Internal Capital Markets and Managerial Power: Implications for Family Groups

Transcription

1 Internal Capital Markets and Managerial Power: Implications for Family Groups March 2014 Markus Glaser Munich School of Management Florencio Lopez-de-Silanes EDHEC Business School Zacharias Sautner University of Amsterdam

2 This working paper reports some of the main results of our paper Opening the Black Box: Internal Capital Markets and Managerial Power, Journal of Finance, LXVIII (4), August But it also presents additional findings that may be useful for family firms and firms with a large controlling shareholder and whose management structure across branches is similar to that of a classic conglomerate or multi-segment corporation. 2 EDHEC is one of the top five business schools in France. Its reputation is built on the high quality of its faculty and the privileged relationship with professionals that the school has cultivated since its establishment in EDHEC Business School has decided to draw on its extensive knowledge of the professional environment and has therefore focused its research on themes that satisfy the needs of professionals. Copyright 2014 EDHEC

3 Executive Summary Capital allocation is a central issue for all firms, but deserves particular attention for family firms as their access to external resources is typically lower putting further pressure to optimally allocate their available capital among competing business units. Although the theory on how firms should allocate their capital has made progress, empirical evidence is still scant. How do firms allocate resources across business units? Do units with better investment opportunities receive larger capital allocations and invest more? Do units run by more powerful and better connected managers receive larger capital allocations? The theoretical literature on internal capital markets can be organised according to two broad views. The first view holds that capital allocation is the result of pooling internally generated cash flows and subsequently distributing funds according to the unit s investment prospects. In contrast, a second view warns about the risks that units run by more powerful or better connected managers may get allocations larger than what is justified by their investment opportunities. In our recent research we have been able to empirically disentangle these two perspectives using a unique dataset of planned and actual capital allocations across business units and measures of power and connections of business-unit CEOs (i.e., personal profile and career at the firm, networking activities, and connections to executive management). Our main contribution is to empirically document whether managerial power and connections make a difference to internal capital allocations, and if so, under what circumstances. We use direct evidence from a newly constructed dataset drawn from the internal accounting system of a large multinational conglomerate and containing information about planned and actual capital allocations to its 20 business units over six years. We complement these data with a second database of formal and informal measures of managerial power and connections for the 40 different business-unit CEOs working at the conglomerate over the sample period. To our knowledge, these are the first such datasets used in the internal capital markets literature. There are three main results that emerge from our findings. First, our access to internal firm documents helps us open up the black box of internal capital markets and provides a detailed picture of the stages of the allocation process and the role played by unit managers. When we compare approved planned and actual capital allocations, we find that managers across the firm use the standardised budgeting process to build buffers into their budgets. These empirical findings, documented for the first time for a large conglomerate, corroborate previous research based on interviews and labouratory experiments. Our detailed analysis of the capital allocation process should be useful to firms that need to reinforce internal budgeting decisions. Second, we show that, although all unit managers try to use their excessive capital budgets to justify additional spending, units run by more powerful and better connected managers obtain higher actual capital allocations at times of financial slack after unexpected cash windfalls to the firm. We find that more powerful unit managers, on average, increase their actual investments 74% more than their less powerful peers. It seems that the approval of investments is easier to come by for powerful and better connected managers when the firm is awash in cash. Third, we provide evidence on the key question of whether the investments following the cashwindfall allocations were good (efficient) or bad (inefficient). We show that the effects of power on internal capital allocation reflect inefficient resource allocations and do not lead to improved unit performance. The evidence is generally consistent with bargaining-power theories, which posit that capital allocations are partly based on power and connections, and with empirebuilding theories, as the larger allocations to powerful managers do not lead to improved performance. 3

4 These findings, although generalisable to all multi-unit firms, seems particularly pertinent for family groups for several reasons. First, as family firms grow and expand across sectors, they typically adopt a group structure similar to that of a multi-unit firm with the family holding in the middle. Indeed, about 20 percent of listed firms belong to family-controlled multi-unit business groups. Second, due to the large concentration of power and control rights among various family members, power and connections are likely to be very important in family firms. It is also common that some unit managers are part of the family or have strong ties with the family, while others do not. In Italy, for example, over 60 percent of the managers are family members. These structural arrangements may create large differences in bargaining power among business unit managers. Finally, family firms have been shown to have less access to external funding sources and may therefore need to be particularly careful about capital allocation. A better understanding of the forces behind capital allocation seems thus useful for firms with limited access to capital. We can think of three groups of implications for family firms. The first set of recommendations refers to the process of internal capital allocation itself. Recent findings show that family firms have less formalised procedures and are more likely to exhibit ineffective internal controls. There is also evidence that they employ less sophisticated capital budgeting techniques when they do not involve outside directors. For these reasons, family groups need to reconsider their processes and try to establish more robust capital budgeting procedures. The need to establish a formalised and homogeneous capital allocation system in family groups is reinforced if we take into account the recent capital budgeting literature findings about the potential misuse of the system by firm CEOs. This literature shows CEOs try to use budgets as a bridge-building tool to elicit cooperation and build alliances with business unit managers, or use their special authority in capital allocation decisions to reject selected projects that do not conform to their biases and goals. To the extent that family owners wield strong power and clout, we could fear that all of these arguments may be frequently used to justify capital allocations that do not necessarily respond to investment opportunities. The detailed capital budgeting procedures illustrated in detail in our recent paper could be used by family firms as a blue print to reform their capital allocations processes. The second set of implications for family firms is related to the relevance of power and connections during cash windfalls. We show that even with detailed and careful procedures for regular capital allocations, powerful managers find ways to use their connections during irregular or ad-hoc cash shocks to boost their capital allocations. This evidence calls for the formalisation of the capital allocation process particularly at times of financial slack. The establishment of rules on how to proceed when cash windfalls arise is essential for family firms which are more likely to lack formalised procedures for such periods. The third and final set of implications of our findings for family firms relates to mechanisms to mitigate inefficient capital allocations. Firms should consider the use of policies to keep capital budgeting in check with productivity. Our results help explain the rationale for some managerial policies observed in certain firms, such as automatic job rotation, headquarters audits, and the use of incentive compensation as a disciplining devise to promote unit managers more truthful reporting of the quality of the projects. All of these measures should bring additional benefits for family firms with less transparent processes. Overall, to the extent that the management teams of family firms and family groups are structured in such a way that differences in power and connections may become more important, our findings are of particular use for such firms as they strive to improve their capital allocation and other internal processes. 4

5 I. Introduction Capital allocation is a central issue for all firms, but deserves particular attention for family firms as their access to external resources is typically lower putting further pressure to optimally allocate their available capital among competing business units. Although the theory on how firms should allocate their capital has made progress, empirical evidence is still scant. How do firms allocate resources across business units? Do units with better investment opportunities receive larger capital allocations and invest more? Do units run by more powerful and better connected managers receive larger capital allocations? In this paper, we seek to answer these questions with direct evidence from a newly constructed dataset drawn from the internal accounting system of a large multinational conglomerate and containing information about planned and actual capital allocations to its 20 business units. Our analysis, although generalisable to all multi-unit firms, seems particularly pertinent for family groups for several reasons. First, as family firms grow and expand across sectors, they typically adopt a group structure similar to that of a conglomerate with the family holding in the middle. Indeed, analysing data for 45 countries, Masulis, Pham and Zein (2011) and find that about 20 percent of listed firms belong to family-controlled business groups. Therefore, the structure of the conglomerate analysed in this paper is very similar to that of large family firms in multiple industry segments. Second, due to the large concentration of power and control rights among various family members (La Porta et al, 2000), power and connections are likely to be very important in family firms. So, the connections of business-unit managers with the family become of great importance. It is also common that some unit managers are part of the family or have strong ties with the family, while others do not. Sciascia and Mazzola (2008) and Mazola, Sciascia and Kellermanns (2013) document that in Italian firms over 60 percent of the managers are also family members. These arrangements may thus create large differences in bargaining power among business unit managers which could substantially impact the allocation of capital across business units. Finally, family firms have been shown to have less access to external funding sources and may therefore need to be particularly careful about capital allocation. A better understanding of the forces behind capital allocation seems thus particularly useful for firms with limited access to capital and external funding. The modelling of internal capital markets has made substantial progress. 1 We can organice the theoretical literature on internal capital markets according to two broad views. The first view emphasises the bright side. This view holds that capital allocation is the result of pooling internally generated cash flows and subsequently distributing funds optimally to units (Weston (1970), Williamson (1975), Matsusaka and Nanda (2002), Maksimovic and Phillips (2002)). In these models, capital allocation is determined mainly by the unit s investment prospects. Through winner-picking methods, internal capital markets add value, as the firm makes larger allocations to units with greater investment opportunities (Stein (1997)). In contrast, the second view highlights the dark side of internal capital markets. The proponents of this view argue that units run by more powerful or better connected managers may get allocations larger than what is justified by the investment opportunities they provide (Meyer, Milgrom, and Roberts (1992), Scharfstein and Stein (2000), Rajan, Servaes, and Zingales (2000), Wulf (2009)). In these bargaining-power models, unit CEOs prefer larger capital allocations (for rent-seeking or empire-building reasons) and wield influence in an attempt to obtain more funds for their units. Influence activities are costly and inefficient because resources are spent trying to affect allocations and because the resulting investment decisions may not be optimal. Some inefficient internal capital markets models focus on the firm CEO as the source of the agency problem. These models predict that CEOs try to allocate more capital to units from which they can obtain higher private benefits of control (Stulz (1990)), expect future favours, or help them get entrenched (Shleifer and Vishny (1989)). 1 - Stein (2003) and Maksimovic and Phillips (2007) provide comprehensive reviews of the theoretical literature. 5

6 Although there has certainly been theoretical progress in the modelling of internal capital markets, empirical evidence is scarce. The current situation is similar to that in the literature on internal labour markets, described by Baker and Holmstrom (1995, p. 259) as being in a phase of too many theories, too few facts and in need of studies of personnel records, supplemented by interviews and institutional facts. As in that literature, the understanding of the workings of capital allocation in firms need to look within the corporation to understand governance structures and their connection to resource allocations. 2 In this paper, we do precisely that: we look inside the firm and put together new data for a multinational conglomerate to analyse its capital allocation process and the means by which more powerful and better connected business-unit managers use this process to obtain larger allocations following cash windfalls. Our approach builds on the previous work by Bower (1970), who used interview transcripts and internal memos of four corporations to better understand the allocation of internal resources. He describes capital allocation within firms as a political process and argues that top management must manage its influence on [such] political processes (p. 305). Our main contribution is to empirically document whether managerial power and connections make a difference to internal capital allocations, and if so, under what circumstances. Moreover, we show that the effects of power on internal capital allocation reflect inefficient resource allocations. As we argue in the conclusion, these findings are of particular relevance for family firms whose access to capital markets may be scarcer. Our analysis relies on two new databases. The first contains five years of quarterly data on planned (i.e., budgeted) and actual (i.e., realised) capital allocations for each of the firm s 20 business units, which operate under five separate divisions. We also have access to quarterly data on planned and actual R&D and marketing expenditures, as well as on assets, sales, EBIT, and personnel. To our knowledge, this is the first such dataset used in the internal capital markets literature. Most of the variables in our dataset are rarely available even at divisional level. 3 An important advantage of our data is that they allow us to analyse individual business units. As in most other conglomerates, the divisions of the firm lack operating activities themselves and act mostly as organisational umbrellas for the underlying business units. Business units are thus the real centres of economic activity in the typical conglomerate: they originate budget requests, receive capital allocations, make investments, and carry out production. We complement these data with a second database of formal and informal measures of managerial power and connections for the 40 different business-unit CEOs working at the conglomerate over the sample period. An initial set of measures is based on management profile data to proxy for managers careers at the firm, their social network at the workplace, and the similarity of their personal profiles to that of the CEO. We also take a survey of business-unit CEOs to put together a second set of measures. The survey, which contains both self-constructed questions and questions previously used in the management and sociology literature, makes it possible to construct proxies that capture more informal aspects such as a unit manager s networking activities and his or her connections to executive management. With these data, we provide some of the first empirical evidence on three main fronts. First, our access to internal firm documents helps us open up the black box of internal capital markets and provides a detailed picture of the stages of the allocation process and the role played by unit managers. When we compare approved planned and actual capital allocations, we find that managers across the firm use the standardised budgeting process to build buffers into their budgets. These empirical findings, documented for the first time for a large conglomerate, corroborate previous research based on interviews and labouratory experiments. 4 Our detailed analysis of the capital allocation process should be useful to firms that need to reinforce internal budgeting decisions Schoar (2002) and Maksimovic and Phillips (2002, 2007). 3 - Segment reporting rules such as US-GAAP and IFRS require firms to report divisional information only on assets, sales, liabilities, depreciation, income, and capital expenditures. In practice, firms do not publish data by business unit and do not provide information on their budgets or investment plans. 4 - Dunk and Nouri (1998) provide a review of this literature.

7 Second, we show that, although all unit managers try to use their excessive capital budgets to justify additional spending, units run by more powerful and better connected managers obtain higher actual capital allocations at times of financial slack in the firm. To measure financial slack, we use the method of Blanchard, Lopez-de-Silanes, and Shleifer (1994), who analyse firms behaviour after unexpected cash windfalls. During our sample period, the conglomerate experiences eight substantial cash windfalls resulting from headquarters sale of equity stakes in other companies whose lines of business are largely unrelated to those of the units. 5 In quarters in which there are cash windfalls, actual investment rates increase across all units and a large share of the variation in this increase is accounted for by the power and connections of the business units CEOs. For all indices of power and connections, we find that more powerful unit managers, on average, increase their actual investments 74% more than their less powerful peers. An explanation consistent with our findings is that approval of investments is easier to come by when the conglomerate is awash in cash and planned budgets are not yet fully spent. Such approval is more likely for units run by more powerful and well-connected managers. 6 Third, and finally, we provide evidence on the key question of whether the investments following the cash-windfall allocations were good (efficient) or bad (inefficient). It is possible that, if the budgeting process is unbiased, power and connections may help overcome budgeting constraints emerging from information asymmetries or the use of hurdle rates. Alternatively, in the absence of budget constraints, differences in managerial power may prevent the cash windfalls from being channelled to the units where the additional resources would be most valuable, ultimately leading to overinvestment by powerful managers. To address this question, we develop tests that exploit the uniqueness of our dataset and use approaches from the internal capital markets literature. 7 Our evidence seems consistent with the view that the larger allocations of cash windfalls to more powerful managers are not efficient and do not lead to improved unit performance. These results support the dark-side view of internal capital markets and suggests that cash windfalls may be an important source of misallocation of capital. The paper is organiced as follows. Section I presents in detail the internal capital allocation process and the financial characteristics of the conglomerate, as well as the planned and actual capital allocation data across business units. Section II presents the evidence of the effects of power on the allocation of windfalls. Section III provides alternative explanations and extensions. Section IV analyses the efficiency of the internal capital allocation. Section V concludes with three sets of implications of our results for family groups and proposals for reforms of their internal systems to try to avoid the dark side of internal capital markets in these firms. II. The Conglomerate and its Capital Allocation A. Organisational Structure and Financial Characteristics The firm of our analysis is a major publicly traded international conglomerate with business units located on two continents. It has production plants in more than 15 countries on four continents and more than 100,000 employees. The organisational structure of the firm takes the multidivisional M-form 8 and the conglomerate operates with a headquarters unit, five product divisions, and 20 business units (see Figure 1). The firm was family-run for several decades but the family now exercises its control through ownership of shares with a professional management team un-related to the family in place. Headquarters coordinates central corporate functions, 5 - Our empirical work shows that planned capital expenditures are not related to the cash windfalls, suggesting that they are indeed not anticipated in the budgets. Although the business units did not consider the windfalls in their previously submitted and approved planned capital budgets, some of the additional cash is available for their investments ex-post. 6 - Our methodology focuses on how power and connections affect the allocation of abnormal cash windfalls, as the relation between regular allocations and power is complicated by the fact that control over extensive resources could be a source rather than a result of power. An additional advantage of our method is that it uses the difference between actual and planned investment as the dependent variable, thereby controlling for the investment opportunities of the units, which should be reflected in the planned capital budgets. Our results are robust to controlling for the reputation or past performance of unit managers, managerial ability, the size of units, and the number of units per division. We also show that they are not driven by endogenous allocations of powerful managers to the units with the best investment opportunities. 7 - Servaes (1994, 1996), Shin and Stulz (1998), Rajan, Servaes, and Zingales (2000), and Ozbas and Scharfstein (2010). 8 - The M-form structure involves the creation of semi-autonomous operating divisions (mainly profit centres) organiced along product, brand, or geographic lines (Williamson (1981, p. 1555)) and was pioneered in large format by the U.S. conglomerates Du Pont and Sloan. 7

8 specifically corporate strategy, internal capital allocation, outside financing, M&A, marketing, investor relations, and legal affairs. It also holds and manages equity stakes in corporations outside the conglomerate. The board of directors, chaired by the CEO, has ultimate responsibility for the firm. The firm did not change its CEO during the sample period. Figure 1. Organisational Structure of the Conglomerate. This figure summarises the organisational structure of the conglomerate. It shows the five divisions and the 20 business units operating within the divisions. The divisions, which have no operating activities themselves, are run by division CEOs. The business units are the operating units and they are run by business-unit CEOs (BU CEOs). The firm does not operate division boards. The five product divisions beneath headquarters have no separate operating activities and assets themselves and act as umbrellas under which the business units operate. The divisions cover five distinct industry sectors and are run by division CEOs, who coordinate the activities of the business units. As in most conglomerates, business units are the real centres of economic activity and business-unit CEOs are responsible for investment, production, and sales. Business units hold all the assets of the divisions and all units have operations beyond the borders of the country where the firm is headquartered. The divisions and their business units have no access to the external capital market and hence cannot raise debt or equity. The division and business-unit CEOs are not part of the board of directors and the firm does not operate division boards. There was no substantial change in the compensation system of the firm during our sample period. To ensure that the conglomerate is not an outlier, we compare a set of its annual financial variables with those of other large conglomerates. As comparison firms, we look at all nonfinancial conglomerates in the Dow Jones 30 and the Euro Stoxx 50. We do so also to show that our firm is unlikely to be financially constrained. Table I shows that our conglomerate s aggregate investment rate is similar to those of the median conglomerate in the Dow Jones 30 and the lower-end conglomerates in the Euro Stoxx 50. Apart from having relatively low leverage and high cash holdings, the conglomerate also has a relatively high dividend payout ratio. 9 Throughout the period of analysis, there is no indication that the firm is financially constrained or in need of cash to finance investment opportunities. As argued in Hovakimian and Titman (2006), these considerations are important, as the cash windfalls, which we exploit in the next section, could be considered not exogenous to the business units investments if the firm were financially constrained The data on annual capital expenditures and total assets are calculated at the consolidated group level, which includes the 20 business units as well as some activities outside the business units (e.g., group-wide activities at the headquarters or joint ventures). The figures reported for the comparison conglomerates are also calculated for at the consolidated group level. The investment rates may differ from those reported in the tables below because the latter are calculated using data of the business units only. Moreover, we use quarterly investment rates in the tables below For a further comparison, we calculate the firm s conglomerate discount. Using the method of Berger and Ofek (1995), we find that the conglomerate discount is 15% over the sample period. This figure is comparable to those documented for other large conglomerates (Berger and Ofek (1995), Lang and Stulz (1994), Hoechle et al. (2012), Glaser and Müller (2010)). In fact, our conglomerate matches the 13 15% discount found for the average US conglomerate in Berger and Ofek (1995). We will later also look at other measures that attempt to assessing the efficiency of the firm s internal capital market (e.g., Shin and Stulz (1998) or Ozbas and Scharfstein (2010)).

9 Table I: Financial Characteristics of the Conglomerate This table shows annual financial ratios of the conglomerate and compares these ratios with those of non-financial conglomerates in the Dow Jones 30 and Euro Stoxx 50 indices. Conglomerates are all firms from the indices with at least two divisions in different two-digit SIC codes. The reported ratios are the averages of annual data over the period from 2002 to For the conglomerates in each of the indices, we also report the median, standard deviation, and the 25th and 75th percentiles. The data on capital expenditures and total assets of the conglomerate are calculated for the consolidated group level, which includes the 20 business units as well as some activities outside the business units (e.g., group-wide activities at the headquarters or joint ventures). The figures reported for the comparison conglomerates are also calculated at the consolidated group level. The investment rates of the conglomerate may differ from those reported in the tables below because the latter are calculated using data of the business units only. Moreover, we use quarterly investment rates in the tables below. Detailed definitions of each variable in the table are provided in Appendix A. B. Internal Capital Allocation: Process and Data To better understand the mechanisms of capital allocation we gained access to internal documents describing the internal budgeting, allocation, and execution processes, and to detailed data on planned and actual allocations. With these documents, and several interviews of managers involved in the budgeting process, we are able to put together the details of the conglomerate s internal capital allocation process. The details are described in Internet Appendix IA.I. The organisational and decision processes are similar to what has been documented for other conglomerates in the literature (see Internet Appendix IA.II). The internal capital allocation process itself consists of two stages which are depicted in Figure 2. In the budgeting or planning stage, the firm develops two documents for each business unit: the strategic outlook and the annual capital allocation plan. The strategic outlook is a threeyear plan and includes general targets for planned investments for all units. The elabouration of the strategic outlook is highly institutionalised and structured. In January and February of each year, business-unit CEOs begin identifying long-term investment opportunities, which are then negotiated with their division CEOs. They then jointly present the plans for their units to the board of directors and negotiate revisions and adjustments. In April, the board makes a final decision on the figures in the strategic outlook. The second phase of budgeting consists of the elabouration of a detailed one-year annual capital allocation plan for each business unit, containing resource allocations, investment budgets, balance sheets and income statements. Starting in June, the division CEOs draw on the strategic outlook to prepare investment targets for the coming year. In July, business-unit and division CEOs negotiate these allocation plans, with preliminary decisions being made in August. They then present the capital allocation plan for negotiation to the board of directors. Finally, in October or November, the board decides on the investment budgets for the coming year. As the divisions have no operating function, the board allocates resources directly to the business units, without separate divisional allocations. 9

10 Figure 2. Decision Process behind Allocation of Capital: Budgeting and Realisation Stages. This figure presents the details of the capital allocation process of the conglomerate. The sources for this figure are internal company documents on the internal budgeting, allocation, and execution process, complemented by interviews with executives and controllers from the firm. Panel A shows the budgeting (i.e., planning stage), and Panel B shows the realisation (i.e., execution) stage. The budgeting stage consists of two different phases: the strategic outlook phase and the annual capital allocation phase. The figure shows how these two phases relate to each other, who the parties are, and when what kinds of decisions are made. A detailed verbal description of the capital allocation process is provided in Internet Appendix IA.I. 10 It is during the realisation or execution stage that concrete investments are made. This final stage of the process is shorter and less formalised, leaving more room for discretion and favours. Decisions here are made much more quickly than during budgeting. All investments below a specific threshold can be performed at the discretion of the business units. Although the annual capital allocation plan is the basis for investments, the firm requires additional approvals for the execution of any project in the plan which involves investments that exceed a specific threshold.

11 In such a case the business unit must prepare a memo, for approval by the division CEO and the board, showing that the investment will generate at least a positive NPV. Although important, the NPV is not the only relevant approval criterion. In fact, other more loosely defined criteria, such as the strategic fit of a project, its cash-flow structure, or the past performance and reputation of the business-unit CEOs behind the project, also play an important role. Taking into account such a range of aspects is not specific to our firm. The survey data of Graham, Harvey, and Puri (2011) confirm that actual allocation decisions in many firms are also based on cash-flow timing, market share, previous returns, and even on manager reputation, manager confidence, and gut feel. The process described above provides the basis for a better understanding of our capital allocation data. It covers quarterly data on planned allocations and monthly data on actual capital allocations for each of the 20 business units from January 2002 to December The planned allocation data are taken from electronic files containing the quarterly figures produced annually in the second phase of budgeting. Our realised allocation data are taken from electronic files based on actual reporting by the business units. Table II presents summary statistics (means) for all 20 business units; all variables are calculated based on quarterly observations. 11 The exact definitions of all variables are presented in the Appendix to the paper. To compare data for planned and actual allocations, the table includes for each variable only the observations that correspond to an exact match between the two forms of data. The table also reports whether differences between planned and actual data are statistically different from each other. The comparisons permit two particular observations. First, with the exception of sales over assets, which seem to be overestimated in several units, actual EBIT and sales growth are not significantly different from planned values in most units. 12 Second, and most importantly for our analysis, planned capital expenditures seem to be above actual capital expenditures for all business units, with the difference being statistically significant in 11 of the 20 units. None of the units seems to under-budget capital expenditures. These data suggest that business-unit managers use the standardised capital allocation process to systematically build slack into their capital budgets. To illustrate the systematic over-budgeting of capital expenditures, Figure 3 plots quarterly planned and actual capital expenditures. The graph shows that the units end up over-budgeting and investing below their investment plans 73% of the time. Confirming the arguments made by Hall (1979, p. 38), our evidence shows that actual investments are usually made within limits of an aggregate capital expenditure ceiling (as imposed by the corporate staff). The management accounting literature calls this pattern budgetary slack. Our findings, which are the first to rely on hard investment data, corroborate previous over-budgeting results that rely on interviews or labouratory experiments (Dunk and Nouri (1998), Onsi (1973), Merchant (1985), Dunk (1993), Young (1985), Chow, Cooper, and Waller (1988), Waller (1988), Stevens (2002), Van der Stede (2000)). However, interview-based measures can lead to biased estimates of budgetary slack and labouratory-based studies may lack external validity. These disadvantages are avoided in our analysis because we have access to hard data resulting from the decision-making process data that form the basis for actual corporate investment decisions Since we have twenty business units and twenty quarters, the maximum number of observations for planned data is 400 unit-quarters. We do not have some data for a few business units at the beginning of our sample period, but the planned data on the main variables used in the paper, including capital expenditures, sales, EBIT, and total assets, is available for at least 359 business-unit-quarters. The coverage of data for planned R&D and marketing expenditures and personnel is not as wide, since these variables were collected only as of 2004 and not at all for two business units The overestimation of sales is puzzling at first glance, as business-unit managers may want to have targets adjusted downward. However, Jensen (2003) points out that the incentive for low-ball targets is not the only factor influencing approved plans. Business-unit managers have to explain to the CEO of the firm why it is impossible to produce a higher net income (p. 381). According to the goal-setting theory in management, the CEO may use high income or sales targets to try to incentivise unit managers. We also cannot rule out that forecast errors may explain the discrepancy between planned and realised sales. Danese and Kalchschmidt (2011), for example, document large cross-sectional heterogeneity in future demand forecast accuracy. 11

12 12 Table II: Business-Unit Planned and Actual Accounting Data This table shows quarterly planned (budgeted) and actual (realised) accounting data for the business units of the conglomerate. The data are obtained from the internal management accounting system of the conglomerate. There are 20 business units in the five divisions of the conglomerate. For each business unit, the numbers reported in each column correspond to the means of quarterly data from January 2002 to December To ensure comparability, the table uses for each business unit and for each variable information if both planned and actual data are available for the same business-unit quarter. The table also reports the significance levels of difference-in-means t-tests between planned and actual data using significance stars. While the later regressions use Winsorised data, this table reports unwinsorised data. n/a indicates that the data are not available. Detailed definitions of each variable are provided in Appendix A. *** indicates significance at 1%, ** indicates significance at 5%, and * indicates significance at 10%.

13 Figure 3. Planned versus Actual Capital Expenditures. This figure provides a scatter plot of planned capital expenditures (x-axis) and actual capital expenditures (y-axis) of the 20 business units of the conglomerate. The figure uses quarterly data from January 2002 to December The figure also contains the 45-degree line to indicate whether planned capital expenditures are above (lower half of the figure) or below (upper half of the figure) actual capital expenditures. Planned capital expenditures are above (below) actual expenditures for 73% (27%) of the observations. III. Managerial Power, Cash Windfalls and Capital Allocation The systematic difference between planned and actual capital expenditures prompts questions about why unit CEOs would ask for more capital than they actually invest. To understand this behaviour, we connect the internal capital allocation process to managerial power and cash windfalls. More specifically, we provide evidence that managers use their overblown capital budgets to justify additional spending at times of financial slack. We show that more powerful and better connected business-unit CEOs obtain substantially higher actual capital allocations for their units when funds from cash windfalls are available. We will later study whether the resulting allocations reflect inefficiencies in internal capital allocations. A. Measuring Power and Connections inside the Firm There were 40 business-unit CEOs working for the firm over our sample period. Only 45% of them were still employed by the firm at the end of the sample period. Our results are based on 53 different manager/business-unit pairs since two managers worked as CEOs of three units and nine managers worked as CEOs of two units at different times. To measure power and connections, we construct six indices for each unit manager. Table III presents averages of these indices for all unit managers in each of the 20 business units. Each index is constructed to take values between zero and one. Management profile data are used in the construction of the first three indices, which are available for all 40 business-unit managers. The detailed description of the variables in each index can be found in the Appendix A. The index of Career at the Firm draws on the management and sociology literature, which underscores the importance of a manager s career at the firm and his or her social network at the workplace (Podolny and Baron (1997), Astley and Zajac (1991). 13 The second index, CEO Similarity, assesses the similarity of a manager s profile to that of the firm s CEO. 14 The third index, Power Index, is formed by averaging the previous two indices to create an aggregate measure In detail, the index is formed by averaging four variables, each normalised to fall between zero and one: (1) the number of months a manager spent in a unit and interacted with a person who later became a member of the board of directors; (2) the number of years a manager has been working at the firm; (3) the number of years a manager has been working in a powerful position at the firm; and (4) a dummy equal to one if a business-unit CEO was appointed during the tenure of the current division CEO. A measure of employment networks similar to the one we use for our first variable is used by Fracassi (2011). For the second and third variables, we assume as in Ryan and Wiggins (2004) or Berger, Ofek, and Yermack (1997) that the power and connections of managers increase as their tenures lengthen and that this effect is particularly pronounced if they are in powerful positions. Finally, the last component of the index helps us measure the connection between the business-unit CEO and the division CEO, since both negotiate for budgets together This index is formed by averaging the following four dummy variables: (1) a dummy equal to one if a manager speaks the native language of the CEO; (2) a dummy equal to one if a manager lives in the country in which the CEO lives; (3) a dummy equal to one if a manager went to the same university as the CEO; and (4) a dummy equal to one if a manager was a student in the same academic discipline as the CEO. The first two variables of the index are inspired by Bandiera, Barankay, and Rasul (2009). We assume that managers are closer to the CEO if they speak the same native language or live in close proximity (see Landier, Nair, and Wulf (2009)). The third and fourth components measure personal connections via shared educational networks and academic backgrounds in an attempt to proxy, as in Cohen, Frazzini, and Malloy (2008) or Duchin and Sosyura (2012), for the relationship between the manager and the CEO. 13

14 Table III: Indices of Managerial Power and Connections This table shows the indices of managerial power and connections for the business-unit CEOs of the conglomerate. There are 20 business units in the five divisions of the conglomerate. The second column of the table shows the number of division CEOs in each of the division of the conglomerate during the sample period (e.g., there were three different division CEOs in division 1 during the sample period). The fourth column shows the number of business-unit CEOs in each of the business units during the sample period (e.g., the third business unit had four different business-unit CEOs during the sample period). The total number of individual division CEOs employed during the sample period is 12 and the total number of business-unit CEOs is 40 (some were employed in more than one unit). The last six columns of the table report the mean values of the six indices of managerial power and connections for all the managers in each business unit and each division of the conglomerate, calculated across the sample period from January 2002 to December n/a indicates that the data are not available. Detailed definitions of each variable are provided in Appendix A. The three remaining indices are drawn from responses to a survey we took with the support of the firm. We distributed a questionnaire to all current and former business-unit CEOs. The survey contained both questions we developed and questions previously used in the management and sociology literature (Podolny and Baron (1997), Forret and Dougherty (2001, 2004)). We guaranteed that the survey answers would be analysed with full confidentiality and anonymity for research purposes only, and that they would not be traceable. In total, 20 business-unit CEOs a response rate of 50% returned the survey. The index Networker measures the extent to which managers engage in networking within the firm. 15 The next variable, Division CEO Connection, measures how well a business-unit CEO is connected to his or her division CEO. 16 Finally, Connected Networker is simply the result of averaging the previous two indices. This index captures the overall degree of the networking abilities of a manager. Table III shows substantial variation both within and across divisions for all power and connections indices. It also shows that there is considerable variation across the six measures within a business unit, possibly indicating that these measures may be capturing different aspects of power and connection. All variations within and across divisions are statistically significant at the 1% level based on Kruksal-Wallis tests This index is formed by averaging the following five dummy variables: (1) a dummy equal to one if a manager is a member of a fraternity; (2) a dummy equal to one if a manager is a member of a social club inside the firm; (3) a dummy equal to one if a manager regularly stops by the headquarters to say Hello ; (4) a dummy equal to one if a manager regularly accepts highly visible work assignments; and (5) a dummy equal to one if a manager regularly participates in highly visible task forces or committees The index is formed by averaging the following variables: (1) a dummy equal to one if a manager named the division CEO as a personal mentor; (2) a dummy equal to one if a manager lunches at least occasionally with the division CEO; and (3) a dummy equal to one if a manager meets the division CEO in person at least every two weeks.

15 B. Managerial Power, Cash Windfalls, and Planned Capital Allocations To study the effects of power and connections in the budgeting phase, Appendix B presents regressions of ex-ante planned investment on our measures of power and connections, cash windfalls (which will be explained below), and a set of measures of investment opportunities. The estimates in Appendix B show that planned investment is significantly related to measures of future growth opportunities (EBIT over assets), indicating that the budgeting process attempts to allocate funds to the units with the best prospects. In terms of economic magnitude, a one standard deviation increase in EBIT over assets implies an increase in planned investment by (0.06*0.0473=) 0.003, which equals about (0.003/0.013=) 24% of the standard deviation of planned investment. These results are in line with neoclassical investment models that suggest that corporate resources should go to the units with the greatest growth opportunities (Stein (1997), Weston (1970), Williamson (1975)). The structured budgeting phase may be designed to neutralise the effects of corporate politics on planned capital budgets, actual capital allocations in the realisation phase may still be distorted by power and connections. We therefore need to take a closer look at the realisation stage, a look that will also help us understand why managers have incentives to over-budget investments. Although it seems difficult to deviate much from planned allocations, in the rest of this section we provide evidence that power may be particularly useful when windfalls arise and ad hoc decisions need to be made. In other words, although the budgeting stage is formally structured to minimise the effects of power, the realisation stage may still allow managers to exercise influence over allocations resulting from cash windfalls. The basis for this conjecture comes from the normative budgeting literature. Predicting when and how power is most likely to matter, Sisaye (1995) argues that in normal times and when resources are scarce, capital budgeting is based on a formalised organisational process where priority is given to technical and economic considerations. By contrast, he suggests that resource allocation decisions become a political process when units bargain over slack organisational resources. Our setting allows us to test this conjecture because our conglomerate experienced eight large exogenous cash windfalls in six different quarters during our sample period. These windfalls resulted from headquarters sales of non-strategic equity stakes in other companies either to cut exposure to those sectors or to exploit market opportunities. As illustrated in Figure 4, the cash windfalls are substantial and are likely to impact the operations of the conglomerate. Depending on the year, the windfalls represent between 24% and 84% of the firm s annual operational cash flow and between 160% and 403% of its annual capital expenditures. As in Blanchard, Lopez-de- Silanes, and Shleifer (1994), large unexpected cash windfalls provide financial slack and give us the opportunity to analyse the behaviour of business units For several reasons, these cash windfalls provide us with a suitable scenario for testing the effects of power on capital allocation. First, the impact of power on the distribution of windfalls, rather than on planned capital budgets per se, mitigates potential reverse causality problems; after all, historically high planned capital allocations could themselves be a source of intra-organisational managerial power. Second, the cash windfalls are largely exogenous to the divisions and their business units because the equity stakes are not part of the unit s assets (they are held and managed by headquarters), and because the decision to sell them is made by headquarters alone. Third, the cash windfalls seem to be unexpected and not anticipated in the budgets: Appendix B shows that planned capital expenditures are not related to the windfalls. Fourth, as Table I shows, the conglomerate is not financially constrained, so the equity sales are unlikely to be triggered by a need of cash to finance opportunities. Finally, our data suggest that the stakes were neither sold for earnings management purposes nor to meet or exceed analysts earnings targets). Roughly 9% of the gross cash windfalls are used for capital expenditure by the business units of the firm. 15

16 Figure 4. Magnitude of Cash Windfalls. The figure documents the size of the cash windfalls relative to the conglomerate s sales, total assets, cash flow from operations, capital expenditures, and EBIT. The conglomerate experienced eight cash windfalls resulting from headquarters sales of equity stakes in other companies. The equity sales occurred in the years 2003 (one sale), 2004 (three sales), and 2005 (four sales). The figure shows the aggregate value of the cash windfalls during each year relative to total aggregate conglomerate values of the same year. Figure 5 provides initial evidence that the cash windfalls seem to be associated with changes in actual unit investment. It shows that, across all business units, actual capital expenditure is substantially closer to the planned budgets during cash-windfall quarters. This effect is statistically significant at the 1% level and uniform across all cash-windfall quarters and most business units. Figure 5. Capital Expenditures in Cash-Windfall and Non-Cash-Windfall Quarters. This figure illustrates the quarterly difference between the business unit s actual and planned capital expenditures over its total assets in cash-windfall and non-cash-windfall quarters. The table reports mean and median values, calculated over the respective business-unit quarters. Negative numbers suggest that actual investment is less than planned investment. The means ( versus ) as well as the medians ( versus ) are statistically different from each other at the 1% level. If managerial power matters for the allocation of cash windfalls, more powerful or better connected unit CEOs should receive larger shares of the windfalls. In other words, the investments ultimately made by their units should be closer to planned investments (or exceed them by more) than the investments made by units run by weaker CEOs. To investigate this hypothesis formally, the regressions reported in Table IV use actual minus planned capital expenditures over total assets as the dependent variable. 18 All regressions control for lagged sales growth and EBIT over assets to This measure is similar to the one used in Rajan, Servaes, and Zingales (2000), but instead of subtracting industry average investment rates from the investment rate of business units, our analysis exploits the availability of planned investment data to adjust investment rates. If planned allocations already reflect the investment opportunities of the business units, our analysis has the advantage that it partially accounts for differences in investment opportunities across units.

17 further account for differences in growth opportunities. In addition, we use business-unit fixed effects and include year dummies. Table IV: Cash-Windfall Regressions: Effects of Managerial Power and Connection Panel A of this table shows OLS regressions for the 20 business units of the conglomerate. The dependent variable in all regressions (shown in the first row) is the quarterly difference between the business unit s actual and planned capital expenditures over its total assets over the period from January 2002 to December We report eight regressions. The first two regressions (1a and 1b) do not include any managerial power index as a control. The following six regressions successively control for the managerial power indices indicated in the second row: Career at the Firm; CEO Similarity; Power Index; Networker; Division CEO Connection; and Connected Networker. The last six regressions also control for the interaction of each of the managerial power indices and Ln(1+Cash Windfall). The coefficient of the interaction term and its standard error are multiplied by 1,000. All regressions, except the second one, also control for Ln(1 + Cash Windfall). All regressions, except the first one (1a), also control for: Sales Growth (lag); EBIT/Total Assets (lag); EBIT Deviation from Plan/Total Assets (lag); and a constant (not shown). The regressions include business-unit fixed effects and year dummies. The number of business-unit quarters and business units in each regression is reported at the bottom of the table. Robust standard errors, clustered at business-unit level, are shown in brackets. The dependent variable and the controls are Winsorised at 1%. Detailed definitions of each variable are provided in Appendix A. *** indicates significance at 1%, ** indicates significance at 5%, and * indicates significance at 10%. Panel B reports the economic effects of the managerial power indices, based on the estimates in Panel A. The economic effects are calculated for an average cash windfall. The change in the dependent variable is calculated both if the managerial power index is low (25th percentile) and if it is high (75th percentile). The increase in the dependent variable is compared to the average quarterly business-unit investment rate (Capital Expenditures/Total Assets), calculated across all non-cash-windfall quarters, which equals

18 The first two specifications in Table IV analyse the impact of cash windfalls and proxies of investment opportunities before introducing our measures of managerial power. Consistent with the pattern presented in Figure 5, the regression estimates in Column (1a) show that the cash windfalls move actual investments closer to those planned. As Panel B of Table IV shows, an average cash windfall increases the difference of actual minus planned investment over assets by Holding fixed planned investment, this estimate suggests that actual investment increases by about this number. Relative to an average quarterly business-unit investment rate in the non-cash-windfall quarters of 0.011, our calculation suggests that an average cash windfall leads to a substantial 62% increase in investment. The estimates in Column (1b) show that EBIT deviation from plan (i.e., actual minus planned EBIT) has a significant positive impact on investment, suggesting that business units that are closer to or above planned EBIT invest more. It also suggests that the internal capital market does not seem to plug cash shortfalls (lower EBITs than planned) of the business units. 19 Most important, the increase in capital expenditures in cash-windfall quarters is far from homogenous from unit to unit and some of the heterogeneity can be explained by different degrees of power and connections of business-unit managers. Each of the Columns (2) to (7) of Table IV includes a different managerial power index, the cash-windfall variable, and an interaction of the two. The estimates of the interaction term show that managerial power and connections have a significant impact on the allocation of cash windfalls. For all six indices, we find that units run by more powerful or well-connected managers receive a substantially larger share of the windfalls. Moreover, we find that even less powerful units receive some of the money from the cash windfalls. The estimated economic effects of power and connections on the cash-windfall distribution are large, as displayed in Figure 6 and calculated in Panel B of Table V. Using Power Index as an illustration of the magnitude, we find that the investment rate of a unit run by a CEO with an index at the 25th percentile of the distribution increases investment by only after a windfall. This change corresponds to an increase of 46% relative to the average non-cashwindfall quarterly investment rate of If, by contrast, a unit is run by a powerful CEO with a Power Index at the 75th percentile, investment increases by , which corresponds to a substantial 66% increase relative to the average non-cash-windfall quarterly investment rate. On average and across all indices, the more powerful and better connected unit managers increase their actual investments by 74% more than their less powerful and connected peers do (33% in terms of percentage points). 20 Figure 6. Economic Effect of Power and Connections. This figure illustrates the economic effects of power and connection. It reports the change in capital expenditures relative to planned capital expenditures (i.e., the change in actual minus planned capital expenditures, standardised by total assets) when our proxies for power or connection are low and high, respectively. A power or connection proxy is considered low (high) if the variable is at the 25th (75th) percentile. The changes in capital expenditures are calculated for an average cash windfall. The figures are calculated based on the estimates in Table V and correspond to those reported in Panel B of Table V This is consistent with the findings in Shin and Stulz (1998), who show for other conglomerates that internal capital markets do not fully insulate a division s investment from its own operating cash flows Glaser, Lopez-de-Silanes and Sautner (2013) provide a series of robustness tests of the main findings shown in this paper. First, they show that our results are robust to the use of different specifications and controls. Second, they run a series of tests to control for managerial reputation and to mitigate concerns that our power and connections measures are proxies for ability. Third, they show that the findings are not the result of the endogenous allocation of powerful managers to units with the best investment opportunities. Finally, tests for the influence of the divisional CEOs in the capital allocation process show only mild evidence in favour, consistent with the fact that business units are largely independent.