Firm Size and the Impact of Profit-Margin Uncertainty on Investment: Do Financing Constraints Play a Role?

I B G t F R S In F D P N 5 h]y 1996 F S A T I PR U I F C P A R V G a P L . N In F D P a p m c s d a c c R p I F D P ( t ac t t a h h a u m s c w t a a

A We s t r i c u a f p f t in d s f i in uncertainty aboutfuture profits depresses investment; in allother industries,increased uncertainty has virtually noeffect (orhasa positiveeffect) on investment. The data setfrom which these findings emerge is a balanced panel, consisting ofannual data from 1958to 1991 for252manufacturing industriesintheUnited States. Thetheoreticalwork onthistopicpointstouncertainty aboutfuture profitflows asoneof theimportantfactors thatdeterminesthe ease withwhich firms can access externalcredit. Theprediction madebythetheoryisthatanincreaseinuncertaintyexacerbates informational asymmetries, and hence makes lenders reduce the flow of credit; this in turn lowers investment in creditconstrained firms. If one is willing to accept firm size as a proxy for access to external credit, then our finding that greater uncertainty lowers investment in small-firm-dominated industries is consistent with the theoretical prediction. . .

Firm Size and the Impact of Profit-Margin Uncertainty on Investment: Do Financing Constraints Play a Role? Vivek Ghosal and Prakash Loungani” I. Introduction The results of recent research on capital market imperfections suggest that there is an important difference between the cyclical behavior of small firms and large firms. For instance, Gertler and Gilchrist (1994)find that a tightening of monetary policy affects real activity in small firms much more than in large firms. Their explanation for this finding is that firm size is a proxy for ability to access (external) capital markets. Small firms are constrained by internal funds because of informational asymmetries; many theoretical studies have shown that such asymmetries can lead to a certain class of borrowers being denied access to external capital markets. This paper provides new evidence on the potential importance of such financing constraints in accounting for cyclical fluctuations in real activity. In particular, we study how investment responds to changes in uncertainty about futureprofits,and whether or notthis response isdifferent in industriesthatare dominated by small firms. The difference that we find is quite stark: In industries dominated by a large number of smallfirms, an increase in uncertainty about future profits depresses investment, but in all other industriesincreased uncertainty has virtuallyno effect (orapositive effect) oninvestment. The data setfrom which these findings emerge is a balanced panel, consisting of annual data from 1958to 1991for 252 SIC 4-digit U.S. manufacturing industries. * The authors are respectively: Associate Professor in the Department of Economics, Miami University, Oxford, Ohio [e-mail: ghosalv@SBAMail.MUOhio.Edu]; and Economist in the Division of International Finance, Board of Governors of the Federal Reserve System [e-mail: lounganp@frb,gov]. This paper ,. represents the views of the authors and should not be interpreted as reflecting those of the Board of Governors of the Federal Reserve System or other members of its staff,

These results are of significance for a number of reasons. First, the theoretical work in this a S a W ( a G a H ( a o p u a f p f o t k f t d t e w w firms c a external credit. The prediction made by the theory is that an increase in uncertainty exacerbates informational asymmetries, a hence makes lendersreduce theflow ofcredit; this in t l i incredit-constrained firms. To the best ofour knowledge. there is noempirical work thattests this prediction. If one is willing to accept firm size as a proxy for access to external credit, then our finding that greater uncertainty lowers investment in small-firm-dominated industries is consistent with the theoretical prediction. Second, the particular impulsethat we consider--a change in uncertainty aboutfuture profits--is very different from that considered in previous work, where the impulse considered has generally been a change in the stance of monetary policy. The fact that we find significant differences in the behavior of the two groups even in response to this impulse (uncertainty) lends support to the use of the “small vs. large firms” distinction in studies of fluctuations in economic activity. The paper is organized as follows. Section 2 reviews the theoretical and empirical work on the link between uncertainty and credit market imperfections. Section 3 isdevoted to describing two important steps that have to be taken in order to carry out our empirical tests. The first is the construction of measures of uncertainty about future profits. The second is to identify a group of industries that are dominated by small firms. An empirical model for investment is specified and estimated in Section 4. We show that the investment-uncertainty correlation is negative for the group that we identify as being dominated by small firms, but zero (or even positive) for the ‘control’group (the set of all other industries). These findings are shown to berobust to: (i)alternate measures of uncertainty; (ii) alternate ways of segmenting industries into the ‘small’and ‘other’categories; and alternate controls for investment opportunities. Conclusions are stated in Section 5. 2. A Review of the Literature 2.1 Uncertainty and Financing Constraints: Theoretical Prediction The ideathat uncertainty affects theorganization ofcapital markets is an oldone. For instance, Hart

(1940) describes how in the presence of uncertainty, capital markets are likely to be become “stepped”or “segmented”,that is, some entrepreneurs would have rely on own funds to finance projects, whereas other entrepreneurs could fund projects via outside equity or by borrowing from middlemen such as bankers. Stiglitz and Weiss (1981) and Greenwald, Stiglitz and Weiss (1984) revived the idea that a certain class of borrowers is likely to face financing constraints when there are informational asymmetries between bo a l In a laterextension, Greenwald and Stiglitz (1990)conduct a more direct theoretical investigation of how investment decisionsare affected byequity and credit rationing atthe firm level. Their model makes the following prediction (p. 19): “Increased uncertainty about future profitability ... increases both the absolute and incremental risk ofbankruptcy underquite general conditions at any levelof investment and firmequity. Thus, firms respondbylowering investmentsincetheycannotabsorbtheincreased risksby issuingmoreequity.” TheGreenwald-Stiglitz prediction providesthebasisfortheempirical testsinthispaper. Likemanyprevious tests of the ‘financingconstraints’theories, we test the theory by exploiting the fact that there are likely to be differences across firms in the extent to which they face financing constraints. Hence, we take the Greenwald-Stiglitz theory as predicting that the impact of increased uncertainty on investment will differ across firms. depending on the degree of access they enjoy to external capital markets. 2.2 Firm Size and Financing Constraints The next step is to come up with a measure of capital market access. Following several notable studiesin thisarea, suchas Fazzari, Hubbard and Petersen (1988),Oliner and Rudebusch (1992)and Gertler andGilchrist (1994), we usefirm sizeasaproxy forcapital marketaccess.’Gertler andGilchrist (p.313-14) argue that 1 Fazzari, Hubbard and Petersen (FHP 1988)suggestthat firms with a low dividendpayout ratio may be the ones that are financially constrained. However, this interpretation has been contested by Kaplan and Zingales (1995) who examine the annual reports or 1O-Kreports for the low-dividend firms in the FHP sample and reach the conclusionthat “thesefirms were financiallyconstrained in fewer than 15%of sample years.” Fazzari, Hubbard and Petersen (1996) respond to this criticism and argue that the Kaplan and Zingales study is based on a flawed definition of financing constrained.

“whilesize per se may not beadirect determinant, it is strongly correlated with the primitive factors that do matter. The informational frictions that add to the costs of external finance apply mainly to younger firms, firms with a high degree of idiosyncratic risk, and firms that are not collateralized. These are. on average, smaller firms.” Since our study uses industry-level rather than firm-level data, we use information on the size distribution offirms andestablishments inordertosegment oursampleintoagroupofindustrieswhere small firms are dominant, and a ‘control’group ofother industries. We then test whether an increase in uncertainty lowers investment inthe groupassumed tobefinancially constrained, and whether ornotthiseffect isgreater than the effect in the control group. 2.3 Investment under Uncertainty: Role Sunk Costs Recent theoretical work on firms’ investment behavior under uncertainty has shown that in the presence ofsunkcosts. wherecapital adjustment costsareasymmetric withdownward adjustment costsbeing significantly greater than upward adjustment costs, an increase in uncertainty is likely to lower investment (see Dixit and Pindyck, 1994).ZFrom this literature it is clear that one should control for the magnitude of sunk costs when investigating any relationship between uncertainty and investment. In the empirical work that follows we show that a negative relationship between uncertainty and investment holds only for industries dominated by a largenumber of small firms, and not for the relatively large firm dominated industries. Can this pattern of results be explained by differences inthe extent of sunk c a t two broad classes ofindustries? While adetailetiexamination of this issue is not undertaken here,3we think that this is explanation for our results is unlikely. Simply put, this is because sunk costs are likely to be much lower in the small-firm dominated industries than in the large-firm dominated industries. We now spell out our line of reasoning in more detail. Industrial Organization theory has emphasized the significance of sunk costs in determining firm size and industry structure. The pioneering contribution by 2Also see Hubbard (1994) and Pindyck (1991), and the reference there. 3Primarily because there are no good measures of true sunk costs. 4

Baumol,Willig andPanzar (1982)highlightedtheroleof sunkcostsasabarrier-to-entry. A sunkcapitalcost requirement to enter into a market creates an asymmetry in the costs and risk faced by an entrant, thereby creating an entry barrier. The implication ofthis lineof reasoning isthat where sunkcosts are high,industry structure is likely to be more concentrated with the presence of relatively fewer large firms--a direct consequence ofentry barriers. Returning to our analysis, since we focus on industries dominated by a large number of small firms,~ it seems unlikely that in these industries sunk costs are high. Therefore, if uncertainty turns out to have an important adverse effect on investment in these small firm dominated industries. then it is more likely that the impact is due to financing constraints.S 2.4 Empirical Work on Uncertaintyand Financing Constraints Mackie-Mason (1988) provides evidence on the factors that influence a firm’sdecision on whether to obtain funding from private or public sources. One of the factors he considers is the forecast variance of a firm’se gr v s in spirit to our uncertainty variable, as we describe in the next section. Mackie-Mason finds that firms with higher earnings variance “were more likely to useprivate sources of funds (p. 94).” His explanation for this finding is that “ifa firm has volatile earnings, outsiders are more uncertain about future prospects and are less willing to buy public security issues, so such firms prefer to financeprivately.”However, otherthan this evidence on the impact of uncertainty onthe choice of financing, we are not aware of a direct test of whether or not increased uncertainty has an impact on 4The data on firm size, number of firms and industryoutput concentrationthat we use are collected over a number of years and represent long-run characteristics (see Section 3.3). Further, industry structure characteristicsare remarkably stableovertime. For example, the cordation between the industrynumber of firmsfor the 1972and 1982Censusyears is 0.94; the correlation for industry four-firm outputconcentration ratio is0.92. These highcorrelationcontinueto holdacrossother Censusyears like 1963or 1987.Thistimeinvariance has been documented elsewhere; see Caves and Porter (1980), Scherer and Ross (1990) and Schmalensee(1989).We stressthe long-runand relativetime-invariantnatureof theseindustrycharacteristics as this important to the argument that our segmentation of into small and relatively large firm dominated industriesis based on true structuralcharacteristicwhich are not subjectto much (if any) cyclical variations. l s However. as we noted earlier. this isonly an indirect way ofcontrolling for sunkcosts and aproper t of the sunk cost hypothesis must wait till we have good measures of true sunk cost. 5

I in t a t f c 3. Data Description and Measurement of Variables We use the following framework for examining the impact of uncertainty on investment. First, we co f e i a t s t s t v t l u a f p N u in p t S B A w d i o co r p i i t g ( i d s f a ( a o in F e t e u i p t d f a in a e e m s t used in many panel data studies. As a check on the robustness of our benchmark results, we present estimates for alternate measures of profit un c s r our firm size measures, and alternate controls for investment opportunities. 3.1 Data Sources With one exception (viz., the aggregate capacity utilization rate), all the data come from the Productivity Database assembled by Wayne Gray and Eric Bartlesman (1991). This data set contains annual data for SIC 4-digit industries over the period 1958-1991.The original source of the data are various issues of the Census of Manufactures and the Annual Survey of Manufactures. Sincethetheoryreviewed abovepertained tofirm-level decisions,ouruseofindustry-leveldataneeds to beju O i r t w o u m t c n j cr v b a a f a t v T c a in es t i c u i T m c used firm- 6There is a set of recent studies that have looked at the impact of price uncertainty on investment, but these studiesdo notinvestigatethe roleof financingconstraints.These includeCaballero and Pindyck (1992), Huizinga (1993) and Ghosal and Loungani (1996.a). 6

level data set. COMPUSTAT. offers researchers a relatively limited amount of time series variation; for instance. Gertler and Himmelberg (1993) have a sample period of 1979to 1989, while Leahy and Whited (1996) use 1981 to 1987 as their sample. Given our chosen m of constructing measures of uncertainty (see Section 3.1). limited time series data poses a serious limitation. These considerations motivated the use of industry level data. We used the following selection rules to decide which industries would be included in the sample: (i)industriesclassifiedas“notelsewhereclassified”or“miscellaneous”weredroppedfromthesample t do not have well defined product markets; and (ii) industries which had missing data on industry four-firm concentration ratiosand firm-size were excluded inordertocreate abalanced panel.The impositionof these selection rules exclusions left us with 252 industries in the full sample. 3.2 Meawring Uncertainty We assume thatfirms useaprofit margin forecasting equation to predictthe levelof future margins. The standard deviation of the residuals from this forecasting equation is used as a measure of the degree of profit uncertainty. This notionof uncertainty isconsistent with boththe theoretical work7 and with previous work on the quantification of uncertainty.8 Our measure of industry profit margins is the following: H= [ (Total Sales Revenue minus Total VariableCosts)/(Total SalesRevenue) ],where total variablecostsinclude labor, materialsand energy costs. 7See Craine (1989), Caballero (1991) and Dixit and Pindyck (1994). 8See Fisher and Hall 0969), Ghosal (1995, 1996.b), Ghosal and Loungani (1996.a), Huizinga (1993), Leahy and Whited (1996), Mackie-Mason (1990) and Winn (1977). All these studies follow the practice of usingthe standarddeviation (or the conditionalstandarddeviation)of some variableof interestas a measure of uncertainty. 7

Hence. H is the short-run profit margin per unit of sales.9 Firms are assumed to forecast IT, and, to the extent that margins are forecastable, this reduces the uncertainty that they face. The forecasting equation is given by(l), where Iltiisthe profit margin of industry “i”and ’’t’’is alinear trend.m E ( c for any deterministic trend in margins and, since we are using annual data, embeds sufficientlagsto capture industryprofitdynamics.11Experimentation showedthatadditionallagsof ~io~were insignificant in virtually all industries. We use the following procedureto create a time-series for the profit-margin uncertaintyvariable.For each industry in our sample. we estimate equation (1) using annual data over fourteen-year overlapping periods starting with 1958: i.e. 1958-71, 1959-72,....1977-91. The standard deviation of the residuals from these regressionsis our measure of uncertainty O(ll)i,t,where “i” and “t”index the industryand time period. Using this procedure we are able to obtaina relatively long time-series--2Oobservationsfrom 1972to 1991-- . This is a fairly commonly used measure in the industrial organization literature. See, for example, Carlton and Perloff (1994, Ch.9), Domowitz, Hubbard and Petersen (1986, 1987), Ghosal (1996.c) and Schmalensee (1989). Carlton and Perloff (p.334-343) and Schmalensee (1989) present a comprehensive discussion of variousmeasuresof profitmarkups.ratesof return and the pitfallsassociatedwithmeasuring them. Our measureII doesnotcontrolforcapitalcosts--which are moreimportantfor measuringtruelongrun profitability. In any case, as discussed by Carlton and Perloff and Schmalensee, quantifying capital costs is difficult due to problems related to valuing capital and assessing depreciation. our general conclusions are robust to alternate specifications of the profit equation (e.g., including an aggregate business cycle control). In Section 4.3 we present results to confirm this, The use of autoregressive models to capture the dynamics of profit margins is quite common. See, for example, Geroski and Mueller (1990), Ghosal (1996.c) and the reference there. The basic results of the paper are not affected if we estimate the profit equation in growth rates rather than levels. 8

on o(II) for each industry.’z Next, we p s s t i r r o t a m F the f s p l9 m ( 2 i t i r w 0 S co w q l g w t c m ( d a i b - 0 ( 1 F t l 1 p l m w 0 T c m ( d across in t fi s c w - ( 1 examined such regression ch for each 1 period over which equation (1) was estimated. In general, the overallfit was quite good and serial correlation was low. T 1p s cr s s @ f t f s i A n s f m t wi v in cr(ll) is the coefficient of variation. The numbers in the row labeled “C.V. @_l)” show that the representativeindustryhas a coefficient of variation of 19%.with the range beingfrom 6Y0 to 57%. Overall, there appearsto be a reasonableamountof variation in @I) both withinand acrossindustries.which isencouragingfrom theviewpointofourproposedempirical examination. Appendix A presents examples of some industries where there a lot of variation over time in the level of uncertainty,and others where there is relatively little variationover time n a(l’1). 3.3 Segmenting Industries into ‘Small’versus ‘Other’ The U.S. Small Business Administration(SBA) providesa list of industriesthat are “dominated”by small businesses. The SBA classifies a small business as one that employs 500 workers or less. This classification was accepted by Congress in 1982as the basis for defining a small business. An industry is 12An alternate approach would be to estimate ARCH models to construct measures of profit margin uncertainty. Our attempts to use the ARCH framework were not successful in the following sense. After imposing all the necessary restrictions for estimating ARCH models (see Hamilton. 1994, Ch. 21), we estimated second-orderARCH modelsfor each of the 252 industriesin our sample. For a very large number of industriesthe estimationfailed to convergealong with problems related to the singularityof the Jacobian. experimentedwithalternatestartingvaluesas wellaschangingtheorderoftheARCH specification;none of these experimentsalleviated the basic problems mentioned above. 9

classified “ b d if at least 60% of industry employment is in firms with fewer than 500 employees.’qTo come up with a set of industries that are consistently dominated by small businesses, the S l b d o t d y p t 1 1 a 1 C Po S t o t m i t S list have high four-firmconcentration ratios (that is, the fraction o a f t f l f t i h H t f t m f w t i a s u our hypothesis. financiallyconstrained--can be outweighed by the fact thattherearea few largefirmsthatdo not face such constrains.To mitigatethe impact of this withinindustryheterogeneity,we create a sample of industriesthat are dominatedby smallbusinesses and have low output concentration ratios. We collected data on industry four-firm concentration ratios, CR4, from the Census of Manufactures over our sample period (to match the time period over which we have data on the uncertainty measure). For our full sample of industries, the cross-industry mean value of CR4 is approximately40%. Using this number. we define a “low”concentration industryas one that has CR4<40% over the sample period.lb Based on this discussion, we can now define four industry groupings: S S R 1 T A a A The S l i a t t S d l di w o d a t S 4 l a3 i c s b d assume that all the component 4-digit industries within this 3-digit grouping are also dominated by small businesses. Davis, Haitiwanger and Schuh (1994) show that there is considerable migration of firms across size categories; hence, to get a clear picture of industries that are truly dominated by small businesses, it is important to examine the size classification notjust at a single point in time, but over a period of years. The SBA classification,based on data over a period of time. satisfiesthis requirement. Some examplesare Canned Seafood (SIC 2091), Roasted Coffee (SIC 2095)and Hard Surface Floor Coverings (SIC 3996), which have industry 4-firm output concentration ratios in the 55% to 90% range. The antitrustliterature(see White 1987,p.16-17) tends to define a market as “non-competitive”when the concentration ratio is in the 50%-60%range. A CR4 cutoff of 40% is a conservative choice because the evidence suggests that the critical CR4 beyond which industries exhibit “non-competitive” behavior appears to be in the 50%-60% range (see White, 1987, p.16-17; Ghosal, 1989; and the reference there). Domowitz et al. (1987, p.389) use CR4=50% as the cutoff for low and high concentration industries. We impose the CR4<40% cutoff over the entire sample period as some Industrieshave a trend in CR4 and so using any one year’sCR4 values 1

I Group 1: ALL industries. Group 2: “SMALL business”industries(based on SBA informationonly). G “ a C i ( i plus concentration ratio data). Group 4: all OTHER industries(basedon SBA informationonly). To g a better feel for the internal structure of the industries in these groupings, we examined the mean numberoffirmsperindustryoveroursampleperiod.~’Forthefourgroups,ALL, SMALL, SMALL&CR4S40 and OTHER, the mean number of firms were 695, 1191,2157 and 515, respectively.It is clear that there is a substantial difference in firm density across the SMALL and OTHER groups. Therefore, our SMALL groups are characterized by a large number of small firms, whereas the OTHER group contains a smaller number of relatively larger firms. In Tables 2 and 3 we present some summary statisticson the uncertainty measure c(H) for the SMALL and OTHER industry groupings. As was the case for the full sample of industries, these statistics show that there is a reasonable amount of within-industry and cross-industry variationin o(H) for both groups, 4. Empirical Results 4.1 SpecifEation We include our measure of uncertainty ~(~)i<~in an empirical investment model for panel data.’8 The dependent variable is the ratio of gross industry investment scaled by the beginning-of-period capital stock, (UK)l,t.In additionto the uncertaintyvariable, c a l v i c f s by capital stock, (CF/K),,,,are the main explanatory variables. There are two theories which motivate the may be misleading. 17We first computed the mean number of firms for each industryoverour full sample,and then computed the group mean number of firms. So the data on the number of firms is a long-run representation,Data on the number of firms were collected from variousissuesof the Census of Manufactures. 18For specification of panel data investment models see Devereux and Schiantarelli (1990), Fazzari, Hubbard and Petersen (1988), Fazzan and Petersen (1993) and the reference there. 11

inclusion of cash flow in an investment model. The first is that cash flow (or earnings) is a signal of the future marginal productivity of capital; the second is that cash flow is a measure of internal funds, a t co b c f a i t i f c i S o which of these two theories is generating the correlation has been the focus of many previous studies; however. it is not crucial here because our tests on financial constraints are based on the impact of the uncertainty variable, rather than the cash flow variable. We did not attempt to construct a cost-of-capital measure or Tobin’s“q”, v::~iableswhich are suggested by alternate m i T b the resultsfrom previousstudiesdo notoffermuch reason forpreferring these measuresto cash flow.Fazzari, Hubbard and Petersen (1988) show that the omission of the “q”variable or the cost-of-capital measuredoes not significantly affect the performance of the investment model.20 An industry-specificfixed-effect, Ui,is included to capture time-invariant influences on an industry’s mean level of investment over the sample period. To capture economy-wide influences on investment that are common to all industries in any given year, we include a set of year time dummies y~.This is an important control. because the time dummies can account for the influence of the myriad shocks--ranging from changes in tax rates to events such as oil price shocks--thatcan affect investment, but are not explicitly included in the empirical model. Lastly. it is a stylized fact that investment spending shows persistence (see Chirinko 1993).As is standard in theempirical literature, we accountfor this by including a lagged dependent variable.Combining l the above features. the investment model is givenby equation (2). Ail variables in equation (2) are measured in logarithms, and so the coefficient estimates can be interpreted as elasticities. 19Furthermore, Tobin’sq is very difficultto construct at the industry level. We am not aware of any study that constructs industry-specific measure of Tobin’sq. Also see Chirinko (1993). See Cummins, Hassett and Hubbard (1994) for some contrasting results. 12

‘ (gi + yt + ~2 + (2) We estimateequation (2) forthe four industrygroupsdescribedin Section 3. Based on our discussionofthe theory, we expect the uncertainty elasticity WI to be negative for Groups 2 and 3; we also expect that investmentshouldbe moreresponsiveto uncertaintyinthesetwo groupsthan inGroup4. In additionto these key hypothesesof intenxt, the resultsof previousstudies lead us to expect that the cash flow elasticities Yz and Y~ will be positive,and largerfor Groups 2 and 3 than for Group 4. T 4 shows the global mean (i.e. the mean over all observations in the sample) and standard de f t industry variablesfor the four groups. Other than the fact that the mean ratio of cash flow c a l b higher in the two categories of “small”firms than in the “other”categories, there is n m di in these summary statisticsacrossthe groups. 4.2 Main Results We use the fixed-effects OLS estimator to obtain estimates of the parameters in equation (2).21 Columns 1-4 of Table 5 present the results of estimating equation (2) for our four groups. The top row indicatesthe industrygroup.The numbersreported are the coefficientestimates of the Y parameters;to save space, the estimates of the industry-fixed effects ( ~i ) and time-fixed effects ( YI) are not r Examining the resultsfor the smallbusinessgroupsin columns 2 and 3, it is evidentthat g u decreasesinvestmentin theseindustriesand the elasticityestimatesare significantat conventionallevels.We a n t t un e g qu l when we imposetheCR4 restriction(column 21Hsiao (1986)showsthatinclusionof laggeddependentvariablesinpaneldatamodelsmay leadto biased estimatesofthedynamiccoefficients.However, Hsiao showsthatthisbias is likelyto be a problemin panels with extremely smallnumber of observationsin the time domain. Our panel has 20 observationsin the time domainand thisbiasis likelyto be very small.Further, usinga strategythatiscommon in the literature(e.g., Fazzari and Petersen. 1993). we verified that our basic conclusions about the impact of uncertainty are unaffected if we exclude lagged investmentfrom the equation. 13

3).The estimatesshow thatthe uncertaintyelasticity ranges fromabout -0.12 to -0.16 in the smallbusiness dominated industries. Turning to the results for the OTHER industries in column 4, we notice a sharp difference: the uncertainty elasticity is positive, relatively small (0.06) and significantlydifferent from zero. We briefly comment on the cash flow coefficients. As in many previous studies, the estimated cash flow elasticitiesfor the relatively financiallyconstrained groups2 and 3 (about0.34) is greaterthat for Group 4 (0.27). However, the quantitativedistinctionshere are not very large. Since our Group 3 definition uses information on both “size”and “concentration”. we conducted a check to see whether one of the two characteristics was the dominant force behind the results on the impact of uncertainty. We created a sample of industries with CR4<40 and no control for size. The coefficient estimate (std.error) on o(H) for this group was -0.059 (0.051). This estimate is quantitatively much smaller than the estimate reported in Column 3, and it is statistically insignificantlydifferent from zero. Hence, the CR4 control by itself is not generating the observed outcome: the small business classification does play an important independent role.n To summarize, the results thus far show that greater uncertainty decreases investment in the small business dominated industries, but not in other industries. Hence, the results support the predictions of the financing constraintstheories. 4.3 Additional Results In this section we present numerous additional results to check the robustness of our basic finding of a negative relationship between profit uncertainty and investment in the small business sector. 4.3(a}. Durable versus Nondurable Goods Producing Industries: The excessive volatility of durable goods In an earlier paper. Ghosal and Loungani (1996.a), examined the impact of “price” uncertainty on current investment in competitive versus oligopolistic industries. Our results indicated a negative impact of price uncertaintyon investment in the relatively competitive industries. 14

industriesrelative to nondurable is well documented. To examine whether some of our results were being driven by such intrinsicproductcharacteristics.we partitionedourfull sampleof industriesinto durablesand nondurable and reestimated the investment equation. For durable goods industries the estimated profit uncertainty elasticity (std. error) was 0.06 (0.034); for nondurable goods industries the estimated elasticity (std.error) was -0.01 (0.035).Thus it does not appear thatthe distinctionbetween small v. large businesses, andthe resultforthe smallbusinessindustries,thatwe reportarebeingdriven primarilyby productdurability characteristics. 4.3(b), An Alternate Measure o-f Uncertainty: To check the robustness of our results, we constructed an alternate measure of uncertainty by estimating equation (3). Equation (3) includes one lag of H and two lags of manufacturing capacity utilization rate CU. The inclusion of CU is motivated by the results in Domowitz, Hubbard and Petersen (1986, 1987)which show that business cycle fluctuationsplay a key role explainingchanges in industryprofit margins.23It could be argued that equation (3) may generate a superior measure of uncertainty as the forecasting equation includes additional and relevant variables in the firms’ information set. The standard deviation of the residuals obtained from estimating equation (3) is our second measure of uncertainy, ~(alt)i,~,We then estimate regressions similar to those reported in Table 5 for the four groups, usingc(alt) in place of o(H). The estimates of the uncertaintyelasticities are presented in the Table 6. It is evidentthat we find the same pattern of differences in the uncertaintyelasticitiesacross the four groupsthat we reported earlier. We conducted a few additionalexperimentsto see if the estimated impact of uncertainty is sensitive Also see Ghosal (1996.c) on this issue. 15

to t sp t p m e T i ( e w a l l in equations (1) and (3). (ii) estimating the profit margins equation in growth rates instead of levek and (iii) using proxies other than CU to capture aggregate conditions. Our general conclusions do not change. 4.3(cL Refinements of the SBA Size Measure: I is well known that the size distribution of firms within an industry is often highly skewed.24In the results reported thus far, we tried to control for this within-industry heterogeneity by conditioning on CR4; here we try a different approach, using data on the size distribution of establishments from the 1 C M ( F e S 4 i t U manufacturing sector, CM p a c d e s ( e Table 7 providesevidencet t s d e m a r p f t s distribution of firms. An advantage of using the CM data is that we now get size classifications at the SIC 4-digit level of disaggregation as compared to the SBA 3-digit classification. A potentially important limitation is that these data are for one year and therefore present only a “snapshot”of the size distribution (in contrast to the SBA classification which is based on data over a number of yea=). Our strategy is to take the SBA SMALL list (that is, our Group 2) and condition on the CM establishment size distributiondata to create the following three even smaller sub-groups: (i) S t g c t i w a i in our category “Group 2: SMALL” @ also satisfytheconstraintthatthe percentage ofestablishments with <50 employeesis “greater l than or equal to” 0.817 (the 50th percentile value). Table 8 providespercentilevalues forothercut-offs. Conditioningon. say.the75thpercentilevalueresultsin very small samples. (ii) SMALL(1OO) -- Industries which are included in SMALL ~ satisfy the constraint that the percentage of establishments with S100 employees is greater than or equal to the 50th percentile value. For early work on this issue, see the classic contribution by Simon and Bonini (1958). Note that the SBA classificationsthat we use a~ based on data over 1979-88.Therefore the 1982 Census of Manufactures roughly representsthe midpoint. 16

(iii) SMALL(500) -- Industrieswhich are included in SMALL ~ satisfy the constraintthat the percentage of establishmentswith <500 employees is greater than or equal to the 50th percentile value. These classificationscreate groupsof industriesthat are likelyto be populatedby even smaller firms than the SBA SMALL category. We then estimate regressionssimilarto those reported in Table 5 for these three new groups, using the two measures of uncertainty, a(~) and c(alt). Table 9 p e t p margin uncertainty coefficients from these six regressions.While the effect is not monotonic,the estimates show that greater profit margin uncertainty continues to have a significant negative impact on current investment in all three small business dominated groups. As shown in Table 1 uncertainty continuesto have virtuallyno (or positive)impacton current investmentin industriesthatare in our OTHER category,even when we conditionfurtheronthe CM data.This conclusionholdsforboththe o(H) and~(alt) measures of uncertainty. 4.3(dL Industr\~ Sales as the Control Variable: In all our specificationsso far we have used industry cash flows (CF/K) as the primary controlvariable.To check whetherour resultsare sensitiveto alternatecontrols, we reestimatedthe investmentequation by replacing (CF/K) with the ratio of industrysalesto capital (S/K). In such an equation, firms’ investment opportunities are assumed to be captured by movements in sales. Table 11 presents estimates of the uncertainty coefficients for the 4 industry groupings. The regressions containcurrentandonelagof (S/K), industryfixed-effectsand yeardummies.The resultscontinueto support our earlier findings from Table 5. We also reestimated the investment equation with (S/K) as the control variable and used the SMALL(50), SMALL(1OO)and SMALL(500), and OTHER(50), OTHER(1OO)and OTHER(500) groups described in Tables 9 and 10, The profit uncertainty elasticities were negative and consistentlystatisticallysignificantfortheSMALL(,) groups,andpositiveand insignificantfortheOTHER(.) groups. Hence, using (S/K) as the control variable preserves our conclusions regarding the adverse impact of profit margin uncertaintyon current investmentin small businessdominated industries.

5. Conclusions Theoretical work points to uncertainty about future profit flows as one of the important factors that determines the ease with which firms can access external credit. The prediction made by the theory is that an i un ex in a a hence makes lenders reduce the flow of credit; this in turn lowers investment in credit-constrained firms. This paper measures the impact of uncertainty on investment in industries dominated by small firms, and compares it with the impact in a ‘c g o in U t m h t firm sizeisaproxyforcapital market access, the e r t p a c w t t find that an increase in uncertaintyaboutfutureprofitmargins lowerscurrent investmentinindustriesdominated bysmallbusinesses, but has no impact in the ‘control’group. Ramey (1993. p. 7-8) has emphasized, even though the small- versus large-firm results seem to offer “very compelling evidence in favor of the hypothesis that there are credit market imperfections,” that does not necessarily imply that such imperfections are important at the aggregate level: 11 ... none of the studies of firms by size classes have shown that the reaction of small firms has an aggregate impact. This is an important link in the argument because one can think of equilibrium forces that would mitigate the aggregate effect. For example, the loss in output from small firms ... might be compensated by a rise in output from large firms.” This describes the situation here, because we find that even though there is a differential impact across size classes, there is no appreciable effect of uncertainty on manufacturing investment as a whole. However, it isworthpointingoutthatsmall firms playamore important roleoutsidethemanufacturing sectorthan within manufacturing. Dennis (1993) estimates that “77 percent of all small businesses fall into broadly defined services. Fewer than one in ten small businesses are manufacturers.”This suggests that evidence from the service s t i u i n o g w n un h i i e i a ab u t i of financing constraints on small firms’investment outlays. 1

References B W J P a RobertWillig.ContestableMarketsand theTheory of IndustryStructure. San Diego: Harcourt Brace Jovanovich, 1982. Bemanke, Ben, and Mark Gertler. “FinancialFragility and Economic Performance,” Quarterly Joumal of Economics, 1990,87-114. Caballero, Ricardo. “On the sign of the investment-uncertaintyrelationship.” American Economic R 81, 1991,279-288. Caballero, Ricardo. and Robert Pindyck. “Investment,uncertaintyand industryevolution.” NBER working paper # 4160, 1992. Caves. R a M Porter. “TheDynamics ofChangingSeller Concentration,”Journalof Industrial Economics, 1980, 1-15. Carlton, Dennis and Jeffrey Perloff. Modem IndustrialOnzanization.New York: Harper Collins, 1994. Chirinko, Robert and Huntley Schaller. “Why Does Liquidity Matter in Investment Equations?,” Journal of Money Credit and Banking, 1995.527-548. Chirinko,Robert. “BusinessFixed InvestmentSpending:A CriticalSurvey ofModeling Strategies,Empirical Results, and Policy Implications.”Working Paper # 93-01, Federal Reserve Bank of Kansas City, 1993. Calomiris, Charles, and R. Glenn Hubbard. “Firm Heterogeneity, Internal Finance, and Credit Rationing,” Economic Journal, 1990,90-104. Cummins. Jason, Kevin Hassett, and R. Glenn Hubbard, “A Reconsiderationof Investment Behavior Using Tax Reforms as Natural Experiments,” Brookin~sPaDerson Economic Activitv, 1994(2), 1-75. Craine,Roger. “RiskyBusiness:The Allocationof Capital.”JournalofMonetarv Economics, 1989,201-218. Davis, Steven. John Haltiwanger and Scott Schuh. “SmallBusiness and Job Creation: Dissecting the Myth and Reassessingthe Facts,” BusinessEconomics, 1994, 13-21. DennisJr., William. ~ Small Business Pn“mer.The NFIB Foundation, 1993.. Devereux, Michael, and Fabio Schiantarelli. “Investment,FinancialFactors, and Cash Flow: Evidence from U.K. Panel Data,” in Hubbard, R. Glenn ed. Asymmetric Information. Comorate Finance. and Investment. Chicago:University of Chicago Press, 1990. Dixit, Avinash, and Robert Pindyck. Investment under Uncertainty. Princeton: Princeton University Press, 1994. Domowitz, Ian, Glenn Hubbard, and Bruce Petersen. “Business Cycles and the Relationship between Concentrationand Price-Cost Margins.”~AND Journal of Economics, 1986, 1-17. 19

Do I Glenn Hubbard, and Bruce Petersen. “OligopolySupergames: Some Empirical Evidence on Prices and Margins.”Journal of IndustrialEconomics, 1987,379-398. Fazzari, Steven. R. Glenn Hubbard, and Bruce Petersen. “FinancingConstraints and Corporate Investment.” Brookinm Paoers on Economic Activitv, 1988, 141-195. Fazzari, Steven. R. Glenn Hubbard. and Bruce Petersen. “Financing Constraints and Corporate Investment: Response to Kaplan and Zingales,” NBER Working Paper # 5462.1996. Fazzari, Steven, and Bruce Petersen. “Working capital and fixed investment: new evidence on financing constraints.”RAND Journal of Economics, 1993,328-342. Fisher, I.N.. and G.R. Hall. “Riskand CorporateRates of Return,”Ouarterly Joumal ofEConomic~,1969,79- 92. Gale. Douglas, and Martin Hellwig. “Incentive-Compatible Contracts I: The One-Period Problem,” Review of Economic Studies, 1985.647-664. Geroski, Paul, and Dennis Mueller. “The Persistence of Profits in Perspective,” in Dennis Mueller cd., ~ wamics of Comoarw Profits. Cambridge: Cambridge Univemity Press, 1990. Gertler, Mark. “Financial Structure and Aggregate Economic Activity,” Journal Monev. Credit. and Banking, 1988,559-588. Gertler, Mark, and Simon Gilchrist. “Monetary Policy, Business Cycles, and the Behavior of Small Manufacturing Firms,” Ouarterlv Joumal of Economics, 1994,309-340. Ghosal. Vivek, and Prakash Loungani. “Product Market Competition and the Impact of Price Uncertainty on Investment.” Joumal of Industrial Economics Vol 44. 1996(a),217-29. G Vivek. “Does Uncertainty Influence the Number of Firms in an Industry?” Economics Letters, 1996(b),229-37. Ghosal, Vivek. “Product Market Competitionand Industry Price-Cost Markup Fluctuations,”Working Paper, Department of Economics, Miami University, 1996(c). l Ghosal. Vivek. “InputChoices Under Uncertainty,” EConomic Inauirv Vol 33, 1995, 142-158. Ghosal, Vivek. “Market Structure,Price-CostMargins and Unionism,” Economics Letters Vol29, 1989, 179- 82. Gilchrist, Simon, and Charles Himmelberg, 1992. “Evidence on the Role of Cash Flow in Reduced Form Investment Equations,” Working Paper, Board of Governors of the Federal Reserve System. Greenwald, Bruce and Joseph Stiglitz. “Macroeconomic Models with Equity and Credit Rationing,” in . Hubbard, R. Glenn. cd., Asvmmetric Information. Comorate FWe. and Investment. Chicago: University of Chicago Press, 1990, 15-42. Gray, Wayne. and Eric Bartlesman. “The Productivity Data Set.” National Bureau of Economic Research, 20

r 1991. Greenwald, Bruce, Joseph Stiglitz,and Andrew Weiss. “InformationImperfectionsin the CapitalMarket and Macroeconomic Fluctuations.’’AmericanEconomic Review, 1984, 194-199. Ha J T S A P N P U P 1 Hart, A.G. Anticipations. Uncertainty and Dvnamic Planning. New York: A.M. Kelly, 1951. Hsiao, Cheng. Analvsis of Panel Data. Cambridge: Cambridge University Press, 1986. Hubbard. R. Glenn. “Investment under Uncertainty: Keeping One’s Options Open,” Journal of Economic Literature Vol 32, 1994, 1816-1831. Hubbard. R. Glenn. (cd.). ~s-vmmetric Information.Comorate Finance. and Investment.Chicago:University of Chicago Press. 1990. Huizinga. John. “Inflation uncertainty, relative price uncertainty and investment in U.S. manufacturing industries.”Journal of Monev. Credit. and Banking, 1993,521-549. Kaplan, Stevenand LuigiZingales. “Dofinancingconstraintsexplainwhy investmentiscorrelated withcash flow?” NBER Working Paper # 5267, 1995. Leahy, John, and Toni Whited. “The Effect of Uncertainty on Investment:Some Stylized Facts,” Journalof onev Credit and Banking Vol 28, 1996,64-83. Mackie-Mason, Jeffrey. “Do Firms Care Who Provides Their Financing?” in Hubbard, Glenn. cd., svmmetn“cInformation. Comorate Finance. and Investment. Chicago: University of Chicago Press, 1990, 63-103. Oliner, Stephen, and Glenn Rudebush. “The Transmission of Monetary Policy to Small and Large Firms,” Working Paper, Board of Governors of the Federal Reserve System, 1992. Pindyck,Robert. “Irreversibility,Uncertainty and Investment,”Journal of Economic Literature Vol 29, 1991, 1110-1148. Ramey, Valerie. “How important is the credit channel in the transmissionof monetary policy?” Camegie- RochesterConference Series on Public Policy, 39, 1993, 1-46. Scherer, F., and D. Ross. IndustrialMarket Structureand Economic Performance. Boston: HoughtonMifflin Company. Schmalensee, Richard. “Inter-Industry Studies of Structure and Performance,” in Schmalensee, Richard and Robert Willig, eds., Handbook of IndustrialOrganization, Amsterdam: North Holland, 1989. Simon,Herbert, and CharlesBonini. “TheSizeDistributionof BusinessFirms,”American Economic Review 48, 1958,607-17. Small BusinessAssociation. The State of Small Business:A ReDortof the President, 1990.

Stiglitz. Joseph, and Andrew Weiss. “Credit Rationing in Markets with Imperfect Information,” ~erican Economic Review, 1981,393-410, White. Lawrence. “Symposium on Mergers and Antitrust,” Joumal of Economic Perwectives, 1987, 13-22. Winn, Darryl. “On the Relations between Rates of Return, Risk and Market Structure, Ouarterlv Journal of Economics 91, 1977,57-63. 22

Appendix Variation in 6(H) over Time Table Al: Some industrieswithverylow variationovertimein @I) I Industry (SIC) Standard deviation of @I) Natural and processed cheese (2022) 0.0009 I Fibre cans and drums (2655) I 0.0006 Book publishing(2731) I0 . 00 I Toilet preparations(2844) 0.0013 Fabricated structuralmetal (3441) I 0.0012 S m p ( 0.0012 Farm m a e ( I0.0009 Ball bearings (3562) I 0.0015 I Transformers (3612) 0.0011 Motors and generators{3621) I0.0008 I Table A.2: Some industrieswith very high variationover time in CS(II) Industry (SIC) Standard deviation of c(H) Canned seafood (2091) 0.0108 Roasted coffee (2095) 0.0094 T c a f ( 0.0080 Organic fibers (2824) 0.0077 Gum and wood chemicals (2861) 0.0117 Black carbon (2895) 0.0087 Electrometallurgicalproducts(3313) 0.0094 Fabricated pipe fittings (3498) 0.0089 Commercial laundryequipment (3582) 0.0079 Carbon and graphite products(3624) 0.0124 23

Table 1 Summary Statistics on C(H) ALL-Industries Mean Std.Dev. Min. Max. Mean a(n) 0.019 0.007 0.006 0.059 C.v. 6(H) 19 9 6 57 . NQE& 1. First we compute the mean valueof a(~) for each industry in the s T g 2 v o f e t 2 i t f s T r l “ c p t c i s s f t v F e t l v i m c 0 f c t c v ( V f ~ i t s The mw labeled“ a presentsthe cross-industrysummary statisticsfor this variable. Forexample, the representative industry in the full sample has a coefficient of variation of cJ(H)of about 19~0. 24

Table 2 Summary Statistics on C(H) “SMALL” Ilusiness Industries I I IM ean Std.Dev. Min. Max. I Mean (s(I_I) I0.019 0.006 0.007 0.034 m I C O(I-I) 19 . 9 7v. 50 I 1 Jotes: See Table 1. Table 3 Summary Statistics on O(H) “’OTHER”Industries Mean Std.Dev. Min. Max. Mean @I) 0.019 0.008 0.006 0.059 C.v. 6(H) 19 8 6 57 “otes: See Table 1. 25

T 4 Panel Data Summary Statistics Statistic a(rI) (UK) (CFIK) Group 1: ALL Industries (Panel Obs.=5040) Mean 0.018 0.069 0.767 Std.Dev. 0.009 0.033 0.526 Group 2: SiWIALL(Panei Obs.=1340) Mean 0.019 0.068 0.829 Std.Dev. 0.009 0.034 0.497 Group 3: SMALL and CR4 S 4090 (Panel Obs.=460) Mean 0.015 0.070 0.872 Std.Dev. 0.006 0.029 0.579 Group 4: OTHER (Panel Obs.=3700) Mean 0.018 0.069 0.744 S 0 0 0 26

I Table 5 Estimation Results Dependent Variable: (UK),, I GROUP 1: I GROUP2: I GROUP 3: I GROUP 4: ALL SMALL SMALL & OTHER CR4<40% 0.014 G(ll.)i, -0.118** -o.159*** 0.059** un (0.023) (0.039) (0.053) (0.029) me ( 0.282*** 0.339*** 0.337*** 0.269*** ca (0.031) (0.066) (0.110) (0.036) ca r (CF/K),t., O.189*** 0.207*** 0.088 0.181*** (0.033) (0,069) (o.113) (0.037) U K0.317*** )0.235*** i 0.373*** ~ 0,349*** .[ l de ( 3) (0.0270) (0,043) . (0.015) 01 va Panel Obs. 5040 1340 460 3700 # Industries 252 67 23 185 Adj-R2 0.2368 0.2318 0.3308 0.2508 Jotes: 1. All specificationsare estimated with industryfixed-effects and year time-dummies. 2. All variables are measured in logarithms:therefore, the reported coefficients measure elasticities. 3. There are 20 time-series observations (1972-1991) per industry in all samples, Heteroscedasticityconsistent standard errors are in parentheses: ***, ** and * imply statistical significanceat the l~o, 5% and 10%level. 4. GROUP 1is the setof all manufacturingindustriesin the sample, GROUPS 2 consists of industries that are “dominated”by small businesses and GROUP 4 consists of all other industries.GROUP 3 is a subsetof GROUP 2. and GROUP is a subset of GROUP 4. Variable definitions: (UK) = Gross investmentscaled by beginning-of-periodcapital stock. (CF/K) = Cash Flow scaled by beginning-of-periodcapital stock. a(n) = Uncertainty about profit margins l-I= Profit margins, constructed as (Total Sales Revenue - Total Variable Costs)/(TotalSales Revenue), where total variablecosts is the sum of labor, materials and energy costs. 27

T 6 Results with Alternate Measure of Uncertainty: O(alt) GROUP {: GROUP2: GROUP3: GROUP4: ALL SMALL SMALL & OTHER CR4S40 a(alt) 0.001 -O.125*** -O.127*** 0.055** (0.018) (0.035) (0.044) (0.023) Panel Obs. 5040 1340 460 3700 # Industries 252 678 23 185 Adj-R2 0.2368 0.2331 0.3292 0.2510 JNote: Only the uncertainty coefficient estimates are reported; the regressions include all the explanato~ variables shown in Table 5. See notes to Table 5. 28

T 7 Percentile Distribution of Number of Firms and Establishments per Industry: ALL Industries I 1 I 25% I50% I 75% I 909o” #Establishments I 64 I 132 I 309 1571 I 1620 #Firms I 49 I 102 260 I636 1524 I I [Estb/Firm] I 1.02 I 1.06 I 1.14 I 1.34 I 1.57 Jote:The numberofestablishmentsperfirm [Estb/Firm]isfairlycloseto 1.Even atthe90thpercentileval (1.57),there is roughequivalencebetween an establishmentand a firm. Thus data on the sizedistributionof establishments appears to be a reasonableproxy for the size distributionoffirms. T 8 Percentile Distribution of Establishments [ “Small” and “Other” refer to the size classes defined earlier ] 25% 50% 75% Small &<50 Employees 0.699 0.817 0.899 Small & S1OOEmployees 0.857 0,903 0.963 Small & <500 Employees 0.983 0.993 0.999 Other & S50 Employees 0.500 0.650 0.782 , Other & <100 Employees 0,645 0.783 0.886 Other & <500 Employees 0.934 0.976 0.994 r

T 9 SMALL plus Conditioning on the Census Distributionof Establishments M Un S S S 6 - - - 1 ( ( ( O - 3 - 1 - 1 ( { ( P O 7 7 8 # In l O t un co a r t r i a e v shown in Table 5. See notes to Table 5. T OTHER plus Conditioning on the Census Distributionof Establishments Measure of Uncertainty OTHER(50) OTHER(1OO) OTHER(500) G(I-I) 0.058* 0.045 0.077** (0.040) (0.040) (0.043) o(alt) 0.057** 0.044* 0.072** (0.033) (0.029) (0.032) Panel Obs. 1900 1780 1660 # Industries 95 89 83 rf IULG3. 1. only theuncertaintycoefficients are reported;theregressionsincludeallexplanatory variablesshown in Table 5. Also see notes to Table 5. 2. From our “OTHER” group we define three sub-groups: (i) OTHER ~ the percentage of establishments with <50 employees is “less than or equal to” 0.650 (the 50th percentile value). We denote this group as “OTHER”. Similarly, (ii) OTHER(IOO) and (iii) OTHER(500). These segmentationscreate industriesthatare populatedby relativelylargerfirms than the “OTHER” group. 30

T 11 Sales-to-CapitalRatio, (S/K), as the Control Variable M G 1: GROUP 2: GROUP 3: GROUP 4: un A SMALL SMALL & OTHER CR4<40% cT(rI) 0.024 -0.087** -o.154*** 0.061** (0.023) (0,038) (0.053) (0.029) c(alt) 0.021 -0.096*** -0.114*** 0.059*** (0.018) (0.033) (0.042) (0.023) I I Panel Obs. 5040 1340 460 3700 I # Industries 252 67 23 185 I . Jote:Onlythe uncertainty coeffici !ntsare reported, rhe regressionscontain year-time dummies, industryfixed effectsand lagged investment(as in Table 5). The Cash-flow variablesare replaced by current and one-lag of industrysales-to-capitalratio. 31