, i.e. bureaucracy, slow decision-making process, unwillingness to deviate from the successful models of the past. To determine which effect is stronger the authors study the relationships of age and patent citations of U.S. firms in the period of The final regression they get includes age, size, claims, R&D, control of technological area and dummy variable for time-fixed effects. The model predictions are as follows: The quality of innovation would fall nonlinearly with firm age. The rate of decrease in the quality with age would be grater in areas with high levels of technological activity. The quality of innovation might increase or decrease with firm size, depending on the phase of the industry. The authors find out that technical quality of innovation falls with firm s age. The learning curve effect cannot offset the burden of age. Patent quality falls with age and size. In sum, authors found out that each additional year each additional year reduces the impact of a 10% increase in R&D intensity on the firm s market value by over 3%. 28

29 Alimin et al. (2010) perform a study on the relationship between organisational competitive advantage and performance, moderated by age and size of firms. The authors focused on the manufacturing companies in Malaysia. They conducted a research on 127 manufacturers listed in the 2008 Federation of Malaysian Manufacturers Directory. It was a cross-sectional study with two-way ANOVA analysis. The authors relate to non-conventional way to define the age of a company. In particular, they find out that some studies define the age of a company as the number of years firm has been engaged in exporting operations or the age of international joint venture formation. The authors propose two hypotheses. First, that older firms have a stronger relationship between competitive advantage and performance. Second, that larger firms have a stronger relationship between competitive advantage and performance. The authors discover that the size of firms does not significantly influence the relationship between competitive advantage and performance. They argue that the reason is the rapid development of technologies, which allows many firms to outsource and thus, keep the size of a company smaller. On the other hand, age does, in fact, have a significant effect on the relationship between competitive advantage and performance, the reason being the experience, which comes with age. Sakai et al. (2010) perform a study on the relation between firms age and evolution of borrowing costs. This paper makes a study on smaller firms from Japan. The authors managed to gather data on 100,000 bank-dependent small firms. This research is unique specifically due to the fact that it is very complicated to gather data on small firms. The authors study, which of three explanations of the evolution of borrowing costs is proved empirically. The first explanation is a reputation of a firm. The better is reputation, the lower are borrowing costs of a company. Reputation is closely related to the age of a company. The second explanation is dynamic of a firm. This means that the bigger a company gets, the more it is concerned of volatility of its profit. Thus, a company starts to reduce its leverage, decreasing a probability of default. This, in turn, leads to lower borrowing costs. This theory, in essence, states that relationship between age and borrowing costs is non-existent, when one controls for size of a firm. The last explanation focuses on lender-borrower relationships. The older a company gets, the longer is the information 29

30 exchange between the company and a bank. However, the final outcome of such relationships is unclear. Some studies state that this will lead to a lower interest rate, due to the increased transparency. Other studies believe that due to longer relationship a firm becomes informationally locked and a bank can abuse it in form of higher interest rates to the firm. As a result the authors find out that the reputational hypothesis is the most plausible. The reputation is by far the most important factor influencing the borrowing costs in small companies in Japan. From small firms in Japan we move to large firms in Australia. Palangkaraya et al. (2009) research the influence of size and age of a firm on productivity. In particular, they investigate how the productivity of large Australian firms evolves over a specific period of time. The authors find out that productivity differs greatly among large firms even from the same industry. Another interesting finding was that larger firms (in a sample of large firms) tend to have better productivity. However, this result does not hold for older firms. As a matter of fact, the authors realised that age has a negative correlation with a firm s productivity. However, the result is not fully robust, as it does not hold in alternative specifications of a model and with alternative data measurement. At last, the authors find out that productivity is a rather sticky factor. For example, low productivity firms tend to stay around the same rank compared to the other companies in the same industry. Therefore, the hypothesis that large Australian firms tend to converge in productivity do not hold. Fagiolo and Luzzi (2006), in turn, investigate if liquidity constraints matter in explaining firm size and growth on the example of Italian manufacturing industry. The authors analyse the Law of Proportionate Effects (LPE) designed by Gibrat in 1931 (Mata, 2008). The law stipulates that the size of a company and its growth rate are uncorrelated. Since then, many empirical studies were searching for support of this theory. However, the results were not as clear. In particular, previous literature investigated growth-size relationship without too much attention to other factors. Notably, the focus was on size, age, industry and other similar factors. The authors explain that the recent studies show that if one takes into account such determinants of firm growth as financial factors, the results start to differ. The authors show the findings of other articles that liquidity constraints tend to negatively affect firm s investment and to increase the probability of default. As liquidity affects investment decisions, then it 30

31 should also have an influence on size and growth dynamic of a firm. Thus, the authors investigate if LPE holds taking into account liquidity constraints. In addition, they research the effects of firm financial constraints on the relationship between age and growth, conditional on size. The authors managed to gather a substantial database of 14,277 observations of Italian manufacturing firms for the period (AIDA database). Using Gibrat s type regressions they find out that liquidity constraints do in fact have a negative effect on the growth of a firm, if one controls for a firm size. As the authors discover, the smaller firms tend to grow more even if one controls for liquidity constraints. In addition, in their study they find out that the age has a negative effect on growth only when one does not control for liquidity constraints. However, if one does control for financial factors, conditional on size, age becomes irrelevant. This and other factors show the evidence against the LPE. Evans (1987) studies the relationship between a firm s growth, size and age. He focuses on manufacturing industries. His study uses the unique database (Small Business Data Base), which covers the period The database is unique, because it has direct and precise measurement of the age of a company. Most of the other studies use different proxies to assess the age of a firm, like the date of IPO or the date of being registered in a certain database. The author finds out that the older firms are less likely to default. The older firms experience decreased growth and variability of growth. This result, according to the author, supports the learning model of Jovanovic (1982). The second finding is that on this sample Gibrat s Law is not supported. This means that small firms experience slower growth decrease compared to large firms. However, this is in line with other studies, showing that small firms tend to deviate from Gibrat s Law more than large firms. Desai et al. (2008) in their paper When Old is Gold: The Role of Business Longevity in Risky Situations study the impact of firms age on perception of consumers. In particular, the authors state that the age of a firm might be a proxy of firm s quality in the eyes of the customers. Furthermore, a firm s age is supposed to allow the consumers to make a prediction regarding the future quality. An interesting point the authors raise is that for the customer, when choosing between the firms with similar performance, the older firm might be preferable. This is due to the fact that although the companies might be at a similar level of quality, the 31

32 successful track-record gives an additional perceived value in a form of a smaller perceived risk. It is worth mentioning that the authors focus on service industry, which has certain implications. To find out if this is true the authors perform a number of experiments. What they discover is that when there is no information available to a customer, except of the age of a firm, the customer tends to prefer older firms. If there is qualitative information available to a customer, then he or she tends to prefer older firms only in case of low involvement. This means that the customers do not invest time in processing available information. In case of high involvement, age is not as much as important. At last, when no qualitative information is available, but a quantitative track record is given to a customer, he or she tends to prefer older firms in case customer s targets of performance and service are relatively low. In case of higher aspiration levels, a consumer prefers younger companies. The authors believe that this is due to a higher perceived upside potential of younger firms. Henderson (1999) studies the dependence of age and a strategy of a firm. He describes the strategy burden as newness, adolescence and obsolescence. Newness means that it is on average harder to implement new routines and processes or to borrow old ones, compared to just continuing the processes already in place. This places young firm in a disadvantageous position. Moving further, another problem firms may encounter is adolescence. This means that companies are at the biggest danger of default just a few years after they were founded. The length is dependent on the amount of resources invested initially. Similarly to newness problem, firms with adolescence problem have the highest risk of default in early stages of development. When one starts to control for size, age parameter might actually change the sign. In particular, one can observe the effect of obsolescence. This means that if one controls for a firm size, one can observe that age makes the companies to become more inertial and their performance decreases. 3 Methodology We try to examine the U-shaped investment curve from lifecycle perspective. We believe that lifecycle analysis will be helpful in investigating trends in investment-cash flow relationships. For this purpose we use the model by Cleary et al. (2007), which we further supplement with lifecycle specifications. This section provides an overview of 32

33 Cleary et al. (2007) model that we are going to use in the analysis, as well as the assumptions and modifications we employ. 3.1 Theoretical Framework We base our analysis on the theoretical framework provided by Cleary et al. (2007) (for more info see Cleary et al., 2007, pp.7-17) The model by Cleary et al. (2007) assumes a firm that can make a non-negative investment I! 0. The investment will generate some payoff in the future, which is defined by stochastic revenue of F( I,! ) in one period.! is a variable that is distributed randomly with density! (" ), and c.d.f.!(" ) over an interval [!,! ]. The following assumptions are made: Partial derivatives F! and F I! are positive meaning that higher! leads to higher revenue and higher marginal revenue on investment I. Authors propose to think of! being some uncertain state of demand for a firm s products and/or services. F II < 0, and E[F(I,!)]! I has a unique maximum value I (the first-best investment) at some level of I. F(0,!) = 0. This assumption means that if investment is zero, revenue is also zero. F(I,! ) = 0. If the firm borrows externally in the bad state (when state of demand is!, the firm will default its debt obligations with positive probability. The model has the following timing (graphic representation of the timeline of the model can be seen in Figure 1 Appendix II): T 1 the firm has internal funds W (W can be both positive and negative), and wants to make an investment I. The firm is considered to be financially constrained, if W < I. In this case in order to be able to make the investment I the firm needs to borrow and amount D = I W, therefore it offers a financial contract to an investor, which can either accept or reject the offered contract. T 2 the firm earns revenue F(I,!), which depends on a realized demand for the firm s products!. The revenue is unobservable to investor. The firm, in turn, realizes whether this revenue is more or less than the amount borrowed. T 3 the firm repays the loan to the investor. The realized payment is equal to R, which can be either equal to D or less than D, depending on the state of demand!. If! is approaching!, then F(I,!) " D, therefore R = D. But if! is 33

34 approaching!, then F(I,!) < D, and R < D. And depending on the realized payment R, the firm is liquidated or is allowed to continue operations. The model specifies liquidation as a stochastic process, meaning that the probability of liquidation being a function of the firm s realized payment is stated in the contract. T 4 if R = D, then the firm is allowed to continue, and it earns an additional non-transferable payoff! 2. If R < D, the firm can be still allowed to continue with the probability!(r) = 1 " (D " R) / # 2, hence it also earns! 2. In the situation of R < D, the firm is liquidated with the probability1! "(R), and the firm s assets are sold for L (a verifiable liquidation value), which is less than! 2. Cleary et al. (2007) assume that the revenue is unobservable, while future payoff! 2 is observable, but non-verifiable. This assumption is made for convenience, in order to avoid complex calculations, which will not actually change the final results. As the firm s assets can be sold for L, the model assumes it to be a market value of assets. Assets may be sold to become a source of repaying the debt to the investor, if the firm violates contractual obligations. The owner of a firm, in turn, values the assets differently the assets are worth! 2 for him/her. Therefore, the owner sees the difference! 2 L as a private benefit from running a firm, or a non-contractible future payoff. Cleary et al. (2007) argue that L does not influence the investment behaviour and debt repayment in any sense in this model, while! 2 is actually a variable that motivates a firm to repay the debt, as a firm is afraid of losing its future payoff. What is more, the model assumes no fixed costs related to investment, as well as no possibility financing investments by issuing risk-free claims. Cleary et al. (2007) also assume that a firm has no debt that is due after a firm gains revenue from its new investment, as it would complicate the analysis, as well as it allows to analyse underinvestment not caused by debt overhang (authors allow though for debt, that is due before new investment is made). Cleary et al. (2007) come up with four propositions concerning the optimal debt contract, relations between internal funds and investment choice, and asymmetric information. Cleary et al. (2007) use the informational assumptions of Diamond (1984) and Bolton and Scharfstein (1990) to derive the optimal debt contract. As it was mentioned before, the revenue is unobservable, hence, a threat of liquidation (thus a threat to lose! 2 ) is used to bring the firm to pay back the debt. 34

35 Proposition 1. The firm s internal funds are assumed to be at least (Cleary et al., 2007, p.9): W : = - $! 2 " L ( E $ F( I,#)! % & ' + L F I,# % 2! 2 ( ) " I & ) [1] ' As it was mentioned earlier, if the firm has a shortage of internal funds, but wants to invest an amount I, which is greater than W, the firm offers the contract to the investor, to borrow D = I W. After a firm repays its debt with a realized payment R, it is allowed to continue if R = D. Otherwise (if a firm defaults on its full repayment, namely if R < D) a firm is allowed to continue only with probability!(r) = 1 " (D " R) / # 2, or is liquidated with probability1! "(R). The threshold between default and solvency is defined by equation [2], while the investor s participation constrained is defined by equation [3]: D = F I, ˆ! ( ) [2] ( ) ˆ! $ F( I,! ) + D " F I,! ' * L! % & # 2 ( ) + (! )d! + 1", ˆ! ( ( )) D = I " W [3] The repayment of the debt cannot exceed the additional non-transferable payoff ( D! " 2 ). The threat of liquidation (thus the threat to lose! 2 ) brings a firm to repay its obligations. It means that the optimal contract is in fact the tool inducing the firm to repay the debt in full, depending on the state of!. A firm sees liquidation as inefficient development of events (as it yields L <! 2 ), therefore the optimal contract minimizes the likelihood of liquidation, leading to a probabilistic liquidation tool (for more detailed information see Povel and Raith, 2004). When a firm decides for which investment to go, I determines the amount that a firm needs to borrow (I W), as well as the required repayment D and the bankruptcy threshold ˆ!. A firm tries to maximize the following formula, by choosing optimal I and D: ˆ" " %! ( F( I," ))# 2 $ (" )d" + % '( F( I," ) & D + # 2 )* $ (" )d" [4] " ˆ" subject to the investor s participation constraint [3]. This formula can be simplified using Proposition 1probabilities: E "# F( I,! ) $% & D( I,W ) + ' 2 [5] 35

36 where D(I,W ) solves equation [3]. Cleary et al. (2007) argue that a set of equations [5], [2], and [3] has a unique solution for I. This solution shows that relationship between I and W is U-shaped. Proposition 2.. At W! I and at W = W, the firm invests 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 (Cleary et al., 2007, p.10. Proof of the Proposition 2 can be found in Cleary et al., 2007, Appendix, p.33). Figure 2. Investment as a Function of Internal Funds. Figure 2 shows relationships between investment (I) and internal funds (W). W can be both positive (if a firm successfully operates) and negative (if the firm experiences financial constraints). I is the firs-best level investment, which is pursued if W! I. The firm invests less, if W < I due to problems of asymmetric information. W and I have monotonic relations until certain level of W = W!, after which the firm invests more, even if internal funds decrease, as more investment leads to more revenue. Source: Cleary et al. (2007, p.11). Figure 2 is a graphic representation of Preposition 2. It shows the relationship between investment and internal funds, which is found to be U-shaped. Until a certain negative level of internal funds! W the relationship is monotonic, a firm s investment decreases if a firm experience more and more severe liquidity problems. Cleary et al. (2007) argue that such relationship is not caused by debt overhang or credit rationing, but rather due to risk of liquidation. This means that a firm s investment is not optimal (first-level investment), because lower investment requires lower debt, hence lower future repayment. If D is small, it leads to higher chances of a firm to stay solvent, 36

37 hence, not default on its obligations. But if we go further on the graph (beyond! W ) the relationship becomes nonmonotonic, which leads to the U-shaped investment curve. As it was mentioned before, lower investment requires lower debt, hence lower future repayment, or in other words, higher investment (holding internal funds equal) requires a greater investor s participation (larger D), hence larger future repayment, which, in turn, leads to a higher probability of going bankrupt. But at the same time larger investment leads to higher expected revenue, which affects a firm in two ways first, directly through higher revenues; second, decreasing marginal costs of debt financing. The higher the revenue from investment, the better chances of a firm to repay any amount of debt, and the more investor receives even if the firm defaults. Cleary et al. (2007) argue that the trade-off between these two effects determines the optimal level of investment of the firm. If W decreases, but the firm wants to keep the current level of investment, it has to borrow the needed amount, hence increasing D, and increasing the likelihood of default. Therefore, a firm s investment needs to be adjusted accordingly to moderate the effect of higher liquidation loss. Hence the tradeoff between cost and revenue effects plays a major role in this situation. In times of high W (but that is below I ) it is optimal to decrease I, if W decreases. This is due to the fact, that if the firm decreases I, it decreases D (if compared to the initial level of I), hence decreasing the likelihood of default. Gain from the reduction in expected liquidation losses outweighs the loss form decrease in I, as it is still close to I. But it is not always optimal for a firm to decrease I in response to decrease in W. As a firm s liquidity level worsens, the probability of going bankrupt increases, and it becomes more important to generate as much revenue as possible in order to be able to survive. Cleary et al. (2007) argue that when I is farther from I, the marginal profitability of investment increases. Hence, further cutting of investment becomes not beneficial for a firm. At the same time, decrease in W leads to higher probability of liquidation, which, in turn, motivates the investor to invest more, as he or she gets the whole revenue in case of default. That is why both parties benefit from increase in I. A firm gains from higher expected revenue, but the investor is more likely to break even, despite lending more money to a firm. As it is seen from the Figure 2,! W is a turningpoint, at which going into even more negative W leads to an increase in I. Such a trade- 37

38 off between revenue and cost effect leads to U-shaped relationship between investment and cash flow. As Cleary et al. (2007) explain the trade-off between cost and revenue effects is derived from the equation [6], which show how a change in I affects ˆ! and thus the probability of default (for more info see Cleary et al., 2007, p.12): d ˆ! di (3) = " ˆ!! $ F I ( I,! )# (! )d! + F I ( I, ˆ! )# (! )d! " 1! $ ˆ! % ˆ! ( 1 $ # (! )d! & '! ) * I, ˆ! Fˆ! ( ) [6] which (as F! > 0 ) is positive if:! # ( I,! )" (! )d! + # F I ( I, ˆ! )" (! )d! $ 1 < 0 [7] ˆ! F I! ˆ! Cleary et al. (2007) admit that Proposition 2 may sound slightly counterintuitive. The reason lies in the effect of financial constraints on the risk premium. The more financially constrained is a firm, the higher is the risk premium, which, in fact, is true for the model, as it can be seen from Proposition 3. The risk premium in the model is defined as: ( ) = D( I,W )! ( I! W ) i I,W I! W Proposition 3. The risk premium increases if W decreases, while either I or I W is held fixed (Cleary et al., 2007, p.13. Proof of the Proposition 2 can be found in Cleary et al., 2007, Appendix, p.35). Cleary et al. (2007) argue that it is important to distinguish between marginal and average costs of debt-financed investment. Marginal cost is non-monotonic, as it may lead to increase or decrease of investment (depending on the state of W), while average cost is monotonic. Hence, it is not correct to see the investment as a function of average cost, as investment may increase, even after increase in average cost. Final proposition concerns the information asymmetry between a firm and an investor, which arises when uncertainty about a firm s future payoff is introduced. As it was mentioned earlier, expected future payoff is denoted by! 2, while liquidation value by L, but their realized values are now stochastic. Both variables are zero with probability!, and! 2 / (1 " #) and L / (1! ") respectively with probability 1! ". It is important to note that the firm recognize! 2, after it realizes the revenue. If future payoff is recognized as zero, there is no reason for the firm to repay the debt to the [8] 38

39 investor. Cleary et al. (2007) stress that this extension of the model helps to capture the situation, in which two identical firms may have different information asymmetry problems. The Optimal debt contract specified in Proposition 1 is fully compatible with this extension if! 2 is large, the firm does not want to lose this payoff, hence is induced to repay the debt, but if! 2 = 0, no payment can be enforced. by: The firm in this situation tries to maximize: E!" F( I,! )# $ % ( 1%! )D +! 2 [9] subject to the investor s participation constraint, which in this situation is defined ( ) ˆ# % ( 1! " ) F( I,# ) + D! F I,# ( + L # & ' $ 2 ) *, (# )d# + 1! " ( ( )) D! I + W = 0 [10] ( ) 1! - ˆ# The result is still the same as it was if W! I the firm invests at the first best level of investment, because no debt is required. When W < I borrowing becomes problematic (if! > 0 ), as information asymmetry increased. The investor now requires additional premium for the risk (thus, anymore D! I " W ). Proposition 4 shows that change in W and in! influences the investment, but in a different manner. Proposition 4. For infinitesimal increases in α, (a) If I W! 0, then I! < 0 ; that is, whenever investment is increasing in internal funds, it is decreasing in the degree of informational asymmetry. (b) For W sufficiently close to I, we have I W! > 0 ; i.e., the sensitivity of investment with respect to the level of internal funds is increasing in!. (c) The risk premium increases for any given I (Cleary et al., 2007, p.16. Proof of the Proposition 4 can be found in Cleary et al., 2007, Appendix, p.36). Figure 3 graphically represents Proposition 4 arguments. It shows that, when there are no problems of asymmetry (! = 0 ), the graph is the same as in Figure 2, but when asymmetric information comes in (! > 0 ) the situation slightly changes. If W is significantly large (W! I ), the right end is unchanged, if compared to! = 0 situation. If W is small but positive, marginal cost of external funds increases, which leads to a situation of less investment. For cases of negative W, revenue effect becomes more important, and the firm increases investment earlier than comparing to! = 0 curve. It is seen from Figure 3 that when! is increasing from 0 to 1, the curve shifts downwards 39

40 and inwards. Figure 3. Investment with Asymmetric Information Figure 3 shows relationships between investment (I) and internal funds (W). The curve α = 0 is the same as in Figure 2. The curve α = 1 depicts situation with more asymmetric information, hence external financing becomes costlier, and the firms invests less, given the same amount of W, as in α = 0 situation. When W is negative, the firm invests even more after a particular point, as revenue generation becomes essential. Source: Cleary et al. (2007, p.17). Cleary et al. (2007) also argue that they could have used a different approach for watching at asymmetric information. The authors use stochastic values for L and! 2 - being them zero or non-zero. The agency problems get worse if L is lower, compared with! 2, hence it has been possible to analyse situation, in which some part of the private benefit could be transferred to the investor. In this case the part of the future benefit, which cannot be transferred, is defined by! = (! 2 " L) /! 2. The authors argue that this approach yields the same results, if adapted to the Proposition 4 scenario. This approach could be pretty interesting, as it is easy to identify possible proxies of!, i.e. intangible assets, R&D, etc., but it would definitely complicate the model, as because L or/and! 2 would vary; the fundamental value of the project would vary as well. Besides four propositions, Cleary et al. (2007) identify three main assumptions of the model. The assumptions are as follows: 1. Presence of capital market imperfections 2. Investment is scalable 3. Cost of debt is determined endogenously (by the investor s participation constraint) 40

41 There is, in fact, the fourth assumption, which is also important for identifying the U-shaped investment curve internal funds may be negative (W < 0). 3.2 Lifecycle Model Theoretical model presented above is a solid foundation for our research. The research itself, however, mostly concerns lifecycle aspect of internal funds vs. investment relationship (several methods applied by Cleary et al. (2007) and Yildiz (2010), though, are used in the analysis for comparison matters). The analysis starts from estimating the following basic equation, which was originally employed by Cleary et al. (2007): I it =!+" 1 CF it +" 2 MB it +" 3 SG it +# it [11] where:! : a constant term I it : investment of a firm i at time t, with i=1,,n; t=1,,t : cash flow of a firm i at time t, with i=1,,n; t=1,,t CF it MB it SG it : market-to-book ratio of a firm i at time t, with i=1,,n; t=1,,t : sales growth of a firm i at time t, with i=1,,n; t=1,,t! it : an error term It is important to mention that investment and cash flow variables were initially controlled for heteroskedasticity by dividing by beginning-of-period net fixed assets. The problem of heteroskedasticity may arise due to the fact that analysed firms are of very different sizes. Please refer to Table 2 in Appendix III for more information about calculation and interpretation of variables. As the paper uses several measures of internal funds, the analysis continues by substituting variable CF it with NLA it in equation [11]: where: I it =! + " 1 NLA it + " 2 MB it + " 3 SG it + # it [12] NLA it : net liquid assets of a firm i at time t, with i=1,,n; t=1,,t The equations are extended by introducing squared variables CF it 2 and NLA it 2 in equation [11] and [12] respectively. Above-mentioned procedures are basically a replication of steps made by Cleary et al. (2007). In order to capture lifecycle effect, we introduce the variable Age it, which is added into equations [11] and [12]. We also examine these trade-offs using Age it 2 41

42 variable, as it would explain if the trend develops at an increasing or decreasing rate. Important to note that we are making these regressions on different subsamples, namely the whole unbalanced sample, positive/negative cash flow and net liquid assets subsamples, and balanced subsample, to see if it shows different results. Furthermore, data sample is split into five 10-year age groups to determine phases of lifecycle development as defined by Miller and Friesen (1984). Basic equations [11] and [12] are regressed on each of five year groups for the whole sample, as well as for positive and negative cash flow and net liquid assets subsamples. To check the robustness of the results we also add dummy variables for age into the main regressions [11] and [12], and analyse the whole data sample and subsamples with age dummy variables. The dummies are created for each age group, e.g. Dummy 1-10 variable is equal to 1, if firm s age is in range from 1 to 10. Otherwise it is equal to zero. Hence, the following equations are estimated for every subsample, namely for the whole sample, for positive/negative cash flow and net liquid assets subsamples, and for balanced subsample: I it =! + " 1 CF it + " 2 MB it + " 3 SG it + " 4 Dummy1!10 it + +" 5 Dummy11! 20 it + " 6 Dummy21! 30 it + " 7 Dummy31! 40 it + # it [13] I it =! + " 1 NLA it + " 2 MB it + " 3 SG it + " 4 Dummy1!10 it + +" 5 Dummy11! 20 it + " 6 Dummy21! 30 it + " 7 Dummy31! 40 it + # it [14] Regressions include only four dummies, as inclusion of all five dummy variables will lead to perfect multicollinearity issues. 4 Data This section explains the data that is used in our analysis. We are continuing the analysis by Yildiz (2010), who replicated the model by Cleary et al. (2007) on the data sample of U.S. manufacturing and mining firms. Therefore, we decided to use a part of the data that is already researched by Yildiz (2010), namely manufacturing sector companies, as this kind of data showed more credible results, as well as manufacturing sector companies are the most widely used data in investment studies (see e.g. Jorgenson et al., 1970a, b; Fazzari et al., 1988; Hoshi et al., 1991; Hubbard et al., 1995; Guariglia, 2008 among others). That is why we are following the same techniques and using the same criteria for data extraction, as performed by Yildiz (2010). In order to construct the sample for the research, the data is taken from COMPUSTAT database (COMPUSTAT data extraction criteria can be found in Table 3 42

43 in Appendix IV). The COMPUSTAT database provides independent financial information on the companies from the United States of America (U.S.) and Canada. The data is taken from Standard and Poor s. The 19 data items are used to calculate variables for the analysis. The description of the items is available in Table 4 in Appendix V. As according to Cleary et al. (2007) and Yildiz (2010), the main variables of the analysis are investment, as a dependent variable, and different liquidity proxies, marketto-book ratio, and sales growth, as independent variables. All variables and the way they are calculated are presented in Table 2 in Appendix III. We use two main proxies of internal funds, namely cash flow (CF/K) and net liquid assets (NLA/K). We are aware of other liquidity proxies, such as cash stock, and sums of cash flow and cash stock, and cash flow and net liquid assets, but we do not use them in the analysis due to the fact that these measurements of internal funds were studied by Cleary et al. (2007) (sum of cash flow and net liquid assets) and Yildiz (2010) (all three proxies), and both papers find pretty similar results as compared to using just cash flow or net liquid assets in the analysis. Both studies consider these additional proxies as secondary (of minor importance) variables. Moreover, according to Cleary et al. (2007) cash stock is not an appropriate measurement of internal funds, as it cannot be negative, therefore is useless in capturing the U-shaped investment curve. On the other hand, Cleary et al. (2007) themselves show with their illustrative example (see Appendix I for details) that there is no need for internal funds to be negative to capture the U- shaped relations between internal funds and investment. But as Yildiz (2010) did not find any significant difference between main and supportive liquidity proxies, we are going to use these (cash stock and sums of CF/K, NLA/K, and cash stock) variables only as secondary ones in order to construct the data sample, as well as to get more insight into financial situations of studied firms. According to Hubbard (1998), cash flow is the most widely used liquidity measurement in the literature, as it accounts for periodical changes in internal funds. However, it does not include accessible funds in the beginning of a period. It is important to mention that net liquid assets is a beginning-of-period variable, hence it accounts for the option, which is missing by cash flow. Net liquid assets, in turn, do not take into consideration immediate changes of internal funds, as well as may incorporate externally generated funds. 43

44 Nevertheless, we admit that additional proxies of liquidity could be useful in such kind of research, which studies the U-shaped curve from lifecycle perspective, as it is interesting to see how all different proxies behave as age of firms changes. But the scope of this particular analysis is limited to cash flow and net liquid assets, hence one of possible suggestions for further research in the field may be inclusion of other liquidity measurements. Besides these main variables (investment, cash flow, net liquid assets, market-tobook, and sales growth), we add another variable, which is essential for our research age variable. To determine the age we used the approach by Loderer and Waelchli (2009). In this approach the age of a firm is the age of a legal entity. A legal entity does not always correspond to what the most of economists mean by a saying a firm. Nevertheless, we continue with this approach as this makes the comparison of the age of firms more comparable, as in such case the data is taken from the same and the only source of information. Loderer and Waelchli (2009) state that the most studies, which use the age of a firm as an important variable in the analysis, define the age of a firm as a number of years after the IPO plus one (to avoid the age of zero). To determine the IPO date we use CRSP and COMPUSTAT databases. The databases have an actual data item called IPO date, but the data is present only for a small number of firms, but for majority of companies the data item is blank (even for some well known companies). In connection with this issue, we define the IPO date to be the date when a firm appeared on these databases. In particular, we chose the earliest date among the two. The earliest available data in CRSP is as of Therefore, the oldest firm at the beginning of our sample in 1991 can be 67 years. Correspondingly, at the end of the sample period (in 2008) the oldest company can be 84 years. Age is a very broad variable, which, arguably, can be a proxy for many other things, thus, leading to spurious regressions. Loderer and Waelchli (2009) show in their paper that age represents much more than declining uncertainty, declining concentration of ownership, industry age or the age of management. Moreover, use of post-ipo age is not a proxy for post-ipo rigidities, according to the authors. Therefore, age is selfsufficient variable and does not require control variables. We will see whether or not our results differ from other literature to confirm that age works well to identify aging of a company in a general sense. 44

45 The initial sample for the analysis consists of 40,358 observations, and 19 data items, from 1990 till 2008, while the first firm-year observations (year 1990) are used to calculate lagged variables, therefore the analysed time period includes 18 years. As we are focusing on the U.S. companies, all non-u.s. firms are deleted from the sample. Further, as we require calculation of lagged variables, all the firms with only one firmyear observation are removed as well. Furthermore, the observation is removed in the following cases: 1. Total assets value (data_6) is equal zero. 2. Sales volume (data_12) is equal zero. 3. Missing values (#VALUE! and/or #DIV/0!) for either of the main variables: investment, cash flow, net liquid assets, cash stock, market-to-book ratio, and sales growth. 4. Sales growth exceeds 100% (Almeida, Campello, and Weisbach (2004) find this step helpful to eliminate observations, which incorporate substantial corporate events, such as M&A, reorganization or transformation of the firm etc.). 5. Observations lying above the 99 th percentile and below the 1 st percentile for the following variables: investment, cash flow, net liquid assets, and cash stock (this step eliminates outliers, which could be caused by data entry mistakes). 6. Missing values (#VALUE! and/or #DIV/0!) for the following variables: payout ratio, leverage ratio, current ratio, return on equity (ROE), and interest coverage ratio (TIE) (this step is executed only by Yildiz (2010), while Cleary et al. (2007) vary sample among different regressions; equal and consistent data sample makes it easier to compare results between different parts of the analysis, that s why we implement the approach by Yildiz (2010)). 7. The data on age is found neither in CRSP nor in COMPUSTAT. As a result, the total number of firm-year observations is 19,180, and all in all 2,092 manufacturing firms are used in the analysis. The descriptive statistics of all variables is provided in the next section. As it was mentioned earlier, we are basically continuing the analysis made by Yildiz (2010), who replicated the model by Cleary et al. (2007) on the sample of U.S. manufacturing and mining firms. We, in turn, try to concentrate on the lifecycle aspect of the U-shaped investment curve, hence we decided to work with already analysed data set. Mining sector firms, as according to Yildiz (2010), showed relatively weak results, that is why we focus our analysis on U.S. manufacturing industry data used by Yildiz 45

46 (2010). We are doing so, because the data is already researched, and U-shaped relationship between investment and internal funds is already found in this data. Nevertheless, we will compare our sample with the one from Yildiz (2010) to make sure that we are on the right track, as well as to avoid data gathering mistakes. Thus, our sample includes only firms from the manufacturing industry as defined in Standard Industry Classification system (SIC codes 20XX-39XX). In particular, this is the industry that consists of companies that use mechanical or chemical transformation of materials to produce their products (the detailed description of manufacturing sector classification is presented in Table 5 in Appendix VI). As discussed above, manufacturing is the most widely used industry in research studying investment behaviour of firms (see e.g. Jorgenson et al., 1970a,b; Fazzari et al., 1988; Hoshi et al., 1991; Hubbard et al., 1995; Guariglia, 2008 among others). Particularly U.S. manufacturing sector is the largest in the world, which, according to current data, is employing about 12 million people, and produces 1.6 trillion USD in manufactured products (NIST MEP, 2010). That all leads to a substantial data sample from a single industry in a single market, which is very important in our case, as enclosure of different sectors of economy or different countries into data sample could lead to industry- or market specific biased results. 5 Empirical Results This section provides the analysis of descriptive statistics of data sample. It further presents graphic based analysis of relations between investment and internal funds. Afterwards, it proceeds with regression analysis and split sample regression analysis. At last, empirical results section provides the analysis of key financial ratios of different age groups. 5.1 Descriptive Statistics As it was mentioned before, the data sample consists of 6 main variables: investment, cash flow, net liquid assets, market-to-book ratio, and sales growth as according to Cleary et al. (2007) and Yildiz (2010), and age variable, which was extracted specifically for the purpose of this research. 46

47 Table 6. Descriptive Statistics Table 6 presents descriptive statistics (means and medians) for data sample used in the analysis. Statistics is calculated for four samples: the whole unbalanced sample (all observations), positive CF/K subsample, negative CF/K subsample, and balanced subsample. Panel A presents means and medians for main variables, which are used in standard regression analysis, and split sample regression analysis. Panel B presents variables for supportive variables, which are used to extract the data sample, as well as for supportive analysis. Sample period is 18 years ( ). Panel A. Means and Medians for Main Variables Sample All observations Positive CF/K Negative CF/K Balanced Variables Mean Median Mean Median Mean Median Mean Median Investment (I/K) Cash flow (CF/K) Net liquid assets (NLA/K) Market-to-book ratio Sales growth Age Number of Firms 2,092 1,613 1, Number of Observations 19,180 14,006 5,174 5,166 Panel B. Means and Medians for Supportive Variables Sample All observations Positive CF/K Negative CF/K Balanced Variables Mean Median Mean Median Mean Median Mean Median Cash stock (Cash/K) Cash flow + Net liquid assets ((CF+NLA)/K) Cash flow + Cash stock ((CF+Cash)/K) Payout ratio Leverage ratio Current ratio Return on equity Interest coverage ratio Number of Firms 2,092 1,613 1, Number of Observations 19,180 14,006 5,174 5,166 Source: calculated by authors. Table 6 shows the summary statistics of the variables used in the analysis. We divide observations according to the Cleary et al. (2007) approach. Division into firmyears with positive and negative cash flows helps to identify financially constrained firms, being the most used approach in the literature (see e.g. Hubbard, 1998). The sample is also divided into unbalanced (all observations, positive and negative CF/K) and balanced (firms with entire time period observations) data sections. The statistics is in line with our expectations, as it does not differ a lot from the one from Cleary et al. (2007) or Yildiz (2010). We have to admit, though, that despite using 47

48 the same data extraction approach as Yildiz (2010); we got a slightly different sample. The total number of firms in our data sample is equal to 2,092, while the total number of observations is 19,180; while Yildiz (2010) has it 2,084 and 19,077 correspondingly. We explain this difference by the fact that the steps of removing observations (see Section 4) could be implemented in a different order by Yildiz (2010) (it especially concerns the 5 th step - elimination of observations lying above 99 th percentile and below 1 st percentile for some of the main variables, because if implemented before some other steps, or after all steps, it could lead to a different final data sample). The other cause of this slight difference in descriptive statistics is that we excluded observations if it was impossible to determine the age of the firm. The main difference between our sample and the one from Yildiz (2010), however, concerns balanced section of the data sample. Number of firms in balanced section of our data sample is 291, and number of observations is 5,166. Yildiz (2010) has the following numbers for the same characteristics 626 and 8,596. Recall that data sample is called balanced if it includes data for all observations for entire sample period in our case, if every firm has all data for all firm-years (18 in our case). It means that dividing number of observations by number of firms we have to get 18. Dividing 8,596 by 626 (data from Yildiz, 2010) we get 13.7, which is not even close to 18; hence, on average, every firm had less than 14 firm-year observations, not 18, as should be in case of balanced data. In our case (dividing 5,166 by 291) we get 17.8 (it is not 18 due to the fact that firms use different approaches in defining fiscal years, meaning that for some companies fiscal year 2007 ended in when in it was 2008 calendar year. But we defined years using fiscal year approach; therefore for some companies the last year in the analysis was 2007, and in total these companies had 17 firm-year observations, not 18 as the majority of firms. We believe that this one-year difference should not cause a problem, as most of the companies had all 18 firm-year observations). Pointing out these differences, we think that it is necessary to replicate several steps of the analysis made by Yildiz (2010) with our data. It is interesting to see if these differences between samples will lead to similar of different outcome, as compared to Yildiz (2010). Reverting to the descriptive statistics results we see that financially weak firms (with negative CF/K) are investing virtually as much as firms with no financial constraints (mean I/K = 0.29 versus mean I/K = 0.27, respectively), a well as the whole sample (mean I/K = 0.28). This conclusion is in line with our expectations that 48

49 financially weak firms invest a lot, despite a shortage of liquidity. Firms with negative cash flow also have high market-to-book ratio with mean of 3.90 (the highest across all subsamples), while sales growth is strongly negative It means that even being currently unsuccessful, these firms have high growth potential that is anticipated by the market (as statistics does not differ a lot from other papers, more thorough analysis of descriptive statistics of main variables can be found in Cleary et al., 2007, and Yildiz, 2010). The variable in which we are mostly interested in is age. As we see from Table 6 the average age for the whole sample is almost 23 years. Average age for firms with positive internal funds is slightly less than 26 years, while firms with negative internal funds have average age of almost 15 years. Balanced subsample consists of mature firms with average age of 36 years. Such allocation of firm-year observations into these subsamples is not a surprise. Financially weak firms are relatively young, with negative sales growth and high market-to-book ratio, as they have ambitions and growth potential. They invest heavily, as investors anticipate ambitions and motivations of young firms and provide financial support. On the other hand, balanced subsample consists of firms, which have firm-year observations for the entire time period of the analysis. It is difficult to satisfy all the criteria of sample construction for young and immature firms; hence the subsample contains firms with average age of 36 years. These firms invest less than young (and less than the whole sample as well) with mean of investment being 0.22; they have an average market-to-book ratio of 1.76, and sales growth of As we described before, we divide data sample into five 10-year age groups (the last group, representing decline phase, includes firm-year observations from Age 41 and till the oldest possible observation with age of 84 years). We associate these groups with five lifecycle stages of Miller and Friesen (1984) as follows: Age 1-10 birth phase Age revival phase Age growth phase Age decline phase Age maturity phase However, this distribution is only tentative. We will search for different features of these groups in our analysis to see if such classification holds. Regardless of our findings, distribution of the sample to age groups will be helpful to see how ageing of a firm affects its investment-liquidity relationship. 49

50 The distribution of firm-year observations is provided in Figure 4. Most of the firms are young and growing companies. More than half of observations, namely 58.2%, are firms with age under 21, while adult firms amount to 41.8% of the whole sample with 2,715 (14.2%) firms of age 21-30; 2,248 (11.7%) firms of age 31-40; and 3,054 (15.9%) firms of age Figure 4. Distribution of Observations for 10-year Age Groups Figure 4 depicts the distribution of firm-year observations of the whole unbalanced sample for five 10-year age groups. Each age group reflects one phase of a firm s development, namely birth, growth, maturity, revival, and decline (Miller and Friesen, 1984). The whole sample consists of 19,180 firm-year observations. Source: plotted by authors. As our data sample covers the time period of 18 years, a single firm can be present at maximum in three age groups in the analysis, but as we use unbalanced data sample for our main analysis, most of the companies present in just one or two age groups. To check if our results are robust, as well as to devote more attention to young firms, as mostly these firms represent financially weak part of the sample, we decided to make another splitting of our data sample on the basis of firms age. The division into five 5-year age groups is as follows: Age 1-5 Age Age 6-10 Age Age Figure 5 depicts the distribution of observations over five 5-year age groups (the last group, as in case of 10-year allocation, consists of firms aged 21 years and more). Almost 42% of all observations, namely 8,017, are located in the last group, which is 50