Diversification, Refocusing, and Firm Value

Transcription

1 Diversification, Refocusing, and Firm Value by Gönül Çolak Henry B. Tippie College of Business University of Iowa Iowa City, Iowa (319) This draft: January, 2003 I am grateful to my dissertation chair, Toni Whited, and my committee, Matt Billett, Tom Rietz, Paul Weller, and Charles Whiteman, for their support and helpful discussions. Thanks also to Simon Lee, Jay Wellman and Janis Zvingelis for their comments and suggestions. Thanks to Jim Grifhorst for his help in data collection. All errors and omissions are my responsibility.

2 Diversification, Refocusing, and Firm Value ABSTRACT In markets we see companies that are either diversifying, refocusing, or doing neither. The reasons behind the diversification decision and its effect on the firm s value are a well-researched subject. However, the factors affecting a firm s refocusing decision and its effect on firm valuation have not been investigated in detail. I estimate the factors affecting diversifying and refocusing decisions using a model of multiple choices. For a sample of diversifying and refocusing firms, between the years 1989 and 1998, I find that diversification occurs generally due to factors associated with industry and economic conditions, and refocusing primarily due to firm-specific reasons. In addition, I explicitly model and estimate refocusing and diversification effects on excess value in the same valuation equation. This allows for inclusion of all firms diversifying, refocusing, and single segment in the sample and thus, eliminates the concern of possible sample selection bias. Including refocusing as an explanatory variable in the valuation equation, together with diversification, is analogous to estimating a multiple regression versus a simple regression. I eliminate simultaneity bias in the parameter estimates of the valuation equation through the utilization of 2SLS. I find no evidence of a diversification discount or a diversification premium. Using different measures of diversification does not change this result. The effect of refocusing is also insignificant.

3 I. Introduction Diversification and refocusing are among the major activities in which firms engage during their life cycle. Both have been and continue to be an important part of a dynamic process through which firms seek to increase their total market value. However, there exist many studies suggesting that while diversification destroys value (Lang and Stulz (1994), Berger and Ofek (1995), Servaes (1996)), refocusing enhances it (Comment and Jarrell (1995), John and Ofek (1995), Daley, Mehrotra, and Sivakumar (1997)). Thus, a rational conglomerate presented with these two alternatives should always choose the latter. Yet, there are many firms that decide to diversify every year. In fact, more than those that refocus. This puzzle has drawn lots of attention from the researchers in financial economics. Hyland (2002) is interested in characteristics causing diversification in general. Others try to justify the relatively lower market valuation of diversified firms by showing that they have inefficient internal capital markets (Stulz (1990), Shin and Stulz (1998), Rajan, Servaes, and Zingales (2000), and Sharfstein and Stein (2000)) and/or higher agency costs (Amihud and Lev (1981), Jensen (1986), and Jensen and Murphy (1990)). Maksimovic and Phillips (2002), on the other hand, find evidence that the diversification discount is due to the equilibrium distribution of comparative advantage across firms in an industry. A significant part of the literature, however, concentrates on issues that cast doubt on the above justifications or on the very existence of the diversification discount. The first group of these studies finds that there are sample selection and measurement error problems associated with the papers explaining the diversification discount. Whited (2001), for example, demonstrates that there is a serious measurement error associated with Tobin s q, and that the investment-q regressions lead to an erroneous conclusion of inefficient internal capital markets. Thus, the tests of efficiency in internal capital markets are not valid, when Tobin s q is used as a proxy for investment opportunities. Mansi and Reeb (2002) use market value of debt, instead of book value, in the calculation of excess value and conclude that there is no discount due to diversification. There is only a value transfer from shareholders to bondholders. Billett and Mauer (2003) find that the value of 1

4 a firm depends on the nature of the specific components of its internal capital market and not on the overall value of that market. Chevalier (2000) raises concerns about sample selection biases that may arise due to the implicit assumption that diversifying firms acquire (or merge with) their new segments randomly. Her findings suggest that the diversification discount may not be a result of diversification, but rather that merging occurred between one or more discounted firms. Moreover, diversification may have occurred because of this discount, implying that the direction of the causality is unclear. Similarly, Graham, Lemmon, and Wolf (2002) conclude that firms diversifying through merger and acquisition observe a decline in their excess value, because the acquired firms are already discounted. This implies that there is a systematic difference between stand-alone firms and the segments of diversified firms, which may lead to downward bias in the excess value measure and hence, the appearance of a diversification discount. Other papers have built their analysis on the endogenous relationship between a firm s decision to diversify and the factors affecting its market valuation. Campa and Kedia (2002) control for this endogeneity by simultaneously estimating a probit model, with firm and industry characteristics as determinants of its diversification status, and a valuation equation, which includes the predicted values from probit estimation as a regressor. They apply the same technique to also determine the value effect of diversification among refocusing firms. Villalonga (2000), on the other hand, resolves this issue by matching each diversified firm with a single segment counterpart, with same size, industry, and propensity score. Again, the propensity score is obtained from a simple probit estimation. Motivated by the desire to explain the diversification discount, most of these studies concentrate on the valuation differences between single segment and diversified firms. However, diversification is only one of the larger set of strategies a firm can pursue to increase its market valuation. It is a strategic move rather than an incident that happens to a firm. Therefore, one should take into account other relevant choices available to a value-maximizing firm, such as refocusing and continuing as is. My study builds on this observation and contributes to the literature in two ways: first, by investigating various factors affecting the refocusing decision, and secondly, by 2

5 explicitly estimating its effect on firm valuation. The factors affecting the refocusing decision are estimated using a multinomial logit model. The decisions to diversify, to refocus, and to continue with the current segment(s) is the choice set; the characteristics affecting each decision are the choice-specific attributes. The types of factors analyzed are firm-specific characteristics, industry characteristics, variables describing the general economic conditions, and exchange related characteristics. The question of Why do firms refocus? is answered together with the question Why do firms diversify? The probability of a firm choosing each strategy is also calculated. The effects of refocusing and diversification are estimated by obtaining the propensities to be refocused and to be diversified from two separate simple probit models, and then including them as regressors in the same valuation equation. A 2SLS estimation procedure with three equations valuation, diversified, and refocused is used to control for the endogeneity of the diversified and the refocused status of the firm. Endogeneity exists, because characteristics determining firm s value are also affecting its decision to be diversified or to be refocused. This approach builds on Campa and Kedia s (2002) methodology, but is more general and, I believe, more appropriate. It is more general, because it accounts explicitly for the effect of refocusing on firm value. If diversification significantly changes a firm s value, then reversing it (i.e. refocusing) must also have a noticeable impact. It is important to estimate both, the diversification and the refocusing effects. Furthermore, it is more informative to estimate an equation with multiple explanatory variables rather than excluding some of them from the regression. My approach is more appropriate, because it allows for inclusion of all firms diversifying, refocusing, and single segment in the sample and thus, eliminates possible sample selection bias. Campa and Kedia (2002) use two different samples: a diversifyingfirms sample and a refocusing-firms sample. In the diversifying-firms sample, the refocusing firms are excluded and vice versa. Choosing a sample of firms on the basis of an endogenous variable leads to an incidental truncation and/or a sample selection bias. I find no evidence to support the existence of a diversification discount or a diversification premium. This result is robust to different diversification measures, such as asset-based Herfindahl index of the firm, sales-based Herfindahl index, and number of different SIC codes assigned to firm s segments. Refocusing also does not significantly 3

6 affect the firm s value. Therefore, prior findings of a diversification discount are primarily due to the existence of a simultaneity bias in the least squares estimates of the coefficients; and the diversification premium documented by Campa and Kedia (2002) may be due to the existence of an incidental truncation bias and/or a sample selection bias in their estimates. Additionally, I find that diversification is carried out primarily due to reasons related to industry and general economic conditions and refocusing primarily due to firm specific reasons. The firm attributes that significantly affect its decision to refocus are size, age, growth, profitability, and R&D intensity. If a firm has higher growth rates, profit margins, and R&D expenditures, it is less likely to refocus. The older and larger it gets, the higher the chances of refocusing. Among the variables indicating industry and economic conditions, only the industry s openness to conglomerates and the industry s average level of R&D expenditures are highly significant in influencing the firm s decision to refocus. Both have a positive effect. The diversification decision is positively affected by the growth of the economy and by the fraction of multi-segment firms in the industry of the firm. Industry s growth and firm s size also have a significant positive effect on this decision. On the other hand, firm s profitability, firm s investment level, and its industry s profitability discourage diversification. As expected, the probability of a firm diversifying is higher than its probability of refocusing: it has to be diversified to be able to refocus. Furthermore, the probability of not changing its number of segments is much bigger than the other two probabilities: 96% of the time it chooses to do nothing. Finally, I conduct an independence test Independence of Irrelevant Alternatives (IIA) to check for the appropriateness of using a multinomial logit model. The IIA test is also important, because it will indicate whether the exclusion of the refocusing option from the choice set (as in Hyland (2002), Campa and Kedia (2002), Villalonga (2000), etc.) is acceptable. If this independence assumption fails, then the probit model estimation will produce inconsistent estimates (Hausman and McFadden (1984)) and thus, erroneous estimation of the propensity to diversify. This, in turn, will lead to inaccurate estimation of the effect of diversification on the firm s value. 4

7 I cannot reject the independence of irrelevant alternatives. Thus, multinomial logit can be used in this context; and estimation of simple probit model would not lead to an inconsistency in the propensity to diversify and in the estimates of the characteristics affecting the firm s diversification decision. The rest of the paper is organized as follows. Section II describes the data and the sample selection criteria. Section III explains the estimation methodologies used. The results are presented in Section IV. Section V presents the estimates of diversification discount when different measures of diversification are used. Section VI concludes. The variables are described in the Appendix. II. Data and Variables The Sample Using the COMPUSTAT Industry Segment files I select a sample of diversifying and refocusing firms between the years 1989 and A firm that increases (decreases) its diversification level by adding (dropping) segment(s) is classified as diversifying (refocusing) firm. The firm years when there is no change in the number of segments are also included in the sample, because not changing its diversification level is also considered a major decision by the firm. To obtain information about the firm s characteristics at the time of the decision the COMPUSTAT Company and CRSP files are used. The time of the decision is assumed to be the year when the change in number of segments appears in the Segment files. Following Lang and Stulz (1994), Berger and Ofek (1995), and Comment and Jarrell (1995), I apply the following sample selection criteria: 1) the firm should have data in both the Company and Segment COMPUSTAT files; 2) for any given year, the firm should not have a segment in financial sector (SIC ), utilities (SIC ), government (SIC ), and non classified establishments (SIC ); 3) firm years with sales less than $20 million are dropped; 4) firm years with missing value of total assets (TA) or sales are excluded from the sample. Some additional criteria will be applied after the description of the excess value variable. 5

8 Excess Value In order to measure the change in firm value due to diversification or refocusing, I use the excess value created by Berger and Ofek (1995). It is a convenient measure for my purposes in that I am only interested in the firm specific variables affecting its value. The industry characteristics or general economic conditions influence all the firms of that industry in the same manner and thus, are of little use in determining the variations in these firms values. To calculate the excess value of a multi-segment firm, I need to know what its valuation would be when all of its segments are valued as separate entities operating on their own. If the current value of a conglomerate is not as high as the sum of its parts, i.e. excess value is negative, then diversification destroys value. If it is higher, then there are certain gains made from diversification. To put a separate value on each segment, the median industry sales (or asset) multiplier of single segment firms is used. The sum of the segment values is the imputed value of the firm. Excess value is calculated by taking the logarithm of the ratio of firm s total capital 1 to its imputed value. ExVal (ExValS) will denote the excess value calculated using asset (sales) multiplier. Following Berger and Ofek (1995), I drop all the firm years for which the sum of the segment asset figures deviates from the firm s assets by more than 25%. If this deviation is less than that, I gross up or down the imputed value of a firm calculated with asset multiplier to adjust for the difference. I exclude from the sample all the observations for which the sum of the segment sales is not within 1% of total sales for the firm. Finally, I drop the extreme observations for which the excess values calculated with either sales or asset multiplier are above or below Sample Statistics The above selection criteria leave me with a total of 6,233 different firms and 29,902 firm years. Table 1 contains detailed information about the distribution of firms across the different diversification profiles. There are 1,219 (651) firms that diversify (refocus), which account for 1,336 (800) firm years. Among diversifying firms, 909 were single 1 Firm s total capital is calculated as Total Capital = Market Value of Common Equity + Long-term Debt + Short-term Debt + Preferred Stock. 6

9 segment before their decision while 310 already had multiple segments. A total of 6,133 firms chose not to change their number of segments at one time or another. They correspond to 27,766 firm years. The number of firms that stayed single segment throughout the whole period is 4,186 (18,111 firmyears). Multi-segment firms that did not engage in any diversifying or refocusing activities throughout the period account for 2,288 firmyears (587 firms). In Table 2, I examine the characteristics of firms with different diversification profile. The table reports average firm size, age, profitability, leverage, growth, investment, number of segments, research and development, and two measures of excess value for firms in each diversification profile. A comparison of the means and medians of these characteristics suggests that refocusing firms are bigger, older, slow growing, and less innovative (measured by R&D intensity) relative to other firms. On the other hand, firms that are satisfied with their current condition (firms that do not change their number of segments) are smaller, younger, and research intensive. One explanation for this pattern could be the inclusion of single segment firms in this group. The diversifying firms have mean characteristics that are, generally, in between the means of the above two firm groups characteristics. Moreover, firms that chose to stay single segment throughout the period are smaller, younger, research intensive, and less leveraged relative to the ones that engaged in diversification and/or refocusing activities. Finally, the mean and median of excess value measures for no-change firms are closer to zero relative to those for diversifying and refocusing firms. The differences in the above variables suggest that a firm s diversification and refocusing decisions might be driven by its characteristics. However, these characteristics also affect the value of a firm, that is, the firm value and the decision are determined simultaneously by the firm features. Therefore, it is important to use an estimation methodology that takes into account this simultaneity. Also, based on the naïve statistics in Table2, one can opine that early on in its life cycle, a firm is reluctant to change its diversification level and is inclined to intensify its research and development and seek growth in its current industry. Firms that do not change their number of segments and are single segment and firms that are always single segment are younger, have higher R&D spending, and relatively higher three-year 7

10 average growth rates. 2 Subsequently, the firm looks for profits and growth in other areas through diversification, resulting in increased size. Growth stays high, but R&D expenses fall relative to its sales. Eventually, the slow growing, oversized firm decides to shed some segments in order to improve its internal capital market efficiency (Rajan, Servaes, and Zingales (2000)) and/or to reduce other agency costs. Variables The variables used in the multinomial logit model can be grouped in four major categories: firm-specific characteristics, industry characteristics, exchange-related characteristics, and variables describing the general economic conditions. Firm characteristics include Size, Profitability, Investment, Age, Leverage, and Growth. Industry characteristics are IndHerf, IndGr, IndRND, IndProf, and NMUL. MAJOREX and SNP are the exchange-related characteristics, and GDP and GDP1 are the variables capturing the general economic conditions. The definitions of these variables are in Appendix I. The variables used in the 2SLS estimation are Size, Profitability, Investment, RND, Age, Leverage, Growth, and SNP. Once and twice lagged values of Size, Profitability, and Investment are also used as regressors. The variables serving as instruments in the 2SLS estimation are the intercept, IndHerf, IndGr, IndProf, NMUL, GDP, GDP1, MAJOREX, SNP, and Leverage; once and twice lagged values of Size, Profitability, Investment, and Leverage. III. Estimation Methodology This section explains the methodologies used to estimate 1) the factors affecting a firm s decisions and 2) valuation effects of these decisions. However, one must first explain how the diversification and refocusing effects are measured. Diversification and Refocusing Decisions and their Valuation Consequences At any point in time a firm faces three choices that will alter its level of diversification: to add segments, to shed some segments, or to continue as is. Accordingly, one can 2 Hyland (2002) and Matsusaka (1995) reach similar conclusion. 8

11 observe that there are diversifying firms in the market, there are refocusing ones, and there are ones not engaging in any of these activities. Thus, some managers must think that their firms would be better off by adding new areas of business; others find it reasonable to get rid of some; and yet others prefer to make no changes. Given that a firm is a for-profit organization trying to maximize its market value, it has to, at least most of the times, act rationally. Behind every rational decision there are reasons. These reasons are closely tied to the firm s characteristics. Therefore, there must be certain identifiable firm characteristics that cause the firm to make that particular decision, and they can be different for each set of firms making a particular decision. Moreover, as a result of its decision a firm will either become more efficient and thus, have higher market valuation, or vice versa. Or it could be that this particular action of the firm is not significant, on itself, in affecting its valuation. Thus, it is important to properly measure and estimate the valuation change due to this decision throughout the years for which its effects are continuing. To do that one needs to differentiate between the decision of the firm, which is being made at a specific time, and the consequences of that decision, which will last for many years. An easy way of achieving this is through utilization of some dummy and choice variables. Table 3 shows few such variables and describes how each one behaves when the number of segments is changing from year to year. Table 3 Year No of Segments Dchange Rchange Dstatus Rstatus Y

12 The variables Dchange and Rchange are dummy variables indicating whether in a given year the firm increased, decreased, or made no changes in its number of segments. Dchange (Rchange) is 1 for the year when the number of segments increased (decreased). For the year when there was no change both of them are zero. Although these two dummy variables are capable of fully describing a firm s decision in a given year, they are not useful in multinomial logit model estimation. 3 Instead I am going to create a choice variable Y, which is 1 if the firm increased its number of segments, 2 if it decreased its number of segments, and 3 if it did not change its number of segments. On the other hand, variables Dstatus and Rstatus are specifying the current status of the firm diversified or refocused, and thus will capture the valuation consequences of the decisions to diversify and to refocus. The consequences of the diversification decision will last as long as the firm stays diversified, even though the firm may change its number of segments, for example, from 3 to 4, or from 3 to 2. If the firm has multiple segments, then the effects of diversification are present. Similarly, a refocusing firm will be affected by its decision, as long as it has fewer segments than in the past. A multi-segment firm does not need to become a singlesegment firm to be considered in the refocused status. If it has fewer segments than the maximum number of segments it ever had, then it is benefiting from its more focused status. Even though it did not decrease its segment number all the way to one, it still learned that it is not efficient to operate with as many segments as before. Therefore, the variable Rstatus describes this past experience of the firm, and will be used to capture the effect of refocusing decision on firm value. 4 Multinomial Logit Model In order to find the firm characteristics influencing its decision to diversify, refocus, or not change its current number of segments, I use a multinomial logit model. It is one of the most convenient estimation techniques for this kind of event. This model allows an 3 They are included in the table for expositional purposes. They help in distinguishing between change variables, like Dchange and Rchange, and status variables, like Dstatus and Rstatus. It is possible to create level variables that will show the level or the magnitude of the firm s diversification or refocusing action i.e. by how much the number of segments increased or decreased in a given year. 4 A simple test rejects at 5% level the hypothesis that the mean excess values of firms that have such past experience and firms that do not are equal. 10

13 individual firm to face multiple choices and each set of firms to have different individual specific characteristics. Firms are classified according to the choice they make. I explain this model in detail later in the section. To formally understand the multinomial logit model, one needs to use a random utility model. Namely, the ith firm faces J choices (in my case J=3), and utility it derives from choice j is U ij = G ij + ξ ij (1) where G ij is a deterministic utility function and ξ ij is an unobserved random variable. I will assume that G ij is linear, G ij = β z ij. The firm will decide on the choice j with the highest utility. This implies that Prob (U ij > U ik ) for all other k j. To have an econometric model one needs to choose a distribution for disturbances ξ ij If one assumes that they are iid with Weibull distribution F (ξ ij ) = exp(exp(-ξ ij )), and that z ij includes only the characteristics of the individuals that make the choice, s/he obtains multinomial logit model (McFadden (1973)). In this model the probabilities of particular choice j being chosen by the individual i are given by β' jzi e Prob ( Y = j) = P j =...(2) J β' k zi e k =1 for j = 1,2,..,J and where Y is the variable indicating the choice made by each firm. z i contains individual specific characteristics only. β J can be set to zero vector (β J = 0) as a normalization and thus: Prob(Y = J) = P J 1 = J β e k= 1 ' k z i...(3) 11

14 Therefore, j logit (or log-odds ratio) can be computed as for j = 1,2,, J-1; P ln P j J = β' z j i...(4) These J-1 equations are the ones that need to be estimated. Each one will have different estimated coefficients β j. The vector z i is common for all choices, but it varies with each firm. Unlike the OLS estimation, here the estimated coefficients vector β j does not represent marginal effects of the regressors. Instead, one has to find the marginal effects of the independent variables on the probabilities through P j z i = Pj [ β j Pk β J k = 1 k ]...(5) Note that the above equation does not guarantee that the signs of estimated coefficients and marginal effects will be the same, however, in most cases, they are. To test for significance of the coefficients I utilize the standard errors of the estimated coefficients. Calculation of standard errors for marginal effects is quite cumbersome and thus, very few studies use them. Plus, the t-statistics of the coefficients are very similar to the t-statistics of the marginal effects, and almost all of the information about the significance of a marginal effect can be conveyed by the t-stats of the coefficients. Finally, I want to determine whether multinomial logit is the right estimation technique in the context of this paper, or I need to use other models of multiple choices, such as a multinomial probit model or a nested logit model. The odds ratio in the multinomial logit model has to satisfy Independence of Irrelevant Alternatives (IIR), which basically states that each logit (P j /P J ) should be independent of the remaining probabilities. Hausman and McFadden (1984) suggest a test, similar to Hausman s specification test, which is based on the idea that if a subset of the choice set is independent of the other probabilities, omitting it from the model altogether should not change parameter estimates systematically. The test statistics is given by 12

15 χ 2 = (ˆ β S β ˆ f ' ) [ Vˆ S Vˆ ] 1 (ˆ β f S β ˆ ) ~ χ 2 ( K)...(6) f where β s indicates the estimators based on the restricted subset, β f indicates the estimators based on the full set of choices, V s - V f is the difference between respective estimates of the asymptotic covariance matrices, and K is the number of parameters estimated. I present the results of this test in Section IV. Next, I describe the methodology used for estimating the valuation consequences of diversification and refocusing. 2SLS Estimation To find the valuation effects of diversification or refocusing, there is a need for a procedure that allows for simultaneous estimation of the parameters of three linear equations. Certain firm characteristics will determine a firm s diversification profile: a diversifying firm or a refocusing firm. Some of these characteristics, alongside with diversified and refocused status of the firm, will also affect its excess value. Therefore, this is a case where certain variables simultaneously determine three endogenous variables (valuation measure, diversified status dummy, and refocused status dummy). If the valuation equation is estimated with OLS, the estimated parameters will have simultaneity bias 5 and will be inconsistent. A common procedure for estimating simultaneous equations is 2SLS. The first and the one with primary interest to me is the valuation equation: (Valuation) V it = X it β + α D Dstatus it + α R Rstatus it + ε it (7) where V it is a measure of firm value, X it is a matrix with the firm characteristics as its columns, β, α D, and α R are parameters to be estimated, ε it is the equation error, Dstatus it and Rstatus it are described above. 5 There are two sources of simultaneity bias: first and more obvious one is described later in this section; second one is due to the way variables are defined same accounting items are used to calculate different variables. For example, Tobin s q (which is calculated similarly to excess value measure firm s total capital is the numerator of both ratios) can serve as a proxy for investment opportunities and thus, it is significant determinant of firm s total investment. In return, firm s investment level affects its excess value. Therefore, investment and excess value variables are determining each other simultaneously. 13

16 The second equation specifies the firm s propensity to be in diversified status in terms of its characteristics. A firm stays diversified, because of certain reasons, such as possible gains due to its bigger size. This equation is intended to capture that. It is a simple probit or a simple logit model. (Diversified) Dstatus it = W it γ + u it (8) Dstatus it = 1 if Dstatus it > 0 Dstatus it = 0 if Dstatus it < 0 where Dstatus it is a latent variable, W it is a matrix with columns containing the variables affecting firm s decision to be in a diversified status, γ are parameters to be estimated, and u it is the error term. Similarly, the firm s decision to stay more focused relative to its previous years is determined by (Refocused) Rstatus it = S it η + ν it (9) Rstatus it = 1 if Rstatus it > 0 Rstatus it = 0 if Rstatus it < 0 where Rstatus it is a latent variable, S it is the matrix of variables affecting firm s decision to be in a refocused status, 6 η is the parameters vector, and ν it is the residual error. If the firm s decision to be diversified or to be refocused affects its value, then the valuation equation cannot be estimated using OLS, because the error term ε it will be correlated with Dstatus it and Rstatus it. To see this more clearly, notice that Dstatus it and Rstatus it are latent variables which are determined by Dstatus it and Rstatus it, respectively, but with an error. This measurement error causes u it and ν it to be correlated with ε it, which means that Dstatus it and Rstatus it are also correlated with ε it. 7 This correlation will violate one of the assumptions of OLS and will lead to biased coefficient 6 In my estimation I take S it to be identical to W it. 7 See Appendix II for a detailed proof. 14

17 estimates. Thus, V it, Dstatus it, and Rstatus it should be treated as endogenous variables in the above system of three equations, and the parameters β, α D, α R, γ, and η should be estimated using 2SLS estimation. Natural instruments for Dstatus it and Rstatus it would be some or all of the firm characteristics in X it. One can also use any industry related variable, since the excess value variable (suggested by Berger and Ofek (1995)) measures firm s valuation relative to the valuation of the median single segment firm in the industry. Thus, it is not affected by the events that have the same impact on all the firms in an industry. Industry characteristics, however, do influence the firm s decision to diversify or to refocus, as I show in Section IV of this study. Lang and Stulz (1994) also find that industry characteristics affect firm s decision to diversify. My industry related instruments are industry s Herfindahl index (IndHerf), growth in sales (IndGr), industry average of profitability (IndProf) and the fraction of all the firms in an industry that are multi-segment (NMUL). 8 The calculations of these variables are described in Appendix I. I also use exogenous variables that are related with the general economy. These instruments do not affect firm s excess value, because of the reasons described earlier; namely, they influence all the firms in the same manner. Following Campa and Kedia (2002), I will instrument the real annual growth of gross domestic product (GDP) and its lagged value (GDP1). Another instrument suggested by Campa and Kedia (2002) is a dummy variable (MAJOREX) that takes the value of 1 when the firm is listed in a major exchange, such as NYSE, NASDAQ, or AMEX, and zero otherwise. Finally, I use the lagged values of size, profitability, investment, and leverage as instruments. Relationship to Other Studies Some prior studies use a sample of only diversifying and single segment firms to determine the factors affecting a firm s decision to diversify (Hyland (2002)), or to estimate the changes in excess value due to this decision (Berger and Ofek (1995), 8 I use the 2-digit SIC code (DNUM) to classify the firms in industries. 15

18 Campa and Kedia (2002)). There are couple of problems with these approaches, however, and this study attempts to address them. First, these studies drop the observations for refocusing firms from their samples. This may induce a bias in the parameter estimates due to truncation of the true distribution. Truncation occurs when the sample used to estimate characteristics of a population is drawn from a subset of this population. Moreover, by excluding refocusing firms, they are in effect biasing their sample towards diversifying firms. Thus, estimated parameters will be biased. Unless characteristics and valuations of diversifying firms are identical to the firms that are left out, this sample bias will be present. My sample includes all firms: diversifying, refocusing, and no-action; single segment and multi-segment. Second, the papers by Berger and Ofek (1995) and Campa and Kedia (2002) find that diversification is significant in determining firm s value. However, neither study includes a variable to account for the effect of the refocusing decision, which means that they implicitly assume it does not have any impact on the firm s valuation. It is difficult to believe that diversification will significantly affect a conglomerate s value and reversing this diversification, i.e. refocusing, will not. Either both of these actions change a firm s value or both of them are done for purposes other than increasing valuation. If indeed both diversification and refocusing changes the firm s market value, then not including dummy variables to account for both of these decisions would lead to a significant missing variable problem (in their case a dummy for refocusing). My study measures the effects of both of these decisions on firm value, and therefore, provides a more complete description of the impact of the diversification activities on the valuation of a firm. The inclusion of refocusing and diversification effects in the same equation is analogous to estimating a multiple regression instead of a simple one. IV. Results Before estimating the valuation consequences of the diversification and the refocusing decisions, I will determine the characteristics affecting these decisions. 16

19 Multinomial Logit Estimation The sample consists of diversifying, refocusing, and no-decision firms. No-decision firms are those that do not change their number of segments for the year. The variable Y would represent the choice (decision) variable. Diversifying firms (Y=1) include firms that add new segment(s). The variable Y is 1 only for the year for which the number of segments of the corresponding firm increased relative to the previous year. The refocusing firms (Y=2) are those that drop some segment(s). Again, Y is 2 only for the year when the firm s number of segments decreases relative to the previous year. Nodecision (or no-change) firmyears are the ones for which there is no change in the segment structure of the corresponding firm. I estimate a multinomial logit model with three choices (J=3) available to the firms. The regressors are certain firm attributes (such as size, profitability, investment, R&D expenditures, leverage, and growth), general economic conditions (real growth of GDP and its lagged value), characteristics of the industry to which the firm belongs (industry s Herfindahl index, industry s growth, industry s profitability, industry s R&D intensity, and the fraction of multi-segment firms in the industry), and dummy variables indicating whether the firm is a member of an S&P Index and/or whether it is listed in a major exchange. Firm attributes will capture the firm specific reasons and the other variables will capture the industry, economic, and market conditions affecting the corresponding decision. The model is equivalent to estimating the two logits described in Eqn. (2): one for diversifying and one for refocusing firms. Table 4 reports the estimates of the coefficients, their t-statistics, and the corresponding marginal effects. The propensities to diversify, to refocus, and to make no changes are also calculated. As expected, the propensity to make no segment changes is very high relative to the other two propensities. Firms do not make changes in their segment structure very often. The propensity to diversify is and propensity to refocus is Firms diversify more often than they refocus. Given that the firm has to be diversified to be able to refocus, this result is expected. Significant firm attributes causing a firm to diversify are its size, profitability, and investment. The sign of the marginal effect of size variable suggests that larger firms are 17

20 more likely to diversify. Current profitability of the firm reduces its willingness to diversify its business activities. Interestingly, profitability is almost significant at a 5% level, but growth is not, even at a 10% level. This is consistent with the notion that firms diversify to increase their cash flows or earnings rather than to buy growth. I found that age is not significant in affecting the diversification decision. This result does not support Matsusaka s (95) claim that diversification is occurring in the late stages of the firm s life cycle. The negative signs of the marginal effect of investment and R&D imply that higher commitment in current activities, in terms of investment in equipment or R&D, prevents the firm from diversifying. Hyland (2002) finds that higher investment increases a firm s probability of diversifying, which is opposite to what I found. I believe that my results are easier to explain in that higher investment would leave the firm little cash to engage in merger and acquisition activities. However, my results agree with Hyland s (2002) findings about the importance of R&D in affecting a firm s willingness to diversify. A firm that does not have much R&D activity would be more likely to search for other firms that do. Among the industry-related variables, industry s profitability and the fractions of multisegment firms in the industry are significant at a 5% level. Firms operating in profitable industries will be less likely to seek profits elsewhere, indicated by negative marginal effect of IndProf. Not surprisingly, the number of diversified firms in an industry increases the propensity of a firm in that industry to diversify. Industry s growth and Herfindahl index are significant at 5% and 10% levels, respectively. Higher industry growth and higher concentration of industry sales in few firms encourages the firms in that industry to diversify. Real growth in the economy encourages diversification activities, indicated by the positive sign of GDP and GDP1. 9 There is a weak indication of the negative effect of SNP variable on the propensity to diversify or refocus. 10 However, neither real growth in the GDP nor listing in a major exchange have an important effect on the decision to refocus. 9 Maksimovich and Philips (2001) reach similar conclusions. 10 I find little evidence for the claim by Campa and Kedia (2002) that inclusion of a firm s stock in one of the S&P indexes would positively affect its decision to diversify or refocus. 18

21 The only industry characteristics significantly affecting the refocusing activities are industry s R&D intensity and industry s openness to diversified firms. Both of them encourage refocusing. However, firm characteristics do substantially affect the firm s decision to refocus. The propensity to decrease its number of segments increases with firm size and age. Again, high level of current profitability and heavy investment in capital and R&D makes the firm reluctant to refocus. As opposed to the diversifying decision, growth of the firm affects its refocusing decision significantly by increasing its willingness to refocus when it is not growing rapidly. 11 The above results suggest that, in general, refocusing is done primarily due to firm specific reasons and diversification primarily due to industry or general economic conditions. The estimated test statistics for the IIA test is (-1.28) in the case when the refocusing (diversification) decision is omitted from the choice set. The corresponding critical value for χ 2 distribution with 17 degrees of freedom is Thus, the decision to diversify (refocus) is independent from the other two decisions, and the multinomial logit model can be used. Prior estimations of the Diversification Discount To check whether my sample would lead to similar results as Berger and Ofek (1995) (from now on BO), I run OLS regressions with the dependent variable being excess value, calculated with either sales or asset multipliers. As shown in Table 5, the coefficient in front of the multisegment indicator is also significantly negative for my sample. For asset (sales) multiplier it is 0.09 (-0.18), which is different from BO s findings of 0.13 (-0.14), but still the diversification discount can be observed in the period between The other estimates, intercept, logta, EBIT/Sales, and CAPX/Sales, have the same sign as in BO. The estimation model of Campa and Kedia (2002) (hereafter referenced as CK) is closely related to mine; therefore, I have replicated their results. I first run OLS regression with CK s variables, calculated using a sample of diversifying firms for the 11 These results support Schlingemann, Stulz, and Walkling s (2002) finding that focusing firms are less profitable, slow growing, and investing less than firms with same number of segments that do not focus. 19

22 period from 1989 to The diversification discount in my sample is 6% (8%) for asset (sales) multiplier, which is comparable to CK's 9% (11%) finding. The signs and significance of the other firm characteristics are similar, with the exception of twice lagged value of logta. Similarly, I find that diversification decreases the excess value of refocusing firms, on average, by 5% (or 14% for sales multiplier). This decrease is 11% (13% for sales multiplier) in CK. Again, my estimates are qualitatively similar to CK s estimates. Thus, CK s sample and mine lead to similar results when OLS estimation is used. However, when I run 2SLS estimation for the same set of regressors as in CK, but with my instruments, 12 I find that diversification is insignificant in determining a firm s excess value. CK find a diversification premium of 19% (or 30% with sales multiplier) for diversifying firms and a diversification discount of 11% (21%) percent for refocusing firms. Their results suggest that while diversification is rewarded by the market with a premium, refocusing firms are punished for not further refocusing their activities to only one segment, that is refocusing firms are discounted, because they still operate with multiple segments even though they refocused. This implies an inconsistent behavior by the market of promoting diversification and discouraging refocusing firms of being diversified. The difference between CK s finding and mine may be due to the difference in the instruments used in the 2SLS estimation or due to the sample period selected. In any case, I confirm that there is no diversification discount for diversifying firms, once the simultaneity bias is removed. Estimating the effects of diversification and refocusing on firm value. In this section I estimate the change in the two excess value measures occurring because of diversified or more focused status of the firm. This change is measured by the coefficient in front of the two dummy variables Dstatus and Rstatus. These two variables describe a current status of the firm being diversified or being refocused. As explained earlier, Dstatus takes a value of 1 if the firm is diversified and 0 otherwise. It captures the valuation impact of the decision to diversify. As long as the firm stays diversified, the 20

23 effects of that decision continue. Similarly, Rstatus takes a value of 1 if the firm has fewer segments than in the past and 0 otherwise. As long as the firm operates with fewer segments than anytime in the past, the consequences of its refocusing decision will continue. Thus, the variable Rstatus is going to be 1 during the years for which these consequences last. I will also use the following variables as regressors: size, profitability, investment and their respective lagged values, RND, 13 age, leverage, growth, and SNP. As suggested by CK, the lagged values of size, profitability, and investment will capture the change in excess value due to past performance, i.e. it is possible that the firm is being rewarded for good performance that continues for years. This consistency may lead to valuation premium. To show that the diversification discount is not an artifact of a particular OLS estimation setup, but rather due to simultaneity bias, I first estimate the parameters using least squares regression with my set of independent variables. Then, using a system of three simultaneous equations described earlier, I eliminate the simultaneity bias by applying the 2SLS method. I run two OLS regressions for my sample, which includes diversifying, refocusing, and the rest of the firms during the period between 1989 and First, the variable Rstatus is excluded from the valuation equation, then the variable Dstatus. The column under OLS(1) in Table 6 shows that the diversification discount is still present, even when all the firms are included in the sample. Its value of 7% (or 13% for sales multiplier) is close to the previously documented estimations. The least squares estimate of the coefficient of Rstatus in column OLS(2) indicates that I cannot observe any significant gain or loss due to refocusing. Next, I simultaneously estimate Equations (7), (8), and (9) with the endogenous variables being excess value measure, Dstatus, and Rstatus. The results are shown in Table 6 (second-stage) and Table 7 (first-stage). I use a total of 18 instruments in the 2SLS estimation. Most of them are independent of excess value, and thus, are not 12 I do not use CK s instruments, because I believe that three of their instruments, namely the historical averages of TA, EBIT, and CAPX, are inappropriate. An instrument has to be uncorrelated with the error term, ε it, in Eqn. (7), but historical averages of the explanatory variables are correlated with ε it. 13 I drop all the observations that have missing RND. 21

24 included in Eqn. (7), but they do affect the propensities of being diversified or refocused. Because of the reasons described earlier, variables such as IndHerf, IndGr, IndProf, NMUL, MAJOREX, GDP, and GDP1, do not have any effect on the excess value measure. The variables that will instrument themselves are Intercept, Leverage, SNP, and lagged values of Size, Profitability, and Investment. To have a sufficient number of instruments for the J-Test, I instrument also the lagged values of Leverage variable. Thus, I have a total of M=18 instruments and K=17 coefficients to estimate in the valuation equation. The estimated coefficients of Dstatus and Rstatus are not significant for both measures of excess value. The previously observed significant negative coefficient of Dstatus is due to simultaneity bias. Therefore, my results do not support the existence of a diversification discount or a diversification premium. Refocusing does not have substantial long-term impact on the firm s excess value. Except for the significant change in the coefficient of Dstatus, all other independent variables are preserving their sign and significance. If one compares their coefficient estimates across the different estimation methodologies, the size of the firm, as well as its profitability and investment, have significant positive effect on excess value. Among the lagged values of these variables, the most notable result is the significance of the positive effects that the past profitability has on the firm s valuation. The estimates of growth and RND coefficients are positive and significant. The market rewards the companies that have higher growth rate and higher R&D expenditure. The effect of leverage is different for the different excess value measures used. 14 Being included in an S&P Index pays off around 20% valuation premium. The valuation of the firm decreases with its age. All of the above estimates are very similar for both OLS and 2SLS estimations. The above results are obtained by dropping all the observations for which RND is missing. To check whether they are substantially affected by this omission, I estimate the same equations, but with the assumption that a missing RND observation is actually an unreported zero R&D expenditure i.e. RND = 0 for all missing RND observations. I find that the OLS estimate of the diversification discount is very similar to my earlier 14 One explanation for this inconsistency could be the way leverage variable is calculated: it is being scaled by Total Asset instead of Sales. This measure is the one commonly used in the literature. 22