The WACC Fallacy: The Real Eects of Using a Unique Discount Rate 1

Philipp Kr ger u
Geneva Finance Research Institute - Universit de Gen`ve e e

Augustin Landier
Toulouse School of Economics

David Thesmar
HEC Paris and CEPR

First Version: February 2011 This Version: September 2011

We greatly appreciate comments and suggestions by Malcolm Baker, Andor Gyrgy, Owen Lamont, o Masahiro Watanabe, Je Wurgler and seminar participants at the NBER Behavioral Finance Spring Meeting, the University of Mannheim, the 2011 European Financial Management Association meetings, the 2011 European Finance Association meetings, the CEPR European Summer Symposium on Financial Markets and HEC Lausanne. Boris Valle provided excellent research assistance. Thesmar e thanks the HEC Foundation for nancial support. Corresponding Author: Philipp Krger. Email: u philipp.krueger@unige.ch, Telephone: +41 (0)22 379 85 69. Augustin Landier, augustin.landier@tsefr.eu, Telephone: +33 (0)5 61 12 86 88. David Thesmar, thesmar@hec.fr, Telephone: +33 (0)1 39 67 94 12.

Electronic copy available at: http://ssrn.com/abstract=1764024

The WACC Fallacy: The Real Eects of Using a Unique Discount Rate

Abstract We document investment distortions induced by the use of a single discount rate within rms. According to textbook capital budgeting, rms should value any project using a discount rate determined by the risk characteristics of the project. If they use a unique company-wide discount rate, they overinvest (resp. underinvest) in divisions with a market beta higher (resp. lower) than the rms core industry beta. We directly test this consequence of the WACC fallacy and establish a robust and signicant positive relationship between division-level investment and the spread between the divisions market beta and the rms core industry beta. Consistently with bounded rationality theories, this bias is stronger when the measured cost of taking the wrong discount rate is low, for instance, when the division is small. Finally, we measure the value loss due to the WACC fallacy in the context of acquisitions. Bidder abnormal returns are higher in diversifying mergers and acquisitions in which the bidders beta exceeds that of the target. On average, the present value loss is about 0.8% of the bidders market equity.

JEL-Classication: G11, G31, G34 Keywords: Investment, Behavioral nance, Cost of capital

Electronic copy available at: http://ssrn.com/abstract=1764024

Ever since the seminal contribution of Modigliani and Miller (1958), a key result of corporate nance theory is that a projects cash ows should be discounted at a rate that reects the projects risk characteristics. Discounting cash ows at the rms weighted average cost of capital (WACC) is therefore inappropriate if the project diers in terms of its riskiness from the rest of the rms assets. In stark contrast, however, survey evidence suggests that performing capital-budgeting using a unique rm-level WACC is quite common. Graham and Harvey (2001) show that a large majority of rms report using a rm-wide discount rate to value a project independently of its risk characteristics. Similarly, Bierman (1993) surveys the top 100 rms of the Fortune 500 and nds that 93% of the responding rms use their rm-wide WACC to value projects and only 35% also rely on division-level discount rates. The potential distortions that rms might face if they discount projects at their rm-wide WACC are prominently underlined in standard corporate nance textbooks. Grinblatt and Titman (2002) note that the WACC of a rm is the relevant discount rate for [...] one of its projects only when the project has exactly the same risk prole as the entire rm. Similarly, Brealey, Myers, and Allen (2005) explain that the weighted average formula works only for projects that are carbon copies of the rest of the rm. Such a gap between the normative formulation of the WACC method (the discount rate should be project-specic) and its implementation by practitioners (rms tend to use their rm-wide WACC for all projects) should lead to specic distortions in the investment policy of rms. The economic magnitude of the bias is potentially large. For example, suppose that a rm invests in a project that pays a dollar in perpetuity. If it takes a discount rate of 10%, the present value of the project is $10. By contrast, a hurdle rate of 8% would imply a present value of $12.5. Hence, underestimating the WACC by only 2 percentage points leads to overestimating the present value by 25%. This paper is an attempt to document and measure such distortions using eld data. First, we use business segment data to investigate if diversied rms rely on a rm wide WACC. To do so, we examine whether diversied companies are inclined to overinvest in 1

Electronic copy available at: http://ssrn.com/abstract=1764024

their high-beta divisions and underinvest in their low-beta divisions. The intuition is the following: A company using a single rm-wide WACC would tend to overestimate the NPV of a project whenever the project is riskier than the typical project of the company. If companies apply the NPV principle to allocate capital across dierent divisions2 , they must have a tendency to overestimate the NPV of projects that are riskier than the rms typical project and vice versa. This, in turn, should lead to overinvestment (resp. underinvestment) in divisions that have a beta above (resp. below) the rm-wide beta. Let us illustrate our empirical strategy with one of our data points: Anheuser-Busch Companies Inc. (ABC). The core business of ABC is brewing; it belongs to the Beer and Liquor industry (Fama-French (FF) industry code 4). In Fiscal Year 2006, this industry represented 81% of ABCs total sales. In this industry, we estimate the asset beta using the industrys stock returns and unlever the resulting equity beta by relying on the industrys capital structure: We obtain an asset beta of 0.12. Besides brewing, ABC operates a large number of theme parks which belong to a totally unrelated industry (Fun, FF code 39) which amounts to about 11% of the rms total sales. In this non-core industry, the estimated asset beta is 0.69, which is much higher than in the core business. If ABC was to use the discount rate of brewing to value investment projects in its entertainment business, it would underestimate the cost of capital by about (0.69 0.12) 7% = 4% (assuming an equity risk premium of 7%). Hence, the theme park division of ABC should invest relatively more than similar divisions. Our statistical tests rest on this logic.3 Using a large sample of divisions in diversied rms, we show in the rst part of the present paper that investment in non-core divisions is robustly positively related to the dierence between the cost of capital of the division and that of the most important
Survey evidence of CEOs and CFOs presented in Graham, Harvey, and Puri (2010) suggests that the NPV ranking is the predominant principle governing capital budgeting decisions. 3 In estimating the cost of equity capital, our empirical analysis relies on the Capital Asset Pricing Model. One issue is the poor empirical evidence in favor of the CAPM (see for instance Fama and French (1993)). All our results carry through with three- or four-factor models, yet the intention of our analysis is descriptive more than normative. Managers use the CAPM extensively for capital budgeting (see Graham and Harvey (2001)), and this generates predictions about the cross-section of investment policies.
2

division in the conglomerate (the core-division). We interpret these ndings as evidence that some rms do in fact discount investment projects from non-core divisions by relying on the core divisions cost of capital. We then discuss the cross-sectional determinants of this relationship and nd evidence consistent with models of bounded rationality: Whenever making a WACC mistake is costly (e.g. the non-core division is large, the CEO has sizable ownership, the within-conglomerate diversity of costs of capital is high), the measured behavior is less prevalent. In the second part of this paper, we document the present value loss induced by the fallacy of evaluating projects using a unique company-wide hurdle rate. To do this, we focus on diversifying acquisitions, a particular class of projects which are large, can be observed accurately, and whose value impact can be assessed through event study methodology. We look at the market reaction to the acquisition announcement of a bidder whose cost of capital is lower than that of the target. If this bidder takes its own WACC to value the target, it will overvalue it, and the announcement will be less of a good news to the bidders shareholders. We nd that such behavior leads to a relative loss of about 0.8% percent of the bidders market capitalization. On average, this corresponds to about 8% of the deal value, or $16m per deal. This nding is robust to the inclusion of dierent control variables. Our paper is related to several streams of research in corporate nance. First, it contributes to the literature concerned with the theory and practice of capital budgeting and mergers and acquisitions. Graham and Harvey (2001) provide survey evidence regarding rms capital budgeting, capital structure and cost of capital choices. Most relevant to our study, they show that rms tend to use a rm-wide risk premium instead of a project specic one when evaluating new investment projects. Relying entirely on observed rm level investment behavior, our study is the rst to test the real consequences of the nding in Graham and Harvey (2001) that few rms use project specic discount rates. More precisely, we provide evidence that the use of a single rm-wide discount rate (the WACC fallacy) does in fact have statistically and economically signicant eects 3

on capital allocation and rm value. Since we make the assumption that managers do rely on the NPV criterion, the present paper is also related to Graham, Harvey, and Puri (2010). This more recent contribution takes a forensic view on capital allocation and delegation of decision making in rms and provides strong survey evidence showing that the net present value rule is the dominant way for allocating capital across dierent divisions. Secondly, our paper contributes to the growing behavioral corporate nance literature. Baker, Ruback, and Wurgler (2007) propose a taxonomy organizing this literature around two sets of contributions: Irrational investors vs. irrational managers. The more developed irrational investors stream assumes that arbitrage is imperfect and that rational managers, in their corporate nance decisions, exploit market mispricing. Relevant to our study, Stein (1996) studies the issue of how capital budgeting should be performed when markets are irrational. Overall, however, our paper is more related to the less developed irrational managers literature. This approach assumes that, while markets are arbitrage free, managerial behavior can be inuenced by psychological biases. So far, this stream of research has mostly focused on how psychological traits such as optimism and overcondence can have distorting eects on managerial expectations about the future and investment decisions (see Malmendier and Tate (2005, 2008), Landier and Thesmar (2009) or Gervais, Heaton, and Odean (2011)). By contrast, far less attention has been paid to whether and how bounded rationality and resulting rules of thumbs can shape corporate decisions. To the best of our knowledge, the present paper is the rst to consider how a simplifying heuristic (using a single company wide discount rate) can have real eects on important corporate policies such as corporate investment and mergers and acquisitions. The reason why rms use a single discount rate might result from lack of sophistication. It is actually not obvious at rst sight why the rm-level cost of capital is not the relevant discount rate for all the projects of the rm. A company that benets from a low cost of capital might feel that nancing risky projects is an arbitrage opportunity. In fact, by changing the risk of the rms cash ows, these projects also 4

modify the expected rate of return that the market expects from the rm (ModiglianiMiller). We also nd several pieces of evidence coherent with the view that the WACC fallacy is related to managerial bounded rationality: The prevalence of this behavior seems to decrease over time, in line with the idea that CFOs are now more likely to have been exposed to modern capital budgeting. Also, the fallacy is less pronounced in larger non-core divisions, in more diverse companies, and when the CEO owns a larger stake in the company. Such evidence is in line with the view that full rationality is costly and that agents become more rational when the gains of doing so increase (see e.g. Gabaix (2010)). Finally, our paper is also related to the extensive literature on the functioning of internal capital markets. Lamont (1997) and Shin and Stulz (1998) provide evidence that internal capital markets exist and reallocate cash ows across dierent divisions. A series of papers, using dierent settings, shows that investment within conglomerates is less sensitive to growth opportunities, which suggests misallocation of funds (Rajan, Servaes, and Zingales (2000), Gertner, Powers, and Scharfstein (2002), Ahn and Denis (2004) or Ozbas and Scharfstein (2010)). Another branch of this literature relates this behavior to value destruction: Berger and Ofek (1995) or Lamont and Polk (2002)). A contribution of our paper is to show that division-level industry betas (and not simply divisions Tobins q or cash-ows) are an important factor in understanding investment distortions within conglomerates. We relate this new type of capital misallocation to the use of a single discount rate, a bias that we call the WACC fallacy. This bias might be related to the politicking argument of Rajan, Servaes, and Zingales (2000) and Scharfstein and Stein (2000): Firms might avoid complex, division-specic, discounting rules as they would facilitate inecient politicking by division managers. The use of a single rm-wide discount rate could be an optimal response to agency concerns in the capital budgeting process. The rest of the paper is organized as follows: Section 1 describes the data. Section II provides evidence on how division level investment in conglomerates is related to rm 5

wide measures of the cost of capital. Section III presents the evidence on diversifying mergers and acquisitions. Finally, section IV concludes.

I
I.1

The data
Sample and basic variables

Our rst battery of tests, which focuses on investment in diversied conglomerates, requires a dataset of conglomerate divisions. To build it, we start with data from the Compustat Segment les, covering the period 1987-2007. From these les, we retrieve segment level information on annual capital expenditures, sales and total assets, as well as a four-digit SIC code for the segment, which we match with the relevant two-digit Fama-French industry (FF48). Within each rm, we then aggregate capx, sales and assets data by FF48 industry. We call divisions the resulting rm-industry-year observations. We then merge these data with rm-level data from Compustat North America, which provide us with rm level accounting information. Whenever the sum of division sales exceeds or falls short of total rm sales (item SALE ) by a margin of 5 % or more, we remove all related rm-division-year observations from the sample. This is done in order to ensure consistency between the Compustat Segments and the Compustat Industrial Annual databases and to reduce the potential noise induced by a rms incorrect reporting of segment accounts. Finally, we merge the resulting division level dataset with rm-level information about CEO ownership from Compustat Execucomp. Such information is available only from 1992 through 2007 and for a subset of rms. Using this merged dataset, we dene a conglomerate rm as a rm with operations in more than one FF48 industry, whereas standalone rms have all their activities concentrated in a single FF48 industry. In Table I, we report summary statistics for all rm-level variables separately for standalone (top panel) and conglomerate rms (bottom panel). All variable names are self explanatory but their exact denition is given in the Appendix. Out of approximately 135,000 rm-year observations, about 120,000 6

observations correspond to standalones (i.e. rms operating in a single FF48 industry) and about 15,000 observations (or approximately 750 rms a year) operate in more than one industry (on average, 2.56 industries). On average, conglomerates are quite focused: About 73% of total sales are realized in the largest division. Unsurprisingly, standalones grow faster, are smaller and younger than conglomerates; conglomerates are more cash ow rich and more levered. [Table I about here.] We also compute the Tobins q using the rms market and book values of assets (see Appendix for details). In line with the existing literature, we see from Table I that conglomerates tend to have lower market-to-book ratios (1.49 versus 1.88), which may reect either slower growth, or the presence of a conglomerate discount. For each conglomerate rm, we then identify the division with the largest sales and label it core-division. Conversely, divisions with sales lower than those of the core-division are referred to as non-core divisions. In Table II we report division-level summary statistics for non-core divisions only. Since there are about 15,000 observations corresponding to conglomerates, and since conglomerates have on average 1.56 non-core divisions (2.56-1), there are about 23,000 observations corresponding to non-core divisions. We dene the Tobins q of a division as the median market-to-book of standalones which belong to the same FF48 industry as the division. The denition of the other variables is straightforward and detailed in the Appendix. On average, Table II shows that non-core divisions are slightly smaller than standalones (log of book assets equal to 4, against 4.3 for standalones). [Table II about here.] We also calculate the Tobins q of each division. Since the division has no market price, we compute the median market-to-book ratio of standalones operating in the same FF48 industry as the division. This has been shown to be a reasonable approximation: 7

Montgomery and Wernerfelt (1988) nd that industry-level Tobins q is a good predictor of rm level Tobins q. For each non-core division, we also dene as QCORE,t the Tobins q of the core division of the conglomerate it belongs to. As we report in Table II, non-core and core divisions have on average very similar Tobins qs.

I.2

Mergers and acquisition data

Our second series of tests relies on a sample of diversifying acquisitions. The sample is constructed by downloading all completed transactions between 1988 and 2007 from the SDC Platinum Mergers and Acquisitions database in which both target and bidder are US companies. The bidders and targets core activities are identied through the SDC variables Acquiror Primary SIC Code and Target Primary SIC Code, which we match to their corresponding FF48 industry categories. We dene a diversifying transaction as a deal in which a bidder gains control of a target which belongs to a FF48 industry dierent from the bidders core activity. We restrict the sample to these transactions only. We keep only completed mergers and acquisitions in which the bidder has gained control of at least 50 % of the common shares of the target. We include transactions that include both private and public targets. We drop all transaction announcements in which the value of the target represents less than 1 percent of the bidders equity market value (calculated at the end of the scal year prior to the year of the acquisition announcement) and also drop all transactions with a disclosed deal value lower than 1 million US-$. Daily stock returns of the bidder are downloaded from CRSP for the eleven day event window surrounding the announcement date of the deal. Finally, we obtain balance sheet data for all bidders from the Compustat North America database. [Tables III and IV about here.] In total, we identify 6,115 of these diversifying transactions between 1988 and 2007 for which the sample selection criteria are satised. Summary statistics of bidder, target and deal characteristics are summarized in Table III. The typical transaction involves a 8

small private or subsidiary target. The average value of the target is slightly less than $200m; 54% of the transactions involve a non-listed target, and only 3% correspond to tender oers. In panel B, we also report the average Tobins q of the bidder and the target, calculated as the median market-to-book of standalones belonging to the same industries as the bidder and target respectively. The dierence is, on average across transactions, not statistically signicant. Table IV reports the number of acquisitions per year in our sample: As expected, there are large year-to-year uctuations of the number of transactions and average valuation, which broadly correspond to the last two acquisitions waves.

I.3

Calculating the cost of capital

For both series of tests, we need to construct an annual industry-level measure of the cost of capital, which we will merge with both relevant datasets (division-level and transactionlevel). We do so by regressing monthly returns of value-weighted portfolios comprised of companies belonging to the same FF48 industry on the CRSP Value Weighted Index for moving-windows of 60 months. We then unlever the estimated industry-level equity beta using the following formula:

A i,t =

Ei,t E i,t , Ei,t + Di,t

(1)

where Ei,t is the total market value of equity within the FF48 industry i in year t, and Di,t is the total book value of debt (see Appendix for denitions of debt and equity values).
E A i,t is the estimated equity beta of industry i in year t. i,t is the beta of assets invested in

industry i in year t. We report average asset and equity betas per-industry in Table A.I. We then merge the information on industry cost of capital with the division-level
A data. For a division or a standalone, i,t is the asset beta of the industry to which the A standalone or the division belongs. We denote as CORE,t the industry-level asset beta of

A the core-division of a conglomerate. Correspondingly, DIV,t refers to the asset beta of a

non-core division. For a conglomerate, we also calculate the average of all division asset
A betas, weighted by total (book) assets of divisions, and call it AV ERAGE,t . We report

estimates of rm and division-level asset betas in Tables I and II. With an average asset beta of 0.55, conglomerates appear slightly less risky than standalones (0.63). Non-core divisions have on average the same asset beta (0.56) as their related core divisions (0.55), so that the beta spread, i.e. the dierence between the beta of a non-core division and the beta of its core is zero, on average. The spread varies, however, a lot: From -0.21 at the 25th percentile to +0.22 at the 75th percentile. Next, we merge the information on asset betas with the diversifying acquisitions data. The relevant information is reported in Table III, Panel B. In contrast to Tobins q, which tends to be similar between bidders and targets, asset betas tend to be significantly smaller for bidders (0.59) than for targets (0.64), with the spread being highest for subsidiary targets.

I.4

Calculating the extent of vertical relatedness between industries

In order to construct a measure of the vertical relatedness between each pair of FF48 industries, we download the Benchmark Input-Output Accounts for the years 1987, 1992, 1997 and 2002 from the Bureau of Economic Analysis4 . We rely on the Use Table of these accounts, which corresponds to an Input-Output (I-O) matrix providing information on the value of commodity ows between each pair of about 500 dierent I-O industries. We match the I-O industries to their corresponding FF48 industry and aggregate the commodity ows by FF48 industry. This aggregation allows to calculate the total dollar value of inputs used by any FF48 industry. The aggregated table also shows the value of commodities used by any FF48 industry i, which is supplied to it by FF48 industry j. For each industry i, we calculate the dependence on inputs from industry j as the ratio
4

see http://http://www.bea.gov/industry/io_benchmark.htm

10

between the value of inputs provided by industry j to industry i and the total inputs used by industry i. We denote this measure by vij . Following Fan and Lang (2000), we dene the vertical relatedness of two FF48 industries i and j as Vij = 1/2(vij + vji ). Vi,j measures the extent to which the non-core division and the core division exchange inputs. In Table II we show summary statistics of VDIV,t . The table shows that the average exchange of inputs between non-core and their corresponding core division is about 4 % in our sample.

II
II.1

Investment distortions within diversied rms

Investment misallocation

Our test rests on the fact that, in conglomerates prone to the WACC Fallacy, nonA A core division investment should be increasing with the beta spread (DIV,t CORE,t ).

Assume a non-core division has a higher beta than the core. Controlling for investment opportunities, if this division uses the cores asset beta to discount future cash ows, it will overestimate any projects NPV, and will overinvest. Conversely, non-core divisions with low betas relative to the core should underinvest. Hence, investment increases with the spread between the beta of the non-core division and the beta of the core. If, however, the rm uses the right cost of capital in each division, then investment should be insensitive to the cores cost of capital. The beta spread should have no impact. One variation of this idea is that, instead of using the cost of capital of its core division, the rm uses a weighted average cost of capital to value non-core division projects.
A In this case, the above prediction remains true as long as CORE,t is a good proxy for the A asset-weighted beta of the rm, AV ERAGE,t . This is a priori reasonable since in conglom-

erates the core division accounts on average for 73% of total sales. But precisely because of this, both stories are dicult to distinguish empirically, although we provide evidence in our robustness checks (see below) that the beta of the core division seems to be more
A relevant. All in all, we choose to use CORE,t in our main specication because we believe

11

A that the results with CORE,t are more convincing as they avoid the multi-collinearity A A concerns that arise when putting DIV,t and AV ERAGE,t together in a regression.

A. Graphical Evidence We rst provide graphical evidence in Figure 1 that non-core division investment is correlated with the beta spread. To do this, we sort observations (non-core division-year) into 10 deciles of beta spread. For each decile, we compute average non-core division investment. In total, we use the three measures of non-core division investment that have been the most commonly used in the conglomerate literature. First, following Shin and Stulz (1998), we rely on raw investment, which is non-core division capital expenditures divided by lagged non-core division assets. Secondly, we use industry adjusted investment which we calculate as a non-core divisions raw investment net of the median raw investment of all standalone rms operating in the non-core divisions FF48 industry in that year (see Lamont (1997)). Finally, following Rajan, Servaes, and Zingales (2000), we also use industry-rm adjusted investment, which subtracts from each non-core divisions industry adjusted investment the rms weighted average industry adjusted investment. The weights used in calculating the average are division to total assets. All three measures are formally dened in the Appendix. Figure 1 shows a somewhat monotonic relationship between all three investment measures and the spread: It seems that non-core divisions with relatively high beta spread (high beta compared to core-division) tend to invest more, whereas non-core divisions with a low spread invest less. [Figure 1 about here] B. Baseline Results We then report multivariate regression results in Table V, in order to control more extensively for observable determinants of investment. The dependent variable in these regressions is raw investment5 . Standard errors are clustered at the rm-level. In unreIn the Appendix (Tables A.IIA.V), we replicate the main results for industry adjusted and industryrm adjusted investment. All coecient estimates are quite similar in terms of economic and statistical signicance.
5

12

ported regressions we also cluster standard errors at the non-core division level, which leaves our results unaected. Column (1) establishes the basic fact by showing that noncore division raw investment depends positively on the spread between the non-core and the core divisions industry betas. The larger the spread, the higher raw investment in the respective non-core division. This is precisely what would be expected if companies discount risky projects using too low discount rates. [Table V about here.] In column (2), we add the main determinants of corporate investment, i.e. industrylevel Tobins qs of the core and the non-core divisions, as well as cash ows (at the rm level) and other potential determinants of division level investment, i.e. Division Size, Firm Size, Firm Age and a measure of Firm Focus. We control for the investment opportunities of both the core and the non-core divisions in order to address the concern that asset betas may correlate with variations in investment opportunities that are not captured by Tobins qs. After including these controls, the coecient estimate for the spread decreases slightly but remains highly statistically signicant. In column (4) we
A A replace the spread by its two separate components CORE,t and DIV,t . The results show A A a negative sign for the coecient estimate for CORE,t and a positive sign for DIV,t . A This suggests that whenever the company has a low risk core activity (low CORE,t ), and

therefore a low hurdle rate, it is inclined to invest more strongly in non-core divisions with
A a higher asset risk. The fact that CORE,t is signicant provides evidence that diversied

companies look at divisions belonging to industries dierent from their core activity with the eyes of their core industrys characteristics. In terms of magnitude, the investment distortion we document is quite important.
A A Assume DIV,t CORE,t = 0.35, which is about one sample standard deviation. This

means the gap in discount rates between the division and its core is approximately 2.5% (assuming a 7% equity risk premium). Given our estimates, we would expect the non-core divisions investment rate to be 0.5 (0.0155*0.35) percentage points higher. This is a nonnegligible eect, equivalent to about 10% of the average raw investment rate in our sample. 13

A A Interestingly, the absolute values of the coecients for DIV,t and CORE,t are of similar

magnitude. This nding is consistent with the WACC fallacy interpretation, since both variables play exact opposite roles in the discount rates misvaluation. Coherent with this idea, it turns out that we cannot reject the null hypothesis that the sum of both coecients is equal to zero. We expect the documented investment distortion to be larger if the projects sales growth is higher. To see this, assume, in the spirit of Gordon and Shapiro (1956), that an investment project in a non-core division pays a cash ow C, with constant growth rate g smaller than the WACC. Then, the present value of the project is given by
C . W ACCg

From this formula, it is obvious that the valuation mistake made by not choosing the right WACC is bigger when g is larger, in a convex fashion.6 Hence, we expect the impact of beta spread on investment to be bigger when the division belongs to a fast growing industry. Column (5) of Table V reports the outcome of such a consistency check. We code an indicator variable for each tercile of lagged industry sales growth (Low, Medium and High Ind Sales Growth) and interact these dummies with the spread in asset betas. The results show that while investment of medium growth non-core divisions are not signicantly more sensitive to the spread than low growth divisions (M edium Ind Sales Growth
A A (DIV,t CORE,t ) is not signicant), high growth divisions turn out to be signicantly

more sensitive to the spread than low growth divisions (t-stat 2.00). The dierence is large too: While for divisions in the bottom tercile of industry growth, the coecient is equal to 0.012, it is equal to 0.020 in the top tercile. The estimated eect is therefore twice as large. This underlines the idea that the investment sensitivity is in fact convex in lagged sales growth. We also test whether the prevalence of the WACC fallacy depends on whether a rm is using internal vs. external nancing.7 . Indeed, it might be that nancial markets
To see this formally, assume that the conglomerate chooses W ACC instead of W ACC, where is C small. Then, the estimated present value of the project is inated by (W ACCg)2 , which is increasing and convex in g. 7 We thank Malcolm Baker and Je Wurgler for suggesting this test.
6

14

(rather than rms managers) are subject to the WACC fallacy and fail to properly adjust the rms cost of capital as a function of the real risks of its investments. In this case, rms might try to take advantage of the markets misperception, by raising abnormally cheap capital to invest in high beta projects. If this mechanism was the driving force behind our results, one should observe higher investment sensitivity to beta spread when a rm is relying on external rather than internal capital. In order to test this, we follow Baker, Stein, and Wurgler (2003) and calculate the rms external nance activity as the sum of contemporaneous debt and equity issues normalized by lagged assets. In column (6) we then interact the spread sensitivity of non-core division investment with tercile dummies based on the measure of external nancing activity. The results show that the WACC fallacy does not depend on the use of new external nance. This conclusion holds irrespective of whether new equity, new debt or the sum of both is used as a measure of external nance activity. In unreported regressions we also nd no link between the magnitude of the beta spread and nancial constraints measured as in Kaplan and Zingales (1997).

II.2

Robustness Checks

A. Average WACC versus Core WACC We report robustness checks in Table VI. First, in column (1), we seek to account for the fact that, as noticed before, instead of using the cost of capital of the core division, conglomerates may be using a rm-wide average asset beta. In column (1), we thus use
A as regressor the weighted average division asset beta (AV ERAGE,t ), where the weights

correspond to the ratio of division to total rm assets. As expected, the coecient estimate is signicantly negative in this specication too. To distinguish between the two stories, we run in column (2) a regression with both the average rm beta and the beta of the core division. In this specication, the average beta of the rm is no longer signicant. The documented investment distortion thus seems to be driven by the use of the core-divisions discount rate rather than the use of an average rm-wide WACC, 15

even though both hypotheses may ultimately be hard to distinguish rigorously. [Table VI about here.] B. Unobservable Dierences in Investment Opportunities In column (3), we seek to directly address the concern that the beta spread could simply be capturing dierences in investment opportunities between the division and the core, which are not fully controlled for by the division-level Tobins qs. We do so by including the gap between the division-levels and the rms overall Tobins q. It leaves the estimated eect of beta spread unaected. In the same spirit, the beta spread may simply capture a measure of diversity of a rms investment opportunities, which has been shown in earlier work to distort investment behavior (Rajan, Servaes, and Zingales (2000)). In column (4), we therefore include as additional control the within-rm standard deviation of industry-level Tobins qs normalized by rm wide q. Intuitively, and consistent with existing evidence, we nd non-core division investment to decrease with the rms diversity in investment opportunities. Yet, the inclusion of the diversity control and the gap between the division and the rms Tobins q does not aect our conclusions. In unreported regressions, we have also sought to replace industry-level Tobins q by an alternative division level measure of investment opportunities, namely lagged division sales growth. Again, our results remain robust in this alternative specication.

C. Controlling for Industry Shocks Another concern with our results is that we may be capturing the impact of upstream integration. Assume, for instance, that a rm produces toys and owns trucks to transport them. It therefore has two activities: Transportation (non-core) and toy-production (core). If the cost of capital in the toy industry goes down, the rm may expand its production capacity, for instance to cater to investor sentiment (for this to hold, note that the beta must be capturing investment determinants not already controlled for in our regressions). To ship the additional production, it will also invest in new trucks. In 16

this setting, investment in a non-core division (trucks) responds to changes in the WACC of the core division (toys), for reasons that have little to do with the WACC fallacy. In column (5) of Table VI, we control for rm-wide investment policy by including the core divisions investment rate. This control variable leaves the coecient estimates for both
A A CORE,t and DIV,t unchanged.

[Table VII about here.] As an alternative way of addressing this concern, we construct a measure of vertical integration between each non-core division and its core, and interact this measures with the beta spread to see if it aects the relation estimated in Table V. For each FF48 industry, we calculate (1) the fraction of this industrys output that is sold to each (downstream) FF48 industry and (2) the fraction of this industrys input that comes from each (upstream) industry. For each industry pair (i, j), Vij is equal to the mean of these two numbers, and therefore measures the extent to which these two industries are integrated both upstream and downstream. For each non-core division, we split the measure of relatedness with the core into three tercile dummies, and interact them with the beta spread. We report the estimates using various specications in Table VII. In column (1), we report the baseline result without interaction. In column (2), we add the interaction terms between beta spread and the dummies. It appears that the measured impact of beta spread on investment does not depend on the extent to which the non-core division is related with its core. In columns (3) and (4), we interact the two measures of growth opportunities with the relatedness dummies. The overall diagnosis remains: Our estimated eect of beta spread on investment is not driven by non-core divisions that are vertically related to the core activity. In all of the previous specications, we do not control explicitly for industry specic shocks to investment. It might be the case that non-core divisions belonging to certain industries invest more for reasons that are specic to their industries and not due to biases in the capital budgeting process. In order to address this concern, we rst include 17

industry-year xed eects in column (6) of Table VI, which leaves our results unaected. Following Lamont (1997), we also rerun our investment regressions using industry adjusted investment as the dependent variable. Industry adjusted investment is calculated as the dierence between raw investment of the non-core division and the median raw investment for standalone rms operating in the same FF48 industry in that specic year. Hence, industry adjusted investment removes shocks to investment that are common to the entire industry and unaected by the capital budgeting distortion we document. The results are presented in Tables A.II and A.III and show coecient estimates for the spread variable that remain strongly statistically signicant. The observed decrease in the magnitude of the estimate is as expected since mean industry adjusted investment is substantially lower than mean raw investment. While average investment is 0.06 for non-core divisions, industry adjusted investment averages at about 0.02 (see Table II). Hence, for a one standard deviation increase in the beta spread, division investment in excess of the average standalone raw investment changes by about 0.00245 (0.0070*0.35) percentage points, translating into a distortion of about 12% of average industry adjusted investment. One last concern might be that, compared to standalone rms in the same industry, diversied rms tend to overinvest in all of their divisions. If this was the case, higher industry adjusted investment would incorrectly be treated as being division specic. In order to correct for the potential rm-wide overinvestment of conglomerates, Rajan, Servaes, and Zingales (2000) propose to further subtract from the divisions industry adjusted investment rate the industry adjusted investment rate averaged across all divisions of the rm. They refer to this measure as the industry-rm adjusted investment rate, which they argue to be the best proxy for the within rm investment bias in direction of a particular division, net of overall excess investment. We again rerun the main investment specications of Table V and Table VI using this second alternative measure. The results are reported in Table A.IV and Table A.V. The coecient estimates for both the spread itself and the components of it remain strongly statistically and economically signicant. 18

D. Excluding Finance and Insurance Our sample includes all industries, even the nancial sector. We have good reasons to believe that even the banking sector can be subject to the WACC Fallacy, which is why we did not remove the nancial sector from our main specications. There might, however, be concerns that investment and investment opportunities are not well captured in Compustat. To check that our results do not depend on the inclusion of the nancial sector, we have replicated the results of Table V, excluding all non-core divisions-year observations that belong to a core division from the Banking and Insurance (FF48 industries 44 and 45). We report these new estimates in Table A.VI. They are, if anything, stronger, consistent with the fact that accounting data are noisier in these industries.

II.3

Evidence of bounded rationality

Taking the wrong cost of capital decreases present value, because it biases the rms investment decisions away from value maximization. On the other hand, computing the adequate cost of capital, making it vary across projects, making sure that division managers do not manipulate it to defend low NPV projects, is also costly. Sometimes, it may therefore be optimal to keep a single WACC for the entire organization. But if the costs of doing so are too high, we expect rms to use dierent WACCs. This bounded rationality view makes predictions as to which rms are more likely to be subject to the WACC Fallacy: The relationship uncovered in Table V should disappear when the benets of taking the right WACC are large. [Table VIII about here.] In this Section, we test the bounded rationality view. We do this by interacting beta spread with measures of the net benet of adopting dierentiated WACCs. Bounded rationality predicts that investment policy should be less sensitive to beta spread when the net benet of using division specic discount rates is high. We report the results of this 19

investigation in Table VIII using four dierent measures of net benet to take the right WACC. In column (1), we rst hypothesize that nancial knowledge of corporate decision makers in charge of making capital budgeting decisions has improved over time. Higher nancial sophistication of managers due to MBA style education could have improved the quality of capital allocation decisions within conglomerate rms, making the cognitive cost of taking the right WACC decrease over time. We therefore expect the sensitivity of investment to beta spread to decrease over time, and test this by interacting four period dummies (1987-1991, 1992-1996, 1997-2001, 2002-2007) with beta spread. The interaction terms indicate that the investment distortion has been strongest between 1987 and 1996. Note that an alternative explanation to that of increased sophistication could be the decrease of the equity premium, multiplicatively reducing the impact of the spread variable.8 In columns (2)(4), we use cross-sectional proxies of net benet of nancial sophistication. In column (2), we use the Relative Importance of a non-core division. The idea is that, when the non-core division is large with respect to the core, valuation mistakes have larger consequences; investment in these divisions are therefore less likely to be suering from the WACC Fallacy. We calculate this measure by scaling non-core division sales by the sales of the core division. Values close to one indicate that the non-core division in question is almost as important as the core-division within the conglomerate. By contrast, values close to zero indicate that the non-core division is negligible vis ` vis the a core division. We then split this measure into terciles (Low, Medium and High Relative Importance) and interact the dummies with the beta spread. We report the regression results in column (2): Investment in more important divisions (High Relative Importance) is less sensitive to beta spread, suggesting that investment in relatively large divisions is less prone to the WACC Fallacy. In column (3), the measure of net benet is diversity of costs of capital, dened at the rm level as the within-rm standard deviation of core
The change in business segment reporting standards initiated by the FASB issuance of SFAS 131 in June 1997 does not seem to have an impact on our results, since our coecients remain statistically signicant also in sub-periods following the change in regulation.
8

20

A and non-core division asset betas (SD(i,t )). Again, the intuition is that taking a single

WACC to evaluate investment projects leads to larger mistakes if costs of capital are very dierent within the organization. As before, we split our measure of diversity into terciles. Column (3) shows that division investment in rms with highly diverse costs of
A capital (High (SD(i,t ))) is signicantly less sensitive to the beta spread. These rms

therefore seem to nd it optimal to use dierent WACCs. Last, we explore in column (4) the role of CEO ownership: Here the intuition is that CEOs with more skin in the game will nd it more protable to avoid value destroying investment decisions and will opt for nancial sophistication in the organization she is running. This is consistent with Baker, Ruback, and Wurgler (2007), who note that in order for irrational managers to have an impact on corporate policies, corporate governance should be somewhat limited. Because of the limited availability of this variable, we only split CEO ownership into two dummies (above and below 1%). We show, in column (4), that the impact of beta spread on division investment is substantially lower for rms whose CEO has a relatively larger stake. Such evidence is also in line with evidence in Ozbas and Scharfstein (2010), who nd that inecient investment in conglomerates decreases with management equity ownership.

III

Eciency eects

In this section, we examine the eciency costs of using the wrong cost of capital. Doing this is dicult in the context of conglomerate investment, since we would need to compare rm values in the cross-section. A more powerful test, which we implement here, consists of looking at market reaction to large investment projects, whose cost of capital we can measure. To do this, we study diversifying mergers and acquisitions. For our purpose, these investment projects have four advantages. First, they are easy to observe: Large standard datasets report the exact date, the size, the industry and the amount invested in these projects. Second, the cost of capital of the acquisition can be computed (using

21

the targets industry cost of capital) and is, for diversifying acquisitions, dierent from the cost of capital of the acquirer, which may give rise to behavior consistent with the WACC Fallacy. Third, we have a reasonable estimate of their NPV, namely, the stock price reaction of the acquirer upon acquisition announcement. Under the assumption that markets are not systematically biased in their reaction, the announcement returns provides an estimate of the NPV created by the project. Last, these projects tend to be large enough so that their impact on the market value of the acquirer is detectable in a credible way.

III.1

Acquisitions subject to the WACC Fallacy

We rst check that acquisitions are subject to the same WACC fallacy as non-core division investment. From Table III, Panel B, we see that, across all deals in our sample, the beta spread is on average about -0.05: Targets have, on average, a higher asset beta than bidders. The dierence is statistically signicant at 1%. This is consistent with bidders using their own discount rate to value targets: This leads to overvaluation from the bidders side, which makes the deal more likely to succeed. From Table III, Panel B, we also observe that this phenomenon is more pronounced when the target is private than when it is public. This is consistent with the bounded rationality hypothesis defended in the previous section: When the target is publicly listed, information about its true value is cheaper to come by (the market quotes a price). As a result, it is easier to make the right estimate.

III.2

Value loss in diversifying asset acquisitions

The fallacy of using inadequate discount rates is expected to have implications for bidder abnormal returns around the announcement of diversifying acquisitions. Whenever bidders use too high a discount rate, they are more likely to undervalue their targets and therefore more likely to propose an oer price for the targets assets below fair value.

22

Shareholders of the bidder should thus perceive a bid from a high beta bidder for a low beta target as relatively better news, since it reduces the likelihood of overpaying for the target. Conversely, a low industry-level beta bidder should have a tendency to overvalue high beta assets. Hence, the bid announcement from a low industry-level asset beta company for a high beta industry asset should be perceived as relatively bad news by stock markets because the bidder is more likely paying too much. We thus expect bidder abnormal returns around transactions to be signicantly higher whenever the bidders WACC is higher than that of the target. [Figure 2 about here] In Figure 2, we plot the mean cumulative abnormal returns of bidders around the announcement of diversifying asset acquisitions conditional on whether the bidders WACC is lower or higher than that of the target. Abnormal returns are calculated as market adjusted returns on the respective event day. We use the CRSP Value Weighted Index as the market benchmark. For both categories of deals, we rst see that announcement returns are positive: This is consistent with Bradley and Sundaram (2006) who nd that announcement returns for acquirers of private rms (the vast majority of the deals in our sample) are positive and statistically signicant (see also Betton, Eckbo, and Thorburn (2008)). More importantly to us, evidence from Figure 2 suggests that the market welcomes bids involving low beta bidders and high beta targets less favorably than bids with high beta bidders and low beta targets. This conrms our hypothesis that low beta bidder tend to overbid for high beta targets. In order to formally test whether this dierence is statistically signicant, we regress both the abnormal return on the announcement day (AR(0)), as well as the seven day cumulative abnormal return surrounding the announcement (CAR(3,3)) on a dummy variable indicating whether the bidders WACC exceeds that of the target. The results from regressions in which the abnormal return serves as the dependent variable are reported in Table IX. Table X reports the regression results for the cumulative abnormal 23

return (CAR(-3,3)). In all specications, we also include year dummies to capture the potential impact of merger waves on announcement returns and control for the size of the transaction, which we calculate as the natural log of the deal value as disclosed by SDC. In order to control for deals that are announced on the same day, standard errors are clustered by announcement dates. In unreported regressions we cluster standard errors by week, month and year, which does not aect our results. [Table IX and Table X about here.] Column (1) of Table IX establishes the main result by showing that bidder abnormal returns on the announcement day of transactions involving low beta bidders and high beta targets are signicantly lower than transactions involving high industry-level beta bidders and low industry-level beta targets. The coecient estimate for the dummy variable is equal to 0.00444 (t-stat of 2.77). This suggests that when the bidder has a lower beta than the target, 0.4% of acquirer value is lost compared to other bids of similar size (target size is the only control in column (1)). Given that bidders have equity value of about $2bn on average, this estimate translates into an estimated excess payment of about $8m, or 4% of the average target value. To control for observable dierences in transactions, that could be correlated with our main variable, we then include known determinants of acquirers announcement returns. In column (2), we seek to control for the fact that some overvalued acquirers may seek to create value by purchasing undervalued targets. The announcement of such a transaction may be interpreted by the market as a signal of overvaluation which would reduce the acquirer stock price. This control is not statistically signicant and does not aect our estimate. In column (3), we control for the fact that the acquisition of publicly listed targets tends to lead to negative announcement returns: Consistently with this, the Target Public dummy comes out negative and signicant, but our estimate remains unchanged. The relative size of the target (inserted in column (5)) is also signicant and positive, but has no impact on our estimate. All in all, some observable controls are statistically signicant, but our estimate resists well. 24

Findings of Table IX are conrmed by results reported in table X, which looks at cumulative returns 7 days around the announcement date (CAR(3,3)). These estimates are larger: The coecient estimate for the dummy variable capturing dierences in the cost of capital between the target and the bidder is now equal, across specications to about 0.8% of the acquirers value. This is in line with the idea that announcements take several days to be impounded into prices. For the average acquirer, whose market capitalization is about $2bn, this suggests a value loss of about $16m, or 8% of the value of the average target (whose average value is approximately $200m).

IV

Conclusion

Survey evidence suggests that many rms use a rm-wide discount rate to evaluate projects (Graham and Harvey (2001)). The prevalence of this WACC fallacy implies that rms tend to bias investment upward for divisions that have a higher industry beta than the rms core division. This paper provides a direct test of this prediction using segment-level accounting data. We nd a robust positive relationship between divisionlevel investment and the spread between its industry beta and the beta of the rms core division. Using unrelated data on mergers and acquisitions, we also nd that the acquirers stock-price reaction to the announcement of an acquisition is lower when the target has a higher beta than the acquirer. The prevalence of the WACC fallacy among corporations seems consistent with managerial bounded rationality. It is actually not so simple to explain to a non-nance executive why it is logically awed for a rm to discount a risky project using its own cost of capital. The costs associated with using multiple discount rates might, however, not be purely cognitive or computational: They might also be organizational, as the use of multiple discount rates might increase the scope for politicking and gaming of the capital budgeting process in a hierarchy, in the spirit of Rajan, Servaes, and Zingales (2000) and Scharfstein and Stein (2000).

25

References
Ahn, Seoungpil, and David J. Denis, 2004, Internal capital markets and investment policy: evidence from corporate spinos, Journal of Financial Economics 71, 489516. Baker, Malcolm, Richard S. Ruback, and Jerey Wurgler, 2007, Behavioral Corporate Finance, in B. Espen Eckbo, ed.: Handbook of Corporate Finance, Volume 1: Empirical Corporate Finance (Elsevier/North Holland). Baker, Malcolm, Jeremey C. Stein, and Jerey Wurgler, 2003, When Does the Market Matter? Stock Prices and the Investment of Equity-Dependent Firms, Quarterly Journal of Economics 118, 9691005. Berger, Philip G., and Eli Ofek, 1995, Diversications eect on rm value, Journal of Financial Economics 37, 3965. Betton, Sandra, B. Espen Eckbo, and Karin S. Thorburn, 2008, Corporate Takeovers, in B. Espen Eckbo, ed.: Handbook of Corporate Finance, Volume 2: Empirical Corporate Finance (Elsevier/North Holland). Bierman, Harold J., 1993, Capital Budgeting in 1992: A Survey, Financial Management 22, 24. Bradley, Michael, and Anant Sundaram, 2006, Acquisition and performance: A reassessment of the evidence, Working Paper, Fuqua School of Business, Duke University. Brealey, Richard A., Stewart C. Myers, and Franklin Allen, 2005, Principles of Corporate Finance (McGraw-Hill: New-York, NY). Carhart, Mark M., 1997, On Persistence in Mutual Fund Performance, Journal of Finance 52, 5782. Fama, Eugene F., and Kenneth R. French, 1993, Common risk factors in the returns on bonds and stocks, Journal of Financial Economics 33, 353. , 1997, Industry costs of equity, Journal of Financial Economics 43, 153193. Fan, Joseph P.H., and Larry H.P. Lang, 2000, The Measurement of Relatedness: An Application to Corporate Diversication, Journal of Business 73, 629660. Gabaix, Xavier, 2010, A Sparsity-Based Model of Bounded Rationality, Working Paper, Stern School of Business, New York University. Gertner, Robert, Eric Powers, and David Scharfstein, 2002, Learning about Internal Capital Markets from Corporate Spin-os, The Journal of Finance 57, 24792506. Gervais, Simon, J.B. Heaton, and Terrance Odean, 2011, Overcondence, Compensation Contracts,and Capital Budgeting, forthcoming in Journal of Finance. Gordon, Myron J., and Eli Shapiro, 1956, Capital Equipment Analysis: The Required Rate of Prot, Management Science 3, 102110. 26

Graham, John R., and Campbell R. Harvey, 2001, The theory and practice of corporate nance: evidence from the eld, Journal of Financial Economics 60, 187243. , and Manju Puri, 2010, Capital Allocation and Delegation of Decision-Making Authority within Firms, Working Paper, Fuqua School of Business, Duke University. Grinblatt, Mark, and Sheridan Titman, 2002, Financial Markets and Corporate Strategy (McGraw-Hill: New-York, NY). Kaplan, Steven N., and Luigi Zingales, 1997, Do Investment-Cash Flow Sensitivities Provide Useful Measures of Financing Constraints?, Quarterly Journal of Economics 112, 169215. Lamont, Owen A., 1997, Cash Flow and Investment: Evidence from Internal Capital Markets, Journal of Finance 52, 83109. , and Christopher Polk, 2002, Does diversication destroy value? Evidence from the industry shocks, Journal of Financial Economics 63, 5177. Landier, Augustin, and David Thesmar, 2009, Financial Contracting with Optimistic Entrepreneurs, Review of Financial Studies 22, 117150. Malmendier, Ulrike, and Georey Tate, 2005, CEO Overcondence and Corporate Investment, The Journal of Finance 60, 26612700. , 2008, Who makes acquisitions? CEO overcondence and the markets reaction, Journal of Financial Economics 89, 2043. Modigliani, Franco, and Merton H. Miller, 1958, The Cost of Capital, Corporation Finance and the Theory of Investment, The American Economic Review 48, 261297. Montgomery, Cynthia A., and Birger Wernerfelt, 1988, Diversication, Ricardian rents, and Tobins q, The RAND Journal of Economics 19, 623632. Morck, Randall, Andrei Shleifer, and Robert W. Vishny, 1990, Do Managerial Objectives Drive Bad Acquisitions?, Journal of Finance 45, 3148. Ozbas, Oguzhan, and David Scharfstein, 2010, Evidence on the Dark Side of Internal Capital Markets, Review of Financial Studies 23, 581500. Poterba, James M., and Lawrence .H. Summers, 1995, A CEO survey of US companies time horizons and hurdle rates, Sloan Management Review 37, 43. Rajan, Raghuram, Henri Servaes, and Luigi Zingales, 2000, The Cost of Diversity: The Diversication Discount and Inecient Investment, The Journal of Finance 55, 3580. Scharfstein, David S., and Jeremy C. Stein, 2000, The Dark Side of Internal Capital Markets: Divisional Rent-Seeking and Inecient Investment, The Journal of Finance 55, 25372564. Shin, Hyun-Han, and Ren M. Stulz, 1998, Are Internal Capital Markets Ecient?, e Quarterly Journal of Economics 113, 531552. 27

Stein, Jeremy C., 1996, Rational Capital Budgeting in an Irrational World, Journal of Business 69, 429455.

28

Figures

A A Figure 1. This gure shows average non-core division investment by deciles of DIV,t CORE,t . We consider three dierent measures of investment, namely Raw, Industry Adjusted and Industry-Firm Adjusted Investment. All three investment measures are dened in the Appendix.

29

Figure 2. Average Cumulative Abnormal Returns (Bidder). Average cumulative abnormal returns of the bidder around the announcement of an asset acquisition conditional on whether the WACC of the bidder exceeds that of the target. Only diversifying acquisitions in which the acquiring rms Fama and French (1997) industry diers from that of the target are considered. Abnormal returns are market adjusted and calculated as the dierence between the acquiring rms daily stock return and the CRSP Value Weighted Market Return on the respective event day. All transactions between 1988 and 2007 fullling the sample construction conditions are considered (N=6,115).

30

Tables
Table I

Firm-Level Summary Statistics

This table reports summary statistics of the employed rm-level variables. Variables based on data from Compustat and CRSP are observed for the period of 1987 to 2007. CEO related variables from Compustat Execucomp are observed from 1992 to 2007 only. All variables are dened in the Appendix, except for betas, which are dened in the text (see section I.3). Standalone Firms have their activities concentrated in a single FF48 industry, whereas Conglomerate Firms are diversied across at least two dierent FF48 sectors. All variables are trimmed at the rst and 99th percentile. Standalone Firms Mean Firm Cash Flowt Firm Sizet Firm Aget Firm Investmentt+1 Leveraget Salest Sales Growtht QF IRM,t A AV ERAGE,t External Financet+1 log(Market Cap)t CEO Share Ownershipt 0.029 4.300 2.042 0.065 0.188 3.937 0.126 1.879 0.631 0.146 4.295 3.491 Median 0.063 4.277 2.079 0.038 0.092 4.104 0.086 1.417 0.600 0.056 4.259 0.470 SD 0.169 2.456 0.983 0.078 0.239 2.658 0.367 1.271 0.339 0.326 2.343 18.581 P25 -0.015 2.631 1.386 0.013 0.000 2.345 -0.042 1.037 0.390 -0.020 2.683 0.105 P75 0.120 5.939 2.773 0.086 0.302 5.749 0.263 2.265 0.830 0.220 5.891 2.400 N 107796 122161 119127 103730 121064 122219 105363 93591 115605 93395 101813 14535

Conglomerate Firms Mean Firm Cash Flowt Firm Sizet Firm Aget Firm Investmentt+1 Leveraget Number of Divisionst Salest Sales Growtht Firm Focust QF IRM,t SD(Qi,t )/QF IRM,t A AV ERAGE,t A SD(i,t ) External Financet+1 log(Market Cap)t CEO Share Ownershipt 0.061 5.928 2.792 0.060 0.229 2.557 5.833 0.102 0.733 1.486 0.174 0.550 0.202 0.070 5.479 2.496 Median 0.075 6.008 3.045 0.047 0.201 2.000 6.041 0.072 0.743 1.222 0.128 0.537 0.171 0.028 5.493 0.290 SD 0.105 2.435 0.933 0.054 0.198 0.867 2.413 0.260 0.173 0.844 0.158 0.236 0.152 0.234 2.505 6.990 P25 0.036 4.140 2.197 0.025 0.074 2.000 4.204 -0.016 0.594 1.007 0.063 0.394 0.095 -0.030 3.624 0.085 P75 0.113 7.721 3.526 0.078 0.328 3.000 7.569 0.178 0.884 1.648 0.232 0.675 0.269 0.116 7.330 1.190 N 16150 16507 16408 16217 16466 16507 16507 15714 16418 14002 13911 16507 16507 15033 14906 3889

31

Table II

Non-Core Division-Level Summary Statistics

This table reports summary statistics of variables at the non-core division-level for the sample period of 19872007. Non-core divisions are divisions that do not have the highest sales within the Conglomerate rm. Divisions are dened by grouping together segments operating in the same Fama and French (1997) industry category. Variables are dened in the Appendix, except for betas, which are dened in the text (see section I.3). All variables are trimmed at the rst and 99th percentile. Mean Raw Investmentt+1 Industry Adjusted Investmentt+1 Industry-Firm Adjusted Investmentt+1 Divison Sizet QDIV,t QCORE,t QDIV,t QCORE,t QDIV,t QF IRM,t E DIV,t E CORE,t A DIV,t A CORE,t A A DIV,t CORE,t Raw Investment Core Divisiont Divison Sales Growtht Ind Sales Growtht VDIV,t Relative Importancet Observations 0.064 0.021 0.002 4.065 1.389 1.379 0.010 -0.063 1.082 1.043 0.559 0.549 0.010 0.070 0.067 0.100 0.041 0.318 23811 Median 0.041 0.001 -0.002 4.261 1.302 1.297 0.004 0.084 1.094 1.080 0.537 0.527 0.010 0.051 0.051 0.108 0.027 0.243 SD 0.075 0.073 0.069 2.538 0.366 0.354 0.433 0.821 0.283 0.308 0.289 0.266 0.350 0.082 0.327 0.085 0.044 0.270 P25 0.013 -0.019 -0.028 2.434 1.137 1.137 -0.207 -0.262 0.923 0.882 0.357 0.375 -0.206 0.025 -0.065 0.049 0.009 0.089 P75 0.085 0.040 0.024 5.866 1.542 1.542 0.217 0.344 1.252 1.232 0.717 0.692 0.220 0.088 0.177 0.152 0.055 0.496 N 23811 23808 23210 23213 23798 23771 23758 20115 23811 23811 23811 23811 23811 23455 17144 23811 23732 23213

32

Table III

Summary Statistics of Deal, Bidder and Target Characteristics

This table shows summary statistics of deal, bidder and target characteristics for our sample of 6,115 diversifying transactions. A diversifying transaction is a completed merger or acquisition in which the bidder has successfully gained control of a target belonging to a dierent FF48 industry. All variables are dened in the Appendix.

Panel A: Deal Characteristics All Deals (N=6,115) Public (N=687) Private (N=3,334) Subsidiary (N=2,094)

A (BIDDER,t

A T ARGET,t

> 0)

(QBIDDER,t QT ARGET,t > 0)

VT ARGET,t

EBIDDER,t1

VT ARGET,t /EBIDDER,t1

Target Public?

Target Subsidiary?

Target private?

33
All Deals (N=6,115) Public (N=687) Target 0.640 (0.342) 1.487 (0.388) Di -0.048 (0.405) -0.001 (0.420) Bidder 0.716 (0.371) 1.561 (0.434) Target 0.724 (0.347) 1.568 (0.429) Di -0.009 (0.408) -0.007 (0.471) Bidder 0.602 (0.415) 1.503 (0.390)

Tender Oer?

All Cash?

All Equity?

Multi-Division Acquirer?

Hostile Bid?

Challenging Bid?

0.41 (0.49) 0.49 (0.50) 186.15 (1082.05) 1905.68 (7708.39) 0.25 (0.57) 0.11 (0.32) 0.34 (0.47) 0.55 (0.50) 0.03 (0.17) 0.31 (0.46) 0.12 (0.33) 0.31 (0.46) 0.00 (0.05) 0.01

0.48 (0.50) 0.50 (0.50) 864.61 (3033.24) 6764.01 (17664.76) 0.37 (0.65) 1.00 (0.00) 0.00 (0.00) 0.00 (0.00) 0.26 (0.44) 0.40 (0.49) 0.31 (0.46) 0.47 (0.50) 0.02 (0.14) 0.06

0.41 (0.49) 0.49 (0.50) 65.73 (211.98) 1090.06 (5094.96) 0.21 (0.48) 0.00 (0.00) 0.00 (0.00) 1.00 (0.00) 0.00 (0.00) 0.25 (0.43) 0.14 (0.35) 0.26 (0.44) 0.00 (0.00) 0.00

0.38 (0.49) 0.49 (0.50) 155.28 (397.44) 1610.35 (4589.79) 0.29 (0.66) 0.00 (0.00) 1.00 (0.00) 0.00 (0.00) 0.00 (0.00) 0.38 (0.48) 0.03 (0.18) 0.33 (0.47) 0.00 (0.00) 0.00

Panel B: Bidder and Target Characteristics Private (N=3,334) Subsidiary (N=2,094)

A i,t

Qi,t

Bidder 0.592 (0.401) 1.487 (0.390)

Target 0.646 (0.346) 1.503 (0.391)

Di -0.044 (0.402) 0.000 (0.422)

Bidder 0.536 (0.379) 1.437 (0.367)

Target 0.603 (0.328) 1.436 (0.364)

Di -0.067 (0.409) 0.001 (0.398)

Standard deviations in parentheses. p < 0.10, p < 0.05, p < 0.01

Table IV

Diversifying Transactions Sample by Calendar Year

This table shows the temporal distribution of the sample of diversifying mergers and acquisitions. Yearly average and standard deviations of the nominal deal value are calculated by relying on the deal value as disclosed by SDC.

# of Acquisitions 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 Total 97 130 120 105 157 227 308 327 461 671 727 425 348 220 252 264 275 358 361 282 6115

Mean Value (Million US-$) 216.55 192.35 79.39 48.55 44.52 55.17 80.14 155.16 127.40 122.99 135.35 266.78 373.22 182.75 142.98 169.39 193.14 399.88 313.49 246.17 186.15

SD 617.18 672.38 319.24 84.94 87.95 152.94 298.45 1130.09 613.60 412.44 522.92 1235.61 1777.27 423.87 456.41 537.68 426.93 2966.29 1544.47 580.86 1082.05

34

Table V

Non-Core Division-Level Investment Regressions

Using business segment data from Compustat (19872007), we construct industry-level divisions by aggregating segment data by Fama and French (1997) industries. A division is dened as the bundle of a rms segments operating in the same FF48 industry. The regressions are run on divisions which do not have the highest sales in the conglomerate (non-core divisions). The dependent variable is Raw Investment, that is division-level capital expenditures in period t + 1 scaled by A division assets in period t. CORE,t is the industry-level asset beta of the core-division (i.e. the division with the highest A sales). DIV,t is the industry-level asset beta of the non-core division. QDIV,t is the divisions industry-level Tobins q. QCORE,t is the industry-level Tobins q of the division with the highest sales in the conglomerate. Both are calculated for a sample of standalone rms belonging to the same FF48 industry. Firm Cash Flow is the rms cash ow scaled by total assets. Division Size is the logarithm of division sales and Firm Size is the log of total assets. Firm Age is the logarithm of the current year plus one minus the year in which the rm rst appeared in the Compustat North America database. Firm Focus is the ratio of the rms core division sales to total sales. Ind Sales Growth is the average sales growth observed in the non-core divisions FF48 industry between t and t 1. Med and High Ind Sales Growth indicate whether industry level sales growth of the non-core division falls in the second, or third tercile of lagged industry level sales growth. External Finance measures the rms contemporaneous equity and debt issues scaled by lagged assets. Med and High External Finance indicate whether the rms contemporaneous external nance activity falls in the second or third tercile. (1)
A DIV,t

(2) 0.0157 (6.71)

(3) 0.0155 (6.53)

(4)

(5) 0.0125 (3.92)

(6) 0.0151 (5.57)

A CORE,t

0.0181 (8.13)

A DIV,t

0.0150 (4.74) -0.0162 (-4.90) 0.0000 (0.01) -0.0022 (-1.19) 0.0018 (0.45) 0.0077 (1.91) 0.0106 (8.01) 0.0213 (13.75) 0.0023 (0.62) -0.0020 (-0.50) 0.0073 (3.07) -0.0015 (-0.61) 0.0915 (11.28) 0.0015 (2.42) -0.0005 (-0.79) -0.0015 (-1.49) 0.0074 (3.09) -0.0015 (-0.63) 0.0889 (10.93) 0.0020 (2.95) -0.0007 (-1.03) -0.0015 (-1.47) 0.0086 (1.60) 0.0738 (37.28) 23814 0.019 0.0630 (11.88) 22676 0.036 0.0566 (8.91) 22289 0.036 0.0077 (3.06) -0.0013 (-0.52) 0.0889 (10.93) 0.0020 (2.99) -0.0008 (-1.09) -0.0015 (-1.47) 0.0085 (1.59) 0.0568 (8.90) 22289 0.036 0.0077 (3.13) -0.0013 (-0.53) 0.0890 (10.94) 0.0020 (2.94) -0.0007 (-1.00) -0.0015 (-1.49) 0.0088 (1.64) 0.0569 (8.76) 22289 0.037 0.0075 (3.02) -0.0020 (-0.84) 0.0889 (10.83) 0.0021 (2.97) -0.0008 (-1.04) -0.0006 (-0.60) 0.0063 (1.13) 0.0449 (6.75) 20334 0.052

A CORE,t

Med Ind Sales Growtht High Ind Sales Growtht

A A Med Ind Sales Growtht (DIV,t CORE,t )

A A High Ind Sales Growtht (DIV,t CORE,t )

Med External Financet+1 High External Financet+1

A A Med External Financet+1 (DIV,t CORE,t )

A A High External Financet+1 (DIV,t CORE,t )

QDIV,t QCORE,t Firm Cash Flowt Divison Sizet Firm Sizet Firm Aget Firm Focust Constant Observations R2

All regressions include year xed eects. t statistics in parentheses. Standard errors clustered at the rm level. p < 0.10, p < 0.05, p < 0.01

Table VI

Non-Core Division-Level Investment Regressions: Robustness Checks

This table shows robustness checks on the specications of non-core division-level investment used in the previous tables. The dependent variable is Raw Investment, that is A division-level capital expenditures in period t + 1 scaled by division assets in period t. CORE,t is the industry-level asset beta of the core-division (i.e. the division with largest A A sales). DIV,t is the industry-level asset beta of the non-core division. AV ERAGE,t is a rm wide average beta and is calculated as the weighted average of division-level asset betas, where the weights correspond to the ratio of division-level to total rm assets. QDIV,t and QCORE,t are the industry level Tobins q of the non-core and core divisions respectively. QDIV,t QF IRM,t is the gap between the industry-level Tobins q of the non-core division and the rm-wide Tobins q. SD(Qi,t )/QF IRM,t is the standard deviation of a rms division-level Tobins qs (core and non-core divisions) in a given year scaled by the overall Tobins q of the rm. Raw Investment Core Division is calculated as capital expenditures of the core division in period t + 1 scaled by the core divisions assets in period t. All other variables are as previously dened. (1) -0.0212 (-3.76) -0.0154 (-3.42) 0.0074 (0.99) 0.0149 (4.05) -0.0073 (-5.38) -0.0267 (-4.85) 0.0893 (6.74) 0.0077 (3.07) -0.0027 (-1.14) 0.0933 (12.12) 0.0924 (12.02) 0.0857 (9.84) 0.0018 (2.40) -0.0004 (-0.55) -0.0003 (-0.07) -0.0008 (-0.80) 0.0618 (10.04) 22840 0.034 YES NO -0.0004 (-0.08) -0.0007 (-0.75) 0.0610 (9.91) 22840 0.035 YES NO 0.0067 (1.14) -0.0021 (-1.74) 0.0611 (8.32) 18955 0.043 YES NO -0.0006 (-0.26) -0.0055 (-2.03) 0.0076 (3.05) 0.0125 (4.21) 0.0125 (4.26) 0.0010 (0.37) 0.0885 (10.01) 0.0016 (2.06) -0.0003 (-0.39) 0.0058 (1.00) -0.0024 (-1.94) 0.0601 (8.24) 18860 0.041 YES NO 0.0083 (3.34) -0.0022 (-0.94) 0.0812 (10.31) 0.0023 (3.41) -0.0010 (-1.42) 0.0071 (1.37) -0.0012 (-1.26) 0.0518 (8.07) 22006 0.045 YES NO -0.0026 (-1.10) 0.0752 (9.64) 0.0021 (3.02) -0.0006 (-0.90) 0.0081 (1.70) -0.0011 (-1.13) 0.0564 (9.81) 22017 0.151 NO YES 0.0799 (6.73) 0.0141 (4.15) 0.0126 (3.68) 0.0156 (4.95) 0.0195 (5.73) -0.0170 (-4.87) -0.0184 (-5.25) -0.0171 (-5.20) -0.0088 (-2.74) (2) (3) (4) (5) (6)

A CORE,t

A AV ERAGE,t

A DIV,t

QDIV,t QF IRM,t

SD(Qi,t )/QF IRM,t

Raw Investment Core Divisiont

36

QDIV,t

QCORE,t

Firm Cash Flowt

Divison Sizet

Firm Sizet

Firm Focust

Firm Aget

Constant

Observations R2

Year Fixed Eects Division Industry*Year Fixed Eects

t statistics in parentheses. Standard errors clustered at the rm level. p < 0.10, p < 0.05, p < 0.01

Table VII

Vertical Integration of Non-Core and Core Division

The specications in this table account for the extent of vertical integration between the non-core and core division. The dependent variable is Raw Investment, that is division-level capital expenditures in period t + 1 scaled by division assets in period t. VDIV,t measures the extent to which the industry of the non-core division and the industry of the core division exchange production inputs. It is the average of (i) the fraction of the non-core divisions industry output that is sold to the industry of the core division and (ii) the fraction of the non-core divisions industry inputs that comes from the core divisions industry. Fan and Lang (2000) interpret this measure as a proxy for the opportunity for vertical integration between two industries. High values of VDIV,t indicate that both industries exchange signicant amounts of inputs while low values indicate vertical unrelatedness. Medium and High VDIV,t are dummy variables indicating whether VDIV,t falls in the second or third tercile. All other variables are as previously dened. (1)
A DIV,t

(2) 0.0150 (3.80) -0.0077 (-1.51) 0.0082 (1.42)

(3) 0.0143 (3.60) -0.0058 (-1.12) 0.0076 (1.27) 0.0082 (3.38)

(4) 0.0138 (3.48) -0.0055 (-1.05) 0.0096 (1.55) 0.0097 (2.79) -0.0009 (-0.17) -0.0044 (-0.78)

A CORE,t

0.0163 (6.75)

A A Medium VDIV,t (DIV,t CORE,t )

A A High VDIV,t (DIV,t CORE,t )

QDIV,t Medium VDIV,t QDIV,t High VDIV,t QDIV,t QCORE,t Medium VDIV,t QCORE,t High VDIV,t QCORE,t Medium VDIV,t High VDIV,t Raw Investment Core Divisiont Firm Cash Flowt Divison Sizet Firm Sizet Firm Aget Firm Focust Constant Observations R2

0.0084 (3.47)

0.0079 (3.26)

-0.0022 (-0.91)

-0.0016 (-0.67)

-0.0044 (-1.45) 0.0099 (2.21) -0.0008 (-0.16)

-0.0048 (-1.59) 0.0101 (2.18) 0.0011 (0.21) -0.0072 (-0.87) 0.0133 (1.44) 0.0890 (6.78) 0.0816 (10.30) 0.0020 (3.01) -0.0006 (-0.87) -0.0010 (-1.04) 0.0055 (1.06) 0.0475 (6.09) 21932 0.048

0.0052 (2.83) 0.0087 (4.20) 0.0893 (6.73) 0.0813 (10.27) 0.0022 (3.35) -0.0009 (-1.34) -0.0012 (-1.21) 0.0071 (1.35) 0.0505 (7.80) 21932 0.044 0.0892 (6.77) 0.0817 (10.34) 0.0020 (3.04) -0.0006 (-0.89) -0.0010 (-1.03) 0.0054 (1.04) 0.0456 (6.89) 21932 0.048

-0.0081 (-1.26) 0.0099 (1.30) 0.0890 (6.78) 0.0816 (10.30) 0.0020 (3.02) -0.0006 (-0.87) -0.0010 (-1.03) 0.0055 (1.05) 0.0490 (6.75) 21932 0.048

All regressions include year xed eects. t statistics in parentheses. Standard errors clustered at the rm level. p < 0.10, p < 0.05, p < 0.01

37

Table VIII

Evidence Consistent with Bounded Rationality

The dependent variable is Raw Investment, that is division-level capital expenditures in period t+1 scaled by division assets in period t. Column (1) includes interaction terms between dummy variables for the three distinct sub-periods (1992-1996), A A (19972001) and (20022007) and DIV,t CORE,t . Relative Importance is a measure of a divisions organizational importance. It is calculated as non-core-division sales scaled by the core divisions sales. Med and High Relative Importance are dummy variables equaling one whenever the relative importance of the divisions falls into the second respectively third A tercile. SD(j,t ) is the within-rm standard deviation of divisional asset betas in a given year, which is calculated for the A asset beats of both the core and the non-core divisions. Med and High SD(i,t ) are dummy variables indicating whether the within-rm standard deviation of asset betas falls into the second or third tercile. High CEO Ownership is a dummy variable indicating whether CEO equity ownership exceeds 1%. Although coecient estimates are not reported in this table, all specications also control for QDIV,t , QCORE,t , Firm Cash Flow, Division Size , Firm Size, Firm Age and Firm Focus. (1)
A A DIV,t CORE,t

(2) 0.0182 (4.64)

(3) 0.0302 (3.23)

(4) 0.0295 (1.82)

0.0264 (5.52) -0.0058 (-1.94) -0.0223 (-7.05) -0.0218 (-5.49) -0.0038 (-0.71) -0.0112 (-1.84) -0.0196 (-3.45)

(1992-1996) (1997-2001) (2002-2007)

A A (1992-1996) (DIV,t CORE,t )

A A (1997-2001) (DIV,t CORE,t )

A A (2002-2007) (DIV,t CORE,t )

Med Relative Importancet High Relative Importancet

A A Med Relative Importancet (DIV,t CORE,t )

0.0003 (0.14) 0.0041 (1.27) 0.0011 (0.22) -0.0094 (-1.88) -0.0008 (-0.49) -0.0042 (-2.16) -0.0017 (-0.17) -0.0185 (-1.95) 0.0007 (0.19) -0.0243 (-2.96) 0.0046 (2.16) -0.0017 (-0.93) 22289 0.038 22288 0.037 22288 0.038 5434 0.038

A A High Relative Importancet (DIV,t CORE,t )

A Med SD(i,t )

A High SD(i,t )

A A A Med SD(i,t ) (DIV,t CORE,t )

A A A High SD(i,t ) (DIV,t CORE,t )

High CEO Share Ownershipt

A A High CEO Share Ownershipt (DIV,t CORE,t )

log(Market Cap)t
A A log(Market Cap)t (DIV,t CORE,t )

Observations R2

All regressions include year xed eects. t statistics in parentheses. Standard errors clustered at the rm level. p < 0.10, p < 0.05, p < 0.01

38

Table IX

Bidder Abnormal Returns as a Function of the Spread in Industry-Level Asset Betas

This table shows cross-sectional regressions in which the abnormal returns of bidders on the announcement day of diversifying mergers and acquisitions are regressed on bidder, target and deal characteristics as well as measures of the targets and the bidders cost of capital. Diversifying acquisitions are dened as transactions in which a bidder buys an asset belonging to an industry dierent from the bidders industry. Industries are dened according to the 48 categories proposed by Fama and French (1997). Abnormal returns are market adjusted and calculated as the dierence between the acquiring rms daily stock return and the return on the CRSP Value Weighted Market Index. All specications are estimated on a sample of 6,115 deals between 1988 and 2007 fullling the sample construction criteria. All regressions include year dummies and standard errors are clustered by announcement dates. All variables are dened in the Appendix. (1) AR(0) 0.00444 (2.77) -0.00251 (-4.45) 0.00155 (1.02) -0.00920 (-3.46) 0.00305 (1.68) 0.00260 (1.45) 0.00859 (1.95) 0.0120 (4.34) 0.00711 (1.62) 0.00307 (1.69) 0.00304 (1.53) 0.00986 (2.14) 0.0120 (4.34) 0.000489 (0.36) 0.00452 (1.16) -0.0110 (-3.86) -0.0116 (-4.06) -0.0127 (-4.38) 0.00145 (0.95) 0.00149 (0.98) 0.00131 (0.87) 0.00137 (0.90) 0.00159 (1.03) -0.0131 (-4.52) 0.00286 (1.47) 0.00930 (2.04) 0.0121 (4.37) 0.0000423 (0.03) 0.00468 (1.19) 0.00513 (2.48) -0.00252 (-4.45) -0.00197 (-3.04) -0.00200 (-3.08) -0.00270 (-4.09) -0.00275 (-4.08) -0.00293 (-4.12) 0.00401 (2.50) 0.00451 (2.79) 0.00448 (2.77) 0.00403 (2.51) 0.00380 (2.42) 0.00314 (2.01) (2) AR(0) (3) AR(0) (4) AR(0) (5) AR(0) (6) AR(0) (7) AR(0) (8) AR(0) 0.00316 (2.02) -0.00293 (-4.12) 0.00158 (1.03) -0.0131 (-4.51) 0.00287 (1.47) 0.00909 (1.91) 0.0121 (4.37) 0.0000487 (0.04) 0.00470 (1.20) 0.00512 (2.47) -0.000771 (-0.07) 0.00245 (0.37) 0.0143 (6.23) 6115 0.008 Yes 6115 0.008 Yes 0.0137 (6.13) 0.0117 (5.16) 6115 0.011 Yes 0.0117 (5.18) 6115 0.011 Yes 0.0116 (5.17) 6115 0.025 Yes 0.0111 (4.97) 6115 0.025 Yes 0.0105 (4.84) 6115 0.027 Yes 0.0105 (4.84) 6115 0.027 Yes

A A (BIDDER,t T ARGET,t > 0)

log(VT ARGET,t )

(QBIDDER,t QT ARGET,t > 0)

Target Public?

Target Subsidiary?

Tender Oer?

39

VT ARGET,t /EBIDDER,t1

All Cash?

All Equity?

Multi-Division Acquirer?

Hostile Bid?

Challenging Bid?

Constant

Observations Adjusted R2

Year Fixed Eects

t statistics in parentheses p < 0.10, p < 0.05, p < 0.01

Table X

Bidder Cumulative Abnormal Returns as a Function of the Spread in Industry-Level Asset Betas

This table shows cross-sectional regressions in which seven day cumulative abnormal returns of bidders around announcements of diversifying mergers and acquisitions are regressed on bidder, target and deal characteristics as well as measures of the targets and the bidders cost of capital. Diversifying acquisitions are dened as transactions in which a bidder buys an asset belonging to an industry dierent from the bidders industry. Industries are dened according to the 48 categories proposed by Fama and French (1997). Abnormal returns are market adjusted and calculated as the dierence between the acquiring rms daily stock return and the return on the CRSP Value Weighted Market Index. All specications are estimated on a sample of 6,115 deals between 1988 and 2007 fullling the sample construction criteria. All regressions include year dummies and standard errors are clustered by announcement dates. (1) CAR(3,3) 0.00813 (2.42) -0.00308 (-3.12) -0.000929 (-0.29) -0.0181 (-3.32) 0.000367 (0.11) 0.00471 (0.54) 0.0264 (5.41) 0.00797 (0.91) 0.000384 (0.11) -0.000648 (-0.19) 0.00102 (0.29) 0.0122 (1.34) 0.0265 (5.43) 0.00472 (1.52) 0.0204 (2.77) -0.0193 (-3.20) -0.0208 (-3.45) -0.0254 (-4.00) -0.00110 (-0.34) -0.00108 (-0.33) -0.00148 (-0.46) -0.00116 (-0.36) -0.000984 (-0.30) -0.0257 (-4.06) 0.000877 (0.25) 0.0118 (1.29) 0.0266 (5.44) 0.00436 (1.39) 0.0205 (2.79) 0.00417 (1.16) -0.00308 (-3.11) -0.00185 (-1.69) -0.00187 (-1.70) -0.00341 (-3.01) -0.00368 (-3.22) -0.00383 (-3.29) 0.00838 (2.46) 0.00914 (2.66) 0.00912 (2.66) 0.00813 (2.38) 0.00694 (2.06) 0.00641 (1.87) (2) CAR(3,3) (3) CAR(3,3) (4) CAR(3,3) (5) CAR(3,3) (6) CAR(3,3) (7) CAR(3,3) (8) CAR(3,3) 0.00640 (1.85) -0.00382 (-3.28) -0.000978 (-0.30) -0.0257 (-3.98) 0.000876 (0.25) 0.0119 (1.22) 0.0265 (5.43) 0.00435 (1.38) 0.0205 (2.78) 0.00418 (1.15) -0.0000812 (-0.00) -0.00132 (-0.05) 0.0274 (6.44) 6115 0.008 Yes 6115 0.008 Yes 0.0277 (6.26) 0.0251 (5.53) 6115 0.009 Yes 0.0252 (5.54) 6115 0.009 Yes 0.0249 (5.50) 6115 0.024 Yes 0.0220 (4.80) 6115 0.027 Yes 0.0216 (4.72) 6115 0.027 Yes 0.0216 (4.72) 6115 0.026 Yes

A A (BIDDER,t T ARGET,t > 0)

log(VT ARGET,t )

(QBIDDER,t QT ARGET,t > 0)

Target Public?

Target Subsidiary?

Tender Oer?

40

VT ARGET,t /EBIDDER,t1

All Cash?

All Equity?

Multi-Division Acquirer?

Hostile Bid?

Challenging Bid?

Constant

Observations Adjusted R2

Year Fixed Eects

t statistics in parentheses. p < 0.10, p < 0.05, p < 0.01

Appendix

Denition of rm-level variables

Firm Cash Flow is the sum of income before extraordinary items (item IB t ) and depreciation and amortization (item DP t ) scaled by total assets (item AT t ). Firm Size is the natural logarithm of the rms total assets (item AT t ). Firm Age is the logarithm of the current year plus one minus the year in which the rm rst appeared in the Compustat North America database. Firm Investment is total rm wide capital expenditures (item CAPX t+1 ) scaled by total rm assets (item AT t ). Leverage is long term debt (item DLTT t ) scaled by total assets (item AT t ). Number of Divisions is obtained by grouping business segments by 2-digit Fama and French (1997) industries and counting the number of dierent industries in which a rm operates in a given year. By denition, the Number of Divisions is equal to 1 in standalone rms. Sales is the natural logarithm of total rm sales (item SALE t ). Sales Growth is the rms total sales growth between periods t 1 and t. Firm Focus is sales of the division with the highest level of sales (core division) divided by total rm wide sales (item SALE t ). It is by denition equal to 1 for standalone rms. QF IRM,t is the rms Market to Book ratio. Market value of assets is calculated as the book value of assets (item AT t ) plus the market value of common equity at scal year end (item CSHO t item PRCC F t ) minus the book value of common equity (item CEQ t ) and balance sheet deferred taxes (item TXDB t ). SD(Qi,t )/QF IRM,t is the standard deviation of a rms division-level Tobins qs scaled by the rm wide Tobins q. We dene division-level Tobins q in the next section of the Appendix. By denition, it is equal to zero for standalone rms. log(Market Cap) is natural log of the market value of rm equity. It is dened as 41

the number of shares outstanding (item CSHO t ) times the price of each share (item PRCC F t ). EF IRM,t is the market value of rm equity. It is dened as the number of shares outstanding (item CSHO t ) times the price of each share (item PRCC F t ). DF IRM,t is the book value of debt, measured as book assets (item AT t ) minus common equity (item CEQ t ) and deferred taxes (TXDB t ). External Finance is the sum of the change in equity issues (CEQ+T XDBRE) and the change in debt issues (CEQ T XDB) scaled by total assets in period t, where denotes the forward dierence operator (both from period t to t + 1).
A SD(j,t ) is the within-rm standard deviation of divisional asset betas in a given year,

which is calculated for the asset beats of both the core and the non-core divisions. By denition, it is equal to zero for standalone rms. CEO Share Ownership is item shrown excl opts pct t from the Compustat Execucomp Database.

42

Denition of division-level variables

All division-level accounting information (assets, capx, sales) comes from the Compustat Segment les, aggregated at the 2-digit (FF48) Fama-French industry level. We rely on the variable ssic1, which measures the closest Primary SIC code for the Segment. Whenever this variable is missing, we use the variable ssicb1. According to the Compustat manual, both variables have the same denition, namely Segment Primary SIC Code but are retrieved from two dierent historical les. We match this 4 digit SIC code to the corresponding FF48 industry. We use an extract of the Compustat Segment les downloaded from WRDS in September 2010. Raw Investment is division-level capital expenditures normalized by total division assets at the previous scal year end, that is Industry Adjusted Investment is
CAP Xi,t+1 ASSET Si,t CAP Xi,t+1 . ASSET Si,t

ASSETi,t+1 , where S SA
i,t

CAP X SA

SA CAP Xi,t+1 SA ASSET Si,t

is the median

investment of standalone rms belonging to the divisions FF48 industry in that year. Industry-Firm Adjusted Investment is Industry Adjusted investment less the weighted average industry adjusted investment across all divisions of the rm. Formally, that is
CAP Xi,t+1 ASSET Si,t

SA CAP Xi,t+1 SA ASSET Si,t

N i=1

wi

CAP Xi,t+1 ASSET Si,t

SA CAP Xi,t+1 SA ASSET Si,t

, where wi,t is the ratio of

division to total rm assets. Division Size is the natural logarithm of division-level sales. QDIV,t is the estimated Tobins q of the division. To compute it, we calculate, each year, the median q of standalone rms that belong to the same FF48 industry as the division. See the previous section for the denition of q at the rm level. QCORE,t is the Tobins q of the core division of the rm to which a given division belongs. For core divisions, QDIV,t = QCORE,t . VDIV,t measures the ows of goods and services between the industries of the non-core and core division. See section I.4 for details. Relative Importance is the ratio between the sales of the non-core and the core division. It is a proxy for the organizational importance of the non-core division.

43

Ind Sales Growth is the mean sales growth between t 1 and t in a given FF48 industry.

44

Denition of M&A related variables

A A (BIDDER,t T ARGET,t > 0) indicates whether the dierence between the bidders and

the targets industry level asset betas is positive. See section I.3 for details on how the asset betas are constructed. VT ARGET,t is the value of the transaction as disclosed by SDC (in Million US-$). Qi,t is the estimated industry level Tobins q of rm i, where i indexes either the target or the bidder. To compute it, we calculate, each year, the median q of standalone rms that belong to the same FF48 industry as the rm. For the denition of Tobins q, see the previous sections. (QBIDDER,t QT ARGET,t > 0) is a dummy variable indicating whether the dierence between the bidders and the targets industry level Tobins qs is positive. EBIDDER,t1 is the scal year end equity market value of the bidder in the year prior to the bid announcement (in Million US-$), which is calculated as (item CSHO t1 ) times share price at scal year end (item PRCC Ft1 ). Target Public? is a dummy variable indicating whether the target is a public company. Target Subsidiary? is a dummy variable indicating whether the target is a subsidiary of either a private or a public company. Target Private? is a dummy variable indicating whether the target is private. Tender Oer? indicates whether the bidder sought control through the process of a tender oer. All Cash? is a dummy variable that takes on the value one when the consideration was entirely paid in cash. All Equity? takes on the value of one whenever the targets shareholders are entirely compensated with shares. Multi-Division Acquirer? takes on the value of one when the bidders business activities are diversied across at least two FF48 industries. Hostile Bid indicates whether the bids nature is hostile.

45

Challenging Bid? challenging bid.

is a dummy variable indicating and whether there has been a

46

Table A.I

Weighted Average Cost of Capital (Industry Level)

E This table shows time series averages of the yearly industry level equity beta i,t , industry level leverage Ei,t /(Di,t + Ei,t ) A , where i indexes FF48 industries. Yearly industry level debt (D ) and equity (E ) and the industry level asset beta i,t i,t i,t are the aggregate debt and aggregate market value of equity observed in the respective FF48 industry in a given year. T T E E A A Time series averages are calculated over the sample period 19872007 as i = (1/T ) t=1 i,t , i = (1/T ) t=1 i,t E and (E/(D + E))i = (1/T ) T (Ei,t /(Di,t + Ei,t )). Yearly industry level equity betas (j,t ) are obtained by regressing t=1 monthly returns of value-weighted portfolios comprised of companies belonging to the same FF48 industry on the CRSP Value Weighted Index for moving-windows of 60 months. E i E ( D+E )i A i

FF48 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48

Industry Agric Food Soda Beer Smoke Toys Fun Books Hshld Clths Hlth MedEq Drugs Chems Rubbr Txtls BldMt Cnstr Steel FabPr Mach ElcEq Autos Aero Ships Guns Gold Mines Coal Oil Util Telcm PerSv BusSv Comps Chips LabEq Paper Boxes Trans Whlsl Rtail Meals Banks Insur RlEst Fin Other

Description Agriculture Food Products Candy & Soda Beer & Liquor Tobacco Products Recreation Entertainment Printing and Publishing Consumer Goods Apparel Healthcare Medical Equipment Pharmaceutical Products Chemicals Rubber and Plastic Products Textiles Construction Materials Construction Steel Works Etc Fabricated Products Machinery Electrical Equipment Automobiles and Trucks Aircraft Shipbuilding, Railroad Equipment Defense Precious Metals Non-Metallic and Industrial Metal Mining Coal Petroleum and Natural Gas Utilities Communication Personal Services Business Services Computers Electronic Equipment Measuring and Control Equipment Business Supplies Shipping Containers Transportation Wholesale Retail Restaraunts, Hotels, Motels Banking Insurance Real Estate Trading Almost Nothing

0.77 0.64 0.76 0.60 0.77 1.48 1.14 0.91 0.90 1.04 0.92 0.90 0.84 0.93 1.10 0.93 0.97 1.19 1.16 0.96 1.20 1.28 1.05 0.95 0.79 0.68 0.55 0.93 0.84 0.64 0.42 1.02 1.03 1.40 1.29 1.55 1.53 0.92 0.77 1.01 0.98 1.05 0.97 1.04 0.82 0.77 0.98 1.04

0.68 0.61 0.50 0.76 0.58 0.56 0.60 0.62 0.66 0.68 0.55 0.79 0.84 0.54 0.54 0.45 0.57 0.40 0.49 0.50 0.55 0.49 0.30 0.53 0.48 0.50 0.83 0.69 0.46 0.63 0.38 0.54 0.62 0.70 0.64 0.69 0.74 0.52 0.43 0.38 0.41 0.63 0.62 0.19 0.17 0.34 0.15 0.53

0.53 0.38 0.38 0.44 0.44 0.82 0.69 0.56 0.58 0.70 0.49 0.70 0.71 0.49 0.61 0.42 0.55 0.48 0.57 0.49 0.66 0.63 0.31 0.50 0.34 0.32 0.44 0.65 0.40 0.40 0.16 0.55 0.63 0.99 0.86 1.10 1.16 0.47 0.33 0.38 0.39 0.64 0.60 0.19 0.14 0.24 0.14 0.57

47

Table A.II

Non-Core Division-Level Investment Regressions (Industry Adjusted Investment)

The dependent variable in all regressions of this table is Industry Adjusted Investment, which is dened in the Appendix. All other variables are as dened in Table V. (1)
A DIV,t

(2) 0.0070 (3.03)

(3) 0.0067 (2.83)

(4)

A CORE,t

0.0051 (2.31)

A DIV,t

0.0067 (2.14) -0.0067 (-2.12) -0.0052 (-2.12) -0.0011 (-0.46) 0.0844 (10.28) -0.0003 (-0.59) 0.0006 (0.96) -0.0013 (-1.33) -0.0050 (-2.02) -0.0011 (-0.46) 0.0818 (9.92) 0.0002 (0.28) 0.0003 (0.51) -0.0012 (-1.28) 0.0090 (1.77) 0.0265 (13.41) 23808 0.004 0.0311 (5.77) 22674 0.017 0.0246 (3.81) 22288 0.018 -0.0050 (-1.89) -0.0011 (-0.45) 0.0818 (9.92) 0.0002 (0.28) 0.0003 (0.50) -0.0012 (-1.28) 0.0090 (1.78) 0.0246 (3.81) 22288 0.018

A CORE,t

QDIV,t QCORE,t Firm Cash Flowt Divison Sizet Firm Sizet Firm Aget Firm Focust Constant Observations R2

All regressions include year xed eects. t statistics in parentheses. Standard errors clustered at the rm level. p < 0.10, p < 0.05, p < 0.01

48

Table A.III

Non-Core Division-Level Investment Regressions: Robustness Checks (Industry Adjusted Investment)

(1) -0.0125 (-2.29) -0.0053 (-1.22) 0.0079 (2.36) 0.0052 (1.45) -0.0075 (-5.43) -0.0183 (-3.34) 0.0864 (6.79) -0.0055 (-2.08) 0.0001 (0.02) -0.0060 (-2.20) 0.0002 (0.08) 0.0854 (9.58) -0.0001 (-0.19) 0.0007 (0.98) 0.0064 (1.15) -0.0020 (-1.70) -0.0009 (-0.96) 0.0264 (4.18) 22839 0.017 YES NO YES NO 0.0285 (3.84) 18954 0.025 YES NO 0.0062 (1.12) -0.0023 (-1.94) 0.0292 (3.93) 18859 0.021 YES NO 0.0817 (9.25) -0.0000 (-0.02) 0.0007 (0.98) 0.0071 (1.53) -0.0009 (-1.00) 0.0268 (4.26) 22839 0.016 0.0070 (1.52) -0.0025 (-0.79) -0.0016 (-0.70) 0.0811 (10.34) 0.0805 (10.29) -0.0004 (-0.18) -0.0055 (-2.10) -0.0044 (-1.69) -0.0020 (-0.83) 0.0743 (9.31) 0.0004 (0.66) 0.0001 (0.19) 0.0078 (1.58) -0.0010 (-1.09) 0.0195 (3.01) 22005 0.025 YES NO -0.0026 (-1.10) 0.0752 (9.64) 0.0021 (3.02) -0.0006 (-0.90) 0.0081 (1.70) -0.0011 (-1.13) 0.0127 (2.21) 22013 0.108 NO YES 0.0799 (6.73) 0.0068 (2.00) 0.0062 (1.80) 0.0073 (2.36) 0.0082 (1.10) (-2.07) (-2.42) (-2.37) -0.0069 -0.0081 -0.0074 -0.0088 (-2.74) (2) (3) (4) (5) (6)

The dependent variable in all regressions of this table is Industry Adjusted Investment, which is dened in the Appendix. All other variables are as dened in Table VI.

A CORE,t

A AV ERAGE,t

A DIV,t

QDIV,t QF IRM,t

SD(Qi,t )/QF IRM,t

Raw Investment Core Divisiont

QDIV,t

49

QCORE,t

Firm Cash Flowt

Divison Sizet

Firm Sizet

Firm Focust

Firm Aget

Constant

Observations R2

Year Fixed Eects Division Industry*Year Fixed Eects

t statistics in parentheses. Standard errors clustered at the rm level. p < 0.10, p < 0.05, p < 0.01

Table A.IV

Non-Core Division-Level Investment Regressions (Industry-Firm Adjusted Investment)

The dependent variable in all regressions of this table is Industry-Firm Adjusted Investment, which is dened in the Appendix. All other variables are as dened in Table V. (1)
A DIV,t

(2) 0.0093 (3.91)

(3) 0.0094 (3.90)

(4)

A CORE,t

0.0056 (2.49)

A DIV,t

0.0072 (2.44) -0.0120 (-3.76) -0.0065 (-3.08) 0.0040 (1.90) -0.0277 (-3.90) -0.0003 (-0.54) 0.0015 (2.59) 0.0002 (0.21) -0.0065 (-3.06) 0.0041 (1.92) -0.0286 (-4.06) -0.0005 (-0.76) 0.0017 (2.88) 0.0002 (0.24) -0.0022 (-0.49) 0.0008 (0.45) 23210 0.002 -0.0014 (-0.28) 22130 0.006 -0.0008 (-0.14) 21744 0.006 -0.0055 (-2.51) 0.0050 (2.31) -0.0286 (-4.08) -0.0004 (-0.60) 0.0016 (2.58) 0.0002 (0.24) -0.0024 (-0.55) -0.0000 (-0.01) 21744 0.007

A CORE,t

QDIV,t QCORE,t Firm Cash Flowt Divison Sizet Firm Sizet Firm Aget Firm Focust Constant Observations R2

All regressions include year xed eects. t statistics in parentheses. Standard errors clustered at the rm level. p < 0.10, p < 0.05, p < 0.01

50

Table A.V

Non-Core Division-Level Investment Regressions: Robustness Checks (Industry-Firm Adjusted Investment)

(1) -0.0104 (-2.07) -0.0162 (-3.69) -0.0050 (-0.70) 0.0067 (1.81) 0.0023 (1.89) 0.0022 (0.43) -0.1513 (-8.90) -0.0056 (-2.56) 0.0046 (2.20) -0.0237 (-3.45) -0.0002 (-0.33) 0.0014 (2.15) -0.0031 (-0.76) 0.0010 (1.25) 0.0066 (1.20) 22279 0.005 YES NO 0.0010 (1.28) 0.0062 (1.13) 22279 0.005 YES NO -0.0031 (-0.76) -0.0021 (-0.44) -0.0003 (-0.28) 0.0025 (0.39) 18487 0.008 YES NO -0.0241 (-3.51) -0.0286 (-3.69) 0.0055 (2.60) 0.0067 (2.82) 0.0054 (2.20) -0.0301 (-3.81) -0.0002 (-0.33) 0.0015 (2.18) -0.0022 (-0.46) -0.0002 (-0.22) 0.0015 (0.23) 18393 0.007 YES NO -0.0057 (-2.58) -0.0078 (-2.95) -0.0063 (-2.36) -0.0056 (-2.59) 0.0054 (2.54) -0.0186 (-2.69) -0.0007 (-1.14) 0.0018 (3.06) 0.0002 (0.05) -0.0001 (-0.10) 0.0094 (1.60) 21518 0.031 YES NO 0.0046 (2.10) -0.0187 (-2.67) 0.0003 (0.38) 0.0015 (2.26) 0.0009 (0.21) 0.0001 (0.13) 0.0036 (0.64) 21526 0.098 NO YES -0.1581 (-9.09) 0.0081 (2.53) 0.0083 (2.58) 0.0060 (2.11) 0.0089 (2.62) (-3.77) (-3.68) (-3.93) -0.0127 -0.0126 -0.0116 -0.0128 (-4.09) (2) (3) (4) (5) (6)

The dependent variable in all regressions of this table is Industry-Firm Adjusted Investment, which is dened in the Appendix. All other variables are as dened in table VI.

A CORE,t

A AV ERAGE,t

A DIV,t

QDIV,t QF IRM,t

SD(Qi,t )/QF IRM,t

Raw Investment Core Divisiont

QDIV,t

51

QCORE,t

Firm Cash Flowt

Divison Sizet

Firm Sizet

Firm Focust

Firm Aget

Constant

Observations R2

Year Fixed Eects Division Industry*Year Fixed Eects

t statistics in parentheses. Standard errors clustered at the rm level. p < 0.10, p < 0.05, p < 0.01

Table A.VI

Non-Core Division-Level Investment Regressions (Excluding Banking and Insurance)

In this table, we exclude non-core division-year observations belonging to a conglomerate with a core division in the banking or insurance sector (i.e. FF48 industries 44 and 45). All variables are as dened in Table V. (1)
A DIV,t

(2) 0.0162 (6.91)

(3) 0.0160 (6.70)

(4)

A CORE,t

0.0190 (8.46)

A CORE,t

-0.0187 (-5.60) 0.0136 (4.29) 0.0064 (2.69) -0.0026 (-1.09) 0.0887 (10.88) 0.0011 (1.71) 0.0001 (0.21) -0.0018 (-1.76) 0.0066 (2.71) -0.0026 (-1.08) 0.0866 (10.58) 0.0016 (2.37) -0.0002 (-0.23) -0.0018 (-1.78) 0.0090 (1.67) 0.0747 (37.56) 23011 0.020 0.0645 (12.08) 22029 0.036 0.0585 (9.14) 21709 0.037 0.0076 (3.01) -0.0016 (-0.68) 0.0865 (10.56) 0.0017 (2.52) -0.0003 (-0.44) -0.0018 (-1.79) 0.0088 (1.62) 0.0594 (9.23) 21709 0.037

A DIV,t

QDIV,t QCORE,t Firm Cash Flowt Divison Sizet Firm Sizet Firm Aget Firm Focust Constant Observations R2

All regressions include year xed eects. t statistics in parentheses. Standard errors clustered at the rm level. p < 0.10, p < 0.05, p < 0.01

52