Add Journal to My Library
Review of Finance
, Volume 22 (1) – Feb 1, 2018

30 pages

/lp/ou_press/growth-option-exercise-and-capital-structure-ZnlylVEWbF

- Publisher
- Oxford University Press
- Copyright
- © The Authors 2016. Published by Oxford University Press on behalf of the European Finance Association. All rights reserved. For Permissions, please email: journals.permissions@oup.com
- ISSN
- 1572-3097
- eISSN
- 1573-692X
- D.O.I.
- 10.1093/rof/rfw062
- Publisher site
- See Article on Publisher Site

Abstract We document that firms decrease their leverage when they convert growth options into tangible assets. We argue that the act of growth option exercise decreases information asymmetry about the firm, which in turn reduces the relative cost of issuing information-sensitive securities such as equity. We show that leverage is negatively correlated with unexpected capital expenditure, our proxy for growth option conversion. The negative relationship becomes stronger when the information environment of a firm deteriorates following a reduction in analyst coverage after a brokerage house merger. Overall, our findings are contrary to standard trade-off and pecking order theories, but are consistent with recent work on signaling and growth options. 1. Introduction How should a firm alter its capital structure when it exercises a growth option? A basic trade-off model would argue that the firm becomes less risky after the exercise of the growth option, which leads to a decrease in the expected bankruptcy costs. As a result, the firm’s optimal leverage ratio should increase. A pecking order view comes to a similar conclusion—the firm should first use internal cash and then debt to finance the investment, with equity being used minimally if at all. In this paper, we show that the intuition from both these perspectives is incomplete. In fact, all else equal, firms decrease their leverage when they engage in capital expenditure, a natural measure of growth option exercise. We build on the argument that, when there is asymmetric information between managers and investors, the optimal exercise of a growth option can communicate information to the market about the quality of the new project (see Grenadier and Malenko [2011] and Morellec and Schürhoff [2011]). As a result, there is effectively a reduction in the relative cost of issuing financial claims that are sensitive to information. This endogenous change in asymmetric information about new projects as a firm expands represents an important dimension to the capital structure trade-off that has been ignored by previous empirical work in this area. A key step in our empirical exercise is to measure the extent of unexpected growth option exercise undertaken by a firm in a given year. The exercise of a growth option implies investment in new projects, and therefore an increase in the capital expenditure of the firm. Broadly, we can think of capital expenditure as consisting of a “predictable” level of investment that the market can forecast and an “unexpected” or surprise amount. Only the surprise component of investment communicates new information to the market. We identify the effect of unexpected capital expenditure on leverage ratios using two empirical designs. In the first design, we consider a comprehensive sample of US firms over the period 1971–2008 and measure unexpected investment as the difference between actual investment in a given year and three measures of a predictable investment level for the firm. First, we take the average capital expenditure ratio of the firm during the sample period as a measure of predictable investment, and estimate the effect of capital expenditure on leverage using a firm-fixed effect specification that captures the effects of investment deviations from the predicted level. Next, we consider the ratio of capital expenditure to assets from the previous year to be the predicted level of investment, and use a first-difference specification to measure unexpected investment. Finally, we estimate an auto-regressive model of expected capital expenditure and take the residual from the model to be unexpected investment. In all three cases, we find a significant negative correlation between capital expenditure and leverage ratios, after controlling for some of the well-known determinants of leverage in the literature. The economic magnitudes are meaningful: depending on the specification, a one standard deviation increase in capital expenditure is associated with a 9–20% decrease in market leverage. Our second research design exploits an exogenous change in the public information environment of some firms. To identify the effects of interest, it is futile for us to look for a random exercise of growth options by firms: by construction, if the market knows firms are behaving randomly, their actions cannot be informative about the quality of the new projects. Instead, we consider variation in the strength of the investment signal resulting from exogenous shocks to the public information environment of some firms. When the quality of public information about a firm worsens, the actions of the firm (including growth option exercise) have a greater impact on investors’ beliefs about the firm. Therefore, the sensitivity of leverage to growth option exercise should strengthen; in particular, growth option exercise should have a more negative effect on leverage after the shock. We follow Hong and Kacperczyk (2010) in considering mergers between two brokerage houses as negative shocks to public information for selected firms. Hong and Kacperczyk document that after such mergers, if both bidder and target have analysts covering the same firm, one of the two analysts is fired. This results in a decline in the information available about the affected firm. Mergers with large targets occurred in the years 1984, 1994, 1997, and 2000, representing staggered shocks over a long period of time. From brokerage house mergers in those years, we form a treatment group of firms that were covered by both bidder and target. The treated firms are matched to a set of comparable control firms that remain unaffected by the brokerage house mergers. We examine the leverage decisions of the matched sample of treatment and control groups for 5 years before and 5 years after a brokerage house merger in a difference-in-differences framework. The key identifying assumption behind this model is that brokerage house mergers and the resulting information decline are exogenous to the firm’s operating, managerial, or financing environment. We compare the difference in the sensitivity of leverage to the capital expenditure ratio for the treatment group before and after the shock to the same difference for the control group. We find strong support for our hypothesis—after the shock, the effect of capital expenditure on leverage is substantially more negative for treated firms when compared with the control firms. In the fixed effects specification, a one standard deviation increase in capital expenditure results in an approximately 7–8% decrease in leverage (book or market) for treated firms. We also examine the financing activity of treated and control firms before and after the shock. We find that, in comparison to control firms, treated firms tilt the financing of investment away from debt and toward equity after the shock. Not only is this consistent with our results on leverage, but also highlights that firms are taking active decisions both in terms of investment and financing. Therefore, our results are not driven simply by some mechanical effect arising from firms being inert. It is important to note that our results are not at odds with the well-known positive relationship between tangibility and leverage (see, e.g., Titman and Wessels [1988] and Rajan and Zingales [1995]). Rather, we show that an abnormally high level of capital expenditure has a negative effect on a firm’s leverage after the tangibility effect has been controlled for. In other words, comparing two firms with similar tangibility, the firm with the higher growth rate in physical assets has lower leverage. Over time, we expect a firm to achieve a steady state with a stable investment level. In such a state, growth option exercise communicates only a limited amount of new information, and the collateral effect of capital expenditure is likely to dominate. Indeed, we show that the relationship between capital expenditure and leverage is concentrated in younger firms, for whom growth option exercise is likely to reveal more information. Our difference-in-differences test based on brokerage house mergers helps rule out several alternative explanations for our results. Nevertheless, we conduct a few additional tests to explicitly rule out some alternatives. Our results are not explained away by market timing (see Baker and Wurgler, 2002), or by the reduced importance of interest tax shields due to higher non-debt tax shields following investment (see DeAngelo and Masulis, 1980). As with other investment-based explanations of capital structure, our results imply that in many contexts one cannot separate investment and financing decisions. That is, the real activities of a firm have an important bearing on its financial structure. The act of investing changes the information asymmetry surrounding the firm’s assets, and hence the optimal mix of debt and equity it should have. These findings have important implications for a large literature in capital structure and security design.1 In particular, we argue that the adverse selection discount to equity should be thought of as an important factor in a trade-off model which includes many of the standard factors including tax shields and bankruptcy costs. An implication for dynamic models of capital structure is that the target leverage ratio for a firm may be a moving target which depends on the assets of the firm at any given point of time.2 Our results complement the work of Lang, Ofek, and Stulz (1996) and Hennessy (2004), who show that debt overhang may reduce capital expenditure. Although their work focuses on distortions in investment decisions due to financing frictions, our work is focused on the effect on financing decisions of information conveyed via investments. Together our papers highlight the intricate inter-temporal links between financing and investment decisions. In their study of seasoned Equity Offerings (SEO) dynamics, Carlson, Fisher, and Giammarino (2006) point out that new investment entails converting risky growth options into less risky assets in place. Consistent with all these papers, we focus on capital expenditure as the key measure of growth option exercise.3 The rest of the paper is organized as follows. We develop our hypothesis in Section 2, and discuss the empirical design in Section 3. In Section 4, we establish a negative correlation between leverage and capital expenditure on our full sample. Results on the matched sample based on brokerage house mergers are contained in Section 5. We discuss some alternative hypotheses in Section 6 and Section 7 concludes. 2. Hypothesis Consider a standard real option model of a firm choosing when to make an irreversible investment, as in McDonald and Siegel (1986) or Dixit and Pindyck (1994). In such a model, if a firm has private information about the value of the project (e.g., based on its knowledge of its own production cost), there is potentially a signaling game in which the price at which the option is exercised may signal the type of the firm. Morellec and Schürhoff (2011) and Grenadier and Malenko (2011) both consider variants of this signaling game, and show that under some conditions there is a separating equilibrium in which the possibility of signaling distorts the investment behavior of a firm. Taking it one step further, Morellec and Schürhoff (2011) show that once the possibility of signaling with financing is added to the model, the implications of a separating equilibrium are nuanced. On the one hand, conditional on separation, the good type faces a lower adverse selection discount to issuing equity, and there is a large set of parameter values for which the good type prefers to finance with equity. However, for some parameter values, it may be cheaper to finance with debt because it leads to less distortion in the optimal investment policy. Overall, in their model, whether the good type prefers the separating equilibrium with equity or with debt depends on the extent of investment distortion under each financing method and on the scale of deadweight bankruptcy costs.4 Motivated by these papers, we postulate that the optimal capital structure for a firm depends on a trade-off over a number of factors, one of which is the adverse selection discount to equity. Other factors in the trade-off may include the tax benefits of debt and the change in expected bankruptcy costs. In a separating equilibrium with equity financing, the new investment is financed with equity. In addition, the reduction in the adverse selection discount to equity tilts the optimal financing mix for existing assets toward greater equity. Both factors imply a reduction in leverage. When the new investment is financed with debt, the second factor remains relevant, so that the implications for leverage are mixed. Taking into account both effects as well as the range of parameter values over which different equilibria are likely to emerge, we formulate our main hypothesis as follows: Hypothesis 1. All else equal, when a firm exercises a growth option, its leverage decreases. Although we consider an asymmetric information setting, our hypothesis is in direct contrast to that of Myers and Majluf (1984). Myers and Majluf (1984) predict that firms will finance their capital expenditure first with internal equity and debt. Only after exhausting these alternatives will they use outside equity. Thus, firms engaged in heavy capital expenditure are likely to experience an increase in leverage. In contrast, Cooney and Kalay (1993) point out that if the new project can have a negative NPV, a good-type firm may be able to separate out and issue equity. Further, Fulghieri, Garcia, and Hackbarth (2015) show that, when there is greater asymmetric information about assets in place than growth options, asymmetric information has a relatively small effect on the right tail of the distribution of firm value. As a result, equity financing can dominate debt financing. Our argument implies that the effect of growth option exercise on leverage depends on the degree to which information asymmetry is reduced when a firm invests. A refinement of our hypothesis provides an auxiliary prediction—the negative effect of growth option exercise on firm leverage should be stronger among the pool of firms that face greater asymmetric information ex ante. We separate our sample into young and old firms to test this prediction. We expect information asymmetry between managers and investors to be greater for younger firms and therefore our results to be stronger on this subset. 3. Empirical Design Our goal is to establish a link between the surprise exercise of growth options and a firm’s leverage. Broadly, we wish to estimate the following regression model: Levit=αi+β GOCit+γ·Xit+yeart+ϵit, (1) where Levit is either the book or market leverage of firm i in year t and GOCit is a proxy for unexpected conversion of growth options into tangible assets by firm i in year t. Here, αi and yeart stand for firm- and year-fixed effects, respectively, and Xit is a vector of control variables. The control variables are described in Section 4.1. A key task in this exercise is to measure GOCit, that is, the extent of the surprise in growth option conversion for firm i in year t. In the canonical real option model, the exercise of a growth option occurs through investment in a new project. The capital expenditure of a firm directly measures its investment in physical assets. Our proxy for GOCit is therefore based on the annual capital expenditure undertaken by firm i between t – 1 and t, scaled by assets at the end of year t – 1 (call this ratio CapExit). An increase in the capital expenditure ratio represents a decision by the firm to expand its scale. However, capital expenditure also includes investment in the maintenance of existing assets and the replenishment of depreciating assets. Thus, CapExit can be represented as: CapExit=PredictableCapExit+UnexpectedCapExit, (2) where PredictableCapExit is the level of investment expected by the financial market. The component of interest to us is UnexpectedCapExit, which represents unexpected investment related to the exercise of a growth option. We use two empirical designs to establish our results. In our first design, we estimate Equation (1) on a large panel of US firms using three different approaches to proxy for the unexpected component of investment. First, suppose a firm maintains roughly a similar level of normal capital expenditure through time (i.e., PredictableCapExit≈PredictableCapExi for all t). Then, a regression with firm-fixed effects will simply subsume the effect of expected investment into a firm-specific intercept. In a within-firm regression, the β coefficient from Equation (1) captures the effect of surprise capital expenditure, and is directly the coefficient of interest to us. Second, suppose the expected level of investment at time t is just its actual level at time t – 1, so that PredictableCapExit=CapExi,t−1. Then, β is determined by estimating Equation (1) in first differences. Third, suppose that the firm’s capital expenditure follows an AR(1) process. After estimating the process, PredictableCapExit is just the forecast level of investment from the regression. The residual is the surprise component UnexpectedCapExit, and is used directly in estimating Equation (1). The first design establishes the relationship between leverage and capital expenditure over a large sample of firms. However, we cannot rule out that the results may be driven by a concurrent change in some other unobservable characteristics of the firm. For example, when a firm undertakes a large capital expenditure project, it may replace its manager at the same time. If the new manager prefers a lower leverage, we would pick up the effect of managerial preference rather than growth option exercise on corporate leverage. Alternatively, it could be that the market-to-book ratio mismeasures future growth opportunities. If the mismeasurement changes at the same time as the firm undertakes capital expenditure, our estimates may be inconsistent. Considering these identification issues, in our second empirical design we use an exogenous variation in the information environment of a firm to establish the causal effect of growth option exercise on leverage. 3.1 Identification via an Exogenous Shock Our underlying model is one with signaling, with the exercise of the growth option being the signal and the firm’s leverage the outcome. An exogenous random assignment of signal (i.e., growth option exercise) across firms does not help us in establishing a causal link from capital expenditure to leverage. If the market knows the assignment is random, by definition the signal has no value and therefore it cannot affect the outcome. Instead, our test relies on exogenous variation in the amount of information communicated by the unexpected conversion of growth options into tangible assets. That is, we consider the effects of exogenous changes in the strength of the signal on the outcome variable. Investors obtain information about a firm through several different channels, including both public sources (such as analyst reports and news stories) and actions taken by the firm. After a negative shock to the public information environment of a firm, the actions of the firm have a greater influence on the market’s beliefs about the firm. That is, if the firm’s public information environment worsens, unexpected growth option exercise should have a greater impact on leverage. Specifically, in Equation (1), the coefficient β should become more negative. Based on this argument, we consider a setting in which the availability of public information on some firms worsens due to exogenous reasons. In particular, we examine mergers of brokerage houses between 1980 and 2000. As documented by Hong and Kacperczyk (2010), in many of these mergers, both acquirer and target had research analysts covering some of the same stocks. After the merger, the combined firm fired the redundant analysts, leading to an overall reduction in the number of analysts covering an affected stock (i.e., a stock that had overlapping analyst coverage). The reduction in the overall number of analysts results in a reduction in the public information available about a subset of firms. Therefore, information asymmetry about the firm increases after the brokerage house mergers.5 Our exclusion restriction relies on the assumption that the merger of two brokerage houses is unlikely to change unobserved characteristics of affected firms in a manner that produces a negative correlation between leverage and capital expenditure. Hong and Kacperczyk (2010) mention that these mergers occur for reasons such as the acquisition of a struggling target or the desire of a foreign bank to expand geographically. These reasons are unlikely to be correlated with the operational and financial characteristics of the firms covered by stock analysts at the acquirer or target. From the set of firms that had dual coverage by analysts at both acquirer and target, we form a treatment group which is matched to a set of control firms in the year before the merger based on industry affiliation, asset size, and analyst following. Details of the shock and the construction of both groups are provided in Section 5. We obtain data on the treatment and the control firms over a 10-year window around the shock. We then compare the effect of capital expenditure on the leverage ratio across the treated and control firms before and after the shock. The resulting difference-in-differences estimator is our coefficient of interest. In this test, we focus on the firm-fixed effects estimation of the effect of capital expenditure on leverage. Specifically, we estimate the following regression model: Levit=αi+δp Post+δr Treat+δpr Post×Treat+δ0 CapExit+δa CapExit×Treat+δb CapExit×Post+β CapExit×Post×Treat+γ·Xit+yeart+ϵit. (3) Here, Treat is a dummy variable set to 1 if firm i belongs to the treatment group and 0 if it belongs to the control group and Post is a dummy variable set to 1 for the years after the shock and zero otherwise. CapExit refers to the capital expenditure of the firm. Essentially, we start with Equation (1), and include the interaction of Treatand Postwith CapExit along with all lower-order interaction terms. As argued earlier, in the firm-fixed effects specification, the β coefficient directly captures the effect of innovation in capital expenditure on leverage. The coefficient of interest is β, which measures the effect of information released through growth option exercise on leverage after differencing out all the above effects. In a nutshell, we consider the difference-in-differences of the slope ∂ Leverage∂ CapEx, using a 5-year window to estimate the slope both before and after the shock. Our main hypothesis is that β is negative; that is, following the shock, growth option exercise has a more negative effect on leverage for treated firms when compared with control firms. 4. Full Sample Results Our full sample consists of US firms covered by the Compustat database over the years 1971–2008. We start our sample in 1971 for two reasons: (i) To avoid the survivorship bias in the Compustat database present in earlier periods (see Davis (1994)), and (ii) Compustat includes flow of funds statements only from 1971 onward; we use these statements to construct measures of financing and operating cash flows. We exclude financial firms and utilities, following much of the empirical capital structure literature. We obtain data on key financial statement variables as well as the market value of equity from the annual Compustat database. We require firms to have non-missing observations on total assets; capital expenditure; leverage; common equity; property, plant, and equipment (PPE); profitability; and the market-to-book ratio to be included in our sample in any given year, since these variables are required for our main empirical specification. A precise definition of the variables used is provided in Appendix A. Following Lemmon, Roberts, and Zender (2008), we require leverage ratios to be between zero and one, and winsorize all other variables at 1% from both tails.6 Table I presents some descriptive statistics for the sample. The summary statistics are comparable to those in prior capital structure studies such as Lemmon, Roberts, and Zender (2008) and Flannery and Rangan (2006).7 Other summary statistics are broadly in line with prior studies as well. As expected, there is a skewness in size, with the mean sales revenue ($1.218 billion) and total assets ($1.368 billion) being significantly larger than their respective medians ($75.8 and $72.3 million, respectively). Notably, the median firm issues a minimal amount of net equity and no net debt in any particular year. The ratio of capital expenditures during a year scaled by beginning-of-year assets, our proxy for growth option exercise, has a median of 5.31% in our sample. Table I Summary statistics for full sample This table shows the means, medians, and standard deviations of some of the key variables in our full sample. The sample consists of all firms in the Compustat database from 1971 to 2008 except for those in the financial and utility industries. Variable definitions are provided in Appendix A. Variable Number of observations Mean Median Standard deviation Sales ($m) 195,019 1,218.36 75.78 7,183.60 Book assets ($m) 195,019 1,368.15 72.35 9,606.16 Book leverage 195,019 0.2524 0.2143 0.2325 Market leverage 195,019 0.2675 0.1939 0.2615 Net equity issued 177,815 0.1764 0.0004 0.7143 Net debt issued 180,620 0.0311 0.0000 0.1691 Tangibility (PPE/total assets) 195,019 0.3711 0.2849 0.3297 Profitability (EBITDA/total assets) 195,019 0.0167 0.1211 0.5088 Market-to-book ratio 195,019 2.2469 1.3289 3.1676 1-year return 158,239 0.1284 0.0163 0.6459 Cap Ex/total assets 195,019 0.0987 0.0531 0.1465 Firm age 195,018 12.7185 9.0000 11.2251 Variable Number of observations Mean Median Standard deviation Sales ($m) 195,019 1,218.36 75.78 7,183.60 Book assets ($m) 195,019 1,368.15 72.35 9,606.16 Book leverage 195,019 0.2524 0.2143 0.2325 Market leverage 195,019 0.2675 0.1939 0.2615 Net equity issued 177,815 0.1764 0.0004 0.7143 Net debt issued 180,620 0.0311 0.0000 0.1691 Tangibility (PPE/total assets) 195,019 0.3711 0.2849 0.3297 Profitability (EBITDA/total assets) 195,019 0.0167 0.1211 0.5088 Market-to-book ratio 195,019 2.2469 1.3289 3.1676 1-year return 158,239 0.1284 0.0163 0.6459 Cap Ex/total assets 195,019 0.0987 0.0531 0.1465 Firm age 195,018 12.7185 9.0000 11.2251 4.1 Capital Expenditure and Leverage 4.1.a. Firm-fixed effects specification In our first test, we estimate Equation (1) by regressing leverage on the capital expenditure ratio and control variables, with firm-fixed effects included. The control variables are based on prior literature (e.g., Rajan and Zingales [1995] and Titman and Wessels [1988]). As a proxy for firm size, we include the log of sales revenue between times t – 1 and t. Profitability over the period t – 1 to t is measured by EBITDA over the year scaled by total assets at time t – 1. The ratio of PPE-to-total assets at time t proxies for the tangibility of the firm’s assets. We measure Tobin’s q by the market-to-book ratio. This variable captures the presence of growth opportunities at a firm. We also include the median leverage in the industry at time t as an explanatory variable to account for other industry-wide factors that may affect leverage. Finally, we include an indicator variable NegEq that takes a value of 1 for firm-year observations with negative book equity values and zero otherwise. Further, to separately assess the impact of Tobin’s q on leverage ratios for these firms, we include an interaction term between q and NegEq.8 The results are presented in Table II. All models include firm- and year-fixed effects. Throughout the paper, we compute robust standard errors clustered at the firm level. Models 1 and 2 present base case estimates with book and market leverage as dependent variables, respectively, and Models 3 and 4 include the previous year’s stock return as an additional regressor. In all four models, we find a statistically significant negative coefficient for the CapEx variable—the leverage ratio falls as a firm increases its capital expenditure, consistent with our hypothesis. Table II Capital expenditure and leverage: full sample with firm-fixed effects This table presents firm-fixed effect regression results of leverage on capital expenditure and control variables for our full sample. The dependent variable is the firm’s book leverage in Models 1 and 3, and market leverage in Models 2 and 4. CapEx/TA measures the ratio of capital expenditure to beginning of the year book value of assets. Industry leverage is the median leverage (book or market) across all firms in the same industry at the same point of time. Neg Eq is a dummy variable set to 1 if the firm has negative book equity in that year. All other variables are defined in Appendix A. All regressions include firm- and year-fixed effects. t-Statistics are displayed in parentheses. Robust standard errors are computed with clustering at the firm level. Dependent Var Model 1 Model 2 Model 3 Model 4 Book leverage Market leverage Book leverage Market leverage Estimate t-Stat Estimate t-Stat Estimate t-Stat Estimate t-Stat CapEx/TA −0.1497 (−20.73) −0.2557 (−32.29) −0.1552 (−19.99) −0.2705 (−31.00) Log Sales 0.0176 (15.54) 0.0227 (18.69) 0.0195 (15.22) 0.0253 (18.64) Profitability −0.0278 (−11.28) −0.0598 (−23.08) −0.0363 (−12.66) −0.0764 (−24.97) Tangibility 0.1183 (24.48) 0.1082 (20.83) 0.1253 (23.64) 0.1257 (21.53) Mkt-to-book (q) −0.0062 (−17.01) −0.0159 (−34.80) −0.0053 (−12.65) −0.0118 (−24.00) Neg Eq 0.2774 (46.42) 0.2296 (44.61) 0.2665 (32.94) 0.2272 (31.67) Neg Eq × q 0.0195 (23.27) 0.0027 (3.54) 0.0166 (8.98) −0.0097 (−5.53) Industry leverage 0.4521 (31.87) 0.4717 (42.81) 0.4615 (30.62) 0.4539 (39.09) 1-year return −0.0118 (−17.89) −0.0445 (−59.94) Firm-fixed effects Yes Yes Yes Yes Year-fixed effects Yes Yes Yes Yes R2 0.671 0.670 0.678 0.700 Number of observation 189,058 189,058 156,241 156,241 Dependent Var Model 1 Model 2 Model 3 Model 4 Book leverage Market leverage Book leverage Market leverage Estimate t-Stat Estimate t-Stat Estimate t-Stat Estimate t-Stat CapEx/TA −0.1497 (−20.73) −0.2557 (−32.29) −0.1552 (−19.99) −0.2705 (−31.00) Log Sales 0.0176 (15.54) 0.0227 (18.69) 0.0195 (15.22) 0.0253 (18.64) Profitability −0.0278 (−11.28) −0.0598 (−23.08) −0.0363 (−12.66) −0.0764 (−24.97) Tangibility 0.1183 (24.48) 0.1082 (20.83) 0.1253 (23.64) 0.1257 (21.53) Mkt-to-book (q) −0.0062 (−17.01) −0.0159 (−34.80) −0.0053 (−12.65) −0.0118 (−24.00) Neg Eq 0.2774 (46.42) 0.2296 (44.61) 0.2665 (32.94) 0.2272 (31.67) Neg Eq × q 0.0195 (23.27) 0.0027 (3.54) 0.0166 (8.98) −0.0097 (−5.53) Industry leverage 0.4521 (31.87) 0.4717 (42.81) 0.4615 (30.62) 0.4539 (39.09) 1-year return −0.0118 (−17.89) −0.0445 (−59.94) Firm-fixed effects Yes Yes Yes Yes Year-fixed effects Yes Yes Yes Yes R2 0.671 0.670 0.678 0.700 Number of observation 189,058 189,058 156,241 156,241 The point estimates of the CapEx variable are about –0.15 for book leverage and about –0.26 for market leverage. These estimates are economically large. For example, a one standard deviation increase in the capital expenditure ratio (14.65 percentage points) corresponds to a decrease of about 3.75 percentage points (i.e., 0.26×0.1465) in a firm’s market leverage ratio based on our estimates. Compared with the sample median market leverage ratio of 19.39%, this represents a decrease of about 20%. Along similar lines, a one standard deviation increase in capital expenditure leads to a 10.2% fall in the book leverage ratios.9 To account for market timing effects, in Models 3 and 4 we include the equity return of the firm over the previous year (measured between time t – 1 and t) as an explanatory variable. As expected, the coefficient of this variable is negative and statistically significant at the 1% level—a higher stock return in the previous year is associated with a lower leverage at the end of the year. Observe that the coefficients of the capital expenditure ratio are slightly larger in magnitude than the comparable coefficients in Models 1 and 2, and continue to be statistically significance at the 1% level. The coefficients of the control variables are statistically significant and have the expected signs. Consistent with the prior literature, most of these control variables are estimated with very high levels of statistical significance (e.g., see Table V of Frank and Goyal [2009] and Table 6 of Graham, Lemmon, and Schallheim [1998]). The leverage ratio increases with industry median leverage, firm size, and the level of asset tangibility and decreases with recent profitability and the market-to-book ratio. Further, negative book equity firms have higher leverage and for such firms the relationship between q and leverage is positive.10 As expected from the previous literature (e.g., Titman and Wessels, 1988), the coefficient of tangibility is positive and statistically significant in all specifications. That is, firms with a greater amount of tangible assets have higher leverage. Our finding that firms with higher capital expenditure have lower leverage does not negate this effect. Specifically, compare two firms that in the previous year had different levels of capital expenditure but are otherwise equal, including on the level of tangibility. We show that the firm with higher capital expenditure has lower leverage. Comparing two other firms which have different quantities of tangible assets, the firm with greater tangibility has higher leverage. Our hypothesis relies on comparing the leverage decisions of two firms that are otherwise equal, with one having a higher level of unexpected investment. We do not make any distinction between the subset of firm-year observations that have a higher than normal level of unexpected capital expenditure and the subset with a lower than normal unexpected capital expenditure. One concern could be that the relationship between capital expenditure and leverage is mainly driven by distressed firms that cut their investments at a time when their leverage is also going up. To address this concern, we re-estimate Equation (1) with a slight modification, in which we include capital expenditure ratios above the firm-level mean and ratios below the mean as separate variables. We find that the coefficients for both high (i.e., above the firm-level mean) and low capital expenditure ratios (below the firm-level mean) are negative and approximately equal in magnitude. Therefore, our results are not driven just by firms with either excessively high or excessively low investment. For brevity, we do not report the detailed results of this test. 4.1.b. First-difference specification Our second test uses a first-difference specification to estimate Equation (1). Here, we use the annual change in the book or market leverage ratio as the dependent variable. All explanatory variables are differenced in the same manner. In this model, the key explanatory variable is the annual change in the capital expenditure ratio. That is, the first-difference model considers the previous year’s capital expenditure ratio as the expected capital expenditure ratio for the firm. Under this interpretation, the coefficient β measures the change in the leverage ratio in response to incremental capital expenditure during the year. The results of the estimation are provided in Table III. The coefficients of ΔCapExit are approximately –0.06 when the dependent variable is the change in book leverage, and –0.12 when the dependent variable is the change in market leverage. In all four models, the coefficient is statistically significant at the 1% level. The results are economically meaningful as well: a one standard deviation increase in capital expenditure during the year (13.65 percentage points) is associated with a decrease of about 1.65 percentage points in the market leverage ratio (about 9% of the sample median). We therefore find strong support for our hypothesis: abnormally high levels of capital expenditure are associated with lower leverage ratios. Table III Capital expenditure and leverage: full sample, first-difference model This table presents first-difference regression results of leverage on capital expenditure and control variables for our full sample. The dependent variable is the annual change in firm’s book leverage in Models 1 and 3, and annual change in market leverage in Models 2 and 4. Models 1 and 2 are estimated on entire sample, Models 3 and 4 exclude observations with negative book equity. ΔX measures the change in variable X from 1 year to the next; for example, ΔCapEx/TA measures the change in capital expenditure to assets ratio over the year. All other variables are defined in Appendix A. All regressions include year-fixed effects. t-Statistics are displayed in parentheses. Standard errors are computed with clustering at the firm level. Dependent Var Model 1 Model 2 Model 3 Model 4 ΔBook leverage ΔMarket leverage ΔBook leverage ΔMarket leverage Estimate t-Stat Estimate t-Stat Estimate t-Stat Estimate t-Stat ΔCapEx/TA −0.0635 (−7.39) −0.1231 (−20.65) −0.0580 (−10.60) −0.1213 (−20.56) ΔLog Sales 0.0107 (3.08) 0.0168 (14.14) 0.0131 (12.43) 0.0178 (15.35) ΔProfitability −0.0119 (−2.49) −0.0309 (−14.19) −0.0155 (−6.82) −0.0328 (−14.91) ΔTangibility 0.0651 (10.22) 0.1045 (26.12) 0.0932 (25.54) 0.1143 (28.73) ΔMarket-to-Book 0.0038 (2.36) −0.0037 (−11.91) −0.0017 (−5.72) −0.0037 (−12.04) ΔIndustry Leverage 0.2425 (17.17) 0.2312 (33.53) 0.2118 (25.61) 0.2286 (33.60) Δ1-year return −0.0148 (−12.43) −0.0410 (−73.96) −0.0092 (−22.09) −0.0391 (−72.39) Firm-fixed effects No No No No Year-fixed effects Yes Yes Yes Yes R2 0.019 0.201 0.052 0.209 Number of observation 140,226 140,226 133,980 133,980 Dependent Var Model 1 Model 2 Model 3 Model 4 ΔBook leverage ΔMarket leverage ΔBook leverage ΔMarket leverage Estimate t-Stat Estimate t-Stat Estimate t-Stat Estimate t-Stat ΔCapEx/TA −0.0635 (−7.39) −0.1231 (−20.65) −0.0580 (−10.60) −0.1213 (−20.56) ΔLog Sales 0.0107 (3.08) 0.0168 (14.14) 0.0131 (12.43) 0.0178 (15.35) ΔProfitability −0.0119 (−2.49) −0.0309 (−14.19) −0.0155 (−6.82) −0.0328 (−14.91) ΔTangibility 0.0651 (10.22) 0.1045 (26.12) 0.0932 (25.54) 0.1143 (28.73) ΔMarket-to-Book 0.0038 (2.36) −0.0037 (−11.91) −0.0017 (−5.72) −0.0037 (−12.04) ΔIndustry Leverage 0.2425 (17.17) 0.2312 (33.53) 0.2118 (25.61) 0.2286 (33.60) Δ1-year return −0.0148 (−12.43) −0.0410 (−73.96) −0.0092 (−22.09) −0.0391 (−72.39) Firm-fixed effects No No No No Year-fixed effects Yes Yes Yes Yes R2 0.019 0.201 0.052 0.209 Number of observation 140,226 140,226 133,980 133,980 4.1.c. AR(1)specification for capital expenditure Finally, we model capital expenditure as a first-order autoregressive process to determine the surprise component of capital expenditure in a given year. We first fit the following model: CapExit=αi+βCapExi,t−1+δt+ϵit. (4) Since the model includes firm-fixed effects with a lagged dependent variable, we estimate the model using the Arellano–Bond GMM estimator, with the first lagged value of capital expenditure as an instrument. Based on the estimates, the residual ϵit (call the residual CapEx_Res) gives us the unexpected component of capital expenditure for each firm-year observation. In the next step, we estimate the key leverage regressions with CapEx_Resas the main proxy for surprise growth option exercise. The results are provided in Models 1–4 of Table IV. Consistent with our hypothesis, we find a negative and significant coefficient for the CapEx_Resvariable. The results are economically meaningful. A one standard deviation increase in CapEx_Res(9.4 percentage points) results in a decrease of about 2 percentage points in market leverage ratio. Compared with the sample median market leverage ratio of 19.39%, this represents a decrease of about 10%. In an unreported test, we include two other explanatory variables in the CapEx model of Equation (4), the market-to-book ratio and cash flows-to-total asset ratio of the firm. We re-estimate Equation (4) with these control variables, and obtain the residual from this model as our proxy for innovation in growth option exercise. The results are very similar. Table IV Capital expenditure and leverage: full sample, AR(1) specification for capital expenditure This table presents regression results of leverage on residual capital expenditure and control variables for our full sample. CapEx_Res is the residual from an AR(1) model of capital expenditure estimated using the Arellano–Bond strategy. The dependent variable is the firm’s book leverage in Models 1 and 3, and market leverage in Models 2 and 4. Industry leverage is the median leverage (book or market) across all firms in the same industry at the same point of time. All other variables are defined in Appendix A. All regressions include year-fixed effects. t-Statistics are displayed in parentheses. Standard errors are computed with clustering at the firm level. Dependent Var Model 1 Model 2 Model 3 Model 4 Book leverage Market leverage Book leverage Market leverage Estimate t-Stat Estimate t-Stat Estimate t-Stat Estimate t-Stat CapEx_Res −0.1372 (−18.86) −0.2221 (−27.90) −0.1315 (−16.95) −0.2148 (−24.92) Log Sales 0.0168 (13.22) 0.0227 (16.57) 0.0188 (13.18) 0.0252 (16.54) Profitability −0.0479 (−13.15) −0.0846 (−21.64) −0.0767 (−15.16) −0.1295 (−23.04) Tangibility 0.1274 (23.74) 0.1039 (18.18) 0.1384 (23.98) 0.1231 (19.47) Mkt-to-Book(q) −0.0067 (−15.47) −0.0182 (−31.47) −0.0058 (−11.35) −0.0138 (−22.70) Neg Eq 0.2705 (42.10) 0.2213 (39.72) 0.2620 (30.37) 0.2211 (28.39) Neg Eq × q 0.0194 (20.08) 0.0038 (4.26) 0.0156 (7.79) −0.0099 (−4.95) Industry leverage 0.4540 (30.27) 0.4737 (40.68) 0.4620 (29.18) 0.4527 (37.35) 1-year returns −0.0103 (−14.57) −0.0422 (−52.72) Firm-fixed effects Yes Yes Yes Yes Year-fixed effects Yes Yes Yes Yes R2 0.682 0.681 0.688 0.710 N 168,128 168,128 141,054 141,054 Dependent Var Model 1 Model 2 Model 3 Model 4 Book leverage Market leverage Book leverage Market leverage Estimate t-Stat Estimate t-Stat Estimate t-Stat Estimate t-Stat CapEx_Res −0.1372 (−18.86) −0.2221 (−27.90) −0.1315 (−16.95) −0.2148 (−24.92) Log Sales 0.0168 (13.22) 0.0227 (16.57) 0.0188 (13.18) 0.0252 (16.54) Profitability −0.0479 (−13.15) −0.0846 (−21.64) −0.0767 (−15.16) −0.1295 (−23.04) Tangibility 0.1274 (23.74) 0.1039 (18.18) 0.1384 (23.98) 0.1231 (19.47) Mkt-to-Book(q) −0.0067 (−15.47) −0.0182 (−31.47) −0.0058 (−11.35) −0.0138 (−22.70) Neg Eq 0.2705 (42.10) 0.2213 (39.72) 0.2620 (30.37) 0.2211 (28.39) Neg Eq × q 0.0194 (20.08) 0.0038 (4.26) 0.0156 (7.79) −0.0099 (−4.95) Industry leverage 0.4540 (30.27) 0.4737 (40.68) 0.4620 (29.18) 0.4527 (37.35) 1-year returns −0.0103 (−14.57) −0.0422 (−52.72) Firm-fixed effects Yes Yes Yes Yes Year-fixed effects Yes Yes Yes Yes R2 0.682 0.681 0.688 0.710 N 168,128 168,128 141,054 141,054 4.1.d. Effect of firm age In our next test, we show that our key results are stronger among young firms. Information asymmetry between managers and investors is a greater problem for such firms. For example, Hadlock and Pierce (2010) find that firm age and size are the two most important predictors of financial constraints. Therefore, we expect that the surprise component of capital expenditure sends a stronger signal to the market about firm quality for these firms. We create an indicator variable young that equals 1 for firms that fall below the 25th percentile in terms of firm age and zero for all other firms. We estimate our main regression model (1) with firm-fixed effects by including two additional variables in the model: Young and Young × CapEx. The results are documented in Table V. We find negative and significant coefficients on the interaction term Young × CapEx, indicating that our results are stronger for younger firms. The coefficient of CapEx/TA remains negative and significant by itself. Together these results indicate that the effect of growth option exercise on leverage is negative for all firms, but it is stronger for younger firms. In Models (3) and (4) of the table, we refine the analysis by estimating the effect of capital expenditure on young as well as small firms. We create an indicator variable Y-small that equals one for firms that are both young (i.e., below 25th percentile in terms of age) and small (i.e., below 25th percentile in terms of annual sales). The results are stronger for young and small firms. Overall, these results show that the effect of capital expenditure on leverage is higher for the subset of firms where information asymmetry is likely to be higher. Table V Capital expenditure and leverage: full sample, effects of firm age This table presents firm-fixed effect regression results of leverage on capital expenditure and control variables for our full sample. The dependent variable is the firm’s book leverage in Models 1 and 3, and market leverage in Models 2 and 4. CapEx/TA measures the ratio of capital expenditure to beginning of the year book value of assets. Young is an indicator variable that equals 1 for firms that fall below the 25th percentile in age distribution. Y-small is an indicator variable that equals 1 for firms that fall below the 25th percentile both in terms of age and size as measured by yearly sales. Industry leverage is the median leverage (book or market) across all firms in the same industry at the same point of time. Neg Eq is a dummy variable set to 1 if the firm has negative book equity in that year. All other variables are defined in Appendix A. All regressions include firm- and year-fixed effects. t-Statistics are displayed in parentheses. Robust standard errors are computed with clustering at the firm level. Dependent Var Model 1 Model 2 Model 3 Model 4 Book leverage Market leverage Book leverage Market leverage Estimate t-Stat Estimate t-Stat Estimate t-Stat Estimate t-Stat CapEx/TA −0.1076 (−11.57) −0.2392 (−22.52) −0.1341 (−16.26) −0.2563 (−27.22) Young −0.0095 (−4.77) −0.0384 (−17.26) Young×CapEx/TA −0.0727 (−7.74) −0.0202 (−2.02) Y-small −0.0120 (−3.88) −0.0348 (−10.93) Y-small×CapEx/TA −0.0623 (−5.77) −0.0236 (−2.18) Log Sales 0.0178 (13.74) 0.0224 (16.38) 0.0185 (14.32) 0.0235 (17.19) Profitability −0.0376 (−13.14) −0.0744 (−24.82) −0.0410 (−13.84) −0.0805 (−25.78) Tangibility 0.1247 (23.66) 0.1248 (21.59) 0.1249 (23.62) 0.1251 (21.52) Mkt-to-book (q) −0.0049 (−11.75) −0.0109 (−22.90) −0.0051 (−12.24) −0.0114 (−23.61) Neg Eq 0.2656 (32.83) 0.2248 (31.50) 0.2662 (32.92) 0.2264 (31.66) Neg Eq × q 0.0162 (8.78) −0.0104 (−6.01) 0.0162 (8.78) −0.0103 (−5.88) Industry leverage 0.4632 (30.64) 0.4501 (38.80) 0.4628 (30.67) 0.4517 (38.94) 1-year return −0.0123 (−18.61) −0.0458 (−61.59) −0.0119 (−17.99) −0.0448 (−60.42) Firm-fixed effects Yes Yes Yes Yes Year-fixed effects Yes Yes Yes Yes R2 0.679 0.702 0.678 0.701 N 156,241 156,241 156,241 156,241 Dependent Var Model 1 Model 2 Model 3 Model 4 Book leverage Market leverage Book leverage Market leverage Estimate t-Stat Estimate t-Stat Estimate t-Stat Estimate t-Stat CapEx/TA −0.1076 (−11.57) −0.2392 (−22.52) −0.1341 (−16.26) −0.2563 (−27.22) Young −0.0095 (−4.77) −0.0384 (−17.26) Young×CapEx/TA −0.0727 (−7.74) −0.0202 (−2.02) Y-small −0.0120 (−3.88) −0.0348 (−10.93) Y-small×CapEx/TA −0.0623 (−5.77) −0.0236 (−2.18) Log Sales 0.0178 (13.74) 0.0224 (16.38) 0.0185 (14.32) 0.0235 (17.19) Profitability −0.0376 (−13.14) −0.0744 (−24.82) −0.0410 (−13.84) −0.0805 (−25.78) Tangibility 0.1247 (23.66) 0.1248 (21.59) 0.1249 (23.62) 0.1251 (21.52) Mkt-to-book (q) −0.0049 (−11.75) −0.0109 (−22.90) −0.0051 (−12.24) −0.0114 (−23.61) Neg Eq 0.2656 (32.83) 0.2248 (31.50) 0.2662 (32.92) 0.2264 (31.66) Neg Eq × q 0.0162 (8.78) −0.0104 (−6.01) 0.0162 (8.78) −0.0103 (−5.88) Industry leverage 0.4632 (30.64) 0.4501 (38.80) 0.4628 (30.67) 0.4517 (38.94) 1-year return −0.0123 (−18.61) −0.0458 (−61.59) −0.0119 (−17.99) −0.0448 (−60.42) Firm-fixed effects Yes Yes Yes Yes Year-fixed effects Yes Yes Yes Yes R2 0.679 0.702 0.678 0.701 N 156,241 156,241 156,241 156,241 4.2 Informational Effects of Capital Expenditure We have shown that capital expenditure is negatively correlated with leverage. In this section, we provide some direct evidence to show that it is also correlated with various measures of information asymmetry of the firm. As a prelude to our natural experiment in Section 5, we consider the effect of capital expenditure on the dispersion of analyst earnings forecasts. If capital expenditure does indeed communicate information about the quality of the firm, we expect that an increase in capital expenditure is associated with a reduction in the dispersion of analyst earnings forecasts. We estimate the following equation: Dispit=αi+ϕCapExit+ψSizeit+yeart+ϵit, (5) where Dispit is a measure of dispersion of analyst earnings forecasts and Sizeit is a measure of firm size. In the regressions, we use the natural log of total assets as our proxy for size.11 Data on analyst forecasts of next year’s earnings per share are obtained from the I/B/E/S summary files. We match each firm-year observation in our full sample with the most recent statistical forecast period of the I/B/E/S dataset. We compute several measures of dispersion. Our first measure is the standard deviation of EPS forecasts among analysts covering the stock. Our second measure is the standard deviation of forecasts divided by the absolute value of the average EPS forecast (i.e., the coefficient of variation). For the third measure, we scale the range of analyst forecasts (the highest minus the lowest forecast) by the average EPS forecast (this measure is only constructed for firm-year observations with a positive average forecast). All measures are winsorized at the 1% level from both tails. The results of the estimation are shown in Models 1–3 of Table VI. We find a strong negative effect of capital expenditure on the dispersion of analyst forecasts across all three measures. That is, when a firm exercises a growth option, analysts converge on their forecasts of future earnings. This result is consistent with growth option exercise communicating information to the market about the quality of the new project being undertaken by the firm. Table VI Capital expenditure and investors’ beliefs: full sample This table presents regression results relating analyst earnings forecast dispersion and forecast error to capital expenditure on our full sample. Model 1 uses σEPS, the standard deviation of 1-year ahead earnings per share forecasts, as the dependent variable. In Model 2, we scale this measure by the absolute value of the mean EPS forecast. Model 3 uses the range in analyst forecasts (i.e., the difference between the highest and lowest EPS forecast) scaled by the mean EPS forecast. This model is estimated using only firm-year observations with a positive EPS forecast. Model 4 uses the consensus forecast error scaled by the mean EPS forecast as the dependent variable. CapEx/TA measures the capital expenditure to assets ratio. Log Assets is the natural log of the book value of assets. t-Statistics are displayed in parentheses. Standard errors are computed with clustering at the firm level. Dependent Var Model 1 Model 2 Model 3 Model 4 σEPS σEPS|Mean EPS| EPS RangeMean EPS Forecast ErrorMean EPS Estimate t-Stat Estimate t-Stat Estimate t-Stat Estimate t-Stat CapEx/TA −0.0410 (−2.51) −0.4968 (−11.30) −0.5676 (−9.70) −1.1532 (−5.50) Log Assets 0.0331 (10.85) −0.0094 (−1.77) 0.0673 (8.41) 0.0742 (2.68) Firm-fixed effects Yes Yes Yes Yes Year-fixed effects Yes Yes Yes Yes R2 0.509 0.226 0.292 0.214 Number of observation 54,986 54,874 57,601 60,640 Dependent Var Model 1 Model 2 Model 3 Model 4 σEPS σEPS|Mean EPS| EPS RangeMean EPS Forecast ErrorMean EPS Estimate t-Stat Estimate t-Stat Estimate t-Stat Estimate t-Stat CapEx/TA −0.0410 (−2.51) −0.4968 (−11.30) −0.5676 (−9.70) −1.1532 (−5.50) Log Assets 0.0331 (10.85) −0.0094 (−1.77) 0.0673 (8.41) 0.0742 (2.68) Firm-fixed effects Yes Yes Yes Yes Year-fixed effects Yes Yes Yes Yes R2 0.509 0.226 0.292 0.214 Number of observation 54,986 54,874 57,601 60,640 As an additional measure of information quality, we analyze the relationship between the consensus forecast error and capital expenditure. The explanatory variable is the absolute value of the difference between the consensus forecast of 1-year ahead earnings and the actual earnings in that year, scaled by the consensus forecast. The results are shown in Model 4 of Table VI. We find a negative association between capital expenditure and the forecast error. In an unreported test, we consider whether growth option exercise leads to a reduction in information asymmetry across investors. We use the Amihud illiquidity measure (Amihud, 2002) as the dependent variable in Equation (5). We find that the coefficient of the capital expenditure ratio, ϕ, is again strongly negative. Overall, our results show that capital expenditure is associated with a reduction in information asymmetry across investors. 5. Brokerage House Mergers: Matched Sample Results Having established a correlation between growth option exercise and leverage, we now estimate a difference-in-differences specification using brokerage house mergers as an exogenous shock to a firm’s information environment. We first provide some details of the exogenous event, brokerage house mergers, and our construction of the treatment and control samples. We then turn to our results on the matched sample. 5.1 Brokerage House Mergers Hong and Kacperczyk (2010, Table III) document fifteen brokerage house mergers between 1984 and 2005, and for each merger provide the number of overlapping stocks that bidder and target both covered. As they document, six of the mergers are particularly large relative to the rest of the sample (both in terms of size of bidder and target, and in the number of overlapping stocks covered). The large mergers occur in the years 1984, 1994, 1997, and 2000. We consider only these merger years. As we are working with annual data, we include all eight mergers that occur in these years. Overall, the mergers we consider account for 94.1% of affected stocks in the Hong and Kacperczyk (2010) sample. Including the remaining 5.9% of affected stocks in our analysis does not change our results, but leads to considerable time-clustering in our observations. We list each of the eight mergers we consider in our study in Table VII. Table VII List of brokerage house mergers This table lists the acquirer and target in the brokerage house mergers we include in our analysis, and the date of the merger. (Source: Hong and Kacperczyk, 2010). Panel A: List of mergers Year Acquirer Target Date of merger 1984 Merrill Lynch Becker Paribas September 10, 1984 1994 Paine Webber Kidder Peabody December 31, 1994 1997 Morgan Stanley Dean Witter Reynolds May 31, 1997 Smith Barney Salomon Brothers November 28, 1997 2000 Credit Suisse First Boston Donaldson Lufkin and Jenrette October 15, 2000 UBS Warburg Paine Webber December 10, 2000 Chase Manhattan JP Morgan December 31, 2000 Paine Webber JC Bradford June 12, 2000 Panel A: List of mergers Year Acquirer Target Date of merger 1984 Merrill Lynch Becker Paribas September 10, 1984 1994 Paine Webber Kidder Peabody December 31, 1994 1997 Morgan Stanley Dean Witter Reynolds May 31, 1997 Smith Barney Salomon Brothers November 28, 1997 2000 Credit Suisse First Boston Donaldson Lufkin and Jenrette October 15, 2000 UBS Warburg Paine Webber December 10, 2000 Chase Manhattan JP Morgan December 31, 2000 Paine Webber JC Bradford June 12, 2000 In each case involving a large merger, the target is a full-fledged brokerage firm with underwriting and trading arms. Hong and Kacperczyk (2010) mention that there are several reasons for these mergers: (i) acquisition of troubled targets in two cases, (ii) geographic diversification by Swiss firms in two other cases, (iii) increased access to retail investors in one case, and (iv) expected “synergy” in the sixth case. None of these reasons appears likely to be related to the operating and financial characteristics of firms covered by the research analysts of either the acquirer or the target. The mergers may therefore be treated as events that are exogenous to the policies of the treated firms. That is, the exclusion restriction necessary for our identification strategy is likely to be satisfied. 5.2 Treatment and Control Firms For each of the eight brokerage house mergers, we classify firms that were covered by both bidder and target analysts before the merger as affected firms.12 As in previous analysis, we exclude financial firms and utilities from the analysis. In the matched sample analysis, the number of treated firms drop further due to our strict matching criteria which we discuss below.13 For each affected firm in the sample, we find a set of control firms that are similar to the affected firm before the shock in terms of industry, size, and analyst coverage. Size and industry are both important dimensions to match on—firms of similar size in the same industry are expected to be similar in terms of factors such as growth opportunities, the information environment, and asset tangibility. Many of these factors directly affect a firm’s leverage decision. Finally, by considering the number of analysts before the shock, we ensure that treated and control firms have a similar public information environment prior to the shock. Our matching procedure is as follows. We obtain data from Compustat for a period of ±5 years around a brokerage house merger for all affected firms. These firm-year observations are excluded from the potential set of control firms. That is, a firm is a potential control firm only if it remains unaffected in the 5 years before and 5 years after a brokerage house merger. For every affected firm, we find all firms in the same two-digit SIC code in the fiscal year ending just before the merger date. From this set, we obtain all firms with total assets within 20% of the assets of the affected firm. From this new subset, we keep as control firms up to five firms that are closest to the affected firm in terms of the number of analysts. We match without replacement: Once a firm enters the control group, we exclude it from further matching. To ensure that our matching results are not driven by a specific sequence of matches, we randomly order the list of affected firms before proceeding with the exercise. As is common with matching procedures, the assumptions we make represent a trade-off between bias and efficiency. Picking control firms from the same industry and close in asset size to the affected firm helps to eliminate bias. The trade-off is that some affected firms cannot be matched, leading to a reduction in the sample size. Similarly, by matching without replacement we increase the set of control firms, providing greater independent variation in the data. However, again we lose some affected firms because the set of potential control firms shrinks each time a firm is matched. Our results are robust to small changes in the assumptions in the matching process. An affected firm enters the treatment group only if we find at least one control firm for it. For some large affected firms, we cannot find any firms in the same industry that are within 20% in size. Overall the median firm in the control sample is slightly smaller in terms of total assets than the median firm in the treatment sample. For each firm in the treatment and control groups, we consider the 5 years after the merger as the “post” period for our analysis. We include the year of the merger in the post period. A treated firm remains in the treatment group for 5 years after the merger. If, during that span, it is affected by a subsequent merger, the second shock is ignored. Table VIII provides the number of treated and control firms for each shock. We have a total of 751 affected firms. After losing about 300 firms in the matching process, we finally end up with a treatment group of 450 firms. We find 1,391 control firms (or just over three firms per treatment firm) to enter our matched sample of treatment and control firms. Table VIII Brokerage merger shock: number of firms in treatment and control groups This table presents the number of firms affected across all brokerage house mergers included for the given year. The second column provides the number of firms that were covered by analysts at both acquirer and target. The third column represents the smaller sub-sample of the dual coverage firms for which we find at least one matching control firm, based on our matching criteria. The fourth column represents the total number of control firms (i.e., those unaffected by the shock) used for the treatment group firms in that year. Year Firms with dual coverage Treatment group Control group Total number 1984 129 69 221 290 1994 151 82 212 294 1997 245 137 410 547 2000 226 162 548 710 Total 751 450 1,391 1,841 Year Firms with dual coverage Treatment group Control group Total number 1984 129 69 221 290 1994 151 82 212 294 1997 245 137 410 547 2000 226 162 548 710 Total 751 450 1,391 1,841 To check that the treatment and control groups are similar before the relevant merger, we plot the kernel densities of their assets, number of analysts, book leverage, and capital expenditure in the year before the merger. Since the number of control firms is different across the treated firms, we weight each treated firm by the number of matches present in the control group. The Epanechnikov kernel densities of the treatment and control samples are shown in Figure 1. As seen from the figure, the size distributions are practically identical across the two groups. The distributions of capital expenditure and book leverage are very similar across the groups even though we did not use these criteria for the matching exercise. Treated firms are followed by a slightly lower number of analysts before the merger.14 Figure 1. View largeDownload slide Distribution of key characteristics of treatment and control firms after matching, brokerage merger matched sample. Notes: The plots give the kernel density functions of the key characteristics of the treatment and control firms after matching on the brokerage merger matched sample. These plots are based on fiscal year data just before the merger of brokerage houses, that is, based on the characteristics as of the matching date. More details on the matching are provided in the paper. Figure 1. View largeDownload slide Distribution of key characteristics of treatment and control firms after matching, brokerage merger matched sample. Notes: The plots give the kernel density functions of the key characteristics of the treatment and control firms after matching on the brokerage merger matched sample. These plots are based on fiscal year data just before the merger of brokerage houses, that is, based on the characteristics as of the matching date. More details on the matching are provided in the paper. Table IX provides summary statistics on the matched sample firms, including both treatment and control groups. The unit of observation here is a firm-year pair. In terms of either sales revenue or total assets, the median firm in the matched sample is about ten times as large as the median firm in the full Compustat sample that we analyzed earlier in the paper. Further, the median firm in the matched sample has a higher market-to-book ratio, profitability, and stock return than the median firm in the full sample. The greater stock return of the matched sample likely reflects the fact that mergers tend to occur when stock prices are high. On other dimensions, median firms across the matched and full samples are broadly similar. Table IX Characteristics of the matched sample based on brokerage house mergers This table shows the means, medians, and standard deviations of some of the key variables for the matched sample of firms based on brokerage house mergers. See the text for details on how treatment and control firms are chosen. The statistics are based on the pooled sample of all treatment and control group firms around the merger date. All annual observations within the range of ± 5 years from the brokerage house merger date are included in the sample. Variable definitions are provided in Appendix A. Variable Number of observations Mean Median Standard deviation Sales ($m) 14,664 2,635.55 724.63 7,127.64 Total assets ($m) 14,664 2,934.42 773.34 10,320.84 Book leverage 14,664 0.2509 0.2229 0.2083 Market leverage 14,664 0.2375 0.1784 0.2294 Net equity issued 13,075 0.0565 0.0008 0.2701 Net debt issued 13,417 0.0395 0.0000 0.1663 PPE/total assets 14,664 0.3223 0.2758 0.2183 EBITDA/total assets 14,664 0.1630 0.1613 0.1452 Market-to-book ratio 14,664 2.0760 1.5420 1.6195 1-year return 14,475 0.2065 0.1091 0.6263 Cap Ex/total assets 14,664 0.0937 0.0652 0.0982 Variable Number of observations Mean Median Standard deviation Sales ($m) 14,664 2,635.55 724.63 7,127.64 Total assets ($m) 14,664 2,934.42 773.34 10,320.84 Book leverage 14,664 0.2509 0.2229 0.2083 Market leverage 14,664 0.2375 0.1784 0.2294 Net equity issued 13,075 0.0565 0.0008 0.2701 Net debt issued 13,417 0.0395 0.0000 0.1663 PPE/total assets 14,664 0.3223 0.2758 0.2183 EBITDA/total assets 14,664 0.1630 0.1613 0.1452 Market-to-book ratio 14,664 2.0760 1.5420 1.6195 1-year return 14,475 0.2065 0.1091 0.6263 Cap Ex/total assets 14,664 0.0937 0.0652 0.0982 5.3 Matched Sample Results We estimate Equation (3) on the matched sample using both book and market leverage as the dependent variable. A number of treated firms also appear as control firms for a different shock, allowing us to include firm-fixed effects in addition to the Treat variable. Year-fixed effects and the Post variable capture the impact of aggregate macro-economic conditions on all firms. Here, Post is a dummy variable equal to 1 if year t is one of the 5 years after a merger and 0 otherwise. Our shocks are staggered over time, further allowing us to separate out the effect of brokerage house mergers from macro-economic conditions. The interaction of the Treatand Postvariable captures the effect of the reduction in analyst coverage on unconditional (i.e., independent of capital expenditure) changes in the leverage of treated firms. We are interested in the incremental effect of capital expenditure on leverage for treated firms when compared with control firms in periods just after the shock. In our model, the coefficient of CapExit captures the average effect of capital expenditure on leverage in the matched sample. As with other mergers, it is likely that brokerage houses merge during periods of good market conditions. To the extent that firms fund their capital expenditure differently across good and bad markets, the effect of CapEx on leverage may be different during the pre- and post-merger periods. We separate out that effect by including the interaction of CapEx and Post. Similarly, if the treated firms in general adopt a different financing strategy than the control group, the unconditional effect of CapEx on leverage is going to be different across the two groups. We include the interaction of CapEx and Treatto separate out this effect. The coefficient of CapEx×Treat×Post is of greatest interest to us. This coefficient estimates the differential effect of the shock on the relationship between the capital expenditure ratio and leverage for firms that are affected by the shock (treatment firms) compared with firms that remain unaffected (control firms). The estimation results are presented in Table X. Models 1–4 present benchmark cases that omit some control variables. We discuss our findings based on Models 5 and 6, which include all the control variables. The estimated β coefficient of CapEx×Treat×Post is –0.16 when book leverage is the dependent variable (significant at the 1% level) and –0.13 when market leverage is the dependent variable (significant at the 5% level). In economic terms, adding up the coefficients of the four CapEx terms, we find that, after the shock, a one standard deviation increase in capital expenditure for a treated firm leads to a reduction in leverage of about 7–8%, compared with a control firm. We therefore find strong support for our hypothesis: the unexpected exercise of growth options leads to reduced leverage.15 Table X Capital expenditure and leverage: brokerage merger matched sample This table presents a difference-in-differences estimation of the effect of capital expenditure on leverage ratios across treatment and control groups for the brokerage merger matched sample. The dependent variable is the firm’s book leverage in Models 1, 3, and 5, and market leverage in Models 2, 4, and 6. Treat is an indicator variable that equals 1 for firms that are affected by the brokerage house merger and zero otherwise. Post equals 1 for 5 fiscal years after the brokerage house merger and zero otherwise. CapEx/TA measures the ratio of capital expenditure to beginning of the year book value of assets. Industry leverage is the median leverage (book or market) across all firms in the same industry at the same point of time. All other variables are defined in Appendix A. t-Statistics are displayed in parentheses. Standard errors are computed with clustering at the firm level. Dependent Var Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Book leverage Market leverage Book leverage Market leverage Book leverage Market leverage Estimate t-Stat Estimate t-Stat Estimate t-Stat Estimate t-Stat Estimate t-Stat Estimate t-Stat Treat −0.0298 (−2.04) −0.0354 (−2.41) −0.0286 (−1.98) −0.0294 (−2.06) −0.0243 (−1.89) −0.0269 (−2.04) Post −0.0057 (−0.78) −0.0044 (−0.55) −0.0060 (−0.88) −0.0044 (−0.62) −0.0064 (−1.02) −0.0040 (−0.58) Treat × Post 0.0059 (0.52) 0.0137 (1.18) 0.0093 (0.86) 0.0077 (0.74) 0.0129 (1.35) 0.0104 (1.09) CapEx/TA −0.1838 (−5.28) −0.3584 (−9.13) −0.1077 (−3.09) −0.1684 (−4.73) −0.0491 (−1.55) −0.1348 (−3.99) CapEx/TA×Treat 0.0593 (0.98) 0.0722 (1.16) 0.0494 (0.82) 0.0387 (0.64) 0.0547 (1.01) 0.0425 (0.76) CapEx/TA×Post 0.1297 (2.60) 0.1410 (2.73) 0.0531 (1.11) 0.0406 (0.90) 0.0259 (0.58) 0.0185 (0.42) CapEx/TA×Treat×Post −0.1572 (−1.93) −0.1669 (−1.94) −0.1508 (−1.94) −0.1210 (−1.60) −0.1619 (−2.38) −0.1319 (−2.01) Log Sales 0.0199 (3.91) 0.0238 (4.61) 0.0196 (4.32) 0.0240 (5.05) Profitability −0.1078 (−4.94) −0.1871 (−9.10) −0.1002 (−5.09) −0.1818 (−9.39) Tangibility 0.1582 (4.69) 0.1315 (3.75) 0.1183 (3.90) 0.1087 (3.25) Mkt-to-book (q) −0.0037 (−2.06) −0.0116 (−7.14) −0.0081 (−4.93) −0.0122 (−7.65) 1-yr return −0.0039 (−1.66) −0.0310 (−14.20) −0.0011 (−0.53) −0.0300 (−14.42) Industry leverage 0.3142 (10.23) 0.3762 (17.45) 0.2909 (10.53) 0.3647 (17.40) Neg Eq 0.2038 (8.50) 0.2531 (11.06) Neg Eq × q 0.0243 (3.24) −0.0361 (−5.17) Firm-fixed effects Yes Yes Yes Yes Yes Yes Year-fixed effects Yes Yes Yes Yes Yes Yes R2 0.702 0.695 0.741 0.762 0.772 0.774 Number of observation 14,664 14,664 14,408 14,408 14,408 14,408 Dependent Var Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Book leverage Market leverage Book leverage Market leverage Book leverage Market leverage Estimate t-Stat Estimate t-Stat Estimate t-Stat Estimate t-Stat Estimate t-Stat Estimate t-Stat Treat −0.0298 (−2.04) −0.0354 (−2.41) −0.0286 (−1.98) −0.0294 (−2.06) −0.0243 (−1.89) −0.0269 (−2.04) Post −0.0057 (−0.78) −0.0044 (−0.55) −0.0060 (−0.88) −0.0044 (−0.62) −0.0064 (−1.02) −0.0040 (−0.58) Treat × Post 0.0059 (0.52) 0.0137 (1.18) 0.0093 (0.86) 0.0077 (0.74) 0.0129 (1.35) 0.0104 (1.09) CapEx/TA −0.1838 (−5.28) −0.3584 (−9.13) −0.1077 (−3.09) −0.1684 (−4.73) −0.0491 (−1.55) −0.1348 (−3.99) CapEx/TA×Treat 0.0593 (0.98) 0.0722 (1.16) 0.0494 (0.82) 0.0387 (0.64) 0.0547 (1.01) 0.0425 (0.76) CapEx/TA×Post 0.1297 (2.60) 0.1410 (2.73) 0.0531 (1.11) 0.0406 (0.90) 0.0259 (0.58) 0.0185 (0.42) CapEx/TA×Treat×Post −0.1572 (−1.93) −0.1669 (−1.94) −0.1508 (−1.94) −0.1210 (−1.60) −0.1619 (−2.38) −0.1319 (−2.01) Log Sales 0.0199 (3.91) 0.0238 (4.61) 0.0196 (4.32) 0.0240 (5.05) Profitability −0.1078 (−4.94) −0.1871 (−9.10) −0.1002 (−5.09) −0.1818 (−9.39) Tangibility 0.1582 (4.69) 0.1315 (3.75) 0.1183 (3.90) 0.1087 (3.25) Mkt-to-book (q) −0.0037 (−2.06) −0.0116 (−7.14) −0.0081 (−4.93) −0.0122 (−7.65) 1-yr return −0.0039 (−1.66) −0.0310 (−14.20) −0.0011 (−0.53) −0.0300 (−14.42) Industry leverage 0.3142 (10.23) 0.3762 (17.45) 0.2909 (10.53) 0.3647 (17.40) Neg Eq 0.2038 (8.50) 0.2531 (11.06) Neg Eq × q 0.0243 (3.24) −0.0361 (−5.17) Firm-fixed effects Yes Yes Yes Yes Yes Yes Year-fixed effects Yes Yes Yes Yes Yes Yes R2 0.702 0.695 0.741 0.762 0.772 0.774 Number of observation 14,664 14,664 14,408 14,408 14,408 14,408 Turning to the other coefficients, we find that the coefficient of Treatis negative. That is, treatment firms have a slightly lower leverage on average than control firms. Leverage is not affected by the shock for either control or treatment firms (the coefficients of Post and Treat×Post are not significantly different from zero). Further, the effect of capital expenditure on leverage is similar for both groups of firms before the shock (the coefficient of CapEx×Treat is not statistically different from zero). The effect is also similar for control firms before and after the shock (the coefficient of CapEx×Post is not statistically different from zero). In line with our full sample result, the coefficient of capital expenditure is negative. These results show that the exercise of growth options result in a decline in leverage ratios and the effect becomes stronger for the treated group after an exogenous increase in asymmetric information. We conduct two robustness exercises related to the brokerage merger shocks. For brevity, we briefly describe these exercises but do not report the entire regression tables in the paper. First, in choosing the control group for the matched sample, we include the firm’s growth rate as a dimension to match on in a propensity score model. Specifically, we estimate a probit model for each year with treatment status as the dependent variable and four explanatory variables: the two-digit SIC code, firm size (the natural log of total assets), the number of analysts following the shock, and the sales growth rate over the previous year. For each treated firm, we then select up to five control firms that fall with ±2.5% of the estimated probability of the treatment firm. The results are essentially similar. Our second robustness test is designed to mitigate concerns about the lack of parallel trend across the treatment and control group prior to the shock. Note that we compare the difference in the sensitivity of leverage to capital expenditure across the treatment and control group around the brokerage house merger date. Since the computation of slope requires multiple years of observation, we cannot plot this variable on an yearly basis to graphically depict the parallel trend. As an alternative, in a robustness test, we include separate trend variables for the treatment and control groups in the regression reported in Table X. The results are unchanged. Thus, it is unlikely that the difference-in-differences results are entirely driven by differential trends across the two groups around the shock. 5.4 Financing Results We have shown that the exercise of growth options has a negative effect on leverage. We now check the active debt and equity financing decisions of treated and control firms in our matched sample using an empirical framework based on the financing deficit regression models of Shyam-Sunder and Myers (1999) and Frank and Goyal (2003). These authors estimate the fraction of the financing deficit that an average firm bridges through the issuance of debt as follows: ΔDit=α+βFinDefit+ϵit. (6) Here, FinDefit denotes the financing deficit of firm i in year t, ΔDit is the net issuance of long-term debt by firm i in year t. The coefficient β estimates the fraction of financing needs funded by debt issuance. By the standard accounting identity equating uses and sources of funds, we can write FinDefit=Divit+Invit+ΔWorkCapit−CashFlowit=ΔDit+ΔEit, (7) where Divit is dividends paid out by the firm at time t, Invit is the net investment (including capital expenditure, acquisitions, and sale of PPE), ΔWorkCapit the change in working capital over the year, CashFlowit the cash flow after interest and taxes, and ΔEit the net issuance of equity. All of the variables in Equation (7) can be scaled by total assets at beginning of the year without affecting the identity. We can further break up the financing deficit into capital expenditure, CapExit, and a remaining component that does not depend on capital expenditure, Non−CapEx Defit=Divit+OtherInvit+ΔWorkCapit−CashFlowit, where OtherInvit refers to components of net investment other than capital expenditure. We estimate the following equation on our matched sample: ΔDit=αi+yeart+β CapExit×Post×Treat+δaCapExit×Post+ δbCapExit×Treat+δ0×CapExit+δpPost+δrTreat+ δprPost×Treat+δnNon-CapExDefit+ϵit. (8) We also estimate Equation (8) with net equity issuance as the explanatory variable. The coefficient β measures the extent of capital expenditure funded by active issuance of debt by the treated firm after the shock. The results are shown in Table XI. The coefficient of interest is again β, the coefficient of the variable CapExit×Treat×Post. In Models 1 and 3, we omit the component of the financing deficit not related to capital expenditure, Non-CapEx Defit. The dependent variable is net equity issuance for Model 1 and net long-term debt issuance for Model 3, scaled by beginning of year assets in each case. The coefficient β is positive (0.3215) and significant at the 10% level in Model 1, and negative (–0.2105) and significant at approximately the 5% level in Model 3. That is, in a difference-in-differences sense, after the shock, treatment firms finance more of their capital expenditure through equity and less through debt when compared with control firms. Therefore, our leverage result arises not through mere inertia, but rather through active financing choices made by firms. Table XI External financing of capital expenditure: brokerage merger matched sample This table shows the relationship between capital expenditure and equity (Models 1 and 2) and debt (Models 3 and 4) issuance by the firms for the brokerage merger matched sample. Net equity (debt) issuance is new issuance of equity (long-term debt) minus repurchases (retirements). The dependent variable in each model is scaled by assets at time t – 1. Cap Ex refers to capital expenditure between t – 1 and t divided by total assets at time t – 1. Non-CapEx Deficit equals total financing deficit minus annual capital expenditure. All regressions include firm- and year-fixed effects. t-Statistics are displayed in parentheses. Standard errors are computed with clustering at the firm level. Dependent Var Model 1 Model 2 Model 3 Model 4 Net equity issued Net equity issued Net debt issued Net debt issued Estimate t-Stat Estimate t-Stat Estimate t-Stat Estimate t-Stat Treat 0.0012 (0.06) 0.0173 (1.30) −0.0278 (−2.31) −0.0197 (−1.53) Post 0.0475 (3.69) 0.0456 (5.00) −0.0398 (−4.48) −0.0415 (−5.05) Treat×Post −0.0185 (−0.92) −0.0313 (−2.53) 0.0302 (2.79) 0.0316 (2.92) Cap Ex/TA 1.4527 (12.27) 0.7240 (11.42) 0.5385 (10.16) 0.2873 (4.39) Cap Ex/TA×Treat −0.3262 (−1.53) −0.1750 (−1.38) 0.0808 (0.89) 0.1177 (0.98) Cap Ex/TA×Post −0.6654 (−6.32) −0.3512 (−4.54) 0.1637 (2.47) 0.2834 (3.94) Cap Ex/TA×Treat×Post 0.3215 (1.69) 0.2906 (2.36) −0.2105 (−1.96) −0.2780 (−2.53) 1-year return 0.0530 (8.29) 0.0143 (3.33) 0.0018 (0.51) −0.0125 (−3.57) Non-CapEx Deficit 0.6136 (31.05) 0.2390 (14.10) Firm-fixed effects Yes Yes Yes Yes Year-fixed effects Yes Yes Yes Yes R2 0.392 0.750 0.185 0.344 Number of observation 12,905 11,873 13,237 11,873 Dependent Var Model 1 Model 2 Model 3 Model 4 Net equity issued Net equity issued Net debt issued Net debt issued Estimate t-Stat Estimate t-Stat Estimate t-Stat Estimate t-Stat Treat 0.0012 (0.06) 0.0173 (1.30) −0.0278 (−2.31) −0.0197 (−1.53) Post 0.0475 (3.69) 0.0456 (5.00) −0.0398 (−4.48) −0.0415 (−5.05) Treat×Post −0.0185 (−0.92) −0.0313 (−2.53) 0.0302 (2.79) 0.0316 (2.92) Cap Ex/TA 1.4527 (12.27) 0.7240 (11.42) 0.5385 (10.16) 0.2873 (4.39) Cap Ex/TA×Treat −0.3262 (−1.53) −0.1750 (−1.38) 0.0808 (0.89) 0.1177 (0.98) Cap Ex/TA×Post −0.6654 (−6.32) −0.3512 (−4.54) 0.1637 (2.47) 0.2834 (3.94) Cap Ex/TA×Treat×Post 0.3215 (1.69) 0.2906 (2.36) −0.2105 (−1.96) −0.2780 (−2.53) 1-year return 0.0530 (8.29) 0.0143 (3.33) 0.0018 (0.51) −0.0125 (−3.57) Non-CapEx Deficit 0.6136 (31.05) 0.2390 (14.10) Firm-fixed effects Yes Yes Yes Yes Year-fixed effects Yes Yes Yes Yes R2 0.392 0.750 0.185 0.344 Number of observation 12,905 11,873 13,237 11,873 In Models 2 and 4, we add the non-capital expenditure portion of the financing deficit as an explanatory variable, and obtain similar results. The coefficient of CapEx×Treat×Post is now 0.2906 in the net equity issuance regression and –0.2780 in the net debt issuance regression. Both coefficients are significant at the 1% level. That is, we again find that, after the shock, when compared with control firms, treated firms tilt the financing of their capital expenditures toward equity and away from debt. Comparing the coefficients of CapEx/TA and CapEx×Treat across Models 2 and 4, firms as a whole finance more of their capital expenditures by issuing equity than issuing debt. The coefficient of 0.2873 for the capital expenditure ratio in Model 4 is consistent with the findings of Frank and Goyal (2003) on the proportion of the financing deficit funded through debt.16 Overall our results provide strong support for our hypothesis that the act of growth option conversion tilts financing decisions in favor of equity claims. 6. Robustness Tests Our difference-in-differences specification in Section 5 rules out any alternative explanation unless it has a differential effect on treated and control firms. Nevertheless, in this section, we explicitly consider a few possible alternative explanations for our results. 6.1 Market Timing Baker and Wurgler (2002) show that managers opportunistically issue equity when the equity price is high in an attempt to time the market. At any given point of time, a firm’s capital structure will reflect the effects of such past market timing decisions. As noted earlier, all our results remain similar when we include past year’s stock returns as an explanatory variable in the model. One may argue that the stock return over the previous year is not enough to capture the effect of market timing, as equity issuances from more than a year ago may still affect leverage in any given year. To address this issue, in an unreported robustness test, we compute an external financing weighted average of the historical market-to-book ratio following Kayhan and Titman (2007). We include this variable as a control variable in our regressions on the full sample. Our results remain robust to the inclusion of this variable. 6.2 Mergers and Acquisitions Many acquisitions are financed by stocks that may result in lower leverage for the acquiring firm. However, our results are not driven by acquisitions. First, we exclude acquisition-related expenditures from the construction of the capital expenditure variable for our study. Second, as a robustness check, over the full sample we re-estimate the regression of leverage on the capital expenditure ratio and other control variables after dropping all observations with very high acquisition expenditure in a year. Our results remain robust, both in the statistical and economic sense, to the exclusion of firm-year observations that fall in the top 20th, 10th, or 5th percentile of acquisition expense (scaled by total assets). 6.3 Non-debt Tax Shields Another alternative hypothesis is that a firm engaging in high capital expenditure is likely to earn substantial tax shields from depreciation. The presence of non-debt tax shields naturally lowers the value of interest tax shields. As a result, the firm reduces its leverage (see DeAngelo and Masulis, 1980). We explicitly include both the depreciation expense and the investment tax credit claimed in a given year (scaled by beginning of year assets in each case) in our leverage regressions to account for this effect. Our results remain similar. For brevity, we do not report the details of these regressions. 7. Conclusion We hypothesize that the unexpected exercise of a growth option signals the quality of the new project to the market. The resulting reduction in information asymmetry between managers and investors implies a corresponding reduction in the adverse selection discount to equity. As a result, the firm’s leverage ratio decreases. We use abnormal increases in investment (or capital expenditure) as an empirical proxy for the unexpected exercise of growth options. On a large sample of Compustat firms over almost a 40-year period, we document that the leverage ratio (book or market) decreases as unexpected capital expenditure increases. We provide three pieces of evidence in the paper to establish that capital expenditure affects leverage by changing the information available about the firm. First, we show that analyst forecasts about the firm’s future earnings converge after the exercise of growth options. Thus, capital expenditure indeed communicates value-relevant information to outside investors. Second, our results are predominantly driven by young firms, which are informationally more opaque. Finally, we consider an empirical design that exploits an exogenous shock to the firm’s information environment which is directly in the spirit of our argument. We show that affected firms exhibit a greater negative sensitivity of leverage to capital expenditure after the shock, in comparison to a control group of firms. We also show that firms actively tilt the financing of capital expenditure toward equity after the shock. Taken together, our results provide evidence for the signaling mechanism underlying our hypothesis. Our results suggest that to explain capital structure it is important to consider changes in the information asymmetry between the managers and outside investors of a firm as it exercises a growth option. Much of the current literature focuses on the risk reduction that follows such option exercise and ignores the effect the conversion has on the information sensitivity of the firm’s assets. Our empirical findings stand in contrast to the predictions obtained from standard models of capital structure, and suggest that endogenous informational effects are an important component of the trade-off faced by a firm when it chooses its financing mix. Footnotes 1 A number of important articles have contributed to this literature. See the survey papers by Myers (2003), Frank and Goyal (2009), and Graham and Leary (2011) for details and references. 2 Strebulaev and Whited (2012) provide a detailed review of the existing literature on dynamic capital structure models of corporate finance. Hackbarth and Mauer (2012) study the interactions between financing and investment decision in a dynamic model, and derive predictions for the optimal debt-to-equity ratio and the priority structure of debt. 3 Lang, Ofek, and Stulz (1996) also consider the growth rate in employees as a measure of firm growth. In an unreported test, we find essentially similar results using both the employee growth rate and the sales revenue growth rate as relevant measures. In our full sample, leverage is strongly negatively associated with both measures, even after controlling for its other known drivers. 4 Morellec and Schürhoff (2011) also show that for some parameter values, the high type may prefer a pooling equilibrium in which both types invest at the same threshold and issue equity. For simplicity, in motivating our hypothesis, we focus on the separating equilibrium. 5 Derrien and Kesckés (2013) show that the increased information asymmetry leads to a reduction in both the level of investment and financing of the affected firms, consistent with the idea that their cost of capital has increased. Irani and Oesch (2013) show that the information asymmetry may be further worsened by managers of affected firms reducing the quality of information disclosure after a brokerage house merger. 6 Our results are essentially unchanged if we also winsorize leverage ratios at 1% from both tails, ignore the restriction that they must lie between zero and one, or simply drop observations with leverage ratios greater than one from the sample. 7 Welch (2011) points out some important issues in the computation of leverage ratios using balance sheet variables, in particular with defining book leverage as total debt/total assets. To facilitate comparison with the prior literature, we continue to use this as our definition of book leverage. We have conducted several unreported tests using alternative definitions; our results stand. 8 Our results remain unchanged if we drop the NegEq dummy entirely from the regression model. 9 In a robustness test, we subtract a firm’s depreciation and amortization expenses for the year (which are arguably predictable) from capital expenditure. The net capital expenditure is then scaled by beginning of year assets and used as an explanatory variable in regression Models 1 and 2 of Table II. We find coefficients of − 0.13 (t-statistic −19.19) and −0.22 (t-statistic −29.53) for the net capital expenditure variable in book and market leverage-based regressions, respectively. For brevity, we do not report the full regression results in the paper. 10 As a robustness exercise, we also consider an unreported specification that adds the interaction of the negative book equity dummy with capital expenditure in the regressions. The coefficient of this interaction term is not significantly different from zero. 11 The results are similar if we use the natural log of sales revenue as the measure of size. 12 The list of affected firms is available at Marcin Kacperczyk’s web site. 13 In one specification reported in Section 5.3 we relax the matching criteria to include all affected firms in the sample. Our results continue to obtain. 14 To ensure that our results are not driven by this small difference in number of analysts across treatment and control firms, in an unreported robustness exercise we include the number of analysts as an additional regressor, and obtain similar results. 15 In an unreported test, we shorten the window around the brokerage shock to ±2 and ±1 years around the shock date. Our results remain robust on these shorter windows as well. 16 de Jong, Verbeek, and Verwijmeren (2010) show that for large financing deficits the proportion financed through debt is significantly lower. References Amihud Y. ( 2002) Illiquidity and stock returns: cross-section and time series effects, Journal of Financial Markets 5, 31– 56. Google Scholar CrossRef Search ADS Baker M., Wurgler J. ( 2002) Market timing and capital structure, Journal of Finance 57, 1– 32. Google Scholar CrossRef Search ADS Carlson M., Fisher A., Giammarino R. ( 2006) Corporate investment and asset price dynamics: implications for SEO event studies and long-run performance, Journal of Finance 61, 1009– 1034. Google Scholar CrossRef Search ADS Cooney J., Kalay A. ( 1993) Positive information from equity issue announcement, Journal of Financial Economics 33, 149– 172. Google Scholar CrossRef Search ADS Davis J. ( 1994) The cross-section of realized stock returns: the pre-COMPUSTAT evidence, Journal of Finance 49, 1579– 1593. Google Scholar CrossRef Search ADS de Jong A., Verbeek M., Verwijmeren P. ( 2010) The impact of financing surpluses and large financing deficits on tests of the pecking order theory, Financial Management 39, 733– 756. Google Scholar CrossRef Search ADS DeAngelo H., Masulis R. ( 1980) Optimal capital structure under corporate and personal taxation, Journal of Financial Economics 8, 3– 29. Google Scholar CrossRef Search ADS Derrien F., Kecskés A. ( 2013) The real effects of financial shocks: evidence from exogenous changes in analyst coverage, Journal of Finance 68, 1383– 1416. Google Scholar CrossRef Search ADS Dixit A., Pindyck R. ( 1994) Investment under Uncertainty , Princeton University Press, Princeton. Flannery M., Rangan K. ( 2006) Partial adjustment toward target capital structures, Journal of Financial Economics 79, 469– 506. Google Scholar CrossRef Search ADS Frank M., Goyal V. ( 2003) Testing the pecking order theory of capital structure, Journal of Financial Economics 67, 217– 248. Google Scholar CrossRef Search ADS Frank M., Goyal V. ( 2009) Capital structure decisions: which factors are reliably important?, Financial Management 38, 1– 37. Google Scholar CrossRef Search ADS Fulghieri P., Garcia D., Hackbarth D. ( 2015) Asymmetric information, security design, and the pecking (dis)order. CEPR Discussion Paper No. DP10660. Graham J., Lemmon M., Schallheim J. ( 1998) Debt, leases, taxes, and the endogeneity of corporate tax status, Journal of Finance 53, 131– 162. Google Scholar CrossRef Search ADS Graham J., Leary M. ( 2011) A review of empirical capital structure research and directions for the future, Annual Review of Financial Economics 3, 309– 345. Google Scholar CrossRef Search ADS Grenadier S., Malenko A. ( 2011) Real options signaling games with applications to corporate finance, Review of Financial Studies 24, 3993– 4036. Google Scholar CrossRef Search ADS Hackbarth D., Mauer D. ( 2012) Optimal priority structure, capital structure, and investment, Review of Financial Studies 25, 585– 609. Google Scholar CrossRef Search ADS Hadlock C., Pierce J. ( 2010) New evidence on measuring financial constraints: moving beyond the KZ index, Review of Financial Studies 23, 1909– 1940. Google Scholar CrossRef Search ADS Hennessy C. ( 2004) Tobin’s Q, debt overhang, and investment, Journal of Finance 59, 1717– 1742. Google Scholar CrossRef Search ADS Hong H., Kacperczyk M. ( 2010) Competition and bias, Quarterly Journal of Economics 125, 1683– 1725. Google Scholar CrossRef Search ADS Irani R., Oesch D. ( 2013) Monitoring and corporate disclosure: evidence from a natural experiment, Journal of Financial Economics 109, 389– 418. Google Scholar CrossRef Search ADS Kayhan A., Titman S. ( 2007) Firms’ histories and their capital structures, Journal of Financial Economics 83, 1– 32. Google Scholar CrossRef Search ADS Lang L., Ofek E., Stulz R. ( 1996) Leverage, investment, and firm growth, Journal of Financial Economics 40, 3– 29. Google Scholar CrossRef Search ADS Lemmon M., Roberts M., Zender J. ( 2008) Back to the beginning: persistence and the cross-section of corporate capital structure, Journal of Finance 63, 1575– 1608. Google Scholar CrossRef Search ADS McDonald R., Siegel D. ( 1986) The value of waiting to invest, Quarterly Journal of Economics 101, 707– 728. Google Scholar CrossRef Search ADS Morellec E., Schürhoff N. ( 2011) Corporate investment and financing under asymmetric information, Journal of Financial Economics 99, 262– 288. Google Scholar CrossRef Search ADS Myers S. ( 2003) Financing of corporations, in: Constantinides G. M., Harris M., Stulz R. (eds.), Handbook of the Economics of Finance , 1 Chapter 4, Elsevier Publishing, Amsterdam. Myers S., Majluf N. ( 1984) Corporate financing and investment decisions when firms have information that investors do not have, Journal of Financial Economics 13, 187– 221. Google Scholar CrossRef Search ADS Rajan R., Zingales L. ( 1995) What do we know about capital structure? Some evidence from international data, Journal of Finance 50, 1421– 1460. Google Scholar CrossRef Search ADS Shyam-Sunder L., Myers S. ( 1999) Testing static trade-off against pecking order models of capital structure, Journal of Financial Economics 51, 219– 244. Google Scholar CrossRef Search ADS Strebulaev I., Whited T. ( 2012) Dynamic models and structural estimation in corporate finance. Working paper, SSRN. Titman S., Wessels R. ( 1988) The determinants of capital structure choice, Journal of Finance 43, 1– 19. Google Scholar CrossRef Search ADS Welch I. ( 2011) Two common problems in capital structure research: the financial-debt-to-asset ratio and issuing activity versus leverage changes, International Review of Finance 11, 1– 17. Google Scholar CrossRef Search ADS Appendix A Variable Definitions Variable Definition σ-EPS Standard deviation of 1-year ahead analyst earnings forecasts 1-year return Stock return over the previous 12 months Book debt Long-term debt + short-term debt Book leverage Book debt/total assets, both at time t Cap Ex/TA Capital expenditure between t – 1 and t/total assets at t – 1 Financing deficit Sum of net equity issued and net debt issued Market leverage Book debt/(book debt + market value of equity), all at time t Market-to-book (Total assets – book value of equity + market value of equity)/total assets all at time t Mean EPS Mean of 1-year ahead analyst earnings forecasts Neg Eq Dummy set to 1 if firm has negative book equity at time t Net equity issued (New equity issued – stock repurchases) between t – 1 and t/total assets at t – 1 Net debt issued New long-term debt financing minus retirement between t – 1 and t/total assets at t – 1 Non-CapEx deficit Dividends + net investment excluding capital expenditure + change in working capital – cash flow, all measured between times t – 1 and t Profitability Earnings before interest, taxes, depreciation, and amortization between t – 1 and t/total assets at t – 1 Tangibility Plant, property, and equipment/total assets, both at time t Variable Definition σ-EPS Standard deviation of 1-year ahead analyst earnings forecasts 1-year return Stock return over the previous 12 months Book debt Long-term debt + short-term debt Book leverage Book debt/total assets, both at time t Cap Ex/TA Capital expenditure between t – 1 and t/total assets at t – 1 Financing deficit Sum of net equity issued and net debt issued Market leverage Book debt/(book debt + market value of equity), all at time t Market-to-book (Total assets – book value of equity + market value of equity)/total assets all at time t Mean EPS Mean of 1-year ahead analyst earnings forecasts Neg Eq Dummy set to 1 if firm has negative book equity at time t Net equity issued (New equity issued – stock repurchases) between t – 1 and t/total assets at t – 1 Net debt issued New long-term debt financing minus retirement between t – 1 and t/total assets at t – 1 Non-CapEx deficit Dividends + net investment excluding capital expenditure + change in working capital – cash flow, all measured between times t – 1 and t Profitability Earnings before interest, taxes, depreciation, and amortization between t – 1 and t/total assets at t – 1 Tangibility Plant, property, and equipment/total assets, both at time t View Large © The Authors 2016. Published by Oxford University Press on behalf of the European Finance Association. All rights reserved. For Permissions, please email: journals.permissions@oup.com

Review of Finance – Oxford University Press

**Published: ** Feb 1, 2018

Loading...

personal research library

It’s your single place to instantly

**discover** and **read** the research

that matters to you.

Enjoy **affordable access** to

over 18 million articles from more than

**15,000 peer-reviewed journals**.

All for just $49/month

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Read from thousands of the leading scholarly journals from *SpringerNature*, *Elsevier*, *Wiley-Blackwell*, *Oxford University Press* and more.

All the latest content is available, no embargo periods.

## “Hi guys, I cannot tell you how much I love this resource. Incredible. I really believe you've hit the nail on the head with this site in regards to solving the research-purchase issue.”

Daniel C.

## “Whoa! It’s like Spotify but for academic articles.”

@Phil_Robichaud

## “I must say, @deepdyve is a fabulous solution to the independent researcher's problem of #access to #information.”

@deepthiw

## “My last article couldn't be possible without the platform @deepdyve that makes journal papers cheaper.”

@JoseServera

DeepDyve ## Freelancer | DeepDyve ## Pro | |
---|---|---|

Price | FREE | $49/month |

Save searches from | ||

Create lists to | ||

Export lists, citations | ||

Read DeepDyve articles | Abstract access only | Unlimited access to over |

20 pages / month | ||

PDF Discount | 20% off | |

Read and print from thousands of top scholarly journals.

System error. Please try again!

or

By signing up, you agree to DeepDyve’s Terms of Service and Privacy Policy.

Already have an account? Log in

Bookmark this article. You can see your Bookmarks on your DeepDyve Library.

To save an article, **log in** first, or **sign up** for a DeepDyve account if you don’t already have one.

All DeepDyve websites use cookies to improve your online experience. They were placed on your computer when you launched this website. You can change your cookie settings through your browser.

ok to continue