Uncertainty about average profitability and the diversification discount

ARTICLE IN PRESS Journal of Financial Economics 96 (2010) 463–484

Contents lists available at ScienceDirect

Journal of Financial Economics journal homepage: www.elsevier.com/locate/jfec

Uncertainty about average profitability and the diversification discount$, $$ John Hund a, Donald Monk b,c, Sheri Tice c, a b c

Jones Graduate School of Business, Rice University, Houston, TX, USA Securities and Exchange Commission, Washington, DC, USA A.B. Freeman School of Business, Tulane University, 7 McAlister Drive, New Orleans, LA 70118, USA

a r t i c l e in fo

abstract

Article history: Received 11 April 2008 Received in revised form 13 January 2009 Accepted 30 April 2009 Available online 11 February 2010

The diversification discount (multiple segment firm value below the value imputed using single segment firm multiples) is commonly thought to be generated by agency problems, a lack of transparency, or lackluster future prospects for diversified firms. If multiple segment firms have lower uncertainty about mean profitability than single segment firms, rational learning about mean profitability provides an alternative explanation for the diversification discount that does not rely on suboptimal managerial decisions or a poor firm outlook. Empirical tests which examine changes in firm value across the business cycle and idiosyncratic volatility are consistent with lower uncertainty about mean profitability for multiple segment firms. & 2010 Elsevier B.V. All rights reserved.

JEL classification: G10 G30 G32 Keywords: Diversification discount Rational learning models Internal capital markets

1. Introduction

$ An earlier version was distributed with the title, ‘‘Rational Learning and the Diversification Discount’’. $$ The SEC disclaims responsibility for any private publication or statement of any SEC employee or Commissioner. This study expresses the authors’ views and does not necessarily reflect those of the Commission, the Commissioners, or other members of the staff. We thank the following individuals for helpful comments: Seoungpil Ahn, Santiago Bazdresch, Greg Brown, Valentin Dimitrov, Charles Hadlock, ˇ Naveen Khanna, Vojislav Maksimovic (AFA Discussant), Luboˇ s Pa´stor, Ramana Sonti, Paul Spindt, and James Weston. Also, we appreciate the feedback received from participants at the 2007 Singapore Conference on Finance, the 2007 FMA Meeting, the 2008 WAFA Conference, and the 2009 AFA Conference; seminar participants at Rice University, Tulane University, University of Minnesota, and Michigan State University; and brown bag seminar participants at UNC-Chapel Hill.  Corresponding author. Tel.: + 1 504 865 5469; fax: + 1 504 862 8327. E-mail addresses: [email protected] (J. Hund), [email protected] (D. Monk), [email protected] (S. Tice).

0304-405X/$ - see front matter & 2010 Elsevier B.V. All rights reserved. doi:10.1016/j.jfineco.2010.02.006

Traditional explanations for the diversification discount (multiple segment firm value is less than the imputed value using single segment firm multiples) rely on agency problems, a lack of transparency, or lackluster future prospects for diversified firms. Rational learning about future average long-term profitability provides an alternative explanation for the diversification discount that does not rely on suboptimal managerial decisions or a poor firm outlook. If diversified firms have less uncertainty about future mean profitability, we predict the following: (1) In the cross-section, diversified firms will trade at a discount relative to single segment firms due to convexity of the discounting function. (2) As firms age, the sales or assets multiples of single segment firms will drop more than the sales or assets multiples of diversified firms as more uncertainty about mean profitability will be resolved for single segment firms than for diversified firms. (3) The difference in the value change

ARTICLE IN PRESS 464

J. Hund et al. / Journal of Financial Economics 96 (2010) 463–484

across time for single segment and diversified firms will be larger during economic booms (when the equity risk premium is low) and smaller during economic recessions (when the equity risk premium is high). (4) After controlling for volatility in profitability, diversified firms will have lower idiosyncratic return volatility than single segment firms due to the idiosyncratic nature of learning. We confirm these predictions in a sample comparable to the previous literature (1978–1997) and in an expanded sample from 1978 to 2005. Though more agency problems, more asymmetric information, and weaker future prospects for diversified firms generate a predicted diversification discount, these explanations do not generate the same dynamic or volatility predictions as the rational learning model. The empirical findings are consistent with rational learning and lower uncertainty about mean profitability for diversified firms as an explanation for the diversification discount. In the rational learning model developed by Pa´stor and Veronesi (2003) mature firms have lower uncertainty about average profitability which leads to lower crosssectional market-to-book ratios, but not higher returns due to the idiosyncratic nature of the learning.1 As investors rationally learn about average profitability, the market value of the firm converges to its book value and market-to-book ratios change through time at a slower rate for mature firms. Mature firms also have lower idiosyncratic return volatility after controlling for the volatility of profitability which is consistent with lower uncertainty about average profitability and the idiosyncratic nature of learning. Using the intuition from Pa´stor and Veronesi (2003), we empirically examine whether lower uncertainty about average profitability for diversified firms is an explanation for the diversification discount. First, we confirm the diversification discount (a negative excess value for multiple segment firms) in our sample. As in Berger and Ofek (1995), we measure firm excess value as a log ratio of firm total capital to an imputed firm value. The imputed firm value is calculated using the median sales or assets multiple for the single segment firms in each segment.2 In our sample diversified firms have an average excess value of negative 9.7%, which is similar in magnitude to the diversification discount reported in the literature. Second, we examine the change in firm excess value over time. If diversified firms have less uncertainty about mean profitability, the drop in excess value should be larger for single segment firms than for diversified firms due to a larger resolution of growth rate uncertainty for single segment firms. Consistent with these predictions, the annual change in excess value is 3% lower for single segment firms (7% lower after controlling for endogeneity via instrumental variables).3 Our finding of a larger drop in

1 See footnote 1 in Pa´stor and Veronesi (2003) for a simple example demonstrating the effect of convexity on market-to-book ratios. 2 See Section 3.2 for more detail. 3 These results could also be explained using arguments from behavioral finance. Our finding of larger annual changes in excess value for diversified firms could be interpreted to suggest that investors are either too pessimistic about diversified firms or too optimistic about

excess value for single segment firms remains after removing firms that enter or exit the sample, addressing the wealth transfer effects between stockholders and bondholders noted in Mansi and Reeb (2002), and using a much broader sample (from 1978 to 2005) than is used in previous literature. The broader sample includes data after the release of Statement of Financial Accounting Standards (SFAS) 131 defined in Financial Accounting Standards Board (1997), a new segment reporting standard designed to increase transparency. Third, we show that the difference in changes in excess value across diversified and single segment firms co-varies with the business cycle in a predictable manner. When the equity risk premium is high future cash flows are discounted at a higher rate, and the discrepancy in uncertainty about mean profitability between multiple segment and single segment firms will have its least effect. On the other hand, when the equity risk premium is low the discrepancy will have its greatest effect. An empirical implication of this effect is that differences in the change in excess value of diversified and single segment firms will be greater during business cycle booms and lesser during contractions. In support, we show that single segment firms have a change in excess value that is 5.6% lower than diversified firms in years not surrounding recessions, but this difference is indistinguishable from zero in the period directly prior to a recession. We report a similar finding using shifts in the aggregate dividend payout ratio as a proxy for shifts in the equity risk premium. The final prediction of the rational learning model is that stocks with lower uncertainty about average profitability will have lower idiosyncratic return volatility after controlling for volatility in profitability. In support, we find that diversified firms have lower idiosyncratic return volatility than single segment firms. In the literature, the diversification discount has been ascribed to many factors. Among the most prominent of these explanations is that agency problems exacerbated by the diversified organizational form result in inefficient internal capital markets which cross-subsidize projects with lower cash flows and/or higher risks than those of their more focused competitors.4 Other prominent explanations are that agency problems cause overinvestment due to access to additional capital as in Jensen (1986, 1988), or there may be a lack of transparency due to diversified firm structure as discussed in Krishnaswami and Subramaniam (1999). We show that an assumption of constant asset returns (including as the simplest case, the constant dividend growth model) will generate zero

(footnote continued) focused firms, in the sense suggested by La Porta (1996) and Lakonishok, Shleifer, and Vishny (1994) for value and growth firms. Brav and Heaton (2002) show that predictions of learning and behavioral models are strikingly similar. However, a behavioral explanation would need to predict the firm valuation effects we document, including the business cycle dynamics and idiosyncratic return results. 4 Influential papers are Shin and Stulz (1998), Rajan, Servaes, and Zingales (2000), Scharfstein and Stein (2000), Lamont and Polk (2002), Dittmar and Shivdasani (2003), and Ahn and Denis (2004).

ARTICLE IN PRESS J. Hund et al. / Journal of Financial Economics 96 (2010) 463–484

changes in excess value if agency problems or asymmetric information are the cause of the diversification discount. Moreover, these traditional explanations do not generate the observed business cycle behavior we find or the observed differences in idiosyncratic return volatility. More recent additions to the literature question value destruction by the diversified corporate form and argue that the diversification discount is endogenous. These papers suggest that firms with poor prospects are more likely to diversify or to be acquired.5 However, these models do not generate the dynamic empirical behavior that we show either. Unless the choice of firm projects is unanticipated and yet, on average in the same direction, these explanations will lead to zero changes in excess values over time. Since the existing explanations for the diversification discount generate static predictions about cross-sectional results, most of the empirical research to date has focused on a static comparison of firm excess values at a particular point in time or has examined changes in firm values surrounding changes in organizational form.6 We extend prior research with a comparison of the dynamic performance of diversified and single segment firms over time. An examination of the change in excess value of diversified and single segment firms in general, and across the business cycle, allows us to cleanly differentiate existing explanations for the diversification discount from a rational learning explanation. A notable distinction of the rational learning paradigm is that it does not rely on the suboptimal performance of managers or on a lackluster outlook for diversified firms. Diversification is neither good nor bad if diversified firms have lower uncertainty about mean profitability. Single segment firms have a higher excess value initially, but they experience a larger drop in excess value over time. The discount may, in fact, just reflect the convexity of the discounting function and differences in uncertainty about average profitability across diversified and focused firms. The remainder of the paper is organized as follows. Section 2 discusses in more detail the implications of various explanations for the diversification discount on the predicted changes in excess value. Section 3 outlines the data and methodology used to construct the measures we use in our empirical tests. Section 4 discusses the main results of the paper generated under the base specification of the model. Section 5 contains robustness checks to address the endogeneity of diversification, entering and exiting firms, wealth transfer effects, and the effects of SFAS 131 and pseudo-conglomeration. Section 6 concludes.

2. Motivation Traditional explanations for the diversification discount predict a value loss at the time of diversification that persists until the firm refocuses. An examination of dynamic changes in the excess values of diversified and single segment (focused) firms over time rather than a static comparison allows us to differentiate between the various explanations for the diversification discount. Using standard pricing models with constant return on assets (the simplest of which is the constant dividend growth model), differences in risk or cash flows between diversified and focused firms will lead to zero changes in relative equity market-to-book (M/B) ratios over time. Thus, these models are incapable of explaining the empirical results we report. 2.1. Changes in M/B in constant ROA models Many explanations for the diversification discount center upon inefficient internal capital markets that allow profitable segments to cross-subsidize low-quality projects. For example, diversified firms may allocate capital inefficiently between divisions based on agency problems between division managers and the CEO (Rajan, Servaes, and Zingales, 2000) or based on agency conflicts between shareholders and the CEO (Scharfstein and Stein, 2000). In these models, headquarters may allocate too much capital to divisions with low expected cash flows. If diversified firms have lower expected cash flows but similar risk, value destruction will be reflected in firm value at the time of diversification. Similarly, diversified firms may overinvest in low-quality projects by selecting projects with higher risk due to enhanced access to free cash flows from other divisions (as in Jensen, 1986, 1988). Here again, value destruction will be reflected in firm value at the time of diversification as riskier investments will have a lower present value. Similarly, if diversification reduces transparency and makes firms riskier (as in Krishnaswami and Subramaniam, 1999), value destruction will be reflected at the time of diversification. Lower firm value persists until the diversified firm reverts back to a less complex (more focused) organizational form. In other words, the predicted change in excess value is zero as long as the diversified firm remains diversified. To see this in the simplest possible setting, consider the change to the equity market-to-book ratio for an allequity firm where market value Mt is simply given by the constant-growth dividend discount model Mt ¼

5 See Fluck and Lynch (1999), Campa and Kedia (2002), Graham, Lemmon, and Wolf (2002), Maksimovic and Phillips (2002), and Villalonga (2004b). 6 Lamont and Polk (2002) and Schoar (2002) are exceptions, but they examine different questions. Lamont and Polk (2002) examine the impact of exogenous changes in within-firm diversity on changes in diversified firm excess value, while Schoar (2002) examines the impact of within-firm changes in relative productivity on excess value.

465

Dt þ 1 ; rg

ð1Þ

where Dt + 1 is the dividend received in the next period, and r and g are, respectively, the expected required return and the growth rate in dividends. For times zero and one and book value Bt, this implies that   M1 M0 D2 D1 1 D2 D1 : ð2Þ  ¼  ¼  B1 B0 ðrgÞB1 ðrgÞB0 rg B1 B0 The change in the market-to-book ratio for all firms will be zero if there is a constant return on assets and a

ARTICLE IN PRESS 466

J. Hund et al. / Journal of Financial Economics 96 (2010) 463–484

constant dividend payout ratio. These two conditions will result in the book value of assets growing at the same rate as dividends, g. Note that this formulation still yields the familiar result in levels, i.e., that higher risk (meaning larger r) firms will have lower market-to-book ratios, and higher growth firms will have higher market-to-book ratios. However, changes in the market-to-book ratio (assuming the conditions above) require unanticipated changes in expected risk or unanticipated changes in growth (e.g., profitability) in order to generate non-zero values. Since the market-to-book ratio is not predicted to change for diversified or focused firms, the market-tobook ratio for diversified firms relative to focused firms will not change. This will be true even if diversified and focused firms have different levels of r or g. An earnings-based model with similar assumptions yields exactly the same results, showing that it is the stability in asset returns that drives the result. Starting with the framework in Fama and French (2006), we can write Mt as Mt ¼

1 X E½Dt þ i  i¼1

ð1þ rÞi

;

ð3Þ

which implies without a loss of generality 1 X M1 E½Di þ 1  ¼ B1 ð1 þrÞi B1 i¼1

and

1 X M0 E½Di  ¼ : B0 ð1þ rÞi B0 i¼1

ð4Þ

As long as we treat expectations as evolving unpredictably, the change in the market-to-book ratio is defined as

D

1 X M E½Di þ 1 B0 Di B1  : ¼ B ð1þ rÞi B1 B0 i¼1

ð5Þ

Then we substitute in the clean-surplus accounting relationship that the change in book value is equal in each period to the difference between earnings at time t, Yt, and dividends paid, Dt Yt Dt ¼ Bt Bt1 ;

ð6Þ

which we can also use to define the return on assets at time t, denoted by k   Yt Dt Bt1 ¼ ð1 þkÞBt1 : ð7Þ Bt ¼ 1 þ Bt1 Recursive substitution of the clean-surplus relation in Eq. (6) into the numerator of Eq. (5) gives the equivalent relation in earnings

D

1 X M E½Yi þ 1 B0 Yi B1  : ¼ B ð1 þ rÞi B1 B0 i¼1

ð8Þ

Incorporating the asset return Eq. (7) for time one gives

D

1 X M E½Yi þ 1 E½Yi ð1 þkÞ : ¼ B ð1 þrÞi B0 ð1þ kÞ i¼1

ð9Þ

Eq. (9) shows that changes in market-to-book ratios will be zero for firms with different levels of risk or profitability when expected next period earnings are equal to the post-distribution constant return on assets. Once again, we emphasize that in models without timevarying returns on assets, differences in risk or cash flows can easily explain the diversification discount, but cannot

explain the changes to excess value that we find. Indeed, within this paradigm, expected changes in the market-tobook ratio for diversified firms relative to focused firms must be zero. 2.2. Changes to M/B in the learning model If diversified firms have lower uncertainty about average profitability, the Pa´stor and Veronesi (2003) rational learning model can explain the diversification discount without relying on the suboptimal performance of managers or on a lackluster outlook for diversified firms. As time passes and investors learn about average profitability, the market value of a firm converges to its book value, and market-to-book (M/B) ratios change through time at different rates for focused and diversified firms. As a result of learning and convexity of the discounting function, diversified and focused firms converge at different rates to a similar market-to-book ratio generating higher changes in diversified firm excess values over time. It is important to note that the convergence assumption to either a stable state where M/B ratios are equal to one or to a state where returns and expected growth rates are constant (essentially where uncertainty is resolved), implies that during some period return on assets will also be evolving. However, since learning about a firm’s average excess profitability is idiosyncratic, it is not priced. Indeed, by construction in Pa´stor and Veronesi (2003), firms with differing uncertainty regarding average profitability are priced using the same stochastic discount factor. In our context, this implies that diversified firms and a portfolio of focused firms chosen to mimic them will not exhibit different expected stock returns; that is by assumption, diversification is not a priced factor.7 Some intuition on the behavior of changes in the equity market-to-book ratio can be found by examining a simplified version of the Pa´stor and Veronesi (2003) model. This simplification can be viewed as the analog to the constant dividend growth model with constant book asset growth rather than constant dividend growth. More formally, the equity growth rate is distributed Nðg ; s2 Þ. Using the ‘‘baby’’ model of Pa´stor and Veronesi (2003), a firm evolves toward a future state at time T where its market-to-book ratio is one. Since the market value at time T is BegT, its value today (at time zero) is BE[e(g  r)T]. This makes the equity market-to-book ratio of firm i   M 2 ¼ E½eðgrÞT  ¼ eðg þ ðT s =2ÞrÞT : ð10Þ B i The higher the uncertainty about book equity growth, captured by s2 , the higher the equity M/B ratio. 7 While recent research appears to confirm that diversification is not priced, this is still an open empirical question. Lamont and Polk (2001) cannot reject the hypothesis that diversified and focused firms have the same realized returns, but note that realized returns are a noisy proxy for expected returns. Recently, Mitton and Vorkink (2010) find that diversified and focused firms may have different returns to compensate for differences in skewness risk.

ARTICLE IN PRESS J. Hund et al. / Journal of Financial Economics 96 (2010) 463–484

If the uncertainty about book equity growth (profitability) for the diversified firm is less than that for the matched portfolio of focused firms,

s2d o s2f ;

ð11Þ

where s2f denotes the variance of the portfolio of focused firms, then diversified firms will trade at a discount relative to their imputed focused analogs, or     M M o ; ð12Þ B d B f which is the diversification discount in levels. Now note that both firms are evolving to the same point (M/B =1) in the same time, but the diversified firm begins below the focused analog if the diversified firm has less uncertainty about book equity growth.8 This difference creates a difference in valuation ratios (market-tobook) and their changes, but does not imply that there are arbitrage opportunities between diversified and focused firms. However, changes in diversified firms’ equity market-to-book ratio will exceed that of their focused analogs in every year until uncertainty is resolved.9 That is,     M M D 4D : ð13Þ B d B f In summary, if diversified firms have lower uncertainty about average profitability, the Pa´stor and Veronesi (2003) rational learning model can explain the diversification discount even if there is no difference in the level of risk or the level of cash flows between diversified firms and their focused analog firms. In this model, rational learning regarding heterogeneous uncertainty about average profitability generates the diversification discount. In the sections that follow, we provide evidence that diversified firms have lower uncertainty about mean profitability. We also show that our predictions for changes in excess value and for idiosyncratic return volatility based on the Pa´stor and Veronesi (2003) model are upheld. 3. Sample and variable construction For our empirical analysis, we take our baseline sample from segment- and firm-level Compustat Industrial Annual files for the period 1978–1997.10 Firm-years are 8 Note that heterogeneity in growth rate uncertainty is a crucial aspect of the model. In the full Pa´stor and Veronesi (2003) model with uncertainty regarding average profitability, learning about average profitability is fully idiosyncratic and uncertainty is eventually resolved. 9 The model assumptions necessary to generate the Pa´stor and Veronesi (2003) results for young versus old firms are the same assumptions necessary to generate our predicted results for focused versus diversified firms if diversified firms have lower uncertainty about mean profitability. They also show that their main results are robust to relaxing several of their model assumptions in Section I, Part C of their paper. 10 Due to data limitations for earlier years, we combine data from a historical source of Compustat files for 1978–1983 with a recent download from the Compustat Current and Research database for 1984–1997. A comparison of data for overlapping years between the two sources shows little difference in number of segments reported.

467

dropped from the sample according to the Berger and Ofek (1995) requirements that firms have no segments in the financial services industry (SIC 6000–6999), total firm sales is at least $20 million, and aggregated firm segment sales is within 1% of firm-level data.11 We also remove regulated utilities (SIC 4900–4941), and after the screens above are complete, firms that do not report sales or fourdigit SIC codes for all of their segments. Due to the complexities associated with using segment data after the new segment reporting rule SFAS 131 went into effect in 1998, our baseline sample stops in 1997. Using a methodology based on Berger and Hann (2003), the post-1998 data are converted to the pre-1998 format, and the main tests are run again using the entire period of 1978–2005 as a robustness check. We leave a detailed discussion of this to Section 5.

3.1. Measures of diversification We use two measures of diversification in our tests. Our first measure of diversification is the commonly used diversification indicator. A firm is considered ‘‘diversified’’ (Div=1) if it reports more than one business segment as of the firm fiscal year end. We generally take this measure at the end of year t 1 but other timing, as indicated by subscripts or table legends, is sometimes necessary. If Div does not equal one, the single-segment firm is considered ‘‘focused’’ and Div=0. This binary variable allows for direct comparison to extant literature. We also use entropy as an alternative measure of diversification. The entropy measure of diversification for firm i is determined at fiscal year end t  1 by Ent i;t1 ¼

n X s¼1

Ps;i;t1 ln

1 ; Ps;i;t1

ð14Þ

where n is the number of four-digit SIC code segments and Ps, i, t  1 is the proportion of sales from segment s of firm i at t 1. As with Div, other timing is indicated by subscripts. Entropy equals zero for firms reporting a single business segment (focused firms), and it is greater than zero for firms reporting multiple business segments (diversified firms). Entropy changes as the distribution of sales across segments changes, even if the number of segments is held constant (see Jacquemin and Berry, 1979, for more details). A change in diversification status is calculated as the first difference in Div or Ent with subscripts indicating the time of measurement. For the diversification indicator, this formulation results in values of 1 for firms focusing from multiple segments to a single segment, zero for firms not changing status, and + 1 for firms diversifying from a single segment into multiple segments. Entropy continuously increases as the degree of diversification increases. (footnote continued) Moreover, our results are robust to using only the historical file covering 1978–1996. 11 For the assets multiplier, we drop firm-year observations that have aggregated firm segment assets not within 25% of firm-level data.

ARTICLE IN PRESS 468

J. Hund et al. / Journal of Financial Economics 96 (2010) 463–484

3.2. Variable definitions Like most papers that examine or attempt to explain the diversification discount, we use excess value which is a variation of the relative equity market-to-book ratio. We calculate excess value (EV) using the standard methodology of Berger and Ofek (1995). Using a log ratio, reported total capital (market value of equity plus book value of debt) is compared to the imputed value for the firm. The imputed value is computed by multiplying the median ratio of total capital to sales (or assets) for focused firms in a segment’s industry by the segment’s reported sales (or assets) and then summing over the number of segments in the firm.12 More specifically, we use the following formulas taken verbatim from Berger and Ofek, page 60:   ! n X V IðVÞ ¼ ; ð15Þ AIi  Indi AI mf i¼1 EV ¼ lnðV=IðVÞÞ;

ð16Þ

where I(V) is the imputed value, V is the firm total capital (market value of equity at the end of the calendar year t plus book value of debt at the end of the firm fiscal year t), AI is the accounting item (sales or assets at the end of the   firm fiscal year t), Indi V=AI mf is the ratio of total capital to an accounting item for the median focused firm in the same industry as segment i, and n is the number of segments in segment i’s firm at the end of the firm fiscal year t. The matched segment median value comes from the finest SIC code level (two-, three-, or four-digit) with at least five focused firms. We use the change in excess value, defined as EVt  EVt  1, as the dependent variable in most of our regressions. Since this methodology uses book value of debt as a proxy for market value of debt, it captures changes in the market value of equity relative to the book value of the firm, but not changes in the market value of debt relative to the book value of the firm. Therefore, the excess value measure is really a measure of excess shareholder value.13 Like Berger and Ofek (1995), we use the market-tosales multiplier in addition to the market-to-assets multiplier. The assets multiplier is more consistent with the equity market-to-book framework used in Pa´stor and Veronesi (2003), but the sales multiplier is less subject to accounting manipulation. Only the sales multiplier can be used with the expanded sample due to extensive variation in how assets are allocated under the new segment reporting standard SFAS 131. We follow Pa´stor and Veronesi (2003) in using firm age as a measure of uncertainty about average profitability. In our tests we examine whether firm organizational form is 12 For the asset multiplier, a special adjustment is made to the imputed value to account for the deviation between segment- and firmlevel reporting. Specifically, the imputed value is divided by the percentage of total segment assets reported to total firm assets reported. 13 The use of this measure raises the possibility that wealth transfers between bond and stockholders distort the results. As is shown in Section 5, the results are robust to this possibility as the findings hold in a subsample of all-equity firms.

an incremental measure of uncertainty about average profitability when we control for firm age. Firm age (Age) is defined as the natural log of the number of years the firm has Compustat data starting in 1978 (the beginning of our sample). We measure volatility in a firm’s profitability process (VolP) following the method used in Pa´stor and Veronesi (2003). VolP is calculated as the residual variance from an AR(1) model using a series of at least ten years of a stock’s annual ROE for which the firm has the same organizational form.14 Mean reversion parameters are adjusted for small-sample bias using the method in Marriott and Pope (1954), and the residual estimated variance is recentered around the bias-corrected coefficients. ROE is defined as income before extraordinary items over stockholders equity and is winsorized at the 2% level in each tail of the distribution. As in Pa´stor and Veronesi (2003), VolP is used as a control variable as it is a determinant of M/B ratios and of return volatilities. Idiosyncratic volatility in profitability also slows learning about mean profitability due to a noisy profitability process.15 Our proxies for profitability are return on equity (ROE), and operating profit margin (EBIT over sales). Investment (capital expenditures over sales) is used as a proxy for future growth, while firm size (natural log of firm total assets) and financial leverage (long-term debt over total assets) are proxies for risk. 3.3. Summary statistics Table 1 reveals that our sample is similar to earlier literature with respect to commonly used variables. Since our analysis examines the annual change in excess value, we require the necessary data to calculate excess value at the beginning and the end of the year for a firm-year to be included in the sample. The following description concentrates on the statistics for the observations using the sales multiplier; however, the assets multiplier results shown on the right half of the table are almost identical. There is a diversification discount in the sample. The average excess value at t  1 (EVt  1) for 14,596 diversified (multi-segment) firm-year observations is 9.7% using the sales multiplier. Berger and Ofek (1995) report exactly the same 9.7%. Focused (single-segment) firms show a 1.4% average excess value. The differences in means (i.e., the diversification discount) and medians between diversified and focused firms are significant at the 1% level. The mean change in excess value (DEV t1 to t ) using the sales multiplier is  1.2% for diversified firms and  4.8% for focused firms. The difference in these values is significantly different from zero at the 1% level. Therefore, the average annual change in excess value is lower for focused firms than for diversified firms. If diversified firms have less uncertainty regarding their average profitability 14 The results are robust to definitions of VolP using five, 15, and 20 years of annual ROE. 15 In the Pa´stor and Veronesi (2003) model, uncertainty about mean profitability is a different concept from volatility in the profitability process.

ARTICLE IN PRESS J. Hund et al. / Journal of Financial Economics 96 (2010) 463–484

469

Table 1 Summary statistics. This table shows descriptive statistics for firm-year observations from 1978 to 1997. Firms reporting one (more than one) segment in the Compustat Industrial Annual Segment file are categorized as focused (diversified). Excess value measures (EV) are calculated using the industry multiplier approach as in Berger and Ofek (1995). DEV is the change in excess value. Age is defined as the natural log of the number of years the firm has Compustat data starting in 1978 (the beginning of our sample). VolP is the residual variance from an AR(1) model fit using a series of at least ten years of a firm’s annual ROE for which the firm has the same organizational form. ROE is earnings divided by book equity. EBIT/Sales is earnings before interest and taxes divided by sales. Size is the log of total assets. LTD/Assets is long-term debt normalized by total assets. CapEx/Sales is the ratio of capital expenditures to sales. Subscripts indicate the time of measurement. Sales multiplier

EVt  1 Focusedt  1 Diversifiedt  1 DEVt1tot Focusedt  1 Diversifiedt  1 Aget  1 Focusedt  1 Diversifiedt  1 VolP Focused Diversified ROEt  1 Focusedt  1 Diversifiedt  1 EBIT/Salest  1 Focusedt  1 Diversifiedt  1 Sizet  1 Focusedt  1 Diversifiedt  1 LTD/Assetst  1 Focusedt  1 Diversifiedt  1 CapEx/Salest  1 Focusedt  1 Diversifiedt  1

Mean

Median

0.014  0.097***

0.000  0.112***

 0.048  0.012***

Std. dev.

Assets multiplier Obs.

Mean

Median

Std. dev.

Obs.

0.545 0.517

30,462 14,596

0.025  0.092***

0.000  0.113***

0.469 0.417

31,787 11,933

 0.035  0.011***

0.372 0.340

30,462 14,596

 0.039  0.012***

 0.026  0.010***

0.334 0.279

31,787 11,933

1.245 1.405***

1.504 1.705***

1.084 1.059

30,462 14,596

1.237 1.369***

1.504 1.504***

1.087 1.066

31,787 11,933

0.138 0.118***

0.075 0.065***

0.151 0.131

15,940 8,441

0.139 0.117***

0.075 0.065***

0.151 0.130

16,482 6,782

0.064 0.077***

0.109 0.112*

0.266 0.223

30,460 14,596

0.059 0.075***

0.107 0.112*

0.271 0.221

31,785 11,933

0.078 0.082**

0.078 0.080*

0.158 0.115

29,515 14,459

0.076 0.078

0.076 0.076

0.175 0.112

30,778 11,829

4.815 5.600***

4.607 5.405***

1.495 1.787

30,462 14,596

4.793 5.549***

4.590 5.366***

1.489 1.770

31,787 11,933

0.192 0.214***

0.150 0.192***

0.191 0.175

30,462 14,596

0.191 0.222***

0.147 0.200***

0.190 0.164

31,787 11,933

0.100 0.081***

0.045 0.046**

0.212 0.152

30,223 14,551

0.103 0.087***

0.045 0.046***

0.239 0.231

31,526 11,890

Using a t-test for means and a Mann-Whitney test for medians,***, **, and * indicate a significant difference from focused firms at the 1%, 5%, and 10% levels, respectively.

(growth rate), the Pa´stor and Veronesi (2003) model predicts a diversification discount and a larger drop in excess value for focused firms. Furthermore, the observed negative average change in excess value for both diversified and focused firms is consistent with an additional prediction of the model. Firms within an industry start out with a ‘‘high’’ market-to-book ratio and then converge to a lower market-to-book ratio as an industry matures and there is less uncertainty regarding the mean asset growth rate. A larger drop in the annual change in excess value for focused firms is inconsistent with explanations for the diversification discount that predict a mean change in annual excess value (conditioned on firm organizational form) equal to zero. Diversified firms are significantly older and have a significantly lower volatility of profitability (VolP). It is not surprising that diversified firms are older since one would expect a more complex organizational form to develop over time.16 Similarly, it is not surprising that diversified firms have a lower volatility of profitability. Profitability

16

Borghesi, Houston, and Naranjo (2007) show that diversified firms are older. They show that part of the diversification discount is

in the divisions of a diversified firm should be less than perfectly correlated, resulting in a lower firm VolP. Another possible explanation is financial synergy where distressed divisions are unable to receive funding as stand-alone firms but receive funding as part of a conglomerate, smoothing profitability across time (see Fluck and Lynch, 1999). Other controls include measures of profitability, size, financial leverage, and capital expenditures. The concern that lower profitability for diversified firms is the source of the diversification discount is not supported by the data. Diversified firms have higher profitability than their focused rivals using either return on equity (ROE) or operating profit margin (EBIT over sales) as a measure of profitability. Focused firms have lower financial leverage but smaller firm size. The former implies slightly lower risk while the latter implies slightly higher risk for focused firms. Focused firms also invest more in capital expenditures. The noted differences in these variables between diversified and

(footnote continued) explained by the fact that diversified firms are older firms with lower future growth rates.

ARTICLE IN PRESS

0.042 0.004 0.057 0.047 0.041 0.040 0.003 0.030 0.047 0.044 0.014 0.030 0.095 0.021 0.055 0.001 0.007 0.024 0.024 0.057 0.177 0.006 0.097 0.090 0.116 0.111 0.034 0.173 0.070 0.122 0.051 0.000 0.067 0.016 0.249 0.011 0.214 0.144 0.008 0.019 0.014  0.256  0.033  0.090  0.132 0.029  0.009  0.086  0.074  0.195  0.084  0.039  0.061 0.056  0.129  0.110  0.029  0.025  0.009  0.003  0.037  0.037  0.017  0.031  0.039  0.018  0.036  0.012  0.002  0.057  0.057  0.064  0.027  0.043  0.057  0.061  0.025 0.022  0.075  0.032  0.044 0.024  0.016 0.027 0.032  0.011  0.016 0.021  0.094  0.046  0.038  0.015 0.012  0.037  0.038 0.084 0.149 0.086 0.137 0.011 0.045 0.133 0.048 0.156 0.056 0.129 0.096 0.033 0.048 0.102 0.187 0.116 0.183 0.100 0.021 0.054 0.062 0.005 0.056 0.073 0.083 0.105 0.124 0.140 0.049 0.186 0.201 0.002 0.004 0.015 0.107 0.158 0.011  0.016  0.004  0.028  0.043  0.041  0.035  0.042  0.047  0.020  0.062  0.021  0.022  0.064  0.067  0.065  0.055  0.041  0.067  0.073

Diversified Focused Diversified Focused Diversified

963 946 905 867 821 787 831 753 714 645 617 596 597 628 643 637 661 644 611 29 26 29 33 38 38 48 50 44 48 35 35 33 31 31 46 40 41 55 20 18 18 12 12 25 26 27 20 28 26 30 24 27 29 41 36 35 54 858 868 888 1,004 1,024 1,125 1,295 1,342 1,478 1,564 1,611 1,614 1,615 1,741 1,946 2,245 2,386 2,613 2,737 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997

Focused Focused Focused

Focused Year t

Focused

Diversified

Diversified

Focused

Diversified

Diversified

Focused

Diversified

to t )

Median (DEV t1 Year t 1

One possible explanation for the finding of a smaller drop in excess value from t  1 to t for firms that are diversified at t 1 is that diversified firms can focus, and focusing has been shown to be correlated with an increase in firm value (Comment and Jarrell, 1995). Similarly, focused firms can diversify, and diversifying has been shown to be correlated with a decrease in firm value. Traditional explanations for the change in excess value that accompanies a change in firm organizational form are based on cash flow and risk differences between diversified and focused firms caused by agency or transparency issues that persist until the firm changes organizational form again. We use a double-sort to examine whether higher changes in excess value are present after controlling for the change in excess value that occurs when firms change their organizational form. The sample observations are first sorted by whether a firm is diversified or focused at time t  1. Conditioned on whether they are diversified or focused at time t 1, the sample observations are then sorted by whether a firm is diversified or focused at time t. The following four categories of firms result from the double-sort: firms that are focused at time t 1 and remain focused at time t; firms that are focused at time t  1 and are diversified at time t; firms that are diversified at time t  1 and remain diversified at time t; and firms that are diversified at time t  1 and are focused at time t. Summary statistics for the double-sort results are shown in Table 2. The baseline sample (1978–1997) is used in this table along with the sales multiple approach to estimate excess value. The first panel in Table 2 shows that relatively few firms change their organizational status in a given year. A refocusing trend is apparent which is also shown in Comment and Jarrell (1995). Changes from the diversified to the focused corporate form occur at a higher frequency than changes from the focused to the diversified corporate form. Our finding that focused firms produce a larger drop in excess value can be seen in the double-sort results shown in the second and third panels of Table 2. Firms that are focused at time t 1 and remain focused at time t have a negative average change in excess value during every sample year. No other category of firms has a negative average change in excess value every year. Firms that are

to t )

4.1. Organizational status double-sorts

Mean (DEV t1

Our summary statistics show that diversified firms trade at a discount and have a smaller drop in excess value. These results are consistent with less uncertainty about average profitability for diversified firms based on the intuition in Pa´stor and Veronesi (2003). In this section we provide evidence that these results hold in a more robust statistical setting.

to t )

4. Empirical model and results

Count (DEV t1

focused firms and their use in prior studies are the reasons we include them as controls.

Diversified

J. Hund et al. / Journal of Financial Economics 96 (2010) 463–484 Table 2 Diversification status changes and change in excess value. This table examines changes in diversification status using a two-way sort. Observations are first sorted by status at t  1 and then sorted again by status at t. Single-segment firms are classified as focused. Multi-segment firms are considered diversified. DEV t1 to t is the change in excess value from t  1 to t, where excess value is calculated using the sales multiplier method of Berger and Ofek (1995). The first column represents year t, spanning 1979–1997.

470

ARTICLE IN PRESS J. Hund et al. / Journal of Financial Economics 96 (2010) 463–484

diversified at time t  1 and remain diversified at time t have a higher average change in excess value than focused firms that remain focused in 15 out of the 19 sample years. It is apparent that there is a value loss for focused firms that diversify and a value gain for diversified firms that focus as has been shown in other studies. However, since focused firms that remain focused have consistent negative changes in excess value, the traditional explanations for the diversification discount may not provide a complete explanation for why focused firms lose value relative to their competitors when they diversify. Even so, we note the impact of a change in form and control for these changes in diversification status in our later regression analysis.

Table 3 Change in excess value on diversification status. The following table contains regression results showing the predictive power of various diversification measures in explaining the change in excess value over the period 1978–1997. The dependent variable, change in excess value, is EVt  EVt  1, where EV is measured using the sales multiplier approach of Berger and Ofek (1995). Div is a dummy variable equal to one for multi-segment firms, and zero otherwise. Ent represents the entropy measure of diversification. Coefficient estimates are shown with t-statistics in parentheses. Subscripts indicate the time of measurement. Standard errors are adjusted for heteroskedasticity and within-year correlation. Model 1 0.035*** (4.84)

Divt  1

DDivt1tot

4.2. Baseline regression model

to t

¼ a þ b1 Divt-1 þ et1

to t :

ð17Þ

The regression output is shown in Model 1 in Table 3 for the sales multiplier approach.17 Focused firms have a change in excess value that is 3.5% lower (3.5% larger drop) than that of their diversified rivals and this result is statistically significant at the 1% level. Standard errors are adjusted for heteroskedasticity and within-year correlation. The results are also robust to including year fixed effects and adjusting the standard errors for heteroskedasticity and within-firm correlation (not shown). Agency-based theories for the diversification discount predict b1 ¼ 0, but that is not what we find in our test. Part of the relative value gain for diversified firms may be due to diversified firms refocusing or focused firms diversifying, as prior research has shown value gains from refocusing and value losses from diversifying. We control for this effect by including the change in diversification status over the year in the specification:

DEV t1

to t

Model 2

Model 3

0.038*** (4.04)

¼ a þ b1 Divt-1 þ b2 DDivt1

e

to t þ t1 to t ;

ð18Þ

where DDivt1 to t ¼ þ 1 if the firm moves from focused to diversified; zero if firm status remains the same; and ¼ 1 if the firm changes from diversified to focused from the end of year t 1 to the end of year t. The estimates of Eq. (18) are shown in Model 2 in Table 3. As expected based on prior research and our results in Table 2, b2 is negative and significant. A shift from a focused to a diversified organizational form is correlated with a drop in excess value, while a shift from a diversified to a focused form is correlated with a rise in excess value.18 However, the change in excess value remains significantly 17 Similar results using the asset multiplier approach are available upon request. 18 Firms may tend to acquire firms with low uncertainty about mean profitability when they diversify and divest segments with low uncertainty about mean profitability when they focus.

DEntt1tot Intercept Adj. R2 N

 0.048*** (  10.09) 0.002 45,058

Model 4

0.031*** (4.15)  0.071*** (  5.93)

Entt  1

In this section we examine whether our main results hold in multivariate tests. In our baseline specification we examine whether diversification status is related to the change in excess value. The specification with subscripts indicating the time of measurement is shown in Eq. (17).

DEV t1

471

 0.046*** (  9.78) 0.003 45,058

 0.046*** ( 9.08) 0.002 45,058

0.030*** (3.23)  0.163*** (-8.75)  0.044*** ( 8.88) 0.005 45,058

Coefficients significantly different from zero at the 1%, 5%, and 10% levels indicated respectively by ***, **, and *.

lower for focused firms at 3.1% (significant at the 1% level). Therefore, after controlling for shifts in firm organizational status, focused firms have significantly lower changes in excess value than their diversified rivals. In the next test, the diversification dummy is replaced with entropy, Ent. Entropy is an alternative measure of diversification. We test whether higher entropy at t 1 predicts higher or lower changes in excess value from t 1 to t. The two equations used to predict changes in excess value are

DEV t1

to t

¼ a þ b1 Ent t-1 þ et1

DEV t1

to t

¼ a þ b1 Ent t-1 þ b2 DEnt t1

to t ;

ð19Þ

e

to t þ t1 to t :

ð20Þ

The regression estimates of Eqs. (19) and (20) are shown in Models 3 and 4 in Table 3. The marginal effects suggest that a 10% increase in entropy at the beginning of the year increases the change in excess value by 0.38% (significant at 1%). After controlling for the change in firm entropy over the year, we find that a 10% increase in entropy at the beginning of the year increases the change in excess value by 0.30% (significant at 1%). The tests in this section show that focused firms have a lower change (larger drop) in excess value over time than their diversified firm matches. These results hold after controlling for changes in firm organizational form during the year. These findings are consistent with the predictions of Pa´stor and Veronesi (2003) if diversified firms have less uncertainty regarding average profitability than their focused firm matches. 4.3. Differences in profitability and risk Based on prior studies, we add additional control variables to our baseline specifications of Eqs. (17)–(20) to capture

ARTICLE IN PRESS 472

J. Hund et al. / Journal of Financial Economics 96 (2010) 463–484

Table 4 Change in excess value on diversification status using additional controls. The following table contains regression results showing the predictive power of various diversification measures in explaining the change in excess value over the period 1978–1997. The dependent variable, change in excess value, is EVt  EVt  1, where EV is measured using the sales multiplier approach of Berger and Ofek (1995). Div is a dummy variable equal to one for multi-segment firms, and zero otherwise. Ent represents the entropy measure of diversification. Age is defined as the natural log of the number of years since the firm first appears in Compustat counting from 1978 (the beginning of our sample). Size is the log of total assets. LTD/Assets is long-term debt normalized by total assets. EBIT/Sales is earnings before interest and taxes divided by sales. CapEx/Sales is the ratio of capital expenditures to sales. Subscripts indicate the time of measurement. Coefficient estimates are shown with t-statistics in parentheses. Standard errors are adjusted for heteroskedasticity and within-year correlation. Model 1 Divt  1

DDivt1tot

0.030 (4.16)

***

Model 2

Model 3

Model 4

***

0.029 (3.94)  0.102*** ( 7.61)

0.027*** (3.07) DEntt1tot  0.218*** ( 11.62) 0.003 0.004* 0.003 0.004* Sizet  1 (1.33) (1.91) (1.20) (1.98)  0.019 0.018  0.017 0.020 LTD/Assetst  1 ( 0.90) (0.88) (  0.85) (0.94) 0.038  0.007 0.038  0.008 EBIT/Salest  1 (1.72) (  0.31) (1.73) (  0.36) *** *** ***  0.121  0.058  0.120  0.058*** CapEx/Salest  1 ( 9.92) ( 3.14) ( 9.99) (  3.14) 0.197*** 0.203*** DSizet1tot (16.97) (17.71) DLTD=Assetst1tot 0.131*** 0.133*** (3.79) (3.87) 0.078*** 0.075*** DEBIT=Salest1tot (3.91) (3.77) 0.186*** 0.184*** DCapEx=Salest1tot (12.07) (11.93) Intercept  0.048***  0.084***  0.045***  0.082*** ( 3.63) ( 5.86) ( 3.37) (  5.88) 0.006 0.040 0.006 0.042 Adj. R2 N 43,700 43,201 43,700 43,201 Entt  1

0.031*** (3.58)

Coefficients significantly different from zero at the 1%, 5%, and 10% levels indicated respectively by ***, **, and *.

differences in profitability and risk across diversified and focused firms which may explain differences in firm marketto-book ratios. We include operating profit margin (EBIT over sales) as a measure of current profitability, investment (capital expenditures over sales) as a proxy for future profitability and growth, and firm size (natural log of total assets) and financial leverage (long-term debt over total assets) as proxies for risk. All variables in the levels are measured at time t1, and we include changes in these variables over year t in additional regression specifications. Table 4 shows that adding the control variables does not impact our main result of lower annual changes in excess value for focused firms. The coefficients on the diversification variables remain similar to the earlier tests and significant at the 1% level. We conclude that our results are robust to the inclusion of standard controls used in the diversification

literature. Results are similar using the assets multiplier, but are not shown for brevity.

4.4. Age and volatility in profitability As shown in Eq. (19) of Pa´stor and Veronesi (2003), firm age and idiosyncratic volatility of profitability are components of the variance of the posterior distribution of growth rate uncertainty.19 Table 1 shows that focused firms have a lower firm age (Age). Younger firms should have more uncertainty about mean profitability. This should result in a higher market-to-book ratio and a larger drop in excess value for focused firms over time. Focused firms also have higher average residual volatility in their profitability process (VolP). If higher residual volatility in profitability for focused firms is due to higher idiosyncratic volatility, learning about mean profitability will be slower for focused firms due to their noisier profitability process. To confirm our intuition that rational learning explains at least some of the diversification discount, we regress the level of the firm market-to-book ratio on the volatility of residual profitability, age, diversification status, and controls. This regression is similar to those run by Pa´stor and Veronesi (2003) but includes diversification status as an additional variable. We find that higher residual volatility, lower firm age, and focused status lead to higher firm market-to-book ratios. All of these coefficients are highly significant, even when including standard controls. These results are not shown for brevity. In Model 1 of Table 5 we use the Pa´stor and Veronesi (2003) variables of Age and VolP and omit diversification status. Consistent with Pa´stor and Veronesi (2003), we find that firm age is positively correlated with the change in excess value while residual volatility in the profitability process is negatively correlated with the change in excess value. In Model 2 we show this result weakens somewhat but remains significant after including changes in profitability and risk as control variables along with levels of profitability and risk. In Models 3 and 4 of Table 5 the diversification variable is added back to the specifications. Diversification remains significant in explaining the changes in excess value after controlling for firm age, VolP, and standard controls for leverage, profitability, and investment. Though the coefficient on firm age remains significant when the diversification status variables are included, the coefficient on the volatility in the firm profitability process (VolP) is no longer statistically significant in Model 3. 19 The posterior distribution also depends upon the (unobservable) initial prior and the mean-reversion parameter in the AR(1) process, and like Pa´stor and Veronesi, we assume no cross-sectional variation in these parameters. The mean-reversion parameter is known to be difficult to estimate in small time-series samples. In our sample, we find that both the mean and median values are 0.386, which is comparable to the mean of 0.397 found by Pa´stor and Veronesi. Moreover, the means of the mean-reversion parameter are virtually identical for diversified and focused firms.

ARTICLE IN PRESS J. Hund et al. / Journal of Financial Economics 96 (2010) 463–484

Table 5 Change in excess value on uncertainty measures and diversification status. The following table contains regression results showing the predictive power of various diversification measures in explaining the change in excess value over the period 1978–1997. The dependent variable, change in excess value, is EVt  EVt  1, where EV is measured using the sales multiplier approach of Berger and Ofek (1995). Age is defined as the natural log of the number of years since the firm first appears in Compustat counting from 1978 (the beginning of our sample). VolP is the residual variance from an AR(1) model fit using a series of at least ten years of a firm’s annual ROE for which the firm has the same organizational form. Div is a dummy variable equal to one for multisegment firms, and zero otherwise. Ent represents the entropy measure of diversification. Size is the log of total assets. LTD/Assets is long-term debt normalized by total assets. EBIT/Sales is earnings before interest and taxes divided by sales. CapEx/Sales is the ratio of capital expenditures to sales. Subscripts indicate the time of measurement. Coefficient estimates are shown with t-statistics in parentheses. Standard errors are adjusted for heteroskedasticity and within-firm correlation. Model 1

Model 2

Model 3

Table 6 Change in excess value on uncertainty measures and additional controls. The following table contains regression results showing the predictive power of various diversification measures in explaining the change in excess value over the period 1978–1997. The dependent variable, change in excess value, is EVt  EVt  1, where EV is measured using the sales multiplier approach of Berger and Ofek (1995). Age is defined as the natural log of the number of years since the firm first appears in Compustat counting from 1978 (the beginning of our sample). VolP is the residual variance from an AR(1) model fit using a series of at least ten years of a firm’s annual ROE for which the firm has the same organizational form. Dividend is an indicator dummy that equals one for firms that pay dividends, and zero otherwise. Div is a dummy variable equal to one for multi-segment firms, and zero otherwise. Size is the log of total assets. LTD/Assets is long-term debt normalized by total assets. EBIT/Sales is earnings before interest and taxes divided by sales. CapEx/Sales is the ratio of capital expenditures to sales. Subscripts indicate the time period of measurement. Coefficient estimates are shown with t-statistics in parentheses. Standard errors are adjusted for heteroskedasticity and within-firm correlation. Model 1

Model 4

0.008*** 0.016*** 0.017*** 0.017*** (4.12) (7.86) (8.40) (8.57)  0.017  0.019* VolP 0.042*** 0.020* (  3.67) ( 1.85) ( 1.60) (  1.69) 0.024*** Divt  1 (8.93)  0.081*** DDivt1tot ( 3.06) 0.023*** Entt  1 (8.03) DEntt1tot  0.204*** (  6.82) 0.001 0.000  0.002**  0.003*** Sizet  1 ( 0.58) ( 0.39) ( 2.42) (  2.77) 0.042*** 0.004 0.000 0.001 LTD/Assetst  1 (  3.52) (0.32) (0.00) (0.12) 0.016 0.030  0.032  0.033 EBIT/Salest  1 (0.77) ( 1.25) ( 1.34) (  1.38) *** *** ** 0.107 0.049  0.038  0.037** CapEx/Salest  1 (  6.42) ( 2.98) ( 2.34) (  2.31) 0.227*** 0.234*** 0.241*** DSizet1tot (14.92) (15.38) (15.73) DLTD=Assetst1tot 0.174*** 0.172*** 0.173*** (5.43) (5.35) (5.43) 0.100*** 0.097*** 0.092*** DEBIT=Salest1tot (3.23) (3.15) (2.98) 0.192*** 0.194*** 0.192*** DCapEx=Salest1tot (7.64) (7.79) (7.76) Intercept 0.008 0.055***  0.057***  0.055*** (  1.34) ( 8.95) ( 9.27) (  9.06) 0.004 0.038 0.040 0.043 Adj. R2 N 23,786 23,598 23,598 23,598 Aget  1

Coefficients significantly different from zero at the 1%, 5%, and 10% levels indicated respectively by ***, **, and *.

473

Aget  1 VolP Dividendt  1

Model 2

0.009*** (4.07)  0.005 ( 0.38) 0.006 (1.51)

0.016*** (7.26) 0.002 (0.18) 0.012*** (2.99)

 0.003*** (  2.64)  0.053*** (  4.13)  0.125*** (  3.81) 0.225*** (4.99) 0.097*** (3.17)  0.110*** (  5.46)

 0.002** ( 2.22)  0.000 ( 0.02)  0.193*** ( 4.35) 0.191*** (4.19) 0.096*** (3.39)  0.048** ( 2.32) 0.221*** (12.27) 0.180*** (4.95) 0.016 (0.41) 0.183*** (7.18)  0.063*** ( 9.15) 0.040 19,693

Divt  1

DDivt1tot Sizet  1 LTD/Assetst  1 EBIT/Salest  1 EBIT/Salest + 1 EBIT/Salest + 2 CapEx/Salest  1

DSizet1tot DLTD=Assetst1tot DEBIT=Salest1tot DCapEx=Salest1tot Intercept Adj. R2 N

 0.017** (  2.46) 0.012 19,761

Model 3

0.017*** (7.65) 0.002 (0.16) 0.009** (2.27) 0.021*** (6.60)  0.086*** ( 3.08)  0.004*** ( 3.41)  0.004 ( 0.32)  0.194*** ( 4.41) 0.191*** (4.19) 0.097*** (3.42)  0.040* ( 1.94) 0.227*** (12.70) 0.179*** (4.91) 0.014 (0.36) 0.184*** (7.28)  0.064*** ( 9.32) 0.042 19,693

Coefficients significantly different from zero at the 1%, 5%, and 10% levels indicated respectively by ***, **, and *.

Table 6 shows that focused firms continue to have lower annual changes in excess value after adding a dividend dummy and including future profitability for year t +1 and year t + 2. It is important to show that the results are robust to the inclusion of these two controls. We would expect diversified firms to have a higher probability of paying dividends since they are older firms and have less volatile cash flows. As shown in Pa´stor and Veronesi (2003), the convexity effect is weaker for firms that pay dividends. Though focused firms in our sample have lower lagged profitability, their higher firm value

may be due to higher expected future profitability. Since the significance level and coefficient on the diversification status variable remain similar with the additional controls, we conclude that the diversification discount and lower changes in excess value for focused firms are at least partially attributable to higher uncertainty about the mean profitability of focused firms.

ARTICLE IN PRESS 474

J. Hund et al. / Journal of Financial Economics 96 (2010) 463–484

4.5. Business cycle We next examine whether the diversification discount varies across the business cycle in the manner predicted by a dynamic discounting explanation. As is discussed in Pa´stor and Veronesi (2006), when the equity risk premium is low (high), firm future cash flows are discounted at a lower (higher) rate, and the convex relationship between uncertainty in average profitability and value strengthens (weakens). Ferson and Harvey (1991) examine the time-series plot of the fitted expected risk premium for NYSE listed stocks across the business cycle. They conclude that the expected risk premium increases during NBER recessions and decreases during NBER booms. Similar findings and conclusions are reached in Fama and French (1989). We use these empirical findings to test if the diversification discount widens during booms and narrows during recessions. The NBER recessions during our sample period (January 1980–July 1980, July 1981–November 1982, and July 1990–March 1991) are incorporated into the data using a dummy variable (Rec) to indicate if a firm’s fiscal year includes any month of a recession. For an added level of detail, we also create dummy variables to indicate if a firm’s fiscal year includes a month in the year prior to a recession (RecB4) and if the fiscal year includes a month in the year after a recession (RecAft). We first use a diversification dummy as our measure of diversification and estimate the effect of diversification on changes in excess value across the business cycle based on the anticipatory nature of the stock market. We add our recession indicators to Eqs. (17) and (18) to capture valuations during and surrounding a recession. We also include interactions of each of the three business cycle dummies with the diversification dummy in each specification. The regression estimates using the sales multiplier are shown in Table 7. Similar results using the assets multiplier are not shown for brevity. Using the results in Model 1 in Table 7, the marginal effect on the change in excess value from being diversified is 5.6% in the years not surrounding a recession, 0.90% in recession years, 0.30% in the year before a recession year, and 2.1% in the year following a recession. The results are similar when including change in diversification status in the regression. The diversification dummy is then replaced with the entropy measure of diversification and the business cycle regression specifications are estimated. The regression estimates are shown in Models 3 and 4 of Table 7. Using the results in Model 3 of Table 7, the marginal effect on the change in excess value from a 10% increase in entropy is 0.67% in years not surrounding a recession, 0.03% in recession years,  0.04% in the year before a recession year, and 0.29% in the year following a recession. An additional test of the business cycle dynamics of the change in excess value is provided by the relationship between the aggregate dividend-price ratio and the equity risk premium. Specifically, if the aggregate dividend–price ratio is positively correlated with the equity risk premium (as in Campbell and Shiller, 1988), then the difference between diversified and focused firm changes in excess

Table 7 Change in excess value over the business cycle—recession dummies. This table examines the effect of diversification status on change in excess value across the business cycle. The sample covers the period 1978–1997. The dependent variable, change in excess value, is EVt  EVt  1, where EV is measured using the sales multiplier approach of Berger and Ofek (1995). Div is a dummy variable equal to one for multi-segment firms, and is zero otherwise. Rec equals one for firms whose fiscal year includes any month considered a recession month by the NBER, and is zero otherwise. RecB4 equals one if a firm’s fiscal year includes a month that is in the year prior to an NBER recession period, and is zero otherwise. RecAft equals one if a firm’s fiscal year includes a month that is in the year after an NBER recession period, and is zero otherwise. Variable interactions are represented by placing an ‘‘x’’ between variable names. Subscripts indicate the time of measurement. Coefficient estimates are shown with t-statistics in parentheses. Standard errors are adjusted for heteroskedasticity and within-year correlation.

Divt  1

Model 1

Model 2

0.056*** (13.04)

0.050*** (10.77)  0.070*** ( 5.78)

DDivt1tot

Model 3

0.067*** (9.56)

Entt  1

DEntt1tot Rect RecB4t RecAftt Rect  Divt1 RecB4t  Divt1 RecAftt  Divt1

0.021* (2.02) 0.034*** (6.39)  0.007 (  1.00)  0.047*** (  3.29)  0.053*** (  5.78)  0.035*** (  3.31)

Rect  Entt1 RecB4t  Entt1 RecAftt  Entt1 Intercept Adj. R2 N

 0.054*** (  10.72) 0.003 45,058

0.021* 0.022* (2.00) (1.93) 0.034*** 0.035*** (6.39) (4.88)  0.007  0.009 ( 1.03) ( 1.15)  0.046*** ( 3.18)  0.051*** ( 5.65)  0.034*** ( 3.33)  0.064*** ( 5.32)  0.071*** ( 3.40)  0.038* ( 2.08)  0.053***  0.053*** ( 10.47) ( 9.98) 0.004 0.003 45,058 45,058

Model 4

0.057*** (7.67)  0.160*** (  8.43) 0.022* (1.92) 0.035*** (4.92)  0.009 (  1.22)

 0.060*** (  4.86)  0.067*** (  3.26)  0.035* (  1.93)  0.051*** (  9.85) 0.006 45,058

Coefficients significantly different from zero at the 1%, 5%, and 10% levels indicated respectively by ***, **, and *.

value should widen when the dividend-price ratio is low, and narrow when the dividend-price ratio is high. Results of regressions shown in Table 8 of changes in excess value on diversification status, the log of the aggregate dividend–price ratio, and their interaction confirm this prediction.20 Fig. 1 plots the slopes and intercepts from the grouped regressions implied by the interaction effect and shows the markedly different relationship of the aggregate dividend–price ratio on changes in excess value

20

We are grateful to an anonymous referee for suggesting this test.

ARTICLE IN PRESS J. Hund et al. / Journal of Financial Economics 96 (2010) 463–484

Table 8 Change in excess value over the business cycle—dividend–price ratio. The following table contains regression results showing the effect of the dividend-to-price ratio on change in excess value over the period 1978–1997. The dependent variable, change in excess value, is EVt  EVt  1, where EV is measured using the sales multiplier approach of Berger and Ofek (1995). Divt  1 is a dummy variable equal to one for multi-segment firms, and is zero otherwise. DDivt1 to t is the change in diversification status from t  1 to t. DP is the log of the nominal aggregate annual dividend-to-price ratio taken from Robert Shiller’s Web site and centered around its mean for tractable interpretation. Aggregate dividends are measured over the year t  1, and the S&P 500 price is measured at the beginning of year t. Variable interactions are represented by placing an ‘‘x’’ between variable names. Coefficient estimates are shown with t-statistics in parentheses. Standard errors are adjusted for heteroskedasticity and within-year correlation. Model 1

Divt  1 DPt  1

Model 2

0.032*** (4.61) 0.027** (2.42)

0.034*** (5.63) 0.043*** (5.33)  0.051** (  2.84)

 0.046*** ( 12.59) 0.002 45,058

 0.045*** (  12.54) 0.003 45,058

DPt1  Divt1

DDivt1tot Intercept Adj. R2 N

Model 3

0.029*** (4.70) 0.043*** (5.28)  0.048** ( 2.70)  0.071*** ( 5.99)  0.044*** ( 12.07) 0.004 45,058

Coefficients significantly different from zero at the 1%, 5%, and 10% levels indicated respectively by ***, **, and *.

when conditioned on diversification status. The difference between focused and diversified firm effects is largest when the equity risk premium is low (as proxied by the aggregate dividend–price ratio), and shrinks dramatically as the equity risk premium rises. These results are consistent with the business cycle dynamics discussed above using recession indicators, where focused firms lose value relative to their diversified matches in booms (when the equity risk premium is low), while there is little difference during recessions (when the equity risk premium is high).21 It is also apparent that focused firm changes in excess value are more sensitive to shifts in the equity risk premium. Since they have more uncertainty about mean profitability, higher sensitivity of firm value to shifts in the equity risk premium is consistent with convexity in the discounting function. Consistent with our explanation about how the values of diversified and focused firms evolve differently through time, the difference in the change in excess value between diversified firms and their focused match firms widens when the equity risk premium is low and narrows when the equity risk premium is high. We find significant results whether we use business cycle or aggregate dividend–price ratios to proxy for the expected equity risk premium. We conclude that the difference in changes

21 It is worth noting that just as with the marginal effects from the regressions in Table 7, diversified firms always gain value relative to their focused match firms over the entire range of our data; it is just that the difference becomes insignificant when the equity risk premium is high.

475

in excess value between diversified and focused firms is consistent with the dynamic predictions of the Pa´stor and Veronesi (2003) model. 4.6. Idiosyncratic return volatility A secondary implication of the Pa´stor and Veronesi (2003) model is that firm idiosyncratic return volatility increases with uncertainty about average firm profitability. If diversified firms have lower uncertainty about profitability, they should have lower idiosyncratic return volatility than focused firms after controlling for residual volatility in profitability. We use firm idiosyncratic return volatility for year t as the dependent variable. This is defined as the firm-year residual (in decimals) from a market model regression of at least ten monthly stock returns on the value-weighted CRSP index for calendar year t. Other variables (with the exception of VolP) are measured for the fiscal year end t. We present crosssectional regression results in Table 9. Model 1 is comparable to a similar model in Pa´stor and Veronesi (2003). Specifications shown in Models 2 and 3 examine the effect of diversification on idiosyncratic return volatility.22 Consistent with the findings in Pa´stor and Veronesi (2003), Model 1 shows that younger firms and firms with higher volatility in their profitability process have higher residual return volatility. Moreover, the negative and significant coefficient on the dividend dummy (Dividend) supports the notion that firms who pay dividends have lower idiosyncratic return volatility. Models 2 and 3 show that diversification measures are negatively associated with residual return volatility after including volatility of operating profitability and firm age as control variables.23 Consistent with our earlier tests, this suggests that diversification is an additional indicator of uncertainty about average profitability over and above that captured by Age. We conclude that diversified firms trade at a discount, have a smaller drop in excess value, and have lower idiosyncratic return volatility after we control for volatility in profitability. These findings are consistent with diversified firms having less uncertainty regarding average profitability. 5. Robustness tests In this section we do four robustness checks. First, we control for the possibility that diversification status and changes in excess value are endogenously correlated by using an instrumental variables approach. Second, we repeat our tests after controlling for the effect of entering 22 We make econometric adjustments according to the methods described in Petersen (2009). Specifically, we include year fixed effects and adjust standard errors for within-year and within-firm correlation. Also, we use an expanded sample (1978–2005; described in Section 5.4) since the definitions of VolP and idiosyncratic return volatility limit available observations. 23 The mean and median idiosyncratic return volatility is approximately 0.10 (or 10%) in the sample. Therefore, the marginal effects support that diversified firms have an idiosyncratic return volatility which is 0.40% lower than that of focused firms, all else held constant.

ARTICLE IN PRESS 476

J. Hund et al. / Journal of Financial Economics 96 (2010) 463–484

0%

Change in excess value

-1% -2% -3% -4% -5% -6% -7% Focused firms Diversified firms -8% 1.75% 2.25% 2.75% 3.25% 3.75% 4.25% 4.75% 5.25% 5.75% 6.25% D/P ratio Fig. 1. This figure plots the fitted change in excess value for the range values of the aggregate dividend–price ratio in the data conditional on organizational form using the coefficients from Model 3 of Table 8. D/P ratio is the log of the nominal aggregate annual dividend-price ratio taken from Robert Shiller’s Web site and centered around its mean for tractable interpretation. Aggregate dividends are measured over the year t  1, and the S&P 500 price is measured at the beginning of year t. Change in excess value, is EVt  EVt  1, where EV is measured using the sales multiplier approach of Berger and Ofek (1995).

and exiting firms. Third, we discuss the wealth transfer effects associated with diversification and show that these effects do not drive our results. Fourth, we extend our sample through 2005 using data after the introduction of SFAS 131 to see if the results are robust to an extended sample period with more transparent disclosure rules. Our results strengthen after employing instrumental variables and are not driven by entering or exiting firms, wealth effects, or a particular time period and accounting methodology. 5.1. Instrumental variables results In this section we use an instrumental variables procedure to control for the possibility that unobservables correlated with diversification status may also be correlated with changes in excess value. In other words, unobservables rather than firm organizational form may be the driving force behind the changes in excess value that we find. If so, inferences are problematic as the coefficient estimates will be biased and inconsistent.24 In our first set of instrumental variables tests we use a generated instrumental variables procedure and instruments similar to those used in Campa and Kedia (2002) and Dimitrov and Tice (2006) to control for potential endogeneity of diversification status. The empirical structure in this paper is very similar to the structure in those papers as there is an industry-adjusted dependent variable (DEV) and a potentially endogenous diversification status dummy, measured at the beginning of the observation year. The generated instrumental variables procedure we employ treats diversification status at the beginning of

Table 9 Residual volatility determinants. The dependent variable for the following regression results is residual return volatility in decimals, which is calculated as the residual variance for year t using the market model. This is done for each year between 1978 and 2005. Age is defined as the natural log of the number of years the firm has Compustat data starting in 1978 (the beginning of our sample). VolP is the residual variance from an AR(1) model fit using a series of at least ten years of a firm’s annual ROE for which the firm has the same organizational form. Div is a dummy variable equal to one for multi-segment firms, and is zero otherwise. Ent represents the entropy measure of diversification. Dividend is a dividend dummy equal to one for firms who pay dividends, and is zero otherwise. M/B is the log of the market-to-book ratio. LTD/Assets is long-term debt normalized by total assets. Size is the log of total assets. ROE is earnings divided by book equity. All control variables are measured at time t, as indicated by the subscripts. Coefficient estimates are shown with t-statistics in parentheses. Year fixed effects are included in each model. Standard errors are adjusted for heteroskedasticity and within-firm and -year correlation.

Aget VolP

Model 2

Model 3

 0.007*** (  7.76) 0.046*** (11.91)

 0.007*** (  7.41) 0.046*** (11.89)  0.004*** (  4.60)

 0.007*** ( 7.62) 0.046*** (11.97)

Divt Entt Dividendt M/Bt LTD/Assetst Sizet ROEt Intercept

24

Our instrumental variables (IV) procedure is designed to control for endogeneity of diversification due to an omitted variables bias where an omitted variable is correlated with changes in both excess value and diversification. Our IV approach is not designed to control for endogeneity due to reverse causality where changes in excess value have a causal effect on diversification.

Model 1

R2 N

 0.028*** (  17.44) 0.002 (1.53)  0.002 ( 0.34)  0.007*** (  15.41)  0.028*** (  8.74) 0.172*** (53.57) 0.356 33,749

 0.027*** (  17.00) 0.002 (1.36)  0.001 (  0.24)  0.007*** (  14.97)  0.028*** (  8.78) 0.171*** (53.60) 0.357 33,749

 0.005*** ( 4.97)  0.027*** ( 17.04) 0.002 (1.33)  0.001 (  0.28)  0.007*** ( 14.85)  0.028*** ( 8.77) 0.171*** (53.56) 0.357 33,749

Coefficients significantly different from zero at the 1%, 5%, and 10% levels indicated respectively by ***, **, and *.

ARTICLE IN PRESS J. Hund et al. / Journal of Financial Economics 96 (2010) 463–484

the fiscal year as endogenous. The procedure starts with a preliminary stage and exploits the binary nature of the endogenous variable. We first create a generated instrument for the potentially endogenous diversification status variable Divi,t  1 using instruments for diversification status. The generated instrument G^ i;t1 is the predicted probability that firm i is diversified at time t  1. Twostage least squares is then applied using the generated instrument as an instrument for diversification status. We adopt the use of a minority interest dummy as a variable to predict diversification status from Dimitrov and Tice (2006). Non-zero minority interest indicates that at some point the firm acquired a majority of the stock of another company. Since some acquisitions result in diversification, we expect the minority interest dummy to be positively correlated with the diversification status dummy. Since the acquisition that created non-zero minority interest could have been completed several years ago, it may not be correlated with contemporaneous unobservables. For example, a firm may have diversified several years ago when it had low-quality managers who happened to like to build empires. Current management, who are higher quality managers, may still be operating with the diversified form but are not destroying firm value as they are higher quality managers. As our second variable to predict diversification status we use an instrument used by both Campa and Kedia (2002) and Dimitrov and Tice (2006). PSDIV, the fraction of industry sales (measured at the two- and four-digit SIC code level) belonging to segments of diversified firms (excluding the subject firm’s segments), captures the overall attractiveness of an industry segment to diversified firms. Since PSDIV excludes the sales of segments from the subject firm, it is not likely to be correlated with unobservables that impact the change in firm i’s excess value. The second-stage instrumental variables results using the generated instrument for diversification status are shown as Models 1 and 2 in Table 10. We use additional control variables from Table 4 for comparability. Focused firms have a change in excess value that is 6.7% lower than that of their diversified rivals and this result is statistically significant at the 5% level. When all of the controls are included in the specification, focused firms have a change in excess value that is 8.7% lower than that of their focused rivals. We next control for the possibility that the entropy measure of diversification is endogenous. We use the minority interest dummy and PSDIV as instruments for entropy and employ two-stage least squares to re-estimate Eqs. (19) and (20). The second-stage two-stage least squares results are shown as Models 3 and 4 in Table 10. Model 3 shows that a 10% increase in entropy at the beginning of the year increases the change in excess value by 0.83% (significant at 1%). When all of the controls are included in the specification, a 10% increase in entropy at the beginning of the year increases the change in excess value by 1.02%. Focused firms have a lower change in excess value relative to their diversified rivals after controlling for endogeneity of diversification status at the beginning of the year. In fact, the results strengthen when we use the instrumental variables procedure to control for this type of endogeneity. If the instruments we use are valid, the

477

Table 10 Change in excess value—instrumental variables. This table reports second-stage results where a multi-segment diversification indicator (Div) and the entropy measure of diversification (Ent) are instrumented in the first stage. The dependent variable, change in excess value, is EVt  EVt  1, where EV is measured using the sales multiplier approach of Berger and Ofek (1995). The sample period is 1978–1997. Size is the log of total assets. LTD/Assets is long-term debt normalized by total assets. EBIT/Sales is earnings before interest and taxes divided by sales. CapEx/Sales is the ratio of capital expenditures to sales. Subscripts indicate the time of measurement. Coefficient estimates are shown with t-statistics in parentheses. Standard errors are adjusted for heteroskedasticity and within-year correlation. Model 1

0.067** (2.71)

Divt  1

DDivt1

to t

Model 2

0.083*** (2.94) to t

Sizet  1 LTD/Assetst  1 EBIT/Salest  1 CapEx/Salest  1

DSizet1

to t

DLTD=Assetst1

to t

DEBIT=Salest1

to t

DCapEx=Salest1 Intercept Adj. R2 N

to t

Model 4

0.087*** (3.16)  0.076*** (  3.79)

Entt  1

DEntt1

Model 3

0.000 0.000 (0.15) (0.21)  0.013 0.022 ( 0.62) (1.01) 0.047**  0.006 (2.27) (  0.25)  0.116***  0.045** ( 10.19) (  2.32) 0.204*** (16.56) 0.121*** (3.21) 0.079*** (3.51) 0.191*** (11.37)  0.050***  0.086*** ( 3.66) (  6.01) 0.004 0.036 39,259 38,789

 0.002 ( 0.66)  0.011 ( 0.51) 0.048** (2.33)  0.112*** (  9.82)

 0.040*** (  3.01) 0.003 39,259

0.102*** (3.29)  0.173*** ( 6.96)  0.002 ( 0.76) 0.025 (1.13)  0.005 ( 0.23)  0.040* ( 2.02) 0.208*** (17.02) 0.124*** (3.31) 0.076*** (3.47) 0.192*** (11.28)  0.075*** ( 5.60) 0.036 38,789

Coefficients significantly different from zero at the 1%, 5%, and 10% levels indicated respectively by ***, **, and *.

results suggest that organizational form is driving differences in uncertainty about mean profitability across diversified and focused firms rather than something correlated with firm organizational form. 5.2. No entering or exiting firms Lower annual changes in excess value for focused firms could be an artifact of firms entering and exiting the sample. Campa and Kedia (2002) show how firms entering (exiting) the sample with positive (negative) excess values can change the interpretation of the diversification discount from one of relative value of the firm to one of industry composition. Entering and exiting firms may create a problem interpreting levels of excess value as well as changes in excess value. For example, if a focused firm enters with a high firm value during year t, it may cause a downward bias in the change in excess value over year t for incumbent firms in the same industry. This is because an entering focused firm with a high firm value

ARTICLE IN PRESS 478

J. Hund et al. / Journal of Financial Economics 96 (2010) 463–484

Table 11 Results without entering and exiting firms. This table shows descriptive statistics and regression results for firms from 1978 to 1997 with entering and exiting focused firm-years removed. Focused firms with data in year t but not in year t  1 are considered ‘‘entering,’’ and focused firms with data in year t but not in year t +1 are considered ‘‘exiting.’’ All excess value calculations are made without the first year of data for entering firms and without the last year of data for exiting firms. Excess value measures (EV) are calculated using the sales industry multiplier approach as in Berger and Ofek (1995). Panel A reports summary statistics. DEV is the change in excess value. Subscripts indicate the time of measurement. Firms reporting one (more than one) segment in the Compustat Industrial Annual Segment file are categorized as focused (diversified). Panel B contains regression results showing the predictive power of various diversification measures in explaining the change in excess value. The dependent variable is change in excess value. Div is a dummy variable equal to one for multisegment firms, and zero otherwise. Ent represents the entropy measure of diversification. Coefficient estimates are shown with t-statistics in parentheses. Standard errors are adjusted for heteroskedasticity and within-year correlation. Panel A: Descriptive statistics Mean

EVt  1 Focusedt  1 Diversifiedt  1 DEVt1tot Focusedt  1 Diversifiedt  1

Median

First quartile

0.011  0.089***

0.000  0.105***

 0.333  0.448

0.346 0.259

0.536 0.520

20,726 10,496

 0.036  0.007***

 0.030  0.010***

 0.231  0.193

0.160 0.174

0.353 0.329

20,726 10,496

Panel B: Change in excess value on diversification status Model 1

Divt  1

0.029*** (3.58)

DDivt1tot

Third quartile

Model 2

Standard deviation

Model 3

0.029*** (3.14)

DEntt1tot

Adj. R2 N

 0.036*** (  7.69) 0.002 31,222

Model 4

0.024*** (3.01)  0.082*** ( 4.04)

Entt  1

Intercept

Observations

 0.035*** ( 7.39) 0.003 31,222

 0.034*** (  7.68) 0.001 31,222

0.021** (2.38)  0.177*** (  6.83)  0.033*** (  7.48) 0.005 31,222

Panel A:*** indicates a significant difference from focused firms at the 1% level using a t-test for means and a Mann-Whitney test for medians. Panel B: Coefficients significantly different from zero at the 1%, 5%, and 10% levels indicated respectively by***, **, and *.

may raise the median focused firm value for the industry at the end of year t. Since it entered in year t, its value will not be reflected in the median focused firm value at the beginning of year t.25 Furthermore, if industry composition is such that there are more focused firms than diversified firms in the industry entered by the highly valued focused firm during year t, average focused firm excess values will drop more than average diversified firm excess values over year t when the results are pooled across all industries, holding all else constant. To allay these concerns we purge our sample of entering and exiting focused firm-years and repeat our baseline tests.26 Focused firms with data in year t but not in year t 1 are considered ‘‘entering,’’ and focused firms with data in year t but not in year t + 1 are considered ‘‘exiting.’’ All excess value calculations are made without

25 Entering diversified firms do not have any impact on the median focused firm value during the year in which they enter and cannot bias excess values in their year of entry. 26 We also run a separate robustness test without the first three years of data for each firm to address the long-run underperformance of initial public offerings (see Ritter, 1991). Our results are qualitatively unchanged. Also, qualitatively and quantitatively similar results using the assets multiplier approach are available upon request.

the first year of data for entering firms and the last year of data for exiting firms. We continue our restriction of requiring at least five focused firms to construct the median focused firm value. Since there will be fewer focused firms, the distribution of median values coming from two-, three-, or four-digit SIC industries will change. If there are less than five focused firms at the four-digit SIC, the three-digit SIC is used. If there are less than five focused firms at the three-digit SIC, the two-digit SIC is used. Therefore, in this context the median focused firm value is more likely to come from a broader industry category. Results for this abridged sample are shown in Table 11. Panel A shows summary statistics that are similar to those for the full sample, although the number of observations falls. Focused firms account for approximately two-thirds of the sample, and a diversification discount of 10% is given by the significant difference (at the 1% level) between mean excess value for focused firms (0.011) and diversified firms ( 0.089). One notable benefit of this sample is that the firms contained in the sample are the same at the beginning and the end of the year, allowing the use of changes in excess value to determine that the diversification discount shrinks from 10% at t 1 to 7.1% at t. Repeating our baseline tests in Panel B reveals that the

ARTICLE IN PRESS J. Hund et al. / Journal of Financial Economics 96 (2010) 463–484

479

Table 12 Results for all-equity firms. This table shows descriptive statistics and regression results for firms from 1978 to 1997 with a ratio of total debt to total capital less than 1% at t  1. Excess value measures (EV) are calculated using the sales industry multiplier approach as in Berger and Ofek (1995). Subscripts indicate the time of measurement. Panel A reports summary statistics. DEV is the change in excess value. Firms reporting one (more than one) segment in the Compustat Industrial Annual Segment file are categorized as focused (diversified). Panel B contains regression results showing the predictive power of various diversification measures in explaining the change in excess value. The dependent variable is change in excess value. Div is a dummy variable equal to one for multi-segment firms, and zero otherwise. Ent represents the entropy measure of diversification. Coefficient estimates are shown with t-statistics in parentheses. Standard errors are adjusted for heteroskedasticity and within-year correlation. Panel A: Descriptive statistics Mean

EVt  1 Focusedt  1 Diversifiedt  1 DEVt1tot Focusedt  1 Diversifiedt  1

Median

First quartile

Third quartile

Standard deviation

Observations

0.239 0.114***

0.245 0.107***

 0.127  0.289

0.698 0.561

0.596 0.584

4,021 663

 0.099  0.024***

 0.066  0.008***

 0.359  0.217

0.178 0.191

0.442 0.381

4,021 663

Panel B: Change in excess value on diversification status Model 1

Divt  1

Model 2

0.075*** (3.48)

Model 3

0.077*** (3.69) 0.026 (0.60)

DDivt1tot

0.112*** (4.55)

Entt  1

DEntt1tot Intercept Adj. R2 N

 0.099*** (  11.34) 0.003 4,684

Model 4

 0.100*** ( 11.49) 0.003 4,684

 0.098*** ( 10.82) 0.004 4,684

0.113*** (4.52) 0.019 (0.25)  0.099*** ( 10.92) 0.003 4,684

Panel A: Using a t-test for means and a Mann-Whitney test for medians,***, **, and * indicate a significant difference from focused firms at the 1%, 5%, and 10% levels, respectively. Panel B: Coefficients significantly different from zero at the 1%, 5%, and 10% levels indicated respectively by***, **, and *.

change in excess value remains lower for focused firms even after removing entering and exiting firms.

5.3. Wealth transfer effects It may be possible for wealth transfers between stock and bondholders to explain our finding of a lower change in excess value for focused firms. As shown in Mansi and Reeb (2002), part of the diversification discount may be an artifact of wealth transfers and measurement error, since diversifying lowers bankruptcy risk and produces a gain for bondholders at the expense of stockholders. Since the excess value measure uses market value of equity but book value of debt, the excess value for diversified firms will be understated as wealth transferred from stockholders will be captured but not the wealth transferred to bondholders. This will produce an apparent diversification discount. In addition, if old debt that matures is replaced with new debt that is issued at par, the reported book value of debt will drift toward its market value. Under this scenario, measurement error in excess value will decrease over time and annual changes in excess value will be larger for diversified firms. To address the concern that wealth transfers between stockholders and bondholders explain our results, we

replicate our tests on a sample of all-equity firms. We define ‘‘all-equity’’ firms as firms with a ratio of debt to total capital less than 1%, as in Mansi and Reeb (2002).27 Consistent with their results, we find that the all-equity firms in our sample trade at a premium. However, the allequity diversified firms in our sample trade at a significantly lower average excess value than the allequity focused firms in our sample, as would be the case if diversified firms had lower growth rate uncertainty. These results are shown in Panel A of Table 12. Furthermore, Panel B of Table 12 shows that our finding of lower annual changes in excess value for focused firms is robust to wealth transfer effects, since the coefficients of interest in our model remain significant at the 1% level. Similar results using the assets multiplier are not shown for brevity. We conclude that our findings of a lower excess value for diversified firms and lower annual changes in excess value for focused firms are not caused by a wealth transfer from stockholders to bondholders.

27 The ratio of debt to total capital is calculated using the following formula of Compustat items: [(data34 + data9 +data56)/(data34 + data9+ data56 +data24*data25)]. Like Mansi and Reeb (2002), we use all of the focused firms in the original sample to identify the median focused firm used to calculate firm imputed values.

ARTICLE IN PRESS 480

J. Hund et al. / Journal of Financial Economics 96 (2010) 463–484

5.4. Expanded sample Beginning in 1998, disclosure of segment information was dramatically changed when the Financial Accounting Standards Board (FASB) introduced SFAS 131, a shift in accounting standards whose stated purpose was to improve reporting accuracy and transparency of firm segment structure. Under the previous standard, SFAS 14, defined in Financial Accounting Standards Board (1976), firms were required to map segments into industry classifications; the new standard requires firms to report segment information consistent with the internal organizational structure of the firm.28 By aligning the reporting requirements and the extant managerial information, SFAS 131 was intended to ease the burden on firms (leading to more timely and accurate reporting), and importantly, to provide a far clearer window into the decision-making structure of the firm. As noted in Villalonga (2004a) and in Berger and Hann (2003), the old reporting standard had been criticized for providing too much managerial discretion in classifying segments, leading to underreporting of the true extent of diversification for some firms. With the benefit of enhanced transparency, however, came the cost of reduced comparability of segments across firms.29 For our purposes, the shift to SFAS 131 and the concomitant change in Compustat reporting creates challenges in computing and interpreting several of our measures. In particular, firms may report multiple geographic or operating segments in similar product lines leading to spurious increases in diversification, and segment sales and assets no longer must conform to GAAP requirements, and may incorporate transfer pricing (in particular, assets are often allocated to corporate overhead segments). While we can do little to ameliorate the biases (if any) caused by departures from GAAP accounting, we attempt to reaggregate post-SFAS 131 segment data in Compustat in order to make it as comparable as possible with the earlier sample.

28 Specifically, the new standard defined an ‘‘operating’’ segment as a ‘‘component of the enterprise that engages in business activities from which it may earn revenues and incur expensesywhose operating results are regularly reviewed by the enterprise’s chief operating decision makeryand for which discrete financial information is available.’’ (Financial Accounting Standards Board, 1997) Note that these revenues and expenses may include those incurred with components of the same enterprise. As shown in Berger and Hann (2003), SFAS 131 increased the number of reported segments thus providing more disaggregated information to market participants. Since nearly all of the preceding papers on the diversification discount use pre-1998 data, our baseline sample stops in 1997. However, since the new segment reporting standard was designed to correct some of the criticisms of the old segment reporting standard, we add 1998–2005 to our sample after converting the data in a method similar to one suggested by Berger and Hann (2003). This extension of our data also provides a powerful quasiout-of-sample robustness check on our preceding results. 29 The FASB specifically acknowledged this motive by stating ‘‘both relevance and comparability will not be achievable in all cases, and relevance should be the overriding concern.’’ In addition, they noted that such a change was requested by the analyst community who ‘‘would assume more responsibility for making meaningful comparisons of those data to the unlike data of other firms that conduct their business differently.’’ (emphasis added)

Essentially, we identify segments within a firm classified as either operating or business type that share four-digit SIC codes and aggregate their sales into a single segment with that SIC code, while preserving the requirement that the sum of segment sales remain within 1% of the total firm-level sales. We drop segments marked as ‘‘corporate’’ that also have zero sales allocated to the segment. We then compute all of our measures on the aggregated data (using sales multiples); we do not present results for asset multiples because of extreme variation in the allocation of assets to segments with zero sales. An additional concern regards the comparability of our aggregated data with the pre-1998 data. Sanzhar (2006) notes the existence of a significant number of pseudoconglomerates, or firms with multiple divisions operating in single (three- or four-digit) SIC codes. Our post-1998 aggregated data by definition eliminates pseudo-conglomerates, and therefore, we perform the aggregation procedure on all of our data (from 1978 to 2005) in order to consistently define the entire sample. Whereas Sanzhar (2006) restricts his attention to only firms with multiple segments in a single SIC code in order to document the existence of an ‘‘organizational’’ discount, we also must identify diversified firms with multiple segments in the same SIC code. As an example of our methodology, consider PepsiCo’s segment reporting from 1985 until 2000.30 From 1985 until 1990, Pepsi reports three segments: Soft Drinks, Snack Foods, and Restaurants, corresponding to SIC codes 2087, 2096, and 5812, respectively. In 1991, however, Pepsi reports five segments: Soft Drinks, Snack Foods, Pizza Restaurants, Fast Food and Mexican, and Chicken Restaurants, but only three SIC codes (2087, 2096, and 5812 for the last three segments). All studies prior to this (and our baseline sample for comparability) would treat this reporting change as an increase in diversification in both the number of segments and in entropy measures. In our aggregated sample, we combine the three restaurant segments back into one, and Pepsi remains as a threesegment firm.31 In 1995, Pepsi adds a fourth segment in SIC 5812 for international sales (baseline segments now equal to six) and then in 1996 again reports only threesegments (Beverages, Snack Foods, and Restaurants); a divestment of restaurant operations leads to a twosegment firm in 1997. Breaking their existing segments into international and domestic segments in a change associated with SFAS 131 (and an increase in the juice business) leads to Pepsi becoming a five-segment firm, which only operates in three SIC codes: 2087, 2096, and 2033. Our aggregated procedure would produce a segment profile for Pepsi which ‘‘undoes’’ the reporting changes, and would report Pepsi as a three-segment firm

30 Berger and Hann (2003) also use PepsiCo as an example; we extend their example slightly to the entire period between 1985 until 2000 to illustrate the prevalence of pseudo-conglomerate segments in the earlier sample. 31 Doing this results in an aggregated sales series for the SIC 5812 segment (restaurants) of 6.225, 7.127, 8.232, and 9.356 billion for the period 1990–1993, rather than three separate segments with lower than industry median sales.

ARTICLE IN PRESS J. Hund et al. / Journal of Financial Economics 96 (2010) 463–484

Table 13 Summary statistics—aggregated sample. This table shows descriptive statistics for firms from 1978 to 2005 used in the analysis. Excess value measures (EV) are calculated using the industry sales multiplier approach as in Berger and Ofek (1995). DEV is the change in excess value. Age is defined as the natural log of the number of years since the firm first appears in Compustat counting from 1978 (the beginning of our sample). VolP is residual variance in the profitability process. ROE is earnings divided by book equity. EBIT/Sales is earnings before interest and taxes divided by sales. Size is the log of total assets. LTD/Assets is long-term debt normalized by total assets. CapEx/ Sales is the ratio of capital expenditures to sales. Subscripts indicate the time of measurement. Firms reporting one (more than one) segment in the Compustat Industrial Annual Segment file are categorized as focused (diversified).

Mean EVt  1 Focusedt  1 Diversifiedt  1 DEVt1tot Focusedt  1 Diversifiedt  1 Aget  1 Focusedt  1 Diversifiedt  1 VolP Focused Diversified ROEt  1 Focusedt  1 Diversifiedt  1 EBIT/Salest  1 Focusedt  1 Diversifiedt  1 Sizet  1 Focusedt  1 Diversifiedt  1 LTD/Assetst  1 Focusedt  1 Diversifiedt  1 CapEx/Salest  1 Focusedt  1 Diversifiedt  1

Median

Std. dev.

Table 14 Change in excess value on diversification status—aggregated sample. The following table contains regression results showing the predictive power of various diversification measures in explaining the change in excess value over the period 1978–2005. To address the changes in reporting after SFAS 131, we aggregate segment-level data at the fourdigit SIC code level to remove pseudo-conglomerates. The dependent variable, change in excess value, is EVt  EVt  1, where EV is measured using the sales multiplier approach of Berger and Ofek (1995). Div is a dummy variable equal to one for multi-segment firms, and zero otherwise. Ent represents the entropy measure of diversification. Size is the log of total assets. LTD/Assets is long-term debt normalized by total assets. EBIT/Sales is earnings before interest and taxes divided by sales. CapEx/Sales is the ratio of capital expenditures to sales. Subscripts indicate the time period of measurement. Coefficient estimates are shown with t-statistics in parentheses. Standard errors are adjusted for heteroskedasticity and within-year correlation. Model 1

Obs.

***

0.026 (4.11)

Divt  1 0.015  0.085***

0.000  0.101***

0.560 0.535

45,952 19,025

 0.048  0.016***

 0.035  0.013***

0.400 0.365

45,952 19,025

1.394 1.599***

1.504 1.872***

1.098 1.101

45,952 19,025

0.179 0.140***

0.089 0.071***

0.200 0.166

26,421 10,981

0.041 0.069***

0.101 0.108***

0.344 0.276

45,948 19,025

0.056 0.074***

0.074 0.079***

0.341 0.282

43,688 18,663

5.035 5.785***

4.823 5.635***

1.575 1.827

45,952 19,025

0.193 0.214***

0.142 0.191***

0.229 0.179

45,952 19,025

0.103 0.077***

0.044 0.044

0.230 0.147

45,499 18,917

DDivt1

to t

LTD/Assetst  1 EBIT/Salest  1 CapEx/Salest  1 to t

DLTD=Assetst1

to t

DEBIT=Salest1

to t

DCapEx=Salest1 Intercept

until it divests its restaurant business, at which point it drops to two segments, and then increases to three segments because of an increase in the real underlying juice operations. Table 13 presents a summary of our aggregated data for comparison with our baseline sample. As expected, the aggregating procedure reduces the proportion of diversified firms in the sample (from 32% to 29%) and the additional years add approximately 20,000 firm-year observations. Unsurprisingly, the median number of segments in diversified firms falls from three to two, and the entropy measure is slightly lower.32 Both the mean and the median diversification discount are 32 Note that this result shows the prevalence of pseudo-conglomerate segments in the baseline data, since the aggregation across SIC codes throughout the sample outweighs the (slight) increase in segment reporting shown by Berger and Hann (2003) in the post-SFAS 131 period.

Adj. R2 N

Model 3

to t

Model 4

***

0.025 (3.82)  0.074*** ( 4.71)

to t

Sizet  1

DSizet1

Model 2

0.032*** (3.90)

Entt  1

DEntt1

*** Indicates a significant difference from focused firms at the 1% level using a t-test for means and a Mann-Whitney test for medians.

481

0.002 0.003 0.001 (0.70) (1.15) (0.57)  0.002 0.026  0.002 (  0.13) (1.57) (  0.10) ** 0.018 0.012 0.018** (2.08) (1.33) (2.07)  0.108***  0.051**  0.107*** (  7.86) ( 2.64) (  7.82) 0.191*** (13.81) 0.065** (2.32) 0.033*** (4.58) 0.169*** (11.41)  0.045***  0.079***  0.042*** (  3.08) ( 5.45) (  2.92) 0.005 0.031 0.005 61,823 60,951 61,823

0.028*** (3.31)  0.176*** ( 5.67) 0.003 (1.10) 0.027 (1.61) 0.011 (1.22)  0.052** ( 2.65) 0.195*** (13.87) 0.066** (2.35) 0.033*** (4.56) 0.168*** (11.23)  0.077*** ( 5.46) 0.033 60,951

Coefficients significantly different from zero at the 1%, 5%, and 10% levels indicated respectively by ***, **, and *.

effectively unchanged, as are the calculated measures of average change in excess firm value, even with our greatly expanded sample and new classification scheme. As with the base sample, higher profitability for focused firms cannot be the explanation for the diversification discount as diversified firms have higher profitability using either return on equity (ROE) or the operating profit margin (EBIT over sales). Diversified firms continue to have a significantly lower VolP and a significantly higher age. Table 14 presents the results of our regression tests applied to the full 1978–2005 sample using controls for differences in profitability and risk. Focused firms continue to have lower annual changes in excess value. There is almost no change in our estimates; virtually all coefficients remain significant at the 1% level and the point estimates are similar. In Model 1 of Table 15 we use the Pa´stor and Veronesi (2003) variables of Age and VolP

ARTICLE IN PRESS 482

J. Hund et al. / Journal of Financial Economics 96 (2010) 463–484

Table 15 Change in excess value on uncertainty measures and diversification status–aggregated sample. The following table contains regression results showing the predictive power of various diversification measures in explaining the change in excess value over the period 1978–2005. To address the changes in reporting after SFAS 131, we aggregate segment-level data at the fourdigit SIC code level to remove pseudo-conglomerates. The dependent variable, change in excess value, is EVt  EVt  1, where EV is measured using the sales multiplier approach of Berger and Ofek (1995). Age is defined as the natural log of the number of years since the firm first appears in Compustat counting from 1978 (the beginning of our sample). VolP is residual variance in the profitability process. Div is a dummy variable equal to one for multi-segment firms, and zero otherwise. Ent represents the entropy measure of diversification. Dividend is an indicator dummy that equals one for firms that pay dividends. Size is the log of total assets. LTD/Assets is long-term debt normalized by total assets. EBIT/Sales is earnings before interest and taxes divided by sales. CapEx/Sales is the ratio of capital expenditures to sales. Subscripts indicate the time period of measurement. Coefficient estimates are shown with t-statistics in parentheses. Standard errors are adjusted for heteroskedasticity and within-firm correlation. Model 1

Model 2 ***

Model 3

Model 4

0.009 0.018 0.019 0.020*** (5.35) (11.02) (11.32) (11.66) VolP  0.033***  0.017**  0.016**  0.017** ( 4.25) ( 2.24) ( 2.19) (  2.28) 0.021*** Divt  1 (8.47)  0.065*** DDivt1 to t ( 3.28) 0.024*** Entt  1 (8.71)  0.156*** DEnt t1 to t (  6.33) 0.009*** 0.014*** 0.010*** 0.010*** Dividendt  1 (2.99) (4.92) (3.48) (3.47)  0.003***  0.004***  0.005***  0.005*** Sizet  1 ( 3.11) ( 4.64) ( 5.72) (  6.11)  0.026*** 0.026*** 0.023*** 0.023*** LTD/Assetst  1 ( 2.87) (3.11) (2.71) (2.79)  0.021  0.026**  0.025**  0.025** EBIT/Salest  1 ( 1.35) ( 2.57) ( 2.44) (  2.49)  0.083***  0.038***  0.032***  0.031*** CapEx/Salest  1 ( 7.21) ( 3.73) ( 3.11) (  3.03) 0.227*** 0.230*** 0.234*** DSizet1 to t (18.67) (18.89) (19.18) 0.172*** 0.171*** 0.172*** DLTD=Assetst1 to t (6.77) (6.73) (6.81) 0.025** 0.025** 0.024** DEBIT=Salest1 to t (2.04) (2.10) (2.10) 0.182*** 0.183*** 0.182*** DCapEx=Salest1 to t (10.05) (10.17) (10.15) Intercept  0.007  0.059***  0.060***  0.059*** ( 1.43) (  11.50) ( 11.67) ( 11.52) 0.003 0.036 0.037 0.038 Adj. R2 N 35,950 35,554 35,554 35,554

Aget  1

***

***

In Models 3 and 4 the diversification variable is added back to the specifications. Once again the diversification variable continues to have explanatory power after including the parameter uncertainty variables that are correlated with diversification. The coefficient on firm age remains significant at the 1% level when the diversification status variable is included. The coefficient on the volatility in the firm profitability process (VolP) is half as large and is significant at the 5% level rather than the 1% level. This is to be expected if diversified firms have lower volatility in the firm profitability process. Characteristics of diversified firms not captured by VolP and age may be correlated with the initial prior or the speed of mean reversion in the posterior distribution of a firm’s average profitability. We conclude that our finding of lower changes in excess value for focused firms is consistent Table 16 Change in excess value—instrumental variables on aggregated sample. This table reports second stage results where a multi-segment diversification indicator (Div) and the entropy measure of diversification (Ent) are instrumented in the first stage. To address the changes in reporting after SFAS 131, we aggregate segment-level data at the fourdigit SIC code level to remove pseudo-conglomerates. The sample covers the period 1978–2005. The dependent variable, change in excess value, is EVt  EVt  1, where EV is measured using the sales multiplier approach of Berger and Ofek (1995). Div is a dummy variable equal to one for multi-segment firms, and zero otherwise. Ent represents the entropy measure of diversification. Size is the log of total assets. LTD/Assets is long-term debt normalized by total assets. EBIT/Sales is earnings before interest and taxes divided by sales. CapEx/Sales is the ratio of capital expenditures to sales. Subscripts indicate the time period of measurement. The number of observations decreases compared to Table 14 due to missing observations of instrumental variables. Coefficient estimates are shown with t-statistics in parentheses. Standard errors are adjusted for heteroskedasticity and within-year correlation. Model 1

DDivt1

and omit diversification status. Once again, we find that younger firms (who tend to be focused) have lower annual changes in excess value. This evidence is consistent with at least part of the diversification discount being attributable to rational learning. In Model 2 we show these results are robust to changes in profitability and risk as control variables, along with levels of profitability and risk.

to t

DEnt t1

LTD/Assetst  1 EBIT/Salest  1 CapEx/Salest  1 to t

DLTD=Assetst1

to t

DEBIT=Salest1

to t

DCapEx=Salest1 Intercept

to t

Model 4

0.081*** (2.89)  0.050** ( 2.65)

to t

Sizet  1

Adj. R2 N

Model 3

0.087** (2.76)

Entt  1

DSizet1 Coefficients significantly different from zero at the 1%, 5%, and 10% levels indicated respectively by ***, **, and *.

0.072** (2.71)

Divt  1

Model 2

 0.001  0.000  0.002 (  0.30) ( 0.06) ( 0.59) 0.003 0.029 0.004 (0.18) (1.69) (0.24) 0.028** 0.018 0.028** (2.64) (1.39) (2.68)  0.099***  0.037*  0.098*** (  7.09) ( 1.73) (  7.25) 0.196*** (13.66) 0.061** (2.14) 0.037*** (2.98) 0.174*** (10.58)  0.047***  0.082***  0.040** (  3.19) ( 5.66) (  2.71) 0.002 0.028 0.003 55,803 54,976 55,803

0.096*** (2.88)  0.138*** ( 4.50)  0.001 (  0.33) 0.030* (1.75) 0.017 (1.35)  0.037* ( 1.73) 0.199*** (13.61) 0.062** (2.17) 0.037*** (2.93) 0.173*** (10.36)  0.075*** ( 5.30) 0.030 54,976

Coefficients significantly different from zero at the 1%, 5%, and 10% levels indicated respectively by ***, **, and *.

ARTICLE IN PRESS J. Hund et al. / Journal of Financial Economics 96 (2010) 463–484

with higher uncertainty regarding average profitability for focused firms. Table 16 presents the second-stage results of our instrumental variables specification applied to the full sample. Instrumental variables regression on the aggregated sample also yields similar point estimates to the earlier results in Table 10. This section provides strong evidence that the results we highlight throughout the paper are not merely an artifact of reporting biases, or the choice of a particular time period. We conclude that our results are exceptionally robust to both an expansion of our sample period and to changes in how we measure segments of diversified firms. 6. Conclusion The discussion about the value effects of diversification is ultimately a discussion on the optimal boundaries of the firm. As such, whether or not diversification creates or destroys value has been the subject of a vast amount of research, most of which has concentrated on identifying and explaining cross-sectional differences between focused and diversified firms. Within this setting it is extraordinarily difficult to differentiate between explanations for the identified value effects, the diversification discount, due to endogeneity both in the decision to diversify and in systematic differences between the types of firms. Our focus is to differentiate traditional explanations for the discount (which do not predict a difference in changes in excess value for diversified and focused firms) from the Pa´stor and Veronesi (2003) rational learning model. In contrast to previous research, we examine changes in excess value conditioned on firm organizational form and how these changes vary across the business cycle. We also examine idiosyncratic return volatility. If diversified firms have lower uncertainty regarding average profitability, the Pa´stor and Veronesi (2003) model can generate a diversification discount, a smaller drop in excess value for diversified firms over time, dynamic movements of excess value across the business cycle, and lower idiosyncratic return volatility for diversified firms. We support all of these predictions in our tests. In addition, we find that diversified firms are older and have lower volatility in profitability which is consistent with lower uncertainty and faster learning about average profitability for diversified firms. While inefficient cross-subsidization of low-quality projects can explain the diversification discount, it cannot account for a smaller drop in subsequent changes in excess value for diversified firms, or the dynamic movement of excess values across the business cycle, or the idiosyncratic return volatility results. Expanding the discussion of the well-documented diversification discount to include additional tests sheds new light on the source of the diversification discount. Lower uncertainty about the growth rate of diversified firms must also be considered as a plausible explanation for at least part of the discount. Traditional explanations for the diversification discount suggest there is suboptimal performance by diversified firm managers, or

483

that diversified firms have a poor outlook. Diversification is neither good nor bad if the diversification discount is due to lower uncertainty about average profitability. Diversified firms have a lower excess value initially, but have a smaller drop in subsequent changes in firm value than focused firms. Furthermore, this model assumes rational behavior on the part of both managers and investors. Volatility in operating profitability (which slows down learning) and firm age may not capture all of the differences in uncertainty across diversified and focused firms in our sample. If differences in uncertainty are the sole source of the diversification discount and we could perfectly measure uncertainty, we should not observe differences in firm excess value, subsequent changes in excess value, or idiosyncratic return volatility between diversified and focused firms after controlling for differences in uncertainty. Firms may tend to acquire firms with low uncertainty about mean profitability when they diversify and tend to divest segments with low uncertainty about mean profitability when they focus. This will cause a drop in firm excess value when a firm diversifies, higher subsequent changes in firm excess value for diversified firms, and lower idiosyncratic return volatility for diversified firms. Provided our instruments for diversification are valid, our IV tests suggest that it is something about organizational form itself (rather than an unobservable correlated with diversification status) that results in differences in uncertainty in mean profitability across diversified and focused firms. A likely cause is some type of cross-division effect. However, we leave further exploration of a crossdivision effect to future research. References Ahn, S., Denis, D., 2004. Internal capital markets and investment policy: evidence from corporate spinoffs. Journal of Financial Economics 71, 489–516. Berger, P., Hann, R., 2003. The impact of SFAS no. 131 on information and monitoring. Journal of Accounting Research 41, 163–223. Berger, P., Ofek, E., 1995. Diversification’s effect on firm value. Journal of Financial Economics 37, 39–65. Borghesi, R., Houston, J., Naranjo, A., 2007. Value, survival, and the evolution of firm organizational structure. Financial Management 36, 5–31. Brav, A., Heaton, J., 2002. Competing theories of financial anomalies. Review of Financial Studies 15, 575–606. Campa, J.M., Kedia, S., 2002. Explaining the diversification discount. Journal of Finance 57, 1731–1762. Campbell, J., Shiller, R., 1988. The dividend–price ratio and expectations of future dividends and discount factors. Review of Financial Studies 1, 195–228. Comment, R., Jarrell, G., 1995. Corporate focus and stock returns. Journal of Financial Economics 37, 67–87. Dimitrov, V., Tice, S., 2006. Corporate diversification and credit constraints: real effects across the business cycle. Review of Financial Studies 19, 1465–1498. Dittmar, A., Shivdasani, A., 2003. Divestitures and divisional investment policies. Journal of Finance 58, 2711–2744. Fama, E.F., French, K.R., 1989. Business conditions and expected returns on stocks and bonds. Journal of Financial Economics 25, 23–49. Fama, E.F., French, K.R., 2006. Profitability, investment and average returns. Journal of Financial Economics 82, 491–518. Ferson, W.E., Harvey, C.R., 1991. The variation of economic risk premiums. The Journal of Political Economy 99, 385–415. Financial Accounting Standards Board, 1976. Financial Reporting for Segments of a Business. Statement of Financial Accounting Standards No. 14, Norwalk, CT.

ARTICLE IN PRESS 484

J. Hund et al. / Journal of Financial Economics 96 (2010) 463–484

Financial Accounting Standards Board, 1997. Disclosures about Segments of an Enterprise and Related Information. Statement of Financial Accounting Standards No. 131, June, Norwalk, CT. Fluck, Z., Lynch, A.W., 1999. Why do firms merge and then divest? A theory of financial synergy. Journal of Business 72, 319–346. Graham, J., Lemmon, M., Wolf, J., 2002. Does corporate diversification destroy value? Journal of Finance 57, 695–720. Jacquemin, A.P., Berry, C.H., 1979. Entropy measure of diversification and corporate growth. Journal of Industrial Economics 27, 359–369. Jensen, M.C., 1986. Agency costs of free cash flow, corporate finance, and takeovers. American Economic Review 76, 323–329. Jensen, M.C., 1988. Takeovers: their causes and consequences. Journal of Economic Perspectives 2, 21–48. Krishnaswami, S., Subramaniam, V., 1999. Information asymmetry, valuation, and the corporate spin-off decision. Journal of Financial Economics 53, 73–112. La Porta, R., 1996. Expectations and the cross-section of stock returns. Journal of Finance 51, 1715–1742. Lakonishok, J., Shleifer, A., Vishny, R., 1994. Contrarian investment, extrapolation, and risk. Journal of Finance 49, 1541–1578. Lamont, O., Polk, C., 2001. The diversification discount: cash flows versus returns. Journal of Finance 56, 1693–1722. Lamont, O., Polk, C., 2002. Does diversification destroy value? Evidence from the industry shocks. Journal of Financial Economics 63, 51–77. Maksimovic, V., Phillips, G., 2002. Do conglomerate firms allocate resources inefficiently across industries? Theory and evidence. Journal of Finance 57, 721–767. Mansi, S., Reeb, D., 2002. Corporate diversification: what gets discounted? Journal of Finance 57, 2167–2183. Marriott, F.H.C., Pope, J.A., 1954. Bias in the estimation of autocorrelations. Biometrika 41, 390–402.

Mitton, T., Vorkink, K., 2010. Why do firms with diversification discounts have higher expected returns? Journal of Financial and Quantitative Analysis, forthcoming. Pa´stor, L., Veronesi, P., 2003. Stock valuation and learning about profitability. Journal of Finance 58, 1749–1789. Pa´stor, L., Veronesi, P., 2006. Was there a Nasdaq bubble in the late 1990s? Journal of Financial Economics 81, 61–100. Petersen, M., 2009. Estimating standard errors in finance panel data sets: comparing approaches. Review of Financial Studies 22, 435–480. Rajan, R., Servaes, H., Zingales, L., 2000. The cost of diversity: the diversification discount and inefficient investment. Journal of Finance 55, 35–80. Ritter, J., 1991. The long-run performance of initial public offerings. Journal of Finance 46, 3–27. Sanzhar, S.V., 2006. Discounted but not diversified: organizational structure and conglomerate discount. Unpublished working paper. University of North Carolina at Chapel Hill. Scharfstein, D., Stein, J., 2000. The dark side of internal capital markets: divisional rent-seeking and inefficient investment. Journal of Finance 55, 2537–2564. Schoar, A., 2002. Effects of corporate diversification on productivity. Journal of Finance 57, 2379–2403. Shin, H.-H., Stulz, R., 1998. Are internal capital markets efficient? Quarterly Journal of Economics 113, 531–552. Villalonga, B., 2004a. Diversification discount or premium? New evidence from the business information tracking series. Journal of Finance 59, 479–506. Villalonga, B., 2004b. Does diversification cause the ‘‘diversification discount’’? Financial Management 33, 5–27.