Abstract
Resource allocation in firms is often done in relative terms. Allocations to each project or, in the case of multibusiness firms, business segments are not made independently but through comparisons among the options. In that context, it becomes particularly important to identify the organizational factors that might influence those processes, as well as the mechanisms that create that influence. In this article, we investigate one of those potential factors—the size of a business segment relative to the rest of the organization—and two possible accounts. One is a naive tendency to spread out allocations evenly over the firm’s segments that would cause managers to relatively ignore differences in size and favor smaller segments over larger ones, holding other variables constant. The second is a tendency to direct larger allocation to the segments with the most political power and clout within the organization, which would normally favor larger segments, as those generally possess more influence. We investigate these competing hypotheses in a cross-section of firms to conclude that both mechanisms are partially at play. We find that both the smallest and the largest of segments are favored in the capital allocation process. Moreover, we find that the segment’s growth and profitability as well as corporate management ownership of the company moderate those effects.
Keywords
Introduction
Every year, corporations devote an enormous amount of time and effort to strategic planning and resource allocation. At almost every organizational level, managers weigh a number of investment options ranging from single projects to entire businesses. Strategic management research has acknowledged the connection between a firm’s strategy and its investment decisions, especially at the corporate level. Recent developments in the resource-based and dynamic capabilities approaches have renewed interest in the impacts of allocation processes on firms’ competitive advantages, as well as in the inefficiencies that might potentially arise in those processes (Berry, 2010; Coen & Maritan, 2011; Helfat et al., 2007; Maritan, 2001; Teece, 2009).
One aspect of these investment processes that has been largely ignored is the fact that corporate resource allocation, especially in organizations that are large and diversified, is done in relative rather than absolute terms. This means that executives, rather than focusing their attention on one investment proposal at a time and assessing its absolute value according to a number of financial and strategic criteria, often decide whether one proposal merits approval over others, or whether one business unit deserves a higher (or lower) allocation than others (Bower, 1970; Gilbert & Bower, 2005). Such comparisons are influenced by the way the process is conducted. Companies generally follow standard procedures on an annual basis to develop and assess investment projects. These procedures may assess a number of proposals jointly during allocation or budgeting committee meetings (Gilbert & Bower, 2005). 1 In a large multibusiness corporation, the “environment of decision” (Barnard, 1938) can become exceedingly complex for decision makers who simply do not have the time and knowledge to analyze a business in all its dimensions and, thus, focus their attention on a limited set of variables (Ocasio, 1997).
Among those variables, two particular sets stand out. First, executives tend to be influenced by the relationships between individual businesses segments within an organization (Banaszak-Holl, Mitchell, Baum, & Berta, 2006; Baum & Ingram, 2002; March & Simon, 1958; Vissa, Greve, & Chen, 2010). Second, allocators tend to be influenced by a few relative qualities of the investment options in those segments, such as past performance (e.g., profitability and growth) or future prospects. For instance, several studies (Berger & Ofek, 1995; Billet & Mauer, 2003; Ozbas & Scharfstein, 2010) have uncovered a tendency to subsidize business segments with poor investment prospects by taking capital from segments with good investment prospects. More recently, Coen and Maritan (2011) posited that investment in a particular business will be driven not only by its set of future opportunities, but also by the asset endowment that the business has accumulated over time. Arrfelt, Wiseman, and Hult (2013) found that business segments that underperform relative to others tend to receive more managerial attention and investment.
Our study aims to increase the understanding of how relative comparisons between different segments of the same firm impact resource allocation decisions. In this case, we focus on one characteristic that has often been overlooked as a driver of strategic decisions, namely, the size of a business segment, both in absolute and, especially, in relative terms. Moreover, since past literature has theoretically and empirically explored the effects on investment of various measures of segment “quality” such as Tobin’s Q (Ozbas & Scharfstein, 2010; Stein, 2003), profitability (Arrfelt et al., 2013; Bardolet, Lovallo, & Rumelt, 2010), and growth (Bardolet et al., 2010), we will investigate the interaction between a segment’s size and its past growth and profitability.
Size is an interesting and useful metric to study for a number of reasons. First, it is easily measurable, much more so than other characteristics. For example, the use of Tobin’s Q in cross-subsidization studies as a measure of businesses’ quality of prospects has raised some methodological issues (Villalonga, 2004; Whited, 2001). One could suggest that the median Q in each business’s industry constitutes a poor approximation to the real value of that business’s prospects. Size, on the other hand, can be measured by accurate proxies such as assets or sales. Secondly, size is also a variable that is easily perceived by a firm’s executives. While the quality of a particular business can be subject to endless discussions based on multiple factors and, thus, be assessed quite differently by managers within the same organization, the size of a business unit is perceived with relatively little noise. Finally, the size of any business segment is a variable that can be manipulated by the company. There is room for increasing it, by merging businesses or adding assets to them, for example. Or, size can be decreased, such as by spinning off or divesting some parts of a company.
In this study, we first analyze data from a cross-section of multibusiness firms to investigate contradicting hypotheses about the effect of relative segment size on allocations. On one hand, Bardolet, Fox, and Lovallo (2011) found evidence of naive diversification in corporate allocations, that is, a tendency for managers to spread investment equally across their firm’s segments, suggesting that they are relatively insensitive to differences among segments. If that insensitivity extends to segment size, it would result in managers allocating disproportionately more capital to their firm’s smaller segments, after controlling for other factors. On the other hand, studies of resource allocation have also documented how agency conflicts and political influences can affect segment investment. For instance, studies on the power balances among organizational units (Duchin & Sosyura, 2013; Glaser, Lopez-de-Silanes, & Sautner, 2013; Hackman, 1985; Kim, Hoskisson, & Wan, 2004) have shown that powerful units are able to attract more resources relative to other units. However, those studies tended to define power in terms of corporate governance relations (e.g., units with the most directors in a keiretsu firm, share ownership, etc.) and only made indirect correlations with unit size (Hackman, 1985). 2 We speculate, as a logical extension of this power hypothesis, that segments with greater relative size (in terms of sales or assets) have more political power and, thus, are able to attract greater normalized capital investment. This hypothesis directly contradicts the naive diversification account hypothesis, and thus, it is important to compare the two when investigating their combined role in the effect of segment size on resource allocation.
We find the combination of cognitive and political effects to be quite complex and nonlinear. In particular, we find that, holding everything else constant, segment capital expenditures, normalized by sales, increase with the relative size of the segment, which lends support to the power hypothesis. However, the smallest (in relative terms) segments in our sample receive proportionally more capital than other businesses in the multibusiness firm. Interestingly, this effect exists after controlling for the absolute size of the segments, so it cannot be attributed to a simple consequence of a small denominator when we use capital expenditures over sales as our measure of investment. Previous literature has neither controlled for absolute size nor examined the differential effects of relative size at the opposite, extreme ends of the distribution. We obtain quite robust results for this nonlinear relationship using both regression analysis and treatment effects from pairwise matching. Therefore, while an initial result of our article is that we cannot replicate Bardolet et al.’s (2011) naive diversification finding, if we control for absolute size, our results suggest that concentrating on a general effect of relative size largely misses the trees for the forest. 3 The smallest 10% of segments drive the previously observed negative correlation between relative size and investment. For the other 90%, relative size is positively associated with normalized capital investment.
On the other hand, executives often rely on measures like past growth and profitability to help them make allocation decisions (Graham & Harvey, 2001). In fact, these measures have traditionally played a large role in assessing the corporate portfolio of businesses. Managers have been taught that growth businesses of any kind, whether profitable or not, tend to be cash-needy (Bardolet et al., 2010). That is, the investment needs of those businesses cannot be covered by self-generated cash, and thus, managers should prune allocations to low-growth businesses and use that cash to fund them. Therefore, the labels generated by portfolio analysis (e.g., “seed business,” “growth business,” “cash cow,” etc.) introduce an additional behavioral pattern that might interact with any effect of relative size, an interaction that we explore in this study. Our empirical analysis provides evidence that executives’ preferences for overinvesting in larger units are mostly driven by the low-growth, high-profitability segments in our sample. In contrast, the tendency to invest in the smallest units is driven by the differential treatment of low-profitability segments.
Resource allocation is still a relatively unexplored area in strategic management, particularly from an empirical approach, where there are only a handful of documented empirical regularities. Our empirical study adds to these and makes a number of contributions. First, it compares and combines two different streams of research on resource allocation and business segment size. In other words, where naive diversification and managerial power studies suggest opposing hypotheses regarding investment and business segment size, our article clarifies the scenarios where one dominates the other, thus helping resolve any confusion. Second, this study furthers the understanding of the impact of segment size on allocation by investigating interactions with past segment growth and profitability. Third, we enrich understanding of the naive diversification phenomenon in resource allocation. Bardolet et al. (2011) proposed a linear relationship that affects all segments equally by augmenting allocations to smaller segments and decreasing them for larger ones. In contrast, our analysis shows that this effect does not hold, on aggregate, when one controls for absolute size, and that overinvestment in smaller segments is concentrated at the far end of the distribution’s bottom tail. Fourth, this article also provides evidence that managerial power, as defined in the literature (Hackman, 1985; Kim et al., 2004; Pfeffer, 1981), extends to relative segment size. While one could readily assume a correlation between the size of a unit and other measures of its influence such as its role in the governance of the firm or even institutional ties, it is not a given that the impact of the former automatically follows from that of the latter. Finally, we find a moderating effect of managerial ownership (i.e., the percentage of company shares that the top five corporate managers own) on these relative size effects. The fact that the observed preference for the extremes decreases with ownership implies that better-aligned managerial incentives can moderate size effects. It is also consistent with an explanation that these effects are not optimal and result from agency problems that senior executives face.
In the following sections, we present, in order, the theory leading to two competing hypotheses for the relationship between investment and relative segment size. We also present two additional hypotheses on the interaction of the previous relationship with segment growth and profitability. Then, we present a final hypothesis about corporate ownership by the top executives of the firm. We then present empirical tests that include regression analysis and matching methods for increasing the robustness of our findings. We conclude with a discussion of our results and avenues for future research.
Theory and Hypotheses
Multibusiness corporations must choose how to allocate capital expenditures to their divisions. Research on resource allocation phenomena has taken either a behavioral perspective (Arrfelt et al., 2013; Bardolet et al., 2011; Bower, 1970) or a rational agency-based one (see Stein, 2003, for a summary of that literature). In the rest of this section, we use insights from both approaches to formulate alternative hypotheses.
From a behavioral perspective, naive diversification—the tendency for even allocation across all segments (Benartzi & Thaler, 2001)—might be a significant part of the cross-subsidization phenomenon observed in the finance literature. Cross-subsidization occurs when corporations shift resources from better-performing divisions to worse-performing ones (Stein, 2003). Bardolet et al. (2011) argued that if firms are biased toward investing 1/n of capital to each of n segments, then capital allocated to each target segment will decrease as n increases, holding other relevant factors constant.
The naive diversification hypothesis has wider implications for corporate resource allocation. Bardolet et al.’s (2011) analysis mostly focuses on the 1/n bias, which may cause managers to underweight quality differences between business segments and allocate resources evenly (and thus provides a cognitive root for the previously observed phenomenon of cross-subsidization from good to bad units). In these circumstances, segments with poor past performance or future prospects may receive a disproportionately high amount of funding. More interestingly for this study, naive diversification may also make managers underweight differences of size. As a result, smaller segments may receive relatively more funding than larger segments (after controlling for other relevant factors such as growth, profitability, future prospects, etc.). This means that segments from different companies that are similar in all attributes except relative size may receive different levels of funding.
One difference between our interpretation of this explanation and Bardolet et al. (2011) is that we control for the absolute size of a segment. There are several neoclassical economic reasons why investment might differ by absolute size. Businesses that are small in absolute terms might require a relatively high amount of resources due to potential growth opportunities (Sengul & Gimeno, 2013) or their greater efficiency (Maksimovic & Phillips, 2002). For these reasons, it is unclear whether we would expect to find evidence of the 1/n bias in our study. That being said, our stated hypothesis is very similar to Bardolet et al.’s.
Hypothesis 1a: Relative segment size is negatively correlated with capital investment. That is, segments that are smaller than the other segments within a company will receive more capital investment than larger ones, holding all else equal (i.e., absolute size, profitability, sales growth, age, industry).
On the other hand, allocation inefficiency in internal capital markets has generally been attributed to agency conflicts between divisional managers and corporate headquarters. Several authors (Rajan, Servaes, & Zingales, 2000; Scharsftein & Stein, 2000; Wulf, 2009) portray divisional managers as rent-seeking agents who spend time and effort trying to lobby headquarters for more money, while corporate managers are principals that use the capital allocation process as an incentive scheme to control those managers. This agency-based account contains some strong assumptions. For instance, one needs to assume that managers of underperforming divisions are powerful enough to merit the principal’s attention and, thus, receive increased allocations. Moreover, one needs to assume that managers of underperforming divisions are more powerful than managers of well-performing ones, in order for cross-subsidization to occur. In fact, more than 35 CFOs of large multibusiness companies with whom one of the authors has held private interviews on resource allocation did not think that this conjecture was plausible enough to justify consideration.
Even if that particular aspect of the agency account is difficult to connect to managerial practice, the basic insight that allocation decisions are influenced by managerial opportunism and political factors is well established. Bower’s (1970) single-firm field study led to a framework—known as the resource allocation process (RAP) model—that describes allocation processes as a competition between business units for a limited pool of resources (Burgelman, 1991). Subsequent extensions of Bower’s model to several strategic processes (Bower & Christensen, 1996; Burgelman, 1983; Noda & Bower, 1996) point to the relationship between corporate and divisional managers as a key source of allocation inefficiency.
On the other hand, considering differences in relative segment size within a firm instead of differences in relative performance might make the agency account more plausible. In this case, there would still be an imbalance of lobbying power between the two types of managers/segments, but it would favor larger segments over smaller ones, rather than poorly performing segments over well-performing ones. Power and influence are an inherent factor in strategic decisions (Pffefer, 1981), especially decisions concerning the distribution of scarce resources (Salancik & Pfeffer, 1974). In resource allocation processes, the relative power of each division has been found to affect the amount that each receives. For instance, an early study by Hackman (1985) found that university departments with higher levels of institutional power (e.g., longer history and visibility inside and outside the university, higher numbers of employees and customers) received more money in the budgeting process than departments with low power. More recently, Kim et al. (2004) found that powerful keiretsu members are given resources to grow, while less powerful ones have a difficult time obtaining those resources, instead being “milked” for profits. Other studies of the influence of power on resource allocation have reached similar conclusions (Duchin & Sosyura, 2013; Glaser et al., 2013). Although the relative size of a business segment (measured by percentage of total assets or sales) has not yet been proposed as a direct source of power, Hackman’s (1985) correlation between a university department’s number of employees and its institutional power suggests that size might indeed lead to increased influence within a corporation. Moreover, the level of institutional power has been linked with the unit’s visibility and age (Hackman, 1985), its ability to provide resources for the organization (Pfeffer & Moore, 1980), and its previous levels of power (Lachman, 1989), all factors that one could easily associate with the unit’s relative size. Thus, we hypothesize that relatively large business units wield higher amounts of power in the competition for resources and attract a relatively large share of resources.
Hypothesis 1b: Relative size is positively correlated with capital investment. That is, segments that are larger than the other segments within a company will receive more capital investment than smaller ones, all else being equal (i.e., absolute size, profitability, sales growth, age, industry).
Generally, managers claim that their capital allocation decisions are driven by the segment’s future opportunities or, at the very least, by the segment’s recent performance (Graham & Harvey, 2001). Hypotheses 1a and 1b addressed the possibility that relative size affects allocations, after controlling for future opportunities (Tobin’s Q) and past performance (profitability and sales growth). However, this does not mean that the effects must be independent of those controlling variables. For instance, segments with high profitability but low growth are very often “self-sufficient” in terms of capital (Bardolet et al., 2010) since their return on assets generally exceeds their growth rate, making them net generators of cash flow within the firm. Classic portfolio analysis frameworks advise executives to use the excess cash from those businesses to fund other segments within the firm that are “cash-needy”; that is, they do not generate enough cash flow on their own to sustain their required investments. Those cash-needy businesses are, in turn, segments with high growth prospects (hence the need for investment) but low profitability (often, but not always, a direct consequence of being at an initial stage of development in the market). Given these heuristics, one could wonder how the preference for segments of certain size within the company would change for different types of segments. In our case, such hypothesizing implies the possibility that the size effects proposed by Hypotheses 1a and 1b might vary according to segments’ growth-profitability types. Therefore, in the following paragraphs, we discuss the potential interactions between segment relative size and its growth-profitability type.
First, the basis for Hypothesis 1a—the 1/n heuristic—is a cognitive bias. Its prediction is based precisely on the inability of individuals—or even groups (Larrick, 2004)—to take into account the relative differences among choices (Benartzi & Thaler, 2001). If managers are mostly anchored on the 1/n allocation, then the difference in capital investment between two businesses of different companies that differ only in relative size (i.e., identical sales, growth, and profitability measures) will not be altered by changes in their growth-profitability type, which leads us to the following hypothesis:
Hypothesis 2a: Hypothesis 1a—the relatively higher capital investment enjoyed by relatively smaller units (all else being equal)—will hold
It is important to stress that this hypothesis is testing an interaction effect. That is, while we find it natural to expect differences in the relative amount of capital that each growth-profitability type gets (after all, portfolio matrix heuristic rules advise managers to redirect cash flows generated by the “cash cows” towards high-growth, low-profitability businesses), Hypothesis 2a looks past such effects. Specifically, we are interested in testing whether the advantage enjoyed by relatively smaller business segments is independent of growth-profitability type.
On the other hand, Hypothesis 1b is an agency-based prediction. It assumes corporate managers are well aware of the profitability and growth of subsidiary business segments but might be susceptible to lobbying from them (Scharfstein & Stein, 2001) or might have strategic reasons to favor certain segment types over others (Sengul & Gimeno, 2013). Segment managers often emphasize the growth and profitability of their business to corporate managers as a justification for receiving more capital investment (Arrfelt et al., 2013). This would be akin to employing an “eat what you kill” rule, in which the segment gets to keep its earnings for future investment. In terms of underlying causes, such rules would reflect procedural justice concerns rather than simple power-based lobbying, and it could partially explain why corporate managers might be more generous toward relatively large, low-growth, high-profit segments. On the other hand, this tendency will be more influential the greater the political power segment managers have. For instance, Kim et al. (2004) found that keiretsu members with stronger ties to the corporate center are more likely to be allowed to invest in diversifying their business, while less powerful members are more likely to be “milked” for profits. Thus, while portfolio prescriptions prod managers to pull investment from low-growth businesses, their actual behavior might be significantly influenced by the segment managers’ ability to leverage their needs (i.e. “we need to increase our growth”) and their past performance (i.e. “we are generating enough cash flow to cover those needs as long as you let us”). For example, some anecdotal evidence (Brass, 2010) suggests that this occurred at Microsoft, where resources that could be devoted to developing opportunities such as tablets or phones were instead absorbed by the mature software business by virtue of its high profitability. 4 For these reasons, we propose the following hypothesis:
Hypothesis 2b: Hypothesis 1b—the relatively higher capital investment enjoyed by relatively larger units (all else being equal)—will be most apparent in the low-growth, high-profitability segment type.
Figure 1 illustrates Hypotheses 2a and 2b by providing a representation of all four growth/profitability quadrants in which we investigate these effects.
Hypotheses by Growth/Profitability Quadrants
Finally, a central feature of the agency account is the possibility to moderate its effect through mechanisms that align the managers’ interests with those of the company. One of the mechanisms is the degree of company ownership that corporate managers have. A higher stake in the firm should make corporate executives more sensitive to efficient allocations and drive them away from self-serving decisions. In fact, Ozbas and Scharfstein (2010) found that “corporate socialism”—that is, the tendency to spread capital over segments of different value beyond what is efficient—is less pronounced in firms where managers have a larger stake in the firm. Hoskisson, Johnson, and Moeser (1994) found similar dynamics at the board level, where smaller ownership leads to less divestment across the firm, presumably because board members have weaker incentives to confront rent-seeking managers. Regarding the relative size effect, a natural prediction that complements the agency account is that when corporate managers possess larger stakes in the firm, they will be less inclined to give in to the lobbying and power of managers of large segments and, as a consequence, will make more efficient allocations. Similarly, those corporate managers might also be less prone to fall into a naive diversification bias over the firm’s differently-sized segments (Larrick, 2004). Therefore, we would predict corporate management ownership to moderate both size effects, as expressed in the following hypothesis:
Hypothesis 3: The effects predicted by Hypotheses 1a and 1b will be mitigated when the top five company managers have a high level of company ownership.
In summary, whereas a cognitive approach predicts larger allocations for smaller segments, the power/agency approach predicts larger allocations for larger segments. We examine these two seemingly contradictory predictions in the next section.
Data and Estimation
U.S. Securities and Exchange Commission (SEC) regulations require all publicly listed companies to report sales, operating profit, depreciation, capital expenditures, and total assets at the business segment level. These data are included in Standard & Poor’s COMPUSTAT database. In order to identify the main activity of each segment, COMPUSTAT assigns a primary and secondary four-digit Standard Industrial Classification (SIC) code to each of a company’s segments, as well as a segment name as reported by the company. One well-known limitation of COMPUSTAT data is that different companies use different criteria in deciding what constitutes a business segment. Moreover, some companies report business segments differently at different points in time. We decided to use a unifying criterion to minimize this problem and used SIC codes to aggregate reported segments at the three-digit industry level. Thus, in our sample, a firm has as many businesses as industries at the three-digit level. This method has the advantage of reducing the noise inherent in the definition of segments in the COMPUSTAT files. 5 Consolidating segments using SIC codes is common in other segment-based studies of capital investment (e.g., Lamont, 1997; Ozbas & Scharfstein, 2010). In order to avoid observations with unreasonably high investment to assets ratios, we require the remaining firms to have total consolidated sales and assets of at least $20 million. From 1991 to 2004, the raw COMPUSTAT data contained 152,287 segment-year observations. After consolidation at the three-digit SIC code level and elimination of segments with incomplete data, 13,639 segment-year observations remained.
Absolute Versus Relative Size and Dummy Regressions
Unlike previous studies on resource allocation, our article examines the effects of both a segment’s absolute size (i.e., total sales) and relative size (i.e., percentage of the firm’s total sales) in capital investment. Making this distinction is quite important—as the two values are correlated, one might conflate their effects. It is conventional, neoclassical wisdom that segments of small absolute size receive a disproportionate share of investment. One possible reason is that small businesses have great growth opportunities because they are at an early stage of development. If this absolute size effect is correct, and one does not account for absolute size, one could spuriously conclude that relative size is inversely proportional to capital investment.
To control for this, we divide our sample into deciles of both absolute size (total segment sales) and relative size (total segment sales/total parent firm sales) and create dummy variables for each decile. That means a segment will have two indicator variables, one for absolute size decile and one for relative size decile. Looking at the correlation between relative and absolute size deciles (r = .282) it seems clear that an nth-decile relative size segment need not correspond to an nth-decile absolute size segment. Moreover, a look at the breakdown of the joint distribution of the two types of deciles (see Appendix A) reveals that, at most, only 28% of one type of decile is contained in another decile. This confirms the distinction between being a small (large) segment in absolute terms (or, more precisely, relative to other segments in the sample) and being a smaller (larger) segment relative to other segments within one’s own firm.
Having split our data into deciles, we start investigating the relationship between the decile-based size indicators and segment normalized investment (defined as segment capital expenditures ÷ sales, as reported by COMPUSTAT). We first perform a descriptive statistical analysis, looking for stylized facts that we can build on later.
Table 1 provides summary statistics of our main variables over the two main sample subsets used in the article. The variables sales, assets, capital expenditures, cash flow, and age are taken directly from COMPUSTAT. Capital expenditures divided by sales will be our main dependent variable. Cross-industry studies of internal capital markets use either assets or sales (they are highly correlated) as valid scaling variables of the unit’s investment (Lamont, 1997; Ozbas & Scharfstein, 2010). Relative segment size is the ratio of segment sales to total firm sales (defined as the sum of all its segments’ sales). Segment profitability is defined as segment cash flow divided by sales. Sales growth is defined as the difference between current and previous year’s sales divided by previous year’s sales. Industry Q in a given year is the median bounded Tobin’s Q of standalone firms in the industry, defined at the three-digit SIC level.
Descriptive Statistics by Group
Note: Observations are by segment and year (COMPUSTAT segment files, 1989–2004). Segment cash flow is defined as segment operating profits plus segment depreciation. Segment sales, assets, capital expenditure, and cash flow are in millions of dollars. Sales growth is the difference in this yearly change in sales over last year’s sales. Profitability is measured as cash flow over sales. Sales growth is the difference between current and previous year’s sales divided by previous year’s sales. Industry Q in a given year is the median bounded Q of standalone firms in the industry. Mean comparison tests between groups are performed without the assumption of equal variance.
p < .10.
p < .01.
p < .001.
In the top panel, we compare the bottom relative size decile to the other deciles. In the bottom panel, we exclude that bottom decile and divide the observations equally above and below the median relative size. Looking over this raw data, we observe that sales-normalized capital expenditures are significantly higher in both relatively small-size groups. However, profitability is also significantly higher for those smaller segments, so it is unclear at this point whether to attribute the extra investment to a size effect or to a simple response to segment performance. Furthermore, we have yet to control for industry and segment fixed effects to make a more clear comparison of capital investment by relative size in the regressions.
Finally, since this study considers both absolute and relative sizes, a natural place to begin is to examine whether previous effects hold if we account for both. Table 2 provides the regressions of Bardolet et al. (2011), specified as
where δ is the coefficient on relative size,
Previous Evidence for the 1/n Effect
Note: Regressions of the effect of relative size (in sales) on capital spending divided by sales for segments of multisegment firms (COMPUSTAT segment files 1990–2004). Industry definitions follow the Input-Output Benchmark Surveys of the Bureau of Economic Analysis. Industry Q in a given year is the median-bounded Q of standalone firms in the industry. Heteroscedasticity-robust standard errors are in parentheses. Standard errors are corrected for clustering at the industry-year level.
p < .05.
p < .01.
p < .001.
Thus, once we control for absolute segment size, we are unable to confirm the result of Bardolet et al. (2011) regarding relative size. We see no aggregate relationship between relative size and normalized capital investment. To better understand the relationships that might be occurring in the data, we continue by performing regression analysis with dummy variables representing the relative and absolute size deciles. In this case, the regression is as follows:
where
Absolute Size Decile Averages
Relative Size Decile Averages
Regarding the relationship between absolute size and investment, we first notice that segments outside the absolute size bottom decile receive less investment than the bottom decile. Second, we also observe a downward trend in absolute size. As a segment’s decile increases, a segment receives less normalized investment relative to the bottom decile. The trend is more pronounced when we use segment fixed effects rather than industry fixed effects. The relationship between investment and relative size is more complicated. The bottom decile of relative size receives the greatest amount of normalized capital investment. However, after accounting for this bottom decile effect, there appears to be an upward trend as higher deciles seem to receive more capital investment than lower deciles. These first approximations suggest at least two interesting patterns. First, as shown in Figure 2, small segments (in absolute terms) receive more investment than large ones. Thus, it will be important that in further analyses we control for absolute segment size to ascertain any relative size effects. Second, Figure 3 suggests that the effect is reversed when looking at relative size. In this case, segment investment gets proportionally larger as the relative size of that segment increases. On the whole, controlling for absolute size, we find that for 90% of segments, the relationship is opposite to what Bardolet et al. (2011) concluded. For most segments, greater relative size actually increases capital investment. The aggregate effect hides the two underlying mechanisms.
The Two-Sided Relative Size Effect
Our preliminary analyses suggest that Hypotheses 1a and 1b may both hold, albeit in different places on the relative size spectrum. We investigate these relationships further in our main regressions, using a version of the typical investment equation proposed in previous studies on capital allocations in multi-business firms (Ozbas & Scharfstein, 2010; Rauh, 2006).
The dependent variable in this regression is the capital expenditure of the focal business segment normalized by its sales. Given that the preliminary analysis of the deciles revealed a specific potential phenomenon in the lowest decile (i.e., larger investment in those units) and a different, more continuous one in the others (i.e., larger investment as relative size increases), we include a separate indicator variable for the bottom decile of relative size (Bottom decile in the equation above) to separate the two effects. Control variables (the
Investment and Relative Size
Note: Regressions of the effect of relative size (in sales) on capital spending divided by sales for segments in multi-segment firms (COMPUSTAT segment files 1990–2004). Industry definitions follow the Input-Output Benchmark Surveys of the Bureau of Economic Analysis. Industry Q in a given year is median-bounded Q of standalone firms in the industry. Heteroscedasticity-robust standard errors are in parentheses. Standard errors are corrected for clustering at the industry-year level.
p < .10.
p < .05.
p < .01.
p < .001.
First, we find a bottom decile effect. Segments in the bottom decile of relative size receive 0.020 to 0.029 additional capital investment/sales. Considering that the average segment in the dataset receives 0.155 capital investment relative to sales, this equates to a 13% to 19% increase in total investment for the average segment. For the average bottom decile business, which has $135 million in capital investment per year, this is roughly $17 to $27 million in additional capital investment per year. This result would seem to offer partial support for Hypothesis 1, that is, that relatively small segments receive proportionally higher allocations, although this only holds for the bottom decile. Second, we also find a relative size effect. The coefficient of relative size is 0.011 to 0.013. To put this value in perspective, if we exclude the bottom decile, the average relative size of a below-median segment is 0.238, while the average relative size of an above-median segment is 0.641 (see Table 1, bottom panel, for both measures). Moving from the average below-median to the average above-median segment (in relative size) is predicted to increase normalized capital investment by 0.004 to 0.005. This equates to a 5% to 7% increase on the average normalized capital investment of a segment. This result offers support for Hypothesis 2, that is, larger segments receive proportionally higher allocations. 7
The other terms in the regression are quite telling. Lagged industry Q and profitability both have significant positive relationships with normalized capital investment. Sales growth and age have little explanatory power. The coefficients are generally the same whether industry or segment fixed effects are used, though the relative size term is not significant when segment fixed effects are used. The loss of significance is not because of a change in the estimated coefficient but because of an increase in the standard error.
Matching Analysis
To check the robustness of these two main effects, namely, additional investment for the smallest and largest segments, matching techniques are used. Matching is a statistical method that allows for the estimation of the hypothesized effect by comparing a “treated” group (for example, segments with relative size above the sample median) and a “nontreated” group (in that example, segments with relative size below the sample median) in a quasi-experiment (Abadies & Imbens, 2011). The goal of the matching procedure is to find, for every unit in the “treated” group, one (or more) nontreated unit(s) with similar if not identical observable characteristics. In our case, by matching treated segments with similar nontreated ones, we establish a comparison of outcomes (i.e., normalized capital expenditures) that tests the regression analysis findings and, moreover, provides an estimate of the effect size. Among other things, we can match on absolute size. This means we can compare two segments of nearly identical absolute size, but with different sizes relative to their firms, to see how this difference affects investment. Following Ozbas and Scharfstein (2010), we use Abadie and Imbens’ (2011) matching procedure, which in our context is more robust and allows us to measure differences in capital spending between unrelated segments and observationally similar standalone firms.
To test for potential overinvestment in bottom-decile relative size segments, segments in the sample are classified within each year according to whether they are in the bottom relative size decile or not. Bottom decile segments are then matched with nonbottom decile segments by exact year and industry, and continuously (i.e., closest available match) by size, age, profitability, and sales growth. The average difference in outcome between the matches is then estimated (see Table 4 for the results). Segments in the bottom decile (controlling for absolute size, age, profitability, and sales growth within each year and industry) get 0.15 to 0.2 additional normalized investment, an increase of 18% to 25% in capital investment compared to the average segment. The coefficient is not much different than the 0.2 to 0.29 value calculated for the initial regressions in Table 3.
Matching Analysis: Lowest Decile vs. Other Segments
Note: Abadie and Imbens’ (2011) bias-corrected estimates of average treatment effect for treated segments with relative size (in sales) in the lowest decile, relative to segments outside of the lowest decile (COMPUSTAT segment files, 1989–2004). Treatment outcome is capital spending to sales ratio. Matching is continuous with respect to sales, age, profitability (cash flow to sales ratio), and sales growth (yearly change in sales over last year’s sales) and exact with respect to industry and year. Number of matches is four. Standard errors are in parentheses.
p < .05.
p < .01.
In order to test for potential overinvestment in relatively large segments, we will follow a similar procedure. Excluding the bottom decile of relative size segments from our analysis (determined the same way as in Table 4), 8 we divide our remaining sample according to whether the segments are above or below the median relative size for each year. These are then matched, exactly by year and industry, and continuously by size, age, profitability, and sales growth. The difference between the matches is then estimated. Table 5 provides the results. Segments with an above-median relative size receive 0.0035 to 0.0045 additional normalized capital investment after controlling for absolute size, age, profitability, and sales growth within each year and industry. This equates to a roughly 5% per year increase in investment for the average segment. Thus, the two main effects we observe in our regressions—a bottom decile and a relative size effect—are robust to matching analysis, with only minor deviations in the estimated coefficient. In the next section, we examine the interactions of these results by relating them to segment profitability and growth.
Matching Analysis: Above-Median vs. Below-Median Segments
Note: Abadie and Imbens’ (2011) bias-corrected estimates for average treatment effect for treated above-median relative size (in sales) segments to control below-median relative size segments (COMPUSTAT segment files, 1989–2004). Segments in the lowest relative size decile were excluded from analysis. Treatment outcome is capital spending to sales ratio. Matching is continuous with respect to sales, age, profitability (cash flow to sales ratio), and sales growth (yearly change in sales over last year’s sales) and exact with respect to industry and year. Number of matches is four. Standard errors are in parentheses.
p < .10.
p < .05.
Effects of Segment Size and Past Performance
Previous literature, as well as managerial practice, recognize that capital allocation decisions are driven by the segment’s future opportunities (Billet & Mauer, 2003; Ozbas & Scharfstein, 2010) or, at the very least, by the segment’s recent performance (Arrfelt et al., 2013; Graham & Harvey, 2001). In the previous sections, we have uncovered evidence that relative size impacts those allocations after controlling for past performance (profitability and sales growth). However, this does not mean that the effects must be constant along those variables. To investigate potential interactions, we study how the relative size effects vary across groups of segments with different combinations of past profitability and growth. In this way, we mimic the categories at the core of the portfolio matrices that managers have been using for decades in their internal investment processes (Bardolet et al., 2010). Thus, we ask questions such as, “Is the relative size effect stronger or weaker when the larger segments are low growth-high profitability? Or is the bottom decile effect consistent across all quadrants of the growth-profitability matrix?”
To that purpose, we calculate sales growth (change in sales from last year to this year/sales last year) and profitability (cash flow/sales) and then classify each segment in our sample as below- or above-median for each of the two variables. We rerun the matching analysis of previous sections for each one of the four cells generated by combining high-low growth and high-low profitability. Table 6 provides the matching analysis for the difference between the bottom decile of relative segment sizes and other deciles. Table 7 provides the matching analysis for below-median versus above-median relative sizes. In this case, matching is continuous with respect to sales, age, profitability, and sales growth, and exact to industry and year.
Matching Analysis: Profitability and Growth for the Bottom Decile Effect
Note: Abadie and Imbens’ (2011) bias-corrected estimates for average treatment effect for treated segments in the lowest relative size (in sales) decile, relative to segments in other deciles (COMPUSTAT segment files, 1990–2004). Treatment outcome is capital spending to sales ratio. Matching is continuous with respect to sales, age, profitability (cash flow to sales ratio), and sales growth (yearly change in sales over last year’s sales) and exact with respect to industry and year. Number of matches is four. Low- and high-profitability bins are based on the annual sample median of profitability. Low- and high-sales growth bins are based on the annual sample median of sales growth. Standard errors are in parentheses.
p < .10.
p < .01.
p < .001.
Matching Analysis: Profitability and Growth for the Relative Size Effect
Note: Abadie and Imbens’ (2011) bias-corrected estimates for average treatment effect for treated above-median relative size (in sales) segments to control below-median relative size segments (COMPUSTAT segment files, 1992–2004). Segments in the lowest decile of relative size were excluded from analysis. Treatment outcome is capital spending to sales ratio. Matching is continuous with respect to sales, age, profitability (cash flow to sales ratio), and sales growth (yearly change in sales over last year’s sales) and exact with respect to industry and year. Number of matches is four. Low- and high-profitability bins are based on the annual sample median of profitability. Low- and high-sales growth bins are based on the annual sample median of sales growth. Standard errors are in parentheses.
p < .10.
p < .05.
p < .001.
Contrary to Hypothesis 2a, which predicts equal effects across all four quadrants, the overall bottom decile effect observed in the data seems to be driven by low profitability segments, regardless of growth. For high-profitability segments, the effect seems to be reversed, especially for high-profitability, low-growth segments, with the bottom decile actually receiving less normalized investment than other deciles. This result is only suggestive—due to a high standard error, it is only significant at the 10% level. Taken together, the results in Table 6 suggest that firms have a special inclination to invest in their smallest segments when these show below-average profitability and slightly underinvest in them when they show above-average profitability. Perhaps corporate managers are willing to tolerate lower profitability out of these relatively small segments because they have less effect on the firm’s bottom line. Along the same lines, the low profitability of a relatively small segment will have less of an effect on a firm’s balance sheet than a relatively large segment, and may be more likely to be excused.
There are also great differences in the magnitude and sign of the relative size effects when isolated to specific growth and profitability types. Consistent with Hypothesis 2b, the overall increase in investment for segments of large relative size appears to be driven entirely by low-growth, high-profitability segments. The increase in investment for this type of segment is 0.021, about 5.5 times the coefficient for the overall effect. This translates into a 25% increase in investment for the average segment. Conversely, for low-growth, low-profitability and high-growth, high-profitability segment types, the difference within the quadrant is negative. In these cases, relatively small segments would receive additional investment relative to their larger counterparts. When only examining low-profitability, high-growth types, there is little difference in investment between relatively large- and small-size segments.
It is interesting to note that this type of segment (low growth, high profitability) is the only type not to show a bottom decile effect and is the type that drives the relative size effect. However, because all these comparisons are done within quadrant, our data analysis to this point has not indicated whether these relatively large-size, low-growth, high-profitability segments are favored by management over any other type of segment out of quadrant. To provide some suggestive answers to this question, Figure 4 displays normalized net cash outflows by quadrant for each type of segment (the bottom decile is included). Management generally takes cash out of both high-profitability quadrants and generally is more likely to take from relatively large-size segments. 9 However, the low growth-high profitability quadrant is the only one where there is a distinct advantage for being relatively large. This finding is consistent with Hypothesis 2b and our preceding analysis.
Normalized Net Cash Outflow Within Corporate Segments by Growth/Profitability Quadrant
Managerial Ownership
Hypothesis 3 makes a prediction about the moderating effect that occurs when the top five managers have a large stake in the company. Perhaps when managers have greater ownership in a multisegment firm, they are less susceptible to size-related biases. Our results in Tables 8 and 9 suggest that this may be the case. For each year, we identify the median management ownership of the firm’s top five managers—as reported in the ExecuComp COMPUSTAT database—and rerun the matching specifications found in Table 4 for segments of firms with below-median ownership, as well as for segments of firms with above-median ownership. Table 8 shows the results of this analysis for the bottom decile effect. Though the differences are not significant, segments in firms with low management ownership exhibit a more pronounced bottom-decile effect than segments with high management ownership. This is a potentially interesting finding that suggests that managers with large stakes in the company are less prone to naively diversify their allocations and, thus, favor the bottom decile. Of course, we must caution the reader that our measure of managerial ownership is crude, and there may be better measures. 10
Matching Analysis: Managerial Ownership Effect on the Lowest Decile
Note: Abadie and Imbens’ (2011) bias-corrected estimates for average treatment effect for treated bottom relative size (in sales) decile segments to control other segments (COMPUSTAT segment files, 1992–2004). Treatment outcome is capital spending to sales ratio. Matching is continuous with respect to sales, age, profitability (cash flow to sales ratio), and sales growth (yearly change in sales over last year’s sales) and exact with respect to industry and year. Number of matches is four. Low- and high-management bins are based on the annual sample median of management ownership. Standard errors are in parentheses. Comparisons between low- and high-management bins assume independence of estimated average treatment effects.
p < .10.
p < .05.
p < .01.
Table 9 presents a similar analysis for the relative size effect. For below-median ownership firms, the coefficients of the effect are generally positive, indicating a greater investment in relatively large-size segments. However, for above-median ownership firms, the effect is reversed, indicating an overinvestment in smaller-size segments. This would offer support for Hypothesis 3, as top managers with high ownership would be less willing to engage in political games and, more importantly, would not let their allocation decisions be influenced by power. On the other hand, only one of the specifications in Table 9 shows significant coefficients and differences, so this evidence is only suggestive.
Matching Analysis: Managerial Ownership Effect on Relative Size
Note: Abadie and Imbens’ (2011) bias-corrected estimates for average treatment effect for treated above-median-relative-size (in sales) segments to control below-median-relative-size segments (Compustat segment files, 1992–2004). Segments in the lowest decile of relative size were excluded from analysis. Treatment outcome is capital spending over sales ratio. Matching is continuous with respect to sales, age, profitability (cash flow over sales ratio), and sales growth (yearly change in sales over last year’s sales) and exact with respect to industry and year. Number of matches is four. Low- and high-management bins are based on the annual sample median of management ownership. Standard errors are in parentheses. Comparisons between low- and high-management bins assume independence of estimated average treatment effects.
p < .10.
Finally, one robustness check on our data analysis so far would be to examine whether the preceding results still hold for different types of spending (e.g., marketing, and research and development). These types of spending are conceivably different than pure capital investment, and businesses of different relative sizes might have different needs.
Unfortunately, our dataset is limited in its number of observations: a little less than 1% and 10% of the businesses in our sample had marketing and research expenditure data, respectively. We also do not know the reasons behind data availability, so the usual caveats about selection apply. That being said, we find a bottom decile effect for both measures that is equal or greater to what we observe for capital. Interestingly, we do not seem to find a positive size effect. There appears to be a negative correlation between normalized research and marketing expenditures in the other deciles. 11 One conjecture is that political power is more influenced by relative size which fluctuates much less year on year than marketing or research and development expenditures. Furthermore, the latter influences the current year’s profitability, whereas capital expenditures are depreciated over numerous years and impact the current year’s earnings much less. Given the issues mentioned above, we cannot say definitively if this relationship is meaningful. We do think it would be an interesting topic for future research if complete data were available. We refer our reader to Appendices B, C, and D for the full analysis.
Discussion
This study partially contests a previous finding on relative size and the use of the 1/n heuristic in corporate capital allocations. It found that effect only in the bottom decile of relative size segments in a corporation. There is no evidence for the effect in other deciles. If it exists, it is swamped by a contradictory effect—the positive correlation of relative size and normalized capital investment. This latter effect is consistent with the idea that executives in large segments have greater ability to influence their allocations. Our results point at an interesting combination of phenomena affecting allocations across segments of different relative sizes. On one hand, the political power hypothesis laid out above is supported, since capital allocations increase with relative segment size. On the other hand, this does not happen to the smallest of segments within corporations, where the 1/n heuristic seems to dominate. Taken together, these opposite pieces of evidence point to a combination of cognitive and political factors that lead to the favoring of extreme relative size types within corporations.
Furthermore, our analysis of the moderating effect of growth and profitability indicates that “larger allocations for larger units” is a phenomenon mostly driven by the low-growth, high-profitability segments in our sample. Moreover, the relative size effect decreases in companies where top management possess a high percentage of shares. This result is consistent with Ozbas and Scharfstein’s (2010) finding that corporate socialism is more common when management has a small equity stake, as both results suggest that weak monitoring and misaligned incentives contribute to poor investment choices.
On a different note, our results might seem a complementary explanation to Maritan’s (2001) finding that investments in existing capabilities tend to follow standard bottom-up resource allocation processes, while investments in new capabilities do not. In the latter case, it is the top managers who take a more direct role in allocation decisions. If one assumes that relatively large units contain most of the firm’s existing capabilities and that relatively smaller units contain most of the developing ones, then our results and Maritan’s mirror each other. However, our data are only suggestive; the data are not fine-grained enough to draw a conclusion. Moreover, Coen and Maritan (2011) developed the idea of resource allocation being influenced by the asset stock or capability existing endowment of the business unit. Again, if one considers that size is positively correlated with that asset stock, our results are consistent with Coen and Maritan’s findings—that relatively large units are favored by the firm’s traditional resource processes.
There are a number of promising directions for future research that are suggested by our results. First, although we document an empirical regularity in corporate resource allocation that is consistent with a bias toward even allocation, it would be useful to more directly investigate the cognitive and social factors that result in a firm’s smallest units receiving relatively large investment. Second, the COMPUSTAT segment data on which we relied has a number of well-known limitations, such as the sometimes inconsistent reporting of segment information by firms. Although we took steps to minimize distortions, there remains some inherent measurement error. Alternative databases that provide more accurate measurements of organizational structure would help test our findings.
Third, it would be interesting to investigate the dynamic aspect of these size effects. A preliminary examination of our sample reveals extremely high serial year-to-year correlations in business units’ shares of a company’s total capital allocations, which suggests that corporations are very slow in changing the allocation balance among their business units. This organizational inertia can impair a firm’s ability to redistribute capital appropriately in response to strategic challenges, and ultimately hinders performance. One possible cause of such inertia could be related to the size effects described here, as larger units would use their influence to maintain their allocation advantages over time. The case of Microsoft, a company where one dominant business unit often absorbed the resources that smaller new ventures needed (Brass, 2010), anecdotally illustrates this behavior. The fact that we find this larger relative size effect mostly concentrated among high-profitability, low-growth units suggests that, over long periods of time, these units might be starving smaller, higher-growth units of resources. Further research on this issue could help identify the distributions of relative size among units that better foster firm growth in the long run.
Finally, it would be interesting to investigate other competitive and organizational factors that moderate the magnitude of the relative size effects documented in this article. For instance, in Rajan et al. (2000), inefficient capital transfers increase when segments are more diverse in their total investment opportunities. The socialism-as-optimal-mechanism model of Bernardo, Luo, and Wang (2006) encompasses the above predictions and further suggests that socialism increases with firm maturity and the degree to which the division manager’s job requires firm-specific human capital. Furthermore, Lamont (1997) showed that firms might also be socialistic when times are tough, cutting investment across the board, even in the business units for which the set of investment opportunities remains unchanged. Although the evidence presented in this article shows that the relative size effects are robust across different cross-sections of our sample, it would be interesting to identify factors that tend to moderate (or exacerbate) those effects.
One potential implication of this study for executives is that if the finding on segments holds for business units—that is, the largest and smallest are favored in corporate allocations—a company seeking allocation efficiency should stay away from having very large or very small segments, especially if those do not contain the best investment prospects.
Further studies of resource allocation are needed, as resource allocation is one of the most understudied areas in the economy. One estimate of the amount of capital allocated by internal capital markets ($640 billion per year) is larger than the amount allocated by capital markets in the U.S. ($621 billion a year; Guedj, Huang, & Sulaeman, 2009). However, far less academic literature has addressed the issue of these internal capital markets. Thus, the relative attention on corporate allocators is small relative to their impact on the economy. Part of the reason for this is, certainly, data availability. Thus, it is important that academics take the time to cultivate new and more fine-grained data on resource allocation.
Footnotes
Appendix
Main Regressions Using Research and Marketing Spending as Dependent Variables
| Dependent Variable | Research Spending/Sales | Reseach Spending/Sales | Marketing Spending/Sales | Marketing Spending/Sales |
|---|---|---|---|---|
| Bottom decile of relative size | 0.02** | 0.02* | −0.02 | 0.13*** |
| (0.01) | (0.01) | (0.09) | (0.04) | |
| Relative size | −0.02**** | −4.15E-03 | −0.62** | −0.03 |
| (0.01) | (0.01) | (0.24) | (0.16) | |
| Lagged industry Q | 3.41E-03 | 1.79E-03 | 0.03 | −0.50E-04 |
| (3.17E-03) | (2.27E-03) | (0.02) | (9.51E-03) | |
| Profitability | 0.04** | 0.03 | 0.19 | −0.17 |
| (0.02) | (0.03) | (0.22) | (0.22) | |
| Age | −2.34E-05 | −7.77E-04*** | −0.01 | 3.18E-03 |
| (2.12E-04) | (2.75E-04) | (7.09E-03) | (3.71E-03) | |
| Sales growth | −1.42E-03 | 7.95E-03** | −0.09 | −5.61E-03 |
| (1.71E-03) | (3.93E-03) | (0.05) | (0.01) | |
| Industry fixed effect (F.E.) | Yes | No | Yes | No |
| Segment F.E. | No | Yes | No | Yes |
| Absolute size decile F.E. | Yes | Yes | Yes | Yes |
| Year F.E. | Yes | Yes | Yes | Yes |
| Observations | 987 | 987 | 108 | 108 |
| R-squared | .68 | .94 | .86 | .98 |
Note: Segments in the bottom decile of relative size (in sales) compared to other segments in multisegment firms (COMPUSTAT segment files, 1990–2004). The general effect of relative size is also measured. The dependent variable is capital spending over sales. Industry definitions follow the Input-Output Benchmark Surveys of the Bureau of Economic Analysis. Profitability is measured as cash flow over sales. Industry Q in a given year is the median-bounded Q of standalone firms in the industry. Sales growth is the difference between current and previous year’s sales divided by previous year’s sales. Industry Q in a given year is the median-bounded Q of standalone firms in the industry. Heteroscedasticity-robust standard errors are in parentheses. Standard errors are corrected for clustering at the industry-year level.
p < .10.
p < .05.
p < .01.
p < .001.
Acknowledgements
The authors would like to thank Craig Fox, Richard Rumelt, and the Australian Research Council, Discovery Grant DP160102290.