Table 13, FHP Sample Summary Statistics 1973-84 Correlations Among Variables FHP Sample Selected Financial Ratio Means FHP Sam ple Group Turnover Statistics FHP Sample Percentage Grou
LI'IXRATURE REVIEW: INVESTMENT POLICY AND FINANCIAI, FACTORS 6
The Basic Irrelevance Argument
Modigliani and Miller (1958) demonstrate that a f m ' s fmancid structure wiU not affect its market value in a world of perfect and complete capital markets An important implication is that reai investment decisions will be made based on available growth opponunities, with no reference to financial factors such as liquidity, leverage or dividend payrnenu This result is the foundation of the neoclassical theory of investment as postulated by Jorgenson (1963) and Jorgenson and HaII (1967) They demonsuate that a f m ' s optimal investment policy can be solved without reference to fmancial factors
This approach assumes that a l l f m s face a cost of capital that is determined by financial markets, independent of the firms' particular linancial structure.
The Q-Theory of Investment
The q-theory of investment, based upon the work of Brainard and Tobin (1968) and Tobin ( 1969)' represents an extension of the basic neoclassical argument The underlying principle of this approach, according to Brainard and Tobin (1968)' is that 'the market valuation of equities, relative to the replacement cost of the physical assets they represent, is the major determinant of new investment Investment is stimulated when capital is valued more highly in the market than it costs to produce it."
Aiternatively, one could Say that investment is encouraged when market yields on equity are low, relative to the real returns on investment in physical assets
Hayas hi ( 1982) presents an important fomulation of the neoclassical mode1 based on the q-theory approach He demonstrates that under the assumption of convex costs of adjusting capital stock, fm investment opportunities can be sumrnarized by the market valuation of the f m ' s capital stock He goes on to prove that under certain assumptions, the ratio of market value of capital stock to its replacement cost (Le the Tobin's q value) will be 'the' underlying variable affecthg investment demand
Hayashi assumes that: (i) managers maximize the expected present value of fiture profits from capital; (ii) capital is the only quasi-fmed factor; (iii) convex costs of adjusting the capital stock; and (iv) new capital resulting fiom investment becomes productive within the year Under these conditions, the value of the f i is given by: s.t K L = ( l - G ) K k - , + I ,
In this equation, i and t denote the fm and time period; K is the beginning of penod capital stock; n is the profit hinction; 0 is an exogenous shock to the profit hinction; C is the cost-of-adjustment hinction: 1 is investment; p, is the tu-adjusted relative pnce of capital goods; A is an exogenous shock to the adjustment cost function; 6 is the constant rate of depreciation; and E(mIR,) is the expectations opentor conditional on the information set Cl available to firm i at time t
The frrst-order condition for maximizing (1) with respect to investrnent is: m rvhere q- = (1 - 6IS[nK (KiJts, 6i,t+S) - CK (Ki,l+s y 11 l t s=O
The right-hand tenn in equation (2) is just marginal q, while equation (3) defines q as the present discounted value of profits from new futed capital investment Hayashi then specifùes adjustment cos& to be linearly homogeneous in investment and capital (so that marginal and average q will be equal) He uses the following convenient parametenzation that adheres to these constraints:
C ( I i f , K , ) = ( a / 2 ) ( 1 , / K i ~ - a j - ~ i ~ ] 2 ~ , ~ (4) This adjustment cosi function allows for a technology shock, d , which may be correkted with the production shock, 0
Substituting the adjustment cost specification in (4) into equation (2) yields the following investment specirication: where is an expectation error As noted previously, under certain conditions, average
Q constructed from fmancial market data may be used as a proxy for marginal q, and the relation between investment and Q cm be expressed as: where b = (1 I a) and Q is the tax-adjusted value of Tobin's q (as in Summers(l98 1)) This is the central equation of the q-theory of investment that descnbes investment behavior for f m s O perating in fnc tionless capital markets
2.2 CAPITAL MARKET IMPERFECTIONS AND THE RELEVANCE OF
2.2.1 'Accelerator7 Models of Investrnent Behavior
The idea that financial structure and output are intemelated has a long histow dating back as far as the time of the Great Depressioo The coliapse of the fmancial system dong with real activity prompted Fisher (1933) to argue that poorly performing fuiancial markets contributed to the seventy of the economic downturn He argued that the high leverage in the economy immediately preceding 1929 had both a direct and an indirect impact on the economy In particular, he noted that the large number of bankruptcies caused by the business downtum, was directly related to aggregate leverage The bankruptcies, in nim f i h e r deepened the recession In addition, the deterioration in the economy led to a redistribution in wealth from debtors to creditors which had a significant indirect impact on the economy He went on to argue that this indirect effect had an even greater impact on the downturn, because it affected alI borrowers, not just those on the verge of bankniptcy The decline in net worth induced borrowers to reduce current expenditures and future commiunents, which sent the economy into steeper levels of deflation, He suggested that the simultaneous deterioration in bo rrower balance sheets and rapidly falling levels of output and prices, offered support for this 'debt-deflation' story
Gurley and Shaw ( 1955) demonstrate the importance of the interaction between fuiancial structure and real activity They argue that 'financial capacity', as measured by borrowers' ability to absorb debt without having to reduce current or future spending commitments, is an important determinant of aggregate demand This implies that balance sheets, which are the key determinants of financial capacity, play an important role in affecting investment levels Strong balance sheet positions have the ability to accelerate business cycles by enhancing spending behavior, while weak balance sheets will have the opposite effect Investment models based on this notion that financial factors can mzgnify initiai shocks to the economy are often referred to as "accelerator" models of investment Subsequent theoretical works, which are discussed below, focus on the contribution of capital market imperfections to this accelerator effect on investment
2.2-2 The Role for Interna1 Funds
Donaldson's groundbreaking 1961 study asserted that firms heavily rely on internal sources for funding This reliance has been consistently observed, demonstrating the significance of internal financing in the corporate landscape.
"Management strongly favored intemal generation as a source of new hnds even to the exclusion of extemal fùnds except for occasional unavoidable 'bulges' in the need for funds." He also observed that even if extemal funds were required, issuing new stock would be the last choice of management
This section outiines the potential importance of intemal funds for fm investrnent policy in the presence of capital market imperfections Figure 1 outlines the basic neoclassical argument graphicaily, similar to the approach used in most introductory finance textbooks Firm investment is plotted along the horizontal axis and the f m ' s weighted average cost of capital (WACC) is plotted along the vertical axis
The neoclassicd mode1 depicts the f m ' s supply curve of hnds (S) as a horizontal line at the f m ' s cost of capital, which is given by the market risk-adjusted red rate of interest The demand curve for capital (D) is downward sloping to reflect the fact that a decrease in the cost of funds wiU increase the fum's desired Ievel of investment, The location of
D is a hnction of the fum's available investment opportunities and an increase (decrease) in these opportunities will shift D to the right (lefi)
The optimal hvestment in capital asseu (1*) occurs at the intersection of D and S, where the marginal return on capital investment equals the market interest rate An
FIGURE 1 Firm Investment and Cost of Capital
Investment decisions are driven by interest rates and investment opportunities A decline in market rates or an increase in available investment opportunities can lead to an increase in desired capital stock The opportunity cost of internal funds is assumed to be equal to the cost of external funds, determined by market rates Internal funds availability does not directly impact investment decisions, which are based on market interest rates and investment availability.
The neoclassical argument assumes b a t fm managers act in the best interest of fum stakeholders It also assumes managers and extemal suppliers of fùnds have the saine information regarding the quantit y and quality of investment op portuniries available to the f m These assumptions serve as a point of departure for models that demonstrate the potential importance of intemal hnds in the investment decision These models argue that fm managers have supenor information regarding fm prospects andor that their objectives do not always cohcide with those of the F m stakeholders This implies the cost of extemal funds will exceed that of internal funds, due to costs associated with adverse selection a d o r mord hazard As a result, the fxrn's cost of capital will increase beyond the point at which internal funds are exhausted (W) and we will observe the fum's supply curve of hnds (S') to be upward sloping beyond W (as depicted graphically in Figure 1)
The resulting capital investment level (1') will be less than the optimal level (1*) that is obtained in fnctionless markets, unless the h ' s intemal resources are greater than or equal to I* In addition, higher marginal information costs wilI result in a steeper upward-sloping portion of the supply curve (S'), which implies increased investment sensitivity to the availability of intemal hnds
Agency Models
Agency models argue that extemal siippliers of funds require higher retums to compensate them for agency costs Agency costs include the costs of monitoring managerial actions and the potential moral hazard associated with management's control over the allocation of investment funds This Iine of reasoning was pioneered by Jensen and MeckIing (1976) in their seminal article regarding principal-agent relationships They argue that agency costs are unavoidable because managers will be encouraged to appropnate corporate resources, in the fonn of perquisites, whenever they are not sole owners of the resources under their controL The total costs consist of monitoring
This is analogous to the condition tbar WA* in Figure 1
The analysis of bank fraud losses identifies three major components: expenditures by the principals, bonding expenditures incurred by the agent, and residual loss The residual loss is caused by inadequate monitoring and bonding processes that fail to control management behavior.
There are also significant agency costs associated with debt financing These consist oE (i) sub-optimal investment decisions which are made due to the impact of debt; (ü) monitoring and bonding expenditures; and (iii) baokniptcy and reorganization costs The first two costs arise due to the incentive effects associated with the use of leverage Equity holders of highly levered firms will prefer that management engage in high risk, hi@ return projects, since the benefits will accrue primarily to them while the brunt of the cost is borne by the f m ' s debt holders Bondholders will attempt to protect iheir interests through the use of monitoring procedures and debt covenants in order to prevent this expropriation of their wealth- The result is that managerid decisions are likely to be sub-optimal, due to the restricted set of actions that will now be available to them
Jensen ( 1986) defmes free cash fi0 w as "cash flow in excess of that required to fund all projects that have positive net present values when discounted at the relevant cost of capital." He argues that "managers have incentives to cause f m s to grow beyond optimal size" since "growth increases managers' power by increasing the resources under their control." He also notes that management compensation is typically tied to growth
As a result, managers wiil avoid large payouts to shareholders, since this would increase the likelihood of having to obtain extemal iünds through capital markets that would scrutinize their behavior Agency costs arise because management m u t be motivated to pay out the cash flows to shareholders, rather t h a investing at rates of retum below the f i ' s cost of capital
Jensen's model demonstrates that firms tend to increase investment when ample cash flows are available During the late 1970s and early 1980s, oil firms exemplified this concept, experiencing significant cash flow increases due to soaring oil prices Instead of distributing these cash flows to shareholders, many oil firms opted to invest in research and development, despite facing returns below the cost of capital Notably, they also embarked on large-scale diversification programs, which ultimately yielded mixed results.
These agency arguments have been extended in recent years Bernanke and Gertler (1989) outline a frnancial accekrator mode1 of investment based on the existence of agency costs They suggest that "higher borrower net worth reduces the agency costs of financing real capital investments Business upturns irnprove net wonh, lower agency costs, and increase investment, which amplifies the uptum; vice versa for downtums." This effect is due to the fact that it is easier for f m s to obtain outside fimds when their balance sheets are healthy, which occurs in a greater proportion of f m s during perùods of strong economic activity
Bernanke and Geder (1990) argue that as net worth decreases, the borrower will have less available hnds to contribute to investment projects This increases the divergence of interests between the borrower and potentid creditors, and results in an increase in agency costs Their model aüows entrepreneurs to undertake costly evaluations of investment projects The evaluations provide them with better information regarding the quality of these projects than is available to extemal providers of funds
This informational asymmetry creates an agency problem that increases the cost of extemal finance and affects the entrepreneurs' ~villingness to evaluate projects in the first place
Borrowers face greater opportunity costs of proceeding with a project as t h e i net worth increases, which makes them more selective This increases the expected profitability of the projects and reduces agency costs When borrower net worth decreases, borrowers have less incentive to engage in costly project evaluations As a result, the quaiity of investment projects fall and agency costs rise This leads to an equilibnum where "both the quantity of investment spending and its expected retum will be sensitive to the 'creditworthiness' of borrowers (as reflected in their net worth positions) Indeed, if borrower net worth is low enough, there c m be a complete collapse of investment"
Gertler (1992) expanded upon prior research by introducing multi-period relationships between borrowers and lenders This implies that credit constraints are influenced not only by a firm's current net worth but also by its anticipated future cash flows Consequently, financial capacity, defined as the maximum debt burden an entrepreneur can bear without project suspension, is determined by future expectations Building on financial accelerator theories, Gertler demonstrated that financial capacity significantly impacts economic growth.
Empirical evidence suggests that net worth and investment expenditures have a positive correlation In particular, studies have found that decreases in cash flow and collateral value lead to reduced investment, regardless of opportunities for investment Furthermore, evidence indicates that increased balance sheet strength results in increased firm investment, indicating that financial health plays a significant role in investment decisions.
Investment Decisions in an Option Theoretic Frarnework 21
A related literature demonstrates that firm investment decisions c m be viewed in the context of option theory Framework This approach assumes the output prices associated with long-term investments may be viewed as stochastic variables, whose future values are uncenain These variables impact the appropnate discount rates as well as the net present value (NPV) of available investment opportunities In this context, Brennan and Schwartz (1985) argue that the "dynamic aspect of the investment decision is closely related to the problem of determining the optimal strategy for exercising an option on a share of common stock."
In the absence of dividends, it is well known that one should never exercise a cal1 option pnor to expiration Ho wever, in the real world, companies may face situations where it will be advantageous to hvest at an early stage Trigeorgis (199 1) identifies two possible situations where a fm may "fmd it justifiable to exercise its real option to invest at an early stage": (i) when the present value of irnmediate cash flows (acting as dividends), exceeds the value of waiting; and, (ii) when the firm c m preempt cornpetitor entry-
McDonald and Siegel (1988) consider the option value associated with postponing irreversible investments They argue that the appropriate investment nile should compare the value of investing today with the value of investing at all possible times in the future
Similar to a fuiancial option, uicreased risk will increase the value of this real option, which provides greater incentive to delay the iovestment expenditure Based on their analysis, f i s that adopt zero NPV projects 'too early', may forgo as much a s 10-20% of the potential value of such investment projects This argument suggests there is value in 'kaiting to invest."
Trigeorgis (1991) suggests that management tlexibility regarding the optimal timing of hvesunent project initiations results in an "expanded NPV" framework This implies the value of a project can be thought of as the sum of the standard NPV of expected cash flows, plus an option premium refiecting the value of this timing option During periods of greater uncertainty and rising interest rates, it may pay fvms to delay positive NPV projects On the other hand, during periods of low uncertainty and low interest rates, the option value may justify entrance into negative NPV projects
This Iiterature shows the potential benefit of delaying investment projects is especially high during periods of high uncertainty This rationale is consistent with the familiar theme that € i s increase their investment outlays in response to declines in their cost of capital Based on the discussion in sections 2.2.1 through 2.2.4 one would expect investment tu increase in response to increases in the availability of intemal funds (which will be less expensive than external funds in the presence of capital market imperfections) This implies it will ofien be advantageous for firms to defer capital speriding until interna1 resources becorne available
Meyer and Kuh (1957) examuied the impact of severai fmancial variables on the invesunent decisions of approximately 750 f m s in tweIve m a n u f a c t u ~ g industries over the 1946-1950 penod They found that increases in sales, profit levels and depreciation expense had a significant positive impact on fm investment These effects were more pronounced during periods of low fm Liquidity- They also found that these f m s were reluctant to mise extemal fuiance These results support the relevance of fuiancial factors in the investment process Meyer and Kuh have been criticized for not controllhg for the availability of growth opportunities, which implies the fmancial variables may appear to be signifcant because they also serve as proxies for growth potential
The empirical studies of Jorgenson and Siebert (1968) and Elliot (1973) presented contrasting views on investment decision-making Jorgenson and Siebert found investment decisions in 15 manufacturing firms aligned with the neoclassical model, emphasizing real factors However, Elliot's subsequent study contradicted this by demonstrating the superiority of the liquidity model in accounting for investment decisions in a larger sample of 173 firms.
Bohn and Reiss (1988) added hie1 to the debate by demonstrating that all standard models of investment can be rejected in cornparison to at least one other modeL
Fazzari, Hubbard, and Petersen (1988) made two significant contributions to empirical research in this field Firstly, they employed the beginning-of-period Tobin's q value for a firm as a proxy for growth opportunities, addressing criticisms of previous studies by reducing the informational content of financial variables designed to proxy net worth Secondly, based on evidence provided by Bernanke, Bohn, and Reiss (1988), they argued that investment models may fail under certain circumstances due to market imperfections These imperfections can create a wedge between the cost of internal and external finance, which affects certain classes of firms more susceptibility.
Under these circumstances, EUio t's fmding that financial affects are relevant for a relatively broad sample of f m s need not be inconsistent with Jorgenson and Siebert's results that r e d factors best explain investment for a group of weil-known mature f m s This notion leads them to depart from previous empincal approaches by focusing attention on the differences in investment behavior exhibited by groups of f r m s that are formed according to thek apparent susceptibility to capital market imperfection pro blems
FHP88 examine two theoretical predictions that arise fiom the discussion in section 2.2.2 and are based on the assumption that equation (6)' as denved by Hayashi (1982)' is represeatative of the neoclassical model They fùist hypothesize that the q-theory of investrnent as specifùed in equation (6), should explain investment relatively weil for f m s with high net worth relative to their desired capital stock Alternatively, one would expect this model to fail for firms with low net worth relative to desired capital stock, who wiU face a high premium for external finance Secondly, they examine the hypothesis that F m liquidity should not affect the hvestment decisions of the high net worth group to the eaent it does for the low net worth group of firms
FHP88 use Value Line data for 422 large U.S manufacturing firms over the 1970-
84 period Their selection criteria are designed to eliminate f m s in a fmancial distress situation in order to focus on the investrnent and fmancing decisions of fvms that have wealth to distribute With this objective in mind, they select only f m s with a complete history of available fmancial information fiom 1969 to 1984 In addition, only f m s experiencing positive sales growth over the entire period were included
They analyze differences in investment behavior by f m s classified accordhg to earnuigs retention" According to FHP88, f m s with higher retention ratios face higher informational asyrnmetry problems and are more likely to be liquidity consuained They argue that if the cost disadvantage of external Fiance is small, retention practices should reveal little or nothing about investment Under this scenario f m s will simply use extemal fmancùig to smooth investment when inremal finance fiuctuates, regardless of their dividend policy However, if the cost disadvantage is signif~cant, f m s that retain in particuiar, FHP88 dassify f m s into the following three groups based on their dividend bebavior over the 1970-84 period: (1) those that have a ratio of dividends to income of less than 0.10 for at least ten years; (2) those that have a dividend-incorne ratio between 0.10 and 0.20 for at Ieast ten years; and (3) al1 other f m s and invest most of their income, may have no low-cost source of investment fmance, and their investrnent should be driven by fluctuations in cash flow
FHP8 8 nin the following regression for the 'q' , neoclassical, and sales accelerator models of investment5:
( I I K ) , =ah + PiIf ( X I K ) i r ] + P 2 [ g ( C F f K ) , l + ~ i t (7) where 1, represents investment in plant and equipment for fm i during period t, K is the beginning of period book value for net property, plant, and equipment, g(CF / K) is a function of current cash flow which mesures f i liquidity, f (X / K) is a function of variables related to investment opportunities, and E, is an error term For example, accordhg to the q-theory of investment, f (X / K) is represented by a f m ' s Tobin's marginal q value
Contrary to the predictions of neoclassical models, financial health plays a significant role in determining the impact of liquidity on investment Companies that rely heavily on internal financing are more sensitive to cash flow fluctuations compared to mature firms with high dividend payouts This suggests a financing hierarchy where internal funds have a cost advantage over external sources Financial effects on investment are particularly pronounced when capital markets are less efficient, implying that financing constraints can significantly influence investment decisions These findings support the notion that liquidity and financial health can materially impact firm behavior even in the presence of efficient capital markets.
Fazzari, Hubbard and Petersen (1988)
Fazzari, Hubbard and Petersen ( 1988) make two important contributions to the empirùcal work in this literature First, they employ the beginning of period Tobin's q value for a fm as a proxy for growth opportunities This alleviates criticisms of previous studies by reducing the informational content of financial variables that are designed to proxy net worth Their second major innovation is based on the evidence provided by Bernanice, Bohn and Reiss (1988) that al1 models of investment fail under certain circumstances Fazzari, Hubbard and Petersen (hereafter FHP88) argue that this result is not surprising if certain classes of f m s are more susceptible to the market imperfections that drive a wedge between the cost of interna1 and extemal fiance
Under these circumstances, EUio t's fmding that financial affects are relevant for a relatively broad sample of f m s need not be inconsistent with Jorgenson and Siebert's results that r e d factors best explain investment for a group of weil-known mature f m s This notion leads them to depart from previous empincal approaches by focusing attention on the differences in investment behavior exhibited by groups of f r m s that are formed according to thek apparent susceptibility to capital market imperfection pro blems
FHP88 examine two theoretical predictions that arise fiom the discussion in section 2.2.2 and are based on the assumption that equation (6)' as denved by Hayashi (1982)' is represeatative of the neoclassical model They fùist hypothesize that the q-theory of investrnent as specifùed in equation (6), should explain investment relatively weil for f m s with high net worth relative to their desired capital stock Alternatively, one would expect this model to fail for firms with low net worth relative to desired capital stock, who wiU face a high premium for external finance Secondly, they examine the hypothesis that F m liquidity should not affect the hvestment decisions of the high net worth group to the eaent it does for the low net worth group of firms
FHP88 use Value Line data for 422 large U.S manufacturing firms over the 1970-
Financial distress situations necessitate meticulous screening to identify firms with sustainable investment and financing strategies To this end, a comprehensive analysis of financial data spanning 1969-1984 is conducted, focusing on firms with consistently positive sales growth throughout the period This stringent selection process ensures that the analysis is confined to firms with a proven track record of financial stability and wealth distribution potential.
Firms with high earnings retention face greater information asymmetry and liquidity constraints, according to FHP88 If external financing costs are low, investment behavior should not be influenced by retention practices However, firms that retain earnings, particularly those with low dividend payouts, may face significant external financing costs and rely on internal cash flow for investment.
FHP8 8 nin the following regression for the 'q' , neoclassical, and sales accelerator models of investment5:
( I I K ) , =ah + PiIf ( X I K ) i r ] + P 2 [ g ( C F f K ) , l + ~ i t (7) where 1, represents investment in plant and equipment for fm i during period t, K is the beginning of period book value for net property, plant, and equipment, g(CF / K) is a function of current cash flow which mesures f i liquidity, f (X / K) is a function of variables related to investment opportunities, and E, is an error term For example, accordhg to the q-theory of investment, f (X / K) is represented by a f m ' s Tobin's marginal q value
Contrary to the predictions of neocIassica1 models, FHP88 determine that the coefficients on the liquidity variable are no t insignificant More importantiy, investment of f m that exhaust all their interna1 finance is found to be much more sensitive to fluctuations in cash How than that of mature, high dividend f m s They attribute these results to a fmancing hierarchy in which intemal funds have a cost advantage over new equity and debt FHP88 also document a diff'erence across f m s in the sensitivity of investment to balance sheet variables measuring liquidity Financial effects on investment appear to be greatest at times wheo capital market information problerns are lùkely to be most severe for high-retention f m s , reiriforcing their thesis that fmancing coostraints in capital markets affect investment These results are robust to a wide
Q mOde1s emphasize market valuations of fm assets as the determinant of investment, sales accelerator models suggest fluctuations in sales or output motivate changes in capital spending, whiIe neoclassical rnodds combine m a u r e s of output and the cost of capital to explain investment demand varợery of estimation techniques and specifkations FHP88 suggest their resub probably understate the me effect of cash flows on investment, since large mature F i s constitute a greater proportion of their Value Line sample than they do in the aggregate economy
Critics of the Tobin's q adjustment for growth opportunities highlight its imperfections Estimating asset replacement costs and using average q as a proxy for marginal q pose challenges Additionally, concerns exist about the reliability of stock prices in reflecting future cash flows Furthermore, the classification scheme based on dividend behavior may be endogenous, as firms relying heavily on internal funding tend to exhibit lower payout ratios.
Subsequent Studies
Despite the criticisms, FHP88 remains the most influential study of this issue in the existing literature Subsequent studies have c o n f i e d their central result by dividing samples according to other a priori measures of fmancial constraint for cornparison purposs For example, Hoshi, Kashyap and Scharfstein (1991) examine the behavior of
145 Iapanese manufacturing f m s that were continuously listed on the Tokyo Stock Exchange between 1965 and 1986 They compare the investment-cash flow sensitivity of
24 firms that are not members of a 'Keiretsu' to 121 f m s that are members of a
6 These criticisms were raised by Poterba (1988) and Blinder (1988) in their original discussions of the paper Refer to Schaller (1993) for an insigbtful discussion of these issues
'Keiretsu' and are presumed to be less fmancially coastrained They conclude that the investment of the coastrained (non-Keiretsu) firms is much more sensitive to Êùrm cash flow, which supports the conclusions of FHP88
Oliner and Rudebusch (1992) utilized two parallel panel sets covering the 1977-83 period The first panel comprised 99 firms, primarily listed on the NYSE, while the second panel included 21 over-the-counter firms obtained from Moody's OTC Industrial Manual.
They run the basic FHP regression after preclassification of f m s according to a varùety of criteria- Their results suggest that investment is most closely related to cash flow for f i s that are young, whose stocks are traded over-the-counter, and exhibit insider trading behavior consistent with privately-held information Schdler ( 1993) categorizes
212 Canadian f m s , over the 1973-86 penod, according to: age; ownership concentration; manufacturing versus non-manufacturing; and group (e-g Bronfman and Reichman groups) versus independent f m s His regression results indicate investment for young, independent, manufacturing fums, with dispersed ownership concentration are the most sensitive to cash flow,
Fazzari and Petersen (1993) add changes in working capital to the basic FHP88 specifcation Since changes in working capital are positively correlated with sales and profits, one would expect h e m to have a positive coefficient in the investment regression However, the existence of financial constraints may cause firms to draw down working capital to mitigate temporarily the effect of an adverse shock to cash flow on investment Using the W 8 8 panel data, Favari and Petersen fmd that the estimated working- capital-investment coefficient is negative for the Iow-payout f m s , which h p l i e s that liquid assets perform a "buffer stock" role for fmancially constrained f m They suggest these results casts doubt on the notion that the estimated effect of cash flow on inves tment largely reflects omitted shifts in investment demand
In a related inquiry, Calorniris, Himmelberg and Wachtel(1995) use bond ratings or access to bond and commercial paper markets to sort f m s according to fmancing costs They fmd that f m s with no ratings or with lower credit ratings (which tend to be srnaDer f m s with lower dividend payout), hold larger stocks of liquid assets and display much more cash flow sensitivity of hvestment in working capital These findings support the existence of a "buffer stock" role for liquid assets for fmancially constrained fiims
An alternative ap proach for testing the relations hip between investment and liquidity is utilized by Whited (1992), and Bond and Meghir (1994) They employ an Euler equation approach to directly test the f ~ s t order condition of an intertemporai maxirnization problem, which does not require the measurement of Tobin's q It is implemented by imposing an exogenous constraint on extemal fiance and testing whether that constraint is binding for a pa~ticular group of F i s Whited uses a sample of 325 U.S manufacturing f m s for the 1972-86 penod, while Bond and Meghir use an unbalanced panel of 626 U.K manufacturing companies for the 1974-86 period Both of these studies fud the exogeneous h a n c e constraint to be particularly binding for the constrained groups of f m s which supports the existence of a fmancing hierarchy arnong constrained f i s
A related body of empirical literature, d e a h g with fm capital structure decisions, is ais0 supportive of the existence of fmiancing heirarchies For exarnple,
Mayer (1 990) examines the sources of indusuy finance of eight developed countries f%om
1970 to 1985 and reveals the following stylized facts regarding global corporate fùnancing behavior: (i) retentions are the dominant source of fmancing in a l l countries; (ü) no countries raise substantial amounts from securities markets in the form of short-term securities, bonds, or equities; (G) the majority of external fmancing cornes from baok loans in all countries; and (iv) s m d - and medium-shed f m s rely more heavily on bank fmancing than larger f m Shyam-Sunder and Myers ( 1995) analyze COMPUSTAT flow of funds data for 157 U.S fums from 1971 to 1989 and find evidence that f m s foilow a pecking order approach to obtaining funds Booth, Aivazian, Dernirguc-Kunt and Maksimovic (1 997) present empincal evidence fro m developing countries over the 1980-1990 period that also supports the existence of a pecking order approach to
The foregoing discussion implies consent regarding the existence of a fiiancing hierarchy that is mos t prevalent among constrained frrns However, Kaplan and Zingales (1997) challenge the generality of this conclusion They perform an in-depth analysis of the 49 low-dividend paying fums identified by FHP as having extremely high investment-cash flow sensitivity Kaplan and Zingales (hereafier KZ) use a combination of qualitative and quantitative information fkom annual reports to rank F i s in terms of the? apparent degree of fimancial constraht In particular, they use data from letters to shareholders, management discussions of operatioos and liquidity (when avaiiable), fiancial statements, notes to those statements for each fm-year, and hancial ratios obtained from the COMPUSTAT database,
Companies are considered financially constrained when external funding constraints hinder their investment decisions, while firms with high cash positions and fewer lender restrictions are typically unconstrained Financial health, characterized by low debt and high cash, is associated with unconstrained companies Despite annual reevaluations, KZ classifies companies into three constraint groups for regression analysis, which assumes that companies maintain their financial status throughout the period.
KZ provide cross-sectional evidence that suggests their classification scheme successfully captures the financial constraint characteristics of fims For example, they categorize a higher percentage of f m s in the fmancially constrained category during the recessionary 1974-75 years In addition, variables such as median cash Bow, Tobin's q, interest coverage, and 'slack' (cash plus unused h e of credit) decrease monotonically across their categories A criticisrn of the onginal FHP paper Fist raised by Poterba
(1988), and examined in greater detail by Gilchrist and Hirnmelberg (1 995), is that their sorting cnterion is correlated with mismeasurement of Q KZ suggest their research design is less subject to this criticism, since their classification scheme is based on direct observation, which should more accurately rneasure the unobservable variable
Contrary to FHP88's prediction that this entire group would face severe fmancial constraints, KZ fmd "in only 15% of f i - y e a r s is there some question as to a firm's ability to access internal or external funds to increase investment In fact, alrnost 40% of the sample frms could have increased investment in every yea of the sample penod." Contrary to previo us researc h, the Ieas t fmanciaily constrained f m s exhibit the greatest investment-cash flow scnsitivity This pattern is found to persist for the entire sample penod, for sub-penods, and for individual years They suggest these controversiai results
"capture general features of the relationship between corporate investment and cash flow", and are not specifc to the sample or techniques utilized They c o n f m the robustness of their results by repeating the analysis usine: (i) alternative definitions of degree of fiancial constraint based on variables such as interest coverage, dividend restrictions, debt covenants, and 'sIack'; (ii) four alternative definitions of investment; and (üi) the Euler equation approach used by Bond and Meghir (1994)
PanelDatasets
Panel data provides multiple observations for several hdividuals over time As a result, it has both a cross-sectional and time series component Blundell, Bond and Meghir (1992) suggest that using panel data for individual f m s to examine investment behavior has several advaotages over aggregate t h e senes studies including: " biases resulting from aggregation across f m s are eliminated; cross-sectional variation contributes to the precision of parameter estimates; several variables of interest c m be measured more accurately at the fm level; and heterogeneity across fvms in, for example, effective tax rates can be explicitly taken into account." More irnportantly, it allows the examination of c r o s s - f i differences in investment behavior
Panel data sets in general are susceptible to two important sources of bias: selectivity bias -and heterogeneity bias Selectivity bias arises due to the selection critena imposed by the researcher in forming his sample The use of such critena implies that the sample is not randomly selected from the population This bias is unavoidable for studies of fm level investment behavior, since the available data sets tend to include large, welI-known fms In addition severai empincal studies, including this one, attempt to focus on f m s that have cash to allocate by imposing selection criteria to e h i n a t e extreme observations This bias cannot be remedied using existing econometric techniques, however, the use of similar selection criteria across the studies implies that cornparison of their results is reasonable16 Heterogeneity bias results fiom differences in regression parameters among cross-sectional unis (fms) and Me-series units The discussion below focuses on methods for dealing with this type of b i s
Pooled Ordinary Least Squares (OLS) Estimation
This approach pools observations from all 'N' cross-sectional units and from d l 'T' tirne periods, resulting in N*T total observations The estimates are then obtained using ordinary Ieast squares (OLS), according to the follo wing specification: y, =a+p.r, + r i i t , i=1, , N und t =1, , T (1 1)
This approach is easy to apply and makes use of all available observations However, it is based on the assumption that all cross-sectional units have the sarne intercepts and slope coefficients, which do not change through t h e
Biases arise in OLS estimates due to differences in intercepts and slopes across individu&, and across time This highlights the importance of accounting for these frm- specific and tirne-specfic effects in the present study, since theory predicts that invesunent behavior will differ across firms and will change through tirne in response to l6 1 WOUM note bat my sample is less subject to this aititism than FHP88 and KZ, since it is mudi larger and is diversifieci aaoss indusmes and by exchange listing This matter is addressed in greater detail in chapter 6 changes in ecooomic conditions The next two sub-sections describe the two basic specifications used for panel data to account for differences across individuals and through Ume units These approaches both maintain the assumption of common dope coefficients across individu& and t h e however, they allow for variation in the intercepts As such, they provide a simple, yet reasonably general alternative to the assumption of common parameters across the sample They are based on the wumption that the effects of numerous ornitted individual tirne-varying variables are unimportant individually, but the sum of these effects may be significant.
Random Effects Estimation
Random effects models treat individual and time specific effects as an additional source of random variation It is assumed that: some of the omitted variables represent factors that are unique to both the individual unợts and time periods associated with the given observations; other factors affect certain individuals in similar fashion tbrough tirne; and, other factors are unique to a given time period and affect all individuals similarly Random effects rnodels assume the individual intercepts are randomly distnbuted around a mean value ( p ), with the random fluctuations consisting of both an individual component (ai ) and a the-v-g component (4) These fluctuations are assumed to have an expected value of zero, with fi'ied variance and are assumed to be uncorrelated with each other and with other error terms
The residuals ( v , ) consist of three components and can be represented as:
Residuals, denoted as a', are assumed to have fixed variances that are independent of each other These models, known as variance components or error-components models, recognize that the variance of y, given x, is the sum of the three individual variances:
The random effects regression model is given by: y, = p + pxir +ai +A, + r i , (1 4)
Generalized l e s t squares estimation provides the best linear unbiased estimates for this model" The use of the random effects model is appropnate, when the effects c m be viewed as random drawings from a population, the researcher is interested in population characteristics and when the number of cross-sectional units (N) is large However, it provides inconsistent estimates when there are omitted variables, which is likely the case l7 Refer to pages 47û475 of Greene (1 993) for computational details of the random effects estimacor for the regressions used in the investment literature In addition, investment theory does no t predict that residuals of the basic FHP regression equation will be uncorrelated with the regressors For these and other reasons, the empirical investment literature has focussed on the use of fixed effects regression estimates, which are discussed below
Fixed effects estimation generalizes the constant intercept and slope mode1 (OLS) by allowing the intercept to Vary across individu& and through tirne This is accomplished by introducing dummy variables to account for the cffects of those omitted variables that are specifc to individual units but rernain constant across time (ai), and for variables that are speciCic to each time period but affect a l l cross-sectionai units the same (A, ) The fuced effects regression mode1 can be expressed as:
The only required distributional assumption is that the error terms are independently and identicdl y distnbuted More irnponantly, these estimates are designed to annihilate the effects of omitted individual-specific and tirne-specfic variables
It may be very cumbersome to maintain N individual dumrny variables and T time dummy variables, particularly when there are a large number of cross-sectional units included in the panel data set As a result, there are two cornmonly used f ~ e d effects estimation techniques, both of whic h transfo rm the actual observations before running regressions using the transformed variables The 'within' or 'demeaned' estirnator subtracts individuai means and time period means kom the actud obsenrations, and then performs OLS on the transformed variables Altematively, one can emplo y the 'first dit3erence9 estimator, which eliminates individual cross-sectional effects by taking first dùfferences of the observations, and uses t h e durnmy variables to account for t h e - specific effects
F i e d effects estimation is very costly in terms of degrees of fieedom lost, and it ignores between unit information However, it also has several advantages that make it well suited for estimating coefficients related to firm investment panel data Decisions regarding levels of investment in capital equipment depend cntically upon initial conditions and expectations of future conditions, due to the magnitude of the associated capital adjustment costs Modeling expectations, which cannot be directly observed, implies omitted variables may be important Further, there is no reason to believe that individual effects will be uncorrelated with regressors, a s assumed in random effects estimation In addition, the prirnary source of available Company information is su bject to the measurement problems associated with using accounting data to measure capital stock and determine Tobin's q Estimating Tobin's q requires the use of market values of equity, which relies on the implicit assumption of efficient capital markets and this may introduce additional measurement problems
The basic investment regression equation used by W 8 8 and several subsequent studies, is given by:
( Z I K ) , =a, + &Q, +&(CFIK), + E , (1 6 ) where Q, is the fùrm's beghning of period Tobin's q ratio and CF I K represents firm cash flow during the period divided by its beginning of period book value of capital assets Given the nature of the investment process, it is like1y that the residual term will contain fm-specific and tirne-specific components In addition, theory predicts that the current value of Q wiU be correlated with current shocks (residuals) This irnpiies that estimators that rely on strùctly exogeneous regressors, such as the random effects estimator, should be avoided The importance of this matter is enhanced by the entry and exit of fkms from available data sources Entry into these databases is usually reserved for companies with public stock listings, while exit generally occurs as the result of bankniptcy or takeover Both entry and exit processes will therefore be related to fm investment decisions, and are likely to be correlated with 'shocks' to the investment equation
To account for unobserved correlations between investment and independent variables, and to capture business cycle influences, this study estimates the basic FHP88 regression equation using firm and year fixed effects Specifically, these fixed effects control for unobserved firm-specific characteristics and time-varying factors that may affect investment decisions.
1 represents investment in plant and equipment during penod t, K is the beginning of period book value for net property, plant, and equipment, CF represents current penod cash flow to the fxm as measured by net income plus depreciation plus the change in deferred taxes; and MIB represents the f m ' s common equity market-to-book ratio based on the previous year's actual market value at year end and is used as a proxy for growth op portunities
The use of market-to-book ratio to proxy for growth o p p o d t i e s follows the approach of KZ This differs fiom FHP88 who calculate 'Q' based on replacement costs and the average market value over the last quarter of the previous year, however, Perfect and Wiles (1994) indicate improvements obtained from the more involved computation of Q are limited In addition, KZ point out that using year end market values c m only be regarded as a methodological improvement, since "the FHP88 rneasure will not distinguish between a f i whose stock price declines from 20 to IO and a fm whose stock price increases from 10 to 20 in the last quarter." Current period cash flow (CF), scaled by 'K', is used to measure the liquidity variable This follows the specification of most previous studies including FHP88 and KZ, and facilitates comparison of resulis with previous evidence
The equation is estirnated using f i e d effects, which is consistent with the preceding discussion, and facilitates comparison with previous studies, whose estimates were obtained using this approach Results are reported for the 'demeaned' or 'within' fixed fm and year estimates, which coincides with estimates presented by FHP88 and
Consistent estimates obtained from alternative panel data estimation techniques, including "first differenced" fixed effects estimates, indicate the absence of significant errors in variables problems This consistency in estimates across different techniques provides evidence of the robustness of the results.
'' OLS estimates have d s o not been reported, however they are also consistent with the reported fixed effecis estimates in t m s of magnitude and observed patterns across groups
3.5 EMPIRICAL LEVELS OF SI[GNI[FICANCE
A major fucus of previous studies has been to compare the investment-liquidity sensitivities across different groups of firms However, traditional tests designed to detect differences in coefficients are not appropriate since the error terms likely violate the required assumptions Traditional tests are generally designed for testing changes in parameters across tirne series data, where it may sometimes be reasonable to assume no heteroscedasticity in the resulting residuals Panel data, with its emphasis on cross- sectional data, likely violates the required assumptions For example, the Chow test requires that the disturbance variance be the same for both regressions, while the standard Wald test requires independence of the error terms These conditions are unlikely to be satisfied by panel data residuals
Due to the inadequacy of existhg tests, conclusions regarding the existence of differences across groups in investment-liquidity sensitivity, have been largely based on observing differences in magnitude and level of significance of the coefficient for the liquidity variable in regression estimates The present study uses simulation evidence to determine the significance of observed differences in coefficient estimates The process uses a bootstrapping procedure to calculate empirical p-values that estimate the likeliho od of O btaining the O bserved differences in coefficient estirnates, if the tme coefficients are, in fact, equal
Estimation in This S tudy
To account for unobserved relationships between investment and independent variables, and capture business cycle influences, the present study employs fixed firm and year effects in the estimation of the basic FHP88 regression equation.
1 represents investment in plant and equipment during penod t, K is the beginning of period book value for net property, plant, and equipment, CF represents current penod cash flow to the fxm as measured by net income plus depreciation plus the change in deferred taxes; and MIB represents the f m ' s common equity market-to-book ratio based on the previous year's actual market value at year end and is used as a proxy for growth op portunities
The use of market-to-book ratio to proxy for growth o p p o d t i e s follows the approach of KZ This differs fiom FHP88 who calculate 'Q' based on replacement costs and the average market value over the last quarter of the previous year, however, Perfect and Wiles (1994) indicate improvements obtained from the more involved computation of Q are limited In addition, KZ point out that using year end market values c m only be regarded as a methodological improvement, since "the FHP88 rneasure will not distinguish between a f i whose stock price declines from 20 to IO and a fm whose stock price increases from 10 to 20 in the last quarter." Current period cash flow (CF), scaled by 'K', is used to measure the liquidity variable This follows the specification of most previous studies including FHP88 and KZ, and facilitates comparison of resulis with previous evidence
The equation is estirnated using f i e d effects, which is consistent with the preceding discussion, and facilitates comparison with previous studies, whose estimates were obtained using this approach Results are reported for the 'demeaned' or 'within' fixed fm and year estimates, which coincides with estimates presented by FHP88 and
KZ 'First daerenced' fixed effects estimates were detennined, but are not reported here- Generaily, the estimates are consistent with the 'within' estimates in terms of magnitude and observed patterns across groups18 Hsiao (1 986), Griliches and Hausman (1986), and Schaller (1993) suggest that obtaining consistent estimates from alternative panel data estimation techniques, provides evidence of no senous errors in variables problems
'' OLS estimates have d s o not been reported, however they are also consistent with the reported fixed effecis estimates in t m s of magnitude and observed patterns across groups
3.5 EMPIRICAL LEVELS OF SI[GNI[FICANCE
A major fucus of previous studies has been to compare the investment-liquidity sensitivities across different groups of firms However, traditional tests designed to detect differences in coefficients are not appropriate since the error terms likely violate the required assumptions Traditional tests are generally designed for testing changes in parameters across tirne series data, where it may sometimes be reasonable to assume no heteroscedasticity in the resulting residuals Panel data, with its emphasis on cross- sectional data, likely violates the required assumptions For example, the Chow test requires that the disturbance variance be the same for both regressions, while the standard Wald test requires independence of the error terms These conditions are unlikely to be satisfied by panel data residuals
Due to the inadequacy of existhg tests, conclusions regarding the existence of differences across groups in investment-liquidity sensitivity, have been largely based on observing differences in magnitude and level of significance of the coefficient for the liquidity variable in regression estimates The present study uses simulation evidence to determine the significance of observed differences in coefficient estimates The process uses a bootstrapping procedure to calculate empirical p-values that estimate the likeliho od of O btaining the O bserved differences in coefficient estirnates, if the tme coefficients are, in fact, equal
Observations are pooled fkom the two groups whose coefficient estimates are to be compared Denoting 'nl' and 'n2' as the number of annual observations available from each group, we end up with a total of 'nl+n2' observations every year Each simulation randomly selects 'nl ' and '132' observations each year from the pooled distribution and assigns them to group 1 and group 2 respectively Coefficient estirnates are then detennined for each group using these observations, and this procedure is repeated 5000 times The empuical p-value is the percentage of simulations where the difference between coefficient estimates ( d i ) exceeds the actual observed difference in coefficient estimates (dsmple ) This p-value tests against the one-tailed alternative hypothesis that the coefficient of one group is greater than that of the other group
( H 1 : d >O) For example, a p-value of 0.01 indicates only 50 out of 5000 simulated outcomes exceeded the sample result, which implies the sample difference is significant, and supports the notion that d > 019 l9 Ernpirical p-values, denoted as P(absolute), were J s o obtained by testing the nuU hypothesis of equal coefficients ( H o :d = O ) againsr the Iwo-tailed alternative hypothesis of non-equality of coefficients
( H l : d t O ) This test is approptiate when theory does not predict which coefficient should be large, and would be appropriate according to the neoclassical theory of investment A p-value of P(absolute)=0.02 indicates that only 1% (or 50 out of 5000) of the sợmulated absolute value ciifferences (Id ) e x ~ e d e d the absolute value of the sample différence ( 1 dsample 1) This irnpiies d O since it is highly unlikely that the observed difference is a random occurrence These pvalues have not ken reported, however, they su bstan tiate the reported 'one-tailed' p-vaiues In particular, these 'two-tailed' pvalues were found to be significant in every case when the one-tailed p-values were found to be significant
FAZZARI, HUBBARD AND PETERSEN (1988) REPLICATION
Data was obtained fiom the 1991 and 1994 COMPUSTAT annual tapes for industrial f i s for the purpose of replicating the original FHP88 study These tapes include information for the previous 20 years, which meant I was unable to obtain financial information pnor to 1972, since the 1991 tapes were the oldest available to me Since the 1972 year-end items are required for regression purposes, my analysis period was reduced to 1973-84 versus the 1970-84 penod used by FHP88 In addition, I was only able to obtain data for 245 frms out of their original sample of 422 f m s (Le 58% of the original number of F i s ) Combining these two factors, 1 ended up with only 464 of the original number of fmn year observations Data availability d s o imposed a great deal of survivorship bias on the sample, since alrnost all of the F i s 1 was able to locate, had a complete history up until the end of 1991 Not surprisingly, my results were not in complete agreement with those of the original study
Group 1 of the original FHP88 study included 49 h s (12% of their total sample of 422), wbose dividend payout ratios were between O and 10% for 10 of the 14 years exarnined 1 was only able to obtain 27 (55%) of these 49 f m s , which resulted in only 44% of the original FHP88 observations (324 versus 735) Group 2 in the original study included 39 furns (9% of their total sample) whose payout ratios were between 10 and
20% for 10 of 15 years, while 1 obtained 19 of these f m s (49%), resulting in 39% of the observations (228 versus 585) Group 3 of the original study consisted of the 334 remaining f m s (79% of their total sample) with higher payout ratios, while 1 obtained
199 of these f i s (60%), resulting in 48% of the original observations (2388 versus 5010)
Aside from, the small number of f m s (245 versus 1080), the nature of the sample obtained for the replication has several other features that make it distinct from the US sample I examine in chapter 6 First, it de& with a completely different tirne period Second, this sample consists of large, well-known companies This is illustrated by the fact that the average net fixed assets figure for the f m s in this sample was $1.88 billion at the end of 1994 This is over twice as large a s that of my sarnple of IO80 U.S f m s , which had a mean net fixed asset figure of $779 million at the end of 1994 In addition, the FHP88 sample is not diversified by industry, consisting of all manufacturing fums, and the majonty of f m s listed their stock on the New York Stock Exchange
The frms in the replication study displayed much higher average annual growth in net fùxed assets than my U.S sample (8.7% versus 3.7%) They also had much higher debt ratios (4 1 Q versus 22%), which is consistent with the larger average fm size Not surprisingly, the replication sample consisted of a much higher proportion of fnms that increased dividends (6 1 W versus 39%) It also consisted of a higher proportion of fiims that decreased dividends (16% versus 7%), which appears curious at Fust glance
However, this is consistent with the fact that rhis sample contained a higher proportion of
F i s in FHP Group 3 ('high' payout) category (79% versus 47%), and a much lower percentage in the FHP group 1 ('low' payout) category (12% versus 38%) This implies
53 the h sin the replication sample would have more opportunhies to cut dividends, since they are generally higher in the frst place
FAZZARI HUBBARD AND PETERSEN (1988) REPLICATION 52
THE CANADIAN SAMPLE 75
Group Characteristics
Firms are classified using three approaches for purposes of this study: (i) according to theû original classification by FHP88 based on dividend behavior during the 1988-94 period; (ii) according to another rneasure based on dividend payout that allows fum classification to Vary every year in response to changing dividend payout ratios; and
(iii) according to the discriminant score approach described in section 3.3.2
The FHP groups were formed similarly to the original FHP88 study Al1 48 firms with dividend payout ratios between O and 10% for 5 of the 7 years examined were assigned to FHP Group 1, the 5 fiims with payout ratios between 10 and 20% for 5 of 7 years were assigned to FHP Group 2 and the 87 f m s with payout ratios above 20% for
A11 financial variables are for the beginning of Gscd year, except for cash Bow and investment which represent ibn cash flow and capital expeaditures during period 't' n e discriminant score (2) is calcuIated using discriminant analysis accordhg to equation (9) A fui1 description of the variables is induded in Appendix 1
Dividend Group 1 includes firms whose dividend per share (DPS) increased in year 't', Dividend Group 2 includes fkms whose DPS decreased in year 't', while Dividend Group 3 inciudes firrns that had no change in
PANEL A Selected Financial Ratio Means (1988-94)
Total Sample Dividend Group 1 Dividend Group 2 Dividend Group 3
(increased dividend (decreased dividend (no change in per share) per share) dividend per siiare)
PANEL B Number of Firms per Dividend Group
The FKP Group 3 contained a significant proportion of firms with at least five out of seven years having negative or excessive payout ratios, indicating a different dividend behavior compared to the original FHP88 sample The current sample exhibits lower dividend payouts, with only 43% of firms classified as high payout versus 79% in the FHP88 sample This divergence can be attributed to the distinct time period and the inclusion of smaller and more diversified Canadian firms in the present study.
The second classification divides the sample into dividend payout groups, but allows fums to be reclassified every year in response to changes in their dividend payout ratio in the previous year This is consistent with the advocated approach of allowing fim status to be determined every period Firm-year observations are delegated to four groups: (i) those with zero dividend payout (the Payû group); (ii) those with O to 30% payout ratios (the Paye30 group); (üi) those with payout ratios greater than 30% (the
The sample consists of 455 observations in the PayO group (32% of the total), 322 in the Payc30 group (23% of the total), 457 in the Payp30 group (33% of the total), and 173 in the Pay Negative group (12% of the total).
The third approach classifies f m s into groups every year according to the fmancid constraint index (Z,,), which is determined using equation (9) The sample is divided into three categories as descnbed above: group 1 fiims which increase dividends
78 and are Ikely not fmancially constrained; group 2 firms which cut dividends and are likely fmancially constrained; and group 3 f m s which do not change dividend payments and are no t used for purposes of the discriminant analysis
Summary statistics for the 1988-94 period are provided in Table 9 for each of these groups The difference between f m s that increase and those that decrease dividends is much more pronounced in this sample than was the case for the FHP88 replication sample, which is consistent with the more heterogeneous nature of this sample Firms that cut dividends appear much more lkely to be fmancially constrained according to traditional financial ratios They have lower current ratios, higher debt ratios, lo wer Fued charge coverage, lower net income margins, lower market-to-book ratios, lower sales growth, and have lower SLACKK values than fvms which increased dividends Tabie 9 also shows the standard ratio pedormance for f r m s that did not increase or decrease dividends, was between the other two groups
Panel B of Table 9 indicates substantid changes in the number of Fims that increase and decrease dividends through the yean The largest number of f i i s increasing dividends (89) occurred in the pre-recessionary year of 1988, while the largest number of f i s cutting dividends (60) occurred in 1991 This provides additional evidence that Frm Fiancial status changes in response to business cycles, and suggests there are benefùts associated with reclassifying fm status every period.
Discriminant Analysis
The discriminant scores are detennined for the Canadian sample using equation (9) of chapter 3 The following beginning of period variables are used to proxy for liquidity, leverage, profitability and growth: current ratio, debt ratio, fixed charge coverage, net incorne margin, sales growth, and SLACWK Univariate significance levels indicate net income margin, sales growth, and debt ratio are all significant at the 1% significance level, while current ratio and fvted charge coverage are signircant at the 13% and 1 1 % levels Correlation coefficients presented in Table 10 indicate a strong correlation between the discriminant (2) score and net income margin (0.84), as well as with the debt ratio (-0.48) These relationships are very similar to those observed in the
FHP replication sample Unlike the FHP replication sample, we also observe a very strong positive correlation between the discriminant score and sales growth of 0.55
The variables do a good job of successfully predicting which f m s will cut or increase their dividends, with group 1 and group 2 f i s being properly classified 64% of the time Summary statistics for the predicted group classification of f i s are presented in Table 1 1 They indicate that fims have been successfully classified according to traditional fuiancial ratios Firms that have been classified as likely to increase dividends (Predicted Group l), appear much more solid than f i s that have been classified as likely to decrease dividends (Predicted Group 2), in terms of all fmancid variables reported
Correlations Among Variables (Canadian Sample)
Al1 financial variables are for the beginning of fiscal year, except for cash tlow and investment which represent firm cash flow and capital expenditures during period 't' Cash flow, investment and slack are al1 scaled by net fixed assers at the beginning of fiscal year 't', The discriminant score (2) is caiculated uskg discriminant anaiysis according to equation (9) A hiIl description of the variables is included in Appendix
Cash Current Debtl F ~ e d Invest- Market Net Sales SlacW Discri- Flow/ Ratio Total Charge ment/ t Income Growth Fixed minant Fixed Assets Cover- Fixed Book Margin (%) Assets Score
Selected Financial Ratio Means (Canadian Sample 1988-94)
Key financial variables are provided at the start of the fiscal year, apart from cash flow and investment, which represent firm cash flow and capital expenditures during period 't' The discriminant score (Z) is computed through discriminant analysis per equation (9) Variable details are presented in Appendix 1 Discriminant analysis categorizes firms into Predicted Group 1, predicted to increase dividends in year 't', and Predicted Group 2, predicted to decrease dividends per share (DPS) in year 't' Firms are classified into FC (financially constrained), PFC (partially financially constrained), and NFC (not financially constrained) groups based on discriminant scores Each year, firms with the lowest discriminant scores (bottom third) are categorized as FC, the middle third as PFC, and the highest third as NFC.
Predicted Predicted FC f i s PFC firms W C f i s
Group 1 Group 2 (financially (partially (not
(likely to (Iikely to consuained) financidl y financiaüy increase DPS) decrease DPS ) constrained) constrained)
Net Fuced Assets S75Sm S769m S626m SlOOOm S659m
The sample is divided into three groups of 67 f m every year according to their
Z,, value Every year, the firms with the highest Z scores are assigned to the NFC group, the ones with the Iowest values are assigned to the FC group, and the remaining f m s are assigned to the PFC group Table 1 1 includes summary statistics for these three groups that c o n f m the effectiveness of this approach in capturing desired cross-sectional properties Similar to the results for the FHP replication sample, the fmancial status of the NFC firms is superior to that of the PFC f m s , while the FC f m s appear to be more constrained than both the PFC and M T fms
Table 12 reports turnover rates for the NFC PFC and FC groups which average
33.8%, 46.5 % and 32.1 % per year These are much higher than those observed for the FHP replication sample Further, 74% (or 149) of the total 201 firms were classified as
W C in at least one year, with figures of 81% and 73% for the PFC and FC groups This c o n f m s that individual firm fmancial status does change ssignificantly from one year to the next In fact, only 1 f r m would have k e n classified as PFC for ail seven years, while only 15 and 12 would have been classified as W C and FC for the entire penod
Table 13 c o n f m s the efficiency of the classification scheme with respect to dividend changes, as the NFC group consists of the smallest proportion of firms that cut dividends, and a larger proportion of frms increasing dividends than the FC firms The NFC group includes a much Iarger proportion of resource companies than the other two groups, which could indicate that the classification scheme is picking up an industry effect In addition, the WC category contains a higher proportion of high payout f m s and a lower proportion of low payout f i s , according to dividend groups formed using the original FHP approach, or using the tirne-varying dividend classification scheme
Group Turnover Statistics (Canadian Sample)
Firms in gr ou^ at least once
# f m s in group for al1 7 vears
# firms in group for 6 of
# f m s in gr0UD for 5 of
# fms in grour, for 1 of
Percentage Group Compositions (Canadian Sample)
This sugge-sts there is some degree of commonality among classification schemes based on dividend behavior and those based upon direct observation of fmancial variables
TotalSampleandDividendPayoutGroups
Fked effects estimates, based on equation (17) of chapter 3 are presented for the entire Canadian sample in Table 14 The evidence suggests that fiim investment decisions are sensitive to investment opportunities as proxied by market-to-book, but are even more sensitive to liquidity The estimated coefficients are 0.023 for market-to-book and 0.034 for the cash flow terrn These differ somewhat from those obtained by Schaller (1993) in his examination of 2 12 Canadian f m s over the 1973- 1986 penod, who obtains fixed effects coefficient estimates of 0.007 and 0.242 for Tobin's q and C F K This may be attributable to the different tirne penods being examined, as the estimates are quite close to the esrimates of 0.043 and 0.058 for Tobin's q and CF/K that are obtained by Cummins, Hassett and Hubbard (1996) for the Canadian f'irms they examine over the
1982- 1992 penodu The adjusted R-squared value of 1 -84% for the entire sample is quite low in cornparison to estimates in previous studies and indicates that we must view the regression results with caution" interestingly the CFlK coefficient estimate obtained by Cummins, Hassett and Hubbard is not signifiant, which is inconsistent with the results of most previous studies
'' For example the adjusted R-square value for the entire sample obtained by Schailer is 20.4% while Cummins, Hassett and Hubbard obtain a value of 1 1.4%
Regression Estimates for the Total Sample and for the FBP Dividend Groups
Within the 1988-94 period, firms characterized by payout ratios between 10% and 20% (FHP Group 2) exhibit higher market-to-book and cash flow/net fixed asset ratios than firms with payout ratios between 0% and 10% (FHP Group 1), as evidenced by p-values of 0.4384 and 0.1894, respectively These p-values indicate a statistically significant difference at the 43.84% and 18.94% levels, respectively.
Market-to-Book Cash Flow/Net Adjusted Number of
Table 14 also includes regression estimates for the FHP dividend payout groups The adjusted R-squared values range k o m 1.07% for FHP Group 3 to 36.10% for FHP
Group 2 The market-to-book ratios are insignifcant for three of the four groups, which may account for the O bserved low R-squared values The CFlK coefficients are positive and significant for ail four groups, which is consistent with the results of previous studies The cash flow coefficients for FHP groups O, 1, 2 and 3 are estimated at 0.109,0.140, 0.292, and 0.024 These estimates suggest FHP Group 2 is the most sensitive to liquidity, followed by FHP Group 1, FHP Group 0, and fmally by FHP Group 3 The low sensitivity exhibited by the high payout group appears to offer some support for the onginal FHP88 results at fxst glace However, despite the magnitude of some of the observed differences, none of them are statistically ~ i g ~ c a n t according to the empirical values
Table 15 presents regression estimates for the the-varying dividend payout groups that are discussed in the previous section Once again, we observe insignificant market-to-book coefficient esthates for three of the four groups PLU of the CFIK coefficient estimates are positive and three are significant at the 5% level, while the estimate for the Payc30 group is significant at the 1 1 % level The coefficient estirnate is highest for the Pay Negative group (0.238), second highest for the PayO group, third highest for the P a p 3 0 group and is lowest for the P a y d 0 group m i s offers some support for the FHP88 results, since the low payout f i m s appear to be more sensitive to liquidity However, as before, despite the magnitude of some of the observed differences, none of them are found to be statisticdly significant The insignificance of
Regression Estimates for Tie-Varying Dividend Payout Croups
Reported coefficients are the 'within' fixeci fum and year estimates over the 2988-94 sample period T- statistics are in parentheses Capital expenditures divided by net fùxed assets is the dependent variable The f i ' s market- tc~book ratio and osh flowlnet fixed assets are the independent variables The Pay Negative group represents the group formed using firm year observations where the nnn's dividend payout was l e s than zero; PayO represents zero dividend payout firm years ; Pay 30 represents payouts between 30 and 100% The empirid p-values are determined using the simulation procedure described in chapter 3 ùhey are estimated based on the nul1 hypothesis that the coefficients are equal for the two groups under consideration- The alternative hypothesis is that the coefficient for the Grst group is p a t e r than that of the second group For example, the pvalue of 0.7W in the market-to-book colurnn for P a p 3 0 versus Pa*, suggests the market-to-book coefficient for the Pap30 group is pater than that for the Pa$ group at the 70.44 % significance level The 0.7446 p-vatue in the next column suggests that the coefficient estimate for Cash FlowMet Fixed Assets is greater for the Pap30 group rhan for the Pa@ group (at the 74.46 % level of significance) P-values in bold indicate a signifiant difference in coefficient estimates at the 5% level
Market-t&Book Cash FlowMet Adjusted Num ber of
89 observed differences across the groups is not surprising given the small number of observations available for some of the groups This reinforces the importance of having an adequate number of firms in a group for cornparison purposes, since it appears the behavior of a few f m s can have a significant effect on overall conclusions
5.3.2 Financial Consbaint and Industry Groups
Regression results for the FC, PFC and NFC groups presented in Table 16 indicate wide variations in the market-to-book coefficient estirnates across the groups, from -0.001 for the NFC group, to 0.058 for the FC group, to 0.083 for the PFC group The estimates are significant for the FC and PFC groups, but not for the NFC group The coefficient estimates for the liquidity variables are positive for dl three groups, however, the estimate for the NFC group is very small (0.00 1) and is insignifcant
Estimates suggest that FC firms exhibit the greatest sensitivity to liquidity, with PFC firms following closely behind In contrast, WC firms display a relatively low level of sensitivity to internal fund availability Statistical analysis reveals a significant distinction between the NFC and FC firm estimates (p-value = 2.70%) However, the discrepancies between the FC and PFC estimates and the PFC and NFC estimates are deemed statistically insignificant, despite notable differences in their values.
Table 13 indicated that the NFC group consists of a higher proportion of resource f i s than the other two groups, which may impact the regression results To examine this matter, 1 divide the sample into the four industry categones described in section 5.1 Regression estimates obtained for the different industry groups are reported in Table 17
Regression Estimates for the Financial Constraint Groups
Reported coefficients are the 'within' Gxed fm and year estimates over the 1988-94 sarnpie period T- statistics are in parentheses Capital expenditures divided by net fixed assets is the dependent variable The fhn's market-to-book ratio and cash flowlnet fixed assets are the independent variables The FC, PFC and NFC groups are formed by sorting al1 firms accordhg to their discriminant scores Every year, the f m s with the lowest discriminant scores (the bottorn third) are categorized as fmancialiy constrained (FC); the next third are categorized as partially financiaiiy constrained (PFC); and the top third are categorized as not frnanàally consirained (NFC) The empiriml p-values are determined using the simulation procedure described in chapter 3 They are estimated based on the nul1 hypothesis bat the coefficients are equd for the two groups under consideration The alternative hypothesis is that the coefficient for the first group is greater than that of the second group For example, the p-value of 0.9382 in the market-tsbook colurnn for
NFC versus PFC, suggests the market-tebook coefficient for the NFC group is greater than that for the PFC group at the 93.82% significance ievel The 0.7600 p-value in the next column suggests that the coefficient estimate for Cash Flow/Net Fixed Assets is greater for the NFC group than for tbe PFC group (at the 76.00% llevel of significance) P-values in bold indicate a signifiant difference in coefficient estimates at the 5% level
Market-to-Book Cash Flow/Net Adjusted Number of
Regression Estimates for hdustry Groups
The reported coefficients represent estimates for the 1988-94 period within each industry and year The dependent variable is capital expenditures divided by net fixed assets, while the independent variables are the firm's market-to-book ratio and cash flow to fixed assets The analysis includes four industry groups: SIC1 (agriculture, mining, resources, and forestry), SIC2 (industrial manufacturing), SIC3 (retail and wholesale), and SIC4 (services) The p-values indicate the significance of differences in coefficient estimates between groups For example, the p-value of 0.7312 for SIC4 versus SIC3 in the market-to-book column suggests that the market-to-book coefficient for SIC4 is significantly greater than that for SIC3 at the 73.12% level.
Market-to-Book Cash Flow/Net Adjusted Number of
The market-to-book coefficient estimates are insignifcant for all four groups, while the cash flow coefficients are a l l positive The CFlK coefficient estimates are si@icant for all industry groups except resource-based companies (the SIC1 group) This may account for the observed insignifcance of the CFIK coefficient for the W C group, since it contains a much higher proportion of these f i s (42%) than is present in the FC group (16%), the PFC group (20%), or the entire sample (26%) The coefficient estimates suggest that resource f m s are less sensitive to intemal hind availability than f m s in other industries, however, the p-values indicate the difference is only statistically significant between the resource fums and manufacturing f m s (the SIC2 group) It is noteworrhy that manufacturing f m s comprise the entire FHP88 sample, and these fums tend to pay higher dividends than resource f m s , on average
Extending the examination of internal fund availability's impact on investment sensitivity to Canadian firms is valuable, as most prior research (excluding Schaller, 1993) has focused on U.S companies The Canadian sample differs from the U.S sample used in the FHP replication, spanning a different time period and representing a broader range of industries Notably, it includes a higher proportion of low payout firms and a lower proportion of high payout firms.
Groupcharacteristics
Firms in the U.S sample are classified using the same three approaches used for the Canadian study: (i) according to the FHP88 classification scheme, based on dividend behavior during the 1988-94 penod; (ü) according to the tirne-varying dividend payout measure described in section 5.2.1; and (üi) according to the discriminant score approach described in section 3.3.2 The FHP groups were formed based on seven year average dividend payout ratios, which is slightly different than the approach used to f o m the FHP groups for the Canadian study
Based on the fust classification approach, the 413 f m s with average dividend payout ratios between O and 10% were assigned to FHP Group 1, the 156 f i m s with payout ratios between IO and 20% were assigned to FHP Group 2, and the 51 1 F i s with payout ratios above 20% were assigned to FHP Group 3 Based on the second approach,
F i - y e a r observations are delegated to three groups: (i) those with zero dividend payout (the Payû group); (5) those with O to 30% payout ratios (the Pay30), however, none of the differences is significant
Within various exchange groups, the liquidity sensitivity coefficient (CFIK) demonstrates a higher sensitivity for firms classified as NFC (non-financial corporations) and PFC (primarily financial corporations) Specifically, among NYSE-listed firms, NFCs exhibit the highest CFIK coefficient (0.186), followed by PFCs (0.116) and FCs (financial corporations) (0.084) These differences between NFCs and other groups are significant at the 2% level, while the PFC-FC difference is significant at the 5.52% level This pattern holds for firms traded on NASDAQ, with all group differences being significant at the 4% level For AMEX-listed firms, the pattern differs slightly, with the highest coefficient observed in PFCs, followed by NFCs and FCs However, only the difference between PFCs and FCs is statistically significant.
Regresion Estimates for Financial Constraint Sub-Croups Within &change Croups
Reported coefficients are the 'within' fùxed hnn and year estimates over the 1988-94 sample period T- statistics are in parentheses Capital expenditllres divided by net fixed assets is the dependent variable The fm's market-to-book ratio and cash flowlnet fixed assets are the independent variables The AMEX Group includes f m s whose shares are iisted on the American Stock Exchange (AMEX); the NASDAQ Group includes f m whose shares trade on NASDAQ; and the NYSE Group includes hrms whose shares trade on the New York Stock Exchange (NYSE) The FC, PFC and NFC goups are formed by sorting f m s within a given exchange group according to their discriminant scores Every year, the f m s in the group with the lowest discriminant scores (the bottom third) are categorized as financially constrained (FC); the next third are categorized as partially financially constrained (PFC); and the top third are categorized as not fmancially constrauied (NFO The empirical pvdues are determined using the simulation procedure described in chapter 3 They are estimated based on the nul1 hypothesis that the coefficients are equal for the two groups under consideration The alternative hypothesis is that the coefficient for the first group is greater than that of the second group For example, the pvahe of in the market-to-book column for NFC versus PFC in the NYSE group, suggests the market-to-book coefficient for the W C group is greater than that for the PFC group at the 36.42 % significance Ievel The 0.0188 p-value in the next column suggests that the coefficient estirnate for Cash FlowMec F i e d Assets is greater for the NFC group than for the PFC group in the NYSE group (at the 1.88 % Ievel of significance) P-values in boid indicate a signifiant difference in coefficient estimates at the 5% level,
Market-to-Book Cash FlowINet Adjusted Number of
Regression Estimates for Financial Constraint Sub-Groups Within Industry Groups
Reported coefficients are the 'within' fixed fkm and year estimates over the 1988-94 sample period, T- statistics are in parentheses, Capital expenditures divided by net fmed assets is the dependent variable- The h ' s market-CO- book ratio and cash flowfnet Gxed assets are the independen t variables The SICI-