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

Transcription

1 The WACC Fallacy: The Real Effects of Using a Unique Discount Rate 1 Philipp Krüger Geneva Finance Research Institute - Université de Genève Augustin Landier Toulouse School of Economics David Thesmar HEC Paris and CEPR December 2010 Preliminary Version Abstract We document investment distortions induced by the use of a single discount rate within firms. According to textbook capital budgeting, firms 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, such as the weighted average cost of capital (WACC), they overinvest (resp. underinvest) in divisions with a market beta higher (resp. lower) than the firm s core industry beta. We directly test this consequence of the WACC fallacy and establish a robust and significant positive relationship between division-level investment and the spread between the division s market beta and the firm s core industry beta. This bias is stronger when the CEO is older and for high growth divisions. It is weaker for larger divisions. We also provide evidence that bidder abnormal returns are higher in diversifying mergers and acquisitions in which the bidder s beta exceeds that of the target. 1 Corresponding Author: Philipp Krüger. philipp.krueger@unige.ch, Telephone: +41 (0) Augustin Landier, augustin.landier@tse-fr.eu, Telephone: +33 (0) David Thesmar, thesmar@hec.fr, Telephone: +33 (0) We thank Boris Vallée for excellent research assistance.

2 Ever since the seminal contribution of Modigliani and Miller (1958), a key result of corporate finance theory is that a project s cash-flows should be discounted at a rate that reflects the project s risk characteristics. Discounting cash flows at the firm s weighted average cost of capital (WACC) is therefore inappropriate if the project differs in terms of its riskiness from the rest of the firm s assets. In stark contrast, however, survey evidence suggests that performing capital-budgeting using a unique firm-level WACC is quite common. Graham and Harvey (2001) show that a large majority of firms report using a firm-wide discount rate to value a project independently of its risk characteristics. Similarly, Bierman (1993) surveys the top 100 firms of the Fortune 500 and finds that 93% of the responding firms use their firm-wide WACC to value projects and only 35% also rely on division-level discount rates. The potential distortions that firms might face if they discount projects at their firm-wide WACC are prominently underlined in standard MBA-level corporate finance text-books. Grinblatt and Titman (2002) note that the WACC of a firm is the relevant discount rate for [...] one of its projects only when the project has exactly the same risk profile as the entire firm.. Similarly, Brealey et al. (2005) explain that the weighted average formula works only for projects that are carbon copies of the rest of the firm. Such a gap between the normative formulation of the WACC method (the discount rate should be project-specific) and its implementation by practitioners (firms tend to use their firm-wide WACC for all projects) should lead to specific distortions in the investment policy of firms. This paper is an attempt to document and measure these distortions. We use business segment data to investigate whether diversified firms 1

3 rely on a firm wide WACC. In doing so, we examine whether companies are inclined to overinvest in their high-beta divisions and underinvest in their low-beta divisions. A company using a single firm-wide WACC would tend to overestimate the net present value (NPV) of a project whenever the project is riskier than the average project of the company. If companies apply the NPV principle to allocate capital across different divisions 2, they must have a tendency to overestimate the NPV of projects that are less risky than the firm average and vice versa. This, in turn, should lead to overinvestment (rep. underinvestment) in divisions that have a beta above (resp. below) the firm s average weighted beta. In line with this prediction, we show that investment in non-core divisions is robustly positively related to the difference between the cost of capital of the division and that of the most important division in the conglomerate (the core-division). We interpret these findings as evidence that firms do in fact discount investment projects from non-core divisions by relying on the core division s cost of capital. In a second step, using M&A data, we provide market-based evidence of the economic inefficiencies induced by the fallacy of evaluating projects using a unique company wide hurdle rate. We study a sample of diversifying mergers and acquisitions in order to examine whether and how differences in bidder and target asset betas affect abnormal bidder returns. We show that abnormal announcement returns of such diversifying transactions are significantly higher if the target company s asset beta falls short of that of the 2 Survey evidence of CEOs and CFOs presented in Graham et al. (2010) suggests that the NPV ranking is the predominant principle governing capital budgeting decisions. 2

4 acquirer. In other words, abnormal announcement returns are significantly higher whenever the acquirer is likely to be undervaluing the target. This finding is robust to the inclusion of a number of different control variables. Our paper is related to several streams of research in corporate finance. First, it contributes to the literature concerned with the theory and practice of capital budgeting and mergers and acquisitions. Graham and Harvey (2001) provides survey evidence regarding firms capital budgeting, capital structure and cost of capital choices. Most relevant to our study, they show that firms tend to use a firm wide risk premium instead of a project specific one when evaluating new investment projects. Relying entirely on observed firm level investment behavior, our study is the first to test the direct consequences of the finding in Graham and Harvey (2001) that few firms use project specific costs of capital. More precisely, we provide evidence that the use of a single firm wide discount rate (the WACC fallacy ) does in fact have statistically and economically significant effects on capital allocation and merger decisions. Since we make the assumption that managers do rely on the NPV criterion, the present paper is also related to Graham et al. (2010). This more recent contribution takes a forensic view on capital allocation and delegation of decision making in firms and provides strong survey evidence showing that the net present value rule is still the dominant way for allocating capital across different divisions. Secondly, our paper contributes to the growing behavioral corporate finance literature. Baker et al. (2007) propose a taxonomy organizing this literature around two sets of contributions: irrational investors vs. irrational managers. The more developed irrational investors stream assumes 3

5 that arbitrage is imperfect and that rational managers, in their corporate finance decisions, exploit market mispricing. 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 influenced by psychological biases. So far, this stream of research has mostly focused on how psychological traits such as optimism and overconfidence can have distorting effects on managerial expectations about the future and investment decisions (see Malmendier and Tate (2005, 2008) or Landier and Thesmar (2009)). By contrast, far less attention has been paid to whether and how bounded rationality and resulting rule of thumbs type of behaviour can shape corporate decisions. To the best of our knowledge, the present paper is the first to consider how a simplifying heuristic (using a single company wide discount rate) can have real effects on important corporate policies such as corporate investment and mergers and acquisitions. The reason why firms use a single discount rate might result from lack of sophistication. It is actually not obvious at first sight why the firm-level cost of capital is not the relevant discount rate for all the projects of the firm. A company that benefits from a low cost of capital might feel that financing risky projects is an arbitrage opportunity. In fact, by changing the risk of the firm s cash-flows, these projects also modify the expected rate of return that the market expects from the firm (Modigliani-Miller). We find two pieces of evidence coherent with the view that the WACC fallacy is related to managerial bounded rationality: Firms managed by younger CEOs, who are more likely to have been exposed to modern capital budgeting, are less prone to overinvest in high beta divisions. Also, when a division becomes larger, the 4

6 overinvestment pattern becomes smaller, suggesting that firms cease to use a unique discount rate when the cost of doing so is higher. this 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. A model of bounded rationality with such feature is e.g. Gabaix (2010), where agents try to avoid using too many parameters to avoid complexity costs in optimizing decisions. Finally, our paper is also related to the extensive literature on the functioning of internal capital markets (see for instance Lamont (1997); Shin and Stulz (1998)). Rajan et al. (2000); Scharfstein and Stein (2000) show that politicking within large organizations can lead to inefficient cross-subsidization between divisions: Divisions with lower investment opportunities obfuscate information about their real needs and manage to extract from management inefficiently large capital allocations at the expense of divisions with better opportunities (conglomerate socialism ). Rajan et al. (2000) show empirically that industry diversity within firms increases transfers toward divisions with below-average investment opportunities. Their proxy of investment opportunities is based on industry-level Tobin s q. Ozbas and Scharfstein (2010) show that unrelated segments of conglomerates invest less than standalone firms in high-q industries. A contribution of our paper is to show that division-level industry betas (and not simply divisions Tobin s q) 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, which we call the WACC fallacy. This bias might be related to the politicking argument, in that more complex, division-specific, discounting rules can potentially facilitate politicking, since divisions can advocate 5

7 through various arguments (strategic choice of industry categorization and beta evaluation techniques) that their discount rate should be lower: In other words, firms might use a unique discount rate precisely because they want to limit the scope for politicking by making rules simple and non-manipulable. The rest of the paper is organized as follows: The next section provides evidence on how division level investment in conglomerates is related to firm wide measures of the cost of capital. Section II presents the evidence on diversifying mergers and acquisitions. Finally, section III concludes. I Investment distortions within diversified firms. A. Does division level investment in non-core divisions depend on the assetbeta of the core-division? To obtain information on a firm s industrial diversification, we aggregate segment data from the Compustat Business Segment database between 1987 and 2007 by segment industries. Segment industries are defined by mapping the segment s primary four digit SIC code to its corresponding Fama and French (1997) industry category (FF48). For each year, we define a division as the bundle a firm s segments operating within the same FF48 industry. We define a conglomerate firm as a firm diversified across more than one FF48 industry, whereas standalone firms have their activities concentrated in a single industry. Table I shows descriptive statistics for all firm-level variables for both standalone and conglomerate firms. [Table I about here.] 6

8 Division sales are defined as the sum of sales of all segments composing the division. Whenever the sum of division sales exceeds or falls short of total firm sales (item12 from Compustat North America) by a margin of 5 % or more, we remove the firm-division-year observations from the sample. This is done in order to ensure consistency between the Compustat Segments and the Compustat North America databases and to reduce the potential noise induced by a firm s incorrect reporting of segment sales. For each conglomerate firm, we then identify the division with the highest level of sales and label it core-division. Conversely, divisions with sales lower than those of the core-division are refereed to as non-core divisions. Next, we construct a yearly industry-level measure of the cost of capital. 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 average leverage and average market value of equity observed in that industry. This procedure allows us to obtain a yearly industry-level asset beta β A i,t = E i,t E i,t + D i,t β E i,t (1), where E i,t is the mean market value of equity at fiscal year end and D i,t is mean industry book value of debt in the respective FF48 industry. At the firm level, the market value of equity is defined as (abs(item25)*item199) and the book value of debt is calculated as book assets (item6) minus common 7

9 equity (item60) and deferred taxes (item74). βi,t E is the estimated equity beta coeffcient from the moving-window regressions. In order to limit the impact of outliers, we set to mising all observations for which a firm level variable deviates from the median by more than five times the interquartile range. Note also that i indexes industries here. If diversified firms are discounting all projects at the core division s cost of capital, their willingness to invest in a non-core division belonging to industry i should depend positively on the difference between the non-core division s (βdiv,t A ) and the core division (βa DIV,t ) asset beta. If they use a weighted average cost of capital, this remains true as (βcore,t A ) is a proxy for the asset-weighted beta of the firm, (β A AV ERAGE,t ). Using (βa CORE,t ) as a determinant of the firm-wide discount rate avoids the multi-colinearity concern that arises when putting (βcore,t A ) and (βa AV ERAGE,t ) together in a regression. Thus, investment in non-core divisions should depend systematically on whether the asset risk of the respective non-core division is lower or higher than that of the core division. On the contrary, if headquarters use divisionspecific hurdle rates, the difference in industry betas should play no role in predicting investment at the division level. [Table II about here.] We calculate division-level investment as a division s capital expenditures scaled by total division assets at the previous fiscal year end. Mirroring the calculation of division sales, we obtain division assets and division capital expenditures by summing the respective variables over all firm segments be- 8

10 longing to the same FF48 industry in a given year. In table II we summarize descriptive statistics of all variables at the non-core division level. With the aim of testing whether investment of non-core divisions depends on the difference between the core and the non-core division s cost of capital, we regress investment of divisions which do not have the highest sales in the conglomerate (i.e. non-core divisions) on the difference in industrylevel asset betas between the non-core division and the core division, that is (βdiv,t A βa CORE,t ). We also control for the classical determinants of corporate investment, essentially industry-level Tobin s q computed for standalone firms, year fixed-effects and firm-level cash-flows. [Table III about here.] The results from pooled cross-sectional regressions clustered at the firm level are reported in Table III. Column (1) establishes the basic fact by showing that non-core division investment depends positively on the spread between the non-core and the core division s industry betas. The larger the spread, the higher the investment in the respective non-core division. This is precisely what would be expected if companies discount risky projects using too low discount rates. In the regression of column (1), the identified effect could be due to an omitted variable bias, since we choose a very parsimonious specification to present the basic fact. In order to address this issue, we add the main determinants of corporate investment, i.e. industry-level Tobin s q of the core and the non-core division calculated for standalone firms (Q CORE,t and Q DIV,t ), as well as Firm Cash Flow to the division investment equation in column (2). 9

11 Following Kaplan and Zingales (1997), we approximate Tobin s q at the firm level by the ratio of the market value of assets to its book value. Market value of assets is calculated as the book value of assets (item 6) plus the market value of common equity (item199*abs(item25)) at fiscal year end minus the book value of common equity (item60) and minus balance sheet deferred taxes (item74). Industry level Tobin s q is calculated as the average Tobin s q in a given FF48 industry-year. We choose industry Tobin s q in order to control for the quality of a division s investment opportunities because Tobin s q cannot be calculated at the division level due to the absence of a market price for a division s assets. Montgomery and Wernerfelt (1988) show, however, that industry-level Tobin s q does in fact explain a large fraction of the variation in firm level Tobin s q. Their evidence justifies industry-level Tobin s q as a reasonable proxy for the investment opportunities faced by a division. Firm Cash Flow is calculated as the sum of income before extraordinary items (item18) and depreciation and amortization (item14) scaled by total assets (item6). We choose to control for the investment opportunities of both the core and the non-core division in order to address the concern that asset betas are merely reflecting differences in valuation or investment opportunities between core and non-core divisions. After controlling for the main determinants of investment and potential differences in valuation, there is little economic justification for neither the spread in asset betas nor βcore,t A and β A DIV,t individually to have any explanatory power for non-core division investment. Yet, the coefficient estimate for the spread remains highly statistically significant. 10

12 Columns (3) and (4) add other potential determinants of division level investment, i.e. Division Size, Firm Size, Firm Age and a measure of Firm Focus to the specification. Firm Focus is defined as the core division s sales divided by total sales. It measures the extent to which the firm is concentrated within its core activity. Our results remain unchanged after the inclusion of these additional control variables. In unreported regressions, we also replace industry-level Tobin s q by a division level measure of investment opportunities, i.e. lagged sales growth. Again, our results remain robust in this alternative specification. In untabulated analysis, we also control for non-core division-industry fixed effects, which does not influence our results either. In column (5) we regress non-core division investment on the two elements of the spread. The results show a negative sign for the coefficient estimate for βcore,t A and a positive sign for βa DIV,t. This suggests that whenever the company has a low risk core activity (low βcore,t A ), and therefore a low hurdle rate, it is inclined to invest more strongly in non-core divisions with a higher asset risk. The fact that β A CORE,t is significant provides strong evidence that diversified companies look at divisions belonging to industries different from their core activity with the eyes of their core industry s characteristics. In terms of magnitude, the investment distortion we document is quite important. If βdiv,t A βa CORE,t = 0.5, meaning that the gap in discount rates between the division and the core-division is equal to half the market riskpremium (say 3.5%), the non-core division s investment rate would be (0.0155*0.5) percentage points higher. This is a relatively large effect taking into consideration that the median non-core division investment rate is about 11

13 4 % in our data (see table II). We also find (in unreported regressions) that the effect of the spread is stronger for highly levered companies: financial constraints seem to make companies more prone to block investment in low beta divisions. One interpretation is that financial constraints increase the hurdle rate up to a level where low beta divisions have very few projects that pass the bar. In another unreported set of regressions, we sort on the size of the company (measured by the log of assets or total enterprise value) and, consistent with evidence in Graham and Harvey (2001) that large firms are more likely to use the net present value rule in allocating capital, the impact of both β A CORE,t and the spread in asset betas turns out to be stronger for large firms. Thus, the bias we document cannot be seen as a small firm effect. Interestingly, we also find that non-core division investment in small firm is not related to the core division s beta. This is, at least to a certain extent, reconcilable with survey evidence in Graham et al. (2010), suggesting that small firms are much more likely to rely on managerial gut feeling in making capital allocation decisions. Hence, it might not be surprising, that an empirical analysis focusing on present value style allocation of capital is unable to dissect significant effects for small firms. [Figure 1 about here.] Next, we document that the effect is stronger in non-core divisions which are relatively less important with respect to the core division. To do so, we calculate the Relative Importance of a non-core division by scaling its 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 12

14 conglomerate. By contrast, values close to zero indicate that the non-core division is negligible vis à vis the core division. Figure 1 shows a histogram of Relative Importance. We then sort non-core division investment by Relative Importance and group the observations into four quartiles. Low quartiles indicate low importance of the division with respect to the core division, whereas higher quartiles indicate that the respective non-core division is almost as important as the conglomerate s core division. [Table IV about here.] Table IV estimates specification 4 from table III for the four quartiles of relative importance. The regression results reveal that the biasing effect of using the core division s asset beta is strongest for relatively unimportant divisions (column (1) and (2)) and vanishes for those that are almost as important as the core-division (column (4)). The finding that the spread (i.e. βdiv,t A βa CORE,t ) does not play a role in explaining the investment of non-core divisions with high organizational importance (fourth quartile of Relative Importance) is particularly interesting, since it suggests that capital allocation in more important non-core divisions is determined by more traditional factors such as investment opportunities and cash flows. This idea is further substantiated by observing that a traditional determinant of investment, i.e. the non-core division s industry-level Tobin s q, becomes significant in explaining division investment for the fourth quartile, while it is not significant in explaining non-core division investment for the first and second quartiles. Taken together, this evidence suggests that the investment distortion induced 13

15 by boundedly rational managers relying on inappropriate discount rates is concentrated within divisions of low organizational importance. [Table V about here.] Table V splits the sample into four sub periods (i.e , , and ). The analysis by sub-period indicates that the investment distortion has been strongest between 1987 and 1996 (See columns (1) and (2)). Interestingly, 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 coefficients remain statistically significant also in sub-periods following the change in regulation. The magnitude of the effect we document has, however, declined recently (see column (4)). This evidence is consistent with an interpretation that capital allocation within conglomerate firms has become more elaborate over time. A potential explanation could be that financial knowledge of corporate decision makers in charge of making capital budgeting decisions has improved over time. Higher financial sophistication of managers due to MBA style education could thus have improved the quality of capital allocation decisions within conglomerate firms. B. Is the investment distortion related to observable characteristics of the executive suite? In order to give more flesh to our interpretation stressing financial education of managers, we estimate the baseline non-core division level investment regression from the previous tables on a subsample spanning from 1992 to

16 for which we observe the chief executive s (CEO) age. We expect the CEO s age to be a reasonable proxy for the age of the overall executive suite. We are able to obtain the CEO s age from the Compustat Excecucomp database for 5,868 division-firm-years. In a similar vein to the investment regressions conditioned on Relative Importance of non-core divisions, we sort on terciles of CEO Age, and estimate the specification for all three terciles. The average age of young CEOs is about 49 years (first tercile), for medium aged CEOs about 58 years and for the oldest about 69 years (third tercile). [Table VI about here] We choose to bunch observations by terciles here because of the smaller sample size. Interestingly, for relatively young CEOs (column (1)), the spread in asset betas is not significantly related to division-level investment. This finding suggests that firms in which younger CEOs and consequently younger executives (potentially holding MBA degrees) are in charge, might be less prone to the WACC fallacy. By contrast, the coefficient estimate for β A DIV,t βa CORE,t is the most statistically significant for the tercile of oldest CEOs (see column (3)). The idea that mature CEOs might not use the most adequate capital allocation tools is also eluded in Graham and Harvey (2001). The authors provide evidence that mature CEOs, in contrast to younger ones, are more likely to rely on payback methods in allocating capital. In a similar vein, our evidence is also somewhat in line with evidence in Bertrand and Schoar (2003) showing that capital expenditures in firms run by CEOs with MBA style education are on average less responsive to cash flows and more responsive to growth opportunities as embodied in Tobin s 15

17 q. Assuming that age is a reasonable proxy for MBA style education, this is precisely what we find: Firms with young CEOs are neither subject to the WACC fallacy, nor is investment strongly related to cash flow. In fact, non-core division investment turns out to be the most conform with textbook recommendations in that it is the most sensitive to investment opportunities (Q DIV,t ). By contrast, non-core divisions in companies run by old CEOs tend to be very cash flow sensitive and also negatively related to Tobin s q, suggesting that capital allocation decisions in these firms are taken differently to what would be recommended by corporate finance textbooks. [Table VII about here] In table VII we test whether these differences are statistically significant by coding two interaction terms between dummies indicating whether the CEO belongs to the second or third tercile of CEO Age. Even though the spread turns out to be significant when non-core division investment is sorted on CEO Age, the interaction terms between the binary variables for medium and high CEO age and the spread in asset betas are no longer significant. Baker et al. (2007) note that in order for irrational managers to have an impact on corporate policies, corporate governance should be somewhat limited. Analogous to the assumption of limits to arbitrage in the behavioral finance literature concerned with irrational investors, the assumption of imperfect corporate governance is required in order for irrational managers to have a lasting impact on corporate policies. In order to measure the likelihood of imperfect corporate governance, we code a binary variable indicating low CEO ownership. Low CEO ownership firms are those for which 16

18 the CEO owns less than 1% of the common equity. Low CEO ownership firms are more likely to suffer from moral hazard problems and thus to deviate from optimal investment rules. We test whether there are significant differences in terms of firms being WACC fallacious conditional on corporate governance by interacting the spread in asset betas with the High Ownership Dummy. Consistent with the argument in Baker et al. (2007), column (2) of table VII reveals that the WACC-Fallacy is less strongly pronounced in firms in which CEOs have high share ownership (t-stat -2.65). This is also in line with evidence in Ozbas and Scharfstein (2010) showing that inefficient investment in conglomerates decreases with management equity ownership. In column (3) we measure a firm s quality of corporate governance by relying on interaction terms based on terciles of the Gompers et al. (2003) (GIM) index. However, when approximating corporate governance by the GIM index, we do not reach the same conclusions. Both the spread and the interacted spread terms for medium (second tercile) and high (third tercile) of the GIM index are not significant. C. What is the relationship between misvaluation and sales growth? In this section we demonstrate that the extent of the investment distortion resulting from using a single discount rate depends strongly on the expected sales growth of the division the project belongs to. First, we use a simple example to illustrate the basic idea. Second, we study whether the documented investment distortion is in fact related to past sales growth. Assume, in the spirit of Gordon and Shapiro (1956), that an investment project in a non-core division pays a constantly growing perpetuity C. The 17

19 growth rate of the perpetuity is denoted by g, which is assumed to be lower than the cost of capital of both the core and the non-core division and best approximated by sales growth. For notational simplicity, we drop index t. If the firm uses the correct divisional cost of capital in order to value the project, the present value of the perpetuity is given by C r f + β A DIV (r m r f ) g (2) If, however, the firm s management is subject to the WACC fallacy, the value of the perpetuity would be fallaciously taken to be C r f + β A CORE (r m r f ) g (3) The extent of misvaluation, or the misevaluation factor is thus given by r f + β A CORE (r m r f ) g r f + β A DIV (r m r f ) g (4), which can be shown to be increasing in g whenever βdiv A > βa CORE. The latter condition holds for the mean of the population (see table II). Hence, we expect the investment distortion to be increasing in the non-core division s sales growth rate. 18

20 [Figure 2 about here] In figure 2, we calibrate the overvaluation factor. The plot of the overvaluation factor as a function of the sales growth rate suggests the relationship between the non-core division investment and the spread in betas is somewhat convex in sales growth. In order to test this prediction, we use a simple proxy for sales growth, i.e. lagged industry-level sales growth. We sort noncore division investment on lagged industry-level sales growth and estimate the relationship by terciles (low, medium and high growth). The results are reported in table VIII. [Table VIII about here] Non-core division level investment is most sensitive to the spread in asset betas for high growth divisions. In fact, the coefficient estimate for β A DIV,t βa CORE,t seems to be increasingly monotonic in the terciles of lagged industry sales growth. This evidence is consistent with the simple argument based on Gordon and Shapiro (1956) and the resulting calibration. In order to test whether the relationship is in fact convex in lagged industry growth, that is significantly more pronounced for divisions belonging to high growth sectors, we code an indicator variable for each tercile of lagged industry sales growth (Low, Medium and High Growth). In column (4) of table VII, we include the spread variable and two interaction terms between the indicator variables for medium and high growth divisions and the spread in asset betas. Medium and High Growth indicate whether the division belongs to the second, respectively third tercile in terms of lagged industry sales growth. The results show that while investment of medium growth non-core divisions 19

21 are not significantly more sensitive to the spread than low growth divisions (Medium Growth (βdiv,t A βa CORE,t ) is not significant), high growth divisions turn out to be significantly more sensitive to the spread than low growth divisions (t-stat 2.02 for High Growth (βdiv,t A βa CORE,t ). This evidence underlines the idea suggested by the calibration that the investment sensitivity is in fact convex in lagged sales growth. D. Robustness Checks. Table IX reports a host of robustness checks of our main finding. First, in column (1) we replace the beta of the core division with an average firm beta βav A ERAGE,t. This firm-wide average asset beta is calculated as a weighted average division asset beta, where the weights correspond to the ratio of division to total firm assets. Should firms discount future cash flows at an average firm-wide WACC, division investment should depend significantly on the average firm-wide WACC. As expected, the coefficient estimate for β A AV ERAGE,t is significantly negative. [Table IX about here.] In column (2) of the same table we run a horse race between the average firm beta and the beta of the core division by including both measures in the equation at the same time. In this specification, the average beta of the firm is no longer significant, suggesting that the average firm-wide beta is dominated by the core-division s beta. The documented investment distortion thus seems to be driven by the use of the core-division s discount rate rather than the use of an average firm-wide WACC. In column (3) we in- 20

22 clude the gap between the division-level s and the firm s overall Tobin s q. In column (4) we control for a measure of the diversity of a firm s investment opportunities, which is essentially the standard deviation of industry-level Tobin s qs normalized by firm wide q. As expected, the more diverse the firm s investment opportunities, the lower non-core division investment. Yet, the inclusion of the diversity control and the gap between the division and the firm s Tobin s q does not affect our conclusions. In column (5) of table IX we include the core division s investment rate as a control variable. This control variable also leaves the coefficient estimates for both βcore,t A and βdiv,t A unchanged. One concern might be that industry-level asset betas capture industry specific effects. In order to address this issue, we include division industryyear fixed-effects in column (6). In this regression, the identification is purely based on the comparison between divisions in the same industry, depending on whether they belong to low beta or high beta industries. The evidence on β A CORE,t in column (6) suggests that our results are driven by cross-sectional variation in β A CORE,t rather than by high WACC or low WACC industries being treated differently in conglomerates. Finally, we control for firm fixed effects (see column (7)). After controlling for firm fixed effects, we still find evidence of an investment bias towards more risky non-core divisions (significant coefficient for β A DIV,t ). By contrast, βa CORE,t is no longer significant. The non-significance of the coefficient βcore,t A is hardly surprising since little time-series variation in β A CORE,t makes identification in fixed-effects estimations difficult to achieve. 21

23 II Efficiency effects. A. Do low industry-level beta firms engage in inefficient diversifying asset acquisitions? In this section we examine whether the fallacy of looking at investment projects belonging to other industries through the eyes of a firm s core activity does in fact generate inefficiencies in terms of firm value. In doing so, we study bidder abnormal returns for a sample of diversifying mergers and acquisitions. We define diversifying mergers and acquisitions as transactions in which a bidder gains control of an asset which belongs to a FF48 industry different from its own core activity. 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 domestic companies. The bidder s and target s core activities are identified through the SDC variables Acquiror Primary SIC Code and Target Primary SIC Code, which are matched to their corresponding FF48 industry categories. We keep only completed mergers and acquisitions (control events), 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 bidder s equity market value (calculated at the end of the fiscal 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 22

24 announcement date of the deal. Finally, we obtain balance sheet data for all bidders from the Compustat North America database. In total, we identify 6,544 of these diversifying transactions between 1988 and 2007 for which the sample selection criteria are satisfied. Descriptive statistics of bidder, target and deal characteristics as well as the temporal distribution of our merger and cquisition sample are summarized in tables X and XI. [Tables X and XI about here.] In a first step, we split the sample conditional on whether the WACC of the bidder exceeds or falls short of that of the target. Assuming that bidders apply some sort of present value method in order to determine their private valuation of the target, we expect bidders from high beta industries to be more likely to undervalue targets from low beta industries. Conversely, low industry beta bidders should tend to overvalue targets from industries subject to a higher cost of capital. The fallacy of using inadequate discount rates is thus expected to have implications for bidder abnormal returns around the announcement of diversifying acquisitions as it is likely to directly affect the bidder s private valuation of the target and hence the offer price the bidder is proposing to pay for the target s assets. Whenever bidders use too high a discount rate, they are more likely to undervalue their targets and therefore more likely to propose an offer price for the target s assets below fair value. 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 23

25 announcement from a low industry-level asset beta company for a high beta industry asset should be perceived as bad news by stock markets because the bidder is paying too much. This should lead to significantly lower abnormal returns than an otherwise equal bid from a high industry-level beta company. Taken together, we thus expect bidder abnormal returns around diversifying transactions to be significantly higher whenever the bidder s WACC is higher than that of the target. [Figure 3 about here] In figure 3, we plot the mean cumulative abnormal returns of bidders around the announcement of diversifying asset acquisitions conditional on whether the bidder s 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. As hypothesized, the mean cumulative abnormal return between event days -5 and 1 is about 100bp lower for bids involving low beta bidders and high beta targets than for transactions involving high beta bidders and low beta targets. In order to formally test whether this difference is statistically significant, 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 bidder s WACC exceeds that of the target. The results from regressions in which the abnormal return serves as the dependent variable are reported in table XII. Table XIII reports the regression results for the cumulative 24

26 abnormal return (CAR(-3,3)). [Table XII about here.] Column (1) of table XII 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 significantly lower than transactions involving high industry-level beta bidders and low industry-level beta targets. The coefficient estimate for the dummy variable indicating whether the difference between the bidder s and the target s WACC is positive turns out to be significantly positive (t-stat of 2.95). 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 affect our results. 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. Note that the average market reactions that we document are positive. This might seem to conflict with Morck et al. (1990) who provide evidence of negative bidder returns to diversifying transactions for a sample of 329 diversifying transactions. We reconcile the conflicting evidence of positive abnormal bidder returns in our sample with the previous evidence on diversifying acquisitions by noting that a large part of the sampled transactions involves private targets. Most often, the literature shows that that bidder returns to acquisitions of private targets tend to be positive (see Betton et al. (2008) or Bradley and Sundaram (2006)). [Table XIII about here.] 25

27 Column (1) of XIII confirms the main result by showing that also the CAR(3,3) is significantly higher for high beta bidder and low beta target transactions. Even though the short run announcement returns might not capture the overall value implications of the transaction, such concerns are less of a concern for the difference in announcement returns, since the CAR from both categories is likely to be affected in the same way by these imperfections. In other words, the imperfections would be differenced away. In columns (2) to (7) of the tables XII and XIII, we control for bidder, target and deal characteristics that the literature has found to significantly influence bidder abnormal returns. As such, column (2) introduces a dummy variable which takes on the value of one whenever the industry-level Tobin s q of the bidder is higher than that of the target. Column (3) controls for whether the target is a private company. Column (4) introduces a dummy variable controlling for whether the transaction is a tender offer. In column (5), we also include a measure of the relative value of the target, which we calculate as the ratio of the deal value, as disclosed by SDC, to the bidder s equity market value at the end of the fiscal year preceding the year of the transaction. Columns (5) and (6) control for the payment method by introducing measures for whether the bidder pays for the target using cash or shares only. Column (7) controls for whether the acquirer is a conglomerate firm in the sense of having activities diversified across multiple FF48 industry categories. Finally column (8) controls for the attitude of the transaction and for whether there is a challenging bid. The coefficient estimate for the dummy variable capturing differences in the cost of capital between the target and the bidder remains highly statistically significant in all regressions, with t- 26

28 statistics ranging between 1.94 and 2.73 for the CAR(-3,3) and t-statistics going as high as 3.07 for AR(0). In unreported regressions, we remove all observations for which the CAR(-3,3) exceeds the median CAR(3,3) by 5 times the interquartile range. The coefficient estimate for (βbidder,t A βa T arget,t > 0) remains virtually unchanged in the regressions in which extreme cumulative abnormal returns are removed from the analysis, suggesting that our results are not driven by statistical outliers. B. Do low beta firms engage in larger diversifying asset acquisitions? As argued in the preceding section, relying on a wrong discount rate can affect a bidder s private valuation of a target. In this subsection we examine whether a firm s cost of capital affects the scale of firms diversifying mergers and acquisitions. We hypothesize that a bidder who is relying on too low a discount rate might engage in bigger diversifying acquisitions. We construct two yearly measures capturing the size of a bidder s diversifying transactions. The first measure (Avg. Size) relates the yearly average deal value of the targets to the bidder s equity market valuation calculated in the year preceding the transaction announcement. The second measure (Sum Size) relates the sum of all deal values in a given year to the bidder s equity market valuation. We expect the bidder s WACC to be negatively related to both of our scale measures. In other words, companies relying on low discount rates are expected, on average, to not only acquire larger assets, but also more firms. [Table XIV about here.] 27

29 In table XIV, we regress both measures on the asset beta and a number of control variables. In line with the argument that using low discount rates leads to higher private valuations of the target, both the scaled average size of the deal and the scaled total sum of deal values turn out to be significantly negatively related to the firm s industry-level asset beta. Since we include only firms that have embarked on mergers and acquisitions in our regressions, these results should be interpreted conditionally. Thus, conditional on being involved in a merger or acquisition, the size of the deal tends to be larger, the lower the firm s asset beta. This is precisely what would be expected when firms tend to overestimate the benefits from acquiring assets belonging to industries different from their core activity. III Conclusion Survey evidence suggests that many firms use a firm-wide discount rate to evaluate projects (Graham et al. (2001)). The prevalence of this WACC fallacy implies that firms tend to bias investment upwards for divisions that have a higher industry beta than the firm s core division. This paper provides a direct test of this prediction using segment-level accounting data. We find a robust positive relationship between division-level investment and the spread between its industry beta and the beta of the firm s core division. Using unrelated data on mergers and acquisitions, we also find that the acquiror s stock-price reaction to the announcement of an acquisition is lower when the target has a higher beta than the acquiror. The prevalence of the WACC fallacy among corporations seems consistent with managerial 28

30 bounded rationality. It is actually not so simple to explain to a non-finance executive why it is logically flawed for a firm to discount a risky project using its own cost of capital. We find evidence that younger CEOs (more likely to have been exposed to modern corporate finance textbooks) and divisions which are relatively important in size are less subject to the WACC fallacy. This is in line with the view that the complexity of using division-dependent discount rates might come at a cost that companies trade-off against its gains (the costs of investment distortions are higher if the division is big). The costs associated with using multiple discount rates might 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 et al. (2000); Scharfstein and Stein (2000). 29

31 References Baker, M., Ruback, R., and Wurgler, J. (2007). Behavioral Corporate Finance. Handbook of corporate finance: Empirical corporate finance, 1: Berger, P. and Ofek, E. (1995). Diversification s effect on firm value. Journal of Financial Economics, 37(1): Bertrand, M. and Schoar, A. (2003). Managing with Style: The Effect of Managers on Firm Policies. Quarterly Journal of Economics, 118(4): Betton, S., Eckbo, B., and Thorburn, K. (2008). Corporate Takeovers. Handbook of Corporate Finance: Empirical Corporate Finance, 2(15): Bierman, J. H. (1993). Capital budgeting in 1992: a survey. Financial Management, 22(3):24. Bradley, M. and Sundaram, A. (2006). Acquisition and performance: A reassessment of the evidence. Working Paper, Fuqua School of Business, Duke University. Brealey, R., Myers, S., and Allen, F. (2005). Principles of Corporate Finance, McGraw-Hill, New York. McGraw-Hill, New-York, NY. Fama, E. and French, K. (1997). Industry costs of equity. Journal of Financial Economics, 43(2): Gabaix, X. (2010). A Sparsity-Based Model of Bounded Rationality. Working Paper, Stern School of Business, New York University. Gompers, P., Ishii, J., and Metrick, A. (2003). Corporate Governance and Equity Prices. Quarterly Journal of Economics, 118(1): Gordon, M. and Shapiro, E. (1956). Capital Equipment Analysis: The Required Rate of Profit. Management Science, 3(1): Graham, J. and Harvey, C. (2001). The theory and practice of corporate finance: evidence from the field. Journal of Financial Economics, 60(2-3): Graham, J., Harvey, C., and Puri, M. (2010). Capital Allocation and Delegation of Decision-Making Authority within Firms. Working Paper, Fuqua School of Business, Duke University. 30

32 Grinblatt, M. and Titman, S. (2002). Financial Markets and Corporate Strategy. McGraw-Hill, New-York, NY. Kaplan, S. and Zingales, L. (1997). Do Investment-Cash Flow Sensitivities Provide Useful Measures of Financing Constraints? Quarterly Journal of Economics, 112(1): Lamont, O. (1997). Cash Flow and Investment: Evidence from Internal Capital Markets. Journal of Finance, 52(1): Landier, A. and Thesmar, D. (2009). Financial Contracting with Optimistic Entrepreneurs. Review of Financial Studies, 22(1):117. Malmendier, U. and Tate, G. (2005). CEO Overconfidence and Corporate Investment. The Journal of Finance, 60(6): Malmendier, U. and Tate, G. (2008). Who makes acquisitions? CEO overconfidence and the market s reaction. Journal of Financial Economics, 89(1): Modigliani, F. and Miller, M. (1958). The Cost of Capital, Corporation Finance and the Theory of Investment. The American Economic Review, 48(3): Montgomery, C. and Wernerfelt, B. (1988). Diversification, Ricardian rents, and Tobin s q. The RAND Journal of Economics, 19(4): Morck, R., Shleifer, A., and Vishny, R. (1990). Do Managerial Objectives Drive Bad Acquisitions? Journal of Finance, 45(1): Ozbas, O. and Scharfstein, D. (2010). Evidence on the Dark Side of Internal Capital Markets. Review of Financial Studies, 23(2):581. Poterba, J. and Summers, L. (1995). A CEO survey of US companies time horizons and hurdle rates. Sloan Management Review, 37(1):43. Rajan, R., Servaes, H., and Zingales, L. (2000). The Cost of Diversity: The Diversification Discount and Inefficient Investment. The Journal of Finance, 55(1): Scharfstein, D. and Stein, J. (2000). The Dark Side of Internal Capital Markets: Divisional Rent-Seeking and Inefficient Investment. The Journal of Finance, 55(6): Shin, H. and Stulz, R. (1998). Are Internal Capital Markets Efficient? Quarterly Journal of Economics, 113(2):

33 A Appendix 32

34 Table I Firm-Level Descriptive Statistics This table reports means and standard deviations (in parentheses) of the employed firm-level variables. Variables based on data from Compustat and CRSP are observed for the period of 1987 to CEO related variables are observed from 1992 to 2007 only. Firm Cash Flow is the sum of income before extraordinary items (Compustat item18 ) and depreciation and amortization (item14 ) scaled by total assets (item6). Firm Size is the natural logarithm of the firm s total assets (data6). Firm Age is the logarithm of the current year plus one minus the year in which the firm first appeared in the Compustat North America database. Firm Investment is total firm wide capital expenditures (data128) scaled by total firm assets (item6). Leverage is long term debt (item9) scaled by total assets (item6). The Number of Divisions is obtained by grouping business segments by Fama and French (1997) industries and counting the number of different industries across which a firm is diversified in a given year. Sales are total firm sales (item12). Sales Growth is the firm s 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 firm wide sales. Q F IRM,t is the firm s Tobin s q, which is approximated by the firm s Market to Book ratio. Market value is of assets is calculated as the book value of assets (item 6) plus the market value of common equity at fiscal year end (item199*abs(item25)) minus the book value of common equity (item60) and balance sheet deferred taxes (item74). SD(Q DIV,t )/Q F IRM,t is the standard deviation of a firm s division-level Tobin s q s scaled by the firm wide Tobin s q. βav A ERAGE,t is the firm s asset weighted average asset beta, where the weights correspond to division-level assets. Division-level equity betas are obtained by regressing monthly FF48 portfolio returns on the CRSP Value Weighted Index for rolling windows of 60 months. Division-level asset betas are obtained by unlevering the yearly industry-level equity betas using the yearly industry average ratio of equity and total firm value. This ratio is calculated as the industry-year average ratio of market value of equity (abs(item25)*item199) to book value of debt (item6-item60-item74) and equity (abs(item25)*item199). All CEO related variables are obtained from Compustat Execucomp. CEO Ownership is equal to the first available of the Execucomp variable shrown excl opts pct and shrown excl opts/(10*abs(item25)). Standalone firms are firms with activities concentrated in a single FF48 industry, whereas Multi-Division firms are diversified across at least two different FF48 sectors. Standalone Firms Mean Median SD P25 P75 N Firm Cash Flow t Firm Size t Firm Age t Firm Investment t Leverage t Sales t Sales Growth t Q F IRM,t βav A ERAGE,t CEO Age CEO Share Ownership Conglomerate Firms Mean Median SD P25 P75 N Firm Cash Flow t Firm Size t Firm Age t Firm Investment t Leverage t Number of Divisions t Sales t Sales Growth t Firm Focus t Q F IRM,t SD(Q DIV,t )/Q F IRM,t βav A ERAGE,t CEO Age CEO Share Ownership

35 Table II Non-Core Division-Level Descriptive Statistics This table reports summary statistics of variables at the non-core division-level for the sample period of Non-core divisions are divisions that do not have the highest sales within the Conglomerate firm. Divisions are defined by grouping together segments operating in the same Fama and French (1997) industry category. All division-level variables are obtained by summing the respective data item from the Compustat Business Segment Database over all segments belonging to the same FF48 industry. Capx are division-level capital expenditures in period t + 1. Assets are division-level assets in period t. Division Size is the natural logarithm of division-level sales in period t. Q DIV,t is the industry-level Tobin s q of the division in period t. Q CORE,t is the industry-level Tobin s q of the firm s core division. Industry-level Tobin s q for both the core and the non-core divisions are calculated as the industry-year average Tobin s q for a sample of standalone firms from the same industry. We calculate Tobin s q as the ratio of the market value of assets (Book assets (item 6) plus the market value of common equity (item199*abs(item25)) at fiscal year end minus the book value of common equity (item60) and balance sheet deferred taxes (item74)) to the book value of assets (item6). βdiv,t A is the industry-level asset beta of the non-core division and βcore,t A is the industry level asset beta of the firm s core division. We construct the industry-level asset betas by regressing monthly returns of the FF48 industry portfolios on the CRSP Value Weighted market index for moving windows of 60 months. These industry-level equity betas are subsequently unlevered using the industry-year average ratio of market equity to total firm value. This ratio is calculated as the industry-year average ratio of market value of equity (abs(item25)*item199) to book value of debt (item6-item60-item74) and equity (abs(item25)*item199). Mean Median SD P25 P75 N Capx/Assets Divison Size t Q DIV,t Q CORE,t Q DIV,t Q CORE,t Q DIV,t Q F IRM,t βdiv,t E βcore,t E βdiv,t A βcore,t A βdiv,t A βa CORE,t Investment Core Division t Divison Sales Growth t Divison Industry-Level Sales Growth t Observations

36 Table III Non-Core Division-Level Investment Regressions Using business segment data from Compustat ( ), we construct industry-level divisions by aggregating segment data by Fama and French (1997) industries. A division is defined as the bundle of a firm s 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 Capx/Assets is divisionlevel capital expenditures in period t + 1 scaled by division assets in period t. βcore A is the industry-level asset beta of the core-division (i.e. the division with the highest sales). βdiv A is the industry-level asset beta of the non-core division. Q DIV,t is the division s industry-level Tobin s q. Q CORE,t is the industry-level Tobin s q of the division with the highest sales in the conglomerate. Both are calculated for a sample of standalone firms from the same industry. Firm Cash Flow is the firm s cash flow 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 firm first appeared in the Compustat North America database. Firm Focus is the ratio of the firm s core division sales to total sales. (1) (2) (3) (4) (5) Capx/Assets Capx/Assets Capx/Assets Capx/Assets Capx/Assets βdiv,t A βa CORE,t (8.15) (6.85) (6.56) (6.41) β A CORE,t β A DIV,t (-4.20) (5.24) Q DIV,t (2.43) (2.28) (2.26) (1.92) Q CORE,t (-0.78) (-1.44) (-1.38) (-1.59) Firm Cash Flow t (12.05) (11.33) (10.98) (10.99) Divison Size t (2.64) (3.16) (3.08) Firm Size t (-1.01) (-1.26) (-1.14) Firm Age t (-1.54) (-1.51) (-1.52) Firm Focus t (1.62) (1.65) Observations R All regressions include year fixed effects. t statistics in parentheses. Standard errors clustered at the firm level. p < 0.10, p < 0.05, p <

37 Figure 1. Distribution of the Relative Importance of Non-Core Divisions. Relative importance is calculated as non-core division-level sales scaled by the sales of the core division. Values close to one indicate that the non-core division is almost as important as the core division within the conglomerate, whereas values close to zero indicate that the non-core division is not important with respect to the core division. 36

38 Table IV Non-Core Division Level Investment Regressions by Relative Importance In this table, the non-core division level investment regression is conditioned on the division s organizational importance. Relative importance is calculated as non core-division sales scaled by the core division s sales. The dependent variable Capx/Assets (division-level capital expenditures in period t + 1 scaled by division assets in period t) is sorted by quartiles of Relative Importance. The first column shows the results for the first quartile of Relative Importance. Non-core divisions included in the first quartile play a subordinate role in the conglomerate. Column 4 shows the results for non-core divisions that have high organizational importance within the conglomerate firm, in the sense of having sales almost as high as the core division. βcore A is the industry-level asset beta of the core-division (i.e. the division with the highest sales). βdiv A is the industry-level asset beta of the non-core division. Q DIV,t is the division s industry-level Tobin s q. Q CORE,t is the industry-level Tobin s q of the division with the highest sales in the conglomerate firm. Both are calculated for a sample of standalone firms from the same industry. Firm Cash Flow is the firm s cash flow 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 firm first appeared in the Compustat North America database. Firm Focus is the ratio of the firm s core division sales to total sales. (Q1) (Q2) (Q3) (Q4) Capx/Assets Capx/Assets Capx/Assets Capx/Assets βdiv,t A βa CORE,t (4.11) (3.90) (3.12) (0.86) Q DIV,t (1.02) (0.62) (1.97) (2.34) Q CORE,t (-0.21) (-0.20) (-1.39) (-0.92) Firm Cash Flow t (2.08) (4.87) (5.25) (4.52) Divison Size t (1.78) (1.68) (0.66) (-1.63) Firm Size t (-0.42) (-1.15) (-0.10) (2.12) Firm Age t (-0.73) (-0.46) (0.54) (-0.43) Firm Focus t (1.26) (0.03) (0.57) (1.28) Observations R All regressions include year fixed effects. t statistics in parentheses. Standard errors clustered at the firm level. p < 0.10, p < 0.05, p <

39 Table V Non-Core Division Level Investment Regressions by Sub-Period In this table, the main investment specification is estimated for four distinct sub-periods of the sample period. Column (1) shows the relationship between non-core division investment and the asset betas for all firm-year observations between 1987 and Column (2) shows the results for the sub-period of 1992 to 1996, column (3) for and column (4) for The dependent variable Capx/Assets is division-level capital expenditures in period t + 1 scaled by division assets in period t. βcore A is the industry-level asset beta of the core-division (i.e. the division with the highest sales). βa DIV is the industry-level asset beta of the non-core division. Q DIV,t is the division s industry-level Tobin s q. Q CORE,t is the industry-level Tobin s q of the division with the highest sales in the conglomerate firm. Both are calculated for a sample of standalone firms from the same industry. Firm Cash Flow is the firm s cash flow 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 firm first appeared in the Compustat North America database. Firm Focus is the ratio of the firm s core division sales to total sales. (1) (2) (3) (4) Capx/Assets Capx/Assets Capx/Assets Capx/Assets βdiv,t A βa CORE,t (4.50) (5.10) (3.41) (2.10) Q DIV,t (2.04) (-0.12) (0.83) (1.37) Q CORE,t (-1.96) (-0.88) (0.99) (-0.62) Firm Cash Flow t (9.84) (4.57) (5.26) (4.40) Divison Size t (0.98) (3.79) (0.32) (2.32) Firm Size t (0.15) (-3.24) (0.43) (-0.31) Firm Age t (0.05) (-1.47) (-0.16) (-2.75) Firm Focus t (2.06) (0.71) (-0.12) (1.02) Observations R All regressions include year fixed effects. t statistics in parentheses. Standard errors clustered at the firm level. p < 0.10, p < 0.05, p <

40 Table VI Non-Core Division Level Investment Regressions by Terciles of CEO Age In this table, we condition the non-core division-level investment regressions on the age of the CEO. We are able to match 5,866 division-firm-year observations between with the Compustat Execucomp database. The non-core investment regressions are run by terciles of CEO Age. The first column shows the relationship for the tercile of the youngest CEOs (Average age: 49 years), the second column shows the relation for second tercile of CEO age ( 58 years) while the last column shows it for the tercile of oldest CEOs ( 66 years). The dependent variable Capx/Assets is division-level capital expenditures in period t + 1 scaled by division assets in period t. βcore A is the industry-level asset beta of the core-division (i.e. the division with the highest sales). βdiv A is the industry-level asset beta of the non-core division. Q DIV,t is the division s industry-level Tobin s q. Q CORE,t is the industry-level Tobin s q of the division with the highest sales in the conglomerate. Both are calculated for a sample of standalone firms from the same industry. Firm Cash Flow is the firm s cash flow 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 firm first appeared in the Compustat North America database. Firm Focus is the ratio of the firm s core division sales to total sales. (LOW) (MEDIUM) (HIGH) Capx/Assets Capx/Assets Capx/Assets β A DIV,t βa CORE,t (0.50) (1.79) (2.39) Q DIV,t (2.06) (0.31) (-2.57) Q CORE,t (-0.20) (-1.38) (-1.47) Firm Cash Flow t (1.87) (3.71) (3.55) Divison Size t (0.20) (0.10) (0.95) Firm Size t (-0.91) (-2.70) (-1.87) Firm Age t (0.38) (1.71) (-0.24) Firm Focus t (0.77) (-0.66) (0.54) Observations R All regressions include year fixed effects. t statistics in parentheses. Standard errors clustered at the firm level. p < 0.10, p < 0.05, p <

41 Overvaluation Factor as a function of lagged industry sales growth Overvaluation Factor Overvaluation Factor as a function of g 0 1.0% 2.0% 3.0% 4.0% 5.0% 6.0% g Figure 2. Calibrated overvaluation factor as a function of lagged industry growth. This figure plots the overvaluation factor r f +β CORE A (rm r f ) g r f +β DIV A (rm r f ) g as a function of g (lagged industry-level growth rate). In the calibration, we chose the following values for the parameters: (rm rf ) = 4%, β DIV A = 1, β CORE A = 0.75 and r f = 2%. 40

42 Table VII Interaction Effects In this table, non-core division-level investment is regressed on all variables used in the baseline specification plus four sets of interaction terms. In Column (1), Medium and High CEO Age are binary variables indicating whether the age of the CEO falls in the second or third tercile. High CEO Ownership is a dummy variable indicating whether CEO equity ownership exceeds 1%. Medium and High GIM Index are indicator variables that take on the value of one whenever the company s Gompers et al. (2003) Index falls in the second, respectively third tercile. Medium and High Sales Growth indicate whether lagged industry level sales growth falls in the second, respectively, third tercile. (1) (2) (3) (4) Capx/Assets Capx/Assets Capx/Assets Capx/Assets β A DIV,t βa CORE,t (1.22) (3.58) (1.48) (3.85) Medium CEO Age (β A DIV,t βa CORE,t ) (0.23) High CEO Age (β A DIV,t βa CORE,t ) (0.16) High CEO Ownership (β A DIV,t βa CORE,t ) (-2.65) Medium GIM Index (β A DIV,t βa CORE,t ) (0.62) High GIM Index (β A DIV,t βa CORE,t ) (1.30) Medium Sales Growth (β A DIV,t βa CORE,t ) (0.60) High Sales Growth (β A DIV,t βa CORE,t ) (2.02) Q DIV,t (0.39) (0.30) (-0.42) (2.19) Q CORE,t (-1.30) (-1.10) (-1.15) (-1.26) Firm Cash Flow t (4.28) (4.06) (6.66) (11.14) Divison Size t (0.86) (0.72) (0.76) (3.41) Firm Size t (-2.92) (-2.79) (-2.34) (-1.61) Firm Focus t (0.44) (0.16) (0.42) (1.37) Firm Age t (0.87) (0.54) (0.43) (-1.41) Observations R All regressions include year fixed effects. t statistics in parentheses. Standard errors clustered at the firm level. p < 0.10, p < 0.05, p <

43 Table VIII Non-Core Division Level Investment Regressions By Lagged Industry-Level Sales Growth In this table, non-core division-level investment is sorted on lagged industry level sales growth. Lagged Industry Sales Growth is the average sales growth observed in the FF48 industry in a given year. The first column shows the non-core division level investment sensitivity for non-core divisions belonging to low growth sectors. Column (3) shows the investment sensitivity with respect to the spread in asset betas for divisions belonging to high growth sectors. (LOW) (MEDIUM) (HIGH) Capx/Assets Capx/Assets Capx/Assets β A DIV,t βa CORE,t (3.63) (3.76) (5.88) Q DIV,t (-0.34) (1.17) (2.61) Q CORE,t (-1.53) (-2.27) (0.56) Firm Cash Flow t (6.75) (7.39) (7.19) Divison Size t (0.87) (2.71) (3.50) Firm Size t (1.12) (-1.42) (-2.97) Firm Age t (-1.99) (-0.40) (-0.84) Firm Focus t (1.09) (2.80) (-0.20) Observations R All regressions include year fixed effects. t statistics in parentheses. Standard errors clustered at the firm level. p < 0.10, p < 0.05, p <

44 Table IX Non-Core Division-Level Investment Regressions: Robustness Checks This table shows robustness checks on the main specification of non-core division-level investment regressions used in the previous tables. β CORE,t A industry-level asset beta of the core-division (i.e. the division with largest sales). β DIV,t A is the industry-level asset beta of the non-core division. βa AV ERAGE,t is a firm 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 firm assets. QDIV,t and QCORE,t are the industry level Tobin s q of the non-core and core divisions respectively. QDIV,t QFIRM,t is the gap between the industry-level Tobin s q of the division and the firm-wide Tobin s q. SD(QDIV,t)/QF IRM,t is the standard deviation of firm s division-level Tobin s q s in a given year scaled by the overall Tobin s q of the firm. This variable measures the diversity of a a conglomerate s investment opportunities. Investment Core Division is calculated as Capx of the core division in period t+1 scaled by the core division s assets in period t. All other variables are as previously defined. is the (1) (2) (3) (4) (5) (6) (7) Capx/Assets Capx/Assets Capx/Assets Capx/Assets Capx/Assets Capx/Assets Capx/Assets β CORE,t A (-3.49) (-4.07) (-4.59) (-4.45) (-2.26) (0.41) β AV A ERAGE,t (-2.64) (1.16) β DIV,t A (6.20) (4.64) (4.71) (4.84) (5.61) (4.25) QDIV,t QF IRM,t (-5.66) SD(QDIV,t)/QF IRM,t (-3.27) Investment Core Divisiont (6.67) (6.93) (6.91) (0.60) QDIV,t (1.84) (1.82) (3.68) (2.37) (2.13) (2.68) QCORE,t (-2.10) (-1.22) (-2.92) (-1.54) (-1.90) (-1.42) (-0.12) Firm Cash Flowt (12.28) (12.20) (9.96) (9.48) (10.49) (9.81) (7.46) Divison Sizet (2.69) (2.93) (3.70) (3.22) (3.12) Firm Sizet (-0.90) (-1.17) (-1.77) (-1.21) (-4.17) Firm Focust (-0.41) (-0.40) (0.93) (0.85) (1.14) (1.46) (3.52) Firm Aget (-0.80) (-0.73) (-1.72) (-1.65) (-1.22) (-1.04) (0.37) Observations R Year Fixed Effects YES YES YES YES YES NO YES Division Industry*Year Fixed Effects NO NO NO NO NO YES NO Firm Fixed Effects NO NO NO NO NO NO YES t statistics in parentheses p < 0.10, p < 0.05, p <

45 Table X Descriptive Statistics of Deal, Bidder and Target Characteristics This table shows descriptive statistics of deal, bidder and target characteristics of our sample of 6,544 diversifying transactions. A diversifying transaction is a completed merger or acquisition in which the bidder has successfully sought control of a target not belonging to its own FF48 industry. βbidder A and βa T ARGET denote the industry-level asset beta of the bidder and target respectively. Q BIDDER and Q T ARGET denote the industry-level Tobin s q ratios of the bidder and the target. (βbidder A βa T ARGET > 0) and (Q BIDDER Q T ARGET > 0) are two dummy variables indicating whether the differences in asset betas and Tobin s q s respectively are positive. Assets BIDDER is the log of the book value of the bidder s assets (item6) in the year prior to the announcement of the bid. E BIDDER is the fiscal year end equity market value of the bidder in the year prior to the bid announcement. V T ARGET is the value of the transaction as disclosed by SDC. arget private? is a dummy variable indicating whether the target is a private company and Tender Offer? indicates whether the bidder sought control through the process of a tender offer. All Cash? is a dummy variable that takes on the value one when the consideration was entirely paid in cash whereas All Equity? takes on the value of one whenever the target s shareholders are entirely compensated with shares. Multi-Division Acquirer? takes on the value of one when the bidder s business activities are diversified across at least two multiple FF48 industry categories. Hostile and Challenging Bid? are two dummy variables indicating whether the bid s nature is hostile and whether there has been a challenging bid. Mean Median SD P75 p75 βbidder A βt A ARGET βbidder A βa T ARGET (βbidder A βa T ARGET > 0) Q BIDDER Q T ARGET Q BIDDER Q T ARGET (Q BIDDER Q T ARGET > 0) E BIDDER V T ARGET V T ARGET /E BIDDER Target private? Tender Offer? All Cash? All Equity? Multi-Division Acquirer? Hostile Bid? Challenging Bid? Observations

46 Table XI Diversifying Transactions Sample by Calendar Year This table shows the temporal distribution of the sample of diversifying mergers and acquisitions. Yearly means and standard deviations of the nominal deal value (In Million US-$) are calculated by relying on the deal value as disclosed by SDC. Year # of Acquisitions Mean Value (Million US-$) SD Total

47 Figure 3. Bidder Cumulative Abnormal Returns (All Acquisitions). Mean 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 firm s Fama and French (1997) industry differs from that of the target are considered. Abnormal returns are market adjusted and calculated as the difference between the acquiring firm s daily stock return and the CRSP Value Weighted Market Return on the respective event day. All transactions between 1988 and 2007 fulfilling the sample construction conditions are considered (N=6,544). 46